Digital Telephony

WI LEY SERI ES I N TELECOMMUNI CATI ONS

AND SI GNAL PROCESSI NG
John G. hoakis. Editor
Northeastem University
Introduction to Digital Mobil Communications
Yoshihiko Akaiwa
Digital Telephony, Sril Etlition
John Bellamy
E lements of I rfonuttion Theory
Thomas M. Cover and Joy A, Thomas
Fundame ntals of Te lecommunicat ions
Roger L. Freeman
P ractic al Data C ommunicat ions
Roger L. Freeman
Radio System Design
for
Telecommunications, Znd Edition
Roger L. Freeman
Telecommunication System Engineering, 3 rd Edition
Roger L. Frceman
Te lec ommunications Transmission Handb ook, 4th Etlition
Roger L. Frceman
Introduction to Communications Engineering, 2nd Edition
Robert M. Gagliardi
Optital C ommunications, Znd Edition
Robert M. Gagliardi and Sheman Ksxp
Active Noise Control Systemt: Algorithm"s and DSP Implementations
Sen M. Kuo and Dennis R. Morgan
Mobile Communications Design Fundamentals, 2nd Edition
William C, Y, Lee
Expen System Applications
for
Telecommunications
Jay Liebowitz
Digital Signal Esilrndtion
Robert J. Mammone, Editor
Digital Communication Receivers: Synchronization, Channel Estimation, and Sigtnl Processing
Heinrich Meyr, Marc Moeneclaey, afld Stefan A, Fechtel
Synchronization in Digital Comntunications, Volume I
Heinrich Meyr and Gerd Ascheid
Business Earth Stations
for
Te lecommunications
Walter L. Morgan and Denis Rouffet
W irele s s I nfo rmat ion N etw o tk
Kaveh Pahlavan and Allen H. lcvesque
Satellite Communicationt: The First
Quarter
Century of Senice
David W. E. Rees
Fundamentals of Te Ie communicat fun N etw orks
Tarek N. Saadawi, Mos'tafa Ammar, with Ahmed El Hakeem
Meteor Burst Communicalions: Theory and Practice
Donald L, Schilling, Editor
Vector Space Projections: A Numerical Approar:h to Signal and Image Processing, Neural Nets, and,
Optict:
Henry Stark and Yongyi Yang
Signaling in Telecommunitation Networl<s
John C. van Bosse
Te Ie communication C ircuit D e s i gn
Putrick D. van der Puije
Worldwide Telecommunications Guide
for
the Business Manager
Walter H, Vignault
P o lynomial S igrwl P roc e s s ing
V. John Mathews and Ciovanni L. Sicuranza
ADSL, VDSL, and Multicatier Modulation
John A. C. Bineharn
( (
L
Digital
Telephony
Third Edition
John
C. Bel l amy
CoPPell,
TX
glgf ef$*. r:rll6lnt jt f
ru adrar*hr
ITnn"f
r.r.R.c.
A Wiley-lnterscience
Publication
JOHN
WILEY & $ONS,l NC.
New York
.
Chichester
r
Weinheim
.
Brisbane
r
Singapore
r
TOronto
-l*
n
5
\oE
.1
r t
' B
b\.ts
2
ooo
This book is printed
on acid-free paper,

Copyright @ 2000 by John Wiley & Sons, Inc,

All rights reserved. Published simultaneously in Canada.
No part
of this publication
may be reproduced, stored in a retrieval system or transmitted in any form or
by any means, electronic, mechanical, photocopying,
recording, scurning or otherwise, except as permit.
ted by Sections 107 or 108 of the 1976 United States Copyright Act, without either the prior
written per-
mission of the Publisher, or authorization thmugh payment
of the appropriate per-copylee
to the
copyright clearance center, 222 Rosewood Drive, Donvers, MA 0t9?3, (s0g)
7j0-g400, fax (50g)
750_
4?44. Requests to the Publisher for permission
should be addressed to the Permissions Depaxtrnent, John
wiley & sons, ftrc., 605 Third Avenue, New
york,
Irry 10158-0012, (212) 8s0-601l, fax
(zl?)
850-
6008, E-Mail: PERMREQ @ WILEY.COM.
For ordering and customer service, call I-800-CALL-WILEY.
lihrary of Congress Cataloging-in-Puhlfuation
Data:
Bellamy, John, l94l*
Digital telephony / John Bellamy.*3rd
ed.
p.cm,- (Wiley
series in t'elecornmunications and signal processing)
"A
Wiley-Lrterscience publication."
Includes index,
ISBN 0-471-34571-7
l. Digital telephone systems. I. Title. IL Series,
TK5103.7.844 2000
6?1.385-dc2l
99-34015
Printed in the United Ststes of America
1 0 9 8 7 6 5 4 3 2 1
To my
father for
passing on the enioyment of being an engineer
CONTENTS
xvll
xtx
Preface
Acknowledgment
Acronyms
/
chapter 1 Background and Terminology
1.1 Telecommunications
StandardOrganizations
3
1.? The Analog Network HierarchY 5
1.2.I Bell SYstem HierarchY 6
I.2.2 Postdivestiture U.S. Network 10
L23 Switching SYstems 12
I-2.4 Transmission SYstems
l8
"1.2.5
Pair-Gain SYstems 24
1.2.6 FDM Multiplexing and Modulation 26
1.2.7 Wideband Transmission Media 28
1.2.8 Transmission Impairments 33
I'2.9 Powerkvels 4l
1.2.10 Signaling 42
1.2.11 Analog Interfaces 46
1'Z.lZ The Intelligent Network 49
1.2.13 Dynamic Nonhierarchical
Routing 51
I.2.14 Cellular Radio Telephone System 52
1.2.15 Voiceband Data Transmission
54
1.3 The Infioduction of Digits 56
1.3.1 Voice Digitization 56
I.3.2 Time Division Multiplexing 58
xxl
VI I
CONTENTS
1.3.3 Dataunder Voice 63
1.3.4 Digiral Microwave Radio ffi
1.3.5 Fiber Optic Transmission
65
1.3.6 Digital Switching 65
1,.3.7 Digital Nerwork Evolution 67
References 69
Problems 7l
Chapter 2 Why Dlgital?
2.1 Advanrages of Digital Voice Networks 7j
.
2.1.1 Ease of Multiplexing 7j
2.1.2 Ease of Signaling i4
2.1.3 Use of Modern Technology 75
2.1.4 Inregration of Transmission
and Switching 77
2.1.5 Signal Regenerarion
78
2.1.6 PerformanceMonitorability
79
2.1.7 Accommodation of Other Services
g0
2. 1.8 Operability at Low Signal-to-Noise/Interference
Ratios 80
2.1.9 Ease of Encryption 8l
2.2 Digital Signal Processing 8l
2.2.1 DSP Applications Bz
2.3 Disadvantages of Digital Voice Networks
g4
2.3.1 Increased Bandwidth 84
2.3.2 Need for Time Synchronization
85
2.3.3 Topologically
Resrricred Multiplexing
g5
2.3.4 Need for Conference/Extension
Bridges
g6
2.3.5 ftrcompatibilities
with Analog Facilities
g7
References 88
Chapter 3 Volce Digitizatlon
3.I Pulse Amplitude Modulation 93
3.1.1 Nyquist Sampling Rate 94
3.1.2 Foldover Distortion
95
3.2 Pulse Code Modulation 98
3.2.1
Quantization
Noise 99
3.2.2 Idle Channel Noise l0Z
3.2.3 Uniformly
Encoded
pCM
103
73
91
CONTENTS iX
3.?.4 Companding 106
3.2.5 Easily Digitally Linearizable Coding 108
3.2.6 Syllabic Companding 116
3.2.7 Adaptive Gain Encoding 119
3.3 Speech Redundancie$ 121
3.3.1 Nonuniform Amplitude Distributions 122
3.3.2 Sample-to-SampleCorrelation
I22
3.3.3 Cycleto-CycleCorrelations
122
3.3.4 Pitch-Interval-to-Pitch-Interval
Correlations
123
3.3.5 Inactivity Factors 124
3.3.6 Nonuniform Long-Term Specnal Densities IZ4
3.3.7 Short-Term Spectral Densities 127
3.4 Differential Pulse Code Modulation 127
3.4.1 DPCM Implementations 129
3.4.2 Higher Order Prediction l3l
3.4.3 Adaptive Differential PCM 131
3.5 Delta Modulation 133
3.5.1 Slope Overload 134
3.6 Adaptive Predictive Coding 136
3.7 Subband Coding 138
3.8 Vocoders 141
3.8.1 Channel Vocoder 142
3.8.2 Formant Vocoder lM
3.8.3 Linear Predictive Coding 144
3.8.4 Enhanced-Excitation Linear Predictive Coding 147
3.9 Encoder/DecoderSelectionConsiderations
151
3.9.1 Voice
Quality
151
3.9.2 Transparency for Nonvoice Signals 152
3.9.3 Tolerance of Transmission Enors 153
3.9.4 Delay 154
3.10 ITU-T Coding Standards
154
References 155
Problems 158
Ghapter 4 Dlgital Tranemission and Multlplexing
4.1 Pulse Transmission 162
4.1.1 Intersymbol Intederence 164
4.1.2 Timins Inaccuracies 164
161
X CONTENTS
4.1.3 InsufficientBandwidth l&
4.1.4 Amplitude Distortion 165
4.1.5 Phase Distortion 165
Asynchronous versus Synchronous Transmission
4.2,.1 AsynchronousTransmission
166
4.2.2 SynchronousTransmission
I6i
Line Coding L7l
4.3.1 Level Encoding L7I
4.3.2 Bipolar Coding 173
4.3.3 Binary N-Zero Substirution 176
4.3.4 Pair Selected Ternary L19
4.3.5 Ternary Coding 180
4.3.6 Digital Biphase 181
4.3.7 Differential Encoding 183
4.3.8 Coded Mark Inversion 183
4.3.9 Multilevel Signaling 184
4.3.10 Partial-Response
Signaling 185
Eror Performance 189
4.4.1 Signal Detection 190
4.4.2 Noise Power 190
4.4.3 Enor Probabilities 191
PerformanceMonitoring
198
4.5.1 Redundancy Checks 198
4.5.2 Signal
Quality
Measurements 201
4.5.3 Framing Channel Errors 203
4.5.4 Performance Objectives 2O3_
4.5.5 Forward Error Correction 2O4
Time Division Multiplexing 207
4.6.I Bit Interleaving ver$us Word Interleaving
4.6.2 Framing 209
4.6.3 DSI Extended Superframe Zl5
4.7 Time Division Multiplex Loops and Rings 216
References 219
Problems 221
Chapter 5 Digltal$witching
ZZs
5.1 Switching Functions ZZE
5.2 Space Division Switching 227
5.2.1 Multiple-StageSwitching
Z3O
165
4. 2
4.3
4.4
4.5
4.6
208
Chapter 6
CONTENTS
5.2.2 Blocking Probabilities: L.ee Graphs 234
5.2.3 Blocking Probabilities: Jacobaeus 238
5.2.4 Folded Four-Wire Switches 242
5.2.5 Pathf,rnding 243
5.2.6 Switch Matrix Control 24
5.3 Time Division Switching 246
5.3.1 Analog Time Division Switching 246
5.3.2 Digital Time Division Switching 247
5.4 Two-Dimensional Switching 251
5.4.1 STS Switching
255
5.4.2 TST Switching 257
5.4.3 No. 4 ESS Toll Switch 262
5.4.4 System 75 Digital PBX 264
5.5 Digital Cross-Connect Systems 265
5.5.1 Consolidation and Segregation 267
5.5.2 DCS Hierarchy 268
5.5.3 Integrated Cross-Connect Equipment 269
5.6 Digital Switching in an Analog Environment
27O
5.6.1 Zero-Loss Switching 270
5.6.2 BORSCHT 272
5.6.3 Conferencing
272
References 273
Problems 274
Dlgltal Modulatlon and Radlo Sy$tsms
6.1 Digital Modulation 279
6.1.1 Amplitude Modulation 280
6.1.2 Frequency Shift KeYing 284
6.1.3 Phase Shift Keying 288
6.I.4
Quadrature
Amplitude Modulation 30I
.
6.1.5 Carrierless Amplitude and Phase Modulation 309
6.1.6 Partial-Response
QAM
310
6.1.7 Trellis-CodedModulation
311
6.1.8 MulticarrierModulation
315
6.2 Filter Partitioning 317
6.2.1 Adjacent-Channellnterference
3I7
6.2.2 OptimumPartitioning
318
6.3 Emission Specifications 320
277
xll coNTENTS
6.4 Radio Systern Design 322
6.4.1 Fade Margins
jZZ
6.4.2 System Gain 323
6.4.3 Frequency Diversity 3.26
6.4.4 Space Diversity 327
6.4.5 Angle Diversity 327
6.4.6 Adaptive Equalization 3ZB
6.4.7 Route Design 3ZB
References 329
Problems 332
,,/ Chapter 7 Network Synchronization
Controland
Management
335
7.1 Timing 336
7.1.1 Timing Recovery: Phase-Locked Loop 336
7.1.2 Clock Insrabitity 337
7.I.3 Elastic Stores 339
7.1.4 Jitter Measurements
j4Z
7.1.5 Systematic Jitter 345
7.2 Timing Inaccuracies 346
7.2.1, Slips 346
7.2.2 AsynchronousMultiplexing
351
7.2.3 Waiting Time Jitter 359
7.3 Network Synchronization 361
7.3.I Plesiochronous
362
7.3.2 Networkwide Pulse Stuffing 363
7.3.3 Mutual Synchronization
3M
7.3.4 Network Master 364
7.3.5 Master-SlaveSynchronization
365
7.3.6 Packetization 366
7.3.7 Network Timing Performance Measurements
366
7.4 U.S. Network Synchronization
370
7.4.1 Synchronization Regions 370
7.4.2 Primary Reference Sources 372
7.4.3 1996 AT&T Synchronization Architecrure
j73
7.5 Network Conhol 373
1.5.1 Hierarchical Synchronization Processes 374
7.6 Network Management 376
coNrENrs xlll
7.6.1 Routing Control 376
'7.6.2
Flow Control 377
References 380
Problems 382
383
-6apter I Fiber Optlc Transmisslon Systems
Ll Fiber Optic Transmission System Elements 386
8.1.I Optical Fiber Fundamentals 387
8.1.? Electrical-to-Optical Transducers 390
8.I.3 Optical-to-ElectricalTransducers
393
8.2 Line Codes for Fiber Optic Transmission 395
8.2.I mBnB Line Codes 396
8.2.2 Bit Insertion Codes 399
8.3 Wavelength Division Multiplexing 401
8.4 Fiber System Design 403
8.4.1 Fiber Connectors and Splices 404
8.4.2 Protection Switching 404
8.4.3 System Gain 405
8.5 SONET/SDH
406
8.5.1 SONET Multiplexing Overview 408
8.5.2 SONETFrameFormats
409
8.5.3 SONET Operations, Administration, and
Maintenance 4l I
8.5.4 Payload Framing and Frequency Justification 4I4
8.5.5 Virtual Tributaries 417
8.5.6 DS3 PaYload MaPPing 422
8.5.7 E4 PaYIoad MaPPing 423
8.5.8 SONET OPtical Standards 425
8.5.9 SONET Networks 426
8.6 SONET Rings 429
8.6.1 Unidirectional
Path-Switched Ring 4Zg
8.6.2 Bidirectional Line-Switched Ring 43I
References 433
Problems 434
Chapter I Digital Moblle Telephony
437
9.1 North American Digital Cellular 437
9,1.1 D-AMPS Transmission Format 438
xlv coNTENTS
9.1.2 D-AMPS Speech Coding 43l.9
9.1.3 D-AMPS Control Channel 439
9.1.4 D-AMPS Error Conrrol 440
9.2 Global System for Mobile Communications 44I
9.2.1 GSM Channel Structure Ml
9.2.? GSM Speech Coding M3
'
9.2.3 GSM Channel Coding and Modulation 443
9.2.4 GSM Mobile Station 443
9.2.5 GSM Frequency Hopping 444
9.2.6 GSM Short Message Service 444
9.3 Code Division Multiple-Access Cellular 444
9.3.1 CDMA Channel Establishment 445
9.3.2 CDMA Multipath Tolerance MB
9.3.3 CDMA
power
Conhol M9
9.3.4 CDMA SoftHandoff 449
9.4 Personal Communication System 450
9.5 Volce Privacy and Authenticarion
450
9.6 Iridium 45I
9.7 Trunked Radio 45?
9.8 Cellular Digiral Packet Dara 453
References 453
Problems 454
Chapter 10 Data and Asynchronous
Transfer Mode
Networks
10.I Message Switching 456
10.2 Packer Switching 458
10.2.1 Packet Formats 460
1O.2.2 StatisticalMultiplexing 461
IO-2.3 Routing Control 46j
10.2.4 Flow Control 466
10.2.5 X..25 46e
10.2.6 Frame Relay 471
10.2.7 TCP/rP 473
10.3 Asynchronous Transfer Mode Networks 474
10.3.1 ATM Cells 474
10.3.2 ATM Service Categories 474
10.3.3 ATM Connections 4i7
455
CONTENTS XV
10.3.4 ATM Switching 477
10.3.5 ATM Applications 484
10.4 Internet Protocol Transport 490
References 492
Problems 494
'/Chapter
11 Dlgital $ubscriber
Accese
495
I 1. I Integrated Services Digital Network 496
11.1.1 ISDN Basic Rate Access Architecture
497
11.1.2 S/Tlnterface
499
I Ll.3 ISDN U Interface 501
11.1.4 ISDN D Channel Protocol 503
Il.2 High-Data*Rate Digital Subscriber loops 503
11.2.1 Asymmetric Digital Subscriber Line
503
rr.2.2 VDSL 507
11.3 Digital Loop Carrier Systems 507
I L3.1 Universal Digital Inop Carrier Systems 507
11.3.2 lntegrated Digital Loop Carier Systems 508
1l .3.3 Next-Generation Digital Loop Carrier
Systems 509
11.4 Fiber in the LooP 510
I L5 HYbrid Fiber Coax SYstems 5l I
I 1.6 Voiceband Modems 512
11. 6. 1 PCMModems 513
II.7 Local Microwave Distribution Service 515
I1.8 Digital Satellite Services 516
References 516
Problems 5I7
/Cnapter tZ Traffic Analysis
519
12.1 Traffic Characterization 520
12.1.1 Arrival Diskibutions 5M
12.1.2 Holding Time Distributions 527
12.2 Loss Systems 530
12.2.1 Lost Calls Cleared 531
12.2.2 Lost Calls Returning 536
12.2.3 Lost Calls Held 539
12.2.4 Lost Calls Cleared-Finite Sources 54I
xv| CoNTENTS
12.2.5 Lost Calls Held-Finite Sources 544
12.3 Network Blocking hobabilities 547
12.3.I End-to-End Blocking
probabilities
549
I2.3.? Over{tow Traffic 551
12.4 Delay Systems 552
12.4.1 Exponential Service Times 555
.
12,.4.2 Constant Service Times 559
l?,.4.3 Finite
Queues
561
12.4.4 Tandem
eueues
566
References 567
Problems 568
Appendlx A Derlvatlzation of Equations
Appendix B Encodlng/Decodlng
Algorlthms for
segmgnted PcM
Appendk G Anatytic Fundamentals
of DigitalTransmisslon
Appendix D Traffic Tabtes
Gloseary
Answers to Selected Problem*
lndex
573
587
579
607
613
631
635
PREFACE
As mentioned in the
preface of the first two editions, the term digffa I telephony tefers
to the use of digital technology
in the message
path of voice communications
net-
works. In this case the tetm dlgital refers to a method of encoding the signal-that
is,
a form of modulation. Hence digital telephony implies voice transmission and switch-
ing applications,
not data communications.
Although the primary focus of this book
is not data communicafions, this edition contains an expanded treatment of data com-
munications networks,
particularly as they relate to
providing voice communications
services in addition to data communications.
This book covers all aspects of digital voice communications
technology and net-
works. It is not a technical book in the traditional, analytical sense of communications
theory. Since numerous books covering communications theory are already available,
this book stresses the application and operational aspects of communications
system
design. Some basic theory is
presented in both
qualitative and, when appropriate,
quantitative terms. The main
putpose, however, is to introduce concepts, terminology,
and how applications influence implementations.
[n most cases the concepts are sup-
ported by citing example implementations in the U.S. telephone network, although ex-
amples from other
(ITU) public telephone networks are also
provided'
The primary audience for this book are
graduate electrical engineers. The electrical
engineering student is most capable ofappreciating occasional references to corffnu-
nications theory and its influence on the
practice. However, because analytical rigor
is waived in favor ofoperational descriptions,
less analytically oriented readers should
have no difficulty understanding
the principles. Chapter 6
(covering digital radio and
modulation) is the most analytical but is easily skipped without losing continuity for
the other chapters. Similarly, Chapter 12
(covering traffic analysis) contains
numerous
equations that are unneces$ary for understanding
the material in other chapters.
When the first edition of Digital Telephony was written
(1980)' public telephone
networks around the world were primmily implemented with analog technology, but
it was clear that digital technology was rapidly taking over' When the second edition
was written
(1990),
the inlemal
portions of the network had, for the most
part, been
converted to an all-digital network. Then and today
(1999) the main remnants of the
xvll
xviii PREFAcE
original analog telephone networks are analog subscriber loops and analog telephones
connected to them.
Although Integrated services Digital Nerwork (ISDN)
technology was developed
as a means of replacing analog loops to complete the transformation of the network to
suppofi end-to-end digital connections, ISDN deployment is below expectations for
several reasons. One of these reasons is a
growing
need formore bandwidth than what
is available from a basic rate IsDN subscriber loop (128 kbps). There is currenrly
much activity within the industry to develop new technologies for medium- and high-
bandwidth digital subscriber access. A new chapter (chapter
l1) has been added to
this edition to specifically address alternative technologies for digital subscriber ac-
CESS.
Another relatively recent application of digital technology added to this edition
involves digital cellular telephones, which first appeared in the marketplace in the
mid-1990s.
Digital mobile radio is enabled by the emergence of low-cost, high-
performance
digital signal processing (DSp)
technology for compressing speech sig-
nals to low bit rates and for providing
sophisticated coding, modulation, and
equalization required for digital radios in a bandwidth,constrained
mobile application.
A complete list of chapter topics is;
chapter l: overview of analog telephone technology followed by an inhoduction
of how digital technology is used ro fulfrll the same functions
chapter 2; Discussion of advantages and disadvantages of digital technology for
voice communications
chapter 3: Descriptions of the most comrnon voice digitization algorithms
chapter 4: Fundamenral$
of digiral wire-line rransmission
and multiplexing
chapter 5: Basic concepts and operations of digital switching machines
Chapter 6: Digital modulation and radio fundamentals
chapter 7; Network synchronization, control, and management requirements
Chapter 8: Fiber optic transmission systems and SONET
Chapter 9: Digital cellular telephone systems
Chapter l0: Data networks
Chapter 1l: Digital subscriber access technology
Chapter 12: Fundamentals
of traffic analysis for designing networks
The appendices cover the derivation ofequations,
pcM
voice coding relationships,
fundamentals
of digital communications theory, and traffic tables.
Coppell, Tems
October 1999
j
ohnc b e I lamy @ ie e e. o r g
Joruq C. BeLLnr4y
ACKNOWLEDGMENT
Once again I am indebted to Wanda Fox and Alcatel USA for allowing me access to
the corporate library for research materials for this edition. I also owe a great deal of
gratitude to Gerald Mitchell of the University of Colorado for thoroughly
reviewing
and enhancing the last chapter on traffic theory'
J.B,
xl x
AAL
ABR
ACD
ACK
ADM
ADPCM
AIN
AMPS
ANI
APC
APON
ARPANET
ARQ
ATM
ATPC
BSZS
BER
BISDN
BLSR
CAC
CAP
CAS
CBR
ccc
CCIS
ccn"T
CCS
CDMA
CDO
CDPD
ACRONYMS
ATM adaptation layer
available bit rate
automatic call distributor
acknowledgment
(Positive)
adaptive delta modulation; add-drop multiplexer
adaptive differential PCM
advanced intelligent network
advanced mobile
phone sYstem
automatic number identification
adaptive
predictive coding
ATM based
passive optical network
advanced research
projects agency network
automatic repeat request
asynchronous transfer mode
adaptive transmit
power control
binary 8 zero substitution
bit-enor rate
broadband integrated services digital network
bidirectional line switched ring
connection admission control
competitive access
provider; carrierless amplitude/phase modulation
channel associated signaling
constant bit rate
(AfUl
clear channel capability
coilrmon charurel interoffice signaling
Consultative Committee
for Intemational Telephony and Telegraphy
(now tTU)
cornmon channel signaling; hundred call seconds
code division multiple access
community dial office
cellular digital
packet data
(for AMPS networks)
xxll AcHoNyMS
CELP code excited linear prediction
CES circuit emulation service
(ATM)
CGSA
Cellular Geographic Service Area
CLASS cu$tom local area signaling services
CLEC competitive local exchange carrier
CLP cell loss priority (ATM)
CMI coded mark inversion
CODEC CODer/DECoder
CPFSK continuous phase
frequency shift keying
CRC cyclicredundancycheck
CSMA/CD carrier sense multiple access/collision detection
CSU channel service unit
CTD cell kansfer delay
CTI computer telephony integration
D-AMPS
digital advanced mobile phone
service
DAVIC digital audio video council
DBS direct broadcast satellite
DCM digital circuit multiplication
DCME digitalcircuitmultiplicationequipment
DECT digital enhanced cordless telephony
DFE decision feedback equalization
DID direct inward dialing
DLC
digital loop carrier
DM
delta modulation; degraded minute
DMT discrete multitone
DNIS dialed number identification
service
DPCM differential pulse
code modulation
DQDB distributed queue
dual bus
DSI
digital signal level I at 1.544 Mbps
DS3
digital signal level3 at,14.736 Mbps
DSI
digital speech interpolation
DSS
digital satellite system
DTE data terminal equipment
DTMF dual tone mulrifrequency (signaling
tones)
DVB
digital video broadcasting group
DVD digital video disc
El European digital signal level I at ?.048 Mbps
E3 European digital signal level 3 at 34.368 Mbps
ECMA EuropeanComputerManufacturersAssociation
EMI elechomagnetic interference
ERMES
enhanced radio message system
ESF
extended super frame
ETSI European Telecommunications
Standards Institute
FDDI fiber dishibuted data inrerchange
ACRONYMS XXIii
FDM frequency division multiplexing
FEC forward error correction
FEXT far end crosstalk
FIFO first in-first out
FRAD frame relay access device
FSK frequency shiftkeying
FTTC fiber to the curb
FTTH fiber to the home
GPS
global positioning sYstem
GSM
global system for mobile communications
HDB3 high density bipolar of order 3
HDLC high-level data link control
HIPPI high
performance parallel interface
HTTP hypertext transport
protocol
IDLC integrated digital loop carrier
IEC International Electrotechnical Commission
IETF internet engineering task force
ILEC incumbent local exchange carrier
IMT international mobile telecommunications
IP internet
Protocol
ISDN integrated services digital network
ISI intersymbol interference
ISO Intemational Standards Organization
ITU International Telecommunications Union
M interactive voice resPonse
JPEG Joint Photographic Experts Group
LAN local area network
LATA local access transPofl area
LD-CELP low-delay CELP
LEC local exchange carrier
LMDS local microwave distribution
service
MAN metropolitan area network
MCM multicarrier modulation
MLCM multilevel coded modulation
MMDS multichannelmultipointdistributionservice
MPEG Motion Pictures Experts Group
MPLS multiprotocol
label switching
MSK minimum shift keYing
MTIE maximum time interval enor
MTSO mobile telephone switching office
MULDEM multiplexer-demultiplexer
NAK acknowledgment
(negative)
NCP network control
point; network control
protocol (ARPANET)
NEXT near end crosstalk
XXIV ACRONYMS
NMT
NNI
NRZ
OFDM
OSI
PABX
PAM
PBX
PCM
PCME
PCR
PCS
PDC
PDH
PHS
PLL
PON
POTS
PRC
PRK
PRS
PSK
PSTN
PVC
Qos
QPRS
QPSK
RADSL
RCC
SDH
SDLC
SES
SF
SIM
SLIC
SMDS
SMR
SMS
SNMP
SOHO
SONET
SRTS
S57
STM
Nordic mobile telephone system
network*to-network
i nterface
nonreturn to zero
orthogonal frequency division multiplexing
open systems interconnection
private
automatic branch exchange (also
pBX)
pulse
amplitude modulation
private
branch exchange
pulse
code modulation
packet
circuit multiplication equipment
peak
cell rate (ATM)
personal
communication system (or
service)
personal
digital cellular (Japan)
plesiochronous
digital hierarchy
personal
handyphone sy$tem (Japan)
phase
locked loop
passive
optical network
plain
old telephone service
primary reference clock
phase
reversal keying
partial
response signaling; primary
reference source
phase
shift keying
public
switched telephone network
permanent
virtual circuit
quality
of service
quadrature partial response
signaling
quaternary phase
shift keying (4-PSK)
rate adaptive digital subscriber loop
radio common carrier
synchronous digital hierarchy
synchronous
data link control
severely errored seconds
super frame
Subscriber Identification Module (GSM)
subscriber loop interface circuit
switched multimegabit data service
specialized mobile radio
short message service
simple network management protocol
small office/home office
synchronous
optical network
synchronous residual time stamp
signaling system version 7
synchronous transfer mode
ACRONYMS XXV
STS
STS-n
SVC
TI
T3
TACS
TASI
TCM
TCPAP
TDM
TDMA
TETRA
TMN
TST
UBR
UDP
UI
UMTS
UM
UPSR
URL
UTP
VBR
VCC
VCI
VPC
VPCI
WAN
WATS
space-time-space digital switching structure
synchronous transport signal-n
switched virnral circuit
TDM transmission system at 1.544 Mbps
TDM hansmission system at 44'736 Mbps
total access communications
system
time assignment speecb interpolation
trelli$ coded modulation
transmission control
protocol/internet
protocol
time division multiPlexing
time division multiple access
Trans-European trunked radio
Telecommunications Management
Network
time-space-time
digital switching $tructure
unspecified bit rate
User datagram
Protocol
unit interval
universal mobile telephone service
user-to-network
interface
unidirectional
path switched ring
universal resource locator
unshielded twisted
pair
variable bit rate
virtual channel connection
virtual channel identifier
virtual
path connection
virtual
path connection identifier
wide area network
wide area telecommunications
services
1
BACKGROUND
AND TERMINOLOGY
Beginning in the 1960s, telecommunications
in the United States and around the world
Uefan undergoing radical changes in several different areas' First, the conventional
*ulog telephone network was being called upon to
provide many new and different
services. most of which emanated from the dataprocessing industry' Second, the mar-
ketplace and the regulatory agencies in the United States stimulated competition in
both old and new areas of traditionally
monopolistic services. Third, digital technol-
ogy emerged to implement many of the fundamental transmission and switching func-
tions wittrin ttre U.S. telephone network and other networks around the world' The
main
purpose of this book is to describe the design, application, and operational as-
pects of tfis new digital equipment. As background,
the technology of the analog tele-
phone network is reviewed to
provide a framework for the introduction of digital
equipment.
limust be emphasized that the introduction of digital technology into the telephone
network was motivated by desires to improve the
quality, add new features, and re-
duce the costs ofconventional
voice services. Digitization ofthe network did not arise
from the needs of the data
processing industry for better data hansmission
services'
Indeed, most of the digital technology
introduced into the network was initially inac-
cessible to data fiaffic, except through analog channels' Of course, a digital network
is a natural environment for data communications
services. As more of the network
became digitized, more rtupport for direct use of the facilities became available for data
applications. Initially, direct digital access exi$ted only for relatively high-end busi-
ness applications. It was not until facilitiesrof the Integrated Services Digital Network
(ISDtiibecame available that end-to-end{switched
digital channels could be used by
individual subscribers for both voice and'datp,l
ny tfre late 1990s numerous other ap-
proaches ro providing digital access to digital facilities became available,
primarily for
Intemet access. These various digital access technologies are described in Chapter 1l'
.A$ a point of historical reference, Figure l. I has been included to show
that the idea
of integrated voice and data is not new. This figure depicts a concept of
a German in-
ventor named Phillip Reis
[1]
to add voice communications to the
prevailing means
of electrical communications at the time-the telegraph' Reis developed the equip-
ment in tfie 1860s and died in 1874-two
years before Alexander Graharn Bell re-
2 BAcKcRoUNDANDTERMtNoLocy
ceived his patent
for the telephone. As indicated, the figure implies altemate use of the
wires for voice or data
cornmunications (i.e.,
integrated
transmission). Reis actually
used the telegraph attachment to signal infbrmation pertaining
to voice tests, an indi-
cation of inadequate voice quality.
To implement simultaneous voice and telegraph communications,
the telephone in
Figure 1.1 would have to have been digital. Because of technology limitations at the
time, such an implementation
was impossible and telephone systems necessarily
evolved with analog technology. one hundred years
later the situation chrurged sig-
nificantly. Telephone equipment developers and service providers
had an abundance
of new technology, and they were challenged
with how to make effective use of it.
This book describes
digital telephone technology from two perspectives.
The first
perspective
describes individual equipments
or subsystems and technical reasons for
transitions from conventional
analog equipment to seemingly less natural digital
counterparts. Thus, one purpose
ofthis book is to describe how digital technology im-
proves
and expands the capabilities
of various subsystems within voice telephone net-
works' Another purpose
of the book is to describe the ultimate benefits derived when
an entire network is implemented with digital techniques. A great
degree of synergism
exists when individual
systems are designed into one cohesive network utilizing aigi-
tal implementations
throughout. The synergistic
effect benefits conventional voice
services and newer $ervices such as ttre in-G:iiiei.
Flgure 1.1 Back to the future: the first integrated voice/data communication
$vstem.
1.1 TELECOMMUNICATIONSSTANDAHDORGANIZATIONS
3
Most of the equipment descriptions urd design examples
presented in this book
come from material authored by engineers at AT&T Laboratories
(now Lucent Tech-
nologies) and other suppliers for the public telephone network. The basic
principles,
however, are by no means unique to the
public telephone network. The concepts and
implementation examples are applicable to any communications
network:
public or
private, voice or data. An inherent attribute of a digital network is that it can, to a large
extent, be designed independently of its application.
Terminals, Transmiesion, and Swttchlng
The three basic elements of a communications network are terminals, tran$mission
systems, and switches. The first part of this chapter
provides an overview
of these ele-
ments as implemented in analog telephone networks. Then, the last
part of this chapter
provides a,lrief overview of digital implementations within the analog network. Fol-
lowing a detailed discussion of the motivation for digital implementations
in Chapter
2, the next four chapters describe the operation and design of the basic elements of a
digital voice network. Chapter 3 discusses digital voice tetminals and the most com-
mon algorithms used to convert analog voice signals into digital bit streams. Chapter
4 presents the basics of digital transmission sy$tems. Fundamentals of digital switch-
ing follow in Chapter 5. Basic digital modulation techniques and their application to
point-to-point digital microwave and digital cellular systems are described in Chapter
6. A discussion of various synchronization and control considerations for digital net-
works is provided in Chapter 7. Chapter I describes fiber optic transmission
systems
and the synchronous multiplexing stiurdard
(SONET)' Chapter 9 discusses
the basic
architecture and operation of
prevailing digital cellular $ystems in use in the United
States and around the world.
The main emphasis of the first nine chapters involves circuit switching as tradition-
ally implemented for voice telephone networks. A circuit-switched
network is onp_that
assigns a complete end-to-end connection in response to each request for service\Each
connection, with its asrrociated network facilities, is held for the duration of the'call.
Chapter l0 describes a different type ofnetwork,
generically referred to as a packet-
switched network, that is particularly suited to servicing data traffic. Included in Chap-
ter 10 is a discussion of Asynchronous Transfer Mode
(ATM), a form of a
packet-switched network. Chapter 1l discusses various technologies and systems for
achieving direct digital access to a digital network
(voice or data). The last chapter
pre-
sents the basics of Faffic theory: the fundamental mathematics for analyzing and
pre-
dicting telecommunications network
performance.
1.1 TELECOMMUNICATIONS
STANDARD ORGANIZATIONS
Prior to the breakup of the Bell System on January l, 1984, telecommunications
stand-
ards in North America were essentially established by the dominant equipment de-
signer and supplier to the Bell System: Bell Telephone Laboratories
and Westem
Electric. Independent telephone companies that
provided local service
to the 207o of
the country not covered by the Bell System relied on the U'S. Independent Telephone
) l
4 encxcRouNDANDTEHMI NoLOGY
Association
[usITA;
larer referred to as rhe u.s. Telephone Association (usTA)]
to
formulate and disseminate standards, particularly
for interconnecting
with the Bell
System.
In anticipation of the divestiture of the Regional Bell operating companies
(RBocs)
from AT&T, the Exchange
carriers standards Association (ECSA)
was
formed in 1983 as a nonprofit trade association to represent the interests of all ex-
change carriers (RBocs
and independents). In February lgg4 the ECSA sponsored
the establishment of the Tl standards committee to formulate new interconnection
standards for the u.s. national
network. The Tl commiftee is accredited by the Ameri-
can National Standards Institute (ANSD
to ensure that standards approvals follow
principles
of openness. Thus Tl committee standards are designated as ANSI Tl.nnn-
date (T
I stands for Telecommunications
standards entity number I
).
Table I . I lists the
major subcommittees within Tl and the respective responsibilities.
Other organizations in North America that establish standards related to telecom-
munications are the Elecffonic Industries Association (EIA),
the Institute of Electrical
and Electronic Engineers (IEEE),
and Bell communicarion$
Research (Bellcore).
Bellcore was an organization chartered to establish standards and qualify
equipment
for the RBOCs. Bellcore has since been reorganized as Telcordia Technologies. The
IEEE is most known for its data communications standards listed in Table 1.2 but has
also established numerous standards for measuring and characterizing telecommuni-
cations eguipment.
Most of the world outside of North America relies on international telecommuni-
cations $tandards committees
e$tablished under the auspices of the International Tele-
communication union (ITU).
In the past, two major entities within the ITu were
established: the International
Telegraph and Telephone Consultative committee
(ccITT)
and the Internarional
Radio consultative committee (ccIR).
ccITT estab-
lished recommendations
for telephone, telegraph, and data transmission circuits and
equipment. ccIR was concerned with coordinating
the use of the radio specrrum.
CCITT and CCIR activities are no longer identified as being distinct from the ITU.
ccITT has become ITU-T and ccIR is now ITU-R. In the united states. use of the
radio spectrum is controlled by the Federal communications comrnission (FCC).
North American standards
and ITU standards have often been incompatible in the
past. Notth American standards established by the Bell System were therefore incor-
TABLE 1.1 T1 Standards Subcommltteee
Committee
Responsibility
T1A1
T1 E1
T1M1
T1 P1
T1S1
T1X1
Performance
and signal processing
Interfaces, power
and
protection
for networks
Internetwork operations, administration,
maintenance,
and
provisioning (IOAM&p)
Wireless/mobile services and sy$lems
Services, architeclures, and signaling
Digital hierarchy and synchronization
1.2 THE ANALOG NETWORK HIERARCHY
TABLE 1.2 IEEE Local Arsa NetworldMetropolltan
Area Nstwork
(LAN/MAN) Data
'Cbmmun
lcstions Standards
802"1
802.2
802.3
802.4
802.5
802.6
802.7
802.8
802.9
802.10
80e. 11
802:12
802. 14
Overview and Architecture, Bridging, Virtual bridged LAN
(VLAN)
Logical Link Control
(LLC)
carrier sense Multiple Access
(csMA) with collision
Detection
(cD) (Ethernet)
Token Bus
(Arcnet)
Token Ring
(lBM Bing)
Queued Packet Synchronous Exchange
(QPSX)
Broadband
Optical Fiber Technologies
Integrated Seryices
Security
Wireless
Demand Priority
Cable TV
porated into CCITT recommendations as a subset. Due to the need for more interna-
iional compatibility, the T1 subcommittees
and ITU-T committees now work closely
together to establish
joint
standards. A first major example of the
joint effort is the
standard for synchronous
fiber transmission
systems refened to as SONET
in the
united states and synchronous
Digital Hierarchy
(sDH) in ITU-T standards.
The International Standards Organization
(ISO) is an organization with
standards
activities in a wide range of subject matters, some of which involve telecommunica-
tions. Technical subcommittees
within ISo work closely with ITU study
groups in
formulating ITU recommendations,
particularly those related to ISDN
protocols that
adhere, as much as
possible, to the ISO data communications
standald for Open Sys-
tems Interconnection
(OSI) Reference
Model.
1.2 THE ANALOG NETWORK
HIERARCHY
Because the analog telephone networlLs of the world evolved over a
period of almost
100
years, a
great arnount of diversity in equipment
implementations also developed'
It is a remarkable achievement that vast networks, like the U.S' network, could accom-
modate the myriad of equipment types and function
properly. In 1980- in the united
States alone. there were 181 million telephones [2],
almost all of which could directly
dial any
public telephone number and have a
good quality connection'
This achieve-
ment wa$ made possible by standardized
interfaces and well-defined
functional hier-
archies. As newer digital equipment was installed, it necessarily
adhered to the
*Although
there is no specific date at which digital technology took over from analog.technology,
1980 is
significant in that it repfesents the time frame in which fiber optics emerged to begin displacing analog
rirlios for intercity transmission, the last stronghold of analog technology in the intemal
portions of the
public
network.
6 BecxcRouNDANDTERMtNoLocy
standardized practices
of the analog network. The fact that the equipment was imple-
mented with digital technology was transparent to the rest of the network.
1.2.1 Bell System Hierarchy
Alexander Graham Bell invented the first practical
telephone in 1876. It soon became
apparent, however, ttrat the telephone was of little use without some means of chang-
ing connections
on an
"a$-needed"
basis. Thus the flrst switching office was estab-
lished in New Haven, connecticut, only two years
later. This switching office, and
others following, was located at a central point
in a service area and provided
switched
connections for all subscribers in the area. Because of their locations in the service ar-
eas, the$e switching offices are often referred to as cenffal offices.
As telephone usage grew
and subscribers desired longer distance connections, it
became nece$sary to interconnect
the individual service areas with trunks between the
central offices. Again, switches were needed to interconnect these offices, and a sec-
ond level of switching evolved. Continued demand for even longer distance connec-
tions, along with improved long-distance
transmission facilities, stimulated even more
levels of switching. In this manner the analog public
telephone network in the United
states evolved to a total of five levels. These levels are listed in Table I .3.
At the lowest level of the network are class 5 switching offices, also called central
offices (cos)
or end offices (Eos).
The next level of the network was composed of
class 4 toll offices. The toll network of the Bell System contained three more levels of
switching: primary
centers, sectional centers, and regional centers.
To illushate
the structure
and motivation for hierarchical networks,
a symbolic,
three*level example is shown in Figure 1.2. In contrast, Figure 1.3 depicts a different
network structure for interconnecting
all of the firstlevel switches; a fully connected
mesh structure. obviously,
the hierarchical network requires more switching nodes
but achieves significant savings in the number of trunks; the transmission links be-
tween switching offices. Detetmination
of the total number of trunk circuits in either
network is necessarily
a function of the amount of traffic between each pair
of switch-
ing nodes. (Chapter
12 provides
the mathematics for determining
the number of trunk
circuits.) As a first approximation, the trunk costs of a mesh can be determined as the
total number of connections (trunk
groups)
N" between switching
off,rces:
TABLE 1.3 Pubtic Network Hierarchy
ot the Bell $ystem (1gSA)
tgl
Functional
Designation
Switch Class
No. i n
No. i n
Bell System lndependents
Total
I
2
Q
4
5
Regional center
Sectional center
Primary center
Toll center
End office
1 0
52
148
508
9803
0
0
20
425
9000
1 0
67
168
933
18,803
1.2 THE ANALOG NETWORK HIERAHCHY 7
N"=+N(N- l )
(l ' 1)
where N is the number of nodes.
Thus the mesh network of Figure 1.3 has 36 connections' aS compared to 12 con-
nections in Figure 1.2. In the case of fiber optic transmission the cost comparison of
Figure 1.2 Three-level switching hierarchy.
Flgure 1.3 Mesh-connected
network.
BACKGROUND AND TERMINOLOGY
the trunks is almost exactly 3 ; I because a single fiber system can provide
more voice
capacity than is needed between any two switches.
A less obvious difference between the networks of Figures 1.2 and 1.3 involves the
method of establishing connections between two offices. In the hierarchical network
there is one and only one route between any two switching nodes. [n the mesh network
most connections would be established on the direct route between the two offices.
However, if the direct route is unavailable (because
of a traffic overload or an equip-
ment failure) and the first-level switches can provide
trunkto-funk connections
(called
tandem switching functions), the mesh network provides
many altematives for
establishing connections between any two nodes. Hence the reliability of a network
architecture must be considered
in addition to
just
the costs. In general,
neither a
pure
mesh nor a purely
hierarchical network is desirable.
Taking these factors into account, Figure 1.4 depicts alternate routing as imple-
mented in the former Bell System. As indicated, the basic backbone hierarchical net-
work was augmented with high-usage trunks. High-usage trunks are used for direct
connections between switching offices with high volumes of interoffice
traffic. Nor-
mally, traffic between two such offices is routed through the direct trunks. If the direct
trunks are busy (which
may happen frequently if they are highly utilized), the back-
bone hierarchical network is still available for alternate routing.
Traffic was always routed through the lowest available level of the network. This
procedure
not only used fewer network facilities but also implied better circuit quality
because of shorter paths
and fewer switching points. Figure 1.4 shows the basic order
of selection for alternate routes. The direct interoffice trunks are depicted as dashed
lines, while the backbone, hierarchical network is shown with solid lines.
):*;;;
- t - t -
, F,
\\\\\\\
l;:;----
I
I
I
f
Figure 1.4 Altemate routing in North American network.
1,2 THE ANALOG NETWORK HIERARCHY
In addition to the high-usage trunks, the backbone network was also augmented
with additional switching facilities called tandem switches. These switches were em-
ployed at the lowest levels of the network and
provided switching between end offices.
Tandem switches were not
pafr of the toll network, as indicated in Figure I .5, but were
(and are)
pafr of what is referred to as an exchange area. Generally speaking, an ex-
change area is an area within which all calls are considered to be local calls
(i.e., toll
free).
Il general terms, any switching machine in a
path between two end offices
provides
a tandem switching function. Thus toll switches also provide tandem switching func-
tions. Within the public telephone network, however, the term tandem refers specifi-
cally to intermediate switching within the exchange area'
The basic function of a tandem office is to interconnect those central offices within
an exchange area having insufficient interoffice fiaffic volumes to
justify
direct
trunks. Tandem offices also
provide alternate routes fbr exchange area calls that
get
blocked on direct routes between end offices. Although Figure 1.5 depicts tandem of-
fices as being
physically distinct from end offices and toll offices, tandem switches
were often colocated with either or both types. Operationally, exchange area switching
and toll network switching in the Bell system were always separated. The
primary rea-
son for the separation wa$ to simplify tandem switching by avoiding billing and net-
work routing. A toll switch had to measure call duration for billing
purposes but a
tandem switch did not. Prior to the introduction of computer-controlled switching,
billing functions were a significant consideration. The operational separation also im-
plied that toll-connecting trunk
groups were separate from tandem trunk
groups. The
flexibility of computer-controlled
switching has eliminated the need for the separa-
tion.
The separation of exchange facilities from toll facilities had an important effect on
the transmission and switching equipment utilized in the respective applications. Ex-
change area connections were usually short and only involved a few switching offices.
Toll connections, on the other hand, coulcl involve numerous switching offices with
Toll netrrvork
Tandom
office
Di.ct trunk
Figure 1.5 Exchange area network.
10 BAcKGRoUNDANDTEHMtNoLocy
relatively long ffansmission links between them. Thus, for comparable end-to-end
quality,
individual analog exchange area equipment did not have to
provide
as much
quality
as did toll network counterparts.
1.2.2 Postdivestiture
U.$. Network
In the decade of the 1980s the structure of the
public
telephone network in the United
states changed significantly as a result ofchanges in the technology and the regulatory
environment. The main technological changes were
(1)
extensive deployment of very
large digital switching machines, (2)
the adaptation of computer-controlled switches
to provide multiple switching functions in one machine (e.g.,
the integration of end-
office, tandem, and toll switching functions), and (3)
the deployment of fiber optic
hansmission systems that could carry very large cross sections of traffic. Notice that
all three of these technological
developments suggest a network with fewer and larger
switching offices. These technological influences on the network topology are dis-
cussed more fully in Chapters 8-10.
The most dramatic and immediate effect on the network occurred on January l,
1984, when the breakup of AT&T officially took effect. Because the breakup involved
divestiture of Bell operating companies (BoCs)
from AT&T, the network irself be-
came partitioned
at a new level. The new partitioning
is shown in Figure I.6, which
depicts AT&T as one of several competing long-distance carriers referred to a$ inter-
exchange carriers
(IXCs)
and local access and transpoft areas (LATAs),
which were
originally the exclusive domain of local exchange carriers (LECs).
In addition to
AT&T, the other two main IXCs are MCI and u.s. sprint. The LECs originally in-
cluded 23 Bocs
(organized
into 7 RBocs), former independenr relephone companies
like GTE, contel, and united relecommunications, and some 1500 mostly small-
town telephone companies. Mergers within the industry have subsequently reduced
the number of LECs and RBOCs.
The number of LATAs in the united states was initially 164, but the number has
changed as adjustments in service boundaries are sometimes made. Because a LATA
entails an area that includes many exchange areas, LECs complete toll calls that kav-
erse different exchange areas within one LATA. The IXCs were not allowed to carry
intra-LATA traffic. similarly, an LEC was not allowed to carry traffic between two
LATAs even when both LATAs might be service areas of a single Boc. only an Ixc
was allowed to carry inter-LATA traffic. To ensure that these service partitions were
adhered to, each IXC interfaced with a LATA at a single point in the LATA, referred
to as a point of presence (PoP). IXC equipment at a Pop could be a switching office
or merely a
junction
for collecting traffic carried elsewhere to be switched.
A major aspect of the modified final
judgment (MFJ)
that specified the divestiture
was the condition of equal access, which meant that an LEC (specifically
a Boc) was
to treat all IXCs equally in regards to exchange access. The conditions ofequal access
meant that access to all end offices in a LATA would be equal in type, quality,
and
price for all IXCs. The LATA nerwork ritructure
[4]
established to accomplish equal
access is shown in Fieure 1.7.
1.2 THE ANALOG NETWORK HIERARCHY 1 1
Figure 1.6 U.S, network partitioning.
POP: Poi nl ol
Ptesence
AT: Accets tandent
TO: Tarrdcnt offi ce
EO: Entl ol l i ce
TIC: Tsndetn i nter"LATA connecti ng
OIG: Di rect i nter-LATA connecl i ng
TCTC; Tdnderfi connecl i ng
Figure 1.7 LATA hierarchy hnd access architecture'
12 BACKGROUND AND TEHMINOLOGY
The design of the LATA network for intra-LATA traffic was left to the discretion
of the LECs. Thus intra-LATA connections can involve multiple switching offices be-
tween end offices. However, connections between an Eo and a PoP could involve at
most one intermediate switching office referred to as an access tandem
(AT).
with re-
spect to the previous Bell system hierarchy, an AT takes the place
of a class 4 toll
switch. However, long-distance billing functions, which were formerly performed in
class 4 switches, are now performed
within the IXC network. Although Figure l.7
shows access tandem and basic tandem switching functions as being distinct, access
tandem functions can be integrated into regular tandem switches if the tandem switch
provides AT features. Foremost among these feahrres are the ability to forward auto-
matic number identification (ANI)
information to an IXC for billing and the ability to
route calls to different IXC POPs depending
qn
presubscription
or per-call
three-digit
carier designations.
In 1997 the FCC issued some rulings with the intent of stimulating competition in
both the local exchange and long-distance networks. under this ruling, LECs that
want to enter the long-distance market can do so if they open their local exchange fa-
cilities to long-distance carriers or other competitive access providers.
A key aspect
of making the local facilities available to competition is the establishment of unbun-
dled pricing
for local seryices; the separation of the cost of the local loop, the local
switching equipment, maintenance, and ancillary services such as 9ll emergency
calling. Another key requirement is number portability,
which allows a subscriber to
change local service providers
without having to change telephone numbers. The in-
troduction of competition for local distribution instigated the use of two tennrr: com-
petitive
local exchange carrier (CLEC)
for the competition and incumbent local
exchange carrier (ILEC)
for the establi$hed carrier.
1.2.3 SwitchingSystems
Manual Swltchboards
The first telephone switching equipment utilized operators at manual swirchboards. The
operators asked a caller for the number they wanted to call and then established the con-
nection by plugging
in a cord between terminal
jacks.
Although switchboards are no
longer used, a legacy of their existence lives on: the use of the terms
"tip
and ring." As
shown in Figure 1.8, one wire of a wire pair
was connected to the tip of a plug
comector
and the other wire was connected to the ring. Ever since, one wire of a wire pair
is com-
Switdrboard
plug
Flgure 1.8 switchboard plug with corresponding
jack
(R,
s, and r are ring, sleeve, and tip,
respectively). (From
Freeman, Fundamentals of releconzmunications, wiley, New
york.)
Switdrboard
jacf
1.2 THE ANALOG NETWORK HIEHARCHY 1 3
monly referred to as the tip and the other is referred to as the ring, even on digital wire
pairs,
which have never used plugs in a swirchboard. On some of the original switchbomds
a thircl connection would be provided by the sleeve conductor shown in Figure 1.8.
Automated Switching
In general terms the equipment associated with any
particular swirching machine can
be categorized as providing one of the following functions:
1. Signaling
2. Control
3. Switching
The basic function of the signaling equipment is to monitor the activity of the incom-
ing lines and forward appropriate status or control information to the control element
of the switch. Signaling equipment is also used to place control signals onto outgoing
lines under direction of the switch control element.
The control element processes incoming signaling information and sets up connec-
tions accordingly. The switching function itself is provided by a switching matrix: an
array of selectable crosspoints used to complete comections between input lines and
output lines. These basic constituents of a switching machine are shown in Figure 1.9.
Electrcmechanicalswitching. Prior to the introduction of digital electronic
switching machines in the late 1970s, switching offices in North America and around
the world were equipped with one of two basic types of electromechanical switches;
step-by-step* and crossbar. As shown in Figure 1.10, crosspoints of a step-by-step
switch are wiper contacts that move in direct response to dial pulses. As the pulses of
the first digit enter the switch, they immediately
"step"
the vertical wiper to a horizon-
tal row corresponding to the first digit. After the
proper row is selected, the wiper is
rotated across another set of contacts until an idle line to the next stage of switching
is located. The next set ofdial
pulses, representing the second digit, then steps the sec-
ond stage in like mannet. The process continues through however many stages are
needed fbr a particular switch size.
As the name implies, a step-by-step switch uses direct progressive control: Succes-
sive segmenrs of a path through the switch are establi$hed as each digit is dialed. Wittr
progressive control, the control elements of the switch are integrated into the switch-
ing matrix. This feature is very useful fbr implementing a variety of switch sizes and
allowing relatively easy expansion. A progre$sive control switch, however, has a
number of significant limitations;
l. A call may be blocked even though an appropriate
path through the switch exists
but is not attempted because an unfortunate
path gets selected in an early stage.
2. Alternate routing for outgoing trunks is not
possible. That is, the outgoing line
is directly selected by incoming dial pulses and cannot be substituted'
*A
step-by-step swirch is also referred to as a Strowger switciz, in honor of its inventor Almon B. Strowger.
1 4 BACKGHOUND AND TEBMINOLOGY
Figure 1.9 Switching system components.
Signaling schemes other than dial
pulses (e.g.,
tone signaling) are not directly
usable.
Number translation is impossible.
In conffast to a step-by-step switch, a crossbar switch is one that used centralized,
coillmon control for switch path selection. As digits were dialed, the conhol element
of the switch received the entire address before processing it. when an appropriate
path through the switch was determined (which
may have involved numbertranslation
or alternate routing), the control element transferred the necessary information in the
form of control signals to the switching matrix to establish the connection. The fun-
damental feature, and advantage, of a common control switch is that control function
implementation is separate from the switch implementation. Common control cross-
SLE E
BANI(
SLE E VE
Ii,IPE R
VE FTI CAL
3.
4.
VE RTI CAL
WI PE H
COMMUTATOB
I USEO I N LI NE
F I NDE RSI
L I NE
WIPE R
Figure l.l0 Step-by-step switching
Laboratories. Reprinted by permission.)
WI PE R
CORDS
elemenl
(Copyright
1977 by Bell Telephone
1,2 THE ANALOG NETWOHK HIERARCHY 15
bar systems introduced the ability to assign logical addresses
(telephone numbers) in-
dependently of physical line numbers.
The crosspoints of a crossbar switch
(Figure
Ll l) are mechanical contacts with
magnet$ to set up and hold a connection. The term crostbdr arises from the use of
crossing horizontal and verlical bars to initially select the contacts. Once established,
the switching contacts are held by electromagnets energized with direct cunent
pass-
ing through the established circuit. When the circuit is opened, the loss of current
causes the crosspoints to be released automatically.
Because of the operational limitations of progressivs control, step-by-step switches
were used
primarily in smaller class 5 switching offices. Crossbar switches, on the
other hand, were used predominantly in metropolitan areas and within the toll net-
work. In $ome cases step-by-step switches were augmented with coillmon control by
receiving the digits into special control equipment. After
processing the request, the
control equipment
generated pulses that set up a connection as if the switch was re-
ceiving dial
pulses directly.
l,lt(
'-
.c'*I.
luc,'F-5
cffiEl
ErilI
xfrss
mlt'f,
Figure LlLll Crossbar switching element.
(Copyright 1977 by Bell Telephone
Laboratories.
Reprinted by permission.)
16 BACKGROUND AND TERMINOLOGY
Stored Program Control. Step-by-step and crossbar switching sysrems used elec-
tromechanical components for both the switching matrix and the control elements. In
some casett the electromechanical control elements in these switches represented ru-
dimentary forms of special-purpose digital computers. The hardwired electromd-
chanical logic, however, had limited capabilities and was virtually impossible to
modify.
A major milestone for telephony was established in 1965 when the Bell System in-
stalled its first computer-controlled switching system: the No. I Electronic Switching
system (Ess).- This switching system uses a stored-program digital computer for irs
control functions. The stored-program control (SPC)
feature of the No. I ESS allowed
the introduction of new features such as abbreviated dialing, call forwarding, call
waiting, and three-way calling.
The introduction of SPC not only provided
significant advantages to end users but
also simplified many administrative and maintenance tasks for the operating compa-
nies. A large parl
of line administration that formerly required many manual modifr-
cations (mainframe
cross-connects) could instead be accomplished with changes in
computer data tables of an SPC switch. Furthermore, physical
line numbers were in-
dependent of the logical
(directory)
line numbers, thus making number changes easy.
other benefits enabled by sPc are automated record keeping, lower blocking prob-
abilities, generation
of traf-fic statistics, automated call tracing, and message unit ac-
counting (per-call
charges as opposed to flat-rate billing for unlimited local calling).
The switching mahix of the No. I ESS
(and
also rhe No. 2 ESS, No. 3 ESS, and
No. IA ESS) is implemented with electromechanical reed relays. Thus the termESS
refers in general to computer-controlled switching and not to the nature of the switch-
ing makix itself. However, AT&T's No. 4 ESS, which was first installed in 1976. is
a high-capacity toll switch using computer control and digital electronics for its
switching matrix. Thus the No. 4 ESS is
"electronic"
in its control and its switching
matrix. Fufthermore, the digital multiplex system (DMS)
switches of Northern Tele-
com, the No. 5 eleckonic automatic exchange (EAX)
of GTE, and the No. 5 ESS of
AT&T also utilize. digital logic circuits for the crosspoint matrix.
( " ' ' f ' t
\
Private Branch Exchanges. In the united states the term private branch ex-
change
(PBX)
refers generically
to any switching system owned or leased by a busi-
ness or organization to provide
both internal switching functions and access to the
public network. Thus a PBX in the united states may use either manual or automatic
control. The term PABX is also used in the united States, and particularly
in other
countries, to refer specifically to automatically controlled PBXs.
The historical development of PBX systems has followed closely that of switches
in the public network. PBXs with computerized control became available in 1963
(be-
fore the No. I ESS) when AT&T's No. l0l ESS was first installed. since that time a
*Computer-controlled
PBXs were available befbre 1965. The No. I E$S represents the first instance of
computer control in the public network hierarchy,
*rra
?
1.2 THE ANALOG NETWORK HIERARCHY 1 7
large number of independent manufacturers have developed computer-controlled
PBXs. In fact, the PBX market has long been one of the most competitive businesses
in telecommunications.
The use of computer control for PBXs introduced numerous new features for users.
Not only were customized calling features
(e.g.,
abbreviated dialing)
provided, but nu-
merous facilities for cost management also became available. Some of the more use-
ful, commonplace features in a PBX are the following:
1. Accounting summaries by individual employee or department
2. Multiple classes of service with priorities and access reshictions to area codes,
WATS lines. and so on
3. Least-cost routing to automatically select tie lines, foreign exchange circuits,
WATS. DDD, and so forth
4. Automatic callback when circuits are available
5. Traffic monitoring and analysis to determine the utilization of existing circuits
or to a$certain blocking probabilities and network cost effectiveness
Centr6x. Many of the features initially provided by PBXs are also offered by oper-
ating companies as Centrex features. As indicated in Figure 1.12, Centrex is abusiness
customer service offering supported by switching equipment in the central office.
Every telephone or data device at the customer premises has a dedicated channel to
the switch in the central office. Originally, each channel implied a dedicated
pair of
wires. It is now more common to use multiplexing techniques
(described in Section
I.2.5) to reduce the hansmission costs. Nevertheless, from the central office
point of
view each Centrex extension has a dedicated appearance at the CO with a unique
pub-
lic network telephone number. A software
partition in the central office treats the Cen-
trex lines as a closed user
group
to
provide the following basic features:
Sit6 I Sit6 2 Site 3
Locll Common Chantel Slgnaling
Figure 1.12 Centrex service to multiple sites.
18 BACKGRoUNDANDTERMI NoLoGY
l. Direct dialing to Centrex extensions from the public network.
2. Station-to-station calling using extension numbers as opposed to full, 7- (or
l0-)
digit public numbers.
3. Common voice user features such as call forwarding, call transfer, call waiting,
call pick up, and three-way calling.
4. Multiple sites with transparent numbering plans and features. With citywide
Centrex the sites can be supported from multiple offices interconnected by
common-channel signaling (CCS)
described in Section 1.2.10.
5. Centralized attendant/message desk with call origination information for
informed processing
of forwarded calls.
6. High availability because CO equipment and directly connected analog phones
are
powered
at the CO with backup power sources.
7. Virtually unlimited growth.
1.2.4 Transmission Systems
Functionally, the communications channels between switching systems are referred to
as trunks. In the past,
these channels were implemented with a variety of facilities, in-
cluding pairs of wires, coaxial cable, and point-to-point
microwave radio links. Except
for special situations, tunk facilities now utilize optical fibers.
Open Wire
A classical picture
ofthe telephone network in the past consisted oftelephone poles
with crossarms and
glass
insulators used to $upport uninsulated open-wire pairs. Ex-
cept in rura] environments, the open wire has been replaced with multipair cable sys-
tems or fiber. The main advantage of an open-wire pair is its relatively low attenuatiofl
(a few hundredths of a decibel per
mile at voice frequencies). Hence, open wire is par-
ticularly useful for long, rural customer loops. The main disadvantages are having to
separate the wires with crossarms to prevent shorting and the need for large amounts
of copper. (A
single open-wire strand has a diameter that is five times the diameter of
a typical strand in a multipair cable. Thus open wire uses roughly 25 times as much
copper as does cable.) As a result of copper co$ts and the emergence of low elecffonics
costs, open wire in rural environments hari been mostly replaced with cable systems
using
(digital)
amplifiers to offset attenuation on long loops.
Paired Cable
In response to overcrowded crossarms and high maintenartce costs, multipair cable
sy$tems were introduced as far back as 1883. Today a single cable may contain any-
where from 6 to 2700 wire pairs.
Figure 1.13 shows the structure of a rypical cable.
When telephone poles
are used, a single cable can provide
all the circuits required on
the route, thereby eliminating the need for crossarms. More recently the
preferred
means of cable distribution is to bury it directly in the ground (buried cable) or use
underground conduit (underground
cable).
1 .2 THE ANALOG NETWoRK HIERARGHY 19
Table 1.4 lists the most conrmon wire sizes to be fbund within paired-cable sys-
tems. The lower gauge (higher
diameter) systems are urted for longer distances where
signal attenuation and direct-cunent
(dc) resisturce can become limiting factors. Fig-
ure l . I 4 rrhows attenuation curves
[5]
for the common gauges
of
paired cable as a func-
tion of frequency. An important point
to notice in Figure l.14 is that the cable pairs
are capable of canying much higher fiequencies than required by a telephone
quality
voice signal (approximately 3.4 kHz).
In the past, the exchange areas of the telephone network used paired cable almost
exclusively fbr short-haul interoffice transmission. Up until the introduction of mul-
Figure 1.13 Multipair cable.
(Copyright 1977 by Bell Telephone Laboratorics. Reprinted by
permission,)
20 BACKGROUND AND TERMINOLOGY
TABLE 1.4 Wire Gauge and Resletance of Common
Palred Cable
Direct-Current
Gauge Diameter
(in.)
Resistance (fy1000
ft)a
30
28
26
24
22
20
1 9
0. 010
0. 013
0. 016
0.020
0.0?5
0.032
0.036
104
oo
41
26
1 6
1 0
eNote
that the loop resistance of a pair is twice the resistanc of a single
wire given in th table.
#
#
g
g
t
E
E
c
o
tl
18
17
18
15
14
13
12
1 l
r 0
I
I
7
6
5
4
3
2
I
0
1,000 l.lfr
Frequercy (Hz)
1, 1d 1. 100 1. 107
Figure 1.14 Attenuation versus frequency of common gauges of paired
cable,
(From
W. D.
Reeve, ,SuDscriDer Loop Signaling Transmission Handbook, IEEE Press, New
york,
Fig.
7-16a. )
1.? THE ANALoc NETwoHK H|FRARCHY 21
Figure 1.15 Single-wire transmission with ground retum,
Figure 1.16 Two-wire transmission.
tiplexing techniques, described later in this chapter, each voice circuit
(trunk)
was car-
ried on a separate pair of wires. Beginning in the early 1960s electronics began to be
used for short-haul interoffice transmission systems using multiplexing techniques to
carry multiple channels on a single
pair of wires.
Tw*Wire Versue Four-Wire
All wire-line transmission in the telephone network is based on transmission through
pairs of wires. As shown in Figure L15, transmission through a single wire
(with a
ground return) is possible and has been used in the past. However, the resulting circuit
is too noisy for customer acceptance. Instead, balanced pairs of wires as shown in Fig-
ure 1.16 are used with signals propagating as a voltage difference between the two
wires. The electrical current produced by the difference signal flowing through the
wires in opposite directions is called a
"metallic
current.'r In contrast, current propa-
gating in the same direction in both wires is referred to as common-mode or longitu-
dinal current. Longitudinal currents are not coupled into a circuit output unless there
is an imbalance in the wires that converts some of the longitudinal signal
(noise or in-
terference) into a difference signal. Thus the use of a
pair of wires for each circuit
pro-
vides much better circuit quality than does single-wire transmission. Some older
switching systemrr used single-wire
(unbalanced)
transmission to minimize the num-
ber of contacts. Unbalanced circuits were only feasible in small switches where noise
and crosstalk could be controlled.
Virtually all subscriber loops in the telephone network are implemented with a sin-
gle pair
of wires.- The single pair provides for both directions of ffansmission. If users
on both ends of a connection talk simultaneously, their conversations are superim-
posed on the wire pair and can be heard at the opposite ends. In contrast, wireJine
(and
fiber) transmission over longer distances, as between switching offices, is best imple-
mented if the two directions of transmission are separated onto $eparate wire
pairs.
*It
is no* commonplace to use fiber for the tbeder portion of a subscdber loop, but the drop to a te sidence
is a single pair per telephone.
22 BACKGROUND AND TERMINOLOGY
Longer distance transmission requires amplification and most often involves multi-
plexing. These operations are implemented most easily if the two directions of ffans-
mission are isolated from each other. Thus interoffice trunks typically use two pairs
of wires or two fibers and are referred to as four-wire $ystems. The use of two pairs
of wires did not necessarily imply the use of twice as much copper as a two-wire cir-
cuit. After 1960, four-wire systems increasingly used some form of multiplexing to
provide
multiple channels in one direction on one wire pair.
Thus, a net savings in cop-
per
could result.
Sometimes the bandwidth of a single pair of wires was separated into two subbands
that were used for the two directions of travel. These systems were referred to as de-
rivedfour-wire systems. Hence, the term/aur-wire has evolved to imply separate
channels for each direction of transmission, even when wires may not be involved. For
example, fiber optic and radio system$ that use separate channels for each direction
are also referred to asfour-wire systems.
The use of four-wire transmission had a direct impact on the switching systems of
the toll network. Since toll network circuits were four-wire, the switches were de-
signed to separately connect both directions of hansmission. Hence, two paths
through the switch were needed for each connection. A two*wire switch, as used in
older analog end offices, required only one path through the switch for each connec-
tion.
Two-Wi ra-to-Fou r-Wl re Conversion
At some point
in a long-distance connection it is necessary to convert from two-wire
transmission of local loops to four-wire transmission on long-distance trunks. In the
past, the conversion usually occurred at the trunk interface of the
(two-wire)
end office
switch. Newer digital end office switches are inherently
"four-wire,"
which means the
two-wire-to-four-wire conversion point
is on the subscriber (line)
side of the switch
as opposed to the trunk side. A generalized
interconnection of two-wire and four-wire
facilities for a connection is shown in Figure I.17. The basic conversion function is
provided by hybrid circuits that couple the two directions of transmission as shown.
Hybrid circuits have been traditionally implemented with specially interconnected
transformers. More recently, however, electronic hybrids have been developed. Ide-
ally a hybrid should couple all energy on the incoming branch of the four-wire circuit
into the two-wire circuit, and none of the incoming four-wire signal should be trans-
ferred to the outgoing four-wire branch.
Hybri d
Hybri d
Figure 1.17 Interconnection of two-wire and four-wire circuits.
1.2 THEANALocNETWoRKHTERARoHy eg
When the impedance matching network Z exactly matches the impedance of the
two-wire circuit, near-perfect isolation of the two four-wire branches can be realized.
Impedance matching used to be a time-consuming, manual proce$$ and was therefore
not commonly used. Furthermore, the two-wire circuits were usually switched con-
nections so ttre impedance that had to be matched would change with each connection.
For these reasons the impedances of two-wire lines connected to hybrids were rarely
matched. The effect of an impedance mismatch is to cause an echo, the power level of
which is related to the degree of mismatch. The effect of echoes on voice quality and
the means of controlling them is discussed later in this chapter.
Loadlng Coils
The attenuation curves shown in Figure I . 14 indicate that the higher frequencies of the
voice spectrum
(up to 3.4 kHz) experience more attenuation than the lower frequen-
cies. This frequency-dependent attenuation distorts the voice signal and is referred to
as amplitude distor"tion. Amplitude distortion becomes most $ignificant on long cable
pairs, where the attenuation difference is greatest.
The usual method of combating amplitude distonion on intermediateJength
(3-
lS-mile) wire pairs is to insert artificial inductance into the lines. The extra inductance
comes from loading coils that ariinserted at 30fi)-,4500-, or 6000-ft intervals. Figure
1.18 shows the effect of loading coils on a 24-gauge loop. Notice that the voiceband
response up to 3 kHz is greatly improved, but the effect on higher frequencies is dev-
astating.
Prior to the introduction of wire-line and fiber carrier systems, loading coils were
used extensively on exchange area interoffice trunks. Loading coils are also used on
the longer, typically rural, subscriber loops. Here, too, carrier system$ have displaced
most of the single
pairs of wires being used on long routes.
Frequency
(kllzl
tr'lgure 1.18 Effect of loading on 24-gauge cable pair.
g t o
E
E,
I
S z o
e
E
EI
E t 0
24 BACKGROUND AND TERMINOLOGY
1.2.5 Pai r-Gai nSystems
Providing service to rural subscribers has always been an expensive proposition
be-
cause of the length of the routes involved and the small number of households to sup-
pott
the cost of both the initial installation and the maintenance. In the past,
a cornmon
means of reducing the co$t$ was to use party lines, which involved sharing of a wire
pair
among multiple households. A
party
line is particularly
useful for satisfying new
service requests on routes with no spare pairs but is obviously objectionable due to the
lack of privacy
and the lack of availability of the line.
A pair-gain sy$tem is an alternate approach to sharing pairs
of wires that is much
more acceptable to the users. This section describes two basic types of pair-gain sys-
tems; concentrators (remote
swirches) and multiplexers (carrier
systems).
--;_
*
Goncentratlon
The first form of a pair-gain
system in Figure L 19 depicts a basic line concentration
system. When viewed ftom the station set end of the system, a
pair-gain
system pro-
vides concentration by switching some number of active stations to a smaller number
of shared output lines. At the other end of the system, deconcentration (expansion)
oc-
curs by switching from the shared lines to individual inputs of the switching office cor-
responding to the active stations. Expanding the traffic back to the original number of
stations ensure$ that the system is operationally transparent to both the switch and the
user. Notice that a definition of which end
provides
concentration and which end pro-
vides expansion is dependent on the point of view.
N surrce$
3
2
/l{ chadnels
Figure 1.19. Pair-gain systems: concentration and multiplexing: (a)
concentration (z
> M);
(b)
multiplexing.
N $rbchandels
N tource$
1.2 THE ANALOG NETWOHK HIERARCHY 25
Since a concentrator is incapable of simultaneously connecting all stations it serv-
ices, a certain amount of blocking is necessarily introduced by concentration. When
the activity of individual stations is low enough, significant amounts of concenffation
can be achieved with acceptable blocking
probabilities. For example,40 stations that
areeachactiveonlyT.SVoofthetimecanbeconcentratedonto
10 lineswithablocking
probability of 0.001 .* This is an acceptable degradation in service since an equally ac-
tive called station is busy 75 times as often.
Notice that a concentration system requires the ffansfer of control information be-
tween the concentrator/expander switch terminals. When one end of the system estab-
lishes a new connectron to one of the shared lines, the other end must be informed to
set up the appropriate reverse comection.
Multlplexing
As shown in Figure 1.14, the inherent bandwidth of a typical wire
pair is considerably
greater than that needed for a single voice signal. Thus, multiplexing can be used to
carry multiple voice channels on a single
pair of wires. The increase in attenuation im-
plied by the higher frequencies is offset by amplifiers in the multiplex equipment and
at
periodic points in the ffansmission lines. The particular multiplexing technique
shown in Figure 1. I9b is a frequency division multiplex system. Another form of mul*
tiplexing, time division multiplexing of digital voice signals, is the
preferred multi-
plexing
approach for digital
pair-gain systems discussed later'
As shown in Figure I . l9b, there is a one-to-one relationship between the customer
lines and the subchannels of the multiplexer. Thus, unlike the concentration system,
there is no possibility of blocking in a multiplexing type of pair-gain system. Also,
there is no need to transfer switching information since the same one-to-one relation-
ship defines the conespondence between customer lines at one end and switching of-
f,rce lines at the other end. A major drawback of multiplexing
pair-gain systems is that
the subchannels are highly underutilized if the sources are relatively inactive. In these
situations a combination of concentfation and multiplexing is normally
justified.
Time Asslgnment Speech lnterpolation
Time assignment speech interpolation
(TASI) is a pair-gain system that dynamically
assigns a channel to a circuit only when there is actual voice activity. Thus, a TASI
sy$tem senses voice activity from a number of sources N; assigns active sources to one
of M channels, where M is typically about half as large as N; and signals the far end
about the connections. Normally, each
participant in a conversation is active for only
407o of the time, which indicates that if M
=
j
N,thereissomeamountofsparecapac-
ity to accommodate excess activity in one direction. If a source begins to talk when all
channels are utilized, the beginning ofthat speech segment
Sets
clipped until a channel
becomes available. Chapter 12
provides mathematical formulations to determine the
probability of clipping as a function of N, M, and the voice activity factor'
Initial TASI applications involved improving the utilization of undersea cable
pairs. The obvious expense of these channels warranted the use of a rather complicated
*A
discussion of traffic analysis is provided in Chapter 12, from which this result can be obtained.
26
( ' )
BACKGHOUND AND TERMINOLOGY
multiplexing technique for the time. The same basic technique has since been used in
numerous applications with digital speech for satellite and landline applications.
These systems are
generally
called digital speech inrerpolation (DSI)
systems
[6].
1.2.6 FDM Mul tl pl exi ng and Modutati on
The introduction of cable systems into the transmission plant
to increase the circuit
packing density of open wire is one instance of multiplexing in the telephone network.
This form of multiplexing, referred to as space division multiplexing, involves nothing
more than bundling more than one
pair
of wires into a single cable. The telephone net-
work uses two other forms of multiplexing, both of which use electronics to pack more
than one voice circuit into the bandwidth of a single transmission medium. Analog fre-
quency division multiplexing (FDM)
has been used extensively in point-to-point
mi-
crowave radios and to a much lesser degree on $ome obsolete coaxial cable and
wire-line $ystems. FDM is also utilized in fiber optic transmission systems, where it
is referred to as wavelength division multiplexing (wDM).-
Digital time division mul-
tiplexing (TDM)
is the dominant form of multiplexing used in the telephone nerworks
worldwide. (Even
the fiber optic system$ that utilize wDM commonly use digital
TDM within the signals of a particular wavelength.)
Frequency Division Multlplexing
As indicated in Figure I.l9b, an FDM system divides the available bandwidth of the
transmission medium into a number of narower bands or subchannels. Individual
voice signals are inserted into the subchannels by amplitude modulating appropriately
selected carrier frequencies. As a compromise between realizing the largest number
of voice channels in a multiplex
system and maintaining acceptable voice fidelity, the
telephone companies established 4 kHz as the standard bandwidth of a voice circuit.t
If both sidebands produced
by amplitude modulation are used
(as
in obsolete NI or
N2 carrier systems on paired
cable), the subchannel bandwidth is 8 kHz, and the cor-
responding carrier frequencies lie in the middle of each subchannel. Since double-
sideband modulation is wasteful of bandwidth, single-sideband (ssB)
modulation
was used whenever the extra terminal costs were
justified.
The carrier frequencies for
single-sideband systems lie at either the upper or lower edge of the corresponding sub-
channel, depending on whether the lower or upper sideband is selected. The A5 chan-
nel bank multiplexer of AT&T used lower sideband modulation.
r...
FDM Htetrarcny
In order to standardize the equipment in the various broadband transmission systems
of the original analog network, the Bell system established an FDM hierarchy as pro-
vided in Table 1.5. ccITT recommendations specify the same hierarchy at the lower
*Optical
technology is customarily defined in terms of the wavelength of the optical signal as opposed to
the corresponding frequency.
'Actually'
the usable bandwidth ofan FDM voice channel was closer to 3 kFIz due to suard bands needed
by the FDM separation filters.
1.2 THEANALOGNETWORKHIERARCHY 27
TABLE 1.5 FDM Hierarchy of the Bell Nstwork
Multiplex Level
Number of Voice
Circuits Formation Frequency Band
(kHz)
Voice channel
Group
Supergroup
Mastergroup
Mastergroup Mux
Jumbogroup
Jumbogroup Mux
1
1 2
60
600
1,200-3,600
3,600
10.800
12 voice circuits
5
groups
10 supergroups
Various
6 master
groups
3
jumbo groups
H
60*1 0B
31 2-552
56rt--3,084
31?,564-1 7,548
56,t-17,548
3,00160,000
levels. Each level of the hierarchy is implemented using a set of standard FDM mod-
ules. The multiplex equipment is independent of
particular broadband transmission
media.
All multiplex equipment in the FDM hierarchy used SSB modulation' Thus, every
voice circuit required approximately 4 kHz of bandwidth. The lowest level building
block in the hierarchy is a channel
group consisting of 12 voice channels. A channel
group multiplex uses a total bandwidth of 48 kHz. Figure I.20 shows a block diagram
of an A5 channel group multiplexer, the most common A-type channel bank used for
first'level multiplexing. Twelve modulators using 12 $eparate carriers
generate l2
double-sideband signals as indicated. Each channel is then bandpass filtered to select
only the lower sideband of each double-sideband signal. The composite multiplex sig-
nal is
produced by superposing the filter outputs. Demultiplex equipinent in a receiv-
ing terminal uses the same basic
processing in reverse order.
Notice that a sideband separation filter not only removes the upper sideband but
also restricts the bandwidth of the retained sigrpl: the lower sideband. These filters
therefore represented a basic
point in the analog telephone network that defined the
bandwidth of a voice circuit. Since FDM was u$ed on all long-haul analog circuits,
l ndi vi dual voi ce
channel i nputs
I
EL
rt-r
l ( x 112
0 4
FI
104 108
l 2 Chennel
mul ti pl ex output
Fl-r
60 68
-
60 64
BdndpEEE
fitter$
Figure 1.20 A5 channel bank multiplexer.
28 BACKGROUND AND TERMINOLOGY
long-distance connections provided somewhat less than 4 kHz of bandwidth. (The
loading coils discussed previously also produce
similar bandwidth limitations into a
voice circuit.)
As indicated in Table 1. 5, the second level of the FDM hierarchy is a 60-channel
multiplex refened to as a supergroup. Figure I.2l shows the basic implementation of
an LMX group bank that multiplexes five flust-level channel groups. The resulting 60-
channel multiplex output is identical to that obtained when the channels are individu-
ally translated into 4-kHz bands from 312 to 552 kHz. Direct translation requires 60
separate SSB systems with 60 distinct cariers. The LMX group bank, however, uses
only five SSB systems plus five lower level modules. Thus two-stage multiplexing,
as implied by the LMX group bank, requires more total equipment but achieves econ-
omy through the use of common building blocks.
(Engineers
at AT&T Laboratories
also developed a multiplexer that generated
a supergroup directly
[7].)
Because a second-level multiplexer packs
individual first-level signals together
without guard
bands, the carier frequencies and bandpass filters in the LMX group
bank must be maintained with high accurqpy. Higher level multiplexers do not pack
the lower level signals as close together. Notice that a ma$ter group,
for example, does
not provide
one voice channel for every 4 kHz of bandwidth. It is not practical to main-
tain the tight spacing between the wider bandwidth signals at higher frequencies. Fur-
thermore, higher level multiplex signals include pilot
tone$ to monitor transmission
link quality
and aid in carder recovery.
' ,.2.7
Wideband Tranemission Medla
Wire pairs
within multipair cables have usable bandwidths that range from a little un-
der I MHz up to about 4 MHz depending on the length, the wire gauge, and the type
of insulation u$ed on the pairs. Multiplexed transmissions on these wire pairs conse-
quently have capacities that range from ?4 channels (on analog N3 or digital Tl carrier
systems) up to 96 channels
(on
obsolete digital T2 carrier systems). In contrast, an ana-
12 Channel
group
i nputB
t-]
60
n l r
380 408 s28
r-1
504 552
t-]
50* s52
Bandpeac
fltterE
Figure 1.21 LMX group bank multiplexer.
1.2 THE ANALOG NETWORK HIERARCHY 29
log L5E coaxial cable $ystem was developed to carry 13,200 voice channels. Optical
fiber systems have been developed that carry in excess of 100,000 voice channels on
a single wavelength. Because optical transmitters
flight-emitting
diodes
(LEDs)
and
lasersl function best in a
pulsed mode of operation, fiber systems are primarily digital
in nature. Digital fiber optic transmission systems are described in Chapter 8. This sec-
tion discusses coaxial cable and
point-to-point microwave radio systems.
Coaxial Cable
Coaxial cable systems were used predominantly to $atisfy long-haul requirements of
the toll network. The first commercial $y$tem was installed in l94l for hansmission
of 480 voice circuits over a 200-mile stretch between Minneapolis, Minnesota, and
Stevens Points, Wisconsin
[7].
To combat attenuation, repeater amplifiers were in-
stalled at 5.5-mile intervals. Considering the maximum capacity of l2 voice circuits
on open wire or cable at the time, the introduction of
"coax"
was a significant devel-
opment. After the first installation coaxial cable capacity was steadily increased by
(l)
using larger diameter cables
(0.375
in.) to r"educe attenuation,
(2)
decreasing the dis-
tance between repeaters, and
(3) improving the noise figure, linearity, and bandwidth
of the repeater amplifiers.
A sumnrary of the analog coaxial cable systems used in the Bell System is provided
in Table 1.6. Notice that each system reserved one
pair of tubes as spares in the event
of failure, a particularly important consideration since each tube canied a high volume
of traffic. Because optical fibers have wider bandwidths, lower attenuation, lower
maintenance, and lower cost, coaxial systems are obsolete.
Mlcrowave Radlo
Much of the impetu.s for terreshial microwave radio systems carne from the need to
distribute television signals nationwide. As the volume of long-distance traffic in-
creased, radio systems also became the most economical means of diskibuting voice
circuits in the long-distance network. Beginning in 1948, when the first system was
installed between New York and Boston, the number of microwave radio sy$tems
grew to supply 607o of the voice circuit miles in the U.S. toll network in 1980
[71.
It
was
just
a few years after that that optical fibers began to take over for high-density
interoffice routes and eventually for the network as a whole.
TABLE 1.6 Coaxlal Cable Systeme ln th Bell Network
System
Designation
Pairs
per
Systeme SignalDesignation
FleDeater
Spacing Capacity
per Total
(miles) Pair Capacity
L1
L3
L4
L5
3/6
5/6
9/1 0
10/ 11
Ma$iergroup
Mastergroup Mux
Jumbogroup
Jumbogroup Mux
I 600 1,800
4 1,860 9,300
2 3,600 32,400
1 10,800 108,000
the number of
pairs
are shown a$ working^otal.
30 BACKGROUND AND TEHMINOLOGY
Microwave radio systems require line-of-sight nansmission with repeater spacings
typically 26 miles apart. The major advantage of radio systems is that a continuous
right-of-way is not required-only small plots of land spaced 20-30 miles apart for
towers and equipment shelters. A major cost of guided transmission, for example, wire
pairs,
coax, or fiber, is the right-of-way costs. In many metropolitan areas, microwave
routes became heavily congested and could not be expanded with allocated common-
carrier frequency bands. In these situations, finding a right-of-way for a coax or even-
tually an optical fiber system was sometime$ the only choice for high-capacity
transmission.-
The frequency bands allocated by the FCC for common-carrier use in the united
states are listed in Table L7. of these bands, 4 and 6 GHz have been the most
popular.
The 2-GHz band has not been used extensively because the relatively narrow allocated
channel bandwidths do not permit
implementation of economical numbers of voice
circuits. The basic drawback of the lI-GHz band is its vulnerability to rain attenu-
ation. However, ll-GHz radios have been used in some
$fiort-haul
applications,,
The microwave radio systems of the analog Belt netwoik are listed in tabte L8.
Notice that each radio system is designed to carry one of the multiplex hierarchies de-
scribed previously.
All of these radios excepr rhe AR-6A used low-index frequency
modulation (FM)
of the signal generated
by the FDM multiplexer equipment. Thus,
the FM radios hansmit the ssB FDM signal as a baseband signal with a bandwidth as
indicated in Table 1.5. FM modulation was chosen to
permit
the use of nonlinear
power amplifiers in the transmitters and to take advantage of FM signal-to-noise ratio
perforrnance.
Examination of Tables 1.7 and 1.8 indicates that 13.3 kHz of bandwidth is utilized
per
voice circuit in TD-2 radios and 14.3 kHz in TH-3 radios. Thus. the use of FM in-
troduced a significant increase in the 4 kHz bandwidth of the individual SSB voice cir-
cuit. In conrrast, the ssB AR-6A radio introduced in lggl provided
6000 voice
circuits in the 30-MHz channels at 6 GHz.
Since a large number of voice circuits are carried by each radio channel, microwave
system$ usually include extra equipment and extra channels to maintain service de-
spite outages that may result from any of the following:
1. Atmospheric-induced multipath fading
2. Equipment failures
3. Maintenance
On some routes, the most frequent source of outage in a microwave radio system arises
from multipath fading. Figure 1.22 depicts a simple model of a multipath environment
arising as a result of atmospheric refraction. As indicated, the model involves two
ray$: a
primary
ray and a delayed secondary ray. Ifthe secondary ray anives out of
phase
with-respect to the primary
ray, the primary
signal is effectively canceled. The
*Micruwave
bands with wide-bandwidth allocations at high camer frequencies are available for some I ocal
distribution services and are discussed in Chapter I l.
1.2 THE ANALOG NETWORK HIERARCHY
TAELE 1.7 Mlcrowave Frequenciee Allocated for
Common-Carrler Uss in the Unlted States
Band
(MHz)
TotalBandwidth ChannelBandwidths
(MHz) (MHz)
31
21 1 0-21 30
21 60-21 80
3700+200
5925-64?5
10,700-1 1
,700
20
20
500
500
1 000
3.5
3.5
20
30
40, 20
amount of cancelation is dependent on both the magnitude and the phase of the sec-
ondary ray.
Quite
often only nominal amounts of fading occur and can be accommo-
dated by excess signal
power in the transmitter, called a fade margin. In some
instances, however, the received signal is effectively reduced to zero, which implies
that the channel is temporarily out of service.
Frequency Diverelty
Fortunately, exceptionally deep fades normally affect only one channel
(carrier fre-
quency) at a time. Thus, a backup channel including a spare transmitter and a spare
receiver can be used to carry the traffic of a faded
primary channel. Selection of and
switching to the spare channel is
performed automatically without a loss of service.
This remedy for multipath fading is referred to as frequency diversity. Notice that fre-
quency diversity also provides hardware backup for equipment failures'
A fully loaded TD-3 radio system used 12 channels: 10 main channels and 2
backup channels for protection. This is
Lefe-rred
to variously as 2-for-10, 1.Q--by12' or
10 X 2
protection switching. Some short-haul systems used l-for-l protection switch-
ingbecause it i$ $implef t-d implement. However, since only half of the required band-
width is actually carrying traffic, systems with l-for-l
protection were only allowed
in uncongested environments.
TABLE 1.8 Bell System Analog Microwave Fsdios
System Band
(GHz) VoiceCircuits Application
TD.2
TD-3
TH.1
TH.3
TM.1
TJ
TL.1
TL-2
AR-6A
4
4
o
6
6
1 1
1 1
1 1
6
600-1500 Long haul
1200 Long haul
1800 ShorVlong haul
2100 ShorUlong haul
600-900 ShorUlong haul
600 Short haul
240 Short haul
600-900 Short haul
6000 Long haul
(SSB)
(
(
32 BAcKcHoUNDANDTERMtNoLoGy
Figure 1.22 Two-ray model of multipath propagation.
Except in maintenance situations, protection
switching must be automatic in order
to maintain service continuity. A typical objective was to restore service within 30
msec to minimize noticeable effects in the message haffic. A more critical require-
ment is to re$tore service before the loss of signal is interpreted by some signaling
schemes as a circuit disconnect. Inadvertent disconnects occur if an outage lasts for
more than 1-2 sec.
Space Diversity
since deep fades only occur when a secondary ray arrives exactly out of phase
with
respect to a
primary
ray, it is unlikely that two paths
of different lengths experience
fading simultaneously. Figure 1.23 depicts a technique, called space diversity, using
different path lengths to
provide prorecrion
against multipath fading. As indicated, a
single transmitter irradiates two receive antennas separated by some distance on the
tower. Although the path length difference may be less than a meter, this difference is
adequate at microwave frequencies, which have wavelengths on the order of tenths of
meters.
Rain is another atmospherically
based source of microwave fading. As already
mentioned, rain attenuation is a concem mosfly in higher frequency radios (l I GHz
and above). unfortunately neither frequency diversity
(at
the high frequencies) nor
space diversity provides
any
protection
against rain fades.
Satellitee
Following the April 1965 launch of the soviet union's Molniya domestic communi*
cation$ satellite and the first international communications satellite, INTELSAT I, the
use of satellites for international telephone traffic grew phenomenally.
The 1g70s and
early 1980s also produced
significant use of satellites in the United States for televi-
sion program
distribution and for corporate voice and data networks. The first domes-
H(-.-=
== =:: ;: :::
:::::::::::--;h
Figure 1.23 Space diversity.
1,2 THE ANALOG NETWOHK HIEHAFCHY 33
tic system in North America was the Canadian Arik A in 1972 followed by Westem
Union's Westar system for U.S. service in 1974
[8].
In one sense a satellite system is a microwave radio system with only one repeater:
the transponder in outer space. In fact, some satellite systems use the same 4- and 6-
GHz frequency bands used by tenestrial microwave radios. In another sense, however,
the broadcast nature of the down link offers additional opportunities for new services
not available from
point-to-point terrestrial systems. Distribution of network televi-
sion programming is one application
particularly suited to the broadcast nature of sat-
etlites. Direct broadcast satellite
(DBS)
transmission to receive-only home receivers
is a prime example.* Two primary examples of DBS systems are the European DVB-T
system
t9l
and the North American digital satellite $ystem
(DSS) system developed
by Hughes Elechonics Systems
(HES).
Another application that is well suited to satellites is communications with mobile
stations. The international maritime satellite
(INMARSAT) system, for example, be-
gan $upporring the maritime industry in 1982 and was augmented with digital
Skyphone
[10]
for aeronautical telephone service in 1989.
One drawback to satellite communications is the inherent propagation delay of the
long transmission
path. For a stationary $atellite, this delay
(not including ground
links) is 250 msec up and down. A complete circuit with satellite links for both direc-
tions of travel therefore implies
greater than a one-half-second round-trip
propagation
time. Delays of this magnitude are noticeable in a voice conversation but not prohibi-
tive. The effects of the propagation delays can be alleviated somewhat by
pairing each
satellite circuit with a
ground-based circuit in the opposite direction. Thus, the round-
trip delay involves only one satellite link.
As is the case with point-to-point microwave radio, fiber optic transmission has
displaced the use of satellites for high-density, domestic and international telephone
communications.
Hence the use of satellites is primarily directed at thin-route tele-
phone and data application$, some mobile sy$tems, and broadcast applications.
1.2.8 Tranemi ssi on l mpai rments
One of the most diff,rcult aspects of designing an analog telephone network is deter-
mining how to allocate transmission impairments to individual subsystems
within the
network. Using subjective evaluations by listeners, certain objectives for end-to-end
fransmission
quality were established in a relatively straightforward manner
[11].
Af-
ter tempering the goals with economic feasibility, the end-to-end objectives
were es-
tablished. However, considering the myriad of equipment types and connection
combinations in the older analog network, designing the individual network elements
to meet these objectives in all cases wa$ a complex problem. A great deal of credit is
due the Bell System for having cleveloped a nationwide analog network with the level
of consistent
performance it had.
*A
DBS system is designed to use rcceiver antenflas that are about
I
rn in diameter. The older home satellite
receiver systems that wefe common in the United States used f- to 5-m anteflnas to rcceive
(intercept)
commercial
program distribution.
34 BACKGROUND AND TERMINOLOGY
The major factors to be considered in establishing transmission objectives for an
analog network are signal attenuation, noise, interference, crosstalk, distortion, ech-
oes, singing, and various modulation- and carrier-related imperfections.
Signal Attenuatlon
Subjective listening tests have shown that the prefened
acoustic-to-acoustic loss
[12]
in a telephone connection should be in the neighborhood of I dB. A study oflocal tele-
phone
connections
[l3]
demonsftates that the typical local call had only 0.6 dB more
loss than ideal. Surveys of the toll network
[4]
indicated that the average analog toll
connection had an additional 6.7 dB of loss. This same survey also showed that the
standard deviation of loss in toll connections was 4 dB (most
of which was attributable
to the local loops).
Since trunks within the toll network used amplifiers to offset transmission losses,
it would have been straightfbrward to design these trunks with zero-decibel nominal
insertion loss. However, as discusrted later, echo and singing considerations dictated
a need for ceftain minimum levels of net loss in most analog trunk circuits.
hterterence
Noise and interference are both characterized as unwanted electrical energy fluctuat-
ing in an unpredictable manner. Interference is usually more structured than noise
since it arises as unwanted coupling from
just
a few signals in the network. If the in-
terf'erence is intelligible, or nearly so, it is referred to as crosstalk.* some of the major
sources ofcrosstalk are coupling between wire pairs
in a cable, inadequate filtering or
carrier offsets in older FDM equipment, and the effects of nonlinear components on
FDM signals.
Crosstalk, particularly
if intelligible, is one of the most disturbing and undesirable
imperfections that can occur in a telephone network. Crosstalk in analog systems is
particularly
difficult to control since voice signal power levels vary considerably (i.e.,
acro$s a dynamic range of40 dB). The absolute level ofcrosstalk energy from a high-
level signal must be small compared to a desired lowJevel signal. In fact, crosstalk is most
noticeable during speech pauses, when the power
level ofthe desired signal is zero.
Two basic forms of crosstalk of concern to telecommunications engineers are near-
end crosstalk (NEXT)
andlitr-end crosstalfr
(FEXT).
Near-end crosstalk refers to cou-
pling from a transmitter into a receiver at a common location. often this form of
crosstalk is most troublesome because of a large difference in power levels between
the ffansmitted and received
signals. Far-end crosstalk refers to unwanted coupling
into a received signal from a transmitter at a dirrtant location. Both forms of crosstalk
are illustrated in Figure 1.24.
Nolse
The most common form of noise iuralyzed in communications systems is white noise
with a caussian
(normal)
di$tribution of amplitude values. This type of noise is both
-Crosstalk
is also used to characterize signallike interferences in nonvoice networks. For cxample, crosstalk
in a data circuit would refcr to an interfering signal being coupled in from another similar data circuit.
I.2 THE ANALOG NETWOFK HIEHAHCHY 35
Flgure I.24 Near-end and far-end crosstalk,
easy to analyze and easy to find since it arises as thermal noise in all electrical com-
ponents. Battery systems used to
power customer loops are also a source of this type
of noise. White noise is truly random in the sense that a sample at any in$tflnt in time
is completely unconelated to a sample taken at any other instant in time. The other
most cornmon forms of noise in the telephone network are impulse noise and quanti-
zation noise in digital voice terminals
(Chapter 3). Impulse noise can occur from
switching transients in older electromechanical switching offices or ftom rotary dial
telephones. Step-by-step switches were the most frequent culprits. More modern elec-
tromechanical switches that use
glass-encapsulated reed relays for crosspoints
pro-
duce much less noise. Whereas white noise is usually quantified in terms of average
power, impulse noise is usually measured in tems of so many impulses
per second.
Impulse noise is usually of less concern to voice quality than background white noise.
However, impulse noise tends to be the greatest concern in a data communications cir-
cuit.
The power level of any disturbing signal, noise or interf'erence, is easily measured
with a root-mean-square
(rms)
voltmeter. However, disturbances at some frequencies
within the passband of a voice signal are subjectively more annoying than others.
Thus. more useful measurements of noise or interference power in a speech network
take into account the subjective effects of the noise as well as the power level. The two
mortt cornmon such measurements in telephony use a C-message weighting curve and
a psophometric weighting curve, as shown in Figure 1.25. These curves essentially
represent filters that weight the frequency spectrum of noise according to its annoy-
ance eff'ect to a listener. C-message weighting represent$ the response of the 500-type
telephone set. As f'ar as perceived voice quality is concerned, only the noise that
gets
passed by the telephone set is imporlant. Notice that disturbances between
I and 2 kHz
are most.
perceptible. C-message weighting is used in North America while
psophometric weighting is the European
(ITU-T) standard.
A standard noise reference used by telephone engineers is I pW, which is l0-12 W,
or
-90
dBm
(dBm
is
power in decibels relative to a milliwatt). Noise measured rela-
tive to this reference is expressed as so many decibels above the reference
(dBrn).
Thus, a noise level of 30 dBrn corresponds to
-60
dBm, or 10=e W of
power' If the
readings are made using C-message weighting, the power level is expressed by the ab-
breviation dBrnC. Similarly, psophometrically weighted
picowatts are expressed by
36 BACKGROUND AND TEBMINOLOGY
0
- 10
-20
250 500 1000 zilro 40(n
Frequency
(Hz)
Figure 1.25 C-message and psophometic
weighting.
the abbreviation pWp. The relationships between various noise power
measurements
are given in Table L9.
The quality
of an analog voice circuit is usually not specified in terms of the clas-
sical signal-to-noise ratio. The reason is that relatively low levels of noise or interfer-
ence are noticeable during pau$e$
in speech, when there is no signal. on the other
hand, high levels of noise can occur during speech and be unnoticeable. Thus, absolute
levels of noise are more relevant than signal-to-noise ratios for speciffing voice qual-
ity.- The objectives for maximum noise levels in the analog AT&T network were 2g
dBrnC for connections up to 60 miles in length and 34 dBrnC for 1000-mile circuits.t
Die|p.rtlon
In a previous
section signal attenuations were considered with the tacit assumption
that a received waveform was identical in shape to a source waveform but merely
scaled down in amplitude. Actually, a received waveform generally
contains certain
distortions not attributable to external disturbances such as noise and interference but
that can be attributed to internal characteristics of the channel itself. In contrast to
noise and interference, distortion is deterministic; it is repeated every time the same
signal is sent through the same path
in the network. Thus distortions can be controlled
or compensated for once the nature of the distortion is understood.
There are many different type$ and sources of distortion within the telephone net-
work. The telephone companies minimized those types of distortion that most affected
the subjective quality
of speech. Later on they also became concemed with distortion
effects on data transmission. Some distortions arise from nonlinearities in the net-
work, such as carbon microphones, saturating voice-frequency amplifiers, and un-
*It
is a "orn rron pracrice
in the industry to specify the quality
of a voice circuit in terms of a
test-tone-to-noise ratio. However, the test tone must be at a specific power level so the ratio, in fact,
*specifies
absolute noise power.
'These
noise power values iue related to a
particular point in a circuit, called ^
zero-transmission-Ievel
paizt, discussed later,
c
!
tl
fl
d
CE
1.2 THE ANALOG NETWOHK HIERARCHY
TABLE 1.9 Relatlonships between Various Noise
Measurement8
To Convert
From
37
To
dBm
dBm
dBm
dBrn
dBc
pW
3 kHz llat
3 kHz flat
3 kHz flat
3 kHz flat
dBrn
dBrnC
dBp
dBrnC
dBp
pwp
Add 90 dB
Add BB dB
Add 87.5 dB
Subtract 2 dB
Subtract 0.5 dB
Multiply by 0.562
matched compandors
(Chapter 3). Other distottions are linear in nature and are usually
characterized in the frequency domain as either amplitude distortion or phase distor-
tion.
Amplitude distortion refers to attenuating some frequencies in the voice spectrum
more than others. The Ioading coils discussed earlier represent one means of eliminat-
ing amplitude distoltion on long voice-frequency
wire pairs. Amplitude
distortion
could also be introduced by spectrum-limiting filters in FDM equipment. Ideally these
filters should uniformly
pass all voiceband frequencies up to 4 kHz and reject all oth-
ers. Practical designs, however, imply the need for gradual attenuation
"roll-offs"
be-
ginning at about 3 kHz. Figure 1.26 shows the attenuation-versus-frequency
respon$e
of a typical analog toll connection of the
past.
Phase distortion is related to the delay characteristics of the transmission medium'
Ideally a transmission sy$tem should delay all frequency components in a signal uni-
Freguency lkHzf
Figure 1.26 Insertion loss versus frequency oftypical toll connection.
c
E
a
3
c
o
*
E
38 BAcKcRoUNDANDTERMtNoLoGy
formly so the proper phase
relationships exist at the receiving terminal. If individuat
frequency component$ experience differing delays, the time-domain representation at
the output becomes distorted because superposition of the frequency terms is altered
at the output. For reasons not discussed here the delay of an individual frequency com-
ponent is usually referred to as its envelope delay. For a
good
explanation ofenvelope
delay see reference
[5].
Uniform envelope delay relates to a phase re$ponse ttrat is directly proportional
to
frequency. Thus systems with uniform envelope delay are also referred to as linear
phase
systems. Any deviation from a linear phase
characteristic is referred to as
phase
distortion. The perceptual
effects ofphase distortion to a voice signal are small. Thus
only minimal attention need be given to the
phase
respon$e of a voice network. The
phase
respon$e and corresponding envelope delay providecl
by a typical analog toll
connection is shown in Figure 1.27.
In addition to the distortions
just
mentioned, analog carrier systems sometimes in-
troduced other frequency-related distortions such as frequency offsets,
jitter,
phase
hits, and signal dropouts. The effects ofthese imperfections and phase
distortion were
adequately controlled for voice traffic but presented
difficulties for high-rate voice-
band data traffic.
Echoee and Slnging
Echoes and singing both occur as a result of transmitted signals being coupled into a
return path and fed back to the respective sources. The most common cause of the cou-
pling
is an impedance mismatch at a four-wire-to-two-wire
hybrid. As shown in Fig-
ure 1.28, mismatches cause signals in the incoming branch of a four-wire circuit to get
coupled into the outgoing branch and retum to the source. In the older network with
two-wire analog end office switches, it was impractical to provide good
impedance
matches at this
point
for all
possible
connections since the two-wire side of the hvbrid
Frquency
(kHrl
Figure 1.27 Envelope delay and phase response of typical toll connection.
1.2 THE ANALOG NETWOFK HIERARCHY 39
Figure 1.28 Generation of echos at two-wire-to-four-wire
interface.
could be connected to many different local loops, each with its own characteristic im-
pedance.
If only one reflection occurs, the situation is refened to as
"talker
echo."
[f a second
reflection occurs.
"listener
echo" results. When the returning signal is repeatedly cou-
pled back into the tbrward
path to
produce oscillations, singing occurs' Basically,
singing results ifthe loop gain at some frequency is greater than unity' Ifthe loop
gain
is only slightly less than unity, a near-singing condition causes damped
oscillations.
Singing and near-singing conditions have a disturbing effect on both the talker and the
listener. Talker echo is usually the most noticeable and ffoublesome.
The degree of echo annoyiunce experienced
by a talker is dependent on both the
magnitude of the returning signal and the amount of delay involved
U
6' 171 ' On short
connections the delay is small enough that the echo merely appears to
the talker as
natural coupling into his ear. In fact, a telephone is
purposely designed to couple some
speech energy
(called sidetone) into the earpiece. Otherwise, the telephone seems
dead to a talker. Near-instantaneous echoes merely add to the sidetone and
go unno-
ticed. As the roundtrip delay increases, however, it becomes necessary to increasingly
attenuate the echoes to eliminate the annoyance to a talker' Hence, long-distance cir-
cuits require significant attenuation to minimize echo annoyance' Fortunately, an echo
experiences twice as much attenuation as does the forward-propagating
signal since it
traverses twice the distance. Intermediate-length connections are typically designed
with 2-6 dB of path attenuation depending on the delay. All transmission
links within
the Bell System were designed with specific amounts of net loss called via net loss
(VNL) that depended on the length of the link and the
position in the hierarchy
[17].
In general, the VNL network design established end-to-end attenuation
in proportion
to the length of the circuit
Connections that
produce more than 45 msec of roundtrip delay
(representing 1800
miles of wire) require more attenuation for echo control than can be tolerated
in the
forward
path. In these cases one of two types of devices was used to control the echo:
an echo suppressor or an echo canceller.
As shown in Figure 1.29, an echo suppressor operates on four-wire circuits by
measuring the speech
power in each leg and inserting a large amount of loss
(35
dB
typically) in the opposite leg when the
power level exceeds a threshold. Thus, a return-
ing echo is essentially blocked by the high level of attenuation. Notice that an echo
riuppressor converts a full-duplex circuit into a half-duplex circuit with energy sensing
being the means of turning the Iine around.
One drawback of echo suPPressors for voice circuits,was that they might clip be-
ginning portions of speech segments. If a
party at one end of a connection begins talk-
40 BACKGROUND AND TEBMINOLOGY
Figure 1.29 Echo suppressor.
ing at the tail end of the other
parfy's
speech, the echo suppres$or does not have time
to reverse directions. Echo suppressors with good performance
are able to reverse di-
rections in 2-5 msec
[16].
For the fastest possible
release of backward attenuation,
split-echo suppressors are necessary. A split-echo suppressor is one that separates the
echo control of each direction so the loss insertion of each direction is closest to the
point of echo occurrence.
A second, and much preferred,
form of echo contror is echo cancellation
[lg,
I
ga].
As shown in Figure 1.30, an echo canceller operates by simulating the echo path to
subtract a properly
delayed and attenuated copy of a transmined signal from the re-
ceive signal to remove (cancel)
echo components. Thus echo cancellation requires
training to determine how much delay and attenuation are needed to simulate the echo
characteristics of the circuit. Notice in Figure 1.30 that echoes are canceled close to
the source so that delays in the echo canceller are minimized. The important feature
of an echo canceller is that it maintains a full-duplex
circuit so clipping does not occur.
Satellite circuits with greater
than 500 msec of roundtrip delay required echo cancel-
lers for acceptable performance.
Because the cost of digital signal processing (DSp)
technology has dropped so dramatically, echo cancellers are now used in any situation
requiring echo control.
Full-duplex voiceband modems (v.32
and rater) incorporate echo cancellers di-
rectly in their receive circuitry. Thus, network-based
echo cancellers are unneces-
sary*-and sometimes undesirable because tandem echo cancelling may not work
properly
if two echo cancellers do not cooperate in the haining process.
Network-
based echo cancellers can therefore be disabled by a modulated 2lfi)-Hz tone
(echo
suppressor$ were also disabled with a 2100-Hz tone) transmitted at the start of a con-
nection
[19].
Another method of echo control involves impedance matching of the hybrids to re-
duce the magnitude of the echo. some electronic hybrids provide
dynamic balancing
to automatically eliminate or reduce echoes. In fact, a coflrmon way of implementing
the impedance matching is to build an echo canceller with near-zero delay. These cir-
cuits eliminate, or greatly
reduce, echoes occurring at the associated hybrid but do not
eliminate echoes that may occur elsewhere in the network. For a detailed discussion
"If
ttre terminal (e.g.,
modem) echo canceller has insufhcient delay buffering for very long echoes, the
network echo canceller at the far end of a corurection may be necessary.
- T l
I L F
1 -
1.2 THE ANALOG NETWORK HIEFARCHY
41
Figure 1.30 Echo canceller'
of all types of echo control, see reference
[20].
In
general, the
procedures used to con-
trol echoes also control singing. On some fairly short connections, however, no echo
conffol is necessary, and singing may become a
problem'
1.2.9 Power Levels
As indicated in previous paragraphs, voice signal
power in a long-distance
connection
needs to be rigidly controlled. The tleliverecl signal
power must be high enough to be
clearly
perceived but, at the same time, not be so strong that circuit instabilities such
as echo and singing result'
To maintain rigid control on the end-to-end
power level of a circuit involving a va-
riety of transmission sy$tems, telephone companies necessarily controlled
the net at-
tenuation and amplification of each analog transmission system. These systems were
designed for a certain amount of net loss
(VFil-) as described
previously.
To administer the net loss of transmission
links, the fransmission levels of various
points in a transmission syrttem are specified in terms of a reference
point- ITU-T rec-
ommendations
call this
point the zero-relative-level
point and the North American
term is a zero-transmission-level
point
(0-TLP). The reference
point may not exist as
an accessible
point but has long been considered to be at the sending end terminal of
a two-wire switch. In North America the sending end of a four-wire switch is defined
to be a
*z-dB
TLP. Hence, a 0-dB TLP is only a hypothetical
point on a four*wire cir-
cuit. Nevertheless, it is useful in relating the signal level at one point in the circuit to
the signal level at another
point in the circuit.
If a O-dBm
(l-mW) test tone is applied at a O-TLP, the power level at any other
point in the circuit is <letermined directly
(in
decibels referred to I mW) as the TLP
value at that point. It should be emphasized, however, that TLP values do not specify
power levels; they specify only the
gain or loss at a
point relative to the reference
point.
Signal
power levels or noi$e
power levels are not normally expressed in terms of
local measured values. Instead,
powers are expressed in tetms of their value at the 0-
TLP. For example, if an absolute noise
power of 100 pw (20 dBrn or
-70
dBm) is
measured at a
-6-dB
TLP, it is expressed as 26 dBrnQ.
(The
"0"
indicates that the
specification is relative to the O-TLP.) If noise
power is measured with C-message
weighting, it is designated as so many dBmC0. Similarly,
psophometric weighted
42 BACKGRoUNDANDTERMI NOLoGY
-zdB _l 3dB
_4d8
TLP TLP
TLP
Figure 1.31 TLp values for Example l.l.
noise is commonly expressed in units of picowatts psophometrically
weighted
(dBm0p
or pWpO).
Example 1.1. using Figure 1.31 for reference, determine each of the following: (a)
the signal power
to be applied at point
B ro determine if points
A and c are at the
proper levels; (b)
the amount of
gain (loss)
a signal experiences when propagating
from A to c; and (c)
the amount of noise that would be measured at c if zj dbrnc of
absolute noise is measured at B and no additionat noise occurs on the B-to-c link.
solution. (a)
Because point
B is a
-l3dB
TLp, the proper
test tone level is
-13
dBm
(0.05
mw). (b)
Because the TLP values drop by 2 dBm, there is 2 dB net loss from A to
c.
(c)
An absolute measuremenr of 27 dBmc at B is 40 dBmc0. This is also 40 dBmc0
at C. The absolute noise power
measured at C would be 40
-
4
=36dBmC.
To put signal and noise powers
in perspective,
a survey of voice signals in the Bell
system
[21]
indicated rhe average speech signal has
-16
dBm0 of power.
hr compari-
son, the noise power
objective for a 1000-mile analog circuit was 34 dBrnc0 (-56
dBmCO). Thus, the noise is 40 dB below the signal power.
1. 2. 10
$i gnal l ng
The signaling functions of a telephone network refer to the means for transferring
network-related
control information
between the various terminals, switching noaes,
and users of the network. There are two basic aspects of any signaling system: spe-
cially encoded electrical waveforms (signals)
and how these waveforms should be in-
terpreted. The most common control signals to a user are dial tone, ringback, and busy
tone. These signals are refered to as call progres$
tones and may never change. The
signaling procedures
used internally to the network are not constrained by user con-
vention and were often changed to suit particular
characteristics of transmission
and
switching systems. As a result, the analog network used a wide variety of signaling
schemes to transfer control information between switching offices.
Signallng Functlone
signaling functions
can be broadly categorized as belonging to one of two rype$: su-
pervisory
or information bearing. Supervisory signals convey statu$ or conffol ofnet-
work elements. The most obvious examples are reque$t for service (off-hook),
ready
1.2 THEANALOG NETWORKHIERARCHY
43
to receive address
(dial tone), call alerting
(ringing), call termination
(on-hook), re-
quest for an operator or feature invocation
(hook flash), called
party ringing
(ring-
back), and network or called
party busy tones. Information bearing signals
include
called
pafiy address, calling
party address, and toll charges. In addition to call-related
signaling functions, switching nodes communicate
between themselves
and network
control centers to provide certain functions related to network management'
Network-
related signals may convey status such as maintenance test signals, all trunks, busy,
or equipment failures or they may contain information related to routing and
flow con-
trol. Chapter 7 discusses some of the basic considerations of network management
for
routing and flow control.
In-Ghannel Slgnaling
Signals are tran;mitted
with one of two basic techniques: in-channel signaling or
common-channel
signaling. In-channel signaling
(sometimes referred to as
"per-trunk
signaling") uses the same transmission
facilities or channel for signaling
as for voice'
Common-channel
signaling, as discussed in the next section, uses one channel for all
signaling functions of a
group of voice channels. In the
past most signaling systems
in the telephone network were the in-channel variety.
In-channel signaling systems can be further subdivided
into in-band
and out-of-
band techniques. In-band systems transmit the signaling information in the same band
of frequencies used by the voice signal. The main advantage of in-band signaling is
that it can be used on any transmission medium. The main disadvantage
arises from a
need to eliminate mutual interference between the signaling waveforms and a user's
speech. The most
prevalent example of in-band signaling was single-frequency
(SF)
signaling, which used a 2600-Hz tone as an on-hook signal for interoffice
trunks' Al-
though normal speech rarely
produces a
pure 26ff)-Hz signal, inadvertent
disconnects
have occurred as a result of user-generated
signals. Two other common examples of
in-band signaling are addressing by dual-tone
multifrequency
(DTMF) signals from
push-button telephones or multifrequency
(MF) signaling between switching offices'
ln-channel but out-of-band
signaling uses the same facilities as the voice channel
but a different
portion ofthe frequency band' Thus out-of-band signaling represents a
form of FDM within a single voice circuit. The most coilmon instance of out-of-band
signaling is dc signaling as used on most customer loops. With this form of signaling,
the central office recognizes the off-hook condition by the flow ofdirect current in the
line. Other commonly used loop signals are dial
pulses generated by a rotary dial at a
rate of 10
pulses per second and a 20-Hz ringing voltage from the central office' All
of these signals use lower frequencies than those
generated in speech' Thus there is no
possibilityof one being mistaken for the other. The major disadvantage
of out-of-band
signaling is its dependence on the transmission
system' For example, SSB carrier sys-
tems fiher out the very low frequencies associated
with each voice channel. Thus, the
on-hooVoff-hook
signal must be converted to something like SF signaling
for FDM
transmission. Out-of-band
signaling is also implemented
with frequencies
above the
cut-off frequency of voice separation filters but below the 4-kHz limit of a channel'
CCITT recommends the use of 3825 Hz for this
purpose.
44 BACKGEoUNDANDTERMINoLoGY
Common-Channel
Signaltng
Instead of sending signaling information over the same facilities that carry the mes-
sage haffic (voice
signals), common-channel
signaling (ccs)
uses a dedicated data
link between the stored-program
control elements (computers)
of switching systems.
Figure 1.32 depicts such a data link between two switching offices. Notice that the per-
trunk signaling equipment associated with the trunks has been eliminated. The data
link sends message$ that identify specific trunks and events related to the trunks. Thus
the type of ccs shown in Figure I.32 is refened to as
"channel-associated"
common-
channel signaling. The main advantages of CCS arel
1. only one set of signaling facilities is needed for each associated trunk group
instead of separate facilities for each individual
circuit.
2. A single dedicated confrol channel allows transfer of information such as
address digits directly between the control elements (computers)
of switching
offices. In-channel system$, on the other hand, must have the control
information switched
from the common control equipment of the originating
office onto the outgoing channel, and then the receiving office must switch the
incoming control information from the voice channel into its common control
equipment' The simpler procedure
for transferring information
directly between
switch processors
is one of the main motivations for CCS.
3. since separate channels are used for voice and control, there is no chance of
mutual interference.
4. Since the control channel of a common-channel system is inaccessible to users,
a major means for fraudulent
use of the network is eliminated.
5. connections involving
multiple switching offices can be set up more rapidly
since forwarding of control information from one office can overlap a circuit set
up through the node. With in-channel systems the associated circuit must first
be established before the control information can be transferred across it. The
ccs data link can al$o operate at a much higher data rate than common analog
network signaling schemes, implying even faster connection setup.
6. The channel used for ccs does not have to be associated with any particular
trunk group.
In fact, the control information can be routed to a centralieed
conhol facility where requests are processed
and from which switching offices
Per Chanrrel
Signeling
Flgure 1.32 Trunk-group-associated
common-channel
signaling.
1.2 THE ANALOG NETWORK HIERAHCHY
45
receive their connection control information. Figure 1.33 depicts a CCS
network that is disassociated
from the mes$age network structure. One
advantage ofcentralized
control is its ability to process requests with knowledge
of networkwide traffic conditions. Centralized control is also atffactive for
.
managing multiple switching offices that are too small to warrant call
processing facilities of their own. The transition from in-channel signaling to
disassociated CCS at the network level is analogous to the lower level transition
from direct progressive control switches
($tep-by-step) to common
control
switches.
The major disadvantages
of CCS are:
1. Control information
pertaining to an established circuit, such as a disconnect,
must be relayed from one node to the next in a store-and-forward
fashion' An
in-channel disconnect signal, on the other hand, automatically
propagates
through the network, enabling all nodes involved in the connection
to
simultaneously
process the disconnect and release the associated facilities.
2. If one node in a common-channel
system fails to relay the disconnect
information
properly, facilities downstream
from the disconnect
will not be
released. Thus, a high degree of reliability is required for the common
channel-both
in term$ of physical facilities
(duplication) and in terms of error
control for the data link.
3. Since the control information ftaver$es a separate
path from the voice signal,
there is no automatic test of the voice circuit as when the voice channel is used
to transfer control information. Initial usage of CCS sy$tems included special
provisions for testing a voice circuit when it was set up'
4. In some instances all trunks in a
group do not logically terminate at the same
switch. Figure 1.34 depicts a multiplexed transmission
link from a PBX to the
_-t
g -
Network
control
c6nter
r/ Control
/' circult
Figure 1.33 Dissociated common-channel
signaling
network'
46 BAcKcRoUNDANDTERMtNoLocy
ccs
( D
Channel )
Ti e Li ne FBX PBX
Figure 1.34 TDM Iink with multiple destinations for the channels.
public
network. As indicated, some of the channels compose a trunk group
to
the local CO but other
channels may represent tie lines to other PBXs or foreign
exchange circuits ro other cos. The digital cross-connect system (DCs)
is a
specialized switching system (described
in Chapter 5) that routes individual
circuits within a trunk group
to individual locations. If CCS is implemented on
the trunk group (as
in an ISDN primary
rate interface, described in chapter l1),
the signaling information for the laffer two trunk groups
terminates at the co.
The CO must then forward the information to the respective destinations
(assuming
those destinations are equipped with CCS). All-in-all rhese laner two
cases are easier to handle if the signaling information
accompanies the message
channels as it does automatically with in-channel signaling.
As a final note, it should be pointed
out that some signaling functions originating
or terminating
with an end user inherently require in-channel implementations.
For
example, data terminals wishing to disable echo suppressors or echo cancellers in dial-
up connections need to send special tones that get
recognized by echo control elements
in the circuit. In similar fashion automatic originating facsimile equipment generate
I 100-Hz tones to allow automatic recognition of the originating equipment type at the
destination.
1.2.11 Anal og Intedaces
The design, implementation,
and maintenance of any large and complex system re-
quire paditioning
of the system into subsystems. Associated with each subsystem is
an interface
that defines the inputs and ouSuts independent of internal implementa-
tions' Well-established interfaces
are a fundamental requirement
to maintain compati-
bility between old and new equipment.
within the u.s. telephone network
DigitEl
Cros*Connect
Systom *""S9#
Local
Serving
Office
1.2 rHE ANALoc NETWoRK HIERAR0HY 47
standardized
interfaces are
particularly necertsary to support competition among
equipment suppliers and service
providers in almost all facets of the network' The
ptiniiput analog interfaces used in the periphery of the network are subscriber loops,
ioop-start trunks,
ground-start trunks, direct-inward-dial trunks, and E&M trunks.
$ubscriher
LooP lntartace
The most common interface in the network involves the two-wire connection of indi-
vidual telephone lines to end office switches: the subscriber
loop intedace. Because
of the nature of industry standard telephones
(2500 station sets) and the electrome-
chanical switches to which they were originally connected, this interface has a number
of characteristics that are difficult to satisfy with modem integrated circuit technology.
The fundamental characteristics
of this interface are;
l. Battery: Application of dc
power to the loop
(48 V normally)
to enable dc
signaling and
provide bias current
for carbon microphones'
2. Overvoltage
Protection: Protection of equipment and
personnel from lightning
strikes and
power line induction or shorts.
3. Ringing:Applicationofa20-HzsignalatS6Vrmsforringerexcitation.Typical
cadence is 2 sec on and 4 sec off.
4. Supervision: Detection of off-hoolq/on-hook
by flow/no-flow of dc current'
5. Test:Access to the line to test in either direction-toward
the subscriber or back
into the switch.
In the case of a digital end office, two more functions are necessarily
required; two-
wire-to-four-wire
conversion
(hybrtil and analog-to-digital coding
(and digital-to-
analog decoding). Taken together, these functions are referred to as BORSCHT'
The
basicimplementation
of a BORSCHT interface is described in the context of digital
switching in Chapter 5.
LooP-Start
Trunks
A loop-start
(LS) h'unk is a two-wire connection between switches
(usually between
a central office and a PBX). From an operational
point of view, an LS trunk is identical
to a subscriber
loop. Thus an LS interface in a PBX emulates a telephone by closing
the loop to draw current for call origination and by detecting ringing voltages for in-
coming catls. To send address information, the PBX interface
generally waits for a few
seconds and assumes that a dial tone is
present before sending DTMF tones or gener-
ating dial
pulses by interrupting loop cunent)tsome
PBXs
provide dial tone detection
so faulty equipment
or connections are mordAasily recognized and addressing can be
sent as soon as the otherend is ready.
|
1
One significant difficulty
with two-#ay LS trunks arises when both ends of the line
seize the line at the same time
(or nearly the same time)' Because both ends of the line
think they are originating a call, the line becomes hung. This situation is commonly
referred io as glare. Ifthe PBX detects a dial tone before sending digits, it will recog*
nize the glare condition by timing out on the wait for dial tone and can then generate
48 BAcKGRoUNDANDTERMtNoLocy
a disconnect to release
the glare
condition but drop the incoming
call. More com-
monly, the PBX blindly sends the address digits and connects the originating
pBX
sta-
tion to the line. Generally,
this means the incoming call gets
connected to the wrong
station. For this reason, LS trunks are normally used only as one-way trunks: either
one-way incoming or one-way outgoing.
Ground-Start Trunks
The aforementioned problem
with glare
on two-way LS trunks can be largely resolved
by augmenting the call origination proces$
to use ground-start (GS) procedures
[22].
When originating a call, the end office applies a ground potential
to the tip lead of the
tip and ring pair
and waits for the
pBX
to acknowledge the seizure by drawing loop
current. when the PBX originates a call, it first applies ground
to the ring teao ana
closes the loop waiting for loop current. (The
co does not apply battery during an idle
state as it does in ar Ls interface.)
The co acknowledges the connect r.quest by up-
plying
battery to the tip/ring pair
and momentarily applying ground
to the tip. (Mai;-
taining low-noise
circuits requires removal of all paths
to ground
during the connected
$tate.) A GS protocol prevents
simultaneous seizures unless the originations
occur
within a few hundred milliseconds
of each other. In contrast, an LS protocol
allows
multiple seizures to occur with windows up to 4 sec (the
silent interval between ring
bursts). Moreover,
a
glare
condition on GS trunks can be recognized
by the interface
equipment so it can be resolved by redirecting the calls to different trunk circuits.
Another advantage of GS trunks is the ability of the CO to signal network discon-
nects to the PBX (the
co remove$ battery). witir I-s trunks the network does nor gen-
erally provide
disconnect signaling so the
pBX
must rely on the end user to hang up.
(A
situation that often produces
hung trunks with data connections.) Furthermore.
when the co places
an incoming call that eventually gets
abandoned, because no one
answers, a co immediately signals the abandonment
by removing ground
from the
tip lead. with LS trunks, abandonecl calls can be recognized only by the absence of
ring voltage, which can take 6 sec.
D i rect4 nward-Dial
Tru n ke
Direct-inward-dial (DID)
trunks are particularly
simple two-wire trunk interfaces be-
cause they are always one-way trunts: incoming only with respect to a
pBX.
As im-
plied
by the name, they allow a serving co to forward the extension number of
incoming calls so a PBX can immediately route the call to a destination without going
through an attendant. In contrast to LS and CS trunks, the
pBX
end of a DID trunk
provides
battery voltage so the co can signal an incoming call by merely closing the
loop to draw current. After the PBX reverse$ battery momentarily (winks)
to signify
it is ready to receive digits, the co either generates
dial pulses
or DTMF tones to send
the extension number (two,
three, or four digits). Ai'ter the designated station answer$,
the PBX reverses battery again to signify the connected state and hotds that state for
the duration ofthe call. DID trunks are also referred to as
..loop
reverse-battery
$uper-
vision" trunks with variations
in the signaling protocol
depending on the type of co
122t.
1.2 THE ANALoc NETwoRK HIERARGHY 49
E&M Trunks
As indicated in Figure 1.35, an E&M trunk is defrned as an intetface to a transmission
sy$tem and not a kansmission system itself. The interface in Figure 1.35 has a four-
wire voice
path, an E lead with an associated
renrm lead
(SG), and an M lead with an
associaled return
(SB). Thus there are eight wires in this interface
(referred
to as a type
II E&M interface). Other types of E&M interfaces are defined with as
few as four
wires
[23]
(a two-wire voice path, an E lead, and an M lead with earth ground returns).
In any type of E&M interface, supervision signaling is always conveyed on the E
and M leads and not on the voice pair (or pairs). The PBX signifies off-hook by closing
the M-SB loop to draw current while the transmission
equipment indicates off-hook
by closing the E-SG loop to draw current. How the transmission
equipment conveys
the supervision
is a function of the transmission link. A variety of timing protocols are
defined for the $tart of addless signaling, which can be in-channel DTMF tones or dial
pulses generated by momentary opens on the respective E and M leads'
-
Althbugh E&M signaling is formally defined as
just
an interface, they are often
used
(withup to four
pairs of wires) as direct connections
between PBXs. Because of
the requirement for multiple
pairs, such applications
usually occur when the PBXs are
located within a single building or campus complex. The availability
of extemal con-
trol leads allows the use of E&M interfaces for special applications such as paging sys-
tems. where the M lead can be used to tum on the loudspeaker'
1.2.12 The Intel l i gent Network
The first common channel signaling facilities of AT&T were installed in
the toll net-
work between a No. 4A crossbar switch in Madison, wiSCOnsin, and a No. 4 ESs
switch in chicago, Illinois, in 1976
[24].
The 2400-bps ccs data links between sPc
switching offices replaced in-channel SFA4F signaling facilities to reduce
costs and
significantly
speed up long-distance-call
connect times. As more and more of the toll
switches were implemented with CCS capabilities, a CCS network evolved
that not
only improved the
performance of the existing
network but also established a
platform
for the introduction of new features. As indicated in Figure 1.36, the CCS network be-
Figure 1.35 TyPe lI E&M interface.
50 BAcKcRouNoANDTERMtNoLocy
Figure 1.36 CCS networMntelligent
network platform.
came a disassociated packet-switching
network that separated control of the network
from the swirching machines themselves.
The packet-switching
nodes of the CCS network are referred to as signal transfer
points (srPs).
The network control points (NCps)
of Figure 1.36 initially represented
network database servers
that defined how to route calls, verify credit cards, or process
special services such as 800 numbers. The same basic structure is also installed within
the LATAs to extend ccs features all the way to the end offices. The original com-
munications protocol
used between ccs entities was ccITT signaling system No. 6
(ccs6).
In the early 1980s this protocol
was replaced
by ccITT signaling system
No. 7 (SS7).
See reference
[25]
for a good
overview of SS7.
As the capabilities of the NCps are expanded from being data base servers ro proc-
essing service requests, the concept of an intelligent network (IN)
126l
begins to take
shape. In its ultimate form advanced services of the network are executed in the NCps.
or service control points (sCPs),
as they are called in the IN, and not in the switching
machines themselves.
The switching machines provide
only the switching fabric; con-
nection commands come from an SCp. The fundamental motivations
behind develop-
ing the IN are:
I ' To enable the deployment of network-based features such as citywide centrex
mentioned previously
2. To allow the introduction
of new features without having to change the
hardware or software of switching vendors
To speed up the design/deployment
cycle ofnew features because onlv the scp
software needs upgrading
To allow customizing of services by the operating companies according to the
needs of their customers
4.
The obvious disadvantage of such centralized control is the vulnerability
of the net-
work to a failure in the SS7 network or the SCP. For this reason basic service is likely
to remain with the switching
machines. In this case, softwarc in the switching ma-
chines recognizes
special service situations as software triggers to involve an scp.
1.2 THEANALOGNETWOHKHIERARCHY
51
Leaving basic service processing in the switching machines also reduces the load on
the SS7 links and the
processing speed in the SCPs.
1.2.13 Dynami c Nonhi erarchl cal
Routl ng
The alternate routing
procedures shown in Figure 1.4 for the hierarchical network of
the Bell System were necessarily simple and straightforward because of the original
reliance on mechanical switching machines. Such simplicity, however, leads to the
following limitations:
l. Even though the topology of the network would allow a large number of routing
altematives, consfiaints on how the routes were selected limited the actual
choices to a relatively small number.
2. The routing
patterns could not be changed according to the time of day or
network traffi c
Patterns.
3. The
proce$s of
progressive setup
precluded retracing a path to a previous
switching node'and trying a new, untried route when a previous choice
led to a
node that was blocked'
4. A completed connection
may have involved a large number of intermediate
facilities using a lot ofnetwork
resources for one connection'
All of these deficiencies resulted from using relatively simple decision logic in each
node with no knowledge of the network status as a whole. The introduction of SPC
switching machines and CCS changed the situation so that more efficient routine
pro-
cedures were possible. Along these lines, AT&T completed the deployment of dy-
namic nonhierarchical routing
(DNHR) in 1987
127,281.
Specific features of DNHR
are:
l . Routing tables in the NCPs list all possible twolink routes in the order of cost
so use of network resources are minimized.
Routing is dynamic to take advantage of traffic noncoincidence'
Blocking at an intermediate node
produces
"crankback,"
which allows selection
of untried routes from the originating node.
Routes that produce too much delay for the echo cancellers
in the circuit
are
excluded.
Reference
[28]
reports that DNHR reduced blocking on Thanksgiving Day fuom34flo
in 1986 toSVo in 1987. After installation of DNHR Mother's Day traffic saw three to
four fimes as many flir$t-attempt completions as before. The implementation of DNHR
demonsffates that a network architecture
is as much a function of the software in the
network as of the
physical topology.
It also demonstrates that centralized
control of a
network is more vulnerable to failures. On January 15, 1990, an obscure bug in the
)
4.
52 BA0KGRoUNDANDTERMINoLoGY
CCS software of the No. 4 ESS switches in AT&T's long-distance
network prevented
completion of many calls during a 9-hour period
[291.
Ironically,
the bug occurred in
software intended to isolate the network from a node failure.
The combination of large digital switches, large-bandwidth (fiber
optic) rran$mis-
sion links, and development of the SS7 network has stimulated a transition from hier-
archical networks to
"flat"
networks wherein switching nodes are interconnected in a
functional mesh-switches are either connected directly to each other or through
cross-connect facilities (DCSs
and SONET rings described later).
1.2.14 Cellular Radio Telephone
System
Prior to 1983 mobile telephone
users in the United States were restricted to using the
services of radio common carriers (RCCs)
that had limited bandwidth and limited re-
sources available for services. The service environment was necessarily one of limited
availability and extreme congestion. October 1983 marks a significant date for mobile
telephony as the time when commercial cellular mobile telephone service started in
chicago, Illinois. As indicated in Figure 1.37, acellular system consists of a number
of radio base stations connected to a
(digital)
switching office refened to as the mobile
telephone switching office
(MTSO).
As a mobile subscriber move$ from one cell to
another, the MTSO automatically switches connections with the respective base sta-
tions to maintain a continuous connection with the public
switched network. The basic
advantages of the cellular architecture are:
l. The ability to reuse allocated channel frequencies within a serving area. Using
a combination of antenna directivity in the base stations and signal attenuation
Figure 1.37 Cellular mobile telephone topology.
1.2 THEANALOGNETWORKHIEHARCHY
53
from distant cells, the same channel can be reu$ed in one of every seven cells
t301.
2. Reduced transmit
power requirements for the mobiles. The power savings
advantage for automobiles
is primarily one of reducing the cost of the
transmitter. For hand-held units the
power savings is important for battery size
and time between recharges. In fact, widespread acceptance and use ofpersonal
radio systems require much lower transmit
power levels and consequently
smaller cells
[31].
3. Reduced occlrrences of multipath
propagation. Shorter distances imply less
chance of signal reflections causing multipath signal degradation.
4. Expandability. A system can be installed with comparatively large cells to
minimize start-up costs. After service requirements
grow and revenue is being
received, the capacity of the system can be expanded by subdividing the
congested cells.
5. Reliability. Because the cells actually have significant overlapping coverage,
if
one cell fails, neighboring cells can
provide service until repairs are made.
The FCC in the United States has defined 728 mobile service areas referred to as cel-
lular geographic service areas
(CGSAs). Each CGSA is allocated 832 radio channels,
which are equally divided between two competing service
providers; one a wireline
carrier and the other a nonwireline carier. The wireline carrier
provides local tele-
phone service in the area but competes with the nonwireline canier
(RCC) for cellular
service. To ensure effective competition, the wireline carrier must not
use facilities
that are shared with local telephone service. Specifically, the MTSOs must be separate
from the local switching offices. Thus, both types of carriers must backhaul
all traffic
to their respective MTSOs. The service
providers can have more than one MTSO in
an area but cannot interconnect them by the switched
public network. They are typi-
cally interconnected
by leased
private lines
(fiber) or digital microwave.
Cellular net-
works have
grown to cover large areas interconnected by dedicated long-distance
facilities that allow $ome cafriers to offer free long-distance calling when using a cel-
lular
phone.
When a mobile unit first activates itself, it scans the channels to determine which
idle channel ofapredefined
set ofcontrol channels has the strongest signal. Using that
channel the unit registers with the system to identify itself and
place calls-
After the
initialization
proce$$, the network continually monitors signals from the mobile and
conffols it to switch channels when necessary.
To complete calls to a mobile, the cel-
lular network
pages for the clesignated subscriber beginning with its home cell
(unless
the cellular control center already knows where the subscriber is located). When a sub-
scriber crosses an MTSO boundary, in addition to a cell boundary, a common-channel
signaling network is used to transfer the call to the new cell and the new MTSO. The
connection to the
public network is unchanged.
Thus, the original MTSO then be-
comes a tandem node to the public network'
Examples of analog cellular system$ are Advanced Mobile Phone service
(AMPS)
[30,
32], developed in the United States and deployed in North America;
Total Access
54 BACKGRoUNDANDTERMINoLoGY
communications system (TAcs),
developed in the united Kingdom and deployed in
several European countries; and the Nordic Mobile Telephone (NMT)
sy$tem, de-
ployed
in Scandinavia. As discussed in Chapter 9, digital cellular systems have been
developed to either replace the analog sy$tems or provide
service altematives.
1.2.15 Voi ceband DataTransmi ssl on
The primary concern of telephone companies, voice service, is sometimes referred to
as Pors: plain
old telephone service. In addition to
pors,
the telephone companies
also provide
a number of special services such as program
distribution of radio and
television, telephoto, teletype, facsimile, and data transmission.
In most cases, these
services are supported by adapting the sources to use the ubiquitous voice-grade tele-
phone
channel. An exception is television distribution, which was an initial stimulus
for the nationwide microwave radio network. To a certain extent, wideband FDM
voice signals were designed to conform to the network TV distribution facilities. Na-
tionwide TV distriburion is now primarily
by sarellites and fiber.
one significant aspect ofa voiceband telephone channel in regards to nonvoice ap-
plications
is the absence of dc and low-frequency transmission.
Although the metallic
circuit provided
in customer loops passes
frequencies down to and including dc, most
of the rest of the network
does not. The equipments that are primarily
responsible for
blocking the low frequencies are ffansformers in two-wire to four-wire hybrids, old
FDM separation filters, and digital voice encoders that purposely
block low frequen-
cies to avoid 60-Hz hum
(that
arises from power
line interference).
Because baseband
signals from facsimile and most data applications have frequency content down to dc,
these signals must be modulated for voiceband fransmission.
A second consideration when using the telephone network for data is the band-
width restriction of approximarely
3 kHz originally imposed by FDM separarion fil-
ters, loading coils, and more recently band{imiting filters in digital voice terminals
(Chapter
3). The main implication
of a resfficted bandwidth is a limitation on the sig-
naling rate, or baud rate, which in turn directly relates to the data rate. A common sig-
naling rate is 2400 symbols per second using carriers between 1700 and 1g00 Hz.
symbol rates of 4800 symbols per
second have arso bern used in lower sideband mo-
dems with a carrier at 2850 Hz.
As mentioned,
user acceptance of voice quality does not require stringent conhol
of the phase
response (envelope
delay) of the channel. High-speed data transmission,
however, requires comparatively tight tolerance on the phase
response to prevent in-
tersymbol interference. Thus analog voice channels used for high-speed data trans-
mission sometimes required special treatrnent. The special treatunent was originally
provided
at additional monthly rates from the telephone companies in the form of c-
type conditioning for leased lines. C-type conditioning was available in several dif-
ferent grades
that provided
various amounts of control of both phase
and amplitude
distortion. As digital signal processing
technology became commercially viable, the
need for signal conditioning became less necessary because modem equalizers accom-
plished
the same function.
1.2 THE ANALOG NETWORK HIEHARCHY 55
Most medium-rate
(synchronous) modems include fixed equalization circuitry de-
signed to compensate for the
phase distortion in a typical connection. For higher data
rates, automatically adjustable equalization is needed. An automatic equalizer fust un-
dergoes a training sequence in which the characteristics of the transmission channel
are determined by measuring the response to known test signals. Then, equalizing cir-
cuitry in the receiver of each modem is adjusted
(and
continually readjusted during
hansmission) to provide compensation for the amplitude and
phase distottions in the
channel. High-speed transmission
over dial-up lines requires automatic equalization
in the modems since the channel characterjstics
change with each connection.
Another form of conditioning, referred to as a D-type conditioning,
provided lower
levels of noise and harmonic distortion than normally
provided in leased lines, even
with C-type conditioning. D-type conditioning usually did not involve special treat-
ment of any
particular line. Instead, the telephone company tested a number of differ-
ent circuits until one with suitable
quality was found. Sometimes this meant avoiding
a cable that included pairs from a noisy switching office, Unlike voice, which is rea-
sonably tolerant of impulse noise, a data circuit is more susceptible to impulse noise
than to the normal background
(white)
noise. The elimination of older equipment
(e.g.,
step-by-step switches) has eliminated
much of the impulse noi$e in the net-
work.
As more and more of the
public network became digitized, fewer and fewer facili-
ties were analog and, hence, fewer analog impairments were likely to occur in a con-
necrion. By the early 1990s virtually all of the internal
portions of the public telephone
network had been digitized. The only remaining $egments of a connection that were
analog were the subscriber loops and the interfaces in the associated end offices. The
principal source of impairment for voiceband modems then became the analog-to-
digiral
(A/D)
conversion equipment in the line interfaces. Recognition of this situation
completely eliminated the need for any type of conditioning
on leased lines
(which
were often digital and not analog) and allowed for the development of V.90 modems
described in Chapter I l.
An important consideration
for data transmission over long-distance circuits in the
analog network was the effect of echo suppressors. As mentioned in Section 1.2'8' an
echo suppressor blocked the signal in a return
path when the corresponding forward
path of a four-wire circuit was active. Thus, operative echo suppressors effectively
precluded a full-duplex operation. Even in a half-duplex mode of operation the echo
suppressor$ might require 100 msec of deactivation time to reverse the direction of
propagation. For these rea$ons the common carriers
provided a means of disabling the
echo suppressors by using an in-channel control signal from the data terminals.
Echo
suppressors were disabled by tran$mitting a
pure tone between 2010 and 2240Hzfor
400 msec. The echo suppressors remained disabled as long as there was continuous
energy in the channel. Thus, the modems could switch to signaling frequencies and
begin full-duplex data transmission after the suppres$ors were first disabled.
If the en-
ergy in the channel was removed for 100 msec, the echo suppressor$ were reactivated.
Hence, rapid line turnaround was required for half-duplex modems utilizing the entire
bandwidth for each direction of transmission. Contemporary modems have echo can*
56 BACKGRoUNDANDTEHMINoLoGY
cellers for full-duplex transmission
and therefore automatically disable echo suppres-
sors
(which
are probably
nonexistent).
1.3 THE INTRODUCTION
OF DIGITS
Voice digitization and transmission first became feasible in the late 1950s when solid-
state electronics became available. In 1962 Bell System personnel
established the first
commercial use of digital transmission when they began operating a T1 carrier system
for use as a trunk group
in a chicago area exchange
t331.
After the Tl system a family
of r-carrier systems (Tl,
Tlc, TlD, Tz, T3, T4) were developed, all of which in-
volved time division multiplexing of digitized voice signals.
The world's first commercially
designed digital microwave radio system was es-
tablished in Japan by Nippon Elecrric Company (NEC)
in 196g
t341.
In the earty
1970s digital microwave systems began to appear in the United States for specialized
data transmission
services. The first digital microwave link in the U.S. public
tele-
phone network was supplied by NEC of Japan for a New
york
Telephone link be-
tween Brooklyn and North staten Island in lgTz
1341.
DigitI microwave systems
were subsequently developed and installed by several U.S. manufacturers for use in
intermediate-length
toll and exchange area circuits.
Bell System's f,rrst commercial use of digital fiber optic transmission
occurred in
september of 1980 on a short-haul route between Atlanta and smyrna, Georgia
[35].
Three years later the first long-haul
system between New
york
and washington, D.c.,
was put into service.
In addition to ffansmission sy$tems, digital technology has proven
to be equally
useful for implementing
swirching functions. The first country to use digital switching
in the public
telephone network was France in 1970
t361.
The firsr application of digi-
tal switching in the public
network of the United States occurred in early lg76 when
Bell system began operaring its No. 4ESS
t37l
in a class 3 toll office in chicago. Two
months later Continental Telephone company began operation in Ridgecrest, Califor-
nia, of a digital toll switch
[38].
The first digital end office switch in the United States
becameoperationalinJuly
of rgii inthesmalrtownof
RichmondHill,Georgia[39].
1.3.1 Vol ceDi gi ti zatl on
The basic voice-coding
algorithm used in T-carrier systems and most other digital
voice equipment in telephone networks around the world is shown in Figure 1.3g. The
first step in the digitization process
is to periodically
sample the waveform. As dis-
cussed at length in Chapter 3, all of the information needed to reconstruct the original
waveform is contained in the samples if the samples occur at an
g-kHz
rate. The sec-
ond step in the digitization process
involves quantization;
identifying which ampli-
tude interval of a group
of adjacent intervals a sample value falls into. In essence the
quantization process
replaces each continuously variable amplitude sample with a dis-
I.3 THE INTRoDUCTION OF DIGITS 57
t l
Outntiz8tion
- l
- t
- t
- l
- 5
- 6
- l
ttn ul|l
Coding
Figure 1.38 Voice digitization
Process'
crete value located at the middle of the appropriate
quantization interval. Since the
quantized samples have discrete levels, they represent a multipleJevel digital signal.
For transmission
purposes the discrete amplitude samples are converted to a binary
codeword.
(For
illustrative
purposes only, Figure 1.38 shows 4-bit codewords.) The
binary codes ale then transmitted as binary
pulses. At the receiving end of a digital
transmission line the binary data stream is recovered, and the discrete sample values
are reconstructed. Then a low-pass filter is used to
"intelpolate"
between Sample val-
ues and re-create the original waveform. If no transmission errors have occurred, the
output waveform is identical to the input waveform except for
quantization distortion:
the difference between a sample value and its discrete representation. By having a
large number of quantization intervals
(and
hence enough bits in a codeword to encode
them), the quantization intervals can be small enough to effectively eliminate percep-
tible
quantization effects.
It is worth noting that the bandwidth requirements of the digital signal increase as
a result of the binary encoding
process. If ttre discrete, multiple-amplitude samples are
transmitted directly, the bandwidth reguirements are theoretically identical to the
58 BACKGRoUNDANDTEHMINoLoGY
bandwidth of the original signal. when each discrete sample is represented by a num-
ber of individual binary pulses,
the signal bandwidth increases accordingly. The two-
level pulses,
however, are much less vulnerable to transmission impairments than are
the multiple-amplitude pulses (or
the underlying analog signal).
1.8.2 Tlme Division Multlplexing
Basically, time division multiplexing (TDM)
involves nothing more than sharing a
transmission medium by establishing a sequence of time slots during which individual
sources can transmit signals. Thus the entire bandwidth of the facility is periodically
available to each source for a restricted time interval. In contrast, FDM systems assign
a restricted bandwidth to each source for all time. Normally, all time slot$ of a TDM
sy$tem are of equal length. Also, each subchannel is usually assigned a time slot with
a common repetition period
called a frame interval. This form of TDM
(as
shown in
Figure 1.39) is sometimes referued to as synchronous time division multiplexing to
specifically imply that each subchannel is assigned a certain amount of transmission
capacity determined by the time slot duration and the repetition rate. In contra$t, an-
other form of TDM (refened
to as sfatlstical, or asynchronousr time division multi-
plexing)
is described in chapter 10. with this second form of multiplexing,
subchannel rates are allowed to vary according to the individual needs of the sources.
The backbone digital links of the public
telephone network (T-carrier,
digital micro-
wave, and fiber optics) use a synchronous variety of TDM.
Time division multiplexing is normally associated only with digital transmission
links. Although analog TDM transmission can be implemented by interleaving sam-
ples
from each signal, the individual samples are usually too sensitive to all varieties
of transmission impairments. In contrast, time division switching of analog signals is
more feasible than analog TDM transmission because noise and distortion within the
switching equipment are more controllable. As discussed in chapter 5, aralog TDM
techniques have been used in some PBXs befbre rtigital electronics became so inex-
pensive that the digitization penalty
disappeared.
T-Carrier Systems
The volume of interoffice telephone traffic in the United States has traditionally
grown
more rapidly than local traffic. This rapid growth put severe strain on the older
Figure 1.39 Time division multiplexing.
1,s THE tNTRoDUcTroN oF Dlclrs 59
interoffice transmission facilities that are designed for lower traffic volumes. Tele-
phone companies were often faced with the necessary task of expanding the number
of interoffice circuits. T-carrier systems were initially developed as a cost-effective
means for interoffice transmission: both for initial inslallations and for relief of
crowded interoffice cable
pairs.
Despite the need to convert the voice signals to a digital format at one end of a Tl
line and back to analog at the other, the combined conversion and multiplexing cost
of a digital TDM terminal was lower than the cost of a comparable analog FDM ter-
minal. The first T-carrier systems were designed specif,rcally for exchange area ffunks
at distances between l0 and 50 miles.
A T-carder system consists of terminal equipment at each end of a line and a num-
ber of regenerative repeaters at intermediate points in the line. The function of each
regenerative repeater is to restore the digital bit stream to its original form before trans-
mission impairments obliterate the identity of the digital pulses. The line itself, includ-
ing the regenerative repeaters, is referred to as a span line. The original terminal
equipment was referred to as D-type (digital) channel banks, which came in numerous
versions. The transmission lines were wire pairs using 16- to 26-gauge cable' A block
diagram of a T-carrier system is shown in Figure 1.40.
The first T1 systems used DlA channel banks for interfacing, converting, and mul-
tiplexing ?4 analog circuits. A channel bank at each end of a span line provided inter-
facing for both directions of transmission. Incoming analog signals were time division
multiplexed and digitized for transmission. When received at the other end of the line,
the incoming bit stream was decoded into analog samples, demultiplexed, and filtered
to reconsfuct the original signals. Each individual TDM channel was assigned I bits
per time slot. Thus, there were (24X8)
=
192 bits of information in a frame' One ad-
ditional bit was added to each frame to identify the frame boundaries, thereby
produc-
ing a total of 193 bits in a frame. Since the frame interval is 125
psec, the basic Tl
line rate became L544 Mbps. This line rate has been established as the fundamental
standard for digital transmission in North America and Japan. The standard is referred
to as a DSI signal
(for
digital signal 1).
A similar standard of 2.048 Mbps has been established by mJ-T for most of the
rest of the world. This standard evolved from a Tl-like $y$tem that provides 32 chan-
nels at the same rate as the North American channels. Only 30 of the channels in the
El standard, however, are used for voice. The other two Are used for frame synchro-
nization and signaling. Signaling and control information for Tl systems are inserted
lnto into each voice channel
(or transmitted separately by CCIS facilities). Digital signal-
ing and conffol techniques for both systems are discussed in Chapter 7'
Figure 1.40 Tl-carrier system,
60 BACKGHOUND AND TERMINOLOGY
The greatly
increased attenuation of a wire
pair
at the frequencies of a DSI signal
(772kHz
cbnter frequency) mandates the use of amplification at intermediate points
of a Tl span line. In contrast to an analog signal, however, a digital signal can not only
be amplified but al$o be detected and regenerated. That is, as long as a
pulse
can be
detected, it can be restored to its original form and relayed to the next line segment.
For this reason Tl repeaters are referred to as regenerative repeaters. The basic func-
tions of these repeaters are:
l. Equalization
2. Clockrecovery
3. Pulse detection
4. Transmission
Equalization is required because the wire pairs introduce certain amounts of both phase
and amplitude distoltion that cause intersymbol intederence if uncompensated. Clock
recovery is required for two basic pur?o$es;
first, to establish a timing signal to sample
the incoming pulses;
second, to transmit outgoing pulses at the same rate as at the input
to the line.
Regenerative repeaters are normally spaced every 6000 ft in a T1 span line. This
distance was chosen as a matter of convenience for converting existing voice fre-
quency
cables to T-carrier lines. Interoffice voice frequency cables typically used
loading coils that were spaced at 6fi)0-ft intervals. Since these coils were located at
convenient access points (manholes)
and had to be removed for high-frequency trans-
mission, it was only nafural that the 6000-ft interval be chosen. One general exception
is that the first regenerative repeater is typically spaced 3000 ft from a central office.
The shorter spacing of this line segment was needed to maintain a relatively strong
signal in the presence
of impulse noise generated
by older switching machines.
The operating experience of Tl systems was so favorable that they were continu-
ally upgraded and expanded. one of the initial improvements produced
Tlc systems
that provide higher transmission rates over Z?-galge cable. A Tlc line operates at
3.152 Mbps for 48 voice channels, twice as many as a Tl system.
Another level of digital transmission became available in 1972 when the T2 system
was introduced. This system was designed for toll network connections. In contrast.
T I systems were originally designed only for exchange area transmi ssion. The TZ sys-
tem provided
for 96 voice channels at distances up to 500 miles. The line rate was
6.312 Mbps, which is referred to as a DSZ standard. The transmission media was spe-
cial low-capacitance 2?-gauge cable. By using separate cables for each direction of
transmission and the specially developed cables, T? systems courd use repeater spac-
ings up to 14,800 ft in low-noise environments. The emergence of optical fiber sys-
tems made copper-based T2 transmission systems obsolete.
TDM Hierarchy
In a manner analogous to the FDM hierarchy, AT&T established a digital TDM hier-
archy ttrat has become the standard for North America. Stafiing with a DSI signal as
I.3 THE INTBODUCTION OF DIGITS 61
TAELE 1.10 Dlgltal TDM Signale ol North America and Japan
Digital Signal Number of
Number Voice Circuits Multiplexer Designation
Bit Rate
(Mbps) Transmission Media
DS1
DSl C
DS2
DS3
D54
48
96
D channel bank
(24
analog inputs)
M1C
(2
DSI inputs)
M12
(4
DS1 inputs)
M13
(28
DS1 inputs)
M34
(6
DSg inputs)
1.544
3. 152
6, 312
44.736
274.176
T1
paired
cable
T1C
paired
cable
T2
paired
cable
Radio, Fiber
T4M coax, WT4
waveguide, radio
672
a fundamental building block, all other levels are implemented as a combination of
some number of lower level signals. The designation of the higher level digital mul-
tiplexers reflects the respective input and output levels. For example, an M12 multi-
plexer
combines four DSl signals to form a single DSZ signal.. Table I.10 lists the
variou$ multiplex levels, their bit rates, and the transmission media used for each. No-
tice that the bit rate of a high-level multiplex signal is slightly higher than the com-
bined rates of the lower level inputs. The excess bits are included for certain control
and synchronization functions discussed in Chapter 7. A similar digital hierarchy has
also been established by ITU-T as an international standard. As shown in Table l. I I'
this hierarchy is similar to the North American standard but involves different num-
bers of voice circuits at all levels.
Di
g
ita I Pal nGal n Syefems
Following the successful introduction of Tl systems for interoffice trunks, most major
manufacturers of telephone equipment developed digital TDM system$ for local dis-
tribution. These systems are most applicable to long rural loops where the cost of the
electronics is offset by the savings in wire
pairs. No matter what the distance is, un-
expected growth can be most economically accommodated by adding electronics, in-
stead of wire, to produce a
pair-gain system. The
possibility of trailer
parks, apartment
houses, or Internet service
providers springing up almost overnight causes nightmares
in the minds of cable
plant forecasters. Pair-gain system$
provide a networking alter-
native to dispel those nightmares.
Digital pair-gain
$ystems are also useful as alternatives to switching offices in
small communities. Small communities are often servicedby small automatic switch-
ing systems normally unattended and remotely controlled from a larger switching of-
fice nearby. These small community switches are refened to as community dial offices
*Because
T2 transmission systems have become obsolete, the Ml2 function exists only in a functional sense
within Ml3 multiplexers, which multiplex 28 DSI signals into I DS3 signal.
62 BACKGHoUNDANDTERMtNoLocy
TABLE 1.11 ITU Digital Hierarchy
Level Number
Number of Voice
Circuits
Multiplexer
Designation Bit Rate (Mbps)
El
E2
E3
E4
E5
30
120
480
1 920
7680
M1 ?
M?3
M34
M45
2.048
8,448
34,368
139.264
565.1 48
(CDOs).
A CDO typically provides
only limited service features to the customers and
often requires considerable maintenance. Because digital pair-gain
systems lower
transmission co$t$ for moderate-sized groups
of subscribers, they are a viable alterna-
tive to a CDO: Stations in the small community are serviced from the central office
by way of pair-gain
systems. A fundamental consideration in choosing between pair-
gain
systems and remote switching involves the haffic volumes and calling patterns
within the small community. The basic techniques of analyzing traffic paftems
and de-
termining trunk group
sizes are provided
in Chapter 12.
The first two digital pair-gain
sy$tems used in the Bell System were the subscriber
loop multiplex (sLM)
system
[40,
4l] and, its successor, the subscriber loop carrier
(slc-40)
system
140,421.
Although these sysrems used a form of voice digitization
(delta modulation) different from that used in T-carrier system$ (pulse
code modula-
tion), they both used standard rl repeaters for digital transmission at 1.544 Mbps.
Both systems also converted the digitized voice signals back into individual analog in-
tetfaces at the end office swirch to achieve system tran$parency. Figures L4l and 1.42
show block diagrams of these sysrem$. Notice that the sLM system provided
both
concentration and multiplexing (80
subscribers for 24 channels) while the SLC-40
was strictly a multiplexer (40 subscribers assigned in a one-to-one manner to 40 chan-
nels).
The sLM and sLC-,{O systems used delta modulation voice coding because it was
simpler than pulse
code modulation as used in Tl systems and was therefore less
costly to implement on a
per-channel
basis-a desirable feature for modular system
implementations. The original Tl systems, on the other hand, minimized electronics
costs by using common encoders and decoders, which precluded implementation
of
Crofibdr
witch
Figure l.4l Subscriber loop multiplexer.
1.3 THE INTHODUCTIoN oF DIGITS 63
'.*Jill"i,**'
Figure 1.42 Subscriber loop carrier
(SLC-40),
less than 24 channels
(an unnecessary feature in an interoffice application). By the late
1970s low-cost, integrated circuit implementations of standard pulse code modulation
became available that led the way to the first
(1979)
installation of the SLC-96, a sub-
scriber carrier rty$tem using voice coding that was compatible with T1 systems and the
emerging digital end office switching machines
[43].
The SLC-96 system
(which is functionally equivalent to four Tl lines) can inter-
face directly with a digital end office and not be demultiplexed into 24 distinct analog
interfaces. Thus this capability, which is referred to as integrated digital loop carrier
(IDLC), greafly reduces the prove-in
distance where the digital carrier becomes less
expensive than separate subscriber pairs. Subsequent enhancements of the SLC-96
$y$tem include use of fiber transmission, support for local switching functions in the
remote module, and configurability of analog interfaces from the central office
[44,
45.1. Many newer copper- and fiber-based digital loop carrier systems have been de-
veloped. These systems are discussed more thoroughly in Chapter 11 in the context of
digital subscriber access.
1.3.3 Data under Voi ce
After the technology of T-carrier systems had been established, AT&T began offering
leased digital transmission services for data communications. This service, known as
Dataphone Digital Service
(DDS),
uses Tl transmission links with special terminals
(channel
banks) that provide direct access to the digital line. An initial drawback of
DDS arose because T-carrier systems were originally used only for exchange area and
shorr toll network trunks. Without some form of long-distance digital transmission,
the digital circuits in separate exchange areas could not be interconnected. AT&T's
original response to long-distance digital transmission was the development of a spe-
cial radio terminal called the 1A radio digital terminal
(IA-RDT)
[46].
This terminal
encoded one DS I signal
(1.544
Mbps) into less than 500 kHz of bandwidth. As shown
in Figure 1.43, a signal of this bandwidth was inserted below the lowest'frequency of
a master group multiplex
(Table
1.5). Since this frequency band is normally unused
in TD or TH analog radio systems, the DSI signal could be added to existing analog
routes without displacing any voice channels. The use offrequencies below those used
for voice signals leads to the designation
"data
under voice" (DUV).
It is important to point out that DUV represented a special development specifically
intended for data transmission and not for voice services. In fact, DUV was used only
to provide long-distance digital transmission facilities for DDS. The emergence of
64 BAcKeRouNDANDrERMrNolocy
3084
Frequency
(kHzl
Figure 1.43 Data under voice.
long-distance fiber sy$tem'r obviously eliminated the need for DUV equipment (and
even the analog radios themselves).
1.3.4 Digital Mlcrowave Radio
In contrast to DUV systems, which piggyback
1.544 Mbps onto an analog radio for
data services, a common-carrier digital microwave system uses digital modulation ex-
clusively to transmit and receive higher level digital multiplex signals for voice traffic.
Digital radios use the same frequency bands allocated for analog radios as listed in Ta-
ble 1.7. Thus a
primary
design requirement of a digital radio was that it must confine
its radiated power
to prevent excessive interference into adjacent, possibly
analog,
chamels. Moreover, the FCC stipulated that the digital radios had to be capable of
pro-
viding roughly the same number of voice circuits as existing analog radios. Table I . 12
lists the minimum number of voice circuits and the resulting bit rate that must be pro-
vided by digital radios in each of the common-canier microwave bands
[47].
Despite the design constraints imposed by compatibility requirements with analog
radios, digital radios proved to be more economical than analog radios in several ap-
plications.
Because of lower terminal cosrs (multiplexing
and demultiplexing), digiral
radio $ystems were generally less expensive than analog radio systems for distances
TABLE 1.12 Minimum Volce Clrcuit Requirements ot Digitat Radlos ln the
United States
E
E
8 . 9
* E
Ee
I
6
cr
Frequency Band
(MHz)
Equivalent
Minimum Number Number of DS1 Resultant Bit Channel
of Circuits Signals Rate (Mbps)a
Bandwidth (MHz)
2110- 2130 96
2160-2180 96
370H200 1152
5925-6425 1152
10, 700- 11, 700 1152
4 6, 144 3. 5
4 6.144 3,5
48 73.7 20
48 73.7
30
48 73.7 40
alhe
actual bit rate is usually slightly greater
owing to xtraframing and synchronization bits. Furthermore, mosl
radio systemB provide
significantly mor6 voice circuits than the minimum.
r.g rHE rNTRoDUcrroN oF Drclrs 65
up to about 300 miles and on longerroutes requiring channel access
(drop-and-insen)
at intermediate
points in the route
[a8].
The major impetus for digital radio in the
United States was the emergence of digital toll switches like
the No.4ESS. The inter-
connection of digital radio signals to digital switches avoided costly lower level mul-
tiplex equipment.
1.3.5 Fi ber Opti c Transmi esi on
Of all the new technology inhoduced into the telephone network, fiber has certainly
had the mo$t
profound effect. Prior to its emergence, transmission engineers would
have considered the combination of exffemely wide bandwidth, extremely low attenu-
ation, and immunity from interference as something akin to perpetual motion. Low at-
tenuation allows long repeater spacings, which equates to low maintenance costs-the
reason for wholesale replacement of long-haul analog radio systems. On December 5,
1986, AT&T completed the last section of a transcontinental fiber system
[49].
By the
end of that decade virtually all of the high-density routes were converted to fiber. Ra-
dio systems are and will continue to be used for carrying low-density traffic where
right-of-way costs are a dominant consideration. However, because these routes are
relatively $hort spurs off of digital fiber arteries, intetface costs imply digital imple-
mentations, thus completing the economic foundation for the all-digital network.
As discussed in Chapter 8, an optical fiber is not an inherently digital transmission
system. However, the interface electronics
(drivers and receivers) and the application
of connecting to digital switches have stimulated their use of digital kansmission.*
Furthermore, considering the amount of bandwidth available there is little incentive
to use analog transmission to conserve bandwidth, as was the original situation with
microwave radios.
1. 3. 6 Di gi t al Swl t chi ng
The original research into digital switching at Bell Laboratories was repofted by Earle
Vaughan in 1959
[501.
Laboratory models were developed to demonstrate the concept
of integrating digital time division multiplex transmission systems with time division
switching systems. Unfortunately the necessary solid-state elechonics had not ma-
tured sufficiently at that time so commercial development of digital switching wa$ not
pursued,
and development of the No. I ESS continued along the lines of space division
electromechanical technology. Almost l0 years later, however, Bell Labs began de-
velopment of a digital toll switch, the No. 4 ESS.
When placed in service in January 1976, the No. 4 ESS provided several new ca-
pabilities for the toll network. First, it was the first toll switch to be designed for
stored-program control at the outset.t Second, its capacity was three times that of the
prevailing electromechanical switch at the time: the No. 44
cross bar. The larger ca-
*Analog
optical fiber systems have been widely used to carry television signals in feedets segments of
.CATV
systems.
lstored-program
confrol was first implemented in the toll network beginning in 1969 by retrofitting
crossbar switches
[5
1].
66 BACKGROUND AND TERMINOLOGY
pacity
of the No. 4 ESS meant that many metropolitan areas could consolidate toll haf-
f,rc into one switch instead of several. Third, the digital time division design of the No.
4 ESS allowed direct connection to digital T-carrier lines. This last feature illu$trate$
the original attraction of digital switching for toll and tandem offices of the network.
By 1976, when the first No. 4 ESS was installed, it was clear that digital transmission
was dominating the exchange area and shorter inteftoll trunks. Thus significant econo-
mies and improvements in transmission quality
resulted by eliminating channel banks
at the interface between digital kunks and a switching system.
The early development of digital end office switches in the United Stares was un-
dertaken by independent equipment manufacturers with the flust system being placed
in service in 1977. These systems were primarily
designed for the smaller switching
offices of the independent telephone companies. Digital switches were
particularly
at-
tractive to rural telephone companies because they could provide
significant copper
savings when directly connected to digital pair-gain
transmission systems. The flust
large digital end office switching sy$tem to be introduced into the North American
network was the DMs-100 provided
by Northern Telecom. Table Ll3 lists digital
switching machines developed for the North American public
telephone network.
The functional essence of a digital time division switching matrix is illustrated in
Figure 1.44. As indicated, all inputs are time division multiplex links. These links may
represent digital pair-gain
system$, T-carrier interoffice trunks, or the outputs of colo-
cated channel banks used to interface analog lines to the digital switch. In any event
the switching mahix itself is designed ro service only TDM input links.
Basically, the switching matrix is required to transfer information arriving in a
specified time slot (channel)
on an incoming TDM link to a specified time slot on an
TABLE 1.13 Dlgltal Central Oflice Swltchlng Systems of North America
Manufacturer Designation
Date of
lntroduction Application Line Size
AT&T 4 ESS
AT&T 5 ESS
C|T-Alcatel E 10-five
GTE 3 EAX
GTE 5 EAX
LM Ericsson AXE10
NEC NEAX.61
ITT System 1210
NorthernTel ecom DMS-10
NorthernTel ecom DMS-100
NorthernTelecom DMS-200
Siemens EWSD
Stromberg Carlson DCO
Vidar lTS4
Vidar lTS4/5
1 S76
1S83
1 982
1 S7B
1982
1978
1979
1 978
1977
1 97S
1978
1 981
1977
1977
1 978
107,000
100,000
100,000
60,b00
145,000
200,000
80,000
26,000
7,000
100,000
60,000
200,000
32,000
7,000
12,768
Toll
Local
Local
Toll/tandem
Local
Local/toll
Local/toll
Local/toll
Local
Local
Toll
Local
Local
Toll/tandem
Local/toll
I FRAME
1,3 THE INTRODUCTION OF DIGIT$ 67
Figure 1.44 Digital time division switching operation.
outgoing TDM link. Since an arbitrary connection involves two different
physical
links and two different time slots, the switching
process requires spatial translation
(space
switching) and time translation
(time switching). Thus the basic operation is
sometimes referred to as two-dimensional switching. Space switching is achieved
with conventional digital logic selector circuits, and time switching is achieved by
temporarily storing information in a digital memory or register circuit.
For convenience, the inputs to the switching matrix of Figure 7.44 arc all shown
on the left, and the outputs are on the right. Since the links are inherently four-wire,
each input time slot is paired with the same time slot on a corresponding output link.
Hence, a full-duplex circuit also requires a
"connection"
in the reverse direction that
is achieved by transferring information from time slot 17 of link N to time slot 3 of
link I in the example shown. Operational and implementation details of a large range
of digital time division switches are
provided in Chapter 5.
1.3.7 Digltal Network Evolutlon
The evolution of the analog telephone network into one that is all digital except for
the access lines is summarized in Figure 1.45. The process began in the 1960s
(a)
with
T1 systems being installed on relatively short haul interoffice trunks within the ex-
change areas. Next, in the early 1970s
(b),
digital transmission was introduced into the
toll network with T2 systems for relatively short routes between toll offices. It was in
the late 1970s
(c)
that digitization really began to take over. Tl coverage expanded
greatly,
digital loop carier
(DLC)
systems came into use,* and digital switches be*
came available at all levels of the network: PBXs (DPBXs), end offices
(DEOs), tan-
dem offices, and toll offices
(DTOs).
Moreover, microwave digital radios
(MDRs)
proved
to be advantageous to use in both the exchange areas and the shorter toll net-
work routes due to low interface costs to digital switches. Thus, the late 1970s
pro-
duced a number of integrated islands where digital switches within a region were
interconnected by digital transmission links but there was little digital connectivity be-
tween the islands.
(Data under voice was installed as an overbuild to analog routes for
*Digital
subscribet caffier systems were actually introduccd in the early 1970s, but these systems utilized
a voice digitization technology (delta
modulation) that was incompatible with the rest ofthe network and
therefore did not figure into the integrated network. These early digital loop carrier systems atE now
obsolete.
68 BACKGHOUND AND TEHMINOLOGY
Exchangn arta
t - O r z - - - - - - - - . -
't}..r-*{",7'g
o
1\n/'1" ,i
Q
o
\
T _ 1 i
t - /
\ - - - -
I ,\r
o) ( c
- - "- - - - *l - /
\ - -
Toll mtwuk
Fibrt
/
-\
Excharqe
'
arot
I
.Tr\
(dl
I
trz
j
I
( 1
I
I
I
l ,
t t * - -
(Dl
DTO
MDR
oTo
DPEX
o
\ - J
. - * * : / \ .
,) .'i
(d)
I
Fibrr
7
DLC /
'r)
,'I
\ * - - - - -
: : - * - ] . /
' \ - - - - -
(d)
I r '
I
PBI ISDil
t6t
Figure 1.45 Digital network evolution.
\ord
oi i x
\--'
REFERENoES 6S
limited-capacity data services.) A fully integrated and interconnected digital network
became a reality in the early 1980s (d)
when fiber optic transmission emerged as the
technology of choice for high-density long-haul routes. Digital connectivity to busi-
ness customer
premises equipment also occurred in this time frame as Tl became the
preferred voice trunk interface for large PBXs.
End-to-end digital connectivity for voice or data services became a reality in the
late 1980s
(e)
wittr the introduction of ISDN basic rate
(ISDN
BR, 28 + D) and ISDN
primary rate
(ISDN
PR, 238 + D) digital connection$ to the customer. In addition, fi-
ber technology became much more ubiquitous as DS3 rate systems eliminated T2 sys-
tems in the toll network and fiber-based systems became the
preferred technology for
loop carrier and feeder systems-even at relatively short distances.
REFERENCES
I S. P, Thompson, Phillip Reis: Inventor of the Telephone, Sylvanus P. Thompson,
London. New York. 1883.
R. Smith and W. Sonnerryille,
"Record'81
Telco Construction Set at $23.5
Billion, Up
5.6Vo,* Telephone Engineer and Management, Jan. 15, 1981, p. 55.
Technical Staff, Bell Laboratories, Engineering, and Operations in the BelI System,Znd
ed., Bell Telephone Laboratories, Indianapolis, Indiana, 1983.
4 D. M. Mardon,
"Plan
for Access*Tandem and Trunking," Telephone Engineer &
Management, Nov. 15, 1983, pp. l4O-142.
5 W, D. Reeve, Subscriber Inop Signaling and Transmission Handbook, IEEE Press,
New York. 1995,
6 S, J. Campanella,
"Digital
Speech Interpolation Systems," in Advanced DiSital
Communication,r; Systerzs and Signal Processing Tethniques, K. Fehrer, Ed.,
Prentice-Hall, Englewood Cliffs, NJ, 1985.
Technical Staff, Bell Laboratori es, Engineering and Operations in the Bell System,Bell
Telephone Laboratories, Indianapolis, Indiana, 1977.
W. L. Pritchard,
"The
History and Future of Commercial Satellite Communications,"
IEEE Communitations M agaarne, May 1984, pp. 22-37 .
Digital Video Broadcasting
(DVB),
Framing structure, channel coding and modulation
for digital tenestrial television, EN 300 744,
(ETSI TM 1545), 1999,
I. Boyd, C. B. Southcott, D. P. Crowe, and P. J. Bolingbroke,
"Spech
Codec for the
Skyphone Aeronautical Telephone Service," BritishTelecommunications
Engineering,
Jul y 1986, pp. 83-91.
A. J. Aikens and D. A. Lewinski,
"Evaluation
of Message Circuit Noise," Bell System
Technical Jountal, July 1970, pp.
879-909.
J. L. Sullivan,
t'A
Laboratory System for Measuring Loudness Loss of Telephone
Connections," Bell System Techniml Joumal, Oct. 1971, pp. 2663-2739.
W. K. Macadam,
"A
Basis for Transmission Performance Objective in a Telephone
Communications System," AIEE Transactions on Comntunications and Electronics,
May 1978, pp.
205-209.
l l
t 2
l 3
70 BACKGROUND AND TERMINOLOGY
14 F. P. Duffy and T. W. Thatcher, Jr,,
"Analog
Transmission
performance
on the
Switched Telecommunications
Network," BelI SystemTechnical Journal, Apr. 1971,
pp. l 3l t - 1347.
l5 W. R. Bennett and J. R. Davey, Data Transmission, McGraw-Hill, New
york,
1965,
16 Members of Technical staff, Bell Telephone Laboratories, Transmission systems
for
Communitatiorrs, 4th ed,, Westem Electric Company, Winston-Salem, North Carolina,
1971.
1l H. R. Huntley,
"Transmission
Design of Intertoll Telephone Trunks,', BeII System
Tethnical Joumal, Sept. 1953, pp. l0l9-1036,
l8 K. Murano, S. Unagami, and F. Amano,
"Echo
Cancellation and Applications,,' IEEE
Communictttions Magazine, Jan. 1 990, pp. 49-55.
18a E, A, Lee and D. G. Messerschmin, Digital communication, Kluwer Academic
Publisher, Norwell, MA, 1993.
19
"A
Technical Report on Echo Cancelling," Report No, ?7, ANSI T14,l.6 Working
Group on Specialized Signal Processing, Nov. 1993.
20 H. G. Suyderhoud, M. Onufry, and S. J. Campanella,
"Echo
Connol in Telephone
Communications," National Telecomtnunications
Conference Record, 1976, pp.
8. 1- l - 8. I
- 5.
21 F. T. Andrews, Jr. and R. w. Hatch,
"Nationa]
Telephone Network rransmission
Planning in the American Telephone and Telegraph company," IEEE Transactions on
Communitations Technology, June 1971, pp. 302-314.
22 cornpatibility Information
for
Telephone Exchange service, Bellcore Technical
Advisory TA-NPL-000:9/2, Bell Communications Research. Issue l. Feb. 1989.
23 Notes on THE NETWORK, American Telephone and Telegraph Company, Network
Planning Division, AT&T, New York, 1980, pp.
44-58.
M
"special
Issue: stored hogram conftol Network," Bell systemTechnical Joumal, sept.
1982.
25 P. J. Kuhn, C, D. Pack, and R. A. Skoog,
"Common
Channel Signaling Networks:
past,
Present, Ftfitre," IEEEE Joumal on seleded Areas in communications, Apr.1994, pp.
383-393,
26 J. S. Doyle and C. S. McMahon,
'-The
Intelligent Network Concept," IEEE
Transactions on Communicafdons, Dec. I988, pp. 1296-130L
27 J. M. Alcinpelu,
"The
overload Performance of Engineered Networks with
Nonhierarchical
and Hierarchical Routing," Bell systemTechnical rountal, sept. 1984,
pp. l 26l - 1281.
28 G. R. Ash and E. oberer,
"Dynamic
Routing in the AT&T Network-tmproved
service
Quality
at Lower Cost,* IEEE Globecom Proceedings,lgSg, pp.
g.l.l-9.1.6,
29 C. Mason,
"Software
Problem Cripples AT&T Long-Distance Network,', Telephony,
Ian. 22, 1990, pp. l0-l l.
30
"special
Issue on Advanced Mobile Phone service
(AMPS),"
Beil system Technical
Joumal,Jan. 1979.
3l D. c. cox,
'?ortable
Digital Radio communications-An Approach to Tetherless
Access," IEEE Communications Magaaine, July 1989, pp. 30-40.
32
"Mobile
station Land station compatibility specificarion," ANsI/EIA./TIA standard
553, Electronic Industries Association, I 989,
PHOBLEMS 71
33 K. E. Fultz and D. B, Penick,
"The
Tl Carrier System," BelI Systern Tethnical Joumal,
Sept. 1965, pp. 1405-1451.
34
..Supply
Record on NEC PCM Microwave Equipment,"
Nippon Electric Company,
Tokyo, May 1974.
35 M. Schwartz,
"Optical
Fiber Transmission-From Conception
to Prominence in 20
Years," IEEE Communications Magazine,May
1984' pp. 38-48'
36 J. B. Jacob,
"The
New Approaches and solutions offered by the Digital ESS.E l0 to
Deal with Switching, Operation, and Maintenance' Functions,"
International
Communic ations Conferenc e P roc e edin g s, I 979' pp. 46.6- I
-46.6-5.
3'l
"special
Issue: No, 4 ESS," Bell System Technical Journal, JulylAng' l98l '
38 J. C. McDonald and J. R. Baichtal,
"A
New Integrated Digital switching system,"
N ational Tele communit at i ons C onfe renc e Re c o rd, 1
976'
pp. 3.2- |
-3'2-5.
39
"First
Class 5 Digital Central Office Goes into Service in Georgia," Telephony, Aug. I
'
1977,pp.M-25.
40 F. T. Andrews, Jr,,
"Customer
Lines Go Electronic," Bell Labs Record, Feb. 1972,pp.
59*64.
4l M, T. Manfred, G. A. Nelson, and c. H. sharpless,
"Digital
Loop carrier systems," Bell
System Technical Joumal, Apr. 1978, pp. I 129-1 142.
42 S. J. Brolin, G. E. Harrington, and G. R. t eopold,
"The
subscriber Loop carier system
Gets Even Better," BeIl labs Record,May 1977
,
pp. ll9-122.
43 S. Brolin, Y. s. cho, w. P. Michaud, and D. H. williamson,
"Inside
the New Digital
Subscriber Loop System," Belt l-abs Record, Apr' 1980' pp. I 10- I I 6.
44 G. E. Harrington and R. w, wyndrum,
"A 'smart'
slc carrier Meets the changing
Nature of services in the Evolving Digital Loop," Bell l-abs Record, sept. 1985, pp.
l 3- 17.
45
"Digital
Interface Between the SLC-96 Digital Loop Carrier System and a Local Digital
Switch," Bellcore Teohnical Requirement$,
TR-TSY-000008, Bellcore, Aug' 1987'
46 R. R. Grady and J. w. tuupT,
"lA
Radio Digital Terminals Put
'Data
Under Voice,"'
Bell Labs Record,May 1974,pp. 16l-166.
4'7 FCC Memorandum Opinion and Order, released January 29' 1975, modifying FCC
Report and OrderofDocket
l93l l, released September 27,1974.
48 W. S. Chaskin,
"Digital
Transmission Economics," Digital Microwave Transmission
Engine e rin g Sympo s ium, Rockwell International, Sept. I 978.
49 R. C. Hansen,
"AT&T's
Digital Network Evolution," IEEE International Conference
on Communications, 1987, pp. 198'l'l
-1.
1.5.
50 H. E. Vaughan,
"Research
Model for Time-Separation
Integrated Communication,"
BelI SystemTechnical Joumal, July 1959' pp. 909-932'
5l B. T. Fought and C. J. Funk,
'Electronic
Translator System for Toll Switching,"
IEEE
Transactions on Communications,
June 1970'
p. 168.
PROBLEMS
l.l Express 33 dBmC of noise in terms of
picowatts and milliwatts of absolute
power.
72
1.2
1.3
t.4
1.5
r.6
BACKGROUND AND TEHMINOLOGY
Twenty-seven picowatts
of noise with a flat spectrum between 0 and 3 kHz is
how many dBmC?
A value of 30 dBrnc0 is how many picowarts
of absolute noise power
at a
-3-dB
TLP?
An idle-channel
noise power
measurement of 2l dBrnc occurs at a
-7-dB
TLp.
Express the noise power
of this measurement in dBrnc0 and determine what
power
measurement this noise would produce
at another point
in the circuit that
is designated as a
-2-dB
TLP.
A transmission
link has 14 dBrn of absolute noise power at a
-13-dB
TLp input
test point
and 27 dBm of absolute noi$e power at a
-3-dB
TLp output te$t point.
How much absolute noise is added on the transmission link?
A transmission link with a
-Z-dB
TLp at the rransmit end and a
-4-dB-TLp
at
the receive end is part of a voice circuit that produces
the following idle-channel
noise measurements:
18 dBrnc0 at the transmit end and 20 dBrnc0 at the re-
ceive end. what is the conrribution (in pwp0)
of the transmission link toward
the total absolute power
at the receive end?
why is echo control unnece$sary on short-delay voice connections? what about
singing?
1.7
WHY DIGITAL?
The first chapter
provides an overview of an analog telephone network and a brief in-
troduction to digital ffansmission and switching technology
introduced to replace
older analog equipment. This chapter discusses the basic technical advantages of digi'
tal implementations that stimulated the replacement of the analog systems.
2.1 ADVANTAGES OF DIGITAL VOICE NETWORKS
A list of technical features of digital communications
networks is provided in Table
2. L These features are listed in the order that the author considers to be their relative
importance for
general telephony. In
particular applications,
however, certain
consid-
erations may be more or less significant. For instance, the last item, ease of encryption,
is a dominant feature favoring digital networks for the military.
Most of the features of digital voice networks listed in Table 2.1 and discussed in
the following
paragraphs pertain to advantages of digital hansmission or switching
relative to analog counteryarts. In some instances, however, the features
pertain only
to all-digital networks. Encryption,
for example, is practical and generally useful only
if the secure form of the message is established
at the source and translated back into
the clear form only at the destination. Thus an end-to-end digital system that operates
with no knowledge of the nature of the traffic
(i.e., provides transparent transmission)
is a requirement for digital encryption applications.
For similar reasons end-to-end
digital transmission is needed for direct transmission of data
(no
modem)-
When a net-
work consists of a mixture of analog and digital equipment, universal use of the net-
work for services such as data transmission dictates conformance
to the least common
denominator of the network: the analog channel.
2.1.1 Ease of Mul ti pl exi ng
As mentioned in Chapter l, digital techniques were first applied to general telephony
in interoffice T-carrier
(time division multiplex) systems. In essence, these systems
traded electionics costs at the ends of a transmission
path for the cost of multiple
pairs
73
74 WHY DIGITAL?
TABLE 2.1 Technical Advantages ot Dlgltar communlcations Networks
1. Ease 0l multiplexing
2. Ease of signaling
3. U$e of modern technology
4. lntegration of kansmission and switching
5. Signal regeneration
6. Performance monitorability
7. Accommodation
of other services
8. Operability at low signal-to-noi$e/intederence
ratios
L Ease of encryption'
of wires between them. (A
hade that is more cost effective every year.)
Although
FDM of analog signals had also been used to reduce cable costs, FDM equipment is
much more expensive than TDM equipment, even when the cost of digitization is in-
cluded. After voice signals have been digitized, TDM equipment costs are
quite
small
by comparison. Since digitization occurs only at the first level of the TDM hierarchy,
high-level digital TDM is even more economical than high-level FDM counrerpans.
It should be pointed
out that TDM of analog signals is also very simple and does
not require digitization of rhe sample values. The drawback of analog TDM lies in the
vulnerability of nanow analog pulses to noise, distortion, crosstalk, and intersymbol
interference. These degradations cannot be removed by regeneration as in a digital
system. Hence, analog TDM is not feasible except for noiseless, distortion-free envi-
ronments.* In essence the ability to regenerate
a signal, even at the expense ofgreater
bandwidth, is almost a requirement for TDM fransmission.
2.1.2 Eaee of Sl gnati ng
control information (e.g.,
on-hook/off-hook,
address digits, coin deposits) is inher-
ently digital and, hence, readily incorporated into a digital Fansmission system. one
means of incorporating
control information into a digital transmission link involves
time division multiplexing
the control as a separate but easily identifiable control
channel' Another approach involves inserting special control codes into the message
channel and having digital logic in the receiving terminals decode that control infor-
mation' In either case, as far as the transmission sy$tem is concerned, control infor-
mation is indistinguishable
from message haffic.
In contrast, analog transmission sy$tems required special attention for control sig-
naling. Many analog transmission
systems presented
unique and sometimes difficult
environments for inserting
control information. An unfortunate result was that many
varieties of control signal formats and procedures
evolved for the analog network. The
control formats depend on the nature of both the transmission system und it, terminal
.Aoulog
TDM has been used in a fbw telephone applications. Farinon's Subscriber Radio System
[l ],
for
example, used'pulse-width-modulated
TDM. As discussed in Chapter 5 some older PBXs also used analog
TDM.
2.1 ADVANTAGES OF DIGITAL VOICE NETWORKS 75
equipment. In some interfaces between network subsystems control information had
to be converted from one format to another. Signaling on analog links therefore rep-
resented a significant administrative
and financial burden to the operating telephone
companies.
The move to common-channel signaling
removed most of the signaling costs a$-
sociated with interoffice trunks but did not change the situation for individual sub-
scriber lines, which must carry signaling on the same facility as the message channel.
The use of digital subscriber
lines (DSLs) reduces the signaling costs relative to analog
subscriber lines, which helps offset the higher cost of a DSL and a digital telephone'
DSLs are a fundamental aspect of ISDN, as described in Chapter I l.
In summary, digital systems allow control information to be inserted into and ex-
ffacted from a message stream independently of the nature of the transmission medium
(e.g., cable, fiber, microwave, satellite). Thus the signaling equipment can
(and
should) be designed separately from the transmission system. It then follows that con-
trol functions and formats can be modified independently of the transmission subsys-
tem. Conversely, digital transmission systems can be upgraded without impacting
control functions at either end ofthe link.
2.1.3 Use of Modern TechnologY
A multiplexer or switching matrix for time division digital signals is implemented
with the same basic circuits used to build digital computers: logic
gates and memory.
The basic crosspoint of a digital switch is nothing more than an AND
gate with one
logic input assigned to the mes$age signal and other inputs used for control
(crosspoint
selection). Thus the dramatic developments of digital integrated circuit technology for
computer logic circuits and memory are applicable directly to digital transmission and
switching fiystems. In fact, many standard circuits developed for use in computers
are
directly usable in a switching matrix or multiplexer. Figure 2.1 shows the basic im-
plementation of a l6-channel, bit-interleaved,
digital time division multiplexer using
common digital logic circuits. As indicated, the multiplexing
function involves noth-
ing more than cyclically sampling the 16 input data streams. Such an operation as-
sumes all of the data streams are synchronized
to each other. As discussed in Chapter
7, the
process of synchronizing the data sffeams requires logic circuitry that is much
more complicated than that shown. Nevertheless, the implementation of TDM is much
less expensive than analog FDM.
Even
greater
advantages of modern technology have been achieved by using large-
scale integrated
(LSD
circuits designed specifically for telecommunications
functions
such as voice encoding/decoding, multiplexing/demultiplexing,
switching matrices,
and special-purpose and
general-purpose digital signal
processing
(DSP). Digital sig-
nal
processing functions are described in Section 2'2.
The relative low cost and high performance of digital circuits allows digital imple-
mentations to be used in some applications that are prohibitively expensive when im-
plemented with comparable analog components. Completely nonblocking switches,
for example, are not
practical with conventional analog implementations,
except in
WHY DIGITAL?
f;TATA*l
76
. r r
B A
F a t t
BA
4 Bi t Gounter
Figure 2.1 Sixteen-to-one TDM multiplexer.
small sizes. In a modern digital switch the cost of the switching matrix itself is rela-
tively insignificant. Thus, for medium-size
applications, the size of the switch matrix
can be increased to provide
nonblocking operations, if desired. The automatic call dis-
tributor developed by collins-Rockwell
[2]
is an early example of a digital switch op-
erating in an analog environment. A digital implementation
was chosen largely
because it could economically provide
a nonblocking operation.
The benefits ofmodern device technology are not confined to digital circuits alone.
Analog integrated circuits have also progressed
significantly, allowing traditional
analog implementations
to improve appreciably. one of the primary
requirements
of
an analog component, however, is that it be linear. It appears, if only because of re-
search and development emphasis, that fast digital components are ea$ier to manufac-
ture than linear analog counterparts. In addition, digital implementations appear to
have an inherent functional advantage over analog implementations. This advantage
is derived from the relative ease with which digital signals can be multiplexed. A ma-
jor
limitation with the full use of LSI components results from limited availability of
external connections to
the device. with time division multiplex techniques, a single
physical pin
can be used for multiple-channel access into the device. Thus the same
technique used to reduce costs in transmission systems can also be used within a local
module to minimize the interconnections and maximize the utilization of very large
scale integration. In the end, a
"switch
on a chip" is possible
only if a great
number of
channels can be multiplexed onto a relatively small number of external connections.
The technological development to have the most significant impact on the tele-
phone
network is certainly fiber optic transmission.
Although fibers themselves do not
favor digital transmission over analog transmission, the interface electronics to a fiber
2.1 ADVANTAGES OF DIGITAL VOICE NETWORKS 77
system function
primarily in an on-off
(nonlinear) mode of operation. Thus digital
transmission dominates fiber applications, although analog optical technology is com-
monly used in analog video distribution.
2.1.4 Integratlon of Transmis$lon and $wltching
Traditionally the analog transmission and switching systems of telephone networks
were designed and administered
by functionally independent organizations.
In the op-
erating telephone companies, these two equipment classes are referred to as outside
plant
and inside
plant, respectively. These equipments necessarily
provide stand-
ardized intdrfaces, but, other than that, transmission equipment was functionally inde-
pendent
of switching equipment.
When TDM of digital voice signals was introduced into the exchange area and
communications
engineers began considering digital switching, it became apparent
that TDM operations
were very similar to time division switching functions. In fact,
as described in later chapters, the first stages of digital switches
generate first-level
TDM signals by nature, even when interfaced to analog transmission links. Thus the
multiplexing operations of a transmission system can be easily integrated into the
switching equipment.
The basic advantage of integrating the two systems is shown in Figure 2.2. The de-
multiplexing equipment
(channel banks) at the swirching offices is unnecessary, and
first-stage switching equipment is eliminated.
If both ends of the digital TDM trunks
are integrated into a digital switch, the channel banks at both ends of the trunk are
eliminated. In a totally integrated network voice signals are digitized at or near the
source and remain digitized until delivered to their destination. Furthermore,
all inter-
office trunks and intemal links of a switching system carry TDM signals exclusively'
Thus first-level multiplexing and demultiplexing are nonexistent except at the
periph-
ery of the nerwork. Although integration of DSI signals into $witching
machines is
Figure 2.2 Integration of transmission and switching:
(a) nonintegrated nansmission and
swirching,
(b) integrated time time division switching and transmission.
78 WHY DIGITAL?
commonplace, integration of higher level signals is complicated by higher level mul-
tiplexing formats (pulse
stuffing) described in chapter 7. A newer multiplexing for-
mat (soNET)
described in chapter 8 has some operational modes thar are more
amenable to direct interconnection
into a switching sy$tem.
Integration of transmission and switching functions not only eliminates much
equipment but also greatly improves
end-to-end voice quality.
By eliminating multi-
ple analog-to-digital
and digital-to-analog
conversions and by using low-error-rate
transmission links, voice qualiry is determined only by the encoding/decoding pnxesses.
In summary, the implementation
benefits of a fully integrated digital network are:
I' Long-distance voice quality
is identical to local voice quality
in all aspects of
noise, signal level, and distortion.
2. since digital circuits are inherently four-wire, network-generated
echoes are
eliminated,
and true full-duplex, four-wire digital circuits are available.
3. Cable entrance requirement$
and mainframe distribution of wire
pairs
is greafly
reduced because all trunks are implemented as subchannels of a TDM signal.
2.1.5 Si gnal Regenerati on
As described more fully in the next chapter, the representation
of voice (or
any analog
signal) in a digital format involves convefting the continuous
analog waveform into a
sequence of discrete sample values. Each discrete sample value is represented
by some
number of binary digits of information. when transmitted, each binary digit is repre-
sented by only one of two possible
signal values (e.g.,
a
pulse
versus no pulse
or a posi-
tive pulse
versus a negative pulse).
The receiver's
job
is to decide which discrete
values were ffansmitted and represent the message as a sequence of binary-encoded
discrete message samples. If only small amounts of noise, interference.
or distortion
are impressed
upon the signal during tansmission, the binary data in the receiver are
identical to the binary sequence generated
during the digitization or encoding process.
As shown in Figure 2.3, the transmission process,
despite the existence of certain im-
perfections,
does not alter the essential nature of the information.
Of course, if the im-
perfections
cause sufficient changes in the signal, detection errors occur and the binary
data in the receiver does not represent the original data exactly.
A fundamental attribute of a digital system is that the probability
of transmission
errors can be made arbitrarily small by inserting regenerative repeaters at intermediate
points in the transmission link. If spaced close enough together, these intermediate
nodes detect and regenerate the digital signals before channel-induced degradations
H:gmretiw
rGgdtr
Ftfttratiw
rlPcrtcr
Figure 2.3 Signal regeneration in a digital repeater line.
2.1 ADVANTAGES OF DIGITAL VOICE NETWORKS 79
become large enough to cause decision effors. As demonshated in Chapter 4, the end-
to-end error rate can be made arbitrarily small by inserting a sufficient number of re-
generation nodes in the transmission link.
The most direct benefit of the regeneration process is the ability to localize the ef-
fects of signal degradations. As long as the degradations on any particular regenerated
segment of a transmission link do not cause errors, their effects are eliminated. In con-
trast, signal impairments in analog transmission accumulate from one segment to the
next. Individual subsystems of a large analog network must be designed with tight
controls on the transmission
performance to
provide acceptable end-to-end
quality.
An individual subsystem of a digital network, on the other hand, need only be de-
signed to ensure a certain minimum error rate-usually a readily realizable
goal.
When an all-digital network is designed with enough regeneration points to effec-
tively eliminate channel errors, the overall ffansmission
quality
of the network is de-
termined by the digitization
process and not by the transmission systems' The
analog-to-digital
conversion
process inherently introduces a loss of signal fidelity
since the continuous analog source waveform can only be represented by discrete sam-
ple values. By establishing enough discrete levels, however, the analog waveform can
be represented with as little conversion error as desired. The increased resolution re-
quire$ more bits and consequently more bandwidth for transmission. Hence, a digital
transmission sy$tem readily
provides a trade-off between transmission
quality and
bandwidth.
(A similar trade-off exists for frequency-modulated
analog signals.)
2.1.6 Performance
Monitorability
An additional benefit of the source-independent
signal structure in a digital transmis-
sion system is that the
quality of the received signal can be ascertained with
no knowl-
edge of the nature of the traffic. The transmission
link is designed to
produce
well-defined pulses with discrete levels. Any deviation in the receive signal, other than
nominal amount$
planned for in the design, represents a degradation in transmission
quality. In
general, analog systems cannot be monitored or tested for
quality while in
service since the transmitted signal sfucture is unknown. FDM signals typically in-
clude pilot signals to measure channel continuity and
power
levels. The
power level
of a pilot is an effective means of estimating the signal-to-noise ratio-only in a fixed-
noise environment. Hence, noise and distortion are sometimes determined by measur-
ing the energy level in an unused message slot or at the edge of the signal passband'
In neither case, however, is the quality of an in-service channel being measured di-
rectly.
One common method of measuring the quality of a digital transmission link is to
add parity, or cyclic redundancy check
(CRC), bits to the message stream.
The redun-
dancy introduced to the data stream enables digital logic circuits in a receiver to read-
ily ascertain channel error rates. If the error rate exceeds some nominal value, the
transmission link is degraded.
Another technique for measuring in-service transmission
qualiry is used in T-cartier
lines. This technique involves monitoring certain redundancies in the signal
waveform
80 WHY DIGITAL?
itself' When the redundancy pattern
at the receiver deviates from normal. decision er-
rors have occurred. A complete description of the line coding format used in T-carrier
$ystems is provided
in Chapter 4. Other methods of measuring transmission quality
in
digital systems are discussed in Chapters 4 and 6.
2.1.7 Accommodatlon
of Other Services
It was previously pointed
out that a digital transmission system readily accommodates
conffol (signaling)
information. This fact is representative of a fundamental aspect of
digital transmission: any digitally encoded message (whether
inherently digital or
convefted from analog) presents
a coillmon signal format to the transmission system.
Thus the transmission system need provide no special attention to individual services
and can, in fact, be totally indifferent to the nature of the traffic it carries.
In an analog network the transmission standard is the 4-kHz voice circuit. All spe-
cial services such as data or facsimile must be transformed to
,,look
like voice." In par-
ticular, data signals must be converted to an analog format through the use of modems.
The standard analog channel wa$ necessarily optimized for voice quality.
In so do-
ing, certain transmission characteristics (such
as the phase
response and impulse
noise) received less attention than more noticeable voice quality
impairments. Some
less emphasized considerations, phase
distortion in particular,
are critical for high-rate
data services. Use of an analog network for nonvoice services often requires special
compensation for various analog transmission impairments. If the analog channel is
too poor,
it may be unusable for a particular
application. In contrast, the main parame-
ter ofquality in a digital system is the error rate. Low-error-rate channels are readily
obtainable. When desired, the effects of channel errors can be effectively eliminated
with error control procedures
implemented
by the user.
An additional benefit of the common transmission format is ttrat traffic from dif-
ferent types of sources can be intermixed in a single transmission medium without mu-
tual interference. The use of a common ffansmission medium for analog signals is
sometimes complicated because individual services require differing levels of quality.
For example, television signals, which require greater
transmission quality
than voice
signals, were not usually combined with FDM voice channels in a wideband analos
transmission system
[3].
2.1.8 Operabillty at Low Slgnal-to-Noiee/tnteileronce
Railos
Noise and inted'erence in an analog voice network become most apparent during
speech pauses
when the signal amplitude is low. Relatively small amounts of noise oc-
curing during a speech pause
can be quite
annoying to a listener. The same levels of
noise or interference are virtually unnoticeable when speech is present.
Hence it is the
absolute noise level of an idle channel that determines analog speech quality.
Subjec-
tive evaluations of voice quality
t4,
5l led to maximum noise level standards of 2g
dBmcO (-62
dBm0) for short-haul systems and 34 dBrnc0 (-56
dBm0) for long-haul
systems. For comparison, the power
level of an active talker is typically
-16
dBm0.
2.2 DIGITAL SIGNAL PBOCESSING 81
Thus representative end-to-end signal-to-noise ratios in analog networks
are 46 and
40 dB for short- and long-haul systems, respectively. Signal-to-noise ratios on indi-
vidual hansmission systems are necessarily higher-
In a digital system speech
pause$
are encoded with a particular data
pattern and
transmitted at the same
power level as active speech. Because signal regeneration vir-
tually eliminates all noise arising in the transmission
medium, idle channel noise is de-
termined by the encoding
process and not the transmission link. Thus speech
pauses
do not determine maximum noise levels as they do in an analog system. As discussed
in Chapter 4, digital fansmission
links provide virtually error-free
performance at signal-
to-noise ratios of 15-25 dB, depending on the type of line coding or modulation used'
The ability of a digital transmission system to reject crosstalk is sometimes more
significant rhan its ability to operate in relatively high levels ofrandom noise. One of
the most troublesome considerations in the design and maintenance of the analog net-
work was the need to eliminate crosstalk between conversations.
The problem was
most acute during pauses on one channel while an interfering channel was at maxi-
mum power. At these times relatively low level crosstalk would be noticeable. The
crosstalk was particularly undesirable if it was intelligible and therefore violated
someone's
privacy. Again, speech
pau$es do not
produce low-amplitude signals on
digital transmission links. The transmission links maintain a constant-amplitude
digi-
tal signal. Thus, low levels of crosstalk are eliminated by the regeneration
process in
a digital repeater or receiver. Even if the crosstalk is of sufficient amplitude to cau$e
detection errors, the effects appear as random noise and, as $uch, are unintelligible-
Considering the fact that a digital system typically needs a greater bandwidth than
a comparable analog system and ttrat wider bandwidths imply
greater crosstalk and
noise levels, the ability to operate at lower SNRs is
pafily a requirement of a digital
system and
partly
an advantage.
2.1.9 Ease of Encryptl on
Although most telephone users have little need for voice encryption, the ease with
which a digital bit $tream can be scrambled and unscrambled
[6]
means that a digital
network
(or
a digital cellular system) provides an extra bonus for users with sensitive
conversations. In contrast, analog voice is much more difficult to encrypt and is gen-
erally not nearly as secure a$ digitally encrypted
voice. For a discussion of common
analog voice encryption techniques,
see references
[7], [8],
and
[9].
As mentioned pre-
viously, ease of encryption stimulated early use of digital voice systems by the mili-
tary.
2.2 DIGITAL SIGNAL PROCESSING
The preceding paragraphs emphasize the advantages of digital technology in imple-
menting the tansmission and switching $ystems of a network. Another significant ap-
plication of digital technology is the area of signal
processing. Basically, signal
82 WHY DIGITAL?
processing
refers to an operation on a signal to enhance or transform its characteristics.
Signal processing
can be applied to either analog or digital waveforms. Amplif,rcation,
equalization, modulation,
and filtering are cofilmon examples of signal processing
functions.
Digital signal processing (DSP) refers to the use of digital logic and arithmetic cir-
cuits to implement
signal processing
functions on digitized signal waveforms. some-
times analog signals are converted to digital representations
for the express purpose
ofprocessing them digitally. Then the digital representations ofthe processed
signals
are converted back to analog. These operations are illustrated in Figure 2.4, where a
sine wave comrpted by noise is digitally filtered to remove the noise. The main ad-
vantages of digitally processing
signal waveforms are listed in Table 2.2.
It is important to point
out that DSP in this context refers to the technology used to
condition, manipulate, or otherwise transform a signal waveform (a
digitizert repre-
sentation thereof). In another context signal processing
refers to the interpretation
of
conffol signals in a network by the control processors
of switching $ystems. In the lat-
ter case the logical interpretation
ofa control code is processed
and not an underlying
signal waveform
2.2.1 DSP Applications
The following four sections identify applications of DSP that either represent lower
cost solutions to functions that have been traditionally implemented with analog tech-
nology or are functions that have no practical
implementation counterpart with analog
technology.
Echo Cancellers
The cost and pedormance
of DSP echo cancellers have improved to the point
that they
can be
justified
for any long-distance
circuit, thereby providing
full-duplex circuits
(no
echo suppression) and no artificial attenuation (no
via net loss). A particularly
critical need for echo cancellation occur$ in high-speed, full-duplex data modems that
incorporate near-end echo cancellation-an
unnece$sary requirement for voice cir-
cuits. Furthermore, low-cost echo canceling enables packet-switched
voice applica-
tions that inffoduce artificial delays that are not accommodated in normal analoe
Anrlog Anelog
input to
digitrl
Dieitd
dFrl
procarEof
Digit l Anrlog
to ouFut
fido0
Figure 2.4 Digital signal processing of an analog signal.
a.a DrcrrAL srcNAL PHocE$slNG 83
TABLE 2.2 Digltal Slgnal Proceeslng Features
Reproducibilitltr The immunity of digital circuit$ to small imperfections and
parasitic elements
imilies that circuits can be
pioducdd with consistent operational characteri$tics without line
adjustments or aging tolerances.
Programmability A single basic structure can be used for a variety of signal type.s and
applications by'merely
-changing
an algorithmic or
parameilic specification in a digital memory.
Time sharing'. A single digital signal
processing circuit can be used for multiple signals by
storing temforary retuftsbf each
process in random-access memory and
processing each
signal in a cyclic
(time-divided) lashion.
Automatic fesfi Since the inputs and outputs of a digital signal
processing circuit are digital.
data, tests can be
perlormed routinely by comparing test responses to data
patterns stored in
memory.
versatititr. Because of the decision-making capahilities of digital logic, digital.signal
processing can
porform many functions that are impossible or impractical with analog
implementations.
intedaces. The aclaptation logic and delay requirements
of a switched network echo
canceller virfually
preclude any type of analog implementation.
Tone Receivers
Detection of DTMF, MF, SF, or other analog tones is easily and economically realized
by converting the analog signals to digital representations
for the explicit
purpose of
detecting the tone. Of course, a DSP implementation is even more economical
when
the tones are already digitized, which is the case within a digital switch. The
program-
mability feature of a DSP circuit is particularly useful for tone receivers because one
hardware implementation can be used for multiple functions by selecting different fil-
ter options
(programs) depending on the application
[10].
Hlgh-Spead Modems
Reliable operation
(low
bit error rates) of high-speed
(e.g., ?8.8-kbps) voiceband mo-
dems
[tI]
over the switched telephone network requires sophisticated
modulation
techniques
(described in Chapter 6) antt sophisticated signal conditioning
referred to
as adaptive equalization. The only practical way to implement these functions is with
DSP circuitry. Reference
il21
describes an early application of DSP to a 14,400-bps
modem. Reference
[13]
describes the use of DSP for adaptive equalization of a 400-
Mbps digital radio. Previous digital radios used analog adaptive equalizers
because
they were cheaper. Very-high-rate digital radios require more sophisticated
equaliz-
ers, which are easier to implement
(perhaps only
possible) with DSP.
Low-Bit-Rate Voi ce E ncod i n g
The realization of low-bit-rate voice encoding algorithms described in Chapter 3 in-
volve$ extensive numerical
processing to remove redundancy in the digitized voice
samples. DSP technology is the only economical means of implementing these algo-
rithms on a real-time basis. References
[14], [15],
and
[16]
describe DSP implemen-
tations for32-,16-, and4.8-kbps voice coders, respectively. Reference
[17]
describes
more of the
general theory and application of DSP to voice compression'
84 WHY DIGITAL?
2.3 DISADVANTAGES
OF DIGITAL VOICE NETWORKS
The first pafr of this chapter discussed the basic technical advantages of digital net-
works. To balance the discussion, this section reviews the basic technical disadvan-
tages of digital implemenrarions
as listed in Table 2.3.
2.3.1 IncreaeedBandwi dth
In the brief introduction to voice digitization presented
in Chapter 1, mention is made
that transmission of samples of an analog waveform requires no more bandwidth than
the underlying waveform (at
least in theory). The bandwidth expansion comes when
the samples are encoded into binary codes and ffansmitted with an individual pulse
for
each bit in the code. Thus a Tl system requires approximately eight times as much
bandwidth as do 24 analog
voice channels since each sample is represented by an 8-bit
codeword and each bit is transmitted as a separate discrete pulse. Although more so-
phisticated
digitization algorithms can be used to encode voice at a lower bit rate than
that used on Tl systems (64 kbps), even the most sophisticated algorithms (described
in chapter 3) cannot provide
comparable voice quality
without at least a rwo-to-one
bandwidth penalty.
In some portions
of the analog network, such as the local loops, the bandwidth in-
crease did not represent much of a penalty
since the inherent bandwidth was
(and
is)
underutilized. In long-haul radio systems, however, bandwidth was at apremium, and
digital systems were relatively inefficient in terms of the number of voice channels
provided.
one mitigating aspect of a digital radio system is its ability to overcome
higher levels of noise and interference, which sometimes provides
compensation for
the bandwidth requirements, particularly
in congested transmission environments
where mutual interference
can become a limiting consideration
[3].
The inherent ro-
bustness of a digital system with respect to interference is one important attribute of
digital cellular sy$tems described in Chapter 9.
The bandwidth penalty
imposed by voice digitization is directly dependenr on the
form of transmission coding or modulation used. With greater
sophistication in the
modulation/demodulation
equipment, greater efficiency in terms of the bit rate in a
given
bandwidth is achievable. Basically, greater
transmission efficiency is achieved
by increasing
the number of levels in the line code. With limited transmit
power.
how-
ever, the distances between discrete signal levels in the receiver are reduced dramati-
TABLE 2.3 Dlsadvantages of Digital lmplementatlona
1. Increased
bandwidth
2. Need for time synchronization
3. Topologically restricted muttiplexing
4. Need for conference/extension
bridges
5. Incompatibilities with analog lacitities
2.3 DISADVANTAGES
OF DIGITAL VOICE NETWORKS 85
cally. Thus, the transmitted signal is no longer as immune to noise And
other imper-
fections as it is with lower information
densities
Using a combination of advanced digital modulation,
lower rate digitization,
and
error-correcting codes,
point-to-point digital radios could
provide voice channel effi-
ciencies comparable to or even better than analog microwave systems. Full develop-
ment along these Iines never occuffed, however, because the emergence of optical
fiber transmission eliminated
the incentive to do so.
2,3.2 Need for Time Synchronlzation
Whenever digital information is transmitted from one
place to another, a timing ref-
erence, or
"clock,"
is needed to control the hansfer. The clock specifies when to sam-
ple the incoming signal to decide which data value was transmitted. The optimum
iample times usually correspond to the middle of the transmitted
pulses. Thus, for op-
timum detection, the sample clock must be synchronized to the pulse arrival times' In
general, the generation ofa local timing reference for detecting the digital signal is not
aimcutt. Chapter 4 discusses some of the design considerations needed
to establish
proper sample clocking in the receiver of a digital transmission linft'
More subtle
problems anse, however, when a number of digital transmission
links
and switches are interconnected to form a network. Not only must the individual ele-
ments of the network maintain internal synchronization,
but also certain networkwide
synchronization
procedures must be established before the individual subsystems can
interoperate properly. Chapter 7 discusses these basic network synchronization
re-
quirements and implementations.
The need for some form of synchronization is not unique to digital nefivorks' Single'
sideband FDM transmission systems
present similarrequirements
for carrier synchro-
nization in analog networks. In analog sy$tems, however, the synchronization
requirements are less critical by about two orders of magnitude
I
I 8] '
2.3.3 Topologically
Restricted Multlplexing
To the
general public, the most apparent use of multiplexing
is broadcast
services for
radio and television. In these systems the airspace is shared by using FDM of individ-
ual broadcast channels. With this system there are no operational restrictions to the
geographic location of transmitters
and receivers. As long as the transmitters confine
theii emissions to their assigned bandwidth and each receiver uses a sufficiently
se-
lective filter to
pass only the desired channel, the network operates without mutual in-
terference. On the other hand, TDM is not nearly as amenable to applications
involving distributed source$ and destinations. Since the time of ardval
of data in a
time slot is dependent on the distance of havel, distributed TDM systems
require a
guard time between time slots. FDM systems also require guardbands between the
channels to achieve adequate channel separation.
The width ofthe FDM
guardbands,
however, is not dependent on the
geographic location of the transmitters.
In a TDM
$ystem the
guard times must be increased as the
geographic separation between trans-
86 WHY DIGITAL?
mitters increases. Furthermore,
each time division source must duplicate the synchro-
nization and time slot recognition logic needed to operate a TDM $ystem. For these
reason$, TDM has been used primarily in applications (e.g.,
interoffice
trunks) where
all of the information source$
are centrally located and a single multiplexer controls
the occurrence and assignment of time slots.
Time division multiple access (TDMA)
satellites and cellular systems are exam-
ples
of applications of TDM for distributed sources. These systems use sophisticated
synchronization
techniques so that each ground
station or mobile unit times its hans-
mission to arrive at the satellite or base station at precisely
defined times, allowing the
use of small guard
times between time slots. Notice that these applications involve
only one destination; a satellite or a base station. If an application involves multiple,
distributed
solurces and destinatians (with
transmission in more than one direction),
larger guard
times are unavoidable. Figure 2.5 shows such an application but uses
FDM instead of TDM. The main engineering consideration
for this system is to ensure
that the FDM channels have sufficient isolation
to allow a high-powered
source to be
adjacent to a receiver with the worst-case receive level. obviously, adequate FDM iso-
lation require$ a certain amount of bandwidth overhead, but it is usually fairly easy to
design filters with adequate isolation for a large range ofsignal levels so distance con-
siderations are minimized.
2.3.4 Need for Conference/Extension
Brldges
The process
of combining
multiple analog signals to form a conference caII or func-
tion as multiple extensions on a single telephone line can be accomplished by merely
bridging the wire pairs
together to superimpose all signals. Nowhere is this more con-
venient than when multiple extensions share a single two-wire line, as indicated in
Figure 2.6. When digitized voice signals are combined to form a conference.
either the
signals must be converted to analog so they can be combined on two-wire
analog
bridges or the digital signals must be routed to a digital conference bridge, as shown
in Figure 2.l.The digital bridge selectively adds the
(four-wire)
signals together (us-
ing digital signal processing)
and routes separate sums back to the conferees as shown.
Multipoint
trmcni$ion
l i ne
FDM srbchrnnels
Figure 2.5 Frequency division multiplexing on distributed multipoint line,
E.3 DISADVANTAGES OF DIGITAL VOICE NETWORKS 87
Figure 2.6 Analog telephones connected to two-wire line'
When conferencing
is implemented
in association with a switching system, the
need for a digital conference bridge is not much of a disadvantage
and in fact can sig-
nificantly improve the quality of a conference
by eliminating echoes and signal loss
caused by
power division. However,
when digital extensions need to have their signals
combined so multiple extensions can be active in a conversation,
the need for a cen-
tralized bridge can be an onerous
problem. Residential telephone wiring typically fol-
lows a daisy-chain
pattern, as indicated in Figure 2.6. Thus the need
to rewire all
outlets and install a centralized conference box is a significant
impediment to the de-
ployment of digital station equipment
in residential applications'
2.3.5 Incompati bl l l ti es
wi th Anal og Facl l l ti es
When digital equipment
was first used in
private and
public telephone networks,
it
necessarily
provided standard analog interfaces to the rest of the network.
Sometimes
these interfaces represented
a major cost of the digitat subsystem. The foremost ex-
ample of this situation arose in digital end offices. The standard analog subscriber loop
interface described in Chapter 1 is
particularly incompatible with electronic switching
machines
(analog or digital). Another aspect of digital switching that complicates its
use in analog environments
is the aftificial delay inserted by a typical digital mahix'
Both of these aspects of digital switching
are discussed in Chapter 5'
One way to eliminate
the problems with the analog interface is to use digital sub-
scriber loops and digital telephones. Unfortunately,
the overwhelming
investment
in
the loop plant for analog telephones complicates
a widespread deployment
of digital
subscriLer equipment.
Most notable of the long-established
Practices
that complicate
Digital
Conference
Bridge
Use of conference bridge for digital telephones.
Figure 2.7
88 WHY DIGITAL?
a transition to digital loops are single wire pairs,
loading coils, bridged taps,- high-
resistance or intermittent
splices, and wiring gauge
changes. The digital subscriber
loop systems described in Chapter I I accommodate most of the above impediments
but do so with very sophisticated DSp circuip.
REFERENCES
I sft-15-76 subscriber
Radio, Technical
Description, Farinon sR systems,
euebec,
Canada,1977.
2 R. J. Hirvela,
"The
Application of computer conkolled
pcM
swirching to Automatic
call Disfibution,"
IEEE communicafions ,sysferns and Technology conference,
Dallas, TX, May 1974.
3 M. R. Aaron,
"Digital
communicarions-The
silent
(R)evorution?"
IEEE
Communications
Magafine, Jan. 197g, pp. 16_26.
4 I. Nasell,
"The
lg62 survey of Noise and Loss on Toll connections."Bel/,svsrern
Technical Joumal, Mar. 1964, pp. 697-718.
5 Technical staff, Bell relephone
L,aboratories, Transmission
systems
for
communications,
westem Electric company, winston-salem, North carolina. Feb.
1970.
6 H. J. Hindin,
"LSI-Based
Data Encryption Discourages
the Data Thief,,, Electronics.
June 21, 1979, pp. 107-120.
7 N. S. Jayant, B. J. McDermott, s. w. chrisrensen, and A. M.
Quinn,
.-A
Comparison
of
Four Methods for Analog speech Encryption," Intemational
communication,
Conference Record, 1980, pp. 16.6, l-16.6.5.
I A. Gersho and R. steele, Ed.,
"special
Issue on Encryption of Analog signals,"rEEE
fournal on Selected Areas in Communicariorrs,
May 19g4.
9 H. J. Beker and F. Piper,
"speech
scrambring," speech Technology,Mar,/Apr.
r9g7,
pp.40-45.
l0 A. Fukui and Y. Fujihashi,
"A
single-chip,4-channel
MFATFC/PB
Receiver,"IEEE
G lobe com C onfe renc e Re mrd, 1987, pp. I 2.6. I
-
I 2.6.4.
I I
"A
Modem operating at Data signaling Rates Up to 2g,gfi) bps for use on rhe General
switched Telephone Network and on Leased
point-to-point
2-wire Telephone_Type
Circuits," ITU-T Rec. V.34, Geneva, Switzerland, Sept. 1994.
12 T. Kamitake, K. uehara, M. Abe, and s. Kawamura,
"rrellis
coding 14.4 kb/s Dara
Modem Implemented
with a single-chip High-speed Digital signat ilocessor,"rEEE
G lobecom Confe rence Record, I 987, pp. I 2.9. I
-
I 2.9.6.
13 H. Matsue, T. shirato, and K. watanabe,
"256
eAM
400 Mb/s Microwave Radio
system with DSP Fading countermeasures,"
IEEE International
conference on
Communications,
I 988, pp.
41.5. l-41.5.6.
14 J' L. so'
"Implementarion
on an NIC
(Nearly
Insrantaneous
companding) 32 kbps
Transcoder using the TMS320cz5
Digital signal hocessor,"
-
IEEE Globecom
Confe rence Record, 1988, pp. 43.4. 1
-43.4.5.
A bridg+'d tap is an unused pair of wires connected at some point
to an in-use pair
as alother extension or
for possible
future reassignment
of a cable pair.
l 5
l 6
t 7
l 8
REFEHENCES 89
J. L. Dixon. V. Varma, and D. W. Lin,
"software
Considerations
in the Design of
a 16
kbps Speed Coder for a TDMA Portable Phone," IEEE Globecom Conference Remrd,
1988,
pp. 26.7.1-26.7 '5.
K. Irie and s. Iai,
,.4.8
kbit/s speech codec using Advanced
Digital signal hocessors
(DSSP),' IEEE Globecom
ConJbrente Record, 1987,
pp' 20'4't-20'4'5'
M. E. Frerking,
Digital signal Processing in communitations systerns,
van Nostrand
Reinhold, New York, 1994'
J. E. Abate, E. W. Butterline, R. A. Carley, P' Greendyk, A' M. Montenegro,
C' D' Near'
s. H. Richman, and G. P. Zampetti,
'?T&T's
New Approach to the synchronization
of
Telecommunication
Networks," IEEE Comruunications
Magazine, Apr. 1989'
pp.
35-45.
VOICE
DIGITIZATION
Because of the interesting
nature of the subject and its usefulness
in a variety of ap-
plications, the field of voice digitization
has been and continues to be an area
of intense
iesearch. This research has
produced many different types of voice digitization algo-
rithms. The choice of a
particular
Upe
is primarily dependent
on the implementation
cost and the
performance requirements
implied by the application.
The algorithm cho-
sen for T1 ,yst*rn,
(companded PcM) provides excellent quality for all types of input
signals
(e.g.,
voice or data) at a moderate data rate
(64 kbps) at what was originally a
*ia"tutt
iost. The algorithms used in the first-generation
digital PBXs used lower
cost coding techniques
(higher rate PCM or delta modulation) because, at the time, a
switching application
was more sensitive to digital conversion
cost and less sensitive
to quality orhatu rate. For example, the first digital PBX in the United States, intro-
auceO Uy Rolm Corporation
in 1975, used uncompanded
PCM at a data
rate of 144
kbps because
it was iheaper than companded
PCM at the time
[1].
Subsequent
adYan-
tages derived from integrating transmission
and switching and a dramatic drop in the
"o*t of Tl-.ompatible
digital voice coders
have made obsolete the use of
"switching-
only" voice digitization
algorithms. Because
of high-volumeproduction,
Tl-compatible
coder/decoder
integrated circuits
(ICs) (codecs) can be used in switching applications
without a cost penalty. In fact, if the digital network were to be designed today, a more
complicated but economically
viable codec with a dala rate significanfly
below 64
kbps would
probably be utilized.
Transmissions applications
with strict bandwidth
limits such as high frequency
(HF) or digital cellular radio require much more sophisticated
voice digitization
algo-
rithms to achieve data rates on the order of 8-16 kbps' As a help in reducing the data
rate, the
performance requirements
of these applications
are also relaxed
as much as
the application allows.
enotttet application for digitized voice is voice storage systems-either
for re-
corded announcements
or for voice messaging'
Digital storage
is particularly apPro-
priate for recorded Innouncements
because the
playback quality does not deteriorate
*ittt ti*" and individual announcements
stored in memory or on a compact
disc
(CD)
can be randomly accessed. Speech Storage
with limited memory is an example of an
application that can u$e very low rate digitization algorithms
with significant
quality
sl
92 votcE DtctlzATtoN
reductions.
The original speak-and-spell
learning machine developed
by Texas In_
struments, for example, stored words with a few hundred bits per
word, representing
a data rate of about 850 bps
[z].
In a similar example requiring
better quality
rhe same
encoding
algorithm (LPC)
was used in the voice Aleft sysrem of chrysler automo-
biles to store 20 sec of speech (40
words) in 32,000 bits of read*only memory-a
data
rate of 1600 bps
[3].
The main reason voice
messaging systems use digital storage is to have random ac_
cess to the individual
messages. To minimize
$torage space, these systems typically
use 8*32-kbps
data rates.
At the upper end of digital storage applications
are high-fidelity
recordings of voice
and music. Many of the same advantages
of digital transmission,
as opposed to analog
transmission,
also apply to digital recordings. Foremost among these advantages is the
abiliry of defining the fideliry at the rime of recording
and maintaining
the quality
in_
definitely by periodically
copying (regenerating)
the digitally stored Informarion
be-
fore inevitable
deterioration produces
bit errors. Thus a high-quality (high-bit_rate)
digital recording of Bing crosby, Ervis
presley,
or Luciano
pavarotti
(depending
on
your
taste in music) can be saved for posterity.
This feat could not be accomplished
with analog recordings
no matter how well cared for or preserved.
As an example of
high-fidelity
audio recording,
compact disc players
[4]
record
two channels of audio
at 705 kbps each.
Speech analysis and synthesis make up another area ofwidespread
research closely
related to voice digitization. [n fact, some of the lowest bit rate voice encoders
and de-
coders use certain amounts of analysis and synthesis to digitatly represent speech. In
its ultimate form, however, analysis and synthesis have unique goals
ana applications
fundamentally
different from those of general
voice digitization. Basically, goals
of
analysis and synthesis are to recognize words
[5]
or produce
machine-generated
speech (e.g.,
text-to-speech)
[6].
one approach to analyzing
speech is to process
waveforms
with the intent of rec-
ognizing speech phonemes-the
basic units of speech from which spoken words are
constructed. once the phonemes
have been identified,
they are assigned individual
codewords for storage or transmission. A synthesizer can then genera:te
speech by re-
creating the combinations
of phonemes.
Analysis of this technique indicates that the
information
content of speech can be transmitted with a data rate of 50 bps
[7].
It must
be emphasized, however,
flrat what is hansmifted is the information
content a$sociated
with the words themselves,
not the more subjective qualities
of speech such as natu-
ralness, voice inflections,
accents, and speaker recognizability.
Thus such techniques,
by themselves,
are not applicable to general
telephony,
which customarily includes
qualities
other than the message content of spoken words.
Efficient encoding
of facsimile images presents
similar opportunities
and limita-
tions. Facsimile
machines typically scan at 200 dots per inch, which implies there are
3.T4mi l l i onbi tsofrawi nformati ononaB.5x
l l -i n.pi eceof
paper.
If thepapercon-
tains only recognizable
text characters at l0 characters and 6 lines per
inch, the same
information
can be encoded
as 561 0 ASCII characters, ot 3g,27o bits, a savings of al-
most 100 to I. Besides being restricted to text-oriented messages,
character-oriented
3.1 PULSE AMPLITUDE
MODULATIOru 93
encoding and decoding
produces the same output character font independent
of the
$ource
(which
could conceivably be hand written).
Facsimile machines achieve one
significant level of coding efficiency
without sacrificing
tfansparency by encoding
rhit" ,pu*" into run length codes. Although this does not reduce the number of bits
in a worst-case
(random-dot) image, it
greatly reduces the number of bits
in the aver-
age image of interesr. similar
processing is possible in voice applications by effi-
ciently encoding silence. However, the voice
problem is more complicated
because
reat-time voice requires reconstructing
the temporal
aspects of the source, restricting
silence encoding to relatively
large intervals'
Another level of speech analysis
involves the actual recognition of spoken
words.
High levels of success have been achieved,
with the two main restrictions that the sys-
tem is trained on the speakers and the speakers are trained to speak with isolated
words. As an example of one implementation
[8]
that tries to achieve the ultimate
goal
of speaker independence,
continuous speech, and large vocabularies, 7l-96$o
recog-
nition accuracy is
possible depending on the level of the
grammar specified.
(A gram-
mar defines allowed $equences
of words.)
Voice digitization techniques can be broadly categorized
into two classesl those
digitally *n*ding analog waveforms
as faithfully as
possible and those
processing
waveforms to encode only the perceptually signihcant aspects of speech and hearing
processes. The first category is representative
of the general problem of analog-to-
iigitut *O digital-to-analog
conversions
and is not restricted to speech digitization.
The three most common techniques
used to encode a voice waveform are
pulse code
modulation
(PCM),
differential
PcM
(DPCM), and delta modulation
(DM)' Except
in special cases, telephone
equipment designed to transparenfly
reproduce
an analog
waveform used one of these techniques.
Thus, when studying these common
wave-
form encoding techniques,
we are, in fact, sfudying the more
general realm of analog-
to-digital conversion
[9].
The second category
of speech digitization
is concerned
primarily with
producing
very low data rate speech encoders and decoders
for narrowband transmission
systems
o. digitul storage divices with limited capacity. A device from this special
class of
techrr:iques
is commonly referred to as a'*vocoder"
(voice coder)'
Very low data rate
1e.g.,
t-ZOO-Ups) vocoder techniques
generally
produce unnatural or synthetic sound-
ing speech. As such, low-data-rate
vocoders do not provide adequate
quality for
gen-
eral telephony.
A
great deal of effort has been expended to develop medium-rate
(e.g.' 8-kbps)
voice coders with natural speech qualities,
primarily for digital cellular
applications'
These coders are implemented
as a combination or hybrid of the low-bit-rate tech-
niques and the waveform coders. Thus, these techniques represent a
third class of
voice digitization algorithm.
3.1 PULSE AMPLITUDE
MODULATION
The first $tep in digitizing an analog waveform is to establish a set of discrete times at
which the input signal waveform is sampled.
hevalent digitization techniques
are
94 votcE DtetTtzATtoN
based on the use of periodic,
regularly spaced sample times. If the samples occur often
enough, the original waveform
can be completely recovered from the sample se-
quence
using a low-pass
filter to interpolate, or
"smooth
out," between the sample val-
ues. These basic concepts
are illustrated in Figure
3.1. A representative
analog
waveform is sampled at a constant sampling frequency/, =
IIT andreconstructed
us_
ing a low-pass filter. Notice that the sampling process
is equivalent to amplitude
modulation
of a constant-amplitude
pulse
train. Hence the techniqu" represented
in
Figure 3.1 is usually referred to as a
pulse
amplitude modulation
qr,lvlr.
3.1.1 Nyqul st
Sampti ng Rate
A classical resulr in sampling
sysrems was established in 1933 by Harry Nyquist when
he derived the minimum
sampling frequency required to extract all information
in a
continuous, time-varying
waveform. This result-the Nyquist criterion-is
defined
by the relation
l,
> (2xBw)
wheref =
sampling frequency
BW
=
bandwidth
of inpur signal
The derivation
of this result is indicated
in Figure 3.2, which portrays
the $pectrum of
the input signal and the resulting spectrum of the
pAM
pulse
train. The
pAM
spectrum
can be derived by observing that a continuous
train ofimpulses
has a frequency
spec-
ffum consisting of discrete term$ at multiples of the sampling frequency. The input
signal amplitude modulates these terms individually. Thus a double-sideband
spec-
trum is produced
about each of the discrete frequency terms in the spectrum of the
pulse
train. The original signal waveform is recovered
by a low-pass filter designed
to remove all but the original signal spectrum.
As shown in Figure 3.2, the reconstruc_
tive low-pass
filter must have a cutoff frequency that lies between BW andf,
-
Bw.
Hence, separation is only possible
iff,
-
Bw is grearer
than BW (i.e.,
iffr > zBw).
PAM samplet
I r t l l t , , , , ,
Lour-Fa$
fi l ter
Figure
3.1 Pulse amplitude modulation.
3.1 PULSE AMPLITUDE MODULATIOI' I 95
/ | \
I npul t pect r um
B w \ t ,
/ . - s w
Figure 3.2 Spectrum of PAM signal.
3.1.2 Fol doverDi etorti on
If the input wavefonn of a PAM system is undersampled
(f, < zBw), the original
waveform cannot be recovered without distortion. As indicated in Figure 3.3, this out-
put distortion arises because the frequency spectrum centered about the sampling
fre-
qo"o"y overlaps the original spectrum and cannot be separated from the original
spectrum by filtering. Since it is a duplicate of the input specffum
"folded"
back on
top of the desired specrum that causes the distortion, this type of $ampling
impairment
is often referred to as
"foldover
distortion'"
In essence, foldover distortion produces frequency component$
in the desired
fre-
quency band that did not exist in the original waveform.
Thus another term for this im-
prrrn""t is
"aliasing."
Aliasing problems are not confined to speech digitization
pr*rrr"r. The
potential for aliasing is
present in any sample data system. Motion
pic-
iure taking, for example, is another sampling system that can
produce aliasing. A com-
mon example occurs when filming moving stagecoaches
in old Westems- Often the
sampling
process is too slow to keep up with the stagecoach wheel movements, and
BW
Input
tpeclrum
Figure 3.3
Output
fi l ter
Di storti on energy
- 2f -3f"
- BW
Foldover spectrum
produced by undersampling an input,
96 VOICE DIGITIZATION
2. 5 kHz
Figure 3.4 Aliasing of 5.5-kHz signal into a 2.5-kHz signal.
spurious rotation rates are produced.
If the wheel rotates 355" between frames, it looks
to the eye as if it has moved backward 5".
Figure
3.4 demonstrates an aliasing process
occurring in speech if a 5.5-kHz signal
is sampled at an 8-kHe rate. Notice that the sample values are identical to those ob-
tained from a 2.5-kHz input signal. Thus after the sampled signal passes
through the
4-kHz output filter, a 2.5-wlz signal arises that did not come from the source. This
example illustrates that the input must be bandlimited,
hefore sampling, to remove fre_
quency
terms greater
than j,[,
even if these frequency terms are ignored (i.e.,
are in-
audible) at the destination. Thus, a complete
pAM
system, shown in Figure 3.5, must
include a bandlimiting
filter before sampling to ensure that no spurious or source-re-
lated signals get folded back into the desired signal bandwidth. The input filter of a
voice codec may also be designed to cut off very low frequencies to remove 60-cycle
hum from power
lines.
Figure 3.5 shows the signal being recovered by a sample-and-hold
circuit that pro-
duces a staircase approximation to the sampled waveform. With use of the staircase
approximation, the power
level of the signal coming out of the reconstructive
filter is
nearly the same as the level of the sampled input signal. The respon$e of the recon-
structive filter, in this case, must be modified somewhat to account for the spectrum
of the wider
"staircase"
samples. (The
modification
amounts to dividing the..flat" fil-
ter spectrum by the spectrum of the finite width pulse.
See Appendix C.)
The bandlimiting
and reconstructive
filters shown in Figure 3.5 are implied to have
ideal characteristics.*
Since ideal filters are physically
unrealizable,
a practical
imple-
mentation
must consider the effects of nonideal implementations.
Filters with realiz-
able attenuation slopes at the band edge can be used if the input signal is slightty
oversampled.
As indicated
in Figure 3.2, when the sampring frequencyf is somewhar greater
than twice the bandwidth,
the spectral bands are sufficiently separated from each other
-An
id-ul filter is one with a frequency-independent
time delay (linear
phase),
no attenuation in the
passband (except
as might be desired for pulse
shaping), an arbitrarily steep cutoff, and infinrte attenuation
everywhere in the stopband.
g.t puLSE
AMPLITUDE MoDULATtotrt
97
, ++
, r l l r , , ,
, \
A=
- t 0
- t l
-lo
SEmFl e
cl (sk
Figure 3.5 End-to-end PAM sYstem.
that filters with gradual roll-off characteristics can be used. As an example, sampled
voice systems typically use bandlimiting
filters with a 3-dB cutoff around 3.4 kHz and
a sampling rate of 8 kHz. Thus the sampled signal is sufficiently attenuated at the over-
lap frequency of 4 kFIz to adequately reduce the energy level of the foldover spectrum.
Figure 3.6 shows a filter template designed to meet ITU-T recommendations
for out-
of-band signal rejection in PCM voice coders. Notice that 14 dB of attenuation
is pro-
vided at 4kllz.
'/L
''ff.t::.:!:i
.10 r-
1.00
{000
rnqurrrcy
(Hz)
Figure 3.6 Bandlimiting
filter template designed to meet ITU-T recommendations
for PCM
voice coders.
6'
-r0
.=
-tl
t'
98 VOICE DIGITIZATION
As mentioned in chapter l, the perceived quality
of a voice signal is not greatly
dependent upon the phase
response of the channel (the
relative delay of individual fre-
quency
components). For this reason the phase responses of the bandlimiting filters
in the encoders and the smoothing filters in the decoders are not critical. Nonlinear
phase
responses in these filters, however, do impact high-rate voiceband data signals
when digitized. Hence a somewhat paradoxical
situation arises when voiceband data
are transmitted over a T-carrier
line: the process
of converting the voiceband data sig-
nal (28,8ffi
bps typical maximum) to a virtually error free 64-kbps digital signal
causes distortion in the lower rate data signal. However, because ofregeneration
the
transmission process
itself does not add to the signal degradation.
By interleaving the samples from multiple sources,
pAM
systems can be used to
share a transmission facility in a time division multiplex manner. As previously
men-
tioned, PAM systems are not generally
useful over long distances owing to the vul-
nerability of the individual pulses
to noise, distortion, intersymbol interference,
and
crosstalk.* Instead, for long-distance transmission
the
pAM
samples are converted
into a digital format, thereby allowing the use of regenerative
repeaters to remove
transmission imperfections before errors result.
3.2 PULSE
CODE MODULATION
The preceding
section describes pulse
amplitude modulation,
which uses discrete
sample times with analog sample amptitudes to extract the information
in a continu-
ously varying analog signal. Pulse code modulation (pcM)
is an extension of
pAM
wherein each analog sample value is quantized
into a discrete value for representation
as a digital codeword. Thus, as shown in Figure 3.7, a
pAM
system can be convefied
into a PCM system by adding an analog-to-digital (A/D)
converter at the source and
a digital-to-analog (D/A)
converter at the destination. Figure 3.9 depicts a typical
quantization process
in which a set ofquantization intervals is associated in a one-to-
one fashion with a binary codeword. All sample values falling in a particular quanti-
zation interval are represented by a single discrete value located at the center of the
quantization
interval. In this manner the quantization process
introduces
a certain
amount of error or distortion
into the signal samples. This error, known as quantiza-
PAM rampler
Di gi tal l y
encoded
Sdmple to
cl ock di gi tal
Figure 3.7 Pulse code modulation.
'As
discussed in Chapter I 1 the emergence of sophisticated DSP equalization algorithms in V,90 modems
enables PAM ransmission on analog subscriber loops.
3.2 PULSE CODE MODULATION 99
Figure 3.8
Quantization
of analog samples'
tion noise,isminimized
by establishing a large number of small
quantization intervals.
Of course, as the number of quantization intervals increases, so mu$t the number of
bits increase to uniquely identify the
quantization intervals.
3.2.1 QuantizationNoise
A fundamental aspect of the design and development
of an engineering
project is the
need for analytical measures of systems
performance. Only then can a system be ob-
jectively
measured and its cost effectiveness
compared to alternate designs. One of the
measures needed by a voice communication engineer is the quality of speech delivered
to the listener. Measurements
of speech
quality are complicated
by subjective aftrib-
utes of speech as
perceived by a typical listener. One subjective aspect ofnoise or dis-
tortion on a speech signal involves the frequency content, or spectrum, of the
disturbance in conjunction
with the
power level. These effects of noise as a function
of frequency are discussed in Chapter I with the introduction of C-message and
psophometric weighting.
Successive
quantization errors of a PCM encoder are
generally assumed to be dis-
tributed randomly and uncolrelated to each other. Thus the commutative
effect of
quantization errors in a PCM system can be treated
as additive noise with a subjective
effect that is similar to bandlimited white noise. Figure 3.9 shows the
quantization
noi$e as a function of signal amplitude for a coder with uniform
quantization intervals'
Notice that if the signal has enough time to change in amplitude by several quantiza-
tion intervals, the quantization errors are independent. If the signal is oversampled
(i.e., sampled much higher than the Nyquist rate), successive samples are likely to fall
in the sarne interval, causing a loss of independence in the
quantization errors'
The quantization elror or distor"tion created by digitizing an analog signal is cus-
tomarily expressed a$ an average noise
power relative to the average signal
power.
Thus the signal-to-quantizing-noise
ratio
(SQR, also called a signal-to-distortion
ratio
or a signal-to-noise ratio) can be determined as
100 votcE DtctlzAloN
l nput ampl i tude
Ouanti zati on
error
Figure 3.9
Quantization
elror as a function of amplitude over a range of quantization
intervals.
sQR=
Etfu)l
( 3. 1)
E{Ly(r)
-r(r)12}
where E{.}
=
expectation or averaging
x(t)
=
srulo* input signal
y(t)
=
decoded output signal
In determining the expected value ofthe quantization
noise, three observations
are
necessary:
l. The error y(r)
-
x(r) is limited in amplitude to
4/?,
where q is the height of the
quantization
interval. (Decoded
output samples are ideally positioned
at the
middle of a quantization
interval.)
2' A sample value is equally likely to fall anywhere within a
quantization
interva1,
implying a uniform probability
density of amplitude l/q.
3. signal amplitudes are assumed to be confined to the maximum range of the
coder. If a sample value exceeds the range of the highest quantization
interval,
overload distortion (also
called peak
limiting) occurs.
If we assume (for
convenience) a resistance level of I o, the average quantization
noise power
is determined in Appendix A as
J
o
o
Quantization
noir" po*er=
fr
d
(3.2)
3.2 PULSE CODE MODULATIOT' | 101
If all
quantization intervals have equal lengths
(uniform quantization), the
quanti-
zation noise is independent of the sample values and the SQR is determined
as*
where v is the rms amplitude of the input. In
particular, for a sine wave input the SQR
produced by uniform
quantization is
sQR(db)= lorogrot-+)
=
10.8 + 201"s,.
FqJ
sQR
(dB)
=
rorogl,1w#)
=7.78
+ro "*,.
[+J
(3.3)
(3.4)
where A is the
peak amplitude of the sine wave'
Exampte 3.1. A sine wave with a l-V manimum amplitr'rde
is to be digitized with
a minimum SQR of 30 dB. How many uniformly spaced
quantization intervals are
needed, and how many bits are needed to encode each sample?
Solution. Using Equation 3.4, the maximum size of a quantization interval is
determined
as
q
= (l )10{3F7' 78)/20
=
0,078 V
Thus 13 quantization intervals are needed for each polarity for a iotal of 26 intervals
in atl. The number of bits required to encode each sample is determined as
trl
=
lo9r(26)
=
4.7
=
5 bits
per sample
When measuring
quantization noise
power, the spechal content is often weighted
in the same manner as noise in an analog circuit. Unforhrnately,
spectrally weighted
noise measurements do not always reflect the h.ue
perceptual quality of a voice en-
.The
SeRs commonly compaxe unfiltered decoder outputs to unfiltered quantization errors. In actual
practice, the decoder output filterreduces the
power level ofboth the signal and the noise. The noise
power
ixperiences a greater feduction than the signal
power, since the uncorrelated noise samples
have a wider
spectrum than the corelated voice samples. Thus filtered signal-to-noise ratios are usually higher thal the
values calculated here by l-2 dB'
102 votcE DlctlzATtoN
coder/decoder. If the spectral distribution
of the quantization
noise more or less fol-
lows the $pectral content of the speech waveform, the noise is masked by the speech
and is much less noticeable than noise uncorrelated to the speech
[l0l.
on the other
hand, ifthe quantization process produces
energy at voiceband frequencies
other than
those contained in particular
sounds, they are more noticeable.
High-quality PCM encoders produce quantization
noise that is evenly distributed
across voice frequencies and independent
of the encoded waveforms. Thus quantiza-
tion noise ratios defined in Equation 3.4 are good
measures of
pcM
performance.
In
some of the encoders
discussed later (vocoders
in particular),
quantization
noise
power
is not very useful. References
tgl, I
I I
l,
and
I
I z] describe other measures of en-
coder speech quality providing
better correlations to quality
as perceived
by a listener.
3-2.2 ldle Channel Noise
Examination
of Equations 3.3 and 3.4 reveals that the SeR is small for small sample
values. In fact, as shown in Figure 3.10, the noise may actually be greater
thrur the sig-
nal when sample values are in the first quantization
interval. This effect is particularly
bothersome during speech pauses
and is known as idle channel noise. Figure 3.I t de-
picts one method of minimizing idle channel noise in
pCM
systems by establishing a
quantization
interval that straddles the origin. In this case all sample values in the cen-
tral quantization
interval are decoded as a constant zero output.
pCM
systems ofthis
type use an odd number of
quantization
intervals since the encoding ranges of positive
and negative signals are usually equal.
The quantization
characteristics required
to produce
the output waveforms
shown
in Figures 3.10 and 3.ll are shown in Figures 3.12 and 3.13, respectively. The first
characteristic (midriser)
cannot produce
a zero output level. The second characteristic
(midtread)
decodes very low signals into constant, zero-level outputs. However, if the
signal amplitude is comparable
to the size of the quantization
interval or if a dc bias
exists in the encoder, midtread quantization
will produce
about as much idle channel
noise as midriser quantization.
As mentioned in chapter l, noise occurring during speech pauses
is more objec-
tionable than noise with equivalent power
levels during speech. Thus idle channel
2.0
r. 0
- 1. 0
-2.O
Figure 3.10 Idle channel noise produced
by midriser quantization.
3.2 PULSE CODE MODULATION 103
Figure 3.ll Elimination of idle channel noise by midtread
quantization.
noise is specified in absolute terms separate from quantization noise, which is speci-
fied relative to the signal level. For example, Bell system D3 channel bank specifica-
tions list the maximum idle channel noise as 23 dBmCO
[13].
3.2.3 Uniformly Encoded PCM
An encoder using equal-length
quantization intervals for all samples
produces code-
words linearly related to the analog sample values. That is, the numerical equivalent
of each codeword is proportional to the
quantized sample value it represents. In this
mamer a uniform PCM sy$tem uses a conventional analog-to-digital converter to
gen-
erate the binary sample codes. The number of bits required for each sample is deter-
mined by the maximum acceptable noise
power. Minimum digitized voice
quality
requires a signal-to-noise ratio in excess of 26 dB
[
14]. For a uniform PCM system to
achieve a sQR of 26 dB, Equation 3'4 indicates that
4*o
=
0' l23A' For equal
positive
and negative signal excursions
(encoding from
-A
to A), this result indicates that
just
over l6 quantization intervals, or 4 bits per sample, are required.*
Flgure 3.12 Midriser
quantizer characteristic.
.This
SeR objective is for minimum acceptable
perform$nce and assumes all degadations occur in a sin gle
encoder, If additional sigrral impairments occur
(such as multiple A/D conversions), the encoder must use
more bits to
ptovide noise margin for other elements in the network.
104 votcE DtctTtzATtoN
Figure 3.13 Midtread quantizer characteristic.
In addition to
providing
adequate quality
for small signals, a telephone system musr
be capable of transmitting a large range of signal ampritudes, referred to as dynamic
range. Dynamic range (DR)
is usually expressed in decibels as the ratio of the maxi-
mum amplitude signal to the minimum amplitude signal:
(3.s)
A typical minimum dynamic range is 30 dB
tt4l.
Thus signal values as large as 3l
times A must be encoded without exceeding the range of
quantization
intervals. As-
suming equally spaced quantization
intervals for uniform coding, the total number of
intervals is determined as 496, which requires 9-bit codewords.*
The performance
of an n-bit uniform PCM system is determined by observing that
(3.6)
where A.u* is the manimum (nonoverloaded)
amplitude.
substituting Equation 3.6 into E4uation 3.4 produces the
pcM
performance
equa-
.
tion for uniform coding:
DR= to**,.fts]
\_
ffirn
/
=zorog,oti"_il
M*^*
q=- T-
(3.7)
This rcsult is derived with the assumption of minimum performance
requirements. Higher performance
objectives (less quantization
noise and grcater
dynamic range) require as many as I 3 bits per sample for
uniform PCM systems. This coding performance
was established when it was likely that multiple
conversions would occur in an end-to-end connection. Now that the possibility
of multiple ArD and D/A
conversions has been eliminated, end-to-end voice quality is much better than it was in the analog network,
seR
=
116 + 6.ozn+ 2o log,o
t^*l
3.2 PULSE CODE MODULATIoN 105
The first two terms of Equation 3.7
provide the SQR when encoding a full-range sine
wave. The last term indicates a loss in SQR when encoding a lower level signal. These
relationships are
presented in Figure 3.14, which shows the SQR of a uniform PCM
sy$tem as a function of the number of bits
per sample and the magnitude of an input
sine wave.
Example 3.2. What is the minimum bit rate that a uniform PCM encoder must
provide to encode a high-fidelity audio signal with a dynamic range of 40 dB? Assume
the fidelity requirements dictate
passage of a 20-kHz bandwidth with a minimum
signal-to-noise ratio of 50 dB. For simplicity, as$ume sinusoidal input signals'
Solutinn. To prevent foldover distortion, the sampling rate must be at least 40 kHz.
Assuming an excess sampling factor comparable to that used in D-type channel banks
(4000/3400), we choose a sampling rate of 48 kHz as a compromise for a practical
bandlimiting filter. By observing that a full-amplitude signal is encoded with an SQR
of 40 + 50
=
90 dB, we can use Equation 3.7 to determine the number of bits n required
to encode each sample:
90
-
1.76
n - G
' "
6. 02
=
15 bits
Thus the required bit rate is
(15
bits/sampleX48,O00 samples/sec)
=
720 kbps
-
40
-30 -20 -
l 0
A/A6.\
(dB)
Figure 3.14 SQR of uniform PCM coding.
e
[ *
s
o
t
EI
.E
40
E
F
o
T e o
o
c
a
106 vorcE DlcrrzATroN
3.2.4 Companding
In a uniform PCM system the size of every quantization
interval is determined by the
SQR requirement of the lowest signal level to be encoded. Larger signals are also en-
coded with the same quantization
interval. As indicated in Equation 3.7 and Figure
3.14, the SQR increases with the signal amplitudeA. For example, a Z6-dB SeR for
small signals and a 30-dB dynamic range produces
a 56-dB SQR for a maximum-
amplitude signal. In this manner a uniform PCM system provides
unneeded quality
for large signals. Moreover, the large signals are the least likely to occur. For these rea-
sons the code space in a uniform PCM system is very inefficiently utilized.
A more efficient coding procedure
is achieved if the quantization
intervals are not
uniform but allowed to increase with the sample value. When quantization intervals
are directly proportional
to the sample value, the SQR is constant for all signal levels.
with this technique fewer bits per sample provide
a specified seR for small signals
and an adequate dynamic range for large signals. When the quantization
intervals are
not uniform, a nonlinear relationship exists between the codewords and the samples
they represent.
Historically, the nonlinear function was first implemented on analog signals using
nonlinear devices such as specially designed diodes
[15].
The basic process
is shown
in Figure 3.15, where the analog input sample is f,rrst compressed and then quantized
with uniform quantization intervals. The effect of the compression operation is shown
in Figure 3.16. Notice that successively larger input signal intervals are compressed
into constant-length quantization
intervals. Thus the larger the sample value, the more
it is compressed before encoding. As shown in Figure 3.15, a nonuniform PCM de-
coder expands the compressed value using an inverse compression characteristic to re-
cover the original sample value. The process
of first compressing and then expanding
a signal is referred to as compandlng. when digitizing, companding amounts to as*
signing small quantization
intervals to small samples and large quantization
intervals
to large samples.
various compression-expansion characteristics can be chosen to implement a
compandor. By increasing the amount of compression, we increase the dynamic range
at the expense of the signal-to-noise ratio for large-amplitude signals. one family of
compression characteristics used in North America and Japan is the p-law
charac-
teristic, defined as
Compressed
di gi tal
codewords
Compre$ion
Li near Expansi on
PCM
decoder
Li n6ar
PCM
encoder
Figure 3.15 Companded PCM with analog compression and expansion.
g,E
PULSE coDE MoouLATtoN 107
rot
t
E
E
t * r t
9 E
E
",,'8
oro $
Input sfirple vrlus
Figure 3.16 Typical compression characteristic.
(3. 8)
wherex
=
input signal amplitude
(
-1
{x { 1)
sgn(x)
=
polarity ofx
F
=
pllrameter used to defirne amount of compression
Because of the mathematical nature of the compression curve, companded PCM is
sometimes referred to as log-PCM. A logarithm compression curve is ideal in the
sense that quantization intervals and, hence,
quantization noise are proportional to the
sample amplitude. The inverse or expansion characteristic for a
p-law compandor is
defined as
. . l n ( l + UL t l )
Fr(r)
=
sen(x)
ffi[y
F' r(v):' s"0)
[iJr(1
+ p;rrr- 1, (3.e)
where
)
=
the compressed value,
=Fp(x) (-l
< y < l)
$gn(y)
-
polarity
ofy
F
='
companding parameter
The first T-carrjer systems in the United States used Dl channel banks
[16],
which
approximated Bquation 3.8 for
F
=
100. The compression and expansion functions
were implemented with the specially biased diodes mentioned previously. Figure 3.17
depicts a block diagram of a Dl channel bank. Notice that the time division multiplex-
ing and demultiplexing functions are implemented on analog PAM samples. Thus the
companding and encoding/decoding functions were shared by all 24 voice channels.
108 vorcE DtetT|zATtoN
I
a
t
-+E
Tl transmi rsi on l i ne
Anal og
Inpul i
Bandl i mi ti ng
fi l terB
Figure 3.17 Functional block diagram of Dl channel bank.
The ability to share this relatively expensive equipment was one of the reasons that
PCM was originally chosen as the means of digitally encoding speech. subsequent de-
velopment of integrated circuit PCM codecs diminished the need to share this equip-
ment, Thus later generation
$ystems could use per-channel
codecs and
provide
more
flexibility in implementing various $ystem sizes. when most of the cost of a channel
bank is in the common equipment,
as in the original channel banks, less than fully pro-
visioned sy$tems are overly expensive.
Each sample produced
by a Dl channel bank was encoded into 7 bits; I polarity
bit and 6 compressed magnitude bits. In addirion, I signaling bit was added to each
channel to produce
an 8-bit codeword for each time slot. since the sampling rate was
I kHz, a 64-kbps channel resulted. Even though the Dl channel banks have been su*
perceded
by newer channel banks utilizing a different coding format, the 64-kbps
channel rate has persisted
as a standard.
3.2.5 Eaeily Dlgltally Linearlzabte
Goding
The success of the flrst Tl sy$tems for interoffice exchange area transndssion paved
the way for fufther use of digital time division multiplex systems. As it became appar-
ent that digital transmission was useful within the toll network, it also became clear
that the form of PCM encoding used in the Dl channel banks was inadequate. In con-
trast to the exchange area, an end-to-end connection through the toll network could
have conceivably involved as mmy as nine tandem connections. Since digital switch-
ing was not in existence at the time that T-carier systems for the toll network were
being developed, each of these tandem connection$ implied an extra DiA and A/D
3,2 PULSE CODE MODULATION 109
conversion. Thus the quality
of each conversion had to be improved to maintain the
desired end-to-end
quality" The D2 channel bank
[
7] was therefore developed with
improved digitized voice quality.
When the D2 channel bank was being developed, digital switching was recognized
as a coming technology, implying that channel banks would be paired
on a dynamic
basis, as opposed to the one-to-one basis in T-carrier systems. Thus a greater degree
of uniformity in companding characteristics would be required to allow
pairing of
channel banks on a nationwide basis. The main features incotporated into the D2 chan-
nel bank
(and ensuing channel banks such as D3, D4, and D5) to achieve the improved
quality and standardization are:
L Eight bits per PCM codeword
2. Incorporation of the companding functions directly into the encoder and
decoder
3. A new companding characteristic
(p255)
The Dl channel banks use I bit per time slot for signaling and 7 bits for voice. Thus
a signaling rate of 8 kbps was established, which was more than necessary for basic
voice service. To provide
a higher data rate for voice, signaling between D2 and all
subsequent channel banks is inserted into the least significant bit position of 8-bit
code-words in every sixth frame. Thus every sixth
p255
PCM codeword contains only
7 bits of voice information, implying that the effective number of bits
per sample is
actually 7f bits instead of 8. The use of 1 bit in every sixth frame for signaling is often
referred to as
"robbed
bit signaling." When common-channel signaling is utilized, the
associated T-carrier systems no longer need to carry signaling information on a
per-
channel basis and a full I bits of voice can be transmitted in every time slot of every
foame.
The compression and expansion characteristics of the Dl channel banks were im-
plemented separately from the encoders and decoders. The D2 charurel bank incorpo-
rates the companding operations into the encoders and decoders themselves. In these
channel banks a resistor array is used to establish a sequence ofnonuniformly spaced
decision thresholds. A sample value is encoded by successively comparing the input
value to the sequence ofdecision thresholds until the appropriate
quantization interval
is located. The digital output becomes whatever code is used to represent the
particular
quantization interval. (See
Appendix B for a detailed description ofthe direct encod-
ing
procedure used in the D? channel banks.) By incotporating the companding func-
tions directly into the encoders and decoders, D2 channel banks avoid certain
problems associated with parameter variability and temperature sensitivity of com-
panding diodes in Dl channel banks.
The D2 channel banks also introduced improved
performance in terms of the effect
of channel errors on the decoded outputs. Of
paramount concern in a PCM system is
the effect of a channel enor in the most significant bit
position of a codeword. Bit er-
rors in other
positions
of a codeword are much less noticeable to a listener. A channel
error in the most significant bit of a codeword produced by a Dl channel bank causes
an output error that is always equal to one-half of the entire coding range of the coder'
1 1O Vo|cE DIGITIZATIoN
The D2 channel bank, on the other hand, uses a sign-magnitude coding format. With
this format, a channel error in the polarity
bit causes an output error that is equal to
twice the sample magnitude
(i.e.,
the polarity
is inverted). In the worst case this enor
may correspond to the entire encoding range. Maximum-amplitude samples are rela-
tively rare, however, so most channel error$ in a D2 coding format produce
outputs
with error magnitudes less than one-half the coding range. Thus, on average, the sign-
magnitude coding format of the D2 and ensuing channel banks provide
superior
per-
formance in the presence
of channel enors
I
I 7].
In addition to a need for improved voice quality,
it also became apparent that as
more of the network began using digital techniques, it would be necessary, or at least
desirable, to implement many signal processing functions directly on digital signals
and not convert them to an analog format for processing. Most signal processing func-
tions (such
as attenuating a signal or adding signals together) involve linear opera-
tions. Thus before processing a log-PCM voice signal, it i$ necessary to convert the
compressed transmission format into a linear
(uniform)
format.
To simplify the conversion process,
the particular companding characteristic with
F
=255
was chosen. This companding characteristic has the property
of being closely
approximated by a set of eight straightJine segments also refened to as chords. Fur-
thermore, the slope of each succes$ive segment is exactly one-half the slope of the pre-
vious segment. The flust four segments of a
p255
approximation are shown in Figure
3.I8. The overall result is that the larger quantization
intervals have lengths that are
binary multiples of all smaller quantization intervals. Because of this properly,
a com-
pressed
codeword is easily expanded into a uniform representation. Similarly, the uni-
form representation is easily converted into a compressed representation.* In fact,
commercially available PCM codecs digitally compress uniform codewords instead of
using direct compressed encoding, as in the D2 channel banks. These techniques use
a uniform encoder with a relatively large number of bits to cover the entire dynamic
range of the signal. As described in Appendix B, the least signifrcant bits of large sam-
ple
values are discarded when compressing the code. The number of insignificant bits
deleted is encoded into a special field included in the compressed-code format. In this
manner digital companding is analogous to expressing a number in scientific notation.
As shown in Figure 3.18, each major segment of the piecewise linear approxima-
tion is divided into equally sized quantization
intervals. For 8-bit codewords the num-
ber of quantization
intervals per
segment is 16. Thus an 8-bit
p255
codeword is
composed of I polarity bit, 3 bits to identify a major segment, and 4 bits for identifying
a quantizing interval within a segment. Table 3.1 lists the major segment endpoints,
the quantization
intervals, and tlre corresponding segment and
quantization
interval
codes.
The quantization
intervals and decoded sample values in Table 3. I have been ex-
pressed
in terms of a maximum-amplitude signal of 8159 so that all segment endpoints
and decoder output$ are integers. Notice that the quantizing step is doubled in each of
'The
inexorable advance of semiconductor technology has obviated much of the ingenuity that weflt into
selecting EDL coding fbrmats. Brute-force table look-up conversion between codes using read-only
memories (ROMs) is now the most cost-effective approach,
3.2 PULSE coDE MoDULAToN 111
31 95 223 478
Linear tignal
Flgure 3.18 First four segments of straight-line approximation to
p255 compression curve.
eight successive linear segments. It is this property that facilitates the conversion to
and from a uniform format. A complete encoding table is provided in Appendix B
along with a detailed description ofthe conversion
process to and from uniform codes.
The straight-line approximation of the
p255 companding curve i$ sometimes re-
ferred to as a lS-segment approximation. The 15 segments arise because, although
there are 8 positive
$egments and 8 negative segments, the 2 segments nearest the ori-
gin are colinear and therefore can be considered as l. When viewed in this manner,
the middle segment contains 3l
quantization intervals with I segment straddling the
origin
(from
-1
to +1 in Table 3.1). Codewords for this middle quantization interval
arise as a positive
value less than +1 or a negative value greater than
-1.
There are, in
effect, a positive zero and a negative eero. As represented in Table 3.1, these values
are encoded as 00000000 and 1ffi00000, respectively. However,
p255 PCM codecs
invert all codewords for transmission. The smaller amplitude signals, with mostly 0
bits in the segment code, are most
probable and would therefore cause les$ than 50%
pulses on the transmission line. The density of
pulses is increased by inversion of the
transmitted data, which improves the timing and clock recovery
performance of the
receiving circuitry in the regenerative repeaters. Thus the actual transmitted code-
wordscorrespondi ngtoaposi ti vezeroandanegati vezeroarerespecti vel y 111111I1
and 01111I I 1, indicating strong timing content for the line signal of an idle channel.
ln the interest of ensuring clock synchronization in the Tl repeaters, p255 PCM
codecs alter tlrc ffansmitted data in one other way. As indicated in Table 3.1, a maximum-
amplitude negative signal is all I
'
s, which would normally be converted to all 0's for
o
.9
ll
$
E
ET
E
I
I 12 vorcE DrcrrzATtoN
TABLE 3.1 Encodlng/Decoding Table for p2SS PCMd
Input
Amplitude
Range
Segment Quantization
Step Size Code S Code Q Code Varue
Decoder
Amplitude
0-1
1-3
3-5
;
29-31
31-35
;
91-95
95-1 03
21F.223
2?3-239
463479
479-51 1
oEo-ool
991- 1 055
r ss1
-ZOr
S
201ts2143
3935-4063
4063-431 I
7903-81 59
000
0000
0001
001 0
1 1 1 1
0000
1 1 1 1
0000
:
1 1 1 1
0000
r r ir
0000
1 1 1 1
000
1 1 1 1
0000
1 1 1 1
0000
1 1 1 1
2
4
001
010
0 1 1
100
1 0 1
1 1 0
1 6
32
64
30
33
9;
99
256
1 1 1
112
219
231
471
495
975
1 023
1 983
2079
3999
4191
8031
:
127
this table displays magnitud Bncoding only. Polarity bits er6 a8signed a
"0"
for
positive
end a
,'1"
for negatlve.
h transmission ell bits are inverted.
transmission. Instead, for the all-0 codeword only, the second least significant bit is
set to I so that 00000010 is transmitted. In effect, an encoding error is produced
to
pre-
clude an all-O codeword. Fortunately, maximum-amplitude samples are exffemely un-
likely so that no significant degradation occurs. (If
the least significant bit were forced
to a 1, a smaller decoding error would result. However, in every sixth frame this bit is
"stolen"
for signaling pu{poses
and therefore is occasionally set to 0 independently of
the codeword. To ensure that
"all-zero"
codewords are never transmitted. the second
least significant bit is forced to a I when necessary.)
Example 3.3. Determine the sequence of codewords for a
p255
PCM encoded
channel bank format representing a half-maximum-power l-kHz digital signal.
3,2 PULSE CODE MODULATION 113
Solution. Since the sampling rate of the standard p255 PCM channel bankis 8 kHz,
a sequence of eight samples can be repeated in cyclic fashion to
generate the l-kHz
waveform. For convenience, the
phases of the samples are chosen to begin at 22.5o .
Thus the eight samples correspond to 22.5"
,
67 .5"
,
II2.5o
,
l5'l .5"
,
202.5"
,
247 .5",
292,.5", and 337.5o. With these phases, only two different magnitudes coresponding
to 22.5o and 67.5' are required. The maximum amplitude of a half-maximum-power
sine wave is 0.707 x 8159
=
5768. Thus the two amplinrdes contained in the sample
sequences are
(5768)sin(22
.5")
=
2207
(5768)sin(67.5o)
=
5329
Using the encoding table in Appendix B, we determine the codes for these sample
magnitudes to be 1100001 and 1110100, respectively. The sequence of eight samples
can now be established as follows:
Sample Phase
(deS) Polarity Segment
Quantization
2 2 . 5 0 1 1 0 0
6 7 . 5 0 1 1 1 0
112. 5 0 1 1 1 0
157. 5 0 1 1 0 0
2 0 2 . 5 1 1 1 0 0
247.5 1 1 1 1 0
2 9 2 . 5 1 1 1 1 0
3 3 7 . 5 1 1 1 0 0
0
1
1
0
1
1
0
0
0
U
0
0
0
0
0
1
0
0
1
1
0
0
1
Note: This sequence defines a 1-kHz test signal at a
power lv6l of 1 mW at the transmission level
point (0
dBmo). However, the actual transmltted data
pattern is the complement of lhat
provid6d above.
because only two amplitude samples are required to
produce
the test ton6, this tono does not test all
encoding/docoding circuitry. In general, a 1004-Hz tone is a better t6st tone since it is not harmonically
related to an 8000-Hz sampling rate and will therefore exercise all encoder and decoder levels.
Pertormance of ap255 PGM Encoder
As mentioned, the main motivation for changing the encoding algorithm of the Dl
channel bank was to
provide better speech
quality for digital toll network transmission
links. The SQR for a maximum-amplitude $ine wave in the first segment of a
p255
codec is determined easily from Equation 3.4 as
SQR(A
:
31)
=
7.78 + 20t"*r.
[*l
\ . -
/
=
31. 6 dB
The SQRs for larger amplitude sinusoids are not as easy to calculate since the
lengths of the quantization intervals vary with the sample size. Thus a general calcu-
lation of the quantizing noise
power involves finding the expected value of the
power
of the
quantization errors;
1 14 VotcE DtctlzATtoN
7
. l _
noise power
=;lW?
r'4
(3. 10)
where p; =
probability
of a sample in ith segment
4;
-
quantization
size for segment i,
=!i+t
for segmented
p255
coding
Using Equation 3.10, we determine the SQR power for a full-range sinusoid as
SQR(A
=
8159)
=
39.3 dB
For comparison, if all quantization
intervals had the maximum length of 256 as in
the upper segment,
trquation
3.4
provides
an SQR of 37.8 dB. The difference of only
1.5 dB reflects the fact that a full-scale sine wave spends 67Vo of the time in the upper
segment where the quantization intervals are maximum (i-e., pt =
0.67). A voice sig-
nal, on the other hand, has a higher peak-to-average
ratio than a sine wave. The aver-
age quantization error is smaller, but so is the average signal power. Hence the SQR
is approximately the same.
The dynamic range of a segmented PCM encoder is determined as the signal power
difference between a low-level signal occupying the entire range of the first segment
and a high-level signal extending to the limits of the code. Thus the dynamic range of
a segmented 255 coder is determined as
DR(A
=
3I toA
=
8159)
=
20l og,o(8159/31)
=
48.4 dB
In summary, an 8-bit
p255
PCM codec provides
atheoretical SQR greaterthan
30 dB
across a dynamic range of 48 dB. For comparison, Equation 3.4 or Figure 3.14 reveals
that a uniform PCM encoder/decoder requires 13 bits for equivalent performance.
(The
extra quality of uniform encoding at high signal levels is unneeded.)
The theoretical performance
of an 8-bit segmented
F255
coder is shown in Figure
3.19 as a function of the amplitude of a sine wave input. Also shown is the theoretical
performance
of an unsegmented p255 coder and a 7-bit
plfi)
coder used in the Dl
channel bank. Notice that the 8-bit coders provide
about 5 dB improvement over the
7-bit coder for high-level signals and even more improvement for low-level signals.
The performance shown for the 8-bit coders does not include the effect ofusing only
7 bits for voice coding in every sixth frame. When this effect is included, the B-bit
coders lose I.76 dB in performance.
The scalloped effect ofthe segmented coder occurs because the quantization
inter-
vals change abruptly at the segment endpoints instead of continuously as in analog
companding. AIso as shown in Figure 3.19, note the required performance of a D3
channel bank and codecs designed to meet this specification
[13].
This specification
assumes all noise measurements are made using C-message weighting. C-message
weighting reduces the effective noise level by 2 dB and therefore improves the SQR.
Thus an ideal 8-bit p255 coder actually exceeds the specif,rcation by more than that
3.2 PULSE coDE MoDULATIoN 115
Full load rignal
=
+3d8m0
Unfiltered
pieccYviBe
linEdr
I-258
Unfiltered
B-bir
#-?55
-50 -40 -30 -20
Signal
porrnr
of rinewave
(dBm0l
Figure 3.19 SQR of
pJaw coding with sine wave inputs.
shown in Figure 3.19.
rilhen
the least significant bit of every sixth frame is used for
signaling, however, the SQR is reduced by a comparable 1.76 dB.
A-Law Compandlng
The companding characteristic recommended by ITU-T for Europe and most of the
rest of the world is referred to as an A-law characteristic. This characteristic has the
same basic features and implementation advantages as does the
p-law characteristic.
In particular, the A-law characteristic can also be well approximated by shaight-line
segments to facilitate direct or digital companding and can be easily convelted to and
from a uniform format. The normalized A-law compression characteristic is defined as
6
:
.E
g
o
, t
E
o
'n
E
3
6
-!
il
c
.gl
th
f*r*r<r
f . . f a u I
lSsn(Jl |
-------
|
l " [ 1 +l n ( A) l
Fa(x)
=
i -
| . I t +t nt t . l l
i ssn( x) l - - |
l .
- ' - [ 1+l n( A) _J
0 < rxr<f
The inverse or expansion characteristic is defined as
( 3. 1 l )
116 votcE DtctlzAlor\
F;t(v)
=
(3. 12)
where y
=
F.A($ and A
=
87.6.
Notice that the first portion
of the A-law characteristic is linear by def,rnition. The
remaining portion of the characteristic (l/A < lxl
<
I) can be closely approximated by
linear segments in a fashion similar to the
trr-law
approximation. In all, there are eight
positive and eight negative segments. The first two segment$ of each polarity (four
in
all) are colinear and therefore are sometimes considered as one straight-line segment.
Thus the segmented approximation of the A-law characteristic is sometimes refened
to as a
"13-segment
approximation." For ease in describing the coding algorithms of
the segmented companding characteristic, however, a ld-segment representation is
used,just as in the case ofthe segmented
p-law
characteristic.
The segment endpoints, quantization
intervals, and corresponding codes for an 8-
bit segmentedAlaw characteristic are shown in Table 3.2. The values are scaled /o a
maximum value of 4096 for integral representations. Figure 3.?0 displays the theoreti-
cal performance
of the Alaw approximation and compares it to the performance
of a
p-law
approximation presented in Figure 3.19. Notice that the A-law characteristic
provides a slightly larger dynamic range. However, theAJaw characteristic is inferior
to the
p-law
characteristic in terms of small-signal quality (idle
channel noise). The
difference in small-signal performance
occurs because the minimum step size of the
A-law standard is 2/4096 whereas the minimum step size of the
p-law
is 2/8159. Fur-
thermore, notice that the A-law approximation does not define a zero-level output for
the first quantization
interval (i.e.,
uses a midriser quantizer).
However, the difference
between midriser and midtread performance
at 64 kbps is imperceptible
tl8l.
3.2.6 SyllabicCompandlng
A significant attribute of the companding techniques described for
pcM
systems is
that they instantaneou$ly encompass the entire dynamic range of the coder on a sample-
to-sample basis. Thus p-law
and A-law companding is sometimes referred to as in-
stantaneous companding. However, because the power level of a speech signal
remains fairly constant for 100 or more 8-kHz samples, it is wasteful of code space to
be able to encode very low level signals adjacent to very high level signals. In other
words, it is unnecessary to allow for instantaneous changes in the dynamic range of a
signal. One technique of reducing the amount of transmission bandwidth allocated to
dynamic range is to reduce the dynamic range of the signal at the source before en-
coding and then restore the original dynamic range of the signal at the receiver while
decoding. when the adjustments to the dynamic range occur on a periodic
basis that
more or less corresponds to the rate of syllable generation,
the technique is referred to
as syllabic companding. Because the dynamic range adjustments occur only every 30
msec or so, there is very little bandwidth needed to communicate the adjustments.
*entrluT@ o<ryr<*fu
pFl(1+|fl(A)l-l
' *ol T
*fustvt(l
3.2 PULSE CODE MODULATION 117
TABLE 3.2 $egmented A-Law Encodlng/Decoding Tablea
Input
Amplitude
Range
Segment Ouantization
Step Size Code S Code Q Code Value
Dcoder
Amplitude
r2
24
1
3
3;
33
63
66
126
132
252
264
504
528
0000
0001 000
3G-32 2
32-34
: 001
62-64
64-8
i 4 010
124-128
1 2B-1 36
: I 01 1
248-256
25tsl272
: 16 100
496-51 2
512-544
i 3e 101
992*1 024
1 024-1 088
: 64 110
1 984-2048
?048-2176
i 1 2 8 1 1 1
3968*4096
1 1 1 1
0000
1 1 1 1
0000
1 1 1 1
0000
1 1 1 1
0000
1 1 1 1
0000
:
1 1 1 1
0000
1 1 1 1
0000
1 1 1 1
95
96
;
1 1 1
112
127
1 008
1 056
201 6
2112
4032
aln
transmission, every oth6r bit is invened.
Syllabic companding was first developed for use on noisy analog circuits to im-
prove the i<lle channel noise. As shown in Figure 3.21, the
power level of low-level
syllables is increased
(compressing the dynamic range) for transmission but attenu-
ated upon reception
(expanding the dynamic range). The process ofattenuating the re-
ceived signal restores the low-power syllable to its original level but attenuates any
noise arising on the transmission link. Thus the receiver signal-to-noise ratio (SNR)
is improved for low-level signals. The amount of amplification applied at the source
is dependent on the short-term
(syllabic) power level of the transmitted signal. Simi-
larly, the compensating attenuation applied at the receiving terminal is determined
from the short-term power level of the received signal.
Syllabic companding on digital systems provides the same basic improvement in
SQRs as it does on noisy analog transmission links. When the digital encoders and de-
1 18 VOICE DIGITIZATIoN
-70
Ftgure 3.20
-60 -50 -40 -30 -20 _lo
Siqnel
powr
of rinswrve
(dBm0l
SQR of A-law PCM coding with sine wave inputs.
30
6
E
:
.E
a
q
fl
g
E
o
.E
.E
c
A
I

.F
6
Input porvr
l evel
(dBm)
* t o [ - ' "
* ' J * f f i u u
(//w^c.*oi*'
--sH
Vari abl e
TrEnsrfiittd
l 6vBl
{dBm}
Ponnr Power
ftreasuromnt meesuteftent
Figure 3.21 Syllabic companding of an analog signal.
Output povt/t
l evel
(dBm)
* 7 . 5 - . r 0
,rr\s-a,
Expanrion
\\
Vari ebl e\ --59
attenueti on
3.2 PULSE CODE MODULATION 1 19
coders are considered as
part of the transmission link, the
process of amplifying low-
level signals before encoding and attenuating them after decoding effectively reduces
the
quantization noise with no net change in signal level. In practice, syllabic com-
panding as implemented in digitized voice terminals does not amplify the signal at the
source and attenuate it at the destination. Instead, an equivalent
process of controlling
the step sizes in the encoder and decoder is used. As far as the transmitted
bit stream
is concerned, it makes no difference if the signal is amplified and encoded with fixed
quantization or if the signal is unmodified but encoded with smaller
quantization in-
tervals. Thus syllabic compandors in digital voice terminals typically reduce the
quan-
tization intervals when encoding and decoding low-power syllables but increase the
quantization intervals for high-power syllables'
Although syllabic companding can be used in conjunction with any type of voice
coding, the technique has been applied most often to differential sy$tems
described in
the following sections. In many of the applications, the adaptation time has been re-
duced to 5 or I0 msec, which is somewhat shorter than the duration of a typical sylla-
ble
(approximately 30 msec). The technique is still generally referred to as syllabic
companding, however, to distinguish it from the instantaneous variety.
To adjust the step sizes in the decoder in synchronism with adjustments made in
the encoder. some mean$ must be established to communicate the step size informa-
tion from the source to the destination. One method explicitly transmits the step size
information as auxiliary information. A
generally more usefirl approach is to derive
the step size information from the transmitted bit stream. Thus, in the ab$ence
of chan-
nel erTors, the decoder and the encoder operate on the same information. This
proce-
dure is analogous to syllabic companded analog systems in which the receiver determines
its attenuation requirements from the short-term power level of the received signal. In a
digital system the bit stream is monitored for certain data
pattems that indicate the power
level of the signal being encoded. Indications of high power level initiate an increase in
the step size, whereas indications oflow levels cause a decrea$e.
Determining the step size information from the transmitted bit stream is generally
better than explicitly transmitting the step sizes for the following reasons. Because
there is no explicit transmission of step size, the transmission of sampled speech in-
formation is never intemrpted, and the speech sample rate is equal to the transmission
rate. Also. the bit sheam does not have to be framed to identify step size infbrmation
separately from the waveform coding. Furthermore, if the step size adjustments are
made on a more gradual basis, the individual increments are small enough that occa-
sional incorrect adjustments in the receiver caused by channel errors are not critical'
However, on transmission links with very high error rates
(one
error in a hundred bits
or so), better decoded voice
quality can be obtained ifthe step size is transmitted ex-
plicitly and redundantly encoded for error correction
[19].
3.2.7 Adaptive Gain Encoding
In the syllabic companding example of Figure 3.21 the dynamic range of a signal is
reduced by a factor of 2 (in dBm). Thus, if an uncompanded signal has 36 dB of dy-
120 votcE DtctIzATtoN
namic range, the encoder sees only 18 dB. The 18-dB reduction implies three fewer
bits are needed for dynamic range encoding. In the limit, if the power level of alt input
signals is adjusted to a single value, no bits ofthe encoder need to be allocated to dy-
namic range encoding. A process that adjusts all signals to a standard value is referred
to as automatic gain control (AGC).
AGC is traditionally used on carrier transmission
systems to adjust all received signals to a standard value, thereby removing variations
in propagation
attentuation. AGC cannot be applied to a source voice signal without
allowances for speech pauses when there is no signal present.
Otherwise, idle channel
noise would be amplified to the average level of active voice. Notice that with AGC
there is no residual information in the power
level of the encoded signal as there is in
syllabic companding. To ascertain the original power level, AGC must be augmented
with adaptive gain encoding (AGE),
as indicated in Figure 3.22.
There are two basic modes of operation for gain encoding depending on how gain
factors are measured and to which speech segments the factors are applied. one mode
of operation, as implied in Fi gureS.ZL,involves
measuring the power
level of one seg-
ment of speech and using that information to establish a gain factor for ensuing speech
segments. obviously, this mode of operation relies on
gradually
changing power lev*
els. This mode of operation is sometimes referred to a$
"backward
estimation."
Another mode of operation involves measuring the power lever of a speech seg-
ment and using the gain
factor thus derived to adapt the encoder to the same segment.
This approach, referred to as
"forward
estimation," has the obvious advantage,Ihat the
encoder and decoder use gain factors specifically related to the speech segments from
which they are derived. The disadvantage is that each speech segment mustbe delayed
while the gain
factor is being determined. Although the availability of digital memory
has made the cost of implementing the delay insignificant, the impact of the delay on
echoes and singing in a partially
analog network must be considered. (As
long as the
subscriber loops are analog, the network is partially
analog.)
Adaptive gain
control with explicit kansmission of gain factors is not without
shortcomings. First, when the periodic gain information is inserted into the transmit-
ted bit stream, some means of framing the bit stream into blocks is needed so gain in*
formation can be distinguished from waveform coding. second, periodic
insertion of
gain information disrupts information flow, causing higher transmitter clock rates that
might be inconveniently related to the waveform sample clock. Third, correct recep-
adiustment
Flgure 3.22 Adaptive gain encoding.
3.3 SPEECH HEDUNDANcIES 12I
tion of
gain factors is usually critical to voice quality, indicating a need to redundantly
encode
gain information.
Reference
[20]
describes a modified form of PCM using forward estimation of gain
factors that is referred !o as nearly instantaneously companded PCM' The need for
transmitting speech segments in blocks is not a disadvantage in the application men-
tioned
(mobile telephone) because repetitive bursts with ertor checking are used as a
means of overcoming shortJived multipath fading. This system
provides a bit rate re-
duction of 30Vo with respect to conventional PCM. Another example of the use of
AGE is the subscriber loop multiplexer
(SLM) system developed by Bell Labs
[21].
The SLM system became obsolete when low-cost PCM codecs became available and
the subscriber carrier systems could be integrated into digital end offices
(with SLC
96 and later DLC systems). All of the encoding algorithms described in the following
sections use syllabic companding or AGE in some form to reduce the bit rate'
3.3 SPEECH REDUNDANCIES
Conventional PCM systems encode each sample of an input waveform independently
from all other samples. Thus a PCM system is inherently capable of encoding an ar-
bitrarily random waveform whose maximum-frequency component does not exceed
one-half the sampling rate. Analyses of speech waveforms, however, indicate there is
considerable redundancy from one sample to the next. In fact, as repofted in reference
t101,
the correlation coefficient
(a measure of
predictability) between adjacent $-kHz
samples is
generally 0.85 or higher. Hence the redundancy in conventional PCM
codes suggests significant savings in transmission bandwidths are possible through
more efficient coding techniques. All of the digitization techniques described in the
rest of this chapter are tailored, in one degree or another, to the characteristics of
speech signals with the intent of reducing the bit rate.
In addition to the correlation existing between adjacent samples of a speech wave-
form, several other levels ofredundancy can be exploited to reduce encoded bit rates.
Table 3.3 lists these redundancies. Not included are higher level redundancies related
to context-dependent
interpretations of speech sounds
(phonemes),
words, and sen-
TABLE 3.3 Speech Hedundancles
Time-domain redundancies
1, Nonuni form ampl i tudedi $tri buti ons
2, Sample-to-samplecorrelations
3. Cycle-to-cycle correlatians
(periodicity)
4. Pitch-interval to
pitch-interval
conelations
5. lnactivity factors (speech pauses)
Frequency-domain redundancies
6. Nonuniform long-term spectral densities
7. Sound-specific shorl-term spectral densities
122 vorcE DrcrrzATroN
tence$. These topics are not covered because techniques that analyze speech wave-
forms to extract only information content eliminate subjective qualities
essential to
general
telephony.
3.3.1 Nonuni form Ampl l tude Di etrl butl ons
As mentioned in the introduction to companding, lower amplitude sample values are
more common than higher amplitude sample values. Most low-level samples occur as
a result of speech pau$e$
in a conversation. Beyond this, however, the power levels of
active speech signals also tend to occur at the lower end of the encoding range. The
companding procedures
described in the previous
section provide
slightly inferior
quality (i.e.,
lower signal-to-noise ratios) for small signals compared to large signals.
Thus the average
qualif
of PCM speech could be improved by further shortening
lower level quantization
intervals and increasing upper level quantization
intervals.
The amount of improvement realized by such a technique is minimal and probably
would not
justify
the additional complexities. The mosr beneficial approach to
proc-
essing signal amplitudes in order to reduce encoder bit rate$ involves some form of
adaptive gain control, as discussed earlier.
3.3.2 Sample-to-SampleCorrelatlon
The high correlation factor of 0.85 mentioned in Section 3.3 indicates that any signifi-
cant attempt to reduce transmission rates must exploit the correlation between adja-
cent samples. In fact, at 8-kHz sampling rates, significant correlations also exist for
samples two to three samples apafr. Naturally, samples become even more correlated
if the sampling rate is increased.
The simplest way to exploit sample-to-sample redundancies in speech is to encode
only the differences between adjacent samples. The difference measurements are then
accumulated in a decoder to recover the signal. In essence these systems encode the
slope or derivative ofa signal at the source and recover the signal by integrating at the
destination. Digitization algorithms of this type are discussed at length in later sec-
tions.
3.3.3 Cycle-to-Cycle Correlatlons
Although a speech signal requires the enrire 300-3400-Hz bandwidth provided
by a
telephone channel, at any particular
instant in time certain sounds may be composed
of only a few frequencies within the band. when only a few underlying frequencies
exist in a sound, the waveform exhibits strong correlations over numerous samples
corresponding to several cycles ofan oscillation. The cyclic nature ofa voiced sound
is evident in the time waveform shown in Figure 3.23. Encoders exploiting the cycle-
to-cycle redundancies in speech are markedly more complicated than those concerned
only with removing the redundancy in adjacent samples. In fact, these encoders more
3. 3 SPEECH REDUNDANCI ES 123
Flgure 3.?3 Time wavefbrm of voiced sound.
or less represent a transition from the relatively high rate, natural-sounding waveform
encoders to the relatively low rate, synthetic-sounding vocoders.
3.3.4 Pitch-lnterval-to-Pitch.lnterval Gorrelations
Human speech sounds are often categorized as being
generated in one of two basic
ways. The first category of sounds encompasses
"voiced"
sounds that arise as a result
of vibrations in the vocal cords. Each vibration allows a
puff of air to flow from the
lungs into the vocal hact. The interval between
puffs
of air exciting the vocal tract is
referred to as the pitch interval or, more simply, the rate of excitation is the pitch. Gen-
erally speaking, voiced sounds arise in the
generation of vowels and the latter portions
of some consonants. An example of a time waveform for a voiced sound is shown in
Figure 3.23.
The second category of sounds includes the fricatives, or
"unvoiced"
sounds. Frica-
tives occur as a result of continuous air flowing fiom the lungs and
passing through a
vocal tract constricted at some
point to generate air turbulence
(friction). Unvoiced
sounds correspond to certain consonant$ such asf
j,
s, and x' An example of a time
waveform of an unvoiced sound is shown in Figure 3.?4. Notice that an unvoiced
sound has a much more random waveform than a voiced sound.
As indicated in Figure 3.23, not only does a voiced sound exhibit the cycle-to-cycle
redundancies mentioned in Section 3.3.3, but also the waveform displays a longer
lgtn
rEpetitive
pattern
poffesponding
to the duration of a pitch interval.
$11p
qtre
of
the
glo$t
Afficient
ways of encoding the voiced portions
9f
speech is to encode one
p,te4+Jp*41--1vpv9{pyr+ an"d.use that encoding,as a template for each successive
pitch
i[GwU"in th.e
qame
sound,-fi!9,!r
in1-elvals
typ,i[a]ly-last
gom
5 to 20 mpes f,or
men and
flrqp
2.5 to 10 mseg for.yomen*"Since a.typica] voiced.sognd
f 4pts
fo,r apprq4i5nately
100 msec, there may be as many as 20-40 pitch intervals in a single sound. Although
iitdh
inierval encoding Can
provide significant reductions in bit rates, the pitch is
sometimes very difficult to detect.
(Not
all voiced sounds
produce a readily identifi-
Figure 3.24 Time waveform of unvoiced sound.
124 VOICE DIGITIZATION
able pitch interval as in Figure 3.23.) rf the pitch gets encodecl erroneously, straxge
sounds result.
An interesting aspect of pitch interval encoding is that it provides
a means of speed-
ing up speech while maintaining intelligibility. By deleting some percentage
of pitch
intervals from each sound (phoneme),
the rate of sound generation is effectively in-
creased in a manner analogous to more rapid word formation. The pitch of the sounds
remains unchanged. In conffast, if the rate of reconstruction is merely increased, all
frequencies including the pitch increase proportionately.
Moderate speedups produce
obvious distortion while greater speedups become unintelligible. Devices designed to
simulate faster word formation have demonstrated that we are capable of assimilating
spoken information much faster than we can generate it.
3.3.5 Inactivity Factor$
Analyses of telephone conversations have indicated that a pafiy is typically active
about 4OVa of a call duration. Most inactivity occurs as a result of one person
listening
while the other is talking. Thus a conventional (circuit-switched)
full-duplex connec-
tion is significantly underutilized. Time assignment speech interpolation (TASD
de-
scribed in Chapter I is a technique to improve channel utilization on expensive analog
links. Digital speech interpolation (DSI)
is a term used to refer to a digital circuit coun-
terpart of rASI. DSI involves sensing speech activity, seizing a channel, digitally en-
coding and transmitting the utterances, and releasing the channel at the completion of
each speech segment.
Digital speech interpolation is obviously applicable ro digital speech srorage $ys-
tems where the duration of a pause can be encoded more efficiently than the pause it-
self. In recorded me$sage$, however, the pauses
are normally short since a
"half-duplex"
conversation is not taking place. DSI techniques have been used to ex-
pand the voice channel capacity of digital TDM links. The inputs are standard PCM
signals that are digitally processed
to detect speech activity. The DSI operation is often
combined with speech compression algorithms to implement digital circuit multipli-
cation
(DCM)
equipment. when a 2 : I voice compression algorithm is combined with
a2.5 : I DSI concentration factor, an overall 5 : I circuit expansion is achieved. De-
pending
on the quality
of speech desired, even greater concentration factors are
pos-
sible. The use of such equipment in a network must be carefully managed to ensure
that voiceband data and digital data channels bypass the DCM operations.
3.3.6 Nonunlform Long-Term Spectral Denslties
The time-domain redundancies described in the preceding
sections exhibit charac-
teristics in the frequency domain that can be
judiciously
processed
to reduce the en-
coded bit rate. Frequency-domain redundancies are not independent of the
redundancies in the time domain. Frequency-domain techniques merely offer an alter-
nate approach to analyzing and processing
the redundancies.
3.3 SPEECH REDUNDANCIES 125
Ftequeilcy
(Hrl
Flgure 3.25 Long-term power spectral density of speech.
A totally random or unpredictable signal in the time domain
produces a frequency
speckum that is flat across the bandwidth of interest. Thus a signal that
produces un-
correlated time-domain samples makes maximum use of its bandwidth'
On the other
hand, a nonuniform spectral density repre$ent$ inefficient use of the bandwidth and is
indicative of redundancy in the waveform'
Figure 3.25 shows the long-term spectral density of speech signals averaged across
two
populations: men and women
[22].
Notice that the upper portions of the 3-kHz
bandwidth
passed by the telephone network have significantly reduced
power levels.
The lower
power levels at higher frequencies are a direct consequence of the time-
domain sample-to-sample correlations discussed
previously. Large-amplitude signals
cannot change rapidly because, on average, they are predominantly made up of lower
frequency components.
A frequency-domain approach to more efficient coding involves flattening the
spectrum before encoding the signal. The flattening
process can be accomplished by
passing the signal through a high-pass filter to emphasize the higher frequencies be-
fore sampling.The original waveform is recovered by passing the decoded signal
through a filter with a complementary, low-pass characteristic. An important aspect
of this
process is that a high-pass filter exhibits time-domain characteristics
of a dif-
ferentiator and a low-pass filter has time-domain characteristics analogous to an inte-
grator. Thus the spectrum-flattening
process essentially means the slope of the signal
is encoded at the $ource, and the signal is recovered by integrating at the destination-
the basic
procedure described
previously for sample-to-sample redundancy removal
in the time domain.
In studying Figure 3.25 it is natural to think that the remarkably low levels of signal
energy at the higher frequencies
(2-3.4 kHz) means that more bandwidth is being al-
located to a voice signal than is really necessary. The error in such a conclusion, how-
ever. lies in the distinction between energy content and information content of the
voice frequency speckum. As any beginning computer
prograflrmer soon leams,- the
*I
am assuming that beginning
prograflmers still encounter older languages or file systems with
seven-character name limitations.
a
I
E
- r o
t
T
o
EL
.H
-20
o
-6
E
-30
I ' H ( \ a
lrHtl
rPUq
lrllsds
126
3.4 DIFFERENTIAL PULSE CODE MODULATION 127
meaning of a
program variable can often be retained even though it is shortened by
deleting all of the vowels. In speech the vowels require most of the energy and
pri-
marily occupy the lower
portion of the frequency band. The consonants, on the other
hand, contain most of the information but use much less power
and
generally higher
frequencies. Hence merely reproducing a high percentage of the original speech en-
ergy is an inadequate
goal for a digital speech transmission or storage system.
3.3.7 Short-Term Spectral Densitiee
The speech spectrums shown in Figure 3.25 represent long-term averages of the spec-
hal densities. Over shorter
periods of time the spectral densities vary considerably and
exhibit sound-specific structureri with energy
peaks (resonances)
at some frequencies
and energy valleys at others. The frequencies at which the resonances occur
are called
formant frequencies, or simply formants. Voiced speech sounds typically contain
three to four identifiable formant$. These features of the short-term spectral density are
illustrated in the spectogram ofFigure 3.26. A spectogram is a display ofspeech spec-
tral energy as a function of time and frequency. The horizontal axis represents time,
the vertical axis represents frequency, and the shadings represent energy levels. Thus
the darker portions in Figure 3.26 indicate relatively high energy levels (formants) at
particular instants in time.
Frequency-domain
voice coders provide improved coding efficiencies by encoding
the most important components of the spectrum on a dynamic basis. As the sounds
change, different
portions
(formants)
ofthe frequency band are encoded. The
period
between formant updates is typically 10-20 msec. Instead of using periodic spectrum
measufements, some higher
quality vocoders continuously track gradual changes in
the spectral density at a higher rate. Frequency-domain
vocoders often
provide lower
bit rates than the time-domain coders but typically
produce less natural sounding
speech.
3.4 DIFFERENTIAL PULSE CODE MODULATION
Differential
pulse code modulation
(DPCM) is designed specifically to take advantage
of the sample-to-sample redundancies in a typical speech waveform. Since the range
of sample dffirences is less than the range of individual samples, fewer bits are
needed to encode difference samples. The sampling rate is often the same as for a com-
parable PCM system. Thus the bandlimiting filter in the encoder and the smoothing
filter in the decoder are basically identical to those used in conventional FCM system$.
A conceptual means of generating the difference samples for a DPCM coder is to
store the previous input sample directly in a sample-and-hold circuit and use an analog
subtracter to measure the change. The change in the signal is then
quantized and en-
coded for transmission. The DPCM structure shown in Figure 3.27 is more compli-
cated, however, because the
previous input value is reconstructed by a feedback loop
that integrates the encoded sample differences. In essence, the feedback signal is an
128 vorcE DtctrtzAnoN
gandl i mi ti ng
fi l ter
CBtimate
Aftumulrtor
Flgure 3.27 F\rnctional block diagram of differential
pCM.
estimate of the input signal as obtained by integrating the encoded sample differences.
Thus the feedback signal is obtained in the same manner used to reconstruct the wave-
form in the decoder.
The advantage of the feedback implementation i$ that quantization
errors do not ac-
cumulate indefinitely. If the feedback signal drifts from the input signal, as a result of
an accumulation of quantization
errors, the next encoding of the difference signal
automatically compensate$ for the drift. In a system without feedback the output pro-
duced by a decoder at the other end of the connection might accumulate quantization
errors without bound.
As in PCM sy$tems, the analog-to-digital
conversion process
can be uniform or
companded. some DPCM systems also use adaptive techniques (syllabic
compand-
ing) to adjust the quantization
step size in accordance with the average power
level of
the signal.
(See
reference
[9]
for an overview of various techniques.)
Example 3.4. speech digitization techniques are sometimes measured for quality
by use of an 800-Hz sine wave a$ a representative test signal. Assuming a uniform
PCM system is available to encode the sine wave across a given
dynamic range,
determine how many bits per
sample can be saved by using a uniform DpcM system.
solution. A basic solution can be obtained by determining how much smaller the
dynamic range of the difference signal is in comparison to the dynamic range of the
signal amplitude. Assume the maximum amplitude of the sine wave is A, so that
"r(f)
=A
si n(2n' 800t)
The maximum amplitude of the difference signal can be obtained by differentiating
and multiplying by the time interval between samples:
Prsvioiti
i nput
#
=
e?n)(A0o) cos(2n
'8oor)
3.4 DIFFERENTIAL PULSE CODE MODULATIOru 129
lAx(t)l*u*= A(2nX800)
[,#l
=
0'628A
\ /
The savings in bits
per sample can be determined as
t r \
loc,
| 0628 l=
0.62 tits
\ /
Example 3.4 demonstratetl that a DPCM sy$tem can use
J
bit
per sample less than
a PCM system with the same quality. Typically DPCM systems provide a full l-bit
reduction in codeword size. The larger savings is achieved because, on average'
speech waveforms have a lower slope than an 800-Hz tone
(see Figure 3.25).
3.4.1 DPCM l mpl ementatl ons
Differential PCM enCoders and decoders can be implemented in a variety of ways de-
pending on how the signal
processing functions are
partitioned between analog and
digital circuitry. At one extreme the differencing and integration f'unctions can be im-
plemented with analog circuitry, while at the other extreme all signal
processing can
be implemented digitally using conventional PCM samples as input. Figure 3.28
shows block diagrams of tfuee different implementations with differing amounts of
digital signal
processing.
Figure 3.28a depicts a sy$tem using analog differencing and integration. Analog-
to-digital conversion is
performed on the difference signal, and D/A conversion fot the
feedback loop is immediately
performed on the limited-range difference code. Analog
$ummation and storage in a sample-anrl-hold
(S/H)
circuit is used to provide integration.
Figure 3.28b shows a system that
perfbrms the integration function digitally. In-
stead of immediately converting the difference code back to analog for feedback, the
difference code is summed and stored in a data register to
generate a digital repre-
sentation of the previous input sample. A full-scale D/A converter is then used to pro-
duce the analog feedback signal for differencing. Notice that the D/A converters in
Figure 3.28b must convert the full amplitude range whereas the D/A converters in Fig-
ure 3.28a convert the more limited difference signal.
Figure 3.28c shows a sy$tem where all signal
processing is performed by digital
logic circuits. The A/D converter
produces full-amplitude-range sample codes, which
are compared to digitally
generated approximations of the
previous amplitude code.
Notice that the A./D convefter in this case must encode the entire dynamic range of the
input whereas the A/D convefters in the other two versions operate on only the differ-
ence signals.
Due to the availability of digital signal
processing components, some of which con-
tain intemal A./D converters, digital
processing
(as in Figure 3.28c) is generally the
most effective mean$ of implementing a DPCM algorithm. In fact, most DPCM ap-
plications involve processing speech signals that have already been digitized into
standard 64-kbps PCM formats. Thus the DPCM implementation usually requires no
130 votcE DtGtlzATtoN
analog processing. As an aid in processing
log-PCM signals, some DSp components
provide
internal
p-law
and A-law conversion functions.
The decoders in all three implementations shown in Figure 3.28 are exactly like the
feedback implementations in the conesponding encoder. This reinforces the fact that
the feedback loop generates
an approximation of the input signal (delayed
by one sam-
ple).
Ifno channel elrors occur, the decoder output (before
filtering) is identical to the
feedback signal. Thus the closer the feedback signal matches the input, the closer the
decoder output matches the encoder input.
(t )
Figure 3.2t DPCM implementations: (a)
analog integrarion; (b)
digital integration; (c)
digital
differencing.
Decoder
Encoder
Oecoder
Encoder
s.4 DTFFERENTTAL
pulsE
coDE MoDULATIoN 131
3.4.2 Hlgher Order Prediction
A more
general viewpoint of a DPCM encoder considers it as a special case of a linear
predictor with encoding and transmission of the prediction error. The feedback signal
of a DPCM sy$tem represents first-order prediction of the next sample value, and the
sample difference is a prediction elror. Under this viewpoint the DPCM concept can
be extended to incolporate more than one past sample value into the prediction cir-
cuitry. Thus the additional redundancy available from all
previous samples can be
weighted and summed to
produce a better estimate of the next input sample. With a
better estimate, the range of the prediction error decreases to allow encoding with
fewer bits. For systems with constant predictor coefficients, results have shown that
most of the realizable improvement occurs when using only the last three sample val-
ues. The basic implementation of linear prediction using the last three sample values
is shown in Figure 3.29. For conceptual purposes this implementation shows analog
differencing and integration as in Figure 3.28a. The most effective implementations
use digital memory, multiplication, and addition in a DSP component
in lieu of the
analog processing shown, particularly because most applications involve already dig-
itized
(PCM) signals.
As mentioned in Section 3.4, analysis of differential PCM systems with first-order
predication typically provides a 1-bit-per-sample
reduction in code length relative to
PCM systems with equivalent performance. Extended DPCM sy$tems utilizing third-
order precliction can provide reductions of
f
-2
bits
per sample
[23].
Thus a standard
DPCM system can provide 64-kbps PCM
quality at 56 kbps, and third-order linear
prediction can
provide comparable
quality at 48 kbps. However, some subjective
evaluations have indicated that somewhat higher bit rates are needed to match 64-kbps
PCM quality.
3.4.3 Adaptive Differential PGM
Relatively straightforward implementations of DFCM can provide savings of I
-2
bits
per sample with respect to standard PCM encoding. Even
greater savings can be
Decoder
Encoder
Figure 3.29 Extension of DPCM to third-order prediction.
132 votcE DtctlzATtoN
achieved by adding adaptation logic to the basic DPCM algorithm to create what is
refened to as adaptive differential PcM (ADpcM).
Many forms of ADpcM have
been investigated
and used in various applications. Two of the most prevalent
appli-
cations are voice messaging and DCM equipment for increasing
the number of voice
channels on a Tl line. with respect to the latter application, ITU-T has established a
32-kbps ADPCM standard (Recommendation
G.721)
[24].
This algorithm has been
extensively tested and characterized to not significantly degrade toll quality
voice cir-
cuits when inserted into the internal portions
of the network. Design considerations of
the standard are:
1. Multiple tandem encodings and decodings berween both
pcM
and analog
interfaces
2. End-to-end signal quality
for voice, voiceband data, and facsimile
3. Effects ofrandom and bursty channel errors
4. Performance on analog signals degraded by loss, noise, amplitude distortion,
phase distortion, and harmonic di$tortion
5. Easy transcoding with p-law
and Alaw
pCM
The 32-kbps rate implies a 2 : I savings in channel bandwidth with respect to standard
PCM. A significant impairment inffoduced by implemenrarions
of the ADpcM stand-
ard is the comrption of modem signals carrying data rates greater
than 49fi) bps
[24].
Voiceband data at rates of 4800 bps and below are adequately supported.
The G.721 ADPCM algorithm is conceptually
similar to rhat shown in Figure 3.29
but more sophisticated in that it uses an eighth-order predictor,
adaptive quantization,
and adaptive prediction.
Furthermore, the algorithm is designed to recognize the dif-
ference between voice or data signals and use a fast quantizer
adaptation mode for
voice and a slow adaptation mode for data.
*q
-*o*****__=-
#
PCM
---*
ADPCM
4
v,
3 3
2
1
4
$ r
2
t
2 3 4 5 I 7 I
Numbsr of Encodlner
(el
Figure 3.30 Average MOS versus number of encodings for
microphone; (b)
carbon microphone.
2 3 4 E
g
7 I
Numbsr of Encodl4r
tbl
PCM and ADPCM:
(a)
linear
g.s
DELrA MoDULATtoN 133
subjective evaluation
[25]
of the G.721 algorithm using the mean opinion score
(MOS)
method of evaluating speech
quality is shown in Figure 3.30.
The MOS
method uses trained listeners to evaluate the speech quality on a scale of I : 5. Figure
3.30 shows the average scores of 32-kbps ADPCM and 64-kbps PCM as a function of
the number of analog tandem encodings. The speech
quality associated with multiple
digital transcodings
(to
and from PCM) does not degrade beyond the first transcoding
as long as there are no digital transmission impairments such as channel errors or slips
(Chapter 7). The
performance of ADPCM
(or any other coder) used to be an important
consideration when the
public network was composed of mixtures of analog and digi-
tal equipment. Because the internal
porxions of the network are all digital,
performance
with multiple analog conversions is no longer a consideration.
(Multiple PCM-to-
ADPCM conversions can still occur when multiple DCM systems are encountered
within a network.)
Because ADFCM at 32 kbps provides good quality at a moderate cost and
power
consumption, it is used in several cordless telephone or low-tier- digital cellular sys-
tems:
Personal Access Communications System
(PACS) (North America)
Second Generation Cordless Telephones
(CT2) (Europe)
Digital European Cordless Telephones
(DECT) (Europe)
Personal Handyphone System
(PHS) (Japan)
3.5 DELTA MODULATION
Delta modulation
(DM) is another digitization technique that specifically exploits the
sample-to-sample redundancy in a speech waveform. In fact, DM can be considered
as a special case of DPCM using only I bit per sample of the difference signal. The
single bit specifies merely the
polarity of the difference sample and thereby indicates
whither the signal has increased or decreased since the last sample. An approximation
to the input waveform is constructed in the feedback
path by stepping up one
quanti-
zation level when the difference is positive
("one")
and stepping down when the dif-
ference is negative
("zero"). In this way the input is encoded a$ a sequence of'trps"
and
"downs"
in amannerresembling a staircase. Figure 3.31 shows aDM approxima-
tion of a typical wavefonn. Notice that the feedback signal continues to step in one
direction until it crosses the input, at which time the feedback step reverses direction
until the input is crossed again. Thus, when tracking the input signal, the DM output
"bounces"
back and forth across the input waveform, allowing the input to be accu-
rately reconstructed by a smoothing filter'
Since each encoded sample contains a relatively small amount of information
(1
bit), DM $ystems require a higher sampling rate than PCM or multibit DPCM systems.
In fact, the sampling rate is necessarily much higher than the minimum
(Nyquist) sam-
*A
low-tier cellular system utilizes simple, Iow-power mobile units, and small cells and only suppofts
pedesfrian speeds,
134 VoIcE DIGITIZATIoN
Figure 3.31 Waveform encoding by delta modularion,
pling
rate of twice the bandwidth. From another viewpoint,
"oversampling"
is needecl
to achieve better prediction
from one sample to the next.
The main attraction of DM is its simplicity. Figure 3.32 shows a basic implemen-
tation of a DM encoder and decoder. Notice that the A/D conversion function is pro-
vided by a simple comparator. A positive-difference
voltage produces
a l, and a
negative-difference
voltage produces
a 0. Correspondingly, the D/A function in the
feedback path,
and in the decoder, is provided
by a two-polarity pulse generator.
In
the simplest form the integrator can consist of nothing more than a capacitor to accu-
mulate the charge from the pulse generator.
In addition to these obvious implementation simplicities, a delta modulator also al-
lows the use of relatively simple filters for bandlimiting the input and smoothing the
output
[26].
As discussed in section 3.1, rhe speckum produced
by a sampling process
consist$ of replicas of the sampled spectrum centered at multiples of the sampling fre-
quency.
The relatively high sampling rate of a delta modulator produces
a wider sepa-
ration of these spectrums, and, hence, foldover distortion is prevented
with less
stringent roll-off requirements for the input filter.
3.5.1 Slope Overload
The conceptual operation of a delta modulator shown in Figure 3.3 I indicates that the
encoded waveform is never much more than a step size away from the input signal.
sometimes a delta modulator,
or any differential system such as DpcM, may not be
able to keep up with rapid changes in the input signal and thus fall more than a step
Eruodr
Figure 3.32 Delta modulation encoder and decoder.
Pul *
gf,ndtbr
3.5 DELTA MODULATION 135
size behind. When this happens, the delta modulator is said to be experiencing
"slope
overload."
A slope overload condition is shown in Figure 3'33.
Basically, slope overload occurs when the rate of change of the input exceeds the
maximum rate of change that can be
generated by the feedback loop' Since
the maxi-
mum rate of change in the feedback
loop is merely the step size times the sampling
rate, a slope overload condition occurs if
( 3. 13)
where x(t)
=
input signal
q
=
steP size
f;
=
sampling frequencY
The design of a DM
(or
DPCM) necessarily involves
a trade-off between two types
of distortion: the more or less random
quantization noise, sometimes referred to as
granular noise, and the slope overload
noise. As indicated in Figure 3-33,
granular
noise is a
predominant consideration for slowly changing signals, whereas slope over-
load is dominant during rapidly changing signals. Obviously,
granular noise is small
if step sizes are small, but small step sizes increase the likelihood of slope overload'
The optimum DM step size in terms of minimizing the total of granular and slope over-
load noise has been considered Uy Abate
[271.
The perceptual effects of slope overload on the
quality of a speech signal are sig-
nificanity difierent from the
perceptual effects
produced by granular noise' As indi-
cated in iigure 3.33, the slope overload noise reaches its
peaksjustbefore the encoded
signal reaches its
peaks. Hence, slope overload noise has sfiong components
identical
irifrequency and approximately
in phase with a major component of the input' Dis-
tortion that is correlated in this manner to the speech signal is effectively
"masked'by
the speech energy and therefore is less noticeable than
"uncorrelated"
distortion. In
fact, overload noise is much less objectionable
to a listener than random or
granular
noise at an equivalent
power level
[28].
Hence, from the point of view of
perceived
speech
quality, the optimum mix of
granular and slope overload noise is difficult to
determine.
l#l
'*
Figure 3.33 Slope overload and granular noise of delta modulation system.
136 vorcE DtcllzAloN
slope overload is not a limitation of
just
a DM system, but an inherent problem
with any system, such as DPCM in general,
that encodes the difference in a signal
fiom one sample to the next. A difference
system encodes the slope of the input with
a finite number ofbits and hence a finite range. Ifthe slope exceeds that range, slope
overload occurs. In contrast,
a conventional
pcM
system is not limited by the rate of
change of the input, only by the maximum encodable amplitude. Notice that a differ-
ential system can encode signals with arbitrarjly large amplitudes, as long as the large
amplitudes are attained gradually.
Many versions of DM for voice encoding were investigated
in the 1960s and 1970s
[29,
30]. All of these efforts focused on ways of implementing
adaptive delta modu-
lation (ADM)
to improve the performance
at a given
bit rate. The intense interest at
that time was related to
the simplicity, good
tolerance of channel errors, and relatively
low cost implementation.
The cost factor is no longer relevant because even relatively
complicated
coding algorithms now have insignificant
costs compared to mosr system
costs. ADM is still used in some old
pBXs,
in some military secure voice radio sys-
tems, and as a mealls of encoding the residual error signal of some predictive
coders
described in the following sections.
3.6 ADAPTIVE
PREDICTIVE
CODING
The differential sysrems described in the previous
sections (DpcM,
ADpcM, ADM)
operate with lower data rates than
pcM
systems because they encode a difference sig-
nal that has lower average power
than the raw input signal. The ratio of the input signal
power
to the power
of the difference
signal is referred to as the prediction
gain.
Simple
DPCM system$ (firsr-order
predictors)
provide
about 5 dB of prediction ga1n.
ADPCM provides greater
levels of prediction gain depending
on the sophistication
of
the adaptation logic and the number of past
samples used to predict
the next sample.
The prediction gain
of ADpcM is ultimately limited by the fact that only a few
fast
samples are used to predict
the input and the adaptation logic only adapts the quan-
tizer, not the prediction
weighting coefficients (the
c's in Figure 3.zg).
Adaptive predictive
coding (Apc)
t3l,
3zl provides
greater
levels of prediction
gain
by adapting the prediction
coefficients to individual speech segments and, in
most cases, using higher orders of prediction (e.g.,
up to 12). If the coefficients
are de_
termined from past history and used to predict
subsequent
speech segments (backward
estimation), l3 dB of prediction
gain
is possible
[10].
If speectr segments are delayed
so predictor
coefficients can be used on the same speech segments from which they
were derived (forward
esrimation), 20 dB of prediction gain
is possible
[33].
A block diagram of a basic Apc encoder/decoder
is shown in Figure 3.34. The in-
put
to the encoder and the output from the decoder are assumed to be uniform
pcM,
most likely representing
conver$ions from and to log
pcM.
The transmitted data
$tream is necessarily composed of blocks containing
three types of information: (l)
the encoded difference signal (residual), (z)
a gain
factor, and
(3)
rhe predicror
coef-
ficients. The most significant difference between this coder and a'DpcM or ADpcM
coder involves the periodic
determination and transmission of the predictor
coeffi-
=
(J
G
E
o
.E
f
tr
o
:E
t
g
c
F
o
E
L
- E
s t
i l ' E
& G
o
T
E
(J
r
E
o
=
F
3
t-'
at
E
I
o
F
E
o
iJ
4)
ts
9J
q)
F
E
4
+
ca?
{)
o!
FFT
.E
,9
H
E
o
. g
E
G
137
138 votcE DtctlzATtoN
cients' Notice that the integrated residual
signal at point
A is identical to the input sig-
nal delayed by one sample (except
for residual quantization
error). Thus the corre-
sponding point in the decoder is the reconstructed
pCM
output.
Explicit transmission of a
gain
factor, as opposed to deriving a
gain
factor from the
transmitted residual, is useful in this application because a block-structured
transmis-
sion format is required for transmitting the predictor
coefficients anyway. Residual
encoding may use any of the waveform algorithms described previously.
Arbiharily
accurate reconstruction
of the input can be achieved if enough bits are used in encod-
ing the residual. of course, the purpose
of the adaptive prediction
is to achieve a low
data rate. single-bit PcM thar conveys only the polarity
ofthe residual provides
ade-
quate pedormance
fbr low-data-rate
applications
[33].
Notice that because of the feed-
back path,
single-bit encoding ofthe residual
essentially produces
a delta modulator
with very sophisticated feedback logic that is periodically
changed to match the shape
and energy level of corresponding (forward-estimated)
speech segments.
A large variety of algorithms have been investigated
for calculating
the predictor
coefficients,
most of which involve extensive
computation. If the apflication is for
voice response systems, the computations
do not have to occur in rial time and are
therefore sometimes
done on a large mainframe
computer. Real-time encoding re-
stricts the algorithm to one that can be realized with a DSp or special-purpose
IC de_
signed to implement a specific coefficient
determining algorithm. A linear predictive
coding
(LPC)
algorithm as described in Section 3.8.3 is a common algorithm because
it provides good performance
and is available in custom parts.
The u.s. defense de-
parHnent
adopted an APC algorithm using a fourth-order LPC coefficient analysis as
a government
standard for secure voice transmission
at 9.6 kbps
[34].
This system
transmits blocks of 240 bits containing I 80 one-bit samples of the residual, 54 bits for
parameters
and gain factors,
5 bits for error corrections
of critical most significant
bits,
and I framing bit.
3.7 SUBBAND CODING
A subband coder is one form of coder using a frequency-domain
analysis of the input
signal instead of a time-domain
analysis as in previously
described .od"r*. As shown
in Figure 3.35, the coder first divides the input spectrum into separate bands using a
bank ofbandpass
filters. The signal passing
through each ofthe rilatively nanow sub-
bands is individually
encoded wirh separate adaprive ApcM,
pcu,
or enpcM en-
coders. After each subband is encoded, the individual
bit streams are multiplexed
for
transmission
to the decoder, where they are demultiplexed,
decoded, and combined to
recon$truct the input.
separately encoding each subband is advantageous
for several reasons. First, by us-
ing separate adaptation
for each band, the quantization
step sizes can be adjusted ac-
cording to the energy level in each band. Those bands with relatively
high energy
levels can be encoded with
relatively coarse quantization.
In this mannei the spectrum
of the quantization
noise is matched
to the short-term
$pectrum of the signal. This
propefly
is very desirable in a perceptual
sense because it allows the speech signal to
3.7 SUBBAND CODING
139
BandpErB
fi l ters
AdaPtive
PCM
Figure 3.35 Subband coder.
mask the
quantization noise.
(The human ear
perceives speech by measuring the short-
term energy level of individual frequency bands. Hence, relatively low noise in a band
with no speech energy is perceptually more significant than
greater noise in a band
with significant speech energY.)
A second advantage
of subband coding is that the bit rate
(quality) assigned to each
band can be optimized
according to the
Perceptual
importance
of each
individual
band. In particular, a relatively
large number of bits
per sample can be used for low
frequencies where it is important to
preserve the
pitch and formant Structure
of voiced
,ound*. At higher frequencies, however,
fewer bits
per sample can be used because
noiselike fricatives do not require comparable
quality in reproduction'
As reported in reference
[10],
subband coders
provide significant bit
rate reduc-
tions compared to the more coilrmon and simpler coding algorithms:
adaptive delta
modulation and adaptive differential
PCM. Specifically,
subband coders at 16 kbps
are reported to be
perceptually equivalent
to ADFCM coders at ?2 kbps. A subband
coder at 9.6 kbps is reported to be eguivalent to an ADM coder at 19.5 kpbs. Extensive
description and
performance analysis
of subband coding are available
in reference
t351.
A particularly significant example
of subband coding is the ITU-T recommenda-
nonl.7}Zfbr wideband speech coding
t36,
371. This standard
provides for encoding
7-kHz speech bandwidth with 64 kbps. Thus it provides a means of significantly
im-
proving voice fidelity when end-to-end
digital channels
are available.
Applications
that can benefit most from the higher fidelity are teleconferencing
and speakerphones'
Because it is not intended for internal equipment of the network, it does
not have
to
proces$ voiceband data signals or support tandem analog conversions.
Tran-
scodings to and from uniform PCM are required in support of conference bridge
affangements.
As shown in Figure 3.36
[36],
the G.722 algorithm
divides the input speech band-
width from 50 to 7000 Hz into two approximately
equal-sized
subbands'
Both sub-
bands are encoded
with ADPCM: the Iower subband at 48 kbps and the upper subband
140 votcE DlctTlzATtoN
l 4 bl t|
l 6k Hr
I
I 04 kbps
oulput
Figure 3.36 G.7227-kIIz
audio codec, (From
reference
t361,
p. 10.)
at 16 kbps. The upper band does not require as high a data rate because it is not as im-
portant perceptually
and has significantly
less energy. The algorithm permits
the
lower band to be encoded at 40 or 32 kbps, which allows
g
or 16 kbps for an auxiliary
data channel in teleconferencing
applications
or for the use of56-kbps channels when
64 kbps is not available. Figure 3.37 shows the Mos performance
olthe G.222 atgo-
rithm for speech and music at three basic rates
[3g].
Excellcnt 5
Good 4
Fai r 3
Poor 2
Bed I
t
6
!
o
,s
E
{
Si gnsl B8ndwi dt h=7kHz
Sprech
Muri c
Hsiearch
Gorl for
7-kHu Spech
o 24& l$ps PCM
r G722:64-. kbps $B-ADPCM
4$
Bit RErB (kbps)
Figure3.37
AverageMosversusbitrateotG.lzzk-kHzaudiocodec.(Fromreference[3g],
p.
15.)
Lowr ilbbEnd
ADPCM dftoder
(3
vrrirntl
Mode indic8tion
3.8 vocoDEHS 141
3.8 VOCODERS
For the most
part, the encocling/decoding
algorithms described
previously have been
concerned
primarily with reproducing the input waveform as accurately as
possible'
Thus they u**u-* little or no knowledge of the nature of the signal they
process and
are basically applicable to any signal occurring
in a voice channel. Exceptions occur
when subband coding and adaptive
predictive coding are designed
for particularly low
bit rates
(20 kbps or iess). At these bit rates the encoders have been closely tailored
to
the statisiics of a speech signal and cannot
provide comparable
quality for other sig-
nals. Differential
systems, such as DPCM and DM, also exhibit a certain
amount of
speech-specific
properties by virtue of their high-frequency
encoding
deficiencies
(slope overload).
The digitization
procedures described
in this section very specifically
encode
speech signats and speech signals only. For this reason these techniques
are referred
to collectively as
"vocOders,"
an acronym for vOiCe coderS. Since these teChniques
are
designed specificatly
for voice signals, they are not applicable
to pottions ofthe
public
telephone
network in which other analog signals
(such as modem signals) must be ac-
commodated.-
The basic
goal of a vocoder is to encode only the
perceptually important aspects of
speech with fewer bits than the more
general waveform encoders. Thus they can be
used in limited-bandwidth
applications
where the other techniques cannot.
Some of the main applications
for vocoders are recorded
(e.g.,
"wrong
number")
message$, encrypted voice transmission over niuT0wband
HF radio, digital cellular te-
Iephony, digital circuit multiplication,
computer
output,
games, and telephony over
tnl fnternet. A particularly interesting, early use of an LPC vocoder
provided multiple
voice channels over a single voice frequency leased line' Using a well-conditioned
leased line to obtain a 9600--bps
circuit, four 2400-bps voice signals were
time division
multiplexed
into a single line
[39].
This is one of the first cases
(early 1980) where
digitization was used to actually decrease the bandwidth
of a voice signal' This system
pr*ovided intelligible voice, but the overall
quality was below telephone
standards'
This
particular system became obsolete when long-distance leased-line costs dropped
to the
point that the sacrifice
in voice
quality was unjustified'
This section describes
three of the most basic vocoding techniques: the channel vo-
coder, the formant vocoder, and the aforementioned
LPC. Many other forms and vari-
ations of vocoders have been
proposed and studied. For a discussion of
some of the
other techniques
and an extensive
bibliography on the subject,
see reference
[10]'
Most commercial
applications for vocoders have concentrated
on adaptations of the
LPC algorithm,
particularly for digital cellular and voice over data networks'
Phtase lnsensitlvltq
A fundamental
requirement for maintaining
good speech
quality involves
preserving
the shot"t-term
power specffum of the signal. The
phase relationship
between individ-
*Vocoders
can be insefted into internal
portions of a network if the interfaces are equipped to detect the
presencc of voiceband modem or fax signals and
process them accordingly'
142 votcE DtGtlzATtoN
t i n { 611
T
cl n {?. drl
Ein
l0rrl
+ -
l i n ( ? o r + f l
Figure 3.38 Effect of phase
shift in the superposition
of two tones.
ual frequency components is perceptually
much less important.
one of the best exam-
ples of the ear's insensitivity
to phase
is demonstrated
when two notes are played
si-
multaneously,
or nearly simultaneously,
on a piano.
The composite
Jound, as
perceived
by a listener, is seemingly no different if one note is struct< stightly later than
the other. In essence, the ear senses how much energy is present
at various frequencies
in the speech spectrum
but does not sen$e the phase
relationship
between individual
frequency components.
The effect of a
phase
shift in one component of a composite signal is shown in Fig-
ure 3.38. The first composite
waveform
is produced
when two individual
frequenry
component$
have identical
starting phases.
The second composite waveform
occurs
when the two frequency terms have starting phases
shifted by-90o with respect to each
other. Notice that the composite
waveforms
are markedly different eventhough
the
difference is imperceptible to the ear. For these reasons the time waveform produced
by a vocoder generally
bears little resemblance
to the original input waveform. In_
stead, the emphasis of a vocoder is in reproducing the short-term
fower
specrum of
the input.
3.8.1 Channel Vocoder
channel vocoders were firsr developed
in l92g by Homer Dudley
[40].
Dudley's
original implementation
compressed
speech waveforms into an *utog signal with a
total bandwidth of about
300 Hz. Based on the original concept,
digital channel vo_
coders have been developed
operating in the range of l_? kbps.
A major part
of the encoding process
of a channel vocoder involves determining
the short-term
signal spectrum as a function
of time. As indicated in Figure 3.3g, ;
bank ofbandpass
filters is used to separate the speech energy into subbands that are
full wave rectified
and filtered to determine
relative power
levels. The individual
power
levels are encoded and transmitted
to the destination.
Notice that this much of
a channel vocoder is very similar to the subband coder discussed previously.
A sub_
band coder, however, typically uses wider bandpass filters, whicir necessitate sam-
pling
the subband waveforms
more often (determining
a waveform instead of
just
a
power
level)' Since a subband coder encodes waveforms, it also includes phase
infor-
mation that is ignored
by a channel vocoder.
In addition to measuring
the signar spectrum, modem channel vocoders also deter-
mine the nature of speech excitation (voice
or unvoiced) and the pirch frequency
of
s
FE
L.:
o
E
E
U
(t\
t t
rti
H
00
E
o
6
.9
::
E
E
g
E E
5 g
F F
o
143
144 votcE DtctlzAloN
voiced sounds. The excitation
measurements are used to synthesize the speech signal
in the decoder by passing
an appropriately selected source signal through a frequency-
domain model of the vocal
tract transfer function. Voiced excitation is simulated by a
pulse generator
using a repetition rate equal to the measured pitch period.
unvoiced
excitation is simulated by a noise generator.
Owing to the synthesized nature of the
excitation, this form of a vocoder is sometimes referred to as a
pitch-excited
vocoder.
As indicated in Figure 3.39, a decoder implements a vocal tract transfer function
as a bank of bandpass filters whose input power
levels are determined by respective
subband power
levels in the encoder. Thus outputs of each bandpass filter in the cle-
coder conespond to outputs ofrespective bandpass filters in the encoder. superposing
the individual bands re-creates, in a spectral sense, the original signal.
Maly variations in the basic channel vocoder have been developed involving the
nature of the excitation and the means of encoding the power
levels. Recent advances
in digital technology have introduced the use of digita-l signal processing
to determine
the input spectrum by way of Fourier transform algorithms in lieu of the bank of ana-
log filters. All forms of vocoders that measure the power
spectral density are some-
times referred to collectively as spectrum channel vocoders to distinguish them from
time-domain vocoders such as the LpC described later.
The most difficult aspect
of most vocoder realizations involves determining the
pitch
of voiced sounds. Futthermore,
certain souncls are not clearly classifiable as
purely
voiced or purely unvoiced. Thus a desirable extension ofthe basic vocoder in-
volves more accurate characterization
of the excitation. Without accurate excitation
information, vocoder output quality is quite poor
and often dependent on both the
speaker and the particular
sounds being spoken. Some of the more advanced channel
vocoders have produced
highly intelligible,
although somewhat synthetic sounding,
speech at 2400 bps
[231.
3.8.2 Formant Vocoder
As indicated in the spectogram of Figure 3.26, the short-term spectral density of
speech is rarely distributed across rhe entire voice band (200-3400
Hz). Insread,
speech energy tends to be concentrated
at three or four peaks called formants. A for-
mant vocoder determines the location and amplitude of these spectral peaks
and trans-
mits this information
instead of the entire $pectrum envelope. Thus a formant vocoder
produces
lower bit rates by encoding only, the most significant short-term componenrs
in the speech spectrum.
The most important requirement for achieving useful speech from a formant vo-
coder involves
accurately tracking changes in the formants. once this is accom-
plished,
a formant vocoder car provide
intelligible
speech ar less than 1000 bps
tl0l.
3.8.3 Linsar Predlctive Codlng
A linear predictive
coder is a
popular
vocoder that extracts perceptually
significant
features of speech directly from a time waveform rather than from frequency specrra,
as does a channel vocoder
and formant vocoder. Fundamentally,
Lpc analyzes a
3.8 vocoDFRS 145
Analvsis
Synthesis
Figure 3.40 Basic model of LPC analysis and synthesis relationship'
speech waveform to
produce a time-varying model of the vocal tract excitation and
transfer function. A synthesizer in the receiving terminal re-creates the speech by pass-
ing the specified excitation through a mathematical model of the vocal hact' By
peri-
odically updating the
parameters of the model and the specification of the excitation,
the synthesizer adapts to changes in either. During any one specification
interval, how-
ever, the vocal tract is assumed to represent a linear time-invariant
process.
The analysis-synthesis
relationship between the encoder/decoder is shown in Fig-
ure 3.40. The analysis
process determines the coefficients of the A(z) matrix to mini-
mize the error e(n) with a given set of speech samples x(n)' If the decoder
knows both
A(z) and e(n), it can re-create
(synthesize) the input samples x(n). In the most basic
form of LPC e(n) is not transmitted to the decoder. Instead, e(rr) is indirectly specifled
as the excitation of a vocal hact model. Notice that e(n) is the excitation of the syn-
thesizer in Figure 3.40.
A block diagram of the basic model for speech
generation is shown in Figure 3.41'
which is also a model of the most basic fbrm of LPC decoder/synthesizer'
The equa-
tion of the vocal tract model shown in Figure 3'44 is defined as:
Excitation model
Vocal tract model
Pirch
period
,l
@
"Voiced"
{
Figure 3.41 Speech
generation model of linear predictive coding.
146 votcE DtctlzAloN
y(n)=Eooy@*k)+Gx(n)
(3. 14)
FI
where)(n) =
zth output sample
er
=
lcth predictor
coefficient
fi
=
gain
factor
x(n)
*
input at sample time n
p
=
order of the model
Notice that the speech output in Equation 3.14 is represented
as the present input value
plus a linear combination of the previous p
outputs of the vocal tract. The model is
adaptive in that the encoder periodically
determines a new set of predictor
coefficients
corresponding
to successive speech segments. Basic LpC does not measure and en-
code difference waveforms or error signals. Instead, the error signals are minimized
in a mean-squated sense when the predictor
coefficients are determined. The ability
to avoid encoding the predictor
errors comes from the fact that LPC uses prediction
parameters
based on the actual input segments to which the parameters
are applied
(forward
estimation). In confiast, most of the predictive
coders mentioned previously
base their prediction
on past
measurements only (backward
estimation). Rapid
changes in the vocal tract or excitation cause models based on
past
mea$urements to
be less accurate. The negative aspect of forward estimation is the delay it inserts into
the encoding process,
which, depending on the application, may or may not be a con-
sideration.
The information
that a LPC encoder/analyzer
determines and tran$mits to the de-
coder/synthesizer consists of
l. Nature ofexcitation (voiced
orunvoiced)
2. Pitch period (for
voiced excitation)
3. Gain factor
4. Predictor coefficients (parameters
of vocal tract model)
The nature of the excitation
is determined, as in other vocoders, by determining
whether strong periodic
components exist in the waveform.
pitch
is determined by
measuring periodicity
when it exists. In addition to measuring pitch
with techniques
similar to those used by other vocoders, an Lpc encoder/analyzer
has particular prop-
erties that aid in
pitch
determination
[41].
The predictor
coefficients can be determined using one of several different compu-
tational procedures.
All procedures
use actual waveform samples as the desired out-
puts
of the synthesizer. using these sample values, a set of p linear equations in p
unknown coefficients is produced.
Thus the coefficients
are determined by inverting
a p-by-p
matrix. since the order of p may vary from 6 to rz, depending on the *peecir
quality
desired, straightforward
matrix inversion represents a significant amount of
computation. Depending
on specific assumptions made in the model, however, the
3.8 vocoDEHS 147
matrices have special
propelties that greatly simpliff the solution of the equations
t4l I.
Although linear
predictive coders
process time-domain representations of speech,
their operation is such that they
provide good estimates of the
peaks of the speech
spectrum. Furthermore, an LPC is capable of effectively tracking
gradual changes in
the spectrum envelope. The overall result is that LPCs
provide more natural sounding
speech than the
purely frequency-domain-based
vocoders
[23].
Most LPC research
has concentrated on encoding speech in the range of 1.2-2.4 kbps'
The u.s. Departrnent of Defense established a 2400-bps, I0th-order LPC
(LPC-
l0) as a standald for secure communications over dial-up telephone lines
[42'
431.
This system transmits blocks of 54 bits every 22.5 msec. The 54 bits are allocated as
7 bits for pitch, 5 bits for energy level
(gain factor),41 bits for 10 coefficients
(voiced
excitation), and I bit for framing.
The performance of the coder with respect to natural speech is provided in Table
3.4. A diagnostic rhyme test
(DRT)
tul4l
is a means of determining the intelligibility
of a speech system as a
percentage of correct word recognitions from a standardized
list of pairs of rhyming words
(e.g., goat vs. coat or thick vs. tick). The DRT-with-
noise test measures the word recognition
percentage when background noise appro-
priate to the application is added to the test words before encoding. The diagnostic
acceptabiliry measure
(DAM)
[45]
rates intelligibility and subjective acceptability us-
ing procedures that eliminate much ofthe dependence on personal preferences ofthe
listeners. The scores are normalized from 0 to 100'
3.8.4 Enhanced-Excitation
Llnear Predlctive Codlng
The basic LPC afgorithm described in the
previous section synthesizes speech in the
decoder using a very simple model for the excitation that requires only about lOTo of
the aggregate data rate. Because voiced signals are modeled by a simple
pulse gener-
ator. an LPC coder is sometimes referred to as a pulse-excited linear
prediction
(PELP)
coder. The simplicity of the model inevitably
produces synthetic sounding
speech. To overcome this shottcoming, numerous techniques have been developed to
enhance the excitation. Three enhanced-excitation algorithms are multipulse LPC
(MPLPC),
residual excited linear
prediction
(RELP), mixed-excitation
linear
predic-
tion
(MELP), and various versions of code-excited linear
prediction
(CELP). Because
the enhanced-excitation
LPC algorithms involve encoding a residual
prediction error
TABLE 3,4 Performance Comparlson of Natural SPeech and LPC'10
Natural Speech
(3.6
kHz)
LPC-10
DHT
DRT with noise
DAM
95o/"
92"/"
65
90o/o
82"/"
48
148 VoIcE DIGITIZATIoN
in some form or another, these algorithms are often referred to as hybrid codecs; a
combination of vocoding and waveform coding.
Multlpulse LPC
As shown in Figure 3.42, where it can be compared with a conventional Lpc system,
MPLPC
[46]
is a conceptually simple exrension of conventional Lpc. where Lpc
uses the residual to determine the pitch period
of voiced signals, MpLpC uses the re-
sidual to determine a sequence of
pulses
to more closely match the residual. In the sim-
plest
case MPLPC uses a fixed number of pulses
in a pulse
kain and determines only
the starting phase
of the train and the amplitudes and polarities
of each pulse within
the train. A significanr side benefit of MPLPC is that it does not have ro identify
whether a speech segment is voiced or not and consequently determine the pitch pe-
riod of voiced signals. Instead, the multipulse determination in each analysis segment
automatically adapts to the nature of the actual excitation. MPLPC at 96fi) bps is used in
AT&T's l4A Arurouncement system for reco'rded messages to telephone usen
[46].
MPLPC is also the algorithm chosen for the skyphone Aeronautical Telephone
service using 9.6-kbps channels on the INMARSAT maritime $atellite
[47].
Another
application of an MPLPC algorithm is the
panEuropean
digital mobile radio system
(GsM)
[48]
that allocates a l3-kbps data rare to voice encoding. The speech codec thar
has been standardized by CEPT for this application is referred to as regular*pulse ex-
citation LPC with long-term prediction (RpE-LTp)
[4g].
This sysrem transmits 260
bit blocks consisting of 72 bits of prediction parameters
and 188 bits of excitation en-
coding. Because the codec does not $upport voiceband data rates at 1200 bps and
above, the system has provisions
for data bypass ofthe codec.
Mlxed-Excltation
LPC
As the name implies, mixed-excitation Lpc (MELP)
[50]
uses a more sophisticated
model for the excitation than either LPC or Mpl-pc. An MELp algorithm has been
ENCODER DECODEH
Volcsd/unvolced
ffil*
4*q'Fs
Cocfficirntr
Input
I
Vocsl
#n'l
-'5;*
Figure 3.42 Multipulse linear predictive
coding. comparison of Lpc
(a)
and Mpl-pc
(D)
encoder/decoder.
3.8 vocoDEFs 149
selected by the U.S.
govemment for $ecure voice applications at 2.4 kbps
[51].
Devel-
opmsnt of this algorithm has paid particular attention to background noise
(as
might
arise in a tank). Enhanced excitation features of this version of LPC are;
1. Mixed
pulse and noise excitation
2. Periodic or nonperiodic
pulse generation
3. Adaptive spectral enhancement
(to improve formant reproduction)
4. Pulse dispersion
(spreading the impulses across more of a pirch interval)
5. Fourier analysis of residual
(to
improve
pitch determination)
Residual Exclhd LPC
The APC algorithm described in Section 3.6 transmits the encoded difference signal
(prediction error or residual) determined in the encoder so the decoder can integrate it
to recover the original input. Notice that the decoded residual waveform in the APC de-
coderof Figure 3.34 acts as the excitation of the
predictionloop. If theprediction
loopuses
an LPC formulation, APC essentially becomes residual-waveform excited LPC.
The formal term residual excited IIC
(RELP) refers to a structure that is identical
to the APC block diagram shown in Figure 3.34 but differs in the manner in which the
residual is encoded. An RELP encoder does not encode the residual directly but pre-
processes it to allow transmission at a lower data rate. The fundamental
premise of the
preprocessing is that the residual consists of a fundamental frequency and multiples
(harmonics)
of the fundamental. Thus, an RELP encoder encodes only the fundamen-
tal. The decoder reconstructs the residual
(in
a frequency-domain
sense) by decoding
the fundamental and adding in the harmonics. In the sense that the process of extract*
ing and encoding the fundamental is
"residual
encoding" and that of decoding and
adding harmonics is
"residual
decoding," the diagram in Figure 3.34 serves as a basic
diagram of RELP.
As commonly implemented, an RELP encoder
[5?]
extracts the fundamental by
low-pass filtering the residual and down sampling to reduce the sampling frequency
to the Nyquist rate. As an example, if a
(DSP-based) low-pass filter has a cutoff of 8ff)
Hz, only every fifth sample of the filtered residual is needed to meet the 1600-Hz
Nyquist rate. In this manner, an RELP decoder is excited by samples occurring at a
1600-Hz rate-approximately the same rate as in MPLPC. Thus the residual decoder
in an RELP receiver is similar to the
pulse generator of an MPLPC synthesizer. In fact,
the regular
pulse
excitation LPC of the European digital mobile system determines
its
pulse excitation by testing each of four sequences obtained by taking every fourth
sample of a filtered residual and selecting the sequence with maximum
correlation to
the residual sequence. Thus even the encoding algorithms of MPLPC and RELP are
sometimes similar. In a comparison of three specific coders
(subband, MPLPC, and
RELP), MPLPC coders
provided the best
performance
[53].
RELP encoding has been used in various mobile radio and satellite applications.
An example of the latter is a 9.6-kbps RELP codec designed to support conversion of
analog FM voice channels to digital voice channels
[54]'
150 vorcE DrGtTtzATtoN
Figure
3.43 CELp encoder block diagram.
Code-Excited LPC
code-excited LPc (CELP)
[55]
is yet
another example of enhanced-excitation
Lpc
whose basic block diagram can be loosely represented by Figure 3.34. As indicated in
the GELP specific block diagram of Figure 3.43, cELp uses a codebook for encoding
residuals. Instead of encoding a residue waveform on a sample-by-sample basis and
using that as excitation in the decoder, CELP a$$umes residuals are not random wave-
forms with independent
samples but rather that a block of residue samples can be rep-
resented by one of a manageable number of waveform templates.
,,Excitation
encoding" in this case involves selecting a codeword in a codebook that minimizes the
overall error in the reconstructed (synthe$ized)
signal.
"Residual
decoding" in this
context implies using the received codeword as an index into the table of codewords
to obtain the residual sequence chosen in the encoder. Because a block ofresidue sam-
ples
can be considered as a residue vector, this form of coding is also referred to as
vector quantization (VQ)
excited LPC
t561.
Maximum coding efficiency is achieved by encoding long sequences (i.e.,
vectors
with many dimensions), which, of course, implies large codebooks. Thus, much of the
research effort of this class of codecs involves establishing a large enough set of vec-
tors in the codebook that all residue vectors can be adequately matched. Next, an ef-
ficient procedure
for searching the codebook is determined to find the vector with the
best match. Because the number of vectors is necessarily large, tree-structured
searches are required, which implies the enhies in the codebook are grouped into hi-
erarchical families of vectors.
Because GELP coders provide good quality
at low bit rates, an extensive amount
ofresearch has been undertaken
to produce
a variety ofCELp algorithms. A parlicu-
larly efficient implementation
determines the excitation codeword as a sum of sequen-
tially determined codewords called vectors. Thus, this technique is referred to as
vector sum excited LPC (VSELP)
[56].
The following list identifies prominent
ver-
Mux
3.9 ENCODER/DECODERSELECTIONCONSIDERATIONS
151
sions of VSELP and CELP in North American applications. Other ver$ions of CELP
coders are included in the list of ITU standard coding algorithms in Section 3.10.
VSELP
(vector
sum excited LPC); used in North American digital cellular systems
(IS-54/IS-136) at 7.95 kbps
[57]
QSELP
(Qualcomm vector $um excited LPc): used in CDMA digital cellular
systems
(IS-95) at l3 kbPs
DoD-CELP Federal standard
(FS-1016)
at 4.8 kbps
[58]
3.9 ENCODEH|/DECODER SELECTION CONSIDERATIONS
This chapter has described several techniques for digitizing voice signals
and has in-
dicated that many other variations are
possible. This section compa.res the
various al-
gorithms in terms of
(l)
voice
quality,
(2) transparency for nonvoice signals,
(3)
tolerance of transmission effors, and
(4)
delay.
3.9.1 Volce Quality
Voice coding applications and corresponding
quality levels have been traditionally
classified as broadcast
quality, toll quality, communications
quality, and synthetic
quality. These categories and their respective ranges of data rates are
qualitatively
shown in Figure 3.44 as obtained from reference
tl
11. The relatively new category of
hybrid coding
(e.g., MPLPC or cELP) has been added. Table 3.5, obtained from ref-
erences
[38]
and
[59],
compares various coders in terms of the
quality measures: DRT
t441,
DAM
[45],
and Mos
t9,
11, 181. Notice that the low-rate coders do reasonably
well on intelligibility
(DRT) but fall off significantly in terms of subjective measure$.
Another
presentation of speech
quality of various coders is shown in Figure 3'45' ob-
tained from reference
t60l
(which contains a
good overview ofITU standard coding
algorithms). This figure essentially contains specific MOS ratings of specific coders.'
Additional comparisons of low-rate coders
(APC, RELP, LPC) are available in refer-
ences
[61]
and
[62].
The caregory of toll
quality, which is traditionally used for
public telephone net-
work voice quality objectives, is no longer a well-defined category. Toll quality in the
analog network could be quantified in terms of
parameters like frequency response
and noise level. Because new, low-rate coders
(e.g., CELPs) do not reproduce wave-
forms, a quantitative analytical measure of
quality is not possible. Suppliers of highly
compressed voice for applications like digital cellular and voice over the Intemet com-
monly use DRT or MOS scores to support the claim of
"toll
quality" under ideal hans-
mission environments
(e.g., no bit errors or
packet loss). This
practice has contributed
to a relaxation of a
"toll
quality" standard'
*MOS
sco.e*
pay
vary by 5% ot so from one study to another'
152 vorcE DtetlzATtoN
Waveform
coding
Toll quality
Broadcast
quality
ynthetic quhlity
8 1 6 3 2 6 4
Data rare (kbps)
Figure 3.44 General speech quality versus ffansmission rate.
3.9.2 Transparency
for Nonvoice Slgnals
The separation between communications quality
and toll quality
coders in Figure 3.44
also separates those algorithms that do not perform
well on nonvoice signals from
those that do. The lower data rates of communications qualiry
coders are achieved by
tailoring the algorithms to the specifics of voice signals, thereby sacrificing modem
and facsimile performance.
For a comprehensive theoretical treatment of voiceband
TABLE 3.5 Speech Ouality Comparleons of Common Goders
)
MOS
4
64-kbps PCM
95
14.4-kbps OCELPl s
3Z-kbps ADPCM (c.721)
94
16-kbps LD-CELP (c.728)
94
6.4 kbps MP-MLO (c.723.1)
13-kbps RPE-LTP (GSM)
9.6-kbps MPLPC (Skyphone)
9.6-kbps OCELP
8-kbp$ CELP
93
4,8-kbps DoD-CELP
99
Z.4-kbps LPC
90
68
70
4.3
4.2
4.'l
4.0
3.9/3.44
3.5
3.4
3.4
3.7
3, 0
2.5
68
67
54
the lower performance
corrosponds to 3qlo
packet
loss.
3.9 ENCODERiDECODERSELECTION CONSIDERATIONS153
S.6cft
AJdly
lxrcfn$ffi
Figure 3.45 Speech quality of standard encoding algorithms.
(From
R. V. Cox,
"Three
New
Speech Coders from the ITU Cover a Range of Applications," IEEE Communications
M a gazine, September 1997.)
data signal digitization, see reference
[63].
As end-to-end digital channels become
more available, there will be less need to suppolt nonvoice applications as analog
voiceband signals.
(See the discussion of V.90 PCM modems in Chapter I I ')
In addition to end user applications, coders installed in the intemal portions of a
network must
provide
acceptable
quality for network-related signaling tones such as
DTMF, MF, and SF. DTMF tones, in
particular, must be accurately reconstructed be-
cause they are used for end-to-end control by users. G.729 at I kbps has special
pro-
visions for carrying DTMF tones. Support for MF and SF signaling
is less of a
problem because their use is confined to specific transmission links that have been
mostly replaced by newer transmission Systems using common-channel signaling.
An additional consideration for voice quality is the
performance of fiome very low
bit rate vocoders in the
presence of audio background noise. If a coder is tuned too
tightly to
process voice signals and voice signals only, it can go off into left field when
speech is superimposed on background sounds such as loud music. The low-bit-rate
coders used in digital cellular applications have to be
pafticularly sensitive to this sinaation.
3.9.3 Tolerance of Transmlssion Errors
Of the waveform coders, the differential systems
(DPCM, ADPCM, DM) are the most
tolerant of transmission elror$ and PCM the least. The threshold of
perceptibility of
random error$ on delta modulation is 10-3. For PCM the threshold is 104. Delta
modulation i$ intelligible at random error rates as high as lOVo, but PCM is unintelli-
154 vorcE DrcrrzATtoN
gible at l7d elror rates. In burst error environments the tolerances of PCM and differ-
ential systems are more nearly the same.
(If
the most significant bit of a
pcM
code-
word is in error, it does not matter if the least significant bits are also in error.)
Those parameters
of synthetic quality
coders that are critical for voice reconstruc-
tion are usually redundantly encoded. Error correction ofcritical parameters
in digital
mobile systems allows intelligible voice at error rates as high as l Zo.
3.9.4 Delay
The effect of encoding and decoding delay of a voice digitization/compression algo-
rithm must be considered in the context of the particular
application. If the application
involves inserting artificial delay of more than 10 msec into local (analogr
connec-
tions, echo/singing control will have to be added. Furthermore, if a significant amount
of delay is added into a long-distance circuit, existing echo cancelers may not have
enough delay capacity to accommodate the delay insertion.
Experience wittr satellite-ba.sed voice connections indicates thatroundtrip delays on the
order of 2ff) msec can be tolerated witlrout sifficant user dissatisfaction. As indicated
in the following, coding/compression algorithms, in themselves, do not approach
this limit.
ADPCM (G.726)
LD-CELP (G.728)
0.125 msec
2.5 msec
CS-CELP (c.729)
t64l
l0 msec
ACELP
(G.723.1)
30 msec (plus 7.5 msec of look-ahead)
Although the roundtrip encoding/decoding delays (which
are double rhe above num-
bers) do not approach the 200-msec threshold, they can add to oflrer system delays to
exceed the maximum desirable delay. System factors such as interleaving for error
correction and packet delay on TDMA mobile systems can add another 30 msec or so
in each direction. If individual compressed voice packets
are carried through an ATM
or packet-switched
network, delays in excess or 200 msec are easily produced. (see
Chapter I 0 for a discussion of these applications.)
3.10 ITU.T CODING STANDARDS
The following list identifies various digital voice coding standards of the ITU:
o.7ll standard for speech codecs that provides
toll
quality
audio at 64 kbps
using eitherA-law or p-law
PCM.
G-721 standard for speech codecs that
provides
toll quality
audio at 32 kbps
using ADPCM.
G'722 standard for speech codecs thar provides
high-quality (program)
audio at
64 kbps using subband ADPCM (SB-ADPCM).
The algorithm uses a l6-
kHz sample rate to capture audio frequencies between 50 and 7000 Hz.
Two ADPCM subbands are used by this srandard to give audio
performance
superior to a single-band ADPCM algorithm operating at the
same bit rate.
REFERENcES 155
G.723 Standardforspeechcodecs thatprovidestoll
quality audio at20or40kbps
using ADPCM.
G.723.1 Standard for speech codecs optimized for modems. It provides toll
quality
audio at 6.4 kbps
(MP-MLQ) or 5.3 kbps
(ACELP).
G.726 Adaptive differentialPCM at 16,24,32, or40kbps (using ?-,3-,4-,ot5-
bit samples).
G.727 An extension of Recommendation G.726 for use in
packetized speech
$ystems. The ADPCM samples are divided into 2, 3, or 4 core bits and 0,
l, 2, or 3 enhancement bits. The core bits
provide for the basic functioning
of the algorithm while the enhancement bits add quality to that
provided
by the core bits. In overload situations, the enhancement bits can be
discarded while the core bits provide basic
quality.
G.7ZB Standard for speech codecs that
provides near toll quality audio at 16 kbps
using low-delay CELP
(LD-CELP).
G.728 encodes five
p-law or A-law
PCM samples into 10-bit, linear
predictive codewords at 1600 codewords
per second.
G.729 Standard for speech codecs that
provides toll
quality audio at 8 kbps using
CELP.
REFERENCES
I J. M. Kasson,
"The
Rolm Computerized Branch Exchange: An Advanced Digital
PBX," Computer Magazine, June 1979, p' 24.
2 G. Frantz and K.-S. Lin,
"speech
Technology in Consumer Products," Speec*
Technology, Apr. 1982, pp. 25-27 .
3 S. Finkelstein,
"Customieed
LPC Vocabulary L.ets Cars
"lalk,"
Speech Technology,
Sept./Oct. 1983, pp. 65-69.
4 M. G. Carasso, J. B. H, Peek, and J. P. Sinjou,
"The
Compact Disc Digital Audio
System," Phillips Technical Review, Vol. 40, No. 6, 1982, pp. 151- 155.
5 L, Rabiner and B.-H. Jtang, Fundamentals of Speech Recogrlition, hentice-Hall,
Englewood Cliffs, NJ, 1993.
6 Y. Sagiska,
"speech
Synthesis from Text," IEEE Communications Magazine, Jan.
1990, pp, 35-41.
7 J.L. Flanagan, Speech Analysis, Synthesis, and Perception, 2nd ed. Springer-Verlag'
New York. 1972.
I K. F. Lee, H. W. Hon, and R. Reddy,
"An
Overview of the SPHINX Speech
Recognition System," IEEE Transactions on Acoustics Speech and Signal Processing,
Jan. 1990, pp. 35-45.
9 N. S. Jayant and P, Noll, Drgital Coding, of Wavefotms: Principles and Applications to
Speech and Video, Prentice-Hall, Englewood Cliffs, NJ, 1984.
l0 J. Flanagan, M. Schroeder, B. Atal, R. Crochiere, N. Jayant, and J. Tribolet,
"Speech
Coding," IEEE Transaction$ on Communicatiorar, Apr. 1979, pp.
'll0-737
-
11 N. Kitawaki and H. Nagabuchi,
"Quality
Assessment of Speech Coding and Speech
Synthesis Systems," IEEE Communications Magazine, Oct. 1988,
pp. 36-214.
156
tz
t 3
I 4
l 5
l 6
t 7
r 8
r 9
20
21
22
24
) \
26
27
?8
29
VOIGE DIGITIZATION
B. J. McDermott, C. Scagliola, and D. J. Goodman,
"Perceptual
and Objective
Evaluation of Speech hocessed by Adaptive Differential PCM," Bell SystemTechnical
Journal, May/June 1978, pp. 1597-1618.
"The
D3 Channel Bank Compatibility Specifrcarion-Issue 3," Technical Advisory
No. 32, American Telephone and Telegraph Company, Basking Ridge, NJ, Oct. lgi7.
M. B, Akgun,
"Transmission
Objectives for a Subscriber Loop System Digital Voice
Codec," paper presented
at the International Symposium-Subscriber Loops and
Services, Ottawa, 1974, pp.
' 7.L1-7.1.7.
H. Nann, H. M. Straube, and C. P. Villars;
"A
Companded Coder for an Experimental
PCM Terminal," BeIl SystentTechnical Joumal, Jan. 1962, pp. 173-226.
K. E. Fultz and D. B. Penick,
'"The
Tl Carrier System," BelI System Technical foumal,
Sept. 1965, pp. 1405-1451.
H, H. Henning and J. W. Pan,
"D2
Channel Bank: System Aspects," Bell System
Technical Journal, Oct. 1972, pp. 1641-1657.
W. R. Daumer,
"Subjective
Evaluation of Several Efficient Speech Coders," IEEE
Transactions on Communicarrions, Apr. 1982, pp. 655*662.
N. S. Jayant,
"Step-Size
Transmitting Differential Coders for Mobile Telephony," Bell
SystemTechnical Joumal, Nov. 1975, pp, 1557-1581.
D. L. Duttweiler and D. G. Messerschmitt,
"Nearly
Instantaneous Companding and
Time Diversity as Applied t0 Mobile Radio Transmission," IEEE International
Communications Conference, I 975, pp. 4*12-40-15.
I. M. McNair, Jr.,
-'Digital
Multiplexer for Expanded Rural Sewice,,, Eelt l_abs Record,
Mar. 1972, pp.
80-86.
H. K. Dunn and S. D. White,
"Statistical
Measurements on Conversational Speech,"
Joumal of Acoustic Society of America, Jan. 1940, pp. 278-288.
J. W. Bayless, S. J. Campanella, and A. J, Goldberg,
"Voice
Signals: Bit by Bir," IEEE
Spectrurn, Oct. 1973, pp.28-34.
M. Decina and G. Modena,
"CCITT
Srandards on Digital Speech hocessing," IEEE
Journal on Selected Areas in Communicarions, Feb. 1988, pp. 227-234.
G. Williams and H. Suyderhoud,
"subjective
Performance Evaluation of the 32 kbit/s
ADPCM Algorithm," I EEE Glob e c om P roc e edin g s, I 984, pp. 23.2.1
-23.2.1.8.
J, A, Greefkes and K. Riemens,
'Code
Modulation with Digitally Controlled
Companding for Speech Transmission," Phillips Technical Review, Vol. 31, 1970, pp.
335-353.
J. E. Abate,
"Linear
and Adaptive Delta Modulation," Pro ceedings of IEEE, Mar, I 967,
pp. 298-308.
N. S. Jayant and A. E. Rosenberg,
"The
heference of Slope Overload to Granularity in
the Delta Modulation of Speech," Bell System Technical Journal, Dec. l97l, pp.
3rr7-3125.
J. A. GreeEkes and F. de Jager,
"Conrinuous
Delta Modularion," Phillips Research
Reports, Vol. 23, 1968, pp.
233-246.
J. A. Greefkes,
"A
Digitally Companded Delta Modulation Modem for Speech
Transmission," Proceedings of IEEE International Communitations,lgT0, p. 7-37 to
7-48.
30
REFEHENCES 1$7
3l B. S. AtaI and M. R. Schroeder,
"Adaptive
Predictive Coding of Speech Signals," Bell
Sy stem Te chnic dI J o umal, Oct. 1970,
pp. I 973
-
I 986.
32 P. Noll,
"On
Predictive
Quantizing
Schemes," BelI System Technical Joumal,
May/June 1978, pp. 1499-1532.
33 M. R. Schroeder,
"Linear
Predictive Coding of Speech: Review and Current
Directions," IEEE Communications Mag,azine, Attg. 1985, pp. 54-61.
34 T. Tremain,
'The
Govemment Standard Adaptive hedictive Coding Algorithm;
APC-04," SpeechTechnology, Feb./IVIar. 1985, pp, 52-62.
35 R. E. Crochiere,
"An
Analysis of 16 kb/s Sub-Band Coder Performance, Dynamic
Range, Tandem Connections, and Channel Errors," Bell SystemTechnical Journal'Qct,
1978, pp. 2927-2951.
36 P. Mermelstein,
"G.722
A New CCITT Coding Standard for Digital Transmission of
Wideband Audio Signals," IEEE Communications Magazine, Jan' 1988, p' 815.
3'7 X. Maine,
"7
kHz Audio Coding within 64 kbit/s," IEEE Joumal on Selected Areas in
Communications, Feb. 1988, pp. 283-298.
38 N. S, Jayant,
"High-Quality
Coding of Telephone Speech and Wideband Audio"'IEEE
Communications Magazine, Jan. 1990,
pp. 10-20.
39 L. Radin,
"Low-Bit-Rarc
Voice Digitizer Hits the Market at 2,400 bls," Speech
Technolo
gy,Sept./Oct. 1983, pp. 92-95.
40 H. Dudley,
"Remaking
Speech," "Ioanral of the Acoustic Society of Amerfua, Oct. 1939'
pp. 169-177.
4l L. R. Rabiner and R. W. Schafer, Digital Processing of Speech Signals, Prentice-Hall,
Englewood Cliffs, NJ, 1978.
42 T. E. Tremain,
"The
Govemment Standard Linear Predictive Coding Algorithm:
LPC- 10," Speech Technolagy, Apr. 1982, pp. a0-49.
43 J. P. Campbell and T. E. Tremain,
"Voiced/Unvoiced
Classification of Speech with
Applications to the U.S. Govemment LPC-I0E Algorithm," IEEE Intemational
Conference on Acoustics, Speech, and Signal Processing, 1986, pp. 473476.
44 W, D. Voiers,
"Evaluating
hocessed Speech Using the Diagnostic Rhyme Test,"
SpeechTechnolagy, Jan./Feb. 1983, pp. 30-39.
45 W. D. Voiers,
"Diagnostic
Acceptability Measure for Speech Communication
Systems," Proceedings Intemational Conference on Acoustics, Speech and Signal
Processing, May 1977, pp.204*207 .
46 B. S. Atal and R.
Q.
Hofacker,
"The
Telephone Voice of the Funre," Bell l-abs Record,
July 1985, pp. 4-10.
47 I. Boyd, C. B. Southcoft, D. P, Crowe, and P. J. Bolingbroke,
"Speech
Codec for the
Skyphone Aeronautical Telephone Service," BritishTelecommunications Engineering,
July 1986, pp. 83-91.
48 J. E. Natig, S. Hansen, and J. deBrito,
"speech
hocessing in the Pan European Digital
Mobile Radio System
(GSM)-system Overview," IEEE Globecom Proceedings,1989,
pp. 29B. l . l -298.1.1.5.
49 K. Hellwig, P. Vary, D. Massaloux, J. P. Petit, C. Galand, and M' Russo,
"Speech
Codec
for the European.Mobile Radio System,* IEEE Globecom Proceedings, 1989' pp'
298.2.1-298.2.r.s.
158 vorcE DrcrrzATroN
50 A. V. McCree and T. P. Bamwell.
"A
Mixed Excitation LPC Vocoder Model for Low
Bit Rate Speech Coding," IEEE Transactions on Speech and Audio Processing, July
1995, pp. 242-250.
51
"Specifications
for the Analog to Digital Conversion of Voice by 2,400 Bir/Second
Mixed Excitation Linear Prediction," Deparffnent of Defense Draft, May 28, lggS
(www.plh.af.miUddvpc/melp.hnn).
52 C. K, Un and D. T. Magill,
"The
Residual-Excited Linear Prediction Vocoder with
Transmission Rate Below 9.6 kbot/s," IEEE Transactions on Communications, Dec.
1975, pp, 1466-1474.
53 K. K. Paliwal and T. Svendsen,
"A
Study of Three Coders
(Sub-band,
RELP, and MpE)
fbr Speech with Additive White Noise," Proceedings IEEE Intemational Conference
on Acoustics, Speech and Signal Processing, 1985, pp. 43.5,I
-43.5.1.4.
54 A. J. Morris and S. Kay,
-'Telephony
Enth Station," IEEE Globecom
proceedings,
1989, pp, 48. I. I
-48.1.1.4.
55 M, R. Schroeder and B. S. Aral,
"Code-Excited
Linear
prediction
(CELP):
High-Quality Speech at Very Low Bit Rates," IEEE International Conferente on
Acoustics, Speech and Signal Processing, 1985, pp, 25.1.1-?5.1.1.4.
56 J. H. Chen and A. Gersho,
"Gain-Adaptive
Vector
Quantization
with Application to
Speech Coding," IEEE Transactions on Communicarrons, Sept. 1987, pp.
gl8-930.
57
"Vector
Sum Excited Linear hediction
(VSELP)
7950 Bit Per Second Voice Coding
Algorithm," Technical Description, Motorola, Shaumburg, IL, Nov. 14, 1989.
58 J. P. Campbell, V. C. Welch, and T. E, Tremain,
'An
Expandable Error-hotected 48fi)
BPS CELP Coder
(U.S.
Federal Standard 4800 BPS Voice Coder\," IEEE Intemational
Conference on Acoustics, Speech and Signal Processing,1989, pp. 735-737.
59 T, S. Rappaport, Wireless Communications: Principles and Practice, Prentice-Hall,
Upper Saddle River, NJ, 1996.
60 R. V. Cox,
'*Ihree
New Speech Coders from the ITU Cover a Range of Applications,"
I EEE Communications Magasrne, Sept. 1 997, pp. 4O47 .
61 R. D. Hoyle and D. D. Falconer,'A Comparison of Digital Speech Coding Methods for
Mobile Radio Systems," IEEE Journal on Selected Areas in Communications, June
1987, pp. 915-920.
62 N. Kitawaki, H. Nagabuchi, and K. Itoh,
"Objective
euality
Evaluation for
Low-Bit-Rate Speech Coding Systems," IEEE Journal on Selected Areas in
Communitatio,ns, Feb, 1988, pp. U2*24'7.
63 J. B. O'Neal, Jr.,
"Waveform
Encoding of Voiceband Data Signals,"
proceedings
of the
IEEE, Feb. 1980, pp. 232-247.
64 C. Schnider and
M. H. Sherif,
"The
Road to G.729.. ITU
g-kb/s
Speech Coding
Algorithm with Wireline
Quality,"
IEEE Corwnunications Magazine, Sept. 1997, pp.
48*54.
PROBLEMS
3.1 Assume a signal con$ists of three tones: one at I kHz, one at 10 kHz. and one at
2l kHz. What tones will be presenr ar rhe ouFut of a PAM decoder if the sam-
PRoBLEMS 159
pling rate is I ? kHz and the input is unfiltered?
(Assume
the output filter cutoff
frequency is 6 kHz.)
3.2 Derive an expression for the average
quantization noise power that occurs when
the decoder output samples are offset from the center of a
quantization interval
by adistance equal to 25Vo of the interval.
(The outputvalues are aLtheTSVa
point instead of the 507o point.) How much degradation in decibels does this off-
set represent
(assuming uncorrelated offsets) ?
3.3 How much does the signal-to-noise ratio of a uniform PCM encoder improve
when I bit is added to the codeword?
3.4 A black-and-white television signal has a banilwidth of about 4.2 MHz. What bit
rate is required if this signal is to be digitized with uniform PCM at an SQR of
30 dB? Use a sampling-rate-to-Nyquist-rate ratio comparable to that used for
PCM voice encoding.
3.5 How much dynamic range is provided by a uniform PCM encoder with l2 bits
per sample and a minimum SQR of 33 dB?
3.6 What is the signal-to-quantizing-noise ratio produced by a segmented
p255
PCM coder when the input signal is a full-range triangular wave?
(Assume
the
repetition frequency is low enough that the bandlimiting filter does not change
the waveform significantly.)
3.7 Given a sample value of 420 mV for a
p255 PCM encoder capable of encoding
a maximum level of 2 V, determine each of the following:
(a)
The compressed
p255 codeword
(b) The linear representation of the compressed code
(c)
The A-law code obtained by converting from the
p255
code
(d) The
pJaw
code obtained by converting back from the A-law code
3.8 Given the following
p255 codewords, determine the (noninverted) codeword
that represents the linear sum
(0 I 10 1001),
(l
01 I 0I I I
).
3.9 Generate an encoding table
(i.e., list the
quantieation intervals and conespond-
ing codes) for the magnitude of a piecewise linear code with segment slopes 1,
j,
i,
and
f.
Assume four equally spaced intervals are in each segment. Assume
all intervals in the first segment are of equal length
(as
in A-law coding).
3.10 What is the signal-to-noise ratio of a full-amplitude sample for the coder of Prob-
l em 3.9?
3.11 What is the signal-to-noise ratio of a maximum-amplitude sample in the first lin-
ear $egment of the preceding coder?
3.12 What is the dynamic range implied by Problems 3.10 and 3.11?
3.13 For the encoder in hoblem 3.9, how many bits are required for unifotm encod-
ing of the same dynamic range and same minimum quantization interval?
3.f4 A uniform PCM system is defined to encode signals in the range of
-8159
to
+8159 in
quantization intervals of length 2. (The quantization interval at the ori-
gin extends from
-1
to +I.) Signals are encoded in sign-magnitude format with
a
polarity bit
=
1 denoting a negative signal.
(a)
How many bits are required to encode the full range of signals?
160 votcE DlcrrzAloN
(b) How many unused codes are there?
(c) Determine the quantization noise, noise power,
and signal-to-noise ratio (in
decibels) of each of the following sample values; 30.2,123.2,
-2336.4,
and
8080.9.
3.15 Repeat
part (c)
in Problem 3. l4 for piecewise
linear p255 PCM.
3.16 Given two A-law piecewise
linear
(noninverted)
codewords
(00110110
and
00101100), determine their linear representations, add them, and convefi back
to compressed representation.
3.17 A D3 charrnel bank uses
"robbed
digit" signaling wherein the least significant bit of
every sixth frame is stolen for signaling. Determine the relative increase in overall
quantization
noise produced
by this process under the following conditions:
(a)
The decoder merely treats the signaling bit as a voice bit and decodes the
PCM sample accordingly.
(b)
The decoder knows which bits are signaling bits and
generates
an output
sample corresponding to the middle of the double-length quantization inter-
val defined by the 7 available bits. (This is the actual technique specified for
D3 channel bank decoders.)
3.18 If 2 bits
per
sample are added to a PCM bit stream, how much can the dynamic
range be increased if the quantization
intervals are adjusted to improve the SQR
by 3 dB?
3.19 How much can the SQR of a digitized voice signal be increased if the bandwidth
is reduced by 30Vo and the dynamic range is unchanged?
3.20 AnA-law PCM encoder with a perfect zero-level threshold detect$ an arbitrarily
small input sine wave because the first quantization
interval does not straddle the
origin. What is the power level of the decoded output signal? Assume the power
level of a full amplitude sinewave is +3 dBm.
3.21 Determine the sequence of four
p255
PCM samples for a 2000-Hz tone at
-6
dBm when the first sample occurs at a phase
of 45".
3.?? Repeat Problem 3.21 forAlaw signals.
3.23 Determine the A-law codewords that correspond to the
p-law
codewords of the
digital milliwatt signal given in Example 3.3. Assume both systems are scaled
to the same maximum signal power.
DIGITAL TRANSM ISSION AND
MULTIPLEXING
A fundamental consideration in the design of digital transmission $ystems is the se-
lection of a finite set of discrete elechical waveforms for encoding the information. In
the realm of digital communications theory these discrete waveforms are referred to
as signals. The same terminology is used in this chapter with the understanding that
signals in the
present context refer to intemal waveforms
(pulses) of a transmission
system and not the control information
(signaling) used to set up and monitor connec-
tions within a network. In communications theory terminology, signal
processing re-
fers to filtering, shaping, and transformations of electrical waveforms, not the
interpretation of conffol signals by the
processor of a switching machine.
A second aspect of digital transmission involves defining certain time relationships
between the individual transmission signals. The source terminal transmits the indi-
vidual signals using predefined time relationships so the receiving terminal can rec-
ognize each discrete signal as it arrives. Invariably the establishment of a proper time
base at the receiver requires transmission capacity above that needed for the digital in-
formation itself. Over relatively short distances
(as within a switching system or a
computer complex), the timing information
(clock) is usually distributed separately
from the information-bearing signals. Over long distances, however, it is more eco-
nomical to incorporate the timing information into the signal format itself. In eittrer
case the timing information requires channel capacity in terms of bandwidth, data rate,
or code space.
This chapter discusses the most common digital signaling techniques fbr wireline
transmission systems. Thus the subject of this chapter is often referred to as line cod-
ing. These techniques are
generally applicable to any transmission system ffansmitting
digital signals directly in the form of pulses (such as coaxial cable or optical fiber).
The fact that these techniques include low-frequency components in their transmitted
spectrum leads to their also being called low-pass or baseband transmission systems.
In Chapter 6 we discuss bandpass transmission sy$tems, that is, radio systems or
voiceband modems that require modulation and carrier frequency transmission. Chap-
161
162 DtcrrAL rRANSMtsstoN AND MULTIpLEXtNc
ter 8 discusses some additional baseband transmission formats commonly used in op-
tical fiber system$.
The following discussions concentrate on system and application level considera-
tions of digital transmission. Analytic details of pulse transmission systems are not
emphasized since material of this nature is available in all works on digital communi-
cations theory. some fundamentals of pulse
transmission systems are presented
in Ap-
pendix
C, where many of the equations presented
in this chapter are derived.
4.1 PULSE TRANSMISSION
All digital transmission systems are designed around some
particular
form of
pulse
re-
sponse. Even carrier systems must ultimately produce
specific pulse shapes at the de-
tection circuitry of the receiver. As a first step, consider the perfectly square pulse
shown in Figure 4.1. The frequency spectrum corresponding to the rectangular pulse
is derived in Appendix A and shown in Figure 4.2. It is commonly referred to as a
sin(x)/x response:
-_
sin(roll2)
1 ' ( ( 0 1
= ( I l +
aTl 2
(4. 1)
where co
=
radian ftequency 2nl,
7
=
duration of a signal interval
Notice that Figure 4.2 also provides
the percentage of total $peckum power at vari-
ous bandwidths. As indicated, 90Vo of the signal energy is contained within the first
spectral null at
f
=
l/T.The high percentage
of energy within this band indicates that
the signal can be confined to a bandwidth of l/T and still pass
a
good
approximarion
to the ideal waveform. In theory, if only the sample values at the middle of each signal
interval are to be preserved,
the bandwidth can be confinedto IlZT.From this fact the
maximum baseband signaling rate in a specified bandwidth is determined as
R**
=
2BW (4.2)
r T
- 1
Z
Time
-**
Figure 4.1 Definition of a square pulse.
4.1 PULSETFANSMISSION 163
t
*ro
. E
E
t
E
fi
tt
fl
-ao
u
-30
Froqumcy (Hrl
Figure 4.2 Speckum of square pulse with duration T.
where R
=
signaling rate,= IlT
BW
=
available bandwidth
Equation 4.2 states a fundamental result from communications theory credited to
Harry Nyquist; The maximum signaling rate achievable through a low'pass band-
width with no intersymbol interference is equal to twice the bandwidth. This rate R^o
is sometimes referred to as the Nyquist rate.
Although discrete, square-shaped
pulses are easiest to visualize,
preservation ofthe
square shape requires wide bandwidths and is therefore undesirable. A more typical
shape for a single pulse is shown in Figure 4.3. The ringing on both sides of the main
part of the pulse is a necessary accompaniment to a channel with a limited bandwidth.
-3r -2 +- - r l T
4T
2T
Ti me.*
Figure 4.3 Typical pulse response of a bandlimited channel.
l ( t l
I
I
\ ,''i*t
t
I
I
I
I
I
I
Frtrtion of out-of-bmd
pomr (<tB)
164 DTct TALTRANSMt SSt oNANDMULTt pLEXt NG
Normally, a digital transmission link is excited with square pulses (or
modulated
equivalents thereofl, but bandlimiting filters and the transmission medium itself com-
bine to
produce
a response like the one shown. Figure 4.3 shows pulse
output in nega-
tive time so the center of the pulse
occurs at f
=
0. Actually, the duration of the
preringing
is limited to the delay of the channel, the filters, and the equalizers.
An important feature of the
pulse
response shown in Figure 4.3 is that, despite the
ringing, a
pulse
can be transmitted once every T seconds and be detected at the receiver
without interference from adjacent pulses.
Obviously, the sample time must coincide
with the zero crossings of the adjacent pulses. Pulse responses like the one shown in
Figure 4.3 can be achieved in channel bandwidths approaching the minimum
(Nyquist)
bandwidth equal to one-half of the signaling rate. Appendix C describes
pulse
transmission design in more detail.
4.1.1 Intersymbol Interference
As the signaling rate of a digital transmission link approaches the maximum rate for
a
given
bandwidth, both the channel design and the sample times become more criti-
cal. Small perturbations in the channel response or the sample times produce nonzero
overlap at the sample time$ called intersymbol interference. The main causes of in-
tersymbol interference are;
l. Timing inaccuracies
2. Insufficient bandwidth
3. Amplitude distortion
4. Phase distortion
4. 1. 2 Ti ml ngl naccuraci es
Timing inaccuracies occuring in either the transmitter or the receiver produce in-
tersymbol interference. In the transmitter, timing inaccuracies cause intersymbol in-
terference if the rate of transmission does not conform to the ringing frequency
designed into the channel. Timing inaccuracies ofthis type are insignificant unless ex-
tremely sharp filter cutoffs are used while signaling at the Nyquist rate.
Since timing in the receiver is derived from noisy and
possibly
distorted receive
signals, inaccurate sample timing is more likely than inaccurate transmitter timing.
Sensitivity to timing errors is small if the transmission rate is well below the Nyquist
rate (e.9., if the transmission rate is equal to or less than the channel bandwidth, as op-
posed
to being equal to the theoretical maximum rate of twice the bandwidth; see Ap-
pendix
C).
4.1.3 InsufflclentBandwidth
The ringing frequency shown in Figure 4.3 is exactly equal to the theoretical minimum
bandwidth of the channel. If the bandwidth is reduced further, the ringing frequency
is reduced and intersymbol interference necessarily results.
4,2 ASYNCHRONOUS VERSUS SYNCHRONOUS TRANSMISSION 165
Some systems purposely signal at a rate exceeding the Nyquist rate, but do so with
prescribed amounts of intersymbol interference accounted for in the receiver. These
systems are commonly referred to as
pafrial-response systems-so called because the
channel does not fully respond to an input during the time of a single
pulse. The most
common forms of partial-response $y$tems are discussed in a later section.
4.1.4 Ampl i tudeDl stoi l i on
Digital transmission systems invariably require filters to bandlimit transmit spectrums
and to reject noise and interference in receivers. Overall, the filters are designed to pro-
duce a specific pulse response. When a transmission medium with
predetermined
characteristics is used, these characteristics can be included in the overall filter design.
However, the frequency response of the channel cannot always be predicted ade-
quately. A departure from the desired frequency response is referred to as amplitude
distortion and causes pulse distortions
(reduced peak amplitudes and improper ringing
frequencies) in the time domain. Compensation for irregularities in the frequency re-
sponse of the channel is referred to as amplitude equaliqation.
4.1.5 Phase Di storti on
When viewed in the frequency domain, a
pulse is represented as the superposition of
frequency components with specific amplitude and
phase relationships. If the relative
amplitudes of the frequency components are altered, amplitude distortion results as
above. If the
phase relationships of the components are altered,
phase distortion oc-
curs. Basically,
phase distortion results when the frequency components of a signal ex-
perience differing amounts of delay in the transmission link. Compensation of
phase
distortion is referred to as phase equalization. For a good tutorial on equalization, in-
cluding a description of an automatic equalizer for data transmission, see reference
[1].
A good description of DSP based equalizers is available in reference
[2].
4.2 ASYNCHRONOUS VERSUS SYNCHRONOU$ TRANSMISSION
There are two basic modes of digital transmission involving two fundamentally dif-
ferent techniques for e'rtablishing a time base
(sample clock) in the receiving terminal
of a digital transmission link. The first of these techniques is asynchronous transmis-
sion, which involves separate transmissions of groups of bits or characters. Within an
individual group a specific
predefined time interval is used for each discrete signal.
However, the transmission times of the
groups are unrelated to each other. Thus the
sample clock in the receiving terminal is reestablished for reception of each group'
With the second technique, called synchronous transmission, digital signals are sent
continuously at a constant rate. Hence the receiving terminal must establish and main-
tain a sample clock that is synchronized to the incoming data for an indefinite period
of time.
166 DIGITAL THANSMISSION AND MULTIPLEXING
Tran$fiitted waveform
l t t | i l 1 l l l l l l r l
ldeol Ea|nple timer for each 8-bit character
Figure 4.4 Asynchronous transmission.
4.2.1 Asynchronous Transmiesion
Between transmissions an asynchronous line is in an inactive or idle state. The begin-
ning of each transmission group is signified by a start bit. The middte of the start bit
is determined, and succeeding information bits are sampled at a nominal rate begin-
ning at the middle of the second bit interval. Following the information symbols, one
or more stop bits are transmitted to allow the line to return to the inactive state.* Figure
4.4 shows an asynchronous mode of operation commonly used for low-speed data
communications.
As shown in Figure 4.4, the detection of each information bit is accomplished by
ideally sampling the input waveform at the middle of each signal interval. In practice,
sample times depart from the ideal depending on how much the start bit is comrpted
by noise and distortion. Since the sample time for each information bit is derived from
a single start bit, asynchronous systems do not pedorm
well in high-noise environ-
ments. of course, more than one start bit could be used to improve the accuracy of the
starting phase of the sample clock, but this would complicate the receiver and add
more overhead for transmission of timing information.
Sample timing enors also arise if the nominal rate of the sample clock in the re-
ceiver is different from the nominal rate of transmission at the source. Even though
the start bit might define the proper starting phase for the sample clock, an offset in
the clock frequency of the receiver causes each successive sample time to drift faflher
from the center ofthe respective signal intervals. Since the very use ofthe term'hsyn-
chronous" implies a free-running clock in the receiver, a certain amount of drift is in-
evitable in all asynchronous systems. The maximum length of each symbol group
or
character is determined by the limits of the initial phase inaccuracies and the maxi-
mum expected frequency difference between the kansmitter and receiver clocks.
-Originally,
stop bits were inserterl to allow electromechanical equipment enough time to reset before the
next character anived. With elechonic equipment thc only purpose
of stop bits is to allow a start bit to
always be a fransition to a space (logic
0).
Receivad waneform
4.2 ASYNCHRONOUS VERSUS $YNCHRONOUS TRANSMISSION 167
The main attraction of asynchronous transmission is the ease with which it deter-
mines the sample times in the receiver. In addition, asynchronous ffansmission auto-
matically
provides character framing and is inherently flexible in the range of average
data rate$ that can be accommodated. For high rates, one character after another is
transmitted. l,ower data rates are automatically accommodated by increasing the idle
time between characters. In contrast, a synchronous receiver must track changes in a
transmitter rate before it can sample the incoming signals
properly. Normally the re-
ceive clock of a synchronous system can be adjusted only quite slowly and only over
a nilrow range. Hence an asynchronous system is more naturally suited to applica-
tions where the data rate varies.
Synchronous transmission systems can suppofi variable information rates, but the
task of adjusting the information rate falls upon higher level
processes
(data link pro-
tocols) that insert null codes into the bit stream. The null codes are used as filler when
a source has nothing to send. This form of transmission is sometimes referred to as
"isochronous."
An isOChrOnous mode Of Operation is required whenever a SynChro-
nous line carries data from an asynchronous source.
The major drawback of asynchronous transmission is its poor performance in terms
oferror rates on noisy lines. Thus asynchronous transmission evolved for applications
where implementation costs dominated
performance considerations. Asynchronous
transmission has been used in voiceband data sets
(modems)
for transmission rates up
to 1200 bps. For digital telephony, something similar to asynchronous
transmission
was proposed for two-wire digital subscriber loops. These systems
provided a full-du-
plex
(four-wire) circuit by transmitting bursts of data alternately in each direction on
a single
pair of wires. Thus these systems are sometimes referred to as
"ping-pong"
ffansmission systems; they are not ffuly asynchronous since each transmission in each
direction occurs at prescribed times allowing timing information in one burst to carry
over into the next burst. The use ofasynchronous tran$mission on long-distance trans-
mission links is obsolete because the cost of electronics for better
performing synchro-
nous transmission is no longer a consideration'
4.2.2 Synchronou$ Transmi s$i on
Tl lines and all other interoffice digital transmission links used in the telephone net-
works use synchronous transmission exclusively. Thus the line-coding format for
these systems must incorporate special considerations to en$ure that each regenerative
repeater or receiver can synchronize a local sample clock to the incoming signaling
rate. Generally speaking, the synchronization requirements imply that a certain mini-
mum density of signal transitions is required to
provide continuous indication of sig-
naling boundaries. Often,
purely random data cannot be relied upon to
provide a
sufficient number of transitions. In these cases ceftain
provisions mu$t be made to in-
sert artificial transitions into the transmitted waveforms. Although these extra transi-
tions imply a certain amount of transmission overhead, the loss in capacity can be
relatively small. Following are descriptions of f,rve techniques for ensuring the exist-
ence of signal transitions for timing recovery:
168 DrcrrAL TRANSMtsstoN AND MULTIpLEXtNG
l. Source coderestriction
?. Dedicated timing bits
3. Bit insertion
4. Data scrambling
5. Forced bit errors
A sixth technique-inserting the transitions into the signal waveforms themselves-
is discussed in Section 4.3.
Source Code Restrlction
One means of ensuring a sufficient number of signal transitions is to re$trict the code
set or data
paftems
of the source so that long, transition-free data sequences do not oc-
cur. Historically, PCM channel banks in the North American telephone network pre-
cluded all-0's codewords since the original line code of rl lines produced
no
transitions for 0's. In the case of 8-bit PCM codewords, the exclusion of a single code-
word represents a loss in transmission capacity of only one part in 256.-
Relying on source coding to ensure sufficient transitions in the line code has one
very significant drawback The transmission link cannot be used for new applications
where the source does not exclude the unwanted data pattems. For this reason, the total
capacity of a Tl line could not be used for random data until a new line code
[binary
eight-zero substitution (B8ZS)]
was introduced.
Dedlcated Timing Blts
As an altemative to excluding
transition-free data patterns, the line itself can peri-
odically insett transition-bearing bits into the data stream. These bits are inserted at
regular intervals, independently ofthe source data, to ensure the existence of a mini-
mum number of signal transitions. Thus some fraction of the channel capacity is dedi-
cated to timing bits.
As an example, the original Dataphone Digital service (DDS)
offering for dara
communications over Tl lines provides
a maximum capacity of 56 kbps for each
channel. when carried within a 64-kbps Tl channel, only 7 of the 8 bits in each rime
slot of the channel are available for the user. Among other functions, the unused bit
in each time slot provides
an assurance that all 8 bits of a time slot are not 0. Several
fiber systems described in Chapter 8 also use explicit timing bits as inherent aspects
of their line codes. The density of timing pulses
in these systems ranges from I in S
bits to I in 20 bits. Notice that insertion of a dedicated riming bit is essentially the same
procedure
used to establi$h timing for asynchronous transmission. In a synchronous
receiver, however, a sample clock is obtained by averaging the timing information
over a large number of timing transitions, not
just
one.
Blt lnsertion
In the preceding
DDS example I bit of every 8 bits in a time slot is dedicated to en-
suring sufficient timing information in the bit stream. Another possibility
for preclud-
"From
an inf'ormation theory point
of view the loss in voice information is even lower since the probahility
of occunence of the all-()'s codeword is much less than I in 256.
4.2 ASYNCHRONOUS VEHSUS SYNCHRONOUS TRANSMISSION 1 69
ing unwanted line patterns is to use
"bit
insertion" only when necessary. As an exam-
ple,
the sotuce data over a Tl line could be monitored for all 0's in the first 7 bits of
a time slot. Whenever the 0's occur, a I could be inserted into the data stream as the
eighth bit of the time slot. At the other end of the line the I following seven 0's is re-
moved automatically. Each insertion delays the source data by the time of I bit, but
otherwise the full capacity of the channel is available.
This procedure is directly analogous to the
"zero-bit
insertion" algorithm used in
the High kvel Data Link Control
(HDLC) protocol. In this protocol a specific data
pattern
called a
"flag"
is used to indicate the end of a data block. The transmitter
must
be
precluded from inadvertently sending the flag as part of the source data. The means
of
preventing inadvertent flags is to insert a 0 following a string of five L's in the user
data. Since a flag contains a string of six I's, zero-bit insertion precludes unintended
flag transmissions. The receiving node of an HDLC data link removes any 0 following
five I's. The receipt of six 1's, however, can only be
part of a legitimate flag
(01
I I 1 I l0). If HDLC data are inverted for transmission, it never produces all 0's
(ex-
cept during the idle state).
Although a bit insertion algorithm allows for more efficient use of a channel than
dedicated timing bits, the insertion
procedure has a number of drawbacks. First, this
proce$$ causes the source data to be delayed every time an insertion is made. Hence a
continuous, real-time transmission application
(such as voice) requires smoothing the
arival rate with data buffers at the destination. Second, the bit insertion
process causes
any character $tructure in the user's data to be become unrelated to the time
slot struc-
ture of a time division multiplexed hansmission link. Thus if user data consist of 8-bit
characters
(like PCM voice samples), character boundaries cannot be maintained with
respect to 8-bit time slots in a Tl transmission link.
Date Scrambling
Many digital transmission systems use data scramblers to randomize the data
patterns
on their transmission links. Although these data scramblers are similar to those used
for encryption, the fundamental
purpose of these scramblers is to
prevent the transmis-
sion of repetitive data
patterns, not to encrypt the traffic. Repetitive data pattems gen-
erate line spectra that can be significantly more degrading from an interference point
of view than continuously distributed spectra produced by random data patterns.
Voiceband data modems, for example, are allowed to operate at higher
power levels
if they include scramblers to randomize the data traffic. Also, digital radio systems are
required by the FCC to not ffansmit line spectra, which essentially means that repeti-
tive data patterns mu$t be excluded.
Even when not required, data scramblers are useful in transforming data sequence$
with low transition densities into sequences with strong timing components. Scram-
bling is not used on lower rate T-carrier $ystems
(T1
and obsolete T2) but was used on
the 274-Mbps T4M coaxial hansmission system
[3]
and current optical fiber systems.
Data scramblers
(with
equal input and output bit rates) do not
prevent long sffings
of 0's in an absolute sense. They merely ensure that relatively short repetition
patterns
are transformed to randomized traffic with a minimum density of transitions. If purely
17O DrcrrAL TRANSMtsstoN AND MULT|pLExtNc
random input data are scrambled, the output data are also purely random and therefore
have a certain stati$tical probability ofproducing arbitrarily long strings of 0's. The
probability of a random occurrence, however, is acceptable when compared to the
probability
of nonrandom sequences representing speech pauses
or idle data termi-
nals. To determine which seemingly random data sequence produces all 0's at the out-
put
of a scrambler, apply all 0's to the corresponding descrambler.
The T4M coaxial transmission system used a data scrambler as the basic means of
producing
adequate timing information. This system could tolerate much longer
strings of 0's because the timing recovery circuits in the regenerative repeaters used
phase-locked
loops that maintain timing over relatively long periods
of time. In con-
trast, the original Tl systems recovered timing with tuned circuits that resonated at the
desired clock frequency (1.544
MHz) when excited by a received pulse.
Because the
tuned circuits have lower effective
Q's
than a
phase-locked
loop, the oscillations drift
from the proper frequency and die out more rapidly. Hence the original rl receivers
could not tolerate as long a string of 0's as could T4M receivers. PhaseJocked loop
clock recovery circuits are used on all later generation
wireline and fiber trmsmission
$y$tems so much longer strings of 0's can be tolerated.
Forced Blt Errorg
A fifrh method of maintaining sufficient timing information in the line signals in-
volves having the transmission terminal at the source force an occasional bit error in
order to intenupt a long, transition-free data pattern. If the transition-free lrequences
are long enough and uncommon enough, the intentional bit errors might be less fre-
quent
than random channel errors on the digital transmission link. Thus the intentional
errors may not represent a significant degradation over and above that that already ex-
ists. Nevertheless, forced bit errors are not generally recommended as part
of a line-
coding procedure
but are mentioned in the interest of completeness.
As mentioned previously,
Norttr American channel banks force a bit error in the
second least significant bit of an all-0's transmission code to ensure sufficient signal
transitions. An important aspect of this procedure is that it is performed
by the source
where the significance of the bit error is known. If the transmission link itself inserted
the bit errors, the effects would not be as controllable, particularly
when a variety of
traffic types are being serviced.
A more subtle problem with forced transmission errors arises if the digital trans-
mission link is used for automatic repeat requerrt (ARQ)
data transmission. An ARe
data communication link is designed to provide
error-free transmission, despite ran-
dom channel errors, by inserting redundancy into the data stream and checking the re-
ceived data for error-free reception. If errors are detected, a retransmission is
requested. when the erTors are not random, but forced by the transmission link, the
ARQ system will become frustrated if it ever encounters the reshicted sequence, no
matter how unlikely it is.* once again, if forced errors are used, they should be incor-
porated into the source as part ofthe $ource code restriction process
and not ag a func-
tion of the transmission link.
*If
the transmission link uses a scrambler, the unlikely sequence will not be repeatecl.
4.s UNE coDrNc 171
4.3 LINE CODING
In the preceding section various techniques for establishing timing information are
described in general terms. The choice of any
particular technique is dependent on
the specific line code in use. This section describes the most common
line codes
used for digital transmission and indicates what additional steps, if any, are needed
to maintain synchronization between the transmitter and receiver. Some line-cod-
ing techniques
provide adequate timing information themselves and do not require
any ofthe
procedures discussed
previously. In addition to synchronization require-
ments, other considerations for choosing a line code are the spectrum of the line
code and the available bandwidth
(particularly at low frequencies), noise and in'
terference levels, synchronization acquisition times,
performance monitoring, and
implementation costs.
4.3.1 Level Encodlng
Conceptually, the simplest form of line coding uses a different signal level to encode
each discrete symbol transmitted. Within a computer system the most common
form
of coding is an on-off code using a 3-V level for a
"1"
and near 0 V for a
"0."
Over a
transmission link, however, it is more eff,rcient in terms of
power
to encode binary data
with an equivalent difference in levels but symmetrically balanced about
0 V. For ex-
ample, the average
power required for equally likely +3- and 0-V encodings is 4.5 W
(assuming 1 fl resistance). With +l.5- and
-1.5-V
encodings, the same enor
distance
is achieved with half the
power requirements
(2.25
W). Communications engineers
commonly refer to the unbalanced code as a unipolar code and the balanced code as a
polar code. A representative sequence of binary data and its corresponding bal-
anced and unbalanced level encoding are shown in Figure 4.5. Notice that the level
of each signal is maintained for the duration of a signal interval. For this reason
I
n [ , ] n n
Uni pol er
(unbal anced)
ri gnel i ng
Figure 4.5 Unipolar and polar
(NRZ)
line codes.
172 DIGITAL TRANSMISSIoN AND MULTIPLEXING
Figure 4.6 Direct-current wander of NRZ signal.
the balanced
(polar)
encoding is also referred to as a nonreturn-to-zero (NRZ)
code.*
As indicated in Figure 4.5, an NRZ signal contains no transitions for long strings
of I's or 0's. Hence one of the
procedures
described previously
to en$ure timing tran-
sitions must be employed if NRZ encoding is used on a synchronous transmission
link.
An NRZ line code is a pulse transmission system wherein the pulse (before
filter-
ing) lasts for the duration ofa signaling interval 7. Hence the frequency spectrum (as-
suming random data) of an NRZ code is the sin(x)/x specfrum of Equation 4.1 ancl
shown in Figure 4.2. As indicated, the frequency $pectrum is significantly nonzero at
zero frequency (dc).
Most wireline transmission links, however, do not pass dc signals
by virtue of their being altemating current coupled with transformers or capacitors to
eliminate dc ground
loops. Furthermore, some systems purposely
remove dc compo-
nents from the signal to allow line powering
ofrepeaters or to facilitate single-side-
bard transmission. The elimination of the low-frequency components in the
waveform causes long strings of I's or 0's to decay gradually in amplitude. Hence a
receiver not only would lose timing information during these strings but also would
lose its amplitude reference for optimally discriminating between a
"r"
lever and a
"0"
level. The effect of low-frequency cutof'f; called dc wander, is shown in Figure 4.6 for
a typical transmission sequence. Notice that following the long string of I's, the output
of the link is such that 1-to-0 effors are more likely than O-to-l errors. Similarly, long
strings of0's increase the likelihood ofa O-to-1 error. This problem
arises not only for
long strings of I's or 0's, but whenever there is an imbalance in the number of I's and
0's. Hence periodic
timing pulses
are not sufficient to remove dc wander.
The existence of low frequencies in a random data signal is the basic reason why
modems are needed for data communications over the analog telephone network.
(Analog
telephone circuits do not pass
direct current.) It is also the reason that NRZ
coding is not often used for long-distance transmission. Direct-current wander is not
unique to data transmission systems. It is a phenomenon that must be reconciled in
television receivers, radar receivers, or radiation detectors.
One technique of offsetting dc wander is referred to as dc or baseline restoration
[a].
As illustrated in Figure 4.7, dc restoration involves passing
received pulses
through a capacitor, detecting them, and then removing the charge on the capacitor be-
fore the next pulse
arrives. charge on the capacitor is removed by driving the voltage
*Some
communications theorists ref'et to a balanced two-level code as a
"bipolar
code." The North
American telephone industry, however, uses the term bipolar torefet to a t}reeJcvel code described in thc
next section,
\
4.s L|NE coDrNc 173
Figure 4.7 Direct-current restoration for unipolar pulses.
to a specific threshold
(0
V in Figure 4.7) and then removing the driving voltage when
the threshold is reached. Since all charge on the capacitor is removed after each pulse,
the baseline or decision reference level is constant at the beginning ofa signal interval.
An obvious disadvantage of this technique is that the signal input must have zero arn-
plitude
or be disabled during the reset time.
A generally
more useful technique for overcoming baseline wander is to use deci-
sion feedback, also called quantized
feedback equalization
[5*7].
In contrast to dc res-
toration, which drives the capacitor voltage to a constant, predetermined level,
quantized
feedback compensate$ for dc wander by locally generating the unreceived
low-pass response and adding it to ttre received signal. To accomplish this, the original
data stream is reconstructed. As shown in Figure 4.8, the reconskucted data stream is
passed through a low-pass filter that generates a pulse
equal to the tail or droop char-
acteristic of the channel. The feedback signal adds to the received signal to eliminate
the droop
(intersymbol
interference). Using a frequency-domain analysis, the feed-
back response is complimentary to the channel response.
Quantized
feedback is used
in ISDN basic rate line interfaces
[8].
4.3.2 Bi pol ar Codi ng
The dc restoration techniques mentioned in the preceding
section simplify pulse de-
tection by creating a low-pass pulse response in the receiver. There are numerous line
Switch
on
+ v
t
Figure 4.8 Decision feedback equalization.
174 DGTTAL THANSMtsstoN AND MULTIpLEXtNc
Figure 4.9 Bipolar
(AMI)
coding,
codes that are specifically designed to not contain dc energy and thereby be unaffected
by dc removal. Bipolar coding solves the dc wander problem
by using three levers to
encode binary data. specifically, a logic 0 is encoded with zero voltage while a logic
I is alternately encoded with positive
and negative voltages. Hence the average volt-
age level is maintained at zero-to eliminate dc components in the signal spectrum.
since bipolar coding uses alternate polarity pulses
for encoding logic 1's, it is also re-
ferred to as alternate mark inversion (AMI).-
Bipolar coding is the basic line-coding procedure
used by Tl lines in the telephone
network. Rather than using full-period pulses, however, TI lines use a 50za duty cycle
pulse
to encode each logic 1. Return-to-zero (RZ) pulses (Figure
4.9) were selected to
minimize intersymbol interference and simplify timing recovery in the regenerative
repeaters of a Tl line
[9].
The power spectrum of a bipolar code is obtained from
I
l0]
AS
s(or)='Plc(co)|2
|
-
2(2p
-
l)cos wT + (2p
-
r)z
(4.3)
I
-
cos ror
wherep
=
probability
of a 1
G(rrl)
=
spectrum of an individual pulse
fr\
sin(roTl4)
-
c(oro)=| *
|
j
forSOVIdurycycl epul ses
[rJ
ar/4
Equation 4.3 is plotted in Figure 4.10 for various values of p. For purery random
data, p
=
j. Recall, however, that source coding for p255
PCM codecs produces
more
I's than 0's in the interest of establishing a shong clock signal. Hence the appropriate
value of p for a Tl voice line is normally somewhat larger than 0.5 and depends on
the amplitude of the voice signal. Low-level signals that remain in the first encoding
segment produce
a value of p approximately equal to 0.65. on the other hand, full-
scale sine waves produce
a value for p that is somewhat below 0.5 since most of the
samples occur near maximum amplitude.
Because a bipolar code uses altemating polarities
for encoding I's, strings of l
's
have strong timing component$. However, a string of 0's contains no timing informa-
tion and therefore must be precluded
by the source. The specifications for Tl line re-
*A
mark is a term arising from telegraphy to refer to the active, or l, state of a level encoded transmission
l i ne,
\
4.3 L|NE CoDING 175
oStT
Frgquency
Figure 4.10 Specnal density of bipolar coding.
peaters state that the repeaters will maintain timing as long as no sfing of greater than
f,rfteen 0's is allowed to occur
I1
11. A string of fifteen 0's can only occur if a 0 framing
bit falls between a 10000000 code in time slot 24 and a 0fi)00001 code in time slot l.
Coda Space RedundancY
In essence, bipolar coding u$es a temary code space but only two of the levels during
any particular signal interval. Hence bipolar coding eliminates dc wander
with an in-
efficient and redundant use of the code space. The redundancy in the waveform also
provides other benefits. The most important additional benefit is the opportunity to
monitor the
quality of the line with no knowledge of the nature of the traffic being
transmitted. Since
pulses on the line are supposed to alternate in polarity, the detection
of two successive
pulses of one
polarity implies an error. This error condition is known
as a bipolar violation. No single error can occur without a bipolar violation also oc-
cuning. Hence the bipolar code inherently provides a form of line code
parity. The ter-
minals of Tl lines are designed to monitor the frequency of occurrence of bipolar
violations, and if the frequency of occurrence exceeds some threshold, an alarm is set'
In T-carrier systems, bipolar violations are u$ed merely to detect channel errors. By
adding $ome rather sophisticated detection circuitry, the same redundancy can be used
for conecting errors in addition to detecting them. Whenever a bipolar violation is de-
tected, an effor has occurred in one of the bits between and including the pulses indi-
cating the violation. Either a pulse should be a 0 or an intervening 0 should have been
a pulse of the opposite
polarity. By examining the actual sample values more closely,
a decision can be made as to where the error was most likely to have occurred. The bit
with a sample value closest to its decision threshold is the most likely bit in error. This
technique belongs to a
general class of decision algorithms for redundant signals
called maximum likelihood or Viterbi decoders
[12].
Notice that this method of error
coffection requires storage of
pulse amplitudes. If decision values only are stored, er-
ror correction cannot be achieved
(only error detection)'
An additional application of the unused code space in bipolar coding is to
pur-
posely insert bipolar violations to signify special sifuations such as time division mul-
176 DIGITALTRANSMISSIoNANDMULTIPLEXING
tiplex framing marks, alarm conditions, or special codes to increase the timing content
of the line signals. Since bipolar violations are not normally part
of the source data,
these special situations are easily recognized. of course, the ability to monitor the
quality
of the line is compromised when bipolar violations occur for reasons other than
channel errors.
4.3.3 Blnary MZero Substltution
A major limitation of bipolar (AMI)
coding is its dependence on a minimum density
of 1's in the source code to maintain timing at the regenerative repeaters. Even when
strings of0's greater than 14 are precluded
by the source, a low density ofpulses on
the line increases timing
jitter
and therefore produces
higher error rates. Binary N-zero
substitution (BNZS)
[
l3l augments a basic bipolar code by replacing all skings of N
0's with a special NJength code containing several pulses
that purposely produce bi-
polar violations. Thus the density of pulses is increased while the original data are ob-
tained by recognizing the bipolar violation codes and replacing them at the receiving
terminal with N 0's.
As an example, a three-zero substirution algorithm
(B3zs)
is clescribed. This par-
ticular substitution algorithm is specified for the standard DS-3 signal interface in
North America
[14].
It was also used in the LD-4 coaxial mansmission system in can-
ada
[15].
In the B3ZS format, each string of three 0's in the source data is encoded with either
00v or BOv. A 00v line code consisrs of 2-bit intervals wirh no pulse (00)
followed
by a pulse
representing a bipolar violation (v). A BOv line code consists of a single
pulse in keeping with the bipolar alternation (B), followed by no pulse (0),
rurd ending
with a pulse
with a violation (V).
With either substitution, the bipolar violation occurs
in the last bit position of the three 0's replaced by the special code. Thus the position
'
of the substitution is easily identified.
The decision to substitute
with 00V or BOv is made so that the number of B pulses
(unviolated pulses)
between violations
(v)
is odd. Hence if an odd number of I's has
been transmitted since the last substitution, 00V is chosen to replace three 0's. If the
intervening number of I's is even, BOv is chosen. In this manner all purposeful
vio-
lations contain an odd number of intervening bipolar pulses.
Also, bipolar violations
alternate in polarity so that dc wander is prevented.
An even number of bipolar pulses
between violations occurs only as result of a channel error. Furthermore, every pur-
poseful
violation is immediately preceded
by a 0. Hence considerable systematic re-
dundancy remains in the line code to facilitate performance
monitoring. Table 4.1
summarizes the substitution algorithm.
Example 4.1. Determine the B3zs line code for the following data sequence:
1010001 10000000010001 . use + to indicate a
positive pulse,
-
to indicate a negarive
pulse,
and 0 to indicate no pulse.
4.3 L|NE CODTNG 177
TABLE 4.1 B3Z$ Substitutlon Rulee
Number ot Bipolar Pulses
(1's)
Since Last Substitution
Polarity of
Preceding Pulse
00-
00+
+0+
-0-
Sotution, There are two
possible sequences depending on whether an
odd or even
number of pulses has been transmitted following the
previous violation:
Substitutions
l 0l
Case I (odd):
+0-
Case 2
(even):
+0*
1
000
00-
+0+
J J
Violations
Example 4.1 indicates that the
process of breaking up strings of 0's by substituting
with bipolar violations
greatly increases the minimum density of pulses in the line
code. In fact, the minimum density is 337o while the average density is
just
over 607o'
Hence the B3ZS format
provides a continuously strong timing component.
Notice that
all BNZS coding algorithms
guarantee continuous timing information with no restric-
tions on source data. Hence BNZS coding supports any application in a completely
transparent manner,
Another BNZS coding algorithm is the 862,5 algorithm used on obsolete
T2 trans-
mission lines
[16].
The 8625 algorithm is defined in Table 4.2. This algorithm
pro-
duces bipolar violations in the second and fifth bit positions of the substituted
sequence.
ITTJ recommends another BNZS coding format referred to as high-density bipolar
(HDB) coding
[l7].
As implemented in the El
primary digital signal, HDB coding re-
places strings of four 0's with sequences containing a bipolar violation in the last bit
position. Since this coding format
precludes strings of 0's
greater than three, it is re-
ferred to as HDB3 coding. The encoding algorithm is basically the same
as the B3ZS
algorithm described earlier. Table 4.3
presents the basic algorithm. Notice that substi-
tutions
produce violations only in the fourth bit
position, and successive
substitutions
produce violations with altemating
polarities'
\ \
l l
+ -
000
+0+
- 0-
000
- 0-
+0+
001
00+
00-
000
00+
00-
178 DIGITAL TRANSMISSIoN AND MULTIPLEXING
TABLE 4.2 BZS Subetltution Rulee
Polarity ol Pulse
lmmediately Preceding
Six 0's to be Substituted
Substitution
+
0 - + 0 + -
0 + - 0 - +
0 0 0 0
- o - + )
+ 0 + - )
0 0 0 0
-
+ ) 0 0 +
)
)
T 0 0
Hxample:
1 0 0
+
*
( 0 -
-
+
( 0 +
0 0
+ U
- 0
1 0
+ 0
- 0
1 1
- +
0 0
( u +
( 0 -
0 0
- U
+ 0
0 1
0 -
0 +
0 0
( o *
( 0 -
0 0
+ - )
- + )
A fundamental feature of end-to-end digital connectivity as provided
by IsDN is
64-kbps transparent channels referred to as clear-channel capability (ccc)
[18].
Two
aspects of a bipolar/AMl line code as used on Tl lines preclude
ccc; robbed sig-
naling in the least signilicant bit of every sixth frame and the need to avoid all-O's
codewords on the channel. Bit robbing for signaling is avoided with common-
channel signaling (also
an inherent requirement for ISDN deployment). Two
means of augmenting Tl lines Io allow transparent channels have been developed.
The first method is zero-byte time slot interchange (ZBTSI)
described in reference
il
91. ZBTSI was used for upgrading some Tl facilities for ccc but is not a
godd
long-term solution.
The second method, which is most desirable in the long run, involves the use of a
BSzs line code for Tl lines. As such, the BBZS line code became the new line code
standard for Tl lines with the introduction of D5 channel banks. (D5
channel banks
can also operate with bipolar/AMl line codes for backward compatibility.) As indi-
cared in Table 4.4, the BSZS algorithm is similar to the B625 algorithm in that each
individual substitution is dc balanced. Notice that there are two bipolar violations, in
positions 4 and 7, in every substitution. The purposeful
introduction of bipolar viola-
tions requires replacement of any equipment that monitors all violations for perform-
ance monitoring.
TABLE 4.3 HDB3 Substltutlon Hutes
Number of Bipolar Pulses (1's)
$ince Last Sub$titution
Polarity of
Preceding Pulse
odd Even
+
000-
000+
+00+
-{0-
4.3 L|NE CODTNG 179
TABLE 4.4 BBZS Subetltution Rules
Polarity of Preceding Pulse Substitution
000-+0t.-
000+-G-+
4.3.4 Pair Selected Ternary
The BNZS substitution algorithms described in the preceding section are examples of
selecting codes in the ternary code space to increase the timing content of a binary sig-
nal. Pair selected ternary
GST) I20l
is another example.
The PST coding
process begins by pairing the binary input data to produce se-
quences of Z-bit codewords. These codewords are then translated into two ternary dig-
its for transmission. Since there are nine two-digit ternary codes but only four
two-digit binary codes, there is considerable flexibility available in selecting the
codes. The most useful of the possible coding formats is shown in Table 4.5. This
par-
ticular format not only ensures a strong timing component but also
prevents dc wander
by switching modes to maintain a balance between
positive and negative
pulses. The
codes are selected from one column until a single
pulse is transmitted. At that time the
encoder switches modes and selects codes from the other column until another single
pulse (of
the opposite
polarity) is transmitted.
Exampte 4.2. Encode the following binary data stream into a PST
line code:
0t 001 l 10101 100.
Soluti.on. There are two possible solutions depending on whether the encoder is in
the
positive or negative mode at the beginning of the sequence:
0l 00 11 l 0 10 l l 00
Case I
(+
mode): 0+
-+
+-
-0
+0 +-
-+
Case 2
(-
mode):
-+ -+
+- +0
-0
+-
-+
One
potential drawback of the PST coding algorithm is that the binary data stream
must be framed into
pairs. Hence a PST decoder must recognize and maintain pair
TABLE 4.5 Pair Selected Ternaly Encoding
Binary Input
+ Mode
-
Mode
00
01
1 0
1 1
*+
0+
+0
0-
-0
+-
180 D|G|TAL THANSMISSIoN AND MULTIPLEXING
boundaries. Recognition of the boundaries is not difficult if random data are being
transmitted since a pairwise misframe eventually produces
unallowed codes
(00,
+ +,
--).
Furthermore, time division multiplex formats typically provide character and
pairwise
framing automatically.
The power spechum of a PST line code with equal probabilities
for l's and 0's is
obtained from reference
[20]
and ploted in Figure 4.11. Also shown is the 8625
power spectrum
[16]
and the conventional bipolar power
specffum.
An important point to notice in Figure 4.l l is that bipolar coding and its extensions
require equal bandwidths. Their only significant difference is that B625 and PST have
higher energy levels as a result of greater pulse
densities. The higher energy levels
have the undesirable effect of increasing crosstalk interference in multipair cables.
However, the degradation from the increased crosstalk is somewhat offset by im-
proved accuracy ofthe recovered sample clock (assuming
all signals in the cable are
using the same line code).
4.3.5 Ternary Goding
Since bipolar and PST coding use a temary code space to transmit binary data, they
do not achieve as high an information rate as is possible
with more efficient use of the
code space. For example, an eight-element ternary code is capable ofrepresenting
38
=
656 I different codes. In contrast, I bits of binary dara produce
only 28
=
256 dif-
ferent codes. The previously described line codes do not take advantage ofthe higher
information content of ternary codes; they select codes for their timing content and
spectral properties.
one temary encoding procedure
involves mapping successive groups
of 4 bits into
three ternary (483T)
digits. since binary words of 4 bits require only t6 of the 27 pos-
sible three-digit ternary codewords, considerable flexibility exists in selecting the ter-
nary codes. Table 4.6 presents
one possible
encoding procedure.
Ternary words in the
middle column are balanced in their dc content. Codewords from the first and third
0.5tT ltT
Frequncy
ofbipolar, B3ZS, and PST line codes for equally likely I's and 0's,
Figure4.ll Specftum
4, 3 LI NE CODI NG 181
TABLE 4.6 Encodlng Table for 4B3T Line Code
Ternary Word
(Accumulated Disparity)
Binary Word
0000
0001
001 0
001 1
01 00
01 01
0 1 1 0
0 1 1 1
1 000
1 001
1010
101 1
1 100
1 101
1 1 1 0
1 1 1 1
- - 0
+ + +
+ + 0
+ 0 +
0 + +
+ + *
+ - +
- + +
+ 0 0
0 + 0
0 0 +
- + -
f - -
- 0 0
0 - 0
0 0 -
- 0 -
0 - -
- - +
0 + -
0 - +
+ 0 -
- 0 +
+ - 0
- + 0
columns are selected alternately to maintain dc balance. If more
positive pulses than
negative
pulses have been transmitted, column I is selected. When the disparity be-
tween
positive and negative
pulses changes, column 3 is chosen. Notice that the all-O's
codeword is not used. Hence a $trong timing content is maintained' Because of the
higher information efficiency, however, the ability to monitor
performance is sacri-
ficed, and framing is required on three-digit boundaries.
4B3T coding is used on the
TI48 span line developed by ITT Telecommunications
[21].
This system
provides T-
carrier transmission for two DSI signals
(48
channels) using a bandwidth that is only
50Vo
grcatm than a Tl bandwidth
(carrying 24 channels)' A
generalized discussion
of
4B3T coding and other ternary coding techniques is contained in reference
[22].
4.3.6 Di gi tal Bi phase
Bipolar coding and its extensions BNZS and PST use extra encoding levels for flexi-
bility in achieving desirable features such as timing transitions,
no dc wander, and per-
formance monitorability. These features are obtained by increasing the code space and
nor by increasing the bandwidth.
(The first spectral null of all codes discussed so far,
including an NRZ code, is located at the signaling rate 1/2.)
Many varieties of line codes achieve strong timing and no dc wander by increasing
the bandwidth of the signal while using only two levels for binary data' One of the
most common of these codes
providing both a strong timing cornponent and no dc
wander is the digital biphase code, also referred to as
"diphase"
or a
"Manchester"
code.
182 DtctrAL TRANSMtsstoN AND MULTtpLEXtNc
'FH
IJ
'tfr
Figure 4.12 Digital biphase (Manchester)
line code.
A digital biphase code uses one cycle of a square wave at a particular phase to en-
code a I and one cycle of an opposite phase
to encode a 0. An example of a digital
biphase coding sequence is shown in Figure 4.12. Notice that a transition exists at the
center of every signaling interval. Hence strong timing components are
present
in the
spectrum. Furthermore, logic 0 signals and logic I signals both contain equal amounts
ofpositive and negative polarities.
Thus dc wander is nonexistent. A digital biphase
code, however, does not contain redundancy for performance
monitoring. Ifin-service
performance
monitoring is desired, either parity
bits must be inserted into the data
stream or pulse quality
must be monitored. (A later section of this chapter discusses
performance
monitoring in more detail.)
The frequency spectrum of a digital biphase signal is derived in Appendix c and
plotted
in Figure 4.13, where it can be compared to the speckum of an NRZ signal.
Notice that a digital biphase signal has its first spectral nill atLlT.Hence the exrra rim-
ing transitions and elimination of dc wander come at the expense of a higherfrequency
signal. In comparison to three-level bipolar codes, however, the digital biphase code
has a lower error rate for equal signal-to-noise ratios (see
Appendix C).
Examination of the frequency
specffa in Figure 4.13 shows that the diphase spec-
trum is similar to an NRZ spectrum but translated so it is centered about 1/I instead
of direct current. Hence digital biphase actually represents digital modulation of a
square wave carier with one cycle per
signal interval. Logic I
's
cause the square wave
to be multiplied by +1 while logic 0's produce multiplication
by
-1.
Diphase is pri-
1/T ztr 3tT 4tT
Ftgure 4.13 Power spectral density of digital biphase.
\
\
\
\
4.3 LINE CODING
183
marily used on shorter links where terminal costs are more significant than bandwidth
utilization. The
"Ethemet"
IEEE 802.3 local area data network uses digital biphase
(Manchester) coding.
4.3.7 Dlfferential Encoding
One limitation of NRZ and digital biphase signals, as presented up to this
point, i$ that
the signal for a 1 is exactly the negative of a signal for a 0' On many transmission
me-
dia, it may be impossible to determine an absolute
polarity or an absolute
phase refer-
ence. Hence the decoder may decode all I's as 0's and vice versa. A common remedy
for this ambiguity is to use differential encoding that encodes a I as a change of state
and encodes a 0 as no change in state. In this manner no absolute reference is necessary
to decode the signal. The decoder merely detects the state of each signal interval and
compares it to the state of the
previous interval. If a change occured, a I is decoded.
Otherwise. a 0 is determined.
Differential encoding and decoding do not change the encoded spectrum of
purely
random data
(equally likely and uncorrelated
I's and 0's) but do double the error rate.
If the detector makes ar effor in estimating the state of one interval, it also makes an
error in the next interval. An example of a differentially
encoded NRZ code and a dif-
ferentially encoded diphase signal is shown in Figure 4.14. All signals of differentially
encoded diphase retain a transition at the middle of an interval, but only the 0's have
a transition at the beginning of an interval'
4.3.8 Coded Mark Inversion
A variety of line codes have evolved similar to the digital biphase code described
pre-
viously. One of these is referred to as coded mark inversion
(CMI) in ITU recommen-
dations
[l7l.
CMI encodes I's
(marks) as an NRZ level opposite to the level of the
o l 0
Figure 4.14 Differentially encoded NRZ and diphase signals:
(a)
differentially encoded
NRZ;
(b)
differentially encoded diphase.
(d)
DIGITAL TRANSMISSION AND MULTIPLEXING
l o l ' l '
l 0 l o I ' l o
I
J l
I
Figure 4.I5 Coded mark inversion.
184
previous
one and 0's as a half-cycle square wave of one par"ticular phase.
Figure 4.15
shows a sample encoding for cMI. There is no dc energy in the signal and an abun-
dance of signal transitions exist as in diphase. Furthermore, there is no ambiguity
between I
's
and 0's- Elimination
of the ambiguity actually leads to a major drawback
of cMI coding; Its error pedormance
is 3 dB worse than diphase.
*
cMI is the specified
interface code for the fourrh-level (Bl)
ITU multiplex signal at 139.264 Mbps. cMI
is also used for electical interfaces of soNET sTS-3c signals described in chapter
g.
4.3.9 Mul tl l evel Si gnati ng
In the line codes discussed so far, two-lever (binary)
signaling has been assumed. In
applications where the bandwidth is limited but higher data rates are desired, the num-
ber of levels can be increased while maintaining the same signaling rate. The data rate
R achieved by a multilevel system is given
as
(4.4)
where .L
=
number of levels that can be freely chosen during each interval
I= signaling interval
The signaling rate llT is often referred to as the symbol rate and is measured in
bauds. within the data communications industry it is common practice
to use
..baud"
as being synonymous with bit rate. strictly speaking, however, the bit rate is only
equal to the baud rate if binary signaling (l
bit per
signal interval) is used. Figure 4. I 6
shows an example of an eight-level transmission format that achieves 3 bits per signal
interval (i.e.,
3 bits per baud).
Multilevel transmission
systems achieve greater
data rates within a
given
band-
width but require much greater
signal-to-noise
ratios for a
given
error rate. One aspect
of wireline transmission that favors multilevel line coding is the lower baud rate for a
given
data rate, which in turn reduces the crosstalk. Hence, in crosstalk-limited
svs-
*The
enor performance
of CMI is 3 dB worse than diphase when bit-by-bit detection is used. The
inetficiency arises because for one-half of an intewal a I looks like a 0. Because CMI has redundancv.
some of the inefficiency can be recovered with maximumlikelihood
ffiterti) tletection.
R
-
ros,(r)
F)
4.3 LINE coDtNc
185
Figure 4.16 Multilevel transmission wittr 3 bits
per signal interva''
tems the signal-to-noise
ratio penalty of a multilevel line code is not as significant. The
TlG T-carrier system developed by AT&T
[23]
uses a four-level line code at the TIC
baud rate
(3.152
MHz) to double the capacity of a TlC $ystem
(from 48 channels to
e6).
Another example of multilevel transmission of pafiicular significance
is the ISDN
basic rate digital subscriber line
(DSL), which uses fbur-level transmission at a $i8-
naling rate of 80 kilobauds to achieve 160 kbps. The primary factors that led to select-
ing a multilevel line code in this application are
( I
)
near-end crosstalk that
cannot be
eliminated by
pair isolation* as in Tl systems and
(2) high levels of intersymbol in-
terference caused by bridged tap reflections. Both of these factors are easier
to control
when lower frequency signals are used
t241.
Additional aspects of this application are
described in the ISDN section of Chapter 11.
4.3.10 Partial-Response $ignaling
Conventional
bandlimiting filters of a digital transmission system are designed to re-
strict the signal bandwidth as much as possible without spreading
individual symbols
so they interfere with sample values of adjacent intervals. One class of signaling tech-
niques, variously referred to as duobinary
[25],
correlative level encoding
[26]'
or
par-
tiai-response signaling
[27],
purposely introduces a prescribed amount of intersymbol
interference that is accounted for in ttre detection circuitry ofthe receivers. By over-
filtering the encoded signal, the bandwidth is reduced for a
given signaling rate, but
the overlapping
pulses produce multiple levels that complicate the detection
process
and increase the signal
power requirements for a given error rate'
Figure 4.17 shows the
pulse response of a typical
partial-response system. If the
channel is excited by a pulse of duration T, channel filters
(defined in Appendix c)
limit the spectrum such that the main
part of the pulse extends across three signal in-
tervals and contributes equally to two sample times. The reason for the term partial
response is now apparent: The output only responds to one-half the amplitude of the
input.
*Pair
isolation involves separating
go and return
pairs into separate cables'
186 DIGITAL TBANSMISSION AND MULTIPLEXING
t
\\N6xt
output
puhe
Figure 4.17 Output pulse of partial-response
channel.
If the inputpulse of Figure 4.17 is followedby
another pulse
of the same amplirude,
the output will reach full amplitude by virtue of the overlap between pulse$.
However,
if the next input pulse
has a negative polarity,
the overlap produces
zero amplitude at
the sample time. Thus, as shown in Figure 4.18, a
partial-response
system with two-
level inputs (+1,
-I)
produces
an ourput with three levels (+l
,
0,
-l).
In similar fash-
ion, Figure 4.19 shows a system with four input levels ( +3, +1,
-1, -3)
and seven
output l evel s (+3,
+2, +1
,0,
-1, -2, -3).
Detection of a partial
response signal (PRS)
is complicated by the additional levels
produced
by the channel filters and the fact that sample values are dependent on two
adjacent pulse
amplitudes. One method of detecting a PRS involves subtracting the
overlap of a previously
detected pulse
from the incoming signal to generate
a differ-
ence signal representing the channel response to the new pulse.
only the overlap (in-
tersymbol interference) at the sample times needs to be subtracted. The difference
samples are nominally equal to one-half the amplitude of the unknown input pulse.
This technique doubles the error rate in the same manner that differentiallv encoded
error rates are doubled.
Another method of detecting partial-response
systems involves a technique called
precoding
at the source. hecoding transforms the input data in such a manner that the
output level at the detector directly indicates the original data without comparison
to
the previous
sample value. In a binary system, for example, a I is encoded with a pulse
of the same polarity
as the previous pulse.
Hence logic I
's
show up at the detector as
either a +l or a
-l
sample. Transmission of a 0 is encoded with a
pulse
of opposite
polarity
to the previous pulse. Hence logic 0's always show up at the detector as a
zero-level signal. similar precoding
techniques exisr for multilevel systems
[zg].
Figure 4.I8 TfueeJevel partial-response
inputs and ourpurs.
4.s LINE coDtNe 187
Figure 4.19 Seven*level
partial-response inputs and outputs'
The partial-response technique
just
described is actually a special case of a more
general class of signaling techniques referred to as correlative level encoding
[26]'
To
describe the properties of
generalized partial-response or correlative level systems,
it
is convenient to introduce a delay operator
D to denote delay equal to one signal in-
terval Z. Physically,
D can be thought of as a delay line of length ?- Two units of delay
is implemented with two delay lines in series and is denoted as D2'
Using this notation, the
partial-re$ponse $y$tem described above is referred to as
l+D PRS: The output represents the superposition
of the input with a delayed version
of the same input. Other forms of overlap are
possible. These systems do not neces-
sarily
produce the overlapping
pulses by over{iltering
an input' An alternative ap-
proach is to overlap and add the
pulses directly in the encoding
process
(hence the term
correlative level encoding). An interesting special case of correlative level encoding
is the I
-
D system shown in Figure 4.20 and extended in Figure 4.21 to
show the ef-
fect of overlapping
Pulses'
The I
-
D encoder uses a single cycle of a square wave across two signal intervals
to encode each bit. Since neither ofthe two individual signals
(
+- or
-+)
produce dc
energy, the encoded signal has no dc wander. Furthermore, the
positive and negative
leveis of the composite signal alternate
in a manner reminiscent of bipolar coding'
In fact, if differential encoding is used
(i.e., if a 0 is encoded with the same
phase
widrh
of one
signal
interurl
I
Overlap
i nto
sdiEcent
interval
#-r
Logic I
Figure 4.20 Individual signal elements for I
Logic
0
-
D correlative level coding'
188 Dt cI TALTRANsMt sst oNANDMULTt pLEXt NG
Figure 4.21 Representative
waveform of I
-
D correlative level encodins.
as the previous
interval and a I as the opposite phase),
this form of encoding is iden-
tical to bipolar coding (assuming
the NRZ levels are replaced by s}vo RZ pulses).
Thus I
-
D correlative level encoding is used to shape the bandwidth rather than
to limit it. The spectra of unfiltered I + D, I
*
D, and |
-
Dz signals are obtained
from referencel2Tl
and plotted
in Figure 4.zz.TheTlD T-carrier
system of AT&T
uses precoded
I + D level encoding (also
called duobinary).
The spectrum of the 1
*
D2 signal is particularly
interesting. Ithas no dc componenr
and an upper limit equal to ll2T: the same upper limit as a maximally filtered NRZ
code with no intersymbol interference. The I
-
D2 partial-response
systems have been
Frequency
{Hzl
Figure 4.22 spectra of unfiltered I + D, I
-
D, ard I
-
Dz correlative encoded sisnals.
/
l/
\
\t
4.4 ERHoR PERFoHMANcE
189
used by GTE in a modified T-carrier system
providing 48 voice channels
[29]
and for
digital transmission
over analog microwave radios using very low baseband frequen-
cies for
pilots and service channels
(where no messaBe spectrum exi$ts)
[30]'
The ob-
solete lA-RDs
(data-under-voice) system of AT&T also used |
-
Dz coding with
four-level inputs to
get 1.544 Mbps into 500 kHz of bandwidth
[31]'
Notice that the I
-
Dz spectrum is essentially the
product of the I + D and the I
-
D spectra. Indeed, aI
-
D2 system can be implemented
by coxcatenating
a I
-
D en-
coder with a I + D channel response:
(1
*
D)0 + D)
=
|
-
Dz'Thus conelative level
system
polynomials are very useful for simultaneously
representing correlations and
spectrum shaping. Since all correlative level systems
use more levels than necessary
for encoding the data, they are inferior to uncorrelated or full-response systems in
terms of error
perfofinance'
A summary of digital transmission
systems identifying various line codes is pro-
vided in Table 4.7. Other line codes used specifically in fiber systems are described in
Chapter 8.
4.4 ERROR PERFORMANCE
The preceding sections of this chapter emphasized the timing and
power specffum re-
quirements o1 various transmission codes. Another fundamental
consideration
in
choosing a line code is the error rate performance in the
pre$ence of noise. Except on
relatively shofi lines, where noise may be insignificant, error
performance require-
mentrt can impact system cost significantly.
If a celtain minimum elror rate is speci-
fied, those coding schemes
providing the desired error rate at lower signal-to-noise
ratios allow regenerative repeaters to be spaced farther apart, thereby reducing instal-
TABLE 4.7 Digital Transmiselon Systems
CountrY or
Designation Administration
Bit Rate
(Mbps) Line Code
Repeater
Media
SPacing
T1
E1
T1C
T148
9148A
T. I D
T1G
T2
LD.4
T4M
AT&T
ITU-T
ATAT
ITT
GTE
AT&T
AT&T
AT&T
Canada
AT&T
1. 544
2.048
3, 152
2.37, ternary
3. 152
3. 152
6.443
6, 312
274.176
274.176
AMYBEZS
HDBs(B4ZS)
Bipolar
4B3T
1
-D2,
duobinary
1 + D ,
duobinary
Four-level
B6ZS
B3ZS
Polar
Twisted
pair
Twisted
pair
Twisted
pair
Twisted
pair
Twisted
pair
6000 ft
2000 m
6000 ft
6000 ft
6000 fr
Twisted oair
6000 ft
Twisted
pair 6000 ft
Low-capacitance
4800ft
twisted
pair
Coax
1900 m
Coax binary
(NRZ) 5700 ft
190 DtctrAl TRANSMlsstoN AND MuLTtpLEXtNc
lation and maintenance.+ Repeater spacing is an important
economic factor in wireline
transmission, in optical fiber transmission, and on point-to-point
radio links.
The error rate results and comparisons presented
in this section are based on white
Gaussian noise. This is the most coilrmon form of noise and the best understood. In
some applications, as within the older electromechanical
telephone network, impulse
noise may be more prevalent.
Thus these analyses do not provide
a complete error rate
analysis for some applications. The intent of this section is to
present
relative error per-
formance comparisons of various line-coding
techniques. For this purpose
a white
noise analysis is most appropriate. If impulses are large enough, they cause enors in-
dependent of the coding $cheme in use.
For the most part,
the following sections present
only the results of the error analy-
ses in the form of graphs
of error rate as a function of signal-to-noise ratios. Appendix
C derives the basic equations used to produce
the results.
4.4.1 Si gnal Detecti on
Invariably,
the detection circuiury of a digital receiver processes
incoming signal
waveforms to mea$ure each possible
discrete signal. In most cases the mea$ures are
nothing more than samples of a filtered receive signal. Depending
on the signal shape
and the level of performance
desired, the receiver uses more sophisticated piocessing.
In any case, the end measurement
of a binary signal nominally produces
one votmge
level for a 0 and another voltage level for a I. A decision of which signal was trans-
mitted is made by comparing the measurement (at
the appropriate timeJ to a threshold
located halfway between these nominal voltages. Naturally,
the error probability
de-
pends
on the nominat distance between the voltages and the amount of fluctuation in
the measurements caused by noise.
Since signal mea$urement$
are normally linear in nature, the error distance between
I
'
s and 0's is proportional
to the received signal voltage. Equivalently,
the amount of
noise power
required to produce
an error is a direct function of the signal-to-noise ra-
tio. The optimum detector for a
particular
signal set maximiees the signal-to-noise
ra-
tio at the input to the decision circuit.
4.4.2 Nolee Power
white Gaussian noise is characterized
as having a uniform frequency spectrum across
an arbitrarily large bandwidth and an amplitude distribution that varies according to a
normal (Gaussian)
probability
distribution. A parameterN,l
conventionally
represents
the power
specfral density of white noise and is the amount of power
measured in a
bandwidth of I Hz. Hence
the rms power
of white noise cominf out of a filter with a
bandwidth BW is (Nr)(BW).t
'On
Tl lines the rcpeater spacing was predetermined
by the locations where loading coils needed to be
.
removed. Thus the error rate performance
essentially determined the maximum oata iate.
TThe
power
spcctral density of white noise is also specifred *
fro
r* " ir"**ia*a rfr-t4 oensity. as a
praclical
matter, there is no difference in the specificarions,'since
a rcal filter has the mathematical
equivalent of identical positive-
and negative-frequency
bands. Thus the measured power
corDng through
a filter with a one-sided (positive-frequency)
bandwidth BW is No BW in either case.
4.4 ERROR PEHFORMANCE
I91
To determine the amount of noise
power present at a decision circuit, it is necessary
to determine ly'p and the effective bandwidth of the detection circuitry' When the de-
tection circuiky consists of nothing more than a filter, the effective bandwidttr
is usu-
ally very close to the 3 dB bandwidth
of the filter. With other more sophisticated
detectors. the effective bandwidth can be more difficult to determine'
As derived in
Appendix C, the effective bandwidth is usually referred to as the effective
noise band-
*idth
(FIBW)
of the receiver. Hence the noise power at the decision circuit is
(No)
(NBW).
4.4.3 Error Probablllties
An enor in detection occurs whenever noise cau$es the signal measurement to cross
the threshold between the two nominal output levels of the detection circuitry'
Appen-
dix C derives this
probability for white Gaussian
noise as
nf
rzd
4,
(4.s)
where v
=
nominal distance
(voltage) to a decision threshold
02
=
noise
power at detector,
= (No)NBW)
Equation 4.5 is nothing more than the area under the
probability density
function
of a normal distribution. As shown in Figure 4'23, the equation represents the error
probability as the
probability of exceeding v/o standard deviations in a normal distri-
bution with zero mean and unit variance N(l' 0).
The error rate is completely determined by the ratio of v to o. Since v is the noise-
free sample voltage and o2 is the rms noise
power, v?/O2 is a signal-power*to-noise-
po*"r rutio at the detector.
This ratio is sometimes refened to as a
postdetection SNR,
iince it is measured after the detection circuitry. It is usually more important to express
error rates in terms of an SNR at the input to the receiver. Figure 4'24 depicts a basic
channel. a basic detection model, and the relation between a
predetection SNR and a
postdetection SNR. For reasons cliscussed
in Appendix C, the most appropriate
pre-
detection SNR for comparing line codes and digital modulation formats is an energy-
Dl o
l *
Figure 4.21t Probability of error for binary signaling.
, ;
\ I I
DfOD(erroD
=
-:--
|
l VZ?t6 "
v
192 DGITAL TRAN$MIssIoN AND MULTIPLEXING
F#
=(;rs)
(';m;) (*)
Noire
Epctral
denrity Ns
v1
i
Equi val Bnt
noi* bendwidth
:
NBW
Noi* power
=
o' = NBI / I ,
' No
Figure 4.24 Signal detection model.
per-bit-to-noise-density
ratro E6/N0. The relationship between E6/N0Nrdthe
signal-
power-to-noise-power
ratio is
g111p:
siEpal
Power
norse power
_dEs1t r )
No NBW
dErlog2l (llT)
(4.6)
NoNBW
where d= pulsedensity
E
=
energy per
symbol
Ea= energyperbi t
logy L
=
number of bits per
symbol (i.e.,
t
=
number of levels)
l/7= signaling rate
NBW
=
effective noise bandwidth of receiver
In determining the signal power
in Equation 4.6, notice the dependence on the
pulse
density d. In a balanced NRZ line code the pulse
density is I, but in many of the
other line codes the pulse
density is dependent on the data and any substitution
algo-
rithms that may be in use. In these codes, increasing the prrlse
density increases the
sNR but does not reduce the error rate. The error rate is determined by the energy-
per-bit-to-noise-density
ratio. In fact, increasing the pulse
density might worsen the
error rate ifintederence
between cable pairs is a significant consideration.
(Interfer-
ence is directly proportional
to the signal power.)
Antipodal
Signaling
The optimum signaling format for binary signaling maximizes the error distance for
a given received
signal power
and simultaneously
minimizes the noise bandwidth.
This condition arises only
when two signal levels are allowed and only when one sig-
4.4 ERROR PEHFOBMANCE
193
nal is the exact negative ofthe other. Since the signal for a I is the exact opposite
(the
antipode) of the signal for a 0, optimum signaling is often referred to as antipodal sig-
naling.Since
no other bit-by-bit signaling scheme is any better, antipodal
perfortnance
is oftin used as a basis for comparisons. Ofthe line codes described
previously, only
balanced twolevel
(NRZ) encoding and digital biphase can
provide antipodal
per-
formance. Figure 4.25 shows the optimum error
performance provided by antipodal
signaling as a function of E/Ne and the SNR.
Error Bate of Level Encoded $ignale
As already mentioned, a balanced, two-level
line code is capable of
providing opti-
*u* "ooi rate
performance. If an unsymmetric
level code is used, such as unipolar,
the same basic detector is used as in symmetric
level encoding. The only difference is
that the decision threshold must be moved from zero to half the amplitude of the
"on"
signal. To maintain the same error distance, the average transmit
power is increased
by a factor of 2. Hence a unipolar code carries a performance
penalty of 3 dB with re-
spect to antipodal
Performance.
Figure 4.25 shows the ideal
performance of a unipolar
(on-oft) code
where it is
compared to antipodal
performance. Notice that for all SNRs the enor rate of the on-
off system is exaitly equal to the error rate of an antipodal system with 3 dB less SNR'
Bipolar Slgnaling
With respect to error
performance, bipolar signaling is basically identical to a unipolar
code. During any
particular signal interval, the receiver must decide between one of
two possible levels: zero or a
pulse with the appropriate
polarity. Hence
the decision
threshold
pertinent to a
particular signal interval lies halfway between
zero and the
amplitudelevel of the allowed
pulse. Nanowing the
pulse to 507o of the signal interval
does not change the theoretical error
performance
(with respect to average
pulse en-
ergy). Thus the error rate curve for on*off keying in Figure 4'25 can be used to deter-
mine theoretical bipolar effor rates.
One consideration
in a bipolar line code contributes to a slightly higher enor rate
than in on-offkeying.
This increase occurs because both positive and negative noise
can cause an erroneous threshold crossing when a zero-level signal is transmitted.
In
contrast, a unipolar code is affected only by positive noise when a lower level signal
is transmitted and only by negative noise when the upper level signal is transmitted'
If the bipolar detector ffeats the elroneous
pulse as a I
(despite a bipolar violation)'
the erroiprobability when 0's are transmitted
is doubled. Thus the overall error prob-
ability is increased by 50Vo if 0's and I's are equally likely.
Because ofthe steepness ofthe curve, an increase in the error rate of 50Vo does not
represent much of a
performance penalty. For example, if the error rate is increased
from 1 x l0-6 to 1.5 x 10-6, the source
power of a line code needs to increase by only
0.2 dB to get back to a I x 10-6 enor rate. At higher enor rates, a larger
penalty occurs
because the curve is not as steep. These effects are demonstrated in Figure 4.25, where
the ideal performance of a bipolar code
(with 507o 0's) can be compared to the per-
formance of a unipolar code,
194 DIGITAL THANSMISSION
AND MULTIPLEXING
SisElll|olrlr-lo-nqllF-pffi rrfi q
LdE)
8 9 l 0 l t 1 2 t 3
En rlv-FoFbit-b{ol$dilrity
rsflo,
4/IVo
(dB)
Figure 4.25 Enor rates of polar (NRZ),
unipolar, and bipolar line codes.
E
E
>
ll
F
6
l o-t F
2 3 f E 8 7 B S r 0 l t 1 2 1 3
4.4 ERROH PEBFORMANCE 195
The fact that bipolar coding incurs a 3.2-dB
penalty (at 10-6 error rate) with respect
to digital biphase is indicative that timing and dc wander
problems are solved by in-
*r*uring the number of signal levels. In contrast' digital biphase incurs a bandwidth
penalry. Not all of the 3.2-dB
penalty for bipolar coding, and its extensions
BNZS and
PST, "an be attributed to removing direct current and adding a consistent
timing
pat-
tem. Bipolar coding contains considerable
redundancy for performance monitoring or
possible error correction.
*
Example 4.3. Assume each section of a T1 transmission system on Z2-gauge cable
is near-end crosstalk limited and operating with a 10-6 error rate. What design changes
are needed to reduce the error rate to 10-8?
Solution. Since the system is crosstalk limited, the enor rate cannot be improved
by increasing the
power out of the regenerative
repeaters. The solution is to space the
repeaters closer together
(ignore the impractical aspects ofthis solution). From Figure
+.i5 it can ue determined that the signal
power must be increased by I '6 dB to improve
the error rate from 10-6 to l0-8. In Figure 4'10 it is seen that the main
lobe of the
bipolar spectrum extends up to 1.544 MHz. However' most of the energy in the
spectrumlies below 1 MHz. Using Figure L14, we determine that the attenuation of
Z-Z-gauge cable is 5 dB/kft at 1 MHz. Hence 1.6/5
=
'32 kft or 320 ft is the reduction
in repeater spacing required.
Example 4.3 demonstrates
a number of important aspects of digital
transmission
systems. iirst, arbitrarily
good transmission
quality can be obtained with only a small
plnalty in transmit
power or repeater spacing.t
Thus, as mentioned
in Chapter 2' a
digitaltransmission
and switching network can be readily designed to impalt no deg-
rJation to voice
quality, except where analog-to-digital
and digifal-to-analog
conver-
sions take place.
Second, the dramatic improvement
in error rate for a relatively small increase in
the SNR implies extreme sensitivity in the opposite direction also. A slight increase
in noise
power or signal attenuation
would cause a large increase in the error rate.
Hence the nominal design of a digital link often
provides considerably better
perform-
ance than normallY necessarY.
Third, the solution to Example 4.3 applies to any digital transmission format using
the same band of frequencies and, hence, the same attenuation
per mile.
Since all line
codes have approximately
the same steepness at a 10-6 error rate, the relative change
in SNR is the same for all systems. Thus a Tl system with suboptimum
detection
would exhibit the same
performance improvement
if the repeaters were
moved 320 ft
closer together.
(Tl
lines are designed to
provide less than a 10-6 error rate to begin
with. See reference
[32]
for a thorough repofl of error rate surveys on Tl lines')
"In
terms of logicJevel decisions, a bipolar code carrrot
provide effor conection, If a pulse is removed to
eliminate abipolar violation, half of the time the efior is correctedbuthalf
of the time another error is made.
Enor corection is
possible only with a Viterbi-like detectot described
previously. In this case a bipolar
code orovides
better
performance
than a unipolar code'
tsome
systems, such as the tadio systems discussed in Chapter 6, require a greater decrease ifl repeatel
spacing to achieve the same improvement in performance'
196 DIGITAL TRANSMISsIoN AND MULTIPLEXING
Arrsrrgl snrqy-pFblt-to-ndire-dondty
iltio, Ebftro
(dB)
Figure 4.26 Error rate of balanced multilevel signals (all
systems providing
an identical data
rate).
E
t
s
I
8 9 t 0 t l 1 2 1 3 t 4 r b 1 8
8 e l 0
Figure 4.27
4.4 ENHOH PERFORMANCE 197
t l 1? l 3 14
I D r u
EulVo ort tlrtdunncl
Error rates of I + D partial-response systems.
198 DG|TAL TRAN$MtsgtoN AND MULT|PLEXING
Multllavel Error Eatee
The multilevel kansmission system shown in Figure 4.16 does not require bandwidth
in excess of a binary system using the same signaling rate, yet it achieves three times
the data rate. The penalty
for multilevel transmission comes in the form of greatly in-
creased signal power
for a given
error rate. For example, the average power of the
eight-level system in Figure 4. l6 is 8.7 dB above the average power
of a symmetric
two-level system with the same error distance. To make matter$ worse, some sysrems
are peak power limited, in which case the eightJevel system has a 12.4-dB disadvan-
tage with respect to a two-level system. The error rates of multilevel systems are de-
rived in Appendix c and plotted
in Figure 4.26 as a function of E6/N11.For those cases
where the peak power
is of interest, the relationship between the peak and average
power
of a multilevel sy$tem is derived in Appendix C as
Peak-to-average(dB)=
l 0l og,o
t.
tl t
(u L)E'/:
At
-
rf
(4'7)
where I is the number of equally spaced levels centered about zero
[e.g.,
+1, +3, +5,
. . . , J
( r
- 1) 1.
The enor rates of I + D partial-response
rrystems are also derived in Appendix c
and plotted
in Figure 4.27. These error rate$ are derived under the assumption that bir
by-bit detection is used. Since partial-rerrponse
systems contain redundancy (conela-
tion) in adjacent samples, better performance
cal be achieved with Viterbi decoders
l l
21.
4.5 PERFORMANCE MONITORING
Two basic techniques exist for directly monitoring the quality of a digital transmission
link: redundancy checks and pulse quality
measurements. Both techniques are de-
signed to provide
an indication of the bit error rate (BER)
of the channel.
4.5.1 RedundancyChecks
Redundancy can be incorporated into a digital signal using one of two common meth-
ods. First, the line code itself may contain redundancy as in bipolar coding. In a ran-
dom, independent
error environment the frequency
of bipolar violations is very nearly
equal to the channel BER (except
for extremely high BERs). Second, logic-level re-
dundancy can be insefied into the data stream in the form of overhead bits. For exam-
ple, parity
bits are inserted into DS3 and DS4 signals for the expre$$ purpose of
monitoring the channel error rate. (The
frame structures of these and other higher level
multiplex signals are
provided
in chapter 7.) cyclic redundancy check (cRC)
codes
are also incorporated into a number of transmission systems as a mean$ of monitoring
BERs and validating framing acquisition. Two prevarent
examples of cRC use are
(l)
extended superframe (ESF)
on T I lines
[33]
introduced with the D4 channel banks and
(2)
optical fiber transmission using the soNET standard described in chapter
g.
4.5 PERFORMANCE MONITORING 199
The use of logic-level redundancy
(either parity bits or CRC codes) is generally
better than the use ofline code redundancy
(such as bipolar violations) because the lat-
ter is dependent on the technology of the transmission link itself. Logic-level redun-
dancy, on the other hand, is unaffected by a change from cable pairs to fiber or radio.
As described later in this chapter, a major impetus for the introduction of ESF into T I
transmission systems was to
get performance monitorability independent of bipolar
line codes. DS I signals are often multiplexed into higher level digital signals and car-
ried on a variety of ftansmission systems such as fiber and radio. The CRC codes in
the ESF framing format
provide the means for end-to-end
pedormance monitoring in-
dependent of whatever transmission systems carry the data stream.
In contrast to line code redundancies,
pafity bits and CRC codes do not provide a
one-to-one indication of the BER. The following equation relates the
parity error rate
(PER)
to the channel BBR:
pER
=
I [T]p(r
*
p)N-i
(i odd)
- t ' J
t=l
"
(4.8)
where N
=
length of a parity field (number of bits over which
panty is generated)
p
=
BER assuming random, independent errors
The relationship between the PER and the BER is plotted in Figure 4.28 for DS3
and DS4 signals. Notice that at low error rate$ the PER is essentially N times the BER
(Np). At high error rates, however, this relationship changes because any odd number
of errors in a frame is indistinguishable
from a single error.
'\ilhen
the BER is high
enough that more than one error in a parity field is likely, the PER is useless as an ab-
solute estimate of the error rate. In these cases the PER indicates only that the BER is
above a threshold approximately equal to 1/N. Because a DS4 frame format contains
a higher density of
pariry bits, DS4 signals can be measured for higher error rates than
can DS3 signals.
Determination of a CRC error rate is simplified because it is unlikely for multiple
errors to not
produce a CRC error. Thus the probability of a CRC enor
(CRCER) is I
minus the
probabiliry that no errors occur;
CRCER=1- ( l - p) N
(4.9)
where N
=
length of the CRC field
(including CRC bits)
p
=
BER assuming random, independent enors
Again, at low error rates Equation 4.9 reverts to N times the BER
(Np). Equation 4.9
is plotted in Figure 4.28 for ESF frames where the CRC fleld is 4614 bits.
(The CRC
field does not include F bits in Table 4.8 except the C bits themselves.)
To have COnfidence in enOr rate measurements, the sample $equence mUSt be long
enough to allow an average of about 10 enors in the sample size
(e.g., the sample size
d
:Y
E
o
g =
0
. $
CN
(4
rTl
t a
b F
F ;
s ;
f E
o v
S E
m e
bI)
a d
& a
d
U}
a
E
t h
t q
E}
Fa
' -
ff
(b
F!
+
{)
b!
EI
.f
b
t
b
q
tr
eler r0r,$
cHs#til-|ed
4.5 PEBFORMANCE MONITORING 201
TABLE 4.8 Extendsd Superframe Framlng Channel Formata
F-Bit Assignment
ESF Frame
Number ESF Bit Number FPS FDL cRc
1
2
3
4
5
6
7
I
s
1 0
1 1
1 2
1 3
1 4
1 5
1 6
1 7
1 B
1 9
20
21
22
23
24
0
193
386
579
772
965
1 158
1 351
15,14
1737
1 930
2123
231 6
2509
27Q2
2895
3088
3281
3474
3667
3860
4053
4246
4439
0
0
I
0
1
1
m
m
m
m
m
m
m
m
m
m
m
m
cB1
c82
c83
c84
cB5
;
aFPS,
frami ng
pattern
sequonc
(. . .001011. . .); FDL, a kbps faci l i ty data l i nk (message bi ts m); CRC' GRC' 6
cyclic redundancy check
(check bits CBl-CB6).
must be I0/BER). Hence, when trying to measure low BERs (e.g., 10*6 or l0-7), the
measurement time may be too long to respond to changing channel conditions such as
radio channel fading.
4.5.2 Slgnal Quality Measurements
The second basic technique for monitoring digital ffansmission
quality is to process
the digital signal directly and measure ceftain
properties related to lhe
error rate. A
simple approach involves merely measuring the received signal
power, a coillmon
technique in analog systems. In a fixed-noise environment this approach is adequate.
However, on transmission links where the noise level can vary or where signal distor-
tions can arise, the quality of the
pulses themselves must be measured.
202 DIGITAL THANSMISSION AND MULTIPLEXING
Docirion threshold Pmudoerror rcgion
Figure 4.29 Pseudoerrordetection.
Figure 4.29 demonstrates the operation of a'!seudo" error detector designed to de-
tect received pulses with abnormal amplitudes at the sample times. In the example
shown, binary data is detected by use of a single threshold located midway between
the normal pulse
amplitudes. Two additional thresholds are included to detect the
presence
of
pulses
with abnormal amplitudes. Sample values falling into the central
decision region a.re not necessarily data enors, but a high pseudo-error
rate is a
good
indication that the channel is not performing properly.
In a random (Gaussian)
noise environment the occuffence rate of pseudo*effors
is
directly related to the actual error rate. Figure 4.30 shows a Gaussian noise distribution
and decision thresholds chosen to produce pseudo-errors
at I 00 times a 10-6 error rate.
Hence an attractive feature of this error rate measurement is that it can measure very
low error rates using comparatively short test intervals. Note, however, that the error
multiplication factor is dependent on the error rate. The technique of estimating a vsry
low error rate by extrapolating from an artificially generated
high error rate is some-
times referred to as a
Q-factor
technique
[34].
In essence, the parameters (mean
and
variance) of a Gaussian distribution are estimated with easily obtained mea$urements
so the much lower error rates can be estimated from the parameters. This technique is
particularly useful in optical fiber systems where extremely low error rates are to be
determined that otherwise require extremely long measurement times.
-1 -0. 2
0 0' 2 1
Figure 4.30 Pseudoerror decision thresholds for error multiplication of 100.
4. 5 PERFOHMANCEMONI TOHI NG203
A major disaclvantage of pseudo-error detection is that pseudo-elror rates are not
accurately related to actual error rates when the noise is non-Gaussian. Channel dis-
tortions, in
particular, cause high pseudo-error rates when the actual error rates are
small. Even though abnormal
pulse amplitudes are
present, enors do not occur unless
noise adds to the distortion. On the other hand, impulse noise might produce high ac-
tual error rates but relatively low
pseudoerror rates.
The correspondence between measured and deduced error rates is improved when
the decision threshold is narrowed. Unfortunately, narrower pseudo-emor decision re-
gions are more difficult to implement and
provide lower
"error"
rate multiplication
factors.
4.5.3 Frami ng Channel Errors
As described in Section 4.6.2. identification of individual channels of a TDM data
sffeam requires overhead bits with a unique data pattem for frame identification. After
a receiving terminal has
"locked"
onto the ftaming
pattern, it ignores occasional bit
errors in the framing pattern but declares an out-of-frame
(OOF) condition when the
erTorrt are consistent. Thus framing bit errors can be used to determine low error rates
but not high error rates that cause false OOF indications and subsequent
loss of fram-
ing. Error rates that are high enough to cause false OOF indications are usually so high
that the link is effectively out of service.
4.5.4 Performance Obiectives
The most basic measure of quality of a digital transmission link is its BER. If the errors
occur in a truly random and independent manner
(i.e.,
they conform to a simple Pois-
son probability model), the average BER completely specifies the performance. Un-
fortunately, elror rate observations
[32]
show that ertors often occur in bursts, and
because bursts affect different services in different ways, specifying the quality of a
transmission link requires more than
just
the average BER. For example, data traffic
is
generally transmitted in blocks that are retransmitted no matter how many error$ oc-
cur in a block. Thus, a burst has little more effect than a single bit error. The rate of
rekansmissions of data blocks is more a function of the frequency of the bursts than
of the long-tefln average BER. Voice traffic, on the other hand, is increasingly de-
graded by all bit errors whether or not they occur in bursts.
Channel errors that occur in bursts can also be much more deleterious to certain op-
erations of a network than are distributed error$ at a similar average rate. A loss of
framing, for example, occurs more often in a burst error environment.' Certain timing
operarions
(pulse
stuffing) described in Chapter 7 are also similarly affected by bursts.
Losses of framing and time synchronization both produce outages
(e.g., continuous er-
*Errors
in a burst are not necessarily contiguous. They merely occur in a short time intewal at a rate that i$
too high to be rasonably explained as chance independent events.
204 DIGITAL TRANSMISSION AND MULTIPLEXING
rors) in the associated haffic channels or even inadvertent disconnects if the condi-
tions last too long.
Common
performance parameter$
used to characterize erTor rates are:
L Error-Free Second
(EFS);
Because data blocks transmitred at 64 kbps require
much less than a second, the percentage
of EFS essentially specifies the
percentage
of time that the channel is available for data transmission
applications. (When
the block transmissions are much shorter than a second. a
shorter meA$urement interval is more appropriate.)
2. Burst Errored Second
(BES):
An errored second with at least 100 errors.
3. Enored Secand (ES): A second with at least one error.
4. Severely Errored Second (SES);
A second with a BER grearer than (10)-3.
5. Degraded Minute (DM): A minute with a BER greater
than 10-6.
The last thrce parameters
are defined in ITU recommendation G.821 for performance
ob-
jectives
of a digital network. The objectives for a 27,5ff)-km reference connection (imply-
ing errorcontributions from many $ources) are ES < 87o, SES <0.ZVo,and DM <l\Vo.
The availability of fiber optic transmission technology with extremely low error
rates (e.g.,
I error in 1012 bits) and higher level communications protocols
that rely on
very low error rates have created a situation where the very low error rates are relied
on, even though they might not be obtained by metallic or radio systems. As discussed
in chapter 10, contemporary data networks
[frame
relay, asynchronous transfer mode
(ATM),
Internet protocol (IP)l
do not provide
error control at the data link layer but
defer the problem
to a higher level process.
4.5.5 Forward Error Correction
On ffansmission channels such as digital subscriber lines and digital radio links, rela-
tively high error rates are sometimes unavoidable. When real-time applications like
voice or video are involved, the effects ofchannel errors cannot be accommodated by
periodically
adding redundant check-sum bits and requesting retransmis$ion of cor-
rupted blocks of data as is commonly done in many data communications protocols.
As mentioned in the previous
chapter, particularly
critical bits of compressed digital
mobile speech are redundantly encoded with enough bits to permit
error correction of
some number of bit errors. Because this basic process
does not require retransmission
of data, it is known as forward error correction (FEC)
t351.
FEC involves adding redundant data bits, like parity
bits or CRC bits, but in
sufficient quantity
to enable error correction instead ofjust error detection. Two
basic forms of FEC are block encoding and convolutional encoding. A block en-
coder attaches error correction redundancy to fixed-length blocks of data, which
implies the need for establishing block framing to identify the fields of redundanr
bits. A convolutional encoder, on the other hand, continuously inserts error cor-
rection bits that are continuously processed
by the decoding circuitry while recov-
ering the original data.
..5
PERFORMANCE MONITORING 205
Reed-Solomon Codes
Although a wide variety of block-coding algorithms have been developed, Reed-
Solomon
(RS) coding is the most
popular form. When a block of source data symbols
of length M is input to an RS coder, an output block of length N symbols is produced,
where N
-
M is the number of check symbols R. An RS code with these
parameters is
commonly referred to as a RS(N, M)
=
RS(N, N
-
,R) code. An RS(N, ry
-
R) code can
conect
+
R symbol errors. In most cartes a symbol consists of an 8-bit byte of data,* so
multiple bit errors in a single byte are no worse than a single bit enor. Digital video
broadcasting
(DVB)
$y$tems use RS(204,188) codes, which means that as many as
eight bytes can be coffupted and be corrected.
Exampte 4.4. Determine the
probability of failure of an RS(204, 188) code
operating in a random error environment with a bit error
probability of 10-3' Assume
each symbol is an 8-bit byte.
Solution. Because the probability of multiple bit errors in a single byte is small,
the probability of a symbol error pr is very close to 8 x l0-3. Using p,, the
probability of failure
(the probability of more than eight symbol errors),
pp can be
determined as
zo4
o,=EfI)'*'
-
P,)'*'
I
= l - T
Lt
r'4
-Pr)to*t
=
0.00005
Gonvolutlonal Coding
Block diagrams of two basic convolutional encoders are shown in Figure 4.31.
Both of these encoder$ are referred to as rate l/2 encoders because the source data
rate is half the channel data rate. During each bit time of a source two output bits are
generated. As illustrated in Figure 4.32, the constraint-zlength coder in Figure 4.314
outputs odd
parity
over bits A and B along with the value of B directly. In Figure
4.31b, odd parity across overlapping fields
(A,
B, C and A, C) are generated. In the
case of odd parity over A, B, C an output value is a I if and only if an odd number of
A, B, C are 1.
An important consideration of a convolutional coder is the constraint length,wlich
specifies the number of instances that a particular source bit gets mapped into a chan-
*Reed*Solomon
codes are sometimes denoted as RS(N' N
-
R, B), wherc ,B designates the length of a
symbol in bits.
f?-)"r'
206 DIGITAL TRANSMISSION AND MULTIPLEXING
( b)
Figure 4.31 Rate l/2 convolutional encoders:
(a)
source constraint length 2;
(b)
source
constraint length 3.
nel bit. In Figure 4.3la the length of the shift registers is 2 bits, which means that this
coder has a source constraint length of 2. Because the output clock is twice the input
rate, the channel constraint length is 4. Similarly, the encoder shown in Figure 4.31b
has a source consffaint length of 3 and a channel constraint length of 6. Longer con-
straint lengths produce better performance.
Because the encoder shown in Figure
4.31a has the shortest possibly constraint length, it is not commonly used but is con-
venient for illustrating the basic operation ofa convolutional decoder. Notice that each
( o)
Sourcedata
0 1 0 1 1 0 0 1 0
F
, . .
h
-{
I
01 11 01 1 0 11 00 01 1 1
Channel data
-
Figure 432 Example input and output data sequences for convolutional encoder in Figure
4.31b.
4.6 TIME DIVISIoN MULTIPLEXING ?;O7
source bit in Figure 4.3la is mapped into 3 channel bits. It is this redundancy that gets
processed to recover the original data.
Example 4.5. Determine the decoding logic to decode received data for the
convolutional encoder of Figure 4.31a. Assume bit-by-bit decisions are made on each
received channel bit and only consider isolated channel errors.
Solution. Two cases are possible for isolated channel errors: an error in a source bit
B or an error in a parity
bit. If a source bit is received in error, two parity errors result.
Thus, when two adjacent
parity errors occur, the most likely scenario is that the
implied data bit needs to be inverted. If a single
parity bit is received in error, the most
likely scenario is that the
parity bit is itselfin enor and no source errors have occurred.
Thus, isolated parity
error$ are ignored.
Example 4.5 illustrates that a convolutional encoder/decoder can easily correct iso-
lated channel errors-at the expense of doubling the data rate. In contrast to the
pre-
vious example and the normal implementation of an RS decoder, a convolutional
decoder usually processe$ the received signal with maximum-likelihood sequence
(Viterbi) detectors
[36].
In essence, a sequence of received signal samples is compared
to all
possible channel input sequences with the closest allowable sequence chosen to
determine the source data. The added complexity
provides signif,rcant benefits. Con-
straint Iength 7 convolutional coding, for example,
provides better than 5 dB of im-
provement in error performance for a
given SNR. Since a doubling of the necessary
bandwidth
(for rate l/2 coding) implies a relative noise increase of 3 dB, a net gain of
2 dB is achieved- Notice further that a convolutional encoder can correct a much
greater number of comrpted bits than can an RS code-as long as the comlpted bits
are separated by something
greater than the channel constraint length. Hence, convo-
lutional coding is well suited to correct independent enors while RS codes work well
in burst error environments. For these reasons combinations of the two codes are often
used.
Maximum-likelihood detection of convolutional code sequencert is very similar to
trellis-coded modulation
(TCM) detection discussed in Chapter 6. The TCM discus-
sion
provides some example measurements that involve the same basic
process re-
quired for Viterbi detection of convolution codes.
4.6 TIME DIVISION MULTIPLEXING
Although FDM of digital transmission signals is
possible and is used in some special
situations, TDM is by far the motit colnmon and economical means of subdividing the
capacity of a digital transmission facility. One application where FDM techniques
have been used for digital signals is on multidrop data communications lines, where
the sources and destinations of the data are distributed along the line. Most telephone
network applications, however, involve clu$ters of channels in the form of trunk
groups between switching offices. In local digital access applications, where sub-
208 DrGrrAL TRANSMtsstoN AND MULTIpLEXtNe
scriber locations are distributed throughout a service area, channels are sometimes
added and dropped on a distributed basis, but only with a limited number of access
points because of maintenance and reliability considerations.*
There are two basic modes of operation for TDM: those that repeatedly assign a
pofiion
of the transmission capacity to each source and those that assign capacity only
as it is needed. The first form of operation is referred to as synchronaus time division
multiplexing (STDM) when necessary, to distinguish it from the
"as-needed"
mode
of operation. Otherwise, TDM is generally
understood to imply the synchronous va-
riety. The as-needed form of TDM is variously referred to as asynchronous time divi-
sion multiplexing (ATDM), Asynchronous Transfer Mode (ATM),
srarisrical time
division multiplexing (Stat-Mux),
or packet switching. Circuit-switched telephone
networks use STDM, whereas data networks typically use ATDM. Discussion of
these latter techniques is deferred to Chapter 10.
The term asynchronous is also used in another TDM context to refer to the multi-
plexing of multiple, independent tributary signals into a single higher rate signal. In
this context,
"asynchronous"
is totally unrelated to either asynchronous transmission
as described at the beginning of this chapter or packetlike multiplexing mentioned in
the previous paragraph. In this third context asynchronous refers to multiplexing of
multiple unsynchronized tributaries into a single high-rate serial stream. This chapter
is concerned only with multiplexing of synchronized tributaries. TDM of unsynchron-
ized tributaries is described in Chapter 7.
4.6.1 Bit Interleaving Versus Word Interleaving
Two different structures of (synchronous)
time division multiplex frames are shown
in Figure 4.33. In the first instance each channel is assigned a time slot corresponding
to a single bit-hence the term bit interleaving. In the second instance each channel
is assigned a longer time slot corresponding to some larger number of bits, referred to
as a word-hence the tennword interleaving.
The decision as to which structure to use in a particular application is primarily de-
pendent on the nature of the sources. [n T-carrier systems each channel produces
a
complete 8-bit word at a time. Hence word interleaving is desirable so all bits can be
transmitted as generated.
ln conffast, higher level digital TDM multiplexers typically use bit interleaving of
the lower level multiplex signals since the lower level signals represent continuous,
I-bit-at-a-time, data streams. The specific formats of the higher level multiplex signals
are described in Chapter 7 when synchronization of bit streams is considered. One no-
table exception to higher level bit interleaving is the word-interleaved structure of fi-
ber-based SONET or SDH multiplexing described in Chapter L
'The
digital network has evolved with ever-increasing levels ofmultiplexing creating serial bit streams with
ever-increasing data rutes. The use of WDM on optical fiber systems represeflts a deparhrre from the
single-stream paradigm
stimulated by two factors: (l) it is impracticalto tansmit the ultrahigh speed serial
channels and (2)
separate wavelengths provide
transparency for diverse applications.
4.6 TIME DIVISION MULTIPLEXING 209
1 Frame
f+
Bit intrlervlng
lilord intcrlearing
'
r ' f f l =i L=Pr =Fr
1 2 3 4 1
Figure 4.33 Bit interleaving and word interleaving of four-channel TDM multiplexers,
4.6.2 Framlng
To identify individual time slots within a TDM frame, a receiving terminal uses a
counter synchronized to the frame format of the transmitter. Just as for synchroniza-
tion of sample clocks, a ceftain amount of hansmission overhead is required to estab-
lish and maintain frame synchronization. In fact, most of the techniques used for frame
synchronization are directly analogous to clock synchronization techniques discussed
previously. Specifically, the basic means of establishing frame synchronization are;
l. Added-digit framing
2. Added-channel framing
3. Unique line signal framing
4. Statistical framing
The main considerations in choosing a framing procedure are time required to es-
tabli$h framing, effects of channel erTor$ in maintaining frame synchronization, rela-
tionships between the line clock and sample clocks derived from the line clock,
transmission overhead, and complexity of the framing circuitry.
The severiry of a loss of framing and the time required to reestablish synchroniza-
tion depend on the nature of the traffic. Since a loss of framing implies a loss of data
on all channels, the mean time between misframes must be as long as possible' For
voice traffic infrequent misframes can be tolerated if frame synchronization is rees-
tablished rapidly enough to minimize the duration of the
"glitch"
in the output speech'
For data traffic the duration of reframe times is not as critical as the frequency of oc-
currence since most data communications
protocols initiate recovery procedures and
retransmit a mes$age no matter how many data bits are lost.
A critical requirement for reframe time in the telephone network comes from the
possibility that various in-channel control signals may be lost and interpreted as dis-
connects. Thus the maximum refiame time on a
particular digital transmission link is
often determined from analog network signaling conventions. A loss of framing is also
used as a
performance monitor for the transmission link and usually sets alatm condi-
I Freme
210 DrcrrAL THAN$MrssroN AND MULTIpLEXtNc
tions, which in turn cause automatic switching to spare terminals or transmission fa-
cilities. Generally speaking, terminals serving 6(X) or more channels require automatic
protection switching
[37].
Out-of-frame conditions occur in two ways. First, the locally derived sample clock
may lose synchronization with the line clock and produce a slip in the counter se-
quence.
Normally, the timing information in the line clock is sufficiently strong to pre-
vent misframes of this type. An exception occurs on radio links when deep fades
reduce the signal power to the point
that clock synchronization is impossible. Receiv-
ing terminals can also lose clock synchronization through no fault of their own or of
the transmission link. If the source clock at the transmitting terminal has too much
jit-
ter or generates abrupt phase
shifts, receiving terminals may be unable to track the
clock signal. Both phenomena
are related to network synchronization practices,
as dis-
cussed in Chapter 7.
A second source of mi$frames is from channel errors creating false out-of-frame
framing pattems. Thus considerable redundancy in the framing pattem
is required to
minimize the
probability
of false misframes. A loss of framing is determined when the
occurrence of framing pattem violations exceeds some shoft-term density threshold.
In all of the framing techniques discussed in the following paragraphs,
special
framing bits or codes are inserted into the information skeam. These insertions do not
have to occur with each frame. lnstead, they can be sent only once for a
predetermined
number of information frames. In this manner the transmission overhead for framing
can be reduced, which is particularly
important in bit-interleaved systems. An individ-
ual information frame contained between framing indicators is sometimes referred to
arr a
"subftame,"
Added-Digit Framlng
One common technique of framing a digital TDM information stream is to
peri-
odically insert a framing bit with an identifiable data sequence. Usually the framing
bit is added once for every frame and alternates in value. This particular
format is the
procedure used to establish framing in the original Dl channel banks. When the Tl
line carries only voice traffic, this framing format is pafiicularly
useful since no infor-
mation bits can sustain an alternating l, 0 pattem. (An alternating pattern
represents a
4-kHz signal component, which is rejected by the bandlimiting filter in the PCM
codecs.)
Framing is established in a receiving Dl channel bank by monitoring frrst one bit
position
within a 193-bit frame and then another, until the alternating pattern
is lo-
cated. With this framing strategy, the expected framing time from a random starting
point with random data i$ derived in Appendix A as
I
averagenumberof
)
I
bi tsto
I
I
determine that an
I
linformation
position
is
I
I
not a framing
nositionJ
= (+N) (2N+ l )
=
M +
|N
bit times
4.6 TIME DIVISION MULTIPLEXING 21 1
(4.10)
where Nis the number of bits in a frame including the framing bit.
For D I channel banks, N
=
193 so that the framing time is 37
,346
bits, or 24. I 88
msec. Also of interest is the maximum framing time. Unfoftunately, fhere is no abso-
lute maximum framing time for a Tl system with random data. It is very unlikely,
however, that the framing time would ever exceed the average search time fbr all bit
positions, or 48.25 msec. This latter measure of framing time is refered to as themaxi-
mum(werageframe time.It i$ the average time required to establish framing, but with
the assumption that all bit positions must be tested for the framing sequence before the
actual framing bit is found. Obviously, the maximum average frame time is twice the
average value from a random statting
point defined in Equation 4. I 0.
The framing time can be reduced by using more sophisticated frame search strate-
gies. One approach examines one bit at a time, as before, but during a reframe the
search begins a few bit positions in front of the present position under the assumption
that short lapses of clock synchronization cause small counter offsets. A second ap-
proach
[38,
39] uses a parallel search by monitoring all bit positions simultaneously
for the framing
pattern. With this framing
procedure, framing is establi$hed when the
last of the N* I information bit positions finally produces a framing
pattern violation.
The probability that all information bit positions produce a framing violation in n or
less frames is derived in Appendix A as follows:
prob(frame time < n)
=
[
*
(*f]t-t
( 4. 1l )
where N is the number of bits in a frame and the probability of I
's
i s j. Using Equation
4. 11, we determine the median framing time by setting prob(frame time < n)
=
i.
Hence
n
=
-logzll -
(+)r/(N-')l
{4.r2)
Setting N
=
193 for the frame length of the Dl channel bank
produces the result that
n
=
8.1 frames, or approximately I msec.
An even more sophisticated framing strategy involves continually monitoring all
bit
positions,
even while the system is synchronized. Then, when a misframe is de-
tected, a new frame position is immediately established or the frame search time is sig-
nificantly reduced. Fufthermore, the continuous search for framing patterns provides
additional information for declaring an out-of-frame condition.
(There is little
point is
discarding the present frame position unless another bit
position exhibits a more con-
sistent fiaming pattern.
)
212 DtctrAl TRANSMtsstoN AND MULTIpLEXINc
The framing pattern
of second-generation channel banks (D3, D4) from AT&T was
changed from the altemating l, 0 pattern to establish a longer sequence for identifica-
tion of signaling frames. As mentioned previously,
these channel banfts provide
I bits
of voice for all time slots except in every sixth frame, which uses the least significant
PCM bit for signaling. The signaling channels thus derived are divided into an A and
a B subchannel, implying each subchannel sends a bit in every twelfth frame. Hence
a l2-bit framing sequence is needed to identify the signaling bits and the two signaling
subchannels. The lZ-frame superframe (SF) structure and associated framing se-
quence are shown in Figure 4.34.
When describing or processing the framing $equence shown in Figure 4.34, it is
convenient to divide the framing bits into two separate sequences. During the odd
frames the framing bit altemates, while during the even frames the framing bit se-
quence
is Offi111000111000. Figure 4.34 indicates that the A signaling frame can be
identified by a 0-to-l transition in the even-numbered frame sequence. Conespond-
ingly, a 1-to-0 transition in the even-numbered frame sequence signifies a B signaling
frame. Frame acquisition begins by finding the alternating bit sequence (with 385 in-
tervening bits). Then, the 000111 framing pattern is located. Another framing se-
quence, extended superframe
(ESF),
for DSI signals is described in Section 4.6.3.
Added-Channel Frdmlng
Added-channel framing is basically identical to added-digit framing except that fram-
ing digits are added in a group
such that an exha channel is established. Hence the
Frami ng
bi r
Freme no,
1
2
3
4
1 2 0
Figure 4.34 Twelve-frame superframe sfucturc of DSI signal with robbed digit signaling,
Frame alignment signal
(F) =
l0l0l0; multiframe alignment signal (M)
=
ffil ll0.
5
6
OFil-T]
offi
l--::]
OFilTl
I-I
rlilTl
f
-l
lFI
Ftr-I T--TI
0FI*--*--T--*---]
| I
l
4.6 TIME DIVISION MULTIPLEXING 213
transmission rate of individual channels is integrally related to the line rate of the en-
tire multiplex.
The fact that the frame boundaries are identified by whole codewords adds consid-
erable
performance and flexibility to the framing process. First, framing can be estab-
lished more rapidly since random 8-bit codewords are very unlikely to appear as
f'raming codes.
(See
the
problems
at the end of the chapter.) Second, the larger code
space simplifies identification of auxiliary functions such as superframe boundaries,
parity bits, or equipment $tatus. In most systems the added channel contains more than
framing bits.
The first-level digital multiplex signal of ITU
(El) is an example of a system
using added-channel framing. The El standard establishes 32 channels per frame
with one channel
providing framing and one more channel dedicated to signaling.
Thus, 30 of 32 channels are available for message channels. Figure 4.35 shows the
frame structure of the El signal. The frame alignment signal
(FAS) is inserted into
the framing channel of every even-numbered frame and a I bit inserted into the
second bit
position in every odd-numbered frame
(to preclude FAS
generation in
those frames). The first bit of every FAS frame may carry a cyclic redundancy
check
(CRC-4) for additional frame alignment integrity and eruor rate monitoring.
The remaining bits of the framing channel are used for CRC-4 alignment or alarm
indications or re$erved fbr other uses.
(See
ITU recommendation G.704 for more
details.)
The signaling channel ofFigure 4.35 depicts the use ofchannel-associated signal-
ing
(CAS), which implies that 4 bits
per 16-frame multiframe are allocated to each
message channel. The positions of the associated signaling channels are determined
with respect to the multiframe alignment signal
(MAS). The 4 bits of each CAS sig-
naling channel should never be 0000 to preclude inadvertent
generation of the
Framing Mesage Signsling
Metcoge
channel channelr ch6nn6l
chennols
0 1
1 6 3 1
Framoo
TTEET-I
"'
lMAsTl
"'
l-]
Framer
f'lil-T l
"'
f
rTG-]
"'
l-l
Framez
fl-
FAsT-l
...
I
,Tu-]
"'
|_*--l
a
,rPreclurion
of FAS
FrEmB rb
I trt |
-l
"'
|
fi'T3t-l
"'
T---l
Ftame o
[EsT--*]
...
liifrHT-l
...
l-l
Figure 4.35 Channel format of ITU primary digital signal
(El).
214 DIGITAL TRANSMISSIoN AND MULTIPLEXING
MAS. The use of these bits is similar to the
"bit-robbed"
ABCD birs of the North
American DS I signal. When common channel signaling is carried on an E1 link, the
CAS channels are replaced by a 64-kbps HDLC signaling link in time slot l6 of the
El frame.
The average frame acquisition time of a multibit frame code is derived in Appendix
Aa s
(4. 13)
where N is the length of a frame including the frame code, /, is the length of the frame
code, and it is assumed that I's and 0's are equally likely.
From Figure 4.35 it can be seen that for the El signal N
=
512 and L
=
7. Thus the
average frame time from a random starting point is determined from Equation 4.13 as
0.5 msec. Again, the
"maximum
average" frame time is twice the average from a ran-
dom starting point, or I msec. Notice that these frame times are much faster than DSI
added-digit frame times because a higher percentage of bits are allocated to framing.
Unique Line Code Framlng
Bipolar coding manages to shape the spectrum
(remove
the dc component) of the line
code by adding extra signal levels to provide more flexibility in selecting signals. The
same technique can be used to establish frame synchronization. In fact, with bipolar
coding and added-digit framing, bipolar violations can be used to identify the framing
boundaries uniquely and rapidly. A particularly
significant example of using line code
violations for framing is the ISDN S/T basic rate interface described in Chapter 11.
Even without added-digit framing, unique line codes can be used to carry informa-
tion while simultaneously conveying frame positioning. If the number of signal levels
is doubled for the framing bit only, the extra levels provide
sufficient code space for
the information but uniquely identify that bit position as a frame boundary. This pro-
cedure does not necessarily increase the effor rate, since for any particular
bit
(infor-
mation or framing), half of the levels can be disallowed.
The main advantage of using unique line codes for framing is that the information
bit positions
cannot generate framing pattems.
Hence framing is established as soon
as a frame bit occurs, and misframes are detected almost immediately.
The main disadvantages ofunique line codes are the added signal processing
re-
quirements (if new levels are established
just
for framing) and the dependence on the
transmission terminals to locate framing. With the other framing techniques described
in this section, the framing pattems
are represented in the data of the bit stream. Hence
the transmission equipment can be changed independently of the multiplex equip-
ment. For additional information on framing techniques and performance
analyses see
references
[40]
and
[41].
Frame rime
:#j.+
(in bits)
4.6 TIME DlvlsloN MULTIPLEXING 215
Statlstlcal Framing
Statistical framing relies on the statistics of data within individual bits of a transmis-
sion scheme. Assuming the source of the data is known, it may be
possible to ascertain
information such as which bit is a most significant bit (MSB) of a PCM codeword and
thereby recover byte framing without an explicit framing bit of any type. Obviously,
this method of framing is only applicable to special applications. One such application
is determining word alignment in ADPCM 7-kHz audio
[42],
as in ITU-T Recoiltmen-
dation G.722.
4.6.3 DSI Extended Superframe
When a Tl line is used as an interoffice transmission link, the operating telephone
companies have access to both ends of the line for maintenance and performance
monitoring. Furthermore, most installations include
protection switching facilities
that can also be used to routinely test a line while the trffic normally assigned to that
line is diverted to a spare line. A significantly different situation arises when a T1 line
is used by a customer as access to the public $witched network or a$ a leased line in a
private network. The explosive use of Tl lines in such applications created the need
for enhanced maintenance features for T1 customers. The ESF as provided in the D5
channel bank provides the desired features.
Foremost among the customer needs is noninfiusive
performance monitoring of an
end-to-end Tl circuit. Monitoring bipolar violations at the customer service unit
(CSU) provides nonintrusive monitoring of the receive signal but
provides no infor-
marion regarding the
quality of the signal at the other end of the line (which terminates
at the provider's facilities). Furthermore, leased Tl circuits
generally involve interme-
diate transmission, multiplexing, and cross-connect equipment. Because the interfaces
of each of these equipments remove bipolar violations, the customer has no access to
performance statistics intemal to theprovider's facilities. Determination of end-to-end
error rate
performance in this environment requires taking at least a portion of the Tl
circuit out of service
(for
which there is usually no spare).
ESF allows cu$tomefs to achieve end-to-end, in-service pelformance monitoring
by redefining the 8-kbps F bit of each DS I frame to consist of a 2-kbps framing chan-
nel, a 4-kbps data link channel, and a Z-kbps CRC channel representing a check sum
over all information bits. Because the CRC channel is carried intact from one customer
location to another, channel enors occurring in any intermediate facility
produce CRC
errors at the far end.
The data link supports numerou$ $ervices, one of which is the ability to interrogate
the far end, or any intermediate equipment, for
performance statistics. Thus the data
link
provides both the customer and the service
provider with an extremely useful tool
for isolating faulty span lines or equipment.
As indicated in Table 4.8 obtained from reference
[33],
the three subchannels are
established by
"extending"
the D4 superframe format to encompass 24 DS t frames.
Because the actual framing bits occur only once every four DSI frames, there are 771
bits intervening between bits carrying the frame pattern sequence
(FPS) 00101 I '
216 DrcrrAl THAN$MrssroN AND MULTIpLEXtNc
The 6 CRC bits (CB I to CB6) of each extended superframe represenr a CRC check
of all4608 information bits in the
previous
superframe.* Besides providing end-to-end
performance
monitoring, the CRC virtually precludes
the chances of false framing on
a data bit po$ition.
Even though static user data can easily simulate the FPS, it is ex-
tremely unlikely that user data can spuriously generate valid CRC codes in successive
superframes.
(Six
bits of random data match a 6-bit code with a
probability
of I in 64.)
The performance parameters
measured and reported by the 4-kbps data link (DL)
are framing bit errors, CRC errors, out-of-frame (OOF)
events, line code (bipolar) vio-
lations, and controlled slip events
(described
in Chapter 7). Individual event$ are re-
ported
as well as event $ummaries. The four performance
summaries reported are:
I. Enored seconds (ESs) (ES =
at least one CRC event)
2. Bursty seconds (BSs) (BS =
2-3lg ESs)
3. Severely errored seconds (SESs) (SES =
>319 ESs or OOFs)
4. Failed seconds (FSs) (l0consecutive
SESs)
ESF CSUs typically determine the above parameters
on 15-min intervals and store
them for up to 24 hr for polling
by a controller
[43].
The sES report conforms to ITU
recommendation G.821. In addition to suppofting remote interrogation of perform-
ance statistics, the data link carries alarm information, loopback commands, and pro-
tection switching commands.
In addition to the
previously
mentioned features, EsF introduces a new option for
per-channel signaling via the robbed signaling bits in every sixth frame. Because an
ESF is 24 frames long, there are four signaling bits in every channel in every super-
frame as opposed to 2 bits in sF format (Figure
4.34). whereas the two signaling bits
in the sF format are designated as A and B bits, the four bits in the ESF case are des-
ignated A, B, C, and D. Three signaling modes are defined: z-state where all bits are
A bits, 4-state where the signaling bits are ABAB, and 16-srate where the signaling
bits are ABCD. The SF format provides the first two signaling modes but not the last.
4.7 TIME DIVISION MULTIPLEX LOOPS AND RINGS
In Chapter 2 it is mentioned that TDM is not a$ amenable to applications with disrrib-
uted $ources and sinks of traffic as is FDM. In this section a particular form of a TDM
network is described that is quite
useful in interconnecting diskibuted nodes. The ba-
sic structure of interest is referred to as a TDM loop or TDM ring and is shown in Fig-
ure 4.36.
Basically, a TDM ring is configured as a series of unidirectional (two-wire)
links
arranged to form a closed circuit or loop. Each node of the network is implemented
with two f'undamental operational feature$. First, each node acts as a regenerative re-
peater merely to recover, the incoming bit stream and retransmit it. Second, the net-
*Calculation
ofthe CRC actually includes F bits that are set to I for purposes
ofCRC calculation only. Thus,
charurel errors in the F bits do not create CRC errors (unless
they occur in the CRC bits themselves).
4.7 TIME DIVISION MULTIPLEX LOOPS AND HINGS 217
Figure 4.36 Time division multiplex loop'
work nodes recognize the TDM frame structure and communicate on the loop by re-
moving and insefting data into specific time slots assigned to each node. As indicated
in Figure 4.36, a full-duplex connection can be establi$hed between any two nodes by
assigning a single time slot or channel for a connection. One node insetts
information
into the assigned time slot that propagates around the loop to the second node
(all in-
tervening nodes merely repeat the data in the particular time slot). The destination
node removes data as the assigned time slot passes by and inserts return data in the
proce$s. The retum data
propagates around the loop to the original node where it is
removed and replaced by new data, and so forth.
Since other time slots are not involved with the particular connection shown, they
are free to be usecl for other connection$ involving arbitrary
pairs of nodes. Hence a
TDM Ioop with C time slots
per frame can suppoft C simultaneous full-duplex con-
nections.
If, as channels become available, they are reassigned to different pairs of nodes, the
transmission facilities can be highly utilized with high concentration factors and pro-
vide low blocking
probabilities between all nodes. Thus a fundamental attraction of a
loop network is that the transmission capacity can be assigned dynamically to meet
changing traffic
patterns. In contrast, if a star network with a centralized swirching
node is used to intercorutect the nodes with four-wire links, many of the links to par-
ticular nodes would be underutilized since they cannot be shared as in a loop configu-
ration.
Another feature of the loop-connected network is the ease with which it can be re-
configured to accommodate new nodes in the network. A new access node is merely
inserted into the nearest link of the network and the new node has complete connec-
tivity to all other nodes by way of the TDM channels.In contrast, a $tar structured net-
work requires transmission to the central node and expansion of the centralized
switching facilities.
?18 DIGITAL TRANSMI$sIoN AND MULTIPLEXING
The ability to reassign channels to arbihary pairs of nodes in a TDM loop implies
that the loop is much more than a multiplexer. It is, in fact, a distributed tansmission
and switching system. The switching capabilities come about almost as a by-product
of TDM transmission. TDM loops represent the epitome of integrated transmission
and switching.
TDM loops have been used within computer complexes to provide
high capacity
and high interconnectivity between processors,
memories, and peripherals
[,14].
The
loop structure in this application is sometimes more atffactive than more conventional
bus structures since all transmission is unidirectional and therefore avoids timing
problems
on bidirectional buses that limit their physical length. Furthermore, as more
nodes are added to a bus, the electrical loading increases, causing a limitation on the
number of nodes that can be connected to a bus. Loops, on the other hand, have no
inherent limits of transmission length or numbers of nodes.
The loop structure of Figure 4.36 is topologically identical to the token-passing
ring developed by IBM and standardized by the IEEE as a 802.5 local area network.
However, a token-passing ring operates differently than a TDM loop in that there is
only one channel. when a node on a ring becomes active, it uses the entire capacity
of the outgoing link until it is through sending its message. In contrast, a node on a
loop uses only specific time slots in the TDM $tructure, allowing other nodes to be si-
multaneously
"connected"
using other time slots. [n essence, a TDM loop is a distrib-
uted-circuit switch and an 802.5 ring is a distributed-packet switch.
A particularly
attractive use of a loop with high-bandwidth links is shown in Figure
4.37. This figure illustrates the use of add-drop multiplexers (ADMs)
that access
whatever bandwidth is needed at a local node but pass
the rest on to other nodes. In
fypical applications the amount of bandwidth allocated to each node is quasi-static: It
is changed only in respon$e to macroscopic changes in traffic patterns, possibly
as a
function of the time of day. This basic operation is genera-lly
referred to as a cross-con-
Figure 4.37 Functional mesh, fiber loop, and ADMs.
REFERENcES 219
Figure 4.3E Use of reverse loop to circumvent link failures in TDM loops.
nect function as opposed to a swirching function, which involves call-by-call recon-
figurations. Both types of switching operations are discussed in the next chapter. An
important
point to note about Figure 4.37 is the ability to utilize a general-purpose
physical topology but define an arbitrary functional topology on top of it.
One obvious limitation of a loop is its vulnerability to failures of any link or node.
The effect of a node failure can be minimized by having bypass capabilities included
in each node. When bypassed, a node becomes merely a regenerative repeater, as on
T-carrier transmission links. Link failures can be circumvented by providing alternate
facilities. Figure 4.38 shows one particular structure using a second, reverse-direcflon
loop to provide backup capabilities in the case of failures. When fully operational, the
network can use the reverse loop as a separate, independent network for traffic as
needed. Whenever a failure occurs, the nodes adjacent to the break establish a new
loop by connecting the forward
path to the reverse
path at both places. Hence all nodes
continue to have full connectivity to any node on the new loop'
A
particular example of the use of the dual reverse loop for both
protection and dis-
tributed
queued access to the channels is the distributed
queued dual-bus
(DQDB)
[45]
system developed by
QPSX
in Aushalia and standardized by the IEEE as an 802.6
metropolitan area network. Further examples of loop structures and applications are
provided in Chapter 8 where SOI.IET rings are discussed.
REFERENCE$
I F, deJager and M. Christiaens,
"A
Fast Automatlc Equalizer for Data Links,"Pftillps
Technical Revielu, Vol. 36,1977,
pp. 10-24.
2 K. Azadet and C. J, Nicole,
"Low-Power
Equalizer Architectures for High-Speed
Modems," IEEE Communicatinns Magaein e, October, 1998, pp' I I 8- 126.
3 F. D. Waldhauer,
"A
Z-Level, 274 Mbls Regenerative Repeater for T4M"' IEEE
Int e mational C ommunications C onfe renc e, I 975,
pp. 48- 1 3-48- t 7.
4 N. Karlovac and T. V. Blalock,
"An
Investigation of the Count Rate Performance of
Baseline Restorers,"
paper presented at the Nuclear Sciences Symposium, Washington,
DC. 1974.
22O DtctrAl TRANSMtsstoN AND MULIpLEXtNG
5 F. D. Waldhauer,
"Quantized
Feedback in an Experimental 280-Mb/s Digital Repeater
for Coaxial Transmission," IEEE Trttnsacfions on Communications, Jan, 1974, pp, l-5.
6 J. Salz,
"Optimum
Mean Square Decision Feedback Equalization," Bell System
Technical Joumal, Oct. 1973, pp. l34l- I 373.
7 J. M. Cioffi, W. L. Abbott, H. K. Thapar, C. M. Melas, and K. D, Fisher,
"Adaptive
Equalization in Magnetic-Disk Storage Channels," IEEE Communications Magazine,
Feb. 1990, pp. 15-29.
I S. Sugimoto, K. Hayashi, and F. Mano,
"Design
of ZBIQ Transceiver for ISDN
Subscriber Loops," IEEE International Conference on Communications,Itrne 1989, pp.
8. 1. 1- 8. 1. 5.
9 M. R. Aaron,
"PCM
Transmission in the Exchange Plant," Bell System Tethniml
Joumal, Jan. 1962, pp. 99*141.
l0 Technical Staff, Bell Telephone Laboratories, Transmission Systems
.for
Communiuttion,r, Westem Electric Co., Winston Salem, NC, l97l
,
p. 668.
ll
"1.544
Mhps Digital Service," Bell System Technical Reference Publication No.
4l45l,May 1977.
12 G. D, Forney,
'"The
Viterbi Algorithm," Praceedings of IEEE, Mar. 1973, pp.268-2i8.
13 V. l. Johannes, A. G. Kaim, and T. Walzman,
"Bipolar
Pulse Transmission with Zero
Extraction," IEEE Transactions on Communications, Apr. 1969, pp. 303-310.
14
"The
D3 Channel Bank Compatibility Specification-lssue 3," Technica.l Advisory
No. 32, American Telephone and Telegraph Company, Basking Ridge, NJ, Oct. lgj7.
l5 B. Johnston and W. Johnston,
"LD-4
A Digital Pioneer in Action," Telesis, Vol. 5, No.
3, June 1977, pp.66-72.
16 J. H. Davis,
'T2:
A 6.3 Mb/s Digital Repeatered Line," IEEE Inrernational
Comrnunications C onferenu, I 969, pp. 34-9-34- 16.
17 Recommendation G.703, CCITT Orange Booft, Vol. 3, No. 2.
18 R. M. Wienski,
"Evolution
to ISDN within the Bell Operating Companies," IEEE
Communications Magafine, Jan. 1984, pp. 33-41.
19 American National Standard: Digital Hierarchy-Formats Specifications, ANSI
Tl.l07-1988, Editorial Revised Apr. 1989.
20 J. M. Sipress,
"A
New Class of Selected Ternary Pulse Transmission Plans for Digital
Transmission Lines," IEEE Transactions on Communitation Technology, Sept. 1965,
pp.
366-372.
2l E. E. Schnegelberger and P. T. Griffiths,
"48
PCM Channels on Tl Facilities," National
Elettronics Conference, 1975, pp. 20 1
-205,
22 P. A. Franaszek,
"Sequence-State
Coding for Digital Transmission," Bell System
Technical Journal, Dec, 1967, pp. 143-157.
23 J. O. Azaret, J. F. Graczyk, W. M. Hauser, and J, V. Mahala,
"New
Improved T-Canier
Doubles Capacity, Cuts Costs," BelI l-abs Rewrd, July 1985, pp. 26-31.
24 J, W. Lechleider,
"Line
Codes for Digital Subscriber Lines," IEEE Communit:ations
Magazine, Sept. 1989, pp.25*31.
25 A. Lender,
-'The
Duobinary Technique for High Speed Data Transmission," IEEE
Transactions on Communication Electronits, May 1963, pp.2l4-218.
26 A. t ender,
"Correlative
Level Coding for Binary Data Transmission," IEEE Spectrum,
Feb. 1960, pp. I 04- I I 0,
PFOBLEMS 221
21 P. Kabal and S. Pasupathy,
"Partial
Response Signaling," IEEE Transactions on
Communications, Sept. 1975, pp. 921-934.
28 A. M, Genish and R. D. Howson,
"Multilevel
Partial Response Signaling,"
International Communicafions Conference, 1967, p' I 86.
29 D, W. Jurling and A. L, Pachynski,
"Duobinary
PCM System Doubles Capacity of Tl
Facilities," IntemationalCommunicationsConferente,
l9TT'pp.32.2-297-32-2-301.
30 T. Seaver,
"An
Efficient 96 PCM Channel Modem for 2 GHz FM Radio," National
Telecommunications Conference, 197 8, pp. 38.4. t
-38'4'5'
3l
"special
Issue: IARDS," BeII SystemTechnical Joumal' Mar. 1984.
32 M. B. Brilliant,
"Observations
of Errors and Error Rates on Tl Digital Repeatered
Lines," Bell SystemTechnical Joumal, Mar. 1978, pp' 7ll*746.
33
..Carrier-to-Customer
krstallation-DSl Metallic Interface," ANSI T1.403-1989,
American National Standards Institute.
34 N. S, Bergano, F. W. Kerfoot, and C. R. Davidson, |EEE Photonhs Technology Letters'
Vol, 5, 1993, pp. 304-306.
35 S. B. Wicker, Error Control Sysferns
for
Digital Communication and Storage,
Prentice-Hall, Englewood Cliffs, NJ, 1995.
36 L. H. Charles I*e, Convolutional Coding: Fundamentals and Applications, Artech
House. Norwood. MA. 1997.
37 Technical Staff, Bell Telephone Laboratories, Eng,ineering and Operations in the BelI
System, Western Electric Co., Indianapolis, 1978.
38 D. E. Dodds, L, R, Button, and S.-M. Pan,
"Robust
Frame Synchronization for Noisy
PCM Systems," IEEE Transactions on Communicatians, May 1985, pp. 465-469.
39 D. E. Dodds, S.-M. Pan, and A. G. Wacker,
"Statistical
Distribution of PCM Framing
Times," IEEE Transactions on Communications, Nov. 1988, pp' 1236-'1241.
40 R. A. Scholtz,
"Frame
Synchronization Techniques,* IEEE Transacfions on
Communications, Aug. 1980, pp' 1204-1212-
4l D. T. R. Munhoz. J. R. B, de Marca, and D. S, Arantes,
"On
Frame Synchronization of
PCM Systems,* IEEE Transactions on Communicarions, Aug. 1980' pp' 1213-1218.
42 M. Anderson and O. Petruclrka,
"Apparatus
and Methods for Recovering Alignment
from a Non-Ideally Encoded Multi-bit Digital Signal," U.S. PatentNo.4,955'037' Sept'
4. 1990.
43 K. Stauffer and A. Brajkovic,
"DS-l
Extended superframe Format and Related
Performance Issues," IEEE Communications Magazine, Apt. 1989, pp' 19-23'
44 C-System General Description, Radio company Report No, 523-0561697-20173R,
Collins Radio Company, Dallas, TX, May I, 1970.
45
"New
hoposal Extends the Reach of Metro Area Nets," Data Communicuriorls, Feb.
1988, pp, 139-145.
PHOBLEMS
4.1 If the transmitter and receiver of an asynchronous transmission system utilize
clock sources that are accurate to one
part in 103, determine the maximum num-
ber of bits in a codeword if the maximum sample timing enor is 207o of a
pulse
222
4.2
DIGITAL TRANSMISSION AND MULTIPLEXING
interval. Assume the sample clock is eight times the bit rate and that the stalt bit
is determined by counting four sample times from the beginning of the start bit.
Using the symbols +, 0, and
-
to represent a positive pulse,
no pulse, and a nega-
tive pulse, respectively, determine the following line code $equences of the bi-
nary data sequence
0l l0 r0000t0001 100000001 0
(a) Bipolar with most recent pulse
being positive
(b)
Bipolar with most recent pulse being negative
(c)
Pair selected temary beginning in the negative mode
(d)
B3ZS with a +0+ substitution having
just
been made
(e) 8625 with the most recent pulse
being positive
4.3 Assume that two identical cable systems are used for digital transmission with
equal pulse amplitudes. One system uses conventional bipolar signaling and the
other uses PST. compare the crosstalk levels of the two $ystems. (Assume
that
l's and 0's are equally probable.)
4.4 A digital transmission system operating at an error rate of 10-6 is to have its data
rate increased by 50Vo. Assuming that the same transmitted power
is to be used
in the second system, what is the new error rate?
4.5 What is the average reframe time of a D3 channel bank (using
SF framing) from
a random starting point?
What is the maximum average reframe time of a D3
chalnel bank?
(Assume
I's and 0's in the me$sage traffic are equally likely.)
4.6 Repeat hoblem 4.5 for the primary
TDM multiplex signal specified by CCITT.
4,7 A TDM system operating at 2 Mbps is to have a maximum average reframe
time of 20 msec. What is the maximum possible
frame length if framing is es-
tablished with a bit-by-bit frame search?
(Assume
that I's and 0's in message
channels are equally likely.)
4.8 A Tl transmission system using a Dl frame format is to have an average reframe
time (from a random starting point)
of l0 msec. How large a block of bit posi-
tions must be examined in parallel to achieve the desired result?
4.9 what is the expected framing time for a Tl line (D3
frame formar) if the framing
strategy is bit by bit and the data sheam has 60% I's and 40Vo O's?
4.10 What is the average pulse
density of 4B3T coding (Assume
I's and 0's are
equally likely.)
4.11 A TDM transmission link using 4B3T coding can transmit 32 64-kbps voice
channels using the same symbol rate as a Tl line (1544
kbps). Assuming a
fixed Gaussian noise environment, how much must the average transmit
power
of the 4B3T system be increased to provide
the equivalent error rate
of a bipolar code?
4.12 Assume that crosstalk interference in a multipair cable system produces
an effect
equivalent to Gaussian noise at an equal power level. Using an error rate of 10-6
as a design objective, determine the effective degradation of the crosstalk on bi-
nary
$olar)
NRZ coding under each of the following conditions. (The
effective
PROBLEMS 2?3
degradation is determined as the increased transmit
power, in decibels, required
to achieve the desired error rate.)
(a) The crosstalk level is I 6 dB below the average signal level, but the crosstalk
is to be overcome on only one
pair (i.e., all other
pairs
$tay at a power level
for 10-6 BER with no crosstalk).
(b)
The crosstalk level is 16 dB below the average signal level, but the effects of
the crosstalk are to be overcome on all pairs' (Flfnt: Use signal-power-to-
noise-power ratios, not E/N^0.)
4.13 Repeat Problem 4.12 for bipolar coding.
4-14 How much does
-1
8 dB of crosstalk degrade the error rate
performance of a
po-
Iar binary NRZ signal for a BER of l0-7? (Assume that all transmitters are at
equal
power levels.) If crosstalk increases 15 dB
per
decade in frequency, what
is the relative crosstalk level of a four-level NRZ code carrying the same data
rate? What is the overall
performance penalty of the four-level system compared
to the two-level system?
4.15 How many distinct codewords of length 4 can be constructed from temary sym-
bols? How many of these codewords contain an equal number of
positive and
negative
pulses? How many codewords can be used to ensure a minimum of one
timing
pulse occurs in each codeword?
4.16 Can a set of ternary codewords of length 8 be constructed to encode binary data
using exactly four pulses per 8-bit word and containing equal numbers of
posi-
tive and negative pulses?
4.17 Given the input $equence +1,
-3,
+1,
-1,
+3, +3,
-3
of signal levels, determine
the sequence of output signal levels for each of the following correlative
encod-
i ngs.
( a) l +Dencoder
( b) 1- Dencoder
(c)
I
-.d
encoder
4.18 What is the
probability of a CRC error in a DSI signal with ESF framing if the
random BER is 10-7?
4.19 What is the minimum and maximum length of a correctable burst of errors in an
RS(204, 188) code where each symbol is an 8-bit byte?
4.20 Determine the probability of failure of an RS(7, 2) code operating in a random-
error environment with a bit error
probability of lQa. Assume each symbol is
an 8-bit byte.
4.21 Determine the
probability of failure of an RS(31, 15) code operating in a ran-
dom-error environment with a bit error
probability of l0-3. Assume each symbol
is an 8-bit byte.
4.22 Determine the parity error
pattern that results in a bit-by-bit constraint length
convolutional decoder if two data
(B)
bits in a row are comrpted and detected in
error but no other received bits are in error. Use the following bit sequence for
reference, where P0 is the first received data bit and B
1
and 82 are in enor. POB0,
P181, P282, P383.
DIGITAL SWITCHING
t'
'"'
''
Of the three basic elements in a communications
network
(terminals,
transmisrion *l-
dia, and switches), switches are the most invisible to the users yet represent the most
impoftant elements in terms of available service offerings. As mentioned in Chapter
l, a major r.nilestone was established in 1965 when stored-program control switching
was first introduced into the U.S.
public telephone network. Stored-program control
provides the means for implementing many innovative user services and for
greatly
simplifying switch adminishation and maintenance.
The use of computers to control the switching functions of a central office led to
the designation
"electronic"
switching
[e.g.,
electronic switching sy$tem
(ESS)
or
electronic automatic exchange
(EAX)]. However, the switching matrices of these
first-generation electronic switches are actually electromechanical in nature. The first
use of electronic switching matrices o.cqrned in France in 1971, when digital switch-
ing was applied to an end office envfriinment.
Ironically, these first digital switches
did not use stored-program control. Digital electronic switching matrices were first in-
troduced into the U.S.
public network in 1976 with AT&T's No. a
EsS-digital
toll
switch.
In the late 1970s digital class 5 switches began to be installed in the United States,
mo$tly in smaller end,offices as replacements for step-by-step switches. At that time
most major_ mettoi_oJif4n switching offices already had stored-program control from
No. I ESS or No. I EAX machines. Because the digital toll and end office switching
machines were initially installed in predominantly analog transmission environments,
their digital matrix provided no direct benefit to network customers. The motivation
for the digital machines wa$ reduced costs for the operating companies: lower main--
tenance, reduced floor space, simplified expansion, and lower manufabturing costs
tu.
By the mid-1980s the interoffice hansmission environment had changed to be al-
most exclusively digital. Thus analog-to-digital conversion costs moved from being
associated with digital transmission links to being associated with analog switches,
thereby further sealing the fate of analog toll or tandem switching technology. At this
time frame, end office ffansmission environments also began tJqe swing to a digital en-
vironment. Interoffice trunks were already digital, digital loop barrier systems became
226 DIGITAL SWITCHING
cost effective in metropolitan applications, remotely conmolled switching modules
with digital fiber interconnect became common for service to outlying communities,
digital cross-connect system$ (DCSs)
were being deployed, and the feeder portion of
the subscriber loop plant
began to use fiber.
The cost penalties
of interfacing to a digital transmission environment and higher
maintenance costs led to the older electromechanical class 5 switches being replaced
by digital machines when expansion or consolidation of an office occurued. The ability
to offer ISDN services was a lesser motivation for changing to digital switches be-
cause low-cost fiber transmission in conjunction with DCS sy$tems allows provision-
ing of these services from other offices.
This chapter describes the basic operation and implementation of digital tifiie divi-
sion switching as applied to PBXs, end offices, toll switche$, and cross connects. Be'
fbre digital switching is discussed, certain basic switching concepts and terminology
are introduced.
5. 1 SWI TCHI NGFUNCTI ONS
Obviously, the basic function of any switch is to set up and release connections be-
tween transmission channels on an
"as-needed
basis," The structure and operation of
a switch depend on particular applications. Three switching categories for voice cir-
cuits are local (line-to-line)
switching, transit (tandem)
switching, and call distribu-
tion.
The most common switching function involves direct connections between sub-
scriber loops at an end office or between station loops at a
pBX.
These connections
inherently require setting up a path through the switch fiom the originating loop to a
specific terminating loop. Each loop must be accessible to every other loop. This level
of switching is sometimes referred to as line switching.
Transit connections require
sening up a path from a specific incoming (originating)
line to an outgoing line or trunk group.
Normally, more than one outgoing circuit is
acceptable. For example, a connection to an interoffice trunk group
can use any one
of the channels in the
group.
Hence transit switching sffuctures can be simplified be-
cause alternatives exist as to which outgoing line is selected. Furthermore, it is not
even neces$ary that every outgoing line be accessible from every incoming line. Tran-
sit switching functions are required by all switching machines in the telephone net-
work. Some machines such as remote concentrators and toll or tandem switches
service only hansit traffic (e.g.,
do not provide
local connections). These concepts are
illustrated in Figure 5. l.
call distributors are often implemented with the same basic equipment as
pBXs.
The mode of operation (software)
is significantly different, however, in that incoming
calls can be routed to any available attendant. Normally, the software of an automatic
call distributor (ACD)
is designed to evenly distribute the arriving calls among rhe ar-
tendants. Although it is not an inherent requirement that every incoming line (trunk)
be connectable to every attendant, call distributors are normally designed to provide
accessibility to all attendants. Furthermore, it is often desirable that nonblocking op-
l\
5.2 SPACF DIVISION SWITCHING 227
Figure 5.1 Local and transit ftaffic switching examples.
erations be provided. (No matter what switch
paths are in use, a new request can be
serviced by any available attendant.)
5.2 SPACE DIVISION SWITCHING
Conceptually, the simplest switching structure is a rectangular array of crosspoints, as
shown in Figure 5.2. This switching matrix can be used to connect any one of Ninlets
to any one of M outlets. If the inlets and outlets ate connected to two-wire circuits, only
one crosspoint
per connection is required'*
Rectangular crosspoint EIIrays ale designed to
provide intergroup
(transit) connec-
tions only, that is, from an inlet group to an outlet
group. Applications for this type of
an operation occur in the following:
1. Remote concentrators
2. Call distributors
3. Portion of a PBX or end office
4. Single stages in multiple-stage
switch that provides transit switching
switches
In most of the foregoing applications, it is not necessary that the inlets be connectable
to every outlet. In situations involving large
groups of outlets, considerable savings in
total crosspoints can be achieved if each inlet can access only a limited number of out-
lets. When such a situation occurs,
"limited
availability" is said to exist. By overlap-
ping the available outlet
groups for various inlet groups, a technique called
"grading"
is established. An example of a
graded switching matrix is shown in Figure 5.3. Notice
*ln
fact, two
(and
sometimes thrce) switching contacts are associated with each ctosspoint of a two-wire
switoh, Since these contacts are pa-rt of a single unit and opetate in unison, they are considered a single
crosspoint.
228 DIGITAL SWITCHING
,y Ourlerr
Figure 5.2 Rectangular crosspoint alray.
that if outlet connections are
judiciously
chosen, the adverse effect of limited avail-
ability is minimized. For example, if inlets I and I in Figure 5.3 request a connection
to the outlet group,
outlets I and 3 should be chosen instead ofoutlets I and 4 to avoid
future blocking for inlet 2.
Graded switching structure$ were often used for access to large trunk groups
in
electromechanical switches where crosspointrr were expensive and individual switch-
ing modules were limited in size. Gradings were also used in individual switching
stages of large multiple-stage switches where more than one
path
to any particular
out-
let exists. Because very large digital matrices can be implemented with full accessi-
bility, graded
switch structures are no longer necessaly.
Intragroup switching, as in line-to'line switching, requires each line to be connect-
able to every other line. Thus full availability from all inlets ro all outlets of rhe switch-
ing matrix is required. Figure 5.4 shows two matrix structures that can be used to fully
interconnect two-wire lines. The dashed lines indicate that coffesponding inlets and
outlets of two-wire switching matrices are actually connected together to provide
bidi-
rectional transmission on two-wire circuits. For purposes
of describing switching ma-
trices, however, it is convenient to consider the inlets and ouflets of two-wire
switching matrices as being distinct.
I
2
3
{
l nl Cl i
_
E
6
I
I
r 2 3 {
Outlctr
Figure 5.3 Graded rectangular switching matrix.
5.2 SPACE DIVISION SWITCHING ?,29
(t) ft)
Figure 5.4 Two-wire swirching matrices:
(a)
square;
(b)
triangular
(folded)'
Both structures in Figure 5.4 allow any connection to be established by selecting a
single crosspoint. However, the square mahix, which is also called atwo-sided matrix,
allows any
particular connection to be established in two ways' For example, if input
link i is to be connected to input linkj, the selected crosspoint can be at the intersection
of inlet i and outlet
j-or
at the intersection of inlet
j
and outlet i. For simplicity these
crosspoints are referred to as
(f,i) and
(J,
i), respectively. In a typical implementation,
crosspoint
(i,7) is used when input i requests service, and crosspoint
(j' i) is used when
input
j
reque$ts service.
In the triangular matrix of Figure 5.4 the redundant crosspoints are eliminated. The
crosspoint reduction does not come without complications, however. Before setting
up a connection between switch input i and swirch inputi, the switch control element
must determine which is larger: d orj. If i is larger, crosspoint
(i,i)
is selected' If i is
smaller, crosspoint
(.7,
i) must be selected. With computer*controlled swirching, the
line number comparison is trivial. In the older, electromechanically
conholled
switches, however, the added complexity of the swirch control is more significant.
I
2
Inlorr i
i
N
t 2 3 '
i
Outlrtt
Flgure 5.5 Four-wire switching matrix'
Inlrtil
out|.t
230 DIGITAL SWITCHING
Switching machines for four-wire circuits require separate connection$ for the go
and return branches of a circuit. Thus two separate connection$ must be established
for each service request. Figure 5.5 depicts a square-matrix structure used to provide
both connections. The structure is identical to the square matrix shown in Figure 5.4
for two-wire switching. The difference, however, is that corresponding inlets and out-
lets are not connected to a common two-wire input. All of the inlets of the four-wire
switch are connected to the wire pair
carrying the incoming direction of transmission,
and all of the outlets are connected to the outgoing pairs.
When setting up a connection
between four-wire circuits i andj, the matrix in Figure 5.5 must select both crosspoints
(i, j)
and
(j,
f). In actual operation these two crosspoints may be selected in unison and
implemented as a common module.
5.2.1 Multlple.$tage Swltching
In the switching $tructures described to this point,
an inlet is connected directly to an
outlet through a single crosspoint. (Four-wire
switches use two crosspoints per
con-
nection, but only one for an inlet-to-outlet
corurection.) For this reason, these switch-
ing structures are referred to as
"single-stage"
switches. Single-stage switches have
the property
that each individual crosspoint can only be used to interconnect one par-
ticular inlet-outlet pair.
since the number of inlet-outlet pairs is equal to N(N
-
lyz
for a triangular array, and N(N
-
l) for a $quare array, the number of crosspoints re-
quired
for a large switch is prohibitive.
Furthermore, the large number of crosspoints
on each inlet and outlet line imply a large amount of capacitive loading on the message
paths.
Another fundamental defrciency of $ingle-stage switches is that one specific
crosspoint is needed for each specific connection. Ifthat crosspoint fails, the associ-
ated connection cannot be e$tablished. (An
exception is the square, two-wire switch
that has a redundant crosspoint for each potential
connection. Before the redundant
crosspoint could be used as an alternate path,
however, the inlet-oriented selection al-
gorithm
would have to be modified to admit outlet-oriented selection.)
Analysis of a large single-stage switch reveals that the crosspoints are very ineffi-
ciently utilized. Only one crosspoint in each row or column of a square switch is ever
in use, even if all lines are active. To increase the utilization efficiency of the cross-
points,
and thereby reduce the total number, it is necessary that any par-ticular
cross-
point
be usable for more than one potential
connection. If crosspoints are to be shared,
however, it is also necessary that more than one path be available for any potential
connection so that blocking does not occur. The alternate paths
serve to eliminate or
reduce blocking and also to provide protection
against failures. The sharing ofcross-
points for potential paths
through the switch is accomplished by multiple-stage
switching. A block diagram of one particular
form of a multiple-stage switch is shown
in Figure 5.6.
The switch of Figure 5.6 is a three-stage switch in which the inlets and outlets are
partitioned
into subgroups ofNinlets and Noutlets each. The inlets ofeach subgroup
are serviced by a rectangular array of crosspoints. The inlet arrays (first
stage) are
n x ft arays, where each of the ft outputs is connected to one of the ft cenrer*smge ar-
5.2 SPACE DIVISION SWITCHING 231
Figure 5.6 Three-stage switching maffix.
rays. The interstage connections are often calledjunctors.
The third stage consists of
ft x n rectangular artays that
provide corxrections from each center-stage
array to the
groups of n outlets, All center-stage arrays we N/n x N/n arrays that
provide connec-
tions from any first-stage alray to any third-stage array. Notice that if all affays
pro-
vide full availability, there are ft
possible paths through the switch for any particular
connection between inlets and outlets. Each of the ft
paths utilizes a separate center-
stage array. Thus the multiple-stage
structure
provides altemate
paths through the
swit"6 to circumvent failures. Furthermore, since each switching link is connected to
a limited number of crosspoints, capacitive loading is minimized'
The total number of crosspoints N1E required by a three-stage switch, as shown in
Figure 5.6, is
(s.1)
where N
=
number of inlets-outlets
n= size ofeach inlet-outlet
group
ft
=
number of center-stage arrays
As is demonstrated $hortly, the number of crosspoints defined in Equation
5. I can
be significantly lower than the number of crosspoints
required for single-stage
switches. First. however. we must determine how many center-stage arrays are needed
to provide satisfactory service.
M
rt!il
Nx=2Nft.-[#l
232 DIGITAL SWITCHING
Nonblocklng
Switches
One attractive feature of a single-stage switch is that it is strictly nonblocking. If the
called party
is idle, the desired connection can always be established by selecting the
particular
crosspoint dedicated to the particular
input*output pair.
When crosspoints
are shared, however, the possibility
of blocking arises. In 1953 charles clos
[2]
of
Bell Laboratories published
an analysis of three-stage switching networks showing
how many center-$tage arrays are required to
provide
a strictly nonblocking operation.
His result demonstrated that if each individual array is nonblocking,
and if the number
of center stages ft is equal to 2n
-I,
the switch is strictly nonblocking.
The condition for a nonblocking operation can be derived by first observing that a
connection through the three-stage switch requires locating a center-stage array with
ar idle link from the appropriate firsr $tage and an idle link ro the appropriate third
stage. since the individual arays themselves are nonblocking, the desired path can be
rtet up any time a center stage with the appropriate idle links can be located. A key
point in the derivation is to observe that since each first-stage array has n inlets, only
n
-
I of these inlets can be busy when the inlet corresponding to the desired connection
is idle. If ft is geater
than n
-
l, it follows that, at most, n
-
I links to cenrer-srage
irrray$ can be busy. Similarly, at most n
-
I links to the appropriate third-stage array
can be busy if the outlet of the desired connection is idle.
The worst-case situation for blocking occurs (as
shown in Figure 5.7) it alln
-
r
busy links from the first-stage array lead to one set ofcenter-stage arays and ifall n
-
I busy links to the desired thfud-stage Errray come from a separate set of center-stage
arrays. Thus these two sets ofcenter-stnge
arrays are unavailable for the desired con-
nection. However, if one more center-stage array exists, the appropriate input and out-
put links must be idle, and that center stage can be used to set up the connection. Hence
i f k=(n
*
1) + (n
-
l ) + 1
=2n-
1, the swi tchi s sri ctl ynonbl ocki ng.
substi tuti ng
t l
l r
L- - J
ArnlhblG
pffh
Flgure 5.7 Nonblocking three-srage switching matrix.
this value of t into Equation 5.1
three-stase switch
5.2 SPACE DIVISION SWITCHING ?33
reveals that for a strictly nonblocking operation of a
Nx=2N(2n-
(5.21
As expressed in Equation 5.2, the number of crosspoints in a nonblocking three-
stage swiich is dependent on how the inlets and outlets are
partitioned into subgroups
of iize n. Differentiating Equation 5.? with respect to n and setting the resulting ex-
pression equal to 0 to determine the minimum reveal that
(for large M) the optimum
value of n is
(N/z)Ltz. Substituting this value of n into Equation 5.2 then
provides an
expression
for the minimum number of crosspoints of a nonblocking
three-stage
switch:
N*(min)
=4N({2N
-
l)
(s.3)
t ) +( 2n- t , [ r OI
where N
=
total number of inlets-outlets.
Table 5.1
provides a tabulation of Nx(min) for various-sized nonblocking three-
stage switchei and compares the values to the number of crosspoints in a single-stage
*qu*" matrix. Both switching structure$ inherently
provide four-wire capabilities,
a
requirement
for digital switches because voice digitization implies four-wire circuits.
As indicated in Table 5.l, a three-stage switching matrix
provides significant re-
ductions
in crosspoints,
particularly for large switches. However, the number of cross-
points for large three-stage switches is still
quite prohibitive. Large switches typically
o*" -o.* than three Stages to
provide greater reductions in crosspoints'
For example,
the No. 1 ESS uses an eight-stage switching
matrix that can service up to 65,000 lines'
The most significant
reductions in crosspoint numbers are achieved not so much from
additional siages but by allowing the switch to introduce acceptably low
probabilities
of blocking.
TABLE 5.1 Croeepoint Requlrsments of Nonblocklng Swltches
Number of Lines
Number of Crosspoints
fot
Three-Stage Switch
Number of CrossPoints
for
Single*Stage Switch
128
51?
2,048
8, 192
32,768
131, 072
7, 680
63,488
516, 096
4, 2 mi l l i on
33 mi l l i on
?68 mi l l i on
16, 2s6
261, 632
4. 2 mi l l i on
67 mi l l i on
1 bi l l i on
17 bi l l i on
234 DIGITAL SWIT0HING
5.2.2 Blocklng Probablilties:
Lee Graphs
Strictly nonblocking switches are rarely needed in most voice telephone networks.
Both the switching systems and the number of circuits in interoffice trunk groups
are
sized to service most requests as they occur, but economics dictates that network im-
plementations
have limited capacities that are occasionally
exceeded during peak-traf-
fic situations. Equipment for the public
telephone network is designed to provide
a
cerlain maximum probability
of blocking for the busiest hour of the day. The value of
this blocking probability
is one aspect of the telephone company's grade
of service.
(other
aspects of grades
of service are availability, tran$mission quality,
and delay in
'tetting up a call.)
A typical residential
telephone is busy 5,l0va of rhe time during the busy hour.
Business telephones are often busy for a larger percentage
of their busy hour (which
may not coincide with a residential busy hour). In either case, network-blocking
oc-
currences on the order of l7o* during the busy hour do not represent
a significant re-
duction in the ability to communicate since the called party
is much more likely to
have been busy anyway. under these circumstances, end office switches and, to a
lesser degree, PBXs can be designed with significant reductions in crosspoints
by al-
lowing acceptable blocking probabilities.
There are a variety of techniques that can be used to evaluate the blocking prob-
ability of a switching matrix. These techniques vary according to complexity-, accu-
racy, and applicability to different network structures. One of the most versatile and
conceptually
straightforward approaches of calculating blocking probabilities
in-
volves the use of probability graphs
as
proposed
by c.
y.
Lee
[3].
Although this tech-
nique require$ several simpliffing approximations, it can provide reasonably
accurate
results, particularly
when comparisons of altemate structures are more important than
absolute numbers. The greatest value of this approach lies in the ease of formulation
and the fact that the formulas directly relate to the underlying network structures. Thus
the formulations help provide
insight into the network $tructures and how these struc-
tures might be modified to change the performance.
In the following analyses we are determining the blocking probabilities
of various
switching rrtructure$
using utilization percentages,
or
,,loadings,"
of individual links.
The notationp will be used, in general,
to represent the fraction of time that a
particular
link is in use
(i.e.,
p is the probability
that a link is busy). In addition ro a utilizarion
percentage
or loading, p is also sometimes referred to as an occupancy. The prob-
ability that a link is idle is denotedby q
=
|
-
p.
when any one of n parallel
links can be used to complete a connection, the com*
posite blocking probability
B is the probability
rhat all links are busyt;
B= p n
(5.4)
'Transmission
and switching equipment in the public
network is normally designed for even lower blocking
.probabilities
to provide for growth in the traffrc volume.
'Equations
5.4 and 5.5 assume each link is busy or idle independently of other links.
s.z $PAoE DlvlsloN swlrcHlNc
205
When a series of n links are all needed to complete a connection' the blocking
prob-
ability is most easily determined as I minus the
probability that they are all available:
B = l
- {
(5.s1
A probability
graph of a three-stage network is shown in Figure 5.8. This
graph re-
lates the fact that any
pafiiculaf connection
can be established
with ft different
pathsl
one tlrough each center-stage alray. The
probability that any
palticular interstage link
is busy is denoted by
p'. The
probability of blocking for a three-stage
network can be
determined as
ff
=
probability that all
paths are busy
= (probability that an arbitrary
path is busy)e
= (probability that at least one link in a path is busy)ft
= ( l - qa) k
where ft
=
number of center-stage
alrays
q'= probability that an interstage link is idle,
-
|
-
P''
If the
probability p that an inlet is busy is known, the
probability p' that an inter-
stage link is busy can be determined
as
(5.6)
(s.7)
p,=fr
tr< F)
where
p =
Mn. Equation 5.7
presents the fact that
when some number of inlets
(or
out-
Iets) are busy, the same number of first-stage outputs
(or third-stage inputs) are also
busy. However, there are
F
=
Vn times as many interstage links as there are inlets or
outlets. Hence the
percentage of intersgge links that are busy is reduced by the factor
p'
The factor
p
is defined as though ft > n, which implies that the first stage of the
switch is
providing $pace expansion
(i.e., switching some number of input links to a
larger number of output links). Actually,
p
may be less than 1, implying
that the first
P ' i e
Figure 5.8 Probability
graph of three-stage network'
236 DtctrAL swtrcHtNc
stage is concenffating the incoming
traffic. First-stage concentration
has been used in
end office or large PBX switches where the inlets are lightly loaded (S_l}vo).In
tan_
dem or toll offices, however, the incoming trunks are heavily utilized, and expansion
is usually needed to provide
adequately low-blocking probabilities.
substituting Equation 5.7 into Equation 5.6 provides
a complete expression for the
blocking probability
of a three-stage switch in terms of the inlet utilizationp:
(s.8)
Table 5.2 tabulates numbers of crosspoints obtained from Equation
5.g for the
same switch sizes presented
in Table 5.1. The number of center arrays was chosen in
each case to provide
a blocking probability
on the order of 0.002. The inlet urilization
in each example was assumed to be 107o. Notice that the designs with small but finite
blocking probabilities
are significantly
more cost effective than nonblocking
designs.
The swirch designs in Table 5.2 assume that the inlets are only l0zo busy, as might
be the case for an end office
switch or a PBX. The dramatic savings in crosspoints for
large switches is achieved by introducing
significant concentration factors
[yp;
into
the middle stage. when the inlet utilization is higher (as
typically occurs in tandem
switches), high concentration
factors are not acceptable, and the crosspoint require-
ments therefore increase. Table 5.3 lists corresponding
crosspoint requirements
and
implementation paraffreters
for inlet loadings of
'l\Vo.
The results presented
in Tables 5.2 and 5.3 indicate that very large switches still
require prohibitively
large numbers of crosspoints, even when blocking is allowed. As
mentioned previously,
very large switches use more than three stages to provide
fur-
therreductions
in crosspoints. Figure 5.9 shows ablock diagram of a five-stage switch
obtained by replacing every center-$tage array in Figure 5.6 with a three-stage array.
This particular
structure is not optimum in terms of
providing
a given level of perform-
ance with the fewest crosspoints,
but it is a useful design because of its modularity.
(Furthermore,
it is a lot easier to analyze than some other five-stage sfucrures.)
If the middle tlree stages of a five-stage swirch as shown in Figure 5.9 are strictly
nonblocking (kz =
Znz
-
I
),
the design provides
a savings of 9704 crosspoints in each
TABLE 5.2 Three-Stage Switch Deslgns for Blocking Probabititlee of 0.002 and Intet
Utlllzation of 0.1
I r r z' 1f t
'=
L' ['
-fii
]
Switch
Size, N
Number of
Crosspoints
Number of Crosspoints
in Nonblocking Design
't28
512
2,048
8, 1S2
32,768
131,072
I
1 6
32
64
128
256
5
7
1 0
1 5
24
41
0.625
0.438
0. 313
0.234
0, 188
0. 160
2,560
14,336
81, 920
491,520
3.1 mi l l i on
21 .5 mi l l i on
7,680 (k= 15)
63,488 (k =
31)
516,0e6 (k=
63)
4,2 mi l l i on (k= 127)
33 million (,( =
255)
268 mi l l i on (k =
51 1
)
5.2 SPACE DIVISION SWITCHING
237
TABLE E.l Three-Stage Switch Deslgns for Blocking Probabllltles of 0.002 end lnlet
UtllltationB of 0.7
Switch
Size N
Number of Number of CrossPoints
in
n
k
B
Crosspoints
Nonblocking Design
128
512
2,048
8, 192
32,768
131, 072
14 1. 75
22 1. 38
37 1. 16
64 1. 0
1 16 0. 91
215 0.84
7, 168
45,056
303,104
2.1 mi l l i on
15.2 mi l l i on
113 mi l l i on
7,680
63,488
516,096
4.2 mi l l i on
33 mi l l i on
268 million
( k= 15)
( k
=
31)
(k
=
63)
(k
=
127)
(k
= 255)
( k =
511)
I
1 6
32
64
128
256
center-stage array of the 32,7681ine, three-stage
switch designs
presented earlier'
Hence a littl* one, I million crosspoints are saved in the 32,768-line
tandem switch
design of Table 5.3. Since the middle stages do not introduce blocking,
the
perform-
ance of this five-stage switch is identical to the
performance of the three-stage design'
Naturally, a more cost-effective design could be obtained by allowing smgll amounts
of bloclcing in the middle stages. The probability
graph of the five+tage switch is shown
in Figure 5-.10. From this
graph, the blocking
probability is determined
as follows:
B
=
ll
-
(sr)z
[t
-
(1
-
d)t' ]]r'
azx*z
;frx ffirr,,,
2 3 4
(s.e)
1 1 1 i l 1
fl 1 r 11
Flgure 5.9 Five-stage swirching
network'
238 DIGITAL
$WITCHING
rz
=
rrfir
rfir
Flgure 5.10
probability
graph
of five_stage network.
where
4r
-
|
-
h
and, qr= I
-
Or.
Even greater
crosspoint reductions
can, of course, be achieved by using more
stages to replace the rather large first- and third-stage
affays. For example, the total
number of crosspoints
in the 32,0001ine
switch can be reduced to less than 3 million.
The 130,000-line
switch is not practical
with electromechanical
switching
makices
but is well within the capabilities of a digiral time division switch.
5.2.3 Blocklng
Probabiliiles:
Jacobaeus
The formulations
of blocking probability
obtained from probability
graphs
enrail sev-
eral simplifying
assumptions. one of these assumptions involves expressing the com-
posite
blocking probability
of the alternate paths as the produci
of the blocking
probabilities
of each individual path.
This step assumes that the individual prob-
abilities are independent.
In fact, the probabilities
are not independent, particutarty
when significant amounts of expansion
are present.
Consider a switching matrix with
k=Zn
-
1. Equation 5.8 produces
a finite blocking probability
even ttrough the switch
is known to be strictly nonblocking.
The inaccuracy
results because when 2n
*
z paths
are busy, the remaining path
is assumed to be busy with a probability
of I
-
(qjr.
In
fact, the remaining path
is necessarily idle.
In general,
when space expansion exists, the assumption
ofindependent individual
probabilities
leads to an erroneously high value of blocking. Ttrelnaccuracy
results
2 Ft
,-1/3-1i.15--\,^.
Pr
5.2 SPACE DIVISION SWITCHING 239
because as more and more
paths in a switch are found to be busy, the remaining
paths
are less and less likely to be in use
(only
a subset ofn ofthe interstage links can ever
be busy at any one time).
A more accurate but not exact analysis of multistage switching matrices was
pre-
sented in 1950 by C. Jacobaeus
[4].
Although the analysis is conceptually stoaightfor-
ward, it does involve a considerable amount of manipulation that is not
presented here.
The resulting equation for a three-stage switch is obtained from reference
[5]
as
(s.10)
where n
=
number of inlets
(outlets) per f,trst-
(third-) stage array
11= number of second-stage
IIIrays
p- inletutilization
In the interest of comparing the two methods, Equations 5'6 and 5.10 have been
evaluated for three-stage switches with varying amount$ of space expansion.
The re-
sults were obtained for inlet utilization of 0.7 and are
presented in Table 5'4'
Table 5.4 reveals that the two analyses are in close agreement
for near-unity expan-
sion factors. In fact, if
p =
l, the two formulations
produce identical results' As ex-
pected, the Lee graph analysis
(Bquation 5.8)
produces overly
pessimistic values for
the blocking
probability when
p
> 1.
As another comparison between the two approaches, Table 5.5 is included to dem-
onstrate the use of Equations 5.8 and 5.10 for switches with significant
amounts of
concenfration made
possible by a relatively low inlet utilization of 0.1.
Table 5.5 reveals that a Lee
graph analysis
(Equation 5.8) consistently underesti-
mates the blocking
probability when concentration exists. Actually, the Jacobaeus
analysis
presented in Equation 5.10 also underestimates the blocking
probability if
large concenhation
factors and high blocking
probabilities are used. When necessary,
more accurate techniques can be used for systems with high concentration
and high
TABLE S.4 Comparison of Blocking Probability Analyses
(P =
0'7)"
Jacobaeu$
Equat i on 5. 10
u=ffipk(2-il,*
Number of Center
Stages, k
Space Expansion,
F
Lee
Equati on 5.8
1 4
1 6
20
24
2B
31D
0,875
1. 0
1. 25
1. 50
1. 75
1. 94
0.548
0.221
0. 014
3. 2 x 10+
3. 7 x 104
8.5 x 104
0.598
0.221
0.007
2. 7 x 10+
7.7 x 104
0, 1 x 10- 12
aswi tch
si ze N = 512; i nl et
group 8i z n
=
16i i nl et ul i l i zati on
p = 0.7.
Di,,lonblocking.
240 DIGITAL SWITCHING
TABLE 5.5 Comparison of Blocking
probablllty
Anatyses (p=
0.1)"
Number of Center
Stages, k
Space Expansion,
F
Lee
Equation 5.8
Jacobaeus
Equati on 5.10
1 0
12
1 4
16
0.375
0.5
0.625
0.75
0.875
1. 0
0, 0097
2. 8 x 104
4. 9 x 10+
5. 7 x 104
4. 0 x 10-10
2. 9 x 10* 12
0.Q27
8. 6 x 10+
1. 5 x 10+
1. 4 x 10- 7
7. 8 x 1f l o
2. 9 x 10-12
aswi tch
si ze N
=
512; i nl et group
si z6 n = 16; Inl r uti l i zati on
p =
0.1
blocking. However, switches with high blocking probabilities
normally have no prac-
tical interest so they are not considered here.
Users of PBXs sometimes experienced high blocking probabilities,
but blocking in
these cases usually arises from too few tie lines to other corporate locations or too few
trunk circuits to the public
network. The subject of blocking in trunk groups
is heated
in Chapter 12.
up to this point,
the blocking probability
analyses have assumed that a specific in-
let is to be connected to a specific outlet. Also, it has been assumed that the requests
for service on the individual
lines are independent. These assumptions are generally
valid for swirching one subscriber line to another in an end office switch or for con-
necting one station to another in a PBx. Neither of these assumptions apply to con-
nections to or from a fiunk group.
when connecting to a trunk circuit, any circuit in a trunk group
is acceptable. Thus
the blocking probability
to a specif,rc circuit is only as important as its significance in
the overall blocking to the trunk group.
The blocking probability
to any particular
cir-
cuit in a trunk group
can be relatively large and still achieve a low composite blocking
probability
to the trunk group
as a whole. If the blocking probabilities
to the individual
trunks are independent, the composite blocking probability
is the product
of the indi-
vidual probabilities.
However, the paths
to the individual trunk circuits normally in-
volve {iome common links (e.g.,
the
junctors
from a first-stage array to all
second-stage arrays). For this rea$on the individual blocking probabilities
are nor-
mally dependent, which must be considered in an accurate blocking probability
analy-
sis.
As an extreme example,
consider a case where all trunks in a trunk group are as-
signed to a single outlet array in a three-stage switch. since the paths from any par-
ticular inlet to all trunks in the group
are identical, the ability to select any idle 61nk
is useless. In practice,
the individual circuits of a trunk group should be assigned to
separate outlet arays.
Another aspect of fiunk groups
that must be considered when designing a swirch
or analyzing the blocking probabilities
involves the interdependence of activity on the
s.z sPAcE DtvtstoN swlrcHlNc
241
individual circuits within a trunk
group. In contrast to individual subscriber
lines or
PBX stations, individual circuits in a trunk
group are not independent in terms of their
probabilities of being busy or idle. lf some number of circuits in a trunk group are
tested and found to be busy, the
probability that the remaining circuits are busy is in-
creased. The nature of these dependencies is discussed more fully in Chapter 12. At
this
point it is only necessary to point out that these dependencies cause increased
blocicing
probabilities if the individual ffunks are competing for common
paths in the
switch. ,l,gain, the effect of these dependencies
is minimized by assigning the individ-
ual trunks to separate inlet-outlet arrays so that independent
paths are involved in con-
nections to and from the trunk
group. This
process is sometimes
referred to as
decorrelating the trunk circuits.
One last aspect of the blocking
probability as a grade of service
parameter that must
be mentioned involves variations in the loading of the network by individual users. In
the design examples for an end office
presented earlier, it was tacitly assumed that all
subscribers are busy l07o of the time during a busy hour' In fact, some subscribers
are
active much more than I 07o of the time, and other subscribers are active less than l07o
of the time. In terms of haffic theory, some subscribers
present more than 0.1 erlangs*
of traffic to the network, whereas others
present less'
When a switch is partitioned into subgroups
(as
all Iarge switches must be) and the
haffic is concentrated by first-stage switching alrays, a few overactive subscribers
in one subgroup can significantly degrade service for the other subscribers
in the
subgroup. It does not matter that the subscribers
in some other subgroup may be
explriencing lower than average blocking. Their essentially
nonblocking service
is no compensation
for those subscribers experiencing a relatively
poor grade of
service.
Operating companies have traditionally solved the
problem of overactive
subscrib-
ers by specifically assigning the most active lines
(businesses) to separate inlet
groups
of the switctr. Sometimes this procedure requires making traffic measurements to de-
termine which lines are most active and reassigning these lines to different
palts of the
switch. These procedures fall into the general category of line administration.
If the
subgroups are large enough, or ifthe designs
provide adequate margin for overactive
ure.s, thir aspect of line administration
can be minimized. One feature of a digital
switch that can be utilized in this regard is the ability to design economical
switches
with very low nominal blocking
probabilities so that wide variations in traffic inten-
sities can be accommodated.
Even a modern digital switch can experience
loading
problems when
confronted
with extreme traffic conditions. An Internet service
provider
(ISP) in a metropolitan
area may atffact an extremely large amount of traffic that all
passes through a single
class 5 switch. Although the connections to the ISP are lines, as far as
the switch is
concerned, they actually represent a trunk
group with very heavy haffic, so much so
that $pecial line adminishation
is required.
*An
erlang is a measure oftraffic intensity specifying the proporlion oftime that a device is busy' A circuit
is said to carry 0.1 erlang if it is busy l0olo of the time.
242 DIGITAL SWITCHING
5.2.4 Folded Four-Wire Switchee
Multiple-stage switches can be used for either two- or four-wire switching operations.
Figure 5.ll depicts a four-wire connection through a four-stage switch. Notice that
two paths
must be establi$hed for the complete connection. A two-wire connection re-
quires only one path
since each outlet is extemally connected to its corresponding in-
let.
The two paths
shown in Figure 5.1 I demonshate a particularly
useful relationship;
One path is a mirror image of the other path.
If the switch diagram is folded about the
vertical center line, the paths
coincide. Hence this method of setting up connections is
sometimes referred to as a folded operation. When all connections in the switch are
set up with a folded relationship, several benefits result.
First ofall, only one pathfinding
operation is needed since the reverse path
is auto-
matically available as a mirror image of the forward path.
In essence, every crosspoint
on one side is paired
with another crosspoint in a corresponding array on the opposite
side of the switch. Whenever one crosspoint of a
pair
is used in a connection, the other
crosspoint in the pair is also used. For example, the third inlet array in the first stage
uses crosspoint (6,4)
to connect its sixth inlet to its fourth outlet (leading
to the fourth
artay of the second stage). The corresponding crosspoint in the third outlet array of
the last stage connects its fourth inlet (coming
from the fourth array in the four-th stage)
to its sixth outlet. [n general,
crosspoint f,7 in one array is paired
with crosspointj, I
in the conesponding array on the opposite side of a switch. Since the availability of
one crosspoint in a pair
ensures the availability ofthe other, the rever$e path
is auto-
matically specified and available.
A second advantage of the folded four-wire operation results because the amount
of information specifying the status of the switch can be cut in half. Only the status of
grth
Inldt
of thlrd arEy of rfiird rnv
El6|rrfr inlit
( 7
l l t
Eldudnfi or|tl|t
of liftdnfi my
of flftrnfi ffr.y
a
t
I
a
I
a
Figure 5.11 Four-wire connection through four-stage switch.
5.2 SPACE DIVISION SWITCHING 245
each pair of crosspoints or associated
junctor
is needed to find an available
path
through the switch.
A third benefit of the folded structure occurs because the blocking
probability is
one-half of the
probability of finding two paths independently. It might seem that
pair-
ing the crosspoints in the described manner would reshict the
paths available for a par-
ticular connection. On the contrary, the crosspoint
pairings merely guarantee that a
reverse
path is automatically available for any selected
path in the forward direction'
The folded operation in the
preceding paragraphs referred to a switch
with an even
number of switching stages. An even number was chosen because the concept is easi-
est to demonstrate when no center stage is
present. The basic approach can be ex-
tended to odd numbers of switching stages if the center stage contains an even number
of anays and is folded about a horizontal line at the center of the stage. In this manner,
crosspoint i,7 in the top center-stage alray is paired with crosspointj, i in the bottom
center-$tage array, and so on.
5.2.5 Pathfi ndi ng
Determining a
path through a single-stage switch is virtually automatic since the nec-
essary crosspoint is uniquely specified by the inlet-outlet
pair to be connected. In con-
trast, availability of more than one
path in a multiple-stage switch complicates the path
selection
process. The catl
processor of the switch must keep track of which
potential
paths for a
particular connection are available in a state store. A pathfinding routine
processes the state store information to select an available
path. Whenever a new con-
nection is established or an old one released, the state store is updated with the appro-
priate information.
Pathtlndlng Times
Pathfinding bperations require the use of common equipment and must
therefore be
analyzed to determine the rate at which connect requests can be
processed. The time
required to find an available
path is directly dependent on how many potential paths
are tested before an idle one is found. Some system$ can test a number of
paths in par-
allel and thereby shorten the processing time. Since the expected number of
potential
paths that must be tested to find an idle path is a function of link utilization,
pathfind-
ing times unfortunately increase when the cornmon control equipment is busiest.
Assume that the
probability of a complete
path through the swirch being busy is de-
noted by p. If each of k
possible paths through the switch has an equal and independent
probability of being busy, the expected number of
paths No that must be tested before
an idle path is found is determined in Appendix A as follows:
( s. l
r )
Example 5.1. What is the expected number of potential paths that must.be tested to
find an idle
path in the three-stage, 8192Jine switch defined in Table 5.2?
N,:H
244 DretTAL swtrcHtNc
solution. As indicated in the table, a $pace expansion factor of 0.234 is used to
provide a blocking probability
of 0.002. Hence the utilization of each interstage link
is 0.1/0.234
=0.427.
The blocking probability
of each path
through the switch is
merely the probability
that one of two links in series is busy. Hence the probability
p
=
|
*
(I
-
O.427)z
=
0.672, and the expected number of paths
to be resred is
Np=
I
-
(0.672)15
|
-0.672
=3. 04
Example 5. I demonstrates that, on average, only 3 of the 15 potential paths
need
to be tested before an idle path is found. However, when the switch is experiencing
greater than normal ffaffic loads, the average number of
paths
tested increases. For ex-
ample, if the input line utilization increases from 10 to l|Vo, the blocking probability
increases from 0.002 to 0.126, and the expected number of
paths
to be tested in the
pathfinding
operation increases from 3 to 4.g.
Although this simple pathfinding
example is, by itself, not particularly
important,
it demonstrates a very important aspect of the loading of common equipment in a
switching system (or
a network as a whole): Linear increases in ttre offered traffic pro-
duce geometric
increases in the utilization of the network resources. If a system is de-
signed and analyzed under nominal traffic conditions and the results are extrapolated
to determine total capacity, greatly
optimistic estimates often occur. In the pathfinding
example the
processing
requirements increase just
when the call processor (stored-
program
conhol) is already loaded with greater than average traffic volumes. For fur-
ther discussion of the effects of traffic loads on the common equipment of a switching
sy$tem see Schwartz
[6].
5.2.6 Switch Matrix Control
When an available path through a cofilmon control switching network is determined,
the control element of the switch transfers the necessary information to the network
to select the appropriate crosspoints. Crosspoint selection within a matrix is accom-
plished
in one of two ways. The control may be associated with the output lines and
therefore speciff which inputs are to be connected to the associated outputs or the con-
trol information may be associated with each input and subsequently speciff to which
outputs the respective inputs are to be connected. The first approach is referred to as
output-associated control while the second is called input-associated control. These
two control implementations are presented
in Figure 5.12.
Input-associated
control was inherently required in step-by-step switches where
the information (dial pulses)
an-ived on the input link and was used to directly select
the output links to each successive stage. [n common control system$, however, the
address information of both the originating line and the terminating line is simultane-
ously available. Hence the connection can be established by beginning at the desired
5.2 SPACE DIVISION SWITCHING 245
Figure 5.12 Switch matrix conffol:
(a)
output associated;
(b) input associated.
outlet and proceeding backward through the switch while selecting inputs to each
stage.
The implementation of both types of digital crosspoint arrays using standard com-
ponents is shown in Figure 5.13. Output-associated control uses a conventional data
selector/multiplexer for each matrix output. The number of bits required
to control
each data selector is log2 N, where N is the number of inlets' Thus the total number of
bits required to completely specify a connection configuration
is M log2 N.
Input-associated control can be implemented using conventional line decoders/de-
multiplexers. The outputs are conmoned using a
"wired-or"
logic function. Thus the
output
gates of each decoder circuit must be open-collector or histate devices if tran-
sistor-transistor-logic
(TTL) is used. The total number of bits required to specify a
connection configuration in this case is N log2 M.
A significant drawback of input-associated control arises from the need to disable
unused inputs to prevent cross connects when another input selects the same output.
Figure 5.13 Standard component implementation of digital
associated control;
(b) input-associated control.
ft)
crosspoint aray:
(a)
output-
Dfir rclictot/mu
ltiPloxor
loga N
Linr dccoder/demultiplexer
log? rY
246 DIGITAL SWITCHING
with output-associated conffol, unused output$ can remain connected to an input
without preventing
that input from being selected by another output. For this reason
and for generally greater
speeds of operation, digital switching networks typically use
output-associated control. Notice, however, that the total amount of information
needed to specify a connection configuration with input-associated
control is less than
that with output control if the number of inputs N is much smaller than the number of
outputs M
(N
log2 M < M log2Il). Furthermore, input-associated
conffol is more flex-
ible in terms of wired-or (e.g.,
bus) expansion.
5.3 TIME DIVISION SWITCHING
As evidenced by multiple-stage switching, sharing of individual crosspoints for more
than one potential
connection provides
significant savings in implementation costs of
space division switches. In the cases demonstrated, the crosspoints of multistage space
switches are shared from one connection to the next, but a crosspoint assigned to a par-
ticular connection is dedicated to that connection for its duration. Time division
switching involves the sharing of crosspoints for shofier periods
of time so that indi-
vidual crosspoints and their associated interstage links are continually reassigned to
existing connections. When the crosspoints are shared in this manner, much greater
savings in crosspoints can be achieved. rn essence, the savings are accomplished by
time division multiplexing the crosspoints and interstage links in the same manner that
transmission links are time
division muttiplexed to share interoffice wire pairs.
Time division switching is equally applicable to either analog or digital signals. Ar
one time, analog time division switching was attractive when interfacing to analog
transmission facilities, since the signals are only sampled and not digitally encoded.
However, large analog time division switches had the same limitations as do analog
time division ffansmission links: the
pAM
samples are particularly
vulnerable to
noise, distortion, and crosstalk. Thus, large electronic switching matrices have always
incorporated the cost of digitizing PAM samples ro maintain end-to-end signal qual-
ity. The low cost of codecs and the prevalence
of digital trunk interconnections
imply
that analog switching is now used in only the smallest of switching system$ (e.g.,
elec-
tronic key systems).
5.3.1 Analog Time Dlvlsion Swltching
Although analog time division swirching has become obsolete, it is a good starting
point
to establish the basic principle
of time division switching. Figure 5.14 depicts a
padicularly
simple analog time division switching structure. A single switching bus
supports a multiple number of connections by interleaving
pAM
samples from receive
line interfaces to ffansmit line interfaces. The operation is depicted as though the re-
ceive interfaces are separate from the respective transmit interfaces. When connecting
two-wire analog lines, the two interface$ are necessarily implemented in a common
s.s TIME DrvrsroN swtrcHtNc 247
- t l r
t - l
-1
i-1
I
r_r,
I
a
Lin6
inGrfe6
[-l
r r i - i
|
- l
I
f \ H
t - l l
i
F\
L-l/
t - l
I
$,vltchlng
i
--
bur
_
Tlme
I
Cvclls
I
divirion
I
qqqqql
I
controt
Figure 5.14 Analog time division swirching'
module. Furthermore, in some PAM-PBX systems, analog samples were simultane-
ously transferred in both directions between the interfaces
[7].
Included in Figure 5.14 are two cyclic control stores. The first control store controls
gating of inputs onto the bus one sample at a time. The second control store operates
in synchronism with the first and selects the apPropriate output line for each input
sample. A complete set of
pulses, one from each active input line, is referred to as a
frame. The frame rate is equal to the sample rate of each line. For voice systems the
sampling rate ranges ftom 8 to 12 kHz. The higher sampling rate$ were sometimes
used to simplify the bandlimiting filter and reconstructive filters in the line interfaces.
5.3.2 Dl gl tal Ti meDl vl si on Swl tchl ng
The analog switching matrix described in the preceding section is essentially a space
division switching matrix. By continually changing the connections
for short periods
of time in a cyclic manner, the conflguration of the space division switch is replicated
once for each time slot. This mode of operation is refened to as time multiplexed
swirching. While this mode of operation can be
quite useful for both analog and digital
signals, digital time division multiplexed signals usually require switching between
time slots as well as between
physical lines. Switching between time slots represents
a second dimension of switching and is refened to as time switching.
In the following discussion of digital time division switching it is assumed, unle$s
otherwise stated, that the switching network is interfaced directly to digital time divi-
248 DIGITAL SWITCHING
sion multiplex links. This assumption is generally justified
$ince, even when operating
in an analog environment, the most cost-effective switch designs multiplex groups
of
digital signals into TDM formats before any switching operations take place. Thus
most of the following discussion is concerned with the intemal structures of time di-
vision switching networks and possibly
not with the structure of an entire switching
complex.
The basic requirement of a time division switching network is shown in Figure
5.15. As an example connection, channel 3 of the first TDM link is connected to chan-
nel l7 of the last TDM link. The indicated connection implies that information arriv-
ing in time slot 3 of the first input link is transferred to time slot l7 of the lasr output
link. Since the voice digitization process
inherently implies a four-wire operation, the
return connection is required and realized by transferring information from time slot
17 of the last input link to time slot 3 of the first ouFut link. Thus each connection
requires two transfers of information, each involving translations in both time and
space.
A variety of switching sffuctures are possible
to accomplish the transfers indicated
in Figure 5.15. All of these structures inherently require at lea$t two stages: a space
division switching stage and a time division switching stage. As discussed later, larger
switches use multiple stages of both fypes. Before discussing switching in both dimen-
sions, however, we discuss the characteristics and capabilities of time switching alone.
A Digital lllemory Swltch
Primarily owing to the low cosr of digital memory, time swirching implemenrarions
provide
digital switching functions more economically than space division implemen-
tations. Basically, a time switch operates by writing data into and reading data out of
a single memory. In the process,
the information in selected time slots is interchanged,
as shown in Figure 5.16. when digital signals can be multiplexed inro a single TDM
format, very economical switches can be implemented with time switching alone.
However, practical
limitations of memory speed limit the size of a time switch so that
some amount of space division switching is necessary in large switches. As demon-
strated in later sections, the most economical multistage designs usually perform
as
much switching as possible
in the time stages.
The basic functional operation of a memory switch is shown in Figure 5.17. Indi-
vidual digital message circuits are multiplexed and demultiplexed in a fixed manner
I
I
t
t
I FNAffiH
Figure 5.15 Time and space division swirching.
5.3 TIME DIVISION SWITCHING 249
Figure 5.16 Time slot interchange operation.
to establish a single TDM link for each direction of travel. The multiplexing and de-
multiplexing functions can be considered as part of the swirch itself, or they may be
implemented in remote transmission terminals. In either case, a byte-interleaved
form
of multiplexing is required. The asynchronous transmission hierarchy
(DSz,
DS3,
DS4) uses bit interleaving and therefore requires back-to-back demultiplexing and
multiplexing operations before switching cur be accomplished. In contrast, the
SONET nansmission format described in Chapter 8 can provide byte interleaving spe-
cifically so it can be more directly interfaced to a digital switching system.
The exchange of information between two different time slots is accomplished by
a time slot interchange (TSI)
circuit. In the TSI of Figure 5.17 data words in incoming
time slots are written into sequential locations of the data store memory. Data words
for outgoing time slots, however, are read from addresses obtained from a control
store. As indicated in the associated control store, a full-duplex connectron between
TDM channel i and TDM channelj implies that data store address i is read during out-
going
time slotj and vice versa. The data store memory is accessed twice during each
link time slot. First, some control circuitry
(not
shown) selects the time slot number
MUX TI ME SLOT I NTEHCHANGE
Figure 5.17 MIDVTSI/DEMLIX memory switch.
250 DIGITAL SWITCHING
as a write address. Second, the content of the conffol store for that particular time slot
is selected as a read address.
since a write and a read are required for each channel entering (and leaving) the
TSI memory, the maximum number of channels c that can be supported by the simple
memory switch is
r25
c : 4
(5.
r ?)
where I ?5 is the frame time in microseconds for 8 kHz sampled voice and f" is the
memory cycle time in microseconds.
As a specific example, consider the use of a t 5.2 nsec of memory. Equation 5. 12
indicates that the memory switch can support 4096 channels (2048 full duplex con-
nections) in a strictly nonblocking mode of operation. The complexity of the switch
(assuming
digitization occurs elsewhere) is quite modest: The TSI memory stores one
frame of data organized as 4096 words by 8 bits each. The control store also requires
4096 words, but each word has a length equal to log2(c) (which
is 12 in the example).
Thus the memory functions can be supplied by a096 x 8 and 4096 x lz bit random-
access memories (RAMs).
The addition of a time slot counter and some gating logic
to select addresses and enable new information to be written into the conffol $tore can
be accomplished with a handful of conventional integrated circuits
(ICs).
This switch should be conrrasted to a space division design (Equation
5.3) that re-
quires more than 1.5 million crosspoints for a nonblocking three-stage switch. Al-
though modern IC technology might be capable of
placing
rhar many digiral
crosspoints in a few ICs, they could never be reached because of pin limitations. As
mentioned in Chapter 2, one of the main advantages of digital signals is the ease with
which they can be time division multiplexed. This advantage arises for communica-
tion between integrated circuits as well as for communication between switchine of-
fices.
Time Stages in General
Time switching stages inherently require some form of delay element to provide
the
desired time slot interchanges. Delays are most easily implemented using RAMs that
are written into as data arrive and read from when data are to be transferred out. Ifone
memory location is allocated for each time slot in the TDM frame format, the infor-
mation from each TDM channel can be stored for up to one full frame time without
being overwritten.
There are two basic ways in which the time stage memories can be controlled; writ-
ten sequentially and read randomly or written randomly and read sequentially. Figure
5.18 depicts both modes of operation and indicates how the memories are accessed to
translate information from time slot 3 to time slot 17. Notice that both modes of op-
eration use a cyclic control store that is accessed in synchronism with the time slot
counter.
s.4 TWo-DTMENS|oNAL swrrcHrNc 251
(rl {bl
Figure 5.18 Time switch modules:
(a) sequential writes/random reads;
(b)
random writes/
sequential reads.
The first mode of operation in Figure 5. l8 implies that specific memory locations
are dedicated to respective channels of the incoming TDM link. Data for each incom-
ing time slot are stored in sequential locations within the memory by incrementing a
modulo-c counter with every time slot. As indicated, the data received during time slot
3 are automatically stored in the third location within the memory. On output, infor-
mation retrieved from the control store specifies which address is to be accessed for
that particular time slot. As indicated, the seventeenth word of the control store con-
tains the number 3, implying that the contents of data store address 3 is transferred to
the output link during outgoing time slot 17.
The second mode of operation depicted in Figure 5-18 is exactly the opposite of the
frrst one. ftrcoming data are written into the memory locations as specified by the con-
trol store, but outgoing data are retrieved sequentially under control of an outgoing
time slot counter. As indicated in the example, information received during time slot
3 is written directly into data store address 17, where it is automatically retrieved dur-
ing outgoing TDM channel number 17. Notice that the two modes of time $tage op-
ration depicted in Figure 5.18 are forms of output-associated control and
input-associated control, respectively. In a multiple-stage design example
presented
later, it is convenient to u$e one mode of operation in one time stage and the other
mode of operation in another time stage.
5.4 TWO.DIMENSIONAL SWITCHING
Larger digital switches require switching operations in
both a space dimension and a
time dimension. There are a large variety of network configurations that can be used
to accomplish these requirements. To begin with, consider the simple switching struc-
ture shown in Figure 5.19. This switch consists of only two $tages: a time stage T fol-
Iowed by a space stage S. Thus this $tructure is refered to a time-space
(TS)
switch.
The basic function of the time stage is to delay information in arriving time slots
until the desired output time slot occurs. At that time the delayed information is trans-
DEtd
Stors
'ims
glot
Countdr
252 DIGITAL SWITCHING
f,'lgure 5.19 Time-space (TS)
swirching matrix.
ferred through the space stage to the appropriate output link. In the example shown
the information in incoming time slot 3 of link I is delayed until outgoing time slot
l7 occurs. The return path
requires ttrat information arriving in time slot 17 of link N
be delayed for time slot 3 of the next outgoing frame. Notice that a time $tage may
have to provide delays ranging from one time slot to a full frame.
Associated with the $pace stage is a control store that contains the information
needed to speciff the space stage configuration for each individual time slot of a
frame. This conhol information is accessed cyclically in the same manner as the con-
trol information in the analog
time division switch. For example, during each outgoing
time slot 3, control information is accessed that specifres interstage link number 1 is
connected to output linkN. During other time slots, the space switch is completely re-
configured to support other connections.
As indicated, a convenient means of representing a conffol store is a parallel end-
around-shift register. The width of the shift register is equal to the number of bits re-
quired to specify the entire space switch configuration during a single time slot. The
length of the shift register conforms to the number of time slots in a frame. Naturally,
$ome means of changing the information in the conhol store is needed so that new con-
nections can be established. In actual practice,
the control store$ may be implemented
as RAMs with counters used to generate
addresses in a cyclic fashion (as in the time
stage control stores shown previously).
lmplementation Complexlty of Time Dlvlslon Switchee
In previous sections, alternative space division switching structures were compared in
terms of the total number of crosspoints required to provide a given grade
of service.
In the case of solid-state electronic switching matrices, in general,
and time division
switching, in
particular,
the number of crosspoints alone is a less meaningful measure
of implementation cost. Switching structures that utilize ICs with relatively large
numbers of internal crosspoints are generally
more cost effective than other $uuctures
5.4 TWO.DIMENSIONAL SWITCHING 253
that may have fewer crosspoints but more
packages. Hence a more relevant design
pa-
rameter for solid-state switches would be the total number of IC
packages. If alternate
designs are implemented from a comlnon set of ICs, the number of
packages may
closely reflect the number of crosspoints.
In addition to the number of crosspoints in space division stages, a digital time di-
vision switch uses significant amounts of memory that must be included in an e$timate
of the overall cost.- The memory count includes the time stage memory arrays and the
control stores for both the time stages and the space stages. In large switches, the num-
ber of space stage crosspoints can be reduced at the expense ofincreasing the amount
of memory required. Thus, athoroughanalysis of implementationcomplexityrequires
knowing the relative cost of a crosspoint relative to the cost of a bit of memory. Be-
cause a crosspoint is closely associated with an extemal connection, it costs signifi-
cantly more than a bit of memory. The use of standard medium-scale integration ICs
leads to ons crosspoint costing about the same as I fi) bits of memory.t Use of custom-
or application-specific ICs with integrated memory can change this factor,
particularly
because large-scale IC packages provide access to many more crosspoints
per pin. For
purposes of illustrating various design trade-offs, the following analyses of implemen-
tation complexity continue to consider crosspoint costs to be lfi) times the cost of a
memory bit. Depending on the implementation approach, this factor may not be accu-
rate, but minimizing the cost of a matrix is no longer much of a concern, except in ex-
tremely large switching systems
(e.g.,
in switches exceeding 100,000 voice channels).
The implementation complexity is expressed as follows:
No
ComPlexitY
=N**ffi
=
number of space stage crosspoints
=
number of bits of memory
(s.
l 3)
where Nx
NB
Example 5.2. Determine the implementation complexity of the TS switch shown in
Figure 5.19 where the number of TDM input lines N
=
80. Assume each input line
contains a single DSl signal
(24
channels). Furthermore, assume a one-stage matrix
is used for the space stage.
Solutinn. The number of crosspoints in the space stage is determined as
Nx
=
8d: 6400
*lt
is worth noting that digital memories are inherently implemented with at least two crosspoints
per
bit.
In this case the crosspoiflts are gates used to
provide write and read access to the bits. These crosspoiflts,
howcver, are much less expensive than message crosspoints that are accessed from extemal circuits,
TSee
the first or second edition of this book.
254 DIGITAL SWITCHING
(The
crosspoints on the main diagonal are included since two channels within a single
TDM input may have to be connected to each other.) The total number of memory bits
for the space stage control store is determined as
Nnx
= (number
of linksXnumber of control words)(number of bits
per
control
word)
= (80x24X7)
=
13.440
The number of memory bits in the time stage is determined as the sum of the time slot
interchange and the control store bitsl
Nsr
= (number
of links)(number of channels)(number of bits per channel)
+ (number
of links)(number of control words)(number of bits per
control word)
= (80X24X8)
+ (80)(24)(s)
=
24,96O
Thus the implementation complexity is determined as
Complexity
=
Nx *
W
:
6114equivalent crosspoints
The implementation complexity determined in Example 5.2 is obviously domi-
nated by the number of crosspoints in the space stage. A significantly lower complex-
ity (and generally
lower cost) can be achieved if groups
of input links are combined
into higher level multiplex signals before being switched. The cost of the front-end
multiplexers is relatively small if the individual DS I signals have already been syn-
chronized for switching. In this manner, the complexity of the space stage is decreased
appreciably while the overall complexity of the time stage increases only slightly. (See
the problems at the end of this chapter.) The implementation costs are reduced pro-
portionately,
up to the point that higher speeds dictate the use of a more expensive
technology.
A significant limitation of the TS structure of Figure 5.19 occurs when a connection
has to be made to a specific channel of an outlet as opposed to any channel of an outlet.
A connection to a specific channel is blocked whenever the desired time slot of the
inlet TSM is already in use. For example, if time slot 17 of the first inlet is in use for
some connection from the first inlet to some link other than link N, the connection
from channel 3 of inlet I to channel 17 of inlet Ncannot be made. Because of this limi-
tation, the TS structure is useful only if the outlets represent trunk groups, which im-
plies
any channel of an outlet is suitable.* Applications that require connections to
specific channels require additional stages for adequate blocking probability perform-
ance.
'A
full-duplex connection requires the reverse connection to be established, which adds restrictions to
which outlet channels can lre used. See the problems
at the end of the chapter.
5.4 TWO-DIMENSIoNAL SWITCHING 255
Multlple*Stage Time and Space Switching
As discussed in the
preceding section, an effective means of reducing the cost of a time
division switch is to multiplex as many channels together as practical
and
perform as
much switching in the time stages as possible. Time stage switching is
generally less
expensive than space stage $witching-primarily because digital memory is much
cheaper than digital crosspoints
(AND gates). To repeat, the crosspoints themselves
are not so expensive, it is the cost of accessing and selecting them from external
pins
that makes their use relatively costly.
Naturally, there are
practical limits as to how many channels can be multiplexed
into a common TDM link for time stage switching. When these practical limits are
reached, furtherreductions in the implementation complexity can be achieved only by
using multiple stages. Obviously, some cost savings result when a single space stage
matrix of a TS or ST switch can be replaced by multiple stages.
A
generally more effective approach involves separating the space stages by a time
stage, or, conversely, separating two time stages by a space stage. The next two sec-
tions describe these two basic structures. The first structure, consisting of a time stage
between two space stages, is referred to as a space-time-space
(STS)
switch. The sec-
ond structure is referred to as a time-space-time
(TST)
switch.
5.4.1 $TS Swltching
A functional block diagram of an STS switch is shown in Figure 5.20. Each space
switch is assumed to be a single-stage
(nonblocking)
switch. For very large switches,
it may be desirable to implement the space switches with multiple stages. Establishing
a
path through an STS switch requires finding a time switch anay with an available
write access during the incoming time slot and an available read access during the de-
sired outgoing time slot. When each individual stage
(S, T, S) is nonblocking, the op-
eration is functionally equivalent to the operation of a three-stage space switch. Hence
a probability graph in Figure 5.2 I of an STS switch is identical to the probability graph
of Figure 5.8 for three-stage space switches. Conespondingly, the blocking
prob-
abilitv of an STS switch is
Figure 5.20 Space-time-space
(STS) switching structure.
256 DIGITAL SWITCHING
Figure 5.21 Probability gmph
of STS switch wirh nonblocking stages.
B=( 1
- q' ?) k
( 5. 14)
wher e4' =l
-
p' =l
-
pl \ ( F=
f r / l l t
ft
=
number of center-stage time switch arrays
Assuming the space switches are single-stage Euray$ and that each TDM link has c
message channels, we may determine the implementation complexity of an STS
switch as*
Complexity
=
numtter of space slage crosspoints
* (number
of space stage control bits
+ number of time stage memory bits
+ number of time stage control bits)/100
=2H,{
+
2/cc log, N + tc(8) + ftc log, c
100
(5. 15)
Example 5.3. Determine the implementation complexity of a 2048-channel srs
switch implemented for 16 TDM links with 128 channels on each link. The desired
maximum blocking probability
is 0.002 for channel occupancies of 0.1.
solution. The minimum number of center-stage time switches to provide
the
desired grade of service can be determined from Equation 5.14 as fr
=
7. Using this
value of ft, the number of crosspoints is determined as (2X7Xl6) =
224. The number
of bits of memory can be determined as
(2X7X128)(4)
+ (7X128X8)
+ (7)(lZB)(7) =
*This
derivation assumes output-associated control is used in the first stage and input-associated contol is
ffiar3l"
third stage, A slightly different result occurs if the space stages are controlled in different
2
a
a
I
s.4 rwo-DrMENsroNAL swtrcHrNc 257
20,608. Hence the composite implementation complexity is 430 equivalent
crosspoints.
The value of implementation complexity obtained in Example 5.3 should be com-
pared to the number ofcrosspoints obtained for an equivalent-sized three-stage switch
listed in Table 5.2. The space switch design requires 8l
,920
crosspoints while the STS
design requires only 430 equivalent crosspoints. The dramatic savings comes about as
a result of the voice signals having already been digitized and multiplexed
(for
trans-
mission purposes). When digital switches were first inserted into an analog environ-
ment, the dominant cost of the switch occurred in the line interface. Digital interface
costs are much lower than analog interface costs, particularly on a
per-channel basis.
5.4.2 TST Swltchlng
The second form of multiple-stage time-space switch is shown in Figure 5.22- the
TST switch. Information arriving in a TDM channel of an incoming link is delayed in
the inlet time $tage until an appropriate path through the space $tage is available. At
that time the information is transferred through the space stage to the appropriate out-
let time stage where it is held until the desired outgoing time slot occurs, Assuming
the time stages provide full availability
(i.e.,
all incoming channels can be connected
to all outgoing channels), any space stage time slot can be used to establish a connec-
tion. In a functional sense the space stage is replicated once for every intemal time slot.
This concept is reinforced by the TST probability graph of Figure 5.23.
Figure 5.22 Time-space-time
(TST)
switching sfiucfure.
2
I
a
a
258 DIGITAL SWITCHING
Figure 5.23 Probability graph
of TST swirch with nonblocking stages.
An important feature to notice about a TST switch is that the space stage operate$
in a time-divided fashion, independently of the external TDM links. In fact, the num-
ber of space stage time slots I does not have to coincide with the number of external
TDM time slots c.
If the space stage is nonblocking, blocking in a TST switch occurs only if there is
no intemal $pace stage time slot during which the link from the inlet time stage and
the link to the outlet time stage are both idle. obviously, the blocking probabiliry is
minimized if the number of space stage time slots I is made to be large. In fact, as a
direct analogy of three-stage $pace switches, the TST switch is strictly nonblocking if
I
=
2c
-1.
The general expression of blocking probability for a TST swirch wirh non-
blocking individual stages
(T,
S, T) is
B
=l r
_
q?l t
wher e4l
=l - pr =l * pl a
cr,
=
rime expansion
(//c)
I
=
number of space stage time slots
The implementation complexity of a TST switch can be derived as
(s.16)
Complexity
=N2
+
N/ log, N + 2Nc(8) + 2Nl log, c
(s. 17)
Example 5.4. Determine the implementation complexity of a 2048-channel TST
switch with 16 TDM links and 128 channels per link. Assume the desired maximum
blocking probability
is 0.002 for incoming channel occupancies of 0.1
Solution' Using Equation 5.16, we can determine the number of internal time slots
required for the desired grade
of service as 25. Hence time concentration of 1/g
=
5. I ?
is possible
because of the light loading on the input channels. The implementation
complexity can now be determined from Equation 5.17 as 656.
100
5.4 TWO-DIMENSIONAL SWITCHING 25S
The results obtained in Examples 5.3 and 5.4 indicate that the TST architecture is
more complex than the STS architecture. Notice, however, that the TST switch oper-
ates with time concentration whereas the STS switch operates with space concentra-
tion. As the utilization of the input links increase, less and less concentration is
acceptable. If the input channel loading is high enough, time expansion in the TST
switch and space expansion in the STS switch are required to maintain low blocking
probabilities.
Since time expansion can be achieved at less cost than space expansion,
a TST swirch becomes more cost effective than an STS switch for high channel utili-
zation. The implementation complexities of these two system$ are compared in Figure
5.24 as a function of the input utilization.
As can be seen in Figure 5.24, TST switches have a distinct implementation advan-
tage over STS switches when large amounts of traffic are
present. For small switches,
the implementation complexities favor STS architectures. The choice of a particular
architecture may be more dependent on other factors such as modularity, testability,
and expandability. One consideration that generally favors an STS structure is its rela-
tively simpler control requirements
[8].
For very large switches with heavy traffic
loads, the implementation advantage of a TST switch is dominant. Evidence of this
fact is
provided by the No. 4 ESS, a TST structure introduced in 1976 to service an
excess of 100.000 voice channels.
{
E
l m
6m
ztr
,.,f
SlSrsotprmlonf
STEcomFhxlty
.i
.2
d
."d"'
rsrtlil.rDililofid
.B
A
,z
Udllrdoft ot lnFut drnndr
p
g
{
$
E
I
Figure 5.24 Complexity comparison of STS and TST switching structures for a blocking
probability of 0.002.
260 DrerrAL swtrcHtNc
Figure 5.?5 Time*space-space-space-time (TSSST)
switching sfllrcture.
TSSST Swltches
When the space stage of a TST switch is large enough to
justiff
additional control
complexity, multiple space strages can be used to reduce the total crosspoint count.
Figure 5.25 depicts a TST architecture with a three-stage space switch. Because the
three middle stages are all space stages, thi$ structure is sometimes referred to as a
TSSST switch. The EWSD switch of Siemens
[9]
uses a TSSST structure.
The implementation complexity of a TSSST switch can be determined as*
Nu* *Nu, *Nrrc
(5. 18)
Complexity
=
Nx *
where Nyg
=
number of crosspoints,
=
ZNft + k
(Nln)z
Nsx
=
number of space stage control store bits,
=
2k(Nln)llog2(n)
+ k(Nln)l log2(N/n)
NBr
=
number of bits in time stages,
=
2Nc (8)
ffnrc
=
number of time stage control store bits,
=
21Vl logz(c)
The probability graph
of a TSSST switch is shown in Figure 5.26. Notice that this
diagram is functionally identical to the probability graph
of a five-stage space switch
shown in Figure 5.10. Using the probability graph
of Figure 5.26, we can determine
the blocking probability
of a TSSST switch as
B
=
{
1
_
(qr)tlr
_
0
_
d)ollt
OuttGt
Spacr tlm6
BtNg rtage
l nl ct
tifte $pece
rtrgtc stEaE
100
The assumed connol orientations by stages are output, output, output, input, and input,
(s.
l9)
s.4 rwo-DrMENsroNAL swtrcHrNc 261
P,
=
Pla
pr=pl l odl
c= Uo
r
u-*' '
Figure 5.26 hobability graph of TSSST switch.
wher e4r * I - pt =l - pl a
Qz = r - Pz = I - p l u 9
Example 5.5. Determine the implementation complexity of a 131,072-channel
TSSST switch designed to provide a maximum blocking probability of 0.002 under
channel occupancies of 0.7. Assume the switch services 1024 TDM input links with
128 channels on each link. Also assume that unity time expansion is used on the space
stages.
Solution. The space switch can be designed in a variety of ways depending on how
many links are assigned to each array in the first
(and
third) space stages. A value of
32 is chosen as a convenient binary number near the theoretical optimum
presented
earlier
(N/2)12.
With this value for n, the only unknown in Equation 5.19 is the number
of center-stage arrays ft, which is determined to be 27 for a blocking probability of
0.0015. The complexity is determined from Equation 5.l8 as
Nx
=
(2X1024)(27) + (27)QDz
=82,944
Nnx
=
2(27X32X I 28X5) +
(27)(32)(r28X5)
=
1,65 8,880
Nsr
=
2(10?4X128X8)
=2,097,152
Nnrc
*
2(1024X128)(7)
=
1,835,008
82,944 + 5,591,040/100
=
138,854 equivalent crosspoints
5.4.3
DIGITAL SWITCHING
No.4 ESS Tol l Swl tch
As shown in Fi gtxe 5.27
,
the basic structure of a No. 4 ESS maFix is time-space-time
with four stages in the space switch
(i.e.,
a TSSSST)
tl0,
I ll. The inputs to the matrix
are l20-channel TDM links formed by multiplexing five DS I signals together. Before
these inputs are interfaced to the matrix, they are passed
through a decorrelator to dis-
tribute the channels of incoming TDM funk groups
across multiple TDM links into
the matrix. Decoruelation is used because the No. 4 ESS is a toll switch in which the
incoming TDM links represent trunk groups from other switching machines. In con-
trast to TDM links of an end office switch formed by multiplexing independent sub-
scriber channels, the channels of a TDM trunk
group
are not independently busy. In
fact, the act ofconcentrating multiple, independent sources onto a fiunk group causes
high levels of correlation between the activity on the individual channels. If the chan-
nels of a trunk group were not decorrelated, they would experience much higher
blocking
probabilities
because they would all be vying for the same
paths
through the
matrix. Decorrelation shuffles the trunk groups so the alternate paths
available for any
particular
connection are more likely to be statistically independent. Notice that, be-
sides shuffling channels, the decorrelator provides
space expansion (7
to 8) and time
expansion (l?0
to I28).
The maximum size of the No. 4 ESS uses 128 deconelators with seven 12O-channel
inputs each. Thus the maximum channel capacity is
(128X7X120) =
107,520 chan-
nels. The space stage is a 1024 x 1024 matrix with four possible paths provided
during
each of the 128 space stage time slots. The probabiliry graph of the No. 4 ESS swirch
is shown in Figure 5.28, from which the following blocking probability
is derived:
B
=
[ l
-
(l
-p, Xt -pzXl *pr)] rz8
(5.20)
where p is the extemal channel utilization, p1
=
Q
l8)(1201128)p, and p2 is the blocking
probability
of the four-stage space switch derived in Appendix A as
pz=zplql+ rlplfi + s0p1q1+ szplfi+zlp\q\,+Bfrqr+ p!
(5.21)
Example 5.6. Determine the blocking probability
and number of paths rhat have to
be searched in a No. 4 ESS with inlet channel occupancies of 0.7 and 0.9.
Solution. The occupancies of the interstage linls (pzr)
are determined to be 0.574
and 0.738, respectively. Using Equation 5.21, the corresponding values ofp2 are
determined as 0.737 and 0.934. Equation 5.20 can now be used to determine the
blocking probabilities
as 0.fi)2 and 0.56 for inlet occupancies of 0.7 and 0.9,
respectively.
The probability
that all four space paths
are busy in a parricular time slot is I
-
erezh
(he
base term in Equation 5.20), which evaluates to 0.952 forp
=
0.7 and
R p
- 9
u r a
o O
!1
(}
I
(t
D(
+ E
g .
i 5
l A r A
x {
i A
d L :'-(
.i
o |J.{
E *
A
z
F
FI
vi
P
EO
tl
iE
o
F
5
U
E
E
I
r
o
263
7
264 DIGITAL SWITCHING
Figure 5.28
0.9955 forp
-
0.9. Equation 5.11 now determines the average number of time slots
(four paths per
time slot) that must be tested to find an idle path
as 2l and 9g, respec-
tively.
Example 5.6 demonstrates that low blocking probabilities
can be achieved even
when individual paths
are highly utilized if there are enough alternatives-a feature
that is most practical
with time division switching. This example also demonstrates the
sensitivity of the call procs$sing
time as a function of load in that path
search times
more than quadruple
when the node becomes overloaded with traffic. Reference
[l
l]
reports that the path
search time is only 72vo of the total processing
load at p =
0.7,
but this example implies that it would become more than 50zo of the load atp
=
0.9 if
other aspects of call processing
were linear functions of the traffic volume (a
risky as-
sumption).
5.4.4 Sy$rem 75 Dtgitat PBX
As originally released, the system 75 PBx of AT&T was a midrange
pBX
that can
suppofr 400 stations, 200 data instruments, and 200 trunks
[12].
The switching matrix
consists of two bytewide TDM buses a$ shown in Figure 5.29. Because the buses op-
erate at 2.048 MHz, there are 256 time slots per bus per
frame. Thus the total number
of time slots on the buses is 5 12, which support 256 full-duptex connections.* All ports
of the switch have access to all time slots on both buses. The dual-bus architecture al-
lows the use of slower electronics and provides
redundancy in the case of failure.
As long as there is an idle time slot on either bus, any source can be connected to
any destination. Thus the blocking probability
is zero from rhe standpoint of matching
*Be"uur-
some of the time slots are used for overhead and continuous distribution of various tones, the
maximum traffic capacity is 7200 CCS
[12],
which relates to 200 connections.
+
pt
Probability graph
of No. 4 ESS matrix.
5.5 DIGITAL CROSS-CoNNECT SYSTEMS 265
Switch BUB A
$Yvitch Bu$ B
Figure 5.29 System 75 matrix architecture.
lOss. Blocking CAn occur, hOwever, as a reSUlt ofcongeStiOn: ifmore Connections are
requested than can be supported by the number of time slots. Blocking analyses of this
type require a different form of mathematics referred to as congestion theory. As pre-
sented in Chapter 12, congestion theory deals with the
probability that the offered traf-
fic load exceeds some value. Matching loss, on the other hand, is concerned with
average or busy-hour traffic volumes.
As an example, congestion analyses in Chapter 12 show that if 800 stations are
37.5Vobwy, on average, the
probability that a station requests service when 400 sta-
tions are already busy is lO-s. Thus the system is virtually nonblocking for voice ap-
plications. Because data connections are often full-period connections, the blocking
probability for voice connections may become more significant in applications involv-
ing intensive voiceband data switching. See the
problem set in Chapter 12 for some
examples.
5.5 DIGITAL CROS$.CONNECT SYSTEMS
A DCS is basicatly a digital switching matrix with an operations interface for setting
up relatively static connections between input and output signals or chamels. Instead
of establishing connections in response to signaling information pertaining to call-by-
call connection requests, DCS connections are established in response to network con-
figuration needs. The most basic function of a DCS is to act as an electronic
patch
panel in lieu of a manual cross-connect facility. Manual cros$-connect frames were
typically installed in switching offices as demarcation
points
between transmission fa-
cilities and switching machines and in wire centers as demarcation
points between
feeder cables and distribution facilities. In both cases, the major
purpose of the cross-
connect frame was to allow rearrangement of transmission circuits and access for test-
ing the circuits in both direcfions.
Figure 5.30 shows a typical manual cro$ri-connect environment while Figure 5.31
shows the same function implemented with a DCS. Manual cross connects are imple-
mented with bridging clips and
punch-down wires. Cross connections within the DCS
are established by entering commands at the management console
(either
locally or re-
7
266 DIGITAL SWITCHING
Menutl
(voice
frequencyl
Cros-Connect
Freme
MGIEl l i c Foci l i ti $
Figure 5.30 Manual cross-connecr sysrcm.
motely). As indicated in the two figures, a major feature of a DCS is its elimination
of back-to-back multiplexing functions when cross connecting individual channels
within TDM transmission links. Additional advantages of a DCS system with respect
to manual cross-connect $vstems are:
r. Automatic Record Keeping. Because the cross connects are under'processor
control, circuit connection reports are readily available through the management
interface. In contrast, records in manual systems were inherently error
prone
and
often out of date.
Clnnnrl
Benk
Conlole
Channl Switchcd Treff ic
Ti6 linrs
Frcilitiet
'*l
DCS
lllol
Fortign Exchange
I DLC
Local
(Di gi tBl l
Swltching
Syrtem
I
M1 3
::
Circuit Switched Traff ic
t l -
\ /
\ /
-\
rr
ltr.r\
Alr
iul
I
I
Local
Switching
System
Flgure 5.31 Elecnonic digital cross-connect system.
5.5 DIGITAL CROSS-CONNECT SYSTEMS 267
2. Remote and Rapid Provisioning. Provisioning is the basic procerts of providing
(or
discontinuing) service to a
particular customer. The basic operations
involved are outside-plant cross connections, inside-plant cross connections,
configuration changes in switching system data base, and customer record
updates in business
(billing)
$ystems. Obviously, the more these
processes were
automated, the faster and more accurately they could be performed.
3. AutomatedTest Access. Testing analog circuits at a manual cross-connect frame
involves physically breaking the connection
(by removing bridging clips) and
attaching the test equipment to the side of the circuit to be tested. All manual
operations are eliminated with an electronic
patch panel
by entering commands
at the management console to connect the desired test tone$ and (DSP)
measurement channels to the circuit under test.
Figure 5.31 depicts two type$ of network traffic: circuit-switched traffic and chan-
nel-switched trffic. Circuit-switched traffic represents locally switched fraffic
(DSO
circuits typically) and channel switched-traffic
refers to leased line equivalents of
digital channels. Channel-switched haffic might terminate at another
public network
switching office as in a foreign exchange
(FX) circuit or at a distant cu$tomer
premise
as a tie line of a
private network. In the latter case, more than one DSO channel might
be concatenated together to form a single higher rate channel referred to as a fractional
TI circuit or M x 64-kbps channel. Channel-switched services typically account for
over one-half of the ffansmission bandwidth between U.S. metropolitan offices
[13]'
The process of separating channel-switched services from circuit-switched services is
often referred to as
"grooming."
Figure 5.30 also shows that universal digital loop car-
rier (UDLC) becomes integrated digital loop carrier
(IDLC)
in a digital environment
(Figure
5.31).
(Chapter
I I describes DLC $y$tems in more detail.)
5.5.1 Consolidation and Segregation
Two basically different DCS
grooming functions are depicted in Figures 5.32 and
5.33: consolidation and segregation. When multiple-access lines carrying traffic des-
tined to a coillmon distant node are
partially filled, the
per-channel co$ts of transpon
to the remote node can be reduced by consolidating the traffic. Conversely, when dif-
ferent types oftraffic originate at a single location, it is desirable to allow a single fa-
Frrthlly Filled
Locsl Acc$6r
Totally FillGd
Nstvvork
50t6
mt\*l-l rm*
30e6--{
*E;
o"'F-
_
30!_.+{_l
t0096
Figure5.32 Consolidation.
268 DIGITAL SWITCHING
DDS Nttrvork
Tio Llnal
Fortign Exchenga
Losil Swltching
Figure5.33 Distribution.
cility (e.g.,
a T1 line) to carry all types of traffic and segregate it at the DCS. Examples
of such traffic are circuit-switched channels, DDS channels, tie lines, multidrop data
circuits, FX circuits, or other special services ttrat might be available only at another
office.
5.5.2 DG$ Hierarchy
The cross-connect system depicted in Figure 5.31 terminates DSI signals and inter-
changes DSO signals, which leads to the designation DCS l/0. similarly, a digital
cross-connect system that terminates DS3 signals and rearranges DSl signals within
the DS3s is referred to as a DCS 3/1. If a DCS, such a$ a DCS 3/0, provides
rearrange-
ment of lower level signals, rruch as DSOs, it does not necessarily mean that it also pro-
vides cross-connect services for intermediate-level signals, such as Dsls. cross
connection of DSls requires transparent mapping of the entire l.5zt4 Mbps, which in-
cludes the framing channel-a capability that may nor be provided
when 64-kbps DSO
channels are the
primary
focus. when a DCS provides
multiple levels of cross con-
nects, the intermediate levels are sometimes designated-as in DCS 3/l/0 for termi-
nating DS3s and rearranging DSls and DS0s. In most cases, it is not necessarJ to
provide
cross-connect services at all levels ofthe digital signal hierarchy for all ofthe
terminations. Figure 5.34 depicts a DCS hierarchy that provides rearrangement
of
lower level signals on only subsets of higher level signals. The higher level (e.g.,
Ds3)
signals that are not cross connected to a lower level DCS may be unchannelized high-
speed
(44.736-Mbps)
signals or transit DS3s that terminate on a DCS 3/l or 3/0 at a
Figure 5.34 DCS hierarchy.
5.5 DIGITAL CROSS-CONNECT SYSTEMS 269
distant location. The primary purpose of the DCS 3/3 is to provide network restoration
or protection switching of the DS3 signals and to
possibly provide time-of-day rear-
rangement of trunk group assignments. SONBT transmission links described in Chap-
ter 8 are cross connected at much higher rates producing additional DCS layers.
5"5.3 IntegratedCrose-ConnectEqui pment
Figure 5.31 depicts distinct multiplexing equipment and two distinct and colocated
swirching systems: the DCS and the local digital switch. Although DCS functions
within the
public network are traditionally implemented with separate equipment, pri-
vate networks often utilize equipment that
provides integrated multiplexing and cross-
connect functions. This equipment has evolved from CPE-based Tl multiplexers
because TSI circuits are insignificant hardware additions to Tl multiplexing hard-
ware. Such equipment are variously refened to as intelligent multiplexers
[14],
nodal
processors, or networking Tl multiplexers. Cross-connect functions are also incolpo-
rated into newer DLC systems because, again, the cost of the TSI function is vidually
nil.
Integration of cross-connect functions with higher level digital signals
(such as
a DS3) has not occurred because the manner in which the higher level signals are
multiplexed
(described
in Chapter 7) is not amenable for direct termination on a
switching system. The newer form of
(synchronous) multiplexing, as specified by
the SONET standard described in Chapter 8, allows integration of higher level
digital multiplexers with cross-connect systems for both public and private
equip-
ment.
Although cross-connect functions are basically nothing more than
"pegged"
or
"nailed-up"
circuit-switched connections, the two functions are traditionally imple-
mented separately because of the following differences in the application require-
ments:
l. A DCS needs to be strictly nonblocking at the DSOlevel, which is generally
uneconomical in large
public network switches.
2. Transparent cross-connection of a 1.544-Mbps signal requires transport of the
framing bit that is not possible in typical digital circuit switches because DSI
interface equipment a$sumes the DS I signals are channelized with framing in
the 193rd bit.
3. DCS functionality does not involve
processing of signaling bits so fully
functional circuit switch interfaces have excess costs when u$ed as interfaces to
cross*connect systems.
4. Cross connecting multiple, concatenated DSO channels
(fractional Tl channels)
requires maintaining order in the concatenated data stream. Although the order
can always be maintained by careful mapping of the individual 64-kbps
connections, it i$ a function that is typically not provided in DSO circuit
switching software. Figure 5.35 shows how the order of bytes in the
concatenated data stream can be transposed by ill-chosen 64-kbps connections'
27O DIGITAL SwITcHING
lffimlng
Oflr
Ditr
Outgolng
Ch{nnel ConnGtloni Ero At to 8t
, Az to Bil As to Bl
Figure 5.35 Transposition of data in concatenated time slots.
5.6 DIGITAL SWITCHING IN AN ANALOG ENVIRONMENT
when digital end office switches (or
PBXs) are installed in an analog environment,
the analog interfaces are necessarily unchanged. Although the digital switch may in-
terface with digital subscriber canier or digital fiber feeder system$, these sysrems
merely extend the analog interface point
closer to the subscriber. This section de-
scribes the basic considerations ofusing digital switching in such an analog environ-
ment. Chapter l l describes digiral end office switching with digital subscriber loops
in relation to the integrated services digital network.
5.6.1 Zero-LossSwi tchl ng
As already mentioned, a well-designed digital transmission and switching system
adds no appreciable degradation to the end-to-end quality
ofdigitized voice. In par-
ticular, the derived analog signal coming out of a decoder can be adjusted to the same
level as that presented
to the encoder at the far end. Curiously, zero-loss transmission
presents
some significant problems
when digital switching is used in the analog fians-
mission environment of a class 5 central office.
Analog end office switches are two-wire switches designed to interconnectbidirec-
tional two-wire customer loops. The voice digitization process,
however, inherently
requires separation of the go
and return signal paths involved in a connection. Thus,
a class 5 digital end office or digital PBx must be a four-wire switch. when inserted
into a two-wire analog environment, hybrids are required to separate the two direc-
tions of transmission. As shown in Figure 5.36, hybrids ateach end of ttre internal dig!
tal subnetwork produce
a four-wire circuit with the potential for echoes and singing.
(Amplifiers
are shown in conjunction with the encoders to offset forward path loss in-
herent in the hybrids.) Instability arises as a result of impedance mismatches at the hy-
brids causing unwanted coupling from the receive portion
to the transmit portion of
the four-wire connection. Impedance mismatches occur because of the variability in
the lengths and wire sizes ofthe subscriber loops. In pafticular,
loaded and unloaded
wire pairs
have markedly different impedance characteristics.
The instability problems
are compounded by certain amounts of artificial delay that
are required in a digital time division switch. Although the delay through a digital
switch
(several
hundred microseconds, typically) is basically unnoticeable to a user,
I
underirNbh
\
I
trEnfhybrid
I
\
coulllns
/
5,6 DIGITAL SWITCHING IN AN ANALOG ENVIRONMENT 271
Flgure 5.36 Four-wire digital switch with two-wire analog interfaces.
it represents the equivalent of as much as 30 to 40 miles of wire. This increased delay
has the effect of lowering the oscillation frequencies that might otherwise be outside
the voiceband and effectively be removed by encoder/decoder filters'
As mentioned in Chapter l, hybrids were traditionally used at the interface to the
toll network where the transmission medium is basically four-wire. In these instances,
the instability of the four-wire circuit was controlled by designing a
prescribed amount
of net aftenuation
(by
way of net loss) into the transmission path of shofier toll network
circuits. On the longer circuits, the echoes and singing are eliminated by echo suppres-
$ors or echo cancelers. Class 5 digital switches could prevent instability in the same
manner: by designing a certain amount of loss into the encoding/decoding
process'
While the necessary amount of signal attenuation
(approximately 2-3 dB in each path
[5])
could be tolerated on local connections, the added loss to some toll connections
would be unacceptably large.
From the foregoing discussion it can be seen that the use of selectable attenuation
is one solution to the instability
problems. The necessary loss is insefied into the talk-
ing path
for local connections but not u$ed on long-distance calls, which already have
loss designed into them. A second solution involves matching the impedance at the hy-
brids more closely. Before the advent of DSP nearly complete elimination of the un-
wanted coupling was prohibitively expensive. However, adequate isolation of the
transmission
paths could be accomplished with
just
two different matching networks:
one for loaded loops and one for unloaded loops
[61.
Impedance matching has been
simplified by the use of advanced interface electronics
(DSP) that contain trainable
and automatic impedance matching circuitry. The impedance matching circuits essen-
tially represent short-delay echo cancelers
[17].
Notice that an all-digital network
(with four-wire telephones) avoids instability
problems because there are no two-wire analog lines. Voice is digitized at the tele-
phone
and uses a separate path from the receive signal all the way to the destination
telephone. Thus an endrto-end four-wire circuit completely eliminates echoes and
singing, allowing all connections to operate on a zero*loss basis'
272 DIGITAL SWITCHING
BORSCHT 5.6.2
In Chapter I the basic functional requirements ofthe subscriber loop interface are de-
scribed. These requirements are repeated here with two additional requirements for a
digital switch interface: coding and hybrid. The complete list of interface require-
ments is unaffectionately known as BORSCHT
[18]:
B:
o:
R:
S:
Battery feed
Overvoltage protection
Ringing
Supervision
Coding
Hybrid
Test
As mentioned in chapter l, the high-voltage, low-resistance, and current require-
ments of many of these functions are particularly
burdensome to integrated circuit im-
plementations.
First-generation digital end office switches reduced the termination
costs by using analog switching (concentrators)
to common codecs. The DMS-100
[19]
of Northern Telecom and the No. 5 ESS of AT&T
[20]
use analog concentration
at the periphery. Integrated circuit manufacturers have worked diligently to imple-
ment the BORSCHT functions in what is called a subscriber loop interface circuit
(sl-rc).
Perline sLICs allow implemenration
of perJine BORSCHT functions.
sLICs can be used in PBx applications with a minimum of other extemal compo-
nents. In cenffal office applications, where lightning protection
and test access are
more demanding, sLICs typically need other components for a complete interface.
5.6.3 Conferencing
In an analog network conference calls are established by merely adding individual sig-
nals together using a conference bridge. Iftwo people
talk at once, their speech is su-
perposed. Furthermore,
an active talker can hear if another conferee begins talking.
Naturally, the same technique can be used in a digital switch if the signals are first con-
verted to analog, added, and then converted back to digital.
As described in chapter 3,1t?55 and the A-law
(ITU)
code were designed with the
specific property
of being easily digitally linearizable (EDL).
with this property,
the
addition function can be performed
digitally by first converting all codes to linear for-
mats, adding them, and then converting back to compre$sed formats. To the user, the
operation is identical to the customary analog summation. For a conference involving
N conferees, N separate summations must be performed,
one for each conferee and
containing all signals but his own. For a description ofthe conferencing algorithm in
the system 75 PBX of AT&T, see reference
[12].
For more general
descriptions of
conferencing implementations
in digital switches, see reference
[21].
C;
H:
T:
REFERENCES 273
Another conferencing technique involves monitoring the activity of all conferees
and switching the digital signal of the loudest talker to all others. Although this tech-
nique is functionally different from a customary analog conference bridge, it is advan-
tageous for large conferences because the idle channel noise of the inactive talkers
does not get added into the output of the conference bridge. High-quality conference
circuits also include echo cancelers so higher signal
powers can be
provided.
REFERENCES
I M. R. Aaron,
"Digital
Communications-The Silent
(R)evolution?" IEEE
Communications Magazine, Jan. 1979, pp. l6-26.
C. Clos,
"A
Study of Non-Blocking Switching Networks," Bell System Technical
Joumal, Mar. 1953, pp.4O6-424.
C. Y. Lee,
"Analysis
of Switching Networks," Bell System Technical Journal, Nov.
1955, pp. 1287-1315.
C. Iacobaeus,
"A
Study of Congestion in Link Systems," Erics son Techniques, No. 48,
Stockholm, 1950, pp. 1-70.
A. A. Collins and R. D. Pedersen, Telecommunications,
A Time For Innovation,Merle
Collins Foundation. Dallas. TX, 1973.
M, Schwartz, Telecommunications Netvvork Protacols, Modeling and Analysis'
Addison-Wesley, Reading MA, 1987.
T. E. Browne, D. J. Wadsworth, and R. K. York,
"New
Time Division Switching Units
for No. l0l ESS," Aell SystemTechnical Joumal, Feb. 1969,
pp.443-476,
S. G. Pitroda,
"Telephones
Go Digital,"IEEE Spectrurn,Oct. 1979, pp.51-60.
N. Skaperda,
"EWSD;
Where It Is," 'lieruen's Telecom Repon, Vol. 12, No. 2-3' 1989'
pp. 56-59.
J. H. Huttenhoff. J. Janik. G. D. Johnson, W. R. Schleicher, M. F' Slana, and F' H'
Tendick,
"No.
4 ESS: Peripheral Systems," Bell System Technical Journal, Sept. 1977'
pp. 1029-1042.
A, E. Ritchie and L. S. Tuomenoksa,
"No.
4 ESS: System Objectives and Organization,"
BeIl SystemTechnital Journal, 1977, pp. 1017-1027.
L. A. Baxter, P. R. Berkowitz, C. A. Buzzard, J. J. Horenkamp, and F' E' Wyatt'
"system
75: Communications and Control Architecture," AT&T Tethnical Joumal,
Jan. 1985, pp. 153-173.
R. K. Berman, R, W. Lawrence, and P. C. Whitehead,
"hoviding
and Managing
Channel Switched Services in an Intelligent ISDN Environment," IEEE Clobecom,
1987,
pp.4.6.1-4.6.5.
J. L, Melsa and H. R. Scull,
"The
Application of lntelligent T1 Multiplexers in Hybrid
and ISDN Networks," IEEE Globecom, 1989, pp. 15.3.1-15'3'6.
J. L. Neigh,
"Transmission
Planning for an Evolving Local Switched Digital Network,"
IEEETransactions on Communications, July 1979, pp' 1019-1024.
R. Bunker, F. Scida, and R. McCabe,
"Tnro
loss Considerations in Digital Class 5
Office," IEEE Transactions on Communications, July 1979, pp' 1013-1018.
I
9
l 0
u
t2
l 3
t 4
15
l 6
274
l 7
18
l 9
20
2l
DIGITAL SWITCHING
D, L. Duttweiler,
"A
Twelve Channel Echo Cancellor,', IEEE Transactions on
Communications, May 1978, pp. 647-653.
F. D. Reese,
"Memo
to Management-You Must Appraise How New Technology Fits
Customers," Telephone Engineering and Management, Oct. I, 1975, pp.
I 16-121.
J. B. Terry, D. R. Younge, and R. T. Matsunaga,
"A
Subscriber Line Interface for the
DMs-l00Digitalswitch," IEEENationalrelecommunicationsconference,lgTg,pp.
28. 3. 1-28. 3. 6.
D. L. Camey, J. I. Cochrane, L. J. Gitten, E. M. Pretl" and R. Staehler.
..The
5 ESS
Switching System: Architectural Overview," AZ& T Technical Jorrrual, Aug. 19g5.
R. J. D'Ortenzio,
"Conferencing
Fundamentals for Digital PABX Equipments,', IEEE
International Conference on Communications, 1979, pp. 2.5-2g-2.5-36.
PROBLEMS
5.1 How many four-wire voice circuit connections can be provided
by a bidirec
-
tional PAM switching bus if the minimum achievable pulse
width is 250 nsec?
5.2 The TS switch of Figure 5.19 uses DSI signals on each rDM rink. what is the
implementation
complexity if groups
of five DS I inputs are first multiplexed to
form 16 input links with 120 channels on each link?
5.3 Determine the Lee graph and Jacobaeus blocking probabilities
of the first switch
in Table 5.5
(ft =
6) if two inlets out of 16 become connected to l-erlang sub-
scribers. (Possibly
these lines could be dial-up porrs
to a compurer.) (Hint:Two
inlets are permanently
busy but the remaining inlets continue to be O.l-erlang
porrs.)
5.4 Repeat Problem 5.3 but instead of two inlets being busy assume that two of the
output links of the firsr-stage module have failed.
5.5 How many crosspoints are needed in a l024line, three-stage space swirch if the
input loading is six common-channel signals per line and the maximum accept-
able blocking probability (using
a Lee graph analysis) is 0.005?
(a)
If n
=
16, N/n
=
64
(b) I f n
=32, N/ n=32
(c)
tf n
=
64, N/n
=
16
5.6 what is the (Lee graph)
blocking probability
of the TS switch in Figure 5.l9 for
channel loading of 0.2 erlang?
(a) Assume each TDM input is a 24-channel interoffice trunk group.
(b) Assume the TDM inputs are derived from 24-channel banks with each analog
interface connected to a dedicated 0.2-erlang line.
5.7 Design an sTS switch for 128 primary
TDM signals of the ccITT hierarchy (30
voice channels per
input). Blocking should be less than 0.002 and the loading is
0.2 erlang per
channel. How many time slot interchange modules are needed?
What is the complexiry of the switch?
5.8 Repeat Problem 5.7 for a TST design.
5.9
PROBLEMS 275
Determine the number of crosspoints Nx and the total number of memory bits
Ns required for a TST switch defined as follows; number of lines is 32, single-
$tage space switch, number of channels
per ftame is 30, and time expansion is
2.
What is the blocking probability of the switch in Problem 5.9 if the channel load-
ing is 0.9 erlang per channel?
How many bits of memory are needed in a time slot interchange circuit for a 60-
5.10
5.l l
channel signal with 9 bits
per time slot?
5.12 Derive Equation 5.15 assuming input-associated control in the first stage and
output-associated control in the third stage.
5.13 Derive Equation 5.18 assuming all control orientations are output associated.
DIGITAL MODUL,ATION
AND RADIO SYSTEMS
Chapter 4 discusses various means of encoding digital information
for transmission
over wireline
(or fiber) transmission
links. In the context of this chapter the data en-
coding described in Chapter 4 is referred to as baseband coding. To impress the same
information onto a carrier for radio tratrsmisSiOn,
a baseband signal is commonly used
to modulate the carrier. Modulation shifts the dc-centered specfum of abaseband sig-
nal to the carrier frequency. In contrast to Chapter 4, which emphasizes binary trans-
mission, this chapter emphasizes multilevel digital modulation to achieve high data
rates wirhin the rigidly defined bandwidth
of a radio channel or analog voiceband data
circuit. As described later in this chapter, most multilevel modulation techniques
can
be implemented by directly modulating a carrier with one of the multilevel baseband
signals described in Chapter 4.
Applications for the digital modulation techniques
described in this chapter cover
point-to-point microwave sy$tems, digital cellular systems, and voiceband modems.
Atthough point-to-point microwave systems are no longer used along heavy ffaffic
routes of ttre U.S.
public network, they are still utilized in thin-route applications in
the U.S. and around the world where cost of a fiber right-of-way is
prohibitive' Fur-
thermore, digital
point-to-point microwave is still commonly used in
private networks
(along
with some old analog systems) and in special applications such a$ digital video
transport from a studio to a transmitter or to and from a digital cellular base station.
Digital cellular transmission between base stations and mobile units uses modulation
techniques that are identical to that used in some original digital
point-to-point sys-
TEInS.
When digital
point-to-point radios were first being developed, the FCC established
certain data rates
[1]
that must be achieved by these radios in the United States to en-
sure a minimum utilization efficiency of the radio frequency spectrum.
Basically,
these minimum rates
(shown in Table 6.1) specify approximately the same number of
voice circuits
(at 64 kbps) as were available in the analog FM radios in use at the time
(19?4).
Although individual voice channels in a FDM signal occupy 4 kHz of band-
width, frequency modulation typically expands the signal bandwidth
by a factor of
277
278 DIGITAL MoDUI.ATIoN AND RADIo SYSTEMS
TABLE 6.1 Informatlon Densltleg Fequired by Fcc for common-carrler Microwsve
Channele Using 64 kbps
per
Volce Circuit
Band (GHz)
Channel BW (MHz)
Bit Rate (Mbps)
Density (bps/Hz)
2.1 10-2.130
3.700-4.200
5.9?5-6.425
6.52ffi.87s
1Q. 7- 11. 7
3. 5
20
30
30
40
6. 144
73.7
73.7
73.7
73.7
1, 8
3. 7
2.5
2. 5
1. 8
about 4, depending on the amount of FM deviation applied. Thus the digital systems
were required to compete with an equivalent voice channel bandwidth of about 16
kHz' If the digital radios were forced to compete with analog single-sideband radios
[2,
3] introduced in the early 1980s, they could not have done so without using lower
bit rate voice encoding.
over and above the minimum bit rates established by the FCC, competition among
digital microwave manufacturers and the economics of manimizing the number of
voice circuits per
radio stimulated the development of more advanced digital modu-
lation techniques to achieve even greater
transmission rates. As an example of signal
processing
advances, the 6G150 MB digital radio of NEC carries a 155-Mbps payload
in 30 MHz of bandwidth-an information
density ofjust over 5 bps/Hz. The first part
of this chapter describes basic modulation
techniques and the transmission efficien-
cies they provide.
The latter sections describe radio system design considerations for
point-to-point
microwave relay system$.
lnformation
Density
A useful parameter
for characterizing the bandwidth efficiency of a digital modulation
system is the information density, defined as
s-#
(6. 1)
where R
=
data rate in bits per
second
BW
=
bandwidth of digital signal in hertz
The units of information
density are sometimes referred to loosely as bits perhertz.
However, as defined in Equation 6.1, the units shourd be bits per
second per
hertz or
simply bits per
cycle. Since bits per
second per
hertz conveys the nature of information
density more completely, it is the preferred
unit.
The bandwidth factor in Equation 6.1 can be defined in a variety of ways. In theo-
retical studies the bandwidth of a signal is usually determined as the width of the ideal
filter used to strictly bandlimit the signal (i.e.,
the Nyquist bandwidth). hr practical
systems, where the spectrum can never be strictly bandlimited, the bandwidth is more
6.1 DIGITAL MODULATION 279
difficult to define. In radio systems a channel is usually specified with a certain
mini-
mum signal attenuation at the band edges. In this case, the information density is easily
determined as the bit rate of a signal meeting these requirements divided by the allot-
ted bandwidth.
When no particulal emission specifications are
provided, however, the bandwidth
required by a digital signal must be defined in another more or less arbitrary manner'
using a 3-dB bandwidth is usually inappropriate since the signal spectrum may drop
rather slowly and thereby allow. significant amounts of energy to spill over into adja-
cent channels. Using a greater attenuation level to define the band edges
would be
more appropriate but is still somewhat arbitrary.
A generally more relevant
criterion
for practical systems defines the bandwidth of a signal as the chamel spacing required
to achieve a specified maximum level of interference into identical adjacent
channels.
Yet another definition of bandwidth, commonly
used in Europe, is the 997o
power
bandwidth.
The theoretical maximum information density for binary signaling is 2 bps/tlz for
a two-level line code or 1 bps/flz for a modulated double-sideband
signal' If a four-
level line code is used to achieve 2 bits
per signal interval, the theoretical
information
density is 4
bps4lz for the line code, or ZbpslHz for a double-sideband carrier signal.
The information densifl of amplitude-modulated
signals can be doubled by using single-
sideband transmission
to effectively achieve the same efficiency as the line codes
(baseband signals).
As a
practical example, a representative
data rate on a dial-up analog telephone line
is 28,800 bps
(usually with l2 bits per signal interval and a 2400-Hz signaling rate).
Since the usable bandwidth of a telephone channel is approximately
3 kHz, a typical
information density of a dial-up line is 9.6 bps/H2..
As a starting point for digital microwave radios, Table 6.1 includes the minimum
bit rates and corresponding
information densities required in common*carrier
bands
of the United States. Competition
and compatibility
with the digital hierarchy led to
realizations of even
greater transmission
efficiencies.
6.1 DIGITAL MODULATION
Digital transmission on a radio link differs from wireline transmission
in two impor-
tant regards. First, the digital information
must modulate a carrier in some manner to
produce the radio frequency
(RF)
signal. In many cases the modulation
proce$s can
be viewed as a special form of amplitude
modulation by a nonreturn-to-zero
(NRZ)
line code signal. Thus the line code repre$ents a baseband signal that amplitude
modu-
lates the carrier in the transmitter and is reproduced by demodulation in the receiver.
Representing the modulation
process in this manner has the advantage
that the RF
spectra can be determined by merely translating the line code
(baseband) spectra to the
$elected carrier frequencY.
*Slightly
higher rates are sometimes achieved with V.34 modems. Rates approaching 64 kbps are also
possible in special
(V.90) applications described in Chapter
1 I .
280 DtctrAL MoDULATtoN AND RADIo sysrEMS
second, a radio link differs from wireline tansmission due to the necessity of
strictly bandlimiting the transmitted signals to prevent
interference into other chan-
nels. Although wireline transmission links automatically filter the line signal to some
extent, explicit filter requirements
sometime$ occur only in the receivers to reject as
much noise as possible.*
Since radio links are bandlimitedin
the hansmitter and zpise
filtered
in the receiver, the end-to-end filter function must be partitioned
between the
two ends. Figure 6. I shows a block diagram of a radio link showing representative
baseband and RF waveforms
along with corresponding frequency spectrums. For
modulation, Figure 6. I shows multiplication
of the carrier by the baseband waveform.
Modulation in a digital radio must ultimately be designed in conjunction with the
firlter functions. For ease of description, however, modulation
techniques are consid-
ered first. Later on, the filtering requirements
of each type of modulation are dis-
cussed. Figure 6.1 also shows the baseband encoding proce$s
to be independent ofthe
modulation process.
For most of the modulation techniques described in this chapter
this view is appropriate. In the case of trellis-coded modulation (TCM)
and carrierless
amplitude and phase (cAP)
modulation, described later, encoding is incorporated into
the modulation process.
6.1.1 Ampl i tudeModul ai l on
Historically, the simplest form of modulation
to generate
and detect is amplitude
modulation (AM).
A conceptual illustration of amplitude modulation is shown in Fig-
ure 6.2. The mathematical definition is
x(t)=[l +am,(r)]cosro"t (6.2)
where a
=
modulation index (0
< a S l)
mn$)
=n-level,
symmetric
NRZ baseband signal normalized to maximum
amplitude of I
0[
=
is the radian carrier frequency,
=
21;1
Amplitude modulation is an example of a class of special modulation techniques
referred to as
"linear
modulation."
Linear modulation implies that the spectrum of the
modulated
signal is obtained by ffanslating the baseband spe{trum to the selected car-
rier frequency band. As shown in Figure 6.2, amplitude modulation by a two-level
digital baseband signal essentially translates the sin(x/x baseband specrrum up to the
carier frequencyf,. other linear modulation techniques are double-sideband
modu-
lation, single-sideband modulation,
and vestigial-sideband
modulation.
Inspection of Equation
6.2 or Figure 6.2 indicates thar if lfi)7o modulation is used
(a =
I
),
no carier is produced
for a logic 0. For obvious reasons, this form of amplitude
modulation is often referred
to as on-off keying,or amplitude shifi keyrng (ASK).
As
'Adherence
to EMI compatibility specifications defined by the FCC (Paxt
15) often requires some amount
of transmit filtering in on-premises wireline transmission systems.
?i
6
o
o
{)
bI)
6
E
,t4
o
(h
\a
P
HO
, f
!
E
E
B
,6
E
E
E
E
F
F
c
CI
-g
t
E
t
!
j
j
j
4
f
ct
f
o
r E
FB
281
282 DIGITAL MODULATION AND RADIO SYSTEMS
] -.t
Figure 6.2 Digital binary amplitude modulation.
shown in Figure 6.3, on-off keying can be obtained by direct multiplication
of a car-
rier with the two-level (unipolar)
line code described in Chapter 4.
Amplitude-modulated
signals are usuarly demodulated with a simple envelope de-
tector. The cost effectiveness
of this detector is the basic reason that commercial ana-
log broadcasting uses amplitude modulation. unfortunately,
the eruor performance
of
digital amplitude modulation in general,
and envelope detection in particular,
is infe-
rior to other forms of digital modulation and detection. For this reason, amplitude
modulation is used only where the cost of the receiver is a significant consideration.
Digital microwave links and digital cellular systems use other forms of modulation
and demodulation to minimize
the enor rate for a given
signal-to-noise
ratio.
Conventional
amplitude modulation provides
suboptimum
error performance
for
two basic reasons. First, if a < r, a discrete (informationless)
spectral line occurs at
the carrier ftequency. Although the existence of this spectral line simplifies canier re-
covery' it increases
the transmitter power
without aiding discrimination between in-
formation signals.
with 1007o modulation (on-off
keying) no line specfra are produced,
but the sys-
tem is still inefficient
in its use of transmifted power.
As discussed in chapter 4 for
two-level line coding, the maximum use of transmitted power is achieved when one
signal is the negative of the other. Thus, a second deficiency of amplitude modulation
arises because a 0 signal is not the exact negative ofa I signal. To achieve optimum
performance,
a symmetric twolevel
baseband signal should directly modulate (mul-
tiply) the carier. As shown in Figure 6.4, this form of modulation produces
two iden-
*kl
=
[ t
*an, (rl l
coro. r
o
frF+
6.1 DtctrAL MoDULATIoN
283
+Sffiffiffi
#'q#q#
#t+t#
Figure 6.4 Phase-reversal keying.
tical signals except for a 180"
phase reversal. Hence this modulation is sometimes
re-
ferred to as
phase reversal keying
(PRK), or more often, two-level
phase shift keying
(2-PSK), f-Iotice that PRK cannot be detected by envelope detection'
Instead, a PRK
signal must be detected by comparing it to a coherent carier reference.
Coherent detection involves comparing the incoming signal to a local carrier syn-
chronized
in phase to that used at the transmitter.
With conventional
amplitude
modu-
lation there is no information
in the phase. With PRK signaling, however, all of the
information is in the
phase. The use of a coherent reference
for PRK signals allows
optimum
(antipodat) elror rate performance. The basic equation defining a PRK signal is
x(t1= r77r111cos oct
(6'3)
where rn2(t) is the binary digital baseband
signal
lm2$)
=
+1 for a I and m2(t)
=
-l
for
a 01. The demodulation
equation is
y(t)
=.r(t)[2
cos oct]
=
lmr(t)
cos octl
[2
cos ro"tl
*
m?(t) + m2(t) cos 2ro.t
(6.4)
where the double-frequency
term is removed with a low-pass filter'
As indicated in Figure 6.4 and Equation 6.4, the coherent demodulation
proce$$
produces a symmetric two-level signal that can be
processed further to detect the data-
-Since
the demodulated baseband signal is, in essence, equivalent to a line code, it must
include timing considerations
as discussed in Chapter 4. [n particular, there must be
enough signal transitions
in the baseband
signal to allow recovery of a sample
clock. However, the baseband signal need not
preclude dc wander since
dc levels
284 DIGITAL MoDULATIoN AND HADIo SYSTEMS
o r r 0 0 t 0 t t 0 0 l
o t t 0 0 l
Phsre
dctoclol
Low
pssg
fllur
Denrodulstol
Figure 6.5 Binary frequency shift keying.
are translated to the carrier frequency
and passed
adequately by the double-sideband
system.-
6.1.2 Frequency
$hift Keytng
In addition to an inefficient
use of signal power,
conventional
amplitude-modulated
signals (not
PRK) have one other undesirable
characteristic. By definition, an ampli-
tude-modulated
signal uses multiple signal levels, which implies AM is quite
vulner-
able to si8nal saturation that narrows
the distance between amplitude levels and
produces
spectrum spreading. A common source of saturation in a radio syslem occurs
in the output power
amplifier of the transmitter. In most cases output amplifiers
are
operated at less than maximum power
to eliminate saturation and other nonlinearities,
so they can accommodate
amplitude-modulated
signals
[4].
_ _Angle-modulated
systems FM or
pM
u.le constant-amplitude
signals not adversely
affected by signal saturation. Hence FM and
pM
can be transmittea at higher powei
levels than AM systems. The ability to use saturating power
amplifiers is one of the
rea$ons why FM was originally chosen for analog microwave radios. This section dis-
cusses digital frequency
modulation, commonly referred to as frequency shift keying
(FSK).
The next section discusses digital phase
modulation,
commonly refened to as
phase
shift keying (PSK).
Both systems provide
a consrant-amplitude
signal. systems
using constant-amplitude
carriers are also referred to as constant-envelJpe
sysrems.
The general
expression for an n-ary FSK signal is
.(,):*,[[*..9J']
(6.5)
'A
single-sideband system does not pass dc energy. Thus if single-sideband
modulation rs used, the
baseband signal must exclude.dc energy from its spectrum. Some double-sideband systems might also
require the elimination of baseband dc energy so that a carrier tone can be inserted intl the center of tlre
passband
without affecting the signal.
trffi#hirfut\ddill
Eq#
Modulator
6,1 DIGITAL MODULATION
285
where ro"
=
radian center frequency
ln,
=
n-level NRZ digital baseband signal
Aro
=
radian difference
frequency between signals
A typical binary FSK signal is shown in Figure 6.5 along with a simple
(but
low-
performance) means of implementing
the modulator and demodulator. The modulator
is a voltage-controlled
oscillator
(VCO)
that is biased to
produce the center frequency
when no modulation is applied. The amplitude a of the symmetric two-level
baseband
signal
produces a frequency deviation of +Arr/2 for a I and
-L{dy?
for a 0.
The demodulator
is implemented
as a
phase-locked loop: a VCO, a
phase detector,
and a loop filter. The phase detector
measures the difference in
phase between
the FSK
signal and the VCO output.
A positive voltage is
produced when the receive signal
leads the VCO, and a negative voltage is produced otherwise. After being filtered to
minimize the effects of noise, the
phase detector output drives the VCO in such a way
as to reduce the phase difference. Ideally, the input control voltage of the
demodulator
VCO will be identical to the input of the modulator
VCO' The loop filter, however,
necessarily slows the demodulator
response to minimize the effects of noise.
Frequency shift keying
generally provides poorer error
performance than PSK,
par-
ticularly for multilevel signaling
in a confined bandwidth.
Older asynchronous
mo-
dems for data transmission
over the analog telephone network used FSK modulation'
Newer systems do not use FSK because the simpler implementation compared
to bet-
ter
performing techniques is no longer significant.
Mlnimum Shift KeYing
One particular form offrequency shift keying that has received considerable
attention
in raaio systems
is minimum shift keying
(MSK)
[51.
Basically, MSK is binary FSK
with the two signaling
frequencies selected so that exactly 180o difference
in phase
shift exists between the two frequencies
in one signal interval' In this manner MSK
produces a maximum
phase differential at the end of an interval using a minimum dif-
i"r"n"e in signaling frequencies.
Furthermore, an MSK signal maintains
continuous
phase at signaling
transitions. For this reason MSK belongs to a class of FSK signals
iefened to as continuous-phase
frequency shift keying
(CPFSK). Figure 6'6 depicts a
representative
MSK waveform. Notice that there is exactly one-half cycle difference
between a I signal and a 0 signal
(?
cycles versus 1'5 cycles).
A mathematical expression
for MSK signaling can be derived from Equation 6.5,
where nn(t)
=
mz(t)
is a symmetric binary NRZ signal and ACO
=
n/7. The result for
any particular signal interval is
fo-
1,6tT
h*2l t
t/r- shnrl l no rrtr
Figure 6.6 Minimum shift keYing.
286 DIGITAL MoDULATtoN AND RADto sysTEMS
f ( T u )
l cosl ' cr
* E* qo)
x( r ) ={
\
l ( T E t . \
f ' o' [ *' -
a+qo)
(logic
1)
(logic
0)
(6.6)
(6.7)
where 1/7is the signaling rate and
S6
is the phase
at the begiruring
of the signal interval
(+n
in Figure 6.6).
The main attraction of MSK is its comparatively compact spectrum. Furthermore,
with an appropriate means of detection, MSK can provide
optimum error performance
in terms of the energy-per-bit-to-noise-density
ratio (E/No).
The expression for the
power
spectral density of the MSK signal defined in Equation 6.6 is
s(rrr): t6rt7[4"==f
[n,
-Ot' )
where
s
=
l(D
-
to.lr. The frequency
spectrum of an MSK signal is plotted
in Figure
6'7, where it is compared to the frequency
specrrum of
pRlt(2-psx;
signating wittr
the same data rate. Notice that the MSK spectrum is more compact and has its first
spectral null at 3/4Iinstead of l/Ifor
pRK.*
Minimum shift keying is actually
just
one example of a general
class of continu-
ous-phase
modulation (cPM)
techniques that maintain a constant amplitude, narrow
power
spectrum,
and good
enor performance.
For a good
overview
and bibliography
of this class of modulation
schemes see reference
t6l.
cpM schemes have not been
used in point-to-point
microwave
applications because they do not provide
high in-
formation densities.
Many versions of CPM have been used in sat"llit" applications
where nonlinear transponder
amplifiers preclude
the use of modulation
iechniques
with multiple amplitude levels
[7].
Gausslan MSK
Like MSK, Gaussian minimum shift keying (GMSK)
produces
a constant-amplitude
and continuous-phase
RF signal. GMSK differs from MSK through the use of a
Gaussian baseband pulse
shape in prace
of a square pulse
shape for MSK. Because the
Gaussian pulse
rises and decays asymptotically
with respect
to a zero response level,
it has a much more constrained bandwidth.
Although it is conceivable
that a GMSK
signal could be generated
by filtering
the modulated signal, a rypical implementation
[8]
utilizes baseband filtering, as shown in Figure 6.8, where it is contrasted to unfil-
tered MSK' The MSK signal is generated
by direct FSK modulation of a carrier with
a baseband signal that is scaled in amplitude to produce
a modulation index or 0.5. A
modulation index of this
value produces
the difference of l80o of phase
shift for the
two data values. one complication
of the bandwidthJimiting
filter of GMSK is the
*MSK
i. achrally more closely related to 4-PSK and therefore rs compaxed to it in a later section of this
chapter.
g
E
E
a
I
I
&
6.1 DIGITAL MODULATION
287
Figure 6.7 Power spectra of unfiltered MSK and 2-PSK signals'
Unfiltered MSK Waveform
NRZ Baseband Waveform
(a)
(b)
Figure 6.8 Comparison of (a) direct modulation
MSK and
(b)
GMSK'
Sigmling r.to
=
D|ta rute
=
l/T
fttctioh of outd+and Pdvir
t
,f,-
ol out*o{-hnd
ponnr
IMSKI
Fruqrrency
(Hrl
Modulation Index
*
0.5
Gaussian Pulse ResPonse Filter
Modul ati on
Index
=
0.5
A
288 DtctrAL MoDULATtoN AND RADto sysrEMS
creation of intersymbol interference
similar to partial-response
sy$tems discussed
later in this chapter. In essence, GMSK trades a small amount of adjacent channel in-
terference for a small amount of intersymbol interference.
GMSK modulation is util-
ized in GSM digital cellular
[9]
discussed in Chapter
g
and in cellular digital packet
data (CDPD)
applications.
6.1.3 Phase
$hi ft Keyi ng
The second category of angle-modulated,
constant-envelope
signals is referred to as
phase
shift keying (PSK).
Actually, one form ofpsK has already been discussed as
phase
reversal keying (PRK),
more corlmonly referred to as 2-psK, indicating that
each signal interval uses one oftwo phases
that are lg0" apart to encode binary data.
Multiple-phase shift keying is also possible.
Four-psK (also
called
epsK)
and
g-psK
are the most common examples of multiple-phase
pSK.
Phase shift keying (which
includes 4-eAM described next) is the mosr popular
modulation technique for intermediate information
density high-performance
applica-
tions. The popularity
is primarily
due to its constant envelope,lnsensitivity
to level
variations, and good
error performance.
Both z-pSK and 4-psK provide
theoretical
optimum error performance
in terms of a signal-to-noise ratio (E/
d.
A general
expression for n-ary phase
shift keying is provided
in Equation 6.g. This
expression
assumes lfi)7a modulation is employed. That is, the phase
shift from one
interval to the next can range anywhere from
-190"
to +1g0.. It is po$sible
to devise
PSK systems with lower modulation indices that allow only transitions to neighboring
phases:
"tt
=cosfco'.O9O)
(6.8)
where Af
=Znln
is the separation between a-djacent signal phases
and mn$) is a sym-
metric n-level NRZ baseband signal with levels
+1,
t3, . . . .
Examples of typical 2-PSK and 4-psK waveforms are shown in Figure 6.g. Ttre
signaling rate for the 4-PSK system is shown to be exactly one-half the z-psr signal_
ing rate so that equal data rates are provided.
The same figure also shows correspond-
ing phasor
diagrams
of the signaling phases
of a cosine wave as defined in Equation
6'8. other phase
orientations
are possible.
The particurar phases
$hown, however, are
convenient
for later discussions
relating 4-psK systems to other types of digital modu-
lation.
Quadrature Signal Bepresentailone
Despite the somewhat
exotic sounding name, quadrature
signal representations
are a
very convenient and powerful
means of describing
psK
signals and many other digi-
tally modulated signals.
Quadrature
signal representations involve expressing
an ar-
bitrary phase
sinusoidal waveform as a linear combination
of a cosine wave and a sine
6.1 DIGITAL MODULATION
289
Ph$e disgrim
h)
Timrwrwform
Fharn dirgrem
Time wwgform
ft)
Figure 6.9 Phase shift keying;
(a)
2-PSK
(b)
4-PSK.
wave with zero starting phases. The derivation
of this representation is
provided by
the trigonometric
identity :
cos(to"t +
0)
=
cos
Q
cos ro./
-
sin
$
sin ro"t
(6.e)
Notice that cos
Q
and sin
Q
are constants over a signaling interval and hence repre-
sent coefficients
for expressing cos(ro"t
+
$)
as a linear combination
of the signals cos
O"t and sin co.f. Since cos rrr"t and sin or./ are 90" out of
phase with respect to each
other, they are ofihogonal
in a phasor diagram and hence are said to be
"in
quadra-
ture.tt
In essence, co$ Oct and sin tDcf represent basis vectors in a two-dimensional
phasor
diagram. The cosine signal is usually referred to as the in-phase or l signal, and the
sine signal is referred to as the out-of-phase
or
B
signal' Table 6.2
provides an example
TABLE 6.2 Quadrature
Slgnal Coefllclente
lor 4'PSK Modulatlon
Quadrature Coefficients
Data Values cos ocf
sin ool
Composite Signal
01
00
1 0
1 1
0.7Q7
-0.707
4.707
0.707
4,707
-0.707
0.707
0.707
cos(ot f+ n/4)
cos(ru"f + 3r/4)
cos(<o"t- 3r/4)
co$(0)cl- ty'4)
290 DGIAL MoDULATtoN AND HADto sysrEMs
TABLE 6.3 Ouadrature Signal Coefficiente for
g-pSK
Modutatlon
Quadrature Coefficients
Data Values
cos ocf
sin o.f Composite Signal
0 1 1
010
000
001
101
100
1 1 0
1 1 1
0.924
0.383
-0.383
-{.924
-0.924
-{.383
0.383
0.924
-{,383
4.924
-{.924
-o,383
0.383
0.924
0.924
0.383
cos(cu"f + dB)
cos(ro"f + 3r/8)
cos(o"t + 5nl8)
cos(ro"f + 7rl8)
cos(ohf
*
7rrle)
cos(ocf- 5nl8)
cos(ro.f- 3nl8)
cos(ro.f
-
dB)
of quadrature
signal representations
for the 4-psK signals presented
in Figure 6.9. Ta-
ble 6.3 provides
a corresponding
specification for an
g-psK
system using signal
phases
as provided
in the phasor
diagram ofFigure 6.10. The phasor
diagram assumes
counterclockwise
rotation, and hence the sine function lags the cosine function by 90o.
Most of the rest of this chapter relies heavily on quadrature
signal representations
to describe various modulation concepts, analyses, and implementationr.
Modulator Implementatlone
A variety of techniques
are possible
for implementing
psK
modulators.
As mentioned
when discussing PRK modulation, a 2-psK modulator can be implemented
by merely
inverting the carier (multiplying
by
-l)
for a logic 0 and by nof inverting for a logic
l. some of the basic techniques
used for generating
multiple-phase
psK
signals are
the followins;
dn 0r.t
Figure 6.10 Phasor diagram of B-pSK signal.
6.1 DIGITAL MODULATION 291
L synthesis of the desired waveforms using digital signal
processing at suitably
low carrier frequencies
(as in voiceband
modems).
2. Generating multiple
phases of a single carrier and selecting between the
phases
depending on the data values.
3. Using controlled delays selected through a switching arrangement to
provide
the desired
phase shifts. Delays are often used to
generate the separate signals
in method 2.
& Generating the PSK signals as a linear combination of
quadrature signals.
None of the foregoing techniques represent a direct implementation
of multiphase
PSK modulation as defined in Equation 6.8. To implement Equation 6.8 directly, a de-
vice is needed that
produces carrier
phase shifts in direct proportion to the levels in
the baseband signal mn(t). Except for the special case of Z-PSK where a multiplier can
be used to
produce
+180o
phase shifts, such devices do not exist'
Direct modulation by baseband signals can
produce PSK signals if a
quadrature
signal implementation
is used, as indicated in method 4' Actually, two multilevel
baseband
signals need to be established:
one for the in-phase
(I)
signal and one for the
out-of-phase
(0)
signal. These baseband signals are referred to as r4(t) and tt?q(t) for
the 1 and
p
signals, respectively. The levels chosen for the two baseband signals cor-
respond to the coefficients
needed to repre$ent a PSK signal as a linear combination
of the / and
Q
signals. For example,
Figure 6.1 1 shows how an 8-PSK signal, defined
in Table 6.3, can be generated by adding two amplitude-modulated
quadrature signals.
A block diagram of the corresponding
implementation is provided in Figure
6. I 2. This
particular form of modulator
is chosen, not so much as a recommendation
for actual
implementation, but because it demonstrates
important modulation concepts and is
useful in analyzing the spectrum requirements
of PSK signaling'
Dem od u I ato r I rr p le m e ntati o n
Owing to the constant envelope,* all PSK systems must be detected with the aid of a
local reference. For Z-PSK systems, the ideal reference is coherent with one of the two
possible phases. When this reference is multiplied
(mixed) with an in-phase signal, a
maximum
positive output is obtained.
When multiplied by the opposite
phase, a maxi-
mum negative output is obtained. In this manner, 2-PSK
provides antipodal
perform-
ance when a local coherent reference is established.
The demodulation
process for
2-PSK is presented in Figure 6.4 and Equation 6.4, where the mixing and filtering
process effectively
recovers the baseband signal ruz(t).
The combination of a mixer and a low-pass filter shown in Figure 6.4 is
generally
referred to as a
phase detector. The
phase detection
property is represented mathemati-
cally as
*Discussions
to this
point have not considered the effects of filtering on constant-envelope
signals. A
heavily filtered PSK signal does not have a constant envelope. However, as long as filtering occurs after
channel nonlinearities
(e,g., power amplifiers),
the most harmful effect of a nonlinearity, spectrum
sprcading, is avoided.
29? DIGITAL MODULATION AND RADIO SY$TEMS
tr(ll
'
frr
.
cB @.r
rq {rl
=
ng
'
ri n o.r
t kl
=
r f ( r l +
r O i r l
i l t = f i Ql
ilO
r
ln9t
Figure 6.11 Generation of 8-pSK
modulated signals.
0r
=
FhrF ol Jrh Intflil
signals by superposition
of quadrature-amplitude-
#r;
=
1o*Ouss{cos(ro"r
+
0)
(2)
cos ocr}
=
lowpass{cos
S
+ cos
Q
cos 2ro"t
-
sin 2qrl
= COs 0
(6. 10)
where lowpass{.} is a low-pass filter function designed to remove twice carrier
terms.
When detecting 2-PSK modulation,
a single-phase detector indicates whether the
received phase
is closer to 0o or to 180". The desired information is directly available
as the polarity
of the phase
detector output cos
Q.
In multiphase systems, however, the
information provided
by a single-phase
detector is inadequate for two reasons. First,
Ampliturlc
moduhtid
Id{ml
Oonnrm erryrlopr
PSK$nd
$n adc
t
Amplltudt
modrlrted
0SCul
Figure 6.12 Generalized PSK modulator structure.
6.1 DIGITAL MODULATION
293
a measure of cos
$
provides no information as to whether
Q
is positive or negative' Sec-
ond, the output of the
phase detector is proportional to the signal amplitude as well as
to cos
$.
Hence the magnitude of the
phase detector output is meaningless unless ref-
erenced to the signal amplitude.
Both of the aforementioned
problems are overcome if a second mixer and filter
measure the
phase with respect to a different reference. As expected, the best perform-
ance is obtained when the second reference is orthogonal to the first. Ify{r) is the out-
put ofthe first
phase detector defined in Equation 6. I 0, then the second
phase detector
output
yQ(r) becomes
lB$)
=
lowpass{cos(o"r +
0)
(2)
sin rrl.rl
= _sin
0
( 6. 11)
The second
phase
detector not only resolves the
positive/negative
phase ambiguity
but also eliminates the need to establish an amplitude reference. All decisions can be
based on the
polarity of a phase detector output and not on the magnitude. As a first
example, consider detection of the 4-PSK signals defined in Table 6.2' Notice that the
first data bit in a pair is a 0 when the
phase angle is
positive
(tt/4
or 3nl4) and a 1 oth-
erwise. Hence the first data bit is completely specified by the polarity of sin
Q:
the out-
put
of the second
phase detector
yA(f). Similarly, the second data bit is a I when the
phase is it/4, indicating that the
polarity ofyr(fl
provides all information necessary
to detect the second bit. The basic implementation
of a 4-PSK
(QPSK)
demodula-
tor/detector is shown in Figure 6.13. AIso shown is a 4-PSK modulator to emphasize
the relationships between the modulator and demodulator.
A 4-PSK system is pre-
$ented specifically because it is a
popular system and because it is a useful foundation
for describing other modulation techniques'
An important concept to notice in *re 4-PSK system shown in Figure 6.13 is that
there are, in e$$ence, two separate data streams. The modulator divides the incoming
bit stream so that bits are sent altemately to the in-phase modulator l and the out-of-
phase modulator
O.
These same bits appear at the output of the respective
phase de-
tectors in the demodulator,
where they are interleaved back into a serial bit $tream. In
this manner two essentially independent
binary PSK channels are established
using
the orthogonal carriers cos r.rlct and sin ocf. The channels are usually referred to as the
I and
Schannels,
respectively. This technique of establishing two channels
within an
existing bandwidth
is sometimes referred to as quadrature multiplexing'
As long as the carriers in the modulator and the references in the demodulator are
truly orthogonal
(coherence
is maintained
for both channels
in the receiver), the 1 and
B
channels do not interfere with each other. Any amount of misalignment
in these re-
lationships causes cros$talk between the two
quadrature channels. Crosstalk also
arises if there is unequalized
phase distortion in the transmission
channel.
At first thought it might seem that quadrature multiplexing increases the capacity
of a
given bandwidth by a factor of 2. It must be remembered,
however, that the binary
PSK signal on each quadrature channel
is a double-sideband signal. Hence the band-
width of the channel, without
quadrature multiplexing,
is only 507o utilized in com-
?,94 DIGITAL MODULATIoN AND RADIo SYSTEMS
parison
to a single-sideband sy$tem. when quadrature
channels are used, a single-
sideband operation is no longer possible
since the sideband separation process
de-
shoys the orthogonality of the two signals. In essence, quadrature
multiplexing
only
recovers the loss of capacity incurred by the double-sideband
specfrum. In fact, $ome
single-sideband
modulators
[10]
possess
a remarkable resemblance
to the
epsK
modulator shown in Figure 6.13.
Demodulation
and detection
of higher level
psK
systems are complicated by the
fact that the use of only two references does not provide
a simple means of detecting
all data bits. There are two basic ways in which simple decisions (positive
versus nega*
tive) can be established to detect all data. One method is to establish more references
in the receiver and measure the phase
of the received signal with respect to the acldi-
tional references.
The second
method is to use only two references
and related phase
detectors and generate
all additional measurements as linear combinations of the first two.
D|tr Input
t 0 1
4-PSK $lgnrl
0 1 0
4-PISK Signal
Dt$ iltput
r 0 1 1
0 t 0
Q
Glunml
(h)
Figure 6.13 Four-PSK modulator-demodulator
structure: (a)
modulator; (D)
demodulator-
detector.
lflf
Frtqudrcy
HrcG
Q
Chrnnd
6.1 DIGITAL MODULATION
295
Figure 6.14 Receiver references for 8-PSK detection.
As an example of the first method, consider the 8-PSK system defined in Table 6.3
and shown in Figure 6. 10. Optimum detection for this system can be obtained if two
additional
phase detector outputs are
provided with respect to references at +ru/4 and
-tc/4.
The two new references are designated A and B as shown in Figure 6.14'
The ouputs of four phase
detectors corresponding
to the four references are deter-
mined as follows;
)0
=
lowpass{cos(ro"r +
0Xz)
sin to"f
}
= -sin
S
yu
=
lowpass
{
cos(co"t +
OXz)
sin(rrl"r +
i
r)
}
=O.707
cos
0
-
0.707 si n
$
yr
=
lowpass{cos(rrr.r
+
0X2)
cos oct}
=
cos
S
yA
=
lowpass{cos(ro"t
+
0X2)
cos(co"t *
i
n)}
=
0;707 cos
$
+ 0.707 sin
Q
(6.12)
As an aid in determining the appropriate
decision logic, these equations have been
evaluated for each of the eight
possible signal
phases and listed in Table 6.4. Exami-
nation of Table 6.4 indicates that the fust data bit is I whenever yg is positive. Simi-
larly, the second data bit is 1 whenever
y1 is
positive. The third data bit is a I whenever
)a,lr,
and y6 are all
positive or when they are all negative. Hence the third data bit is
determined as a logical combination
of phase detector outputs' In summary,
Dr =
Q
Dz=I
Dt =AI B+AI B
(6. 13)
where Di is the lth data bit and
0,
/, A, and B are logic variables representing
positive
outputs from yg,
/r, )4,
andyr' respectively.
296 DtcrrAL MoDULATtoN AND HADto sysrEMs
TABLE 6.4 Elght-P$K Phaee Detactor Outpurs
YA Yt YB Yn
Data Phase
011
010
000
001
101
100
1 1 0
1 1 1
nl8
3rll8
5nl8
7r18
-hil8
-5nl8
-3?r/8
4tlB
-0.383
4.924
-0.924
4,383
0.383
0.924
0.9?4
0.383
0.383
4.383
{.924
-0.924
-0.383
0,383
0.924
0.924
0.924
0.383
-0.383
-0.924
-0.924
{.383
0,383
0.924
0.924
0.924
0.383
4.383
-0.924
4.924
4.383
0,383
Another method of detecting 8-PSK signals, which avoids two extra references
and
phase
detectors, is revealed in the phase
detector Equation 6.12. The exha mea$ure-
ments yd
and y6 can be determined as
h=
0.70'lyr
-
O.707yn
t6=
0.7O7y, + 0.707y,
(6. 14)
Hence the
Ja
and
)r
measurements can be obtained as linear combinations of the
quadrature
channel phase
measurements y7 and
Jg,
and no additional phase
detectors
are required. The resulting 8-PSK demodulator/detector is shown in Figure 6.15. No-
tice that implementation
of the linear combinations in Equation 6.14 can ignore the
magnitude of the 0.707 multipliers since only the sign of the result is needed.
The linear combinations in Equation 6.14 essentially represent a rotation of the
quadrature
channel basis vectors through an angle ofr/4 radians. By changing the an-
gle
of rotation, the linear combinations needed for other phase
measuremen6
are eas-
Figure 6.15 Eight-PSK demodulator-detector
using only two references,
6.1 DIGITAL MODULATION 297
ily determined. Hence all phase-detector-based demodulators can be implemented
with two
phase detectors and as many linear combinations as needed. The general
equation for the linear combinations
produced by a rotation of a radians is
v " = v ^ c o s c L - y , s l n g
y. =
v^ $rn fl, + V, cOS
g, (6. 1s)
Beference and Clock Recovery
The demodulators di$cussed in the preceding paragraphs all require a local, phase-co-
herent, carrier reference for their operation. Furthermore, multiple-pha$e systems re*
quire
at least one more reference in quadrature with the first one. Recovery of any
reference is complicated by the fact that phase shift keying is a double-sideband -rap-
pressed carier modulation system. In other words, there is no discrete spectral line at
the carrier frequency as there is with some other types of modulation. In fact, the FCC
has ruled specifically that no spectral lines are allowed in the fransmitted signal. The
absence of a spectral line at the carrier frequency is overcome by using one of several
nonlinear
processing techniques
[1].
After one coherent reference is established, a
quadrature reference is obtained by delaying or differentiating the flust.
Clock recovery is obtained in several ways. Once carrier recovery is accomplished
and demodulation occurs, the clock can be obtained by locking onto transitions in the
baseband signal. Baseband clock recovery occurs as though the signal has never been
modulated and demodulated.
In contrast, the clock can sometimes be recovered di-
rectly from the modulated signal. If a PSK signal is heavily filtered, the envelope is
not constant and, in fact, contains amplitude modulation at the signaling frequency.
Hence the clock can be recovered, independently of signal demodulation, by envelope
detecting a heavily filtered PSK signal
[2].
Clock transitions are often assured by
purposely shifting the carrier reference half-
way between signal points in every interval. Using 4-PSK, for example, the references
would be shifted by 45o. During any one interval there are still only four possible sig-
nal states separated by 90o. However, the signal states alternate between being aligned
with the l and
Q
axes and being aligned at 45', as shown in Figure 6'9' Because a
phase
shift is ensured with every symbol interval, both the
(filtered)
carier envelope and the
recovered baseband signal have amplitude transitions. Without the reference shifts,
proper clock recovery requires a minimum density of data transitions as described in
Chapter 4.
D itfe re nti al Detectl o n
AS an alternative tO recovering a Coherent reference, some systems merely compafe
the phase in the present interval to the phase in the
previous interval. The signal re-
ceived in the previous interval is delayed for one signal interval and is used as a ref-
erence to demodulate the signal in the next interval. Assuming that the data have been
encoded in terms of phase shift, instead of absolute
phase positions, the data can be
decoded properly. Hence this technique, referred to as
"differential
dete{tion," inher-
ently requires differential encoding.
298 DIGITAL MODULATION AND RADIO SYSTEMS
In general,
PSK systems require differential encoding since the receivers normally
have no means of determining whether a recovered reference is a sine reference or a
cosine reference.- Furthermore, the polarity of the recovered reference is ambiguous.
Thus error probabilities
for PSK systems are doubled automatically because of the dif-
ferential encoding process. Differential detection, on the other hand, implies an even
greater
loss of performance
since a noisy reference is used in the demodulation proc-
ess. Typically, differential detection imposes a penalty
of I to 2 dB in signal-to-noise
ratio
[10].
PSK Specta
By far the easiest way to determine the spectrum of a PSK signal is to analyze the base-
band waveforms applied to the quadrature
channels. Owing to the orthogonality of the
two channels, the signals are uncorrelated, and the composite spectrum is merely the
sum ofthe individual (identical)
spectra.
In either ?-PSK or 4-PSK systems the baseband signal is a symmetric two-level
NRZ waveform. The corresponding spectrum is the common sin(x)/x spectrum shown
in Figure 4.2. High-level sysrems (8-PSK
or greater)
use symmetric multilevel NRZ
baseband signals similar to that shown in Figure 4.16. The multiphase
pSK
baseband
signal is somewhat different since unevenly spaced levels as def,rned in Table 6.3 are
used.
As mentioned in chapter
4, a multilevel NRZ signal has the same spectrum as a
two-level signal. Hence all conventional PSK systems produce
a spectrum that fol-
lows the sin(x)/x response defined in Equation 4.1 but translated to the carrier fre-
quency.t Figure 6.16 shows the PSK spectrum for two-, four-, and eight-phase
systems designed to provide
the same data rate. Hence the higher level systems signal
at lower rates and have proportionately
narTower spectra.
PSK Error Pertormance
The error performatrce
of any digital modulation system is fundamentally related to
the distance between points
in a signal space diagram. For example, a z-psK sy$tem,
as represented in the phase
diagram of Figure 6.9, is capable of optimum error per-
formance since the two signal points
have maximum separation for a given power
level (radius
of the circle). In other words, one 2*psK signal is the exact negative of
the other' Hence 2-PSK modulation provides
antipodal error performance
as defined
in Chapter 4.
The error performance
of a multiphase PSK system is easily compared to a 2-psK
system by determining the relative decrease in the error distance (voltage
output of a
properly
referenced phase
detector). In addition to the error distance, however, the
relative values of the noise bandwidths must also be considered.
(Recall
that the noise
bandwidth effectively determines the variance of the noise samples.)
*Absolute
phase
can be determined if a particular pattern
in the data sream such as a framing pattern
is
,unambiguously
related to some particular phase of the caxrier.
'The
baseband levels must be unconelated to each other to produce
a sin(.r)/.r spectrum, If
phase
transitions
from one interval to the next are restricted in some manner, the baseband levels are correlated. and a
different spectrum results.
6.1 DIGITAL MODULATION ?99
l - I ?
3T 2t 3r
.l
3
_E-
3T 2T 3T
Figure 6.16 Spectrum of unfiltered PSK signals carrying equal data rates.
The general expression for the distance between adjacent
points in a multiphase
PSK system is
(6.r6)
where N is the number of
phases.
A general
expression of the bit error probability (or bit error rate) of an N-phase
PSK system is derived in Appendix C as
6
3
F
E
s
E
E
T

t
T
(6. 17)
where
Equation 6.17 reveals that, with respect to EblNg, 4-PSK provides the same error
per-
formance as does 2-PSK. Thus, as mentioned earlier, both systems provide optimum
performance, but 4-PSK utilizes half as much bandwidth. The 4-PSK system has an
error distance that is 3 dB smaller than the error distance of Z-PSK. However, the
shorter error distance is offset by a 3-dB decrease in the noise bandwidth
(indicating
a 3-dB reduction in noise power at the detector). For the 4-PSK system to have the
same noise power at the detector it would have to experience a 3-dB greater noise
spectral density. Hence conventional signal-power-to-noise-power ratios
(SNRs) can
be misleading parameters for comparing digital modulation system$. Howevet, as
mentioned in Chapter 4, error rate
performances in terms of SNRs are desired when
d= 2 ,i"
[_N)
,=,*F)u"*rt"'["+J
Pu
=
*fu
erf (z)
\
"o*\
I
300 DtctrAL MoDuLAloN AND RADto sysrEMS
8 e 1 0 r 1 1 2 t s 1 4 1 8 t 8 1 7
Enrqy-prrdt-E-noh-dindty
mio
A/Vo {dBl
Figure 6.17 Error rates of PSK modulation systems.
E
o
t
=
E
t
6.1 DtcrrAL MoDULATToN 301
determining the effects of interference or when specifying error rate$ with respect to
measurable quantities. Figure 6.17 displays the error performance of 2-, 4-,8-, 16-,
and 32-PSK as a function of E6lNs. Appendix C
provides the relationships needed for
enor rates in terms of SNR.
6.1.4 Quadrature Ampl i tude Modul ati on
As described
previously,
a convenient means of representing phase shift keying with
four or more phases involves the use of quadrature signals. In the case of 4-PSK, the
quadrature signals represent two separate channels by virtue of the independence of
the baseband signals for each
quadrature channel. In higher level PSK systems, the
level of a baseband signal for the 1 channel is not independent of the baseband level
for the
Q
channel
(see
Table 6.3 or Figure 6. I I
).
After the baseband signals have been
established, however, the quadrature channel modulation and demodulation processes
are independent for all PSK systems.
Quadrature
amplitude modulation
(QAM)
can be viewed as an extension of multi-
phase PSK modulation wherein the two baseband signals are generated independently
ofeach other. Thus two completely independent
(quadrature)
channels are established
including the baseband coding and detection processes. In the special case oftwo lev-
els
(fl)
on each channel, the system is identical to 4-PSK and is often referred to as
such. Higher level
QAM
systems, however, are distincfly different from the higher
level PSK $ystems. Figure 6.18 shows a signal constellation of a I6-QAM system ob-
tained from four levels on each quadrature channel. The dots represent composite sig-
nal points
while the hatch marks on the axes represent amplitude levels in each
quadrature
channel. Figure 6.19 shows a basic
QAM
modulator and demodulator
structure along with,a representative waveform for l6-QAM.
Notice that, in contrast to PSK signals, the
QAM
signal shown in Figure 6.I 8 does
not have a constant envelope. A constant envelope is maintained with PSK modulation
by restricting the combination of levels on the quadrature channels. A
QAM
system
does not restrict the combinations since the levels on each channel are selected inde-
pendenfly. Thus
QAM
modulation cannot be used with saturating amplifiers.
Figure 6.18 Signal constellation of I6-QAM modulation.
30? DIGITAL MoDULATIoN AND HADIo SYSTEMS
Modulrtor
Oetnodulrtor
Ftgure6.19
QAMmodulator-demodulator.
The spectrum of a
QAM
system is determined by the spectrum of the baseband sig-
nals applied to the quadrature
channels. Since these signals have the same basic struc-
ture as the baseband PSK signals,
QAM
spectrum shapes are identical to
psK
spectrum shapes with equal numbers of signal points.
Specifically, l6-eAM has a
spectrum shape that is identical to I6-PSK, and 64-QAM has a $pectrum shape iden-
tical to tr-PSK.
Even though the spechum shapes are identical, the error performances
of the two
system$ are quite different. with large numbers of signal points,
eAM
systems always
outperform PSK systems. The basic reason is that the distance between signal points
in a PSK $y$tem is smaller than the distance between points in a comparable
QAM
system. Figure 6.20 compares the signal poinrs of a I6-QAM sysrem with the signal
set of a 16-PSK system using the same peak power.
r6-oi.u
r6-P$K
Figure 6.20 Comparison of I6-QAM and I6-PSK signal sets.
6,1 DIGITAL MODULATION
The general
expression for the distance between adjacent signal points in a
peak-amplitude
QAM
system with /, levels on each axis is
303
unit
Equations 6.16 and 6.18 reveal that an n-ary
QAM
system has an advantage over an
n-ary PSK system with the same
peak power level. In terms of average power levels,
the
QAM
system has an even greater advantage. The following equation, derived in
Appendix C, provides a
general expression for the peak-to-average ratio of a
QAM
svstem:
d=
{ 2
L - l
Peak
power
-
L(L
-
I)2
Average
power
zZ!!l
1zi
-
t1?
(6. r 8)
(6.re)
Example 6.1. Determine the enor
performance of a I6-PSK system relative to a
I6-QAM system with the same peak power level. Also determine the relative
performance with respect to identical average powers.
Solution. Since the two systems provide identical numbers of signal points, they
signal at the same rate and require the same bandwidth for a
given data rate. Thus the
relative error performance is completely determined by the relative distances between
signal
points. (When different signal rates are used, the effect of different noise
bandwidths in the receiver$ mu$t also be considered.) Evaluating Equations 6.16 and
6.18 indicate$ that 16-QAM has a 1.64-dB advantage over 16-PSK for a given peak
power. Equation 6.19 indicates that I6-QAM has a peak-to-average power ratio of
2.55 dB. Since PSK $ystems have unity peak-to-average ratios, the advantage of a
16-QAM system over a 16-PSK is 4.19 dB for equal average
powers.
The results of Example 6. I show that I 6-QAM is a significantly better modulation
format than 16-PSK for applications like voiceband models that have no technology
limits.- In digital microwave applications the
performance advantage of I6-QAM is
diminished by available amplifier technology, but, nevertheless, it is at 16 signal
points that PSK modulation defers to
QAM
modulation for
performance reasons.
Some first-generation digital radios used 8-PSK modulation
[13]
and some used
'Voiceband
modems are not resticted by technology but by signal power limits to prevent intetference with
other signals in the network. It is ironic that these rcstrictions arise primarily from old FDM analog radio
systems in which a high-powered signal in one channel could create inter{erence
(crosstalk) into other
channels. In a predominantly digital network the main signal limitation would be the saturation point of
PCM encoders (and possibly crosstalk in subscriber pairs). It is also ironic that increasing the signal power
ofa voiceband modem (V.34 and earlier) to falljust short ofthe PCM saturation point would not improve
performance in a mostly digital network. Performance of high-speed modems in this case is
primarily
determined by multiplicative noise crcated by PCM companders
[18].
304 DIGITAL MoDULATIoN AND HADIo SYSTEMS
: 21. 96d8
Figure 6.21 256
QAM
and stepped square 256
QAM.
I6-QAM
[a]
with comparable performance.
Second-generation radios use 64-eAM
[5]
followed by third-generation radios with 256-QAM
[16,
t7].
Reliable operation of high-density modulation formars like 256-eAM requires ex-
tremely linear amplifiers to tolerate the wide range of signal amplitudes and extremely
accurate adaptive equalizers to remove small percentages
of intersymbol interference
that arise when a high-amplitude pulse
is kansmitted adjacent to a low-amplitude
pulse. Amplifier and equalizer performance requirements are basically a function of
the dynamic range of the signal amplitudes*;
Dynamic roilg
=
(6.20)
The dynamic range of a square
QAM
modulation format can be improved by modi-
fying the signal set to eliminate corner points
that produce
high peak powers.
one such
technique referred to as 256-stepped square
QAM
(256-sseAM)
tlgl
is shown in
Figure 6.21, where it can be compared with the signal set of conventional 256-eAM.
Example 6.2. Determine the dynamic range of both the conventional 256-eAM
signal set and the 256-SSQAM signal set shown in Figure 6.21.
Solution. Using integral units along each
quadrature
axis of l, 3, 5,7,9,1 l, 13, and
15 for 256-QAM, the dynamic range (DR)
is
/ t q2- r - t s2\
DR(256-QAM): l Ol osr
|
' - ^"
I
o l
1 2 + 1 2 |
\ , /
=
23.52 dB
'Ifthe
quadrature carriers are perfectly recovered, intersymbol interference would only be a functio n ofthe
dynamic range along each axis. To make an allowance for quadrature
carrier phase
enor, which causes
interfercnce between I and
p
channels, the composite signal power
is used.
lorog,ot"Tl
.
6,1 DIGITAL MODULATION 305
The highest amplitude signal point of 256-SSQAM is the
(5,17) point.
Thus the dy-
namic range of the 256-SSQAM signal is
DR(256-ssQAM)
=
lo log,o
l'Hl
\ -
-
J
: 21. 96
dB
The result of Example 6.2 shows that the
peak power of 256-SSQAM is 1.56 dB
lower than the peak power
of conventional 256-QAM for the same signal
point sepa-
ration. Thus in a peak-power limited application, 256-SSQAM can operate with 1.56
dB more signal separation than can conventional 256-QAM.
When
other, less signifi-
cant factors are considered, reference
[20]
reports that a 140-Mbps 256-SSQAM radio
at6 GHz has a 2-dB performance advantage.
The theoretical enor rate equation for
QAM
modulation, derived in Appendix C,
is identical to the multilevel baseband enor rate equation with the same number of lev-
els. This equation is used to
produce effor rate curves for 4-, 16-, 64-, and 256-QAM
modulation in Figure 6.??. Table 6.5 compares the most coillmon forms of digital
modulation used in point-to-point microwave radio systems.
Offeet Keylng
Because of the interdependence of the baseband signals in highJevel PSK modula-
tion, the signal transitions on the
quadrature channels necessarily coincide. However,
since
QAM
systems independently modulate the
quadrature channels, they are not
consffained to align the signal intervals on the two channels. When the signal intervals
overlap each other by 507o
(Figure 6.23), the mode of operation is referred to as offset
keying. Offset keying is commonly used on 4-QAM systems
(more
commonly re-
ferred to as 4-PSK or
QPSK
systems). The main advantage of offset keying lies in the
ability of the reference recovery circuitry to become synchronized to the incoming car-
rier at lower signal-to-noise ratios than conventional
(aligned)
QAM
and
QPSK
sys-
tems
[21,22],
QAM Representation of Mlnimum Shlft Keying
The
preceding discussions of
QAM
sy$tems have assumed the use of baseband signals
with NRZ level encoding. A more general
view of
QAM
allows arbitrary
pulse shapes
in generating the baseband signals. One particularly interesting
pulse shape is a half-
sinusoid
pulse;
A(t):cosff
F I r
</ <+ zl
\ -
"
)
where ? is the duration of a sienal interval.
(6.21)
306 DGtrAL MoDUt-ATtoN AND RADto sysrEMs
9 10 l t 12 13 t 4 t 5 t 8 r7 18 rg 2{t 2l
Avrage rnergy-per-bit- to
-
mis
-
dsrrlty rutio Eb /,Vo
(dBl
Figure 6.22 Error rates of
QAM
modulation systems.
t0-?
l 0-3
10-l
to4
l(r7
l0-
-
1o-4
e
E
o
=
5
t
o
d
10-6
\
\
\
\ \
\
\
\
\
\
\
\
\
\
\
\
zEE-I evel
\
\
\
\
\
\
\
\
\
\
\
\ \
\
\
\
\
\
\
\ \
\
\
\
\
\ l
0l
+
\
t6-i
\
32-lcvil
\
\
6.1 DIGITAL MODULATION 307
TABLE 6.5 Comparison ol Varlous Digital Modulatlon Technlquss Based on Equal
Data Rstes
Sig nal-to-Noise^Ratios
for BER
=
10* (dB)
System
Designation
Information EblNo on the
Density
(bp#Hz)
Channel
SNH at Decision
Circuit
Peak-to-Average
Hatio (dB)e
2-PSK
4.PSK,4-QAM
QPR
8-PSK
16.QAM
16- QPR
16.PSK
32.QAM
64.QAM
1
2
Zb
q
4
4b
4
5
6
10, 6
10. 6
1?, 6
14. 0
14. 5
16. 5
18, 3
17.4
18. 8
13. 6
13, 6
17.6
' t8.8
20.5
24.5
24.3
24.4
26,6
0.0
0.0
2.0
0. 0
2.55
4.55
0.0
2.3
3.68
aRetio
of maximum steady-state Bignal
power
to averag signal power
with random data. MBasured on a
channel with square-rool liltor
partitloning.
Dln
a strict Bns, th6 signal bandwidth of a panial-response
systm is no narrower than the th6orBtical
(Nyquist) bandwidth of a conesponding fulFresponse sybtem. As a
practical
mannr, however, the partial-
re8pons systems require about 17ol, lsss bandwidth
[21].
If half-sinusoidal pulse shaping of duration 7 is used on both channels of an offset
keyed
QAM
system, the quadrature channel signals p(r) and
q(r)
can be expressed as
follows;
Q(t)
=
aq'-[ff)-i" *"t (6.22)
p(t)
=
ai*,[r$J""-
+r
f
+t=,=+t)
( 0<r <I )
t r ( r l = Dr ( , | ' c E
l ^l
^l ^l , l . l ' l . l " l " l ol . l ' l ^l
rln(dJl
Figure 6.23 Offset keyed 4-PSK signaling.
308 DGITAL MoDULATtoN AND RADto sysrEMS
where a; and c, are data values (tl) for the / and
0
channels, respectively, cos(ntlLT)
is shaping on the.l channel, and sin(nrl?T) is shaping on the
p
channel.
(The
sine func-
tion is used to account for offset keying.) Adding the two quadrature
signals together
and applying some fundamental trigonometric identities produces
the result
x(t): p(t) + q(t)
l '
nr)
=ai cos[ +r +7J
( a, =no)
( m \
=
4i cos
[*t
-
rrj
(a,+
au)
(6.23)
Equation 6.23 is essentially identical to Equation 6.6 defining minimum shift keying.
Hence, except for a logicJevel transformation of data values, offset
QAM
with
sinusoidal pulse
shaping is identical to MSK
15,241.
The relationship is further dem-
onstrated in Figure 6.24, which shows how an MSK signal is generated by offset
keyed quadrature
channel modulation with sinusoidal pulse
shaping.
I-Channel brrohnd: Dr
(rl
,5rltl
=
Df
(tl
'
cor
(o.rl
Q-Chennel
bftsbrnd: Dg
lrl
.$a (tl =
Dq
(rl
.
rin
(<o.rl
$(t l
=
sr{t l +. So (rl
^
l r , l
^I t
w,
=2t f ,
l r l r . l r " l r " l ^l o
lo
*
2'l5lt
l ^l ^l o
fi
=a.isrt
Figure 6.24 MSK signaling produced
by offset keyed
QAM
with sinusoidal pulse shapes.
6
3
F
T
E
t
f,
I
6.1 DterrAL MoDULATToN 309
Figure 6.25 Power spectra of MSK and 4-PSK signals.
These foregoing results show that MSK is very closely related to offset-keyed 4-
PSK. The only difference is in the use of half-sinusoidal baseband pulse shapes for
MSK and square pulse
shapes for 4-PSK. Because of this close relationship, it is in-
teresting to compare the amplitude spectra of the two systems in Figure 6.25. As in-
dicated, the MSK spectrum has its first spectral null at a 507o higher frequency than
4-PSK's first null. Other than this, the MSK spectrum is more compact than the 4-PSK
spectrum and is significantly lower in amplitude at frequencies outside the main lobe
of the spectrum. For thig reason MSK is an attractive modulation technique where con-
stant envelopes and little or no filtering are desired. One such application is on digital
satellite links with frequency division subchannels
[7].
In applications requiring trans-
mitter filters, MSK has no
particular advantage over 4-PSK and usually requires a
more complicated modulator.
6.1.5 Carrl erl ess Ampl i tude and Phase Modul atl on
Carrierless amplitude and
phase (CAP)
modulation
[25]
is a specialized form of
QAM.
As indicated in Figure 6.26, no explicit modulation of a carrier
(or carriers) oc-
curs. Instead, two separate modulated signals are directly generated with DSP filter
functions: one for an in-phase channel and one for a quadrature channel. The in-phase
filter convefts f channel data directly into a filtered DSP cosine wave while the
quad-
rature filter converts
0
channel data into a filtered DSP sine wave. In this manner the
symbol rate is locked to an implied carrier embedded into the filter algorithm.
(Typi-
cally, the embedded carier frequency is equal to the baud rate so there is one cycle of
a
"carrier"
in each symbol interval.) After fhe in-phase and
quadrature-phase signals
Frnumcrr {Hzl
310 DIGITAL MoDULATIoN AND HADIo SYSTEMS
Figure 6.26 CAP modulator block diagram.
are added, the result is converted to an analog signal and f,rltered to smooth the DSp
sampling frequency.
CAP modulation is used in one version of asymmetric digital subscriber line
(ADSL)
[26]
and for a 51.84-Mbps unshielded twisted-pair ATM srandard
[27].
ADSL applications are described in Chapter 11.
6.1.6 Partl al -ResponseQAM
Another popular modulation technique is quadrature partial-response
signaling
(QPRS).
As commonly implemented
[28],
a
QPRS
modularor is nothing more than a
QAM
modulator followed by a narrow-bandpass filter that
"over
filters" the quadra-
ture signals and produces
controlled intersymbol interference in each channel. The
most common application of
QPRS
involves two levels on each channel before filter-
ing and three levels afterward (see Chapter 4).
This system is essentially a 4-PSK system with partial-response
filtering to in-
crease the information density. As shown in Figure 6.27, the effect of partial-re$ponse
fiItering is to produce nine signal points
from the original four. In a similar manner, a
16-QAM partial-response
system, with four levels on each channel before filtering,
has seven levels afterward and 49 signal points in all
t291.
Edtor" filtfiing
Afti. tlltfiing
Figure 6.27
QPRS
signal constellations.
\-
6.1 DIGITAL MODULATION 31 1
Heired codne chonncl
Cdinc chrnnol
r
-i
Truncetion
l ol l
=
OdB
(due
to
prkingf
NBw
=
1/ r
NBW
=
Z/tT
(2dB
bolort l/n
Sigr|nl
sspsr|tiod
=,/Ve
Signrl
rGpdratlon
=t/Iatz
Truncrtlon
l on
=
2dB
Figure 6.2E Comparison of
QPSK
and
QPRS
systems with equal data rates.
Although the modulators of a
QPRS
system can be conventional
QAM
modulators,
the demodulator/detectors must be modified to account for the extra levels in the
waveform. After the signal is demodulated, the detection processes for each channel
are independent and identical to the baseband PRS detection procedures described in
Chapter 4.
Figure 6.28 compares a
QPRS
system to a
QPSK
(4-PSK)
system in terms of equal
peak power out of the modulator. The average transmit
powers are different, how-
ever, because the partial-respon$e system overfilters the signal to reduce the trans-
mitted bandwidth. In
practice, the information density is increased by about 177o
t231.
As discussed in Chapter 4, partial-response filtering cuts the distance between sig-
nal points in half, indicating a 6-dB reduction in error performance. However, the
noise bandwidth of a PRS receive filter is lower than the noise bandwidth of the cor-
responding full-response system so that some of the enor distance degradation
is re-
covered. For the filter systems shown in Figure 6.28 the net
performance loss of a
QPRS
system is 4 dB. Notice, however, that with respect to
power on the channel the
performance loss is only 2 dB.
6.1.7 Trellls-Coded Modulation
The 2-dB
pedormance disadvantage of PRS mentioned in the previous section as-
sumes that data are detected by making hard decisions one bit at a time. As described
in Chapter 4, however, a PRS signal has redundancy that overlaps two bit intervals. If
the detection circuitry processes the signal spanning two intervals before making hard
decisions the performance penalty
can be recovered. Processing a redundant signal in
312 DrcrTAL MoDULATtoN AND RADto sysrEMS
this way is an example of a maximumlikelihood
or viterbi decoding algorithm
[30]
for redundant signals. If the redundancy is extended to more levels and more intervals
and optimally processed,
even greater
coding gains
are possible. As typically imple-
mented, these extensions use extra signal points with reskicted sequences
(transitionrr
between points).
When the allowed sequences are represented with state ffansition
diagrams, they form a trellis. Hence the term trellis-coded modulation (TcM)
t3
l, 3zl.
As mentioned in chapter4, TCM is very similarto convolutional coding. The ma-
jor
difference is that convolutional coding adds redundant symbols (increases
the
bandwidth) but TCM adds redundant signal levels (increases
the signal space). Both
$ystems are detected in similar fashions using trellis diagrams. The decoding algo-
rithms of both $ystems essentially determine the most likely sequence of states of the
tran$mitted codes, The transitions between states then determine the data.
As a TCM example, consider the expansion of a 4-PSK signal set to an B-psK sig-
nal set as shown in Figure 6.29. Although there are eight signal points in the
g-psK
constellation, only four points
can be freely chosen in any one interval. which four
points
can be chosen is dependent ofthe signal points
chosen in previous
intervals. A
TCM demodulator/decoder then determines which of the allowed signal sequences
most closely matches a received $equence before making hard data decisions.
Figure 6.29 shows that if data decisions are made one interval at a time. without
processing
the redundancy, the error performance
of ttre expanded signal set is 5.33
dB worse than the original signal set (a
consequence of the noise power
margin of ad-
jacent
signal points being reduced from 0.5 to 0.146). Also shown in Figure 6.29 are
the noise margins of more widely separated signal points.
TCM recovers the 5.33-dB
penalty,
and more, by ensuring all allowed sequences include some of the larger dis-
tances. Furthermore, noise in only one interval is not likely to produce
a decision error.
when four-state coded 8-PSK uses the transition rellis shown in Figure 6.30, the per-
formance approaches a 3-dB improvement with respect to uncoded 4-psK
t3ll.
To understand Figure 6.30, it is important to realize that it represent$ state transi-
tions in the encoding process,
not signal waveforms. The use of the states in this man-
4 (157. 5)
or2= si nl ae.s") =
0.146
d3=si nt +s") =0. S
dsz=si nt oz. b") =0. 854
doz=si nt ggo) =1. 0
5
(202.5)
Figure 6.29 EighrPSK signal points
and error disrances.
6,1 DIGITAL MODULATION 313
Figure 6.30 Four-state trellis for 8-PSK modulation.
ner is
just
a convenient means of representing
past history.
(All past histories are en-
capsulated into
just
four states.) Transitions from one state to another are labeled with
the 8-PSK signal point chosen for encoding 2 bits of data. Notice that only four of the
8-PSK signal points can be freely chosen and that these
points have the sarne $epara-
tion as uncoded 4-PSK. The significant aspects of the decoding
proces$ are suflrma-
rized as follows;
1 . An allowed transition between two states can occur with either of two signals,
which implies sequence information does not help discriminate between those
two particular signals. However, the two signals in
question are chosen to have
maximal separation
(noise margin 1) so redundancy is unnecessary.
Transitions that originate in different states and terminate in any
particular state
are encoded with signals having noise
power margins of 0.5.
All sequences that begin and end in common states are at least three intervals
long with minimum noise
power margins of 0.5, 0.146, and 0.5. Thus the total
noise margin between any two minimum-length sequences is 1.146.
It may be necessary to discriminate between seguencerl that are longer than three but
in all cases the endpoints involve transitions with noise power margins of 0.5' Thus
the noise margin is always greater than l, which is 3 dB
greater than the noise margin
of uncoded 4-PSK. Determination of the exact error rate of a TCM system is much
more involved than when interval-by-interval
detection is used. Reference
[31]
shows
2.
3.
314 DGTTAL MoDULATIoN AND HADto sysrEMS
that the bit error rate of coded 8-PSK asymptotically approaches 3*dB improvement
with respect to 4-PSK as the noise margin would indicate.
Example 6.3. A receiver for a trellis-coded modulation system as shown in Figure
6.30 detects a
phase
sequence of 20o, 220", and 10o. Determine the allowabre
$equence of three signals with the closest distance to the given received sequence.
Assume the starting state is state B.
solution. To begin with we can determine the closest encoder signal phases as
22.5",202.5o,22.5",whi chcorrespondtosi gnal poi nts
l ,5, l .Exami nati onofFi gure
6.30 indicates that this is a disallowed sequence because a 5 signal cannot follow a I
signal. If data were decided on a symbol-by-symbol basis, an error would certainly be
made. By tracing the trellis from state A, a list of allowable state sequences can be
determined as provided
in Table 6.6. For each transition, the most likely signal of each
pair of signals that can produce
a particular
transition is indicated. The symbol errors
(in
degrees) for each of these signals is then determined followed by the total sequence
error. As indicated, the most likely state sequence is cAA (signal
sequence: 1,6,0).
Although the result of Example 6.3 indicates one particular
sequence is more likely
than any other, there are two other allowable sequences (l4l
and 162) that are fairly
close to the most likely signal sequence. Notice that these two sequences have end
states (C and B) that are different than end state A ofthe selected sequence. Thus, there
is more information to come as to which is the most likely sequence. Because states
A and B have no coillmon allowed signals, the next symbol will provide
additional
discrimination between 160 and 162. The very next signal will not help discriminate
between 160 and 141 but the signal following that will. A thorough determination of
the most likely transmitted signal sequence needs to consider other possible
starting
phases and, consequently, previous
signal values.
(This
is an exercise left to the stu-
dent.)
Coded 8-PSK TCM has been used in satellite circuits
[33]
where nonlinearities dic-
tate the use of a constant-envelope
signal. TCM with higher density
eAM
signal sets
TABLE 6.6 Sequence and Error Detrmlnatlon for Example 6.3
States
Signals Symbol Enors Total Enor
cBc
DDD
CAA
DCB
CBD
DDC
CAB
DCA
141
351
160
370
143
353
16?
372
?,5, 62.5, 12.5
9?.5,17.5, 12.5
2. 5, 27. 5, 32. 5
92.5,72.5,32.5
2.5, 62.5, 102.5
92.5, 17.5, 102.5
2. 5, 27. 5, 57. 5
92.5,72.5,57.5
77.5
122.5
62.5
197. 5
167. 5
212.5
87.5
222.5
6.1 DIGITAL MODULATI0N 315
is used in high-speed voiceband
QAM
modems such as the 14.4-kbps V.33
[3a]
and
28.8-kbps V.34
[35].
TCM can also be used in conjunction with MSK modulation
t361.
TCM has been utilized in point-to-point microwave radios such as the 155-Mbps
6G150 MB radio of NEC. The modulation format is 128-QAM. NEC refers to TCM
as multilevel coded modulation
(MLCM).
6.1.8 Mul ti carrl erModul ati on
In all of the modulation formats described to this
point it has been implicitly assumed
that modulation occurs on a single carrier, or possibly on two quadrature carriers op-
erating at the same frequency. Multicarrier modulation
(MCM) involves dividing the
data stream into multiple, lower rate sffeams that are transmitted in parallel on multi-
ple
carrier frequencies. Although MCM has been utilized in some special applications
in the
past
[37],It
is the emergence of DSP technology that allows economic imple-
mentation of multiple modulators and demodulators for commercial applications. The
most common means of implementing MCM utilizes fast Fourier transforms
(FFTs),
as shown in Figure 6.31. Source data are demultiplexed
into separate data streams for
each subchannel. The data for each subchannel are then independently encoded into
discrete-level complex numbers representing the amplitudes of quadrature channel
carriers. An inverse FFI transforms the complex $pectrum components into a time-
domain waveform that is converted to analog and transmitted.
The basic steps of the process are reversed in the receiver where the incoming time-
domain waveform is sampled and fed to an FFT that recovers a complex spectrum'
The spectrum components are then processed as individual
QAM
(or possibly PSK)
signals to decode the data, which are then multiplexed back into a composite data
stream. Not included in Figure 6.31 are ancillary functions of equalization and clock
recovery. Normally, one frequency component is dedicated as a pilot signal that car-
ries clock information fbr all subchannels. Amplitude equalization involves merely
scaling the FFT component amplitudes at the output of the FFT. Amplitude equaliza-
tion may not even be required if PSK modulation is utilized because, even though
there may be amplitude distortion in the channel, within each narrow subchannel the
InvstEe
Encode FFT
Figure 6.31 Multicarrier modulation utilizing fast Fourier transforms.
316 DtctrAL MoDULAIoN AND RADto sysrEMS
distortion is essentially flat, which implies that PSK data can be recovered by merely
determining the phase
of each complex frequency term.
Multicarrier modulation with an FFT implementation is commonly referred to as
discrete multi-tone (DMT)
in North America and as orthogonal frequency division
multiplexing (OFDM)
in Europe. (The
term
"orthogonal"
occurs because the fre-
quency
components of an inverse FFT are harmonically related and therefore have
zero-valued cross-correlation products.) The terms DMT and OFDM are interchange-
able with the exception that in some OFDM applications it is understood that all sub-
channels utilize the $ame form of modulation with the same number of bits per
channel. In DMT systems the modulation of the subchannels is more general
so that
different data rates can be carried on different subchannels depending on the transmis-
sion quality
of the respective subchannels. A principal
application of DMT is ADSL
[38]
standardized by ANSI committee TlEl.4. OFDM is specified by European Tele-
communications standards Institute (ETSD
as the modulation format for Digital
Video Broadcasting (DVB)
t391.
One of the most attractive features of DMT is the inherent ability to match an in-
formation signal spectrum to a channel response. An example of such a system is
shown in Figure 6,32 that is representative of an ADSL application on subscriber wire
pairs. A significant impairment of using existing wire pairs for high-bandwidth clata
is the possible presence
of bridged taps. ISDN basic rate installations require removal
of bridged taps. In an ADSL implementation bridged rap$ are accommodated by de-
tecting their presence
during channel characterization
and then transmitting only as
much information in the affected subchannel(s) as can be reliably supported. Notice
that a conventional wideband data signal would experience significant distoftion if
ffansmitted on the channel
of Figure 6.32. Thus, a significant amount of amplitude
(and probably phase)
equalization would be required.
Bridged tap notch
Channel
response
Narrowband
interference
Information
density per
subchannel
Frequency
Figure 6.32 Use of DMT modulation to match subchannel data capacity to transmission
channel.
6.2 FILTER PARTITIONING 317
Figure 6.32 shows that a DMT system assigns information densities to individual
subchannels based on respective signal-to-noise
ratios determined during channel
characterization.
Included in the
process is detection ofnanowband interference that
causes a subchannel to be eliminated altogether. Notice that a conventional
wideband
system might be totally disabled by the narowband interference.* For more details on
DMT implementations see the ADSL discussion in Chapter I l.
6.2 FILTERPARTITIONING
The transmitting and receiving equipment of a digital radio system typically contains
several filters that limit the signal spectrum to some degree or another. Since the end-
to-end frequency response of the channel must conform to certain baseband
pulse re-
sponse objectives, the desired composite filter function must be
partitioned among the
individual filters. Normally, a number of the filters can be designed to
provide their
respective functions without significantly impacting the channel
pulse response. For
example, a mixing
process produces a sum and a difference of the input frequencies.
Only the sum is wanted when mixing upward, or only the difference is wanted when
mixing downward. Usually the undesired terms can be eliminated by a filter that does
not significantly affect the pulse shape of the underlying signal. The following discus-
sion assumes that only two filters significantly influence the baseband
pul$e response;
one in the transmitter and one in the receiver.
6.2.1 Adlacent-Channel Interference
One basic
pil?o$e of the radio channel receive filter is to minimize the amount of
noise present
at the detector. A second purpose of this filter is to reject energy in ad-
jacent
radio channels. Energy from an adjacent channel that does not get rejected is
referred to as adjacent-channel interference.
In this discussion, we assume that the
spectrum shapes in the adjacent channels are identical to the desired spectra. This situ-
ation is shown in Figure 6.33, which depicts a number of frequency-division-multi
plexed digital channels.
As indicated in Figure 6.33, adjacent-channel
interference occurs as a result oftwo
phenomena. First, the receive filter passes
unwanted
power P1 because the adjacent
signal is not completely truncated to
prevent overlap into the desired channel. The sec-
ond source ofinterference, P2, occurs because the receive filter does not
provide inf,t-
nite attenuation of power properly belonging in the adjacent channel. Unwanted
power P1 is minimized by narrowing the transmit filter, while P2 is minimized by nar-
*Totally
avoiding the effects of high-energy narrowband interference in a DMT sylttem is not as simple as
it might seem, If the interference is
present at the input to the A,/D converter in the recei ver, the interference
may cause A/D saturation
(orincreased quantization noise ifa companded converter is utilized). Thus, total
avoidance of narrowband interference requires front-end norching of the signal. A front-end notch could
also be used in a conventional wideband system followed by decision feedback equalization to
accommodate the inserted distonion
[401.
318 D|GITAL MoDULATIoN AND RADIo SYSTEMS
rowing the receive filter. Since channel pulse
response considerations constrain the
composite filter function to some minimum width, one component of the interference
cannot be reduced without increasing the other. Hence the total filter function must be
partitioned
in some marlner to minimize the sum of Pr and
p".
6.2.2 Optl mum Parti ti onl ng
The optimum filter design for any particular
application may depend on a number of
factors including legislated emission specifications, available technology for power
amplifiers, the availability ofcross polarization
for adjacent*channel isolation, and the
relative effects of noise versus adjacent-charurel interference. In the absence of exter-
nal constraints, a classical result attributed to Sunde
t4ll
and also presented
in refer-
ences
[42]
and
[43]
determines optimum partitioning
as one marching the outpur of
the transmit filter to the square root of the desired channel response. The desired out-
put
spectrum Y(or) is obtained as
Y(or)
=
Holo);I1r*(ol)X(ro)
where X(ro)
=
channel input spectrum
flrx(rn)
=
transmit filter response
/1nx(ot)
=
receive filter response
Then the optimum partitioning
is obtained as
(6.24)
Adi*rnt
rlgnrl
Adleent
rl0nrl
De$rrd
tignrl
Rffihrfi filtrr.E|pofite
Figure 633 Adjacent-channel interference.
6.2 FILTEH PARTITIONING 319
lflo*(o)l
=lY(rt)llt?
lar*(ru)l=H
(6.25)
Equation 6.25 def,rnes the filter amplitude re$ponses to minimize the adjacent-channel
interference under the condition tttat the adjacent channels contain identical signals
with identical
power
levels. In addition, if the transmitted spectrumffty(ro)X(ro) is the
complex conjugate of the receive filter response Hs;q(o), the receive filter is matched
to the channel spectrum, and the best
possible error performance is obtained with re-
spect to signal
power
on the channel.
Partitioning as defined by Equation 6.25 is shown in Figure 6.34 for a pulse
input
and a raised cosine output
(see Appendix C). Even though only baseband signal spec-
tra are shown, the concept is
extended easily to
passband filter functions.
Although Equation 6.25
provides a sound theoretical basis for determining the op-
timum filter
partitioning, practical considerations
of a
particular application may re-
quire deviation from the
"optimum."
One problem that arises in digital microwave
radios is related to the peak at the band edges of the transmit filter in Figure 6.34. With
a
passive fllter this peak can be obtained only by inserting loss into the midband re-
sponrte. If the transmit power can be increased as compensation, no ill effects result.
However, the output level of many digital microwave radios is limited by the technol-
ogy of microwave frequency power amplifiers
(a
few watts typically). Since these ra-
dios are device
power limited, midband insertion loss cannot be overcome and
therefore subtracts directly from the received signal level.
In device-power-limited applications the optimum hansmit filter is one having a
flat response in the passband. To achieve the desired channel response at the detector,
the receive filter must then be
peaked at the band edges. Channel insertion losses at
this point do not degrade
performance $ince both the signal and fhe noise ate attenu-
ated equally. However, the peaks have the undesirable effect of increasing the receiver
noise bandwidth and the P1 component of adjacent-channel interference. Although
these increases represent a
poorer perfonnance than that obtained by theoretically op-
timum partitioning, the degradation is not a$
Sreat
as the insertion loss needed to
achieve the
"theoretical
optimum."
As one fulther note, it should be mentioned that
ilob. Nnd
Infirfafdftot
Flgrrre 6.34 Theoretical optimum filtering for square-pulse excitation and raised-cosine
output.
320 DrcrrAL MoDULATtoN AND HADto sysrEMS
not all theoretically optimum transmit filters have a peak at the band edges. In
particu-
lar, partial-response
systems do not require band edge peaking (see
Appendix C).
Another aspect of the optimum design to be kept in mind is that optimum error per-
formance is achieved with respect to signal power
on the channel, and not with respect
to unfiltered transmitter power.
It is this feature that allows any amount of attenuation
to be inserted into the charurel at the transmitter and not degrade the optimum perform-
ance. If performance
is measured with respect to unfiltered transmit power,
the best
transmit filter may be different from that defined in Equation 6.25. Not only would
less midband attenuation be desirable, as discussed, but it might also be desirable to
widen the transmit filter response. This decreases the truncation loss in the kansmit
filter so the receive filter can be narowed to decrease the receiver noise bandwidth.
As an extreme example
of how widening the transmit filter can improve error rate
performance,
consider removing the transmit filter and incorporating it into the re-
ceiver. For a
given
output at the power
amplifier, the signal at the detector is un*
changed. The receiver noise bandwidth, however, is reduced because the composite
filter is nilrower than the original receive filter. Hence a higher signal-to-noise ratio
is present at the detector.
The penalty
incurred for removing the transmit filter, of cour$e, is a greatly in-
creased P1 component ofadjacent-channel interference. Ifadjacent channels do not ex-
ist or if they are adequately isolated by cross-polarization (i.e.,
if the system is noise
limited), the performance
can be improved by moving some of the transmitter ffunca-
tion losses to the receiver until adjacent-channel interference matches the noise. Keep
in mind, however, that if the sy'rtem is adjacent-chamel interference limited, the op-
timum partitioning
is indeed defined by Equation 6.25.
6.3 EMISSION SPECIFICATIONS
One unavoidable consideration when determining the filter functions of a transmitter
and receiver is the out-of-band emission specifications established by the FCC in the
united states or the ITU-R in other parts
of the world. In many cases the emission
specifications dictate a narower transmit filter than the theoretical optimum. Thus the
partitioning
aspect of the filter designs may be
predetermined.
It is somewhat ironic that the FCC emission specifications were intended to control
adjacent-channel interference but, in some cases, actually cause the interference to in-
crease by forcing the use of a wider than optimum receive filter. The FCC specifica-
tions, however, were intended to protect
adjacent analog radio channels from
out-of-band digital emissions. They were not selected with adjacent digital channels
in mind.
In January 1975, when the FCC established the out-of-band emission specifica-
tions, two separate specifications were established; forradios operating below 15 GHz
and for radios operating above l5 GHz. The emission limitations for operation below
l5 GHz are more restrictive than those for operation above 15 GHz because the lower
frequencies were heavily used for analog FDM-FM radios, which are more sensitive
to interference. The higher frequencies have not been used extensively because of vul-
6.3 EMISSTON SPECIF|CATIONS 321
nerability to rain attenuation. These frequencies are used for short-distance, special-
putpose applications and are not as congested as the lower frequencies. Initially, the
most
popular bands for digital microwave were I I and 6 GHz.
The emission limitations for operation below 15 GHz are
A
=
35 + l 0l og,oB + 0.8(P
-
50)
(P > 50)
(6.26)
where A
=
power in 4-kHz band relative to mean output power (dB)
B
=
authorized bandwidth
(MHz)
P
=
percent removed from carrier frequency
(50
is the band edge)
In addition, the attenuation A must be at least 50 dB everywhere outside the band but
does not need to exceed 80 dB at any point. Notice that the emission limitations are
not $pecified in terms of absolute
power levels, but only relative to the transmitted
power. Thus these emission limitations do not constrain the output
power level of the
radio. Microwave output powers are often conshained by microwave amplifier tech-
nology.
(A
limit of 10 W of ouryut
power does exist when technology can reach it')
Figure 6.35 displays the emission mask for I 1-GHz radios using a 40 MHz of band-
width. Also shown is the power spectrum of a 90-Mbps 8-PSK radio using a signaling
rate of 30 MHz. To be compatible with the emission specification, the $-PSK
power
spectrum is shown in terms of
power per 4-kHz band, The difference between the sig-
nal spectrum and the FCC mask represents the minimum attenuation required of the
transmit filter. The filtered 8-PSK signal thereby
provides an information density of
90/40
=
2.25 bpslHz. This is the basic modulation and frltering format used by Col-
lins/Rockwell in their 9O-Mbps digital radio, the MDR-I1
[13]'
The emission limitation for microwave bands above 15 GHz is defined as
PorYrr in {kHr bmcl rclstir'c to
mcln outFut potw
{dBl
FCC rmidon limitrtiom
SX $.ct.um Fowr/{
kHr
(3O
irHt $gilllne .rtrl
h
Frnqucncy
(MHrl
-60
* 70
-H)
Figure 6.35 FCC emission mask at I t GHz and a 90-Mbps 8-PSK spectrum.
322 DIGITAL MODULATION AND RADIO SYSTEMS
A
=
l l + 101og, oB +0. 4( p- 50) ( p>50)
where A
=
attenuation in l
-MHz
band below mean output power (dB)
B
=
authorized bandwidth (MHz)
P
-
percent
removed from carrier frequency
(6.27)
The attenuation A must be at least I I dB but does not have to exceed 56 dB.
In conjunction with the emission limitations, the FCC has stipulated that no dis-
crete lines exist in the transmitted spectrum
[44].
Thus no carrier components can exist
and no repetitive data patterns
are allowed to occur. The repetitive data pattems
are
effectively eliminated by using a scrambler in the transmitter and a descrambler in the
receiver. As mentioned already, PSK and
QAM
are forms of double-sideband sap-
pressed
carrier modulation so that the carrier terms are eliminated automaticallv as
long as modulation is continuous.
6.4 HADIO SYSTEM DESIGN
The foremost design requirement of a
point-to-point
radio system for telephony is op-
erational dependability, usually referred to as availability. Availabitity is expressed as
the percentage
of time that a system provides
a specified minimum quality
of service.
with analog systems, minimum performance
is determined by the noise power in the
received signal. The performance
of a digital system is determined by the bit error rate.
Typical objectives for bit error rates range from l0-3 for voice trafTic to 10-6 or l0-7
for data traJfic. Recall that a bit error rate of l0+ corresponds to the threshold ofper-
ceptibility fbr errors in a PCM voice signal. Typical design objectives for microwave
radio systems speciff availability on rhe order of
gg.g\vo
[45,
46]. Hence the maximum
acceptable accumulation of outage, due to all causes, is on the order of z hr per year.
Radio system availability is dependent on equipment reliability and path propaga-
tion characteristics. The high-availability
objectives of a typical radio system mandate
redundalt equipment and often require redundant path$.
The need for redundant paths
is determined by the likelihood of atmospheric-induced
outages
(rain
attenuation or
multipath interference). Rain is a dominant consideration at higher carrier frequencies
(above
lt GHz), and multipath interference must be considered at all frequencies.
Multipath fading is dependent on prevailing
climate and terrain.
Redundant radio equipment typically operates in either a backup or a hot-standby
mode with automatic protection
switching. The transmission path is backed up with
spare channels (freguency
diversity) or spare paths (space
diversity receivers). In ex-
treme ca$e$ a backup route may even be utilized.
6.4.1 Fade Margins
The main technique used to circumvent atmospheric-induced
outages is to provide
an
exffa $trong signal at the receiver during normal path
conditions. The difference be-
6.4 RADIO SYSTEM DESIGN 323
tween the normal received
power and the power required for minimum acceptable
per-
formance is referred to as the fade margin. Greater fade margins imply less frequent
occurrences of minimum
performance levels. Radios operating in higher frequency
bands generally require
greater fade margins because they are more susceptible to rain
attenuation. A 50-dB fade margin is typical for a digital radio at I I GHz, while a 40-
dB fade margin is typical for lower microwave frequencies.
The amount of fade margin actually required for a particular route depends on the
probability of multipath-induced fades and heavy rainfall occurrences. Thus drier cli-
mates permit lower fade margins, thereby allowing
greater
repeater spacing. In some
mountain-based microwave links in the western United States, microwave hops can
be 100 miles long. By comparison, the average hop in other
parts of the country is less
than 30 miles long.
When large fade margins are provided, the received signal power during unfaded
conditions is so strong that bit errors are virtually nonexistent. Nevertheless,
problems
with extra strong signals do exist. Namely, automatic
gain control in a receiver must
operate over a wide dynanric range. If the maximum signal level into the demodulation
and detection circuihy is not controlled, saturation is likely to degrade petformance,
especially in high-density modulation formats such as 64- or 256-QAM, where infor-
mation is encoded into the signal amplitudes.
To minimize dynamic range requirements in a receiver and reduce interference be-
tween systems, adaptive transmitter
power control
(ATPC)
is sometimes used
[47].
ATPC uses a feedback data link from a receiving station to control the output power
of a transmitting station. Thus when excess power is unnecessary, it is not used' ATPC
is commonly used in digital mobile telephone $ystems where interference control is a
primary concern.
6.4.2 System Gain
One of the most impoltant parameters used to characterize digital microwave system
performance is the system gainA,. System
gain is defined to be the difference, in
deci-
bels, of the transmitter output power and the minimum receive
power for the specified
error rate;
(6.28)
A r =
where P.1
=
transmitter output power
Pn
=
rceive power for specified error rate
The minimum acceptable receive power is sometimes referred to as the threshold
power and is primarily dependent on the receiver noise level, the signal-to-noise
ratio
required by the modulation format, and various system degradations such as excess
noise bandwidth, signal distortions, intersymbol interference, carrier recovery offsets,
fr,\
r0 ros,o
['p;j
324 DIGITAL MoDULATIoN AND RADIo SYSTEMS
timing
jitter,
and coupling and filter losses that either attenuate the signal or increase
the noise level.
Noise power
in a receiver is usually do:ninated by thermal noise generated
in the front-
end receiver amplifier. In this case, the noise power
can be determined as follows:
pN
=
(FXNfl(B)
=
F(kTo)B
where F= thereceivernoise figure
No
=
the power spectral density of the noise
B
=
the receiver bandwidth'
ft
=
1.38(10)-23 is Boltzmann's constant
7o
=
the effective receiver temperatures in degrees Kelvin
(6.2e)
Equation 6.29 essentially states that the receiver noise power
is determined by the
spectral noise density of the receiver input resistance and the additional noise inho-
duced by the amplification process (noise
figure F). Normally, a reference tempera-
ture of 290 K is assumed so that the thermal noise density (ftTo)
is 4 x 10-21 wFIz.
The noise figure of any device is defined as the ratio of the input signal-to-noise
ratio to the output signal-to-noise ratio;
_
(S/Mi,
r = -
(s/N)out
(6.30)
In effect, the noise figure specifies the increase in noise power
relative to the increase
in signal power.
since all physical devices inhoduce noise, the noise figure ofany sys-
tem is always greater than I and is usually expressed in decibels. If a system has no
gain or attenuation, the noise figure is exactly equal to the ratio ofoutput noi$e to input
noise. Noise figures of low-noise microwave amplifiers typically range from z to 5
(3-7
dB). Radio receiver noise figures are typically 6-10 without a low-noise ampli-
fier.
combining Equations 6.28 and 6.29 and incorporaring a term D for the degradation
from ideal performance produces
the following general
expression for system gain;
(6.3
r)
where SNR is the theoretical signal-power-to-noise-power ratio required for the maxi-
mum acceptable error rate and D includes all degradations from ideal performance.
-Typically,
B is assumcd to be the minimum theoretical bandwidth for the particular modulation format in
use. Excess bandwidth required by practical
implementations is then incorporated rnro a sysrem
degradation factor.
A,=ro**,,[ffi*rJ-"
6.4 RADIO SYSTEM DESIGN 325
Notice that the SNR term in Equation 6.31 refers to signal-power-to-noise-power ra-
tios and not EblNO. The relationship betwen SNR and E6lNs is
provided in Appendix C.
The system
gain, in conjunction with antenna
gains and path losses, determines the
fade margin:
Fade margin
=
A, + 6r + GR + 20 log,n fu
-Af -
A0
=
transmitter antenna
gain (dB)
=
receive antenna
gain (dB)
=
tran$mitted wavelength
=
antenna feeder and branching loss
(dB)
(6.32)
=
free*space attenuation
(distance d must be in the same units as l,)
=
20logro
(4nd)
The directivity, and consequently the
gain of, an antenna is directly
proportional to the
size of its aperture and inversely
proportional to the square of the transmitted wave-
length. In determining the receive power, however, it is actually only the area of the
antenna that is important and not the directivity or
gain. Thus radio system designers
conveniently consider transmit and receive antenna
gains as contributing to signal
power but include a wavelength normalization
(20log1s
l,) to relate the gain of the re-
ceiving antenna back to the size ofits aperture.
In addition to
providing increased antenna
gain, greater directivities also reduce
multipath
problems. The longer secondary paths, which arise from
greater emanation
angles, have lower
power levels when the directivity
is increased. Unfortunately,
an-
tenna gains are limited by
practical considerations in several regards: Economically
sized towers can support only limited-sized antennas, mechanical alignment
is diffi-
cult, and directional stability of both the antenna and the path is limited.
The feeder and branching losses 41 included in Equation 6.32 arise because single
antenna systems typically carry several channels for separate radios. Furthermore, re-
liability considerations usually dictate that spare transmitters and receivers be avail-
able for protection switching. The process of combining signals for transmission or
distributing them after reception inherently introduces various amounts of attenuation
or splitting of signal power.
Example 6.4. Determine the system
gain of a l0-Mbps, z-GHz digital microwave
repeater using 4-PSK modulation and an output
power of 2.5 W. Assume the excess
bandwidth of the receiver is 307o and ttrat other departures from ideal performance
amount to 3 dB degradation. Assume a noise figure of 7 dB for the receiver, and the
desired maximum error rate is 104. Also determine the fade margin assuming antenna
gains
of 30 dB each and a path length of 50 km. The branching and coupling losses
are 5 dB.
Solution.
From Figure 6.17, the required value of E6lN0 for 4-PSK modulation can
be determined as 10.7 dB. Using Equation 3.42
(in
Appendix C), it can be determined
where 6i1
Gq
L
Af
Ao
326 DIGITAL MoDULATIoN AND RADIo SYSTEMS
that the signal-power-to-noise-power
ratio at the detector is 3 dB higher than E6lr/6.
Thus, the required SNR is 13.7 dB.
since 4-PSK modulation provides
2 bps/rlz, the signaling rate is 5 MHz, which is
the theoretical minimum (Nyquist)
bandwidth. Equation 6.31 can now be used to de-
termine the system gain:
1 3 . 7 * 7 - 3 * 1 0 l o s l . 3
: l 1 6 d B
At a carrier frequency of 2 GHz, the wavelength is 3 x 108/2x lOe
=
0. l5 m. Thus the
fade margin can be determined from Equation 6.31:
Fade margin
=
I I 6 + 60 + 20 log,o(0. l5)
-
S
-
Z0 log,n(4ru5 x lOa)
=
38.5 dB
6.4.3 Frequency Diverslty
As mentioned previously,
neither the transmitting and receiving equipment nor the
path
is normally reliable enough to provide
an acceptable level of system availability.
Frequency
diversity is one means of providing
backup facilities to overcome both
types of outages. A deep multipath-induced
fade occurs when a signal from a $econ-
dary path arrives out of phase with respect to the primary
signal. Since the phase shift
produced
by a path
is proportional
to frequency, when one carrier fades, it is un-
likely that another carrier fades simultaneously.
Frequency diversity involves the
use of a spare transmitter and receiver operating in a normally unused channel. Since
separate hardware is used, frequency diversity also provides protection
against hard-
ware failures.
The simplest means of implementing frequency diversity is to use one-for-one
(l
: l)
protection
switching as indicated in Figure 6.36. one-for-one protection
switching
implies that one spare channel is provided
for each assigned message channel. when
high-spectrum
efficiency is required, it is generally necessary to have only one spare
for a group
of N channels (
I ; N protection
swirching). In fact, the FCC has stipulated
that, in some frequency bands, a system must be implementable in one-for-Nconfigu-
rations. The main impact of a I ; Nprotection system is the complexity of the switch-
ing unit and the need to switch back to the assigned channel in the first available hop
so that a single spare channel can be reused repeatedly on a long roule.
Frequency diversity normally does nothing to alleviate rain outages since all
channels in a particular
frequency band are simultaneously affected. when rain is
a pafticular problem,
it can be overcome only by using higher transmit powers or
shorter hops.
)
6.4 RADIo SYSTEM DESIGN 327
-'Wrt
Figure 6.36 Protection switching with one-for-one frequency diversity.
6.4.4 Space Diverslty
Space diversity is implemented, as shown in Figure 6.37, by vertically separating two
receive antennas on a single tower. The resulting difference in the two
paths is nor-
mally sufficient to provide independent fading at the two antenna$. Space diversity is
the most expensive means of improving the availability, particularly if separate receiv-
ers for multiple channels are used for each antenna.
The cost can be minimized, how-
ever, by combining the two received signals in a
phase-coherent manner for input to
a common receiver
t451.
This technique
provides less hardware
backup than when
completely
separate receivers are used for each antenna.
6.4.5 Angle Diverslty
Because multipath fading is produced by multiple incident rays alTiving at slightly dif-
ferent angles,
protection from fading can be achieved by discriminating
on the angle
Figure 6.37 hotection swirching with space diversity receivers.
328 DIGITAL MoDULATIoN AND RADIo SYSTEM$
of arrival' Angle diversity utilizes two side-by-side receiving antennas with slightly
different angular elevations to provide
the discrimination.
Although angle diversity is
generally
not as effective as space diversity, it can improve the availability ofdigital
signals
[48,
49] in applications where the tower requirements
of space diversity are
impractical or disallowed.
6.4.6 Adaptlve Equallzation
Since multipath effects are frequency dependent, not all frequencies in a particular
channel simultaneously
experience the same amount of fading. Thus multipath fading
can produce
not only a
general
attenuation ofthe received signal but also the equiva-
lent of in-band amplitude (and phase)
distortion.
Distortion of the spectrum amplitude
produces
a general
degradation in the error performance
over and above the
fenalty
incurred by the general
attenuation of the signal. In wideband digital radios, fre-
quency-selective
fading (as
opposed to flat fading) has proved
to be the dominant
source of multipath outage$. Fortunately, better performance
can be achieved if the
spectrum amplitude is equalized (adjusted
to a uniform level).
Since atmospheric-induced
multipath interferena varies with time, removal of mul-
tipath-induced
amplitude distortion requires adaptive equalization.
A common ap-
proach
to adaptive amplitude equalieation merely samples the energy at selected
frequencies in the received signal spectrum. A compensating filter shape is then in-
serted into the signal path
to adjust all energy samples to a common level. This basic
technique is used in digital radios manufactured by Bell Northern
[45]
and Rockwell
International
[50].
Reference
[50]
reports a 7-dB improvement
in effective fade mar-
gin
can be achieved with an adaptive equalizer.
Because channel distortions produce
intersymbol interference
in the received base-
brurd signal, time-domain equalization with adaptive transversal filters or decision
feedback equalizers is also used. Reference
[51]
contrasts the relative merits of both
approaches and provides
an extensive list of references. The use of high-density
modulation formats such as 256-QAM places
very stringent requirements
on the adap-
tive equalizers so these radios commonly use both types of equalizers
[47].
Frequency-selective
fading, which comrpts only a
portion
of a wide-bandwidth
channel, is particularly
difficult to equalize because it changes rapidly and requires
complicated equalizer structures (in
the frequency domain or the time domain). The
problem
can be greatly
simplified if, instead of using a single carrier, multiple carriers
(i.e.'
MCM) within the allotted bandwidth are used to create mulriple subchannels. With
this approach, fading within any subchannel is relatively flat and easy to equalize. The dis-
advantage of this approach is the cost of replicated circuiry for each subchannel. As re-
ported
in
[5?],
however, atmospheric-induced
outages are greatly
reduced.
6.4.7 Route Design
The layout of a point-to-point
microwave relay system involves numerous considerations
of the local terrain, prevailing
atmospheric conditions, radio fiequency interference with
other symbols, and interference from one hop to another within a single system.
REFERENCES
329
Path Length
The foremost consideration
in setting up a single hop of a microwave
system is that
line-of-sight
transmission is required. Using antenna heights of about 60 m, the cur-
vature of the earth limits the transmission distance to approximately
50 km. Longer
line-of-sight distances are possible if taller antenna towers are used. However,
plscti-
cal considerations of mechanical stability limit the height of a tower,
particularly since
longer distances imply larger and more directive antennas.
When antennas are rigidly mounted on the top of a building or the side of a moun-
tain, mechanical stability may not be a problem but the $tability of the
path itself may
become a limiting factor. Under some atmospheric conditions, radio waves can be re-
fracted to the point that a narrow beam completely
misses the receiving antenna.
Bending of the
propagation path can also cause a f,rxed obstacle to intermittently ob-
struct transmission,
implying that clearance between
the notmal
path
and nearby
physical obstructions is required. Thus there is always a practical limit on how narrow
the beam can be.
Local teruain also influences the
prevalence of multipath
fading. Nearby bodies of
water contribute
significantly to multipath conditions during late evening or early
morning hours when there is litfle wind. Direct transmission ovor water is usually very
difficult because of reflecfions off the water surface. Transmission over water often
requires higher antennas and some means of blocking direct reflections.
Cochannel
lnbrterence
Since there are a limited number of channels
available for point-to-point microwave
systems, the same channelS must be used Over and over again. ReuSe of microwave
fiequency bands is enhanced by the directivity of the antennas and the
general need
for line-of-sight
reception. In many metropolitan areas, however, there is so much traf-
fic converging into one particular area that it is impossible to completely isolate two
systems using the same channel. This type of interference
is refened to as cochannel
interference.
Cochannel
interf'erence results from converging
routes or from overreach of one
hop into another hop of the same system reusing a channel. Sometimes reflections or
atmospheric refractions can contribute to overreach,
even when direct line of sight is
not
present. Overreach
is usually controlled by zigzagging
the hops so that a beam
from one tran$mitter misses all subsequent
receive antennas in a route. The reduced
use of
point-to-point microwave
for long-distance ftaffic has reduced the cochannel
intetference
in point-to-point systems.
However, cochannel interference
is a major
consideration
in the design and deployment
of digital cellular systems, as discussed in
Chapter 9.
REFERENCES
I Federal Communications
Commission, Report and Order, Docket No. 19311' FCC
7 4-985, released SePt. 27, 197 4.
2
..Special
Issue: AR6A Radio System," Bell system Tethnical Joumal,
Dec. 1983.
330 DIGITAL MODULATION AND RADIO SYSTEMS
W. A, Conner, R. J. Girvin, and W. R. McClellan,
,.Frequency
Control of the MSR_6
single sideband system," IEEE International conference on communicarions. 19g3,
pp,
E2. 5. 1-E2. 5. 5.
Y' Yoshida, Y. Kitahara, and s.
yokoyama,
"6G-90
Mbps Digital Radio system with
I6-QAM Modulation," IEEE International
conference on comrnunicafiozs,
19g0, pp,
52.4.r-52.4.5.
s. Pasupathy,
"Minimum
shift Keying: A spectrally Efficient Modulation,"
/EEE
Comrnunications
Magaline, July 1g79, pp. l4-ZZ.
c.-E. sundberg,
"continuous
phase
Modulation," IEEE communications Magafine,
Apr. 1986, pp.25-38,
Y. Morihiro,
s. Nakajima, and N. Furuya,
"A
100 Mbir/s hotorype MSK Modem for
satellite communications," IEEE Transactions on communications,
oct, 1979, pp.
l 512- 1518.
"hactical
GMSK Data Transmission,"
MX-coM Application
Note, rggg
(
www.mxcom.com).
K. Murota and K. Hirade,
"GMSK
Modulation for Digital Mobile Radio Telephony,"
IEEE Transactions
on Communications,
July lg8l.
R. E. Ziemer and w. H. Tranter,
principles
of communications, Houghton Mifflin,
Boston, 1976.
E. Panayirci and E, Y. Bar-Ness,
"A
New Approach for Evaluating
the
performance
of
a symbol Timing Recovery system Employing
a General Type of Nonlin Earity
,,'
IEEE
Transailions
on Communicarrans.
Jan. 1996.
c' R. Hogge, Jr.,
'carrier
and clock Recovery for
g-psK
synchronous Demodulation,"
IEEE Transactions
on Communications,
May 1979, pp. 529_533.
P' R. Hartmann and J. A. crosset,
"A
90 MBS Digital Transmission system at I I GHz
Using 8 PSK Modulation," IEEE International conference on communications.lg16.
pp.
l 8- 8- 18- 13.
I. Horikawa, T. Murase, and
y.
saito,
"Design
and
performances
of 200 Mbivs 16
QAM
Digital Radio system,* IEEE Transactions
on computers, Dec. 197g, pp,
r953-1958.
J. A' crosser and P. R. Hartmann,
"64-eAM
Digital ttadio Transmission sysrem
Integration and Performance,"
IEEE International comrnunications
conference.lgg4.
pp. 636-641.
w. A' conner,
"Direct
RF Modulation 256eAM Microwave system,"lEEE
Globemm
P roteedings, 1988, pp. 52.7.1,52.7.6.
H. Matsue, T. shiraro, and K. watanabe,
"256
eAM
400 Mb/s Microwave Radio
system with DSP Fading countermeasures,"
IEEE Intemational
conference on
Communications,
June 1988, pp. 41.5. l-41.5.6.
K. Pahlavan and J. L. Holsinger,
"A
Model for the Effects of
pcM
compandors on the
Petformance
of High speed Modems," IEEE Grobecom
proceediigs,
19g5, pp.
28.8.1*28.8.5.
T. Ryu' J. uchibori, and
y. yoshida,
'A
stepped square 256
eAM
for Digitar Radio
system," IEEE Intemational
conference on communicarions, 19g6, pp. 1477-l4gl.
l 0
l 1
l 2
l 3
t 4
l 5
r 6
t 1
t 8
l 9
20
7.1
2?
HEFERENCES
331
M. Tahara, T. Deguchi, S. Mizoguchi, and Y' Yoshida,
"6
GHz 140 Mbps Digital Radio
System with 256-SSQAM Modulation,"
IEEE Globecom
Proceedings, 1987'
pp'
38.3.1-38.3.6.
S. A. Rhodes,
"Performance
of Offset-QPSK
Communications
with Partially Coherent
Detection." National Telecommunications
Conference
Remrd, Nov. 1973'
pp'
32A-l-32A-6.
M, K. Simon and J. G. Smith,
"Offset
Quadranrre
Communications
with
Decision-Feedback
Carrier Synchronization,"
IEEE Transadions on Communftations,
Oct. 1974,
pp. 1576-1584.
23 J. W. Bayless, R, D. Pedersen, and J. C. Bellamy,
"High
Density Digital Data
Transmission," National Telecommunications
Conference
Record, 1976,
pp.
51. 3- l - 51. 3- 6.
S. A. Oronemeyer and A. L. McBride,
"MSK
and Offset
QPSK
Modulatton,"
IEEE
Transactions on Communicafians,
Aug, 1976' pp. 809-820.
R. D. Gitlin, S. K. Rao, J. Werner, and N. Zervos,
"Method
andApparahrs forWideband
Transmission of Digifal Signals Between, for Example, a Telephone Central Office
and
Customer
Premises," U.S. Patent No.4,9U'492' May 8' 1990.
B. Saltzberg,
"comparison
of single-carrier
and Multitone Digital Modulation
for
ADSL Applications," IEEE Communication't
Magazine, Nov' 1998,
pp. l14-122'
,,Mid-range
Physical Layer Specification for Calegory 3 Unshielded Twisted-Pair"'
afohy-0018.000,
ATM Forum, Sept. 1994'
c. w, Anderson and s. G. Barber,
"Modulation
considerations for a 9l Mbit/s Digital
Radio," IEEE Transactions
on Communications,
May 1978,
pp. 523-528'
D. P. Taylor and M. Shafr,
.,Fade
Margin flnd outage computation of 49-QPSR
Radio
Employing Decision Feedback Equalization,"
IEEE Intemational
Conferente on
Communications,
1983, pp. 1453- 1458'
A. J. Viterbi,
"Error
Bounds for Convolutional
Codes and an Asymptotically
Optimum
Decoding Algorithm," IEEE Transactions
on lffirmation Theory, Apr' 1967'
pp'
260-269.
3l G. Ungerboeck,
"Trellis-Coded
Modulation
with Redundant Signal Sets Fart
Introduction."
IEEE Communicatktns
Magaeirre, Feb. 1987' pp. 5-l l'
G. Ungerboeck,
"Trellis-coded
Modulation with Redundant signal sets Part II: state
of the Art," IEEE Communications
Magarine, Feb. 1987' pp. t2-21 .
R. J. F. Fang,
"A
Coded 8-PsK system for 140-Mbivs Information
Rate Transmission
over 80-MHz Nonlinear Transponders," Proceedings
7th Intennational Conference
on
Digital Satellite Communications,
1986, pp' t2-16'
T. Kamitake, K. Uehara, M. Abe, and S. Kawamura,
'"Trellis
Coding l4'4kb/s Data
Modem Implemented with a single-chip High-speed
Digital signal hocessor,"
/EEE
Globemm Proce edings, I 987'
pp. I 2.8. I
-
I 2.8.6'
.,A
Modem operating at Data signaling Rates up to 28,800 bps for use on the General
swirched Telephone Network and
on Leased Point-to-Point Z-wire Telephone-Type
Circuits," ITU-T Rec. V.34, Sept, 1994.
Ramin sadr and J. K. omura,
"Generalized
Minimum shift-Keying
Modulation
Techniques," IEEETransactions
of Communications,
Jan. 1988,
pp. 32-40'
25
30
33
332 DIGITAL MoDULATIoN AND RADIo SYSTEMS
37 J. A' C' Bingham,
'-Multicartier
Modulation for Data Transmission: An Idea Whose
Time Has Come," IEEE Comtnunitations Magazine,May 1990, pp. 5_14,
38 J. M. cioffi,
"Asymmetric
Digita-l subscriber Lines," in The cornmunications
Handbook, J. Gibson, Ed., CRC
press,
Boca Raron, FL, 199g, pp. 450_479.
39 Digital Vidco Broadcasting (DVB):
Framing stucture, channel coding and modulation
for digital terresrrial television, EN 300 744, vl.Z.l, ETSI. lggg_01.
40 B. Saltzberg,
"Comparison
of Single-Canier
and Multitone Digita.l Modulation for
ADSL Applications," IEEE Comrnunications
l4agaline,Nov. lg9g, pp. llL-llL.
41 E. D. sunde,
"Ideal
Binary
pulse
Transmission by AM and FM," Eel/ sys temTechnical
Joumal, Nov. 1959, pp. 1357-1426.
42 w. R. Bennet and J. R. Davey, Data Transmission,
McGraw-Hill. New
york.
1965.
43 R. w. Lucky, J. salz, and E. J. weldon, Jr.,
principles
of Data communications,
McGraw-Hill. New
york.
lg68.
44 Federal Communications
Commission, Memorandum
Opinion and Order modifying
FCC Repon and Order of Docket 193 I l, released January Zg,1975.
45 G. H. M. de witte,
"DRS-B:
system Design of a Long Haul 9l Mb/s Digital Radio,"
I E E E N ational Te lec ommunicat io n s C onfe renc e, I
g7 g,
pp.
3
g,
l l
_3 g.
1 .6.
46 H. Kostal, D. R. Jeske, and v. K.
prabu,
"Advanced
Engineering Methods for Digital
Radio Route Design," IEEE Intemational
communitations conference, l9g7, pp.
l 9B.6.l _198,6,6.
47 c' P. Bates, P. L. Penney, and K. L, seastrand,
"Digital
Radio Technology in the AT&T
Network," IEEE Intemationar
communications
conference, 19g7, pp,
l 98. 5. l
- 198. 5. 7.
48 E. A. Sweedyk and P. Balaban,
"A
Multipath Fading Model for Tenesrrial Microwave
Radio," IEEE Glohecom
proceedings,
lgB8, pp. 52.6.1_52.6.1
.
49 s. H. Lin,
"opening
New vistas for Microwave Radio,"
gel/co
re Exchang,e,July/Aug.
1988, pp. 9-12.
50 P. R. Hartmann and E. W. Allen,
"An
Adaptive Equalizer for Correction of Multipath
Distortion in a 90 MB/s B-psK system," IEEE Intemational
conference on
Communications,
1979,pp, 5.6. l-5.6.4.
5l J. K. chamberlain, F. M. clayton, H. sari, and
p.
vandamme,
,.Receiver
Techniques
for Microwave Digital
Radio," IEEE cownunitations Magazine, Nov. 19g6, pp.
43-54.
52 Y, Nakamura, H. ohtsuka, s. Aikawa, and H. Takanashi,
'Advanced
rechniques for
super Multi-carrier
Digital Microwave Radio with Trellis-coded
2s6
eAM
Modulation," IEEE Globecom
proceedings,lggg,
pp.
l l.3.l_l1.3.6.
PROBLEMS
6.1 To prevent
the transmission of line spectra, digital radio terminals use data
scramblers to randomize
the data patterns.
Furthermore,
differential encoding is
normally required for proper
data detection. Both functions inffoduce error mul
-
tiplication. If the combined effect of these operations causes an average of five
6.2
6.3
6,4
6.s
PROBLEMS 333
decoded enors for every channel enor, what is the effective
penalty in transmit
power for a 4-PSK system at BER
=
10-6? At BER
=
10-3?
If a digital radio receiver is experiencing
thermal-noise-induced errors at a rate
of I
per 106 bits, what is the new error rate ifthe path length is decreased from
30 to ?5 miles?
Derive Equation 6.18.
what is the minimum
theoretical bandwidth of an 8-PSK signal carrying 4800
bps?
A 32-QAM signal set is implemented by eliminating the four corner
points of a
36-QAM signal. what is the minimum error distance in terms of the
peak
signal
power? How does this answer compale
to the error distance for antipodal sig-
naling?
(Express the answers in decibels')
what is the
peak-to-average ratio of a 32-QAM signal
(Problem
6.5) assuming
transmission of random data?
How does 32-QAM
(Problem 6.5) eruor
performance compare to 32-PsK enor
performance?
(a)
In terms of average signal
powers?
(b) In terms of
peak
signal
powers?
A carrier transmission
system using 4-PSK modulation
provides an error rate of
l0-6. tf the modulation is changed to 16-PSK to naffow the transmission band-
width
(data rate unchanged),
how much must the hansmit
power be increased to
maintain the same error rate?
Repeat Problem 6.8 assuming that the bandwidth
is unchanged and the data rate
is doubled.
6.10 What is the error rate of an ideal I6-QAM signal with a signal-to-Gaussian-noise
ratio
(SNR)
of 18 dB?
6.6
6.7
6.8
6.9
6.11
6.r2
6.13
6.14
What is the error rate of hoblem 6.10 if interference is also
present at a level
that is 21 dB below the received signal?
(Assume the effect of the intefference
is identical to Gaussian noise at the same rms power.)
How much must the transmit
power in Problem 6.1 I be increased to off'set the
effect of ttre interference?
(Assume
the interference level is fixed.)
Repeat Problem 6.1?, but assume that the interference
increases in proportion to
the increase in transmit
power. (Signal powers in the adjacent channels
are in-
creased along with the desired signal
power.)
Repeat Problem 6.3 assuming a starting state of
D-
N ETWORK SYNCH
RON IZATION
CONTROL
AND MANAGEMENT
In Chapters 4 and 6 some synchronization
requirements of transmission
systems are
discussed.
These requirements
involve carier recovery for coherent detection of
modulated signals, clock recovery for sampling incoming data, and framing proce-
dures for identifying individual channels in a TDM signal format. All of these consid-
eratrons are inherent in digital transmission
systems and, for the most part, operate
independently of other equipment in a network. An instance of one subsystem's
de-
pendency on another was noted for Tl lines. Tl source data mu$t include minimum
densities of 1's to maintain timing on the original transmission link. In contrast, other
line codes are described
that maintain clock synchronization
independently of the
source data.
This chapter discusses network-related
synchronization
considerations
for in-
terconnecting various digital transmission
and switching equipment.
Foremost
among these considerations is synchronization
of switching equipment' When
digital switching equipment was first installed in the public telephone network, the
interfaces were analog transmission systems. Hence, each switching machine
could operate with an autonomous frequency source
(clock) that converted all
voice signals into digital signals with precisely the same data rate
(nominally 64
kbps). These switching matrices were designed to carry one channel rate and one
channel rate only. The advent of subsequent digital switch interconnection
re-
quired switches to carry digital channels
originating someplace else in the net-
work-from
a different frequency
source. Thus, network synchronization
requirements arose.
When individual synchronous transmission
and switching equipments
are inter-
connected to form a network, certarn
procedures need to be established that either syn-
chronize the clocks to each other or provide for their interoperability
when each
subsystem uses independent clocks. Following the discussion of network clock syn-
chronization, the concept of synchronization
is extended to other aspects of network
control.
336
7. 1
NETWORK SYNCHRONIZATION CONTROL AND MANAGEMENT
TIMING
All digital systems inherently require a frequency source, or
..clock,"
as a means of
timing internal and external operations. operations timed from a single frequency
source do not require particularly
stable sources since all commonly clocked elements
experience timing variations in common. A different situation occurs when transfers
are made from one synchronous equipment to another (as
from a ffan$mitter to a re-
ceiver). Even if the clock of the receiving terminal is
"synchronized"
to the transmit-
ting terminal on a long-term or average basis, short-term variations in either clock may
jeopardize
the integrity ofthe data transfer. Thus it is generally
necessary to use fre-
quency
source$ (oscillators)
in both the transmitter and the receiver that are as stable
as is economically feasible.
7.1.1 Tlming Recovery: Fhaee-Locked
Loop
A common means of synchronizing a receiver clock to a transmitter clock uses a
phase-locked
loop (PLL)
as shown in Figure 7. l. A phase
detector continuously meas-
ures the phase
difference between the incoming clock and a locally generated
clock.
The phase
detector in Figure 7.1 merely measures the difference in the zero crossings
between the two signals. When the zero crossing of the line clock precedes
the zero
crossing of the local clock, a positive
voltage is generated;
otherwise, a negative volt-
age is produced.
The output of the phase
detector is filtered to eliminate as much re-
ceive noise as possible,
and then the phase
measurement
adjusts the frequency
of the
voltage-controlled
oscillator (VCo)
to reduce the phase
difference.
Some amount of
noise or interference inevitably passes
through the phase
detector and the filter, caus-
ing erroneous adjustments in the vco frequency. As time passes,
however, a fre-
quency
offset produces
ever-increasing phase shifts. When the phase
difference builds
up, it is easier to detect, and the appropriate changes in the VCo occur. Hence the local
Figure 7.1 Phase-locked
loop clock recovery circuir.
rMrNG 337
clock maintains the desired average frequency but inherenfly produces $hort-term
vari-
ations ofphase and frequency as it
"hunts"
the underlying frequency ofthe line clock.
The line clock in Figure 7.1 is shown to have a transition in every clock interval, a
situation that does occur with Manchester or diphase-type
line codes. With some line
codes
(bipolar
or AMI in particular) there are no clock transitions during intervals of
binary 0's. In these cases either the clock extraction circuitry inserts artificial transi-
tions exffapolated
from
previous intervals or the
phase detector
is disabled during in-
tervals when no pulse is detected.
7.1.2 Cl ock Instabi l l tY
The variation in the output frequency of the VCO described above is an example of
clock instability. All clocks have a certain amount of instability-even
free-running
oscillators.
An important aspect of clock instability is its frequency:
the rate at which
the clock frequency changes from being too high to being too low. The frequency of
the instability can be directly observed as the frequency spectrum of the VCO control
voltage in a PLL clock recovery circuit. When the VCO control voltage varies slowly,
the variations are referred to as clock wander. When the variations are more rapid, the
instability of the clock is refened to as
jitter.
The most generally accepted dividing
point between wander and
jitter
is 10 Hz. Pure wander on a Tl line therefore
produces
positive phase errors for more than 77,200 bit intervals
(0.05 sec) followed by more
than 77,200 bit intervals of only negative
pha$e errors.* The main sources of clock in-
stability
(both
wander and
jitter)
in a network are;
1. Noise and interference
2. Changes in the length of transmission
media
3. Changes in velocity of propagation
4. Doppler shifts from mobile terminals
5. Irregular timing information
Nolse and lnErterence
If the low-pass filter of a PLL is designed with a very low cutoff frequency, it can filter
out almost all of the noise and interference on a transmission
link that would otherwise
comrpt the timing recovery. There are three main reasons why arbimarily
low pass fil-
ters cannot be used. First, the ability ofthe PLL to acquire synchronization
(e.9.,
lock
on to the underlying
clock) is inversely related to the PLL bandwidth' If the VCO be-
gins oscillating at the wrong frequency and the bandwidth
is too narrow' the PLL may
never
pull the oscillator to the frequency of the line clock. Sometimes, this
problem is
accommodated
by using two bandwidths: a wide one to acquire synchronization and
a narrow one that is selected after lock is achieved'
+As
discussed later, observing wander in this way requires extremely stable clocks as a reference. The
normal PLL clock recovery circuitry cannot be used to observe wandet because the VCO tracks the
relatively long-term
phase offsets, Thus the clock recovery citcuit does not filter out thc wanderbut
passes
it on.
338 NETWoRKSYNcHHoNI ZATI oNcoNTHoLANDMANAGEMENT
A second consideration that generally precrudes
the use ofvery nanow filters, even
after acquisition, is that the source may vary in frequency that cannot be tracked by a
slowly responding
PLL. In this case, the recovered clock does not track the ideal sam-
ple
times leading to high error rates or, worse yet,
the
pLL
loses synchronization
al-
together and has to reacquire lock.
The third limiting factor for low-bandwidth
PLLs is the instability of rhe VCO ir-
self. If the vco begins to drifr in frequency,
very low bandwidth filters preclude
ad-
justing
the VCO input voltage soon enough to
prevent
bit errors or
possible
loss of
synchronization.
Because operational
considerations dictate certain minimum PLL bandwidths.
noise and interference on the transmission link always cause the recovered clock to be
more impaired than the source clock. The
pLL
does, however, eliminate that portion
of a disturbance with frequency content above the bandwidth of the PLL. Thus distur-
bances with low-frequency
content are the most difficult to deal with. systematic
jit-
ter, as
produced
by particular pattems
of intersymbol
interference, can have an
arbitrarily low frequency content and is discussed more fully in a later section.
An important consideration in the design of a digitrrl transmission link is the accu-
mulation ofjitter in tandem clock recovery circuits. If a recovered clock is used to time
the transmission of outgoing data, as in a regenerative
repeater, some amount of in-
coming
jitter
is imbedded into the outgoing clock. The clock recovery circuit in the
next receiver tracks its incoming
clock but introduces even more
jitter
due to noise and
interference on the second
section. Thus
jitter
accumulates at every regenerative
re-
peater
using its received line clock as its transmit clock. If there is a large number of
regenerative
repeaters, thejitter can accumulate
to a point where subsequent clock re-
covery circuits have difficulty tracking the receive clock, produce
sampling error$,
and possibly
lose lock.
Changee ln Length of Transmiesion
Media
Path length changes occur as a result of thermal expansion
or conffaction of guided
transmission
media or of atmospheric bending of a radio path.
while a path
is increas-
ing in length, the effective bit rate at the receiver is reduced because more and more
bits are being
"stored"
in the medium. similarly, as the path
shortens, the bit rate at
the receiver increa$e$ because the number of bits stored in the transmission link is de-
creasing. After the path
has stabilized, the receive signal returns to the nominal data
rate. The mo$t significant changes in path
length occur with communications
satel-
lites' Geostationary satellites produce path length variations of approximately
?fi) miles,
which cause propagation
time variations of approximately
I msec
I
I
I
.
path
length changes
also occur in guided
transmission media such as copper wire and optical fibers. These
changes are referred to as diurnal changes because they occur once a day.
Changes in the Vetocity of
propagation
Temperature
changes not only cause expansion and contraction of wireline transmis*
sion media but also can change those propagation
constants of the media that deter-
7.1 rMrNG 339
mine the velocity of
propagation. The resulting change in received clocked stability,
however, is much less than that
produced by the change in path length
[2]'
The propagation velocity of radio waves in the atmosphere also changes with tem-
perature and humidity. Although these velocity changes are more significant than
those occurring in wirelines, they are still smaller than the path-length-induced vari-
ations. Notice that a change in
propagation velocity is effectively equivalent to a
change in
path length since the number of bits
"stored"
in the transmission
path is
changed.
Doppler Shlfts
The most significant source of
potential timing instability in a received clock occurs
as a result of Doppler shifts from airplanes or satellites. For example, a Doppler shift
induced by a 350-mph airplane amounts to an equivalent clock instability of 5 x 10*7'
Digital mobile telephone receivers must accommodate
Doppler shifts equivalent to
clock instabilities of about one part in 107.* Again, Doppler shifts occur, in es$ence,
as a result of path changes.
I neg u lar Tl m ing lnfo rmetlo n
As discussed in Chapter 4, a fundamental
requirement of a digital line code is that it
provide sufficient timing information to establish and maintain a receiver line clock.
If the timing information is data dependent,
jitter
in the recovered clock increases dur-
ing periods of relatively low density timing marks. The magnitude of the
jitter
is de*
pendent not solely on the density of timing marks but also on the timing
(data)
patterns. In an ideal repeater, only the density would matter. In
practice, however,
various imperfections
lead to
pattern-dependent
jitter
[3].
As discussed later in this chapter, higher level digital multiplexers insert overhead
bits into a composite data stream for various
purpo$es. When the higher rate data
stream is demultiplexed,
the arrival rate of data within individual channels
is irregular'
This inegularity
produces timing
jitter
when
generating new line clocks for the lower
rate signals. This source ofjitter
(waiting
time
jitter)
is often the most troublesome and
is discussed in more detail later.
7.1.3 Elastic Stores
The timing instabilities described in the
preceding paragraphs essentially represent
changes in the number of bits stored in a transmission link. In the case of noise- and
interference-inducedjitter, the change in
"bits
stored" occurs because data are sam-
pled a little earlier or a little later than nominal. Since the outgoing data of a regenera-
tive repeater are transmitted according to the recovered clock, a
phase offset in the
clock means the delay through the repeater is different from when there is nomisalign-
ment in timing.
*If
you *e
pa*noid about cooperation between cellular operators and law enforcement authorities, you may
not want to use your cell
phone while speeding.
340 NETwoHKSYNcHHoNIZATIoNcoNTRoLANDMANAGEMENT
Ifphase offsets in successive regenerative repeaters coincide, a net change of sev-
eral bits of storage in a long repeated transmission link occurs. Since these exffa bits
enter or leave the hansmission link over relatively short periods
of time, the accumulated
jitter
rnay represent a relatively large, but short lived, instability in the receive clock.
Because regenerative repeater$ use incoming sample clocks as output clocks, sus-
tained timing differences between inputs and outputs do not exist. The endpoints of a
transmission link, however, may interface to a local clock. In this case a difference be-
tween a received and a relatively fixed local clock must be reconciled with an elastic
store' An elastic store is
a data buffer that is written into by one clock and read from
by another. If short-term instabilities
exist in either clock, the elastic store absorbs the
differences in the amount of data transmitted and the amount of datareceived. An elas-
tic store can compensate only for short-term instabilities that produce
a Iimited differ-
ence in the amounts of data transmitted and received. If sustained clock offsets exist.
as with highly accurate but unsynchronized
clocks, an elastic store will eventually un-
derflow or overflow.
TDld-Swltch ln|errtace
A typical need for an elastic store occurs when a digital transmission link is interfaced
to a digital time division switch. As shown in Figure 7.2,theelastic store is placed
be-
tween the incoming digital transmission link and the inlet side of the switch. In most
instances the digital switch provides
timing for all outgoing TDM links so thar no tim-
ing discrepancies exist between these links and the swirch. For the time being, assume
that the far end of the digital link derives its clock from the receive signal and uses
that clock to time digital transmissions returning to the switch. This is the situation that
arises when a remote channel bank is connected to a digital swirch through Tl ]ines
and is commonly referred to as channel bank
"loop
timing." when loop timing is used,
the line clock on the incoming link of the switch is nominally
synchronized to the
switch clock. However, for rea$ons discussed previously,
a certain amount of insta-
bility in the incoming clock necessarily
exists. The elastic store absorbs these insta-
bilities so that purely
synchronized data are available for the swirch.
Figure 7.2 Interface between TDM nansmission link and a digital switch using an elastic
store.
TrMrNc 941
In essence, the loop formed by the transmission
links and the elastic store maintains
a constant and integral number of clock intervals between the inlet and outlet of the
switch. Thus, from a timing point of view, the inlets and outlets operate as though di-
rectly connected to each other using a common source of timing.
Removal of Accumulated Jitter
Another application
for an elastic store is shown in Figure 7.3, where it is used to re-
move transmission-induced
timing
jitter
in a regenerahve repeater. Normally, a regen-
erative repeater establishes the transmit timing directly from the locally derived
sample clock. In Figure 7.3, however, the transmit timing is defined by a separate local
clock. The elastic store absorbs the shoft-term instabilities in the receive clock, but the
long-term frequency Of the transmit clock is controlled by maintaining a cefrain
"av-
erage level of storage" in the elastic store. Thus the transmit clock is synchronized to
the line clock on a long-term basis, but not on a short-term basis. Ifthe elastic rltore is
large enough to accommodate all transient variations in the data rate, high-frequency
instability of the input clock is removed.
All regenerative repeaters, regardless of the mechanism used to recover timing, de-
rive their output clocks by averaging the incoming timing information over a period
of time. Tuned circuits average the incoming clock for relatively few signal intervals;
phase-locked loops do so for many intervals. In all cases a certain amount of storage
or delay is implied. An elastic store is merely a mechanism to increase the available
delay so that output timing adjustments can be made more
gradually. Notice that an
elastic store always inserts significant amount$ of artificial delay into the data
path'
To remove arbitrarily low frequency instabilities
(wander), aftihalily large elastic
stores
(and arbitrarily stable VCOs) are required. Thus a
jitter-removing elastic store
should be inserted into achain of repeaters only when
jitter
accumulation
threatens the
ability of a regular regenerative repeater to maintain synchronization'
Jitter is not
just
a
phenomenon of a digital transmission link but also occurs in digi-
tal storage $y$tems. For example,
jitter
removal like that shown in Figure 7.3 is used
in laser disc
ptayers to eliminate
blurring of the visual images
[4].
Elastlc Store ImPlementdtiong
The required size of an elastic store varies from a few bits to several hundred bits for
high-speed, long-distance communications
links. Figure 7.4 shows one means of im-
plementing a small elastic store utilizing a series-to-parallel
converter, a register, and
Figure 7.3 Jitter-removing
regenerative repeater.
342 NETWoRK
sYNcHRoNIzATIoN coNTRoL AND MANAGEMENT
Input rignrl
Ortput clock
OrrtDUt drtr
Figure 7.4 Basic implementation of an elastic store.
a parallel-to-series
converter. As indicated, incoming data are transferred into the reg-
ister as soon as each word is accumulated in the series-to-parallel convefier. Some
time later, data in the register are transferred to the output parallel-to-serierr
converter
as a complete word is shifted out. Notice that transfers to the parallel-to-series
con-
verler are independent
of the incoming clock. As long as ouFut transfbrs occur be-
tween input transfers, no data are lost, and short-term
jitter
is absorbed by varying
delays through the elastic store.
Normally, some control circuitry (not
shown) is needed to initialize the elastic store
so that the first transfer into the register occurs midway between output transfers. This
process
means some incoming data are initially discarded by the series-to-parallel
reg-
ister until the desired transfer time occurs.
The relative times of the parallel
transfers into and out of the holding register pro-
vide a direct indication
ofthe relative phase
ofthe input and output clocks. Thus the
parallel
transfer clocks contain the information needed to generate
VCO control volt-
ages if the elastic store is being used to remove accumulated transmission
jitter.
The basic structure shown
in Figure 7.4 can be extended to implement larger elastic
stores, as shown in Figure 7.5. The only change involves the substitution of a f,rrst-in,
first-out (FIFO)
buffer for the holding register of Figure 7.4. This data buffer i$ de-
signed specifically to allow input transfers under the control of one clock while out-
put$
are controlled by a different clock. Normally, the FIFO buffer is initialized by
inhibiting output transfers until it is half firll. In fact, some commercially availabll
FIFO buffers have an output signal specifically indicating when it is half full or
greater.
7.1.4 Jitter Measurements
A simple circuit for measuring timing
jitter
is shown in Figure 7.6. As indicated, it is
nothing more than a phase-locked
loop (pLL)
with the output of the phase
comparator
providing
the measurement of the timing
jitter.
Normally, the bandwidth of the low-
pass
filter (LPF)
is very small so the vco ignores short-term jiner
in the timing signal.
If there is no
jitter
at all, the output of the phase
comparator is constant and no signal
is passed
by the high-pass selection filter (HpF).
Input
clock
7.1 TrMrNc 343
Inrutfita
Output clock
Output
(htr
Figure 7.5 FIFO implementation of an elastic store.
Very low frequency
jitter
ciurnot be measured by the circuit in Figure 7.6 because
the VCO tracks slowly changing
phase shifts. Low-frequency
jitter
may be of no con-
cern, however, because it can be hacked by a PLL. Higher frequency
jitter,
on the
other hand, is more apt to cause sampling errors or a loss of lock in the clock recovery
of a repeater. Thus the spectral content of the
jitter
as well as its magnitude is of inter-
est. Besides not being able to measure low-frequency
jitter,
the circuit of Figure 7.6
cannot operate if it cannot lock onto the fundamental clock signal, Section7.4.7
de-
scribes another way to measure
jitter
and other timing impairments within a network.
Phase
jitter
is commonly specified by communications
theorists as a power meas-
urement in units of radians squared or cycles squared.
(One cycle
=
Zru radians') As
indicated in Figure 7.7, phase
jitterPoweris then ameasure of the variance of the num-
ber of clock cycles or unit intervals
(UIs)
stored in the transmission
link. In a
physical
sense
jitter
"power"
has little meaning
because it represents timing variations, not
power. Some
physical
justification for expressing
jitter
as a power can be obtained by
observing that the Ims power o| of the
phase detector output signal is propottional to
the rms phase
jitter
ofr:
2
at
Jitt0f
Clock rlgnal
Irput
clalt
Phi|e lock loop
Figure 7.6 Circuit tbr measuring timing
jitter'
344 NETWORK SYNCHRONIZATION CONTROL AND MANAGEMENT
Prohbility dkfibution of nuftbGr
of ryrtolr in trrnrlt
lntunbncou| numblr
=
/V
Ar,6rgp
-
ilo
Trsn|ftrhtfi *ith
it|bh riming surE+
Figure 7.7 Phase
jitter
modeled as the
transmission link.
Rtcdrnr
variance in the number of symbols
',stored"
in the
ot,=K3si
ry2)
( 7. 1)
where Ka is the phase
detector gain
factor in volts per radian.
Example 7.1. Given an rms phase jitter
of 10.7 dB relative ro one uI, what is the
standard deviation of the phase offset?
Solutinn. The variance of the signal phase
is determined as ofr
:
l0 exp(10.7/10)
=
11.76 UI squared. Hence, the standard deviation is (ll.?6)u2 =
j.4j
UI
(symbol
intervals)' Since 687o of a normal probability
distribution lies within one standard
deviation, the phase
of this signal is within
+3.43
symbol intervals for 687o of the time.
One percent
of the time the signal phase
will be outside 2.6 standard deviations. or
+8.9
symbol intervals.
If phase jitter
arises as a result of additive Gaussian noise on a stable signal, the
phase noise can be approximated as
4:#
(rad2)
(7.2)
where ofr
=
additive noise power
Pr
=
signal power
Equation 7.2 is the basic equation ofphasejitter produced
by additive noise on a
continuous sinusoid
[5].
when timing is exhacted from a data signal, the timing in-
formation is usually not continuous. The distinction is not important because
jitter
produced
by additive noise is normally insignificant compared to other sources
[6].
For an analysis of
phase jitter
produced
by regenerative repeaters operating on ran-
domly occurring timing transitions, see reference
[7].
TrMrNc 3rt5
TABLE 7.1 Maxlmum lnstablllty of D$l Customer lnterlace
Frequency Band Maximum Peak-to-Peak Variation
< 1 0 H 2
< 10 Hz
10 Hz-40 kHz
I kHz-40 kHz
28 Uls in 24 hr
5 Ul s i n 15 mi n
5 Ul s
0, 1 Ul s
As an example of a particular instability specification,
Table 7.I lists the wander
and
jitter
specifications of a DSI digital carrier
(Tl)
interface to the
public network
[8].
Notice that the higher frequency
jitter
specification allows deviations in
pulse cen-
ters of only 5Vo, which effectively specifies the accuracy of the data sample clock'
Larger variations are allowed for lower frequency instabilities because the clock re-
covery circuits can track the changes to maintain a good sample clock.
7.1.5 $ystematl cJi tter
An original analysis of
jitter
in a chain of digital regenerators was reported by Byrne'
Karafin, and Robinson
[3].
Figure 7.8 shows the basic model of their analysis' Each
of the regenerative repeaters in a T-carrier line extracts timing from the received wave-
form and
passes that timing on to the next regenerator as a transmit clock. Because of
implementation imperfections
(primarily intersymbol intetference) in the timing re-
covery circuits,
jitter produced by repeaters is dependent on the data
patterns' One
worst-case
pattern produces an exheme phase lag. Another
pattem produces an ex-
treme phase lead. When the datfl
pattern shifts ftom one worst case to the other, a phase
ramp occurs. Because every repeaterha$ the same basic implemenfadon, the
jitterpro-
duced by individual repeaters tends to be coherent. The systematic nature ofthisjitter
makes it the most signiflrcant source of accumulated line clock
jitter
at the end of a
chain of repeaters.
ptt$7nt
Figure 7.8 Model of systematic
jitter
in a string of regenerative repeaters;
(D2
=
phase shift
produced by worst-case data pattem for phase lead;
(Dt
= phase shift produced by worst-case
data pattem for phase lag.
I
I
Phuo dritt
$rilih rr
Stvitch
indlvlilirl
ln drtr
ttlpetif
7
346 NETWoRKSYNCHRoNIZATIoNcoNTRoLANDMANAGEMENT
As indicated in Figure 7.8, the last repeater in the chain experiences a large phase
ramp equal to the number of repeaters times the phase shift of each individual repeater.
This phase
ramp represents an abrupt change in the clock frequency that may cause
bit errors or a complete loss of synchronization. Thus there is a limit to the number of
repeaters that can be used without
jitter
removal. For more analyses of
jitter
accumu-
lation including the combined effects of systematic
jitter
and rzurdom perturbations,
see references
[9J, [0],
and
[11].
7.2 TIMING INACCURACIES
In the preceding
section the nature of certain instabilities or transient variations in tim-
ing was discussed. Although these variations represent shifts in the frequency of a line
clock, the shifts are only temporary and can be absorbed by elastic stores. In some in-
stances digital communications equipment using autonomous frequency sources must
be interconnected. when this happens, the clock rates of the two systems are never ex-
actly the $ame, no matter how much accuracy is designed into the frequency $ources.
An offset in the two clocks, no matter how small, canaot be reconciled by elastic stores
alone.
In the preceding
section, channel bank loop timing was mentioned as an example
of how remote terminals are synchronized to a digital switch. When the remote termi-
nal is another digital switch using its own frequency source as a reference, a difTerent
situation results. As shown in Figure 7.9, the outgoing clock for each direction of
transmission is def,rned by the local switch clock. Thus the incoming clock at each
switch interface contains not only tran$mission-line-induced jitter
but also a small and
unavoidable frequency offset.
7.2.1 Slips
As indicated in Figure 7.9, the interface of each incoming digital link necessarily con-
tains an elastic store to remove transmission link timing
jitter.
The elastic store at the
first digital switch is written into by the recovered line clock but read
from
the local
rate R1. If the average rate of the recovered line clocklR2 is different fromRl, the elastic
Figure 7.9 Connections between autonomously timed digital switches.
7,2 TIMING INACCURACIFS 347
store will eventually underflow or overflow, depending on which rate is larger. When
R2 is greater than R1, the elastic store at the first digital switch overflows, causing a
loss of data. If R2 is less than R1, the same elastic store underflows, causing extraneous
data to be insefied into the bit sffeam entering the switch. Normally, the extraneous
data are a repetition of data bits already transferred into the switch. Disruptions in the
data stream caused by underflows or overflows of an elastic store are referred to as
"slips."
Uncontrolled slips represent very significant impairments to a digital network be-
cause they generally cause a loss of frame synchronization. Thetefore, slips are al-
lowed to occur only in prescribed manners that do not upset framing. One
general
approach to controlling the slips is to ensure that they occur only in the form of a repe-
tition or deletion of an entire frame. Thus the time slot counter$ and framing logic as-
sociated with the multiplex group remain synchronized. Conholled slips comprising
entire frames can be assured by using elastic stores with at least one frame of storage.
As a slip occurs, the storage level in the elastic store is effectively increased or de-
creased by a full frame. Rather than actually inserting or deleting a frame of informa-
tion, the desired effect is achieved more easily by indexing address pointers in a
random-acce$$ memory. Such a system is shown in Figure 7.10.
The elastic store in Figure 7.10 operates by sequentially writing input information
into memory addresses corresponding to individual TDM channels. Data for individ-
ual output channels are obtained by reading the same addresses in the same sequential
manner. Ideally, if there is no offset between the clock rates, the read times of each
channel occur midway between write times for corresponding channels. The elastic
store then has the capability of absorbing transmission delay variations up to one-half
of a frame time.
Memory
wrltS*
Memory
reodr
Double
reEd
Figure 7.10 Hlastic store operation with a one-frame memory,
348 NETWORK SYNCHRONIZATION CONTROL AND MANAGEMENT
The timing diagram in Figure 7.10 depicts an exaggerared timing offset in which the
switch clock R2 is greater than the incoming clock lt1. As indicated, the read times carch
up gradually
with the write times until a
"double
read" occurs. At that time the information
retrieved for each charurel is a repetition of the information retrieved for the previous out-
going frame. Although write and read times for only one channel are shown, the corre-
sponding times for all other channels have the same relationship. Thus all charurels slip
together. Notice that R1 is greater
than R2, a slip occurs when a
"double
write" on all
channels causes the information in the previous incoming frame to be overwritten.
The elastic store operation depicted in Figure 7.10 is very similar to the operation
of a time slot interchange memory described in Chapter 5. This relationship is exploit-
able in a TST switch where the inlet memory can provide
both the elastic store func-
tion and the time switching function. When the two functions are combined, slips
generally
occur at different times for different channels. Nevertheless, individual
charnels maintain proper frame alignment since each channel is transferred through
the inlet memory using dedicated memory addresses.
One attractive feature of using the inlet memory as an elastic store is that, when set-
ting up a new connection, an internal switching time slot can be chosen so that the inlet
memory read is halfway between inlet memory writes for the particular
channel. Thus
a slip in that connection will not occur for a long time, probably not until long after
the connection is released. (With a clock inaccuracy of one pafl
in 108, ttre time be-
tween slips in any one channel is 3.5 hr.)
One potential problem
with the elastic store in Figure 7.10 occurs when write and
read times nearly coincide. When both accesses to a single channel occur one after the
other, transient timing instabilities can cause the two accesses to cross back and forth
with respect to each other. Thus slips caused by double reads may follow slips caused
by double writes and vice,versa. To remedy this situation, some amount of hysteresis
is needed in the counter adjustment process. The hysteresis, in turn, implies that addi-
tional storage is needed to defer the occurrence of one type of slip after a slip of the
other type has recently occurred.
One means of implementing an elastic store with the desired hysteresis is to use two
frames of storage as shown in Figure 7.11. Forconvenience, the elastic store is divided
into an A-frame memory and a B-frame memory. The counter logic again accesses the
memories in sequential fashion except that frames are written alternately into the A
and B memories. Under normal operation, the memories are accessed in the same way
for output data. When a slip is imminent, however, control logic causes the output
channel counter to be reset so that the A memory is read twice in a row. This situation
is depicted in the timing diagram of Figure 7.1I, which again assumes that R2 is
Sreater
than R1. The imporrant point
to be noticed in the timing diagram is that after
the counter adjustment produces a double read of memory A, the write and read times
of each individual memory are approximately one frame time apart. Thus another ad-
justment
can be deferred until the write and read accesses again drift one full frame
time with respect to each other. The structure and mode of operation shown in Figure
7.11 describe the elastic store used for DSI signal interfaces to AT&T's (now
Lu-
cent' s) No. a ESS
II2].
7. 2 TI MI NGI NACCURACI ES
349
ilhmory
r$lt||
l,/hmofy
rgrd
A
Double
t?ral
Figure 7.ll Elastic store with a two-frame memory.
9llp Bate Qbjactlves
If the difference between an elastic store's input data rate and its output data rate is
M,- the time between slips is
where N is the number of bits that
get dropped or repeated whenever a slip occurs.
Normally, a slip involves a full frame of data, in which case the time between slips
is determined as
where AF is the difference in frame rates.
As long as slips are controlled so they do not disrupt framing,l their only effect is
an infrequent repetition or deletion of the information within affected TDM channels.
The audible effect of slips on a digitized voice signal is an occasional
"click."
Only
one slip in 25 produces
an audible click in PCM voice
[].
Voice signals can therefore
tolerate several slips per minute
[13].
*Typicatly,
clock offsets are specified in relative terms (e,g,, onepart in 106). A clock tlat is accurate to P
parts per million (ppm) has a maximum offset of AR
=
ll
.
P/10".
TWith
single-frame slips, framing is disrupted in thc sense that framing
pattem is shifted, Thus, the frame
sequence has to be reacquired, but this process is simplified by the fact that the location of the framing bits
and the integrity of the message channels are maintained.
A7={
AR
47=-l -
^F
n, $
r
I
awntr
I
l \
H
A
VP
fi 2
J I
I
A
B
J }
P/S
A
350 N ETWORK SYNCHRONIZATION CONTHOL AND MANAGEMENT
A more subtle and troublesome aspect of slips occurs when a digitized channel car-
ries voiceband data. High-speed data modems for the analog telephone network use
QAM
modulation with coherent detection in the receiver. Since these modems are par-
ticularly sensitive to phase
shifts, they are particularly
vulnerable to slips. An 8-bit slip
in a digitized modem signal using a carrier of 1800 Hz causes an instantaneous phase
shift of 8l
o.
obviously, a phase shift of this size causes a data error, but more impor-
tantly, it upsets the carrier recovery circuitry in the receiver and causes multiple errors.
A single slip can upset the operation of some voiceband modems for several seconds
[ 14,
15] .
characterizations of the effects of slips on Group 3 facsimile equipment
[16]
reveal
that a single slip can cause the loss of four to eight scan lines without an error report.
Sometimes the loss of the lines is not immediately evident to the reconstructed image.
Diagonal lines, however, readily reveal missing vertical space.
Encrypted traffrc
(voice
or data) is more susceptible to slips since the encryption-
decryption process
usually relies on bit-synchronous scramblers and unscramblers.
When the bit count is altered by insertion or deletion of bits in a time slot, counters in
the source and destination become unsynchronized. At berrt, the decryption process
causes every slip to be audible. At worst, unintelligible speech or data result until the
unscrambler is resynchronized. Another important aspect of encrypted communica-
tion is transmission of encryption keys or indexes to encryption keys. If synchroniza-
tion is lost for some reason, it may be necessary to resynchronize the encryption keys,
thereby compromising security.
When a digital transmission link is being used to transmit data directly, the effect
of a slip may not be any more significant than a single channel error. Most data com-
munications receiving equipment requests a complete retransmission of any block of
data not satisfying certain redundancy checks. Thus one error is as bad as many errors
or a complete loss of data. The effect of the slip will be more significant, however, if
the communications protocol
[e.g.,
the DDCMP of Digital Equipment corporation
(compaq)l
relies on byte count procedures
to delimit message blocks. Insefiion or de-
letion of data by the network causes the receive counter to become unsynchronized,
and the normal exchange of information is disrupted until the loss of synchronization
is recognized.
From the foregoing considerations for data transmission, the slip rate objective for
the AT&T network and adopted by Bellcore and ANSI for North America was ser ar
one slip in 5 hr for an end-to-end connection
[17-19].
Since slips can occur at multiple
points within a network, the objective for slips at individual trunk and switching in-
terfaces was $et at one slip every 20 hr.
Example 7.2. Determine the relative accuracy requirement$ of two independent
clocks to maintain a mutual slip rate objective of one slip in 20 hr. Assume a frame
rate of I kHz as in PCM voice sisnals.
solution. The slip rate objective implies that the frame rate produced
by one clock
can be different than the frame rate produced
by the other clock by no more than
7.2 TIMING INACCURACIES 351
I
*
=
r; * * *
=
1.39 x 10-5 slips per second
Since ilrere are 8000 frames
per second, the relative accuracy is determined as
1. 39x l 0- 5
" n. - ,
ffi
=
1.7 x lo-e sliPs/frame
Hence the clocks must be accurate to 1.7 parts in lOe.
Because Example 7.2 determines only a maximum relative inaccuracy, the abso-
lute inaccuracy of each individual clock must be less than
( I .7 x 10-e)/2, or 0.85
parts
in 10e.
7.2.2 Asynchronou$ Mul tl pl exi ng
In the
preceding section certain aspects of network synchronization were discussed
that implied the need for clock synchronization to prevent a loss of data by way of
slips. In this section a procedure referred to as
"pulse
stuffing" is discussed that avoids
both slips and clock synchronization. The term
pulse stuffing can be somewhat mis-
Ieading since it implies that
pulses are inserted into the line code to make timing ad-
justments.
Actually, pulse stuffing involves only the data stream and is independent
of the line code or modulation system in use. Pulse stuffing is a term commonly used
in North America while the $ame concept is refened to as
"justification"
in Europe.-
The basic concept ofpulse stuffing involves the use of an output channel whose rate
is purposely higher than the input rate. Thus the output channel can carry all input data
plus some variable number of
"null,"
or
"stuff,"
bits. The null bits are not part of the
incoming data. They are inserted, in a prescribed manner, to pad the input data skeam
to the higher output rate. Naturally, the extraneous null, or stuff, bits must be identi-
fiable so that
"destuffing"
can recover the original data stream.
The practice of pulse stuffing arose when the initial digital TDM hierarchy
(Table
1.10) was defined. At this time there were only isolated digital transmission links
within the network that precluded
synchronizing them to a common clock. When it
came time to combine lower rate tributaries
(e.g., DS ls) into higher level signals
(e.g.,
DS2s or DS3s), the multiplexing
procedure necessarily had to accommodate tributar-
ies operating at $lightly different rates. The generic
term for combining these unsyn-
chronized signals is
asynchronous multiplexing. In this context,
"asynchronous"
refers to multiplexing ofunsynchronized tributaries into a higher level signal
(using
pulse
stuffing). It doerr not refer, in any way, to a means of transmission. The higher
level signal is always carried on a synchronous transmission link.
As the amount of digital equipment in the network grew and more and more of it
became interconnected, both the means and the necessity for a different form of mul-
*Justification
is the
pnnting industry practice that aligns the right sirle oflines oftext by inserting a variable
amount of space within the line. As you
will see in the following discussion, aligning individual tributaries
to thc rate of a higher level multiplexer is conceptually the same process,
352 NETWoRKSyNcHRoNTzATToNcoNTRoLANDMANAGEMENT
tiplexing arose. The chosen approach is refened to as synchronous multiplexing
(SONET)
in North America and Synchronous Digital Hierarchy (SDH)
in the rest of
the world. The principles of SONET and SDH are described in conjunction with fiber
systems in the next chapter.
P ulse-Stuff i ng Concepts
As a starting point for understanding the need for pulse
stuffing, consider the simple
two-channel, bit-interleaved multiplexer in Figure 7.12. As indicated, within any
string of even-numbered bits in the multiplexer ouryut, the number of bits carried in
each subchannel is necessarily identical. Thus the rates of the subchannels are also
identical. If the two input clata streams are running at different rates, the output can be
synchronized to one ofthe channels but not both. Thus slips would necessarily occur
in at least one of the tributaries.
As a simplified example of pulse
stuffing, Figure 7.13 shows a two-channel, bit-
interleaved format as before but with the additional detail needed to allow adjustments
of the information flow within each subchannel. As indicated, the multiplexed output
is formatted into 10-bit master frames with 5 bits assigned to each subchannel. The
frrst 3 bits in each subchannel of each master foame always carry data from the respec-
tive tributaries. The fourth bit in each subchannel
(C1
and Cj specify whether the last
bits (S1
and Sj carry data or are stuffbits. When C1 is a
"1,"
a bit is stuffed; otherwise
s
1
carries tributary data. Hence each ma$ter frame can carry 3 or 4 bits from each tribu-
tary. Il on average! each tributary sends 3.5 bits during a ma$ter frame, variations of
+l4Vo
in the tributary clock rates can be accommodated.
An important point to notice about an asynchronous multiplexer is
that
the output
frame structure is unrelated to the frame skucture of the lower level inputs. As far as
the higher level multiplexer is concemed, each inprrt signal is merely a serial bit stream
with no
particular
structure assumed. Framing bits in the lower level multiplex signals
are transmitted right along with the information bits. After the higher level signal is
demultiplexed and the tributaries are unstuffed, framing of the lower level signals
must be established for further demultiplexing.
Although pulse stuffing can be implemented with a variety of higher level framing
formats, the generally most desirable features of a pulse-stuffing
format are identified
as follows:
1. The use of fixed-length master frames with each channel allowed to stuff or not
to stuff a single bit in the master frame
ffi
Figure 7.12 Two-channel multiplexer showing equal output data rates for each input,
7.2 TIMING INAccURACIES 353
I Frame
c E D B A
Figure 7.13
Simplifiedpulse-stuff,rng example.
Redundant $tuffing specifi cations
Noninformation bits distributed across a master frame
Timing offsets are
generally quite small, so only small adjustments from a single oc-
casional stuff bit are required. Thus large numbers of tributary bits can be combined
into a master frame with one specific bit position identified as the S bit. Nominally,
approximately one-half of the master frarnes contain the maximum number of infor-
mation bits NM, and the other half contain Ny
-
l information bits.
The purpose of a pulse-stuffing operation is to
prevent a loss of data when two in-
terconnected digital transmission links are unsynchronized with respect to each other.
If single bit errors can cause a stuffbit to be interpreted as information
(or
vice versa),
the basic objective is lost. Furthermore, notice that an erroneous interpretation of a
stuff code causes embedded, lower level multiplex signals to lose framing. For these
reasons, the interpretation of C bits must be encoded redundantly. Under the a$rtump-
tion that channel errors are random, the probability of misinterpreting a stuff code
i s
-
p)'
where p
=
probability of a channel error
n
=
number of correctable stuff code errors
(2n
+ I bits in a stuff code)
Information bits should be distributed across a master frame for several reasons.
First, by sepamting these bits as much as possible, enors in redundant $tuffing speci-
fication bits
(C bits) are more likely to be independent. If the specification bits are too
close together and burst errors are prevalent, the redundant encoding is of little use.
Second, by distributing noninformation bits, the irregularity of information flow is
minimized. When higher level multiplex signals are demultiplexed, a clock for each
individual lower level signal needs to be derived from the irregular information rate
in each channel. Generation of a suitably stable clock synchronized to the information
rate is simplif,red if information bursts or gaps are minimized. Furthermore, elastic
stores needed to smooth out the information rates are smaller when the length of in-
formafion gaps is minimized.
2.
3.
Po:rfti,';,-','
354 NETWORK SYNCHRONIZATION CONTROL AND MANAGEMENT
M12 Multiplexer
An example of a higher level multiplexing format is provided
in Figure 7.14. This is
the format used for 6.3l2-Mbps DsZ signals in the North American digital hierarchy.*
A DSz signal is derived by bit interleaving four DSI signals and adding the appropri-
ate overhead bits.
A DS2 master frame is 1176 bits long. Of these there are 1148 information bits
(287
per channel), I I framing bits
(I%,
Mr, Fo, Fr), l2 stuffing conrrol bits (Cr,
Cz, C:, C+),
4 S bits (Sr, Sz, Sr, S+), and an alarm bit X. Since an S bit can be a null bit or an infor-
mation bit, each channel can send 287 or 288 bits in a master frame. An S bit is
designated as an information bit if all three of the corresponding C bits are 0. The S
bit is a null
(stuff)
bit if all three corresponding c bits are 1. obviously, this encoding
procedure
allows for single error correction in the stuffing control bits.
The first level of framing is established by the alternating F6, F1, F0, . . .
pattern.
Notice that exactly 146 bits separate the Fs and Fl bits. Another level of framing for
identifying the C and S bits is established by the Ms and M1 bits. A fourth M bit (X)
is not used for framing and therefore can be used as an alarm service digit. Similar
frame structures exist for other higher level digital signals. Figures 7.15-7.18 show
the structures for DS3, DS4, DSIC, and the E2 second-level digital signal of the ITU,
respectively.
Example 7.3. Determine the minimum and maximum input channel rates
accommodated by an Ml2 multiplexer. Also determine the rate of DSI misframes
caused by an erroneous interpretation of a stuffed bit. Assume the bit error rate is 10-6.
Solution. The maximum information rate per
channel is determined as
6.31?:?8s
=
1.5458 Mbps
l t 76
The minimum information rate per
channel is determined as
6.312x287
ffi6
=
1'5404 MbPs
Since there are three possible combinations of two errors in the C bits, the probability
of misinterpreting an S bit is closely approximated by 3 x (10-6)2 =3
x l0-r2. The du-
ration of each master frame is 117616.312
=
186
Fsec.
Thus the rate of misframes per
DSI si enal i s
3 x 10-12
"#r":0'016x
10{ misframes per
second
which is equivalent to one misframe every 2 years.
'DS2
signals are no longer transmitted as individual signals. They only exist as an intermediate level
between 28 DSls and a DS3 sienal.
+--'-
Zgt-Bit1;btilffi
aE
/tB
i l C- * r - r O
7.2 TIMING INAcCURACIES 355
Erulf
bitr
Figure 7.1C Frame format of DSZ digital signal. Stuffing occurs in channel i when the
previous Ci bits
=
lll; X is an alarm bit that equals I for no-alarm condition. Framing is
established by the F6F1F6 . . . sequence with 146 intervening bits.
Example 7.3 demonstrates that the tolerance of a 1.544-MHz DSI clock is
-3.572,
to +1.796 kHz. Thus the relative accuracy between the DSI and DS2 clocks must be
I.79611544, or only I part in 860. This relatively large timing tolerance is much
greater than what is required for reasonable clock and line instabilities. The timing ad-
justment
capabilities and unsymmetric tolerance range were chosen out of a desire to
(l)
minimize DS2 reframe times,
(2) provide a line clock that is a multiple of 8 kHz,
and
(3) minimize waiting time
jitter
[20].
C8O blt ilbfrfile
Figure 7.15 Frame format of DS3 digital signal; P is even parity over all message bits in the
previous master frame. Stuffing occurs in channel i when the previous Ci bits
=
lll, The X
(alarm) bits and the P bits must be l l or 00 so the MoMrM6 sequence can identify the end of
the master frame.
f
!
F
l
E
T
E
-
C1
s
E
t
l l
$
E
s
356 NETWoRKSYNCHRoNIZATIoNcoNTRoLANDMANAGEMENT
Figure 7.16 Frame format of DS4: Ci bits
=
I l l implies stuff the eighth message bit position
for channel i following the last ci; Pr is even parity over the 192 previous
odd-numbered
message bits; P0 is even parity over the 192 previous even-numbered message bits.
Elastlc Slore Size Requlrements
A functional model of an Ml ? multiplexer is shown in Figure 7.19. Associated with
each lower level (DSl) input is an elastic store to hold incoming data until it is trans-
ferred to the higher level
(DS2)
output. The elastic stores serve two
purposes:
to re-
move the arrival
jitter
of the incoming data and to hold data for the proper time slots.
In addition to generating framing, the control logic of the multiplexer monitors the
318 bir NUbfrEmG
rtro-i-c,-33-ro cr
q
Figure 7.17 Frame format of DSIC digital signal. Stuffrng occurs in channel i when the
previous
Ci bits
=
lll; X is an alarm bit that equals I for no-alarm condition. Framing is
established by the FoFrFo. . . sequence with 158 intervening bits.
E
E
tl
o
E
g
o
ti
{
5
({
tl
o
6
q
E
:E
Mr
Ml
Stuft bit
7.2 TrMrNc rNAccuRActEs 357
21?Ei tsbl m
Fflri digffi$t rign|l
200ilmgr blt
irBtff
tffir-
8{8 bltr
Figure 7.18 Frame format of second-level digital signal of ITU-T
(E2),
Stuffing occurs in
channel i when the previous Ci bits are lll.
$torage level
(which
serves a$ a phase comparator) of each elastic store and initiates a
stuffing operation whenever the elastic store is less than half full. Conversely, no stuff-
ins occurs when the elastic store is more than half full.
Since stuffing can occur at only certain times and only at a certain maximum rate,
the elastic tltore must be at least as large as the peak jitter (peak phase offset) of the
incoming signal. As discussed at the beginning of this chapter,
jitter
accumulates
along the entire length of a repeatered transmission link. Thus longer links require
larger elastic stores if slips are to be prevented.
As an example of the relationship between line length
(number
of repeaters) and
number of bits of elastic storage needed by an M12 multiplexer, refer to Figure 7;20.
This figure was submitted by AT&T to the CCITT special study group on
jitter
[21].
The analysis is an extension of the systematic
jitter
analysis of reference
[3].
DS-l
inputs
DS-Z outout
Sdl6ctor codtrol
framing inrrtion
dnd rtuffing
Storag6
l#el
Eign|lr
Framing gonaration
Etuffing contfol
Figure 7.19 Functional diagram of a Ml2 multiplexer
358 NETWoRKSyNcHRoNtzATtoNcoNTRoLANDMANAGEMENT
e5 3.0
Jltur rlopq
(kHr)
Figure 7.20 Maximum number of regenerative repeaters as a function of elastic store size and
jitter
slope; ,4
=
number of cells in elastic store.
The abscissa of Figure 7.20 is the maximum-phase slope
produced
by a clock re-
covery circuit in the presence
of a worst-case shift in data patterns (worst-case
system-
atic
jitter).
Since phase
slope is nothing more than frequency offset, the required
elastic store size can be determined as the maximum phase
slope times its maximum
duration. Since the total phase
slope is proportional
to the number ofrepeaters, Fig-
ure 7.?0 displays the maximum number of repeaters per
storage cell in an elastic
store versu$ the
jitter
slope of an individual repeater when making worst-case tim-
ing transition.
As an example, f,ust-generation Tl repeater$ produce
a worst-ca.se slope of 2.4 ktlz.
Ml2 multiplexers allocate 5 bits of storage to input phase jitter (3
more bits are included
for implementation ease and waiting time
jitter).
From Figure 7 .20 it can be seen that
the ratio of N-* to A is 56, which implies that N*"*
=
56 x 5
=
280 repeaters.*
*The
performance
of long Tl lines is not as much of a concem as it once was because most long-distance
DSl circuits ate now embedded in frber links, which have much greater repeater spacing,
{
E
e
li
o
o
o
o
E
fl.
o
E
o
cl
o
E
7.2 TrMrNc rNAccuRAcrES 359
7.2.3 Waitlng Time Jitter
When demultiplexing a higher level TDM data stream, it is necessary to generate a
clock for each derived subchannel.
Because the subchannels are transfened
(or
trans-
mitted) as a synchronous data stream, the derived clock must be continuous. Deriva-
tion of subchannel clocks is complicated by the insertion into TDM data streams of
overhead bits that create gaps in the bit anival times. Inegularity in the data anival
rate caused by these
gaps is referred to as waiting time
iitter.
Most of the overhead bits
(e.g.,
framing bits,
parity bits, stuffing control C bits) oc-
cur on a regular and
predictable basis. The waiting time
jitter
caused by these
gaps
(sometimes referred to as mappingjitter+) can be eliminated easily with an elastic store
and an output clock derived from the incoming line clock. For example, a single PCM
channel clock at 64 kbps can be derived from a 1.544-Mbps Tl line clock by multi-
plying by I and dividing by 193. Mapping
jitter
as it occurs in a EI signal mapped into
an E2 signal is depicted in Figure 7 .21 . In this figure the phase of the tributary data
mapped into the higher level signal is shown relative to an unmapped
(continuous)
data clock at the same average rate. Notice that the phase of the tributary falls behind
the reference during
periods of data gaps but catches up during data fields because the
average tributary data rate during these fields is slightly higher than the reference. No-
tice further that the fill level of an elastic store does not get perfectly reconciled at the
end of a single E2 master ftame. This result occurs because, on average, a nonintegral
number of bits of a tributary are calried in a master frame.
In contrast to mapping data rate variations, waiting time
jitter
produced by
pulse
stuffing is significantly more difficult to deal with. The difficulty arises because wait-
ing times
produced by stuffed
pulses
are irregular and unpredictable.
For this reason
the subchannel output clocks derived from a pulse-stuffed TDM line must be derived
independently and only from the average arrival rafe of each channel's data-not
from
the higher level TDM rate!
Output clocks from M12 demultiplexers are
generated using
jitter-removing
elastic
stores, as shown in Figure 7.22.If larye elastic stores and very slowly adjusted output
clocks are used, most of the
jitter
can be removed. Unfortunately, waiting time
jitter
has frequency components down to zero frequency so the
jitter (e'g',
wander) can
never be eliminated entirely. However, the
jiner
can be conf,tned to as low a band of
frequencies as desired by using a large enough elastic store. Figure 7.23 depicts time
interval errors resulting from the pulse stuffing-destuffing
process. The key
point to
notice is a full bit of offset between the source data clock and the mapped data can oc-
cur at any time but an adjustment can only be made when the next stuff opportunity
occurs-hence the term waiting time
jiner.
Waiting time
jitter
is basically a function of how often
pulses
are stuffed, but it is
also dependent on the ratio of actual stuffs to stuffing opportunities.
If the input clock
"Some
references include mapping
jitter
to include waiting time
jitter.
Here, mapping
jitter
is used to
represent repetitive data rate variations that occuf when both the tributary and the higher level transport
signals are respectively at their
pre,cise rates. Thus, waiting time
jitter
only occurs as a result of frequency
adjustments,
360 NETWoHKSyNcHRoNtzATtoNcoNTRoLANDMANAcEMENT
Phffi of tn'hrrsry m cmpositc rig*t rcl8tivc b tqFt EigFrl iD Ulg
X: otrcrtad lr&
Frfio
I; lfrffilrim bit pilidm fn I rrfrltry sr Tffig a{iahflr (.||d)
blr
poridn
'
Ewy_@lpilh AM c*df, Zoj hfffi|timtir p*
ffirieyftr wDoilftb rd,fdwlirftd|thf, til
!H
hhEyin m S bit poiria
strpaftrg o r!.d.
Ftgure 7.21 Mapping Jitter of a CCITT El in a CCITT E2 sisnal.
is
jitter
free, the output
jitter
peaks
when one-half of the opporhrnities are used. From
the point of view of maximum tolerance for clock offsets, a stuffing ratio of
{
is ideal.
To reduce the waiting time
jitter,
however, stuffing ratios of approximately
{'are
often
used. For a thorough analysis of waiting time
jitter,
see reference
t2zl.
As an example
of waiting time
jitter
dependence on stuffing (iustification)
ratios, see Figure 7 .24 ob-
tained from reference
[21].
The abscissa ofFigure 7.24 represents the ratio of stuffs
to opporhrnities
while the ordinate is
jitter
power produced
by a single pulse-stuffing
process.
The
jitter
power is expressed in decibels relative to one slot squared (a
slot is an older term for a unit interval, ul). curve A shows the output
jitter
pro-
Figure 7.22 Functional diagram of Ml2 demultiplexer/desynchronizer,
thrcshold
7.3 NETWORK SYNCHRONIZATION 361
Tlms iileffsl
rrorE AdJurtuent
H
lnput clock phase
Figure 7.23 Time interval errors produced by adjustment waiting times.
(dBl
0dB
=
t sl otr
- 1 2
cclTT.4773A
0.4 05 0.6
Juttification ratio
0.7 0.s
0.9 1.0
Tri butsry
j i tter
-
0.10 tl ot r.m.r.
TributEry
iitter
-
0.25 rlot r.m.r
Figure 7.24 Waiting time
jiner
dependence on
justification
ratio.
duced when the input
jitter
is
-20
dB
(0.
t UI rms). Curve B shows the output
jitter
when the input
jitter
is
-72
dE
(0.25 UI rms). Figure 7.24 showsjitter produced
by a single multiplexer. From measured data
122],
a good order-of-magnitude es-
timate of the waiting time
jitter
accumulated by N, tandem M12 stuffing-destuff-
ing operations is
E
o
E
* r E
.;
5

E
h
B
-20
o
o
5
o
A:
BI
ol
4=ffiruti'
(7.6)
7.3 NETWORK SYNCHRONIZATION
As discussed in the preceding section, whenever a digital transmission link is con-
nected to a digital switch, it is desirable to synchroniee the two systems by having the
, A
'1
\ -h
J E
7
362 NETWoRKSYNCHHoNIZATIoNcoNTHoLANDMANAGEMENT
transmission link obtain its timing from the switch. An obvious exception to this mode
of operation occurs when a digital kansmission link is connected to a digital switch
on both ends. Generally, a hansmission link in an all-digital network derives its timing
from
just
one of the switches to which it is connected. If the other switch is not syn-
chronized to the first in some manner, an unsynchronized interface necessarily results.
This section is concemed with network synchronization as a whole, not simply the
synchronization of a single interface. Basically, network synchronization involves
synchronizing the switches of the network. The transmission links can then be syn-
chronized automatically by deriving timing directly from a switching node.
There are two basic reasons for paying
assiduous attention to the timing require-
ments of a digital network. First, the network must prevent
uncontrolled slips that
could produce
misframes, inadvertent disconnects, and cross connects. It is generally
very difficult or very expensive to prevent
slips altogether. Thus a second aspect of a
network timing plan requires establishing a manimum rate of controlled slips as pafi
of the end-to-end circuit quality objectives.
Synchronizing private networks is sometimes difficult because the network topolo-
gies
are not designed with network synchronization in mind, and the switching equip-
ment (PBXs)
me not designed to provide
synchronization to other nodes.
Furthermore, the private
networks often interface to multiple carders in multiple lo-
cations. Determining which signals to synchronize to, particularly
on a dynamic basis
when a reference signal becomes unavailable, is exceptionally difficult.
There are six basic approaches used, or considered for use, in synchronizing a digi-
tal network:
1. Plesiochronous
2. Networkwide pulse
$tuffing
3. Mutual synchronization
4. Network master
5. Master-slave clockins
6. Packetization
7.3.1 Pl esi ochronous
A pJesiochronous
network does not synchronize the switches but merely uses highly
accurate clocks at all swirching
nodes so the slip rate between the nodes is acceptably
low. This mode of operation is the simplest to implement
since it avoids distributing
timing throughout the network. A plesiochronou$
network, however, implies that the
smaller switching nodes carry the cost burden of highly accurate and redundant timing
source$' As a compromise, large networks can be divided into subnetworks for timing
purposes and use plesiochronous
operations for inter-subnetwork synchronization
and
some other, more cost-effective, means of providing
intra-subnetwork synchroniza-
tion. As described in section 7.5, the public
telephone network in the united states
uses plesiochronous
synchronization at the upper levels.
7.3 NETWORKSYNCI{HONIZATnN 363
Plesiochronous timing is also used to synchronize international digital network in-
terconnections. In recommendation G.811
[23],
the ITU has established the stability
objectives for clocks of all intemational
gateway digital switches. The stability objec-
tive of one part in l01l implies that slips between international gateway switches will
occur at a rate of one
per 70 days.
(This assumes one clock is positive one part in 10lt
and another clock is negative one pafr in 1011.)
7.3.2 Networkwide Pulse Stuffing
If all internal links and switches of a network were designed to run at nominal rates
slightly higher than the nominal rates ofthe voice digitization
processes, all voice sig-
nals could
propagate through the network without slips by stuffing the information
rate up to the local channel rate. None of the clocks would have to be synchronized to
each other, and relatively coarse clock accuracies could be tolerated. At every inter-
face between systems running under different clocks, however, the individual chan-
nels would have to be unstuffed from the incoming rate and stuffed up to the local or
outgoing rate. In essence, the TDM links of the network would provide TDM channels
through which user data flows at lower and variable rates, the differences being ab-
sorbed by internal pulse
stuffing.
ln contrast to
pulse-stuffing operations of higher level multiplexers where all chan-
nels in a lower level digital signal are stuffed as a group, switching operations imply
that each channel must be stuffed independently. The need for separate pulse-stuffing
operations is illustrated in Figure 7.25, which depicts two voice signals being switched
into a common TDM outlet link. Obviously, the bit rate rt3 of both output channels is
identical. If the two channels originate in portions of the network running under dif-
ferent clocks R1 and R2, the
pulse-stuffing adjustments must be made separately for
each channel. The complexity of stuffing and unstuffing every 64-kbps channel at
every network element would have been extremely expensive when the digital net-
work began to take shape.
Figure 7.25 Swirching two channels with different rates onto a common TDM output
364 NETWoFKSyNcHRoNt zATt oNcoNTHoLANDMANAGEMENT
Although the availability of low-cost logic could minimize the cost aspect of net-
workwide pulse
stuffing, other problems
would occur. First, the 64-kbps clocks for
speech reconstruction would be required for each channel and would contain rela-
tively large amounts of waiting time
jitter.
second, the network would no longer pro-
vide byte framing so the channel recovery process
would have to also include byte
framing logic. Byte boundaries of PCM data can be readily determined from statistical
data patterns
in the bit positions (e.g.,
the polarity
bit), but other applications for the
channels may require explicit byte boundary identifiers.
7.3.3 Mutual Synchronizatlon
The two preceding
sections discuss modes of operation for the network that do not in-
volve synchronization
ofindividual clocks. This section and the next two describe net-
work timing plans that synchronize each individual clock to a common frequency. The
first method, mutual synchronization, establishes a coilrmon network clock frequency
by having all nodes in the network exchange frequency references as shown in Figure
'l
.26.Each node averages the incoming references ancl uses this for its local and trans-
mitted clock. After an initialization period,
the network clock normally converges to
a single stable frequency. under certain conditions, however, the averaging process
can become unstable
[24].
The main attractiveness of a mutually synchronized network is its ability to remain
operational in spite of a clock failure in any node. The main disadvantages are the un-
cenainties of the exact average frequency
and unknown transient behavior. Mutual
synchronization has not been considered for the North American telephone network.
In Great Britain, however, a hierarchical timing structure was once considered that
utilized mutual synchronization within some portions
of the network
[25].
7.3.4 Network Master
Another method of synchronizing the network is shown in Figure 7.27. with this
method a single master clock is tranrtmifted to all nodes enabling them to lock onto a
common frequency. As indicated, all network nodes are directly connected to the net-
work master, implying the need for a separate transmission network dedicated to the
SN
Flgure 7.26 Mutual synchronization:
SN, switching node.
7.3 NETWORK SYNCHHONIZATION 365
Figule7.27 Network master synchronization: SN, switching node,
distribution of the reference. Reliability considerations also imply that alternate
paths
be
provided to each node. Because ofcost considerations for the separate timing net-
work and reliability problems with reference distribution, a networkmaster with direct
transmission to each node is undesirable.
Something similar to a master synchronized network is evolving through the use
of Global Positioning System
(GPS) satellites for timing distribution to network
nodes. Switching system$ of telecommunications networks around the world are using
GPS and other satellite systems to synchroniee their switching office clocks. As the
costs of the GPS receivers and suitably stable oscillators drop in cost, more and more
network nodes ate being timed from this master
(highly
accurate) source. Because
CDMA digital mobile systems also lock to GPS, lower cost synchronization systems
have also been developed that obiain GPS timing from the CDMA base stations in lieu
of directly receiving GPS signals that often require outside antennas to receive from
multiple GPS satellites simultaneously.
7.3.5 Master-Slave Synchronizatlon
The main drawback to network ma$ter synchronization as described in the preceding
section is its need for separate and reliable transmission
facilities to every node. Figure
7.28 shows a network configuration that disseminates a master reference by way of
the message links themselves. A network reference frequency is transmitted to a few
selected higher level switching nodes. After these nodes synchronize
their clocks to
the reference and remove tansmission link-induced timing
jitter,
the reference is
passed on to lower level switches by way of existing digital links. The next lower level
switches, in turn, synchronize to an incoming link from the higher level and pass tim-
ing on to another level of switches by way of their outgoing digital links. The
process
of
passing the reference downward from one level to the next is referred to as
"mas-
ter-slave synchronization.
"
Since all switching nodes in the network are synchronized either directly or indi-
rectly to the same reference, they all run at the same nominal clock rate. Thus slips
should not occur. However, because of the different paths through which timing is dis-
seminated, short-term frequency differences can occur between some nodes' If these
366 NETWoRKSYNcHRoNIZATIoNcoNTHoLANDMANAGEMENT
Indlrcctly
rynchronlrrd
linkr
. - - - ; _
Figure 7.2t Master-slave synchronization.
nodes are synchronized indirectly, as shown in Figure 7.28, infrequent slips might oc-
cur. Furthermore, reliability considerations imply that backup clocks must be pro-
vided in all switches should the clock distribution system fail. when this happens,
slips become more likely, but only after relatively stable backup clocks have had
enough time to drift from the common reference frequency.
AT&T and the united
states Independent Telephone Association (usITA)
origi-
nally selected master*slave synchronization for the switched digital network in the
united states
[26].
The reference ftequency was locared in Hillsboro, Missouri, from
which selected No. 4 ESS switching centers received their timing by way of dedicated
transmission facilities. synchronization
of all other switches occuned by way of ex-
isting digital transmission links. As discussed in Section 7.4, the original plan has been
changed to use plesiochronous
synchronization at the highest level. Furthermore, as
mentioned in the previou$
section, the availability of Gps timing sources is leading to
more and more nodes at the higher level and, consequently, fewer and fewer slave nodes.
7.3.6 Packetlzation
The synchronization
discussions of the five preceding
sections have assumed implic-
itly that a synchronous, circuit-switched network was being considered, since prevail-
ing digital voice networks operate in that manner. For completeness, however, another
form of network must be mentioned-a packet-switched
network.
As discussed in chapter 10, packet-switched
networks break up messages into
identifiable blocks (packets
or cells) of data. In between the blocks. the hansmission
links carry either idle codes or control messages. If all messages (control
and data) are
separated by a nominal interval of idle transmission, elastic stores can be reset in
preparation
for the next block. As long as each block is limited in length, the elastic
stores can absorb clock differences and avoid losses of data. (In
essence, slips occur
in the idle codes.)
7.3.7 Network Timing Performance Measuremcnts
After choosing a synchronization
architecture for a network, it is necessary to be able
to measure the quality
of the timing signals within the network to qualify
equipment,
7.s NETwoRK sYNcHRoNrzATloN 367
determine performance margins, isolate faulty equipment or transmission links, and
possibly evaluate design altematives. ANSI and ITU standards use the concept of time
interval error
(TIE)
and maximum time interval error
{MTIE)
for these
purposes' An
additional measure of clock
performance is a time variance
(TVAR).
Maxlmum Time lnterval Error
A TIE is the difference in delay between a timing signal and an ideal timing signal
measured at the end of a
particular time
period (i.e.,
the error at the end of a time in-
terval). An MTIE is the maximum
peak-to-peak varjation in TIE values that occur
within a specified time interval. These concepts are illustrated in Figure 7.29. Figure
7.29a displays the TIE that occurs when measuring a
perfectly stable but inaccurate
timing signal. Because the frequency ofthe signal under test differs from the reference
frequency by a constanf value
(A/), the TIE is directly propol'tional to the measure-
ment interval S. The enor is determined by counting clock cycles in both signals and
expressing the difference as the time required by the test signal to catch up
(or fall
back) to the curent reference count. Thus the TIE
produced by a constant-frequency
offset is
_
s(/+
4fl
-
s(/)
f
J
(7.7)
where An is the accumulated difference in clock cycles, A/is the frequency offset, and
S is the measurement interval in seconds.
Because the TIE in Figure 7.294 increases monotonically with S, the MTIE and the
TIE are identical. In contrast, Figure 7.29b depicts the TIE produced by a timing signal
with no long-term frequency offset but some shott-term instabilities
fiitter
and wan-
der) as would occur if the reference was relayed through a network. As indicated, the
TIE varies as a function of S but is bounded. The MTIE is the largest peak-to-peak dif-
ference in TIE values in a measurement interval S. The determination of MTIE with
no long-term frequency offsets is also shown in Figure 7.30, where
periodic samples
of time interval errors are depicted.
Figure 7.29c displays the more
general case where the signal under test contains
bottr instability and a frequency offset. In ttris case the choice of the value for S is critical.
If S is too small,
jitter
will mask the
presence of the offset. If S is too large, the frequency
offset dominates the TIE and MTIE measurement$.
Thus periodic measurements need to
be recorded for a time history to fully characterize the timing imperfections.
The first three examples in Figure 7.29 assume the use of a
"perfect"
reference with
which to make the mea$urements.
If an
"imperfect"
reference is used, meaningful
measurements are obtained as long as the TIE variations are significantly
greater than
T''=T
='H)
NETWOHK SYNCHRONIZATION CONTROL AND MANAGEMENT
(d)
Timc
-r>
s
Figure 7.29 TrE and MTIE for various clock imperfections: (a) pure
offset;
(D) pure jitter;
(c)
offset and
jitter;
(d)
TE difference measuremenr.
StablE
PErfEGt lnaccilratG
Source Sourcl
Ti me +
PErfoct
Soutcs
Time +
Tim6
-rF
s
7.3 NETWORK SYNCHRONIZATION 369
Figure 7,30 Maximum time interval error example,
the imperfections of the reference. Sometimes
the relative
performance between two
timing signals is of interest. For example, a PBX with Tl connections to two different
places (or carriers) is more dependent on the timing differences between the two trans-
mission links than on the absolute performances. Figure 7.29d depicts measuring the
relative perforrnance of two such links. In this case, all measurements are meaningful,
even arbitrarily low frequency wander, which can cause slips but is difficult or impos-
sible to measure in an absolute sense. If relative measurements indicate timing prob-
lems exist, absolute measurements may be needed to isolate the problem source.
Time Varlance
Neither TIE nor MTIE measurements
convey any information regarding the fre-
quency content of the
jitter (other
than that conveyed by the measurement
interval ,f).
A more general statistical characterization of thejitter requires representing thejitter
magnitude as a function of frequency, or equivalently, as a function of time between
TIE samples. The time variance
(TVAR)
is such a measure. TVAR values are deter-
mined as the expected vaJiance of second-order differences between TIE samples that
are separated by a time t, where t varies from zero (or
some fraction of a second) to
some maximum observation
period. TVAR values are customarily measured in units
of time squared
(e.g.,
nanoseconds squared). The formula for calculating
TVAR val-
ues from TIE samples;;(i
=
1, . . .
,19
is
TVAR(I)
=
o:(r)
:.tgltltt)tl
N*3n+l fn-r
'2
-ffi -
3n + 1)
I
l E
{"r*r"*o
-zxv*k**r*)
|
(7' 8)
L v '
' r r . - L r
F l
l m
l
where t
=
nto
(to is the sampling
interval) and the obserrvation
period is Nto'
The use of second-order differences
removes the effect of a dc offset or of a linear
phase ramp in the TVAR samples. Thus, there is no need to ulie a synchronized refer-
370 NETWoHKSYNCHHoNIZATIoNcoNTHoLANDMANAGEMENT
ence to deterrnine the TIE samples before determining the TVAR values. However, if
TVAR values for long observation periods
T are to be determined, a stable reference
is needed to preclude
reference wander from influencing the measurement values.
Time Devlatlon
Time deviation (TDEV)
measurements
are merely the square root of the TVAR meas-
urements. Thus, TDEV and TVAR values have the same relationship
as the classical
standard deviation and variance of a probability
distribution. TDEV is customarily
measured in nanoseconds.
7.4 U.S. NETWORK
SYNCHRONIZATION
The original plan
for synchronization
of the u.s. network formulated by the Bell sys-
tem and the u.s. Independent Telephone Association was to use master-slave
syn-
chronization with a single master clock
[26].
Due to the breakup of the network into
multiple independent companies and to difficulties in reliably distributing a highly ac-
curate master, the synchronization architecture for the United States was changed to
a plesiochronous/hierarchical
design in the lare l9g0s
[13,
lg, I
g].
Although the hier-
archical design is still in use, it is gradually
changing to a
..flatter"
design by incorpo-
rating top-level functionality in more and more nodes.
7.4.1 Synchronization
Regions
As shown in Figure 7.31, the public
network is partitioned
into synchronization re-
gions that are intemally synchronized
with a master-slave timing hierarchy that es-
tablishes different levels of timing quality:
sffatum I to stratum 4. stratum I clocks
have the highest quality
while stratum 4 clocks have the lowest. Timing for each re-
gion is established by a primary
reference source
(pRS)
at stratum l. stratum I clocks
are free-running clocks with inaccuracies no greater
than one part
in 10il. Some re-
gions
may have their own PRS while others may use a synchronization
signal from
another party (e.g.,
an interexchange
carrier such as AT&T).
For the most part,
the synchronization regions correspond to LATAs. Every region
must have at least one stratum 2 clock, which is typically associated with an access
tandem switch. Toll offices within LATAs may also have stratum 2 clocks. All toll
switches within AT&T network contain stratum ? clocks
[13].
Except when they use
a coilrmon PRS, the synchronization
regions are independently
syncfuonized. Thus
connections
between the regions (using
interexchange
carriers like AT&T) are typi-
cally plesiochronously
timed. within a single region nodes are synchronized in a mas-
ter-slave hierarchy as indicated.
The accuracy requirements
of the four levels of stratum clocks are provided
in Ta-
ble 7 .2. These accuracies pertain
only to situations in which the nodes are operating
in a free-running
mode. Normally the nodes are synchronized to higher level clocks
so the long-terrn accuracy is haceable back to the respective PRS. In addition to listing
free-running
accuracies, Table 7 .2lists accuracies that must be met durine holdover
E
t
xqr f ; t
a o J 4
c
E 6 ( J E
o
.9
o
F ( ,
U O
o
0
o
_ t h
E ( J
l ; 6
:i;
t
q i
* [
\
i
I
6
N
ti
h
+
F
E
o
{
-
o
z
(t
t*
I
uo
371
572 NETWoRK syNcHHoNtzATtoN
coNTRoL AND MANAGEMENT
TABLE 7.2 Stratum Clock Accuracy Requlrementa
Level
Free-Run Accuracy
24-Hr Holdover Accuracy
Stratum 1
Stratum 2
Stratum 3E
Stratum 3
Stratum 4
Not applicable
t 1 x10- l oper day
+1 x 10-s (day 1)
< 255 slips in day 1
Not required
states. The holdover mode for stratum 2 and 3 clocks requires these nodes to gradually
transition to the free-running state when they lose their reference signals.* A stratum
2 clock, for example, loses the previously
established frequency at the rate of I x I 0-r0
per day, implying it would take 160 days to reach the worst-case free-running
accu-
racy' Stratum 4 clocks, however, have no holdover requirements
so they enter the free-
running state immediately
upon losing their reference(s).
Network synchronization requirements also specify how the various nodes respond
to degradation or complete failure of their references. Stratum 2 and 3 clocks must
bridge short intemrptions in the reference with minimum specified time-keeping
er-
rors
[l9].
Stratum 3 clocks, which typically have a primary reference and a secondary
reference, must perform
a very gradual
switchover to the secondary reference when
the primary fails. Abrupt switchovers, as often occur in stratum 4 nodes (pBXs),
can
cause phase
transients in the output clock, which can in turn disrupt synchronization
in all downstream devices (e.g.,
within a private network).
7.4.2 Primary Reference
Sources
The ANSI MTIE specification for the accuracy of a primary
reference source is shown
in Figure 7.32. Notice that
jitter (instability
that can be observed in less rhan 0.05 sec)
is not specified. wander, as measured over 500-sec time intervals. is limited to 3000
nsec. The asymptote for long-term timing errors corresponds to inaccuracies of 1 x
10* 1r .
In addition to being designed with highly accurate (cesium
beam or rubidium)
clocks, all PRSs must be continuously verified with universal coordinated time
(urc).
such verificarion can be achieved by using a UTC-based navigation sysrem
such as Loran-c or GPS. MCI uses Loran-c and Gps to directly synchronize its
pRS.
AT&T uses GPS to monitor (verify)
the long-term accuracy of each
pRS
node estab-
lished in Norttr America.t The typical accuracy of these nodes is much better than the
ANSI or ITU requirement
[13].
*Holdou-t
operations are typically implemente<l by storing the last valid conection value (e.g.,
filtered
phase
detector output) and maintaining that value as an offset conection to a highly stable
*(temperature-compensated)
oscillator.
'Originally'
AT&T had 16 PRS nodes. Subsequent large-scale usage of GPS has led to the establishment
of hundreds of PRS in the AT&T network.
+1
x 10-11
+1
. 6 x 10+
t4.6 x 10+
f4.6 x 104
i32 x 104
7.5 NETWoRK CONTROL 373
105
^ r d
o
E
E' *
t
t0?
l 0l
t
l 0- l
104
Ob86Nation
PBriod
(sec)
Figure 7.32 Permissible MTIE versus observation period at the output of a primary reference
source.
7.4.3 1996 AT&T Synchronlzation
Architecture
Beginning in 1996 AT&T began a move away from a two-tiered synchronization
ar'
chitecture to a single-level architecture
with a sfiatum 2 clock in every office synchro-
nized to GPS
[?7].
Significant features ofthis architecture are:
l. The synchronization
distribution network has been eliminated.
2. Synchronization is independent of the network topology so the traffic-carrying
network can be changed without affecting synchronization.
3. Each node is monitored by two adjacent nodes and each node monitors two
adjacent nodes for
performance verification'
4. Performance verification involves the use of both MTIE and TDEV.
5. DSI timing signals are derived from SONET optical signals.
7.5 NETWORK CONTROL
The synchronization
procedures described in the
preceding section represent methods
for controlling the timing between transmission
and switching systems. In this section
. . . . . . . .
MTt E
-
aTrT, 10-il
(t o-?
s
+
X
-
3000 nsec
(proritionrlly)
10?
374 NETWoBK$yNcHBoNIzATtoNcoNTHoLANDMANAGEMENT
synchronization is discussed in a more general
sense. Instead ofjust time or frequency
control, the synchronization
concept is extended to higher level functions ofconnec-
tion control and network control as a whole. Fundamental to the control concept is the
interaction between two processes (e.g.,
the exchange of information from one switch-
ing machine to another to set up or monitor a connection).
A particularly
useful means of defining the interaction
of two processes
is a state
transition diagram. The main purpose
of the state transition diagram is to abstract the
operational state$ of a process
and indicate what events cause ffansitions from one
$tate to another. when these events are messages (signaling
tones or ccs messages)
from another process,
the state transition diagram efTectively def,rnes how two com-
municating proce$$es
interact.
7.5.1 Hlerarchical
Synchronization
processes
As an example of a more generalized
concept of synchronization,
Figure7.33has
been
included to demonstrate
three distinct levels ofcontrol for a conventional
telephone
connection using digital transmission and multiplexing.
The lowest level process
shown in Figure 7.334 depicts nothing more than the clock synchronization proces$
required to transmit and receive digital information. There are only two states to the
process
in both the transmitter and the receiver. The purpose
of clock synchronization
is to cause ffansitions between the two states in the receiver to coincide with transi-
tions in the clocking process
of the transmitter. To accomplish this, a certain amount
of transmission
capacity is required in the form of line code transitions.
Figure 7.33b depicts a higher level synchronization process
involving the framing
of a time division multiplexer. Both processes
represent a modulo-N counter, where
N is the number of channels
in the TDM frame. The two processes
are synchronized
(framed)
by utilizing some of the transmission capacity to send framing patterns.
once
the receiver acquires framing, the counter in the receiver counts in synchronism with
the counter in the transmitter
so that individual TDM channels are properly
identified.
Figure 7.33c provides
state transition diagrams of a somewhat more complicated
but easily understood process. The figure depicts the connection control ofa conven-
tional telephone call. The state transition diagram ofthe first process
represents a sub-
scriber placing
a call
(going
off-hook). The second state transition diagram represen6
the sequence of states the control element in the local switch goes
through to set up
the connection.
As indicated, the process
begins by the originating subscriber going
off-hook and
waiting for the dial tone. When the switch recognizes the off-hook signal (current
flow
in the line), it connects the subscriber line to a digit receiver that returns a dial tone.
The subscriber then dials the address ofthe desired telephone and enters another wait
state' Upon receiving the last digit of the address, the switch control processes
the re-
quest.
once the status of the called party
is determined, a busy tone or a ringback tone
is returned
to the originating subscriber. A busy tone prompts
the subscriber to hang
up (go
on-hook) while a ringback signal causes the subscriber to stay in the wait state
until the called party
answers or until the caller
"times
out" and abandons the call.
7.s NETWoRK coNTRoL 375
(c)
Figure 7.33 State transition diagram of synchronizatiQn
processes:
(a)
timing
process;
(b)
tiaming proce.qs;
(c)
telephone connection
process,
When the called party answers, both
processes enter the connected state and com-
munication between the end users begins. The end users then
get involved in yet an-
other level of
"synchronization."
Voice telephone
users begin by exchanging
greetings and identities to establish a
"connection"
between their thought
processes to
communicate
on a mutually understood
subject. The message exchange
process also
requires synchronization
so that only one person talks at a time. Hence various forms
of control signals are needed to
"turn
the line around." Although being somewhat sub-
tle in nature, these control signals represent transmission overhead in the same sense
as control signals within the network. A talker may indicate his end of transmission
Rectlvtr
H
Traffimlttor
fl
fl}
(a)
Originating rub,ecriber
Suvltch cofitrol
376 NETWoRKSyNcHRoNtzATtoNcoNTRoLANDMANAGEMENT
by asking a question,
by inflections
in the voice, by the message itself, or more com*
monly by a pause.
Data communications equipment goes
through the same basic procedures
in order
to establish connection$
and exchange information. In this case, the procedures
are de-
fined more formatly and, consequently, are more restrictive. The formal rules of com-
munication
between data communications
equipment are usually referred to as a
-'protocol."
Data communications protocols
typically include a definition of certain
control codes, code interpretations,
message framing, turn-around procedures
for half_
duplex lines, error control, message sequencing, fault control, and recovery.
Automated fault control and recovery procedures
for communications
networks
can become quite involved
and difficult to implement reliably. When individual voice
circuits malfunction (e.g.,
become noisy or disconnected), the recovery procedures
are
left to the users. They merely redial the connection
and take up where they left off.
However, Iarge trunk groups
or switching systems must be designed for higher levels
of dependabitity
and maintainability.
The dependability criterion ensures that failures
or malfunctions rarely occur or that they are circumvented
automatically by protection
switching' High levels of maintainability
ensure that failures are repaired quickly
when they occur. within switching systems, most of the instructions
and memory
words of the processor
are dedicated to hardware and software performance
monitor-
ing, recovery procedures,
and maintenance diagnostics.
7.6 NETWORK
MANAGEMENT
In addition to controlling individual
connections and equipment, a communications
network must also manage its facilities on more macroscopic
levels. The basic goal
of
network management is to maintain efficient operations during equipment failures
and traffic overloads. The main considerations
are routing conffol and flow control.
7.6.1 Routing
Control
Routing conffol refers to procedures
that determine
which paths
in a network are as-
signed to particular
connection$.
If possible,
connections should use the most direct
routes at the lowest levels
of the network. The direct routes are obviously desirable
because they use f'ewer network facilities and generally provide
better ffansmission
quality'
However, economic considerations
often limit the capacities of the direct
routes so that alternate routes are needed to maintain suitably low blocking prob_
abilities between one switching machine and another.
lf a t.unk group
between two switching machines contains enough circuits to pro-
vide an acceptably low blocking probability,
a significant number of the circuits in the
group
are idle during average traffic loads. A more economical
design allocates a lim-
ited number ofheavily utilized trunks in the direct route and provides
alternate routes
for overflow (alternately
routed) traffic. In this manner the users are able to share
larger portions
of the network. chapter 12 presents
basic examples of how a network
7.6 NETWoHKMANAGEMENT 377
can be engineered to minimize the transmission
facilities while providing a given
grade of service
(blocking probability). As discussed in Chapter l, the use of cenmal-
ized control for the network, with common-channel
signaling,
provides signiflcant ef-
ficiencies of operation in congested
networks.
7.6.2 Flow Gontrol
ln the preceding section, altemate routing is discussed as one aspect of managing traf-
fic in a communications
netwolk. Routing algorithms
are concerned
only with the
utilization of paths or directions
of travel within a network. Another requirement
of
netWork management
is to control the amount of traffic in a network' Managing the
rate at which traffic enters a network is referred to as flow control.
A network without
effective
flow control
procedures becomes very inefficient or ceases to function en-
tirely when
presented with excessively
heavy traffic conditions'
The
generalized performance of a large, uncontrolled
network is shown in Figure
7 .34 as a function of the offered traffic. As indicated, when light traffic conditions ex-
ist, the network caffies all traffic requests
presented. As the load increases,
however,
some of the offered haffic is rejected because no appropriate circuits are available
for
particular connections;
that is, blocking exists. As the input load increases
even fur-
ther. a network with no flow control eventually begins to carry less traffic than it does
when Iighter loads are
presented. If the offered load increases even more, the network
may even cease to carry any tralfic at all.
The reason that the volume of carried traffic decreases when the offered traffic ex-
ceeds some critical value is that partially completed
requests tie up network resources
while trying to acquire other resources tied up by other partially completed
requests'
Thus a form of dynamic deadlock occurs. Prior to the development of centralized net-
work control, this situation would often arise on busy calling days
(e.g.,
Mother's
Day). In a network with distributed control all sources of traff,rc are serviced by suc-
cessively seizing trunks to intermediate switching
nodes until the destination
is
reached. If heavy traffic exists, request$ emanating from all sides of the network en-
,9
E
E
I
Figure 7.34 Traffic caffied versus traffic offered for a network with no flow contol.
378 NETWoRKsYNcHRoNIzATIoNcoNTRoLANDMANAGEMENT
counter congestion somewhere
near the middle. At that time, the partially
completed
connections are tying up facilities needed by other requests. In the limit, when ex-
tremely heavy traffic exists, all network resources
are held by partially
completed re-
quests
and no complete connections
can be established!
Another example of the need for flow control is automobile haffic at a metropolitan
intersection'
Have you ever encountered an intersection
in which your
lane of traffic
was blocked by cross traffic backed up into the intersection?
In essence. the driver
blocking your direction
of travel seized a cornmon resource
(the
middle of the inter-
section) without being able to obtain the next resource required (the
other side ofthe
intersection).
Bumper-to-bumper
traffic in one direction
can significantly
degrade the
throughput in other directions. With heavy traffic in all directions, total throughput
can grind
to a halt until the congestion
is relieved from the periphery
inward.
The fundamental
principle
demonstrated
by these examples is th;t efficient use of
the common resources of a heavily loaded network requires some form of flow con-
trol. In the automobile example, smooth operation of an inter$ection
depends on each
driver looking ahead and not entering an intersection
unless he can g*t
ull tt * *uy
across. A telecommunications
network must use the same basic principle (it
is hoped
with more discipline).
The control elements at the periphery
or ihe networt must be
aware of the intemal status of the network and conhol the flow of haffic from its
sources.
More than one level
of flow control is sometimes implemented
within a network.
In a data communications link, some form of flow conhol is required to keep a source
terminal from overloading the terminal at the other end of the link. The receiving ter-
minal uses a reverse channel (sometimes
with a lower bit rate) to inform the source
when to cease and when to begin transmissions.
This level of flow control in a circuit-
switched network involves the terminals themselves and is of no concern to the net-
work since the traffic flows within an established connection.
of more concem ro a
circuit-switched
network
is how to control the flow of connection requests into the in-
terior of the network. In setting up a long-distance
connection, the first few circuits
required should not be seized unless there is a reasonable
chance that all ofthe circuits
necessary to complete
the connection
can be obtained. Partially completed
circuits
only degrade the network
capacity by increasing congestion
without satisfying a serv_
ice request.
Network flow control i$ greatly
simplified with common-channel
signal_
ing support for cenffalized network control. The following paragraphs
describe basic,
uncentralized
flow control
techniques and how cenhalized
conffol simplifies their im-
plementation.
Trunk Directional
ization
The operation
of tuunl< circuits can be classified according to two different ways of
controlling seizures for particular
calls. Two-way
trunks can be seized at either end.
one-way trunks,
on the other hand, can be seized only at one end.
(Notice
that this has
nothing to do with the direction of message transfer on established connections.
which
is always in both directions.)
when one-way
trunking is used, the trunk group
is usu-
ally partitioned
into one group
that can be seized at one end and one group
that can be
7.6 NETwoRKMANAGEMFNT
379
seized at the other end. Two-way trunk
gfoups are obviously more flexible in servicing
fluctuating traffic
patterns, but they are more difficult to control since the possibility
of simultaneous
seizures
(called glare) at both ends must be resolved'
A useful feature to incorporate
into two-way
trunks is the ability to directionalize
them by marking them busy at one end and effectively
creating one-way trunks. With
this mechanism, a distant, overloaded switching
node can be relieved of additional
in-
coming traffic while
providing it sole access to the trunk
group for outgoing traffic'
Thus the overloaded node relieves its congestion
while inhibiting new anjvals.
When the network as a whole experiences
heavy haffic loads, trunk directionali-
zation can be used to reduce the flow of connect requests
into the interior of the net-
work while establishing
one-way trunks from the interior to the
periphery. Thus
connect requests that manage to get to the interior
have a much better chance of ob-
taining all facilities
required for the complete
connection'
Cancellation
ol Alternate Routing
Alternate routing of traffic accommodates localized
overloads by transferring
traffic
to underutilized
routes. During networkwide
overloads,
however, altemate routing
is
undesirable
for two reasons. First, alternate routes imply that a
greater number of
transmission
and switching
facilities are needed for a connection.
If these same facili-
ties could be a.rsigned to two or more direct connections,
the total number of links
per
call could be reduced and the network could carry more traffic.
Second, the
probability that an alternately
routed call can acquire all the necessary
resources is relatively low. Trying to set up a connection
with a large number of fa-
cilities is undesirable
ifthe
probability ofgetting all ofthe facilities is low
(particularly
so, if some facilities are fruitlessly tied up while less demanding
requests ale pending)'
Code Blocking
Code blocking rbf"rt to artificially blocking calls intended
for specific destination
codes. If the calls are blocked at originating end offltces before they acquire internal
network facilities, the destinations
are relieved of incoming traffic
without tying up
facilities that may be needed
for outgoing requests
from the specified areas.
The method of flow conhol
is particularly useful
in times of natural disasters,
which typically stimulate
large numbers
of calls both into and out of the area of the
disaster. [n these events a network control
center can initiate code blocking for all, or
a large
percentage, of the calls into the area. The
principle of
giving preference to the
outgoing calls serves two
pufposes. First, no network facilities are seized unless there
is a reasonable
chance of obtaining all facilities necessary.
It is the runks into or out of
the disaster arrea that are the focal point of network congestion. Once one of these fiunks
is seized, the rest of the connection
can
probably be established. Second,
code blocking
is
useful because outgoing calls are probably more impoftant than incoming calls.
Centralized
Connectlon Eontrol
All of the flow control
procedures described
previously are designed to eliminate
sei
zures of common resoulces if the desired connection
has a low completion
probability.
380 NETwoRKsyNcHHoNtzATtoNcoNTRoLANDMANAGEMENT
Because of the distributed nature of network control implied by these operations,
these
control procedures
are necessarily probabilistic.
To maintain
a certainamount
of net-
work efficiency, the network is purposely
operated
at less than maximum capacity.
A more desirable
mode of operation, from a throughput point
of view, is to allocate
network facilities from a single centralized
conhol node. Since this central node has
access to the status of all network resources, no facilities
are assigned
to a pafiicular
request unless all facilities needed for the complete connection
are available.
Network
hansmission
links are assigned in a manner that is analogous
to the assignment of in_
temal links of common control switches.
Complete centralizecl
confrol of a network as large as the public
telephone network
is obviously infeasible from the point
of view of maintaining status oi all interstage
links within end office switches and from the point
of view ofiurvivability
of the ne1-
work when the control node tails. However, many aspects of centralized control have
been implemented in North America and around the world with common-channel
sig_
naling (CCS)'
For example, INWATS call requests are routed ro a cenrral node that
determines
if the destination
is busy or not. If the destination is busy, the originating
end office is instructed
to return the busy tone without any of the int#al mansmission
links ever being seized' This mode of operation is particularly
useful for 800 numbers
oNwATS) that occasionally
experience very heavy traffic flow because of national
televi sion announcemenrs.
without ccs, the previous
mode of operation
was to return a busy tone ail the way
through the network from the place
at which the busy circuit or subscriber
is located.
Thus the path
through the network was tied up during the time the busy tone was being
returned' CCS allows the originating
office to return the busy tone so internal network
facilities
can be released and reassigned
immediately
upon detecting the busy condi_
tion.
REFERENCES
I M. Decina
and U- deJulio,
"International
Activities
on Network Synchronization
for
Digital communication,"
IEEE Intematiotutl
communications
conference,lg7g,
2 J. R, Pierce,
"synchronizing
Digiral Networks," Bel/ system Techntual
rournal, Mar.
1969, pp.
615-636.
3 c. J' Byrne, B. J. Karafin, and D. B, Robinson,
"systematic
Jitter in a chain of Digitar
Repeaters,"
Bell System Technical
Journal, Nov. 1963, pp. 2679_27 14.
4
"Digitally
Dejirtering LaserDisc
players,"IEEE
Spectrum,Feb. 1990, p. 14.
5 F. M. Gardner, PhaselockTechniques,2nd
ed., Wiley, New
york,
1979.
6 E. D. sunde,
"self-Timing
Regenerative
Repeaterc,,, Bell system Technical
rournal,
July 1957, pp.
891-938.
7 D. L. Duttweiler,
"The
Jitter
performance
of
phase-Locked
t oops Extractrng Timing
from Baseband
Data
waveforms," BerI system Technical
Journal, Jan. 1976, pp,
37-58.
8
"carrier-to-customer
Installation-DSl
Metallic Interface," ANSI Tl. 403_l9gg,
American National Standards Institute,
New
york,
19g9.
10
l l
12
r3
t 5
l 6
t 7
l 8
19
REFERENCES
381
9 H. Meyer, L. Popken, and H. R. Mueller,
"synchronization
Failures in a chain of PLL
Synchronizers ,"
IEEE Transadions
on Communicatiazs,
May 1986,
pp' 436-445'
P. R. Trischitta and E. L. Varma, finer in Digital Transmission Systems, Artech House,
Norwood,
MA, 1989.
P, R. Trischitta and P. Sannuti,
"The
Accumulation
of Pattern-Dependent Jitter for e
Chain of Fiber Optic Regenerators," IEEE Transattions on Communications,
June
1988, pp. 76r-765.
J, F. Boyle, J. R. Colton,
C. L' Dammann,
B' J' Karafin, and H' Mann,
"No'
4ESS:
Transmission/Switching
Interfaces
and Toll Terminal Equipment," Bell System
Technical Jountal, Sept. 1977
,
pp' 1057-1097
-
J. E. Abate, E. W. Butterline, R. A. Carley, P. Greendyk, A' M. Montenegro,
C' D' Near'
S. H. Richman, and G. P. Zampetti,
"AT&T's
New Approach to the Synchronieation
of
Telecommunication Networks,"
IEEE Communications
Magazine,
April 1989'
pp.
35-45.
H. Drucker and A. C. Morton,
"The
Effects of Slips on Data Modems,"
IEEE
International Conference
on Communitations,
1987, pp. 12'4.I-12'4'3'
J. F. Ingle,
..Identification
of Digital Impairments in a voiceband
channel," IEEE
Int e rnational C onfere nc e on C ommunicarions,
1 989, pp. I 2'3' I
-
I 2' 3'5'
J. E. Abate and H. Drucker,
"The
Effect of slips on Facsimile Transmission,"
IEEE
Int e mati onal C onferenc
e o n C ommunicalions,
1 988, pp' 32'3.1
-32'3'4'
J. E. Abate, L. H. Brandenburg, J. c. Lawson, and w. L, Ross,
"The
switched
Digital
Network Plan," Bell SystemTechnital Joumal, Sept. 1977,
pp. 1297-1320'
..Digital
synchronization Network
Plan," Bellcore TA-M1,000436, Issue 1, Bell
Communications
Research, Morristown, NJ' Nov' 1986.
,.synchronization
Interface standards for Digital Networks,"
ANSI Tl. 101*1987'
American National Standards
Institute, New York, 1987.
Technical
staff, Bell Telephone Laboratories,
Transmission systems
for
Communications, Bell Telephone Laboratories, Winston-Salem'
NC' 197I '
,,Impact
of Jitter on the second order Digital Multiplex at 6312 kbit/s," AT&T
Submittal to CCITT study group on
jitter,
Green Book, Vol. 3, pp. 861*869'
D. L. Dutrweiler,
..waiting
Time Jitter," Bell systemTechnical Joumal,
Jan. 1972, pp.
165*207.
23
"Plesiochronous
Operation of International
Digital Links," CCftT
Recommendation
G.LLI, Orange Eook, Geneva, Switzerland'
1976.
24 J. P. Moreland,
"Performance
of a System of Mutually
Synchronized Clocks," Bell
Sy s te m T e thnical J o umal, Sept. 1 97 l, pp. 2449
-MM'
ZS
p,
A. Mitchell and R. A. Bourler,
"synchronization
of the Digital Network
in the United
Kingdom,"
IEEE Intemational conferences
on communications,
1979,
pp. l 1. 2. 1- 11. 2. 4.
26 C. A. Cooper,
"synchronization
for Telecommunications
in a Switched Digital
Network." I EEE Transactions
on Communications, July I9?9, pp' 1 028-1 033'
2':. C. Olszewski,
"AT&T
Synchroniaadon
Network
Architecture and Operations,"
hesentation
to 1999 NIST-TD{I
Workshop on Synchronization
in
Telecommunications
Systems,
Boulder, CO, March 9-ll' 1999'
zl
22
382 NETWoRKsYNcHRoNIzATIoNcoNTHoLANDMANAGEMENT
PROBLEIT,IS
7.1 Determine
the size
of an elastic store needed to accommodate
a velocity shift of
*10fi) km/hr that lasts for l0 sec if the data rate is l0 Mbps. (The
speed of light
i s3xl 08m/ sec. )
7.2 How many bits are needed in an elastic store designed to interface
the El digital
signal of ITU-T
to a digital switch? what is rhe maximum slip rate if the line
clock and switch clock differ by +50 to
-50
ppm (the
maximum recommended
offsets)?
7.3 what is the maximum (ggvo
probability)
phase
offset (in
signal intervals) pro_
duced by a
jitter
power
of + I 0 dB relative to I radz?
7.4 Determine
the rate
at which DSr signals in a DS2 multiplex lose framing be_
cause stuff codes
of a DS2 signal are incorrectly interpreted.
Assume the chan_
nel BER is l0-3. Assume the BER is 10-6
7'5 What would the rate of incorrect DSZ stuff code occurrences
be if a 5-bit stuff
code were used instead of the 3-bit stuff code? Assume the channel BER is l0-3.
7.6 A digital transmission link is to be used to hansmit blocks (packets)
of data with-
out slips. If the transmission
link is autonomously
timed with,"spe"t to the re_
ceiving terminal,
what is the maximum
allowable block length if the clocks vary
by
+50
ppm
each and an elastic store of 16 bits exists in the receiving
terminal?
Assume the elastic store is initialized
to half full between each block.
7.7 Assume the systematic jitter
from a single repeater produces
a symmeffic, worst
case phase
slope of 300 rad/sec for I msec. what is the peak-to-peak jitter
in
decibels relative
to a unit interval at the end of a line with 200 such repeaters?
7.8 Repeat Example 7.j for DSZ signals in a DS3.
7.9 A
jiuer
power
of 20 dB relative to I radz is observed at the receiving end of a
digital microwave link. What is the probability
that the phase
offset will exceed
14.0 symbol intervals?
7.10 Determine
the amount of phase
shift injected into a 2400-Hz carier signal by a
slip of one PCM sample.
7.tl
7.12
Determine
the TIE and the MTIE at the end of a 10-sec interval produced
by a
DS3 signal that has a constant offset of one part
in 106.
Repeat Problem
7. I I but assume a l
-MHz
jitter
component is added that has a
peak-to-peak
amplitude of 8 uls. Assume the starting phase
of the
jitter
compo-
nent is 0o.
7.13 Compare the slip rate of $tratum 3E with stratum 3 clocks in holdover
condi-
tions.
7.14 what is the oscillator accuracy (in ppm)
implied by the holdover requiremenr
of
a stratum 3 clock?
FI BER OPTI C
TRANSMISSION
SYSTEMS
The evolution of the common equipment
portions of the
public telephone
network
from analog technology
to digital technology began and ended with digital transmis-
sion systems:
Tl systems for short-haul
trunks and fiber optic systems
for long-haul
kunks. If fiber optic technology
had not emerged, extensive digitization of the long-
distance network would not have occurred as rapidly. High-bandwidth
coaxial cable
transmission
systems are too expensive to universally
replace radio systems, and ana-
log radios are more efficient than digital
radios in terms of voice circuits per pre-
viously available bandwidth.
Even though high-density
digital modulation techniques
and sophisticated
voice compression
algorithms
can make up for the bandwidth
inef-
ficiencies of digitized voice, any ultimate advantage of digital radios over analog ra-
dios is not nearly
great enough to walTant wholesale replacement
of the analog radio
backbone.
Fiber transmission
systems, on the other hand,
provide such dramatic
sav-
ings in equipment
and operational
costs that wholesale
deployment of high-density
routes with fiber systems was economically
justified'
As a note of historical
interest, Figure 8.1 shows a diagram of a
"photophone,"
pat-
ented by Alexander Graham BeIl in 1880. BeIl developed several
techniques for
modulating a light source
(the
sun) so as to directly vary the resistance
of a selenium
cell detector and thereby create an analog electrical
current for the speaker-
The modu-
lation mechanism shown in Figure 8.1 is a mirror that is vibrated by acoustic energy
to deflect more or less light to the receiver. It was almost exacfly 100
years later that
a suitable uansmission medium and suitable sources made optical transmission
practical-
The par"ticular characteristics of optical fibers that make them so useful for frans-
mission sy$tems are low loss, high bandwidth,
small
physical cross section, EMI im-
munity, and security.
Fiber Attenuation
As a result of the invention of the laser in 1960, materials scientists began searching
for optical transmission media that could be usefully applied as a communication
sys-
tem. The first instance of a
practical optical
fiber was announced l0
years later in 1970
383
384 FIBEH OPTIC TRANSMISSION
SYSTEMS
l . d. DELr.
Apprntur
for Bi gnrl hg
rl d 0onmul ortl l 3,orl hl
pLotopLoar.
t{o. ?3$rl 9g.
Frl rl trd D.c_ r_ rl rn
j 1
l :
Figure 8.1 Alexander Graham Bell's photophone.
[l].
This announcement
described a silica-based
fiber with
.,only"
20 dB/km attenu_
ation. In
just
a little less than l0 more years!
commercially
viable optical fibers with
0.2 dB/km attenuation had been developed
[2].
such remarkably low attenuation at_
tracted immediate
attention because it meant intercity transmission
links could be
traversed
with very few repeaters,
implying a dramatic
savings in equipment
and
maintenance.
As shown in Figure
8.2, the attenuation of an optical fiber is very dependent on the
wavelength of the light signal in use. Two wavelengths
that experience particularly
low attenuation in contemporary
fibers are 1300 and 1550 nm. Rlpresentative
attenu_
ations at these wavelengths
are 0.35 and 0.2 dB/km, respectively.
with an easily
achievable
net loss of 20 dB the latter attenuation value allows 100 km between re_
peaters
or amplifiers.
Because 0.2 dB/km is close to the minimum theoretical
attenu-
ation at 1550 nm, additional reductions in attenuation cannot be realized without
going
to higher wavelengths.
If fibers and associated
electronics can be developed to
operate at higher wavelengths,
much lower levels of attenuation may be achieved in
the future. The main application of such systems would be in submarine
cables where
the avoidance
of repeaters is most desirable.
FIBER OPTIC TRANSMIS$ION SYSTEMS 385
t o
5
t o
5
I
. 5
E 2
{ i
fA
E . 5
.s
8 . 2
5
6 . r
d
. t
8oo looo tt*
***l#r,
(n.l16ffi
18oo
Figure t.2 Fiber attenuation ver$us wavelength.
Bandwidth
When compared to elechomagnetic
transmission
media, the bandwidth of an optical
fiber is mind boggling: a single optical fiber operating at 1300 or 1550 nm of wave-
length has a
potential bandwidth of 20 THz
(20 x totz Hz), which is enough for 312
million 64-kbps channels. Bandwidth limits of fiber optic transmission
sy$tems are
mostly determined by the electro-optic drivers and receivers or the electronic inter*
faces to these devices. As described in Section 8. I . l, multimode fibers have an inher-
ent distance-dependent
bandwidth limitation, but this particular limitation
is avoided
in single-mode
fibers. Single-mode fiber systems also have a distance-dependent
bandwidth limitation. but this limitation
is as much a limitation of the optical sources
as it is a limitation of the fiber.
Small Physical Croee Sectlon
One of the most beneficial features of Tl systems is their ability to relieve over-
crowded cable ducts with a 12+o- I savings of copper
pairs compared
to two-wire ana-
log transmission. Fiber systems
provide the same advantage with respect to Tl
$ystems, only to a greater degree. A 25-pair copper cable has approximately the same
cross section as a fiber cable with 24 fibers and some copper pairs for line-powered
repeaters or amplifiers. The latter can easily carry 100,000 voice circuits. Further-
more, fiber sy$tems can sometimes be upgraded
for more capacity by merely installing
higher speed elecrronics
or wavelength division
multiplexing
(WDM). Thus, after
copper cable is replaced by a fiber cable, duct congestion
is unlikely to reappear
unless extensive
use of broadband
services
greatly expands the need for fiber fa-
cilities.
Electromagnetic lnterference lmmunity
Because glass has a very high resistance to electricity, the f,rber itself cannot pick up
interfering signals or noise or propagate harmful or damaging
transients to personnel
and equipment at the endpoints. The immunity to interference also means crosstalk is
386 FIBER oPTIc TRANSMIssIoN SY$TEMS
not a
problem
between multiple-fiber cables. In systems utilizing line-powered repeat-
ers, however, some amount of copper is included in the cable to carry power,
im-
plying
that the immunity, particularly
with respect to hazardous voltages, is
compromised.
The immunity of fibers to external interference
such as noise and crosstalk implies
there is no crosstalk-induced limit from high powered
transmitters
or more sensitive
receivers' Receiver
sensitivity
is ultimately limited by internal noise in the photode-
tectors, and transmit power
has technology limits due to spectrum spreading in the
sources and nonlinearities
in the fiber. Until these limits are reached, however,
the ab-
sence of crosstalk between fibers allows the capacity or transmission
distance of a fi-
ber system to be increased by merely upgrading the electronics, not the fiber.
Although crosstalk from one fiber to another is nonexistent,
crosstalk can arise be-
tween separate channels within WDM channels of a single fiber.
Security
Because optical fibers radiate no energy, noninvasive eavesdropping
ofthe signal is
impossible. Furthermore,
invasive taps are more diffrcult to implement
than are wire-
line taps, which merely require bridging across the conductors with a sufficiently
high
impedance to remove a usable but unnoticeable amount of signal energy. A similar
process
is possible
with optical fibers, but it requires bending the fiber a very precise
amount to allow a small amount of energy to escape and be amplified by a tap. Not
only does this process
allow passively
tapping a fiber, but also signals can be injected
into the fiber through such a bend. This technique has been used as a means of locally
testing the effectiveness
of a fiber splice and has been considered as a means of im-
plementing
distributed, passive
taps for fiber distribution to the home
[3].
8.1 FIBER OPTIC TRAN$MISSION
SYSTEM ELEMENTS
As shown in Figure 8.3, the basic elements of a fiber optic transmission
system are the
electrical-to-optical
transducer in the transmitting terminal, the optical fiter itself, the
optical-to-electrical
transducer in the receiver,
and signal processing
circuitry for am_
plifitcation,
clock recovery,
and data detection.
Regenerative repeaters require optical-
to-elechical
conversion for the signal processing
functions and conversion back to
optical for transmission.
Direct optical amplification with erbium-doped fiber am-
plifiers (EDFAs)
are used in lieu of repeaters, particularly
for wDlftransmission
links.
Flgure 8.3 Fiber optic transmission system elements.
8.1 FIBEH OPTIC TRANSMISSION SYSTEM ELEMENTS
Jsket
387
Figure 8.4 Optical fiber construction.
The system depicted in Figure 8.3 is not necessarily a digital one. Although all of
the fiber applications for telephony have utilized digital signals, analog hansmission
is possible. Analog FM modulation of optical signals has been successfully applied to
feeder applications
of CATV systems
[4].
Furthermore, sy$tems that carry 80 chan-
nels of AM video on fiber to residences have been developed, a capability that al-
lows analog television set$ to receive fiber-based cable TV without a digital video
decoder.
8.1.1 Optical
Fiber Fundamentals
The basic construction
of an optical fiber is shown in Figure 8.4. Both the core and
the cladding are tran$parent to the desired light signal but the cladding is designed with
a lower index of refraction, which causes most light waves in the core to be reflected
back into the core. As shown in Figure 8.5, whether or not a ray is reflected back into
the core is dependent on the angle at which it strikes the sore-cladding boundary.
If
the angle is too sharp, the ray is not reflected but
passes through the cladding and
is absorbed by the opaque,
protective
jacket.
The sharp angles can occur at two
places;
(l) near the source where all of the source'$ output is not focused into the
center of the fiber and
(2)
at bends, splices, or other imperfections in the fiber.
tllultimode Flbere
Although it is usually desirable to deliver as much signal energy as possible to the re-
ceiver. waves that reflect back and forth before reaching the receiver may be undesir-
able if they experience too much delay with respect to the
primary ray traveling down
ttre center of the frber. An optical fiber that allows rays to arrive at the receiver via mul-
tiple
paths is referred to as a multimode fiber. Multimode
fibers have core diameters
Jack6t
Chdding
Core
Figure 8;5 Fiber with multimode propagation.
388 FTBER oplc rRANSMtsstoN sysrEMs
that are large compared to the wavelength of the signal. A typical multimode fiber will
have a 50-pm core diameter and a 125-pm cladding diameter (such
a fiber is desig-
nated as a 5oll25 fiber). The delayed rays cause pulse
spreading referred to as multi-
mode dispersion. The significance of the spreading depends on the width of the pulse
or, conversely, the data rate being used. Multimode dispersion therefore creates an in-
herent operational limit defined as a bandwidth-di$tance
product (BDp).
The BDp of
a typical step-index multimode fiber is 13 Mbps-km
[5].
The previously
described multimode fiber is referred to as a step-index fiber be-
cause the index of refraction in the core is constant with a step change in the index oc-
curring at the core-cladding boundary.
Multimode dispersion can be significantly
reduced by varying the index of refraction
within the core so that a high value occurs
in the center and a low value occurs at the edge. Because the speed of propagation
of
light is higher in lower indices of refraction, rays that reflect back and forth within the
core ffavel at an average speed that is greater
than a primary
ray that remains entirely
within the center. Thus, if the index of refraction is carefully graded
wirhin the core,
all rays can be made to arrive at the receiver with the same amount of delay. such a
fiber is refened to as a graded-index
fiber. A representative BDp of a graded-index
fiber is 2 Gbps-km
[6].
Notice that this is an improvement of more than two orders of
magnitude over typical step-index multimode fibers.
Example 8.1. Derermine the loss limir and the multimode dispersion limit of a
graded-index
multimode fiber system operating at 820 nm and providing
a bandwidth
of 90 Mbps (enough
ro carry two DS3 signals). Assume that the difference between
the available output power
from the source and the input power
required by the
receiver for an acceptable maximum error rate is 42 dB.
solution. From Figure 8.2, the attenuation of a multimode fiber operating at
g20
nm
is approximately 3 dB/km. Thus,
. . 42
Losshmi t =8. 0=t +t t t ' t
Using 2 Gbps-km as a typical BDP of a graded-index
multimode fiber, the multimode
dispersion distance limit is determined
as
Dispersion limit
=
#
=
ZZ.2 km
The results of Example 8.I are representative
capabilities of first-generation
fiber
systems such as the FT3c system of AT&T
[7].
In actual practice,
repeater spacing
would be less than the l4-km loss limit to allow margin for, for example, compo-
nent tolerances, splicing, and aging. The first FT3c system, which began service
between New York city and washington, DC in February 19g3, had repeater spac-
8.1 FIBER OPTIC TRANSMISSION SYSTEM ELEMENTS 389
ing of 7 km-the
location of repeaters in a coaxial cable system replaced by the fiber
system.
Slngle-Mode
Fibars
As the first-generation
fiber systems were being btought
into service in the long-
distance network, optical technology
had already advanced to the point that deploy-
ment of second-generation
systems was underway. One key technology of the second
generation was the use of single-mode fibers
(SMFs)
that
provide two distinct advan-
tages. First, SMFs have smaller diameter cores
(8 pm typically) that restrict propaga-
tion to a fundamental mode and therefore eliminate multimode dispersion.
Second,
SMFs have less internal
(Rayleigh)
scattering,
which reduces the attenuation by about
50Vo
16,81.
For example, the attenuation of an SMF operating at 820 nm is about 2
dB/km, as opposed to 3 dB/km for a multimode fiber.
Another key development of second-generation
$ystems involved the use of longer
wavelengths
(1300
nm), which, as shown in Figure 8.2, experience significantly
less
attenuation than the 800-900-nm wavelengths
used in first-generation systems. Use
ofthese wavelengths required new technology for sources and receivers, as described
i n Secti ons 8.1.2 and 8.1.3.
Chromatic
Dlsperaion
The combination of eliminating
multimode dispersion and using a wavelength with
much lower signal attenuation revealed another ffansmission
limitation referred to as
chromatic
dispersion. Chromatic dispersion arises when a photonic signal contains
more than one wavelength
and the individual wavelengths
propagate at different
speeds. Thus, chromatic dispersion
is the photonic equivalent ofphase distortion
(also
called delay distortion)
in electromagnetic
(wireline) propagation. Dispersion levels
are quantified by a dispersion coefficient
of picoseconds per nanometer
per kilometer'
A representative value for the SMF at 1550 nanometers is 16
psec/nm km.
The effect of chromatic
dispersion is minimized by using one or more techniques.
First, operation can be centerecl about a wavelength that exhibits a small amount of
chromatic dispersion.
Typical silica fibers, for example,
produce approximately
15
times less chromatic dispersion at 13fi) nm than at 1550 nm. Second, an optical source
should be chosen that is as $pectrally
pure as
possible (e.g., has a naffow spread of
wavelengths). Third, nalrow retum-to-zero
(RZ) pulses can be used to
prevent in-
tersymbol interference.
Fourth, a dispersion-compensating
fiber can be included in the
transmission
path that has a dispersion slope in opposition to the main fiber. When
just
the first two techniques are used, BDPs of 250 GHz-km are
possible
[6]'
Exampte 8.2. Determine the loss limit and the chromatic dispersion
limit of a
high-performance SMF optical fiansmission
system operating at 1300 nm and
providing a bandwidth of 417 Mbps
(enough
to carry nine DS3 signals). Assume a
narrowband $ource is used with an output
power that is 42 dB
gteater than the
minimum acceptable receive
power
(i'e.,
assume the system
gain to be 42 dB).
390 FIBEH oPTIc TRANSMISSIoN
$YSTEMS
solution. As indicated in Figure 8.2, the attenuation of a single-mode fiber
operating at 1300 nm is approximately 0.35 dB/km. Thus,
. . 42
Loss l l mi t
=
0j 5
=
120 km
Using 250 Gbps-km as the BDp of a silica, single-mode fiber,
Chromatic dispersion limit
=
#:600
km
When the dispersion limit is much higher than the loss limit, as is the case in Exunple
8.2, it indicates that the optical sources are providing
a niurower specffum than is nec-
e$sary. Thus, the system design could be changed to use less expensive sources or the
power
output of the source could be increased to get
a longer loss limit. (Increasing
the power
output of the $ource may widen its spectrum and thereby reduce the BDp.)
The system parameters
of Example 8.2 are representative
of the second-generation
FT series G systems deployed by AT&T
[9].
ttre Fr series G systems had maximum
repearer spacing of 48 km (29
miles), indicating that the initial systems could be in-
stalled with significant
margin for reliability and future upgrades. Many of the routes
were, in fact, upgraded to 1.7 Gbps without changing the fiber
tIO,
I il. Even at the
higher data rates, the FT series G routes had significant loss margin for splicing and
the insertion of wavelength division multiplexers
as another means of upgrading the
system. wavelength
division multiplexing is described in section 9.3.
Example 8.2 shows that attenuation is the limiting distance factor in 1300-nm sys-
tems operating at low and intermediate
data rates. To operate with even longer dis-
tances, Figure 8.2 indicates that 1550 nm should be used. with typical silica fibers,
the limiting factor at 1550 nm would be chromatic dispersion. To ouirco-" this limi-
tation, two other types of fiber have been developed; a dispersion-shifted single-
mode fiber (DS-SMF)
that has minimal dispersion ar 1550 nm and a dispersion-
flattened single-mode
fiber (DF-sMF)
that has less dispersion
across a range of
wavelengths. Figure 8.6 contrasts the dispersion of both of these types of fibers with
a conventional SMF' Even
though Figure 8.6 indicates zero dispersion exists at 1300
nm on the sMF and at 1550
nm on the DS-SMF, some affount of dispersion has to be
assumed due to fiber manufacturing
variations and uncertainty
of the actual center
wavelength produced
by a source. values of l-3.5 psec/km
nm are typically used as
the dispersion coefficient
at the
"zero"
dispersion points.
8.1.2 Electrical.to.Optlcal
Transducers
Two basic types of semiconductor
devices convert electrical signals into optical sig-
nals and can have the output
coupled into an optical fiber: laser diodes (LDs)
and light-
emitting diodes (LEDs).
LDs generally provide
befter performance
in terms of higher
output power'
Breater
bandwidth, and narrower signal spectrum. LEDs, on the other
8.1 FIBER oprc rHANsMlsstoN sYsrEM ELEMENTS 391
20
E
E r o
J
t
E 0
.E
.t
-to
a
-20
Wrydength
(nml
Figure 8.6 Dispersion of SMF and DS-SMF fibers.
hand, are less expensive,
require only simple interface circuitry, afe more tolerant of
environmental
conditions, and are
generally more reliable. Thus LDs are used for long
transmission distances and LEDs are used when ruggedness and interface co$ts are
more important than performance.
To achieve maximum distances between repeater$,
it is necessary to minimize
chromatic dispersion by using narrowband sources. Various forms of LDs have been
developed that come close to achieving
a spectrally pure output. Figure 8.7 shows the
typical output spectrum of one the most popular types of
"single-ftequency"
devices:
a Fabry-Perot laser diode. Another type of diode with a very nalrow spectrum
is a
diskibuted-feedback
LD
(DFB-LD). As indicated, the dominant mode of a Fabry-
Perot diode is typically 10 dB stronger tltan the adjacent modes. To maintain the spec-
trum indicated, it is generally nece$sary to tightly confol the operational environment
(bias points, modulation level, and temperature). Otherwise, modulation-dependent
effects broaden the spectrum or even cause very shoft duration
modes shifts that
pro-
- 2 - l o + t + 2
Wwsler|g8h lrom Cenur warbltngth
(nml
Ftgrrre 8.7 Representative
laser output $pectrum.
6
s
$
D
-2
d
E
.E
392 FIBER OPTIC TRANSMISSION
SYSTEMS
TABLE 8.1 Repreeentatlve
parameters
of Varloue Optical Sources
Device Type
Launched Output FWHM Spectrum
Wavelength (nm)
Power (dBm)
Width'(nm)
Si LED
Ge LED
InGaAsP LED
DFB LD
DFB LD
IUDFB LD
850
1 300
1 300
1 300
1 550
1 550
- 16
- 19
- 10
-5
-5
+2
50
70
120
1. 0
0.4
0.8
duce a dominant mode at one of the adjacent wavelengths. This latter phenomenon
is
referred to as mode partition
noise
[12],
which can cause a halving of the BDp of a
system
[6].
The spectral width of an optical source is commonly specified as the full-
width at half-maximum (FWHM),
which represents the spread between the wave-
lengths at which the specrrum is ar half the magnitude of the peak.
To achieve greater
distances in loss-limited systems, either a more sensitive re-
ceiver or a higher powered
source is necessary.*
Laser diodes provide
the highest
available powers,
as indicated in Table 8. l, which lists representative parameters
of
various LD and LED source$.
Systems that are dispersion limited cannoi be improved
by higher output powers
because the intersymbol interference dominates the imper-
fections in the received signal.
The effective transmit power
depends on the coupling efficiency between the
source and the fiber. The coupling efficiency may be as low as l%o in the case of an
unfocused LED connecred
to a single-mode fiber. High-efficiency
couplings of 6ovo
are possible
with focused sources. The launched power
values provided
in Table
g.I
include the coupling efficiencies.
Bandwi dth-Dlstance
P rod ucte
The bandwidth-distance
factor of a fiber system resulting from chromatic dispersion
is determined from the fiber dispersion coefficient and the spectral width of the source.
The maximum acceptable amount of pulse
spreading with an NRZ line code is typi-
cally specified to be one-foufth ofa bit interval T. Thus.
DL AL!0.257
where D
=
dispersion coefficient of fiber (psec/km
nm)
Z
=
distance (km)
A.l,
=
spectrum width of source
(nm)
7= duration of a bit interval (psec)
( 8. 1)
Transmit power
cannot tle increased arbitrarily without encountering nonlinear phenomena
in the fiber
itself. The optical power
threshold where nonlinear effects begin to occur may bJas low as l0 mW
[6],
8.1 FIBER OPTIC THANSMISSION SYSTEM ELEMENTS 393
Using the relationship that the bandwidth
(B) is the reciprocal of the bit interval, the
BDP is determined as
NRZ-BDP=Bi
Gbps-km
(8.21
A50Vo RZ line code allows twice as much pulse spreading because energy in one in-
terval is half a bit period from energy in the next interval. Thus
<
250
-
DAl.
RZ-BDP.
#
Gbps-km
(8.3)
The RZ line codes are obviously prefened in dispersion-limited sy$tems but are infe-
rior to NRZ line codes in loss-limited systems because the receiver must detect
pulse
with half as much energy
(assuming
the
peak transmit
power is held constant).
Example8.3. Determine the BDP of a 1550-nm
SMF system and a 1550-nm
DS-SMF $ystem using the 1550-nm DFB LD source identified in Table 8.1. Assume
an NRZ line code and a
"zero"
dispersion value of 3.5
psec/km nm to allow relaxed
tolerances for the fiber and the source operating wavelength.
Solution, From Figure 8.6, the dispersion coefficient of an SMF fiber at 1550 nm
is 16 psecflrm nm. From Table 8.1, the spectral width of the source is 0.4 nm. Thus,
?qn
NRZ-BDP=ffi=39GbPs-km
Using the given
"zero"
dispersion value, the BDP of the DS-SMF system is deter-
mined as
NRZ-BDP=5ft=
179 Gbns-km
8.1.3 Optical-to-Electrlcal Transducers
Two basic fypes of
photodetectors are available as transducers to convert the optical
energy at the receiver to electrical energy for amplification and other
processing such
as clock recovery and data detection. Silicon based p-i-n diodes were the first
photode-
tector$ to be used, operating in systems using wavelengths of 800-900 nm. These di-
odes are inexpensive and reliable and provide good performance. Their major
drawback is that they do not operate at higher wavelengths where fiber loss is mini-
394 FIBER OPTIC THANSMISSION
SYSTEMS
TABLE 8.2 Rspresentative Parameters
of Optlcal Detectors (BER =
1g-tt1a
Device Type Wavelength (nm)
Receive Power (dBm)
Data Rate (Mbps)
Si pFn
Si APD
InGaAs pr:n
InGaAs APD
InGaAs APD
InGaAs APD
InGaAs pr-n
InGaAs APD
850
850
1310
1310
1 310
1 550
1 550
1 550
-48.0
-58.0
-35.0
-43.0
-26.0
-37.5
-37.0
-33.0
50
50
420
420
8000
678
1 200
4000
aThe
detector sensitivities assume an NF|Z line code. The sonsitivities decrBase by
g
dB for
(S0%)
BZ codes.
mized. Germanium devices function at the higher wavelengths and are more sensitive
but are more temperature dependent and less reliable.
The second basic type of photoderector
is an avalanche photo
diode (ApD),
which
enhances the receiver sensitivity because it operates with internal gain. (A p-f-n diode
does not have internal gain
and therefore requires all amplification to be externally ap-
plied,
which raises the noise level.) The amplification inherent in the ApD's conver-
sion from an optical signal to an electrical signal is useful because it means the ApD
can be 10-15 dB more sensitive in detecting low-level signals at a given error rate. A
major drawback of an APD is ttrat it needs a high supply voltage to operate and is very
sensitive to temperature. APDs tend to have somewhat lower GBps thanp-i-n diodes,
which restricts their use in very high data rate systems. Table
g.2
lists various types
of photodetectors
and provides
representative performance parameters
for each.
6
E
.E
,E
6
a
-60L
.001 . 0 1 . 1 . 2 . 4 t 2 4
OEtr R{te
(Gbm}
Figure 8.t Receiver sensitivity versus data rate.
8.2 LINE coDES FOR FIBER oPTIC THANSMISSIoN 395
Both types of receivers utilize direct detection, which means that they merely mea$-
ure the
presence or ab$ence of optical energy to detect data. Because higher data rates
imply less energy per bit, the sensitivity decreases with data rate. As a first approxi-
mation, a doubling of the data rate implies the receiver becomes 3 dB less sensitive.
The general formula for determining the change of sensitivity
(in decibels) as a func-
tion of a change in the data rate is
Sensitivity R
=
Sensitivity Ro +
where R
=
new data rate
Ro
=
data rate at
given sensitivity
Figure 8.8 plots the receiver sensitivity ofrepresentative
p-i-n
diode detectors and
APD detectors as a function of the data rate for both NRZ and RZ line codes.
8-2 LINE CODES FOR FIEER OPTIC TRANSMISSION
Line codes for fiber optic transmission
(FOT)
systems must satisfy the same basic re-
quirements
as the line codes in electrical
(wireline)
transmission sy$tems described in
Chapter 4. Some fiber systems use the line codes described previously. A few line
codes developed specifically fbr fiber applications are described here. Optical sources
and detectors are primarily
used in nonlinear modes of operation with significant gain
and threshold variations, which implies they are best suited to operating in only two
states: on or off. Hence, a two-level line code
(on-off
keying) is most natural. Con-
sidering the wide bandwidth available, multilevel line codes are usually unnecessary
and extra timing transitions can be included in the line code without significant pen-
alty. An exception to the availability of
"free"
bandwidth occurs in submarine systems
where maximum repeater spacing is achieved by minimizing the bandwidth of the op-
tical signal. These systems are typically dispersion limited because the need for ultra-
high reliability precludes the latest single-frequency lasers with unproven longevity.
A
particularly useful property ofdirect detection optical receivers is the absence of
any polarity ambiguity, which means the transmitter and receiver are inherently dc
coupled so differential encoding is unnecessary. Even though the optical signal is in-
herently dc coupled, the dc balance of the line code may still be a consideration be-
cause the gain
of some photodiodes (particularly APDs) is sometimes dependent on
the dc level of the line code. Furthermore, it is usually desirable to ac couple amplifier
stages in the receivers
[3],
which means electrical pulse amplitudes at the detecting
comparator$ are dependent on optical
pulse
densities. In all, these constraints are typi
cally easier to deal with than their counterparts in wireline sy$tems, so line codes for
fiber systems generally tolerate
Breater
amounts of dc variation. Notice that, in fiber
systems, both a Manchester and a CMI line code have unvarying dc levels equal to
one-half of the
pulse amplitude. A number of other line codes used in fiber systems
allow some variation in the average dc level. Because these variations are controlled,
(8.4)
tot"*,.[ftJ
396 FIBER opTtc rRANsMtsstoN sysTEMS
the line codes are sometimes refemed to as
"dc-constrained"
line codes as opposed to
being perfect
or near-petfect
"dc-balanced"
line codes. The dc variations that do exist
in these line codes either are too small to cause problems
or are accommodated by pro-
viding some form of dc restoration
[14].
For applications that do not require maximum repeater spacing or maximum data
rates typically use Manchester (diphase)
or cMI line codes because both of these
codes provide
strong timing content and a completely constrained dc level (502o
on).
An early example of the use of cMI is a high-speed intraoffice link developed by
NTT of Japan
[5].
Manchester and CMI line codes experience the worst case of both the loss limit and
the dispersion limit. The worst-case loss limit arises because the receiver must detect
optical pulses
with duration equal to half the bit interval (i.e.,
with a sensitivity of an
RZ code). The dispersion limit of a Manchester or CMI line code is actually half of
the NRZ dispersion limit because the symbol rate is essentially doubled. (Each
bit in-
terval is divided in half, with the need to derecr a
pulse
or no pulse
in both halves.)
Thus, the Manchester or CMI BDP can be determined as
MC-BDP
=#
Gbps (8.s)
From the previous
discussion it can be seen that when performance
is more impoftant
than low cost, some other line code is needed. Three basic line codes are commonly
used: scrambled NRZ, scrambled RZ, or some variation of a code referred to as an
mBnB line code, described below. The choice of an NRZ or RZ line code is primarily
based on whether the system is loss limited or dispersion limited. Loss-limited sys-
tems use NRZ line codes whereas dispersion-limited systems use RZ codes. A scram-
bler is typically used to enhance timing recovery and control dc wander in a statistical
sense (but not in an absolute sense). Often, overhead data patterns
in the data stream
provides
some assurance that the data stream cannot exactly match the scrambler se-
quence
so a ceftain minimum number of transitions are assured. This is the case in
SONET transmission systems described in Section 8.5.
8.2.1 mBnB Li ne Codee
As an alternative to relying on scramblers for ensuring timing transitions, a class of
line codes has been developed that encodes rn binary data bits into blocks ofn binary
line bits
(mBnB),
where nr < n. Because only 2* data codes must be selected from 2,
codes in each block, there is coding flexibility for conholling timing and dc wander.
For example, if m
=
4 and n
=
5, sixteen data codes must be chosen from the set of
thirtytwo 5-bit line codes. A specific choice of these codes is provided
in Table
g.3,
which depicts the data codes and control codes selected fbr the fiber-distributed data
interface (FDDD
standard
[6].
The significant a$pects of the coding assignments in Table 8.3 are:
g,z
LINE coDEs FoR FIBEH oFTrc TRANSMtssroN 397
TABLE 8.3 FDDI4BSB Llne Codee
Line Gode
Decimal Binary Symbol Name
Functional Assignment
0
1
2
3
4
5
6
7
E
I
1 0
1 1
12
1 3
1 4
1 5
1 6
1 7
1 8
1 9
20
21
22
23
24
25
26
27
28
29
30
31
00000
00001
0001 0
0001 1
001 00
001 01
00110
00111
01 000
01001
01010
01011
01100
01101
0 1 1 1 0
0 1 1 1 1
1 0000
1 0001
10010
10011
' 101
00
10101
1 0 1 1 0
1 0 1 1 1
11000
11001
1 1010
1 1 0 1 1
11100
1 1 1 0 1
1 1 1 1 0
1 1 1 1 1
o
V
V
V
H
L
V
R
V
1
4
5
V
T
E
-l
K
B
q
2
3
A
B
J
S
c
D
E
F
o
I
Quiet
Violation
Violation
Violation
Halt
Violation
Re$et
Violation
Violation
Violation
Line state symbol
Oisallowed
Disallowed
Disallowed
Line state symbol
Start delimiter
Disallowed
Control indicator
Disallowed
Data symbol 0001
Data symbol 0100
Data symbol 0101
Disallowed
End delimiter
Data symbol 01 10
Data symbol 01 1 1
Disallowed
Start delimiter
Data symbol 1000
Dala symbol 1001
Data symbol 001 0
Data symbol 001 1
Datasymbol 1010
Data symbol 101 1
Start delimiter
Control indicator
Datasymbol 1100
Data symbol 1 101
Data symbol 1110
Data symbol 1111
Data symbol 0000
Line state symbol
Set
Set
l dl e
L The code space is used for control as well as data.
2. When transmitting data, the minimum pulse density is 40Vo and there can never
tre more than three intervals without a
pulse.
3. The dc component is constrained to fall between a minimumof 40Vo pulses and
a manimum af 60Vo
pulses, a range that is one-fifth of an unconstrained random
data stream.
398 FIBER oPTIc THANSMISSIoN SYSTEMS
TABLE 8.4 Digital Blphaee (Manchester)
lB2B Line Code
Line Code
Binary Functional Assignment
0
1
2
e
00
01
1 0
1 1
Disallowed
Data symbol 0
Data symbol 1
Disallowed
The features ofthe 4B5B
line code described above are achieved at the expense of
a25vo inwease in the line data rate. By way of comparison, a digital biphase (Man-
chester) line code and the cMI line code described in chapter 4 can both be repre-
sented as lB2B line codes wirh the coding assignments provided
in Tables 8.4 and
g.5,
respectively. Notice that both of these codes come at the expense of a lfi)7o increase
in the line datq rate.
Because the 4B5B line code defined for FDDI is intended for a specific application,
it contains codes for control as well as for data. The 5B68 codes given
in Table 8.6
are intended for transmission only and therefore do not allocate code space for control.
Notice that the 5B68 code assignments are made in such a way that the dc level is
fixed at 507o pulses but the maximum run length of no pulses is 6. Thus, tighter control
of the dc level comes at the expense of increasing the worst-case duration between
pulses.
5868 and TBBB line codes have been used exten$ively in Europe. Examples
of 7B8B usage are a 565-Mbps rerrestrial system developed by British Telecom
[17]
and a 280-Mbps NLI submarine sysrem developed by src of Great Britain
[18].
The 4B5B and 5B6B examples use output blocks that are only I bit
greater
than
the input blocks. considerably more coding flexibility is achieved when the ourput
block is more than I bit longer than the input block. As an example, a 6B88 code al-
lows all 64 input codes to be encoded with an output code containing exactly four I's
in every code. (The number of combinations of four I's in 8 bits is 70.) Thus dc wander
can be maintained without having to altemate between low-density and high-density
codes but coding efficiency is sacrificed.
TABLE 8.5 Coded Mark Inversion (CMl)
1B2B Line Code
Line Code
Binary Functional Assignment
0
1
2
3
00
01
1 0
1 1
Data symbol 1
(if
11 previously
sent)
Data symbol 0
Disallowed
Data symbol 1
(if
00 previously
sent)
8.2 LINE coDE$ FoB FIBEH OPTI0 THANSMISSIoN 399
TABLE 8.6 5868 Line Code Aesignrnentss
58 lnput 68 Output 5B Input 68 Output
00000
00001
0001 0
0001 1
00100
001 01
001 10
0011' l
01 000
01 001
01010
01011
01 100
01101
0 1 1 1 0
0 1 1 1 1
011101/ 100010
1 01 1 1 0/01 0001
01 01 1 1/1 01 000
0001 1 1
1 01 01 1/01 01 00
001 01 1
001 1 01
001110
110101/ 001010
01 001 1
01 01 01
010110
01 1 001
01 1010
011100
1 1 01 1 0/001 001
1 1 1 01 0/0001 01
1 0001 1
1 001 01
100110
101001
101010
101100
0l 1011/ 100100
1 1 0001
110010
110100
101101/ 010010
111000
1 1 001 1/001 1 00
101101/ 0' t 0010
1 001 1 1/01 1 000
1
0000
1
0001
10010
10011
10100
10101
1 0 1 1 0
1 0 1 1 1
1 1000
11001
1 1 0 1 0
1101 1
1 1 1 0 0
1 1 1 0 1
11
' t
10
1 1 1 1 1
aA
code 1/code 0 output must alternate between the code with four 1's and the code wlth two 1's to mainlain
dc balanc.
An 8B 108 coding affangement
fl
9, 201 has been defined as the fiber channel trans-
mission standard
(ANSI
X3.230-1994). As indicated in Table 8.7, each input byte is
separated into a 5-bit field and a 3-bit field that are respectively encoded using a 5B68
and 3B4B algorithm. In some cases, two codewords are provided for encoding par-
ticular input data. The choice of one code in a
pair (which are complements of each
other) is made to maintain dc balance. [f the
previously transmitted unbalanced code
had more I
's
than 0's, the code with fewer 1's is chosen. If the previously transmitted
unbalanced code had more 0's than I's. the code with fewer 0's is chosen.*
8.2.2 Bit Insertion Codes
The mBnB block codes described in the previous section have one significant disad-
vantage: They are difficult to implement on very high speed data streams.
(In lower
speed applications, decode logic or table lookup in
a small read-only memory is triv-
ial.) Very high speed links therefore use another type of code referred to as bit inser-
tion codes. These codes are actually special cases of mBnB codes with n
=
m + 1 and
the codes selected so the coding/decoding logic is greatly simplified.
The first bit insertion code to be considered is the MB lP code, which merely inserts
an odd parity
bit after every rz bits of source data. Odd
parity ensures that at least one
*Two
special cases exist that may alter the rule when 001 I I is 5B6B encoded or 0t I is 3B4B encoded.
400 FIBEB OPTIC THANSMISSION SYSTEMS
I is included in the
parity
field of m + I bits. As a point
of reference, notice that the
digital biphase (Manchester)
line code is a 18 lP line code. An example of the use of
a 24BlP line code is the Trans Pacific submarine cable system (Tpc-3) put
into serv-
ice in December 1988 by AT&T and KDD of Japan
[21].
An even simpler bit insertion code described by engineers of NTT in Japan
[22]
is
the mB I c code, which merely inserts a bit after every ru source bits-the value of
which is the complement of the iilrmediately preceding
bit. Thus, the added bit always
forces a data transition and is very easy to implement. An obvious disadvantage of the
simplicity is the loss of performance monitorability over the first rn
-
1 bits of a block.
Again notice that a diphase line code is a degenerate case of an mB lc line code with
ru
=
L An example of the use of a l0BlC line code is the F-1.6G sysrem of NTT in
Japan
[23].
58 Input 68 Output 58 Input
68 Outpul
0 (00000)
1 (00001)
2
(00010)
3 ( 00011)
4 (001
00)
5
( 00101)
6 (001 1 0)
7
( 001
11)
I
(01ooo)
I
( 01001)
10 ( 01010)
11 ( 01011)
12 ( 01100)
13 ( 01 101)
14( 01110)
1 5 ( 0 1 1 1 1 )
100111/ 011000
01 1 1 01i 1 0001 0
101101/ 010010
1 1 0001
110101/ 001010
1 01 001
01 1 001
111000/ 000111
111001/ 000110
1 001 01
01 01 01
110100
001 1 01
101100
01 1 100
010111/ 101000
16 (10000)
17 ( 10001)
1B ( 10010)
19 ( 10011)
20
( 10100)
21 ( 10101)
22 ( 10110)
23 ( 10111)
24 (1 1000)
25( 11001)
26
( 11010)
27 ( 1
101 1)
28( 11100)
2 e ( 1 1 1 0 1 )
30 ( r 1110)
3 1
( 1 1 1 1 1 )
011011/ 100100
10001 1
01001 1
110010
001 01 1
101010
011010
1 1 1 01 0/0001 01
1 1001 1/001 100
100110
010110
1 1 01 1 0/001 001
001110
1 01 1 1 0/01 0001
011110/ 100001
101011/ 010100
38 Input
48 Outout
0
(000)
1 ( 001)
2 (010)
3
( 01
1)
4 ( 100)
5 ( 101)
6
( 110)
7 ( 1 1
1 )
01 00/1 01 1
1 001
01 01
1 1 00/001 1
0010/1 101
101 0
01 10
0001/1 1 1 0
TABLE8.7 88108 FlberChannet Codtng
8.3 WAVELENGTH DIVISION MULTIPLEXING 401
8.3 WAVELENGTHDIVISIONMULTIPLEXING
Wavelength division multiplexing
(WDM)
is the basic technique used to establish
multiple, independent optical channels on a single fiber. The concept of WDM is il-
lustrated in Figure 8.9 showing the coupling
(multiplexing) of separate sources into a
fiber and the separation
(demultiplexing) of the signals out of the fiber. Prismatic re-
fraction is depicted as the basic coupling mechanism at both ends of the link. Refrac-
tion is usable only for channels with relatively large wavelength separations.
Diffractive
grating
devices, on the other hand, can operate with channel spacing on the
order of 1-2 nm
[24].
All WDM devices share the common characteristics of being
purely passive
and being reversible so any particular device can
perform either mul-
tiplexing or demultiplexing functions. Other than some optical insertion loss and some
crosstalk from imperfect separation, the multiplexing/demultiplexing operations are
transparent to the individual channel signals. The insertion losses of the diffraction
grating devices can vary from I to 7 dB depending on fiber size (multimode vs. single
mode) and number of channels
[24].
When the f,rrst fiber routes were installed, WDM wa$ not used because a single
high-speed transmitter and receiver was generally less expensive than separate pairs
of lower speed transrnitters and receivers and accompanying WDM devices. Thus,
WDM would typically be installed as a subsequent upgrade to increase the capacity
of a system without laying new fiber. For example, the original 90-Mbps, 820-nm
FT3C system installed in 1983 by AT&T in the northeast corridor was upgraded in
1984 by adding an additional 180 Mbps on a 1300-nm carrier
[25].
In addition, refer-
ences
[9]
and
[0]
describe how the FT Series G can have its capacity increased by
multiplexing a 1550-nm system onto an existing 1300-nm system and/or adding
higher speed electronics without taking the existing system out of service.
The
passive property of WDM devices is a dominant attraction for some applica-
tions such as fiber to the home. Allocating individual fibers from a cenffal office to
each home is expensive, impJying the need for some form of fiber sharing. The outside
plant environment
(-40-70"C)
and the need for remote power for active components
imply that passive WDM devices are prefened. The use of WDM in this application
is sometimes referred to as the
passive photonic loop
(PPL)
[26].
WDM is fundamentally identical to FDM as used on electrical
(copper)
or electro-
magnetic
(radio)
transmission systems. In fact, using the following relationship allows
wavelengths to be related to frequency:
v
=,fl (8.6)
Flgure 8.9 Wavelength division multiplexing.
4OZ FIBER oplc rHANsMtsstoN sysrEMs
where
y=
velocityof light,
=
3 x 108 m/sec
,f
=
frequency (Hz)
L.
=
wavelength (m)
optical systems are defined in terms of wavelength as opposed to frequency because
optical $ources are traditionally specified in terms of the wavelengths of their outputs.
Nevertheless, systems with extremely close wDM channel spacing (on
the order of
0.04 nm) are sometimes referred to as FDM systems
[27].
systems that use electronic multiplexing to produced
electrical FDM signals
for modulation of a single optical carder are referred to as subcarrier multiplexing
systems. Because the individual channels of these systems are typically close to-
gether
in frequency, and hence in wavelength, passive
separation is usually infea-
si bl e.
Example 8,4. Determine the difference in wavelength of two optical signals
separated by 2GHz and centered at 1500 nm.
Solution. The frequency (in free space) ofa 1500-nm signal is
f=#-=2oo,oooGHz -
1500x l 0-o
Thus the upper and lower frequencies are determined as 200,001 and 199,9g9 GHz,
respectively. The corresponding wavelengths are
I r =
3 x l OE
199.999 x l 0e
=
1500.007 nm
^ 3 x 1 0 8
n':
too.oot ^, ,t
=
1499'992 nm
and the diff'erence in wavelengths is 0.015 nm.
Example 8.4 shows that even when subcarriers are separated by zGHz,the result-
ing wavelength difference is small-too small for passive
demultiplexing.
Wavelength demultiplexing
with passive
diffuaction does not inherently cause a
loss of signal power
in the individual channels. An altemative approach to demulti-
plexing
is shown in Figure 8. 10, which involves power
splitting of the received signal
followed by wavelength filtering to extract individual channels. This approach is pas-
sive and functionally identical to diffractive separation but is wasteful of optical
power. Its main advantage is that power
splitting can be implemented as
passive
taps
distributed along the fiber roure. Thus, it is a useful technique in applications (like lo-
cal area networks) where distances are less important than flexibility in network to-
pology.
8.4 FIBEH SYSTEM DESIGN 403
Figure 8.10 WDM with power splining/filtering.
Dense Wavelength Division Multiplexing
Major advances in WDM technology occurred in the late 1990s, primarily in response
to the needs of common carriers to dramatically expand the bandwidth of their fiber
facilities for high-bandwidth data networking. In contrast to
previous systems, which
carried a small number of WDM channels, the newer systems carried a large number
of closely spaced wavelengths. These systems are generally referred to as dense wave-
length division multiplexing
(DWDM)
systems. An example of a first-generation
(1996) DWDM system is the MultiWave 1600 $ystem from Ciena Corporation. This
system provided 16 channels spaced 0.8 nm apafr in the region of 1550 nm. DWDM
systems are enabled by optical amplifiers
(EDFAs)
that transparently amplify all
wavelengths in the band and by fiber Bragg gratings fabricated into glass fiber for de-
multiplexing and filtering in a receiver.
DWDM usefulness is not confined to retrofitting of existing plant. DWDM may be
the only possible
way to achieve extremely high bandwidths. A 4O-channel system op-
erating at 2.488 Gbps
(OC-48) per channel provides an aggregate bandwidth of 100
Gbps-a difficult speed for a single TDM channel, particularly on dispersionlimited
fibers and with other high-speed limitations such as polarization mode dispersion. An
additional advantages of DWDM is the inherent transparency of individual wave-
lengths, which allows transmission and interoperability of mixed types of services.
Yet another advantage of DWDM is the inherent reliability provided by separate elec-
tronics for each wavelensth interface.
8.4 FIEER SYSTEM DESIGN
The examples in the previous sections highlight only the most basic aspects of fiber
optic technology. This section provides system-level considerations for the design of
practical FOT systems. As is the case with any transmission technology, the system
design is strongly dependent on system availability requirements and cost of mainte-
nance. At the low end of these requirements are easily accessible f,rber connections to
individual workstations within a single building. At the other end of the spectrum of
reliability requirements are submarine cables that are obviously very expensive to re-
pair. Repairs are minimized by using only ultrareliable components. Even though sub-
marine cables are immune to bacftfioes* and
gophers, they must contend with oceanic
'Some
outside plant
enginecrs ofthe telephofle companies wouldprobablynot be too surprisedifabackhoe
could somehow get
to a submarine cable.
404 FTBER oplc rRANsMtsstoN sy$rEMS
counterparts: fishing trawlers and sharks (which
seern attracted to electromagnetic ra-
diation). AT&T developed a
"seaplow"
for burying submarine cables in areas of large-
scale fishing activity
[28].
About the only advantage that submarine applications have
in their favor is the constant, relatively low temperature environment for repeaters/am-
plifrers.
8.4.1 Fiber Connectors and Sptices
Three basic techniques ofjoining fibers are connectors, mechanical splices, and fusion
splices
[29].
connector$ are used for terminal connectionrr, patch panels,
or other in-
stances where loss (0.4-l
dB) is less important than ease of maintenance and recon-
figuration.
Mechanical splices involve mechanical alignmeut and clamping of two adjoining
fibers. There are many different types of mechanical splicing equipment that provide
various levels of loss depending on the fiber size
(multimode
or single mode) and in-
stallation time. Splices can be made on individual fibers or in bulk on fiber bundles or
ribbons. Two basic splicing procedures
are possible: passive
or active. Passive splices
are faster because the craftsperson merely places the fibers in the alignment sleeves
and clamps the fibers in place. Active alignment involves passing
a signal through the
splice and making fine mechanical adjustments to minimize the loss before the final
clamp is applied. A single fiber splice described in reference
[30]
provides
an average
of 0'2 dB loss when passively installed and 0.05 dB loss when actively installed. The
installation times are 5 and 7 min, respectively.
Fusion splices provide
the lowest losses because they essentially form one continu-
ous fiber that is almost as good
as a single-drawn fiber. In fact, submarine fibers are
pretested
at the factory for tensile strength and fused at the weak points where breaks
occur
[28].
This process
ensures that the fibers can withstand the tensile stress encoun-
tered when laying and recovering the fiber cable but introduces slightly higher levels
of average attenuation in the fiber. For example, the I 04-km repeaterless system in-
stalled between Taiwan and the island of Peng Hu has 0.24 dB/km ar 1550 nm includ-
ing all splices
[28].
Fusion splicing equipmenr is available rhar is completely
automated with high-resolution TV used to align the fibers and estimate the splice loss
by measuring the misalignment after fusion
[29].
Fusion losses can vary between 0.01
and 0. I dB. For undersea applications, only values close to 0.0 I dB are accepted.
8.4.2 Protection Switchlng
Protection switching within an FoT system is basically the $ame as in other high-
capacity transmission $ystems wherein one spare line, including transmitters, repeat-
ers, and receivers, can be switched into service if one of Nmain lines fails (l : N
protection).
There are, however, a few unique considerations for protecting fiber sys-
tems, arising primarily
because of the high capacity of the sy$tems. First, a system can
be installed with lower
cost
(lower
performance)
optoelectronic devices with an up-
grade
for higher performance
anticipated at a later date. In this case, the protection
8.4 FIBEH SYSTEM DESIGN 405
switching procedures and packaging concepts should allow one-at-a-time upgrading
of electronics without disrupting service
[11].
A more significant and difficult aspect of
protection switching arises because fiber
routes tend to carry large-traffic volumes representing aggregations of traff,rc that
would otherwise be carried by separate facilities on $eparate route$. One-for-N
pro-
tection of most fiber systems is of no use when a backhoe cuts an entire cable. Protec-
tion from the loss of a complete facility requires network-level routing diversity, a
feature of a network that tends to disappear when using fiber and large switching sys-
tems to their maximum economic advantage-large hubs fed by large trunk groups.
A network architecture that
provides 1 ; I route diversify and no hub dependence is a
SONET ring described later in this chapter.
8.4.3 System Galn
The fust step in determining the repeater spacing of an FOT section is to determine
the difTerence between the launched output power of a hansmitter and the receive
power required for a designated minimum BER. In analogous fashion to the same term
introduced for radio systems, this parameter is referred to as the system gain. Notice
that the system gain includes coupling losses at the optical source and optical detector.
The system
gain must be greater than the sum of all the losses in the
path. Sources of
loss include inherent attenuation of the fiber, splicing/connector losses, bending
losses, WDM device losses, and photonic switch losses if any. Splicing losses that oc-
cur at regular intervals during the installation of the fiber are often added into the fiber
attenuation so there is only one distance-dependent
parameter involved in determining
repeater spacing. The difference between the sy$tem gain and the sum of all the nomi-
nal losses is the loss margin. In any
particular installation, the loss margin allows for
equipment manufacturing tolerances,
powerpenalties due to dispersion, unanticipated
splices, component aging, and possible WDM upgrades.
Example 8.5. Determine the system gain,
the BDP, the dispersionlimited repeater
spacing, and the loss margin for an FOT system with the following parameters: data
rate
=
565 Mbps, line code
=
5BdB RZ, wavelength
=
1550 nm, source
=
-5
dBm
DFB-LD with 0.4 nm FWHM, fiber= SMF, detector=InGaAs APD, repeater spacing
=
65 km, and splicing lo$ses
=
0.2 dB/km.
Solutinn, The use of the 5BdB line code implies the Iine data rate of 565(6/5)
=
67t
Mbps. The receiver sensitivity for 678 Mbps is determined from Figure 8.8 or Table
8.2 as
-34.5
dBm. Thus.
System gain
=
-5 -
(-34.5) =
29.5 dB
Because an RZ line code is in use.
406 FIBEH oplc rRANSMtsstoN sysTEMS
BDP:
: ; so?,
=73. 6Gbps-km
17 xO. 4
BDPspaci ng
=m:
l 09km
Path loss
=
(0.2 + 0.2X65)
=
26 dB Loss margin
=29.5 -
26
=
3.5 dB
Example 8.5 is representative of the 565-Mbps system developed by Fujitsu of Japan
for carrying an E5 digital signal (565.148
Mbps)
[3
I
].
The fact that the repearer spac-
ing is 607o of the BDP limit implies that dispersion penalties
are small (estimated
to
be 1.2 dB).
8.5 SONET/SDH
The first generations
of fiber optic systems in the public telephone network used
pro-
prietary
architectures, equipment, line codes, multiplexing formats, and maintenance
procedures.
Some commonality with other systems in the network came from suppli-
ers who also supplied digital radio system$. In these cases, the multiplexing formats
and maintenance protocols
emulated counterparts in the radio $ystems, which also had
proprietary
architectures. The only thing in common with all of the radio and fiber sys-
tems from all of the suppliers was that the interface to the network wa$ $ome number
of DS3 cross-connect signals. Proprietary multiplexing formats for multiple DS3 sig-
nals evolved because there was no higher level standard compatible with the applica-
tions. A DS4 signal, which is composed of six DS3 signals, requires too much
bandwidth for radio systems and carries a larger cross section of channels
(4032)
than
needed in many applications.
The Regional Bell operating companies and interexchange carriers (IXCs),
the us-
ers of the equipment, naturally wanted standards so they could mix and match equip-
ment from different suppliers. This became particularly
important as a result of
competition among the IXCs who desired fiber interfaces to the local exchange carri-
ers (LECs)
but did not want to necessarily buy from the same suppliers as the LECs.
(It might be necessary for an lxc ro interface with a different supplier at each LEC.)
To solve these problems,
and others, Bellcore initiated an effort that wa$ later taken
up by the Tlxl committee of the Exchange carriers standards Association (ECSA)
to establish a standard for connecting one fiber system to another at the optical level
(i.e.,
"in
the glass").
This standard is referred to as the synchronous optical network
(soNET)
t32,
331. In the lare srages of the development of this srandard, ccITT be-
came involved so that a single international standard exists for fiber interconnect be-
tween telephone networks of different counties. Internationally, the standard is known
as the synchronous digital hierarchy (sDH)
t341.
The soNET srandard addresses rhe
following specific issues:
l. E$tablishes a standard multiplexing format using some number of 5l.84-Mbps
(STS-1)
signals as building blocks.
2. Establishes an optical signal standard for interconnecting
equipment from
different suppliers,
8.5 SONET/SDH 407
TABLE 8.8 SONET $ignal Hlerarchy
North American Designation
Electrical Signal Optical Signal Data Rate (Mbps) ITU-T Designation
STS-1
STS-3
srs-12
STS.24
STS-48
STS-S6
STS-19?
oc-1
oc-3
oc-12
oc-24
oc-48
oc-96
oc-192
51. 84
155.52
622.08
1244.16
2488,32
4976.64
9S53.28
STM.1
STM-4
STM-8
STM-16
STM-Sz
STM.64
3. Establishes extensive operations, administrations, maintenance, and
provisioning (OAM&P)
capabilities as part of the standard.
4. Defines multiplexing formats for carrying existing digital signals of the
asynchronous multiplexing hierarchy
(DSl,
DSlC, DSz, DS3).
5. Supports CCITT
(ITU-T)
digital signal hierarchy
(E1,
E2, E3, E4).
6. Defines a DSO identifiable mapping format for DSI signals.
7. Establishes a flexible architecture capable of accommodating other applications
such as broadband ISDN with a variety of transmission rates. Wide-bandwidth
signals
(greater
than 5l.84 Mbps) are accommodated
by concatenating multiple
STS-1 signals. A STS-3c signal, for example, is an 155.52-Mbps signal that is
treated by the network as a single entity.
The data rates and signal designations of the SONET hierarchy are
provided in Table
8.8. At the lowest level is the basic SONET signal referred to as the synchronous trans*
port signal level I (STS-I). Higher level signals are referred to as STS-N signals. An
STS-N signal is composed of Nbyte-interleaved STS-1 signals. The optical counter-
part of each STS-N signal is atr optical carrier level N signal
(OC-N).
Table 8.8 also
includes ITU nomenclature for the SDH, which refers to signals as synchronous trans-
port modules N
(STM-N).
Because corlmon applications of the ITU signal hierarchy
cailrot efficiently use a 51.84-Mbps signal, the lowest level STM signal is a 155.52-
Mbps
(STS-3c)
signal.*
Although the SONET specification is primarily concerned with OC-N interconnect
standards, STS-1 and STS-3 electrical signalsr within the SONET hierarchy are useful
within a switching office for interconnecting network elements
(e.g.,
multiplexers,
switching machines, and cross-cormect systems)
t35,
361.
.A
51,84-Mbps STS-I SONET signal is sometimes referred to as a STM-O SDH signal.
TAn
STS-1 electical signal uses a B3Zs line code and a STS-3 electrical signal uses a CMI line code.
408 FIBER oPTIc TRANSMISsIoN SYSTEMS
44.738 Mbpe
51,84O MbpB
sTs-t
81"840 Mbpl
CEPT' 4
139,?64 Mhpr
Figure 8.11 Functional block diagram of SONET multiplexing.
8.5.1 SONET Multiplexlng
Overview
The first step in the soNET multiplexing process (shown
in Figure
g.ll)
involves
generation
of a 51.840-Mbps sTS-l signal for each tributary. The sTS-l signal con-
tains the tributary (payload)
traffic plus
transport overhead. As indicated in rhe figure,
a variety of tributary type$ are accommodated;
1. A single DS3 per
srs-l that can be a srandard asynchronous DS3 signal
generated
by an M13 or M23 multiplexer. Asynchronous DS3 inputs are passed
transparently through the system to a DS3 output. Because this transparent
option exists, arry 4.736-Mbps
signal can be carried within the payload
envelope.
2. A
group
of lower rate tributaries such as DSl, DSlc, DS2, or El signals can be
packed
into the STS-I payload.
3. A higher rate (wideband)
signal can be packed
into a multiple number of
concatenated srs-l signals. Prevalent examples of higher rate signals are
139.?64-Mbps fourth-level multiplexes of ITU or a broadband ISDN signal at
150 Mbps. Each of these applications requires three STS-1 signals concatenared
together to form an sTS-3c signal. Higher levels of concatenation (to form
srs-Nc signals) are possible for higher rate tributaries. concatenared srs-l
signals contain intemal control bytes that identify the signal as a component of
Byte
Interlosrrgd
Mux
8.5 soNET/$Dn
tt09
a higher speed channel so the integrity of the concatenated data can be
maintained as it
passes through a network.
An STS-N signal is created by interleaving bytes from N STS- l signals that are mu-
tually synchronized. All timing
(frequency)
adjustment is done in generating each of
the individual STS-I signals. STS-I signals that originate in another SONET node
with a possibly different frequency are rate adjusted with the equivalent ofbyte stuff-
ing (described
later) to become synchronized to the clock of the local node. No matter
what the nature of the tributary traffic is, all STS-I kibutaries in a STS-N signal have
the same high-level format and data rate.
Optical carrier level-N signals are
generated by first scrambling the STS-N signal
(except for framing bytes and STS-ID bytes) and then converting the electrical signal
to an optical signal. Other than scrambling, the OC-N signal is generated with direct
conversion to an optical signal. Thus, the data rates, formats, and framing of the STS-
N and OC-N signals are identical.
A SONET $ystem is defined as a hierarchy of three levels-sections, lines, and
paths-as indicated in Figure 8.12. Each of these levels has overheadbandwidth dedi-
cated to administering and maintaining the respective level. As indicated in Figure
8. 1 1, one of the overhead functions provided within an STS-N signal involves calcu*
lation and transmission of a
parity byte for the entire STS-N signal. Parity is also de-
fined for the other levels ofthe architecture as described in the following section.
8.5.2 SONET Frame Formats
The frame format of an STS-1 signal is shown in Figure 8.13. As indicated, each frame
consists of9 rows of90 bytes each. The first 3 bytes ofeach row are allocated to trans-
poft overhead with the balance available for
path overhead and
payload mapping. The
Path
Terminating
Equipment
Line Scction
Terminating Torminatirg
Equipment Equipment
Sction
Terminating
Equipfiont
Line PEth
TBrminsting Terminating
Equipment Equipment
Figure 8.12 SONET system hierarchy.
410 FIBER oPTIc TRANSMIsSIoN SYSTEMS
Sction
O\rrrhrld
Llno
Owrheld
I Rowe
3 Columns 87 Columns
Figure 8.13 STS-I frame format.
tran$port overhead is itself composed of section overhead and line overhead. Path
overhead is contained within the information payload
as indicated.
The 9 rows of 87 bytes
(783
bytes in all) in the information payload
block is re-
ferred to as the envelope capacity. Because the frame rate is 8 kHz, the composite data
rate of each STS- I signal can be represented as the sum of the transport overhead rate
and the information envelope capacity:
STS-1 rate
=
overhead rate + information
envelope rate
=
9 x 3 x I x 8000+9 x 87 x 8 x 8000
=
I-728 x 106 + 50.112 x 106
=
51.840 Mbps (8.7)
The internal format of the envelope capacity is dependent on the type of tributary traf-
fic being canied. one aspect of the envelope format that is common to all types of traf-
fic is the 9 bytes of path
overhead indicated in Figure 8.13. The actual location and
purpose
of this overhead are described in the next two sections.
As a specific example of a higher level (srs-N)
signal, Figure 8.14 depicts the de-
tails of an srs-3 signal that also represents the STM-I signal format in rru terminol-
ogy. Transmission of the bytes occurs row by row and left to right. Thus, the first 3
bytes of an srs-3 frame are the three framing bytes Al, Al, Al. Most of the secrion
and line overhead functions within an sTS-3 signal are carried in the srs-l number
I overhead. Thus many of the corresponding bytes of the other srs-l signals are un-
used and are so designated with an asterisk. Notice, however, that path overheacl is in-
cluded in the information envelope for each of the STS-I signals.
After a frame of an srs-N signal is scrambled, a parity byte (BIp-8)
is generated
that provides
even parity
over corresponding bits in all bytes of the sTS-N frame. This
parity
byte is inserted into the section overhead of the first STS-I signal of the next
STS-N frame.
Tranrpon
Ovshord
sTs-l
lnformstion Payload
At ta. c1
81 El FT
Dt 02 D3
J1
B3
c2
Gl Path
F2 Owrhear
H4
B
z4
z5
Hl HZ H3
82 K1 K2
D4 D6 D6
07 D8 D9
D10 Dl t D12
z1 t2 E2
Trunsport Overhead ST$"3 lnformrtlon Payload
\ l A r A r A 2 A 2 A e C l C l C l
81 El FI
D l * r D 2 D 3
Jl Jl Jl
83 83 83
c2 c2 c2
Gl Gl Gl
F2 F2 F2
H4 H4 H4
z3 z3 23
z4 7A 24
z5 z5 z5
H3 HI Hl Hl H2 H2 H2 H3 H3
82 82 82 t(1
r +
l(2
D 4 D S r r D e
07 D8 D9
D 1 0 " D l l r * p 1 2 t
21 z1 21 zt z2 22 E2
:1)
(2) (3) (1) (2) (3) (1) (2) (3)
I Columns
( 1) ( 2) ( 3) ( 1) ( 2)
261 Columns
8,5 SONET/SDH 411
Flgure 8.14 STS-3 frame format.
8.5.3 SONET Operations, Adminietratlon, and Malntenance
The SONET standald places significant emphasis on the need for operations, admini-
shation, and maintenance (OAM)
of an end-to-end system. As shown in Figure 8.15,
the OAM architecture is based on the section, Iine, and path Iayers described
pre-
viously. OAM standardization is a requirement for mixing equipment from multiple
vendors and ease of management of all levels of a system (an individual repeater sec-
tion or an end-to-end
path).
Sectlon Overhead
The functional allocation of the 9 bytes of section overhead in each STS-1 frame
shown in Figure 8.13 are;
Al Frami ng byte
=
Fd hex
(l 1110110)
A2 Framing byte
=
28 hex (00101000)
Cl STS-l ID identifies the STS-l number
(1,
. . .
,1f)
for each STS-l within
an STS-N multiplex
B I Bit-interleaved parity byte providing even
parity over previous STS-N
frame after scrambling
El SectionJevel64-kbps PCM orderwire
(local
orderwire)
Fl A 64-kbps channel set aside for user purposes
D1-D3 An 192-kbps daia communications channel for alarms, maintenance,
control, and administration between sections
The fact that there is such a richness of maintenance support at the section level
(from
one repeater to another) is indicative of the recognized need for extensive OAM fa-
cilities and the availability of economical technology to
provide it.
E
a?
$
(t
p
u
h
G
o
ts
o
F
rrl
z,
(n
ln
od
P
AD
ll
412
8.5 sONEr/SDn 413
Llne Overhead
The functional allocation of the l8 bytes of line overhead in each STS-I frame shown
in Figure 8.13 are as follows:
Hl*H3 Pointer bytes used in frame alignment and frequency adjustment of
payload data; the functions of these bytes are defined in detail in Section
8.5.4
82 Bit-interleaved
parity for line-level error monitoring
Kl, K2 Two bytes allocated for signaling between line-level automatic
protection switching equipment
D4*D12 A 576-kbps data communications channel for alarms, maintenance,
conffol, monitoring, and administration at the line level
ZI.ZZ Reserved for future use
E2 A 64-kbps PCM voice channel for linelevel orderwire
Notice that the line-level OAM facilities are similar to those available at the section
level with the addition of the
protection switching signaling channel and Hl, H2, and
H3
pointer bytes use for payload framing and frequency adjustment.
Path Ovarhead
As indicated in Figure 8 . I 3
,
there are 9 bytes of
path overhead included in every block
(9 x
87 bytes) of information
payload. The important aspect of this overhead is that it
is inserted when the tributary data are
packed into the synchronous payload envelope
(SPE)
and not removed
(processed)
until the tributary data are unpacked. Thus, it
pro-
vides end-to-end OAM support independent ofthe path through the synchronous net-
work, which may involve numerous intermediate multiplexers, cross-connect
switches, or add-drop multiplexers. The exact location of these 9 bytes within the
pay-
load envelope is dependent on pointer values defined in the next sectron. The functions
of the path overhead bytes are:
Jl A 6a-kbps channel used to repetitively send a 64-byte fixed-length
string so a receiving terminal can continuously verify the integdty of a
path; the contents of the message are user
programmable
83 Bit-interleaved
parity
at the path level
CZ STS
path signal label to designate equipped versus unequipped STS
signals and, for equipped signals, the specif,tc STS
payload mapping that
might be needed in receiving terminals to interpret the payloads
Gl Status byte sent from path-terminating equipment back to
path-originating equipment to convey status of terminating equipment
and
path enor
performance (received
BIP error counts)
F2 A 64-kbps channel for path user
H4 Multiframe indicator for
payloads needing frames that are longer than a
single STS ftame; multiframe indicators are used when packing lower
rate channels
(virtual
tributaries) into the SPE
Z3-ZS Reserved for future use
414 FIBER OPTIC TRANSMISSION SYSTEMS
Payload Framlng and Frequency Ju$tification
8.5,4
Payload Framing
The location ofthe 9 bytes ofpath overhead in the sTS-l envelope is not defined in
terms of the srs-l transport framing. Instead, the
path
overhead is considered to be
the first column of a frame of data referred to as the spE, which can begin in any byte
position
within the sTS-l payload
envelope (see
Figure 8.16). The exact location of
the beginning of the sPE
(byte
Jl of the
path
overhead) is specified by a
pointer
in
bytes Hl and H2 of the STS line overhead. Notice that this means that an SPE typically
overlaps two STS-1 frames.
The use of a pointer
to define the location of the SPE frame location provides
two
significant features. First, SPE frames do not have to be aligned with higher level mul-
tiplex frames. It may be that when first generated,
an SPE is aligned with the line over-
head at the originating node (i,e.,
the poinrer value is 0). As the frame is carried
through a network, however, it ardves at intermediate nodes (e.g.,
multiplexers or
cross connects) having an arbitrary phase
with respect to the outgoing hansport fram-
ing. If the sPE had to be frame aligned with the outgoing signal, a full spE frame of
storage and delay would be necessary. Thus, the avoidance of frame alignment allows
SPEs on incoming links to be immediately relayed to outgoing links without artificial
delay. The location ofthe sPE in the outgoing payload
envelope is specified by setting
the Hl, H2 pointer
to the proper
value (0-782).
The second advantage of the pointer
approach to framing spE signals is realized
when direct access to subchannels such as DSls is desired. Because the pointer pro-
vides immediate access to the start of an SPE frame, any other position
or time slot
within the SPE is also immediately accessible. If the tributary uses a byte-synchronous
mapping format, individual channel bytes have fixed positions
with respect to the start
of the sPE. This capability should be compared to the procedures
required to demul-
tiplex a DS3 signal. In a DS3 signal there is no relationship between the higher level
framing and the lower level DS2 and DSI framing positions.
In essence, two more
frame recovery processes
are needed to identify a DSO time slot. The use of pointers
Frame 0
9 Row*
Frame I
I Eowe
87 Columnr--+
.fl
H;
F
{ h
=
'Psth
Overherd
87 Columnr
] , " . -
Figure 8.16 Representative location of SPE.
8.5 SONET/SDH 415
in the SONET architecture eliminates the need for more than one frame recovery proc-
ess when accessing byte-synchronous lower level signals.
Freq uency J uetiflcatl on
Although it is generally intended that SONET equipment be synchronieed to each
other or to a cofilmon clock, allowances must be made for the interworking of SONET
equipment that operates with slighfly different clocks. Frequency offsets imply that an
SPE may be
generated
with one clock rate but be carried by a SONET transport run-
ning at a different rate. The means of accommodating a frequency offset is to accept
variable SPE frame rates using dynamic adjustments in the SPE
pointers. Pointer ad-
justments
allow SPE frames to float with respect to the transport overhead to maintain
a nominal level of storage in interface elastic stores. Figure 8.17 shows the basic
means of accommodating a slow incoming SPE. If the elastic store begins to empty,
positive
byte stuffing is invoked to skip one information time slot
(the
slot immedi-
ately following the H3 byte) and simultaneously
incrementing the pointerto delay the
SPE frame by one byte.
Negative byte stuffing, to accommodate a fast SPE clock, requires sending an extra
SPE byte whenever the elastic store begins to fitl. As indicated in Figure 8.18, the H3
slot carries the extra byte of data, which requires the pointer to be simultaneously de-
cremented, thereby advancing the SPE frame by I byte. To protect against errors in
misinterpreting byte-stuffing
operations, the need to increment or decrement a
pointer
is redundantly encoded,* and the new
pointer value is transmiffed for a minimum of
three frames following the frame in which the
pointer adjustment occurs. This implies
that a l-byte adjustment can be made once every four frames
(once
every 500
psec).
For an analysis of the effects of channel errors on SONET
payload pointers, see ref-
erence
[37].
Example 8.6. Determine the range of SPE data rates that can be accommodated by
the byte-stuffing operation described above.
Solution. Four SPE frames nominally contain 4 x 9 x 87
=
3132 bytes of data. Thus,
the nominal SPE rate is 8 x 3132 x 2000
=
50.112 Mbps. When
positive byte stuff,rng
is used to accommodate a slow incoming SPE rate, 313I bytes of data are transmitted
in four frames. Thus, the lowest, slip-free rate is
Minimum sPE rate
=
8 x 3131 x 2000
=
50.096 Mbps
When negative byte stuffing is used to accommodate a fast SPE rate, 3133 bytes of
data are transmitted in four frames. Thus, the highest, slip-free rate is
Maximum SPE rate
=
8 x 3133 x 2000
=
50.128 Mbps
.During
the a
jusrnent
frame only, 5 even-numbered bits of the pointet value are invefied to indicate a
negative stuff
(data
in byte H3). When a
positive stuff occurs, 5 odd-numbercd bits of the pointer arc
inverted,
416 FIBER oplc rBANSMtsstoN sysrEMS
STS-Frsme
I
-----l*-- -
I
Strn of ST*t Synchronour Emretope
0 p
Frlme z
FredG n + |
Pwl th
1
bi$ invtrted
Frome n + 2
Pf f i
-P+
|
Ftlme n * 3
50Orr
Figure 8.17 Positive STS- I pointer justifi
cation operation.
Example 8.6 demonstrates that the SONET clock accuracy required for maintaining
sPE data is 50.112 t 0.016
Mbps-a very wide torerance of f320 ppm. In compari-
son, a soNET node is specified to maintain a minimum timing accuracy of z0 ppm
if
it loses its reference. Thus, the frequency of timing adjustments was chosen more
from a desire to simplify the
process
than from
just
assuring a margin for slip-free
operations.
The use of byte stuffing to accommodate timing differences introduces waiting
time
jitter
into SPE payloads, just
as bit stuffing introduces waiting time
jitter
into DS I
signals being asynchronously multiplexed into DS2 or DS3 signals. If the spE is car-
rying
DSl traffic, the effect of byte sruffing at the spE rate is about one-fourrh of a
bit interval at the DS 1 rate. (Because
the SPJE rate is 32 times the DS I rate, the duration
of an sPE byte is one-fourth the duration of a DSI bit.) Notice that this amount of
Poimgr
Vr l ue ( P|
I
-l*
I
Start of $TS-l Synchronous
B.s $oNET/SDH
417
STS-l
Framd
0rtg
Freme a
Frame il +
I
Pwi t h
D
bits ifiYErtd
Freme a * 2
Frerne r + 3
500 gr
Figure 8.18 Negative STS- I pointer
justification
operation.
phase jitter
is comparable to the amount introduced by bit stuffing a DS2 signal into
a DS3 signal.
8.5.5 Virtual Trlbutaries
To facilitate the transport of lower rate digital signals, the SONET standard uses sub-
STS-I payload mappings referred to as virtual tributary
(VT)
structures, as shown in
Figure 8.19. This mapping divides the SPE frame into seven equal-sized subframes or
VT blocks with 12 columns
(108
bytes) in each. Thus, the subframes account for
7 x 12= 84 columns with the path
overhead and two unused columns
(reserved
bytes
R) accounting for the remainder of the 87 columns in an SPE. The rate of each VT
structure is determined as 108 x I x 8000
=
6.912 Mbps.
(\l
c{
(l
FI
ct
gr
f E 5 6 l D E
ro
ro
f c 5 6 0 ! l
al
.8
E - -
3
E F
tr
b
5
o
C'
F
q EB6 f l T f f i f i H
(\|
o
rJt
+
(rI
(\|
q
o
. F
H
c { E
V L H
s ' F
E r r
= 5
#
v =
So
EL
.F
t ! +
E , F
( 9 E
l- rrl
. 5 E
i c n
c\
I
J
F o O
E
FD
F i i
(o
ro
+
b s
E
i n
E
f;a
llr
h :
418
S.5 SONET/SDH 419
TABLE 8.9 SONET Virtual Tributarles
Tributary Type
VT
Designation
Number of Number in VT Maximum
Columns
per VT Group Number in SPE
DS1
E1
DSl C
D52
wl . 5
w2.0
VTs.O
VT6.O
4
3
2
1
28
21
14
7
The VT structures can be individually assigned to carry one offour types ofsignals.
Depending on the data rate of a
particular
signal, more than one signal may be caried
within a VT structure as a VT group. All signals within a VT group must be of the
same t)?e, but VT
groups within a single SPE can be different types. The particular
lower rate signals accommodated as VTs are listed in Table 8.9. The last column in-
dicates how many of the lower rate $ignals are carried in a single SPE if all seven VT
groups are the same type.
VT-SPE
payloads
are allowed to float within an STS-I SPE in the same fashion as
pointers to SPE payloads are allowed to float at the STS-1 level. Thus, a second level
of pointer logic is defined for VT payloads. Again, a floating VT-SPE allows for mini-
mal framing delays at intermediate nodes and for frequency
justification
of VT-SPEs
undergoing transitions between timing boundaries. High-rate VT-SPEs are accoillmo-
dated by inserting an information byte into V3 while slow-rate VT-SPEs are accom-
modated by stuffing into the information byte immediately following V3 when
necessary.
The mapping format for a VTl.5 is shown in Figure 8.20. Each VTI.5 uses three
columns of data to establish 108 bytes in a VT1.5 payload. There are four such pay-
loads in a l2-column VT group. The Vl, V2, V3, V4 bytes of the
payload have fixed
positions within the STS- I payload. The remaining 104 bytes of the VT1.5 signal con-
stitute the VTI.5 payload, the start of which is the V5 byte
pointed to by Vl and V2.
Figure 8.21 depicts two different mappings for a VTl.5 payload: an asynchronous
mapping and a byte-synchronous mapping.
Aeynchronous Mapping
The asynchronous operation is identical in concept to the bit-stufflng operation de-
scribed in Chapter 7. The DSI bit stream is inserted into the information bits
(I)
with
no relationship to the VT-SPE frame or byte boundaries. As indicated, there are two
stuffing opporhrnities (Sr and 52) available in every four-frame superframe. Thus, the
VTl.5 superframe carries 771,772,or 773 information bits depending on the value of
the stuff control bits C
1
and C2. The nominal number of information bits in each frame
is 193 x 4=772. Nominal frames carry information in 52 while stuffing in 51.
Because the asynchronous operation is compatible with the asynchronous
network,
it is the format used in most SONET applications. The major advantage of the asyn-
chronous mode of operation is that it
provides for totally transparent fransmission of
3
4
b
1 2
42O FIBER oPTIc TRANSMISSIoN $YSTEMS
eCoUmfiB
)offiool
nmootl0
ffil1
l--fr4.l
rooooffi
I Rorva
g
Rorys
i l EeR.un
ffi
ffi
n
n
tr
tr
l) A Wl.f rm lhree mluffir of m SPE fr f 08 bytca in a
j00
Dre. f,perFrffio.
?) Vl
,
V2, V3, V4 bytcr hsve ffrfid locdiilB h I $pts idmified by tho hf t$o b&B dH4.
3) Vt ed V2 poid' to V5
qftirh
is rhe fr* tryte ofba flocing WiJ Sff.
4) SEE oncrtod bl,tff V5, .E!, 26, Z7 mr h idediaal rd*ive porilima
of m SFE .
Figure E.20 Superframe sffucture for VTI.S kibutaries.
the tributary signal in terms of information and in terms of information rate. The major
disadvantage of the asynchronous mode is that 64-kbps DSO charurels and signaling
bits are not readily extracted.
Eyte-Sy n c h ro n o u s M u lti
p
I exl n g
In contrast to the asynchronous mapping, the byte-synchronous payload
mapping
shown in Figure 8.21b allocates specific bytes of rhe payload
to specific bytes (chan-
nels) of the DSI tributary signal. Hence, this mode of operation overcomes the main
drawback of the asynchronous mode in that 64-kbps DSO channels and signaling bits
within the payload
are easily identified. In fact, when the DSl tributary arises from
legacy applications, the signaling bits of a DSI are moved from the least signif,rcant
bit
(LSB)
of every sixth frame of respective channels and
placed
in dedicated signal-
ing bit positions
within rhe vr-sPE. Thus byte-synchronous
multiplexing offers an
additional feature of converting from in-slot signaling to out-slot signaling for DSI
signals.
Al important aspect of the byte-synchronous format shown in Figure 8.21b is the
absence of timing adjustments for the source DS I signal. Thus, the DS I interface nec-
essarily requires a slip buffer to accommodate a DSI source that may be unsynchron-
ized to the local soNET clock. Although slips in byte synchronously mapped DSI
v5
R R R R R R I R
lffi I htts
(246)
J2
c . G o o o o I
R
lga I bftB
(246)
z6
c . G o o o o I R
194 | bltl
(2/sd)
z7
c . c ' R R R
q S R
lm I bfts
(24d)
t " -
24 Eyteg
2 Byte8
24 BytEs
2 Eyteg
24 BytB
2 B!,toB
24q,tss
1(}{Bytee
s.5 soNET/sDn 421
\15
R
Pn
$
S" Sn$r F R
ClS0 clHnneh
(1-24)
J2
R P^ sr s" s" g, F R
DSO channels
(1-24
ze
e P " S ' S " 9 " $ r F
R
DSo channl8
(1-24't
z7
F. P" Sr Sr 8. g, F R
DSO chsnnolB
(r.24)
Sr, Sr" $r, 8r
-
Chmel-asso+iEtrd
sigtalingbiB
P1, Ps
=
Si8nrlingphuc indicdsr bits
F=DSl *aftinebit
R bih $e ItuEd srtrbitr
(uspociffed)
No, oflixod informdim bils
=
4(19?)
+3(1,
=
77 I
$1, $1
-
Surfflinfrnuatim bitr
Cr, c!
-Sutrmgor*rot
bits
R bitc ffe SxEd rtlfftte (uupdcifiod)
O bits arsrctcrved fm futurupurpo*e
(a) (b)
Figure 8.21 DSI mappings in VTl.5 SPE:
(a) asynchronous;
(b)
byte synchronous.
signals may occur at the SONET network interface
(e.g.,
SONET
gateway), slips can-
not occur within the SONET network because internal nodes rate adjust the VT1.5
payloads with pointer adjustments.-
El Mappinge
El signals are mapped into VT2 signals with the same basic procedures used for DS ls.
As shown in Figure 8.22, the VT2 signal is composed of four columns of bytes in an
STS- l that produce
a total of 144 bytes. After removing the Vl, V2, V3, and V4 bytes,
the VT2
payload has 140 bytes. Formats for asynchronously mapped EIs and byte syn-
chronously mapped Els are shown in Figure 8.23. Notice that the byte-synchronous
mapping for a 30-channel
El carries channel-associated
signaling in slot 16-the
form of out-slot signaling designed into El signals at their inception. The same basic
format supports common-channel
signaling, which is sometimes referred to as a
'
The original specifications for SONET included a locked timing format for VT-SPEs that eliminated the
VT
pointers so DSO channels could be identified directly within the STS-1. This mode of operation has
since been abandoned,
422 FIBER OPTIC THANSMISSION SYSTEMS
4Columne
I Rorrs
noom0l
gRUfYE
EoomdO
9RsrYs
)ffi|1
eRorw
Ndcr: l) A VTz'|H fo|f oolumr o,f m $pE fu 144 bytes itr r Jfi) rxe, speftfi.
2) Vl, VA V3, V4 byta have fi xed locd,iqs in r SPE i.ft'niffd by th lrd two tfrs of H4.
3) Vl sd V2 psid' to Vi wtid is the fir* byAa ofihe f,o*hg Vfit $fB
4) SFE ovcrfad b]ics V5, J2, 26, Z7 ooon h iders,icnl rddilo porttime
of m SpE .
Figure 8.22 Superframe skucture for VTZ tributaries.
3l-channel El format. In this case channel 16 is the ccs channel and channels l-15
and 17-31 are the bearer channels. Thus, the multiplex mapping is not changed,
just
the nomenclature of the channels and the spE type designation in the vr path
over-
head byte V5.
8.5.6 DS3 Payload Mapping
The previous
section describes several alternatives for packing
virtual tributaries into
an sTS-l envelope. when all seven vrs in an envelope are vrl.Ss, a total capacity
of 28 DSls is provided-rhe
same as a DS3 signal. Thus one method of carrying a
DS3 signal involves demultiplexing it into its consrituent DSI (or DSZ signals) and
packing
the constituents as virtual fributaries. This approach is attractive in that the
virtual tributaries are individually
accessible for cross-connect or add-drop multi-
plexer
systems. If the application does not need to access the individual tributaries. it
is simpler to pack the DS3 signal directly inro an sTS-1, as indicated in Figure
g.24.
The payload
mapping in Figure 8.24 treats the DS3 signal simply as a 44.736-Mbps
data stream with no implied intemal structure. Thus, this mapping provides
transpar-
ent transport of DS3-rate data streams.
Each row ofa nine-row sPE envelope contains 87 x 8
=
696 bits, which can carry
621 or 622 DS3 data bits depending on the value of the C bits. Notice rhat this format
I
H{-l
toooootr
tr
E
H
n
il
n
n
I
ffi
ffi
v5
R R R R R R R R
256 | btts
@6)
R R R R R R R R
R R R R R R R R
c . G o o o o R R
258I blt$
(sa6)
R R R R R R R R
R R R R R R R R
c . G o o o o R R
256 | blte
(ffi|
R R R R R R R R
R R R R R R R R
C . G R R R R R
q
$ t t I I I | | |
248 | bltE
(31))
R R R R R R R R
No.offixed informdim btc
:
4t256)
-l -
1023
$r, Sr
-
Strffinfqfidiqr
bits
Cr, Q
=
Stuffhrg odrol bit$
R bits arc fillcd friffbiu
(uspecificd)
O bits are rBffved fm finm
PuPocee
8.5 SONET/SDH 423
\,8
H B R R R R R R
Unused elotO
Chahnels 1-31
R R R R R R R R
R R R R R R R R
R R R R R R R R
Unusd sk*O
Chanreb l-31
R R R R R R R R
R R R R R R R R
E P ^ R R R R R R
Unusd alot 0
Chanrtels t.91
R R R R R R R R
R R R R R R R R
R R R R R R R R
Unu*d stoto
Channcls 1+1
R R R R R R R R
CA$bitgindund
16
Pl, Po
:
SiEnrlinEFa$E bdiodor bits
R bite ffcfixed *uffbttc
(lE+Eifiod)
t**
Saqfs
SByt
T
SBytE
SaBytes
4Bytee
f l q B
lgE-
1/0 B!/t68
(b) (a)
Figure 8.23 El mappings in VT2 SPE;
(a) synchronous;
(b)
byte synchronous'
has five C bits, which allows for single and double bit error correction. The
path over-
head
(POH)
bytes carry the 9 bytes of POH as defined in Section 8.5.3.
Determination
of the minimum and maximum DS3 rates is left as an exercise at the
end of the chapter.
8.5.7 E4 Payload Mapping
one example of a soNET supenate mapping is shown in Figure 8.25 for a 139.264-
Mbps fourth-level ITU-T signal
(E4).
This signal is packed into a 155-52-Mbps
STS-3c
(or
STM-I) signal. Figure 8.25 shows only the synchronous
payload envelope
424 FIBER OPTIC TRANSMISSION SYSTEMS
87(8)
-096
Bit$
POH: Path overhead
h^e I
R; Fixed etuff bit
(realed)
69
C: Stutf controt bit s
l: lnformstion (paytoad)
bit 621
S; Stuff opportunity bit l
O; Ovsrhead communications
bit 2
Totalbie 6S
NotF: $ contains information
when CCCCC -
00000.
S is a stuff bit whn CCCCC
=
fl 111.
Figure 8.24 Asynchronous 44.736-Mbps (DS3)
payload
mapping.
l+87(3)=261 Bytes
Prth ovrhead bVte l
Information (payload)
byte 241
CRRRRROO
5
RRRRRRRR
13
i i i i i i sR
1
Toral bytec IEI
Stuft control blt
Fixd stuff bit
Overhead communlcations bir
Stuff opportunity bit
Infomation bit
S contslne informstion when CCCCC:00000.
$ ie s atuff bit whon CCCCC *
t1 tlt.
Figure 8.25 Asynchronous 139.264-Mbps (E4) payload
mapping.
1
I
POH:
t :
X:
Y;
Z:
C:
R:
o;
S;
i :
Noto:
8 1 1 1 1 1 1 1 1 1
l 1 1 r r 1 1
POH R c I R CCRBRRRR
R ccRRooRs
I
POH R c I R CCRRRRRR I R ccRRooRs I
x 121 Y r 2l Y t 2l Y 121
\
POH 131
x 121 Y 12r z t zl
FOH 131
x t 2l Y 121 z 121
B,s soNET/sDu 425
(SPE-3c),
not the 9 bytes of section and line overhead in each row. Notice that there
is only one column of POH within the SPE-3c envelope. The POH bytes carry the 9
bytes of overhead as defined in Section 8.5.3.
The payload mapping in Figure 8.25 treats the 139.264-Mbps signal a$ a transpar*
ent data sfteam with no implied internal structure. Each row of a nine-row SPE-3c en-
velope contains 87 x 3
=261
bytes, which can carry 1934 or 1935 data bits depending
on the value of the C bits. Notice that this format also has five C bits, which allows
for single and double bit error colrection.
8.5.8 SONETOpti cal Standards
The optical interface standard
[36J
defined for
"mid-span-meet"
of $ONET equip-
ment allows for either NRZ or RZ line codes on single-mode fibers. Generation
of the
OC-N signal fromthe STS-N signal requires a scrambleras shown inFigure 8.26. The
scrambler is synchronized to each STS-N frame by presetting the shift register to all
I's immediately after transmitting the last Cl byte of the STS-N section overhead.
Thus, the frame codes
(A1,A?) and STS-I ID
(Cl)
code are not scrambled. A mini-
mum level of timing content is assured by the Al, A2, and Cl bytes along with the
static overhead bits of the STS-N frame that are anticoincident with the scrambler se-
quence. Because the scrambler is
preset at the $ame
point of every frame, every bit po-
sition in successive frames experience the same scrambler value. Thus, when static
overhead is
"exclusive
ored" with the scrambler, the same data values arise.
(The
scrambler merely converts fixed overhead
data into a different
pattern of fixed data.)
The BER objective is I x l0-I0 fbr optical sections of 40 km or less. Equipment
from separate manufacturers can be freely interchanged for applications
with dis-
tances up to 25 km. Longer distances may require
joint
engineering.
(Transmitters and
receivers from separate suppliers must be
jointly
specified to support longer dis-
tancefr.)
SONET systems are specified to operate with central wavelengths at 1310 nm with
sMF fibers or at 1550 nm with Ds-sMF fibers. operation at l3l0 nm with Ds-sMF
fibers or at 1550 nm with SMF fibers is not disallowed but must be
joinfly
engineered.
A range of laser wavelength tolerances and maximum allowable
spectral widths is
DEIE In
Figure t.26 SONET scrambler.
4?;6 FIBEH OPTIC TRANSMISSION SYSTEMS
TABLE 8.10 Fepreeentative Maximum Spectrat Widths of SONET Sources",0
SONET Source
Bate
(Mbp$)
<25k m
< 4 0 k m
oc-1
oc-3
oc-9
oc-12
51 .84
155.5?
466.56
622.08
30nm
1 Snm
1Onm
Bnm
25. 0 nm
10. 0 nm
7. 5 nm
6. 0 nm
eThe
spectral width of a $ource is determined as the wavelBngth dif{erence between th6 p6ak
mod and the
farthBst mode thet iB 10 dB b6low the peak.
DThese
specifications are for lasBrs oprating within 10 nm of th6 cntral wavlBngths (1310 end 1550 nm).
Lasef,s with greater
deviation lrom the cntral wavelength are allowed but haG nariower spcrrat wldth
spcifications to compensate tor
orater
fibr di8perslon.
-
specified for both 1310 and 1550 nm. Table 8.l0
provides
representative values of the
specifications.
8.5.9 SONET Networks
A basic block diagram of a soNET network is shown in Figure
g.27.
Gateway net-
work elements (GNEs) provide
interfaces to external (asynchronous)
digital signals.
These signals are mapped (synchronized)
and unmapped (desynchronized)
by the
gateway
using the appropriate mapping formar. At this point
only bit stuffing is used
to synchronize the asynchronous
tributaries to SONET. No pointer
a-djustments occur
in the GNE. As the sTS-N signals propagate
tluough the network, pointer
adjustmenrs
in pointer processing (PP)
interfaces nury be applied at internal network elements
(NEs),
but the lower level interface mappings that occur at the GNEs are untouched.
If a particular
NE accesses vr payloads,
vr payloads
in rhe same vr group
rhar pass
through the node may experience vT pointer
adjustments. otherwise, vlr pointer
ad-
justments
do not occur (only
the sTS-l level signals are rate adjusred). The fottowing
paragraphs
summarize pointer processing
aspects of a SONET network:
I' Pointer
justification
events (pJEs)
never occur in an originating GNE.
2. A desynchronizer
experiences continuous
pJEs
only as a result of a
synchronization
difference between the originating GNE and the terminating
Figure t.27 soNET network elements: s, synchronizer;
pp,
pointer processor;
D,
desynchronizer,
8.5 soNEr/sDn
427
GNE.* Synchronization differenceVfailures
at intemal nodes of a SONET
network produce continuous
pointer adjustments, but these
get removed when
the SPE passes through a node that is synchronized to the source GNE.
3. PJE bursts occur for two possible reasons. The first is a result of a reference
switch and a subsequent
phase adjustment of a node's local clock to align it with
the
phase of the new reference. Bursts can also occur as a result of clock noise
in multiple nodes
producing near-simultaneous
pointer adjustments.
In order for
all of these adjustments to
propagate to a desynchronizing
gateway, all of the
elastic stores in the path
must be at the appropriate threshold. This can only
happen if the source GNE has previously produced some abnormal behavior
such as a loss of a reference or sustained a rather large amount of wander.
4. A pointer adjustment at the SPE level does not affect a VT signal unless it is
passed to a node that accesses the VT and that particular adjustment haPpens to
cause a pointer movement at the VT level. Even when this occurs, the VT
pointer adjustment must
pass through the network
(without
absorption) to the
desynchronizing
gateway to affect the outgoing tributary signal. On average'
one of every 30 PJEs at the STS-1 level
produces a PJE at the VTl.5 level.
Figure 8.28 shows a block diagram of an SPE synchronization circuit
(PP)
depict-
ing two halves of pointer processing: one half exhacts
(desynchronizes) the SPE pay-
load from a received signal and the other half synchronizes the SPE to the local STS-I
frame rate. The RX
pointer processing block extracts the
payload data from the re-
ceived signal and
passes it to the elastic store. The TX pointer processing block moni-
tors the fill level of the elastic store and makes pointer adjustments to maintain a
nominal level of storage. The size of the elastic store only needs to be on the order of
8 bytes in length, not a full frame. The ability to use a relatively small elastic store
(as
compared to frame-length elastic stores in the asynchronous network) is one of the fea-
tures of a
pointer-based synchronization architecture: The
payloads are allowed to
float with respect to the STS-l frame boundaries'
Frequency of Palnter Justlllcation Events
If all NEs of a SONET island use a timing reference that is traceable to a common pri-
mary reference source
(PRS),
PJEs occur only as a result of distribution-induced clock
wander that produces no sustained frequency offset. Thus, when all NEs are synchro-
nized to the same reference, PJEs occur at random times and have equal numbers of
positive and negative values over the long run.
Continuous PJEs occur only when there is a reference failure at some NEs within
a SONET island or the island is intentionally designed to operate in a
plesiochronous
mode. If the reference failure occurs at some internal node of the SONET island, the
resulting PJHs are removed at the next node in the
path that is still locked to the same
reference as the gateway NE. Thus, a tributary desynchronizef at a GNE must deal
*This
statement assumes that the terminating GNE synchronizes incoming SPEs to a local clock before they
urrive at the desynchronizer.
Odrr SPEe
Ovtrterd datr
428 FIBEH oPTIc TRANSMIsSIoN SYSTEMS
Flgure t.28 Block diagram of SPE synchronizing equipment: RX, receiver; TX, transmitter.
with continuous PJEs only when there is a reference failure in either the originating
GNE or in the rerminating
GNE.
SONET Detynchronizers
soNET desynchronizers
are necessarily designed with very low clock recovery band-
widths to smooth the effects
of (l) isolated pointer
adjustments, (2)
continuous pointer
adjustments, (3) pointer
adjustment bursts, or (4)
combinations of the latter two. A
pointer
burst is defined as the occurrence of multiple pointer
adjustments of one po-
larity occurring within the decay time of the desynchronizer
circuit
(i.e.,
the reciprocal
of the desynchronizer closed-loop PLL bandwidth). Thus. it is ironic rhat as the clock
recovery bandwidth is narrowed to smooth the effect of a burst, the probability
of a
burst occurrence is increased (by
definition only). Extremely nrurow
pLL
bandwidths
are easiest to implement using digital filtering techniques commonly referred to as bit
leaking. Bit leaking is essentially a mechanism for converting byte-sized pointer
ad-
justments
into bit- (or
fractional-bit-) sized timing adjustmenrs.
Figure 8.29 shows a block diagram of a microprocessor-controlled
DSl desyn-
chronizer. The microprocessor
is used to perform
long-term
averaging ofphase ad-
Ml3 multlplerer
Figure 8.29 VTl.5 desynchronizer
hardware functional components.
DSI
ertrrctlon
8.6 SONET RINGS 425
justments
in lieu of dedicated
logic that requires large counters and wide word sizes
for low-bandwidth DSP filtering. The first function of the rnicroprocessor is to deter-
mine the average DSI payload frequency offset represented by all frequency adjust-
ment events
(bit stuffs, vT
pointer adjustments, and sTs
pointer adjustments). After
the average frequency adjustment is determined, a stuff ratio value is calculated
that
allows insertion into a DS3 signal as shown.
(The
M12 stage is embedded in the M13
multiplexer.)
The elastic store fill level is used for very long term adjustments in the
output frequency that arise from finite
precision limits of the DSP calculations and for
accommodating
variations in the DS3 clock, which is typically not synchronized to the
SONET line clock.
8.6 SONET HINGS
As has been mentioned earlier in this book, the development of large switching ma-
chines and transmission systems with extremely
large cross sections has impacted
telecommunications network architectures with a trend toward fewerhierarchical
lev-
els. An undesirable consequence
of this trend is increased dependence
on the opera-
tional status of individual switching machines and fiansmission
paths. A SONET
self-healing ring, or more simply a SONET
ring, is a network architecture that specifi-
cally addresses network survivability. Two basic types of self-healing
rings are shown
in Figure 8.30: a unidirectional ring and a bidirectional ring. The main difference be-
tween the two types of rings is how the two directions of a duplex connection are e$-
tablished.
In a unidirectional ring a single time slot of the entire ring is assigned to both halves
of a connection.
As indicated in Figure 8.304, traffic is normally carried only on the
(unidirectional) working
path with the counterrotating
path used for
protection. In the
example, an STS-1
(out of an OC-48) might be carried directly from A to B, but the
returning STS-1 would be carried from B through C and D to A. A bidirectional
ring,
on the other hand, establishes both halves ofthe duplex connection
over the shortest
path in the ring. Thus, no fiber is identified as a pure working fiber and another as a
pure protection fiber. Because bidirectional
rings
provide shorter round trip delays for
most connections and allow reuse of time slots on the ring, it is the
prefened mode of
operation for interoffice networks. Rings for subscriber access applications do not
carry much traffic between ADM nodes and therefore are more suited to a unidirec-
tional mode of operation.
8.6.1 Unidirectlonal
Path-Switched RIng
As shown in Figure 8.31, a unidirectional
path-switched ring
(UPSR)
[38]
transmits
the same information from A to B in both directions around the ring. Normally, only
the working
path is accessed by the receiving node; If a failure occurs, a node can $e-
lect the data on the protection channel. Notice that in the example shown selection of
the
protection path
actually
leads to a shorter path for the connection
from A to B'
430 FIBEH oPTIc TRANSMISSIoN SY$TEMS
Protc{ion
psth
(b)
Figure 8.30 (a)
Unidirectional
and
(b)
bidirectional rings.
8.6.2 Bldirectlonat
Line-$witched
Ring
Bellcore defines rwo versions
of bidirectional line-switched
rings (BLSRs)
I39l:
a
two-fiber BLSR and a four-fiber
BLSR. on a two-fiber BLSR protecrion
is provided
by reserving bandwidth in each of two counterrotating fiber paths (Figure
g.32).
If all
traffic is to be protected,
only 507o ofthe total system capacity can be used. Under nor-
mal conditions connections
between two nodes utilize the shortest path
between the
nodes' If a fault in either direction of hansmission occurs, the nodes adjacent to the
fault perform
ring switches as indicated. A ring switch involves switching traffic from
working channels of the failed facility to spare channels of the other facility on the side
of the node on which the fault occurs. The protection-swirched
traffic propagate$
all
the way around the ring, being ignored by intervening nodes, until it is switched back
(*)
8.6 SONET RINGS
431
Wofting Path
Protoction Path
Path Terminatlon
Figure E.31 UPSR
protection swirching.
to the working channels
by the other node next to the fault. Notice that all nodes
(in-
cluding the nodes adjacent to the fault) communicate on working channels
in the same
manner as they did before the
protection switching. That is, the path terminations are
not part of the protection path. The main impact of the protection switch is an increase
in delay for affected traffic
(and
a momentary insertion of exkaneous
data when the
switch occurs).
On a four-fiber BLSR
(Figure
8.33) two pairs of fibers are
provided for each direc-
tion of kansmission-one bidirectional
working
pair and another
pair for protection
of the first
pair. Thus, working and protection channels are carried on different
physi-
cal facilities. Again, connections are normally set up to use the shortest distance of
ion Switches
. if:-*-
West Esst
nodes nodes
West
East
nodes nodes
Figure 8.32 Two-Fiber BLSR protection switches
Working Pair/
432 FIBER OPTIC THANSMISSION
SYSTEMS
Protgdion
Rlng Protsdlon
$rritches
\
wo*ing e"ii
F' P. Kapron, D. B. Keck, and R. D. Maurer,
"Radiation
Losses in Glass opticar
Waveguides," App lied Physics l*rrers, Nov. I
g70,
pp.
4Z34ZS.
T' Miya, Y. Terunuma, T. Hosaka,
and T. Miyashita,
"urtimate
Low-Loss
single Mode
Fibre at 1.55 pm," Electronic
l*tter.r, Feb. 1979, pp. 106-10g.
D. Large,
"The
Star-Bus Network: Fiber Optics to the Home," CED Magazine, Jan.
r 989.
R. olshansky, v. A. Lanzisera, and
p.
M. Hil,
"subcarrier
Murtiplexed Lightwave
systems for Broadband Distribution,"
Journal of Lightwave
Tethnilogy, sep-t. 19g9,
pp.
1329-1342.
Figure 8.33 Four-Fiber
BLSR protection
switches. Note that a ring switch and a span switch
cannot coexist without channel
switching and a reduction in capacity,
travel for each side of a connection.
If a failure occurs on only a working facility, pro-
tection switching
occurs similar to
"span
switching" of a point-to-poini
system: The
traffic is merely switched to and from the protection
facility by nodes adjacent to the
fault. However, if a fault affects both the working and the proiection
facilities, a ring
switch is needed as shown.
Again, protection-switched
traffic propagates
all the way
around the ring without being accessed by intervening nodes. eil traffic accesses still
occur on the working channels
even though the same information
is passing
through
the nodes in the protection path.
A four-fiber BLSR obviously requires more facilities rhar a two-fiber BLSR but
has numerous
advantages. First, the protected
capacity of the system is twice as large.
second, fiber failures
on only the working pair
can be accommodated
by a span switch
with minimal
disruption to traffic. Third, multiple separate failures can tccur on work-
ing pairs
and be accommodated
by multiple span switches. Fourth, the presence
of a
spare pair
simplifies
maintenance testing and possible
upgrading of facilities. For
these reasons,
a four-fiber BLSR is generally
favored.
REFERENCES
*\
\
Prt""tion
p"i,
REFERENoES
430
5 T. Li,
,'Advances
in optical Fiber communications:
An Historical Perspectle," IEEE
Joumal on Selected Areas in Communicatfons,
Apr. 1983' pp. 356-372,
6 P. S. Henry,
"Lightwave
himet" IEEE lourrutl of
QuantumElectronics'
Dec' 1985' pp'
1862-1877.
7 J. R. Stauffer,
',FT3C-A
Lightwave
system for Meffopolitan
and Intercity
Applications,"
IEEE Journal on Selected
Areas in Communications,
Apr. 1983' pp.
413- 419.
8 P. S. Henry, R. A. Linke, and A. H. Gnauck,
"Introduction
to Lightwave Syst'ems," in
OpticalFiberTransmission
l l, S. Mitler and I. P. Kaminow; Eds., Academic Press, San
Diego, 1988, pp. 781-832.
9 D. C, Gloge and I. Jacobs,
"Tenesnial
Intercity Transmission
Systems," in Optical
Fiber Transmission II, s, Miller and I. P. Kaminow, Eds., Academic hess, san Diego,
1988.
l0 D. L. Howells,
"High
Capacity Lightwave
Technology Comes of Age," AT&T
Technol ogy,Vol .3, No.4, 1988, pp. l 0*15'
1l W. C. Mara, J. S. Linnell, N. M. Denkin, and K. Ogawa,
"FT
Series G Lightwave
Digital Transmission
system Architecture and upgrade capability,"
IEEE
Int e rnational C onferenu on C ommunicat ions, 1987, pp. I 9B' 3. I
-
I 98. 3'5.
l2 K. Ogawa,
"Analysis
of Mode Partition Noise in Laser Transmission Systems,*
IEEE
Joumal of
Quantum
Electronrcs, May 1982, pp' 849-855.
13 J. E. Goell.
..An
Optical Repeater with High-Impedance Input Amplifrer," Bell,$ystem
Technical Joumal, Apr. 19?4, pp. 629-&3'
14 N. Yoshikai, K. I. Katagiri, and T. Ito,
"mBlC
Code and Its Performance in an Optical
communications system," IEEE Transactions on, communications,
Feb. 1984,
pp.
163-168.
15 K. Hagishima, Y. Kobayashi, and K. Aida,
"High-speed
cMI optical Intraoffice
Transmission
System: Design and Performance," IEEE Joumal on Selected
Areas in
Comtnunications, Dec. 1986, pp. 145 I
-
1457'
16 F. Ross.
"Overview
of FDDI: The Fiber Distributed Data Interface,"
IEEE .lournal on
SelectedAreas
in Communications,
Sept. 1989' pp. 1043-1051.
l7 P. Cochrane, R. Brooks, and R. Dawes,
"A
High-Reliability 565 Mbit/s Trunk
Transmission
System," IEEE Joumal on Selected Areas in Communicatiorts,
Dec.
1986, pp. 1396-1403.
l8 R. L. Williamson and M. Chown,
"NLl
Submarine System," /EEE foumal on Selected
Areas in Communications, Apr. 1983' pp' 454-458'
19 P. A, Franaszek and A. x. widmer,
"Byte-oriented
Dc Balanced
(0.4)
88/108
Partitioned Block Transmission
Code," U.S. Patent 4,488,'739, Dec. 4' 1984.
20 A. F. Benner, Fib re Channel Gigabit Communications and.I/Ofor Computer Networks,
McGraw-Hill, New York, 1996.
21 Y, Niro, Y. Ejiri, and H. Yamamoto,
"The
First Transpacific
Optical Fiber Submarine
Cable System," IEEE Intemational Conference on Communicafians,
June 1989' pp.
50. 1. r - 50, 1. 5.
22 N. Yoshikai. S. Nishi. and J. L Yamada,
"Line
Code and Terminal Configuration fbr
Very Large-Capacity Optical Transmission
System," IEEE Journal on Selecteil Areas
in Communicafians, Dec. 1986, pp. 1432-1437.
434 FTBEH oplc rRANSMtsstoN
sysrEMs
23 H. Kimura and K. Nakagawa,
"F-1.6G
system overview,"
Review of the Electrical
Communications Laboratories
NTT Japan, Vol. 35, No. 3, I 9g7, pp.
ZIV_ZZS.
24 s' s. wagner and H. Kobrinski,
"wDM
Apprications
in Broadband
Telecommunications
Networks," IEEE communications Magazine,
Mar. 19g9, pp.
22_30.
25 R. J. sanferrare,
"Terrestrial
Lightwave systems," Ar& T Technical Journal, Jan7qeb.
1987.
26 S. S. Wagner, H. L, Lemberg, H. Kobrinski, L. S. Smoot, and T. J. Robe,
,.A
passive
Photonic Lnop Architecture
Employing wavelength-Division
Multiplexing,',
IEEE
Globecom Proceedings, l
gSS,
pp. 48. l, t
-48.
1.5.
27 N. K. Dutra,
"[I-v
Device Technologies
for Lightwave
Applications,l' AT&z
Technical
Joumal, Jan./Feb. 1989, pp.
5-l
g.
28 P. R. Trischitta
and D. T. s. chen,
"Repeaterless
undersea Lightwave systems
,-
IEEE
Communications
Magafine, Mar. I 989, pp. I 6_2l.
29 s. c. Mettler and
c. M, Miller,
"optical
Fiber splicing
,,,
in optical Fiber Transmission
I I
'
s' Mller and I. P. Kaminow, Eds., Academic hess, san Diego, r 9gg, pp. 263-300.
30 J. B, Haber and J. W. Rogers,
..Fiber
in rhe Loop," AT&T Technology,
Vol. 3, No. 4,
1988, pp. 2-9.
31 N. Yamaguchi, K. Yarnane, and T. Kihara,
"565-Mbit/s
optical Fiber Transmission
System," Fujitsu
Science and Technology Journal, Mar. 19g9, pp.
60_63.
32
"Synchronous
Optical Network (SONET)-Basic
Description Including Multiplex
Structure, Rates and Formats," Tl.l05-lgg5, ANSI, New
york,
1995.
33
"synchronous
optical Network (soNET)-payload
Mappings," T1,105.02-1995,
ANSI, New York. 1995.
34 Recommendation
G.707,
"Network
Node Interface for the synchronous
Digital
Hierarchy (SDH),"
ITU-T,
Geneva, Switzerland,
Mar. 19g6.
35 N. B. sandesara, T. H. Jones, and A. G. Edwards,
..soNET
Intra-office
Interconnecr
Signal," IEEE Glob e com
p
roce edin g
s, I 9
g g,
pp. 30.5. I
_30.5.7.
36
"synchronous
optical Network (soNET)
Transport
systems: common Generic
Criteria," GR-ZS3-Core, Bellcore, Morristown, NJ. Dec. 1995.
37 N. Zhang, K,-H. Liu, and E. c.
posner,
"Reriable
payroad pointer protocol
for
Synchronous
Optical Network," IEEE International
Communicatittns
Conference
P roceedings,
I 989, pp.
l4.Z.l
-l
4.2.j.
38
"soNET
Dual-Fed undirectional
path
swirched Ring (upsR)
Equipment Genedc
Criteria," GR- 1400-Core, Bellcore,
Jan. 1999.
39
"soNET
Bidirectional
Line-swirched
Ring Equipment Genenc criteria,"
GR-l 230-Core, Bellcore, Dec. 1993.
PROBLEMS
8.1 Determine
tle attenuation
g,
in decibels per kilometer such that the loss limit of
a Sfi)-Mbps fiber system is exactly equal to the dispersion limit. Assume the
transmitter output is 30 dB above the receiver threshold for the desired error rate
and that the system has a BDp of
g0
Gbps-km.
PROBLEMS 435
8.2 An 850-nm,
graded-index, multimode, So-Mbps fiber system with 2 dB/km loss
in the fiber is to be used for a token-passing
bus local area network. Assume the
system uses the first entries in Tables 8.1 and 8.2 as a source-detector
pair. If
the BDP of the fiber is 500 Mbps-km, how many passive taps with 0.4 dB of loss
can be inserted
per kilometer without affecting the distance between transmitters
and receivers.
8.3 What is the distance limit
(without taps) of the system in hoblem 8.2 if the data
rate is reduced to l0 MbPs?
8.4 A
graded, multimode
fiber exhibits 100 psec/km nm of chromatic dispersion at
850 nm. Determine the BDP of a NRZ system using an LED having 40 nm of
specrral width.
8.5 A WDM SMF system using 1300 and 1550 nm is dispersion
limited at both
wavelengths.
How must the spectral bandwidths of the two $ources be related?
(AssumeD
=
3.5 pseclkm nm at 1300nmandD
=
17'5
psec/kmnm at 1550 nm')
8.6 What is the minimum and maximum pulse density of mBlp and mBlC line
codes?
8.7 What is the longest string of no
pulses that can occur in mBlP and mBlC line
codes?
8.8 A 486B line code can encode all4 bits of source data with 6-bit line codes hav-
ing exactly three pulses in every code. How many dc-balanced codes are left
over for link control?
8.9 Determine
the minimum and maximum data rate of an asynchronously mapped
DSI signal that can be caried within a VTI-5 envelope.
8.10 Determine
the minimum and maximum datarates of an asynchronously mapped
El signal that can be carried within a VT2 envelope.
8.11 Determine the minimum and maximum data rates of an asynchronous
DS3 pay-
load that can be carried within an STS-1 envelope.
8.12 Determine the minimum and maximum
B[ data rates that can be accommodated
within an STS-3c envelope.
8.13 Determine the average frame acquisition
time of an STS-1 signal assuming all
nonframing
bits have equally likely random values of 0 or 1'
8.14 Repeat Problem L13 for an STS-3c signal'
8.15 Determine the
payload capacity of a STS-3c
(STM-1) signal.
DIGITAL
MOBILE TELEPHONY
A basic concept ofa cellular system is to provide ever-increasing capacity by dividing
cells into smaller and smaller sizes to increase frequency reuse. Unfortunately,
the cell
division concept has proven to be impractical
in terms of finding suitable locations for
base station antennas and for
getting repeated construction authorizations from
gov-
erning organizations. The explosive demand for mobile telephones
in the early 1990s
within the United States and elsewhere in the world helped stirnulate the development
of new Sy$tems to accommodate
the demand.
Two basic approaches have been
pur-
sued: expanding the channel capacity of existing $y$tems and allocating
new fre-
quency bands to cellular mobile phone service.
All of the new systems utilize digital
transmission in lieu of the analog FDM transmission used by analog cellular systems
such as AMPS in North America, TAcs in Great Britain, and NMT in scandinavia.
Commercially viable digital cellular systems are enabledby the availability
of low-
cost digital signal
processing technology to provide solutions to problems in several
key areas. First, it is necessary to compress a digital speech signal into a low enough
bit rate that digital voice transmission
does not impose a spectrum
penalty with respect
to analog system$. As discussed in Chapter 3, speech compression algorithms
have ad-
vanced to the
point that digitization
can, in some cases,
provide
Sreater
spectrum ef-
ficiency than analog sy$tems. Second, the application
of digital transmission
to a
mobile environment requires a sophisticated
equalizer to overcome the effects of dy-
namic multipath transmission impairment$.
Third, the susceptibility
of the speech
compression
algorithms to channel errors requires sophisticated error conection and
control measures. Lastly, the low-bit-rate
voice coders and the digital system archi-
tectures introduce significant artificial delay into the voice channel, which imposes the
need for echo cancelers for acceptable
voice quality'
9.1 NORTH AMERICAN DIGITAL CELLULAR
North American Digital Cellular
(NADC), also known as US Digital Cellular
(USDC)
or Digital-AMPS
(D-AMPS), repre$ents a digital upgrade from the
previously de-
ployed analog cellular system referred to as Advanced Mobile Phone Service
437
438 DIGITAL MoBILE TELEPHoNY
(AMPS).
The D-AMPS sysrem is designed to be compatible with AMps. In fact. a
D-AMPS
$ystem installation
can coexist with an AMPS installation,
thus allowing a
graceful
migration from an all-analog service to an all-digital service. An analog_to-
digital migration is supported by a dual-mode phone
that can operate as an AMps
phone in one call and as a D-AMPS phone on the next call. D-AMpS is standardized
by EIA/TIA as Interim Standards IS-54 and 15-136.
9.1.1 D.AMPS Transml ssi on
Format
The most significanr
aspect of maintaining compatibility with AMps is the need to
adhere to the AMPS FDM channel structure. This channel structure uses 30-kHz-wide
channels in the range 824-8g4MHz.
within each 30-kHz FDM channel IS-s4 defines
six digital channels operating in a time division multiple-access (TDMA)
mode of op_
eration, as shown in Figure 9.1. Transmission from a base station to the mobiles is ac-
complished
with a continuous
TDM stream with six time slots. Transmission
from
each of the mobiles occurs in data bursts that are timed to arrive at the base station in
separate, nonoverlapping
time slots synchronized to the outgoing time slots. Associ_
ated with each burst from a mobile is a guard
time to prevent
overlap and provide
a
traxsmitter ramp-up preceding
the data. The guard
time between time slots is mini-
mized by adjusting
the transmit time of the mobiles with control messages from the
base station. These adjustments
are dynamic to accommodate
mobility.
The TDMA digital transmission
frame format within each 30-kHz channel con-
tains six time slots with a total of l9zl4 bits. The repetition rate of each frame is 25
frames per
second, which leads to an aggregate bit rate of 49.6 kbps in the 30 kHz of
bandwidth. The modulation format is r/4 shifted, differentially
encoded, quadrarure
phase
shift keying. This format is essentially
4-psK modulation
with two four-point
constellations offset from
each other by nl4 radians. By alternating between constel-
lations, a symbol transition of at least n/4 radians is always assured-a property
that
helps in symbol clock recovery.
Full-rate voice coding utilizes two time slots in each frame for the voice informa-
tion. Thus, the system capacity with full-rate voice coding is three times that of an
AMPS system since there are three TDM voice channels within each FDM channel.
Baee
, r ilrobiles
station
*-
fJ ff f-J
+-
*--
| r-f
f-r l+E
Fig're 9.1 TDMA transmission format of NorJh American dieital cellular.
9.1 NORTH AMERICAN OIGITAL CELLULAR 439
If half-rate voice coding gets implemented in the future, the capacity expansion will
be sixfold.
9.1.2 D-AMFS $peech Godl ng
The speech-coding algorithm is vector sum excited linear predictive
(VSELP)
coding
[],
which is described in Chapter 3. The D-AMPS VSELP algorithm
processes the
speech waveform in segments of 20 msec duration. Each speech segment is repre-
senred by 159 bits. Since there are two VSELP frames in each TDMA frame, the raw
data rate of the voice is 2 x 159 x 25
=
7950 bps. To the raw bit date is added 5050
bps ofredundancy
encoding for error correction and detection to
produce a composite,
aggregate data rate of 13 kbps for a voice channel. As shown in Figure 9.2, there are
260 data channel bits and 64 overhead bits in each time slot. Table 9.1 describes the
basic use of each data field within the time slots,
9.1.3 D-AMPSControl Ghannel
In addition to providing a threefold capacity expansion,
the digital nature of D-AMPS
provides other advantages that are not
possible or at least difficult to achieve in an ana-
log system such as AMPS. The first of these is use of the CDVCC channel to ensure
that a base station maintains connections with intended mobiles. AMPS utilizes a
similar feature with supervisory audio tones
(SATs). A SAT is a tone at 5970, 6000'
or 6030 Hz that is inserted and removed from the audio signal specifically
for detect-
ing fades and ascertaining connection
integrity. The availability
of only three tones
and the complexity of inserting, detecting, and repeating these tones are significalt
limitations of AMPS.
A more significant advance
provided by the digital nature of D-AMPS involves the
use of the SACCH channel embedded
in each time slot. Because this channel is always
present, it is quite useful for communicating conkol and supervision
information
while speech is actively in
progress. Specific, advantageous uses ofthis contol chan-
sYl{c EACG}I DATA
CDYCC DATA HFVD
Brrt thtlon to Mobl|t
Moblloto Brrt sftd|on
f igure 9.2 Time slot format of North American digital cellular.
t2 rs0
It0
lfr
12 t2
n tt
F G DATA 8YI{G DATA SACCH
eDvcg DATA
440 DtctrAL MoBtLE rELEpHoNy
TABLE 9.1 Data Fietd Functions of D-AMPS TIme Stots
CDVCC
DATA
G
R
BSVD
SACCH
SYNC
coded digilal v9rifhation
color code. A unique code sent by a base station and
returned by each mobile for base station conlirmation of connection intogrity.
Application bearer channel bit$ (voice
or data). can also be used for a fast
associated control channe_l (FAcc)
when there is no active application or a
situation arises when application transmission
needs to be usurped.
Guard time. Mobile transmitter is off.
Ramp tim. Mobile transmitter ramps up to assigned power
level.
Reserved (unused).
slow associated control channel. A continuous channel used to send conhol
and superuisory information.
synchronization channel. Used for synchronization, equalizer training, and time
slot identification.
nel involve authentication, additional connection integrity,
transmit power
control,
channel quality
mea$urement
reports, handoffs
to a new cell, keypad depressions, and
calling number identification.
The SACCH control channel is also used for timing ad-
justments
specific to the TDMA operation.
A particular
example of the usefulness
of the SACCH channel is its support of mo-
bile assisted hand off (MAHO).
If a mobile with an established connection moves
from one cell to another, the process
of handing the mobile off is performed
with much
more control and reliability
than is possible
in AMps. The MAHO process
begins by
the base station telling the mobile to make channel quality
mea$urements
on the cur-
rent chatrnel and on candidate
channels for a potential
handoff. Channel quality
meas-
urements involve received
signal power
levels and bit error rates (BERs).
The TDMA
nature of D-AMPS facilitates
measurements
of candidate channels by tuning to the
candidate frequency during an inactive time slot. After each set of measurements
the
mobile sends the results to the base station (via
SACCH) whereupon the base station
can determine if a handoff is
justified.
9.1.4 D.AMPS Error Control
There are three mechanisms
incorporated into D-AMPS for mitigating
the effects of
channels errors: elror correction, error detection, and interleaving. Enor correction is
implemented with a half-rate convolution coder for the perceptually
most significant
bits of the voice. There are 77 such bits out of a frame size of 159 bits. The half-rate
convolution
coding process
therefore adds 77 more bits to the channel. Of the 77 bits,
12 are
particularly
important. A 7-bit CRC check sum is added for rhese bits to deter-
mine if any of these 12 bits are received in error. when a cRC enor occur$. certain
critical parameters
from the previous
error-free frame are used to reconsg.uct speech
to avoid use of aberrant values. If several CRC errors are received in successive
frames, the reconstructed
speech is muted. The general
term for these operations is bad
frame
masking.
The third error control mechanism involves
separating the data in a single speech
frame, interleaving
it with data from adjacent speech frames, and transmitting it in two
s.a GLoBAL sysTEM FoR MoBtLE coMMUNtcATloNS 441
time slots. This
process reduces the possibility that a burst of errors will circumvent
the error corection capabilities ofthe convolutional coding. A drawback to interleav-
ing is the delay it adds to the channel, which must be accounted for in echo cancelers.
9.2 GLOBAL SYSTEM FOR MOBILE COMMUNICATION$
Global System for Mobile Communications
(GSM)
[2]
is a cellular mobile commu-
nications system developed in Europe afld standardized by the European Teleconrmu-
nication Standards Institute
(ETSI).
GSM has subsequently been adopted worldwide
as the intemational digital mobile standard. Initial work on GSM standardization be-
gan in 1982. The first field trial of a GSM system occuffed in 1991.
The International Telecommunication Union
(ITU)
allocated frequency spectrum
at 935-960 MHz for the downlink
(base
station to mobile) and 890-915 MHz for the
uplink
(mobile
to base station). Even though some of this spectrum wa$ being used by
established analog systems, there was no attempt to be compatible with the existing
frequency
plan. (The
incumbent
analog systems in the various European countries
were incompatible with each other and a main goal of GSM was to establish a coflrmon
standard so backward compatibility was not a consideration.)
9. 2. 1 G$MChannel St ruct ure
Starting with an open-spectrumplan
provided GSM system designers with more free-
dom in system design than was available to D-AMPS designers. The most significant
difference in GSM with respect to D-AMPS is the use of 200-kHz-wide digital RF
channels as opposed to the 30-kHz-wide
D-AMPS RF channels. Each GSM RF chan-
nel operates at 270.833 Mbps using GMSK modulation. As mentioned in Chapter 6,
GMSK modulation is fairly closely related to 4-PSK modulation as used in D-AMPS.
GMSK modulation does, however, require more bandwidttr than tightly filtered 4-PSIE
as evidenced by an information density of 270.8831200
=
1.35 bps/Hz for GSM and
48.6/30
=
1.6? bps/Hz for D-AMPS. The GMSK modulation format of GSM provides
a constant-envelope RF signal that is more efficient for RF power generation than is
tightly filtered 4-PSK modulation used by D-AMPS. This efficiency is most important
for hand-held battery life..
As shown in Figure 9.3, a GSM RF channel utilizes digital TDMA with eight full-
rate voice channels, in contrast to three full-rate voice channels suppor.ted by a single
D-AMPS RF channel. The ability to terminate more TDMA channels on a single
transceiver
provides a cost advantage for GSM base stations.
(The
cost of a 270'833-
kbps TDMA transceiver is no different than the cost of a 48.6-kbps
TDMA trans-
ceiver.)
The GSM system carries eight full-rate voice channels in 200 kHz of bandwidth,
which amounts to 25 kHz per voice channel, a specffum efficiency identical to Euro-
*As
discussed below, GSM uses Iess speech compression than does D-AMPS, which Ieads to a higher rate
digital channel for voice and in tum requites more ftansmit
power.
442 DIGITAL MOBILE TELEPHONY
Baee
| .
Mobilee
r-l
G
+-
|
f f l { -
Figure 9.3 TDMA transmission format of GSM.
pean analog FDM systems of the time. Thus, the introduction
of the GSM system did
not provide
spectrum efficiency improvements
as D-AMps did. A GSM system does,
however, provide
cellular system efficiencies in that digital transmission, in general,
and strong error correction, in particular,
allow operation at lower signal-to-noise ra-
tios' Greater noise or interference
tolerance leads to longer transmission
distances
and/or greater
amounts offrequency reuse.
The burst period
of a GSM system is l?0/26/8 =
15126 ms. This burst period is de-
rived from a 120-ms superframe
consisting of 26 TDMA frames and
g
bursts per
TDMA frame' Twenty-four frames of a 26-friune super frame are allocated to traffic
(e.g.'
voice) transmission while one of the frames is allocated to a SACCH control
channel for each traffic channel. The last TDMA frame of a superframe is reserved.
A unique aspect of GSM, with respecr ro D-AMps, is that a TDMA burst format is
used in both directions of transmission,
as opposed to only on the uplink from the mo-
bile to the base station. The format of these bursts is shown in Figuie 9.4, where it can
be seen that there are 148 bits ofdata and an idle guard
time corresponding to the pe-
riod of 8'25 bits. The burst transmission rate of a taffrc channel can now be determined
as 156.25/r5D6 =
270.833 kbps. The fields within the burst are identified in Table 9.2.
The use of multiple bursts in the downlink direction,
as opposed to continuous
ffansmission,
is advantageous in that it inherently
allows tuming off the base station
transmitter
during idle channels, which in turn reduces the total amount of interference
between cells in a widespread
and congested installation. An advantage of continuous
transmission, as used in D-AMPS, is the relative ease of implementation
and greater
performance
of the digital receiver in the mobile.
15/26
=0.577
msec
Figure 9.4 Time slot format of GSM burst.
9,2 GLOBAL SYSTEM FOR MOBILE COMMUNICATIONS 443
TABLE 9.2 Data Field Functlons of G$M Tlme Slot
Flag
Guard
Tail
TCH
Train
A single bit used to signify voice or FACCH content in an associated TCH field
ldle period
of 8.25 bits interuals timing margin between bursts
3
"0"
bits for equalizer training
Field for transporting bearer data or FACCH data
Field of fixed data
pattern used to train equalizers and acquire a data clock for
the entire burst
9.2.2 GSM Speech Codi ng
GSM uses regular pulse excited-linear predictive
coding with a long-term predictor
loop
(RPE-LTP)
[3].
The RPE-LTP algorithm is described in Chapter 3. Speech is di-
vided into 2O-msec samples, each of which is encoded as 260 bits, giving a total bit
rate of 13 kbps. This is the original, full-rate speech-coding algorithm. An enhanced
full-rate
(EFR)
speech-coding algorithm has been implemented by some North
American GSM 1900 operators. EFR is said to
provide improved speech
quality
using
the existing l3-kbps bit rate.
9.2.3 GSM Channel Codi ng and Modul ati on
GSM utilizes error control mechanisms similar to D-AMPS. First of a]l. the 260 bits
of a speech frame ate divided into three classes:
Class Ia, 50 bits-most sensitive to bit errors
Class lb, 132 bits-moderately sensitive to bit errors
Class II, 78 bits-least sensitive to bit errors
Class Ia bits have a 3-bit CRC added for error detection. If an error is detected, the
frame is replaced by a slighfly attenuated version of the previous correctly received
frame. The 50 Class Ia, 3 CRC, 132 Class Ib, and a 4-bit tail sequence
(a total of 189
bits) are processed by a half-rate convolutional encoder for error correction. The out-
put of the convolutional encoder is added to the 78 Class II bits to produce an ag-
gregate speech frame of 456 bits. Thus, the redundantly encoded speech rate is
456tO.OZ0
=
22.8 kbps.
To further protect
against the burst errors, each sample is interleaved. The 456 bits
of output by the convolutional encoder are divided into eight blocks of 57 bits, and
these blocks are spread across eight consecutive time slot bursts. Since a time slot burst
carries two 57-bit blocks, each burst contains raf,fic from two different speech samples.
9.2.4 G$M Mobi l e Stati on
The GSM mobile station
(MS) consists of the mobile equipment
(the
terminal) and a
smart card called the Subscriber Identity Module
(SIM). The SIM
provides personal
mobility, so that the user can have access to subscribed services irrespective of a spe-
cif,rc terminal. By inserting the SIM card into another GSM terminal, the user is able
444 DIGITAL MOBILE TELEPHONY
to receive calls at that terminal, make calls from that terminal, and receive other sub-
scribed services.
The mobile equipment is uniquely identified by the Internarional Mobile Equip-
ment Identity
0MEI).
The SIM card contains the Intemational Mobile Subscriber
Identity (IMSI)
used to identify the subscriber to the system, a secrer key for authen-
tication, and other information. The IMEI and the IMSI are independent, thereby al-
lowing personal
mobility. The sIM card may be protected
against unauthorized use
by a password
or
personal
identity number.
9.2.5 GSM Frequency Hoppl ng
The mobile station is inherently frequency agile, meaning it can move between a trans-
mit, receive, and monitor time slot within one TDMA frame, all of which are normally
on different frequencies. GSM makes use of this inherent frequency agility to imple-
ment slow frequency hopping-the mobile and BTS transmit each TDMA frame on
a different carder frequency. The frequency-hopping algorithm is broadcast on the
broadcast control channel. since multipath fading is dependent on carier frequency,
slow frequency hopping helps alleviate the problem.
In addition, cochannel interfer-
ence is more random than when fixed-frequency allocations exist.
9.2.6 GSM Short Message $ervice
Short Message Service
(SMS)
is an integrated bidirectional messaging service that al-
lows GSM cellular subscribers, and various PCS offerings, to send and receive data.
Individual messagerr (with
GSM) can be up ro 160 bits in length. Because the sAccH
is used for sMS data transmission, messages can be received or tran$mitted during a
voice call. Initial applications of SMS focused on alphanumeric paging
services with
fundamental differences: SMS is bidirectional and message delivery is guaranteed.
subsequent applications served by sMS are voice mail notification, e-mail delivery,
stock quotes,
and downloading/updating of SIM cards.
9.3 CODE DIVISION MULTIPLE.ACCESS CELLULAR
The two digital cellular systems discussed in the previous
sections, D-AMps and
GSM, utilize a combination of frequency division multiplexing and time division mul-
tiplexing+ as a method of partitioning
a block of allocated frequency spectrum into in-
dividual communication
channels. This section discusses systems with a
fundamentally different approach to channel definition-code division multiple ac-
cess (CDMA)-that
belong to a class referred to as spread spectrum cornmunications
systems
[a].
The term spread spectrum refers to the fact that transmission bandwidth
*Within
a particular
GSM cell it is conceivable that a single FDM channel supporting eight TDMA channels
is sufficient for sufficiently low traffic situations. Irr this case, the
particular
cell utilizes only TDMA.
Nevertheless, the mobiles still suppott FDM operations so they can move to different cells and switch
freouencies.
9.3 CODE DIVISIoN MULTIPLE.AG0ESS CELLUI.AR 445
used by an individual channel is much wider than the inherent bandwidth required by
the message being transmitted. Spread $pectrum systems have traditionally been used
in military applications where the increased complexity of implementation is
justified
by two particular features. First, it is relatively difficult to detect the presence of a
spread spectrum signal because the signal energy is spread acro$$ a wide band and is
often masked by background noise. Second, it is more difficult to
jam
a spread spec-
trum signal because the
jamming
signal energy must be spread across a wide band as
opposed to being focused into a relafively narrow band. The performance
of a spread
spectrum receiver is comparable to the pelformance of receivers for traditional nar-
rowband signals as long as the spread spectrum receiver knows and can synchronize
to the method being used to spread the spectrum.
Two general categories of spread spectrum communications are frequency-hopping
systems and direct-sequence modulation systems. A frequency-hopping system is one
in which hansmission at any particular instant is confined to a relatively narow band
of frequencies commensurate with the inherent bandwidth of the message. Instead of
staying within one particular band as a conventional communications system does, a
frequency-hopping system
jumps
between narrow bands of frequencies within a large
block of spectrum in some prescribed manner. As mentioned in the previous section,
a GSM cellular system has the ability to operate with frequency hopping
(specifically,
slow frequency hopping).
9.3.1 CDMA Channel Establishment
A contemporary cellular CDMA system, as developed by
Qualcomm
and stand-
ardized by EIA/TIA in Interim Standard IS-95
[5],
uses direct-sequerce rtpectrum
spreading. One particular method of implementing a direct-sequence spectrum
spreading system is shown in Figure 9.5. In this system the source data is
"exclusive
ored" with a relatively long digital codeword. In essence, the
"exclu$ive
or" process
replaces a I of the source data with the
given codeword while a 0 of the source data is
replaced with the bitwise complement of the codeword. If a codeword contains n bits,
the occupied spectrum of the hansmitted signal is n times as wide as if the source data
Chip Rste Clock
Oanier
Figure 9.5 Direct-sequence generation of spread spectrum signal.
446 DIGITAL MOBILE TELEPHONY
were directly transmitted. Signals of other CDMA channels occupy the same band of
frequencies but do so with different speckum-spreading codes, which allows separa-
tion of the signals in the signal processing
circuinry of a receiver.
The basic process
of separating CDMA channels in a receiver involves correlating
a received signal with each of the various codewords (i.e.,
channels) assigned to the
cell. The correlation process produces
a correlation measurement by subtracting the
number of mismatches in a codeword from the number of matches. Table 9.3 lists an
example set of codewords with particularly
useful correlation properties.
As indicated
in the table, a codeword has seven matches with itself and no mismatches for a net cor-
relation of +7. A source data value of 0 produces
no matches and seven mismatches
for a net correlation of
-7.
The measurement weights for all other codewords are either
+ 1 or
-
I
,
depending on the values of the source data. If all seven channels defined in
Table 9.3 are active in a single cell, the worst-case interference between the codes pro-
duces an interference value ofeither +6 or
-6.
Thus, the desired data can be recovered
with a discrimination threshold of 0 for each channel.
(The
worst-case net measure-
ment values are +l for a I and
*l
for a 0.) Notice that same correlation propefties
exist
for all channel codes with respect to the other codes in the table.
Example baseband waveforms for the seven-channel CDMA sy$tem of Table 9.3
are shown in Figure 9.6. Notice the channel 0 receiver measurement is positive,
which
implies a data value of l. In this example interference from the adjacent channels ac-
tually enhanced the channel 0 measurement from an expected value of 7 to 9.
Discriminating between nominal measurement values of +7 for a desired signal
and +6 for worst-case interference is obviously very tenuous, particularly
because the
individual channels will be received with different power
levels. It i$ importanr ro
note, however, that an interference value of +6 can arise only if the data values of all
intedering channels destructively coincide. on average, the composite interference
has an average value of0. The length ofa spectrum-spreading code in IS-95 is actually
64 bits long, which means that the example worst-case interference would be 63 in re-
lation to a desired channel value of 64. The chances of destructive interference from
all 63 channels, or even a large number of channels, is astronomically small.
TABLE 9.3 Example Dlrect-$aquence Spectrum-Spreading
Godesa
Channel Number Channel Code
Number of
Matches
Number of
Mismatches Net Correlation
0
1
2
3
4
c
6
1 1 10010
01 1 1001
101 1 100
0101110
00101 1 1
100101 1
1 100101
7
3
3
3
3
3
3
0
4
4
4
4
4
4
+7
-1
- 1
-1
- 1
- 1
-1
aspectrum-spreading
code of dsirBd channel: 1 1 1 0010.
9.3 CODE DIVISIoN MULTIPLE-ACCESS CELLUTAR 447
Chnffid ftti
Numbcr Vf,hr$
(0) {
(1) 0
fr)
'l
(3) 0
({) o
(5) 0
(6) |
Figure 9.6 Example seven-channel CDMA encoding and decoding.
Example 9.1. Determine the
probability
of maximum interference of a 64-channel
CDMA system with 64-bit spreading codes. Also determine the effective
signal-to-interference power ratio of the same CDMA sy$tem. Assume all channels
operate at the same effective power level at the receiver and that all channel codes have
a cro$s correlation of+l bit.
Solution. The probability
of 63 destructive interferers is merely the
probability of
occurrence of 63 equally likely binary events:
prob(max interference)
= (0.5)63 =
I x 10-1e. The value of a desired receive signal is the autocorrelation of a codeword
with itself and can therefore be represented as a value of 64. The interference level is
the sum of 63 binary random numberrt with equally likely values of tl. Even though
a single interferer does not produce a Gaussian probability
distribution, the sum of a
large number of independent random vmiables approaches a Gaussian distribution
(central
limit theorem). The mean and variance of an individual interferer are easily
determined to be 0 and l, respectively. The mean and variance of a sum of 63 such
variables are 0 and 63, respectively. The signal-to-interference ratio is now
determined as
F'42
SI R= l 0t ogr ot i =
l sdB
As presented in Chapter 6, the 18-dB SIR result of Example 9.1 is quite
sufficient
to support an acceptable elror rate if the effective
power level of all channels can be
Chf,nnloCdi
"-.'
1
*i--j
Clilnnl0 Mtsrruttffil
448 DIGITAL MoBILE TELEPHoNY
maintained to be equal. In actual practice,
a CDMA deployment does not use all
pos-
sible codes in a single cell,
just
as an FDM sy$tem does not use all frequencies in one
cell. Thus, the amount of interFerence in a cell is limited by the number of codes as-
signed to the cell and a lesser amount of interference from adjacent cells (assuming
the adjacent cells do not use identical, synchronous spectrum-spreading
codes). The
effect of varying power levels in the interfering channels is covered in some problems
at the end of the chapter.
9.3.2 CDMA Multlpath Tolerance
A primary
advantage of a CDMA transmission system is its robustness in the presence
of multipath conditions. The basic reason for multipath tolerance can be appreciated
by examining the example codes given
in Table 9.3. Notice that each code is a cyclic
shift of all other codes in the table. Because the selected codes have low correlation
with each other, a delayed version of any particular
code has the same low correlation
with an undelayed version of itself. Thus, the effect of a multipath delay of more than
one spread spectrum bit (referred
to as a chip) is no more than the effect of the inter-
ference fiom another cDMA channel, even if the delayed version is at the $ame power
level as the primary
signal.
The effect of a multipath condition on a D-AMPS channel may be much more dele-
terious. Because a D-AMPS
system operates at a relatively nEurow band of frequen-
cies, it is possible
that a complete fade occurs for a particular
channel at a particular
physical
location. If the user moves through the physical
location, the effect of the fade
is a momentary dropout. If the user dwells at the location, the connection can be lost
unless a handoff occur$ (to
a new frequency and/or an adjacent cell). The slow fre-
quency-hopping
feature of GSM ameliorates the effect of a complete multipath fade
to a greater
degree, but not as effectively as a cDMA system. In GSM there may be
momentary dropouts, but they are not long enough to cause a dropped connection and
do not require a handoff, even if a user dwells in a physical
location where one par-
ticular frequency is totally lost. In essence, the frequency-hopping process
ofGsM is
equivalent to repeated and automatic handoffs to different frequencies.
From a somewhat philosophical point
of view, the basic intent of a GDMA system
is to equalize the performance
of all channels in the system. In a FDIvI/TDMA system
it is likely that some channels operate with very high performance
while others operate
at very low performance
or cannot be used at all. The existence of high-performance
channels does not compensate for the existence of low-performance
channels. Thus,
a system that equalizes the performance
of all channels has a greater
total capacity.
The primary
reason for variable channel performance
in FDIWTDMA is multipath
fading. An FDM/TDMA transmitter typically operates with a certain amount of ex-
cess
power
referred to as fade margin. The excess power is not much of a
problem
with
adjacent-channel intetference because it is relatively easy to isolate FDM channels
with frequency guard bands and TDM channels with time guard bands. cochannel in*
terference from one cell to another is the crux of the
problem.
If a
particular
channel
9.s coDE DrvrsroN MULTTpLE-AccEss cELLUIAR 449
is operating at an exce$$
power level, that channel cannot be reused except at a rela'
tively larger distance.
9.3.3 CDMA Power Control
Effective transmit power control of the mobile units in a CDMA system is
both a re-
quirement and a benefit. It is a requirement because a mobile transmitter that is close
to a base station receiver will obliterate other mobiles in the same cell that are farther
away. This is referred to as the near-far problem of CDMA. If the transmit power lev-
els of all mobiles in a CDMA sy$tem are controlled to be no higher fhan absolutely
necessary, the channels can be reused more often. Although power control is used in
FDM/TDMA systems, it is not possible to operate with bare minimum power levels
because the system cannot respond fast enough to adjust the power levels for fast mul-
tipath fades. A side benefit of using minimum
power
levels in a CDMA mobile is in-
creased battery life.
The power
levels of a mobile are controlled in two ways; open loop and closed
loop. In the open-loopmode, amobile can determine its transmitpowerlevelby meas-
uring its received power level under the assumption that transmission losses are equal
in both directions. This assumption is reasonable for a CDMA system but not for a
FDM/TDMA system because the latter are much more $usceptible to independent
frequency-selective fading
(multipath).
Closed-loop power
control involves base sta-
tion measures of mobile received power and adjustments to the mobile power levels
with the control channel. Power conftol in IS-95 CDMA is described in references
[6*8].
Table 9.4 lists
the basic parameters of the IS-95 CDMA digital cellular system for
the 800-MHz cellular band. The speech compression algorithm uses
QCELP
(Qual-
comm code excited linear prediction). The first commercial installation of CDMA oc-
curred in Hong Kong in 1995.
9.3.4 CDMA Soft Handoff
A unique feature of a CDMA system is the ability of a mobile to simultaneously re-
ceive from more than one source. Because each cell in a CDMA cellular network
TABLE 9.4 lS-95 CDMA Moblle Telephone Sy8tem Parameters
Channel bandwidth
Voice-Coding Rate
Error control overhead
Aggregate channel rate
Code length
$pread spectrum channel rate
Modulation format
1.25 MHz
9.6 kbps
(maximum)
9.6 kbps
(downlink)
19.2 kbps
(uplink)
19.2 kbps
(downlink)
28.8 kbps
(uplink)
64 chips
1.2288 Mbp$
QPSK
(base to mobile), otfset QPSK
(mobile
to base)
450 DIGITAL MOBILE TELEPHONY
transmits in a common frequency, a common RF receiver inherently receives the
spread spectrum signal from all adjacent base $tations. Signals from multiple base sta-
tion$ can then be acquired with multiple baseband code correlators or by time sharing
a single correlator with the separate codes. In a soft handoff operation the same voice
signal is distributed to selected cells adjacent to a currently active cell. An active mo-
bile can then compare the
quality
of the signals and switch to the best one before dis-
connecting from the established base station.
A
"make-before-break"
operation is not feasible in an FDIIfITDMA 'rystem
wherein adjacent cells utilize separate frequencies that require separate RF receivers.
FDIVf/TDMA system$ inherently use hard handoffs, which require disconnecting from
one base station before connecting to a new one. Notice, however, that a soft handoff
operation in a CDMA system increases the background interference because of the
multiple active signals for a single connection.
9.4 PEHSONAL
COMMUNICATION SYSTEM
A personal
communication system (PCS) is a cellular system operating in a band of
frequencies at 1.9 GHz. The original concept for PCS included multiple, new features
and services beyond those offered by a basic cellular system. Some of the envisioned
features were single telephone number for multiple service$ (voice,
data, fax) and user
mobility for home or office use and location determination. Although some commer-
cial PCS offerings provide
some new user features, initial North American FCS sys-
tem$ afe basically cellular systems utilizing a new band of frequencies.
when the FCC allocated
the PCS frequencies for the united states, they did so
without stipulating which type of system should be deployed. Thus, any organization
that bids for and receives a franchise for PCS spectrum is free to choose whatever
type of system it want$ for providing
service to the public. As a result, North American
PCS systems have been developed with three different transmission formats:
D*AMPS, GSM, and CDMA. The D-AMPS implemenration follows EIA/TIA stand-
ard 15-136, which is basically a revision of IS-54 that incorporates digital control
channels. (IS-54
defines the use of an analog control channel for compatibility with
AMPS.)
9.5 VOICE PBIVACY AND AUTHENTICATION
Ensuring privacy ofconversations
and
preventing
fraud are two critical aspects ofcel-
lular telephone systems that are addressed more completely in the digital systems than
in the original analog systems. An FDM analog system is particularly vulnerable to
casual eavesdropping because a relatively simple scanner can be used to identify an
active channel and tune into the conversation. If the scanner has two receivers. the
eavesdropper can listen to both sides ofthe conversation, assuming the eavesdropper
is in a high enough location or close enough to the active mobile that it can receive
the uplink signal.
e.6 rRtDruM 451
Digital signals are inherently more complicated to intercept because an eavesdrop-
per may need to monitor the connection establishment proces$
and not
just
tune in af-
ter a conversation has started. Explicit encryption
parameters
for optional voice
privacy is established during call establishment, or pos$ibly during a conversation, by
exchanging privacy control messages in the control channel. An eavesdropper must
receive the relevant information $ent in both directions before eavesdropping is pos-
sible. Even without explicit encryption, an eavesdropper must be located somewhere
near the base station to
properly receive both sides of a TDMA conversafion. At other
locations within a cell the eavesdropper will encounter overlapping time slots that will
inflict bit errors into the eavesdropper's received data. A CDMA system produces a
similar effect when the
phase of two specffum-spreading codes from two separate mo-
biles coincide.
Authentication of a valid mobile station is significantly $trengthened
in digital sys-
tems. A major problem in analog cellular systems is the process of cloning, in which
disreputable persons monitor call establishment handshake procedures
to acquire
valid mobile equipment electronic serial numbers
(ESNs)
and
program
them into
counterfeit mobile units that are then used to place calls. Part of the improved fraud
prevention involves maintaining a more up-to-date database of valid mobile ESNs.
The strongest pnrt of the fiaud prevention
involves determining authorization codes
based on
past call history in both the base station and the mobile. The authorization
code determined by the mobile and sent to the base station must coincide with the
authorization code calculated by the base station before service is allowed. Additional
steps may require entry of access codes by the user or, in the case of GSM, the mag-
netic data card called the Subscriber Identification Module
(SIM).
9. 6 l Rl Dl UM
Iridium is a satellite-based system for telephone and two-way paging services. The
satellite system is a Low Earth Orbit Satellite
(LEOS)
$ystem, which means that signal
powers and antenna sizes can be reduced with respect to conventional geostationary
satellites.
(The
Iridium sy$tem orbits are less than 500 miles, versus the 23,000 miles
for geostationary satellites.) In addition to enabling lower power ffansmission, a
LEOS avoids the
propagation
delay of a
geostationary
satellite, which is a minimum
of 500 msec. A basic disadvantage of a LEOS system is the continuous movement of
the satellites with respect to ground
locations. Because of this, Iridium provides a large
number of satellites
(66)
so continuous coverage from at least one satellite is always
available.
Iridium phones are dual-mode phones.
A
phone first makes an attempt to place a
call over a terresffial cellular system* but will default to the satellite network if local
cellular coverage is not available. When communicating through the Iridium system,
a user first gets connected to the nearest available satellite. From there the communi-
cation might retum to the ground or be relayed through multiple satellites before re-
.Multiple
versions of hanclhelds are available to operate on mMA, GSM, or CDMA cellular networks,
452 DIGITAL MoBILE TELEPHoNY
TABLE 9.5 lrldlum System Parameters
Number of satellites
Satellite orbit
Orbit
period
Transmission protocol
RF modulation
Speech data rate
Radio frequency
66
485 miles (780
km)
100 mi n
FDMA/TDMA
QPSK
2,4 kbps
1.6 MHz handheld to satellite
23 GHz satellite to $atellite
19 GHz uplink to satellite
29 GHz downlink from $atellite
turning to the ground.
The second ground
link might be direct to another Iridium user
or it may involve a base station with interconnection to a
public
telephone network.
System aspects of the Iridium system are provided
in Table 9.5.
9.7 TRUNKED RADIO
The term trunked radio generally
refers to Private Mobile Radio
(pMR)
communica-
tions networks. Originally, users of PMR equipment were allocated specific operating
frequencies dedicated to each user (or
organization). Such allocations were obviously
inefficient in terms of bandwidth utilization when the users did not have continuous
need for radio communications. Significant improvements in efficiency are achieved
when the $eparate channels are placed in a
group
and shared by a larger group
ofusers
on an as-needed basis. When a user of a trunked radio sy$tem needs service, the radio
equipment accesses an idle channel that becomes temporarily assigned to that user. As
soon as the users of a particular channel cease to transmit, the channel becomes avail-
able for other users. Access to a communications
channel utilizes a control channel
and a centralized controller for resource allocation.
A trunked radio system is not a cellular system but does utilize a central node re-
ferred to as a repeater. The repeater receives a signal on one frequency, shifts it to an-
other frequency, and transmits the signal on the new frequency. Thus, end u$ers do not
communicate directly with each other.* Transmission through the repeater is more ef-
fective because the tower is located at a high point
in the coverage area utilizing a rela-
tively high transmit power
that facilitates communications between end users who are
likely to not have an adequate path
between them.
In a general sense, trunked radio systems are PMR systems that provide
radio com-
munications directly between users without use of the public
telephone network.
However, the distinction between a trunked radio system and a cellular system has be-
*$ome
ffunked radio equipment does support a
(special)
two-way mode of operation in which the users
cofiununicate directly with each other. This mode of operation is
generally
reserved for communication s
in outlying areas where the rcpeater cannot prbvide seryice.
HEFERENCES 453
come bluned with U.S. offerings from companies like Nextel and Oeotek. These com-
panies utilize technology developed by Motorola that augments the use of spectrum
previously used for private radio service referred to as Specialized Mobile Radio
(SMR).
SMR applications typically involve fleet operations such as taxi cabs and de-
livery vehicles in need of dispatch services wherein multiple mobiles simultaneously
hear transmissions on a courmon channel. SMR radios utilize analog FM/FDM trans-
mission with 25-kHz channels.
The new equipment, generically
refened to as Enhanced Specialized Mobile Radio
(ESMR),
upgrades analog SMR transmission to digital TDM transmission in the same
manner that D-AMPS upgrades AMPS. ESMR transmission protocols allow for six
digital TDM channels in each 25-kHz channel. ESMR systems suppor"t conventional
dispatch applications and other similar specialized $ervices such as private group call-
ing. Most importantly, for this chapter, an ESMR system can provide connections to
the public telephone network-referred to as an interconnect feature. With this fea-
ture, an ESMR system can function as a cellular telephone system.
9.8 CELLULAR DIGITAL PACKET DATA
Cellular Digital Packet Data
(CDPD)
[9]
is a system designed to
provide data services
up to 19.2 kbps as an overlay of an AMPS installation. The primary service concept
is to ofTer fixed and mobile data services independent of the need to upgrade an AMPS
sy$tem to a D-AMPS system. CDPD uses AMPS channels that are not being used
for voice. Because it is a packet-oriented data service, multiple users share a single
30-kHz channel-a significant savings in spectrum usage art compared to the use
of connection-oriented voice chamels for data. Principal applications for CDPD are
mobile Internet access and credit card verification.
Access to the shared channel is accomplished with Digital Sense Multiple Access
with Collision Detection
(DSMA/CD)
which is similar to CSMA/CD of other radio
systems and Ethernet LANs. CDPD uses GMSK modulation with RS(63,47) forward
error conection.
REFERENCES
"Vector
Sum Excited Linear Prediction (VSELP) 7950 Bit Per Second Voice Coding
Algorithm," Technical Description, Motorola, Schaumburg, Illinois, Nov, 14, 1989.
M. Rahnema,
"Overview
of the GSM System and Protocol Architecture,"
IEEE
Communications Magazine, Apr. 1993, pp. 92-100.
P.Vary, K. Hellwig, C, Galland, M. Russo, J. Petit, and D. Massaloux,
"Speech
Codec
for the European Mobile Radio System," in IEEE GLOBECOM 1989, Nov, 1989, pp.
29.8.2.
R. Dixon, Spread Spectrum Systemt with Commercial Applications, Wiley, New York,
1994.
"Mobile
Station-Base Station Compatibility Standard for Dual-Mode Wideband
Spread Spectrum Cellular System," EIA/TIA./IS-95, Washington, DC, July 1993.
454 DIGITAL MOBILE TELEPHONY
A. Salamasi and K. S. Gilhousen,
"On
the System Design Aspects of Code Division
Multiple Access (cDMA)
Applied to Digital cellular and
personal
communications
Networks," Praceeding of the Forty-First IEE VehicularTechnology Conference, May
1991, pp. 57-62.
A, J. Viterbi and A. M, Viterbi,
"Erlang
Capacity of a Power Conrolled CDMA
System," IEEE Journal on Selected Areas of Comtnunitations, Nov. lgg3, pp.
882-890,
K. S. Gilhousen, L M. Jacobs, R. Padovani, A. J. Viterbi, L. A. Weaver, and C. E.
Wheatley,
"On
the Capacity of a Cellular CDMA System," IEEE Transactions on
Vehitular Technology,May 1991, pp.
303-3 12.
A. K. Salkintzis,
"Packet
Data over Cellular Nerworks: The CDpD Approach," IEEE
Communications Magazine, June 1999, pp. 152-159,
PROBLEMS
9-1 what is the available bearer rare of a half-rate digiral channel in a D-AMps sys
-
tem?
9.2 what is the data rate of the slow associated control channel in a D-AMps sys-
tem?
9.3 what is the available bearer rare a full-rate digital channel in a GSM sy$tem?
9.4 what is the data rate of the slow associated control channel in a GSM system?
9.5 Determine the receiver output measurements for channels I and 2 for the GDMA
example of Figure 9.6.
9.6 what is the effective signal-to-interference ratio of a single cDMA uplink chan
-
nel operating at a distance that is twice as far from the base station as 62 other
channels? Assume a code length of 64, cross correlations of
+1,
and all trans-
mitters operate at identical power
levels. (a)
Assume all interferers are active.
(b) Assume half the interferers operate with a,25qo data rate because of no voice
activity.
9.7 In a CDMA system with a code length of 64 determine the signal-to-interference
ratio of a single uplink CDMA channel if there are 16 active interfering channels
operating at an effective receive power level that is 12 dB higher that the desired
channel.
1 0
DATA AND ASYNCHRONOUS
TRANSFER MODE NETWORKS
The primary focus of this book is to describe how and why digital electronic technol-
ogy is used to communicate voice, a fundamentally analog signal. Naturally, the digi-
tal telephone network can accommodate inherently digital information
(data), but only
if data can accept or be adapted to service characteristics established for voice. This
chapter discusses digital technology and networks specifically directed to supporting
data communications seruices. Ironically, the later part of this chapter describes how
the data-oriented
networks are adapted to carry voice.
Communication of discrete messages
(data)
is actually an ancient practice. In early
times communication over distances
greater than the range of the human voice was
provided by sight or sound of discrete signals
(e.g.,
heliographs, smoke signals, flags,
and horns). Furthermore, the first practical communication system to use electricity,
the telegraph, is inherently digital. As telegraphy evolved from the original, manually
based systems to fully automated systems, they developed into what are commonly re-
ferred to as me$sage-switching networks. Modem message switching is discussed in
the first section of this chapter.
As the need for modem electronic data communications arose, it was only natural
that the public
telephone network would be used for data kansmission services. The
availability overshadowed numerous technical shortcomings of a network designed
primarily for voice communications services. The main deficiencies of a conventional
telephone network for data transmission are:
1. Need for signal hansducers
(modems)
on analog access lines
2. Limited data rates
3. High error rates
(in
the older analog network)
4. Inefficient circuit utilizations
As data communications requirements increased, so did the
justification
for more
cost-effective data communications solutions. One solution to reducing data transmis-
455
456 DATAANDASYNcHHoNoUSTFANSFERMoDENETWoRKS
sion costs was to improve circuit utilizations through the use of packet-switching
net-
works. The technology of
packet
switching was pioneered
by the Advanced Research
Projects Agency of the u.s. government.
This agency developed a network referred
to as the ARPANET
[].
In addition to ARPANET, which was used only by
govern-
ment, educational, and industrial research institutions, a number of public packet-
switching networks were also developed in the United States and around the world.
ARPANET developments eventually evolved into what is now known as the Internet.
A second approach to improving data communications involves developing sepa-
rate networks specifically designed for digital ffansmission (no
analog circuits with
modems). The first major enterprise of this type in the United States was a nationwide
digital microwave network developed by Digital rransmission corporation (DA-
TRAN). Because the data communications marketplace of the time could not support
a separate, dedicated network for data, DATRAN ran into financial difficulty and de-
clared bankruptcy in 1977.
Another, more successful, data-networking
approach was the Dataphone Digitat
Service (DDS)
offering of AT&T. This service utilizes digital technology of the tele-
phone
network for strictly data applications. DDS circuits are dedicated to data serv-
ices, but the facilities and routes are shared with telephone network facilities. A major
hurdle for DDS is achieving digital access to DDS circuits. If a subscriber is ourside
the range of digital transmission facilities of a DDS serving office, a voiceband mo-
dem over an analog line is required.
A fourth approach to satisfying data communications
services involved developing
means for directly accessing the digital ffansmission and switching equipment of the
telephone network. The first widespread approach of the telephone companies for pro-
viding universal digital access is the Inregrated services Digital Nerwork (ISDN).
ISDN provides
digital access to the digital facilities of the telephone nerwork for voice
or data services on a call-by-call basis. ISDN digital subscriber lines and other
methods of digital access to digital networks such as the Internet are described in
Chapter I L
10.1 MESSAGE
SWITCHING
As one telegraph system after another was installed in the countries around the world,
nationwide communications
networks evolved. A message could be sent from one
point
to another even if the two points were not serviced by a common telegraph line.
In this case, telegraph operators at intermediate points
would receive a message on one
line and retransmit it on another. When a telegraph office had several lines emanating
from it, the process
of hansferring a message from one line to another was, in essence,
a switching function.
The process
of relaying, or switching, a message from one telegraph line to another
became semiautomated when teletypes with paper
tape punches
and readers were de-
veloped. An incoming message could be punched
automatically onto a
paper
tape by
one teletype and subsequently read by another teletype for transmission on the appro-
10.1 MESSAGESWITCHING 457
priate
outgoing line. The process of transferring a message from one line to another in
this manner led to these systems being referred to as torn-tape message switches.
One of the world's largest me$sage switches was completely automated in 1963
when Collins Radio Company of Cedar Rapids, Iowa, installed a computer-based
me$$age switch for the airline companies of North America. This system and the more
recent successors eliminate paper tape transfers by storing incoming messages direcfly
into a computer memory
(disk
file) and forwarding them automatically to the appro-
priate output line when available. Hence this mode of operation is often refened to as
store
-
and-forward message switching.
Included with each message is a header containing an address and possibly routing
information so the message proce$sor at each node can determine to which output line
to $witch the message. As indicated in Figure 10.1, the
processor
in each node main-
tains message queues for each outgoing link. These queues are normally serviced on
a first-come, first-served basis. However, priority information can $ometimes be in-
cluded in each header to establish different classes, or grades of service, thereby al-
lowing tirne-critical messages to be placed at the head of a queue.
A message-switching network is fundamentally different from a circuit-switching
network in that the source and destination do not interact in real time. In fact, most
message-switching networks could deliver a message on a delayed basis if a destina-
tion node is busy or otherwise unable to accept traffic. In a message-switching net-
work there is no need to determine the status of the destination node before sendins a
message, as there is in circuit switching.
Figure 10.1 Message-switching network.
458 DATAAND ASyNcHHoNous THANSFER MoDE NETwoRKS
Message-switching networks are also fundamentatly different from circuit-switching
networks in their response to traffic overloads. A circuit-switching network blocks or
rejects excess traffic while a message-switching network normally accepts all traffic
but provides
longer delivery times as a result ofincreased queue lengths.
Another importaxt distinction of a message-switching network is that the transmis-
sion links are never idle while traffic is waiting to use them. In a circuit-switching
net-
work, a circuit may be assigned to a particular
connection but not actually carrying
traffic. Thus, some of the transmission capacity may be idle while some user$ are de-
nied service. In contrast, utilization of the transmission links of a message-switching
network is directly related to the actual flow of information. Arbitrarily high utiliza-
tion efficiencies are possible
if increased store-and-forward queuing
delays are ac-
ceptable' Chapter 12 provides basic results of
queuing
theory that relate utilization
efficiency to queuing
delay.
10.2 PACKET SWITCHING
The circuit-switched telephone network is ill-suited to interactive data traff,rc because
it is fundamentally designed for less frequent service requests with comparatively
long holding times
(3-4
min on average). Neither the control elements in the switches
nor the capacity of the signaling channels are capable of accommodating frequent re-
quests for very short messages. The result is that connection setup time may be many
times greater
than the holding time of a data message. Obviously, more efficient utili-
zation of the network requires greater
control channel bandwidth and increased call
processing
capacities in the switches. Beyond this, however, interactive data traffic
with low-activity factor$ requires a network operation that is fundamenta-lly
different
from a conventional circuit-switched
network, The most appropriate mode of opera-
tion for traffic that comes in bursts is more closely related to a message-switched net-
work than to a circuit-switched network.
Figure 10.2 depicts both the conceptual structure and the conceptual operation of
a packet-switched
network. A single message at the source is broken up into
..packets"
for transmission through the network. Included in each packet
is a header containing
address and other control information. Each packet
is relayed through the network in
a store-and-forward fashion similar to a message-switching network. At the destina-
tion node, the packets
are reassembled into the original contiguous message and de-
livered.
The main feature of a packet-switching
operation is the marurer in which the tran$-
mission links are shared on an as-needed basis. Each packet
is transmitted as soon as
the appropriate link is available, but no transmission facilities are held by a source
when it has nothing to send. In this manner, a large number of relatively inactive
sources can share the transmission links. In essence, link utilization is improved at the
exPense of storage and control complexity in the nodes. A circuit-switched network
ha$ conffol overhead associated with connection setup but very little control thereaf-
ter' In contrast, packet-switching
nodes must process
the header information in each
packet
as it arrives. Thus a long message in a packet-switched
network requires more
ro,? PAoKETSWTToHTNG 459
Flgure 10.2 Packet-switching network.
conhol overhead than if it were serviced in a circuit-switched network. Considering
the declining cost of digital memory and
processing,
the increased control complexity
becomes less and less significant as digital technology advances. As discussed later in
this chapter, one particular variation of packet switching, Asynchronous Transfer
Mode
(ATM),
is designed to specifically support hardware implementations of con-
trol intensive functions, thereby supporting very high traffic volumes with low delay.
As the traffic load in a packet-switched network increases, the average transmis-
sion delay increases correspondingly. ln contrast, a circuit-switched network either
grants service or rejects it. There is no
graceful
degradation in service. Conversely,
when only a few circuits are in use in a circuit-switched network, much network trans-
mission capaciry is idle. When there is a light load on a packet-switched network, the
active users benefit by shorter than usual delay times. Hence from a user's grade-of-
service point of view, the two network types are fundamentally different.
Using automatic repeat request
(ARQ
error control, packet-switching networks
(e.g.,
ARPANET) traditionally
provided essentially error-free transmission for each
node-to-node transfer. This process requires the receiving nodes to monitor redundant
check bits appended to each packet to determine if the
packet was received correctly.
When eruors are detected, a retransmission is requested
(message
NAK). Hence trans-
mitting nodes must hold all transmitted
packets in memory until a positive response
(message
ACK) is retumed by the receiving terminal. Fufihermore, an entire
packet
is usually received and checked for errors before forwarding it to another node.
Customers typically access packet networks by way of leased lines or dial-up con-
nections. Dial-up connections are used by infrequent users, while leased lines are pre-
ferred by heavy users to achieve constant availability, higher dara rates, and possibly
lower error rates.
Despite the similarity to a me$sage-switching operation, a
work is different in two impodant re$pects:
fonsrrd nod*r
460 DATAANDASYNCHRoNoUSTHANSFERMoDENETWORKS
l. The store-and-forward
delay through a
packet-switchecl
network is relatively
short. Thus interactive communications can occur in much the same manner as
if a dedicated, end-to-end circuit is established.
2. A packet-switched
network doe$ not provide
$torage of messages, except in an
incidental manner while relaying packets from one node to another. The
network is designed to provide
switched communication between two nodes.
both of which
are actively involved in the communications process.
A
packet-switching
network does not normally store a me$sage for later delivery
to an inactive or busv terminal.
One reason for breaking messages into packets
is to allow transmission of the first
segment of a long message while other segments are in transit. If the entire me$$age
had to be received at each
node befbre forwarding it to the next node (as
in message
switching), the delays through the notles might be too large. Another reason for break-
ing the mes$ages into packets
arises from operational simplifications derived from
storing, processing,
and transmitting smaller and possibly fixedJength blocks of data.
In addition, if long message$ are transmitted intact, short messages experience exces-
sive delays when queued behind long messages. Packetization allows shon messages
to get
through a transmission link without waiting behind long messages. This same
principle
occurs in multiprogrammed
computers, which use time slicing to allow short
jobs
the opportunity of finishing before previously
started long
jobs.
one more moti-
vation for packetization
is that when a ffansmission block is too long, it is unlikely that
the entire message will be received correctly. Packetization provides
a means of re-
transmitting only those portions
of a message that need to be retransmitted.
10.2.1 Packet Formats
The format of a packet in a packet-switching
network can vary significantly f'rom one
network to another. Some formats include numerou$ fields fbr control infbrmation
while other system$ rely more heavily on special control packets
to transmit control
information. Generally speaking, the conffol information associated with a particular
message or link is included in the header of a mes$age packet. Less frequent, network-
related control information is communicated tlrough special control packets.
As indicated in Figure 10.3, a packet
contains three major fields: the header, the
message, and the redundancy check bits. some packet$
may not contain a message
field if they are being used strictly for control purposes.
Although a variery of tech-
niques for generating redundancy
checks are possible,
the most popular technique
uses cyclic redundancy checks (cRCs).
Basically, a cRC is nothing more than a set
of parity
bits that cover overlapaing fields of message bits. The fields overlap in such
a way that small numbers of errors are always detected and the probability
of not de-
tecting the occurrence of 2 large number of errors is l in 2M, where M is the number
of bits in the check code.
A header typically contains numerous subfields in addition to the necessarv address
field. Additional fields sometimes included in a header are:
IO,2 PACKETSWITCHING
/161
Hcdtr Trxt
cRc
rR
To
iddtr|8
Op
coda
From
EddrG$
$tq
BytE
count
I ,
Figure 10.3 Typical packet format,
An operation code to designate whether the packet is a message (text) packet
or
a control packet.
In a sense this field is a
pat"t of the destination address, with the
address specifying the control element of a switching node.
A source address for recovery
purpo$es or identification of
packets at a
destination node that is capable of simultaneously accepting more than one
message.
3. A sequence number to reassemble messages at the destination node, detect
faults, and facilitate recovery procedures.
4. A length code to indicate the length of a packet when less than a standard size
packet is
transmitted.
Some protocols insert special delimiters
(flags)
at the end
of a
packet and therefore do not use a length count.
10.2.2 Statittical Multiplexlng
The digital time division multiplexing techniques described in the
previous chapters
provide multiple channels by periodically assigning a time slot to each channel.
The
time slots are assigned whether or not the respective sources have anything to send.
Channel assignments may be altered on a connection-by-connection basis, but for the
duration of a
"call"
a particular time slot is dedicated to a respective connection. Be-
cause of the continuous manner in which time slots occur for each channel, this form
of multiplexing is sometimes referred to as synchronous time division multiplexing
(STDM).
In this section we describe another form of multiplexing, variously referred
to as asynchronous time division multiplexing
(ATDM)-
or statistical time division
multiplexing
(statmux).
This form of multiplexing is mentioned here because of its
close relationship to packet-switching
techniques.
Statmuxes operate with framing formats that are basically identical to STDM fram-
ing formats. The major difference is that a statmux
periodically redefines the length
of its frames to change the number of time slots and, hence, the number of channels.
Whereas an STDM system
permanently
assigns a time slot to each of its sources, a
"Here
is alother context-sensitive use of the terms synchronous and asynchronous. Il the context of d igital
telephony,
"synchronous
multiplexing" refers to combining tributary signals that have been synchroni zed
to each other as in upper layer SONET multiplexing. Corrospondingly,
"asynchronous
multiplexing"
refers to the use of pulse stuffing to accofimodate unsynchronized tributaries,
z.
462 DATAANDASyNcHHoNousTHANSFERMoDENETwoHKs
statmux assigns a time slot only when a source becomes active. A time slot is elimi-
nated
(the
frame shoftened) when the respective lrource becomes inactive.
statmux $ystems have been primarily
used to
provide
line sharing for a multiple
number of interactive
terminals communicating
with a host computer. when only a
few sources (terminals)
are active, each source has a relatively high data rate channel
available to it. As the number
of active sources increases, the frame length increases
so individual
channel rates decrease. some systems limited the number of active
sources to ensure certain minimum data rates.
The purpose
and performance
of statmux systems are very similar to the purpose
and performance
of a packet-switching
link. The main difference is that a
packet-
switching link hansmits larger blocks of data with a header included in each block.
Each time slot of a statunux
$ystem is shorter and contains only source data. Figure
10.4 contrasts the basic operation of message switching, packet switching, and statis-
tical time division multiplexing.
The message switch transmits each message in its en-
tirety in a f,rrst-come, first-served
manner.
packet
switching breaks messages up to
allow interleaving of packets
from other sources. Thus short mes$ages never ger
queued
behind long messages such as file transfers. The statmux system breaks the
messages up into even finer blocks
(words)
of data and adds periodic
frame rlefinition
messages so that receiving terminals can properly
identify the individual time slots
and switch the incoming data accordingly.
As indicated in Figure 10.4, a
packet-switching
operation becomes very similar to
a statmux operation if the size of the packets
is small. In fact, commercially available
irhrr|[e rwitohing
B B B 8 A 8 C B . C
trII trDtrDEDtrDEDEtrEDtrU
Ptrket sinching
f igure 10.4 comparison of message-switching, packet-switching,
and statistical
multiplexing.
B
-
c
-t
A
HI
iEdGflnltiod
10,? PACKET swrTcHrNc 463
statistical time division multiplexers can be used to build up a packet-switching net-
work. In addition to the multiplexers, however, a packet-switching node needs suffi-
cient storage to accumulate an entire packet, check it for errors,
process it, and
retransmit it.
Strictly speaking, a statistical multiplexer does not accumulate an entire
message or
packet. It switches the incoming data as they are received. Error control
functions
(redundancy
checking and requests for retransmission) are implemented be-
tween endpoints of a
"connection"
instead of between nodes. In summary, a statmux
is strictly a multiplexer/demultiplexer. A packet-switching node provides multiplex-
ing functions as well as message-level and networklevel conhol functions.
10.2.3 Routl ng Control
Much discussion and experimentation have been undedaken regarding various
proce-
dures for routing packets from sources to sinks through a
packet-switching network
[2].
All routing techniques allow for a certain amount of adaptation or alternate routing
to circumvent line or node failures. The various techniques differ, however, in how
fast they respond to failures and whether or not they circumvent network congestion
as well as equipment failures. The basic techniques are the following:
l. Dynamic routing
2. Virrual circuit routing
3. Fixed-path routing
Each of these techniques can be implemented in a variety of ways and can assume
some ofthe characteristics ofthe other routing control
procedures.
Dynamlc Routing
Dynamic routing is implemented on a distributed basis with network nodes examining
the destination address ofeach received packet to determine the appropriate
outgoing
link. The outgoing link is selected by
processing
locally stored information to deter-
mine which path provides minimum delay to the destination. The routing criteria are
routinely updated to include both the operational status
(health)
and the size of the
queues in the neighboring nodes. The routing decisions are adjusted rapidly enough
that individual packets of a single message may follow dffirent paths through the net-
work. Dynamic routing, with its ability to respond quickly to changes in network to-
pology or traffic conditions, is one of the original features espoused
for packet
switching. In fact, dynamic routing was $ometimes considered to be inherent in the
definition of a packet-switching network.
In spite ofthe obvious attraction ofbeing able to adjust to rapid fluctuations in traf-
fic
patterns,
dynamic routing has a number of significant drawbacks.
One implication
of allowing successive packets in a message to follow different routes through the net-
work is that packets may arrive at a destination out of sequence. Although sequence
numbers are used to rearrange the
packets properly, the reassembly
process is compli-
464 DATAANDASYNCHFIoNoUSTHANSFEFMoDENETwoFKS
cated, particularly
since the destination does not know if a missing packet is merely
delayed or lost entirely.
Another drawback to
dynamic routing is the possibility
of oscillation occuring in
the routing decisions. Ifthe bandwidth assigned to updating the routing control algo-
rithms is too small, a lightly loaded node will attract more traffic than it can handle
before neighboring nodes are informed of the change in the traffic flow. In this in-
stance, a
packet
might even wind up at a node from which it has previously
been sent.
Purely distributed control, in general,
and dynamic routing, in particular,
also pre-
sent difficulties with respect to flow control in a
packet
network. As mentioned in
chapter 7, alternate routing in the switched telephone network is sometimes sus-
pended when the network as a whole becomes overly conge$ted (alternate
routes re-
quire
more resources). obviously, the same principle
applies to a packet-switching
network' Flow control in packet
networks is discussed in later sections.
Dynamic routing is most appropriate
for
small networks or in a military environ-
ment where survivability
of the network in the presence
of muttiple-node failures is a
requirement' A military network i$ typically more concerned with reliable 4d timely
completion of a few important messages than with achieving the highest possible
throughput from a given amount of resources.
Virtual Circults
A virtual circuit network
embodies some of the basic aspects of both circuit switching
and packet switching. The transmission capacity is dynamically assigned on an
,.as-
needed" basis, but all packets
of a multipacket message follow the sarne route through
the network. Before interactive communication
begins, a route is established through
the network, and all participating
nodes are informed of the
"connection"
and how to
route the individual packets
that follow. From then on, all packets flowing between
the two end points follow the same route through the network. In essence, a virtual cir-
cuit is a logical concept involving addresses and pointers in the nodes ofthe network
but no dedicated transmission facilities. At the end of a connection
(or
..session"
in
data communications
terminology), a virtual circuit is released by a
..disconnect"
me$-
sage propagating
through the network.
Except in the case ofpermanent virnral circuits, separate connections, or sessions,
involving the same two endpoints do not necessarily use identical paths
thrcugh the
network' Each virtual circuit is established during the call establishment phase
de-
pending
on the traffic pattern$
at the time. Thus a virtual circuit network can respond
to network failures or changing
traffic patterns,
but in a longer time frame than a dy-
namically routed network. When virtual circuits are changed from one connection to
the next, the mode of operation is sometimes referred to as a switched virtual circuit
networkby direct analogy to conventional circuit swirching.
Virtual circuits can be established using either distributed or cenffalized conffol.
when distributed control is used, the call establishment message propagates
through
the network with each node making a local decision as to which outgoing link should
be selected.
10.2 PACKET swtrcHtNc 465
As discussed in Chapter 7 concerning conventional circuit-switched networks, cen-
tralized call establishment has the basic advantage of being able to set up circuits with
a networkwide view of network status and traffic
patterns. The TYMNET
network of
the United States
[3]
and the DATAPAC network of Canada
[4]
are examples of vir-
tual circuit,
packet-switched networks with centralized conffol.
Since the call establishment
phase ofa virtual circuit represents excessive overhead
for single-packet me$sages, a virtual circuit mode of operation is obviously most use-
ful when a network services a relatively
large number of multiple-packet messages or
$essions. Thus a dual-mode network is suggested requiring vittual circuits for longer
messages and immediate transmission with dynamic routing for single-packet mes-
sages. In this instance, the single-packet messages are usually referred ta as data-
grams.
One of the main advantages of a virtual circuit operation is its ability to provide
more orderly control of packet delivery. If a node in a virtual circuit never forwards a
packet pertaining
to a particular connection until the previous packet has been ac-
knowledged, packets cannot arrive out oforder. A second advantage ofa virtual circuit
is the reduced addressing requirements of individual packets. Once a virtual circuit has
been established, complete destination addresses are no longer needed in the packets.
In place of destination addresses, virtual circuit identifiers can be used that are local
to each link. In essence, virtual circuit identifier$ are pointers to memory addresses in
the call processors of the packet-switching nodes or to look-up tables in ATM nodes.
The designated memory addresses contain the pertinent information conceming the
actual source, destination, and routing of the
packets. Another important feature of a
viftual circuit operation is its inherent ability to aid in flow control, as discussed in the
next section.
The main disadvantage of a virtual circuit operation is the possibility
of
greater
transmission delays. When a path for a virtual circuit is established, it is chosen to
minimize the delay through the network under the traftic
pattems at that time. If the
traffic pattems
change, packets pertaining to a
particular
virtual circuit may experi-
ence long queuing delays on some links while alternate links are more lightly loaded.
Yum and Schwartz
[5]
report that analyses of routing techniques for several small net-
work configurations indicate a packet delay improvement of lA-ZlVa is
possible when
adaptive routing is used in lieu of a fixed routing rule.
When first considered, call establishment might seem to be a disadvantage of a vir-
tual circuit network. Actually, however, flow control considerations require some type
of query/response mes$age to determine the status of the destination nodes before a
$ource is allowed to begin sending a
packet
$trearn.
Thus the control overhead and the
delay associated with setting up a
"connection"
is usually a fundamental requirement,
even in a dynamically routed network.
Fixed-Path Routlng
Fixed-path routing embodies the same basic concepts of a virnral circuit network ex-
cept successive connections, or sessions, between any two endpoints always use the
466 DATAANDAsyNcHRoNousTHANsFEBMoDENETWoRKS
same path'
In essence a fixed-path network is one that assigns pemanent
virtual cir-
cuits to each pair
of endpoints.
One attractive f'eature of a fixed-path network is the absence of the call estab-
lishment phase
of a virtual circuit network. However, unless the necessary resources
are perrnanently
allocated, a
"clear-to*transmit"
message is needed before sending
packets. Inactive virtual circuits do not tie up network resource$ in the same manner
as conventional circuits,
but each virtual circuit does require transmission capacity
and store-and-forward
buffers in a probabilistic
sense. If minimum grades
of service
(delay
times) are to be guarauteed,
a network mu$t limit the number of virtual circuits
existing at any particular
time. Hence clear-to-tranrrmit
signals are needed befbre a
fixed-path circuit becomes active. of course, a network can provide
two modes of op-
eration: permanently
active (hot
line) virtual circuits and virtual circuits activated and
deactivated as needed.
An obvious disadvantage of a
purely-fixed-path
network, as described so far, is its
vulnerability
to node or link failures. To circumvent this problem,
a network control
center usually assigns semipennanent paths
through the network that are changed only
when necessary for failure survivability or maintenance. Some older message-switching
networks were implemented with fixed-path
routing in the strict sense. This mode of
operation was more appropriate to merrsage-switching
networks because message de-
livery was less time critical and could be deferred while repairs were undertaken for
inoperative
equipment.
10.2.4 Flow Control
As discussed in ChapterT concerning conventional circuit-switched
networks, routing
and flow control are two closely related operational requirements
of any communica-
tions network. The same basic principlb
fbr controlling the flow in circuit-switched
networks also applies to packet-switched
networks. Namely, whenever
the network is
experiencing excessive sffess, either from a loss ofcapacity due to failures or from an
abnormally heavy demand fbr services, new service requests must be blocked at the
periphery
of the network before they tie up common reriources and compound the con-
gestion.
In a packet-switched
network, the common resources are store-and-forward
buffers and transmission links.
Flow control in a packet
network is primarily
concerned with buffer managemenr.
For example, if all store-and-forward
buffers in adjacent nodes become filled with
packets
destined to each other, the nodes are unable to receive additional packets,
and
a deadlock exists. Recall that, in pure packet
switching, a node does not release a buff-
er as soon as it transmits a packet.
The buffer is releasecl when an acknowledgment is
returned from the adjacent node. If a receiving node has no available buffers, it cannot
accept a new packet
and therefore cannot acknowledge it. (Frame
relay ancl ATM net_
works described later do not provide
node-by-node error control so they do not save
copies of messages in internal nodes of the network.)
Flow control requirements imply that interface nodes in a packet-switched
network
are aware of overload conditions
and refuse new requests for service until the conges-
10.2 PACKET SWITCHING 467
tion is relieved. A parficularly attractive feature of a virtual circuit network is that the
call establishment
phase provides
an automatic means of determining whether or not
a particular requertt should be serviced. If no path
through the network can be estab-
lished because the interior nodes are too conge$ted, the request is rejected. On the
other hand, if a virtual circuit is established, there is reasonable expectation that the
entire request will be serviced in a timely manner.
Unless network nodes are very conservative in accepting request$ for new virtual
circuits, the ability to set up a circuit does not
guarantee the avoidance of excessive
congestion or deadlocks. A node accepts a new virnral circuit based on an expected
load and its capacity to service that load. If the tratfic volume and pattems happen to
exceed the expected load, excessive delay or congestion is possible.
Virtual circuits are an effective means of conrolling the flow of multiple-packet
mes$ages, but they represent too much overhead for single-packet or datagram flow
control. If a virtual circuit network must support a significant number of single-packet
mes$ages, it can allow immediate transmission of these mes$ages and forego the call
establishment
phase. In terms of transmission
overhead, a datagram is not much
greater than a call establitrhment packet. Thus, from this point of view, the packet
might as well be sent immediately and be considered its own circuit setup message.
In terms of store-and-forward buffers, however, a single-packet
message is much dif-
ferent from a call establishment
packet. A call establishment
packet requires a certain
amount of storage and processing by the call
processor
of each network node it
reaches, but it does not compete for store-and-forward buffers as does a me$sage
packet. Ifnecessary, a call establishment message can be ignored by an overloaded node,
and the originating node times out waiting for the network respon$e and reissues the re-
quest. The time out should be long enough that the network has had a chance to rela,r.
In confrast to call establishment packets, if a message packet is ignored by an over-
loaded node, congestion migrates to the node that last transmitted the packet, since this
node is holding a copy of the rejected
packet in a buffer. The buffer cannot be released
until an acknowledgment is received. Hence datagram$ cannot be allowed to enter the
network unless a reasonable chance for complete
passage exists. When the network is
heavily congested, attempts to set up virtual circuits might also be suspended.
A conventional circuit-switched network is unconcerned with flow control be-
tween the endpoints of a connection since, once the circuit is established, the activity
or inactivity ofthe endpoints
has no effect on other connections
or on the network as
a whole. End users necessarily administer flow control between themselves so the
source does not overrun the receive buffers of a sink. These procedures concern only
the endpoints.
In contrast, the very nature of a packet-switching network implies direct involve-
ment with endpoint activity. If a source pumps excessive traffic into a network, other
users experience degraded
performance. Hence interfaces to a
packet-switching net-
work necessarily include flow control for respecfive sources. Source flow control es-
tablishes a maximum data rate for a network, If a sink accepts data at a lower rate for
a sustained period of time, this fact must be communicated to the network node serv-
ins the source in order to slow the source down.
468 DATAANDAsyNcHRoNoUsTRANSFEHMoDENETwoRKS
store-and-forward
buffers of a packet-switching
network are needed for commu-
nications purposes
and are not used as a storage medium for messages. If a packet-
switching network also provides
message-switching
services, message srorage
functions should be implemented
separately from the communications
buffers. Then.
the packet-switching
network is used to transfer data to the message $torage facility.
Another implication of maximizing store-and-forward
buffer utilization is the need
to wait until a sink is ready to accept data before a source begins sending. Ifa sink is
not ready, packets get
stuck in store-and-forward
buffers at the far end of a network
and cause congestion. Thus some form of request-to-transmit/clear-to-transmit
se-
quence is needed before message transmission begins. This requirement is inde-
pendent of the routing algorithms employed in the network. Hence, in actual practice,
the setting up of a virtual circuit may not represent a time penalty.
single-packet meri$ages (datagrams)
can be an exception. If they are ffansmitted
without a clear-to-transmit signal, they may be discarded at the destination node if the
sink is inactive or has no receive buffers available. The destination node then needs to
return a rejection message to the source indicating the status of the sink. In this man-
ner, the message itself is a requerrt-to-transmit
signal. The ARpANET used a flow con-
hol strategy wherein single-packet messages serye a$ their own requests for buffer
storage but multiple-packet messages require preallocated
buffers in groups
of eight
at the destination
[6,
71.
Flow control in TYMNET is implemented in a different manner because of its ex-
clusive use of virtual circuits for all messages. Before any node in the network can
send a packet
to a neighboring node, it must receive a clear-to-transmit signal from
the neighboring node for the particular
virtual circuit. The clear-to-transmit
signal is
an indication that a specified number of store-and-forward buffers are being held in
reserye for that particular
virtual circuit. After a node sends the specified number of
packets, it sends no more until another clear-to-transmit
signal is received (indicating
the previous packets have been forwarded or more buffers have been allocated to the
virtual circuit). By using the same node-to-node
method of flow control at the inter-
face between the network and the users, a very solid end-to-end flow control strategy
is established.
The TYMNET flow
control strategy is somewhat conservative in that it may allo-
cate store-and-forward
buffers to one virtual circuit while another virtual circuit has
more use for them
(possibly
causing a decrease in line utilization). This conservatism.
however, provides
a number of useful features:
l. Networkwide flow control is established automatically
by the flow control
within each virtual circuit.
2. Under light traffic conditions, packet
flow within each virtual circuit adjusts
automatically
to the maximum rate that the source and sink can support. If only
a few virtual circuits exist, a relatively large number ofbuffers can be assigned
to each circuit, allowing the retum of more frequent clear-to*transmit signals.
3. If a sink stops accepting packets
for some reason, this condition propagates
back
to the source by way of cessation of clear-to-transmit signals. Thus the source
10.2 FACKET SWITCHING 469
stops sending when all allocated buffers are fuIl. This principle of operation
is
sometrmes referred to as back
pressure.
As long as a node never overcommits its buffers, store-and-forward lockups
cannot occur. Ifseveral sinks stop accepting data, the $tore-ard-forward buffers
assigned to the particular virtual circuits get filled and become unavailable.
Other virtual circuits, however, can maintain transmission through their own
assigned buffers.
The mechanism is fail-safe in the $ense that positive indications
(clear-to-transmit
signals) are needed before
packets are forwarded to
neighboring nodes. When network links are overloaded, flow control bandwidth
requirements are minimal. If flow control signals stop altogether,
packet
transmission stops.
10.2.5 X.25
X.25 is an ITU standard protocol established
in 1976 for accessing a public packet-
switching network. The data link layer of X.25 is link acce$s procedure balanced
(LAPB)
using high level data link control
(HDLC). HDLC is a bit-oriented protocol
based on the
previously defined Synchronous Data Link Protocol
(SDLC)
established
by IBM for Synchronous Network Architecture
(SNA) networks. HDLC has beconte
the data link
protocol of choice for numerous data communications applications.
The
balanced aspect of LAPB refers to a
peer-oriented operation between the two ends of
the link. In an unbalanced mode of operation a primary node controls the link for one
or more secondary
nodes.
The basic format of an HDLC
packet is shown in Figure 10.5. As indicated, packets
are delimited by a stafling and an ending fiag
(01
I I I I 10). The data stream between
the flags is precluded from including an inadvertent flag by azero bit insertion algo-
rithm. Whenever a string of five ls occur in the data stream a 0 is inserted by the source
node. The receiving node removes a 0 occurring after five ls. If a 1 is
present after
five ls, it must be the sixth I of a flag and is therefore not removed.
The address field is typically 8 bits long but can be extended in increments of 8 bits.
The information field can be any number of bits long, but specific applications typi-
cally define a maximum length. The information field is nonexistent in a conhol
packet. The frame check field contains a l6-bit CRC but can be 32 bits long in some
Fleg AddruE$ Control lnformation I
Frums ch6ck Flag
0 1 1 1 1 1 1 0
01 11 11
0 N(S) r t f N(R)
l-forffat mntrol fi6ld
Figure 10.5 HDLC frame format,
4.
5.
47O DATAANDASYNCHHoNoUSTFANSFER
MoDENETWoRKS
applications. The control field is 8 or 16 bits long. Informarion
in the control field is
formatted in various ways
depending on whether the packet
is an information packet
or a control packet.
control formats are designated by a leading l. Figure 10.5 shows
the format for a information packet (I-format)
as designated by the leading 0 in the
control field. The P/F field is a single bit used for polling.
The N(S) and N(R) fields contain sequence numbers used for both error conffol and
tlow control. The length of N(s) and N(R) is 3 or 7 bits depending on the size of con-
hol field in use.* For illusftative purposes,
the following discussion assumes the use
of 3-bit fields. Every time a source sends a packet
containing information, it incre-
ments N(S) modulo 8. When a destination receives a packet
with a valid check sum
and the next expected sequence number, it returns that sequence number to the source
in its outgoing N(R) field. Thus, a source node knows that the destination has correctly
received every hansmitted packet
up to and including the packet
with the sequence
number N(R) in its incoming packets.
If one or more packets
have been received in
error, the value of N(R) returned to a destination does not change until the source times
out and reffansmits the packets
beginning with the last received N(R) value plus I.
[All
packets
after a lost or comrpted packet
are retransmitted because the destination
ignores packets
with an out-of-sequence
value in the incoming N(S)].
with a 3-bit length for N(s) a source can have a maximum of seven packets
out-
standing at any particular
time. (A particular
sequence number cannot be reused until
an acknowledgment for the previous
one has been received.) Beyond the maximum
number of outstanding packets
determined by the length of the sequence fields, par-
ticular applications may be configured for a smaller number of outstanding packets
for
flow control and reducing the amount of memory needed by a source to hold copies
of unacknowledged packets.
Whatever the value of the maximum number of outstand-
ing packets
is, a destination is expected to be able to receive that number as a burst.
The destination
does not have to immediately acknowledge
a valid received packet
if
it is congested
and wants to free up rerrources before receiving more packets.
If a sys-
tem is configured with a maximum number of outrrtanding packets
that often exceeds
the capacity of a receiver, the system works but the source spends a lot of time retrans-
mitting packets
that were discarded by the destination because it had
no resources
(memory)
to store them,
x.25 permits
a user on an x.25 network to communicate with a number of remote
locations simultaneously.
connections occur on logical channels of two types:
switched virtual circuits (svcs)
or permanent
virtual circuits (pvcs).
svcs require
a connection establishment process
before data can be transferred. A PVC is similar
to a leased line in that the connection is permanently
established by network admini-
stration. Therefbre, data can be sent without connection setup.
To establish a connection
on an svc, the originator sends an x.25 call request
packet,
which includes the address of the remote destination. The destination decides
whether or not to accept the call (the
call request packet
includes the originator's ad-
dress and possibly
other information
that the destination
can use to decide to accept
-Use
ofHDLC on high-speed fiber links has led to the use ofeven larger sequence numbers because a large
number of frames can be in fiansit on the link.
10.2 PACKET SWITCHING 471
the call). A call is accepted by returning a call accepted
packet or rejected by returning
a clear request packet.
10.2.6 Frame Relay
As local area networks and file $ervers evolved to provide data
processing functions
within a corporation, some meims of interconnecting LANs at separate sites became
necessary. Figure 10.6 illustrates three ways of accomplishing LAN interconnection
to implement a colporate wide area network
(WAN).
The fust method utilizes a public
(X.25) packet-switching network to relay messages between the LANs. This approach
is a useful one for limited amounts of inter-LAN taffic. It also facilitates interconnect-
ing two or more dissimilar LANs
(such as Ethernets, token rings, or token buses)-
Public
packet network
(a)
LAI{ I
LA}-I2
Frame relay network
(c)
Figure 10.6 LAN-Io-LAN interconnectron methods:
(a)
through x,25 packet-switching
netowrk;
(b)
through leased line;
(c) tfuough frame relay network.
Lsasd
digital lire
(b)
472 DATAANDASYNCHRoNoUSTHANSFERMoDENETWoRKS
because the X.25 interface of each LAN acts as a data link layer protocol
converter.
The main drawback is the cost and slow response for high-bandwidth
ffansfers.
The second approach of interconnecting
the two LANs is to use a leased line. ob-
viously this approach is more cost effective if large amounts of data are transferred be-
tween the two locations. The LANAVAN interface equipment used to connect to the
leased line will vary depending on the nature of the individual LANs and how the
LANs and WAN are administered.
Functional possibilities
for the LAN-WAN inter-
face are bridges, routers, or switches. The speed chosen for the leased line inevitably
requires compromising performance
objectives with cost.
The third method of interconnecting
the LANs is to use a frame relay service of a
common carrier. Frame relay services were developed specifically
to address LAN in_
terconnection and are available from all major common carriers. Frame relay is a
packet-rtwitching
protocol
but is faster than traditional X.25 networks because it does
not provide
error control. Error control at the data link layer (e.g.,
HDLC of x.25) re-
quires
receipt of an entire packet
before it can be forwarded from one link to another.
Frame relay also supports shortened
address processing
with a data link connection
identifier (DLCI)
field in rhe header which identifies a
pvc.
The frame relay prorocol
is defined by the ANSI Tl.618 standard and ITU Recommendation
e.922.
The PVC feature of frame relay permits
the functional
equivalent of a leased-line
connection but much more cost effectively for high-rate, bursty traffic. Furthermore,
a single frame relay access device (FRAD)
can achieve connectivity to multiple re-
mote LANs by using a different DLCI for each remote LAN.
Frame relay nodes can process
the simpler address format and switch an incoming
message directly to an outgoing line as it is being received. The elimination of error
control- at the data link layer of a network protocol
reflects the fact that the underlying
transmission
facility, fiber, is essentially error free. In this environment it is more ef-
ficient to move effor control
to a higher level (e.g.,
an application layer). when infre-
quent
errors occur, the application
can invoke error control as appropriate to the
application.
Example 10.1. Determine the amount of transmission
capacity required to
refransmitframesof
1000bitsof
datainanetworkwith
l0tandemlinkseachof
which
has bit error probability
IO"t. (a)
Assume link-by-link enor conrrol. (b)
Assume
end-to-end
enor control. (c)
Repeat the calculations for a bit error probability
of l0-5.
Solution,
(a)
(b)
with link-by-link
error control the probability
of a bit error in a frame is 1000 x
10-8
=
l0*s. The expected number of bits of transmission capacity required to
rerransmir is l0-5 x 1000
=
0.01 bit/link.
with end-to-end error
conffol the probability
of a comrpted frame is
10 x tO-s
=
l0+. The expected retransmission
capacity required is lO-a x
1000
=
0.1 biilink.
CRCs are included in frames of frame relay data, but a frame relay node does not request a retransmission
if an error is detected.
10.2 PACKET SWITCHING
473
(c)
With a bit enor rare of 10-5, the respective calculations
are 10 bits/link for
link-by-link conffol and 100 bitsflink
for end-to-end control.
Example 10. I illustrates that with very Iow bit error
probabilities both forms of error
conffol
require insignificant amounts of transmission capacity for frame retransmis-
sions. Thus, from a transmission
capacity point of view there is no reason to use link-
by-link effor control. On transmission
links with higher rates, a$ in Tl lines with
marginal error
perfbrmance, link-bylink
error control is obviously desirable,
pruticu-
larly in a packet network where
packets might be larger than 1000 bits.
10.2.7 TCP/IP
The original networking
protocol of the ARPANET was designated Network Control
Protocol
(NCP).
While this protocol addressed the needs of a single network,
it could
not be used as a means of communicating from one network to another.
The desire to
develop a protocol that would allow communication from a node in one network to a
node in another network led to the eventual development of Transmission Control
Protocolflntemet Protocol
(TCP/IP)
[8].
TCP/IP replaced NCP in ARPANET hosts on
January I, 1983.
TCP/IP is a
protocol that can be used within a network, but more importantly, it
can be used as a prOtocol to Communicate hetween two netwOrks'" Thus, the Internet
is composed of many networks that communicate with each other using TCP/IP. Fore-
most among these many networks are LANs. The Internet is
yet another way for in-
terconnecting LANs. ftr conftast to fhe creafion of a corporate WAN using frame relay,
LAN interconnection via ttre Internet is primarily intended to cross co{porate boundaries.
As defined by the Internet Engineering Task Force
(IETF), TCP/F uses variable-
length packets that are
processed with basically one class of service:
best-effort rout-
ing. A user datagram
protocol
(UDP)
replaces
TCP when real time services such as
voice or video are carried. UDP has no reffansmissions. Before Voice on the Infernet
(VoIP)
can be realized for something
approaching universal
service, the routers must
be extencled to support a new class of service: one with controlled delay, delay vari-
ation, and delivery guarantees. A first attempt to address the
quatity-of-service issue
was a propo$ed modification
of TCPIP with a Reservation
Protocol
(RSVP). This
ap-proac:h was determined unsatisfactory because it required changing all nodes in a
network simultaneously
[9].
A more
practical approach is a
"differentiated
service"
(DiffServ)
enhancement to TCP/IP.
When considering
the viability and expected
quality of service of packetized
speech,
it is worthwhile to consider the difTerence between IP telephony
and Internet
telephony. IP telephony refers to carrying
packetized speech
in an IP network. Ifthis
network is wholly contained within one organization, commonly referred to as an in-
tranet, it is more likely that a relatively high level quality of service can be provided.
Packetized speech carried over the Internet that traverse$
multiple, autonomou$ net-
*Th"
m,r*t *pp*unt aspect of TCP/IP to a uscr is the Universal Resoutce Locator
(URL), with the following
format: filetype://www.address.tld/resource. The address.tld
(.top level.domain) identifies a location on
the web while everything after thc back slash identities a resource
(e.g.,
a frle) inside the location.
474 DATAANDASyNcHRoNousTRAN$FEHMoDENETWoHKS
works is much more difficult to rnanage, particularly
when economic considerations
for allocation ofcosts and revenues
are considered. For a description ofmany ofthe
technical considerations
forpacketized
voice see reference
[10].
10.3 ASYNCHRONOUS
THANSFER
MODE NETWORKS
In contra$t to TCP/IP, the Asynchronous
Transfer Mode (ATM)
network architecture
incorporated
features for supporting real-time traffic such as voice and video in the in-
itial implementation.
The principal
aspects of ATM directed to real-time
$uppofi are
short, fixed-sized packets (cells),
short headers,
and no link-byjink
error control.*
ATM is a standardized architecture
of packet-oriented
transmission and switching
originally proposed
for a Broadband Integrared
services Digital Network (BISDN).
ATM has since been expanded
in scope to support a wide variety of seruice types;
wideband,
narrowband, bursty, non-rear time, and real time. The synchronous
TDM
(circuit-switched)
network that evolved primarily
for voice services supports the same
services with external adaptations,
but the adaptations come from u u*i"ty of suppli-
ers necessitating
the need for multiple, nonintegrated,
nonsiandardized
equipment and
support. ATM standardizes the wide range of services by defining quality-of-service
re_
quirements
fbr various naffic qrpes.
The quality-of-service
pu*-rt
ru specifically
in_
tended far voice services
are maximum delay, delay variation, and cell losi probability.
10.3.1 ATM Cel l s
The ATM transmission
format consists of fixed-size cells of 53 bytes each. As shown
in Figure 10.7, there are five overhead bytes within each cell, wtrich teads to 4g bwes
of payload per cell' The use of fixed-sized
cells for all applications
facilitates hard-
ware-only implementations
of switching functions (queuing
and transfer;.
Cell Loss Prlority
The cell loss priority (CLP)
bit in the cell header is used to identify rwo basic classes
of service with respect
to discarding of cells within a network. when a cell with a cl-p
value of I arrives at a heavily congested network
element, that cell can be discarded to
relieve congestion for higher priority
traffic (e.g.,
cells with cl-p
=
0). cells with a cl-p
value of 0 are never discarded
unless atl cells with lower priority
are aheady discarded.
10.3.2 ATM Service
Gategoriee
ATM services are categorized (by
the ATM forum
Il
l, 121) into the following caregories
related
to the statistical nature
of the data rate requirements
of the respective
sources and
the quality
of selvice (Qos)
that the network can provide
for those $ervices:
-TCP/IP
also defets eror control to the application but has variable-sized packets
and a relatively
complicated header format.
10,3 ASYNCHRONOUS TRANSFER MODE NETWOHKS 476
Bits
8 7 6 5 4 3 2 1
Floru Control VidualPath ldentifier
Virtual
Path tdentifier
rnel ldentifier Viduaf Char
Payload
Type CLF
Header Enor Check
Payload
Payload
3
4
Byr"*
5
I
5S
Flgure I0.7 ATM cell format; CLP, cell loss prioriry'
Constant-bit-rate
(CBR)
services
Variable-bit-rate
(VBR) services
Available bit rate
(ABR)
services
Unspecified
bit rate
(UBR) services
Co n stant-Bit-Bate Servl cee
An application requesting a CBR service
is guaranteed a specified
data rate assuming
the service request is accepted
(i.e.,
ttre network can
preallocate sufficient
resources
for the request). Thus, a CBR service is similar to a synchronous TDM service. CBR
services are specifically designed fo emulate circuit switching. A CBR application
cannot exceed the specified rate
(i.e.,
the peak cell rate) but need not necessarily send
data at the specified rate. In this sense, CBR services are different from synchronous
TDM services-if a source has
periods of time in which there are no data available,
the channel capacity may be released for other applications'
Va riable-Eit-Rate Se rulcee
VBR services are separated into two distinct categories: real time
(rt-VBR) and non-
real time
(fft-VBR). The rt-VBR services are for bursty applications
with tightly con-
strained delay and delay variation requirements sqch as voice or video. The main
QoS
476 DATAANDASyNcHRoNousTRANSFEFMoDENETwoRKS
parameter
of an rt-vBR service is the maximum cell transfer delay (maxcTD).
The
nIt-VBR
services are for
bursty data transmission
applications with no particular
de-
Iay requirements.
whenever
an rt-vBR
or nrt-vBR service request is accepted, the foltowing three
parameters
are agreed upon for source transmission statistics:
Peak cell rate
(pCR)
Sustainable cell rate
(SCR)
Maximum burst size (MBS)
As long as the source adheres to the above statistical parameters,
the network is com-
mitted to carrying the traffic.
Avallable Blt Rate Serulces
ABR services are intended for non-real-time
applications that operate between a mini-
mum and maximum data rate specified by a minimum cell rate (MCR)
parameter
and
a PCR parameter'
An essential aspect of ABR services is the need for feedback signals
within the network to control the flow of data from the ABR sources. The network dv-
namically
allocates additional resources to ABR services (beyond
those for MCn,up_
port)
when the additional resources are available
and so informs the ABR sources with
the feedback mechanism.
Unspeclfled
Blt Rate Seruices
UBR services
are inherently intended for non-real-time
applications wherein no ripe-
cific quality
of service is desired or implied. Neither delay parameters
nor cell loss ra-
tios (cLRs)
are specified. If a network
becomes overly ctngested, cells from uBR
applications
are discarded by the network (as
opposed to confrolling the flow of the
UBR source)' when a UBR application negotiates
a connection, a
pcR
parameter
value may be provided
by the network, which provides
the application
with an indi-
cation of the throughput that can be expected. The following inoicates
which
eos
pa_
rameters are relevant
to the ATM classes of service.
QoS Parameter
Gell loss ratio (CLR)
Cell transit delay (CTD)
Cell delay variation (CDV)
Peak cell rate (PCFI)
Sustained cell rate
(SCR)
Burst tolerance (BT)
Flow control
CBH
nrt-VBR N-VBH
X
X
X
X
X
X
X
X
X
X
X
X
X
X
10.3 ASYNGHHONOUSTRANSFERMODENETWORKS
477
10.3.3 ATM Gonnectlong
ATM networks are connection oriented,
which means that a source must make a re-
quest for service
before transmitting cells. Network control determines aroute through
the network and allocates resources
(in a statistical or
probabilistic sense) in response
to each service reque$t. In cOntrast, LANs typically operate in a connectionless
mode
wherein a $ource sends unconskained data to a destination whenever the need arises'
ATM supports two types of connections: virtual channel connections
(VCCs)
and
virtual
path connections
(VPCs). A VCC is a virtual circuit between two user$ repre-
senting a single stream of cells. When there are multiple VCCs between two common
endpoints,
a VPC can be created as a bundle of VCCs. The use of VPCs simplifies
net-
work management and the operations of intermediate nodes of an ATM network be-
cause a VFC is
processed as a single entity as opposed to individual VCCs. Thus, a
VPC is analogous
to a trunk
group in a circuit-switched network'
10.3.4 ATM Switching
The basic requirement of an ATM switching
fabric is to transfer cells arriving on one
physical link to another link. Because incoming headers must be processed before an
incoming cell can be fiansferred
through a switch and because an outgoing link is
likely to be busy, ATM sffuctures always
provide
some amount of delay in the switch-
ing
process. Thus, ATM switches
provide both space and time switching functions in
a similar manner as TDM circuit switches described in Chapter 5.
A significant difference, and complication, of an ATM switch servicing streaffr
traffic such as voice or video is that, in contrast to TDM channels,
ATM cells can ar-
rive at irregular
intervals. This phenomenon greatly increases the likelihood
that a
par-
ticular cell encounters blocking on a shared interstage link of a switching network. The
basic solution to this problem is to add cell buffers a$ a queue for blocked cells until
the necessary path is free. Simple first-in,
first-out
(FIFO) queuing operations
are
readily implemented in hardware but better network efficiency occurs if the FIFO dis-
cipline is modified to allow a cell in the
queue with an open
path to bypass ablocked
ennl
fl3,
141. Thus, more complicated
hardware implementations are desirable.
The principal aspects of an ATM switch that are different
from a
TDM circuit
switch are:
t . There is no
preassigned incoming time slot so header information needs to be
extracted
and processed to know where to transfer incoming
cells'
There is no preassigned outgoing time slot for outgoing
cells so the delay
involved in transferring cells is a variable
queuing delay associated
with waiting
until the outgoing link is idle.
The header information in an outgoing
cell is
generally different than what is
received so the switch must insert appropriate header information into the traffic
stream.
2.
478 DATAANDAsYNcHHoNoUSTHANSFERMoDENETwoRKs
ATM ltlemory Switch
As is the case of TDM circuit switches, ATM switches use memory switching to the
maximum
extent po$sible.
As long as the memory speeds support the traffic, the most
economical
design is one in which all traffic is written into ana read from a cornmon
memory. In addition to the switching fabric, a high-speed
ATM switch must imple_
ment other cell processing
functions in hardware. These functions
are header e*t ac-
tion, header error checking, table lookup for route information
using received virtual
path
indicator (VPD
or vcc, recording of traffic usage statisti"r, ""ll queuing/rejec_
tion, insertion of outgoing
vpl/vcc, and outgoing checksum generation.
Example 10.?. Determine
the memory
speed required
for an ATM switch fabric
using a shared-memory
archite*ure
in support of 12 srs-3 (oc-3)
bidirecrional
ports'
Also determine the number of voice connections
that can-be carried by this
switch. Assume that an active voice signal requires 12 kbps and that the activity factor
is 40Vo.
solution. Exceptforthe
gbytes
ofpathoverhead,
alr byres ofan srs-3 payloadcan
be used for ATM cells.
Thus, each srs-3 input rink provides
260 x
g/s3 -
44
cells/frame. (A partial
frame is not carried so cell boundaries can be aligned with the
start of the srs-3 payload.)
If a S3-byte-wide memory is used for storing the cells, rhe
total access rate for l2 bidirectional porrs
is 12 x z x,44 xg000 =
g.++t
rralt.
since each ATM cell provides
4g bytes ofpayload,
the number
ofvoice connec_
tions that can be supported by one sTS-3 is a4 x 4g x
g
x
g0fi)/12,000/0.4
=
2g,160
connections.
The total capacity of the switch is 337,g20 voice calls.
It is interesting
to compare
the results of Example 10.2 with a circuit switch design
for 64-kbps voice channers.
using a standard soNET multiprexing
format, a single
srs-3 canies three DS3 signals (2016
voice channels).
sirnitarty,*a srs-3 (e.g.,
an
srM-1)
carries a single B$ signal with 1920 voice channels. In either case, the capac_
ity of each ATM link is over 10 times the capacity of the conventional
circuit-switched
architecture
using the same data rate. The advanfege
comes from a combination of
compressing
the speech and taking advantage
of voice inactivity.
A memory switch for 12 x 2016 =
z4,rgzchannels
requires an access rate of 3g7
MHz, which is much greater
than the
g.44g-MHz
access rate of the ATM switch, even
though the total capacity is much less. The reason for the high access rate ofthe circuit
switch is that only 8 bits are being switched at a time. rr muttipte bytes were accumu-
lated and switched
through a wider memory, as in the eru case, ttre access rate of
the circuit switch would be reduced accordingly.
circuit switches have not been im_
plemented
with wider ffansfers
to
prevent
the insertion
of extra circuit delay. Voice
interfaces to ATM networks
necessarily require
an echo canceler to accommodate
the
delay.
Example 10'? is a nonblocking
switch as far as internal
operations of the switch are
concerned' However, if a large burst of kaffic arrives at oni particular
output port,
it
may be that some cells are discarded because the queue
for that particular
port
is full.
10.3 ASYNCHRONOUSTRANSFERMODENETWORKS
4Ig
Notice that if traffic is being queued for some ports, other outgoing ports are operating
below their capacity.
The amount of variation in the traffic flow on outgoing ATM links is a direct func-
tion of the connection admission control
(CAC) process involved in setting up con-
nections. If all connections are VPCs and these VPCs have relatively constant arrival
rates
(because the aggregate data rate of multiple, independent variable-rate
channels
is fairly constant), the transmission resources of the network will be efficiently utilized
with modest queue sizes. If the traffic arrivals are highly variable, larger queues (and
consequently
larger delays) are required to achieve high transmission link utilization.
When there is a mixture of traffic types, the time-critical
types are
given priority so
cell loss and long delays are confined to non-real-time-critical cells.
Spac*Memory Switch
The implementation of large ATM switches requires multiple stages of space and
times switching
(e.g.,
buffering) in the same manner that large time division switches
described
in Chapter 5 are implemented. One particularly useful architecture involves
a front-end TDM bus as a $pace switch followed by separate output
queues' aS shown
in Figure 10.8. The inputbus is synchronized to series-to-Parallel
(S/P)
circuits ofthe
input
ports so that a time slot is assigned for every received cell. Thus, the bus must
operate at rhe total speed of all incoming signals. Each header is decoded via a table
look-up to select the appropriate output so the output memories store only cells for re-
spective output
ports.
The main attraction
of the multiple-output-memory
architecture shown in Figure
10.8 is its ability to scale to a large range of switch sizes. Furthermore, broadcasting
and multicasting are easily supported by having the header decode logic enable simul-
taneou$ writes into multiple buffer$. Unfortunately, this architecture does not do much
to solve memory speed
problems. Output accesses
only need to support the output
port
transmission
rate, but input accesses must suppofr successive time slots on the bus.
Memory speed restrictions can be somewhat
alleviated with high-speed caches acting
as speed buffers between the bus and larger, slower output buffels, as shown in Figure
10.9. The caches do not add much cell delay but do repre$ent a source ofcell loss when
one fills up because of a burst to the respective
output line'
Figure 10.t Input bus/output buffer ATM swirch.
480 DATA AND ASYNCHBONOUS
THANSFEH MODE NETWORKS
Figure 10.9 Input bus/output buffer ATM switch with caches.
Memory-Space
Switch
Another version of a two-stage
ATM switch is shown in Figure 10.10 wherein all in-
coming cells are available to all output ports
via separate paths
for each input. Each
input path in each output port module is terminated in a separate cell buffer. Associ-
ated with each buffer is a table look-up function
that selects only those cells destined
to the respective output port$.
Output data are selected from the $eparate buffers as a
cornmon queue,
thereby pedorming
a space switch function.
The main advantage of the configuration
shown if Figure I 0. I 0 is that all data paths
and memories
operate at the external link speed. The basic disadvantage is the need
for AP buffers (and
AP header processors),
where N is the number orplrts.
cell loss
occurs when an incoming
cell encounters
a full buffer. The probability
of this occur-
ring is conkolled by how much traffic for a
particula-r
output is accepted from a
par-
ticular input. The use of multiple, individual queues
leads to smailer group
sizes,
which leads to more total
memory for a given
cell loss probability.
However, because
the memories
operate at relatively
low speeds, the cost impact is minimizea.
Notice
that delay probabilities
are determined by the total traffic ur.*pt-d for an ourput port
in the same miutner as previous
con_figurations.*
Me m ory-$ pace-l,ilem
o
ry
Switc h
ATM switch architectures
with single stages of space and memory switching are im-
practical
in terms of speed and/or hardware requirements
when very large $ysrem$ are
desired. As is the case with circuit-switching
architecture$, more practic-aVeconomical
structures require multiple memory stages or multiple space stages-or
both. one
such architecture
is a counterpart
to a TST switch, the memory-space-memory
(MSM)
implementation
shown in Figure 10. I L The inpur memorystage
of an MSM
buffers complete cells of information that are queued
for transfei
through the space
stage and then buffered again for transfer from the output memory.
successive cells
of a particular
connection are not necessarily transfened through thi space smge at any
palticular predetermined
time. As each cell is received into an inpui buffer, a queue
entry is made for rhe appropriate transferthrough
the space stage. Support ormuitiple
*This
assumes that the separate buffers associated with each output link are managed as a single FIFO queue
for output purposes.
10.3 ASYNCHFONOUSTBANSFERMODENETWORKS
481
Flgure 10.10 Multiple-memory ATM swirch.
classes of service reqUire multiple input queues (usually sharing a coilunon memory
element).
Before a cell in an input buffer can be transferred through the space stage, the de-
sired ouput stage must be available for input. If the desired output memory is busy,
the cell carnot be transferred and head-of-line blocking occurs.
If another cell in the
input
queue can be selected when one or more of the head-of-line cells are blocked,
the impact of the blocking is minimized.
Such operations are ea$y to implement in
computer-controlled store-and-forward swirches but must be implemented in hard-
ware in a large, high-speed ATM switch. An altemative approach to determining if
blocking exists before a hansfer is made is to blindly fransfer cells through the switch
and have blocked cells reenter an input
polt
though
a recirculation
path, as shown
in
Figure 10.12. Determining the delay/cell loss performance of such a system is very
complicated
and dependent
on the
probability of blocking and the number of cells that
can enter the recirculation buffer(s).
Figure 10.11 Memory-space-nremory
ATM switch.
482 DATAANDASYNCHRoNoUSTRANSFEFM0DENETWoRKS
Figure l0.l? Memory-space_memory
switch with recirculation.
For best performance,
the input cells from all input lines of an MSM switch should
be distributed across all input buffers using a front-end shuffle distribution
switch
[15]'
This process
decorrelates the enhies in an input queue
from the cell arrivals of
any one particular
input line. Recall that a similar operation is advantageous
in TST
switches to reduce blocking probabilities
for individual
channels of a busy trunk
group.
.
The design of a space stage for an MSM switch matrix can follow any one of a va-
riety of approaches, including a single-stage square matrix or a multipie-stage
clos
network, as described in chapter 5. with either of these designs the probability
ancl
treatment of output blocking can be minimized by running the space stage faster than
the cell rate of the extemal ports (i.e.,
with the MSM equivalent
of time expansion).
The existence of expanded opporhrnities
for space stage transfers reduces the prob-
ability of contention for an output buffer and minimizes
the delay when contention
oc-
curs. A very significant consideration
in the space stage design is the complexity of
hardware control logic, particularry
when multiple service
Irioriti",
and multiple
paths
through the switch are accomrnodated.
Banyan Network
Swltch
one approach to implementing
modular space stage structures is a banyan network
U6l.
As indicated
in Figure 10.13, abanyan nerwork is implemenred
by inrerconnecr-
ing 2 x 2 switching elements in multiple, recursive
stages. Notice that the
g
x
g
skuc-
ture is implemented
as two 4 x 4 structures and an additional front-end srage to select
either ofthe two 4 x 4 structures
from each input. Repeated replications and first-stage
additions allow implementation
of arbitrary sized networks.
A banyan network provides
full accessibility
in that any input can be connected to
any output. Notice that there is a unique path
between any input and any output. How-
10.3 ASYNCHRONOUSTRANSFEFMODENETWORKS
483
(a)
(b)
Figure 10.13 Banyan swirching network:
(a)
2 x 2 switching element connection statesl (b)
8 x I network.
ever, the networks are not nonblocking. If the first input is connected
to the first out-
put, for example, the second input can only be connected to one ofthe bottom four out-
puts. One method of overcoming the blocking aspects of the banyan network involves
cascading tWo such sections to
produce what are
generally referred to as Benes net-
works
[17].
A Benes network is strictly nonblocking if rearangement of connections
is utilized. Another augmentation of a banyan network is to concatenate a Batcher
sorting network
[18]
with a banyan network. The front-end sorting
process orders the
cells so they can all be transferred through the switch without blocking. Another modi-
fication to a banyan network to address blocking
problems is to add
parallel networks
referred to as switch
planes to provide more paths to an output. This latter approach
is identical to adding more center stages to a multiple-stage switch described in Chap-
ter 5. Unfortunately, this approach compromises the simplicity
of unique path routing.
The basic attraction of a banyan network is the modular simplicity of its control.
When a cell it to be transferred through the network, the table look-up process pro-
duces a sequence ofbits for internal routing through the network. The first bit defines
the selection of the first-stage element.
The second bit deflnes the selection of the sec-
ond stage and so on. The routing information is appended to the front of the cell and
the composite message is transferred through the switch one stage at a time, with lead-
ing bits deleted as they are used. Thus, a banyan network utilizes direct
progressive
control reminiscent of an electromechanical step-by-step
switch, described in Chapter
5. Unfortunately, modifications to the basic banyan architecture
to provide multiple
paths compromises the simplicity and distributed aspect of the hardware control.
A wide variety of other ATM switch architecfures have been
proposed and devel-
oped
[19-23].
-Et*
Cross
r-t
-l-
Slraight
484 DATAAND ASYNcHRoNoUS THANSFER MoDE NETWoHKS
10.3.5 ATMApplicarions
Although the basic ATM architecture is primarily
directed to servicing data commu-
nications, the ATM architecture includes provisions
for other services such as voice,
video, and multimedia.
Support for the various services is included in various versions
of ATM adaptation layers (AALs).
service adaptation occurs at the periphery
of a net-
work in edges switches. Internal nodes of an ATM network
(core
switches) are uncon-
cerned with the nature of the traffic other than supporting
the
eos
objectives of the
general
classes of service. The role of AALs are shown in the functional layers of Fig-
ure 10.14 of the (ATM)
Multi-service
switching Forum (MSF)
[?4].
The generat
classes of service and the adaptation layers to support these services are identified as
follows:
AALI: Transport of cBR rraff,rc (program
audio, video, and emulation of
TDM-based
circuits (i.e.,
DSl, El).
AAL2: Protocol standard for low-bit-rate
and time-dependent
variable-bit-rate
(VBR-rt)
connection-oriented
traffic (compressed
voice ancl packetized
video).
AAL3/4: Protocol standard for supporting
both connectionless
and
connection-oriented
vBR traffic
[primary
application, switched Multimesabit
Data Service (SMDS)1.
AALS: Transport
of vBR traffic ancl signaling messages (packet
data, Ip, frame
relay, LAN emulation).
Circuit Emulatlon
Serulces
A circuit emulation service (cES)
t25l
is used to transport TDM-based
channels such
as DSO, DSl, or El signals. cESs are processed
according to an AALI protocol.
A
Switching
plane
Extemel interfaces
Figure 10.14 ATM functional implementation
layers.
Applications
Gontrol
plf,ne
Adaptation
plsne
10.s ASyNcHFoNoUSTHANSFERMoDENETWoBKS
485
critical aspect of a CES service involves
recovery of the source clock at the egress
point of the ATM network. Figure 10.15 depicts three different clock recovery sce-
narios dependent on the nature of the network synchronization. The most desirable
situation is shown in Figure 10.154, in which the entire network, including the data
source, is synchronized to a common
primary reference clock
(PRC). In this case the
output data clock is derived from the PRC and therefore has the same average value
as the source. As long as the ATM network meets cell delay variation
(CDV)
criteria,
the cell buffer in the egress AALI equipment absorbs cell arrival variations and no
data are lost. If cell arrival variations exceed CDV requirements, data slips may occur.
As is the case in TDM networks the slips should be controlled
to consist of integral
numbers of frames of the underlying data
(e.g.,
193 bits for DSI or 256 bits for El) to
preclude loss of framing in the underlying circuit.
Figure l0.l5b shows a network configuration wherein the ATM network is syn-
chronized to a common PRC but the data source is timed from some other reference.
For this situation, ATM networks utilize synchronous residual time stamps
(SRTSs).
SRTSs are periodically inserted into the overhead ofa cell data stream at an ingress
AALI. The egress AALI can determine the average data rate of the source by dividing
the amount of source data received between successive SRTSs by the time interval de-
fined by the SRTSs.
Notice that the use of SRTSs depends on the ingress and egress
equipment being synchronized to a common PRC.
Figure 10.15c depicts the most undesirable
case in which the ingress and egress
nodes are controlled by separate references
(PRCI
and PRCZ). In this case, adaptive
clocking is utilized to recover the source data clock. Adaptive clocking involves a
phase-locked loop
(PLL) with the fill level of the cell buffer
providing a
phase meas-
urement. A very low bandwidth
filter is used to determine the long-term average fill
level, which in tum establishes the recovered output clock of the VCO. Jitter in the
derived clock is easily controlled hy using a large-cell-buffer and narrow-bandwidth
PLL. However, low-frequency clock variations
(wander) caused by irregular cell ar-
rival times are unavoidably ransferred to the derived clock. Depending
on the appli
cation, the wander may or may not cause downstream
problems. If a DSI signal, for
example, is immediately
demultiplexed into analog signals, no impairments to voice
channels will arise but high-speed voiceband
modems may be adversely
affected (be-
cause of impairments to the carrier reference phase). If the DS I signal is instead passed
to a switching
machine, the wandermay contribute to slips in the elastic store interface
of the switch. Again, slips do not impair voice signals but significantly
comrpt voice-
band data.
LAN Emulatlon
To provide compatibility with LAN
protocols and established applications, the ATM
Forum created a
protocol,
called LAN Emulation (LANE), which defines how Eth-
ernet
(IEEE
802.3) and token ring LAN
(IEEE
802.5) data can be carried over an ATM
network. Figure 10.16 show how an ATM network can
provide LAN-to-LAN connec-
tivity as well as connectivity directly to workstations
(ATM
to the desktop).
The con-
nectionless nature of contemporary LANs is simulated in a manner that first
DATA AND ASYNCHRONOUS
TRANSFER MODE NETWOFKS
(a)
(b)
lPRcTl
.;
(c)
Figure 10.15 Synchronization of circuit emulation services: (a)
fully synchronized network;
(D)
unsynchronized
source (sRTs
derived clock); (c)
unsynchronized
ATM gateways (adaptive
clocking).
10.3 ASYNCHRONOUSTRANSFERMODENETWOBKS
487
Figure 10.16 LAN interconnection using ATM LAN emulation.
determines if a connection
(SVC
or PVC) to an addressed station already exists. If so,
an ATM message is sent to the station. Otherwise, an SVC is first established
and then
the message is sent. A LAN Emulation client
(LEC)
is established in every LAN at-
tachment to the ATM network. Several types of LAN Emulation servers are used to
configure the LECs and provide contemporary LAN functions $uch as multicast and
broadcast services.
Low-Bit-Rdte Voice
Although voice can be carried as 64-kbps PCM channels in a CES mode, great savings
in bandwidth are achieved by transmitting compressed voice. Transport of com-
pressed voice is
particularly appropriate for ffansmission to and from digital cellular
base stations where voice is already compressed or about to be compressed. Similarly,
transport of a
group of compressed voice channels between PBXs of a
ptivate
network
is advantageous. As shown in Figure 10.17, individual voice channels can be serviced
as VBR-rt
(AAL2)
switched virtual circuits by packing multiple frames of low-bit-
rate $peech into ATM cells. The
particular example shown in Figure 10.17 assumes
G.729 Conjugate Structured Algebraic Code Excited LPC
(CS-ACELP) voice com-
pression
with 80 bits per lO-msec processing ftame. Notice that cell
packing puts 4.8
ACELP frames into an ATM cell, which increases the voice latency by as much as 40
msec in each direction of transmission.
The voice latency can be reduced by forming a trunk group and
packing multiple
compressed
voice packets into individual 48-byte cell
payloads. A single cell can then
carry four lO-byte CS-ACELP
packets in a frame and not add a lot of delay to the in-
EnterpriseLAtl
488 DATAANDAsyNcHRoNoUsTRANsFEHMoDENETwoRKS
Ftgure 10.17 Packing low-bit-rate voice into ATM.
dividual channels. This approach is particularly
useful for fixed network transport of
digital cellular traffic, which inherenfly involves multiple connections
to and from
base stations.
Example 10.3. Derermine the probabiliry
that the delay of an ATM voice cell
exceed$ l0 msec if the cell is carried on a DSI access link to an ATM network.
Assume the access link is 907o utilized on averase.
solution. The solution
to this problem
utilizes queuing
theory as presented
in
chapter 12 where more problems
of this type are provided.
It involves determining
.
the probability
that the DSl acce$s queue
contains enough cells to represent l0 msec
of transmission time. Assuming all but the framing bit of the DSI signal is used to
carry ATM cells, the service time
(f.)
of a single cell is determined
as
5 3 x 8
t
=
rgt; Booo
=
276
Psec
Therefore, l0 msec of delay represents lo/0.2j6= 36.2 cell times. Equation 12.25 of
Chapter l2 is used to get a solution:
P(>t)
=
Ps-(l*P)t/rn
p(>10
msec)
:
(0.9)e
(14'e)36'2
=
0.025
Example 10.3 indicatesthat2.5vo
of the cells will be delayed by more than l0
msec' Ten milliseconds
of delay on an acce$s link is quite
acceptable considering there
are presumably
only two access links in a connection. Delays of this magnitude are
not acceptable on internal links of an ATM network, but the internal links have very
high bandwidths
and as such very short service times so
queuing
delay of voice cells
10.3 ASYN0HHONOUSTRANSFEHMODENETWORKS 489
is not much of a consideration-as long as voice cells get
adequate
priority and the
CAC limits the number of voice calls on a link.
The solution to Example I0.3 involves a couple of simplifying shortcuts. The first
of these is the use of Equation 12.25, which is intended for exponential service times
as opposed to constant services times implied by fixedJength cells. As discussed in
Chapter 12, constant-service-time analyses are generally not appropriate for ATM cell
traffic because the cells are not statistically independent. A Iarge file or extended du-
ration talkspurt creates a burst of cells that have an effect comparable to a variable-
length message-hence the use of an exponential service time model.
A second simplification of the solution of Example 10.3 is an assumption of ho-
mogeneous traffic-all voice cells. Multiple voice sources are reasonably well be-
haved from a traffic statistics point of view because no one source can
produce an
instantaneous burst of cells as can a file transfer. When a mixture of traffic types with
a mixture of arrival statistics and service priorities exist, the analysis is quite involved
and typically requires simulation
(as
opposed to time-honored analytical solutions).
Other factors that influence the analyses are the
queue size, the queue discipline
(pri-
orities and order of service), cell discard algorithms, flow control algorithms, and the
specific CAC algorithm
(what percentage of a link is allocated to real time versus dis-
cretionary traffic). An extensive amount of research has been undertaken to analyze
the performance of various ATM switch structures and
processing
algorithms. Refer-
ences
[26-28]
are some
good
examples.
The condition of 907o loading of voice channels in Example 10.3 is required to
achieve a controlled amount of delay.If 700Vo loading of voice channels is attempted,
as implied by Example 10.2, the average queuing delay is unbounded. The channel can
still be 1007o loaded if the additional 107o is discretionary, non-real-time traffic. The
solution requires voice traffic to be given priority in a manner that never sends discre-
tionary traff,rc if there are any voice cells in the
queue. Otherwise, the discretionary
traffic impacts the
queuing
analysis. Cell loss for the discretionary
(e.9.,
best effott)
haff,rc will be significant if a link is allocated at close to 100% of capacity.
Traffic Shaping
Traffic shaping is a term used to denote the control of $ources of an access node so
that they conform to a panicular set of traffic statistics
(e.g.,
SCR and PCR). Traffic
shaping may involve flow control for some types of fiaffic or cell discarding
for oth-
ers. Because the statistics of a group
of sources is less variable than the statistics of
individual sources, haffic shaping can be implemented with less impact on individual
traffic streams if a
group of sources
(VCC$)
are packed into a
group and shaped as a
VPC. If a VPC creation
process has access to the sources of the individual VCCs
(e.g.,
64-kbps PCM voice channels), additional
possibilities for traffic shaping occur. In this
case the compression processes for the individual VCCs can be controlled to achieve
a constaflt composite data rate with speech quality degradation distributed across each
of the channels as opposed to degradation occurring to an individual connection as a
result of cell loss or excessive delay. For a description of other techniques of traffic
shaping and an associated analysis ofdelay and cell lo$s, see references
[29
and 30].
490 DATA AND ASYNCHHONOUS THANSFEF MODE NETWORKS
TABLE 10.1 Uncompressed Blt Rates of Dlgltlzed Standards
$tandard Pixels/Line LineVFrame Frames/sec
pixels/sec
Mbps
NTSC
PAL
HDTV
640
s80
1 920
480
575
1 080
9.216 221
16.675 400
62.208 1493
This principle
can also be applied to MpEG-z video encoding of multiple, real-time
video signals
t3l,
321. Maintaining a constant composite data rate of multiple inde-
pendent
channels complicates the multiplexing process
because a figure of quality
most be prepared
and compared across the multiple sources to determine which source
is least affected by a lower bandwidth allocation. Undoubtedly, this process
also adds
delay to the channels since double-pass encoding may be required.
Video
Table l0' I identifies three major television standards and the bit rates required to digi-
tally encode them with no compression and using 24 bits perpixel.
The three standards
are NTSC for North American broadcast TV, PAL for European broadcast TV, and
high-definition
TV
(HDTV).
Two ver$ions of digital compression encoding have been defined by the Motion
Pictures Expert Group: MPEGI and MPJG2. MPEGI is intended for vHS-quality
video and audio. MPEG2 addresses higher quality requirements of broadcast-quality
video and audio as well as HDTV. Because of the encoding complexity, MpEG2 is
primarily
used in broadcast applications where the encoding cost is shared across mul-
tiple destinations. (MPEC2
decoders are relatively inexpensive.)
Data rates are vari-
able depending on the nature of the source material (amount
of movement). The mean
bit rate of MPEG-I is I.544
Mbps while MpEc-? has a mean bit rate of 5 Mbps.
The ATM Forum has defined how MpEG-2 TS packets
are to be packed
into a
AALS-CBR frame. The use of AALI layer is advocated by the rru
oru-T
J.gz).
However, ITU-T has agreed to incorporate ATM Forum defined AALS-cBR packing
into its document. According to the ATM Forum, AALS services are adequate for
MPEG-2 streams because MPEG-2 includes a prograrn
clock reference (FCR)
and
therefore does not need the SRTSs provided
by AALI.
10.4 INTERNET
PROTOCOL TRANSPORT
The Internet evolved from a packet-switching
network consisting of host computers
connected with leased lines to Intemet Protocol (IP)
routers that were, in turn, inter-
connected with leased lines. As this evolution occurred, high-speed digital networks
also evol ved. Figure I 0. I 8 depicts the situation wherein these two developments have
come together in various ways. What is shown is five different means of providing Ip
transporJ using various combinations of networks: ATM, soNET, frame relay, and
direct physical
layer fiber.
30
50
30
|
" , l
l A t l 4 I
f
sotrEr@
10.4 TNTEBNETpnoTocoLTHANSpoRT 491
I
t r
l l
F
l l
F
I
lsoltEr/spHll
Am,,I
llF' -".t' yll
p
I
Fiber plant physical
layer
F'igure 10.18 Intemet transmission layer altematives.
Other than IP transport directly over fiber, IP over SONET
[33]
is the most etfi-
cient. The most inefficient situation is IP on top of ATM. A straightforward approach
to reconciling the two protocol layers is to use a relatively static configuration for the
ATM network with permanent virtual circuits. ln this manner the IP network thinks it
is using leased lines. An AAL5 interface converts the IP packets into fixed-length cells
and ships them through the ATM network to an ATM destination node. At the desti-
nation node the
packet is recon$tructed and passed to the IP layer. Ifthis IP node is the
final destination, all is well and good. More often than not this is not a final destination
because a direct virtual connection between all nodes in a large network is very diffi-
cult to manage.
(Tl1e
number of paths grows arr the sguare of the number of nodes and
nodes are constantly being added and removed.) When the,first ATM node is an in-
termediate node, the IP packet
is
passed back down through an AALS interface to get
back into the ATM network. and so on.
An altemative procedure involves having the ATM network set up SVCs for each
packet
so the ATM destination coincides with the IP destination. This approach avoids
the multiple trips up and down the adaptation layers but introduces an extreme amount
of control overhead within the ATM network.
To minimize IP over ATM problems, Cisco systems has developed a networking
procedure referred to as
"tag
switching" that merges IP and ATM protocols. A tag is
essentially an ATM destination address that gets attached to each
packet (address
and
all) at a tag edge router. The composite message is passed through the ATM network,
which provides tag switching, until it gets reconstructed
and delivered to the destina-
tion where the tag is removed. The simplification occurs because a tag is easier to
process than IP adfuessing and routing. The IETF has standardized the basic concept
of tag switching as multiprotocol label switching (MPLS).
As indicated on the far right of Figure 10. 18, the overhead of multiple network lay-
ers is avoided by directly connecting one IP node to another IP node with a dedicated
fiber link. There are, however, limitations with direct connections. lP networks have
been traditionally implemented with leased circuits that are managed and maintained
by a common carrier. These leased-line services include provisioning, performance
monitoring, and protection switching. Common-carrier provisioning allows rearrang-
492 DATAANDASYNcHRoNoUsTEANSFERMoDENETWoHKS
ing network interconnections
using SONET cross-connect equipment.
performance
monitoring allows automated identification and isolation of faulty or marginal equip-
ment. Physical layer protection
switching assures an application with a reliable con-
nection despite equipment failures or cable cuts. when direct physical
layer
interconnect is used, the IP application must implement application layer protection
switching to circumvent transmission link failures.
Another, possibly
more significant limitation with Ip
"over
glass" is the potential
for a lack ofcarrier-$upported signal regeneration. Although transparent optical hans-
mission (on
fibers or wavelengths) is a seemingly atmactive service offering, a trans-
parent
channel, by its definition, cannot be supported with regeneration. The cariers
could offer transparent optical channels for limited distance applications, but long-
distance applications through optical multiplexing and switching sy$tems cannot be
realized without regeneration. Regeneration is what allows a telephone user to
"hear
a pin
drop."
HEFERENCES
I R. E, Kahn,
"Resource-sharing
computer communications Network s,"
proceedings
of
IEEE, Nov. 1972,pp. 1397-1407.
2 H. Rudin and H. Miiller,
"More
on Routing and Flow Control," National
Telecommunications
Conference, 1979, pp. 34.5. l-34.5.9.
3 J' Rinde,
"Routing
and Control in a Centrally Directed Network,"l{af ional Computer
Conference, 197 7, pp.
603-608.
4 s. c. K. Young and c. I. McGibbon,
"The
control system of the Datapac Network,"
Inte rnat i ona I C onfe rent e on C ommunicatio n s, lg7 6, pp,
I 37- I 4 I .
5 T. s' Yum and M. schwartz,
"comparison
of Adaptive Routing Algorithms for
Computer Communications Networks," National Telecommunications
Conference-
1978, pp. 4. 1. 1- 4. 1. 5.
6 F, E. Heart, R. E. Kahn, S. M. Omstein, W. R. Crowther, and D. C. Walden.
.'The
Interface Message Processor for the ARPA computer Network," spring Joint
Computer Conference, 1970, pp. 55 1-556.
7 L. Kleinrock,
"Principles
and l,essons in Packet communications,"
proceedings
of
IEEE, Nov, 1978, pp. 1320-1329.
I Internet standard 3,
"Requirements
for Internet Hosts: IETF RFC 1122," Internet
Engineering Task Force, lg9B.
9 J' McQuillan,
"Beyond 'Best-Effort'
Routing," Business communitations Review,
May 1998.
l0 M. Hamidi, O. Verscheure, J.
p.
Hubaux, L Dalgic, and
p.
Wang,
..Voice
Service
Interworking for PSTN and IP Networks," IEEE communications Magazine, May
1999, pp.
104-111.
I I The ATM Forum TC,
"Traffic
Management specification version 4.0," April 1996.
12 N. Ghani, s. Nanukul, and s, Dixit,
"ATM
Traffic Management considerations for
Facilitating Broadband Access," IEEE communications Magazine, Nov. 199g, pp.
98-105.
I J
L4
15
t 6
t 7
r 8
l 9
20
2l
24
25
26
27
28
REFERENcES 493
P. Newman,
"A
Fast Packet Switch for the Integated Services Backbone Network,"
IEEE Joumal on Selected Areas of Corununications, Dec. 1988.
K. Sarkies,
"The
Bypass
Queue
in Fast Packet Swirching," IEEE Transactions on
Communications, May 1991, pp. 766-7'74.
F. M. Chiussi, J. G. Kneuer, and V. P. Kumar,
"Low-Cost
Scalable Switching Solutions
for Broadband Nefworking; The ATLANTA Architecture and Chipset," IEEE
Communications M agaline, Dec. I 997,
pp. 44-53.
S. C. Knauer and A. Huang,
"STARLITE:
A Wideband Digital Switch," IEEE
Glohecom, Nov. 1984.
V. E. Benes, Mathematical Theory of Connecting Networks, Academic Press, New
York, 1965.
K. E. Batcher,
"Sorting
Networks and Their Applications," Proteedings of Spring Joint
Computer Conference, 1968, pp. 307-314.
Y-S. Yeh, M. G. Hluchyj, and A. S. Acampora,
"The
Knockout Switch: A Simple,
Modular Architecture for High-Pedonnance Packet Switching," IEEE Joumal on
Selected Areas of Communications, Oct. 1987, pp. 1274-1283.
E. W. Zegura,
"Architectures
for ATM Switching Systems," IEEE Communications
Magazine, Feb. 1993. pp. 28-37.
A. Pattavina,
"Nonblocking
Architectures for ATM Switching," IEEE Communicatians
Magazine, Feb. 1993, pp. 38-48.
M. de Prycker, Asynchronous Transfer Mode Solution
for
Broadband ISDN, Prentice
Hall, Englewood Cliffs, NJ, 1995.
R. Y. Awdeh and H. T. Mouftatr,
"Survey
of ATM Switch Architectures," Computer
Networks and ISDN Systems, Vol. 27, 1995, pp. 1567-1613.
"Multiservice
Switching Forum Implementation Agreement," msf-architecture-Ol.00,
ATMForum, Nov. 1998.
Circuit Emulation Senice Interoperabiliry SpecificationVersian2.0, af-vtoa-fi)?8.000,
ATM Forum. Ian. 199'7.
B. Steyaert, Y. Xiong, and H. Bruneel,
"An
Efficient Solution Technique for
Dirlcrete-Time
Queues
Fed by Heterogeneous TrafFrc," International Joumal of
Communications ,9ystens, Mar./Apr. 1997, pp. 73-86.
A. Ia Corte, A. Lombardo, and G. Schembra,
'?nalysis
of Packet Loss in a
Continuous-Time Finite-Buffer
Queue
with Multimedia Traffic Stream," Intemational
fournal of Communitations, Mar./Apr. 199'1,
pp. 103*1 15,
J. Choe and N. B. Shroff,
"A
Cental-Limit-Theorem-Based Approach for Analyzing
Queue
Behavior in High-Speed Network," IEEE/ACM Transactions on Networking,
Oct. 1998, pp. 659-671.
K, Sriram, T. G. Lyons, and Y, T. Wang,
'Anomalies
Due to Delay and Loss in AALZ
Packet Voice Systems: Performance Models and Methods of Mitigation," IEEE fournal
on Selected Areas in Communications, Jan. 1999, pp.4-17.
K. Sriram and Y, T. Wang,
"Voice
over ATM Using AAL2 and Bit Dropping;
Performance and Call Admission Control." IEEE Joumal on Selected Areas in
Communications, Jan. 1999, pp. 18-28.
494 DATA AND ASYNCHRONOUS TRANSFEH MODE NETWOHKS
S. Gringeri, K. Shuaib, R. Egorov, A. L,ewis, B. Khasuabish, and B. Basch...Traffic
shaping, Bandwidth Allocation, and
Qualilty
Assessment for MpEG video Distribu-
tion over Broadband Networks," IEEE Nen+ork, Nov./Dec. 1g98.
M' Krunz,
"Bandwidth
Allocation strategies for Transporting variable-Bit-Rate video
Traffic," IEEE Communications Magazine, Ian. 1998, pp. 40*46.
33 J. Manchester, J. Anderson, B. Doshi, and S. Dravida,
..Ip
over SONET." /EEE
Communications Mag,azine, May I998, pp. 136-142.
PROBLEMS
l0.l Determine the average transmission capacity required to rehansmit
50-kbyte
mes$ages if they are transmitted intact across a single transmission link with a
bit error rate of 10-6. what is the average retransmission
capacity required if
the messages are packetized
into ?-kbyre packets? (Ignore
the size of the
packet
headers.)
10.? Repeat Problem 10.1 with 2 bit error probability
of l0-a.
10.3 Determine the total number of bits in an HDLC frame if an information
field
contains nothing more than 4 bytes of all 1
's.
Assume the minimum size for all
fields of overhead. Include the opening and closing flag in the calculation.
10.4 Determine the number of 2 xz switching elements in a 32 x 32 banyan switch
matrix.
r0.5 Determine the transmission efficiency that results when a single channel of
G.723.r compressed voice is packed
into ATM cells and transmitted with
minimal delay. G723.1 utilizes 30 ms processing
frames. Assume the higher
of the two standmd rates (6.3
kbps). How much delay is added ro rhe voice
channel if the ATM
cells are packed
with close to l00zo of capacity? (Ignore
speech activity con siderations.
)
3 l
32
1 1
DIGITAL SUBSCRIBER ACCESS
By the early 1990s the public telephone networks of the world had converted to digital
technology in virnrally all of the internal transmission and switching equipment. To-
ward the latter half of the 1990s new digital applications for traditional analog tech-
nology focused on
providing
customer acces$ to the digital network. The Integrated
Services Digital Network
(ISDN)
digital subscriber loops, defined in the mid 1980s,
represent one approach to digital access. Although ISDN has been deployed in many
markets, it has not achieved widespread penetration for various reasons. This chapter
contrasts ISDN access with altemate approaches for accessing the digital network.
Broadly speaking, there are four basic transmission technologies used in
providing
local digital access: copper wire pairs, fiber, coaxial cables, and wireless. Copper ac-
cess utilizes telephony wire pairs
by replacing analog ffansmission with digital trans-
mission for both voice and data
(e.9., ISDN), adding digital transmission onto
conventional analog voice transmission (e.9.,
ADSL), or carrying digital signals in the
existing analog loop using modems. The rapid advancement of modems for realizing
transmission rates of 33.6 kbps in a V.34 voiceband modem or roughly 56 kbps for a
V.90 modem is a main reason for less than expected ISDN usage.
Wide-bandwidth transmission on copper pairs installed for analog POTS service is
complicated by several factors. First, loading coils on longer loops must be removed.
Second, it may be necessary to remove bridged taps for adequate performance. ISDN
basic rate acces$ typically requires bridged tap removal, but newer approaches such
as ADSL utilize more sophisticated channel characterization and equalization tech-
niques to accommodate bridged taps. Crosstalk and interference at high frequencies
are other considerations. A few wire pairs in a cable can car4i the high-frequency sig-
nals, but if a large number of pairs
try to be active at the same time, crosstalk becomes
a significant limitation. Interference from external sources such as AM radio stations
is a consideration at higher frequencies (e.g.,
I MHz). Some higher rate access systems
are designed to detect narrowband interference and mask it out.
Coaxial cable access involves adding data transmission onto unused frequency
bands of a cable TV system. Fiber access has the potential for providing very wide
bandwidths but requires installation of new transmission facilities dedicated to the
new services. Wireless access involves microwave distribution svstems like mul-
4S5
496 DIGITAL SUBSCRIBER AccESs
tichannel multipoint dishibution service (MMDS)
and local microwave distribution
system
(LMDS)
or satellite-based services such as DSS.
11.1 INTEGRATED SERVICES DIGITAL NETWORK
In addition to the digitization of the internal portions
of public
telephone networks, a
lesser known but also significant change involved the development of common-channel
signaling (CCS)
for network control. Both the digitization and the use of CCS srafred
at internal portions
of the network and migrated toward the periphery.
Except for some
special data service offerings and a few network-based features derived from the sig-
naling network, these facilities provided
no direct benefit to the end users. As shown
in Figure I I . I
,
ISDN i s a service offering that extends access to both of these facilities
to the end user. Access to the digital transport facilities occurs on 64-kbps bearer (B)
channels while access to the signaling network occurs on l6- or 64-kbps signaling (D)
channels. Major features or benefits made available by these channels are listed in Ta-
bles 1l.l and 11.2, respectively.
Two levels of digital access to the ISDN network have been standardized: basic rate
access and primary rate access. As shown in Figure I1.2, the (worldwide)
basic rate
interface (BRI)
$tandard is also referred to as a 2B + D interface. In North America.
the primary rate interface (PRI)
standard is sometimes referred to as 238 + D while
the ITU-T counterpart is 30B + D. The North American
pRI
is fundamentally
a L544-
Mbps DSI signal with the D channel replacing one of the 24 message channels (usu-
ally the last one). To achieve a 64-kbps clear channel capability, a BSZS line code is
used to eliminate one's density requirements and common-channel
signaling frees up
the signaling bits so the full 64-kbps bandwidth is available for user data. The ITU-T
I
- - - - - . - /
Figure 11.1 Integrated Services Digital Network access to circuits, channels, leased lines, and
common-channel signaling.
I SDN
AccEte
Li ne
/ a \ -----
f-ffiif-
-*
1 1..I INTEGRATED SERVICES DIGITAL NETWoRK 497
TABLE 11 .1 Features of ]SDN B Channels
1 .
2.
3.
4.
5.
o.
End-to-end four-wire digital circuits: no loss or echoes for voice circuits using digital
instruments
$hared network access for voice, data, and leased lines
Flelatively high bandwidth data channel$ (64 kbps)
Lower error rates than typical voiceband modems
I n"seruice performance
monitoring
Possible expansion of speech bandwidth because elimination of tandem encodings allows
greater one-time quantization
Brrors
Signaling simultaneous with active connections
Calling number identification
Far-end superuision
User-to-user message transfer
Telemetry for fire alarms, $ecurity, meter reading, etc,
Access to
packet-switching
network
$upport for nelwork $eruices such as multiple directory numbers sharing one or more B
channels, trunk group blocking statistics, and identification of calling number for
abandoned or blocked calls
PRI is a 2.048-Mbps El digital signal with the D channel occupying the signaling
channel
(time
slot 16). Because a single D channel can support more than one PRI,
248 and 3lB interfaces are allowed for additional PRIs in a group
of PRIs.
11.1.1 ISDN Basic Rate Access Architecture
An ISDN basic rate access line is a standard copper pair that has been specially con-
ditioned to support a bidirectional 160-kbps aggregate data rate. Transmission tech-
nology required for basic rate access is generally refened to as the digital subscriber
loop (DSL).
Complications arise when using existing analog pairs. The principal con-
siderations are bridged taps and wire gauge
changes, both of which cause reflections
that impact higher speed digital signals. To allow flexibility in the selection and de-
ployment
of the DSL, the ITU-T basic rate specification
I
I
]
does not define a two-wire
transmission $tandard. Instead. it establishes an interface standard that assumes the
presence
of a network termination module that converts any chosen transmission sys-
tem to the standard interface. In the interest of supporting deregulated customer prem-
ises equipment, the Exchange Carriers Standard Association in the United States
established a basic rate transmission standard
[2]
so CPE equipment could connect di-
rectly to the transmission link or select network termination modules f,rom alternate
vendors. Figure 11.3 depicts the architecture and associated terminology of a North
American BRI.
TABLE 11.2 Features of ISDN D Channels
1 .
2.
3.
4.
5.
A
7.
498 DIGITAL SUBSCHIBEF ACCESS
Prim8ry Rete Accgrs
l 2
B I B l .
. . l B
l D
238 +D
I SDN
N6twgrk
-rTpT
BTil
2 B + D
Ba$ic Rate AccBBB
Figure ll.2 Basic rarc and primary
rate access to ISDN.
Module Definitions
NTI : A network termination
module for layer I functions that provides physical
and electrical termination of the transmission link only. In essence, the NTI
isolates the user from the transmission technology but does not demultiplex or
process D channel messages.
NTZ: A second level of network termination that implements functions associated
with layers 2 and 3 of the osl protocol
stack. Thus, NTZ equipment extracts and
processes
D channel messages. Representative
NTZ equipment includes
pBXs,
multiplexers, or LAN gateway$.
TEl: Type I terminal equipment such as a digital telephone that complies with the
ISDN S interface recommendation.
714.' Terminal adapter used to convert from an arbitrary (R)
interface to the ISDN
S interface.
TEZ: A non-ISDN terminal that requires a terminal adapter to interface to the
ISDN S interface. Prevalent examples of a TE2 equipment are analog
telephones or asynchronous (RS-232)
data terminals.
Reference Points
U; Interface to the two-wire transmission line.
I SDN
Network
Figure 11.3 Basic rate interface architecture.
11 1 TNTEcRATED sEnvtcEs DrcrrAL NETWoHK 499
CCITT ISDN interface defined in Recommendation I.430.
Interface to NTZ equipment identical to a T interface.
A non-ISDN interface such as an analog tip and ring.
11.1.2 S/T Interface
The S/T interface is defined in ITU-T recommendation 1.430 to be supplied by net-
work termination equipment (NT2A{TI). It is intended for customer
premises instal-
lations only.
(No
overvoltage protection is prescribed.) The most significant aspects
of the S/T interface are;
T:
S;
R.'
1 .
2.
J -
4.
5.
Four-wire facility
(one pair for each direction)
One kilometer maximum required distance
Alternate space inversion line code
(which
is the inverse of an AMI line code:
see Figure l l.4)
Point-to-point or point-to-multipoint confi gurations
Data rate of 192 kbps with 48 kbps of framing, control, and synchronization
The frame rttructure at reference points S and T is shown in Figure 11.5. As indicated,
each 250-msec frame contains 48 bits. Thirty-eight of these bits are cornmon to both
directions of transmission and are defined as follows:
16 bits in first B channel
(Bl)
16 bits in second B channel
(82)
4 bits in the D channel
I bit in the framing channel F
I bit in the auxiliary framing channel F4
The remaining l0 bits are assigned different functions depending on the direction of
transmission. From the TE to the NT all remaining 10 bits are defined as L bits, which
are used to maintain dc balance. The definition of the 10 bits from the NT to the TE
are:
2 L bits for maintainins dc balance
l r l o l o l o l t l o l r l o l
_JL__n__ -l_r
1r-
---Lr
Figure 11.4 Altemate space inversion line code.
NT to TE
500 DIGITAL SUBSCHIBER ACCESS
48 Bits ln 260
F L Brrr tl
gr
Br tr Br rr E D A F^
r E?tt!tt2B? | 2r?t2 f, u H !t rl rl rt rr
B t B 2 t 0 L F t .
FJ. _J. _J
Figure 11.5 S/T frame structure,
4 E bits that echo D bits received from the TE(s)
I A bit for activation
I N bit, which is the complement of the Fo bit
I M bit far multiframe identification
1 S bit for S channel
Figure I I.5 indicates that the framing bit F is always a binary 0. Even though a
posi-
tive voltage level is indicated, either a positive
or negative voltage is allowed so the
receivers are not sensitive to wiring polarity. A transmitter always produces
the same
level, however, so the receiver always receives the same polarity in every framing bit.
As an aid in rapid acquisition of the framing pattem, the framing bit always represenm
a line code violation (it is the same polarity
as the previous
0). To maintain dc balance,
an L bit with the opposite polarity
of the F bir always follows the F bit. The first 0 in
a data block following the L bit is encoded with the same polarity
as the L bit, which
implies another line code violation. Direct-current balancing of this violation is the
purpose of the L bit at the end of each data block, which also assures that the next
(fixed-polarity)
F bit produces a line code violation.
The reason for the additional L bits in the frame from the TE to the NT arises be-
cau$e more than one TE can be connected to the S intedace as a passive bus (Figure
I 1.6). Because the TEs transmit independently of each other, each individual trans-
mission (D
channel bits and B channel bytes) is individually dc balanced.
Passive bus operations are also the reason for the existence of the NT-to-TE E bits.
Multiple station access to the D channel is controlled by having a terminal wait for an
idle code on the NT-to-TE D channel before hansmitting on the TE-to-NT D channel.
when a terminal begins D channel transmission, it monitors the incoming E bits. If an
incoming E bit does not match the previously
transmitted D bit, that terminal stops
transmitting and waits to seize the channel at a later time. Two levels of priority
are
defined for accessing the D channel. signaling informarion is the highest priority
L t t t l l l l l l l l l l l l l L o L r ^ L l r t t B t ! 2 8 2 8 2 8 2 ! l L 0 L B l E l E t l t r l E l B l u L D L 1 2 u l t i t 1 2 u u | 0 L D t t L
l r -
TEl
s
a
a
I
NTt
u
TET
J
TEl
1 1.1 INTEGHATED SERVICES DIGITAL NETWORK 501
Ftgure ll.6 S-bus connections.
while user
packet
messages are the lower priority.
All terminals on a passive
bus have
equal priorities within each level.
Example 11.1. Determine the distance limit imposed by the need to echo E bits in
a BRI S/T inteface. Assume the speed of transmission of a signal on a pair
of wires
is one-third the speed of light in a vacuum. Ignore transmitter and receiver filter delays
and assume no appreciable delays in the NT logic circuitry.
Solutian. From Figure I I.5 it can be seen that the minimum delay between a
terminal hansmitting a D bit and receiving it back in the following E bit is seven bit
times
(this
is the TE to NT D bit following the first 82 byte). At a 192-kbps dara rare
the duration of a bit is 5.2
psec. Thus, the total round-trip propagation
time is 7
x 5.2
=
36.4 p sec. Assumi ng no appreci abl e ci rcui try del ays i n the NT,
Maximum wire length
= (36.4 x 10*6) x
+
(3 x 108) m
Because round-trip propagation involves both directions of transmission
Maximum distance
=
]
x 3.64
=
1.82 Lrn
Example 1l.I shows that the BRI standard has a fundamental distance limitation
that is not far above the minimum transmission distance specification of 1 km.
11.1.3 ISDN U Interl ace
Prior to the e$tablishment of a standard U interface by the Exchange Carriers Stand-
ards Association in North America a number of basic rate transmission sy$tems were
developed by telephone equipment suppliers around the world and
put
into service.
504 DrcrrAL suBscHtBER AccESS
one of these involves the use of time compression multiplex (TCM).
TcM, developed
by NEC in Japan
[3],
provides
full-duplex transmission on a single pair of wires by
alternately transmitting bursts of data in each direction. For this reason it is sometimes
referred to as
"ping-pong"
tralsmission. one big advantage of rCM transmission is
that near-end crosstalk is avoided because a station is never receiving while transmit-
ting' The big disadvantage is that the burst data rate must be more than twice the de-
sired data throughput.
AT&T in the united states also developed a BRI transmission system for the
No. 5 ESS end office switching system available with generic
releases 5E4 and
5E5
t4l.
This system has a 160-kbps dara rate utilizing a 50zo duty cycle AMI line
code. Full-duplex transmission is achieved by simultaneous transmission in both
directions using hybrids and echo cancelers (ECs)
to separate the two signals, as
indicated in Figure I I.7. Beginning with generic
5E6 the No. 5 ESS supporrs both
the AT&T (Lucent)
u interface (referued
ro as a 5E4l5E5 u interface) and the
ANSI U interface.
Like the 5E4l5E5 U interface, the ANSI U interface uses simultaneous ffansmis-
sion in both directions with echo cancelers and a data rate of 160 kbps. The major dif-
ference is the use of a four-level line code referred to as 2B1e (two
binary digits in
one quaternary
digit). Thus, the symbol (baud)
rate on the line is 90,000 symbols/sec.
Because the line code itself
does not prevent
dc wander, dc restoration is necessary.
Reference
[5]
describes an implementation with adaptive quantized
feedback as being
the best approach. A 2BlQ line code was chosen primarily
because the lower symbol
rate minimizes the two dominant transmission limitations in this application; in-
tersymbol interference and near-end crosstalk
[6].
The frame format and superframe structure of the ANSI U interface are shown in
Figure 11.8. Each frame consists of 240 bits containing lg framing bits, 216 payload
bits
(12
fields of I I 2B + D data bits), and 6 overhead bits. Because the frame rate is
667 frames/sec the data rate is 160 kbps. The 6 overhead bits are organized as a block
of 48 bits in an eight-frame superframe. Functions included in the overhead bits are
24 bits of embedded operations channel, I activation bit, I deactivation bit, I far-end
block error bit, l2 cRC bits, and 9 fixed I bits. All bits except the framing bits are
scrambled fbr transmission.
Figure 11.7 ISDN DSL TX/RX block diagram.
11.2 HIGH-DATA-RATE DIGITAL SUBSCRIBER LOOPS 503
rsw
sw
svv
sw
SW
svlf
sw
sw
28+D
28+D
28+O
28+D
28+D
28+D
28+O
28+D
28+D
28+D
28+D
28+D
28+D
2B+D
28+D
28+D
28+D
28+D
28+ D
2E+D
28+D
?B+D
28+D
28+D
Mr - M"
Mr - M.
Mr - M"
Mr - M.
Mr - M.
Mr - M"
Mr - M"
Mr - Mt
SW
-
Sync Wor d
=
+3+3- 3*3- 3+3- 3+3+3
I SW: I nvwud $W
=
- 3- 3+3+3+3- 3+3- 3- 3
2B+D* l B,
l
B"
l
D
l ( l 8l 8f
2)
Mr-M"
*
Ovdrlrosd bitr
Data ar encoded as 0O
=
-3.
01
=
-1,
11
:
+1,
10
=
+3
Figure 11.8 ANSI U interface frame and superframe structure.
11.1.4 ISDN D Channel Protocol
The D channel
protocol
is defined in two separate series of ITU-T recomrnendations;
the I series and the
Q
series. The data link layer
(LAPD)
is defined in L441 or
Q.921.
This protocol is similar to LAPB of the X.25 standard except
Q.921
allows more than
one logical link.
(Therefore
separate
"connections"
can exist for signaling, packet net-
work, or far-end terminals.) The main functions of the data link layer are message se-
quencing, error checking and retransmission, and data layer link recovery. The
network layer of the D channel protocol is defined in I.451 or
Q.931
. This layer pro-
vides connection setup, alerting, routing, and release of ISDN calls. When a B channel
accesses a
packet network, the X.25 protocol is used while in the connected state.
11.2 HIGH.DATA.RATE DIGITAL SUBSCRIBER LOOPS
A basic rate ISDN digital subscriber loop provides an aggregate, bidirectional data
rate of 160 kbps on a single pair of wires. This section describes several transmission
techniques that allow much larger bandwidths on copper wire pairs. These new tech-
niques are enabled by the availability of low-cost, high-performance digital signal
processing.
Several versions of high-speed digital subscriber
lines have been devel-
oped. The various versions are collectively referred to as xDSL
(see Table I 1.3).
11.2.1 Asymmetric Dlgital Subscriber Line
ADSL allows for high data rates to the subscriber and moderate to low data rates from
the subscriber to the network, ADSL technology was originally conceived as a means
of delivering switched digital video services over a copper loop
[7],
which obviously
do not require high data rates from the subscriber. Although video applications did not
18 1 8 t 8
I
z
3
4
5
6
7
E
504 DIGITAL SUBScFIBEH AccEsS
TABLE 11.3 Verslone of Dlgltal Subscrlber Lines
DSL
ADSL
HDSL
SDSL
VDSL
Digital Subscriber Line (ISDN
basic rate)
Asymmetric DSL (9
Mbps downstream,
640 kbps upstream)d
High-bit.rate DSL (T1lE1
service on two pairs)
Single-line DSL
(T1/El
service on one
pair)
Very high bit rate DSL (52
Mbps downstream, 2.8 Mbps upstream)d
'Listd
data rates ere maximum posslble
values. Downstream ls toward the subscriber. Upstr6em is toward
the network.
materialize, the asymmehic data rate is also suited for Internet access. The philosophy
behind the asymmetric data rates is that subscribers typically need to reieive high-
bandwidth data (for
lntemet file downloads) but normally need to rransmir (query)
at
a relatively low data rate. The subscriber's receive data rate on ADSL varies between
1.5 and 9 Mbps while the subscriber transmit rate varies between 16 and 640 kbps.*
The specific data rate utilized depends on the transmission quality
of the particular
wire pair.
In contrast to ISDN channels, which represent extensions ofthe digital facilities of
the public
switched telephone nerwork (psrN),
ADSL channels are separated from
the public
nerwork at rhe line inrerface of the
psrN.
As shown in Figure I l.g, ADSL
lines terminate at a telephone company central office (or
remote terminal) where the
data streams are forwarded to and received from a facility that is separate from the
telephone network. Transmission between the line interface and an intemet service
provider,
for example, is typically provided
with an ATM connection.
ADSL has two major advantages over IsDN access. First, ADSL data rates pro-
vided to the subscriber are significantly higher than the lZg-kbps ISDN basic rare.
second, ADSL piggy
backs digital transmission
on a standard *atog telephone wire
pair. Thus, existing analog telephones are retained on ADSL but are either replaced
by digital phones
or are connected
through conversion
devices when ISDN is utilized.
Two versions
of ADSL
transmission links have been developed: carrierless ampli-
tude and phase (CAP)
modulation and discrete multitone
(DMT)
modularion. CAp is
the first version deployed but DMT has been selected as the standard
[g].
Because
DMT makes more intensive use of Dsp, a DMT implementation
typically requires
more power-a
significant consideration for remote terminal deployment. cAp, on
the other hand, its not generally
considered to be as flexible as DMT in achieving
maximum data rates on some wire pairs
or in some interference
environments. For a
thorough comparison
of the two alternatives, see reference
[9].
DMT lmplementdtion
Basic parameters
of the standard ADSL DMT implementation
are provided
in Table
I 1.4. A block diagram of an end-to-end DMT implementation
is shown in Figure
I I.10. As mentioned in chapter 6, a DMT implementation
utilizes an inverse FFT as
*A
less ambitious vetsion of ADSL rcferred to as G.Lite only attempts to achieve 1.544 Mbps clownstream
and 384 kbps upstream.
P$TN
Switcl't
metrix
Line interfaces
Subscriber
loops
Splitters
volce
volCc
Figure 11.9 ADSL network configuration.
TABLE 11.4 AD$L DMT lmplementatlon Paramters
11.2 HIGH.DATA.RATE DIGITAL SUBSCRIBER LOOP$ 505
multiplexer
Subchannel $eparation
Maximum bits/$ubchannel
4.3125 kHz
15
512 samples
32 samples
s44 (512 + 32)
2,208 MHz (51? x 4312,5)
246.377 x 10a sec (5,+4/2,208 x 106)
276kHz
(subchannel 64)
64 samples
4 samples
68 (64 + 4)
276 kHz (64 x 4312.s)
?46.977 x 104 sec
(68/0.276 x 1Oo)
69 kHz
(subchannel
16)
Downlink
Number of subchannelsa
255
FFT sample size
Cyclic prefix
Total number of samples
Sample rate
FFT frame duration
Pilot frequency
upl i nk
Numberol subchannels
31
FFT sample size
Cyclic
prelix
Total number of samples
$ample rate
FFT frame duration
Pilot frequency
the number of subchannFl8 dBpends on the amount of matgin tor PoTS filterlng and
on choice of echo cancling or FDM to isolate the two directions ol transmission.
506 DIGITAL sUBscRIBER AccESs
a modulator and an FFT as the corresponding demodulator. The following paragraphs
identify the basic funcrion of each block in Figure I I . 1 0:
PRS source: A pseudorandom
sequence generator provides
a prescribed
sequence
for characterizing the channel during a training period.
Characteristics
of the
channel that are determined
during the training periorl
are attenuation and phase
distortion across the band, noise/interference
levels across the band. and the
information
capacity of individual
subchannels.
Channel Allocation: Provides assignment of data bits to individual
subchannels
according to the subchannel capacity determinations
obtained during training.
FEC: A combination
of Reed-solomon
and convolutional
coding.
IFFT
QAr4
Modulation:
conversion of data values to quadrature
channel signal
amplitudes and conversion to a time-domain waveform using an inverse FFT.
D/A : Digital-to-analog
conversion.
TX Filter: Bandpass filter to prevent
interference into the voiceband
and to smooth
the discrete DSP samples.
POTS Splitter" Used at both ends of the line to separate the analog voice from the
data. (Simpler
versions of ADSL incorporate the splitter function in the modems
to simplify installation.)
AGC.' Automatic gain
control to adjust overall receive level.
sktpe/Delay Equaliption'
A front-end equalizer ro partially
flatten the frequency
response
and equalize extreme delay variations in the chaanel
[10].
,4./D.' Analog-to-digital
conversion.
Source data
Wire pair
Figure ll.l0 ADSL DMT block diagram
11,3 DIGITAL LOOP CAFHIEB SYSTEMS 507
FFT
QAM
Demodulation: FFT conversion of time-sampled waveform to frequency
domain where data values are related to amplitudes of quadrature carriers.
Clock Recovery: AID sample timing obtained by locking to pilot frequency. The
desired sampling rate is eight times the pilot frequency so 1-of-8 phase
ambiguity has to be determined by monitoring framing/synchronization bit
integrity.
Frequency-Domain Egualizer: Multiplication of complex
(quadrature)
frequency $pectrum by amplitude and phase equalization pararneters obtained
during training period.
Data Detection and Interleaving: Slicing of
quadrature amplitude values to
decode data and subsequent generation of composite stream identical to original
source data.
11. 2. 2 VDSL
VDSL is an expanded version of ADSL to achieve even higher bandwidths on
par-
ticularly short lines as would be available from remote terminals of fiber to the curb
systems. A primary motivation for VDSL is potential distribution of HDTV signals.
Although several different modulation techniques have been proposed for VDSL, a
DMT version rieems to be favored
[11].
11.3 DIGITAL LOOP CARRIER $YSTEMS
As discussed in Chapter 1, the primary purpose of a digital loop carrier
(DLC)
system
is to reduce or eliminate copper
pairs fiom a central office to the vicinity of a
group
of subscribers. Even though the transmission link from the central off,rce to the DLC
remote terminal is digital, the transmission links from the remote terminal to the sub-
scribers are typically conventional analog loops. Thus, the main purpose of a DLC is
not to provide digital subscriber access. Nevertheless, some DLC systems (particu-
larly fiber-based DLC systems) provide options for T1, ISDN, or xDSL digital inter-
faces. Moreover, the fact that fhe DLC remote terminal is relatively close to the
subscriber locations facilitates the use of these interfaces.
(A short copper drop from
the remote terminal has low attenuation and crosstalk and is less likely to have loading
coils, bridged taps, or multiple sections of wire with varying gauges.)
Although a DLC
is a natural mechanism to provide enhanced services through ISDN or xDSL digital
interfaces, mechanical
packaging and
power considerations of these interfaces repre-
sent unique requirements with respect to strictly POTS applications.
11.3.1 Universal
Digltal Loop Carrier Systems
As shown in Figure I1.11, a universal digital loop carier
(UDLC)
system
[12]
can be
interfaced to any switching system: analog or digital. The interface between the local
508 DtetrAL suBscRlBER AccEss
Customer
loops
COT
RT
Figure 11.11 Universal digital loop carrier system.
switching system (end
office) of the public
network and the central office terminal
(coT)
involves individual circuits (e.g.,
individual
analog rip and ring connections).
The multiplexed
digital transmission links between the COT and the remote terminal
(RT)
can be wire pairs
or fiber. Each interface of a COT is paired
with a corresponcling
subscriber intetface at the RT so the use of a UDLC is transparent to both the switch
and the subscriber. In its simplest mode of operation, the uDLC uses pure
multiplex-
ing between the COT and the RT so that there i$ a one-to-one correspondence
between
a particular
TDM channel and the coT/RT interface pair.
some systems can also be
configured with concentration
wherein the Cor/RT pairs are dynamically assigned
transmission channels. If the number of requested
conversations exceeds the number
of channels, blocking occurs. The possibility
ofblocking introduces nontransparency
and implies that some means of returning a reorder tone (fast
busy) is needed in the
RT.
UDLC installations
are configured to match each particular
interface of the central
office switch with a complimentary
interface in the RT. A fully capable system must
provide
a wide variety of interfaces such as loop-start line, ground-start
pBX
trunk,
foreign exchange lines, and coin telephone interfaces. In some early systems the con-
frguration process
involved nothing more than physically
installing matching inter-
faces in respective
equipment slots of the CoT and RT. More recent.systems typicatly
utilize line units with multiple-service
capabilities. These systems can be configured
electronically (i.e.,
no straps) with either a local or a remore manasement
interlace.
11.3.2 Integrated Digitat Loop Carrier Systems
Whenever a universal digital loop carrier system is interfaced with a digital switch, ob-
vious inefficiencies
occur in terms of back-to-back demultiplexing-multiplexing
and
D/A-A/D conversion.
As shown in Figure 11.12, an integrated digital loop carier
0DLC)
system eliminares the inefficiency by directly connecring
the DLC TDM link
to the digital matrix. Typically rhe direct digital connections
are either DSI or El
crorts-connect
signals. Thus, a fiber-based
DLC system will typically interface
with
the switch through multiplexing/demultiplexing
equipment as some number of digital
crostt-connect
signals.
11.3 DIGITAL LOOP CARRIEH SYSTEMS 509
Customer
loops
Figure 11.12 Integrated digital loop carier system.
From a functional point of view an early IDLC systems was nothing more than a
distributed switching system wherein some line interfaces of the switch are moved to
remote locations. Most cenfial office switch manufacturers
provide such a capability-
sometimes as simply as using channel banks for analog line interfaces that can be colo-
cated with the matrix or remoted. Central office switch vendors also offer remote
switching modules wherein some
portion of the matrix itself is remotely located.
These configurations provide remote concentration for efficient use of the connecting
transmission link and, in some cases,
pmvide local switching in the remote module so
connections between two
por"ts of a remote module do not use the transmission link.
(If
remote local switching is not
provided, a connection between two ports
of a remote
module requires two channels of the hansmission link.)
Remote modules of a particular switching system vendor are often implemented
with proprietary signaling and management
processes that preclude use of DLC equip'
ment from other vendors
(except
when a DLC vendor develops the ability to emulate
a particular switch vendor's remote modules). In the interest of opening IDLC appli-
cations to competing vendors, Bellcore established an IDLC standard referred to as
GR-303
[13,
14] that compliant switching system vendors must support
(possibly in
addition to a proprietary IDLC capability). Included in the GR-303 specification are
definitions for signaling,
provisioning, testing, alarm surveillance, and performance
monitoring.
Due, in part,
to a move to unbundle
LEC local loop services GR-303 has assumed
a much broader scope than
just
an IDLC application. Because a GR-303 capability
in-
cludes being able to define and administer a myriad of switch interface types, the GR-
303 $tandard can be used for interfacing other types of equipment such as xDSL
equipment. ETSI has established a similar IDLC standard for intemational
(ITU)
switching equipment refened to as a V5 interface
[5,
16].
11.3.3 Next-Generation Digital Loop Carrler Syetems
The term next-generation digital loop carrier
(NGDLC)
has been adopted within the
industry to refer to DLc sy$tems fhat adhere to GR-303 and
provide
additional con-
figuration options and interfaces. There is no
precise definition of what constitutes an
NGDLC system. Four basic attributes are adherence to GR-303, optical fiber trans-
mission capabilities
(e.g.,
SONET/SDH),
generally larger line sizes, and an ability to
interface with a cenhal office operational support system for diagnostics, alarms, and
510 DIGITAL SUB$CRIBER AccESS
Figure ll.l3 Next-generation digital loop carrier.
remote provisioning
[l4].
other aspects commonly available in NGDLC systems are
depicted in Figure I 1.13.
The most important
aspect of an NGDLC $ystem from a services point
of view is
the availability of new services such as copper or fiber digital subscriber interfaces in
addition to the conventional
POTS interfaces. The most desired digital interfaces are
Tl/El, primary
rate and basic rate ISDN, ADSL, and vDsL. other digital inrerfaces
may also be provided
for services like broadband data and digital video. Although the
diversity of interfaces belies use of a single multiple service line intedace for total
electronic provisioning,
the system must report inconsistencies
between installed
hardware and the elechonic
database. A particularly
desirable feature of the system
shown in Figure I l. I 3 is drop-and-insert
capability, which allows distributed access
to a single backbone digital route. GR-303 identifies star configurations, linear ADM
distribution, and ADM rings as desirable topologies. NGDLC sysrems thar support ad-
vanced interfaces or topologies necessarily
use a COT to separate out the special serv-
ices from the Pors. In essence, the cor performs
cross-connect operations to groom
and distribute various types of traffic.
11. 4 FI BER I N THE LOOP
Fiber in the loop (FITL)
is a generic
term that refers to one of three more specific de-
scriptions of the use of fiber for local distribution. The flrst category is fiber to the
cabinet or fiber to the neighborhood.
These systems are often installed by local tele-
phone
companies as part
of the feeder portion
of their local distribution facilities. Tra-
ditional copper pairs
for voice service extend from the cabinet to subscriber
residences. Because a relatively
long copper pair
is used for
,.the
last mile,,' they do
not provide
much opportunity for enhanced service offerings. Their use andjustifica-
tion are strictly based on reducing rhe cosr of
pOTS
disFibution.
The second caregory of FrrL is a fiber-to-the-curb (FTTC)
system. As the name
implies, these systems are designed to reach within 1000 feet of a subscriber residence.
An FTTC system is generally
intended to provide
enhanced services such as video or
high-speed
data using ADSL or VDSL. Distribution
of the enhanced services from rhe
"curb"
location is canied over wire pairs
or coaxial cable. These systems are essen-
tially identical to advanced DLC sysrems with optical rransport.
The third category of FITL is f,rber ro the home (FTTH).
These sysrems obviously
off'er oppornrnities for extremely large bandwidths to the home but have signifircant
11. s HYBRTDFTBERcoAXSYSTEMS 511
deployment obstacles. First, installation in established neighborhoods is expensive be-
cause underground installations
(under
streets and driveways) are normally required.
Second,
providing network power to subscriber equipment is a major problem. The
lack of
power for enhanced service applications is not much of a consideration but
power to telephones is. Local exchange carriers go to great effotts to ensure inde-
pendence from commercial power systems for both their office switching systems and
the connected subscribers.
(A
central office typically maintains enough batteries to
keep a system up for ?4 hours if commercial
power is lost. If power is out for longer
than this, diesel generators
are available.) A further complication of providing tele-
phone service over FTIH is the need to convert a digital voice channel to analog for
interfacing to conventional telephones-an additional expense and power problem.
Batteries are often offered as an alternative to line powered telephones but logistical
problems remain; who maintains the battery and what happens if the battery is dead
when commercial
power is lost and someone wants to diat 911?
11.5 HYBRID FIBER COAX SY$TEMS
The cable TV systems installed around the country can be augmented with down-
strearn data transmission to subscribers by merely adding
"cable
modems" utilizing
unused or displaced TV channel bandwidth. Upstream transmission from the sub-
scriber to a cable TV head end is much more difficult. Although many cable TV sys-
tems were designed and installed with upstream transmission as an option, the
bandwidth available to the upstream channels
is generally limited and often subject to
very high noise and interference levels. Upstream transmission limitations can be
somewhat alleviated by utilizing a conventional telephone connection with voiceband
modems with data rates up to 28.8 kbps. The telephone modem connection is used in
the same basic manner as the upstream channel of an ADSL while relative high-
bandwidth downstrearn data are carried on the cable. The disadvantages of this solu-
tion include the need for a subscriber telephone line, the cost of large numbers of
telephone channel connection$ into a service
provider, and the need to coordinate dial-
up telephone comections with particular cable channel users.
The basic configuration of a hybrid fiber coax
(tFC)
$ystem is depicted in Figure
I 1.14. The coaxial cable portions of CATV system$ are configured as tree-and-branch
topologies with all customers receiving the same, multichannel broadcast signal. Am-
plifiers are in$erted wherever the signal level
gets
unacceptably low from attenuation
and branching loses. Bidirectional amplifiers are shown in Figure 11.14 under the as-
sumption that this sy$tem is an application with a return path from the residences. In
TV-only applications the return path (if there is one) is used for premium channel se-
lection. In expanded $ervice applications the return path carries voice or data with fre-
quency division multiplexed cable modems in a band from 5 to 42 MHz.
The optical fiber transmission
portion of an HFC sy$tem repre$ents a replacement
of relatively long haul coaxial cable sections with numerous amplifiers. For this rea-
son, the optical transmitters and receivers are designed to carry a wideband analog sig-
nal. Notice that the optical links are shared by a large number of cu$tomers
(anywhere
512 DIGITAL SUBS0RIBER AccEsS
Fiber node
Figure 11.14 Hybrid fiber coax system configuration.
from 100 to 1500). upgrading an IIFC system for new $ervices typically requires
Sreater
penetration
of the optical frber portions
so that fewer households
are connected
to a common coaxial cable segment. In the limiting situation, wherein each household is
connected through a dedicated coaxial cable, an FIFC system becomes an FTTC system.
Downstream
digital services can utilize cable modems that typically pack
30-40
Mbps into a 6-MHz analog TV channel. 64-eAM modulation is commonly
used. In
newer FIFC system.s, new downsfieam digital services can be canied at fuquencies above
450 MIJZ while the band from 54 ro 450 MHz is reserved for haditional analog Tv.
A major impediment to upgrading an IIFC system for return channel services is the
shared use of a coaxial cable segment common to some number of households. The
network termination
within each home is passive
and bidirectional,
which means that
all noise and interference within a home is passed
onto the cornmon cable to all other
homes. Thus, a single source of interference
can disrupt the signal to all other homes
served by the common coaxial cable. Furtherrnore,
the noise and interference
of all
households are additive, indicating
the need to limit the number of households served
by a single coax segment. An additional drawback of the shared cable is the need for
some form of encryption for content security. TDMA return channels help minimize
the interference problem
by blocking all output energy from a residence except when
an allotted time slot occurs.
using an HFC system for POTS has the same basic drawbacks as an FTTH sy$tem
in that there is no inherent facility for line powering
rhe telephones. Thus, FIFC might
not be used for primary ("lifeline")
POTS distribution but could be quite
effective in
providing
secondary telephone
applications. The main attribute of HFC for enhanced
services is the ability to provide
dynamic assignment of high-rate digital downstream
channels
and relatively low rate full-period upstream channels.
11. 6 VOI CEBAND
MODEMS
voiceband modem technology
improved
dramatically in the early 1990s with the cul-
mination of 33.6 kbps becoming
standard with ITU recommendation
v.34
[17].
The
1 1 ,6 VOICEBAND MODEM$ 513
rapid advance of voiceband modem
performance was due to two primary factors; the
availability of economical DSP technology for equalization/echo canceling and the
improved
quatity of the network in terms of lower noise and distortion resulting from
the near-all-digital implementation, In an all-digital network the only significant
source ofnoise is the quantization noise ofthe analog-to-digital conversion.
Example 11.2. Determine the theoretical maximum data rate of a perfectly
equalized voiceband modem undergoing a single A/D and D/A conversion.
Solution. The signal-to-quantizing-noise ratio (SQR) is derived in Chapter 3 to be
on the order of 36 dB, which conesponds to a
power ratio of 3981. Using this value
in Shannon's theorem for the theoretical capacity of a channel yields
C= lVlogz(l + SNR) bps
=
3100 l ogr(l + 39Bl )
=
37 kbps
where the channel is limited by the transformer coupling, 60 Hz elimination frlters,
and D/A smoothing filters to extend from 300 to 3400 Hz.
11. 6. 1 PGM Modems
As the
previous problem indicates, a V.34 voiceband modem provides data rates that
are near the theoretical limit imposed by
quantization noise alone. Recognition that the
principal
source of noise in the end-to-end connection is the
quantization noise of the
A./D converters leads to alternative modem implementations that directly utilize the
digital 64-kbps channel and eliminate the quantization noise
[18].
These modems are
commonly referred to as PCM modems.
As shown in Figure 11.15, a V.90 PCM modem relies on the digital network to
carry an unaltered digital signal from a digital source to a digital-to-analog con-
version device
(codec)
at an analog subscriber interface. The codec converts the PCM
codewords to PAM samples that are detected by the receiving customerpremises mo-
dem and converted back to the original PCM data. Successful data detection requires
the receiving circuitry to adequately equalize the combined distottion of the D/A
smoothing filter and the transmission link, to know the quantization levels of the
codec, and to become synchronized to the D/A conversion clock. The equalization
and
quantization requirements are determined during an initialization proce$$ while clock
synchronization requires processing of data transitions in the multilevel received
waveform.
It might seem that a PCM modem could
provide a data rate of 64 kbps. However,
several factors restrict the data rate to something less than 64 kbps. The first of fhese
is the bandpass filter in the D/A codec
(for
60 Hz elimination and sample smoothing).
A second conskaint
is the
possibility
that the digital path through the network might
514 DIGITAL SUBSCHIBER AccEsS
Tl or ISDN digitat lnterface
m1orrm 10mfi 00 10t10011
Subscrlber loop
F
+'*.-.H*
mlor roo 1fiI}l roo 1ol tmii
Figure 11.15 Downstream V.90 modem concept.
include a digital pad
for changing the signal level of rhe assumed analog signal.* A
third constraint is rhe possibility (in
North America) that robbed bit signaling might
be in use on one or more of the digital links.
The fact that the overall bandwidth of the channel is slightly over 3kHz implies that
the maximum, intersymbol, interference-free
pulse
rate is
just
over 6 kHz. Thus, the
actual sample rate of 8 kHz implies that some amount of intersymbol interference is
inevitable (assuming
the PCM samples are independent).
The lack of a low-frequency
transmission response is accommodated
by V.90 mo-
dems by utilizing every eighth
pcM
sample sorery fbr dc restoration. Thus, the maxi-
mum data rate is 56 kbps. If robbed bit signaling is present,
its effects are minimized
by determining, during initialization, which frames in the six frame sequences are sig-
naling frames and then using only 7 bits per sample in those frames und
g
bits p*r ru*_
ple in the nonsignaling
frames. Digital pads
can be accommodated
by detectins their
presence
and modifying the digital codewords
according to the particular
amolunt of
attenuation being inserted by the pad.
The v'90 uplink direction is implemented
as a convenrional
(v.34)
modem.
thereby implying asymmetry in the data rates. It is conceivable that
pAM
signaling
could be used in both directions,
but the uplink is more complicated to implement
and
is often unnecessary because most applications (e.g.,
Intemet access) are inherently
asymmehic in the data rates required. It is also possible
to utilize analog PAM on both
-Dgital
pads
are often included in the codecs for digital setting of gain
levels to desired voiceban d sigflal
Ievels.
Dlgital network
1 1,7 LOCAL MICHOWAVE DISTFIBUTION SEHVICE 515
ends of the connection (as opposed to
just
one end), but these implementations are
considerably more complicatd
[19,
20].
The main attraction of PCM modems is that they provide almost the same data rate
as an ISDN B channel but do not require changes in the line interfaces or special treat-
ment of the customer loop
(e.g.,
the elimination of bridged taps). When the customer
loop is a short drop from a remote terminal of a digital loop carrier, the maximum data
rate of 53-56 kbps is assured.
11.7 LOCAL MICROWAVE DISTRIBUTION SERVICE
The FCC has allocated two separate microwave frequency bands for digital radio ac-
cess communications services. The first of these is Multichannel Multipoint Distribu-
tion Service
(MMDS),
which operates at 2 GHz. MMDS is essentially a wireless cable
sy$tem and, as such, provides only one-way transmission. Reverse channel commu-
nications rcquires dial-up modem connections through the telephone network.
Local Microwave Distribution Service
(LMDS)
operates at 28 GHz and provides
cell-based, two-way communications. Bandwidths available with LMDS are depend-
ent on a particular service
provider's frequency plan. Typically, these
plans provide
bidirectional data rate$ on the order of 1.5-50 Mbps, although the total allocated band-
width of almost 1 GHz can be partitioned to provide higher rate$ or asymmetric rates
if desired. International LMDS allocations have been made in a range of 25-31 GHz.
These systems are also refbrred to as Local Multipoint Communications Systems
(LMCS) in Canada and Brazil.
The principal applications supported by LMDS are voice, video, and high-bandwidth
data. The immense amount of bandwidth available is the main attraction. The major
cost of an LMDS system is associated with the subscriber radio equipment. Thus, after
the cell sites are established, additional costs are incurred only as subscribers begin
paying for service. This situation is in contrast with new fiber-based facilities that re-
quire major, up-front investment before revenues are realized. The most immediate
opportunity
for LMDS is to offer an alternative to high-speed digital leased lines for
businesses.
High-speed Internet access and HDTV to individual residences are secon-
dary opportunities.
The major disadvantages of LMDS are the need for FCC licensing, line-of-sight
transmission, distance limits of 2-3 miles, extreme attenuation in heavy rainfalls,
po-
tential interference from other services such as satellites in the same bands, and the
need for local power in all subscriber
(telephone) sites. Fade margins or dynamic
power control on the order of 40 dB overcome all but the mo$t severe rainfall rates.
The requirement tbr line-of'-sight transmission means that a single base station can-
not communicate with all locations in a cell
(except
maybe in West Texas or the Aus-
tralian outback). Overlapping cells provide greater coverage, but some locations are
sure to be in the shadow ofall base stations unless reflectors can be installed. Line-of-
sight transmission can be further compromised by the appearance of buildings or trees
after initial deployment.
516 DtctrAL suBscRtBER AccESS
11.8 DIGITAL SATELLITE SERVICES
Direct Satellite Service (DSS)
has recently become a viable alternative to cable TV
services. Because the DSS
system utilizes digital transmission for its TV channels, it
is straightforward for a DSS provider
to offer data communications
services to the TV
subscribers. However, because DSS is a one-way communications service, dial-up
modem connections through
the telephone network are required for two-way data
communications. DSS is also limited ro the toral bandwidth (400
MFIz) of the satellite
transponders, which must be shared by the Tv ffansmissions
and any allocated data
channels.
In contrast to DSS, which uses geostationary
satellite orbits, Low Earth orbit sat-
ellites oflridium and Teledesic provide
opportunities for lower cost, bidirectional user
terminals. As mentioned in Chapter 9, Iridium is primarily
oriented to relatively nar-
row band applications such as voice, messaging,
and paging.
Teledesic, on the other
hand, is oriented to wide-bandwidth data applications.
REFERENCES
Recommendation I.430 ISDN user-Network Interface; Layer I Recommendations,
Fascicle IlL9,CCITT Blue Book,1988, pp. lll-240.
American National standards
for Telecommunications,,.Integrated
services Digital
Network-Basic
Access Interface for use on Metallic Loops for Application on the
Network Side of the NT-Layer I Specification," ANSI TI.60l-1988.
R. Komiya, K. Yoshida, and N. Tamaki,
"The
Loop coverage comparison between
TCM and Echo canceller under various Noise considerations,"rEEE Transactions
on
Communications,
Nov. 1986, pp. 1058-1067.
sESs@. switch ISDN Basic Rate Inrerface specification, 5E5 Generic
program
Issue
I.00, AT&T, Dec. 1987, pp. II-B-t/15.
s. sugimoto' K. Hayashi, and F. Mano,
"Design
of 2Ble rransceiver for ISDN
subscriber Loops," IEEE Intemational
conferente on communication,Iune
19g9, pp,
8. 1. r
- 8.
1. 5.
J. w. Lechleider,
"Line
codes for Digital subscriber Lines," IEEE communications
Magazine,
Sept. 1989, pp.25-32.
w' chen and D. waring,
"Applicability
of ADSL to support video Dial rone in the
Copper Lttop," IEEE Communications Magazine, May 19g4, pp,
102_109,
"Network
and customer Installation
Interfaces-Asymmetric
Digital subscriber Line
(ADSL)
Metallic Inrerface,"
ANSI Tl.4t3-1995, Aug. 1995.
B' saltzberg,
"comparison
of single-carrier and Multitone Digital Modulation for
ADSL Applications,"
IEEE Communications
Magasine, Nov. 199g, pp. ll4_lLL.
w. Y. chen, DSL simulation Techniques and standards Development
for
Digital
subscriber Line,sysferns, Macmillan
Technical
publishing,
Indianapolis,
l9gg.
J. M. Cioffi, V. Oksman, J. J. Werner, T.
polletr, p.
M.
p.
Spruyt, J. S. Chow, and K. S.
Jacobsen,'-very-High-speed
Digital subscriber Lines," IEEE communications
Magazine, Apr. 1999, pp.72-79.
r0
n
PROBLEMS 517
12
"Functional
Criteria for Digital Loop Carrier Systems," Technical ft.eference,
TR-NWT-000057, Bellcore, Morristown, NJ, Jan. 1993.
t3
"Integrated
Digital Loop Carrier System Generic Requirements, Objectives and
Interface," Generic Requirement, GR-NWT-000303, Bellcore, Morristown, NJ' Dec'
I 998.
14
"Generic
Operations Interfaces Using OSI Tools; Information Model for Integrated
Digital Loop Carrier and Fiber in the l,oop Systems," Generic Requirements,
GR-NWT-002833, Bellcore, Morristown, NJ, Dec. 1996'
Recommendation G.964
(06194),
"V-Interfaces
at the Digital Local Exchange
(LE)-VS.l-Interface (Based on 2048 kbiVs) for the Support of Access Network (AN)'
ITU, Geneva Switzerland,
Recommendation G.965
(03/95)-V-Interlaces at the Digital Local Exchange
(LE)-VS.2 Interface (Based on 2048 kbit/s) for the Support of Access Network
(AN).
ITU, Geneva, Switzerland.
G. D. Forney, L. Brown, M. V. Eyuboglu, and J. L. Moran,
"The
V'34 High-Speed
Modem Standard," IEEE Communications Maga4ine, Dec. 1996, pp' 28-33.
B. Townsend,
"High
Speed Communications System for Analog Subscriber
Connections," U.S. Patent 5,801,695, Sept. l, 1998.
I. Kalet, J. E. Mazo, and B. R, Saltzberg,
"The
Capacity of PCM Voiceband Channels,"
Intemational Communications Conference.
1993, pp. 507*51 l.
E. Ayanoglu, G. Golden, R. Jones, J, Mazo, and D. Shaw,
"High
Quantization-Level-Sampling
Modem with Equalization Arrangement," U.S.
5,528,625, June 18, 1986.
PROBLEMS
11.1 Determine the maximum loop length of a ping-pong transmission system util-
izing 8-kHz data bursts of 50 psec duration in each direction. Assume the ve-
locity of propagation is one-third the speed of light.
ll,2 Determine the maximum theoretical data rate that can be achieved by a voice-
band modem passing through the
public network and a single universal digital
loop carrier system. Assume there are no signal imperfections in the digital
poftions of the facilities.
r5
l 6
t 7
l 8
19
Speed
Patent
r1-3
11.4
Determine the range of data rates achievable with a V.90 modem
passing
through six Tl facilities with
"robbed
bit" signaling.
Assume a multipair cable system is used to carry bidirectional high-bandwidth
AMI signals and that near-end crosstalk coupling from one pair to another
is
IVo
(-2O
dB). If the system is crosstalk limited, determine the maximum num
-
ber ofactive
pairs for a bit error rate of 10-6.
Repeat Problem I1.4 with a 3-dB interference margin for noise and other im-
t',t#Fi
It#t",*
"
x
h't
11.5
perfections.
\t'l\\
1 2
TRAFFIC ANALYSIS
Except for station sets and their associated loops, a telephone
network is composed of
a variety of common equipment such as digit receivers, call
processors, interstage
switching
links, and interoffice trunks. The amount of common equipment designed
into a network is determined under an assumption that not all users of the network
need service at one time. The exact amount of common equipment required is unpre-
dictable because of the random nature of the service requests. Networks conceivably
could be designed wifh enough common equipment to instantly service all requests
except for occurrences of very rare or unanticipated
peaks. However, this solution is
uneconomical because much of the common equipment
is unused during
normal net-
work loads. The basic
goal of traffic analysis is to provide a method for determining
the cost-effectiveness of various sizes and configurations of networks.
Traffrc in a communicatrons network refers to the aggregate of all user requests be-
ing serviced by the network, As far as the network
i$ concerned, the service reque$ts
arrive randomly and usually require unpredictable service times. The flust step of traf-
fic analysis is the characterization of traffic arrivals and service times in a
probabilistic
framework.
Then the effectiveness of a network can be evaluated
in terms of how
much traff,rc it carries under normal or average loads and how often the haffic volume
exceeds the capacity of the network.
The techniques oftraffic analysis
can be divided
into two
general categories: loss
fiystems and delay systems. The appropriate analysis category for a
particular system
depends on the system's treatment of overload traffic. In a loss system overload traff,tc
is rejected without being serviced. In a delay system overload traffic is held in a queue
until the facilities become available to service it. Conventional circuit switching op-
erates as a loss system since excess traffic is blocked and not serviced without a reffy
on ttre part ofthe user. In some instances
"lost"
calls actually represent a loss ofreve-
nue to the carriers by virtue of their not being completed.
Store-and-forward message or packet switching obviously
possesses the basic
characteristics of a delay system. Sometimes, however, a
packet-switching operation
can also contain certain a$pects of a loss $ystem. Limited
queue siees and virnral cir-
cuits both imply loss operations during traffic overloads. Circuit-switching
networks
also incorporate certain operations ofa delay nature in addition to the loss operation
519
520 THAFF|c ANALysts
of the circuits themselves. For example, access to a digit receiver, an operator,
or a call
proce$sor
is normally
controlled by a
queuing
proces$.
The basic measure of performance
for a loss system is the probability
of rejection
(blocking
probability).
A delay system, on the orher hand, iJ measured in terms of
service delays' Sometimes the average delay is desired, while at other times the prob-
ability of the delay exceeding some specified value is of more interest.
.
some of the analyses presented
in this chapter are similar to those presented
in
chapter 5 for the blocking probabilities
of a switch. chapter 5 is concerned mostly
with matching loss-the probability
of not being able to set up a connection througir
a switch under normal
or average traffic volumes. This chapter, however, is mostly
concerned with the probability
that the number of active sources exceeds some speci-
fied value' Typically, the specified value is the number of trunk circuits in a route.
12.1 TRAFFICCHARACTERIZATION
Because of the ratrdom nature of network
traffic, the following analyses involve
cer-
tain fundamentals
of probability
theory and srochastic pro""r*r*.
L tni, heatment
only the most basic assumptions
and results of traffic analysis are presented.
The in-
tent is to provide
an indication
ofhow to apply results ofhaffic analysis, not to delve
deeply into analytical formulations.
However,
a few basic derivations
are presented
to
acquaint the user with assumptions in the models so they can be appropriately
applied.
In the realm of applied mathematics,
where these subjects are nlatea more for-
mally, blocking probability
analyses
are referred to as congestion
theory and delay
analyses are refened to as queuing
theory. These topics are also commonly
referred
to as traffic flow analysis. In a circuit-switched
network,
the
..flow"
of messages is not
so much of a concern
as are the holding times of cornmon
equipment. A circuit-
switched network
establishes
an end-to-end circuit involving
various network facili-
ties (transmission
links and switching
$tages) that are held for the duration of a ca1l.
From
a network point
of view, it is the holding of these re$ources
that is important,
not the flow of information
within individual
circuits.
on the other hand, message-switching
and packet-switching
networks
are directly
concerned with the actual flow of information,
since in these systems traff,rc on the
transmission links is directly related
to the activity of the sources.
As mentioned
in Chapter 7, circuit switching does involve certain aspects of traffic
flow in the process
of setting up a connection.
connect requests flow from the sources
to the destinations acquiring,
holding, and releasing
certain resources in the process.
As was discussed, controlling
the flow of connect requests
during network overloads
is a vital function of network management.
The unpredictable
nature of communications
traffic
arises as a result of two under-
lying random processes:
call arrivals and holding times. An arrival from any particular
user is generally
assumed to occur purely
by chance and be totally independent
ofar_
rivals from other users. Thus the number of arrivals during any particular
time interval
is indeterminate.
In most cases holding times are also distributed randomly.
In some
applications
this element of randomness
can be removed
by assuming constant hold-
tz.t rBAFFtc cHAHAcrEFlzATloN
521
ing times
(e.g., fixedJength packets). In either case the traffic load presented to a net-
work is fundamentally dependent on both the frequency of arrivals and the average
holding time for each arrival. Figure 12.1 depicts a representative situation
in which
both the arivals and the holding times of 20 different sources are unpredictable. The
bottom of the figure depicts activity of each individual source while the top displays
the instantaneous total of all activity. [f we assume that the 20 sources are to be con-
nected to a hunk group, the activity curve displays the number of circuits in use at any
particular time. Notice that the maximum number of circuits in use at any one time is
16 and the average utilization is a little under I I circuits. In general terms, the trunks
are referred to as servers, and a trunk group is a serYer group'
Traffic Measurementg
One measure of network capacity
is the volume of traffrc carried over a period of time.
Traffic volume is essentially
the sum of all holding times carried during the interval.
The traffic volume represented in Figure 12.1 is the area under the activity curve (ap-
proximately 84 call minutes).
A more useful measure of traffic is the traffic intensity
(also
called traffic flow).
Traffic intensity is obtained by dividing the traffic volume by the Iength of time during
which it is measured. Thus traffic intensity represents the average activity during ape-
riod of time
(10.5
in Figure 12. 1). Although traffic intensity is fundamentally dimen-
sionless
(time divided by time), it is usually
expressed in units of erlangs,
after the
Danish pioneer tlaffic theorist A. K. Erlang, or in terms of hundred
(century) call sec-
onds
per hour
(CCS).
The relationship between
erlangs and CCS units can be derived
by observing that there are 3600 sec in an hour:
Es
I
E.E
! s
E E
t
Tims lfiinu$rl
Flgure 12.1 Activity profile of network traffic
(all
calls canied).
522 TRAFFIC ANALYSIS
I erlang
=
36 CCS
The maximum
capacity of a single server (channel)
is l erlang, which is to say that
the server is always busy. Thus the maximum capacity in erlangs of a group
of servers
is merely equal to the number of seryers. Because
traffic in a loss sysLm experiences
infinite blocking probabilities
when the traffic intensity is equal to the number of serv-
ers, the average activity is necessarily
less thal the number of servers. similarly, delay
rrystems
operate at less than full capacity, on average, because infinite delays occur
when the average load approaches
the number of servers.
-
Two important palameters
used to characterize maffic are the average alival rate
L and the average holding time t*. If the traffic intensity l, is expressed in erlangs, then
a
=fu*
( 12. 1)
where I' and f* are expressed in like units of time (e.g.,
calls per
second and seconds
per
call, respectively).
Notice that traffic intensity
is only a measure of average utilization
during a time
period
and does not reflect the relationship
between arrivals and holding times. That
is, many short calls can praduce
the same traffic intensity as a few long ones. In many
of the analyses that follow
the results are dependent only on the trathc intensity. In
some ca|tes, however,
the results are also dependent
on the individual
arrival pattems
and holding time distributions.
Public
telephone networks
are typically analyzed in terms of the average activity
during the busiest hour of a day. The use of busy-hour traffic measurements
to design
and aralyze telephone networks representrr
a compromise
between designing for tle
overall average
utilization (which
includes virtually unused nighnime
hours) and de-
signing for short-duration
peaks
that may occur by chance o, *, u resurt of rv com-
mercial breaks, radio call-in conte$ts, and so on.
Busy-hour traffic measurements
indicate that an individual
residential telephone is
typically in use between 5 and l0zo ofthe busy hour. Thus each telephone represenm
a traffic load of between 0.05 and 0.l0 errangs. The average holding time is between
3 and 4 min, indicating
thar a typicar telephone is involved
in one oi two phone
calls
during the busy hour.
Business
telephones usually produce
loading patterns
different from residential
phones'
First, a business phone
is generally
utilized more heavily. second, the busy
hour ofbusiness
traffic is often different from the busy hour ofresidential
traffic. Fig-
ute 12.2 shows a typical hourly variation for both sources of traffic. The trunks of a
telephone network
are sometimes
designed to take advantage of variations in calling
patterns
from different
offices. Toll connecting
trunks from residential
areas are often
busiest during evening hours, and trunks from business areas are obviously
busiest
during midmorning
or midafternoon.
Traffic engineering
depends not only on overall
traffic volume
but also on time-volume
traffic patterns
within the network.
A certain afirount of care must be exercised
when determining the total traffic load
of a system from the loading of individual
lines or trunks. For example, since two tele-
12.1 TRAFFICCHARACTEFIEAT|0N
523
r i l l Ar ' r 2 3 .
t 6 t t t l 0 l l l 2H I t 3 4
I 6 t I
ftuq ol dof
I
oti(|;uomrdi*ia
t
ofi* . *.ll.,ttot or'i"
Figure 12.2 Traffic volume dependence on time of day.
phones are involved in each connection, the total load on a switching system is exactly
one-half the total of all traffic on the lines connected to the switch. ln addition,
it may
be important to include certain setup and rclease times into the average
holding times
of some common equipment. A lQ-sec setup time is not particularly significant for a
4-min voice call but can actually
dominate the holding time of equipment used for
short data messages. Common equipment setup times also become more significant
in the presence ofvoice traffic overloads. A greaterpercentage ofthe overall load is
represented by call attempt$
since they increase at a faster rate than completions.
An important distinction
to be made when discussing traffic in a communications
network is the difference between the offered haffic and the carried traffic. The of-
fered trffic is the total traffic that would be carried by a network capable of servicing
all requests as they arise. Since economics
generally precludes designing
a network to
immediately cany the maximum
offered traffic, a small percentage of offered traffic
typically experiences network blocking
or delay. When the blocked calls are rejected
by the network, the mode of operation
is referred to as blocked calls cleared or lost
calls cleared. In essence, blocked calls are assumed to disappear and never return. This
assumption is most appropriate for trunk
Broup$
with alternate
routert. In this case a
blocked call is normally serviced
by another tnrnk
group and does not, in fact, retum.
The carried traffic ofa loss system is always less than the offered traffic. A delay
system, on the other hand, does not reject blocked
calls but holds them until the nec-
essary facitities are available. With the assumption that the long-term average of of-
fered traffic is less than the capacity of the network,
a delay system caries all offered
:
I
F
t
o
524 TRAFFIc ANALYSIS
ffaffic. If the number
of request$ that can be waiting for service is limited,
however, a
delay system also takes on properties
of a ross system. For exampre, if the queue
for
holding
blocked arrivals is finite, requests arriving when the queue
is full are cleared.
12.1.1 Arri val Di strl buti one
The most fundamental
assumption of classical traffic analysis is that call arivals are
independent.
That is, an arrivar from one source is unrelated to an arrival from any
other source. Even though this assumption
may be invalid in some instances,
it has
general
usefulness for most applications.
In those cases where call arrivals tend to be
correlated, useful results can still be obtained by modifying
a random arrival analysis.
ln this manner the random
arrival assumption provides
Jmathematical
formulation
that.catr
be adjusted to produce
approximate
solutions to problems
that are otherwise
mathematically
intractable.
Negatlva
Exponenilal
lnterarrlval
Timeg
Designate
the average call arrival rate from a rarge group
of independent
sources (sub-
scriber lines) as 1,, Use the following
assumptions:
l. Only one anival can occur in any sufficiently
small interval.
2. The probability
of an arrival in any sufficiently
small iriterval is directly
proportional
to the length of the interval. (The probability
of an arrival is l" lr,
where Ar is the interval length.)
3. The probability
of an arrival in any particular
interval is independent
of what
has occurred
in other intervals.
It is straightforward
[1]
to show that rhe probability
disnibution of interarrival times
is
P()(I"t)
=
e-M (r2. 2)
Equation l?'2 defines the probabiliry
that no arrivals occur in a randomly
selected in-
terval t' This is identical
to the probability
that f seconds elapse from one arrival to the
next.
Example 12.1.
Assuming each of 10,000
subscriber lines originate
one cail per
hour, how often do two calls arrive with less than 0.01 sec between themt
Solutinn.
The average anival rate is
l,
=
3600/10,000 =
2.78 arrivals/sec
From EquationlZ.z,the
probability
of no arrival in a O.0r-sec interval is
12.1 TRAFFICCHARACTERIZATION 525
PoQ'0215): e4'o?78
=
0.973
'lhus
2.7 Vo of the arrivats occur within 0.01 sec of the
previous arrival. Since the arri-
val rate is ?.78 arrivats per second, the rate ofoccurrence ofinteranival times less than
0.01 sec is
2.78 x0.027
=
0.075 times/sec
The flrst two assumptions made in deriving the negative exponential arrival distri-
bution can be intuitively
justified
for most applications. The third assumptron, how-
ever, implies certain aspects of the sources that cannot always be supported. First,
certain events, Such as televiSion
commercial breaks, might stimulate the sources to
place their calls at nearly the same time. In this case the negative exponential distri-
bution may still hold but for a much higher calling rate during the commercial.
A more subtle implication of the independent arrival assumption involves the num-
ber of sources, not
just
their calling
pattern$. When the
probability of an anival in any
small time interval is independent of other arrivals, it implies that the number of
sources available to
generate requests is constant. If a number of arrivals occur imme-
diately before any subinterval in
question, some of the sources become busy and can-
not
generate requests. The effect of busy sources is to reduce the average arrival rate.
Thus the interarriva'l
times are always somewhat larger than what Equation 12.2 pre-
dicts them to be. The only time the arrival rate is truly independent of source activity
is when an infinite number of sources
exist.
If the number of sources is large and their average
activity is relatively
low, busy
$ources do not appreciably
reduce the arrival rate. For example, consider an end office
that services 10,000 subscribers with 0.1 erlang of activity each. Normally,
there are
1000 active links and 9000 subscribers available to generate new artivals. Ifthe num-
ber of active subscribers increa$es by an unlikely 507o to 1500 active lines, the number
of idle subscribers reduces to 85fi), achange of only 5.6Vo. Thus the arrival rate is rela-
tively constant over a wide range of source activity.
Whenever the arrival
rate is fairly
constant for the entire range of normal source activity, an infinite $ource assumptron
is
justified.
Actually, some effects of finite $ources
have already been discussed
in Chapter 5
when analyzing blocking
probabilities of a switch. It is pointed out that Lee graph
analyses overestimate the blocking
probability because, if some number of interstage
links in a group are known to be busy, the remaining
links in the group are less likely
to be busy. A Jacobaeus analysis
produces a more rigorous and accurate solution to
the blocking
probability, particularly when space expansion
is used. Accurate analy-
ses ofinterarrival times for finite sources are also possible. These are included
in the
blocking analyses to follow.
Poisson Arrlval Dlstributlon
Equation 1.2.2merely
provides a means of determining the distribution of interarrival
times. It does not, by itself,
provide the
generally more desirable information of how
526 TRAFFIC ANALYSIS
many arrivals can be expected to occur in some arbitrary time interval.
using the same
assumptions presented,
however, the probability
ofj arrivals in an interval
, can be de_
termined
[]
as
Equation 12.3 is the well-known
poisson
probability
law. Notice that whenj =
0, the
probability
ofno arrivals in an interval r is
ps(r),
as obtained in Equation
12.2.
Again, Equation
12.3 assumes
arrivals are independent
and occur at a given
aver-
age rate 1,, irrespective
of the number
of arrivals oicuning just
prior
to an interval in
question'
Tfius the Poisson probability
distribution
strould onty be used for arrivals
from a large number
of independent
sources.
Equation l?.3 defines the probability
of experiencing exactlyj
ardvals in, seconds.
Usually there is more interest in determining
the probability
ofj or more arrivals in r
seconds;
r,{M)
=p
r-u
p=Itu)
=Er,fUt
ej
./:l
:
r
-Er,flrt
r'{
=
I
_p*lftf)
where P;(l,r) is defined in Equation 12.3.
Example
12.2. Given a message-swirching
node that
arrivals per minure,
what is the probability
that eight or
arbitrarily chosen 30-sec interval?
(12. 3)
(r2.4)
normally experiences
four
more arrivals occur in an
solution, The average number of arrivals in a 30-sec interval is
The probability
of eight or more arrivals (when
the average is 2) is
P>8(2)
=
i
",,r,
r=8
Lr =a x#=?
I2.1 TRAFFIC CHARACTERIZATION 527
7
=t - Lr , ( z)
r{
Exampte 12.3. What is the probability that a 1000-bit data block experiences
exactly four errors while being transmitted over a transmission link with a
lit
enor rate
(BER)
of l0-5?
Solation, Assuming independent errors
(a questionable assumption on many
transmission links), we can obtain the probability of exactly four errors directly from
the Poisson distribution. The average number of errors
(arrivals) lut
=
101 x tO-s
=
0.01. Thus
=r - a{ r
=
0.0011
prob(4errors)
=
P4(0.01)
=
ry
e-{'or
-
4. 125 x 16-
10
prob(4errors)
:
[t
oo*Jtt
t
-
p)ee6
* { )
" ' ' r!
)
An alternative solution can be obtained
from the binomial probability law:
=
4. I 01 x 10-10 wherep
=
lfl-s
As can be seen, the two solutions of Example 12.3 are nearly identical.
The close-
ness of the two answers reflects the fact that the Poisson
probability distribution is
often derived as a limiting case of a binomial
probability distribution. Because it is
easier to calculate, a Poisson distribution is often used as an approximation to a bino-
mial distribution'
12.1-z Holdlng Time Distrlbutions
The second factor offfaffic intensity as specified
in Equation 12.1 is the average
hold-
ing time tr-. In some cases the average of the holding times is all that needs to be known
about holding times to determine
blocking probabilities in a loss system or delays in
a delay system.
In other cases it is necessary to know the probability disribution of
the holding times to obtain the desired
results. This section describes the two most
commonly assumed holding time disnibutions: constant holding times and exponen-
tial holding times.
528 rRAFFtc ANALysts
Constant Holdlng Times
Although constant holding times cannot be assumed for conventional
voice conver$a-
tions, it is a reasonable
assumption for such activities as per-call
call processing
re-
quirements,
interoffice
address signaling, operator assistance, and recorded -*ssag"
playback'
Furthermore,
constant holding times are obviously valid for transmission
times in fixed-length packet
networks.
when constant holding
time messages are in effect, it is straightforward
to use
Equation 12.3 to determine the probability
distribution
of active channels. As$ume,
for the time being, that all requests are serviced. Then the probability
ofj channels be_
ing busy at any particular
time is merely the probability
that; arrivals occurred
in the
time interval
of length r* immediately preceding
the instant in question.
since the av-
erage number of active circuits over all time is the traffic intensity A
=
l,f*, the prob-
ability ofj circuits being busy is dependent only on the traffic intlnsity:
P;(i.r-)
=
Pj(A)
(12. 5)
where l,
=
arrival rate
fm
=
constflrlt holding time
A
=
traffic inten$ity (erlangs)
Exponentlal
Holding Tlmes
The most commonly
assumed holding time distribution for conventional
telephone
conversations
is the exponential holding
time distribution:
P(>t)
=
s-t/t^
(12. 6)
N - a
=7"
yhere
r* is the average holding time. Equation 12.6 specifies the probability
that a
holding
time exceeds the value r. This relationship
can-be derived fro* u few simple
assumptions concerning
the nature of the call termination process.
Its basic
justifica-
tion, however, lies in the fact that observations
ofactual uoice "onuersations
exhibit
a remarkably
close conespondence
to an exponential
distribution,
The exponential
distribution possesses
the curious property
that the probability
of
a termination is independent
of how long a call has teen in progress.
Thai is, no matter
how long a call has been in existence,
the probabiliry
ofit Lsting another r seconds is
defined by Equation 12'6. In this sense exponential holding times represent the most
random process possible.
Not even knowledge
ofhow long a call hasieen in progress
provides
any information
as to when the call will terminaie.
combining
a Poisson arrival process
with an exponential
holding time process
to
obtain the probability
disribution
of active circuitJ is more compilated
than it was
for constant holding times because calls can last indefinitely.
rhe final result, how-
ever' proves
to be dependent on only the average holding time. Thus Equation 12.5 is
12.I TRAFFIC CHARACTERIZATION 529
valid for exponential holding times as well as for constant holding times
(or any hold-
ing time distribution). Equation 12.5 is therefore repeated for emphasis; The prob-
ability of
j
circuits being busy at any
palticular instant, assuming a Poisson arrival
process and that all requests are serviced immediately, is
(r2.1)
where A is ttre traffic inten$ity in erlangs. This result is true for any distribution of
holding times.
Example 12.4. Assume that a tnrnk
group has enough channels to immediately
carry all of the traffic offered to it by a Poisson process with an arrival rate of one call
per minute. Assume that the average holding time is 2 min. What
percentage of the
total traffic is carried by the first five circuits, and how much traffic is carried by all
remaining circuits?
(Assume that the haffic is always
packed into the lowest
numbered circuits.)
Sol uti nn. Thetraffi ci ntensi ty(offeredl oad)of
thesystemi sA=
I x2=2erl angs.
The kaffic intensity carried by I active circuits is exactly i erlangs.
Hence the traffic carried by the fust five circuits can be determined as follows;
As
=
1Pr(2) +2Pr(2) + 3Pr(2) + 4Pa(2) + 5Pr(?)
PIA)=fte-A
^ . Z x z ? . 3 x z 3 . 4 x z a ,
s x z s
)
z+
zl
*- 3i _*- T-
st
)
= e
=
l.g9 erlangs
All of the remaining circuits carry
2- 1. 89=0. l l er l ang
The result of Example 12.4 demonstrates the
principle of diminishing returns as the
capacity ofa system is increased to carry
greater and greater percentage$ ofthe offered
traffic. The first five circuits in Example 12.4 carry 94.57o of the traffic while all re-
maining circuits carry only 5.5Vo of the ftaffic. If there are 100 sources, 95 exka cir-
cuits are needed to carry
the 5.5Vo.
12.2 LOSS SYSTEMS
Example l2'4 provides
an indication
of the blocking probabilities
that arise when the
number of servers (circuits)
is less than the maximum possible
traffic load
(number
of
sources). The example demonstrates
that
g4.5%o
of the traffic is carried by only five
circuits' The implication is that the blocking probability,
if only five circuits are avail-
able to carry the traffic, is 5.svo. Actually, Example 12.4 is carefully worded to indi_
cate that all ofthe offered traffic is carried but that only the traffic carried by the first
five circuits is of interest. There is a subtle but important
distinction between the prob-
ability that six or more circuits are busy (as
can be obtained from Equation 12.7) and
the blocking probabiliry
that arises when only five circuirs exisr.
The basic reason for
the discrepancy is indicatecl in Figure 12.3, which depicts the
same traffic pattern
arising from z0 sources as is shown previously
in Figure tz. t. nlg-
ure l2'3, however, assumes that only l3 circuits are available to carry ttre traffic. thus
the three arrivals at t
=
2.2,2.3,
and 2.4 min are blocked and assumed to have left the
system. The total amount of traffic volume lost is indicated by the shaded area, which
is the difference between
all traffic being serviced as it arriverand
traffic being carried
by a blocked calls cleared system with l3 circuits. The most important feature to no-
tice in Figure I2.3 is that the call arriving at r
=
z.g is not blocied, even though the
original profile
indicates that it arrives when all 13 circuits are busy. The reason it is
not blocked is that the previously
blocked calls left the system and therefore reduced
the congestion for subsequent
arrivals. Hence the percentage
of time that the original
traffic prof,rle
is at or above 13 is not the $ame as the blocking probability
when only
13 circuits are available.
530 TRAFFIC ANALYSIS
Tim6
(rninutEl
Figure 12.3 Activity profile
of blocked calls cleared
(
I 3 channels).
gs
E. E
E f i
. t E
P E
= E
G
12.2 LOSS SYSTEMS 531
12.2.1 Loet Calls Cleared
The first
person
to account fully and accurately for the effect of cleared calls in the
calculation of blocking
probabilities was A. K. Erlang in 1917. In this section we
discuss Erlang's most often used result; his formulation of the blocking prob-
ability for a lost calls cleared system with Poisson arrivals. Recall that the Poisson
arrival assumption implies infinite sources.
This result is variously referred to as
Erlang's formula of the first kind, El,y(A); the Erlang-B formula; or Erlang's loss
formula.
A fundamental aspect of Erlang's formulation, and a key contribution to modern
stochastic process theory, is the concept of statistical equilibrium. Basically, statistical
equilibriunr implies that the probability of a system's being in a particular state
(num-
ber of busy circuits in a trunk group) is independent of the time at which the system
is examined. For a system to be in statistical equilibrium, a long time must
pass (sev-
eral average holding times) from when the $ystem is in a known state until it is again
examined. For example, when a trunk
group first begins to accept traffic, it has no busy
circuits. For a short time thereafter, the system is most likely to have only a few busy
circuits. As time
passes, however, the system reaches equilibrium. At this
point the
most likely state of the system is to have A
=
I'rr busy circuits'
When in equilibrium, a $ystem
is as likely to have an arrival as it is to have a ter-
mination. If the number of active circuits happens to increase above the average A, de'
partures become more likely than arrivals. Similarly, if the number of active circuits
happens to drop below A, an arival is more likely than a depadure. Thus if a system
is
perturbed
by chance from its average state, it tends to return.
Although Erlang's elegant formulation
is not
particularly complicated, it is not
pre-
sented here because we are mostly interested in application of the results. The inter-
ested reader is invited to see reference
[2]
or
[3]
for a derivation ofthe result:
B=Erl,A)=
Mffi
(12. 8)
where N
=
number of servers (channels)
,4
=
offered traffic intensity, \,t^
(erlangs)
Equation
12.8 specifies the probability of blocking for a system with random arri-
vals from an infinite source and arbitrary holding time distributions. The blocking
probability of Equation 12.8 is plotted in Figure 12.4 as a function of offered traffic
intensity for various numbers of channels. An often more usefUl presentation of Er-
lang's results is provided in Figure 12.5, which present$ the output channel utilization
for various blocking
probabilities and numbers of servers. The output utilization
p rep-
resents the traffic carried bv each circuiu
j
h
g
s - g
T U
F @
E =
= ! t
t ;
E . E
6 H
t b
? =
E
J X
E H
o t r
F o .
E b o
L ' t r
.it
X
q
cl
P
HO
tl
g
^rilHrqod
0sHoolg
\
I \ \
l \ \
\
\
\ \\
\ \'
\
\ \ \
\ \ \
I
\
I
\ \
\ \ \
\
\
\
\
\
\
t
I
rr
JT
(\
\
a '
_ Q
aj
Jt
\
i\
$ \
l e
\
\
\
\
\
\ \
\
\ I
\ \
\
\ \
\ \
\ \
\
\ \
\
\
,r\
\
\
\
o ( ' r e h q u l r t t ' ? t ! - 1 o
j t c t o o o o o o o
i
q
h
8 r i
d H
u
E
q
U ' =
o i i
C !
t r H
r E ; =
t t r
( J o
d
$ E
a , H
L ! t
rlr
.5
. a =
E E
f t r t
E H
v t
14. E
' ( J

*
O E
i o
U?
ol
Fi
P
b0
l - ,
(uotterilrrn
tauueqc lndtno)
d
TRAFFIC ANALYSIS
( 1- B) 4
P=- F
(12.e)
whereA
N
B
(l
-B)A
offered traffic
number of channels
blocking probability
carried traffic
Blocking probabilities
are also provided
in tabular form in Appendix D.
Example
12.5. A Tl line is to be used as a tieline trunk group
berween two
pBXs.
How much haffic can the
trunk group
cany if the blocking ptouuuitity
is to be 0.1?
What is the offered traffic intensitv?
Soluti'on- From Figure 12.5 it can be seen that the output circuit utilization
for
B
=
0.I and N
=
24 is 0.8. Thus the carried traffic intensity is 0.g x24 =
19.2 erlangs.
since the blocking probability
is 0. l, the maximum level of offered traffic is
l q ?
A=ffi =2l .3erl angs
Example 12.6. Four clusters of data terminals are to be connected to a computerby
way of leased circuits, as shown in Figure 12.6. In Figure lz.6athetraffic
from the
clusters uses separate group$
of shared circuits. In Figure lz.6b the traffic from all
clusters is concentrated
onto one common group
of circuits. Determine
the total
number of circuits required in both cases when the maximum
desired blocking
probability
is 57o, Assume that 22 terminals are in each cluster and each terminal is
active llVo of the time. (Use
a blocked
calls cleared analysis.)
Figure 12.6 Data terminal network of Example 10.6: (a)
four separate groups; (b)
all traffic
concenfrated
into one group.
1a.a Loss sYsrEMS 535
Solution. The offered traffic from each cluster is22x0.l
=2.2
erlangs. Since the
average number of active circuits is much smaller than the number of sources, an
infinite source analysis can be used. Using Table D.l, the number of circuits required
for B
=
SVo at a loading of 2.2 erlangs is 5. Thus the configuration of Figure 12.54
requires a total of 20 circuits.
The total offered traffic to the concentrator of the configuration of Figure 12.6b is
4 x2.2
=
8.8 erlangs. From Table D.1, l3 circuits are required to support the
given
traffic load.
Example 12.6 demonstrates that consolidation
of small traffic groups into one large
traffic group can
provide
significant savings in total circuit requirements. Large
groups
are more efficient than multiple small
groups because it is unlikely that the
small groups will become overloaded at the same time
(assuming independent arri-
vals). In effect, excess traffic in one
group can use idle circuits in another
group. Thus
those circuits that are needed to accommodafe traffic peaks but are normally idle are
utilized more efficiently when the,traffic is combined into one
group. This feature is
one of the motivations mentioned in Chapter 10 for integrating voice and data traffic
into a common network. The total savings in ffansmission co$ts is most significant
when the individual traffic intensities are low. Hence it is the peripheral area of a net-
work that benefits the most by concentrating the traffic.
The greater circuit efficiency obtained by combining txaffic into large groups is
often referred to as the advantage of large
group
sizes. This efficiency of circuit utili-
zation is the basic motivation forhierarchical switching structures. Instead of intercon-
necting a large number of nodes with rather small trunk
groups between each pair, it
is more economical to combine all traffic from individual nodes into one large trunk
group
and route the traffic through a tandem switching node. Figure 12.7 contrasts a
mesh versus a star network with a centralized switching node at the center. Obviously,
the cost of the tandem switch becomes
justified
when the savings in total circuit miles
is large enough.
(a)
ft)
Flgure 12.7 Use of tandem switching to concentrate ffaffic:
(a)
mesh;
(b)
star.
536 THAFFIC ANALYSIS
Example 12.7. what happens ro rhe blocking probabiliries
in Figure rz.6a and b
discussed in Example 12.6 when the traffic intensity increase$ by 50va?
soluti.on. If the traffrc intensity of each group
increases from z.zto3.3erlangs,
the
blocking probability
of the configuration of Figure 12.6a increaserr from 5Zo to almost
l 4Va.
In the configuration of Figure 12.6b a 507o increase in the traffic intensitv
causes
a 400Vo increase in the blocking probability (from
5 to ZOVo).
Example 12.7 demonstrate$
some important consideration$
in network design. As
indicated,
blocking probabilities
are very sensitive to increases in traffic intensities,
particularly
when the channels are heavily utilized. Because large trunk groups
utilize
their channels more efficiently, they are more vulnerable to traffic increases than are
a number of smaller groups
designed to provide
the same grade
of service. Further-
more, failures of equal percentages
of transmission capacity affect the performance
of
a large group
more than the performarce
of several small groups.
In both cases the vul-
nerability of the large group$
arises because large groups
operate with less spare ca-
pacity
than do multiple small groups.
A second aspect of blocking analyses demonstrated in Example l Z.7 is that the cal-
culated results are highly dependent on the accuracy ofthe traffic intensities. Accurate
values of traffic intensities
are not always available. Furthermore, even when accurate
traffic measurements are obtainable, they do not provide
an absolute indication of how
much growth
to expect. Thus only limited confidence can be attached to calculations
of blocking probabilities
in an absolute sense. The main value of these analyses is that
they provide
an objective means of comparing various network sizes and configura-
tions' The most cost-effective design for a given grade
of service is the one that should
be chosen, even if the traffic statistics are hypothetical.
If a network is liable to expe-
rience wildly varying traffic patterns
orrapid growth,
these factors mustbe considered
when comparing design alternatives. A network with a somewhat larger initial cost
may be more desirable if it can absorb or grow to accommodate unanticipated
traffic
volumes more easily.
12.2.2 Lost Catts Returning
In the lost calls cleared analyses
just
presented,
it is assumed that unserviceable re-
quests
leave the system and never retum. As mentioned, this assumption is most ap-
propriate
for trunk groups
whose blocked requests overflow to another route and are
usually serviced elsewhere. However, Iost calls cleared analyses are also used in in-
$tances where blocked calls do not get
serviced elsewhere. In many of these cases,
blocked calls tend to return to the system in the form of retries. some examples are
subscriber concenhator
systems, corporate tie lines and
pBX
trunks, calls to busy tele-
phone
numbers, and access to WATS lines (if
DDD altematives
are not used). This
1z.z Loss sYsrEMS 537
section derives blocking
probability relationships for lost calls cleared systems with
random retries.
The following analysis involves three firndamental assumptions regarding the na-
ture of the retuming calls:
l. All blocked calls return to the system and eventually
get
serviced, even if
multiple retries are required.
The elapsed times between call blocking occurrences and the generation of
retries are random and statistically independent ofeach other.
(This
assumption
allows the analysis to avoid complications arising when retries are correlated to
each other and tend to cause recurring traffic
peaks
at a
particular waiting time
interval.)
The typical waiting time before retries occur is somewhat
Ionger than the
average holding time of a connection. This assumption essentially states that the
$ystem is allowed to reach statistical equilibrium before a retry occurs.
Obviously, ifretries occur too soon, they are very likely to encounter congestion
since the system has not had a chance to
"relax."
In the limit, if all retries are
immediate and continuous, the network operation becomes similar to a delay
system discussed in later sections of thi$ chapter. In this case, however, the
system does not queue tequests-the sourcerl do so by continually
"redialing."
When considered in their entirety, these assumptions characterize retries as being
statistically indistinguishable from first-attempt traffic.* Hence blocked calls merely
add to the first-attempt call arrival rate.
Consider a system with a first-attempt call arrival rate of 1,. If a percentage B of the
calls is blocked, B times L retries will occur in the future. Of these retries, however, a
percentage
I will be blocked again. Continuing in this manner, the total arival rate
l,i after the system has reached statistical equilibrium can be determined as the infinite
series
l,'
=
l, +Bl,+ R7"+ B37r- . . .
L
(12. 10)
I - B
where B is the blocking probability from a lost calls cleared analysis with traffic in-
ten'tity A'
=
l,'fm.
Equation 1 2. 10 relates the average arrival rate l,/, including the retries, to the first-
attempt arrival rate and the blocking probability in terms of l,'. Thus this relationship
does not provide a direct means of determining l,' or d since each is expressed in terms
of the other. However, the desired result can be obtained by iterating the lost calls
*First-attempt
traffic is also referred to as demand traffic: the service demands assuming all arrivals are
serviced immediately. The offered haffrc is the demand traffic plus the refries,
538 THAFFIc ANALysts
cleared analysis of Equation 12.8. First, determine an estimate of B using L and then
calculate l,'. Next, use l,' to obtain a new value of B ancl an updated valui of 1,,. con-
tinue in this mamer until values of l,' and B are obtainecl.
Example 12.8. what is the blocking probability
of a
pBX
to a central office trunk
group
with I 0 circuits servicing a first-attempt
offered traffic load of 7 erlangs? What
is the blocking probability
if the number of circuits is increased to 13? Assume random
retries for all blocked calls.
solution. It can be assumed
that the 7 erlangs of haffic arise from a large number
of PBX stations. Thus an infinite source analysis is
justified.
The blocking probability
forA
=
7 erlangs and N= l0 servers is about 87a. Thus the total offered load, including
retries, is approximately 7.6 erlangs. with N
=
l0 and,4'= 7.6, the blocking
probability
is llvo. Two more iterations effectively produce
convergence atA,
=
g
erlangs and B
=
lTvo.rf the number of circuits in the trunk group is increased to 13,
the blocking probability
of a lost calls cleared system is 1.57o. Thus a first approximation
to the retuming traffic intensity is 7/0.985
=
7.1 erlangs. Hence the blocking probability
including all retuming traffic increases only slightly above the l.5Zo.
Example l2'8 demonstrates that the effect of retuming traffic is insignificant
when
operating at low blocking probabilities.
At high blocking probabilities,
however, ir is
necessary to incorporate
the efTects ofthe returning traffic into the analysis. This re-
lationship
between lost calls cleare{ and lost calls retuming is shown in Figure 12.g.
when measurements
are made to determine the blocking probability
of an outgo-
ing trunk group,
the measurements
canxot distinguish between first-attempt
calls (de-
mand traffic) and retries. Thus if a significant number of retries are contained in the
measurements,
this fact should be incorporated into an analysis of how many circuits
must be added to reduce
the blocking of an overloaded trunk group.
The apparent of-
fered load will decrease as the number of servers increases becausl the number ofre-
u. I tt,z u.ir u,+ u.[ o.o 0,7 0,9
Of|rttd rilftTo Inrfirlry
Frr
ctunrut
(ritfigll
Figure 12.8 Blocking probability
of lost calls reruming.
t.0
.5
E
.100
E
.080
{
E .010
.F
.om
J
t
E
.oot
0J 0.,0 0.[ 0.8 0.7
Of|rttd tilffh Inrfirlry
Frrctunrd
(ritfigll
12,2 LoSS SYSTEMS 539
tries decreases. Thus fewer additional circuits are needed than if no retries are con-
tained in the measurements.
12.2.3 Lost Galls Held
In a lost calls held system, blocked calls are held by the system and serviced when the
necesrrary facilities become available. Lost calls held systems are distinctly different
from the delay systems discussed later in one important respect; The total elapsed time
of a call in the system, including waiting time and service time, is independent of the
waiting time. In essence, each arrival requires service for a continuous period of time
and terminates its request independently of its being serviced or not. Figure 12.9 dem-
onstrates the basic operation of a lost calls held system. Notice that most blocked calls
eventually get some service, but only for a portion of the time that the respective
sources are busy.
Although a switched telephone network does not operate in a lost calls held man-
ner, some system$ do. Lost calls held systems generally
arise in real-time applications
in which the sources are continuously in need of service, whether or not the facilities
are available. When operating under conditions of heavy ftaffic, a lost calls held system
typically provides
seryice for only a
portion
of the time a particular source is active.
Even though conventional circuit switching does not operate according to the theo-
retical model of lost calls held, Bell System traffic engineers have used it to calculate
blocking probabilities for trunk groups
[4].
A lost calls held analysis always produces
a larger value for blocking than does Erlang's loss formula. Thus the lost calls held
analysis produces
a conservative design that helps account for retries and day-to-day
E , :
E s
E
E G
- g
Figure 12,9 Activity profile
sf lost calls held,
540 THAFFIC ANALYSIS
variations in the busy-hour calling intensities. In contrast, CCITT recommendations
[5]
stipulate Erlang-B formulas should be used in determining blocking probabilities.
One example of a system that closely fits the lost calls held model is time assign-
ment speech interpolation (TASD.
A TASI sy$tem concentrates some number of voice
$ources onto a smaller number of transmission channels. A source receives service (is
connected to a channel) only when it is active. If a source becomes active when all
channels are busy, it i$ blocked ard speech clipping occurs. Each speech segment
starts and stops independenfly
of whether it is serviced or not. TASI systems were
originally used on analog long-distance transmission links such as undersea cables.
More modern counterparts of TASI are referred to as digital circuit multiplication
(DCM)
systems. [n contrast to the original rASI systems, DCM systems can delay
speech for a small amount of time, when necessary, to minimize the clipping. In this
case, a lost calls held analysis is not rigorously
justified
because the total time a speech
segment is
"in
the sy$tem" increase$ as the delay for service increases. However, if
the average delay is a small percentage
of the holding time, or if the coding rate of de-
layed speech is reduced to allow the transmission channel time to
"catch
up," a lost
calls held analysis is stilljustified. Recall that controlling the coding rate is one tech-
nique of traff,rc shaping used for transporting voice in an ATM network,
Lost calls held systems are easily analyzed to determine the probability
of the total
number of calls in the system at any one time. Since the duration of a source's activity
is independent of whether it is being serviced, the number in the system at any time is
identical to the number of active sources in a system capable of carrying all traffic as
it arises. Thus the distribution of the number in the system is the Poisson distribution
provided
earlier in Equationl2.3. The probability
that i sources requesting service are
being blocked is simply the probability
that i + N sources are active when N is the num-
ber of servers. Recall that the Poisson distribution essentially determines the desired
probability
as the probability
that i + N arrivals occurred in the preceding
f. seconds.
The distribution is dependent only on the product
of the average arrival rate l" and the
average holding time tm.
Example 12.9. what is the probability
that a talk$purt experiences clipping in a
TASI system with l0 sourceri and 5 channels? with 100 sources and 50 channels?
Assume that the activity factor of each talker is 0.4. (Ignore
finite sources..l
solution, For the first case, the clipping probability
can be determined as the
probability
that five or more sources are busy in a
poisson
process
with an average of
A
=
O.4 x l0
=
4 busy servers. Using Equati on 12.7,
,
/ , ,
prob(crippingl :
f,
r,r+)
=
"*l
#* #
*
+.#.fr l:
o.ru
Fi
l , " '
)
With 100 sources, the average number of busy circuits is,4
=
0.4 x
100: 40. A speech
segment is clipped if 50 or more talkers are active at once. Thus the clipping prob-
ability can be determined as
1z.a LosssysTEMs 541
99
prob(clippingl
=
f,
rr{+O)
=
0,04
'Fso
Example 12.9 demonshates that TASI system$ are much more effective for large
group sizes than for small ones. The 36Vo clipping factor occuning with 5 channels
produces
unacceptable voice quality. On the other hand, the 4Vo
clipping
probability
for 50 channels can be tolerated when the line costs are high enough.
In reality, the values for blocking probabilities obtained in Example 12.9 are overly
pessimistic
because an infinite source assumption was used. The summations did not
include the case of all sources being active because there needs to be at least one idle
source to create an arrival during the time congestion. A more accurate solution to this
problem is obtained in a later section using a finite source analysis.
12.2.4 Lost Calls Cleared-Finite Sources
As mentioned previously,
a fundamental assumption in the derivation of the Poisson
arrival distribution, and consequently Erlang's loss formula, is that call arrivals occur
independently of the number of active callers. Obviously, this assumption can be
jus-
tified only when the number of sources is much larger than the number of servers. This
section presents
some fundamental relationships for determining blocking prob-
abilities of lost calls cleared systems when the number of sources is not much larger
than the number of servers. The blocking probabilities in these case$ are always less
than those for infinite rtource system$ since the arrival rate decreases as the number of
busy sources increases.
When considering finite $ource systems, traffic theorists inffoduce anotherparame-
ter of interest called time congestion. Time congestion is the
percentage
of time that
all servers in a
group are busy. It is idertical to the probability that all server$ are busy
at randomly selected times. However, time congestion is not necessarily identical to
blocking probability (which
is sometimes refened to as call congestion). Time con-
gestion merely specifies the probability that all servers are busy. Before blocking can
occur, there must be an arrival.
In an infinite lrource system, time congestion and call congestion are identical be-
cause the percentage ofarrivals encountering all servers busy is exactly equal to the
time congestion.
(The
fact that all servers are busy has no bearing on whether or not
an arrival occurs.) In a finite source system, however, the percentage
of arrivals en-
countering congestion is smaller because fewer arrivals occur during periods when all
$ervers are busy. Thus in a finite riource system, call congestion (blocking probability)
is always less than the time congestion. As an extreme example, consider equal num-
bers of sources and servers. The time congestion is the probability that all seners are
busy. The blocking probability
is obviously zero.
The same basic techniques introduced by Erlang when he determined the loss for-
mula for infinite sources can be used to derive loss formulas for finite sources
[3].
Us-
542 TRAFF|c ANALysts
ing these techniques, we find the probability
of n servers being busy in a system with
M sources and N servers is
( 12. 1
1)
where t is the calling rate per
lzle source and f* is the average holding time. Equation
12.1I is known as the truncated Bernoullian distribution and also as the Engset dishi-
bution.
setting n
=
N in Equation 12.1 1 produces
an expression for the time congestion:
(r2.r2)
using the fact that the arrival rate when N servers are busy is (M
-M)/r4
times the ar-
rival rate when no servers are busy, we can determine the blocking probability
for lost
calls cleared with a finite source as follows;
u
=
[t"
t)t^"-r
( 12. 13)
ELF,I-
tlrrrJ'
t ' '
)
which is identical to P1,, (the
time congestion) for M
-
I sources.
Equations l2.ll, 12-12, and 12.13 are easily evaluated in terms of the parameters
ir,' and t-. However, Ll and r. do not, by themselves, speci$, the average activity of a
source' In a lost calls cleared system with finite sources the effective offered load de-
crease$ as the blocking probability
increases because blocked calls leave and do not
retum' When a call is blocked, the average activity of the offering source decreases,
which increases the average amount of idle time for that source. The net result is that
U decreases because the amount of idle time increases. If the average activity of a
source assuming no traffic is cleared is designated a$ p
=
It*, the value of l/t can be
determined as
rL[Y)rrr-r'
} j r * : =* - j
" '
I
-
P( l - B)
(r2.14)
12.? LosssysTHMs 543
where B is the blocking
probability defined by Equation 1?.13.
The difficulty with using the unblocked $ource activity factor p to characterize a
$ource's offered load is now apparent. The value of li t* depends on B, which in furn
depends on l"'I*. Thus some form of iteration is needed to determine B when the
sources are characterized by
p (an
easily measured parameter)
instead of il. If the total
offered load is
considered to be Mp, the carried traffic is
A"*ie*l
=MPI
-
B)
(
I 2. 1s)
A table of traffic capacities for finite sources is provided in Appendix D.2, where
the offered Ioad A
=Mp
is listed for various combinations of M, N, and 8. Some of
the results are plotted in Figure 12.10, where they can be compared to blocking prob-
abilities of infinite source systems. As expected, infinite $ource analyses (Erlang-B)
are acceptable when the number of sources M is large.
Example 12,10, A group
of callers generate request$ at a rate of f,rve calls per
hour
per
telephone
(including
incoming and outgoing calls). Assuming the average holding
time is 4 min, what is the average calling rate of each idle source? How many callers
can be supported by a l2-channel concentrator/multiplexer if the maximum
acceptable blocking probability is lVo'!
Solution, Since each caller is typically active for ?0 min ofevery hour and places
an average of five calls during the 40 min of inactivity, the calling rate for idle $ources
M
=
5l4O
=
0.l?5 calls
per
minute. The offered load for M sources, assuming all traffic
is carried, is 0.33 M. Table D.2 must be searched to find the largest M such that 0.33
M is less than or equal to the maximum offered load for B
=
lVo and N
=
12. Using
interpolation for M
=
2l reveals that 12 servers can support 7.11 erlangs at B
=
LVo.
Since 2I x 0.33
=
6.93 is the offered load, 2l sources is an acceptable solution. If 22
0.4 0.8 0.6 0,7 0.9 0.9
Offered traffic
pr rErvsr
(orlangBl
!
5
c
o
C
E
t
d
ffi
Figure 12.10 Blocking probability of lost calls cleared with finite sources.
544 THAFFIo ANALysts
sources are used, the offered load of 7.26 erlangs is higher than the 7.04 erlangs
obtainable from interpolation in Table D.2 as the maximum offered load for B
=
lvo.
It is worthwhile comparing the result of Example 12.10 to a result obtained from
an infinite source analysis (Erlang-B).
For a blocking probability
of lvo,Table D.l re-
veals that the maximum offered load for 12 servers is 5.88 erlangs. Thus the maximum
number of sources can be determined as 5.88/0.333
=
t/.94. Hence in this case an in-
finite source analysis produces
a result that is conservative by l5Vo.
12.2.5 Lost Calls Hetd-Flnlte
Sources
A lost calls held system with finite sources is analyzed in the same basic manner as a
lost calls held $ystems with infinite sources. At all times the number of calls
"in
the
system" is defined to be identical to the number of calls that would be serviced by a
strictly nonblocking
server group. Thus Equation 12.11 is used to derermine the prob-
ability that exactly n calls are in the system:
(r2. 16)
Because no calls are cleared, the offered load per idle source is not dependent on B:
r%[Y)rr,t-r'
[y),",-,
ql
+ l"'r*)tr
r,,*=*h=f;
(12.r7)
combining Equations l?.16 and 12.17 produces
a more useful expression for the
probability
that n calls are in the system:
+=[ f ; )ot ' -p)M-n
( 12. 18)
If there are N servers, the time congestion is merely the probability
that N or more
servers are busy:
Pr*
EP,
n4'l
(1,'t-)"
(12. 19)
12.2 LOSS SYSTEMS 545
The blocking probability, in a lost calls held sense, i$ the
probability
of an arrival
encounterins try'or more calls in the svstem:
Br,
=
ff^
f
,prob(anival
lru sources are busy)
average arrival rate
(12.20)
where p
=
offered load per source
M
=
number of sources
N
=
number of servers
Exampte 12.11. Determine the
probability
of clipping for the TASI systems
described in Example 12.9. In this case, however, use a lost calls held analysis for
finite sources.
Solution, In this exarnple we are concemed only with the
probability
that a speech
utterance is clipped for some
period
of time until a channel becomes available. Thus
Equation 12.20 provides
the desired answer using p
=
0.4 for the offered load per
source. In the first case, with 10 sources and 5 channels,
t)o' ct
-
p)M
r-' r
9
/rt\
Bn
=
E | ] lto.+fto.o)%n
=
o.z7
,=s
(-/
In the second case with 100 sources and 50 seryers.
.
=,1f;)(o'4)n(0
6)ee-'= o'023
The results of Example 12.11 show again that a TASI system requires large group
sizes to
provide low clipping probabilities. When compared to the results of Example
12.9, these results indicate that an infinite source analysis overestimates the clipping
probability in both cases
(0.36
versus 0.27 and 0.04 versus 0.023). Notice that the
per-
centage error in the infinite $ource analysis is almost identical for the IO-source
system and the l0O-source system. Hence the validity of choosing an infinite
source model is more dependent on the ratio of sources to servers than it is on the
number of sources.
=iU
546 TRAFFTc ANALYSIs
$ource activlty
(erlangsl
F'igure l2.Il Clipping probability
and clipping duration of TASI.
Example I2. I I merely determines the probability that a speech segment encoun-
ters congestion and is subsequently clipped. A complere analysis of a TASI (or
DCM)
system must consider the time duration of clips in addition to their frequency of oc-
cuffence. In essence, the desired information is represented by the amount of traffic
volume in the clipped $egments. weinstein
[6]
refers to the clippert $egments as
"frac-
tional speech loss" or simply
"cutout
fraction." This is not the silne as the lost traffic,
since a conventional lost calls held analysis considers any arrival that encounters con-
gestion
as being completely
"lost"-even
if it eventually receives some service. The
cutout fraction is determined as the ratio of untransmitted traffic intensitv to offered
traffic intensitv:
F
E
E
5
E
I
ll
s
$
s
E
o
g
E
L
o
.s
o
.s
g
o
.E
4
.t
o
E
iz.g NErwoFrK BLocKtNc
pFroBABtLtlES
547
(r2.21)
where M
=
number of sources
A
=
ofTered load,
=
Mp
N
=
number of servers
Pn
=
probability of n calls in the system
(Equation
12. 18)
Example 12.12, Determine the average duration of a clip in the two TASI systems
of Example 12.11. Assume the average duration of a speech $egment is 300 msec.
(The
average length of a speech segment is dependent on the activity threshold, which
also influences the activity factor p.)
Solution. In the first case,
M
n",:i.E (,
-1v)P"
n=N+l
100
u",:+,*E,,n
-
50)
fifl),r.-t*.6)rtx]-
=0.00r
a,,=II
tn
-
s)
f1,0.)(0.4r(0.6)10-n
:0.05e
n + t \ - /
Thus, on average, 5.9Va, ot 17.7 msec, of every 300-msec speech segment is clipped.
Since?7%o of thesegmentsexperienceclipping(Example
12. l l),theaverageduration
of a clip for clipped segments is 0.059/0.27
=
22Vo, or 66 msec-an obviously intol-
erable amount. In the second case for 100 sources and 50 channels.
Thus in this case only 0.1% of all speech is clipped, which implies that when clipping
occurs, 300 x 0.001/0.023= 13 msec of the segment is lost.
Example 12.12 shows that large group sizes not only greatly reduce the clipping
probability
of TASI systems but also reduce the duration of the clips. The relationship
of clipping probabilities
and clipping duration
(fractional
speech loss) to group
size
and source activity is provided in Figure I ?.I l. As shown, the clipping
probability is
extrcmely sensitive to s(turce activity (offered load). For a discussion of the effects of
clipping on speech quality, see references
[7]
and
[8].
12.3 NETWORKBLOCKING PROBABILITIES
In the preceding
sections basic techniques ofcongestion theory are presented to deter-
mine blocking probabilities
of individual trunk groups. In this section techniques of
calculating end-to-end blocking probabilities of a network with more than one route
548 TBAFFIc ANALysts
between endpoints is considered. In conjunction with calculating the end-to-end
blocking probabilities,
it is necessary to consider the interaction of raffic on various
routes of a network. Foremost among these considerations is the effect of overflow
ffaffic from one route onto another. The following sections discuss simplified analyses
only. More sophisticated techniques for more complex networks can be obtained in
references
t9l, t101,
and
I
l].
12.3.1 End-to-End Blocking Probabilities
Generally, a connection through a large network involves a series of transmission
links, each one of which is selected from a set of alternatives. Thus an end-to-end
blocking probability
analysis usually involves a composite of series and
parallel prob-
abilities. The simplest procedure
is identical ro the blocking probability (marching
loss) analyses presented
in chapter 5 for switching networks. For example, Figure
I 2. I 2 depicts a representative set of alternative connections through a network and the
resulting composite blocking probability.
The blocking probability
equarion in Figure 12.12 contains severar simplifying as-
sumptions. First, the blocking probability (matching
loss) of the switches is not in-
cluded. In a digital time division switch, matching loss can be low enough that it is
easily eliminated from the analysis. In other switches, however, the matching loss may
not be insignificant. when necessary, switch blocking is included in the analysis by
considering it a source ofblocking in series with the associated ffunk groups.
when more than one route passes
through the same switch, as in node c of Figure
12'12, proper
treatment of correlation between matching losses is an additional com-
plication.
A conservative approach considers the matching loss to be completely cor-
related. In this case the matching loss is in series with the common link. On the other
hand, an optimistic analysis assumes that the matching losses are independent, which
implies that they are in series with the individual links. Figure 12.13 depicts rhese two
approaches for including the matching loss of switch c into the end-to-end blocking
probability
equation of Figure l2.r2.In rhis case, the link from c to D is the common
link.
B=pr 11- qpqsl
Figure 12.12 hobability graph for end-ro-end blocking analysis.
1
-
qz{t
1
-q+
fr%-
d
;
bo"o
,o,
Qr*= 1- (1
-0r9")Q
g=1- { r of i g
Figure 12.13 Incorporating swirch*matching loss into end-to-end blocking analysis:
(a)
independent switch blocking; (b) correlated switch blocking.
A second simplifying assumption used in deriving the blocking probability equa-
tion in Figure 12.I ? involves assuming independence for the blocking probabilities
of
the trunk groups. Thus the composite blocking of two parallel routes is merely the
product
of the respective probabilities (Equation
5.4). Similarly, independence im-
plies that the blocking probability
of two paths-in series-is I minus the product of
the respective availabilities (Equation
5.5). In actual practice individual blocking
probabilities
are never completely independent. This is particularly
true when a large
amount of traffic on one route results as overflow from another route. Whenever the
first route is busy, it is likely that more than the average amount of overflow is being
divefied to the second route. Thus an alternate route is more likely to be busy when a
primary route is busy.
In a large prrblic network, trunks to tandem or toll switches normally carry traffrc
to many destinations. Thus no one direct route contributes an overwhelming amount
of overflow traffic to a panicular
trunk
group.
In this case independent blocking prob-
abilities on alternate routes are
justified.
In some instances of the public network, and
often in private
networks, overflow traffic from one route dominates the traffic on tan-
dem routes. In these cases failure to account for the conelation in blocking prob-
abilities can lead to overly optimistic results.
Example 12.13. Two trunk
groups
are to be used as direct routes between tws
switching systems. The first group
has 12 channels and the second group
has 6
channels. Assume I0.8 erlangs of haffic is offlered to the l2-channel group and
overflows are offered to the 6-channel group when the first group is busy. What is the
blocking probability
of the first group, and how much traffic overflows to the second
group? Using the overflow traffic volume as an offered load, determine the blocking
probability of the second kunk group. What is the probability
that both trunk groups
are busy? Compare this answer to the blocking probability
of one t8-channel tunk
group.
12.3 NETWoRK BLoGKING PRoBABILITIES 549
l
-wfl"-
{
@- q\
E/
\0,
ffito'
B= 1-(1-sl Ssxl
-Q2f t e)
550 THAFFI0 ANALysts
solution. using a lost calls cleared analysis with an infinite source, we
determi nethatbl ocki ngof
thefi rsrgroupi s l | vo(A= l 0.B,N= 12).Thereforethe
overflow traffic is 10.8 x 0.ltsI.62 erlangs. The blocking probability (assuming
random arrivals?) of the second group is o.Svo (A =
1.62, N
=
6). The probability
that both trunk groups
are busy simultaneously
can be determined (assuming
independence?)
as
B=0. 15x0. 005=0. 00075
In contrast, the correct blocking probability
of an l8-channel trunk group with an of-
fered load of 10.8 erlangs is
B: 0. 013
The question
marks in the solution of Example 12.13 point
to two sources of error
in the determination of the first blocking probability
value. one error is the assump-
tion ofindependence ofblocking in the two trunk groups.
A second error results from
the use of an analysis predicated
on purely
random (poisson)
arrivals for over{low
traffic into the second trunk group.
Resolution ofthis error is discussed in the next sec*
tion.
separating the 18 channels of Example 12.13 into two groups
is an obvious artifice.
This example is useful in that it demonshates an extreme case of correlation between
blocking probabilities
of two trunk groups.
when correlation exists, the composite
blocking probability
of a direct route and an alternate route should be determined as
follows:
B
=
(8,) (8211) (12.22)
where 81
=
blocking probability
of group I
B2ll
=
blocking probability
ofgroup 2 given
that group
I is busy
In the artificial case of dividing a trunk group
into two subgroups, the conditional
blocking probability
can be determined as
Brl l
=B(MN1)
;
prob (N
servers are busy when N, are known to be busy)
_
P*
_
AN/NI
- ! 1 V
p
- \ a N
r
-n=Nr' n un=71;An/nl )
(r2.23)
where Pn is the probability
rhat exactly n of N servers are busy
(Equation
l?.3).
la.s NETWoRK BLoCKTNG
pRoBABtLrrES
551
Evaluating Equation 12.23 for A
=
10.8, Nr
=
12, and N
=
18 reveals that the ap-
propriate
conditional
probability B2ll for Example 12.1 3 is 0.033. Thus the composite
blocking probability per Equation 12.22is B
=
0.15 x 0.033
=
0.005. The remaining
inaccuracy
(0.005
versus 0.013 actually) is due to nonrandom characteristics ofover-
flow trffic.
Equation 12.23 is valid only for the contrived case of an altemate route carrying
overflow traffic from only one
primary route. It can be used, however, as a worst-case
solution to situations where overflow from one route tend$ to dominate the traffic on
an alternate route.
The correlations between the blocking probabilities of individual routes arise be-
cause congestion on one route
produces overflows that tend to cause congestion on
other routes. Extemal events stimulating networkwide overloads also cause the block-
ing
probabilities
to be correlated. Thus a third assumption in the end-to-end blocking
probability equation of Figure 12.12 is that traffic throughout the network is inde-
pendent. If fluctuations in the traffic volume on individual links tend to be conelated
(presumably because of extemal events such as television commercials, etc.), signifi-
cant degradation in overall performance results.
12.3.2 OverflowTratfic
The second source oferror in Example 12.13 occurred because an Erlang-B analysis
used the average volume of overflow traffic from the first group
to determine the
blocking probability of the second trunk group. An Erlang-F analysis assumes traffic
arrivals are
purely
random, that is, they are modeled by a Poisson distribution. How-
ever, a Poisson anival distribution is an erroneous assumption for the traffic offered
to the second trunk group. Even though arrivals to the first group may be random, the
overflow process tends to select groups of these arrivals and
pass them on to the sec-
ond trunk group. Thus instead ofbeing random the arrivals to the second
group occur
in bursts. This overflow effect is illustrated in Figure 12.14, which portray$ a typical
random ardval pattern to one hunk group and the overflow pattern to a second group,
If a significant amount of the traffic flowing onto a trunk group results as overflow
from other trunk groups,
overly optimistic values of blocking probability arise when
all of the traffic is assumed to be purely random.
The most cornmon technique of dealing with overflow traffic is to relate the over-
flow traffic volume to an equivalent amount of random traffic in a blocking
prob-
ability sense. For example, if the 1.62 erlangs of overflow traffic in Example 12.12 is
equated to 2.04 erlangs of random toaffic, a blocking probability of LSVI is obtained
for the second trunk group. (This
is the correct probability
of blocking for the second
group since both
groups
are busy if and only if the second group is busy.)
Thi s method of treating over{low traffic is refened to as the equivalent
random the-
ory
[12].
Tables of traffrc capacity are available
[3]
that incorporate the overflow ef-
fects directly into the maximum offered loads. The Neal-Wilkinson tables used by
Bell System traffic engineers comprise one such set of tables. The Neal-Wilkinson ta-
bles, however, also incorporate the effects of day-to-day variations in the traffic load.
552 THAFFIC ANALYSIS
Olnrflofr rtylnh to rmnd trunk
Foup
l i l l
Figure 12.14 Bursty characteristic of overJlow traffic.
(Forty
erlangs on one day and 30 erlangs on another is not the $ame as 35 erlangs on
both days.) These tables are also used for trunk groups
that neither generate
nor re-
ceive overflow traffic. The fact that cleared traffic doe$ not get
serviced by an alternate
route implies that reffies are likely. The effect of the rehies, however, is effectivelv
incorporated
into the value of B by equivatent randomness.
12.4 DELAY SYSTEMS
The second category oftelefaffic analysis concern$ systems that delay nonserviceable
requests until the necessary facilities become available. These systems are variously
referred to as delay systems, waiting-call systems, and queuing
system$. call arrivals
occurring when all servers are busy are placed
in a
queue
and held until service com-
mences. The queue
might consist of storage facilities in a physical
sense, such as
blocks of memory in a message-switching node, or the queue
might consist only of a
list of sources waiting for service. In the latter ca$e, storage of the messages is the re-
sponsibility of the sources themselves.
using the more general
tnrm queueing
theory, we can apply the following analyses
to a wide variety of applications outside of telecommunications.
Some of the more
coiltmon applications are data processing,
supermarket check-out counters. aircraft
landings, inventory control, and various forms of service bureaus. These and many
other applicatrons are considered in the field ofoperations research. The foundations
of queuing
theory, however, rest on fundamental
techniques developed by early tele-
communications
traffic researchers. In fact, Erlang is credited with the first solution
to the most basic type of delay system. Examples of delay system analysis applications
in telecommunications
are message switching, packet
switching, statistical time divi-
sion multiplexing, multipoint data communications, automatic call distribution, digit
receiver access, signaling equipment usage, and call processing.
Furthermore,
many
AfflYrh to flil trunk
gru.rp
Drprturd frotn llnt tunk
$uup
frprclty of fim mrnk
ero{F
12.4 DELAY$YSTEMS 553
PBXs have features allowing queued acce$$ to corporate tie lines or WATS lines. Thus
some systems formerly operating as loss systems now operate as delay systems.
ln
general, a delay operation allows for
greater
utilization of servers
(fransmission
facilities) than does a loss system. Basically, the improved utilization is achieved be-
cause peaks in the arrival process
are
"smoothed"
by the queue. Even though arrivals
to the sy$tem are random, the $ervers sse a somewhat regular arrival pattern. The effect
of the queuing process on overload traffic is illustrated in Figure 12.15. This figure
displays the same traffic
patterns presented earlier in Figures I2.1,12.3, and I2.9. In
this case, however, overload traffic is delayed until call terminations produce available
chamels.
In most of the following analyses it is assumed that all traffic offered to the system
eventually
gets
serviced. One implication of this assumption is that the offered haffic
intensity A is less than the number of servers N. Even when A is less than N, there are
two cases in which the canied traffic might be less than the offered traffic. First, some
sources might tire of waiting in a long queue and abandon the request. Second, the ca-
pacity for storing requests may be finite. Hence requests may occasionally be rejected
by the system.
A second assumption in the following analyses is that infinite $ources exist. In a
delay system, there may be a finite number of sources in a
physical
sense but an infi-
nite number of sources in an operational sense because each source may have an arbi-
trary number of requests outstanding
(e.9.,
a
packet-switching
node). There are
instances in which a finite source analysis is necessary, but not in the applications con-
sidered here.
An additional implication of servicing all offered traffic arises when infinite
source$ exist. This implication is the need for inJinite queuing
capabilities. Even
Ee
E
E.E
g s
E C
6
F G - + - - r - + - - - '
Figure 12.15 Activity profile of blocked calls delayed
(13
servers).
554 THAFFIC ANALYSIS
though the offered traffic intensity is less than the number of servers, no statistical
limit exists on the number of arrivals occurring in a short period
of time. Thus the
queue
of a
purely
lossless system must be arbitrarily long. In a
practical
sense, only
finite queues
can be realized, so either a statistical chance of blocking is always pr+
sent or all sources can be busy and not offer additional traffic.
When analyzing delay systems, it is convenient
to separate the total time that a re-
quest
is in the sysrem into the waiting time and rhe holding time. In delay systems
analysis the holding time is more corlmonly referred to as the service time. In contrast
to loss systems, delay system performance
is ggnerally
dependent on the distribution
of service times and notjust the mean value Im. Two service time distributions are con-
sidered here; constant service
times and exponential service times. Respectively,
these
distributions represent the most deterministic and the most random service times pos-
sible. Thus a system that operates with some other distribution of service times per-
forms somewhere
between the performance
produced
by these two distributions.
The basic purpose
of the following analyses is to determine
the probability
distri-
bution of waiting times. From the distribution, the average waiting time is easily de-
termined. sometimes only the average waiting time is of interest. More generally,
however, the probability
that the waiting time exceeds some specified value is of in-
terest. In either case, the waiting times are dependent on the following factors:
I. Intensity and probabilistic
nature ofthe offered traffic
2. Distribution of service times
3. Number of servers
4. Number of sources
5. Service discipline of the queue
The service discipline of the queue
can involve a number of factors. The first of
these concerns the manner in which waiting calls are selected. commonly, waiting
calls are selected on a first-come, first-served (FCFS)
basis, which is also referred to
as first-in, first-out (FrFo)
seryice. sometimes, however, the server system itself does
not maintain a queue
but merely polls its $ources in a round-robin fashion to determine
which ones are waiting for service. Thus the queue may be serviced in sequential order
of the waiting sources. In some applications
waiting requests may even be selected at
random. Furthermore,
additional service variations arise if any of these schemes are
augmented with a priority
discipline that allows some calls to move ahead of others
in the queue.
A second aspect of the service discipline that must be considered is the length of
the queue.
If the maximum queue size is smaller than the effective number of sources,
blocking can occur in a lost calls sense. The result is that two characteristics
of the
grade
of service must be considered: the delay probability
and the blocking prob-
ability' A common example of a system with both delay and loss characteristics is an
automatic call distributor with more access circuits than attendants (operators
or re-
servationists). Normally, incoming calls are queued
for service. under heavy loads,
12.4 DELAYSYSTEMS 555
however, blocking occurs before the ACD is even reached. Reference
[4]
contains
an analysis of a delay system with finite queues and finite $ervers.
To simplify the characterization of
pafiicular
systems, queuing theorists have
adopted a concise notation for classifying various types of delay systems. This nota-
tion, which was introduced by D. G. Kendall, uses letter abbreviations to identify al-
ternatives in each of the categories listed. Although the discussions in this book do not
rely on this notation, it is introduced and used occasionally so the reader can relate the
following discussions to classical
queuing theory models. The interpretation of each
letter is specified in Figure 12.16.
The specification format presented
in Figure 12,1 6 actually represents an extension
of the format commonly used by most queuing theorists. Thus this format is some-
times abbreviated by eliminating the last one or two entries. When these entries are
eliminated, infinite case specifications are assumed. For example, a single-server sys*
tem with random input and negative exponential seryice times is usually specified as
Mll,lUl. Both the number of sources and the permissible queue length are assumed in-
finite.
12.4.1 Exponential Servlce Times
The simplest delay system to analyze is a system with random arrivals and negative
exponential service times: M/IVI/I.{. Recall that a random arival distribution is one
with negative exponential interarrival times. Thus in the shorthand notation of queu-
ing theorists, the letter M always refers to negative exponential distributions
(an
M is
used because a purely random distribution is memoryless).
Input rfclflcttlon
Servlce timE distrihrtiorl
Numbtr of *rrrot
Gonffrl {nq s$urhpt;ofi;
Purely rrndom
Gcnlfd
{no
ffiumptiofil
;{egltiw rxponintirl
Corltrnt
Flnlte number
{ ;
Gl
Mr
D:
N:
Numbff of rourcd Jur
rinittnumuer
I
L*
Inrinitc
I
or*"
r"r$fi
{
L' Flnite lensrttr
| |
L*'
Inflnitc lefisth
t t
3 / 4 / E
Figure 12.16
Queueing
$ystem notation.
l 2 l
556 THAFFIo ANALysts
In the IM/lvVl system and all other systems considered here, it is assumed that calls
are serviced in the order of their arival. The following
analyses al$o assume that the
probability
of an arrival is independent of the number of requests already in the queue
(infinite
sources). From these assumptions,
the probability
that a call experiences con-
gestion
and is therefore delayed was derived by Erlang:
where N
=
number of servers
A
=
offered load (erlangs)
B
=
blocking probability
for a lost calls cleared system (Equation
l2.g)
The probability
of delay p(>0) is variously refened ro as Erlang's second formula,
Ezn(A); Erlang's delay formula;
or the Erlang-c formula. For single-server
$ystems
(N=
l) the probability
of delay reduces to p,
which is simply rhe outpur urilization or
traffic carried by the server. Thus the probability
of delay for a single-server sysrem
is also equal to the offered load l,t- (assuming
fu. <
l).
The distribution
of waiting times for random arrivals, random service times. and a
FIFO service discipline is
p(>t)
:
p(>0) "-{N-A)t/to,
(r2. ?5)
wherep(>0) =
probability
of delay given
in Equation 12.24
fm
=
flYrfl8o service time of negative exponential service time distribution
Equation 12.25 defines the probability
that a call aniving at a randomly chosen instant
is delayed for more than t/t^ service times. Figure 12.17 presents
the relationship
of
Equation 12.25 by displaying the traffic capacities of various
numbers of servers as a
function of acceptable delay times. Given a delay time objective r,/f., Figure l2.l7a
displays the maximum traffic intensity if the delay objective is to be exceeded for only
lovo of the arrivals. similarly,
Figure lz,.r7b displays the maximum traffic intensity
if the delay objective is exceeded for only lzo of the arrivals. Notice that atp(>r)
=
0.01, the server systems do not approach their maximum
capacity (number
of servers)
unless the acceptable delay is several times larger than f*.
By integrating Equation 12.25 over all time, the average waiting time for all arri-
vals can be determined
as
;
P(>O)r,n
f = -
N_ A
(r2.24)
(12.26)
Notice thatTis the expected delay for all arrivals. The average delay ofonly those ar.
12.4 DELAYSYSTEMS 557
It/t-l
r.0
,rrr.u
2.0 e.s
(yr-l
Figure 12.17 Traffic capacity of multiple-server delay systems with exponential service
times;(a)probabilityof exceeding
LpFt)=10To;(b)
probabilityof exceeding t,p(>t)=lVo.
rivals that get delayed is commonly denoted as
E
I
I
B
.B
E
F
OF
1,0
1.6 2.0 aE
(t)
$
o
!
u
o
E
* - t *
v - N- a
(r2.27)
Example 12.14. A message-switching network is to be designed for 95Vo
utilization of its hansmission links. Assuming exponentially distributed message
lengths and an arrival rate of 10 messages
per minute, what is the average waiting
time, and what is the probability that the waiting time exceeds 5 min?
Salutinn. Assume that the message-switching network uses a single channel
between each pair of nodes. Thus there is a single server and a single
queue for each
transmission link. Since p is given to be 0.95 and l,
=
10 arrivals
per
minute, the
558 TRAFFIc ANALYSIS
average service time can be determined as f*
=
0.g5l10
=
0.095 min. The averase
waiting time
(not
including
the service time) is easily determined as
;=
o'q5 x-q:095
=
l.Bo5 min
I
-
0.95
Using Equation 12.25, we can determine the
probability
of the waiting time exceeding
5 min as
p(>5)
=
(0.95)e-tt-o'sr)5/0.0es
-
0.068
Thus 6.87o of the messages experience queuing
delays of more than 5 min.
Example 12.15. Derermine rhe number of digit receivers required to support 1000
telephones with an average calling rate of two calls per hour. Assume the diating time
is exponentially
distibuted with an average service time of 6 sec. The grade
of service
objective is to return dial tone within I sec ofthe off-hook signal for 99Zo ofthe call
attempts. compare the answer obtained from a delay system analysis to an answer
obtained from a loss system analysis at B
=
lVo . If the blocking probability
is less rhan
l%o, fewer than lVo of the calls are delayed.
Solution. The calling rate l, and the offered haffic intensiry,4
are easily determined
as 0.555 calls per second and 3.33 erlangs, respectively. since the number of servers
N cannot be solved for directly from the equations, Figure lz.ljb is used to obtain a
value of eight servers for t/t^=t.
Examination of Table D.1 reveals thatgg.Svo of the call attempts can be serviced
immediately
if there ate nine digit receivers. Thus in this case the abilitv to delav serv-
ice provides
a savings ofonly one server.
Example 12.15 demonstrates rhat a blocking probability
analysis produces
ap-
proximately
the same results as a delay system analysis when the maximum accept-
able delay is a small percentage
of the average service time. The two results are almost
identical because, if a digit receiver is not imrnediately
available, there is only a small
probability
that one will become available within a short time period. (With
an aver-
age service time of 6 sec, the expected time for one of eight digit receivers to be re-
leased is 6/8
=0.75
sec. Hence the delay operation in this case allows a savings of one
digit receiver.)
Because
a digit receiver must be available within a relatively
short time period
after
a request is generated,
digit receiver group
sizing is often determined strictly from a
blocking probability
analysis. The fact that digit receiver access is actually operated
as a delay system implies the grade
of service is always better than that calculated.
12.4.2 Constant
Service Timee
This section considers delay systems with random arrivals, constant service times, and
a single server (IWD/I).
Again, FIFO service disciplines
and infinite $ources are as-
12.4 DELAYSYSTEMS 559
sumed. The case for multiple servers has been solved
[3]
but is too involved to include
here. Graphs of multiple-server systems with constant service times are available in
reference
[5].
The average waiting time for a single server with constant service times is deter-
mined as
where p
=,4
istheserver utilization.NoticethatEquation l2,28producesan average
waitingtimethatisexactlyone-halfofthatfor asingle-server systemwithexponential
servicetimes. Exponential servicetimes causegreater averagedelays becausethere
aretworandomprocessesinvolved increatingthedelay.Inbothtypesofsystems,de-
lays occur when a large burst of arivals exceeds the capacity of the servers. With ex-
ponential service times, however, long delays also arise because of excessive service
times ofjust a few arrivals.
(Recall
that this aspect of conventional message-switching
systems is one of the motivations for breaking messages up into
packets
in a packet-
switching network.)
If the activity profile of a constant service time system
(M/D/l)
is compared with
the activity
profile of an exponential service time system
(lWIWl),
the M/D/l $ystem
is seen to be active for shorter and more frequent periods of time. That is, the Ivl/ii{,/l
system has a higher vanance in the duration of its busy
periods. The average activity
of both system$ is, of course, equal to the server utilization
p. Hence the
probability
of delay for a single-server system with constant service times is identical to that for
exponential service times: p(>0)
=
l"t .
The probability of congestion for larger N is relatively close to that for exponential
service times. Thus Equation 1?.25 can be used as a close approximation forp(>0) for
multiple-server systems with arbitrary service time distributions.
For single-server systems with constant holding times, the
probability of delay
greater than an arbitrary value f is
p(>r)
=p[>(ft+
r)t*l
k
= t _ ( 1 _ p ) E
I4
pili
-
t/t^1itP$-t/t^)
-
Pf-
' =
ze*p)
(r2.28)
k
=1-(l -p)er' i ef f
where ft
=
largest integral quotient of t/t*
r
=
remainder of t/t^
P
=
server utilization,
=
fu-
(r2.2e)
560 THAFFIc ANALysts
Comparisons
of the waiting time distributions for single-server systems with ex-
ponential
and constant service times are shown in Figure l2.lg. For each pair
of
curves' the upper one is for exponential
service times and the lower one is for constant
service times. Since all other
service time distributions produce
delay probabilities
be-
tween these extremes, Figure 12. 18 provides
a direct indication of the range of possi-
ble delays.
Example 12.16. A packet-swirching
node operates with fixed-length packets
of
300 bits on 9600-bps lines. If the link utilization is to be 90%, whatls the averase
q
E
o
o
x
o
g
o
E
.E
o
&
DElay time. r/r-
Figure 12.18 Delay ptobabilities
ofsingle-server
$ystems (exponential
and constant service
times).
12.4 DELAYSYSTEMS 561
delay through a node? What percentage of packets encounter more than 0.35 sec of
delay? What is the average delay if the offered load increases by 107o?
Soluti,on. Message lengths of 300 bits and a data rate of 9600 bps imply that the
fixedJength service time is 300/9600
=
0.031 sec. From Equation 12.28, the average
waiting time is
0.9 x 0.031
f
=F=0. 14
sec
2(1
-
0.e)
The total average delay through the node, excluding
processing, is obtained by adding
the average waiting time to the service time:
Average delay
=
0.140 + 0.031
=
0.171 sec
Since the service time is 0.031 sec, 0.35 sec of delay occurs when the waiting time is
0.35
*
0.031
=
0.319. This corresponds to 0.319/0.031
=
10 service times. FromFigure
12.18, the probability of delay for t/t^- 10 is approximately 0.12. Thus lZVo of the
packets
experience delays of
greater
than 0.35 sec. An increase of 10Vo in the traffic
intensity implies that the new offered load is 0.99 erlang. From Equation 1?.30, the
average waiting time becomes
-
0.99
x 0.031
f
=- : =
1. 53 seC
2(t
-o.ee)
Thus when the offered load increases by only l0%, the average delay through the node
increases ninefold to a value of 1.53 + 0.03 I
=
I .56 sec !
Example I 2. 16 demonstmtes the same characteristic for heavily utilized delay sys-
tems that was demonstrated for loss sy$temsr The performance is very sensitive to in-
creases in traffic inten$ity. Thus, as discussed in Chapter 10, flow control is a critical
aspect of a packet-switching operation, particularly when there are real-time delivery
objectives.
12.4.3 Fi nl te Queues
All of the delay system analyses
presented so far have assumed that an arbitrarily large
number of delayed request$ could be placed in a queue. In many applications this as-
sumption is invalid. Examples of systems that sometimes have significantly limited
queue
$izes are store-and-forward swirching nodes
(e.g., packet switches and ATM
switches), automatic call distributors, and various types of computer input/outPut de-
vices. These systems treat arrivals in three different ways, depending on the number
"in
the $vstem" at the time of an arrival:
562 TRAFFIo ANALYSI$
I . Immediate
service if one or more of N servers are idle
2. Delayed service if all servers are busy and less than 1- requests are waiting
3. Blocked or no service if the queue
of length Z is full
In finite-queue
systems the arrivals getting
blocked are those that would otherwise
experience long delays in a
pure
delay system. Thus an indication of the blocking
probability
of a combined delay and loss sy$tem can be determined from the prob-_
ability that arrivals in pure
delay systems experience delays in excess of some speci-
fied value. However, there are two basic inaccuracies in such an analysis. First, the
effect of blocked or lost calls cleared is to reduce congestion for a period
of time and
thereby to reduce the delay probabilities
for subsequent anivals. Second, delay times
do not necessarily
indicate how many calls are
"in
the system." Normally, queue
lengths and blocking probabilities
are determined in terms of the number of waiting
requests, not the amount of work or total service time represented
by the requesd.
with constant service times,
there is no ambiguity between the size of a queue
and its
implied
delay. with exponential service times, however,
a given
size can represent
a
wide range of delay times.
A packet-switching
node is an example of a system in which the queue
length is
mo$t appropriately determined
by implied service time and not by the number of pend-
ing requests. That is, the maximum queue
length may be determined by the amount
of store-and-forward
memory
available far variablelength
messages and not by some
fixed number of messages.
For a system with random input, exponential service times, N servers. an infinite
source, and a maximum queue
length of z (wMlNl*lL),
the probability
ofj calls in
the svstem is
PrA)=hni
INIri-
0<j <N
N< j s N+ L
(12.30)
where 4
=
offered load (erlangs), =
f/*
N
=
number of servers
I
=
maximum number in the queue
Here, Pq(A) is chosen to make the sum of all
p,(A)
=
l;
"',o)=[=i#.,-1,,---tl
=[-i#.#]#J
(12.30a)
12.4 DELAYSYSTEMS 563
The time congestion, or probability of an arrival being delayed or blocked, can be
determined foom Equation 12.30 as
P^(A)AN*L
B= P- , ( A) =
" '
'
N!N"
G I
- n , I t
p(>r):PN(A)
I
\
)
xte-*dx
i +" ' u' / ' *
from which the average delay can be determined as
(r2.3r)
(
r 2.32)
(12. 33)
(r2.34)
N_ A
Again, Equation 12.34 is identical to Equation 12.26 for an infinite
{ueue
(L = m).
Single-Server Equations
Because most queuing applications involve single-server configurations, the previous
equations are listed explicitly for N
=
l:
,
Prob(7 calls in system) (12.30):
ntG):rn(P)d
(r2.3s)
P.$)
=[,
'p+p*"I'=[E,'l'
N+L
P(>o)=EPIA)=Pru(A)++
j = w
l - P
where
p
=A/N
is the offered load per server.
The loss, or blocking probability, is determined as
lp(>O)
-Pff+r(A)lt*
It is worth noting that if there is no queue (I
=
0), these equations reduce to those of
the Erlang loss equation
(12.8).
If t is infinite, Equation 12.31 reduces to Erlang's de-
lay formula, Equation 12.24. Thus these equations represent a
general formulation that
produces
the pure loss and pure
delay formulas as special cases.
The waiting time distribution
[3]
is
(12"36)
564 TRAFFIG ANALysts
Probability
of delay (12.31):
p(>0):
Probability of loss (12.32):
Po(p)p(l
-
p*t)
p(l
I
- p
o _ { t - p ) p a l
u----=-: -
l - p* "
(r2. 37)
(12. 38)
Average delay (12.34):
;
tp(>o)
-
Pr*,(p)Jr*
p(l
-
pz)r-
l - p
( t _pXl
_pLr z )
(12.3e)
The blocking probability
of a single-seryer system (N =
1) is plotted
in Figure
12.19. when using Figure 12.19, keep in mind that the blocking probability (Equation
12.38) is determined by the number of waiting calls and not by the associated service
time' Furthermore,
$ince the curves of Figure 12.19 are based on exponential service
times, they overestimate the blocking probabilities
of constant holding time system$
(e.g.,
fixed-length packet
networks). However, if fixed-length packets
arise primarily
from longer, exponentially
dishibuted message$, the arrivals are no longer inde-
pendent,
and the use of Figure 12. 19 (or
Equation t 2.38) as a conservative analvsis is
more appropriate.
ATM Cell Queues
Analysis of queuing
delays and cell loss in an ATM switching node is complicated.
The cells have a fixed length of 53 bytes so it would seem that a constant service time
analysis would be appropriate. This assumption is valid for voice traffic inserted onto
wide-bandwidth
signals such as 155-Mbps sTS-ls. In this case rhe service time is
much shorter that the duration of a speech burst (e.g.,
2.7 psec
versus several tens of
milliseconds). Even though correlated arrivals occur from individual sources. the ar-
rival times are separated by many thousands of service times so they appear inde-
pendent.
When ATM voice is carried in CBR trunk groups,
a different situation results. In
this case the service times
of the voice cells may be only slightly smaller than the in-
terval between voice cell generation,
and the average delay would indicate that two or
more cells from the same source could be present in the queue
at one time. Thus, a
queuing
analysis that assumes exponentially distributed service times is more appro-
priate
even though the variable-length talk spurts are broken up into fixed{engttr ciffs.
Example 12.17. A 64-kbps cBR virtual path in an ATM network is to carry lg
voice signals that are compressed to 7.25 kbps during voice spurts. Assuming a speech
activity factor of 407o, determine each of the following;
I2.4 DELAYSYSTEMS 565
1. 000
100
010
0. 001
Figure 12.19 Blocking probability of single server loss/delay system
(exponential service
times).
(a) Output channel loading
(b) Service time
(c) Probability of delay with an infinite queue
(d) Probability of delay with a queue of length 20 ATM cells
(e)
Average time in
queue for exponential service times with an infinite
queue
(f;
Average time in queue for exponential service times with a finite queue of 20
(g)
Average time in
queue for constant service times
(infinite queue)
(h) Probability of cell loss
(assuming
exponential service times)
f,
I
2
3
5
10
tt
o'
-o
(s
-o
e
s.
(D
.g
.s
IJ
g
- 0 .
Offered baffic
(erlangs)
566 THAFFIC ANALYSIS
Solution
(a) An ATM cell consists of48 bytes ofpayload
and 5 bytes ofoverhead.
Thus, rhe
offered load to the channel
*
(18
x 7.25 kbps x 0.4) (53/4g)/64
kbps
=
0.9 erlangs.
(b)
The service time of a cell is 53 x
g/64
kbps
=
6.625 msec.
(c)
The probability
of delay from Equation 12.24 is 0.9.
(d)
The probability
of delay from Equarion 12.37 is 0.g9.
(e) The average queuing
delay from Equation 12.26 is 59.6 msec.
(0
The average queuing
delay from Equation 12.34 is 5g.2 msec.
(S) The average queuing
delay from Equation l?.29 is 29.g msec.
(h)
The probability
of cell loss from Equation 12.39 is 0.012.
The results of Example
12.17 illustrare
several imporrflnt points.
First, (c)
and (d)
indicate that the probability
of delay in a finite-queue
system is smaller than that of an
infinite-queue
system-becau$e
some traffic is rejected. with a reasonably
sized
queue
the effect on the delay probability
is small. similarly, a comparison
of (e)
and
(f) illustrates
that the average queuing
delay in a reasonably sized queue
is not much
different than that of an infinite queue.
As discussed earlier, there is a significant
dif-
ference between the average delay of a system with exponentially
distributed
service
times as opposed to constant service times. Before assuming that a consrant service
time analysis should be used, the average delay must be compared
to the delay be_
tween ardvals of voice
cells. The duration of a speech segment carried in the ATM
cells of Example 12.17 can be derermined
as 4g xgl72s0=
53 msec. The average de-
lay of a constant service time analysis (29.8
msec) indicates that cells from a particular
source are usually serviced before a subsequent
arrival from the $ame source. but cer-
tainly not always. If Equation
12.34 is used to determine the average delay, the 5g.2_
msec result indicates
that, on average, an arriving cell from a source encounters a
previous
cell from the same $ource. Thus the assumption of independent arrivals is
marginal. To be safe, a cell loss calculation assuming
exponential service times ac-
counts forcorrelated
arrivals.
The cell loss probability
of l.zvo is marginal from a
voice quality point
of view. Notice that this value of cell loss relates to the CBR gate-
way (AALI
adapration layer), which is presumably
the only significant source of cell
loss.
The solution to example problem
12.17 necessarily used a simple model for the
traffic and the queue.
A thorough analysis of an ATM or packet-switching
network is
much more complicated.
Factors that must be considered
are the switching node ar_
chitectures and queue
imprementation (e.g.,
shared queues
versus dedicated queues;,
server disciplines (e.g.,
priorities),
cell discard algorithms,
call admission
conhol al_
gorithms,
and traffic statistics. Some of these issues are covered in references
[16-lg].
12.4.4 Tandem
eueues
All ofthe equations provided
in previous
sections for delay system analysis have dealt
with the performance
of a single queue.
In many applications
a service request under-
goes
several stages ofprocessing,
each one ofwhich involves queuing.
Thus it is often
REFEHENcES 5S7
Locrl rrrlvrlr
Arrinh
lrom
pr|Yiodt
qu6lC
Figure 12.20 Tandem queues.
desirable to analyze the
performance of a system with a number of queues in series.
Figure 12.20 depicts a series of
queues that receive, as inputs, locally generated requests
and outputs from other
queues. Two principal examples of applications
with tandem
queue$ are data processing systems and storc-and-forward switching networks.
Researchers
in queuing theory have not been
generally
successful
in deriving for-
mulas for the performance of tandem
queues. Often, simulation iS used tO analyze a
complex arrangement of interdependent
queues arising in systems like store-and-for-
ward networhs. Simulation has the advantage
that special aspects of a network's op-
eration-like routing and flow control-can be included in the simulation model. The
main disadvantages of simulation are expen$e and, often, less visibility
into the de-
pendence of system
perforrnance on various design
parameters'
one tandem
queuing problem that has been solved
[19]
is for random inputs and
random
(negative
exponential) holding times for all queues. The solution of this sys-
tem is based on the following theorem:
In a delay system with
purely random arrivals
and negative exponential holding times, the instants at which calls terminate is also a
negative exponential distribution.
The significance of this theorem is that outputs from an IvI/IvIA.{ system have sta-
tistical properties that are identical to its inputs. Thus a queuing process in one stage
does not affect the arrival
process in a subsequent Stage, and all gueues can be ana-
lyzed independently. Specifically, if a delay $ystem with N servers has exponentially
distributed interarrival
times with average 1/1,, and if the average
service time is f.,
calls leave each ofthe servers according
to exponentially distributedintercompletion
times with average l/LN.
Although independent analysis of tandem queues can be rigorously
justified only
for purely random arrivals and service times, independence is often assumed in other
cases. Before using such assumptions, however, the systems in
question should be ex-
amined closely to determine
if the state of one
queue can influence the operation of
another
queue in the system.
REFERENCES
I W. Feller, An Introduction to Probability Theory and lts Applications, Wiley' New
York. 1968.
2 A. A. Collins and R. D. Pederson, Telecommunicatians-a
Timefor Innovation,Metle
Collins Foundation, Dallas, TX, 1973.
Output
quctre
568 rRAFFtc ANALysts
3 R. syski, Introduction
to congesfion Theory in Telephone,sysfems,
oliver and Boyd,
London. 1965.
4 Technical Staff, Bell Telephone Laboratories,
Engineering and Operations inthe Bell
System, Westem Electric, Indianapolis, lg7?.
5
'Determination
of the Number of Circuits in Automatic
and Semi-Automatic
operation," ccITT Recommendarion
E. 520, orange Book,vor. II.2, r976,p.zrr.
6 c. J. weinstein,
"Fractional
speech Loss and ralker Activity Model for
packet
switched speech," IEEE Transactions
on communications
Technology. Aug. I 97g, pp.
t253-1256.
7 H. Miedema and M. schachtman,
"TASI
eualiry-Effect of speech Detection and
Interpolation,"
Bell SystemTechnical Joumal, July 1962, pp.
l4i5_1473.
I G. Szarvas and H. Suyderhoud,
'-Voice-Activated-Swirch
Performance
Criteria,"
Comsat Technical Review, Spring I
g80,
pp. I Sl
*177
.
9 S. S. Katz,
"Improved
Network Administation
hocess Utilizing End-to-End Service
Considerations," Intemational
Teletraffic Conference, lg7g,
10 R. Dayem,
"Alternate
Routing in High Blocking communications Netw orks.,, National
Telecommunications
Conference, lg7g,pp. 2g.4. I
_2g.4.6,
l l P. R. Boorstyn and H. Frank,
"Large*scale
Network Topological optimization,"
rEEE
Transactions
on Communicaflrrzs,
Jan. lg77
,
pp,
Zg_47 .
12 R. I. wilkinson,
'Theories
for Ton rraffic Engineering in u.s.A.,- Beil system
Technical Journal, Mar. 1956.
13
"calculation
of the
Number of circuits in a Group carrying overflow Traff,rc," ccITT
Recommendation
E.52r, orange Boak, vol. 2, No. 2, Geneva, switzerrand, p.
2rg.
14 J. A. Morrison,
"Analysis
of some overflow hoblems with
eueueing,,'
Bell system
Technical Joumal, Oct.1980, pp.
1427-1462.
15 Telephone Tragic Theory, Tables, and charts, siemens Alctiengesellschaft,
Munich,
t970.
16 Y.-s. Yeh' M. G. Hruchyj, and A. s. Acampora,
"The
Knockout switch: A simple,
Modular Architecture for High-perfonnance
packet
switching,,, IEEE Journal on
Selected Areas in Communications,
Oct.l9g7, pp. ll74_llg3.
17 K. sriram, T. G. Lyons, and r.-T. wang,
"Anomaries
Due to Delay and Loss in AAL2
Packet voice systems: Performance Models and Methods of Mitigati on,,' IEEE roumal
on Selected Areas in Cornmunitations,
Jan, I
g99,
W,
4_17 .
18 K. Sriram and
y.
T.
yang,
..Voice
over ATM Using AALZ and Bit Dropping:
Perfbrmance and call Admission
control," IEEE Joumal on selected Areas in
Communications,
Jan. 1999, pp. 18-28.
19 L. Kleiruock and R. Gail,
eueueing
sysferns,
problems
and solutions, wiley, New
York. 1996,
PROBLEMS
l2.l A central-office-to-PBX
funk group
contains four circuits. Ifthe average call
duration is 3 min and the busy-hour offered haffic intensity is 2 erlangs, deter_
mine each of the followine:
12.5
12.2
12.3
12.4
12.6
12.7
12.8
tz.9
12,10
PHOBLEMS 569
(a)
Busy-hour calling rate
(b)
Probability that two arrivals occur less than I sec apart
(c)
Blocking probability
assuming
a lost calls cleared operation
(d)
Amount of lost traffic
(e)
Proportion of time the fourth circuit is in use
(assuming fixed-order selec-
tion)
A Tl line is used to cany traffic from a remote concentrator to a central office,
How many l0 CCS subscribers can the concentrator system suppofr at0.SVo
blocking. Compare
an infinite source analysis to a finite source analysis'
As-
sume blocked calls cleared.
Two switching
offices experience 20 erlangs of average busy-hour traffic load
between them. Assume a single TI line provides 24 direct trunks between the
offices. How much busy-hour traffic over{lows to a tandem switch?
A PBX with 200 stations has five trunks to the
public
network'
What is the
blocking probability if each station is involved in three external calls
per 8-hr
working day with an average duration
of 2 min
per call? Assume the average
calling rate is constant during the day
(no busy hour) and blocked calls return
with random retries. What is tlre offered load? What is the demand traffic?
How many dial-up input-output
(VO) ports are needed for a computer center
to support 40 users with a blocking
probability limit of 57o? Assume each u$er
averages four calls per day with an average $ession duration
of 30 min. If three
u$srs remain connected all day, what is the
grade of service for the remaining
37 users?
A 24-channel trunk
group is divided into two
Sroups
of 12 one-way
trunks in
each direction.
(A
one-way trunk is one that can only be seized at one end.)
How many erlangs of traffic can this system support at O-SVI blocking? How
many erlangs can be supported if all 24 trunks are two-way
trunks? (That is,
every trunk can be seized at either end.)
The following 10 A.M. to 11 A.M. busy-hour
Erlang
(E)
statistics have been
observed on a 3z-channel interoffice trunk
group. what is the overall blocking
probability? What is the blocking probability for the same busy hour if day-to-
day flucfuations are averaged together?
Monday Tuesday Wednesday
20E 19E' 22E'
Thursday
19E
Traffic measurements on a PBX-to-central-office trunk
group indicate that dur-
ing the busiest hour of the day the trunk$ are 807o utilized. If there are eight
trunks in the
group, what is the blocking
probability, assuming blocked calls
do not retum? How many trunks must be added to achieve a maximum block-
ing probability
of SVo?
Repeat Problem 12.8 assuming blocked calls return with random retries.
A small community with 400 subscribers is to be serviced
with a community
dial office switch. Assume
that the average subscriber originates 0.1 erlang of
traffic. Also assume that 20Vo of the originations are local
(intracommunity)
Friday
30E
570 TRAFFIC ANALYSIS
calls and that 807o are transit calls to tlre serving central office. How many er-
langs of traffic are offered to the community-dial-office-to-central-office g1nk
group?
How many frunks are needed for 0.5vo blocking of the hansit traffic?
12'11 For the comrnunity
of Problem 12.10 determine the number of concentrator
channels required iflocal calls are not switched locally but are merely concen-
trated into pair-gain
$ystems and switched at the central office.
12.12 Repeat
hoblems 12.10 and 12.l I if
g0%
of rhe originations
are inrracommu_
nity calls andZOVa are transit calls.
u.13 A group
of eight remote farm houses are serviced by four lines. If each of the
eight families utilizes their telephones for I0zo of the busy hour, compare the
blocking probabilities
of the following configurations:
(a)
Four party
lines with two srations per
line
(b)
An 8-to-4 concentration sy$tem
12.14 A PBX provides queuing
and automatic calr back for access to outgoing wATS
lines. If there are
z0 requests per
hour for the wATs lines, and if the average
call is 3 min in length,
how many WATS lines are needed to provide
derays of
less than I hr for 90Zo of the requests?
f2.15 A call processor
has 507o of its time available for servicing requests. If each re-
quest
requires 50 msec of processing
time, what arrival rate can be supported
if only I 7o of the service requests
are delayed by more than I sec? Assume that
processor
time is sliced into iOO-msec time slots. (That
is, 500 msec are allo_
cated to call processing
and then 500 msec to overhead functions..l
12'16 A group
of 100 sources offers messages with exponentially diskibuted
lengths
to a 1200-bps line. The average
message length is 200 bits, including overhead,
and each source generates
one mesrlage every 20 sec. Access to the line is con_
trolled by message-switching
concentration
with an infinite queue.
Determine
the following:
(a)
Probability
of entering rhe queue
(b)
Average queuing
delay for all arrivals
(c)
Probability
of being in the queue
for more than I sec
(d)
Utilization
of the transmission link
12.17 An airline company uses an automatic call distributor
to service re$ervations
and ticket purchases.
Assume that the processing
time of each inquiry is ran-
domty distributed
with a 40-sec
average. Also assume that if customers
are pur
on hold for more than 2 min, they hang up and cail another airrine. If each of
200 inquiries per hour produces
g30
worth of sales, on average, what is the op_
timum number of reservationists?
Assume each reservationist
costs the com-
pany
$20/hr
(including
overhead).
12'18 A radio station talk show solicits the listening public
for cornments on the in-
eptne$$ of government (I
assume this will be a topical subject for the life of this
book). Assume
that each caller talks for a random length of time with an aver-
age duration of I min. (Either
the show is unpopular
or the public
has given
up
PRoBLEMS 571
on the govemment.) How many incoming lines must the radio station have to
keep the idle time below 5% if the call arrival rate is 3 calls/min?
12.19 Repeat Example I 2' 17 for a
queue length L
=
40'
(Although a rigorous solution
requires calculation of a ZQ-term summation, only fhe first few terms are sig-
nificant.)
APPENDX
A
DERIVATION
OF EQUATIONS
A.1 QUANTIZING
NOISE POWER: EOUATION 3.2
The
probability density function of a noise v sampled is assumed
to be uniform;
The average or expected value of noise
Power
is determined
as
l +
-i q<n
<i q
p( n) =i q
lO
otherwise
euantization
noir*
po*"rJf
[n1jrt
*
=[#)n
A.2 NRz LINE GODE:
EQUATION 4.1
[ t
t r t s+7
,tt)
=
{O
other'wise
rUo))=j
71t1e-i'ntdt
574 AppENDtx A
=[i'Jt"'-"' -
e-i$T/z)
=671' i n(a44)
(aT/2)
Note: FQw) is the specfum
of a single purse;
rF(7co)z
gnl
isthe power
spectral den-
sity of a random purse
train assuming positive
and negative pulses
are equalty likery
and occur independently.
4.3 Dl cl TAL
BIpHASE:
FtcURE 4.19
- l T<t <O
o <t <l r
otherwise
0 (t/z)r
| . . . .
I
F(7'to)=
!
"-*dr-
J
**i*dt
_(t / ?\ T
0
_
rl(
/2laT
_
s-iTrz)ar * 11
A.4 FBAME
ACQUISITION
TIME
OF SINGLE.BIT
FRAME
CODE:
EOUATION
4.10
Framing is established
by successively
examining one bit position
after another until
a sufficiently long framing pattern
is detected.
In this derivation it is assumed that the
framing pattern
alternates
I's and 0's. Furthermore,
it is assumed that when beginning
to test a
particular
bit position
for ftaming, the value of the first appearance
is saved
and compared
to the second appearance.
Thus the minimum
time ao reject an invalid
framing position
is one frame time. If we denote
byp the probability
of a I and by
s
I
r
_
p the probability
that a 0 is received
first, the average number of ftames re-
quired
to receive
a mismatch is
^r=l:,
=[ * ) "
=[.,,')'*'[T)
FRAME ACQUISITION TIME OF SINGLE"BIT FRAME CODE: EQUATION 4,10 575
Ao
=
(l) (probability of mismatch
at end of first frame)
+ (2) (probability of mismatch at end of second frame)
+
(3) (probability of mismatch at end of third frame) . . .
=
(1)4 + (zxl
*
q)p
+
(3Xl
-
q)(r
-
p)q
+( aXl
- i l 20- i l p+. . .
=
(1)4 +
Q)pz
+
(3)pqz
+
{4)p3
q
+
(5)p'q3 + . .
=(q+2p?
+ps\ [ I +(Z)ps+(3)pzqz + (4)p3q3 +. . . ]
=
[7
-
p + Zpz + p(l-p)1(1 +
pq + pzqz + psqt + . . .)'
l +p 3
=
tr*P
1 + p
=' =-
r _pq
Similarly, if a I is received ftrst, the expected
number of frames before receiving a
mismatch is
l + a
At=T_fi
The overall average number of frames required to detect a mismatch is
A= qAo* pA1
_ I
+Zp g
1 - p q
If we assume a random starting
point in a frame with N bits, the average number of
bits that must be tested before the true framing bit is encountered is
Frame time
= (l
/Ztl)(A.
M
+ 1/2N
=1/ zN( A. N+
1) bi t t i mes
576 AppENDtx A
If I's and0's are equally likely (p =q=l),A =
2 so frame time
=Ap
+ t/2Nbittimes
(Equation
4.10).
A.5 FRAME
ACQUISITION
TIME OF SINGLE.BIT FRAN,IING
CODE
USING A PARALLEL
$EARGH: EQUAT|ON
4.11
This framing algorithm
as$umes that all bit positions
in a frame are simultaneously
scanned for the framing pattern.
If we assume that an alternating-bit
frame code is used
and that I
's
and 0's in the information
bits are equally likely, th-e probabiliry
rhat a par_
ticular information
bit does not produce
a framing violation in n frames is
/ r Y
o^=l i l
The probability
that a framing violatidn iuJ u**n received in n or less frames is
I
-p,'
The probability
that all N
-
I information
bits in a frame produce
a framing
violation in rz or less frames is
o.=f,-fri]-'
(4.l
r)
A.6 FRAME ACQUISITION
TIME
OF MULTIBIT FRAME
GODE:
EQUATION
4.13
N= Iength
of frame including framing code
,L
=
length of framing
code
p
=
G)L:
probability
of matching frame code with random data
The expected
number of frames examined
before a particular
frame position
matches
the frame code follows:
l - P ) + " '
. )
mis-
A
= (OXl
-p)
+ (l)p(t
-
p) + (Dpz(
:
(l
-
p)p(I
+ Zp
-t
3p? + 4p3 + . .
=
(1
-p)p(l
+p + pz +p3 + . . .)?
/ ' r \ 2
=( r - p) p
l + |
[ ' - pJ
_ P
l - p
A.7 PATHFINDING TIME: EOUATION s.11 577
The average number of bits that pass before the frame position is detected
(assuming
a random starting
point and the test field is moved one bit position when a mismatch
occurs) is
given as
r.d/otzt'*t
.
t
,,
=T:769-1''
(4. 13)
Iy'afe.. Equation 4.13 with Z
=
I is not identical to Equation 4. l0 because Equation 4.13
assumes a fixed frame code while Equation 4.10 assumes an alternating code.
A.7 PATHFINDING TIME: EQUATION 5.11
Assume that all paths through a switch are independently busy with probability p. Let
the probability that a path is not busy be denoted by q= |
-p.
The probabilityp; that
exactly i
paths are tested before an idle one is found is the probability that the first
i
-
| are busv and the ith is notr
P'= P(i-t)n
The expected number of
paths tested before an idle path is found is
ruo
*
(1)a +
(Z)pq
+
(3)pzq+
. . . +
(k)pk*Lq
+
(k)pk
where the last term represents the expectation that all
possible paths k are unavailable.
A closed form for A is determined as
r=ftruJraxnn*it
At =( 1* pXl + 2p+3p2 +. . . + kpk- r +kpk1
=
I + p+ pz + pj +, . . +pk- '
I
k (
r
\
t - p
'
[ t
- oJ
=l :
Po
l - p
(s. l l )
578 APpENDtx A
Figure A.l.
probabiliqr
graph
of No, 4 ESS four_stage
space switch.
A,8 LESS SPACE
STAGE BLOCKING
PROBABILITY
5.21
The most straightforward
way to calculate the blocking probability
of the four-stage
space shucture shown in the probability
graph
of Figure A.I is to list all elementary,
mutually exclusive probability
terms, determine their probabilities
of occurrence,
and
generate
the sum of those that represent blocking. Because there are eight links, which
are.assumed to be independently
busy or idle, there are zs
=
256 elementary
terms.
Rather than laboriously list them all, the analysis can be greatly
simplified, with a risk
oJ miscounting blocking terms, by grouping
the terms according to the number
of busy
links- The following table lists the groups
and the correspondinf numbers
of terms that
block and do not block.
Number of
Busy Links I
Combinations
(:)
Number that Do Not
Block
Number That
Block
0
1
2
3
4
5
6
7
I
1
I
28
56
70
56
28
B
1
256
0
0
2
1 6
50
52
2B
I
1
157
1
I
26
40
20
4
0
0
0
99
Entries in the last two columns are determined
by analyzing the topology of the net_
work. For example, when two links are busy, onry 2 of the 2g combinations produce
blocking (
I and 2 and 7 and 8). when three links are busy, I 6 of the combinations pro-
duce blocking. To determine the remaining
enhies in column 3, it is easier to deter-
mine the number
of combinations
that do not block and subtract from the total. For
example,
when only three links are idle, there are 4 of 56 combinations that do not
block. using the entries in column 3, the blocking probability
is determined as
B
=
Zpz q6 + l6p3 qs + S}pa qa + Sbps qt
t Zgp6 qz +
gp7
q
+ pg
wherep is the probabiliry
that a link is busy and q
=
|
_
p is the probability
that it is
idle' All ofthe interstage links are equally loaded
because there is no concentration or
expansion in the stages.
APPENDX B
ENCODI NG/ DECODI NG
ALGORITHMS
FOR
SEGMENTED
PCM
8.1 EIGHT-BIT p255 CODE
The encoded representations of
p255 PCM codewords use a sign-magnitude
format
wherein I bit identifies the sample polarity
and the remaining bits specify the magni-
tude of the sample. The 7 magnitude bits are conveniently partitioned into a 3-bit seg-
ment identifier S and a 4-bit
quantizing step identifier
Q.
Thus the basic structure of
an 8-bit
p255 PCM codeword is shown
in Figure B.l.
In the following descriptions of encoding and decoding algorithms, it is assumed,
for convenience in using integer representations, that analog input signals are scaled
to amaximumamplitude of 8159. Furthermore, all amplitudes
and segmentidentifiers
are assumed to be encoded using conventional binary representations. The actual en-
coders used in T1 transmissiOn systems, however, complement the codewords to in-
crease the density of I's in a transmitted
bit stream.
8.1.1 Al gori thm 1: Di rect Encodl ng
(Tabl e 8.1)
polar.
[o for uositive sample values
tty Dtt r
=
ll
for negative sample
roalues
Given a sample value with a magnitude.x, the first step in the magnitude
encoding
process is to determine the segment identifier S. The major segments are identified by
the segment endpoints: 31, 95, 223,479,991,2015,4063, and 8159' Thus 'S can be
I I I
v\-rJ\-1/--/
s Q
Eight-bit
p225 PCM code format.
P
Figure B.l.
580 APFENDIX B
TABLE 8.1 Piecewlee Linear
Approxlmation
to
FZSS
Compoundlngd
Quantization Endpoints by Segment Code S
000 001
1 1 1 1 1 0 101 100 011 010
Quantization
Code Q
0
1
2
3
4
E
6
7
I
I
1 0
1 1
1 2
1 3
1 4
1 5
0
1
3
5
7
I
1 1
1 3
1 5
1 7
1 9
21
23
25
27
29
31
31
35
39
43
47
51
55
59
63
67
71
75
79
83
87
91
95
95
103
1 1 1
1 1 9
127
135
143
151
159
167
175
183
191
199
2Q7
215
223
223
239
255
271
287
303
319
335
351
367
383
399
415
431
447
463
479
479
51 1
543
s7s
607
639i
671'
703
735
767
799
831
863
89s
927
959
991
991
1 055
1 1 1 9
1 183
1347
131 1
1375
1439
1503
1567
1631
1 695
1759
1 823
1887
1 951
2015
asample
values are relerenced to a
Iujl--scle
yaly6
oj 8159. Nogative samples are encoded in sign-magnitud6
formet with a polarlty
bit of 1. In actuel transmission
the cod6s alre inverted to in"reaie tn" o"nsrty of 1,s when
low signal amplitude$ are ncoded. Anelog output samples ere docoded as th6 center of the encodd
quantization
interval.
Quantization error is i'he difference between the teconstructoo out[ut vatu6 ancl th.
origir-lal input sample value.
, !
determined
as the smallest endpoint
that is greater
than the sample value x. Here, s is
equal to the smallest a such that
x < 6 4 - T - 3 3
a = 0 , I , . . . , 7
After the major segment containing the sample value has been determined,
the par-
ticular quantization
interval within the major segment must be identified. As a first
step a residue r is determined as the difference
between the input amplitude and the
lower endpoint
of the segment:
S = 0
S = I , 2 , , . . . 7
The value of
Q
can now be determined as the quantization
interval
containing the
residue r. Here,
fl
is equal to the smallest b such that
, 5=0
S = 1 , 2 , . . . , " 1
where b
=
0, l, . . .
,
15. Notice that this process
identifies quantization
intervals in
segment ^s
=
0 as having upper endpoint$ at l, 3, 5, . . .
,
3t while the other segments
'=fi-
(32.zs-
33)
l zu* t
r<[ 1zs+rxb+
l )
8.1 EIGHT-EITp255CODE 581
have quantization endpoints fhat are multiples of 4, 8, 16,32,64, 128, 256 for S
=
1,
2, 3, 4, 5, 6, 7, respectively.
After,S and
O
have been determined, they are binary encoded into 3 and 4 bits, re-
spectively. The concatenation of S and
O
produces
a 7-bit word that can be conven-
iently represented as an integer between 0 and 127. In essence, this integer identifies
one of the 128
quantization intervals of a compressed signal amplitude.
The decoding proces$ involves assigning the designated
polarity to an analog out-
put sample at the midpoint of the nth quantization intrval n
=
0, l, ' ' .
,
127' Using
the values of S and
Q
directly, we can determine a discrete output sample value as
yn= (2Q
+ 33X2r)
-
33
where n is the integer obtained by concatenating the binary representations of ,5 and
a.
Example 8.1.
An input sample of +242 produces the following codeword:
0 , 3 , 1 = f f i
The decoder output becomes
t $ =( 2 .
1 +3 3 X2 3 ) - 3 3
: 247
which is the midpoint of the forty-ninth
quantization interval from ?39 to 255.
A.1.2 Algorithm 2: Linear Code Conversion
The fundamental reason for using a
p-law compounding characteristic with p255 is
the ease with which the segmented approximation can be digitally converted to and
from a uniform code. This section describes the basic algorithms that implement the
conversions. Thefirst algorithmprovides themeans of implementing 4p255 PCM en-
coder using a 13-bit uniform encoder followed by digital logic to provide the compres-
sion function. The second algorithm indicates how to implement the decoder function
offirst expanding a compressed code into a 13-bit linear code to be used in
generating
the ouQut samples.
Just as in algorithm 1, the polarity bit P is determined as
o
_
I0
for positive sample values
'
-
11
for negative sample values
s82 APPENDIX B
The simplicity of converting from a linear code to a compressed code is most evident
if the linear code is biased by adding the value 33 ro the magnitude of all samples. No-
tice that this bias shifts the encoding range from 0*gl5g ro 33-gl9z. The addition
process
can be performed
directly on the analog samples befare encoding or with digi_
tal logic after encoding. In either case, the general
form of all biased linear code pat-
terns and the corresponding
compressed codes are as follows:
p255
Encoding Table
Biased Linear Input Code
Compressed
Code
0 0 0 1 w x y z a
0 0 l w x y z a b
0 l w x y z a b c
l w x y z a b c d
w x y z a h c d e
x y z a b c d e f
y z a b c d e f s
z a b c d e f g h
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 1
0 0 1 w
0 1 w x
l w x y
0 0 w x y z
0 1 w x y z
1 O w x y z
l 1 w x y z
0 0 w x y z
0 1 w x y z
1 O w x y z
l 1 w x y z
0
0
0
0
From the foregoing
table it can be seen that all biased linear codes have a leading
I that indicates the value of the segment number ,s. specifically, rhe value of s is equal
to 7 minus the number
of leading 0's before the I. The value of
e
is directly available
as the 4 bits (w,
x, y, z) immediately following
the leading 1 . All trailing bits (a-h)
are merely ignored when generating
a compressed code.
In reverse fashion the following table indicates
how to generate
abiased linearcode
from
a compressed code. An unbiased output can be obtained by subtracting
33 from
the biased code:
p255
Decoding Tabte
Compressed
Code
Biased Linear Qutput Code
0
0
0
0
l w x y z 1
w x y z 1 0
x y z 1 0 0
y 2 1 0 0 0
2 1 0 0 0 0
1 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 1
0 0 0 0 0 1 w
0 0 0 0 1 w x
0 0 0 1 w x y
0 0 1 w x y z
0 1 w x y z 1
1 w x y z 1 0
0 0 w x y z
0 1 w x y z
1 0 w x y z
1 1 w x y z
0 0 w x y z
0 1 w x y z
1 0 w x y z
l 1 w x y z
Both of these tables indicate
that 13 bits of the linear code are used to represent the
magnitude of the signal. In chapter 3 it is mentioned that a
p255
pcM
coder has an
amplitude
range equivalent
to l2 bits. The discrepancy occurs because the first quan-
tization interval has length I while all others in the first segment are of length z. Thus
the exha bit is needed only to specify the first quantization
interval. Notice furrher that
the least significant bit in the tables carries no information
but is included only to fa-
cilitate the integer relationships.
kr particular,
the least significant bit in the encoding
table is completely ignored when determining
a compressed code (assuming
that the
B.r EIGHT Brr A-LAW coDe 583
bias is added to the analog sample). Furthermore, the least significant bit in the output
codes is completely specified
by the segment number S. It is a 0 for segment zero and
a I for all other segments.
Example 8.2.
An input code word of +242 is biased to
produce
a value of 275. The binary repre-
sentation of 275 is
0000 I 000 I 00 1 1
(bi asedl i nearcode)
Hence from the encoding table S
=
011 and wxlz
=
fi)01, and the compressed code is
ffi
(compressedcode)
Using the decoding table, this compressed code
produces the following biased lin-
ear output code:
0000 1000 I 1000 ( bi asedl i near out put )
The decimal repre$entation of the foregoing code is 280, which corresponds to an
unbiased output equal to +247.
8.2 EIGHT BIT A-LAW CODE
The following compounding algorithms
for segmentedAlaw codes use the same ba-
sic
procedures
a$ those presented for the
p255 codes. One difference that does occur,
however, involves the elimination of a bias in the linear code for conversion to and
from a compressed code. Another difference occurs in the use of the integer 4096 as
the maximum amplitude of a sample in an Alaw representation. If desired, the scale
factors for the two systems can be brought into close agreement by doubling the AJaw
scale to 8192.
As mentioned in Chapter 3, the segmented Alaw code is usually
referred to as a
l3-segment code owing to the exisknce of seven
positive and seven negative seg-
ments with the two segment$ near the origin being colinear. In the following descrip-
tions, however, the first segment for each
polarity is divided into two
pal'ts to produce
eight positive and eight negative segments. This point of view permits a code format
that is identical to the
p255 code format. Thus, a compressed codeword consists of a
sign bit P followed by 3 bits of a segment identifier S and 4 bits of quantizer level
Q.
8.2.1 Al gorl thm 1r Di rect Encodl ng
The segment endpoints of an A-law code are 32,64,128, 256, 5I2, l0Z.,2048, and
4096. Thus for a sample with magnitude x the major segment identifier ,5 can be de-
termined as the smalle$t a such that
584 APPENDIX B
TABLE 8.2 Segmented A-[aw Encodlng Table
Ouantization Endpoints by Segment Code
1 1 1 1 1 0 101 011 010 001
Quantization
Code
0
1
2
3
4
5
6
7
B
I
1 0
1 1
1 2
1 3
14
15
0
2
4
6
B
10
12
14
16
1 8
20
22
24
26
28
30
32
32
34
36
38
40
42
44
46
48
50
52
5,4
56
58
60
62
84
64
68
72.
76
80
84
BB
92
96
100
104
108
112
1 1 6
120
1?4
128
128
136
144
152
160
168
176
184
19e
200
208
216
224
232
240
248
256
256
272
288
304
320
336
3s2
368
384
400
416
432
M8
464
480
496
512
512
544
576
608
640
672
704
736
768
800
832
864
896
928
960
992
1024
Not6i Every oth6r bit is inverted for tran8mi8slon.
x < 3 2 ' 2 "
a = 0 , l , . . .
, 7
After S has been determined, the residue r can be obtained as
,
=[ *
l x -
1 6 . 2 s
S = 0
S = 1 , 2 , . . . , 7
The value of
B
can then be determined
as the smallest b such that
S = 0
S = 1 , 2 , . . . , 7
S = 0
S = I , 2 , . . . , 7
Exampla
8.8.
An input sample of +121 produces
the following codeword:
'.{#[..'i,
Just as in the case for
p255
coding, an Alaw magnitude can be represented as an in-
tegern
=
0, l , .. ' ,121 deri vedfromtheconcatenati on
of 3 s-bi ts and4
p-bi ts.
An
output magnitude
I
can then be expressed as
[ zB+
r
, , =l t t , n+
16f )
8.2 EIGHT BIT A.LAW CODE
ffi
=+46i ndeci mal
The decoder output becomes
585
)+e
=
2s(14 + te
|)
_ |
tr,,
which is the midpoint of the quantization interval from 120 to 7M.
8.2.2 Algorithm 2: Linear Code Converslon
The following tables indicate how to convert a lZ-bit linear code directly into a com-
pressed.A-law code. The algorithm is basically the same as for the p255 conversion
except that biasing the linear code is unnece$sary and a first segment code does not
have a leading L Thus the segment numberS can be determined as 7 minus the number
of leading zeros as before. The
p
field data are obtained as the 4 bits
(wxyz)
immedi-
ately following the leading l, except when ,S
=
0, in which case the
Q
field is contained
in the 4 bits following the seven leading 0's.
A.Law Encodlng Tabl
Linear Code Compressed Code
The following table provides
the means of
generating a linear codeword directly
from a compressed codeword. The output value conesponds to the middle of the quan-
tization interval designated by S and
p.
A-Law Decoding Table
Compressed Qode
Linear Output Code
0 0 0 0 0 0 0 w x y z a 0 0 0 w x y z
0 0 0 0 0 0 1 w x y z a 0 0 1 w x y z
0 0 0 0 0 1 w x y z a b 0 1 O w x y z
0 0 0 0 1 w x y z a b c 0 1 1 w x y z
O 0 0 l w x y z a b c d 1 O 0 w x y z
0 O l w x y z a b c d e 1 0 1 w x y z
O l w x y z a b c d e f l l 0 w x y z
l w x y z a b c d e f g l l l w x y z
0 0 0 w x y z 0 0 0 0 0 0 0 w x y z 1
0 0 1 w x y z 0 0 0 0 0 0 1 w x y z 1
0 1 O w x y z A 0 0 0 0 1 w x y z l O
0 1 1 w x y z 0 0 0 0 1 w x y z 1 0 0
1 0 0 w x y z 0 0 0 1 w x y z 1 0 0 0
1 0 1 w x y z 0 0 1 w x y z 1 0 0 0 0
' f 1 O w x y z O 1 w x y z 1 0 0 0 0 0
1 1 1 w x y z 1 w x y z 1 0 0 0 0 0 0
586 AppENDtx B
Each of these tables relates 12 bits of magnitude in a linear code to a compressed
code with 7 bits of magnitude. Notice, however, that the least significant bit of the en-
coder is always ignored. Thus the encoder needs only 1l bits of resolution if all of its
outputs are immediately compressed.
If any signal processing (such
as adding two sig-
nals together) is to take place
before compre$$ion, however, the extra bit is useful in
reducing the composite quantization
error.
Example
The previously
used sample value Lzr is represented in binary form as
000001 I 11001. From the
encoding rable,S =
010 and
e
=
I I 10. Thus the compressed
code word is
ffi
using the decoding table, the linear ourput can be determined as 000001 I l l0l0,
which is 122 decimal.
APPENDX
C
ANALYTIC FUNDAMENTALS OF
DIGITAL TRANSMISSION
C.l PULSE SPECTHA
This section presents
the f'requency $pectra of common pulse
waveforms used for digi-
tal transmission. These are square pulses as generated
at a source. Since the spectra of
square pulses have infinite frequency content, the spectra presented here do not cor-
respond to
pulse responses at the output ofa channel where the pulse
shapes have been
altered by bandlimiting filters. In the next section channel output pulse responses are
described. Then the necessary combinations of input pulse
shapes and filter designs
to
produce particular output pulses are considered.
The various pulse shapes and corresponding frequency spectra are presented
in Fig-
ure C.1. In deriving the spectra, the following conditions and assumptions were made:
1. All pulses have equal energy.
2. All systems signal at rutfr IlT.
3. The waveform$ shown are used to encode a 1.
4. Opposite
polarities are used for a 0.
5. It is equally likely for l's and 0's to occur and to occur at random
(complete
independence).
C. 2 CHANNELOUTPUTPUL$E RESPONSES
Although a digital transmission system can be designed to produce
a variety of output
pulse responses, the most common is defined as
sin(nr,zl) cos(ant/7)
v--\0 |
-
' /rc\r/
IEI/T I-(Zat/T)z
where 1/T is the signaling rate and cr, i$ an excess bandwidth factor between 0 and 1.
Equation C.l represents the response of a raised-cosine chamel-so-called because
the frequency spectrum conesponding to yrc(t) in Equation C.l is
(c.l )
588 APPENDTx c
NRZ l l nc
code
Dlelul
biphrre
| + D Oorrahtivr
ficodhg
|
*
pr
Cotolrtiw
Encodlng
- | r
t r
f l t l
*
l l r l { +
r
: 2
- l
F
(lr^rl -
*3
, f r t -
r
- t r <r <o
__r
0( r { i ?
- 2 - r t t
r r pl *S*n' r Sr
_ t
|
-Dcornhth'.
I l' ,
'ncodrns
TE]_
t't:t:*'-r:1,**'o
-3 -2 -1
I ? 3
rqr.,r=ffipII
l"
r;-r
-3 -2 -l
r ? 3
F(r.l
-
tf;t .rnr
trtt
_l _
I
fl d
";
_ _
I
?
-F*' *-f
E* r * T
-3 -z ..r
r
l*ot
-t$rn
t'n *n rfl
Figure C.1. Spectra of common pulse
shapes.
F,"(f)
=
I
C.2 GHANNELOUTPUTFULSEHE$PONSES 589
, o * I - o
' { ' -
2 T
1 - o * r f l < 1 * o
2T
- ' ' ' -
2T
=il'.*'[#-r*r]]
:"",'[#
q#)
=
Q otherwise
(c.2)
The origin ofthe apellation
"raised
cosine" is apparent in the third line ofEquation
C,2,
The parameter u in Equations C.l and C.2 is referred to as an excess-bandwidth
parameter.
If o
=
0, the spectrum defined in Equation C.2 is exactly equal to the theo-
retical rninimum bandwidth ll2T for signaling rate llT. As fl increases from 0 to l,
the excess spectrum width increases to 1(X)7o. Raised-cosine channel specFa are illus-
trated in Figure C.2 for several values of cr.
Practical systems are typically designed for excess bandwidths of 307o or more for
several reasons. First,
"brick
wall" filters needed to produce the infinite attenuation
slopes implied by fl
=
0 are physically unrealizable. Second, as shown in Figure C.3,
the time-domain pulse response for small values of u, exhibits large amounts of ring-
ing. Slight errors in the sample times cause significant degradations in performance
due to intersymbol interference. Third a slight deviation in the signaling rate from the
design rate also produces significant intersymbol interference.
It must be emphasized that Equation C.2 defines the desired spectrum at the output
ofthe channel (the input to the decision circuit). Thus the desired response results from
a combination of the input pulse spectrum and the channel filter responses. Often, the
input pulse
$pectrum arises from a square pulse
of duration T:
-1
l o = 0
I
- l
tr'igure C.2.
I
I
* * "-.-+
. 5
I
i
t
I
I
t
+
0.
Y,"lfl
Raised-cosine spectrums fbr various values of c.
590 APPENDTx c
.l
tnput pura"
Ouput
puber
Normrlirrd time I r/Tl
Figure C.3. Raised-cosine pulse
responses for various values of s.
[ t
t d<+r
r,(r)
=
lo
otherwise
(c.3)
The frequency spectf,um corresponding
to x"(r) in Equation c.3 is the
,.sin(.r)/x"
spectrum also referred to as a
"sine"
function:
(c.4)
when the channel input spectrrrm is as defined in Equation c.4, the filter function
of a channel to produce
a raised-cosine ouryut is determined as
*"cn=I4ffi
=
f sinc(n/T)
H(n=ffi (c.5)
The channel filter functions implied in Equation c.5 are shown in Figure c.4 for
the same values of fl, shown in Figure c.2. Figure c.2 displays channel output spectra
while Figure C.4 displays corresponding frequency-domain transfer functions for the
channel.
Normrlirrd f lugurrty
l.f T
I
figure C.4. Channel filterresponses needed to produce raised-cosine outputs when excited
by sin(4172)/(rn72) pulses.
! *l t l
C,2 CHANNELOUTPUTPULSEHESPoNSES 591
I
-'5
no-"Lr*u.rlv
urr
Ftgure C.5.
"Optimum"
ffansmit and receive filter functions for raised-cosine response with
s
=
0,3 and sin(ro72)/(coZl?) excitation.
As mentioned in Chapter 3, the design of a smoothing filter in a digital voice de-
coder sometimes requires a modification like that defined in Equation C.5. In fhe case
mentioned, the ideal
"flat"
filter for reconstructing speech from narrow impulselike
samples should be modified by dividing the flat re$ponse by the $pectrum of the fi-
nite-width samples (Equation
C.4). When the samples are narrower than the signaling
interval, the sin(x)/x response is essentially flat across the bandwidth of the filter.
However, when the samples are made to last for the entire duration of the sample in-
terval, the filter response should be
"peaked"
to compensate for sin(x)/x roll-off.
As discussed in Chapter 6, a channel filter function is usually partitioned between
the transmitter and the receiver. The so-called
"optimum"
filter
partitioning
occurs
when the receive filter response is the square root of the desired output response
Y*(fl, and the transmit filter response is whatever is necessary to han$form the channel
input spectrum into the complex conjugate of the receive filter response. (The
output 'ryec-
trum of the transmitter is also the square root of the desired channel response.)
rilhen
the
channel input is a
pulse of duration 7, the
'bptimum'
filter functions are defined as
H**(.f)
=
{I' "(,f)}r/2
HwU)=#
(c.6)
(c.7)
Notice that the transmit spectrum resulting from HnE(f) in Equation C.7 is equal
to the square root of the desired output response, no matter what the input pulse
shape
is. Hence, when optimum partitioning
is used, the transmit spectrum and the receiver
design are independent of the channel excitation.
The transmit and receive filter functions for square-wave excitations
(Equations
C.6 and C.7) arc shown in Figure C.5.* Notice that the transmit filter function has
peaking at frequencies other than direct current. The implication for midband attenu-
ation is one of the reasons why
"optimum"
partitioning
may not be optimum in a sys-
tem with source
power
limitations.
*This
discussion assumes baseband transmission. The principles
are easily extended to carrier-based
systems by translsting the filter functions to the carrier frequency.
592 AppENDtx c
C.2.1 Optlmum Filterlng for Minimum Shift Keylng
As discussed in chapter 6, minimum shift keyed (MSK)
modulation can be repre-
sented as quadrature
channel modulation with baseband excitation defined as cosine
pulse
shapes:
.,,=f-[*J
ttts+T
otherwise
(c.8)
(c.10)
( c. l l )
(c.12)
The transform ofx"(r) is
v, A f
r ' ) cosn/ r
x.ul
=[zrJ
|
-ffi
(c'e)
When a raised-cosine
ouQut response and optimum filter partitioning
is desired,
the MSK filter functions are
H*"(f)
=lYn(frlt/z
ro(f)=w
4 cos(nt/T\
Y,\I)=
n|__eiffi
The transmit spectrum and the receiver of an optimally partitioned
MSK system are
identical to counterparts in an optimally partitioned,
offset-keyed 4-psK system!
C.2.2 Partlal-Response
Systems
As another digital hansmission system design example, consider a I + D partial-re-
sponse system. The desired time response of the channel is defined in Equation C.12
and the associated frequency spectnrm in Equation C. 13. The pulse
response of a I + D
partial-response
channel is shown in Figure C.6. Notice that a single pulse
contributes
equally to the response at two successive sample times but crosses zero at all other
sample times:
fcos
nlr
Y.ffi =i
Io
nt+i
otherwise
(c.13)
C.? cHANNELoUTPUT PULSE BESPoNSES 593
1
- 1
I
r -
I
I
I
F---tnprt
putrr
I
puls6Jrrlrl
-2.5 -r.6
t . 5
2. 5
Noffirlirid tiffi6 {tr?}
Figure C.6. Output pulse of I + D partial-response
channel.
Optimum filter
pafiitioning
is again achieved with a receive filter having an ampli-
tude response equal to the square root ofthe desired output respon$e Y"(/). Ifthe chan-
nel is excited by square pulses
of duration ?, as defined in Equation C.3, the optimum
filter functions are
Hnx(fl
=[Y"ff)]L/z
(c.r4)
3.S
vl s+
vts+
(c. r5)
H*(f)
-Lv,(f)1"'
X'fl
where Xr(f) is defined in Equation C.4.
The optimum filter functions defined in Equations C.14 and C.l5 are shown in Fig-
ure C.7 along with the desired output response of the channel. Notice that, unlike the
full-response
(raised-cosine)
systems, optimum partitioning
of a
partial-response sys-
tem does not require peaking of the transmit filter.
o.5 1
Normelitrd f ruqurncy l.f Tl
Figure C.7. Spectra of 1 + D PRS channel and
"optimum"
filter responses for
sin(aT | 2) |
(aT
| 2) excitation.
- t
c.3
594 APPENDIX C
ERROR RATE ANALYSES: BASEBAND SYSTEMS
C.3.1 Bi naryTransmi ssi on
consider the receiver model of the digital transmission system shown in Figure c.g.
The receiver consists of two parts:
signal processing
circuitry and a data detector
(de-
cision circuit). For the time being, assume that the output of the signal processing
cir-
cuitproducesapulseof amplitude+vwhena I istransmittedandapulseof
amplitude
*v
when a 0 is transmifted. obviously, the detector merely examines the polarity
of
its input at the sample times defined by the sample clock. A decision error occurs if
noise at the sample times has an amplitude greater than v and a polarity
opposite to
the transmitted pulse.
The most commonly analyzed type of noise is assumed to have a Gaussian or nor-
mal probability
distribution. Thus the
probability
of enor
ps
can be determined as
po
=
^-l- l
r-'tzo' dt
(c.16)
-
\txr o
"v
where oz is the rms noise power
at the detector. Using the enor function,
,
l t z
erf
a=| l e-r- dt
.vr,
;
Equation C.16 is sometimes rewritten as
P" =| ( 1- er f z) (c. 18)
where.e
=
v/"12o. The error probability
Ps can also be expressed in terms of the com-
plimentary
error function :
;t4#
(c.r7)
Noire
Figure C.8. Digital receiver model,
ERROB BATE ANALY$ES: BASEBAND SYSTEMS 595
P"=f erf cz
(C. 19)
where
er f c4- 1- er f z -
{ 2o
In lieu of evaluating the integral in Equation C.17 or C.19
(which
has no closed
form solution), the error functions can be approximated as
p-z
srfs
g
-.:F-
( e' vn;
(e
>> l;
C.3.2 Multllevel Transmlssion
The error rate of a multilevel baseband system is easily determined by an appropriate
reduction in the error distance. If the maximum amplitude is V, the error distance d be-
tween equally spaced levels at the detector is
d = *
""=['=;)F}.'{.t#)
(c. 21)
where L is the number of levels. Adjusting the error distance V of a binary system to
that defined in Equation C.21
provides
the error rate of a multilevel system as
(c.22)
where the factor
(L
*
1)/Z, reflects fhe fact that interior signal levels are vulnerable to
both
positive
and negative noise and the factor Illog2 L arises because the multilevel
system is assumed to be coded so symbol errors produce
single-bit errors
(log2.L
is the
number of bits per
symbol).
Equation C.22 reveals that, with respect to
peak signal-to-noise ratios at the detec-
tor, L-level transmission incurs a
penalty of 20log10(/-
-
l) decibels. If Vis increased
by a factor L
-
I, the error rate of the llevel system is identical to the enor rate of the
binary system
[except
for the factors lftogyL and
(f
-
1/L, which typically represent
only a few tenths of a decibell.
Equation C.22 relates error rate to the
peak signal power IE. To determine the error
rate with respect to average power, the average power of an L-level system is deter-
mined by averaging the
power
associated with the various pulse amplitude levels:
596 APFENDTx c
(Il)"",
. . F ]
7 ? 2
3L[*l
.[#J
' \ / \
=ffiT^Qi-t)? (c.23)
where the levels
v
r _,
{ +t , +3, . . . , *
( r -
1) }
are assumed to be equally likely.
C.3.3 Energy.per-Blt-to-Noiee-Den$ity
Railos
The foregoing error rate equations relate PB to the signal energy at the sample times
and the rms noise power
at the detector. When comparing various digital modulation
formats, multilevel systems in particular,
it is more relevant to relate error perform-
ance to signal power
and noise power
at the input to the receiver (in
front of the signal
processing
circuitry). As a first step in developing an error rate equation based on sig-
nal-to-noise ratios at the receiver input, the noise power
at the detector is determined.
The variance ofthe noise o2 used in the previous
equations is exactly equal to the
rms power
that would be measured at the detector in the absence of a signal. This noise
power
can be determined analytically as
o":
I
lHffi(|No)t' df
(c.24)
=No' Br v
(c.2s)
where-.rNs is the one-sided noise power
spectral density in watts per hertz and
Brv
=
Jo
=lH(fllz
df is the noise-equivalent
bandwidth or simply the noise bandwidth
of the receiver filter function
Hll).
In Equation c.?5, the noise source is assumed to be white. That is, a uniform spec-
tral density exists across the entire band of interest. This noise may exist in the tans-
=
ND
J
IHU)I' df
0
C.3 EHHOR RATE ANALYSES: BASEBAND SYSTEMS 597
mission medium itself, or it may occur in the
"front-end"
amplifier of the receiver. If
the rms noise power passing through a $quare
(ideal)
frlter of bandwidth B is meas-
ured, a reading of (IV$(B)
watts would be obtained. Thus B1y represents the bandwidth
of a perfectly
square filter that passes the same amount of noise as the receiver filter
H(l)
lH(f)
may be decidedly nonsquare in its amplitude responsel.
The receiver function H(fl necessarily provides
a compromise between two con-
flicting objectives. First, it must minimize the amount of noise passed to the detector
(i.e.,
minimize Ery). Second, the difference between sample values
(+Vand
*I4
must
be maximized. Obviously, the signal-to-noise ratio at the detector
(Vld)
should be
maximized to minimize the error probability.
A classical result of digital communica-
tion theory states that V/o is maximized when
HIJ) is'?natched" to the received sig-
nal.
When viewed in the time domain, a
"matched
filter" is implemented by correlating
(multiplying)
the received signal with each of the receivable (noise-free) pulse shapes.
The outputs of the correlators are integrated across a signal interval to determine the
overall average correlation during the interval. The integrator with the maximum out-
put indicates the most likely symbol to have been tr-ansmitted.
In most systems all signals or pulses have the same shape but differ only in ampli-
tude and polarity.
Thus a single matched filter can be used. Detection merely involves
comparing the matched frlter output to appropriate decision levels. The output of a sin-
gle correlator ft(r) and its integrator is expressed as
where s(r) is the signal or pulse shape being measured. Notice that Vis, in essence, a
measure of the energy in the signal over the signaling interval 7.
When analyzed in the frequency domain, amatched filterresponse f(/) is the com-
plex conjugate of the channel pulse spectrum S(f). Thus the matched filter output
Y(/) can be expressed in the frequency domain as
T
V=) s( t ) h( t ) dt
0
=
J
ls(r)F dr
0
YU)=H(f)' SU)
=
S.(fl
.
,S(fl
(c.26)
(C.zt1
Frequency-domain representations are most convenient when the transmit signal is
stictly bandlimited, implying that the duration of the
pulse
rcsponse is theoretically
unlimited. In this case the energy in a pulse (e.g., raised cosine) is directly proportional
598 APPENDTx c
to the detector voltage at the optimum sample time. Hence optimum detection is
achieved by merely sampling the output of the receive filter at the proper
time.
Using the relationship of Equation C.25 and rhe parameterE5
to represent symbol
energy, we express the binary error rate Equation C.I 9 as
P"
=
j
erf(z)
(c.28)
where
z2
=
Es/(Nfiil.
Notice that for a given
system (fixed
81,,), the error rate is dependent only on the
ratio of the symbol energy E5 and the noise density Ns. This ratio is commonly referred
to as a signal-to-noise ratio, although it is not a signal-power-to-noise-power
measure-
ment. Equation c.?8 is the preferred
form of the enor rate equation for comparing dif-
ferent modulation schemes. In a binary scheme, the symbol energy Es is equal to the
energy per bit Es.
As an example of a specific sy$tem, consider a baseband raised-cosine
channel
with optimum partitioning.
The output spectrum of the hansmitter is the square root
of the raised-cosine spectrum Y*(fl defined in Equation c.2. The matched receiver fil-
ter also has a square root of a raised-cosine response (Equation
C.6). Hence the noise
bandwidth of the receiver can be determined as
;
BN=
J
tHU)t' df
0
(l +fl )/27
=
!
tY*u)t df
0
(independent
of u)
(c.?9)
As defined in Equation c.l and shown in Figure c.3, the (normalized)
peak
sample
value at the detector of a raised-cosine pulse is L Using unnormalized pulses
of am-
plifude Es, we can determine the error rate of a binary (+Es,
-Es)
raised-cosine chan-
nel as
Pu
=
|
erfc(z)
(c.30)
where
e2
=
(EslNg)f
and Zis the duration of a signal inrerval.
Although Equation C.30 was derived for a raised-cosine channel, it is more general
in that it is applicable to any binary system using antipodal signaling. Thus Equation
c.30 is plotted in Figure 4.23 as the best perfonnance
achievable by *y digital hans-
I
2T
c.3 EHFORRATEANALYSES:BASEBANDSYSTEMS 5gg
mission system detecting one pulse at a time.
(Lower
error rates are possible
when re-
dundant signals or enror-coffecting codes are used.)
The enor rate performance presented
in Equation C.22 for multilevel systems is
based on noise power
at the detector. As long as the signal bandwidth is identical for
all systems, Equation C.22 is valid since the noise bandwidth of the receiver is inde-
pendent
of the number of levels. When the signaling rate is held constant, however,
the bit rate increases with the number of levels. To compare multilevel systems on the
basis of a
given
data rate, the signaling interval 7 and, hence, the noise bandwidths
must be adjusted accordingly.
If 7 is the signaling interval for a two-level system, the signaling interval
I
for an
/,-level system providing
the same data rate is determined as
Tr =Tl ogr L (c. 31)
Using the noise bandwidth of a raised-cosine filter in Equation C.29, we extend Equa-
tion C.22 to multilevel systems as
"=[,#)[#)-'*o
v/(L
*
L)
.
<r@rn
Equation C.3? can be simplified and
presented
in a more customary form by using the
relationship that energy per symbol Er= Ealog2 L
=
IPTI, where E6 is the energy per
bit:
"=[,#J[tt)*'r.r
(c.33)
(l ogrL)r/z
(Eo\"
r =
1- 1 l 1r
I
\ ' 0 , /
Equation C.33 is plotted in Figure 4.26 for 2, 4, 8, and 16 levels. These curves rep-
resent the ideal relative
performances
of multilevel baseband systems at a constant
data rate. The bandwidth requirements of the higher-level systems decrease in propor-
tion to log2L.
600 APPENDTx c
Equation c.33 and Figure 4.26 represent the performance
of multilevel systems
with respect to E6lNp (common
data rate but different bandwidths). The following re-
lationship can be used to determine error rates with respect to signal-to-noise ratios at
the decision circuit (different
data rares but common bandwidth):
Slgp:
siTal
Powet
nolse power
_(E)(rogzL)(r/T)
(No)0/2r)
/ F \
=(2)(l og2z)l #
|
l-"0
i
where ll?Tpis the minimum
"Nyquist"
bandwidth of the signal.
The sNR obtained in Equation c.34 repre$ents the ratio of signal power
at the sam-
ple
time to noise at the detector. This ratio is sometimes refened to as a postdetection
sNR because it is the signal-to-noise
ratio at the output ofthe signal processing
cir-
cuiffry.
Some communications theorists use predetection
signal-to-noise
ratios in deter-
mining elror rates. Since predetection
SNRs are measured prior
to bandlimiting the
noise, a noise bandwidth must be hypothesized to establish a finite noise power.
com-
monly, a bit rate bandwidth ( l/7) or a Nyquist bandwidth (l/zT)
is specified. The latrer
specification produces
SNRs identical to that in Equation C.34. Exceptions occur with
double-sideband modulation using coherent demodulation (e.g.,
2-psK) where the
predetection
sNR is 3 dB higher than the postdetection
sNR.
(All
signal power is co-
herent to the demodulator carrier reference, but only halfofthe noise is
.,coherent.")
C.3.4 Partlal-Response
Systems
The enor rate equation for a I + D partial-response
system is obtained by incorporat-
ing the following modifications to a full-response
system:
l. The error distance is exacfly one-half the error
full-response system (see
Figure C.6).
2. The noise bandwidth of the receiver (assuming
obtained by integrating IIsx of F4uation C.14:
(c.34)
distance of a corresponding
$quare-root partitioning)
is
| /27
B"=J cos
MTdf
0
_ l
nT
(c.35)
C.4 CARRIEF TRANSMISSION
SYSTEMS 601
Thus the noise bandwidth of a 1 + D patial-respon$e system is 2 dB lower than the
noise bandwidth of a raised-cosine
(full-response)
system obtained in Equation C.29.
Since the error distance is reduced by 6 dB,
partial-response filtering incurs a net pen-
alty of 4 db-with respect to unfiltered signal
power
at the source. In terms of channel
powers,
the
panial-response system incurs a smaller penalty
owing to greater spec-
trum truncation in the hansmit filter.
The difference in channel powers of the two sy$tems can be obtained by integrating
the respective channel power spectra. If square-root partitioning is used, the charurel
power of the partial-response system is 2 dB below the channel power of the corre-
sponding full-response systems. In this case, the difference in channel powers i$ ex-
actly equal to the difference in noise bandwidths because the receiver filter responses
are matched to the respective channel spectra. That is, integration ofchannel spectra
has essentially been accomplished in Equations C.29 and C.35.
When spectrum truncation loss in a transmitter is considered, a square-root pafii-
tioned partial-re$ponse system is only 2 dB worse than that of a corresponding full-
response system. Of course, the partial-response system requires less bandwidth than
the full-response $ystem. For completeness, the erTor rate equation of a square-root
partitioned I + D
partial-response system is
(c.36)
where
{
= (rE/4)(Eb/No)t
/2(logz
L)trz rrt
-
1) and E6 is the energy
per
bit on the chan-
nel .
Equation C.36 is identical to Equation C.33 except for the factorrd4 resulting from
the lower error distance, the lower noise bandwidth, and the lower channel
power of
the partial-re$ponse system.
C.4 CARRIER TRAN$MISSION $YSTEMS
C.4.1 Filter Design
Except for a few relatively uncommon frequency-modulated systems, digitally modu-
lated carrier systems can be designed and analyzed with baseband-equivalent chan-
nels. Carrier-based filters are derived by kanslating the baseband filters to bandpass
filters centered at the carrier frequency. The output
pulse
response of the channel is
determined by the composite of the baseband-equivalent filters. Thus pulse shaping
can be achieved by filtering the baseband signals or the modulated signals. Partition-
ing of the channel filter function is dependent on the application. In all cases the com-
posite channel response is identical to that defined in Equation C.2 for raised-cosine
channels or Equation C.13 for
partial-response systems.
"=['#Jf-r)-'r-t.r
602 APPENDIX c
Q.4.2 Error Bate Anatysis
Error rate analyses of baseband systems can be direcfly applied to carrier $ysrems-
under one important condition: Coherent modulation and demodulation must be used.
For example, coherent demodulation of a 2-pSK signal y(r) =
cos[fitf +
Q(I)]
involves
implementing the following equations;
S(t)
=
lowpass{cos[rol +
Q(r)]
2 cos(rrlr)]
=
lowpass{
[(cos
Q(r)
cos
(Dr
*
sin
S(r)
sin rryl2 cos rof
]
=
lowpass{cos
$(rxl
+ cos rrlr)
-
sin
S(r)
sin 2cor}
=
cos
S(r)
for
Q(t)
=
6
for
S(t)
=
7s
(c.37)
Notice that coherent demodulation involves multiplying the received signal by a local
carrier that is exactly in phase
with the respective incoming signal. Hence coherent de-
modulation is closely related to matched filter detection as presented
in Equations
c.26 and c.27. To complere optimum detection of a digitally modulated signal, the
baseband equivalent of the receiver filter function must also be matched to the carrier
envelope: cos
Q(r).
When coherent demodulation (also
called coherent detection) is
used, the error rate performance
is identical to the analogous baseband system. Hence
the error rate of a coherently detected z-psK system is provided
in Equation c.19 or
c.28.
The correspondence of coherent demodulation to matched filter detection is illus-
trated in Figure c.9. For convenience, digital biphase (diphase)
is compared to an
NRZ baseband signal. The basic principle
also applies to sine wave carriers at any fre-
quency-
The important point to notice in Figure C.9 is that the output of the coherent
demodulator (or
equivalently, the matched filter) is identical to the NRZ sisnal. Fur-
I t l t
l o l r l o l r l
Dffi
I-l
r-l t-l
Mlxcr
(defitodulfiorl
_Jr
[-r
T, H| [ l f ' - " - l J+
f f i r |
hodrm
l-r t-t _-r I
Figure C.9. Coherent demodulation (detection)
of digital biphase signal.
c.4 cARRIEB THANSMtsstoN sysrEMS 603
thermore, the noise power
density coming out of the coherent demodulator is identical
to the noise power density of the baseband system. (Positive-weighted
white noise is
statistically no different than negative-weighted white noise.)
Figure C.9 demonstrates that coherently demodulated carrier-based signals
pro-
duce the same signal-to-noise ratios at the detector as baseband systems, despite the
fact that
(double-sideband)
carrier systems require twice as much bandwidth. Coher-
ent demodulation leads to a receiver noise bandwidttr equal to the baseband-equivalent
noise bandwidth because only one-half of the noise power in the carrier signal band-
width is passed by the coherent demodulator.
(Noise
in the carrier signal bandwidth
that is out of
phase with respect to the coherent reference is ffanslated to twice the car-
rier frequency and therefore eliminated by the low-pass filter, sin ot
'cos
o)f
=
I
sin 2rot.)
C.4.3 QAM Error Rates
The error rate equation of a coherently detected
QAM
system is identical to the error
rate of the conesponding multilevel baseband system applied individually to each
quadrature
channel. Thus the eror performance of a 16-QAM system is provided in
Equation C.33 for I
=
4 levels. In a
QAM
system coherent demodulation causes one-
half of the noise power in the carrier signal bandwidth to show up at the /-channel de-
tector and one-half shows up at the
Q-channel
detector. Of course, the total signal
power is divided in two so the signal-to-noise ratio at the individual detectors is iden-
tical to the carier signal-to-noise ratio
(i.e., predetection
SNR is equal to
postdetection
SNR). The error rate performance
of
QAM
systems is
plotted
in Figure 6.20 in terms
of E/Ns. To relate those results to signal-to-noise ratios, use Equation C.34 using L
as the number of levels on each quadrafure channel.
C.4.4 PSK Error Rates
The error rate of a multilevel PSK system is derived most easily by using quadrature
channel representations for the signals. For example, Figure C.10 displays the regions
of decision errors for a repre$entative signal phase in an 8-PSK signal. The received
signal is processed by two orthogonal phase detectors to produce quadrature signals
Y{t) and Yp(t) in Equations 6.10 and 6.11, respectively.
If the transmitted phase is rd8
(corresponding
to data values 011), a decision error
results if noise causes Ig(0 to go positive at the sample time
[yo(t)
is
positive
down-
ward to represent the
p
(sine)
channel as lagging the 1
(cosine)
channell. The normal-
ized error distance is sin
(n/8).
A decision error also results if noise causes
-Ia(t)
to
exceed Y{t), indicating the phase is greater than
-nl4.
This latter condition can be rep-
resented as a negative value for the hansformed signal Y3(t)
=O.lOiYBft)
+ A.7O7Y6[I).
fHere,
Y6(f) is a
projection of the received signal onto a
-nl4
basis vector
(see
Figure 6. l3)1. Examination of Figure C.l0 reveals that the error distance for this
second type of error is also sin
(d8).
Since the noise variance
yB(f)
is identical to the
604 APPENDIX c
Feglon 2 whero
0lO ir dstsctod
Rcgion I nlrcre
l l l hdstsrsd
Figure C.10. Regions of decision error for 8-pSK signal at ldg
(01
I
).
noise variance of r{f) (by virtue of the 0.707 multipliers), both types of errors are
equally likely.
In general
terms, the error distance of a
psK
system with N
phases
is
v' sir{n/M, where v is the signal amplitude at a derector
(i.e.,
the radius of a
psK
signal constellation). A detection error occurs if noise of the proper polarity
is present
at the output of either of two phase
detectors. A detection error, however, is assumed
to produce
only a single-bit error. The general
expression for the theoretical error rate
for PSK modulation is now determined by modifying Equation C.22 as
"":['chJ*'oo
sin(n/MV
^
r/Zo
The signal amplitude V can be expressed as
where
(c.38)
(c.3e) v=frorros,
r
[+)"'
"=[",[,+)'"
and the rms noise voltage o as
(c.40)
c.4 CARRIER TRANSMISSIoN SYSTEMS 605
for noise in a Nyquist bandwidth.
Combining Equations C.38, C.39, and C.40 relates PSK error rate$ to energy per
bit to noise density on the channel:
(c.41)
/ r r \ . . ^( a, 1"
z
=
sin
| ; lttosz
lOt
/2
| + |
\
' /
l .toj
Equation C.4l is plotted in Figure 6.16 for PSK systems with various numbers of
phases.- To determine error rates with respect to signal-power-to-noise-power ratios,
use the following:
(c.42)
(N> 2)
For 2-P$K systems, the error rates as specified in Equation C,38 or C.41 should be divided by 2 because
only one
phase detector is needed and it produces errors for one
polarity of noise only.
""=['=lo)w'r
sNR=t"-,"h1J
APPENDIX D
TRAFFIC TABLES
Table D.l is a table of maximum offered loads A for various blocking probabilities B
and number of servers N. The blocking probabilities are for infinite $ources with lost
calls cleared
(Erlang-B,
Equation 10.8).
Table D.2 is a table of maximum offered loads A for various blocking probabilities
B, number of servers N, and finite number of sources M. The offered load A is deter-
mined as Mp, where p is the average source activity assuming no calls are cleared. The
blocking probability for finite sources is determined from Equation 12.13.
Note: The following tables were obtained from Telephone Traffic Theory, Ta-
hles, and Charts, Siemens Aktiengesellschaft, Munich, 1970.
TABLE D.l Maxlmum Ofiered Load Versus Band M
MB 0.01 0.0s 0.1 0.5 1. 0 40 30 20 15 1 0
1 .0001 .0005 .001
2 ,014 .032 .046
3 .087 .1s2 .194
4 .23s .362 .439
5 ,452 .649 .762
,00s .010 .020
.105 .1s3 .223
.340 ,45s .602
.701 ,869 1.09
1, 13 1. 36 1. 66
1. 62 1. 91 2. 28
2.16 2.50 2.94
2.73 3.13 3.60
3.33 3,78 4.34
3.96 4.46 5.08
4.61 5,16 5.84
5.28 s.88 6.61
5.96 6.61 7.40
6.66 7.35 8.20
7.38 8.11 9.01
8.10 8.88 9.83
8.83 9.65 10.7
9. 58 10. 4 11. 5
10, 3 11. 2 1e. 3
1 1 . 1 12. 0 13. 2
.728 .9S6 1.15
1. 05 1. 39 1. 58
1. 42 1. 83 ?. 0S
1.83 2.30 2.56
2.26 2.80 3.09
.176 .250
,796 1.00
1. 60 1. 93
2.s0 2.9S
3.45 4.01
4. 44 5. 11
s.46 6.23
6.50 7.37
7.s5 8.52
8.62 9.68
9.69 10.9
10.8 12.0
11, 9 13. 2
13.0 14.4
14. 1 15, 6
15. 2 16. S
16. 3 18, 0
17.4 19.2
18,5 20.4
19. 6 21 . 6
.429 .667
1.45 2.00
2.63 3.48
3.89 $.42
5. 10 6, 60
6. 51 8. 19
7,86 9.80
9, 21 11. 4
10, 6 13. 0
12.O 14.7
13. 3 16. 3
14.7 18.0
16. 1 19. 6
17.5 21.2
1B.g 22.9
20,3 24.5
21.7 26.2
23.1 27.8
?4.5 29,5
25.9 31 :2
. 053 . 11' l
.381 .595
.899 1.27
1.62 2.05
2.22 2.BB
6
7
I
o
10
1 1
1 2
1 3
14
1 5
1 6
1 7
1 8
1 9
20
2.72
3.21
3"71
4.24
4.78
2.96
3.74
4.54
5.37
6.22
7.08
7,95
8.83
9.73
10, 6
1 1 . 5
12. 5
13.4
14. 3
15.2
3.76
4. 67
5. 60
6.55
7.51
8.49
9.47
10. 5
1 1 . 5
12. 5
13, 5
14. 5
15. 5
16. 6
17. 6
3.33 3.65
s.88 4.23
4.45 4.83
5.03 5.45
s.63 6.08
5.34 6.25 6.7?
s.91 6.88 7.38
6.50 7.52 8.05
7,09 8.17 8.72
7.70 8.S3 9,41
608 APPENDIx D
TABLE D.1 (Continued)
NIB 0.01 0.05 0.1 0.5 1.0 2 5 10 15 20 30 40
21 8.32 9.50 10.1 11,9 12.8 14.0 16.2 187 20.8 22.8 27.9 32.8
22 8.95 10,2 10.8 12.6 13.7 14.9 17.1 19.7 ?1 .9 24j ?a.7 34.5
23 9. 58 10. 9 11. 5 13. 4 14. 5 15. 8 18. 1 aQJ 23. 0 25, 3 30. 1 36. 1
24 10.2 11.6 12.2 14.2 1s.3 16,6 19.0 21.8 24.2 26.5 31,6 37.8
25 10.9 1?,3 13.0 15.0 16.1 17.5 20.0 22.8 ?5,3 277 33.0 39.4
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
11.5 13,0 13.7 15.8 17.0 18.4 20.9 23.9 26.4 28.9 34.4 41.1
1?.2 13.7 14.4 16.6 17.8 19.3 21.9 e4,9 27.6 30.2 35.8 42.8
12.9 14.4 15.2 17.4 18.6 20.2 22.9 26.0 28.7 31.4 97.2 44.4
13.6 15.1 15.9 18.2 19.5 21.0 ?3,8 27.1 29.9 32.6 38.6 46.1
14.2 15.9 16.7 19.0 20.3 21.9 24.8 28.1 31 .0 33,8 40.0 47.7
14.9 16,6 17.4 19,9 21.2 22.8 2S.8 2s.2 32.1 3s.1 41.s 49.4
15.6 17.3 18.2 20.7 22.0 23J 26.7 30.2 33.3 36.3 42.9 51.1
16.3 18.1 19.0 21,5 22.9 24.6 277 31.3 U.4 s7.5 44.9 52.7
17.0 18,8 19.7 22.3 23.8 25.5 28.7 s2.4 35.6 38.8 45.7 54.4
17.8 19.6 20,5 23.2 24.6 26.4 29.7 33.4 s6J 40.0 47.1 56.0
18.5 ?0,3 21.3 24.0 25.5 27.3 30,7 34.s 37.9 41.2 48.6 57.7
19.2 21.1 2e.1 24.8 26.4 28.3 31.6 35.6 39.0 42.4 50.0 5e.4
19,9 21.9 22.s 25.7 27.3 29.2 32.6 36.6 4Q.2 4s7 51.4 61.0
20.6 22.6 23.7 26.s 28.1 30.1 33,6 s7.7 41.3 44.9 52.8 62.7
21.4 23.4 24.4 27.4 29.0 31.0 34.6 38.8 42.5 46.1 54.2 64.4
22.1 24.2 25.2 28.2 29.9 31.9 3s.6 39.9 43.6 47.4 55,7 66.0
22.8 25.0 26.0 29.1 30.8 32.S 36.6 40.9 44.8 48.6 57j 67.7
23.6 25.7 26.8 29.9 31.7 33.8 37.6 42.0 4s.9 49.9 s8.5 6e.3
24.3 26.5 27.6 30.8 32.s U.7 38.6 43.1 47.1 51.1 59.S 71.0
25.1 27.3 28.4 31.7 33.4 3s.6 39.6 44.2 48.2 s2.3 61.3 727
?5.8 28.1 29.3 32.5 34.3 36.s 40.5 45.2 49.4 53.6 62.8 74.3
26.6 28.9 30.1 33.4 35.2 37.5 41.5 46.3 50,6 s4.8 64.2 76.0
27.3 29.7 30.9 34.2 36,1 38.4 42.5 47.4 51.7 56.0 65.6 77.7
28.1 30.5 31.7 35.1 37.O 39.3 43.s 48.s 52.9 57.3 67.0 7s.3
28.9 31.3 32.s 36.0 37.9 40.3 M.5 49.6 s4.O s8.s 68.5 81.o
29,6 32.1 33.3 36.9 38.8 41.2 45.5 50.6 55.2 59.7 69.9 82J
30.4 3?,9 34.2 37.7 39,7 42j 46,s 51.7 56.3 61.0 71.3 84.3
31.2 33.7 35.0 38.6 40.6 43.1 47.5 52.8 57.5 62.2 72J 86.0
31,9 34.5 3s.8 39.5 41.5 44.0 48.5 53.9 58.7 63.5 74.2 87,6
32.7 35.3 36.6 40.4 42.4 44.s 49.s s5.0 59.8 64.7 7s.6 89.3
33.5 36.1 97.5 41.2 43.3 45.9 50.5 56.1 61.0 65.9 77.O 91.0
34.3 36.9 38.3 42.1 M.2 46.8 51.5 57.1 62,1 67J 78.4 92.6
3s.1 37.8 39,1 43.0 45.1 47.8 52.6 58,2 63.3 68.4 79.8 94.3
35.8 38.6 40.0 43.9 46.0 48.7 53.6 59.3 64.5 69.7 81.3 96.0
36.6 39.4 40.8 44.8 46.9 49.6 54,6 60.4 65,6 70.9 827 97,6
37.4 40.2 41.6 45.6 47.9 50.6 55.6 61,5 66.8 72j 84.1 99.3
38.2 41,0 42.5 46.s 48.8 51.s 56.6 62.6 68.0 7s.4 8s.s 101.
39.0 41 .9 43.3 47.4 49.7 52.5 57.6 63.7 69,1 74.6 87.0 103.
39.8 42.7 44.2 48.3 50.6 53,4 58.6 64,8 70.3 75.s 88.4 104.
40.6 43.5 45,0 49.2 51.5 54.4 59.6 65.8 71.4 77.1 89.8 106.
APPENDIX D 609
TAELE D.l (Gontlnuedl
ME
66
67
68
69
70
71
72
73
74
75
0.01 0.0s 0.1 0.5 1.0
41.4 44.4 45.8 50.1 52.4 55.3 60.6 66.9 72.6 78.3 91.2 108,
42.2 48.2 46.7 51.0 53.4 56.3 61.6 68.0 73.8 79.6 92.7 109.
43.0 46.0 47.6 51.9 54.3 57.2 62.6 69.1 74.9 80.8 94.1 111.
43.8 46,8 48.4 52.8 s5,2 58.2 63.7 7Q.2 76.1 82.1 95.5 113.
44.6 47.7 49.2 53.7 s6.1 59.1 64.7 71.3 77.3 83.3 96.9 114,
45.4 48.5 50.1 54.6 57.0 60.1 65.7 7?.4 78.4 84,6 98.4 116.
46,e 49.4 50.9 55.5 58.0 61.0 66.7 73.5 79.6 8s.B 99.8 118.
47.0 50.2 s1.8 56.4 58.9 62.0 ts7.7 74.6 80.S 87.0 101, 119.
47.8 51.0 52.7 57.3, 59.8 62.9 68"7 75.6 81.9 88.3 103. 121.
48,6 51.9 53.6 58.2 60.7 69.9 69.7 76.7 83,1 89.5 104. 123.
76 49.4 52,7 s4.4 59.1 61,7 64.9 70.8 77,8 84.2 90.8 10s. 124.
77 50.2 53.6 55.2 60.0 62.6 65.8 71.8 78.9 85.4 92.0 107. 126,
78 s1.1 54.4 s6,1 60.9 63.5 66,8 72.8 80.0 86.6 93.3 108. 128.
79 51.9 55,3 56.9 61.8 64,4 67.7 73.8 81,1 87.7 94.5 110. 129.
80 52.7 56.1 57.8 62,7 65.4 68.7 74.8 82.2 88.9 95.7 111. 131.
81 s3.5 5S.9 58.7 63.6 66.3 69.6 75.8 83.3 90.1 97.O 113. 133"
82 54.3 s7,8 59.5 64.5 67.2 70.6 76.9 S4.4 91.2 S8,2 114. 134.
89 55.1 58.6 60.4 65.4 68.2 71.6 77.9 85.5 92.4 99.5 115. 136,
84 56.0 59,5 61.3 66.3 69.1 72.5 78.9 86.6 93.6 101. 117, 138.
8s s6.8 60.4 62.1 67.2 70.0 73.5 79,9 87.7 94.7 102. 118. 139.
86 57.6 61 .2 63,0 68.1 70.9 74,5 80.9 88.8 9s.9 103. 120. 141 ,
87 58.4 6e.1 63.9 69.0 71.9 75.4 82.0 89.9 97.1 104. 121., 143.
88 s9,3 62.9 M.7 69.9 72.8 76.4 83,0 91.0 S8.2 106, 123. 144.
89 60.'1 63.8 65,6 70.8 73.7 77.3 84.0 92.1 99,4 107. 124. 146,
90 60.9 fr.6 66.5 71.8 74.7 78,3 8s.0 93.1 101. 108. 126. 148,
40 30 20 1 5 1 0
91
92
93
94
95
96
s7
98
99
100
61.8 65.s 67.4 72.7 75.6 79.3 86.0 94.2
62,6 66.3 68,2 73.6 76.6 80.2 87.1 95.3
69.4 67.2 69.1 74.5 77.5 81.2 88.1 96,4
64,2 68.1 70.0 75.4 78.4 82.2 89,1 97.5
65.1 68.9 70,9 76.3 79.4 83.1 90.1 98.6
65.9 69.8 71.7 77.2 S0.3 84.1 91.1 99.7
66.8 70.7 72.6 78.2 81.2 85.1 92.2 101.
67.6 71 .5 73.5 79.1 82.2 86.0 93.2 102.
68.4 72.4 74.4 80.0 83.1 87.0 94.2 103.
69.3 73.2 75.2 S0.9 84.1 88.0 95.2 104.
102. 109, 127. 149.
103. 111, 128. 151.
104. 112. 130. 1s3.
105. 1 13. 131. 154.
106. 114. 133. 156,
108. 116.
109. 117.
110. 118.
1 1 1 . 1 1 9 "
' 11?.
121 .
134. 158.
135. 1$9.
137. 161.
138. 163.
140. 164.
a/V
is the nufibef of servers. Th6 numerical column hedings indicate blocklng
probability (%).
610 APPENDIX D
TABLE D.2 Maxlmum Offered Load Versus B, A|, and Flnite $ourcee llla
30 20 1 5 1 0
1. 0 0.5 0. 1 .01 M N
2
' r t ' 3
' 4
5
0002 ,0010 .0020 ,0100 .0200 .0400 .100 .202 .307 .417 .6ss
.0002 .0008 ,001s .0075 ,0151 .0303 .077 .1s9 .246 .341 .559
.0001 ,0007 .0013 ,0067 .0134 .0270 ,069 .143 .224 .312 .s19
. 0001 . 0006. 0013. 0063. 0126. 0254. 065
. 136 . 213 . 298 . 498
8 2,15 2.70 2.98 3,76 4j 7 4.63 5.39 6.20 6.89 7.56 9.01
9 1.8s 2.36 2.62 3.36 3.7s 4.22 s.00 5.8s 6.56 7.25 8.74
10 1.70 2j 7 2.42 3.13 3.52 3.99 4.78 5.64 6.97 7.07 8.s8
15 1.39 1.80 2.02 2.68 3.05 3.51 4.31 5.21 5.97 6.70 8.2s
2Q 1.28 1,67 1.88 2.52 2.88 3.33 4J4 s.Os s.81 6.s6 8.13
30 1.19 1.56 1.76 2.38 2.74 3,18 3.99 4.90 5,68 8.44 8.03
3 .030 .067 .095 .212 .300 .425 .678 .980 1.23 1.47 1.s7
4 .023 .0s2 .074 .167 .238 .342 .s60 .832 1.07 1.30 1.78
5 .021 .046 .065 ,149 .213 .308 .510 .787 .gs7 1.22 1.70
6 .019 .043 .061 .139 .200 .289 .482 .731 .955 1.18 1,65
7 .018 .041 .058 .133 .191 .277 .464 .707 .928 1.1s 1.6?
4 .186 .317 .400 .685 .864 1.09 1.s0 1.95 2.31 2,65 3.35
5 .148 .254 .322 ,561 .715 .918 1.30
,t.72
2.OB 2.42 3.13
6 .131 .227 .288 .505 .648 .837 1.20 1.62 1.97 2,g1 3.02
7
j 22
.211 .268 .473 .609 .790 1.14 , 1.5s 1.90 2.24 2.e5
8 .116 .201 .255 .452 .583 .759 1.10 1.51 1.85 2.19 2.90
9 ,111 .194 .246 .437 .s65 .737 1.Q7 1.47 1.82 2.16 2.86
10 .108 .1S8 .240 .426 ,551 .720 1.05 1.45 1.79 2.13 2.84
15 .100
j 74
.222 .396 .514 .67s ,994 1.38 1.72 2,06 2.76
5 .500 .748 .889 1.33 1.59 1.89 2.24 2.98 3.43 3,86 4.76
6 .408 .617 .737 1.12 1.36 1.64 2.15 2.71 3.16 s.6o 4.51
7 .365 .554 .665 1.02 1.24 1.52 2.01 2.56 3.02 3.46 4.38
8 .340 .517 .621 .963 1.17 1.44 1.9? 2.47 2.93 3.37 4.30
9 .323 .492 .592 .922 1.13 1.39 1.86 2.4' t 2.87 3.31 4.?4
10 .310 .4' t4 .571 .892 1.09 1,35 1.82 2.36 2.82
g.27
4.20
r5 .280 .429 .518 .816 1.00 1.25 1.71 2.24 2.7Q 3.14 4.08
. 9s1 1. 31 1. 51 2. 08 2. 40 2] 7 3, 39 4. Q4 4. 58 5. 09, 6. 17
.794 1.11 1.28 1.80 2.10 2.45 s.o7 s.7s 4.28 4.80 5.91
. 716 1, 00 1. 16 1. 66 1. 94 2. 29 2. 90 3, 56 4j 2 4. 65 5. 77
.668 .940 1.04 1.56 1.84 2.18 2.78 3.45 4.01 4.55 5.68
.695 .896 1,09 1.50 1.77 2.11 2.71 3.s7 3.94 4.48 s.6.1
.592
.,839
.979 1.42 1.68 2.01 2,60 9.27 3.84 4.38 5.52
,556 .791 ,924 1.35 1.60 1.92 2.51 3.18 3.74 4.2g 5.44
.525 ,748 .876 1.28 1.53 1.84 2.42 3.09 3.66 4.21 5.37
1.51 1.97 2.21 2.90 3.26 3.69 4.38 5.12 5.73 6.32 7.59
1.28 1.70 1.91 2.55 2.90 3.32 4.o2 4.78 5.41 6,02 7.g2
1.17 1.55 1.76 2.37 2.71 3.12 3.82 4.59 5.24 5.S5 7.17
1.09 1.46 1.6s 2.25 2.58 2.98 3.69 4.47 s.12 5.74 7.O7
.926 1.2s 1,43 1.97 2.29 2.68 3.38 4.17 4.84 s.48 6,84
.865 1.17 1.34 1.87 2j 7 2.56
g.A6
4.OS 4.72 S.i 7 E:74
. 813 1. 10 1. 27 1. 77 2. O7 2. 45 3. 15 3. 94 4. 62 s. 28 6. 66
6
7
I
1 0
1 2
1 5
20
7
I
I
10
1 5
20
30
APPENDTx D 611
TABLE D.2 (Continuedl
30 20 15 10 t . 0 0.5 0. 1 .05 .01 M ,v
1 0
I 2.85 3.48 3.80 4.65 5.09 5.59 6,41 7.30 8,05 8.80 10.4
10 2.49 3,08 3.37 4.20 4.64 5.14 6.00 6.92 7.71 8,48 10.2
11 2.25 2.85 3.14 3.94 4.38 4.89 5.76 6.71 7.51 8.2S 9.99
12 2.16 2.70 2.97 3.77 4.20 4.17 5,59 6.56 7.37 8.17 9.88
15 1,93 2.43 2.70 3.46 3.89 4.40 5.29 6.29 7.13 7.94 9.69
20 1.76 2.24 2A9 3.23 3.65 4.16 5,06 6.07 6.93 7.76 9.s4
30 1.63 2.08 2.32 3.03 3.45 3.95 4.86 5.89 6.77 7.61 9.41
10 3.59 4.30 4.84 5.57 6.03 6.56 7.44 8.39 9.22 10.0 11.9
11 3.18 3.84 4.17 5.O7 5.54 6,09 7.OO 8.01 8.87 5.72 1' 1.6
12 2.94 3.57 3.89 4.75 5.26 5.81 6,75 7.78 8.66 9.52 11.4
13 2.78 3,39 3.71 4.s9 5.06 s.61 6.57 7.62 8.51 9.39 11.3
14 2.66 3.26 3.57 4.44 4.91 5.41 6.43 7,50 8.41 9.2S' | 1.2
16 2.50 3.08 3.38 4.33 4.70 5.27 6.34 7.33 8.25 9.15 11.1
18 2.39 2.95 3.25 4.09 4,56 5.13 6.11 7.21 8.15 9.06 11.0
20 2.31 2.87 3.15 3.99 4.46 5.02 6.02 7.13 8.07 8.99 11.0
30 2j 2 2.64 2.91 3,73 4.19 4.76 5,77 6.90 7.87 8.81 10.8
11 4.38 5.15 5.52 6.49 6,98 7.54 8.47 9.49 10.4 11.3 13.3
12 3,91 4.64 5.00 s.97 8.47 7.04 8.02 9.09 10,0 11.0 13.0
13 3.63 4.34 4.6S 5.65 6.16 6.74 7.75 8"86 9.81 10.8 12.8
14 3.45 4.13 4.47 5,43 5.94 6.54 7,56 8.69 9:66 10' 6 12.7
15 3.31 3.98 4.32 5.27 5,78 6.38 7.41 8.56 9.55 10.5 12.6
16 3.20 3.86 4.19 s.14 5.6s 6.2s 7.30 8.46 9.46 10.4 12.6
1B 3.04 3.68 4.01 4,95 s.46 6.07 7.13 8.31 9.33 10,3 12.s
2a 2.93 3,56 3.88 4.81 5.32 5.93 7.O1 8,21 9.23 10.2 12.4
25 2.75 3.36 3,68 4.s9 5.10 5,72 6.81 8.04 9,08 10.1 12.3
30 2.65 3,25 3.56 4.47 4.98 s.59 6.69 7.93 8.99 10.0 12.2
12 s. 19 6. 01 6. 41 7. 44 7, 95 8. 53 9. s0 10. 6 11. 6 12. s 14. 7
13 4,68 5.46 5.85 6,88 7.40 8.01 9,04 10.2 11.2 12,2 14.4
14 4.37 5.13 5.51 6.54 7.O7 7.69 8.76 9,94 11.0 12.0 14.2
1s 4.15 4.90 5.27 6.30 6.84 7.47 8.56 9.77 10.8 11.9 14.1
16 3.99 4.72 5.09 8.12 6.66 7.30 8.40 9.63 10.7 11.7 14.0
17 3.86 4.s9 4.95 5.98 6.52 7.17 8.28 9.s3 10.6 11.7 14.0
18 3.76 4.48 4.84 5.86 6,41 7.06 8.18 9,44 10.5 11.6 13.9
20 3.60 4.31 4.66 5.68 6.23 6.88 8.03 9.31 10,4 11.5 13.8
2s 3.36 4.04 4.39 5.40 5.95 6.62 7.79 9.10 10.2 11.3 13.7
30 3.32 3.90 4.24 s.24 5.79 6.46 7.64 8.98 10,1 11.2 13.6
13 6.03 6.90 7.31 8.39 8.92 9,52 10.s 11.7 12.7 13.8 16.1
14 5.47 6,31 6.72 7.80 8.35 B.9B 10.1 11.3 12.4 13.4 1s,8
15 5.13 5.95 6.35 7.44 8.00 8.65 9.77 11.O
' 12.1
13,2 15.7
16 4.88 5.69 6.09 7.18 7.75 8.42 9.56 10,8 12.0 13.1 15.5
17 4.7Q s.50 5.90 6.9S 7.56 8.24 9.40 10.7 11.8 13.0 15,5
18 4.56 5.35 s.74 6.83
7.41 8.09 9.27 10.6 11.8 12.9 15.4
20 4.34 5,' t2 5.51 6.60 7.18 7.87 9.08 10.4 11.6 1e.8 1s,3
25 4.03 4.77 5.16 6.24 6,83 7.54 8.78 10,2 11.4 12.6 15.1
30 3.85 4.s8 4.96 6,04 6.63 7.35 8.61 10.0 1 1 ,3 12.5 1s.0
11
12
the numerical column headings indlcate blocking
probability (%).
GLOSSARY
2BlQ. Four-level line code used in ISDN basic rate access lines.
997o Power bandwidth. Bandwidth containing
99Vo of the energy of a signal'
Abbreviated diating. ,See Speed dialing.
Access tandem.
"switching
system within a LATA used as an access
point by
long-distance carriers."
Add-drop multiplexer
(ADM).
Network element that accesses only a
portion of a
higher level digital signal to extract and insert local traffic at an intermediate
point
of a route.
Added-channel framing. TDM framing format utilizing an additional channel with
the same rate as the message channels for the express
purpose of defining
frame
boundaries.
Added-digit framing. TDM framing format utilizing an additional bit in every ftame
with a
prescribed repetitive sequence for the express
purpose of defining frame
boundaries.
Advanced mobile phone service
(AMPS).
Analog cellular mobile telephone
standard of NorJh America.
Alternate mark inversion
(AMI)
signal. Three-level
signal with values +V, 0'
-y.
A
space
(logic
0) is encoded with 0 V. A mark
(logic
1) is encoded with altemating
values of +Vand
-t/,
which ale nonnally only 507o duty cycle pulses. Also refered
to as a bipolar signal.
Alternate mark inversion violation.
In AMI coding, the occunence of two
successive marks with the same
polarity. Also refened to as a bipolar
violation.
Amplitude distortion. Distortion of a transmission signal caused by nonuniform
passband
attenuation as a function of frequency.
Amplitude modulation
(AM).
Modulation of the amplitude of a carrier wave with
the amplitude of a baseband signal.
Antipodal signaling. Technique
of encoding binary signals so that the symbol
for a
I is the exact negative of the symbol for a 0. Antipodal signaling
provides optimum
error performance in terms of the signal-to-noise ratio.
613
614 GLoSSAFY
Asynchronous
network. A network in which the clocks of the transmission
links
and switching systems are not synchronized to each other.
Asynchronous
transfer mode
(ATM).
packet-switched
method of time division
multiplexing (labeled
multiplexing) utilizing fixed packet (cell)
sizes of 53 bytes.
Asynchronous
transmission. Mode of communication characterized
by start-stop
transmissions
with undefined time intervals
between transmissions.
Each
transmission burst generally
contains a single word or byte of information.
Automatic call distributor (ACD).
switching sy$tem used to evenly distribute
incoming calls to a number
of stations on a first-come, first-served basis.
Applications include operator assistance
and airline reservations.
Automatic number identification (ANr).
process
of identifying
and forwarding
a
calling number to network call conffol services.
Automatic
repeat request (ARo.
Error correction process
involving
the use of
redundant check bits to detect comrpted blocks of data and trigger requests for
retransmission
of the same.
Availability. (l)
With respect to switching
systems: the number of outlets available
from a particular
inlet. (2)
with respect to equipment in general:
the percentage
of
time the equipment is providing
acceptable
service.
Balanced code. Line code with equaily occurring positive
and negative energy ro
preclude
a dc component in the frequency spectrum.
Baseband.
Literally, the frequency
band of an unmodulated
signal. A baseband
signal is an information-bearing
signal that is either transmittea
as is or used to
modulate
a carier.
Baud rate. unit of signaling speed (symbols
per
second). For binary signaling the
data rate is the same as the baud rate. For multilevel transmission
the data rate is
equal to the baud rate times log2(L), where L is the number of levels.
Bipolar coding. ,See Altemate mark inversion.
Bis. second version. Ter means three. For example, v.27bis and v.27ter
are the
second and third versions
of the V.27 standard, respectively.
Bit leaking. Process of. converting
byte-sized timing adjustments into multiple
bit-sized or fractional
bit-sized timing adjustments.
Blocked
calls cleared (BCC).
Service discipline in which unserviceable
requests are
rejected by the system without service. Also called losr calls cleared (LCC).
Blocked calls held (BcH).
service discipline in which unserviceable
requesm sray
"in
the system" without being serviced but having a portion
of their desired service
time elapse until service begins. Also called losr calls held
(LCH).
Blocking-
Inability of a calling party
to be connected to a called parry
because either
all circuits are busy or there is internal (matching
toss) blocking within a switch
matrix.
Bridged tap. Extra pair
of wires connected in shunt to a main cable pair.
The extra
pair is normally open circuited but may be used at a future time to connect the main
GLOSSAFY 615
pair to a new customer. Shor.t-bridged taps do not affect voice frequency signals but
can be extremely detrimental to higher ftequency digital signals.
Building integrated timing supply
(BITS)- Master timing supply for all equipment
in a building.
Busy hour. The 60-minute
period of a day
(and
sometimes
of a weet) in which the
average offered traffic load is highest.
Call admission control
(CAC).
ATM control
process
responsible for negotiating
quality-of-service
parameters for new connections dependent on the traffic loads
of the network at the time.
Call congestion. Blocking probability of a trunk group.
Call forwarding. Process of signaling a switching system to reroute calls intended for
a
particular number to some other number on a temporary basis.
Call waiting. Process of inserting a short tone into the voice path of an active user
informing the user that another call is waiting to be connected.
Catling number identiticafion
(CM).
Service whereby a telephone network
provides the identity of a calling number
(or person) with incoming ringing'
Carrier sense multiple accesdcollision detection
(CSMA/CD).
Access
protocol
for a common bus or radio link in which sources
flust ny to detect the
presence of
a carrier and then begin transmitting when the facility is idle. If another source
begins transmitting during a
period of time corresponding to the maximum
transmission delay, acollision
is detected
and all actiye
nodes cease transmission
for a random arrount of time.
Carrierless amplituddphase
(CAP)
modutation.
QAM
implementation technique
using direct
(DSP) generation of filtered quadrature signal waveforms followed by
DSP summation for the composite signal.
CCS. Measure of traffic intensity expressed
as so many hundred call seconds
per
hour; 36 CCS
=
I edang.
Cell delay variation
(CDV).
An ATM
QoS
parameter that specifies the maximum
range ofcell
propagation delay through a network'
cell transfer delay
(cTD). An ATM
QoS
parameter specifying the average
propagation time for a cell to be transferred
from its source to its destination.
Central office. Public network switching office
(and particularly the switching
machine
within it) to which subscriber lines are connected. Also referred to as an
end office or class 5 office. In a more general sen$e, any switch in the public
network is sometimes referred to as a central office.
Centrex service. Private switching service
provided by a local exchange carrier to an
organization utilizing
hardware and software
partitions of a central office switch.
CEPT-|. Term used to designate the 2.048-Mbps first-level digital signal of the
fTU-T digital hierarchy.
(More
often referred to a$ an EI signal')
Channel-associated signaling
(CAS).
Use of $eparate
signaling channels for each
message channel. The individual signaling channels may be separate from the
616 GLoSSARY
message channels (as
in El) or derived within the message channels themselves (as
i n Tl ).
channel bank. Equipment
that converts multiple analog interfaces
to and from a
time-division-multiplexed
digital bit stream (24
analog io one DS I or 30 analog to
one El). Channel banks are also used for FDM applications.
channel service unit (csu)-
Equipment installed on customer premises
at the
interface to phone
company lines to terminate a DDS or Tl ciicuit to provide
network protection
and diagnostic capabilities.
Also called a customer service unit.
circuit Emulation
service (cEs).
An ATM service in which constant-bit-rare
virtual circuits use AALI adaptation to emulate an end-to-end physical
circuit.
Circuit switching.
The principle
of establishing
an end-to-end connecrion
between
users of a network. The associated facilities
are dedicated
to the particular
connection and held for the duration of the call.
clear-channel
capability.
Ability to provide
a transparenr
64-kbps channel through
a North American network (using
BSZS and common-channel
signaling).
clock. Time base used to control the transfer of digital information.
C'message
weighing. Selective atlenuation
of voiceband noise in accordance wifh
the subjective effects
as a function
of frequency (noise
weighting filter
characteristic used in North America).
codec(coder-decoder).
Integrated circuit providing
analog-to-digital
and
digital-to-analog
conversion of voice signals.
coded mark lnversion (cMr).
Two-level
line code in which a binary 0 is coded
with a positive
level and an equal-magnitude
negative level for half a unit time
interval each. A binary I is coded with a full-period positive
or negative level in an
alternating manner to maintain
dc balance.
coherent demodulation.
Demodulation
using
a carier reference that is
synchronized
in frequency
and phase
to the carrier used in the modulation process.
common-channel
signaling. use of a separate channel dedicated
to tran$port of
signaling information pertaining
to a multiple number of message channels.
community dial office (cDo).
small, normally unanended switching
system that is
used in small communities
and is controlled from a rarger centrar o}fice.
companding.
Process of compressing
a signal at the source and expanding it at the
destination to maintain
a given
end-to-end
dynamic range whiie reducing
the
dynamic range between the compressor-expander.
concentration.
Process of switching some number of lighfly
used channels
or
sources onto a smaller number
of more heavily used channels.
Conference
call. Call in which three or more stations are interconnected
in a manner
that allows all parries
to talk and be heard by all other parties.
connection admission
control (cAc).
An ATM function that determines whether
a vifiual circuit (vc)
connection request should be accepted or rejected.
constant (or
continuous) bit rate (cBR).
An ATM class of service that supports
transmission
of a continuous
data rate.
GLOSSARY 617
Constant envelope.
A (frequency or phase) modulated carrier with no
the carrier amplitude.
Controlled stip. Slip involving the repetition
or deletion of an integral number of
TDM frames.
Crossbar switch. Rectangular array of crosspoints arranged so that an input
connected
to a row of crosspoint input$ can be switched to any output
connected to a column of crosspoint outputs.
Crosstalk. Unwanted signal transfer from one circuit into another'
Custom Local Area Signaling Services
(CLASS).
Signaling service for analog
telephone
lines provided between the first two rings of an incoming telephone call'
Principal features enabled by CLASS services are Caller ID, Call Return, Call
Screening, Preferued Call Forwarding, Priority Ringing, and Repeat Dialing'
Datagram. Single packet message
(from the point of view of a packet network).
Dataphone Digital Service
(DDS).
Service offering of common-carder telephone
companies
providing digital channels
at 2.4, 4.8, 9.6, 56, or 1544 kbps.
dBm. Power level in decibels relative to I mW'
dBrnC. Power level of noise with C-message weighting expressed in decibels
relative to reference noise. Reference
noise power is
-90
dBm
=
10-l? W.
dBrnC0. Noise power measured in dBmC but referenced to the eero-level
transmission level
point.
Decision feedback equalization
(DF'E). Equalization technique involving the use of
deftcted
data values to produce simulated channel responses in a receiver that are
then subtracted
from subsequent received signal Samples
to cancel channel
distortion.
Delta modulation. Technique for digitally encoding an analog signal by continuously
measuring and transmitting only the
polarity of the signal slope. The analog signal
is reconstructed as a staircase approximation
of the input'
Demand traffic.
First attempt traffic.
(offered
traffic
not including retries.)
Desynchronizer. Circuitry
that extracts
a tributary digital signal from a composite,
higher level signal and derives the source clock frequency of the tributary.
Digitat advanced
mobile phone
service
(D-AMPS), Digital mobile telephone
standard
of North America, also known as tS-I36 TDMA.
Digital circuit muttiplication
{DCM).
Process of increasing the number of voice
signals carried on a digital transmission link through the use of speech compression
or silence removal or both.
Digitsl loop carrier
(DLC).
A TDM system
for transporting multiple digital signals
to and from multiple subscriber lines.
Digitat
pad. Digitat circuit that changes PCM sample values for the
putpose of
adjusting
(attenuating) the power level of the encoded analog signal.
Digital satetlite system
(DSS).
Direct broadcast satellite
(DBS)
system developed
by
Hughes Electronics Corporation.
618 GLOSSAHY
Digital speech interpolation (DSI).
Dieital equivalenr
of rASI wherein digital
speech signals are monitored and connected to a hansmission
channel only when
voice activity is present (a
form of DCM).
Digital
video broadcasting (DvB)
group.
European
digitat rv standard for TV,
audio, and data. DvB can be broadcast via satellite or terrestrial systems.
Direct broadcast
satellite (DBS)-
One-way TV broadcast service from a sarellite to
a small l8-in. dish antenna.
Direct inward
dialing (D.rD).
process
of a public
telephone network providing
pBX
extension numbers
with incoming
calls.
Double-sideband
modulation.
Modulation
technique in which a baseband signal
with no dc energy directly modurates a carrier to produce
both upper and lower
sidebands
but no carrier energy.
Dual'tone multifrequency (DTMF)
signaling.
Generic name for pu$h-button
telephone
signaling equivalenr to the Bell Sysrem's TOUCH_TONE.
Dynamic
rflnge. Range of power
levels (minimum
to maximum)
achievabre by a
signal or specified for equipment operation.
E&M lead signaling. Interface
between a switching
system and a transmission
system utilizing pairs
ofwires for signaling
that are separate from the voice pairs.
outgoing control signals from each facility are carded by respective M-leads
to
E-leads of the other facility.
Echo canceler. Device that removes talker echo in the return branch of a four-wire
circuit by subtracting
a delayed version of the signal hansmitted
in the forward
path.
Echo suppressor. Device that is activated by voice energy in one path
of a four-wire
circuit to insert a high amount of signal loss in the other path
ior the purpose
of
blocking an echo.
Echo. Reflected
and delayed
signal. common sources of echo in the telephone
network are electrical reflections
at four-wire to two-wire
conversions and acoustic
reflections
with speakerphones.
Edge switch.
An ATM switch providing
gareway
interfaces
to non-ATM
communication
links.
Elastic store. First-in, first-out data buffer that accepts data under conftol of one
clock and outputs data under control of another clock.
Electronic
automatic
exchange (EAx).
Designation
of stored-program
control
switching machines
developed by cenerar relephone and Erectronic*s.
Electronic
switching
system (ESs).
Designation
of stored-program
conhol
switching machines
developed by AT&T.
End office. class 5 switching office. Arso referred to as a central office.
Envelope
delay. Derivative
of channel phase
response
with respect
to frequency.
Ideally,
the phase
response
should be linear, indicating
ttrat att frequencies
are
delayed equally.
GLOSSAHY 619
Envelope detector. Method of detecting
amplitude modulation by tracking only the
peak values of a carrier wave.
Equalization. Practice of compensating for transmission distortions with fixed or
adaptive circuitry.
Erlang. Measure of traffic intensity. Basically, a measure of the utilization of a
resource
(e.g.,
the average number of busy circuits in a kunk group, or the ratio of
time an individual circuit
is busy).
Error second. A l-sec interval containing at least I bit error.
Exchange area. Contiguous area of service defined for administrative
purposes that
typicatly comprises an enrire town or city and includes the immediate
countryside
and suburbs. An exchange area may have one end office or many end offices
interconnected by trunks and tandem
offices.
Expansion. Switching
of a number of input channels onto a larger number of output
channels.
Extended superframe format
(ESF).
A Tl framing
format that embeds diagnostic
capabilities and a facility
data link into the 8-kbps DSI framing channel.
Failed seconds
(FS).
An
(ESF) transmission link failure defined as 10 consecutive
severely errored seconds.
Far-end crosstalk
(FEXT). Unwanted energy coupled from one channel or circuit
into another circuit and appearing at the far end of the transmission link.
Fiber distributed data interface
(FDDI). High-bandwidth fiber transmission
system for interconnecting local area networks.
FIFO. First-in, first-out service discipline for a queue.
Flow control
(in
data network).
Procedure for controlling the rate of transfer of
packets from one node to another.
Foreign exchange circuit
(FX).
Extension of service from one switching office to a
subscriber normally serviced by a different switching office,
Forward error correction
(FEC). Error correction technique with sufficient
redundancy embedded into the signal for correcting errors without retransmission.
Four-wire circuit-
Circuit using two separate channels for each direction of
transmission. When wireline transmission is involved, each direction of
transmission is provided by a separate
pair of wires.
Fractional speech loss. Fraction
of speech that
Sets
clipped in a TASI or DSI system
because all channels are in use when talk spurts begin.
Frame aliSnment. Identification of the beginning and end of a TDM frame using a
framing daia
pattern,
Frame. Set of consecutive time slots in a TDM format in which the position of
individual time slots can be identified
by reference to a frame alignment signal'
Frequency diversity.
In radio systems the use Of one or more back'up transmitters,
channels, and receivers to
protect
against
atmospheric
(multipath) fading-
620 cLossAny
Frequency
division muttiplex (FDM).
Partitioning
the bandwidth
of a rransmission
link into separate channels of lesser bandwidttr into which separate signals
are
inserted
and extracted.
Frequency
modulation (FM).
Moduration process
that varies
the frequency
of a
carrier signal in accordance
with a baseband information
signal.
n'requency
shift keying
ff'sK).
Digital
modutation technique
in which
data are
represented
by discrete frequencies.
n'ricative.
speech sound produced
with continuous
air flow through
one or more
vocal tract restrictions
to cause air turbulence (friction).
Example fricative sounds
are s, f, t, x, orj,
Full'duplex.
Transmission
in two directions simultaneously (also
referred
to as
simply a duplex operation).
Gateway network
element (GNE).
A soNET node with an interface
to
unsynchronized
tributary signals.
Gaussian noise. Background
noise with a uniform frequency
spectrum across a band
of interest
and having
amplitude sample values thit foliow a normal (gaussian)
probability
distribution.
Glare. simultaneous
seizure of both ends of a two-way trunk by two switching
machines trying to set up two separate
connections.
Global positioning
system (Gps).
Global satellite
navigation
system commonry
used by telecommunications
sy$tems
as a time and friquency
synchronization
source.
Half'duplex
transmission.
Transmission
in both directions
but only in one direction
at a time.
HDB3. Modified AMI (bipolar)
line code in which strings of four 0,s are encoded
with an AMI violation in the last bit.
Head'of-line
hlocking.
Blocking of cells or packets
in a FIFO queue
because the
next enffry cannot be serviced
while other entries in the queue
could be serviced.
Highway.
A TDM path
inside a digital (TDM)
switching
machine.
Hook flash. Momentary
depression
of the switchhook
to alert afl operator
or
equipment,
but not so long as to be interpreted
as a disconnect.
Hot standby. Redundant
equipment
kept in an operational
mode as backup for
primary
equipment.
usually, automatic
switchingio
the standby equipment
occurs
when the primary
equipment fails.
Huffman
coding.
process
of assigning
variabreJength
digitar codewords
to a
message set wherein the length of a codeword is inversely proportional
to the
probability
of occurrence
of respective
messages. Thus a freiuentty
hansmitted
message requires fewer bits than the average rength
of all messages.'
Hybrid fiber coax (HFc).
cable terevision
distribution
system involving
rhe use of
fiber in the feeder portion
of the network
from a head-end to fiber nodes followed
by coax drop$ to residences
in neighborhoods.
GLossAHY 621
Hybrid.
Device used to connect a two-wire, bidirectional circuit to a four-wire
circuit.
Idle channel noise. Noise occurring during silence
intervals of speech'
Idle character. Control character
inserted into a synchronous transmission link when
there is no information to be sent'
Impulse noise. Short-duration
spike of noise that is much larger than randomly
disnibuted
(Gaussian) noise.
Inband signaling. Signaling tran$mitted within the same channel
and band of
frequencies used for message toaffic.
Integrated digital network
(IDN). Network in which digital
TDM transmission
links are directly
interfaced into digital TDM switching
matrices.
Integrated services digitat network
(ISDN).
Integrated digital
network in which
subscriber interfaces are
provided with direct access ofa signaling channel and one
or more digital 64-kbps
channels.
Intercept. Process of diverting
a caII from an intended
station to an operator or
recorded announcement.
International
Telecommunications Union
(ITU).
International standards
organization that subsumed the CCITT in March of 1993
(renamed ITU-T)-
Internet. Connected
set of networks using TCP/IP suite.
Intersymbol interterence
(ISD.
Interference in a transmission system caused by a
symbol in one signaling
internal being spread out and overlapping the sample time
of a symbol
in another signal interval.
Jitter.
Short-term variations of the significant
instant$ of a digital signal
from their
ideal
positions in time. Short-term variations are often defined to correspond to
frequencies above 10 He.
Joint
photographic experts
group (JPEG). ISO standards
group that defines how
to compress
still pictures.
Justifrcation
(pulse
stuffing). hocess of adapting the rate of digital signal by adding
non-information-carrying
bits at a source interface and extracting the same bits at
a destination interface.
Justification
ratio. Ratio of the
justifrcation (pulse-stuffing) rate used to synchronize
a hibutary signal to the maximum
possible
justification
rate allowed by a higher
level multiplex format.
(A justification
ratio of
{
implies one-third of the rate
adjustrnent
opportunities are null bits and two-thirds of the opportunities are
fributary databits.)
Key system. Small, customer
premises telephone switching system that allows users
to directly select telephone lines
(and receive dial tone from a cenffal office).
LAN Emulation. An ATM service offering that emulates Ethernet
or token ring
LANs.
Line administration. See Load balancing.
622 GLossARy
Line code. Set of electrical (or
optical) waveforms
chosen to represent
data for the
purpose
of transmission.
Load balancing.
Adjusting
the assignment
of very active customer lines so that all
groups
of customer lines in a multistage
switch receive approximately
the same
amount of traffic.
Loading
coils. Lumped-element
inductors inserted
at periodic
points
in cable pairs
to flatten out their voice
frequency response.
Although loading
coils improve voice
frequency ffansmission,
they severely attenuate
higher frequJncies
as required for
digital transmission.
Local access and transport
area (LATA).
service area of a local exchange carrier
(local
operating
company).
Longitudinal
current.
Electrical current passing
along a
pair
of wires in a common
direction.
Contrast Metallic current.
Loop
start. Off-hook signaling protocol
involving the flow of dc current between
the
tip and the ring of a subscriber loop.
Loop timing. Synchronizing
the hansmit
timing of a bidirectional
link to the
received
timing of the same link.
Lost calls cleared (also
loss system).
Mode of operation in which blocked calls are
rejected
by the network and may or may not return.
Lost catls held. Mode of operation
that holds blocked
call requesrs
until a channel
becomes available. The portion
of a call that gets
blocked is lost.
Mfil
{ltiplexer.
Asynchronous
network
multiprexing equipment
using bit stuffing
for 28 individual
1.544-Mbps
Dsl signals to create a1/..ii6-lvrbps
ns3 signal.
Main distrlbuting
frame (MDF).
Framework
used to cross connect ourside plant
cable pairs
to central office switching
equipment. The MDF provides
protection
and test access to the outside plant
cable pairs.
Management
lnformation
base (MrB),
coilection
of software
defined objects that
can be accessed via
a network rnanagement protocol (e.g.,
sNMp or TMN).
Master
clock. Particularly
accurate and stable frequency
source from which various
nodes in a network
extract their operating frequency (clock).
Master frame.
set of consecutive
TDM frames
that are identified by a master frame
alignment signal (MAS).
Matching
loss. Blocking within a mulristage switch mahix resulting
from at least
one link in all suirable paths
through the matrix being busy.
Maximum
time interval
error (MTrE).
Largest time interval error occurring
in all
possible
measurement
intervals of length
s within a measurement
period
T.
Message switching.
practice
of transporting
complete messages from a $ource to a
destination
in non-real
time and without interaction
between
source and
destination,
usua.lly in a store-and-forward
fashion.
Message
unit accounting.
Activity-based
billing as opposed to flat-rate
billing,
which is independent
ofusage.
GLOSSARY 623
Metallic current. Current
flowing in opposite directions in a wire
pair
as a result of
a difference signal
propagating along the pair. Contrast Longitudinal current'
Minimum shift keying
(MSK). Digital frequency shift modulation wherein the
magnitude of a frequency shift is the minimum amount required to
produce
a 180"
of
phase shift in a symbol interval.
Modem. Contraction of the terms modulation
and demodulation. A device used to
generate
"voicelike"
data signals for transmission over telephone lines. A modem
is referred to as a
"data
set" in older Bell System terminology.
Multifrequency
(MF)
signaling, Signaling method used for interoffice applications'
MF signaling uses two of six
possible tones to encode 10 digits and five special
auxiliary signals.
Multipath. Transmission
phenomenon in which a single transmit signal follows two
or more separate
paths to a receiver
with differing delays.
Muttiplexing. Process of combining multiple signals into a composite signal for
transmission over coilrmon facilities.
Mutually synchronized network. Technique for synchronizing a network in which
all nodes derive their operating frequency as the average of their own
free-rUnning clock frequency and the frequencies of some number of other nodes
in the network.
Near-end crosstalk
(NEXT).
Unwanted energy coupled from a ffansmitter in one
circuit into a receiver of another circuit at the same location
(near end).
Network element
(NE).
Internal node of a SONET network whose intetfaces are
signals of the SONET hierarchy.
Network management. Network management is the function of supervising a
communications network to ensure maximum utilization
of the network under all
conditions. Supervision requires monitoring, measuring, and, when necessary,
action to control the flow of traffic.
Next
generation
digital loop carrier
(NGDLC). A DLC system capable of using
SONET transmission links and
providing basic compliance
to the Bellcore GR 303
standard.
Nonblocking. Switching network that always
has a free path from any idle incoming
trunk or line to any idle outgoing kunk or line.
Nonreturn to zero
(NRZ).
Line code that switches directly from one level to
another. Each level is held for the duration of a signal interval.
Nyqulst rate.
(1) Minimum sampling rate required to extract all information in an
analog waveform. The Nyquist rate is equal to twice the bandwidth of the signal
being sampled.
(2)
In a baseband channel
with a bandwidth B, the maximum rate
(28)
that
pulses can be transmitted without intersymbol interference.
Off hook. State or condition that a telephone
receiver is requesting seryice or in use.
Also a supervisory signal to indicate acfive status of a telephone
or line.
Offered traffic. Amount of traffic carried by a system assuming the system has
inJinite capacity and therefore does not block or delay any calls.
624 clossAny
On hook. Inactive status of a telephone or line.
One-way trunk. Trunk circuit that can be seized at only one end.
Open systems interconnection (OSI).
Communications protocol
reference model
introduced
by ISo consisting of seven layers (physical,
link, network,
transport,
session, presentation,
application).
out'of-band signaling.
signaling technique thar u$es the same path
as message
traffic but a portion
of the channel bandwidth above or below that used for voice.
Pair-gain system. subscriber
transmission system that serves a number of
subscribers with a smaller
number of wire pairs
using concentration,
multiplexing,
or both.
Parity. Process of adding a redundant
bit to a
group
of information bits to maintain
either odd or even numbers of l's in the composite group
of bits. A parity
error
results if an odd number
of l's is detected when even parity
is transmitted
or vice
versa.
Partial-response
signallng (pRS).
use of controlled intersymbol
interference
to
increase the signaling rate in a given
bandwidth.
Permanent
virtual circuit (pvc).
virtual
circuit (x.25),
virtual connection (Frame
Relay), or vinual channel connection (ATM)
that emulates a leased-line
connection,
Personal
communicstions
system (pcs).
Digiral mobile cellular telephone sysrem
operating in l9fi)-MHz
frequency bands in North America (1g50-1910
MHz for
mobile to base starion 1930-lgg0 MHz for base srarion to mobile).
Per-trunk signaling.
Method of signaling
in which rhe conrrol information
pertaining
to a pafricular
call is transmitted
over the same circuit
(channel)
that
carries the call.
Phase distortion. Signal distortion resulting from nonuniform
delay offrequencies
within the passband.
Phase reversal keying (PRK).
special case of phase
shift keying involving
only two
phases
l80o apart.
Phase shift keying (PSK).
Form of digital modulation
that uses Zn distinct phases
to
represent n bits ofinformation
in each signal interval.
Plesiochronous
digital hierarchy (pDH),
Designation
of the original digital
multiplexing
hierarchy
utilizing free-running
clocks at all leiels
of the
multiplexing hierarchy.
Pleslochronous.
Method
of network
synchronization
involving the use of
independent (unsynchronized
but highly accurate) clocks at the same nominal rate.
Pointer burst. In soNET, the occurrence
of multiple pointer justification
events
(PJEs)
of one polarity
within the time constant of a desynchronizer
circuit.
Pointer
justiflcation
event (pJE).
change in a pointer
value to accommodate phase
drift.
GLOSSARY 625
Polar signaling. Two-level line coding for binary data using balanced
(symmetric)
positive and negative levels.
Power spectral density. Distribution of signal power as a function of frequency.
Primary reference source
(PRS).
Top level
(Stratum 1) clock source with an
accuracy of+l v lfl-tt.
Priority ringing. Feature for incoming call discrimination in which a telephone
rings
with a special ringing pattern for calls from a select set of
phone numbers,
Private branch exchange
(PBX).
Switching equipment used by a company or
organization to
provide in-house switching
and access to the public network.
Protection switching. Practice of transferring message taffic from one fransmission
link to a spafe transmission link when the former is degraded in some fashion or
for maintenance te$ts or equipment upgrades.
Provisioning. Allocating or deallocating facilities in suppotl of a
particular service
offering
(tuming a seryice on or off;.
Psophometric weighting.
Noise-weighting filter recommended by CCIfi.
Pulse amplitude modulation
(PAM).
hocess of representing a continuous analog
waveformwith a series of discrete-time samples. The amplitudes
of the samples are
continuous
and therefore
analog in nature.
Pulse code modulation
(PCM).
Process of converting PAM samples
into discrete
levels represented by digital codewords.
Pulse stuffing. See Justification.
QPRS. Quadrature
channel modulafion using partial-response signaling on each
channel.
Quadrature
amplitude modulation
(QAM).
Independent amplitude modulation of
two orthogonal channels using the same carrier frequency.
Quantlzation
noise. Difference between the discrete sample value represented by a
digital code and the original analog sample value.
Raised-cosine channel. Digital transmission channel with a
particular type of
pulse
response that produces no intersymbol interference at the sample
times of adjacent
signaling intervals. The designation
"raised
cosine" is derived from the form ofthe
analytical
frequency spectrum
(l
+ cosine or cosine squared).
Recirculation. Passing a cell back to an input stage of a multiple-stage ATM switch
when the cell would otherwise be discarded because it encountered blocking
in an
output stage.
Reframing time. Amount of time it takes to detect and synchronize to a frame
alignment signal.
Regeneration. Process
of recognizing and restoring
a digital signal before
perturbafions of the signal accumulate to the point that bit elrors occur'
Regenerative repeater. Device used to detect, amplify, reshaPe, and retransmit a
digital bit stream.
626 GLossAFty
Return loss. Difference in decibels between reflected and incident
energy at a signal
reflection point.
Ringback. Signaling tone returned by switching equipment to a caller indicating
that
a called party
telephone is being alerted (ringing).
Ringing voltage.
Low-frequency
ac signal used to activate the ringer of a telephone
(typically,
a20-Hz signal at 90 V rms).
Robbed bit signaling. North American practice
of using the least signifrcant
pCM
bit
of every sixth frame for channel-associated
signaling.
Router. Store-and-forward
packet-switching
element that processes
packet
headers
and determines
an appropriate
output rink for each packet
using routing tabres and
routing protocols.
severely errored
seconds (sES).
An
(ESF)
transmission link performance
measure
defined as the occurrence of 320 or more errored seconds.
Sidetone. Portion of a talker's signal that is purposely
fed back to the earpiece so that
the talker hears his or her own speech.
signaling. Exchange of electrical information (other
rhan by speech) specifically
concemed
with the establishment
and control of connections
and manasement in
a communication
network.
simple network management protocol (sNMp).
standard protocol
for perforrning
network management
functions (created
by the Intemet community).
Simplex transmission.
Mode of operation involving
transmission in one direction
only.
singing. Audible oscillation of a telephone circuit caused by a net amount of gain
in
a four-wire
$egment of the circuit.
Single mode fiber. Fiber that is narrow enough to preclude
ffansmission of an optical
signal along more than one path.
single-frequency (sF')
signating. Method
of conveying dial-pulse
and supervision
signals from one end of a trunk or line to the other using the presence
or absence
of a single specified frequency. A 2600-Hz
tone is commonlyused.
slip. Deletion or repetition
of data in a data sffeam caused by an over{low or an
underflow of an elastic store due to variations in the write and read rates. A
controlled slip is a deletion or repetition of an entire frame of data so the position
of the framing bits is undisturbed.
Slopeoverload.
Saturation
oftherangeofdifference
signalsinadifferential
encoder
caused by a signal with a large, rapid change in amplirude,
soNET island. collection of one or more soNET GNEs and NEs that create a
continuous SONET path
for a digital signal.
space diversity. In radio systems, the use of two receiving antennas and possibly
two
separate receivers
to provide
protection
against atmosphere-induced
signal
attenuation (fading).
GLOSSARY 62f
Space expansion. Use of more center stages in a multistage switch than there are
inputs to a first-stage array or outputs from a last-stage array.
Span line. Repeatered T1 line from end to end but not including channel banks.
Speed diating. Use of a short address code to represent an often*called telephone
number. The common control computer in a PBX or end office providing
such a
seryice hanslates the short code into the desired
number.
Spread spectrum. Process of distributing an information-bearing signal across a
bandwidth that is much wider than the inherent bandwidth of the signal.
Star network. Network with a single node to which all other nodes are connected'
Statistical multiplexor
($tat-mux). Multiplexer that assigns kansmission capacity
to a tributary only when the source of the tributary is actively transmitting
information.
Step-by-step
(SXS)
switch. Elecffomechanical switching system using step-by-step
switch elements to progressively set up a connection as each digit is dialed.
Stored-program control
(SPC).
Computer-controlled switching.
Stratum clock. Definition
of free-run accuracy of a clock depending on stratum level
in synchronization network.
Stuff Rstio. See Justification ratio.
Supervisory signal. Signal used to indicate the status of a
equipment on the line.
Switched multimegabit data service
(SMDS). Public, packet-switched wide-area
data communications service offered by common carriers.
Switchhook. Hook or buttons upon which a telephone handset rests when it is not
being used.
Symbol rate. See Baud rate.
Synchronizer. Circuitry that inserts a nibutary digital signal into a higher level
multiplexed signal and
performs justification (pulse stuffing) in the proce$$.
Synchronous network. Network in which the clocks of separate nodes operate at
identical
rates.
Synchronous transmission. Mode of digital transmission in which discrete signal
elements
(symbols)
are transmitted at a fixed and continuous rate.
Synchronous. Mode of operation in which two or more pieces of equipment of
network perform their operation$ at
preci$ely the same rate for an indefinite amount
of time. The rate of operation is derived by distribution of a timing
reference.
Tag switching. Fast packet-switching process that integrates network layer
(layer
3)
routing and data link layer
(layer 2) switching to simplify address and
protocol
processing.
Talker echo. Ponion of a talker's voice that is retumed to the talker and heard by the
talker. The amount of annoyance to a talker is dependent on both the amount of
delay and the amplitude of the returning signal.
or to control
628 cLos$ARy
Tandem oflice. In general,
any intermediate
switch used to establish
a connection.
In specific terminology, a tandem office is a switch used to interconnect
end offices
in an exchange area.
TASI. Time assignment speech interpolation,
the practice
of concentrating
a group
of voice signals onto a sma]ler group
of channels by dynamically switching activl
voice signals to idle channels.
Ternary coding. use of all states of a three-level
code to send more than I bit of
information
in a single symbol. This is in contrast to bipolar
coding, which uses
three levels, but only one of two in any particular
interval. one method
of
interfacing
binary data to a ternary line code is to encode 4 bits with three ternarv
symbols (483T).
Tie line. Dedicated circuit
connecting two private
branch exchanges.
Time compression
multiplex. Transmitting
in two directions
transmission link using altemating directions
of transmission
transmission).
on a single
("ping-pong"
Time congestion. Ratio of time that all facilities of a sy$tem are busy (congesred).
Time congestion refers to the status of the system and does not necessarily
imply
that blocking occurs.
Time deviation (TDEV).
Square root of the time variance (TVAR).
Time division multiplexing (TDM).
sharing a rransmission
link among multiple
users by assigning time intervals
to individual users during which they have the
entire bandwidth of a system.
Time expansion. Use of more time slots on internal links of a switch than exist on
external links.
Time interval error (TrE).
variation
in time delay of a given
signal relative to an
ideal timing signal over a particular
time period.
Time variance (TVAR).
Measure of the expected time variation of a signal as a
function of time separation.
Traffic carried. Amount of raffic canied (by
a group
of circuits or a group
of
switches) during any period
is the sum of the holding times expressed in hours
(cclrr).
Trafric
engineering. Network management
activity that analyzes average and peak
traffic loads to determine how to design or reconfigure
a network
to maximize the
traffic capacity of the network
and to ensure an acceptable value of blocking
probability.
Traflic flow. Amount of traffic (in
erlangs or ccs) carried or offered to a
transmission link or switching sy$tem.
Transhybrid loss. Amount of isolation (in
decibels) between go
and rerurn parhs
on
the four-wire side of a four-wire to two-wire
hybrid.
Transmission level point (TLP).
specification, in decibels, of the signal
tr)ower
ar a
point
in a transmission system relative to the power
of the same signal at a
(hypothetical)
zero transmission level point (0_TLp).
GLOSSARY 629
Transmultiplexer
(transmux).
Equipment that convefls FDM voice signals to TDM
voice signals and vice versa.
Transversal
equalieer. Time-domain equalizer utilizing a tapped delay line and
weighting coefficients at each of the taps to remove intersymbol interference.
Traveling class mark. Code that accompanies connection setup messages indicating
the nature of the service request and any special provisions that may be desired'
Trunk. Circuit or channel between two switching systems.
Two-wny trunk. Trunk circuit that can be seized at either end of the circuit.
Two-wire circuit. Circuit consisting
of a single
pair of wires and capable of
simultaneously carrying two signals in opposite directions.
Unipolar.
Binary line code using single polarity pulses and zero voltage for the two
coding levels.
Unit interval
(UI).
Nominal difference in time between pulse positions at a specif,ied
data rate.
Unspecified bit rate
(UBR).
Class of service of ATM without specific quality
assurances sometimes referred to as
"best
effort" service.
User datagram
protocol (UDP). An Internet protocol without network delivery
guarantees. (I-ost
or errored
packets are not retransmitted.)
Variable bit rate
(VBR). Class of service of ATM with quality assurances as long as
the source adheres to
preestablished average and peak data rates'
Vector
quantization. Choosing one codeword representative of one discrete
waveform of a set of discrete waveforms such that the selected waveform has the
best match to a segment of the signal being encoded.
Vestigiat sideband transmission. Form of single-sideband transmission that
includes a vestige of the deleted sideband and a small amount of carrier energy-
Virtual circuit. Path through a packet-switched network through which all packets
associated with a
particular
"connection"
flow.
Waiting time
jitter.
Timing
jitter
caused by waiting for a timing adjustment
opportunity beyond the time when the need for an adjustment arises'
Wander.
Long-term variations of the signif,rcant instants of a digital signal from their
ideal positions in time, where long term implies
phase oscillations of frequency less
than or equal to a demarcation
point (typically l0 Hz) that is specified for each
interface rate.
Wavelength division multiplexing
(WDM). Use of multiple optical wavelengths
to
carry multiple optical signals.
White noise. Noise with a flat frequency spectrum across a band of interest in which
samples taken at twice the bandwidth or lower have zero correlation.
Wide area telecommunications services
(WATS). Service that permits customers
to make
(OUTWATS)
or receive
(INWATS
or 800 Service) long-distance calls
and to have them billed on a bulk basis rather than individually.
X.25. International protocol standard for accessing packet-switched networks.
ANSWERS TO SELECTED PROBLEMS
1.1 2000 pWC, 2(10)-6 mWC
1.3 500 pwC
1.5 250 pW at
-3
dB TLP
3.r l kHz,2 kHz, 3 kHz
3.2 2.43 dB degradation
3.4 49.4 Mbps
3.6 38.5 dB
3.8 0/110/0110
3.10 23.5 dB
3.14
(a)
13;
(b)
33
(c)
Quantization
Sample Noise Noise Power SQR
(dB\
30.2 .2 .04 43.6
.64 43.7
. 16 75. 3
. 81 79.1
3.17
(a) 3 dB
(b)
1.76 dB
3.19 SQR increases by 20.6 dB
3,21 0/l I 0/000 1, 0/ I 10/000 l, 1/1 1 0/000 1, and I /1 10/000 I
(noninverted)
4.1 37.5 bits
4.3 1.76 dB
4.5
(a)
Average frame time
=
.097 seconds.
(b)
Maximum average reframe time
=
0.193 seconds.
4. 7 141bi t s
4.9 94 ms
4.ll
(a)
1.99 dB
(b) 3.78 dB
4.15
(a)
81
(b)
19
(c)
18
123.2
-.8
-2336.4
.4
8080.9 .9
631
632 ANSwERSToSELEcTEDpHoBLEMS
4.17
- 1
+l
-3
+l
+ 2 + 6 0
+ 4 0 - 6
+2 +4
-6
(a)
(b)
(c)
l +D
l -D
l -D2
|,
-4
a
+4
0
0
a
+2
5.1 250
5.2
(a)
621
s.3 (a)
.11
(b)
.2s
5.5 (a)
55,296
s.7 (a)
4l (b)
10,828
5.9 (a) Total number of memory bits
=
48,0fi).
(b)
Complexity
=
1504.
5.11 900 bits
6.1
6.2
6.4
6.5
6.7
6.9
6.11
6.13
(a)
.5 dB degradation
New BER i s l (10)-8
1600 Hz
(a)
lTVo of peak
signal power.
(b) 15.3 dB reduced error distance.
(a)
7 .14 dB advantage for 32-QAM.
(b)
4.78 dB advantage for 32-QAM.
1l.2 dB degradation in eruor distance
1.5(161*r
3 d B
.188 dB/km
l 6 km
Al" at 1300 nm is 5 times larger than Al, ar 1550 nm.
(a)
2m for mBlP (b)
m + I for mBlC
(a)
Minimum DSl rate
=
1.542 Mbps.
(b)
Maximum DSI rate
-
1.546 Mbps.
(a)
Minimum rate
=
44.712 Mbps.
(b)
Maximum rate
=
44.784 Mbps.
68.7
psec
7.1 195 Bits if the receiver knows the nominal data rate.
370 Bits to accommodate peak+o-peak
shifts.
7.3
+1.3
UI with 99% probability
7.5 5.37(10)-5 misframeVsec (once
every 5.2 hoursT
7.7 ?5.6 dB relative to I UP
7.9 l .ZVo
7.11 (a)
TIE
=
+4.736
g1
(b)
MTIE
=
44.7?i61JI.
8.1
8-3
8.5
8.1
8.9
8. 11
8.13
ANSWERSTOSELEoTEDPHOBLEMS 633
9.1 6500kbps.
9.3 I1.4 kbps
9.5 Channel I output
=
+7; channel 2 output
=
+9
9,7 SIR
=
12
dB
10.1
(a) 2.5 kbits/message
(b)
100 bits/message
10.3 Total number of bits is 86 bits.
f0-5
(a)
0.aa5
ft)
delay is increased by 30 milliseconds
11,f 1250 meters
11.3 56 to 62.67 kbps
l2.l
(a) 40 calls/hour
(b)
.989
(c)9.5Vo (d)
.19 E
(e) .453
12.3 t.4E
12,5
(a) 15 ports (14 ports with finite source analysis)
(b)
B
=
97o
12.7
(a) 2.2Vo
(b)
rEo
12.8
(a) B
=
307a
(b)
add 6 channels
12.10
(a) At
*.it
=
32 E.
(b) 46 circuits required for B < .57o.
12.13
(a)
B
=
lOVo
(b)
B
=
.787o
12,14 2 WATS lines
12.16
(a)83.3Vo (b).833
sec
(c)30.6Vo (d).833
Abandoned calls, 48 Automatic protection
switching, 210,322, 411
Abbreviated dialing, 1 6, 17 Automatic repeat request (ARQ), 1 70, 459
Access tandem, 12
Adaptive transmitter power conhol
(ATFC),
323 Backward estimation, 120, 136, 146
Advanced mobile phone service
(AMPS),53,70, Bad frame masking,440
437
-454
Bandwidth distance
prnduct (BDP),
388, 390,
Advanced Research Projects Agency (ARPA/ 393,406,435
ARPANET),456,468,473 Banyan network,482,483,494
Aliasing,95,96 Baseline restoration, 172
Alphanumeric pagiflg,444 Batcher, K. E.,483,493
American National Standards Institute
{ANSI),
4, Battery, 35, 47,272, 511
22t , f , 81 Baud, 54, 184, 185, 309, 326, 502
Analog bridge, 86 Bell, Alexander craham, 6, 383
Analog interfaces,47,62,63,87
,132,2' 10
Benes, V. E., 483, 493
Analogradio,6,63-65,85, 189,284,303,320, Bemoullian,542
383 Binomial probability
disribution, 527
Asynchronous transfer mode (ATM),
3, 154,204, Bipolar violation, 175-178, 193-199, 214,215
208, 310, 33 1,
,159*504,
540, 564-568
Bit insertion. 168
adaption layers (AALs),484-493,
566,568 Bit leaking,428
available bit rate, 475,476 Blocking
calladmissioncontrol,479,489
blockedcallsclemed,523,530,569
cell discarding,489,566
Clos, 232
cell loss,480,489
Jacobaeus, 238,239. 273,274,525
cell packing,487
t,ee graph, ?0, 155, 221, 234-239,2ffi,262,
cell transfer delay, 476
271,274, 525
circuitemulation,485,487 probability,16,17,25,217,234-275,376,
constant bit rste,415,484 481-570
quality
of service (QoS), 476, 484 BORSCHT, 47, 272
switching,477,484,564
Bragg gratings,403
synchronous rcsidual time stamps, 485 Bridged tap, 88, 185, 316, 495,497
,507 ,515
taffrc shaping,489 Bridging clips, 267
unspecified bit rate, 47 5, 416 Burst errored second, 204
variable bitrate, 475, 476,484,487 Busyhour, 234,241,265,522, 540, 569, 57O
virtual path connections, 477 Busy tone, 42, 43,374,380
Automatic call distributor
(ACD),
76,226,555,570 Byte stuffing, 409, 415, 416
Automatic gain
conbol
(AGC),
120, 506
Automatic number identification
(AM),
12 Cable modems, 5l l, 512
INDEX
635
636 INDEX
Cable TV, 65, 387, 5l I, 516
Call congestion, 541
Call distribution, 226
Call forwarding, 16, 18
CalI tracing, 16
Call trarsfer, 18
Call waiting, 16, l8
Catrtier recovery, 28, 282, 297, 335, 350
CCIR (International
Radio Consultative Commit-
tee),4, 320
CCS (hundred
call seconds), 521, 522,569
Cellular digital packet
data (CDPD), 288,453,454
Cellular geographic
service area (CGSA),
53
Central limit theo,rem. 447
Central office terminal (COT),508,
510
Centralized attendant, I I
Centrex, 17, 18,50
Channel banks
analog
A5. 26
LsE,29
LMX. 28
digital
Dl, 107-1 14, 2t0, 211, 222, 4tr
Dl A, 59
D2, 109, 110, 156, 187_189.223
D3, 103, 109, tl4, 156, 160,212,220,222.
4l l
D,l, 109,212.4r3
D5, 109. 178. 215
Circuit switching, 3, 269, 4#, 47 5, 480, 5 I 9, 520,
s39
Class of service. 473, 476
Clear channel capability, 178,496
Clipping, 25, 40, 540, 541, 545, 541
Clock recovery, 60, 297
Clos, Charles, 232, 273, 482
C-message weighting, 35, 41, 99, 1 14
Coaxial cable, 18, 26-30, 16t*176
Code blocking, 379
Code division multiple access (CDMA),
151, 365.
M5-454
Coherent modulation, 283, 291. 297, 321
Coin telephone, 508
Compact disc,9l,92
Companding, 37,91, 106-128, 317
AJaw, 115, 116, 130, 132, 154, 155, 272
instantaneous, I l6
nearly instantaneous, l2l
F255,
109-114, 159, 160, 174,272
syllabic, ll6
Competitive local exchange canier (CLEC),
12
Compressed voice, I 51, 154, 478
,484,487,494
ConcenEation,24
Conferencing, n2,214
bridge, 87, l40' 2'72, 273
digiral, 87
Congestion theory, 265, 52Q,547
Constant envelope, 288, 291, 301, 309, 314, Ml
Constant holding times, 528, 529
Constant service times, 489, 554-566
Constraint length, 2O5, 2W, 223
Cordless phones
crz, 133
DECT. I33
Crossconnect switches, 46, 52, 21 8, 226, U5,
265-26e,362. 492. 510
Crosstalk, 34,293,495
CSMA/CD,453
Customer service unit
(CSU),215
Cutout fraction, 546
Datagrams, 465, 467, 468
DATAPAC,465
Dataphone digital service (DDS),63,
168, 268.
456
Data under voice (DW),
64
DATRAN,456
dBrnC/dBmC0, 36, 42, 7 2, 101
dc balance, l8l, 395, 399. 500
dc restoration, 172, ll3,396, 502, 514
dc wander, 172-194,396, 398, 502
Decision feedback, 173, 317, 328
Decorrelation, 262, 482
Degraded minute,2O4
Delay systems, 519, 522, 539, 552^562
Delay variation, 413, 47 4
Demand traffrc, 537, 569
Desynchronization, 426, 427
Differential detection, 298
Differential encoding, I 83
Diffractive grating,
40I
DiffServ.473
Digital cellular, 3, 54, 81, 84,91,93, 133, 141,
15 1, 1 53, 277
-288,
329, 437 488
code di vision multiple access, l5l
,
365
,
444-
452
global
system for mobile communications,
148, 157,288,441454
Noflh American digital cellular, l5l, 437
Digital circuit multiplication (DCM), tU, l3.L,
133, 141, 540, 546
Digital loop carrier (DLC),
67, 121,26i,269.
507r509
integrated digital loop carrier (IDLC),63,267
,
508. 509. 517
next generation
digital loop carrier (NGDIf),
509, 510
subscriber loop carier
(SLC), 62
subscriber loop multiplex
(SLM),
62, l2l
universal digital loop carier system (UDLC),
267. 507. 508
Digital pad,514
Digital phones,504
Digital radio, 3, 53,64-85, 169,204,278-333,
382, 406, 456, 515
Digital sense multiple access with collision detec-
tion (DSMA/CD),453
Digital signal
processing (DSPJ,40, 82-88, 130-
138, r49, 165, 267, 271, 1W, 315, 330, 4?9,
504. 506. sr3
Digital speech interpolation (DSI),
124
Digital video broadcast
(DVB),
33, 69, 205, 316,
J J I
Direct bruadcast satellite
(DBS), 33
Direct inward dial (DID),48
Direct progressive control, I 3, 45, 483
Ditect satellite service
(DSS),
33,496,516
Distortion
amplitude, 23, 37, 60, 132, 165, 328, 506
delay, 389
foldover,95, 105, 134
harmonic,55, 132
phase, 37, 55, 164, 328, 506
quantization,
57
Diversity
aflgle, 327
ftequency, 3 l, 32, 326
space, 32,328
Doppler, 337,119
Dual tone multifrequency (DTMF),43-49, 83, 153
Dynamic power confrol, 5I5
Dynamic range, I03, I 14, 304, 349
Eb/No, 192
Echo,23, 34, 39-55,78-87, 120, 154,270-273,
437, 441, 41 8, 500-502, 5 13
cancellation, 39, 40, 46, 51, 82, 154, 27 1, 273,
437
,44t,502
listenet.39
suppression, 39, 40,46, 55, 82
Elastic storcs,339
Electxomechaflical switching, 1 3
Elecnonic Industries Association
(ElA),
4, 438,
445, 450
Elechonic serial numbers, 45 1
Emergency calling
(91 l), 12, 511
Encryption, 73, 7 4, 81, l4l, 169, 350, 451,
5I2
Bngset distribution, 542
Enhanced specialized mobile radio (ESMR),453
Envelope delay, 38, 54
Envelope detection, 282, 297
607
Equalization, 55, 82, 164, 173, 3M, 315, 317,
328,437
,495,507, 5 13
amplitude equaliz-ation, 315, 507
frequency domain,507
phase, 165,507
quantieed feedback, 173
Erlang, A. K., 454, 521-563
Erlang's delay fonnula, 556, 563
Erlang's loss formula, 531, 54I
Erlang's second formula, 556
Enor contol, 45, 80, 204, 37 6, 440-47 4
convolutional coding, 205, 207
,221,
440,443
correction, 1 19, 138, 154, 17 5, 195, 204, 354,
423453
cyclic redundancy check
(CRC),79,
198, 204,
213-216, 223, 331, 440, 443, 460, 470
detection, 203,1tr;0, 41
forward error corrcction, 2O4, 506
Reed-Solomon (RS),
205, 207
,223,
453,506
Errored seconds. 20/'. 216
Error free second, 204
European Telecommunication Standards Institute
(Ersl),
69, 316, 331, 44I, 509
Exchange area, 9-77
,1OB
Exchange Carriers Standards Association
(ECSA), 4, 406,
s01
Exponential distribution, 525, 528, 567
Exponential holding times, 528., 529, 567
Exponential service times, 489, 554, 555, 559,
560, 562, 564-566
Extended superframe
(ESF),
199, 212-216, 223
Fa*imile, 46, 54, 80, 92,132, r41, 152, 350
Fade margin, 31, 323, 325, 326, 328, 448, 515
Fa<les, 30-32, l2l, 201, 210, X22-329, 444-449
Failed seconds sfate. 216
Fast fourier tansform (FFI),
315, 316, 506,507
FCC mask,321
Feeder, 22, 69, 226, 265, 270, 325, 387, 5 10
Fiber
dispersion,3S9
dispersion shifted single mode (DS-SMF), 390,
393.4?,5
distributed data interface
(FDDI),
398,434
fiber channel, 399
in the loop (FITI"),
510
multimode, 385, 387, 388, 389, 401, 4M, 435
to the home
(FTTH),
386, 5l I
,
5 l2
to the curb
(FTTC),
510, 512
FT Series G, 401, 433
FT3C, 388, 401, 433
Filter partitioning, 31 8, 320
Flag, 169,469,494
638
Flow conhol, 43
,376-379,4fu-470, 476, 489,
561. 567
Foreign exchange (FX), 17,46, 267,508
Formant frequencies, 127
Forward estimation, 120, 121, 136, 146
Fractional speech loss, 546,547
Frame alignment signal (FAS),
213
Frame relay,466,472
Ftame relay access device (mAD),472
Frarning
added digit,210
bitenors, m3,216
byt e, 215, 364
loss oi 203, 209,210.485
statistical. 209
unique line code,214
Frequency agility,444
Frequency hoppinE,444, 445, M8
Frcquency
justification,
414, 415
Fricatives.123
Full width half magnitude (FWHM),
3,gZ,Ns
Gaussian, I 90, I 9 l, 202, 2O3, 222, 286, 333, 344,
47
Geostationary satellites, 338, 451, 516
Gl are, 47, 48, 379
Global positioning
system (CPS), 365, i66, 372,
373
GR-303, 509, 5t 0
Grade of service, 234, 241, 252, 256, 258, 377,
457,466,536, 554, 558. 569
&ading,227
Grooming, 267, 510
Guard time, 85, 86,438,442
Hierarchical network, 6, 8, 5l, 52, 535
High definition TV (HDTV),490,
507, 515
High level data link conffol (IIDLC), 169,214,
449,470,472,494
Hook flash,43
Hybrid, 22, 23, 38, 40, 4'1, 54, 93, 148, 15 1, 270-
2t2
Hybrid fiber coax
(HFC),
5l I
, 51 2
Impedance matching, 23,,10, 271
Incumbent local exchange carrier, 12
Lrfinite source, 525, 531-569
Infbrmation capacity, 506
Information density, 85, 278-288, 3 10-317, 441
INMARSAT,33, I48
Institute of Electrical and Electronic Engineers
(rEEE),4
Integrated sewices digital network
flSDN),
l,
456,474-516
basic rate interface (BRI),69,
173, 185,214,
316. 495-510
broadband integrated services digital network
(BrsDN),474
D channel. 189.496-503
L430. 499. 516
I.,l4l,503
I . 451, 503
NTI/I.IT2. 498. 499
primary rate interface (PRI),46,
69,496
Q.9?1,
so3
Q.922,472
Q, 931, 503
S interface.498.500
S/T basic rate interface, 2 I 4
S/T interface. 499. 501
TElrrEz,498
U interface.50l. 502
Intelligent network, 49, 70
INTEL$AT,32
Interexchange carrier (IXC), 10, 12,406
Interference
adjacent channel, 288, 317-320, 448
cochannel.329
electromagnetic (EMI),
280, 393
intersymbol, 54, 60, 74,98, 163-188, 288
multipath,322
mutual,43,,14, 80, 84, 85
narrow band. 317,495
krtemational mobile equipment identity (IMEI),
444
International mobile subscriber identity (MSI),
444
Intemational Standards Organization (ISO),
5
International Telecommunication
Union
(IIU),
4,
4l
Intemet, l,61, 141, U\453-515
access, 1, 453, 504, 515
Engineering Task Force (IETF),473
protocol (IP), 204, 473, 474,49D-494
service provider (ISP),
61, U,l,5M
telephony (IP
telephony), 473
TCP/IF.473
user datagram protocol (UDP)
473
I ri di um. 45l . 516
Jitter 38, 176,210, 324.337-358
rnapping,359
measurements, 342
phase,
343, 344, 358, 417
removing,359
systematic, 338
waiting time, 339, 358-360, 3M, 416
Justification, 343, 351, 360,419, 510, 528
Kendall, D. C.,555
Laser disc, 341
Li ghtning protec tron, 27 2
Limited availability, 5?, 7 6, 227
Line administration, 16, 241
Line code violation, 214,5m
Line coding, 80, 81, l6l-174, 190,282
2Bl Q, 220, 502, 516
altemate mark inversion
(AMI),
1,74-178,337,
499. 502. 517
alternate space inversion, 499
antipodal, 193, 283, 291
,298,333
bipolar, 172*198,223
binary N-zero substitution
(BNZS),
176-181
8325, 176, 177,222
B6Z3, 177
,
178, 1.80,222
B8ZS, 168, 178
coded markinversion (CMI),
183, 184,395-
398,407, 433
conelative level encoding, 185-1 88
digital biphase
(diphase),
1 81
-1
84, 337, 396,
4m
duobinary, 185, 188
high density bipolar 3
(HDB3),
177
Manchester coding, 1 83, 337, 395-398, 400,
494
mBlC, 400, 433, 435
mBl P. 399, 435
mBnB, 396, 399
3848. 399
4858. 398
5868, 398, 399,405
6888. 398
8Bl 0B. 399
non-retum-to-zero
(NM),
172, l8l-193, 223,
21 9
-288,
298, 305, 392-396, 425, 435
panial response, 165, 185-187, 198,288, 305-
3i l . 320
return-to-uero
(RZ),
174, 188, 389-396,405,
425
temary, lEO
pair selected
(PST),
179-181, 195
483T, tg0,Z22
unbalanced. 171.399
unipolar, l7l, 193, 195,282
Link access procedure
balanced
(LAPB),
469,503
l,oading coils, 23,28,17,54, 60, 88, 190,
.195,
507
Lccal access and transport area
(LATA),
10, 12
Local axea networla
(LANs),
471
-473,
485, 498
distributed queued dual bus
(DQDB),
219
emulation
(LANE),485
ethemet, 183,453,485
INDEX 639
token passing
ring, 218,485
802. 3, 183, 485
802. 5, 218
842.6,119
local exchange canier
(LEC),
10, 406,
487, 509
Local microwave distribution service
(LMDS),
496, 515
Local multipoint communications systems
(LMCS), 515
Longitudinal currcnt, 2 I
Loop timing, 340, 346
Loss systems, 519, 553, 554, 561
Lost calls cleared, 523, 53 1-569
Lost calls held, 539, 54,0
Loudest talker, 273
Low earth orbit satellite,451, 516
Ml2 multiplexer, 61, 354, 356-359,361,429
Manual swirchbosxds, 12
Matching loss,265
Maximum average frame tine,2l1, 214,222
Maximum likelihood, I 75, 184, 207, 312
Maximum time interval enor
(MTIE),
367-373,
382
Mesh, 6-8, 52, 535
Message sequencing, 376, 503
Message switching, 455462, 552, 559
Message unit accounting, 16
Metallic curent, 2l
Mobile assisted hand off
(MAHO),440
Mobile telephone swirching offrce
(MTSO),
52,
53
Modems
v. 32, 40
v. 33, 315
v. 34, 88, 279, 303, 315, 331, 495, 513, 5r4,
5L'7
v.90, 55, 153,279,495,5 13, 514, 5 17
Moduletion
analog
amplitude, ?6, 94, 98, 279-283,297,301,
305
double sideband, 94
frequency
(FI\4),
30, 221, n7,284, 331,
387,453
i ndex. 280. 286
linear,280
phase,284,286
pulse amplitude
(PAM),94-98,
107, 158,
M6, 274, 513, 514
single sideband, 85
suppressed carrier, 297
,
322
640
Modulation
(Ca
ntinued)
digital
carrierless amplitude and phase (CAP),
280,
309.504
continuous phase frequency shift keying
(CPFSK),285
discrete multi-tone (DMT),
316, 317,504,
507
frequency shift keying (FSK),
284-286
gaussian minimum shift keying (GMSK),
286, 330, 44r, 453
minimum shift keying (MSK), 285, 286,
308, 309, 315, 330, 331
multilevel coded (MLCM),
315
ott/off keying, 282
orthogonal frequency division multiplexing
(OFDM), 316
phase
reversal keying (PRK),
283-290
phase
shift keying (PSK), 283-302,438,441
quadrature
amplitude (eAM),
88, 288, 301-
328
quadrature partial
response signaling
(QPRS), 310, 3i l
trellis, 207, 312,314
MPEGIA{PEGz,490
Multichannel multipoint distribution service
(MMDS),496,
515
Multiframe alignment signal, 2 I 3
Multipath, 30-32, 53, I2l, 322-329, 437, 444-
449
Multiplexing
frequency division multiplexing (FDM),
26*
86, 207, 21 6, 303, 3 17, 402, 437 453,
5t7
.
guardbands,44S
hierarchy,27, 28
mastergroup, 28, 63
supergroup,28
time division, 58, 444, 461
analog,74
asynchronous, 5S, 208, 351, 461
hierarchy, 74,351
l oop, 216-218
ri ng, 216
statistical (statmux),
208, 461-463, 552
synchronous, 3, 58, 2O8, 352,420,46I
Multiprotocol label switching (MPLS), 491
Net loss, 34,39, 41, 42,82,271,384
Network cofltrol point (NCP),
50
Network contol protocol (NCP),473
Network management, 43, 376,377, 407, 411,
413.477 - 520
Noise
background, 1 47
-153,
445
Gaussian, 190, 1 91, 202, 222, 333, 3M
idle channel, 72, 81, 102-120, 273
impulse, 35,55, 60, 80, 190,203
mode partition,
392
quantization
noise,35,99-l19, 128, 135, I38.
159, 160, 317. 513
thennal noise, 35, 3U' 333
white noise, 35.99. 190
Noise bandwidth, l9l, 192, 298-320
Noise figure, 29, 324, 3?5
Noise power,
36-42
Nordic mobile telephone (NMT),
54, 437
Normal distribution. I 9l
Nyquist sampling rate, 94, 99, 149, 159*165, 305,
326
Offered load, 244, 265, 377, 529-569
Offset keying, 305
Open systems interconnection (OSI),
5, 498, 5 17
Open wire, 18
Overflow traffic, 519. 548-553
Overvoltage protection,
272
Packet switching, 208, 456472, 490, 51 9, 552-
566
Paging, 49, 4a+, 451, 516
Pair isolation, 185
Parity, 199,409
Passive photonic
loop, 401
Pathfinding, U2-24
Peak cell rate,475-490
Performance monitoring, 79, l7 l-182, 195, I 99,
2rs,2r6,376.400. 509
Permanent virhral circuits, 4#,466, 491
Petsonal access communications system (PACS),
133
Personal communication system (PCS),
444, 450
Personal handyphone system (pHS),
133
Phase locked loop (PLL),
336-343,381,428,485
Phonemes,92
Pick up, 18,385
Pilot, 28, 79,315, 5M
Pitch, 123, l4l 139,144-149
Plain old telephone service (POTS),
54,495,506,
507. 510. 512
splithr,506
Point of presence (POP),
10, 12
Poisson, 203,526,527,529,531, 540, 541, 550.
551
Power amplifiers, 30, ?84, 291, 318, 319
Precoding, I 86
Primary reference clock (PRC),485
641
himary rcference source
(PRS), 186, 187, 311,
3't0,372, 427
,
506
Private branch exchange (PBX), 16, 17
,4'l-49
Private mobile radio, 452
Prcb(delay),556
Propagation delay, 33, 451
hotection switching, 404
Provisioning, 267
Pseudo random sequence (PRS),506
Psophomeric, 35, 99
Pulse code modulation
(PCM)
modem, 153, 513,
515
Pulse *haping, 96, 307, 308
Pulse stuffing, 78, 35 1-364, 46 1
Quadrature
multiplexing, 293
Quadrature
signal representations, 290
Queueing,
458, 488, 520, 552, 555, 567
delays, 458, 465, 558, 564
finite, 561
tandem,566
M/ D/ l , 559
MA{/I, 555, 556, 559
MA,I/N. 555. 562, 567
MA.4INI*|L.562
Radio common carrier
(RCC),52
Radio system availability, 322
Rain attenuation,30, 32, 321-323, 515
Raised cosine, 319
Random refies, 537, 538, 569
Rayleigh fading, 389
Reed telays, I 6, 35
Refraction, 329, 387, 388, 401
Reframe time, 209, 222,355
Regeneration, 59, 60, 74-Bl, 98, l l 1, 167*176
Reis, PhiIIip, l, 69
Remote concentr ator s, 227
Remote provisioning, 5 10
Remote switching, 24, 62, 509
Remote terminal,508
Repeater spacing, 30,65, 190, 195, 323, 388-406
Residual error, 136, 138
Returning traffic,538
Ringback, 42,43,374
Ringing voltage, 43, 47, 48
Routing, 376,463,4&
alternate, 13,379
best effort, 4?3
crankback,51
dynamic,463,4M
fixed path,463
packels,463
satellite, 26,32, 54,75, 86, 148-154
Molniya,32
Westar,33
Scrambling, 81, 170, 322, 396,410,425
Seaplow,404
Second generation cordless telephones, 133
Segregation,26T
Sequence numbers,470
Service control point
(SCP),50
Severely errored seconds, 21 6
Shannon' s theorcm,513
Short message service
(SMS),444
Sidetone, 39
Signaling, 13, 14, 42-88, I 84-33 1, 500
A and B bi t s, 212, 216
channel associated, 213
comrnon channel, 18,43-53,75, 109, 153,
[l 8, 214, 377-380, 421, 496
E&M. 47. 49
ground start (GS),48
loop start (LS),
47, 48,
508
per tuk,43
robbed bit, 109, 178, 216, 514
wink,48
Signaling system
(SS7),
50, 52
Signaling tones, 153, 31 4
multifrequency (MF),43,49,
83, 88, 153
single frequency
(SF),
43, 49,83, 153,212,
216, 222, 336, 391
sin(.r)/x, 162, ll2, 28Q, 298
Singing, 34, 38, 39,4I, 120, 154, 27o,nl
Skyphone, 33, 69, 148, 157
Slips, 216, 341
-350,
362
buffer,420
controlled, 349,485
rate, 349,350
Slope overload, 135, 136, l4l
Soft handoff.450
Source code restriction, I 68
Space expansion
,235-262,
525
Space switching, 67, 257
Span line, 59, 60, 181, 215
Span switching,431
Specialized mobile radio (SMR), 453
Spectogram, 127, 14
Spectrum spreading, 284, 291, 386, 445, 446,
448,451
Speech compres sion, 124, 437, 441, 49
Speech pauses, 34, 80, 8l, 102, 120, 122,124,170
Speech recognition,46, 86,93, 147
Speech storage,9l
Splices, 88, 387, 390,404,405
Spread spectrum
,
445, 448, 450
Star network, 217, 5lO
642
Statistical equilibrium, 531, 537
Stored program
control ($PC),
16,
,14,49,
51,
225,24
Stuffing ratios, 360, 429
Submarine cables, 384, 403, 540
NLl . 398. 434
Trans Pacific.400
Subscriber identity module (SIM),
M, 451
Subscriber loops
analog, line interfaces, 272,5M
digitar, 75, 87, 185, 204, 27o,456,495,49' 1,
503, 516
asymmetric digital subscriber line (ADSL),
3 10, 3 16, 317, 331, 495, 503-5 16
interfaces,5I0
vDsL. 507. 510
XDSL, 503, 507, 509
Superframe, 212, 21 1, 215, 419, 502
Supervision, 47, 272
Supervisory audio tone ($AT),439
Swirched multimegabit data service (SMDS),4&4
Switched viilual circuit (SVC), 46/.,470,487
Swirches, network
class 3, 56
class 4, 6, 12
class 5 (end
office),9-87, 121,225,226,Mt,
. 262,270,379, 380, 502, 508, 5?5
crossbar, I3-16,49,66
DMS, 16, 66, 272, 274
electronic switching system (ESS),
16,49,52,
65, 7 L, 225, 233, 259
_27
4, 502
No. 1, 16, 65, 225, 233
No. 101. 273
No. l A. l 6
No. 2, 16
No. 3. 16
No. 4, I 6, 49-7 l, 225, 259, 262, 273, 366,
381
No. 5, 16, 272, 274. 502
No. 5 EAX, 16
EWSD, 260, 273
stcp-by*step, 13-55, 225, 244, 483
t ol l , 6, 9, 10, 67, 236
Switching, time division, 246
analog,246,252
digital, 66, 67
,226,238, U7
,253,340, 548
rnemory switch, 248, 478-480
sTs.255
TSSSST,262
TSSST. 260. 261
TST, 255*260, 27 5, 348, 48Q, 482
Synchronization, 85, 89, 221, 331,3#-381.427
clock, I I l, 209-211,335, 351, 374.513
frame,59,209,214,347
hieraxchical, 374
net work, 85, 210, 335, 351, 362, 485
isochronous, I 67
mutual,36?.364
network master,362
plesiochronous,
362, 363,370, 381
stratum. 370*373.382
Synchronous data link protocol (SDLC),469
Synchronous digital hierarchy (SDH),
5, 352,
4Q6,407,434
Synchronous optical network (SONET),
3,5,52,
78, lM, 219, M9, 269, 352, 373, 396, 405_
4l l
asynchronous mapping, 419, 420
concatenation, 267, 269, 408
gateway
network element (GNE),
426-428
line overhead, 41 Q414, 425
path overhead, 410425, 47 8
pointer
adjustments, 4 I 5, 426, 427
pointer
burst, 428
rings, 52, 219,405,429
bidirectional line swirched (BLSR),
431, 432
unidirectional path switched (lJpSR),
429,
435
section overhead, 410, 411,425
synchronous payload
envelope (SPE), 41 3, 414
fransport overhead, 4O8-410, 415
virtual tributary (VT),
417,429
Synchronous transmission, I 67
System availability, 322, 326, 403
System gain,
323,405
T-carrier, 59, 63, 132, 167
-178,
2lO, 215, 222,
268, 335, 337, 340
El, 59, 177, 213, 359, 382, 40' t
_41
t, 421, 435,
484. 497. 510
82, 330, 354, 359,407.413
H, t84, 407, 423. 436. 478
repeaters, 60, 62, I I l, 358
TI 48. 181
Tt c, 56, 60, 185
Tl D. 56. 188
Tl c, 185
T2, 56, 60, 6t, 67, 69, 169, 177,220
T3. 56
T4, 56
T4M, 169, t70,2l9
Tl committee,4
TI , 4, 5, 56-67, 84, 91, t 08, I I I , 132, 156, t 67_
174
Tag switching,49I
Talker echo. 39
Tandem clock recovery, 338'
Tandem switching, 8, I, 12,66, 67,226,236,237,
370. 535. 569
Telephoto,54
Television distribution, 29, 33, 54
Tic lines, L7
,46,
U0,268, 553
Time assignment speech interpolation (TASI),
25,
124, 540-547, 568
Time compression multiplex (TCM), 207, 280,
312- 314, 502, 516
Time conge$tion, 541, 542,563
Time division multiple access (TDMA),
86, 89,
154,438-452,512
Time expansion, 258, 259, 261, 26?, 482
Time interval error
(TIE),
359, 367,369,382
Time slot intetchmge
(TSI),
l'1 8, 249
-254,
269,
275,348
Time variance
(TVAR),
367
,
369,370
Tip and ring,48,499, 508
Toll network, 6, 9-34, 49, 60-69, 108, ll3, 271
Traffic intensity
,
241, 521-561
Transmission level point (TLP),
41, 42, 7Z
Transpondcr, 33, 286
Transvemal filters, 328
Trunk directionalization, 379
Trunked radio,452
TYMNET.465.468
U.S. Independent Telephone Association
(usITA),4,
366, 370
U.S. Telephone Association
(USTA),
4
Unit interval
(UI),
343, 344,361
,382
Universal coordinated time, 372
v5, 419, 422, 509, 5r7
Vector quantization, 150
Via net loss
(VNL),
39,41, 82
Video, 77, 2Q4,2' 17
,387 ,474-515
Virtual circuit, 464-469, 477
,487 ,491
Viterbi detectors, 175, 184, 195, I 98, 201,220,
3l 2. 33 I. 454
Voiceband data, 38,40,55, 83, 98, 124, 132, 140,
r4tt, 161, 167, 265, 277, 303, 350, 485, 51 I
Voice coding
adrlptive delta modulation (ADM), 136, I40,
156. 429. 510
adaptive difTercntial PCM
(ADPCM),
132-
140, 153- 156, 215
adaptive predictive coding
(APC),
136, 138,
r 49, 151, 157
analysis and synthesis, 92
channel vocoder, 141-14
delta modulation, 62, 69, 9l, 93, 133-136,
r4t . 153. 2M
diff'erential pulse code modulation
(DPCM),
93, I 27- 136, 141, 153
easily digitally linearizable coding
(EDL),
108,
110,27?.
formunt vocoder. 141. 144
643
Iinear predictive coding, 92, 138, 141*158,487
algebraic code excited
(ACELP),
154, 155,
487
code excited
(CELP),
150-1 58
multipulse
(MPLPC),
147-149, 151
pulse excited (pelp),
I 47
Qualcomm
code excited
(QCELP),
l5l,449
residual excited (RELP), 147, 149, 151. 158
vector sum excited
(VSELP),
151, I 58, 439,
453
pulse
code modulation (PCM), 71, 88-l14,
128, 136, 140, 159
prediction gain, I 36
subband, 138, 139, 142
vocodem, 93, 102, 123, I 27, 141-153
c. 704. 213
G. 7l l . 154
G. 12t , t 32, 133, 154
G.722, 140, 154, 157
,?15
G. 723. 154, 155, 494
G. 723. r, 154, 155, 494
c.726. 154. 155
G. 727, 155
G.728. 154. 155
G; 729, r54, 155, 158, 487
G. 8t l . 363. 381
G, 821. 2M. 216
Voice messaging, 91, 92, I 32
Voi ce qual i t y, 80, 99, 102, 113, 133, 146, 151,
443,547
broadcast, 15l
communications, l5l, 152
djagnostic acceptability measure (DAM), I47,
151
diagnostic rhyme test
(DRT).
147, l5l
mean opinion score (MOS),
133, l5l
signal-to-quantizing ratio
(SQR),
I02-1 15,
159. 160. 513
synt het i c, l 5l , 154
t ol l . 132. 151-155
Voltage controlled oscillator
(VCO),
28 5, 33 6-
343, 485
WATS, 17, 380,553, -570
Wavelength division multiplexing (WDM),
26,
385-405.434.415
Weinstein, C. J.,546, 568
Wire gauge changes,497
x.25.469-472. 503
Zero bit insertion, 169, 469
Zero byte time slot interchange, 178
Zero loss transmission, 270
ghrltntrnf
sftrtfi;r
Btr
- er r t l hy
-
r TTTTT
[T.T.E.