18

Improving Throughput and Availability of Cellular

Digital Packet Data (CDPD)
Jay M. Jacobsmeyer
Pericle Communications Company
P.O. Box 50378
Colorado Springs, CO 80949

Abstract
channel, called the channel stream , is a
Cellular Digital Packet Data (CDPD) is a shared medium employing digital sense
packet radio system designed to provide multiple access (DSMA) with collision
reliable 19.2 kbps data service over the detection (CD). The system uses its own base
North American cellular phone system. The station, called a mobile data base station
chief obstacle to high speed data transmission (MDBS) and communicates with data
on CDPD is multipathfading. Time-varying mobiles, called mobile end systems (M-ESs).
amplitude and phase force CDPD to operate Calls are routed by intermediate systems
at a throughput far below that of equivalent (ISs), also called routers. CDPD is described
non-fading channels. A new adaptive data in more detail in a specification [1], published
rate technique can improve throughput on in July 1993. Service will be available in
CDPD by more than a factor of 5. This several cities in 1994.
technique adapts the data rate by varying the
number of information bits, k, of a klk+1 CDPD is a wireless channel and is therefore
trellis code emplo ying M-ary DPSK or subject to a time-varying carrier-to-
CPFSK modulation. This adaptive trellis interference ratio (Cll). The two main causes
code ensures that we can raise the data rate of this time-varying CII are Rayleigh fading
without expanding the bandwidth. A novel and shadow loss.
soft-decision Viterbi decoder performs error
correction, signal estimation, and rate 2.0 Effects of Rayleigh Fading on
changing in a single device. The built-in Receiver Performance
estimator automatically estimates the channel
carrier-to-interference ratio with a standard Multipath fading on cellular channels causes
deviation of 25 dB. An added benefit of the time-varying amplitude and phase that
adaptive rate scheme is an increase in the seriously hamper high speed digital
service area of 17% for an omnidirectional communications. To maintain acceptable bit-
error-rates, CDPD must operate at an effective
cell and 9% for a directionalcell.
data rate that is far below channel capacity. In
other words, a fading channel requires
1.0 Introduction significantly greater power to achieve the same
error rate performance as the equivalent non-
Cellular Digital Packet Data is designed to fading channel.
provide cost-effective data service over
existing cellular radio channels. CDPD To see why this is so, consider the Rayleigh
operates independently of the voice system, fading mobile radio channel. Although the
using either dedicated 30 kHz channels or average signal energy of the fading channel is
hopping between idle channels. The CDPD the same as the average signal energy of the
228
non-fading channel, the bit-error-rate (BER) 37 dB (5,000 times) more power than the
for the Rayleigh fading channel is several non-fading channel.
orders of magnitude higher. For example, the
probability of bit error for binary frequency Several techniques are used in an attempt to
shift-keyed (FSK) modulation on the non- recover the performance lost to fading. The
fading additive white Gaussian noise most common are worst case design,
(AWGN) channel is [6] diversity, and error-correction coding. Worst
case design requires large fade margins
(sometimes greater than 40 dB) and is
obviously inefficient. Diversity is a bit more
(1) effective (-20 dB improvement for dual
diversity), but often impractical. Frequency
where EblNo is the ratio of energy per bit to diversity requires additional spectrum, a
noise power spectral density. But the valuable resource that is usually not available,
probability of bit error for FSK on the and space diversity is impractical on handheld
Rayleigh fading channel is [6] radios.

A third approach is error correction coding.

(2) CDPD employs a good burst-error correcting
code, the Reed-Solomon (RS) code. The
These functions are ploned in Figure 1. particular code used is a RS (63,47) code over
GF(64). This code can correct up to 8 symbol
errors (a symbol is 6 bits), but the CDPD
specification [1] recommends that the code
10 ·'
correct no more than 7 errors so the remaining
parity symbols can be used to provide bener
error detection performance. Since the Reed-
10 ·' Solomon code provides both error correction
and error detection, a cyclic redundancy check
~ 10 ·)
(CRC) is not required.
l
~
s CDPD employs Gaussian minimum shift
1
0:
10· keying (GMSK), a binary modulation
technique. Most mobile receivers will use
non-coherent detection, and the error
;~l~f~f~~i~ . ;:.;!.;i; .~;;~~~~Jli~;i ;.
10·)
performance will be approximately the same
as non-coherent binary FSK.
. "t". ···1··'7··t· .. ··i·..t..t··t···:··+··t··· ··+··j···t·· -.-
10 • ,
S 10 15 20 2$ JO JS 40 4S SO
Although the Reed-Solomon code is a
powerful burst-error correction code, it is not
failsafe, and many Reed-Solomon blocks will
be uncorrectable because the fading channel
Figure 1 - Probability of bit error for AWGN and
will introduce more than 7 symbol errors. To
Rayleighfading channels
handle this problem, CDPD will implement an
automatic repeat-request (ARQ) scheme and
Note that at a bit-error-rate of 10-5 , the will re-transmit uncorrectable blocks. The
Rayleigh fading channel requires almost CDPD protocol makes it impractical to
 229
retransmit individual blocks, so an entire noise. The main source of this co-channel
frame must be retransmitted if one or more interference is the first tier of co-channel cells,
Reed-Solomon blocks contain uncorrectable shown in Figure 2 for a 7-cell reuse pattern.
errors. A CDPD frame holds a minimum of 4
Reed-Solomon blocks and the protocol
behaves like a go-back-N ARQ system with
N = 4. From [10], we know that the
throughput for a go-back-N CDPD system is
approximated by

Ravg = (19.2 k) P
cd
Pcd
+ (l-P)N
cd
(!n)
where Pcd is the probability that a RS block is
correctly decoded, and (kin) is the code rate
of the RS code (3/4 in our case).

Computing Pcd for a Rayleigh fading channel

is no simple task because of the strong time
correlation of symbol errors. We resorted to
simulation with a software simulator meeting
the requirements ofTIA IS-55A [11].
Figure 2 - Co-Channellnterference in a 7-Cell
The results are shown in Table 1. Note that at Reuse Panem
a typical cellular CII of 17 dB, CDPD
produces less than half of its rated throughput. As a mobile user travels through the cell, he or
she will experience a time-varying carrier-to-
Table 1 - Go-Back N ABO Performanceof cprp interference ratio caused by terrain and made-
made obstructions. This loss is known as
f.tI1:UJ. fai B..avg shadow loss. At cellular frequencies,
7.0 dB 0.06 240.0 bps
shadow loss varies about 100 times more
9.5 0.21 880
12.0 0.41 2,160 slowly than Rayleigh fading. The main effect
14.5 0.61 4,020 of shadow loss on a conventional cellular
17.0 0.83 7.930 system is a loss of service in certain regions of
19.5 0.92 10,540 the cell due to low CII.
22.0 0.96 12,340
To quantify this effect, we need a relevant
Table 1 serves as a baseline for CDPD
performance and it will be compared in measure of the CII in co-channel cells. The
Section 4.0 to adaptive data rate performance. most appropriate measure is the probability
that a randomly selected point in the cell of
3.0 Effects of Shadow Loss and Co- interest (the center cell in Figure 2) will have a
Channel Interference on Cell Coverage CII of umin dB or greater. In other words,
we want the probability that a mobile user has
Cellular radio is a frequency reuse system and a CII ~ umin dB, assuming the mobile
its performance is usually limited by co- user's position is uniformly distributed over
channel interference rather than Gaussian the cell's area. Typical values of umin for
230
analog voice are 17 or 18 dB. sum of the variances. Therefore, the carrier-
to-interference ratio, U, is normally
The following derivation of this probability is distributed with mean Co - io and variance
similar to Lee's derivation in [2], but we think
our derivation is easier to understand.
(5)
The probability P(U ~ umin) is written as
For convenience, we can write the probability
P(U Sulr) in terms of the unit normal
P(U > U min) = 1- P(U s Umin) distribution, <Il(x),

Jr
RO
= 1- P(U::; U min I r)/R (r) dr
o (3)

where where
fR(r) = probability density function for the
distance from the base station to the user
<Il(x) = _ =1 [ e_l. dy
'127t -
2
(1)
location, R, and
RO = the radius of the cell and u(r) is the mean ell at a distance r
from the base station. Assuming that the
From [9 , pp. 191] , we know that the interference is roughly constant throughout the
probability density function for R is given by cell [4], the mean ell at a distance r from the
base station can be written as
!R (r) = -2r
R5 (4) u (r) = u (R 0) - Ylog(;0 )
(8)
where we have approximated the cell as a
circle of radius RO' where y is the propagation constant (typically
y = 40). Substituting (8) into (6) and (4) and
The conditional distribution of U, P( U S; (6) into (3), we get
ulr), is found as follows: Because of terrain
shadowing and man-made obstacles, the
carrier, C, and the interference, I, at any P(U > .... )
2
= 1-""'1'
R.
r' (.. . -.(R.)+Y"'~*)J
~
a
r dr
point in the cell are random variables. One of
the most widely accepted probability Making a change of variables, s = rIRO, we
distribution for e and I is the log-normal arrive at the following expression :
distribution [2], [4] . Therefore, if eand I
are in dB, they are normally distributed with
mean Co and iO and variance oc 2 and 0/ 2,
respectively. But we are interested in the
random variable, U = e-I.
where dU =u(RO) - umin' This expression
can be integrated using numerical
From [9], we know the the sum of two
in te gration. 1 Equation (9) is plotted in
independent normal random variables is also a
normal random variable with mean equal to lThe function has a discontinuity at s = 0 which
the sum of the means and variance equal to the will cause most numerical integration routines to fail.
 231
Figure 3 (at the end of this paper) for CJc = 6 Partitioning a cell into 3 sectors with 1200
dB (small cell) and CJc = 8 dB (large cell). beamwidth antennas will increase the CII at
the cell boundary. Assuming perfect
Next, we need the value of Au = u(RO) - antennas, only cells 1 and 2 of Figure 4
umin for an omnidirectional cell site. contribute to co-channel interference.
Assuming a propagation constant of y = 40
(i.e., 40 dB per decade path loss), the carrier
power at the edge of the cell is proportional to
the cell radius to the minus fourth power,
-4
C oc R 0

And the carrier-to-interferenceratio is

-4 •
Ro

(10)

where di is the path distance from the center

of cell i to the edge of our cell. Referring to
Figure 2, we can calculate the path distances
to the edge of our cell. These distances are
shown in Table 2.
Figure 4 - Co-Channel Interference with
Using these distances and (10), we find that Directional Cells
the CII at the edge of the cell is 17.8 dB.
Therefore, Au is 0.8 dB for a umin = 17 dB. Using (10), we can show that sectorizing
Note from Figure 3 that this results in a increases the CII at the cell boundary by over
coverage of 74.5%, which is far less than the 6.2 dB. From Figure 3, we see that this
usual design goal of 90%. increase in CII raises the area covered with
greater than umin = 17 dB to 87.5%.
Table 2 - Co-Channel Cell Distances

Distance di
We will show in the next section that an
~
1 5.3 R o adaptive data rate modem lowers umin and
2 4.4 R o therefore increases the coverage area over a
3 3.6 R o conventional CDPD modem.
4 4.0R o
5 5.0 Ro
6 5.6R o 4.0 Adaptive Data Rate Performance

Our approach is to adapt the system data rate

to the time-varying signal-to-noise ratio at the
We recommend the midpoint rule. This expression mobile radio receiver. To understand why this
can be integrated analytically. but the derivation is
approach makes sense, we must re-examine
omitted in the interest of brevity . The analytical result
is in terms of the error function. so numerical the fading characteristics of the mobile radio
integration or approximation is required regardless. channel.
232
The vast majority of bit errors on mobile radio Assuming ergodicity-, we can conclude that
occur during deep fades. As the signal-to- we lose only 63% of our throughput, which is
noise ratio drops below the system threshold, equivalent to a 4.3 dB loss in power over a
the probability of bit error rises rapidly and Gaussian noise channel. Or put another way,
errors are clustered in bursts coinciding with we gain more than 32 dB over conventional
the fade. Even in the absence of intersymbol FSK in Rayleigh fading. We see this result
interference, burst errors still occur because plotted in Figure 5 below.
the data rate is too high for the instantaneous
signal-to-noise ratio . In other words , the
energy-per-bit is too low.
10 ·'
The most obvious way to increase energy-per-
bit is to boost the transmitter power. But
power is expensive, and most systems already 10"

operate at the highest practical transmitter

power. The next logical choice is bit rate. ~ 10"
Since Eb = PtlR (where P, is the transmitter ...
-":

'&
power in watts and R is the bit rate in bits per -~
second), we can double the energy -per-bit
simply by halving the bit rate. An adaptive ) 1 0~

data rate system exploits this phenomenon by

10 ·'
operating at high bit rates when signal-to-
noise ratios are high, and low bit rates when
signal-to-noise ratios are low. But whenever
we lower the bit rate, we degrade throughput.
So the obvious question is: Do we get a net
gain or a net loss? The answer is a net gain, a
fact that was proved mathematically by Bello Figure 5 - Performance of a simple on/off
and Cowan in 1962 [7]. adaptive datarate system

For example, consider the simplest adaptive To achieve this result, we need a system that
data rate scheme: one that operates at rate R operates at a constant data rate of R bits per
bits/s when the received signal is above a second (bps) for 37% of the time and drops
predetermined threshold, and 0 bits/s when instantaneously to 0 bps at the onset of a deep
the received signal is below this threshold. fade. Of course, accurate and timely channel
estimation is required to approach this
performance in a practical system.
To quantify the performance of this scheme,
we recall that the power on a Rayleigh fading Next consider a more sophisticated adaptive
channel is exponentially distributed [6] and data rate system with n available non-zero
assuming that the minimum required signal- rates , r j • rz' ... ,rn. Each rate has a
to-noise ratio is also the mean signal-to-noise corresponding signal-to-noise ratio per bit,
ratio, the probability that we drop below this (Eb/NO)i' i = 1,2,3, ..., n These signal-
minimum is simply to-noise ratios are the thresholds that
2 An assumption backed by experimental evidence,
P = I- e -1 = 63% see Lee [4].
 233
determine when the system can raise or lower and 3 bits/symbol (coded 16-DPSK).
the data rate and still maintain reliable
communications. We normalize the adaptive Table 3 - Sjgnal-lO.nojse Thresholdsfor
system by letting AdaptiveTCM <BER = \0"5)

(EiJNO}i Modulation ~ Regujred~

(EiJNO}mean DQPSK 1(2. 8.3dB

8-DPSK 2{3 10.5 dB
16-DPSK 3/4 14.4dB
Then we can write the probability mass
function for R for an n-rate system as
The improvement in throughput of the trellis-
coded adaptive rate scheme over fading binary
FSK is dependent on the mean signal-to-noise
ratio on the channel. A typical cellular signal-
j = 1,2, 3, ..., n- 1 to-noise ratio is 17 dB. Substituting the rates
2
PR (n} =P(R =rJ =e-a· and signal-to-noise ratio thresholds from
Table 3 into Equation (11), we get
And the average bit rate for an n-rate system Ravg = 2.3 R = 44.2 kbps (12)
on the Rayleigh fading channel is simply

n-I
for a mean E biN0 of 17 dB where R = the
CDPD signaling rate of 19.2 kHz.
«; = L
i= 1 (11) From Table 1, we see that the improvement
over binary FSK with Reed-Solomon error
Now, what type of adaptive rate scheme correction is a factor of 5.4.
should we use, and what are the possible rates
and thresholds? The throughput of this adaptive data rate
technique is compared to conventional CDPD
We could design a system that increments the as a function of signal-to-noise ratio in Figure
channel symbol rate by a factor of 2 whenever 6.
the signal-to-noise ratio increases or decreases
by 3 dB , but such a system would require 57.6
bandwidth expansion and the CDPD channel
48.0
is bandlimited. Adaptive data rate
...
:§' 38.4
A better approach is to use a constant symbol .S
rate and adapt the data rate by varying the
number of bits per channel symbol. We ..
;; 28.8
~
" 19.2
~"
should also employ trellis-coded modulation
(TCM) to improve the power efficiency of the 9.6
system.

The Eb/NO thresholds for a 3-rate adaptive o 2 4 6 8 10 12 14 16 18 20 22

TCM scheme with M-ary DPSK are shown Mean signal-to-noise ratio,E tiNO.dB
in Table 3 below.
Figure 6 - Adaptive data rate and CDPD throughput
The allowable bit rates are 1 bit/symbol (coded ifc =881 MHz, velocity =30 mph)
DQPSK), 2 bits/symbol (coded 8-DPSK),
234
Note that the adaptive rate improvement is a The rate changing algorithm and signal
decreasing function of signal-to-noise ratio. estimator are integral parts of a soft-decision
For example, the gain at 7 dB is a factor of Viterbi decoder, simplifying the
30, but the gain at 7 dB is factor of 5.4. implementationand reducing size.

An added benefit of adaptive data rate is a gain We are addressing several implementation
in geographical coverage. Let's assume that a issues, including the following:
conventional CDPD modem requires a GIl of
17 dB to provide reliable service. In Table 3, • Mobile access and overhead losses
we showed that an adaptive data rate system • Signal estimation
requires only 8.3 dB Eo/NO or Gil to operate • Memory & memory control
at its lowest data rate. 3 For an • Feedback channel effects
omnidirectional cell with 17.8 dB Gil, t1u = • Irreducible error floor
9.5 dB. From Figure 3, we see that such a
system can raise the area covered by at least 5.2 Mobile Access. CDPD is a shared
umin GIl from 74.5% to 91.2%. For a medium. When a mobile is monitoring the
directional cell (3 sectors), the adaptive rate system before access, CDPD is a one-tOoN
system raises the service region from 87.5% system. When a mobile gains access to the
to 96.8% of the cell's area. channel stream, the exchange of data berween
the mobile and the base is a one-to-one
5.0 Implementation Issues system. Busy/idle status, synchronization,
and other control irtformation are interleaved
5.1 Introduction. Adaptive data rate schemes with the data, but this irtformation occurs in
make sense from a theoretical view, but can separate channel symbols. Our modem
high performance adaptive data rate modems operates at a constant symbol rate and varies
be realized in practice? We think so. the number of bits per channel symbol to vary
the information rate. With proper framing
Pericle Communications Company is information, the modem can change the data
developing an adaptive data rate modem for rate on a symbol-by-symbol basis.
cellular radio." The modem combines trellis
coded modulation (TCM) and adaptive data Thus, we can send control irtformation at the
rate operation in a single device. The trellis lowest non-zero rate of 1 bit per symbol, but
encoder operates at a constant symbol rate and the user information can be sent at a variable
varies the bit rate by adapting the number of rate of 0, I, 2, or 3 bits per symbol. For
bits per channel symbol to the channel example, consider a generic packet radio
conditions. The code rate of the encoder is system with user data preceded and followed
k/k+ I, k = I, 2, 3. The encoder is a by control information as shown in Figure
constraint length 3 convolutional code. To 7. 5 In the discussion that follows, we will
date, we have investigated M -ary DPSK refer to the forward link (base to mobile). The
modulation, but the modem architecture will modem works equally well on the reverse link
accommodate other bandwidth efficient (mobile to base).
modulation techniques such as M -ary
continuous phase frequency shift keying
(CPFSK).

3 Assuming the interference is Gaussian distributed . 5This is not how CDPD works . We are using a
4Patent pending. simplified scheme to aid the explanation.
 235
O&a I COl'llroJ I 0 •• I Cont rol I o.~ error correction. For error detection, the
Viterbi decoder may be used (although this
gets complicated) or we may use a 16-bit
Col'lu&mR,o
CRC. Details such as these are beyond the
scope of this paper. The main point is that we
Figure 7 - Frame Structure Showing
can adapt the rate dynamicallyto accommodate
Variable Rate Transmission a one-to-N packet radio system.

In this example, we have 32 symbols of 5.3 Si~a1 Estimation. An adaptive data rate
control information preceding each 384 system also requires an accurate and timely
symbol block of user data. Within the block estimate of the received signal-to-noise ratio.
of user data, we can adapt the data rate almost In the system under development, the
continuously to the instantaneous channel estimator is built into the Viterbi decoder and it
conditions. Of course, we require a return estimates signal-to-noise, not just signal. An
path (i.e., a full duplex link) to pass rate example of the estimator's operation is shown
information to the base station. in Figure 9 at the end of this paper. At typical
cellular doppler frequencies, the estimator
If the data transfer is one-way, the entire tracks the signal well with a standard deviation
return path can be used for passing rate
of 2.5 dB. Since the Viterbi decoder can
information. In this case, we could
theoretically change the rate every symbol, correct many error patterns, an infrequent bad
but some time delay is necessary to transmit estimate does not necessarily result in a bit
and receive the rate information sent on the error. In Figure 9, we see several cases
return path. A minimum delay of 3 symbols where the estimator overestimated the signal,
is more realistic. but no error occurred.

If the data transmission is two-way, rate When the signal is below the threshold for our
information must be included as overhead in lowest data rate (l bit per symbol),
the user's data stream. One way to do this is information transfer ceases and training
shown in Figure 8. symbols are sent. The estimator continues to
measure the channel and directs the transmitter
I. 3&4 a, fT'lOOlt (32 IOMalotil - -- - .., to resume information transfer as soon as the
I • I • signal rises above threshold.
+ ~
o-.~ (2 bC.) 12bb (1 ~tl
5.4 Memory. For a conventional constant
0... 1'0 bftIIl
data rate system, the delay to receive an N-bit
Figure 8 - Slot Structure Within A Data Block packet is simply T = NIR where R is the
data rate in bits per second. When a constant
Note from Figure 8 that the data block carries data rate system employs a block error
correction code or an interleaver, the delay is
between 32 and 96 slots where each slot
increased by the encoding time or interleaver
comprises 10 data bits and 2 overhead bits.
depth, 0, so that the total delay is T = 0 +
Only 2 overhead bits are required since we N/R. In either case, the delay is a constant,
have only four possible rates. The overhead independent of the channel conditions.
is 17%, which is less than the overhead
required by the Reed-Solomon code. Since Adaptive data rate systems are different.
we will use trellis-coded modulation, the Packet delay is a random variable that is a
Reed-Solomon code would not be required for function of the amplitude and doppler
236
characteristics of the channel. Consequently, random phase rotation that occurs during a
additional memory is needed to provide a symbol period. The extent of the problem
buffer between a constant data rate source and depends on the symbol rate and the doppler
the variable data rate modem. On CDPD, frequency of the channel. We would prefer
overflow of the buffer is not a problem low doppler and high symbol rates to
provided good control is used . CDPD minimize this problem. One advantage of an
already assembles packets in the mobile adaptive data rate system is that it does not
terminal's memory and can feed one packet at transmit information during the deep fades
a time to the adaptive rate modem. when the random phase rotations are greatest.
Consequently, the irreducible error floor for
5.5 Feedback Channel. Another obvious an adaptive data rate system will be lower than
obstacle to adaptive data rate communications for a conventional system.
is the need for a reliable feedback channel. On
mobile radio, the feedback channel will also 6.0 Conclusions
fade and will therefore introduce a time-
varying signal, noise and delay. CDPD is a wireless communications system
and is therefore subject to the time-varying
In the modem, a reliable data rate is estimated signal levels that are inherent to wireless
by the channel estimator and a word, Kj, channels. Actual throughputon CDPD will be
representing the desired data rate, is less than half the rated throughput of 19.2
transmitted on the feedback channel to the kbps because of Rayleigh fading. We can
forward channel transmitter. Propagation improve the throughput of CDPD by a factor
delay adds a 2 symbol delay to the processing of 5 by adapting the data rate to the
time. An additional delay is built in to the instantaneous signal-to-noise ratio at the radio
estimator since it must have a non-zero semple receiver. An adaptive data rate system will
size, n, to make an estimate. also increase the service area of a cellular base
station by over 17%.
The data rate word is subject to fading and
noise and errors will occur in accordance the Pericle Communications Company is
channel statistics. When an error occurs on the developing an adaptive data rate modem for
feedback channel, the transmitter and receiver cellular applications. The modem combines
are operating at different data rates and a loss rate changing, signal estimation , and error
of rate synchronization occurs. When rate correction in a single device. An efficient
synchronization is lost, the bit-error rate is decoding algorithm ensures the modem can be
almost 112, and a packet transmitted during built cost-effectively.
this time will almost certainly require re-
transmission. 7.0 Acknowledgement

All of these feedback channel effects have Dr. Mark A. Wickert developed the Rayleigh
been simulated. Simulations show that the channel simulator used to generate the CDPD
ultimate effect of the feedback channel is a performance measures used in this paper. Dr.
loss in throughput of approximately 15% over Wickert is an associate Professor with the
theoretical. Electrical and Computer Engineering
Department at the University of Colorado at
5.6 Irreducible Error Floor. On fading Colorado Springs.
channels, differential PSK modulation suffers
from an irreducible error rate caused by the
 237
8.0 References [7) P.A. Bello and W. M. Cowan, "Theoretical study
of onloff transmission over Gaussian multiplicative
[1) Cellular Digital Packet Data System Specification, circuits; Proceedings of the IRE Nat ional
Release 1.0. July 19, 1993. Communications Symposium. Utica, NY, Oct 1962.

[2) W.C .Y. Lee, "Data Transmission via Cellular [8) J. K. Cavers, "Variable-rate transmission for
Systems," 43rd IEEE Vehicular Technology Rayleigh fading channels," IEEE Transactions on
Conference, May 18-20, 1993, Secaucus, New Jersey. Communications, vol. COM-20, no. I , pp. 15-22,
Feb. 1972.
[3) V. H. MacDonald, "The cellular concept," The
Bell System Technical Journal, January 1979. [9) S. M. Ross , A Erst Course in Probability, New
York: MacMillan , 1984.
[4) W.C.Y. Lee, Mobile Cellular Telecommunications
~,New York: McGraw-Hill, 1989. [10) S. Lin and DJ. Costello, Error Control Coding
Fundamentals and Applications, Englewood Cliffs ,
[5) N. J. Boucher, The Cellular Radio Handbook 2d NJ: Prentice Hall, 1983.
ful., Mendicino, CA: Quantum, 1992.
[11) TIA IS-55A, "Recommended Min imum
[6) R.E. Blahut, Digital Transmission of Information. PerformanceStandardsof 800 MHz Dual-Mode Mobile
Reading Massachusetts: Addison Wesley, 1990. Stations," September 1993.

~__-_..=_=._,.'• ~:_=_=_=_=_=f-_=_:",:_=__-.:_~.=~(]j===6=dl=B=J== _+=+__ ! L .

--t--t-~+==+==+==~==J==~
_==__!: ,'!. !, .li.
. i
' . j I ! I r 1

-I o 2 3 4 5 6 7 8 9 10 II 12 13 14 15 16 17 18 19 20

/l.u u (R O>- umin in dB

Figure 3 - Probability that ell > Umin

 tv
VJ
00

15 I I

10

5
~
.5
.§ o
~
IU
.!:2
~
-5
Sl
~. ~
Vj - 10

Signal

- 15 "1 51. Dev, of error = 2.5 dB ~ Estimate

Bit errors =0
Throughput = / .4/ 7 bits
.20 Iii' il"" I I' iii'" ii" i f Iii i i i i. i i i i. i 'I i i i ii" iii'" ' I ' i' 'I"" I·" , i i i i i i Ii' i 'I" i i i i i i i i i i i ' Iii iii
o 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850 900 950 1000
Time in symbol periods

Figure 9 - Estimator Performance