ELINT The Interception and Analysis of Radar Signals The Artech

ELINT
The Interception and Analysis of Radar Signals

For a listing of recent titles in the Artech House
Radar Library, turn to the back of this book.
ELINT
The Interception and Analysis of Radar Signals
Richard G. Wiley
Library of Congress Cataloging-in-Publication Data
Wiley, Richard G.
ELINT: the interception and analysis of radar signals / Richard G. Wiley.
p. cm.—(Artech House radar library)
Includes bibliographical references and index.
ISBN 1-58053-925-4 (alk. paper)
1. Radar. I. Title. II. Series.
TK6575.W55 2006
621.3848—dc22 2005058854
British Library Cataloguing in Publication Data

Wiley, Richard G.
ELINT: the interception and analysis of radar signals.—(Artech House radar library)
1. Electronic intelligence 2. Radar 3. Signal detection I. Title
621.7’34
ISBN-10: 1-58053-925-4
Cover design by Igor Valdman
 2006 ARTECH HOUSE, INC.

685 Canton Street
Norwood, MA 02062
All rights reserved. Printed and bound in the United States of America. No part of this
book may be reproduced or utilized in any form or by any means, electronic or mechanical,
including photocopying, recording, or by any information storage and retrieval system,
without permission in writing from the publisher.
All terms mentioned in this book that are known to be trademarks or service marks
have been appropriately capitalized. Artech House cannot attest to the accuracy of this
information. Use of a term in this book should not be regarded as affecting the validity of
any trademark or service mark.
International Standard Book Number: 1-58053-925-4

Library of Congress Catalog Card Number: 2005058854
10 9 8 7 6 5 4 3 2 1
Contents
Preface xiii
CHAPTER 1
Electronic Intelligence 1
1.1 Electronic Intelligence Defined 1
1.2 The Importance of Intercepting and Analyzing Radar Signals 2
1.3 Limitations Due to Noise 4
1.4 Probability of Intercept Problems 4
1.5 Direction Finding (DF) and Emitter Location 5
1.6 Inferring Radar Capabilities from Observed Signal Parameters 5
1.7 Receivers for Radar Interception 6
1.8 Major ELINT Signal Parameters 6
1.9 The Impact of LPI Radar on ELINT 7
References 8
CHAPTER 2
ELINT Implications of Range Equations and Radar Constraints 9
2.1 Range Equations 9
2.2 Radar Constraints 12
2.2.1 Range Resolution Related to Bandwidth 13
2.2.2 Spread Spectrum: Radar Versus Communications 14
2.2.3 Moving Targets and Integration Time Constraints 15
2.2.4 Constraints on Time-Bandwidth Product or Pulse Compression
Ratio 16
2.2.5 Constraints on Doppler Resolution 16
2.2.6 Frequency Agility 19
2.2.7 PRI Agility 19
2.2.8 Power Constraints 23
2.2.9 Pulse Compression Modulation Constraints 23
2.3 Some ELINT Implications of Future Radar Designs 23
2.3.1 Bistatic and Multistatic Radars 24
2.3.2 Radar Trends 24
2.3.3 Wideband Active Adaptive Array Radars 26
2.4 Summary of Radar Design Constraints and Trends 26
2.5 High-Power Microwave Weapons 27
References 28
v
vi Contents
CHAPTER 3
Characteristics of ELINT Interception Systems 29
3.1 Intercept System Characteristics and Functions 29
3.2 Frequency Coverage 30
3.3 Analysis Bandwidth 31
3.3.1 Wideband Radar Signal Trends 33
3.4 Dynamic Range 35
3.4.1 Dynamic Range Requirements 37
3.5 Sensitivity 39
3.5.1 Noise Figure Measurement 40
3.5.2 Y-Factor Measurement 41
3.5.3 Some Sensitivity Measures 43
3.5.4 Output SNR and Receiver Applications 44
3.5.5 Threshold Detection 45
3.5.6 Sensitivity and the Received Pulse Density 51
3.6 The Ultimate Limits to ELINT Parameter Measurements 52
3.7 ECM and ELINT Receivers 56
3.8 Crystal Video Receivers 56
3.8.1 Crystal Video Applications 59
3.8.2 Postdetection Signal Recording and Sorting 60
3.8.3 CV System Design Considerations 60
3.9 Superheterodyne Receivers 63
3.9.1 Superhet Performance 65
3.9.2 Sweeping Superhet Receivers 67
3.9.3 Tuning Considerations 70
3.9.4 Other Heterodyne Receivers 72
3.10 Instantaneous Frequency Measurement Receivers 74
3.10.1 Limiters Applied to IFMs 74
3.10.2 The Simultaneous Signal Problem 75
3.10.3 CW Signals and IFMs 81
3.10.4 Digitizing the IFM Output 82
3.11 Other Receivers 85
3.11.1 Channelized Receivers 85
3.11.2 Acousto-Optic (Bragg Cell) Receivers 89
3.11.3 Microscan Receivers 90
3.12 System Considerations 93
References 94
CHAPTER 4
Probability of Intercept 97
4.1 Background 97
4.2 Developments in the Theory Behind POI 98
4.2.1 Intercept Description 98
4.2.2 Implications of Today’s Environments/Operations on Intercept
Time 98
Contents vii
4.2.3 Mathematical Models 100

4.2.4 Recent Developments on POI 108
4.3 Summary 114
References 114
CHAPTER 5
Antennas and Direction Finders 117
5.1 Omni-Directional Antennas 117
5.1.1 Omni-Directional Antenna Applications 117
5.1.2 Parameters for Omni-Directional Antennas 118
5.2 Directional Intercept Antennas 121
5.3 Direction Finding 124
5.4 Instantaneous Direction Finding 126
5.4.1 Amplitude Comparison AOA Measurement 127
5.4.2 Phase Interferometers 131
5.4.3 Bearing Discriminators 139
5.5 Arrays, Lenses, and Subspace DF Methods 144
5.6 Short Baseline TDOA for AOA 145
References 147
CHAPTER 6
Emitter Location 149
6.1 Introduction 149
6.2 Emitter Location Estimation 150
6.3 Deriving the Location Covariance Matrix 152
6.4 Angle of Arrival Location Analysis 153
6.5 Time Difference of Arrival Location Analysis 155
6.6 Time/Frequency Difference of Arrival Location Analysis 159
6.7 Geometric Dilution of Precision 163
6.8 Incorporation of Measurement Error 164
6.9 Summary 167
References 168
CHAPTER 7
Estimating Power at the Transmitter 169
7.1 Power Estimation Through ELINT 169
7.2 Distance to the Horizon 170
7.3 ERP Errors Due to Antenna Pointing Errors 171
7.4 Estimating the Distance to the Radar 174
7.5 Multiple Signal and Multipath Problems 176
7.6 Summary of Power Measurement Requirements 178
7.7 Sample ERP Calculations 179
References 181
viii Contents
CHAPTER 8
Antenna Parameters 183
8.1 Polarization Defined 183
8.2 Elliptical Polarization 183
8.3 Stokes’ Parameters 185
8.4 Measuring Polarization 187
8.4.1 Polarization Pattern Method 188
8.4.2 Phase-Amplitude Method 189
8.4.3 Multiple Antenna Method 190
8.5 Cross-Polarization 191
8.6 Propagation Effects 194
8.7 System Aspects of Polarization 195
8.8 Antenna Beam Shape 196
8.9 Basic Antenna Pattern Relationships 197
8.10 Beam Patterns from ELINT 199
8.11 Beam Patterns of Array Antennas 205
8.12 Antenna Beam Summary 208
References 209
CHAPTER 9
LPI Radar and the Future of ELINT 211
9.1 What Is LPI Radar? 211
9.2 Radar and ELINT Detection of Signals 212
9.3 Matched Filter Theory 213
9.4 One Interception Strategy: Noncoherent Integration 213
9.5 ESM and Radar Range Compared 214
9.6 Some Pulse Compression Modulation Constraints 217
9.7 Interception Techniques Using the Envelope of the Received Signal 218
9.8 Narrowband Channels and Frequency Modulated Signals 222
9.9 Predetection Processing Methods to Detect Linear FM and Other LPI
Signals 226
9.10 ELINT Receiver Requirements for Interception of Low Peak Power
Signals 232
References 233
CHAPTER 10
Antenna Scan Analysis 235
10.2 Some Principles of Searching 236
10.3 Relationships Among Scan Rate, Maximum Unambiguous Range,
and Energy on Target 237
10.4 Fan Beam Scanning: Circular and Sector 239
10.5 Pencil Beam Scanning: Raster, Helical, and Spiral 245
10.6 Tracking Scans and Monopulse 246
10.7 Electronic Scanning 251
10.8 Scan Measurement and Analysis Techniques 252
Contents ix
10.9 A Three-Dimensional Search Example 253

References 254
CHAPTER 11
Intrapulse Analysis 255
11.2 Pulse Envelope Parameters 257
11.3 Envelope Parameter Measurements 258
11.3.1 Rise and Fall Times 260
11.4 Some Radar Performance Limits Related to Pulse Envelope 264
11.5 Multipath Effects 266
11.6 Intrapulse Frequency and Phase Modulation 268
11.6.1 Choosing the Receiver Bandwidth 271
11.7 Intentionally Modulated Pulses 272
11.8 Incidental Intrapulse Shape—Uses and Causes 275
11.9 Comparing Wave Shapes 278
References 280
CHAPTER 12
Pulse Repetition Interval Analysis 281
12.2 Common PRI Categories 282
12.2.1 Constant PRI 282
12.2.2 Jittered PRIs 284
12.2.3 Dwell and Switch PRI 285
12.2.4 PRI Stagger 285
12.2.5 Sliding PRIs 286
12.2.6 Scheduled PRIs 286
12.2.7 Periodic PRI Variations 287
12.2.8 Pulse Groups 287
12.3 Time Interval Measurements 287
12.3.1 SNR Limitations 288
12.3.2 Limitations Due to Pulse Amplitude Changes 289
12.3.3 Improving Interval Measurements 292
12.3.4 Digital Thresholding 294
12.4 PRI Analysis Techniques 295
12.4.1 Raster Displays 296
12.4.2 PRI Sounds 297
12.5 PRI Analysis Theory and Practice 298
12.5.1 Statistical Techniques 300
12.5.2 Delta-T Histogram 303
12.6 Interpreting the Results 307
12.6.1 Delay Line PRI Generators 308
12.6.2 Crystal Oscillators and Countdown Circuits 308
12.7 PRI and Range Velocity Ambiguities 313
12.8 MTI Radar Blind Speeds 313
x Contents
12.9 Moving Target Detection 314

References 315
CHAPTER 13
Deinterleaving Pulse Trains 317
13.1 Pulse Sorting 317
13.2 PRI-Based Gating 319
13.3 Deinterleaving Algorithms 320
13.4 Delta-T Histogram Applied to Deinterleaving 322
13.5 The Pulse Train Spectrum 331
13.6 Combining Pulse Bursts 332
13.7 Raster Displays and Deinterleaving 333
13.8 Measuring Deinterleaver Performance 333
References 334
CHAPTER 14
Measurement and Analysis of Carrier Frequency 335
14.1 Pulsed Signal Carrier Frequency 335
14.1.1 Frequency Measurement Accuracies 335
14.1.2 Doppler Shifts 337
14.1.3 Drift Measurement 337
14.1.4 FM Ranging in Radar 338
14.2 Intrapulse Frequency or Phase Modulation 338
14.2.1 Analysis of Predetection Data 342
14.3 Coherence (Short-Term RF Stability) 342
14.3.1 RMS Phase Fluctuation 346
14.3.2 RMS Frequency Fluctuations 347
14.3.3 Signal Repeatability 350
14.3.4 Effects of Variations in ␶ 0 352
14.3.5 Frequency-Domain Stability Measures 353
14.3.6 Bandwidth Limitations on Allan Variance Measurements 354
14.3.7 Noise Limitations on Allan Variance Measurements 356
14.3.8 Frequency Stability Measures for Power Law Spectra 357
14.3.9 Sinusoidal FM and Linear Frequency Drift 359
14.3.10 Short-Look Problem 359
14.4 Frequency Character of CW Signals 361
14.5 Pulsed Signal Example 361
14.6 Measuring Coherence 364
14.7 Effects of Drift 365
References 365
CHAPTER 15
Determining ELINT Parameter Limits 367
15.2 Histograms Used to Determine Parameter Limits 367
15.3 Types of Histograms 370
Contents xi
15.4 Two-Sigma Parameter Limits 373

15.5 Histogram Analysis Techniques 375
15.5.1 Parameter Limits Example 375
15.5.2 Intercepts Separated into Accuracy Classes 377
15.5.3 Most Probable Values 378
15.6 Analysis Problems 379
15.6.1 Signal Identification Errors 379
15.6.2 Transforming Parameters and Their Accuracies 379
15.7 Histogram Analysis Summary 381
References 382
CHAPTER 16
ELINT Data Files 383
16.2 Signal Identification 385
16.3 ELINT Data for Radar Warning Receiver Design 386
16.4 ELINT Data for Simulation and Training 388
16.5 Adding Non-ELINT Data 388
16.6 Summary 389
Reference 389
APPENDIX A
Spectrum Widths: 3-dB and First Nulls for Trapezoidal Pulses 391
A.1 Introduction 391
A.2 Rectangular Pulses 391
A.3 Trapezoidal Pulses: Equal Rise and Fall Times 391
A.4 Trapezoidal Pulses: Unequal Rise and Fall Times 393
APPENDIX B
Some ELINT Considerations of FM Signals 395
B.1 Introduction 395
B.2 Effects of Sinusoidal Interference on Phase- and Frequency-
Demodulated Signals 395
B.3 Signals with Sinusoidal Frequency Modulation 398
B.4 Carson’s Rule 400
B.5 Modification of Carson’s Rule for ELINT Applications 401
B.6 FM Demodulation Degradation by RF Band Limiting 403
B.7 Effects of Noise on FM Demodulated Signals 408
References 412
APPENDIX C
A Frequency Hop Radar Example 413
C.1 Probability That One or More Pulses Occur at the Desired Frequency 413
C.2 Probability That Exactly One Pulse Occurs at the Desired Frequency 414
C.3 Probability That Exactly k Pulses Occur at the Desired Frequency 415
xii Contents
C.4 Probability That Several Pulses Occur at the Desired Frequency Less
Than G Pulses Apart 415
C.5 Probability Distribution of the Interval Between Two Pulses 418
C.6 Determining an Optimum Receiver Sweep Rate 427
APPENDIX D
History and Fundamentals of the IFM 429
D.1 The Broadband Microwave Frequency Discriminator 433
References 439
APPENDIX E
Emitter Location Partial Derivatives 441
About the Author 445
Index 447
Preface
More than 20 years have passed since the publication of Electronic Intelligence:
The Analysis of Radar Signals (Artech House, 1983) and Electronic Intelligence:
The Interception of Radar Signals (Artech House, 1985). Much of what those
books contained remains valid today. With this new book, I have again avoided
emphasizing specific hardware and software currently available because it changes
too rapidly. My hope is that this book remains useful for the next 20 years. This
book is partially based on my earlier works, Electronic Intelligence: The Analysis
of Radar Signals (Artech House, 1983), Electronic Intelligence: The Analysis of
Radar Signals, Second Edition (Artech House, 1993), and Electronic Intelligence:
The Interception of Radar Signals (Artech House, 1985).
It is appropriate to list those people who introduced me to the exciting field
of ELINT and who continue to provide insights and inspiration. Professor E. M.
Williams of Carnegie-Mellon University introduced me to the field. Dr. Thomas
F. Curry provided my early training. Both are now gone but often remembered.
There is no way to adequately recognize the contributions of Robert B. Shields,
my friend and business associate for more than 45 years. He always grasps the
‘‘big picture’’ and asks the right questions at the right time.
For their contributions to this book, I am grateful to Arthur Self for his work
on Chapter 4 and to Charles Estrella for his work on Chapter 6.
I am also grateful to the many people who work in ELINT and who have
shared their experiences and insights with me. This includes not only my coworkers
at Research Associates of Syracuse, Inc., but also the several thousand people who
have attended continuing education courses I have presented over the past 20
years—many for the Association of Old Crows. The struggle to answer their
questions is one of the motives in writing this book.
The continued support of my wife, Jane, and the good start in life provided
by my parents Mildred and Asa (1911–1995) were key to the creation of this book.
I am also thankful for the support of my six children, their spouses, and our 13
grandchildren. Many of them are now in their teens and approaching adulthood.
It is my hope and prayer that vigilance through ELINT and other means helps
keep them and all children everywhere in safety and freedom.
‘‘Lord God Almighty, in whose Name the founders of this country won liberty
for themselves and for us, and lit the torch of freedom for nations then unborn:
Grant that we and all the people of this land may have grace to maintain our
liberties in righteousness and peace. . . .’’
—Collect for Independence Day, page 241, The Book of Common Prayer (1979)
According to the use of the Episcopal Church
xiii
CHAPTER 1
Electronic Intelligence
1.1 Electronic Intelligence Defined
Radar has been called the greatest advance in sensing remote objects since the
telescope was invented in 1608 [1]. Electronic intelligence (ELINT) is the result of
observing the signals transmitted by radar systems to obtain information about
their capabilities: it is the remote sensing of remote sensors. Through ELINT, it is
possible to obtain valuable information while remaining remote from the radar
itself. Clearly, ELINT is most useful in hostile situations; otherwise, the information
could be obtained directly from the radar user or designer. The value of ELINT
is that it provides timely information about threatening systems, such as radars
that guide aircraft or missiles to targets. ELINT also provides information about
defensive systems, which is important in maintaining a credible deterrent force to
penetrate those defenses.
ELINT refers to the information gained from the interception of signals of
interest. In addition to radar, other types of signals referred to as sources of ELINT
include beacons and transponders, jammers, missile guidance, some data links,
altimeters, navigation emissions, and identification friend or foe (IFF). In this broad
sense, ELINT can refer to nearly any noncommunications emission. This book is
concerned largely with ELINT derived from radar signals.
Receiving radar signals is often not difficult because the available power density
is quite high. The power transmitted by a radar is proportional to the fourth power
of the range at which it is to detect a target; whereas the power available at an
ELINT receiver is proportional to the reciprocal of the square of the distance from
the radar. An analogous situation is a searchlight being used to spot an aircraft.
An observer on the aircraft can see the searchlight from far off, even if the light
is not pointed directly toward the aircraft. By comparison, an observer on the
ground can see the aircraft only when the searchlight beam is aimed at the aircraft
and when the plane is closer than it was when the observer on the aircraft first
saw the searchlight. This aspect and other radar constraints are explored in detail
in Chapter 2, and the topic of low probability of intercept (LPI) radar is explored
in Chapter 9.
While the ELINT observer may enjoy a large range advantage over the radar,
ELINT signals can still be quite weak, and analyzing them is not a simple matter.
The purpose of the chapters that follow is to describe the interception and analysis
processes for various radar signals.
1
2 Electronic Intelligence
1.2 The Importance of Intercepting and Analyzing Radar Signals
The first large-scale use of radar occurred in World War II. No sooner had radar
been deployed (by both sides) than the effort to counter it spawned electronic
countermeasures (ECM), which included the interception of the radar’s signals. In
today’s terminology, this is called electronic attack (EA). In the early 1940s, ‘‘the
basic countermeasures principle was simple. A ship or aircraft carried a rapidly
tunable receiver which could scan the wavelength bands in which enemy radars
might be operating. If one was detected, its strength and characteristics gave clues
as to the dangers involved and what the next tactic should be. If the operator was
flying over enemy territory, he could switch on a jammer, a powerful transmitter
which would overload the radar receiver rendering it ineffective, or he could dump
bales of aluminum ‘chaff’ from the airplane, or he could dump and jam’’ [2]. A
side effect of the large amounts of chaff dropped was that thousands of cows died
in Germany from eating the foil as they grazed.
The following illustrates the fundamental role of ELINT in preventing disas-
trous technological surprise [3]:
In early 1942, the RAF Coastal Command used L-band radar as an aid for locating
German U-boats recharging batteries on the surface. The overall effectiveness of
the RAF in this task was quite good until the U-boats began using L-band search
receivers. These receivers allowed the submarine to hear transmitted radar signals
at a range greater than that over which the radar echo could effectively be returned.
The U-boat therefore had time to dive before actually being sighted by the searching
aircraft. In turn, general effectiveness of the RAF antisubmarine effort decreased.
The Allies, realizing what had happened, installed new S-band search radars aboard
their aircraft during early 1943. As a result of the effectiveness of new equipment
the intercept rate rose sharply. German submarines sitting on the surface, listening
to L-band search receivers, became vulnerable targets for S-band radar directed
aircraft.
As the U-boat sinkings increased, the Germans tried frantically to determine
the method of detection the Allies were using. Since reports from surviving subma-
rines stated that no radiation had been heard in their L-band search receivers prior
to the attack, it was thought that perhaps an infrared detection device of some
type was being employed. Considerable effort was spent in an attempt to combat
a non-existing infrared threat. U-boat activity was greatly reduced by the time the
German High Command realized that new S-band radar was in use.
This is an example of weapon (L-band radar), a countermeasure (L-band
search receiver), and an improvement (S-band radar) providing a clear margin of
technical supremacy.
There is another point to be considered. To be sure, the use of S-band radar
employing magnetrons and extending the useable frequency by a factor often
provided a definite advantage. However, had the Germans had information as to
what was being used, the time lag until they were able to develop an effective
S-band search receiver would have been greatly reduced. It is obvious that the lack
of information by the opponent side is the basic requirement of the so-called ‘‘secret
weapon.’’
This point is mentioned here because illustrated in this example is one of the
important roles of electronic reconnaissance. Had the Germans been conducting
1.2 The Importance of Intercepting and Analyzing Radar Signals 3
an extensive reconnaissance program at the time, it is probable that they would

have intercepted S-band signals from magnetron oscillators in the development
and testing stages during flights over England. The development of the magnetron
was, of course, the solution to the problem of generating high power for 10
cm radar, and simple crystal receivers for reconnaissance purposes were indeed
available, if the Germans had cared to use them in this application. High sensitivity
is not necessary for intercepting high-power sources.
Of course, one can find more recent examples of the value of radar interception
than World War II. In the Vietnam War, there were heavy losses of aircraft because
of the large-scale use of surface-to-air missiles. New tactics and the use of radar
warning receivers designed with the aid of ELINT data helped reduce U.S. aircraft
losses and allowed the North Vietnamese air defense to be destroyed with much
lower losses on the U.S. side [4]. Similarly, the SA-6 missiles supplied by the Soviet
Union to the Arabs in the October 1973 conflict with Israel proved a major
factor in destroying low flying Israeli aircraft. New tactics and jammers eventually
overcame the advantage achieved by the Arab combatants through technological
surprise [4].
Radar uses have proliferated, particularly for military applications, but also
for civilian air traffic control, harbor surveillance, weather monitoring, and so on.
Prudence requires knowledge of the military capabilities of potential adversaries,
and this means continued use of receivers to listen for radar transmissions and all
of the other aspects of ELINT. For this reason, the accumulation of knowledge of
radar signals has a peacetime role in proportion to the development of modern
weapons, many of which incorporate radar target detection and tracking.
Consider the following remarks of Soviet authors, Rear Admiral Peronmov and
Captain First Rank-Engineer A. Partala from their essay ‘‘A Look at Development of
Means of Electronic Warfare’’ [5]:
The military effectiveness of electronic countermeasures, even with emphasis on

real life experimentation, must be regarded as confirmed only under the condition
that basic characteristics of the electronic weapons of the opponent have been
revealed with sufficient accuracy. But the possibility of dependably uncovering and
obtaining this information is an extremely difficult task, according to American
specialists; ‘‘. . . Experience in battle with application of means of ‘electronic war-
fare’ in Southeast Asia,’’ writes the American press, gave evidence that even the
most modern reconnaissance [intelligence] is not able to secure timely disclosure
of all nomenclatured radio-electronic means, their tactical-technical characteristics,
and special military employment. And, therefore, from the beginning of military
operations various surprises are probably inevitable. For prevention of large mate-
rial losses and tactical operational failures, questions about preparation for ‘elec-
tronic warfare’ cannot really be left without attention.
In [5], these remarks are interpreted to mean: ‘‘If you don’t know the exact details
of the electronic threat and fail to prepare effective countermeasures in advance,
you will be shot down or sunk.’’
ELINT has an important role in maintaining defensive capabilities and pre-
venting surprises—in this very real sense, knowledge is power, and where hostile
radar is concerned, ELINT provides a great deal of knowledge. The path to that
knowledge begins with antennas, receivers, and strategies for search, which are
major topics of this book, and continues through the measurement of signal parame-
ters and the recording and reporting of the interceptions for more detailed analysis.
The data gathered is typically about a potential rival’s defense network, such
as radars, surface-to-air missile (SAM) systems, and aircraft. ELINT can be gathered
from ground stations near the hostile country, from ships near the coast, from
aircraft near the airspace, or from satellites in space. This may lead to incidents
where aircraft or ships stray into hostile waters or airspace. In one particularly
dangerous incident, a Chinese fighter collided with a U.S. reconnaissance aircraft
and forced it to land in China.
ELINT data is valuable in the event of a conflict. Knowledge of the location
and type of SAM and antiaircraft artillery (AAA) systems allows planning to avoid
heavily defended areas, making use of flight profiles which will give the penetrator
the best chance of avoiding hostile fire. It also enables intelligent jamming of an
enemy’s defense network. Knowledge of the whereabouts of ships, command and
control centers, surface-to-air missile systems, and other assets of the enemy permits
them to be attacked if need be. ELINT is important to stealth operations because
stealth aircraft are not totally undetectable and also need to know which areas to
avoid. Of course, ELINT data is an important part of network-centric warfare.
One way it is used is ‘‘traffic analysis’’ of ELINT signals. In ELINT, it is not the
message that is of interest as it is in communications intelligence (COMINT);
rather, it is the type of signal, the location of its transmitter, and the timing of the
transmissions relative to other events which may be taking place.
1.3 Limitations Due to Noise

Although in the usual situation the intercept receiver enjoys a considerable range
advantage over the radar, this does not mean that the ELINT signal analyst has
the luxury of a high signal-to-noise ratio (SNR). The situation is frequently just
the reverse, as the ELINT site attempts to receive signals located at great distances
or beyond the radar horizon—possibly due to national boundaries that cannot be
crossed. It is therefore typical for an ELINT station to be equipped with very low
noise front-end amplifiers. High gain antennas that can be steered may also be
used at the price of decreased probability of interception. It is also typical to locate
sites at high altitudes and to use airborne platforms. The potentially large ratio of
ELINT range–to–radar range for long-range radar systems often means that the
limiting factor in receiving signals from a radar is the reduction in signal strength
due to the curvature of the Earth.
In many cases, the ingenuity of the ELINT signal analyst is severely taxed by
a low SNR. The limitations due to low SNR arise repeatedly in the following
chapters. No matter what clever signal analysis technique is devised, the fundamen-
tal limits due to noise remain.
1.4 Probability of Intercept Problems

The ELINT receiver has an advantage in the R 2 versus R 4 path loss over the
radar. However, the ELINT operator usually does not know exactly how to ensure
1.6 Inferring Radar Capabilities from Observed Signal Parameters 5
reception of the signal from the radar. The direction of arrival and carrier frequency
(RF) of the signal are not known. It is also not known when the radar’s main beam
may be pointed in the direction of the ELINT receiving antenna. As a result, the
ELINT operator must search for the radar’s signal in much the same way the radar
operator searches for targets. The ELINT receiver may be tuned in radio frequency
(RF) across a band repeatedly as the ELINT antenna is pointed in different direc-
tions. Intercepting the radar’s main beam then requires that the ELINT receiver
be tuned to the proper frequency and have its antenna pointed toward the radar
at the same time the radar ‘‘looks’’ in the direction of the ELINT site. This problem
makes interception of the radar’s sidelobes a great convenience if there is a sufficient
signal level to do so. The problems and strategies for searching for radar signals
have greatly influenced intercept receiver design, as described in Chapters 3 and 4.
1.5 Direction Finding (DF) and Emitter Location
Of special importance to operational ELINT (OPELINT) is the location of emitters.

This is usually approached by means of angle-of-arrival (AOA) measurements for
coarse location and also for pulse sorting. More precise location makes use of
precise time of arrival (TOA) and frequency of arrival (FOA) measurements at
multiple platforms. These aspects are discussed in detail in Chapters 5 and 6.
1.6 Inferring Radar Capabilities from Observed Signal Parameters
The basic function of ELINT is to determine the capabilities of the radar, so that
decisions can be made as to what threat it poses. Radar functions are generally to
make measurements of the range, velocity, and direction of targets of interest to
the radar. Consider radar measurement of target range. A simple pulsed radar has
a maximum unambiguous range measurement capability given by the speed of
light multiplied by half of the pulse repetition interval (PRI). Therefore, if the PRI
is measured, the maximum unambiguous range can be determined. This provides
a clue to what detection range the radar achieves. It is to the radar’s advantage to
use the shortest PRI commensurate with the range measurement capabilities neces-
sary to perform the radar’s mission. True assessment of radar maximum range
measurement capability requires determining all of the quantities in the radar range
equation. Some of these are impossible to know through ELINT (such as the
required signal level at the radar receiver). Therefore, intelligent estimates or knowl-
edge from other sources must be used.
Of course, inferring radar capabilities becomes very complicated as the radar
becomes more complex. Nevertheless, it is always true that the radar’s transmitted
waveform determines the ambiguity diagram for the radar. ELINT can probe the
character of the ambiguity function by a close examination of the radar’s trans-
mitted signal. This brief outline of the procedures useful in inferring radar capabili-
ties from ELINT is intended to point out the motivation for the signal analysis
techniques described in Chapters 7, 8, and 10 through 15.
1.7 Receivers for Radar Interception
The subject of intercept receivers is quite broad because there is no single design
or type appropriate for all uses. Generally, intercept receivers can be divided into
warning receivers and ELINT receivers.
Radar warning receivers are designed to give nearly immediate warning if
specific threat signals are received (e.g., illumination of an aircraft’s warning receiver
by the target tracking radar of a surface-to-air missile system). The warning receiver
typically has poor sensitivity and feeds into a near-real-time processor that uses a
few parameter measurements to identify a threat. Usually, rough direction (e.g.,
quadrant or octant) is determined for the threat and the operator has a crude
display showing functional radar type, direction, and relative range (strong signals
displayed as being nearer than weaker ones). This type of receiver does not provide
the kind of output that is analyzed using the methods described here. Rather, the
identification algorithms used in the warning receiver might be based on the results
of ELINT.
The ELINT receiver is usually more sensitive compared to the warning receiver
and often has a variety of filters and demodulation adjuncts that are used to get
the best SNR possible. Modern ELINT receivers and radar warning receivers
(RWRs) are becoming more nearly equal in sensitivity and capability due to the
widespread adoption of digital techniques. The outputs of the receiver may be
analyzed using off-line analysis tools based on software. The on-site analysis results
may be forwarded to centralized analysis centers for further processing.
ELINT receivers must cope with a wide variety of signal parameters. The signals
of interest extend over a wide dynamic range—wider than any receiver design can
hope to handle due to the tremendous range differences possible and the different
radar effective radiated power (ERP) values that may be encountered. The ELINT
receiving equipment must cover a wide frequency range to be able to search for
new radars in any part of the spectrum. Coverage to 100 GHz may be needed.
The modulation bandwidths of the signals can range from very short pulses to
those of frequency modulation (FM) and continuous wave (CW) signals. The
receiving equipment may also be operated remotely with automatic digitizing and/
or retransmission to analysis sites. ELINT receivers may be found in a variety of
locations, including ships, aircraft, balloons, and ground locations. Often this places
them in proximity to local radar sets or jammers. In this case, it is generally
necessary to provide for blanking of the ELINT receiver during the transmission
cycle of these local emitters. Receivers are discussed in Chapter 3.
1.8 Major ELINT Signal Parameters
The typical radar signal can be analyzed according to its power (Chapter 7), antenna
(Chapters 8 and 10), and carrier modulation characteristics (Chapter 14). There
are many receiver characteristics of interest as well, but these cannot be obtained
directly from ELINT. The power characteristics include any variations in the power.
The antenna characteristics include the observed polarization beam patterns and
the angular motion (scanning) of the beam. The modulation characteristics can be
1.9 The Impact of LPI Radar on ELINT 7
divided into the more usual pulsed types and CW types. Pulse modulation includes
both the envelope and intrapulse frequency or phase modulation (Chapter 11),
and pulse interval characteristics (Chapter 12). The radio frequency, or carrier,
characteristics are of importance, including the carrier frequency used, as well as
any intentional variations such as frequency agility. Also of interest are measures
of incidental variations in the frequency (or phase), which are measures of RF
coherence.
When measuring any of these parameters, two problems must be addressed:
separating the signal of interest from the rest of those signals in the environment,
and statistically determining parameter ranges for radar sets of a given type. Brief
discussions of these two subjects are included in Chapters 13 and 15. Finally,
Chapter 16 briefly describes some uses of ELINT data files including their use in
radar warning receivers.
1.9 The Impact of LPI Radar on ELINT
Since ELINT has such value, a natural direction for radar design would be to
reduce the ability of a would-be listener to receive the radar signal. Such radar
systems are said to have low probability of intercept features. These may include low
sidelobe antennas, infrequent scanning, reducing the radar power when tracking a
closing target (as range is reduced, the radar power is also reduced), making use
of waveform coding to provide transmitting duty cycles approaching one (to reduce
peak power while maintaining the required average power), and using frequency
hopping to force the interceptor to consider more of the spectrum in attempting
to characterize the radar.
The problem of LPI radar design is fundamentally quite difficult. Search via
radar is analogous to using a searchlight to spot an airplane at night. Energy is
sent out, reflected by the target and used by the observer to spot the airplane. An
LPI-type searchlight design still would require that the airplane be spotted, but
also that the searchlight itself could not be seen by an observer. Because of the
fundamental nature of this problem, it requires careful analysis. LPI radar is consid-
ered in some detail in Chapter 9.
The problems inherent in LPI radar design are such that electronically inter-
cepted signal information will continue to be available; however, new approaches
for interception of LPI-type radars will be needed.
The design and use of intercept receivers is heavily influenced by the design of
radar systems: today’s intercept receivers reflect the radar signals in use today;
tomorrow’s intercept receivers will reflect tomorrow’s radar signals. From the
earliest days of radar until recent times, radar designers have not been directly
concerned with the activities of would-be interceptors. Their aim has been to
produce well-designed radars capable of performing specified functions without
regard to the activities of ELINT operations. Of course, radar designers and opera-
tors have always been concerned with chaff and jamming or electronic attack, and
have taken steps to reduce their effects through electronic counter-countermeasures
(ECCM) or electronic protection (EP) [6].
If the LPI philosophy of radar design influences the signal emissions of tomor-
row’s radar systems, the designers of intercept receivers must respond (as best they
can) to be successful, where possible, in spite of the LPI design approaches which
may be adopted. (The LPI radar designer makes a strategic mistake by assuming
that intercept receiver designs will not change in response to introducing LPI radar
into the signal environment.)
Another influence on radar design is efforts to reduce the radar cross section
of targets. This makes LPI design more difficult because the lower the target’s radar
cross section, the more average power required by the radar to detect the target.
Radar systems are evolving in the direction of multiple functions and multiple
modes. Future radars may well have both LPI and low target cross section modes
among many others in their repertoire.
References
[1] Barton, D. K., Radar Systems Analysis, Dedham, MA: Artech House, 1976, p. 1.
[2] Kraus, J., The Big Ear, Powell, OH: Cygnus-Quasar Books, 1976, Chapter 12, p. 18.
[3] Schlesinger, R. J., Principles of Electronic Warfare, Englewood Cliffs, NJ: Prentice Hall,
1961.
[4] Johnston, S. L., Radar Electronic Counter-Countermeasures, Dedham, MA: Artech House,
1979, Section 1.3.
[5] Stone, N. L., ‘‘Soviet Perceptions: The Soviets on Electronic Countermeasures for Surface
Ships,’’ Defense Science and Electronics, March 1983, p. 59.
[6] Maksimov, M. V., et al., Radar Anti-Jamming Techniques, Dedham, MA: Artech House,
1979.
CHAPTER 2
ELINT Implications of Range Equations

and Radar Constraints
The effects of the one-way range equation of ELINT and the two-way range
equation of radar on signal strength must be understood and explored in order to
appreciate the typical situations encountered in ELINT and electronic warfare
(EW). Similarly, the constraints placed on radar waveforms must be understood
in order to correctly interpret the functions and applications of the signals trans-
mitted by radar and also to be aware of the signal characteristics expected to be
encountered by ELINT. In many ways, understanding these aspects of ELINT is
what separates one who only observes signals from one who both observes and
analyzes signals.
2.1 Range Equations
In free space, the signal received from a target by the radar receiver can be expressed
as [1]
P T G T G R ␭ 2␴
SR = 3 4
(2.1)
(4␲ ) R R L T L R
where:
S R = signal power at radar receiver (W)

PT = transmitter power (W)
G T = gain of radar transmit antenna in the direction of the target
GR = gain of radar receive antenna in the direction of the target
␭ = wavelength (m)
␴ = target radar cross section (square meters)
R R = radar range to target (m)
L T = radar losses from transmitter to antenna
L R = radar losses from antenna to receiver
9
10 ELINT Implications of Range Equations and Radar Constraints
The received signal is inversely proportional to the range to the target to the
fourth power because the signal undergoes spreading on the way to the target and
again on the way back from the target. The radar equation is often referred to as
the two-way range equation. In free space, the signal level received by the ELINT
receiver from the radar transmitter is given by [2]:
PT G TE GE ␭ 2
SE = 2 2
(2.2)
(4␲ ) R E L T L E
where:
S E = signal level at ELINT receiver (W)

G TE = gain of radar transmit antenna in the direction of the ELINT receiver
GE = gain of ELINT antenna in the direction of the radar transmit antenna
R E = range from radar transmit antenna to ELINT antenna
The other symbols are defined under (2.1).

The signal undergoes spreading from the radar transmitter to the ELINT
receiver, and so it is inversely proportional to the second power of the range. This
equation is often referred to as the one-way range equation. Properly speaking,
these are not range equations but rather received signal level equations. When
solved for the range in terms of the signal strength received, they become expressions
for the range.
The qualification ‘‘in free space’’ points out that the effects of the atmosphere
and the Earth’s horizon have not been included. Note that the one-way range
equation may also be used to compute the signal level from a jammer at a victim
receiver.
A significant aspect of these range equations is that the power level transmitted
by pulsed radar transmitters in order to detect targets at long range is very high.
This allows ELINT receivers to detect radar signal at very long ranges even when
observing the sidelobes of the radar’s transmit antenna.
To simplify the discussion, suppose that the ELINT receiver requires a signal
level that is a factor ␦ times the signal level needed by the radar receiver; that is,
S E = ␦ (S R ) (2.3)
Combining (1.1), (1.2), and (1.3) and solving for the ratio of the ELINT range
to the radar range give
RE
RR
= RR 冋
4␲ 1 G TE GE L R 1/2
␦ ␴ G T GR L E 册 (2.4)
To get an idea of the typical situation, it is useful to assume some values for
the quantities in (2.4). Suppose the radar target cross section is ␴ = 1 m2 and that
2.1 Range Equations 11
the losses in the radar and ELINT receive paths are equal. Suppose that the radar
uses the same antenna for transmit and receive so that G T = GR and that the
ELINT receiver is in the sidelobes of the radar’s transmit antenna where G TE = 1.
Also, to eliminate the need to search for ELINT signals at various angles, assume
that the ELINT antenna is omni-directional and that GE = 1. Equation (2.4)
becomes
RE RR
=
R R GR 冋册
4␲ 1/2
␦
(2.5)
In (2.5), R R is in meters because the radar target cross section of 1 m2 is

included inside the square root.
The factor 1/␦ is the sensitivity advantage of the radar receiver over the ELINT
receiver. Generally speaking, the ELINT receiver requires a higher level of signal
power because it is not matched to the radar signal, whereas the radar receiver is
nearly matched to the signal transmitted. There are several reasons for this state
of affairs. The ELINT receiver’s function is to detect the presence of a radar signal
and also to measure its parameters such as amplitude, carrier frequency, angle of
arrival, time of arrival, pulse duration, and so forth. The function of the radar
receiver is to detect the presence of a target echo and to determine its range, velocity,
and angle. The radar receiver often makes use of multiple echoes to detect the
presence of a target; today’s ELINT receivers, however, often depend on detecting
each individual radar pulse and then associate these individual pulses into pulse
trains after detection. The result is that the factor ␦ is generally larger than 1,
meaning that the ELINT receiver has poorer sensitivity (requires a stronger signal)
than the radar receiver. One major difference is that the ELINT receiver may not
know the carrier frequency of the signal and must search over broad bands to find
signals of interest (SOI). To minimize search time, the ELNT receiver may choose
to use a very wide bandwidth. This degrades its sensitivity relative to the radar
receiver—except for small Doppler shifts it can be tuned to the exact carrier
frequency. For example, a radar receiver may have a bandwidth of 1 MHz to
receive echoes from its 1-␮ s transmitted pulses. The ELINT receiver may have a
bandwidth of 10 MHz to preserve the details of the pulse shape and ease the
problem of tuning to the center of the transmitted spectrum. Furthermore, the
radar may use the echoes from 10 pulses prior to deciding that a target is present
while the ELINT receiver is making a ‘‘pulse present’’ decision on every pulse. In
this case, the ELINT receiver sensitivity is approximately 100 times (or 20 dB)
poorer than the radar receiver.
Sometimes the ELINT receiver is in the direction of the main beam of the radar
antenna. (This is the situation for a radar warning receiver carried by targets of
interest to the radar.) In this scenario, the expression for the ratio of the ELINT
range to radar range becomes
RE
RR
= RR 冋册
4␲ 1/2
GR ␦
(2.6)
Figure 2.1 shows (2.5) and (2.6) for a typical radar antenna gain of 30 dB and
a value of ␦ of 20 dB. Suppose that the ELINT receiver is in a region where the
sidelobes of the radar transmit antenna are G TE = 1. As can be seen from point
A in Figure 2.1, these sidelobes of a radar able to detect a 1-m2 target at 100 km
have a ratio of ELINT range to radar range of about 35.4. In free space, those
sidelobes of the radar are detectable at 3,540 km even when the ELINT receiver
is 20 dB less sensitive than the radar receiver. If the ELINT receiver is in the main
beam of the radar, the main beam could be detected in free space at a range in
excess of 100,000 km.
This example illustrates the situation prevalent in ELINT much of the time:
namely that signals from long-range radars are quite large and these signals can
be received at long range, often limited only by the line of sight. The range equations
provide insight into why ELINT is of interest in a number of aspects of electronic
warfare. Low values of the range ratio are the goal of designers of LPI radars.
Some aspects of LPI radar are explored in Chapter 9.
2.2 Radar Constraints
ELINT signals of interest include radar signals of all types. Sometimes, people
concerned about ELINT attribute properties to radar signals that are contrary to
the constraints under which radar systems must function. Avoiding this pitfall is
an important aspect of ELINT work. Understanding the fundamental limitations
faced by radar designers and the associated ELINT implications is important.
Consider this statement: Radars of the future could transmit noise waveforms over
Figure 2.1 ELINT to radar range ratio.

gigahertz bandwidths and be undetectable by ELINT receivers. Should ELINT

equipment be developed to intercept and process this kind of signal? Probably
not—because signals like this would not be useful for tracking or search radars in
military applications. Assume that the technology for transmitting noise waveforms
over wide (> 1 GHz) bands and the digital signal processing (DSP) hardware to
store and correlate the transmitted waveform with the echoes received is available.
Will such radars be deployed? The answer depends on the job the radar is designed
to do and the characteristics of the target and target motion it must detect or track.
To be ‘‘undetectable’’ by ELINT means that the power must be low. After all, a
high power noise-like waveform would be detectable like the signal from a jammer.
Low peak power means the radar must use a long integration time in order to get
sufficient energy back from the target. Long integration time may restrict the radar’s
ability to keep up with target motion (the target could move out of a range or
Doppler cell if the integration time is too long). Transmitting a wide instantaneous
bandwidth means that the radar has fine range resolution. If the range cells are
smaller than the range extent of the target, the target echo will be spread across
many range cells (a bandwidth of 1 GHz provides range cells every 15 cm). This
reduces the energy received in a single range cell. For example, if one wanted to
recover all the energy from a target 15m long, it would be necessary to coherently
add the energy of 100 range cells. Therefore, the ability to transmit and receive a
particular waveform does not by itself make it a good choice for a radar designer.
ELINT work must be done with these factors in mind.
2.2.1 Range Resolution Related to Bandwidth

Range resolution in radar is inversely proportional to the bandwidth of the signal
(assuming that it is processed coherently) [3, 4]. The fundamental relationship is
c
⌬R = (2.7)
2B
Here c is the speed of light, and B is the bandwidth of the signal during the
coherent processing interval—also called its instantaneous bandwidth.
For example, to distinguish between two fighters in tight formation 30m apart
in range, bandwidth must be about 5 MHz. Figure 2.2 shows the relationship of
range resolution to bandwidth.
If one postulates a value of B = 1 GHz, the radar has a range resolution of
15 cm. This means that the target echoes are resolvable in 15-cm range increments
called range cells. The echoes from a 75-m target are spread across 500 range cells.
This spreading of the echoes across a multiplicity of range cells reduces the
apparent radar cross section (and thus reduces the SNR available) in a single range
cell. For this reason, radar designs generally have range resolution appropriate for
their function. This leads to choosing coherent bandwidths of 10 MHz or less.
(10 MHz corresponds to range resolution of 15m.) In this sense, there is no such
thing as a ‘‘spread spectrum’’ radar—what is transmitted is also received, and the
resulting range resolution is determined by the bandwidth. What this means for
Figure 2.2 Range resolution related to radar coherent bandwidth.
ELINT is that the coherent bandwidth of radar signals is likely to remain the same
as it is now provided the radar performs the same task. (See Table 2.1.)
Tactical radars (early warning, fire control, target acquisition, target track,
airborne intercept) have bandwidth requirements typical of applications 1 and 2
in Table 2.1, that is, under 10 MHz. Some tactical radars and multifunction radars
may have modes for imaging (application 3 in Table 2.1).
2.2.2 Spread Spectrum: Radar Versus Communications

In radar, synchronizing the received signal with the transmitted signal is used to
determine the range. In communication, after synchronization is achieved, the
spreading code can be removed and the narrowband message data may be recovered.
This is the reason for the term ‘‘spread spectrum.’’ The transmitted spectrum is
wider than that required by the message. In radar, use of wider coherent bandwidth
than required for its function requires reassembling the target echoes across many
range cells into a single value. It is conceivable that this could be done by future
radar systems; however, such a direction is not evident in current radar signals.
But that is the sort of radar that would qualify as spread spectrum in the sense
Table 2.1 Instantaneous Bandwidth Required for Selected Applications

Range Resolution Required Resolution (m) Bandwidth (MHz)
1. Count A/C in attack formation 30 5
60 2.5
2. Detect missile separation at launch 15 10
3. Imaging of ships, vehicles, and aircraft 0.5–1 150–300
4. High-resolution mapping 0.15 1,000
that a wider bandwidth is transmitted than is used to detect the target. For ELINT
today, the radar’s coherent bandwidth is determined by the radar’s function—
bandwidth is not widened or narrowed without altering the radar’s range resolution.
2.2.3 Moving Targets and Integration Time Constraints

If a radar is to detect targets moving in a radial direction (toward or away from
the radar), the amount of time the target will be present in a given range cell is
determined by the target velocity and the range resolution. This limits the coherent
integration time of present day radars to
␦ R ⌬R
TCV = < (2.8)
v v
Here TCV is the maximum coherent integration time for a constant velocity
target with radial velocity v, and ␦ R is the change in range during that time. If
the target is accelerating in the radial direction, the maximum integration time is
now a quadratic function of both velocity and acceleration
v − [v 2 + 2a(␦ R)]0.5 v − [v 2 + 2a(⌬R)]0.5

TACC = < (2.9)
−a −a
In this expression, TACC is the coherent integration time available for an acceler-
ating target of velocity v and acceleration a. Figure 2.3 shows range versus time
for a constant velocity (dotted curve) and for a target that is accelerating at
2 g’s (19.507 m/sec/sec) (solid curve). For integration times on the order of 10 to
100 ms and acceleration of a few g’s the effect of acceleration on integration time
is small. For example, for 2-g acceleration and 300-m/s velocity, the constant
Figure 2.3 Range versus time.

velocity approximation yields 0.5s integration time for 1-MHz radar bandwidth
(range resolution of 150m), while the calculation with acceleration included yields
an integration time of 0.492s.
What this means for ELINT is that the radar’s coherent integration time is
limited by the range resolution chosen and the characteristics of the target’s motion.
During the coherent integration time, the radar waveform bandwidth determines
the range resolution of the radar. Furthermore, the center frequency of the carrier
must constant over then integration time if Doppler processing is used.
2.2.4 Constraints on Time-Bandwidth Product or Pulse Compression Ratio

Because range resolution is determined by bandwidth and integration time is deter-
mined by velocity, there is a natural limit on the product of the instantaneous
bandwidth and the duration of the coherent processing interval or pulse width.
This is called the time-bandwidth product. The radar’s pulse compression ratio is
limited to no more than its time bandwidth product. By combining (2.7) and (2.8),
it is easy to see that this is limited to
BT <
BV
a 冉√ 1+
ac
2
BV
−1冊 ––→
a →0
c
2v
(2.10)
This limit is usually quite large—if there is no acceleration and v is 300 m/s,
then BT < 500,000. Most pulse compression systems have a much smaller value
of BT—typically between 10 and 1,000. Pulse compression is used to increase
average power (or pulse duration) while maintaining the required range resolution.
The pulse compression ratio is the ratio of the width of the transmitted pulse
to that of the ‘‘compressed’’ pulse at the radar receiver output. For example,
BT = 100 allows 1-␮ s pulse duration at the receiver output when the transmitted
pulse duration is 100 ␮ s. Some radar systems with pulse compression in excess of
100,000 have been built, but these are long range systems used to observe ICBM
flights or for discrimination of warheads from decoys in an ICBM attack scenario.
The limit on BT is affected by target acceleration. Equation (2.8) assumes constant
velocity. If the target accelerates toward the radar, it will leave the range cell sooner
and this affects the maximum BT allowed. Figure 2.4 shows the effect of target
acceleration on the BT limit when the acceleration is in the same direction as the
velocity. If the target was approaching the radar but slowing down, the limit on
BT would increase instead of decrease.
The higher the radial velocity and acceleration of the target, the more con-
strained the time-bandwidth product. The time-bandwidth product of deployed
radar systems is determined by the nature of the radar’s targets, not by the state
of the art in signal generation and processing alone.
2.2.5 Constraints on Doppler Resolution

If the radar coherently integrates the echoes in one range cell for the entire integra-
tion time, the minimum Doppler filter bandwidth, B f , is approximately the recipro-
Figure 2.4 Limit on time-bandwidth product.
cal of the integration time T, which is either TCV for constant velocity targets or
TACC for accelerating targets:
1
Bf ≈ (2.11)
T
However, if the target is accelerating, the Doppler shift changes. Clearly there
is a relationship between acceleration and the time the Doppler shift of the moving
target remains within the Doppler filter bandwidth.
2aTfo 2aT
⌬ facc = = < Bf (2.12)
c ␭
Here ⌬ facc is the spread of Doppler frequencies caused by the target acceleration
a during the integration time T at carrier frequency fo . Because the coherent
integration time is approximately equal to 1/B f , substituting B f = 1/T into (2.12)
gives the maximum allowable coherent integration time and the minimum Doppler
filter bandwidth as
√ √
␭ 2a
T< , Bf >
2a ␭
To avoid large straddling losses, practical radar designs allow a considerable

margin—for example, by moving the factor of 2 from inside the square root to
outside of the square root.
Clearly, the dispersion of Doppler frequencies due to acceleration must be no

greater than the bandwidth of the Doppler filter during any one coherent integration
interval T. Figure 2.5 is a graph of (2.11) and (2.12) to show this relationship for
target velocity = 300 m/s, RF = 10 GHz. The spread of Doppler frequencies during
the time the target is in one range cell is shown in Figure 2.5 for accelerations of
1, 2, 5, and 10 g’s. The Doppler filter bandwidth must be no wider than the spread
of Doppler frequencies expected. Figure 2.5 also shows the maximum radar signal
bandwidth. For the case where acceleration has a minimal effect on the integration
time, the maximum acceleration of the target can be expressed in terms of the
radar signal’s bandwidth as
v2
a max = 2B 2 (2.13)
c (RF)
The effect of long integration times is to require small target acceleration values.
This means that the radar designer must choose a signal bandwidth that suits the
range resolution requirements of the radar function and that the integration time
must be limited to values that suit the target motion expected. Long integration
time prior to detection implies either slow targets with little acceleration or else
very poor range resolution. High acceleration targets require wider signal band-
widths. For example, an aircraft target approaching at 300 m/s and maneuvering
at 3 g’s needs a radar signal bandwidth of at least 2.5 MHz at 10 GHz. Arbitrarily
long integration times cannot be used by practical search radars to detect targets
because integration time is limited by expected target radial velocity and accelera-
tion along with the RF used. It means that radar signals exhibit relatively constant
characteristics during their coherent integration times—an important consideration
for ELINT. Tracking radars can extend the coherent integration time when target
velocity and acceleration are approximately known. Correlating with all possible
Figure 2.5 Doppler spread and maximum signal bandwidth.

target velocities and accelerations by search radars requires huge processor

throughput and is generally not practical today.
2.2.6 Frequency Agility

From one coherent processing interval to the next, the radar can change its carrier
frequency without changing its range resolution properties. This type of wideband
signal is called frequency agility. The agility band is limited by the radar designer’s
ability to obtain sufficient transmit power and to maintain the antenna beam width
and pointing angle over that band. Typical designs have an agility band of about
10% of the center frequency. (For example, a 1-GHz agility band centered at
10 GHz.) What this means for ELINT is that narrowband receivers have a low
probability of intercepting the complete radar transmission. If it is sufficient to
intercept only portions of the radar transmission, narrowband receivers can be
slowly tuned across the radar band and the entire agility band can still be determined
if the signals is present for enough time. If not, then a wideband ELINT receiver
is needed. These may be of the channelized type with sufficient channels to cover
500 to 1,000 MHz at once. Frequency agility (FA) is used primarily to make the
radar signal more difficult to recognize and to jam, but it also provides a way to
reduce fluctuations of target echoes by, in effect, averaging the observed radar
target cross section over several different frequencies. Depending on ELINT system
architecture, it may be useful to have sufficient instantaneous bandwidth in the
ELINT receiver to cover the entire frequency agility band at one setting of the
receiver. The FA bandwidth does not affect the radar’s range resolution. It does
provide enhanced target detection because the target’s cross section is examined
at several frequencies. Note also that at present the center frequency is not changed
during the coherent processing interval or over the time that Doppler processing
is performed. That is, if more than one pulse is used for Doppler processing, which
is normally the case, the center frequency or RF carrier frequency cannot be changed
for the length of the processing interval. And the processing interval determines
the Doppler resolution. When FA is used with Doppler processing, the frequency
is changed on a pulse-burst to pulse-burst basis, not a pulse-to-pulse basis.
2.2.7 PRI Agility

Modern multifunction radar systems make use of multiple pulse repetition intervals
values during one look at the target. It is a requirement of today’s pulse Doppler
radars that the PRI remain constant during each coherent processing interval. For
moving target indicating (MTI) radar designs, there is usually a sequence of PRI
values that must be completed during one processing interval. This repeated
sequence is known as stagger, and ELINT analysts call the period of the stagger
the stable sum. This is because when consecutive PRIs are added, the sum is constant
when one adds together the PRIs that make up the stagger period—regardless of
which PRI is selected as the starting point for the sum.
MTI radars operate by subtracting (in amplitude and phase) the echoes from
one PRI from those in the next PRI. Stationary targets have the same phase and
amplitude and thus cancel. Echoes from moving targets generally do not have then
same amplitude and phase and so do not cancel. However if the target moves an
integer multiple of half wavelengths in one PRI, the phase of the second echo is
shifted by a multiple of 360° from the first and the echoes cancel. These speeds
are called blind speeds. Changing the PRI changes the blind speeds and the PRI
sequence is selected so that the target can be detected regardless of its speed
moving target detection (MTD) radar systems, used at present in air surveillance
applications. A Doppler filter bank is used to divides the frequency region between
the pulse repetition frequency (PRF) lines into several filter bands (for example,
8). This improves the ability to detect moving targets in clutter. In the original
MTD design, the MTI cancellation process was used ahead of the Doppler filter
bank due to limited dynamic range of available analog-to-digital (A/D) converters.
In modern designs, an MTI canceller need not be included. Of course, there is still
a blind speed problem, but now the PRI must remain constant for the number of
pulses required to excite the filter bank (and canceller if any). This leads to a series
of repeated constant PRIs (say, 10 pulses at one PRI and then 10 pulses at another
PRI, and so forth). Multiple PRIs are required to trade between range and velocity
ambiguities and also to make visible target ranges and velocities eclipsed by trans-
mitted pulses (in time) or spectral lines (in frequency).
For constant PRI and RF, the maximum unambiguous range (R u ) and the
maximum unambiguous velocity (Vu ) are given by
c (PRI)
Ru = (2.14)
2
c
Vu = (2.15)
2(RF)(PRI)
Examples at 10 GHz:
PRI 1,000 ␮ s, Vu = 15 m/s and R u = 150 km

PRI 100 ␮ s, Vu = 150 m/s and R u = 15 km
PRI 10 ␮ s, Vu = 1,500 m/s and R u = 1.5 km
As can be seen, the product of unambiguous range and velocity is a constant.

This means that the total ambiguity is fixed but changes in PRI can increase the
unambiguous range but decrease the unambiguous velocity and vice versa.
c2
R uVu = (2.16)
4(RF)
The product R uVu can be made as large as desired by using a low operating
frequency; however, low frequencies may make the antenna unwieldy. This relation-
ship is shown in Figure 2.6 for common radar frequencies.
Another reason for PRI agility is that if the range to the target is any integer
multiple of unambiguous range, then the echo returns when a pulse is being trans-
mitted and it cannot be received (it is said to be eclipsed). Similarly, if the Doppler
Figure 2.6 Range/velocity related.
shift is any multiple of the PRF (= 1/PRI), the target echo is obscured by the
transmitted spectral lines and returns from stationary clutter and it cannot be
received (it is said to be at a blind speed). Radar designers try to choose PRI
sequences that allow the radar to detect targets at all ranges and velocities of
interest. Another reason for PRI variation is for protection against electronic attack.
Random variations of PRI (jitter) can be used to prevent jamming through
anticipating when a pulse will arrive from the radar. This prevents the generation
of false targets at ranges closer than the target. False targets can be generated at
ranges beyond the real target range by transmitting a false pulse soon after receiving
a pulse from the radar. For this type of jamming, random PRI variations do not
matter.
What this means for ELINT is that the PRI sequences observed in pulsed radar
systems are highly constrained by the radar functions. Furthermore, the entire
sequence of PRIs must be completed during times the antenna beam is illuminating
the target. The sequence might be spread over several radar scans. But note that
the total time available to detect the target is limited. If the radar performance
requirement is to detect the target in one scan, then the entire PRI sequence must
be completed during a single beam dwell. Radar designers use a variety of PRIs
for different functions. This leads to the multiplicity of modes observed by ELINT.
Radars are said to operate in one of three PRF regimes: low PRF, high PRF,
and medium PRF. Low PRF means that target echoes at all ranges of interest return
before the next pulse is transmitted. Range measurements are then assumed to be
unambiguous. This is the usual case for long-range search radars and all types of
simple ranging radars. The actual value of the PRF depends on the range at which
the radar is to detect targets. Therefore, there is no particular numerical value for
a low PRF. For typical low PRF designs the PRF is in the hundreds of hertz. For
this type of radar, velocities of interest usually produce Doppler shifts many times
the PRF and are thus highly ambiguous. Of course, if the radar operating frequency
is low enough, Doppler shifts of interest may be smaller than the PRF to provide
both sufficiently large unambiguous velocity and range.
High PRF means that the target echoes for all velocities of interest have Doppler
shifts less than the PRF. Velocity measurements are thus always unambiguous. The
values used as PRFs are determined by the RF and by the velocities of interest. For
typical designs, the PRF may be in the hundreds of kilohertz. This is the usual case
for airborne radars that depend on Doppler filtering to distinguish targets from
clutter. For most ranges of interest, the echoes are delayed by many times the PRI
and therefore the range measurements are highly ambiguous.
Medium PRF means that several PRls elapse before the target echoes return
and that the Doppler shifts for most velocities are more than several times the
PRF. As a result, both the range and velocity measurements are ambiguous.
One radar may have all three of these modes of operation. If the PRF is
high, the range measurements are not only ambiguous, but the echoes have some
possibility of being eclipsed by the transmitted pulses. The fraction of the ranges
eclipsed is given by the ratio of the pulse duration to the PRI. Even if there is no
interest in the range itself, the radar designer must make sure that targets are not
missed by being eclipsed. For this reason, several PRls are used so that no matter
what the range, for at least some of the PRls the target will be detectable. An
analogous statement can be made concerning blind velocities. There is some band-
width (B e ) centered about multiples of the PRF in which the transmitted energy
(or ground clutter) obscures the Doppler-shifted target echoes. The fraction of the
velocities obscured is the ratio of this bandwidth to the PRF of the radar. Again,
the radar makes us of several PRFs. This is necessary to ensure that the Doppler-
shifted echoes are visible for at least some of the PRFs. To detect a target it must
be visible in both range and velocity. The fraction of the targets visible at an
arbitrary range and velocity is given by the product of those not eclipsed in velocity
and those not eclipsed in range:
F = (1 − PD/PRI)(1 − B e /PRF)
Substituting PRF = 1/PRI, differentiating F with respect to the PRI, setting the
result equal to zero, and solving for the optimum PRI gives
PRI 0 = (PD/B e )1/2
A typical value for B e at X-band is 3 kHz, and a typical pulse duration is

3 ␮ s. The optimum PRI in this case is 31.6 ␮ s. (In this case, the optimum PRF is
31.6 kHz.) This optimum PRI could be estimated through ELINT analysis. The
pulse duration is a standard ELINT parameter and the bandwidth B could be
estimated through high-resolution spectral analysis. This is a typical PRF for a
medium PRF radar. While the PRI must be switched among several values to
detect targets and resolve ambiguities, the PRIs might be near this value. A typical
multimode radar might use a set of eight PRIs during one look at a target. The
design could be such that the target can be detected on at least three of the eight
PRIs regardless of the range and velocity of the target.
2.2.8 Power Constraints
The maximum radar detection range depends on the total energy returned from
the target during the integration time. Energy is the product of the average power
and the integration time. Average power is the product of the peak power times
the duty factor. The radar duty factor is the ratio of the pulse duration to the PRI.
However, unless pulse compression is used, radar range resolution is the pulse
duration times the speed of light divided by 2. This means that narrow pulses
provide fine range resolution whether you want it or not.
In addition, narrow pulses reduce the duty factor when the PRI is determined
according to the unambiguous range and velocity requirements.
The use of short duration pulses makes it difficult to get sufficient energy back
from the target. Pulse compression is used to increase the average power while
retaining the maximum range capability of long pulses and the range resolution
of short pulses.
As illustrated later, a CW signal can have much lower peak power than a
pulsed signal but have the same energy.
2.2.9 Pulse Compression Modulation Constraints
The use of energy detection receivers is necessary if the radar waveform is noise-
like to the interceptor. While such waveforms can be used, their time sidelobes
and Doppler properties may be far from ideal in any specific time interval; it is
only over a long-term average that their ambiguity function approaches the ideal
‘‘thumbtack’’ shape. Up to now, nearly all radar designers prefer to control their
ambiguity function by using specially designed, deterministic pulse compression
waveforms such as linear FM (LFM) or binary phase shift keyed (BPSK). Polyphase
shift keyed and nonlinear FM waveforms are also common in the radar literature
but have not been deployed extensively. For example, the ‘‘best’’ BPSK waveforms
are the Barker codes. These have time sidelobes at zero Doppler of maximum
amplitude of 1. However the longest known binary Barker code is only 13 bits
long. It has been shown that even length Barker codes with length greater than 13
do not exist; however, there are longer sequences of polyphase Barker codes.
2.3 Some ELINT Implications of Future Radar Designs
The constraints discussed in the sections above could be somewhat relaxed through
the use of sophisticated signal processing (i.e., by combining the returns in many
range cells prior to announcing target detection). Faster and cheaper processing is
the direction of current developments. Nevertheless there is a cost associated with
such changes in radar design and such changes will not occur without an impelling
need plus significant development time and sufficient funding. The ELINT commu-
nity must maintain close contact with radar R&D activities to keep abreast of
trends in radar design.
2.3.1 Bistatic and Multistatic Radars

As a general rule, performance of radars systems with receiving antennas located
at different places than the transmitting antennas is inferior to monostatic radar.
Bistatic and multistatic designs are interesting because of covertness, survivability,
flexibility, and economy. The most recent wave of interest was in the mid-1980s,
spurred by technological advances. The noncooperative coherent bistatic radar uses
a transmitter of opportunity (such as a TV transmitter). Through the early 1990s,
bistatic radar for surveillance had been reduced to practice but there was no
deployment. The use of transmitters of opportunity will always present technical
problems because the waveforms are not designed for radar use. In fact, the transi-
tion to HDTV will have an adverse effect on using TV transmitters because the
HDTV waveform is less suitable for radar use than current TV waveforms.
2.3.2 Radar Trends

Counter stealth and LPI, wideband and ultra wideband are among the trends in
radar design. In counter stealth (that is, radars to detect stealthy targets) there is no
evidence of a new class of radars currently under development using technological
breakthroughs to detect low RCS targets. Most radars today have some counter
stealth capability at short range, and greater power-aperture product would
improve that capability. The critical system problems are netting of radars deployed
more closely in space and the costs associated with it. The many air surveillance
radars in existence might be modified to operate with closer spacing and with
higher PRFs to provide the Doppler capability to better distinguish small moving
targets from birds and other clutter. PRFs are changed on a burst-to-burst basis
to uncover blind speeds. (S-band PRFs tend towards 2,500 pps for 60-km unambigu-
ous range and 125-m/s blind speed; L-band PRFs tend towards 1,500 pps for
100-km unambiguous range and 170-m/s blind speed.) The key to modification of
existing radar receivers to counter stealth is to maintain full sensitivity at shorter
range without saturation. This means that sensitivity time control may need to be
removed or modified.
LPI radar trends are to reduce peak power (and increase the duty factor) to
avoid detection by current ELINT and ES/ESM receivers designed for single pulse
detection (based on peak power). This is discussed further in Chapter 9. Low
sidelobes are important to avoid detection of the radar signal except in the main
beam; especially as a defense against antiradiation missiles (ARMs). Power manage-
ment and atmospheric attenuation shielding are often considered as LPI techniques
but are of very limited benefit in search radar. Power management is useful in
altimeters and some tracking systems. Atmospheric shielding is of limited use due
to the two-way path length of the radar. The radar path has more attenuation
than the one-way path to the intercept receiver unless the interceptor is located
more than twice the range from the radar’s target. Of course, if high altitude ELINT
collectors are used, the atmospheric effects are largely eliminated as the density of
the atmosphere decreases.
The ELINT challenge is frequency agile, noise-like signals of modest bandwidth
transmitted with high duty factor, as shown in Figure 2.7. The instantaneous
bandwidth is not likely to exceed that of conventional radars having the same
Figure 2.7 Modern frequency agile radar with 100% duty factor.
function and in every application will probably be less than 500 MHz. Randomness
in the radar’s pulse compression waveform makes it necessary for the ELINT
receiver to use energy detection techniques. Today’s radars typically transmit the
same pulse compression code or waveform in every coherent integration interval
during one mode. Tomorrow’s radars may choose to transmit different waveforms
from one coherent processing interval (CPI) to the next.
Wideband and ultra-wideband (UWB) radar are loosely defined in terms of
the fractional bandwidth used by the radar. F high and F low are the upper and lower
3-dB points of the radars bandwidth. The fractional bandwidth, FB, is given by
2(F high − F low )

FB = (2.17)
F high + F low
Wideband radar has a fractional bandwidth greater than 10%, ultra wideband
greater than 25% [5]. Applications include underground probing (e.g., mine detec-
tion), very short range (< 3m) such as proximity sensors and automatic braking
devices, and short range (< 100m) such as intrusion alarms and through the wall
detection. Due to limited energy on target, it is not likely that long range UWB
radar will be deployed. The use of ELINT receivers to detect such signals must be
weighed against the cost of doing so relative to the threat posed by these signals.
Due to their current short range, monitoring these signals is best done at short
range also. In this case, the need to sort them from among a myriad of other signals
is reduced and their high peak power makes detection relatively easy at short range.
2.3.3 Wideband Active Adaptive Array Radars

The concept of a shared antenna aperture is attractive. Such antennas could provide
radar, communications, and ELINT capability in one array. The aperture could
be subdivided into sections to allow different beam widths and scan patterns
simultaneously in different bands. A complete transmitter and receiver are required
at each antenna element along with the processing power to combine them coher-
ently in different ways for different functions. A fully adaptive array of thousands
of active elements appears to be too complex at present. There is no way to invert
such large matrices in near real time. This area of radar research is called space
time adaptive processing (STAP), and is the focus of many research efforts, mainly
for airborne radar applications. (One viewpoint is that STAP is a generalized
sidelobe canceller). However, partially adaptive arrays are practical. The radar
constraints given above apply to active phased array radars. The main purpose of
the active array is to provide better control of beam shape and to aim it in a desired
direction more quickly (without the effects of mechanical inertia). But the target’s
range and velocity characteristics still place constraints on the motion of the beam,
time on target, and so on. STAP and active arrays will eventually revolutionize
radar design. The question is when will it be deployed—that is, when will it become
practical and cost effective? If the number of array modules is small enough,
applications within the near term are likely and may already be deployed. For large
arrays, the answer seems to be probably not in the next decade.
2.4 Summary of Radar Design Constraints and Trends
The one-way range equation provides an advantage to the ELINT system (except
for very short range radars). The ability to use nearly matched filter processing in
the receiver gives the advantage to the radar. This is explored in more detail in
Chapter 9.
Radar signals are designed to achieve certain range, range resolution, velocity,
and velocity resolution properties, which are largely determined by the radar func-
tion and by the characteristics of its targets. These properties cannot be changed
solely for the purpose of denying the opponent’s ability to perform the functions
of ELINT—radar performance must also be maintained!
The ELINT community must maintain close contact with the radar community
in order to anticipate what the radar signals of the future will be like. There is
little need to develop ELINT technology (as applied to radar signals) in order to
intercept waveforms unable to perform radar functions. Note that noise radars
and impulse radars have many of the same constraints as conventional radars.
Some of the trends in radar signals are summarized below.
1. Instantaneous bandwidth < 100 MHz (more likely < 10 MHz for tactical
radar).
2. Frequency agility bandwidths ~10% of center frequency typical, 25% upper
limit until wideband shared apertures are deployed.
3. PRI agility increases as radars become computer controlled and perform
more functions. Dwell/switch (high PRF for pulse Doppler and low PRF
2.5 High-Power Microwave Weapons 27
for MTD), stagger for MTI, random jitter for protection from some types
of jamming. Programmed or scheduled PRIs when computer control of
multifunction radar is used (probably in all new designs for sophisticated
radar. Sliding PRIs useful for elevation searching or for adaptive tracking
of targets. All of these PRI variations are applicable when computer control
of multifunction radar is used.
4. Short duration pulses (< 50 ns) are not likely for robust tactical radars,
except for short range applications.
5. Bistatic/multistatic deployment will be limited. Even if these radars are
deployed, ELINT is only collected from the transmitter. Receive only sites
are not targets for ELINT. The principal ELINT concern is if exotic wave-
forms are chosen for transmission. Note that the constraints described here
still apply.
6. Power management is of little use for long-range search radars.
7. Low sidelobe antennas are likely to be increasingly used as protection against
ARMs and standoff jamming.
8. A major challenge is high duty factor (or CW) radar with modest coherent
bandwidth but wide frequency agility.
2.5 High-Power Microwave Weapons
Leaving the subject of radar for the moment, consider the high power microwave
(HPM) weapons being developed to produce field strengths strong enough to
damage sensitive receivers. One approach is to transmit very short but high ampli-
tude pulsed signals. Designers of these waveforms need not be concerned with
radar constraints; however, the physics of atmospheric effects and breakdown place
limitations on the amplitude of such pulses at the source. An explosively pumped
flux compression generator (FCG) is the technology used in electromagnetic bombs.
A lethal radius of a few hundred meters from a distance of a few hundred meters
is likely. The term lethal means that electronic equipment is damaged or at least
upset. The terawatts of power in the dc pulse generated is 10 to 1,000 times greater
than that in a typical lightning strike.
From the ELINT point of view, there are several questions:
1. Is it necessary to harden ELINT receivers to withstand this type of electronic

attack? (This part of EW is called electronic protection or EP.)
2. Is it necessary to intercept and make parameter measurements on microwave
bursts from such weapons?
3. Will similar wideband signals made up of very narrow pulses be used to
transmit data?
4. Will similar wideband signals made up of very narrow pulses be used as
radar waveforms?
The high power density and wide bandwidth of these beam weapons makes it
easy for an ELINT receiver to detect them. A typical wideband radiometer could
be expected to produce a signal well above the threshold at ranges much greater
than the lethal range; the problem to be solved, however, is to distinguish these
wideband, low duty factor pulses from lightning, ignition noise, and other types
of wideband interference. For this it may be necessary to use multiple-pulse data
to detect periodic components. Alternatively, multiple ELINT sites may collect data
from a single pulse to locate the source of the emission in the manner currently
used by lighting strike detection and location systems.
Wideband communications signals using very short pulses may be used to
transmit missile guidance data or otherwise be a part of the ELINT signal environ-
ment. These could be very difficult to detect if random pulse spacing known only
to the receiver but not to the interceptor is superimposed on the data. Wideband
radar signals using short pulses are less of a problem to the ELINT community.
In addition to the problem of excessively fine range resolution as the pulse duration
becomes short, there is the need to get the same amount of energy back from the
target as when a longer duration pulse is used. This means that the transmitted
pulses are of higher amplitude and can be more easily detected.
References
[3] Pace, P. E., Detecting and Classifying Low Probability of Intercept Radar, Norwood, MA:
Artech House, 2004.
[4] Wiley, R. G., Electronic Intelligence: The Interception of Radar Signals, Dedham, MA:
Artech House, 1985, Chapter 2.
[5] Wiley, R. G., Electronic Intelligence: The Analysis of Radar Signals, 2nd ed., Norwood,
MA: Artech House, 1993, pp. 2–4.
CHAPTER 3
Characteristics of ELINT Interception

Systems
3.1 Intercept System Characteristics and Functions
The characteristics that are important to intercept system designers reflect the
functions which such systems are expected to perform, namely, (1) to intercept
signals, (2) to recognize the type of transmitter, and (3) to measure signal parame-
ters. The purposes of making detailed parameter measurements include determining:
• Technical features of the radar’s design and construction;

• The radar’s performance capabilities;
• The radar’s weaknesses for jamming (including deception), or other kinds
of attack.
The intercept system characteristics of interest include those shown in Table

3.1. In general, interception requires high sensitivity and dynamic range covering
large angular regions and frequency bands, but the only information expected is
the existence of the signals. The dynamic range need not be excessively wide, since
some saturation can be tolerated. For the recognition function, the receiver demands
can be relaxed somewhat if the interception process has allowed reducing the
angular sector and frequency coverage. Then only a few signals will be received
Table 3.1 Receiver Characteristics for Various Functions

Function
System Characteristic Interception Recognition Measurement
Probability of intercept High Moderate Low
Sensitivity High Moderate High (for improved
SNR)
Dynamic range Moderate Moderate Large
Coverage band Wide Moderate Moderate (for
conventional radars)
Analysis bandwidth Narrow Moderate Wide
Exotic environment performance High High High
Dense environment performance High Moderate Low
Single pulse performance High Moderate Low
Multiple simultaneous signal High Moderate Low
performance
Processor/signal sorter requirements High Moderate Low
29
30 Characteristics of ELINT Interception Systems
at once. Usually, recognition requires measuring the basic signal parameters

(typically RF, PRI, pulse duration, and scan period) to accuracies on the order of
0.1 to 1%. Naturally, any unusual modulations must be noted to aid in the recogni-
tion process.
For the measurement function, the probability of intercept need not be high
for signals in general, but must be acceptable for only that particular signal on
which measurements are to be made. The sensitivity must be sufficient to provide
the high SNR required for precision measurements (generally, the variation due to
noise in a parameter measurement is proportional to the reciprocal of the square
root of the SNR). The dynamic range should be large in order to permit obtaining
a high SNR without saturation in addition to allowing characterization of nulls
and lobes in the antenna pattern during scanning. Conversely, the coverage band
need only allow for accurately tuning to the carrier of the signal, but the analysis
bandwidth should be wide enough to investigate intentional or incidental variations
in frequency or phase. Generally, a dense environment can be reduced by using
directive antennas and frequency selectivity, so that the measurement receiver need
not have the ability to deal with large numbers of signals at once. Likewise, it is not
necessary to employ sophisticated signal sorters and processors since the reception
process attempts to narrow the field of view, frequency range, and dynamic range
to match the signal under scrutiny. Clearly, an intercept system can perform all of
these functions, but if it does so simultaneously—that is, attempts to intercept,
recognize, and measure all signals from all directions and all frequencies at once—
numerous compromises and complications are inevitable.
Describing the requirements for ELINT receivers leads naturally to consider-
ation of various aspects of the signal environment on which the receivers must
operate.
This includes some consideration of radar signal modulation as well as the
geometry of the intercept situation.
The best approach for an ELINT system is to have very flexible capabilities
so that adapting to the functions of interception, recognition, and measurement
can be done with ease as the situation demands. This requires a variety of directional
and nondirectional antennas with a variety of polarization, several types of wide-
band and narrowband receivers, a variety of demodulation and analog-to-digital
conversion capabilities, and numerous software packages for analysis and display
as well as storage devices. In a sense, the ELINT system covers the whole state of
the electronic arts, and the system operator should have knowledge in each of these
areas. Such diversity is apparent in this chapter.
3.2 Frequency Coverage
Of course, frequency coverage of the antenna system and receiver front-end must
include all of the frequencies used by the transmitters of interest. In many cases,
the frequencies in use are not known, and thus wide frequency coverage is the rule
for intercept systems. Since the total range of frequencies potentially used by radar
and radar-related systems is so broad (ranging from below 30 MHz to above
100 GHz), the coverage must be divided into bands compatible with practical
components (e.g., antennas, amplifiers, mixers, filters, and detectors). The typical
system makes use of octave bands (i.e., the ratio of the upper band edge frequency
to the lower band edge frequency is 2:1). This is useful in that any harmonics
which may be generated by nonlinearities will be outside the band. Broader bands
are possible and practical, particularly in receivers which do not use frequency
conversion (heterodyne conversion), such as crystal video or instantaneous fre-
quency measurement (IFM) types (also termed direct detection receivers). In the
microwave region, coverage of several octaves in one antenna and receiver can be
achieved (e.g., 1 to 10 GHz), if desired.
Both radar designers and intercept receiver designers are subject to the con-
straints of the state of the art in components. However, the designer of a particular
radar, who must design a high power transmitter as well as a receiver, is generally
subject to more constraints than the intercept receiver designer when it comes to
the band covered by the radar’s transmitted signals. First, the high power restricts
the component performance (e.g., gain × bandwidth product of amplifiers tends
to be a constant), and second, the need to maintain control over the radiated
spectra for electromagnetic compatibility restrains even the most innovative radar
designer. Thus, the intercept receiver can usually accommodate the frequencies
used by a single radar system or type. The interceptor wishes to maintain enough
flexibility to cover a variety of radar types, and this complicates the interceptor’s
problem considerably. Generally speaking, the interceptor’s antenna system band-
width and receiver coverage bands will be selected based on what can be achieved
economically, rather than based on the extremes of a particular radar’s tunability
or spectrum width.
Common frequency bands and their letter designations are shown in Figure
3.1. The two different letter designations shown are both in use. The older designa-
tions have the benefit of many years of common use. The more orderly new band
designations are used widely in the electronic countermeasures community and
have a logical pattern. Since both designators are in use, such phrases as ‘‘I-band-
old-X-band’’ are needed to avoid confusion. The best approach is to simply give
the frequency range in hertz (or megahertz or gigahertz) and eliminate any doubt
about what is meant.
3.3 Analysis Bandwidth
The analysis bandwidth of a receiver is usually the instantaneous predetection

bandwidth available for analysis of the signal spectrum structure. In a superhetero-
dyne receiver, it refers to the bandwidth at the intermediate frequency (IF). In some
cases, if the signal envelope is to be analyzed, the term ‘‘analysis bandwidth’’ may
refer to the postdetection bandwidth—particularly, in a direct detection receiver.
It will be clear from the context of the analysis problem exactly how the analysis
bandwidth is defined: it is the bandwidth which restricts the signal processing or
measurement at hand.
The analysis bandwidth of an intercept receiver is usually not ‘‘matched’’ to
the bandwidth of a particular radar signal (which would improve the SNR), but
is wider for two reasons:
Figure 3.1 Frequency bands.
1. The intercept receiver must be capable of analyzing many different radar

signals having a variety of bandwidths. Hence, it usually is as wide as
needed for the widest bandwidth signal expected. Modern signal processing
approaches may apply multiple bandwidths to the same data.
2. The analysis may be concerned with features of the radar signal incidental
to the radar’s performance and contained in portions of the signal spectrum
beyond the region used by the radar receiver (e.g., pulse top ripples whose
period is a fraction of the pulse duration).
Conventional pulsed radars do not normally pose difficult problems for the
intercept receiver designer. Once again, since both the radar and interceptor use
similar components, the interceptor can maintain an analysis bandwidth at least

as wide as the widest radar signal. Radars may make use of both pulse compression
and frequency agility, especially wideband radars used for target discrimination.1
Such radars pose some additional problems for the intercept receiver designer.
3.3.1 Wideband Radar Signal Trends2

Ultimately, the bandwidth requirements for intercept receivers are established by
the waveform designs used in radar systems. Hence, a discussion of the directions
being followed by radar designers is important to the design of intercept receivers.
Wide-bandwidth pulsed signals are used in a number of radar applications.
Some of the applications are listed in the approximate order of their popularity in
Table 3.2. Wideband signals are used in radar applications to improve radar range
resolution, which depends on the reciprocal of the transmitted signal bandwidth.
A radar designer can simultaneously increase detection range (average transmitted
power) and maintain (or improve) the range resolution through pulse compression
techniques without increasing the peak power transmitted by the radar.
As noted in Chapter 2, in a conventional radar, bandwidth (range resolution)
is inversely proportional to the pulse width so that the range resolution becomes
worse as the pulse becomes wider. For a fixed peak power, then, the designer can
improve detection range only by increasing the pulse width, if the PRI is fixed.
Pulse compression allows simultaneous detection range and range resolution
improvement.
There are several ways to achieve pulse compression, as listed in Table 3.3.
One class of techniques is to modulate the frequency within the pulse in one of
the ways listed. A second method is to use phase modulation within the pulse. By
far the most common frequency modulation technique is to impart a linear fre-
quency ramp on the transmitted pulse.
Frequently used terms in pulse compression systems are time-bandwidth prod-
uct and pulse compression ratio. The time-bandwidth product is the product of
the transmitted pulse width (T) and the transmitted signal bandwidth (W). For a
conventional radar, TW is approximately 1. In a pulse compression radar, it is
Table 3.2 Radar Applications of Wideband Signals

Search-surveillance
Tracking (missile and aircraft)
Missile measurements and instrumentation
Ground mapping
Decoy discrimination, reentry, measurements, ABM
Radar imaging
Ionospheric propagation experiments
Active homing (seekers)
High-resolution battlefield surveillance
Artillery and mortar locating
Missile guidance
1. Target discrimination means using the fine structure of the target echo to determine information about
the size and shape of the targets.
2. The author is grateful for the help of Mr. Robert B. Shields in preparing this section.
Table 3.3 Wideband-Pulsed Signals

Pulse Compression Techniques
Frequency modulation on-pulse
Linear frequency deviation (up or down)
Stepped frequency
Nonlinear frequency deviation
Binary frequency coded
Phase modulation on-pulse
Phase reversal (180°)
Polyphase
Other Wideband Technique
Very short pulses (0.5 to 150 ns)
rare to find a time-bandwidth product less than 10, and some radars have products
above 100,000. Figure 3.2 is a scattergram showing pulse duration versus frequency
deviation for a number of radar designs. Lines of constant TW are also shown.
Pulse compression ratio is defined as the ratio of the uncompressed pulse width,
␶ , to the compressed pulse width, ␶ c . The pulse compression ratio is approximately
Figure 3.2 Scatter diagram of frequency deviation versus pulse duration for some linear FMOP
radars.
equal to the time-bandwidth product. However, it is always somewhat less than

TW since, in the processing, amplitude weighting is employed in the receiver to
reduce the time sidelobes. The weighting necessarily widens the compressed pulse
so that ␶ /␶ c < TW.
Some trends in radar design include widespread use of frequency modulation
on-pulse (FMOP), with increasing amounts of frequency deviation. Phase modula-
tion on-pulse (PMOP) is also being increasingly employed. Except for cases where
the phase modulation is used for missile guidance in a radar-directed weapons
system, the purpose of phase coding is generally the same as the frequency modula-
tion: to improve range resolution. Phase reversal systems are the most common, as
observed in Table 3.4. Bit lengths as short as 12 ns are being utilized experimentally,
although the bit length range of 0.1 to 6 ␮ s is more representative.
Quadrature phase coding (Taylor codes) and Frank polyphase coding represent
two other types of coding used to compromise the advantages and disadvantages
of frequency modulation and phase modulation. The Frank polyphase code actually
uses a quadratic phase progression (modulo 2␲ ) and hence imitates linear frequency
modulation [1].
3.4 Dynamic Range
The idea of dynamic range is simple to convey: it is the ratio of the largest signal
amplitude that the receiver can process to the smallest signal amplitude that the
receiver can instantaneously process. However, the words ‘‘can process’’ are trou-
blesome to define. The smallest signal that can be analyzed depends on what the
analysis process is like and what its purpose is. Clearly, a precision phase measure-
ment within the pulse requires a larger signal than simply stating that a pulse has
Table 3.4 Parameters for Some Phase Modulation Radars

Phase Coding RF (GHz) Pulse Width (␮ s) Bit Length (␮ s)
180° reversal 0.02 2,000 20.0
0.42 6 0.46
0.43 390 6.09
— 780 6.09
0.60 52 4.0
5.6 0.2125 0.0125
9.35 18.1 0.203
16.25 2.0 0.154
9.0 CW 125 (511 bits)
9.0 CW 1.5 and 0.3 (63 and 255 bits)
10.0 CW 2.19
10.0 CW 0.0133 (8,000 bits)
3.0 6.6 0.5
5.5 6.0 0.1
8.75 CW Unknown (range resolution = 25m)
10.2 CW Unknown (range resolution = 35m)
Other
Quadrature — 25.6 0.2
Frank — 39.2 0.0625
Polyphase — 78.4 0.108
crossed a threshold. Likewise, the largest signal that can be tolerated also depends
on the measurement being made. Making a precise ERP measurement when the
radar’s main beam is pointed toward the intercept receiver requires that saturation
be minimized; whereas a frequency measurement can be made after a limiter and
saturation is of little importance if only one signal is present. Receiver dynamic
range must be measured under a variety of conditions, but the conditions must be
carefully delineated to avoid confusion and to be certain that a particular receiver
can perform a particular function. In a simple test the minimum signal might be
determined in the same way as the sensitivity is specified, and the maximum signal
level might be that for the ‘‘1-dB compression point.’’ This point is illustrated in
Figure 3.3. In the linear portion of the input-output transfer characteristic, a 1-dB
change in the input level produces a 1-dB change in the output level. As the input
level increases, various components in the receiver begin to saturate and the output
level is unable to continue to increase in the same proportion to increases in the
input level. When the input level reaches a point at which the output level is 1 dB
less than it would have been had there been no saturation, the 1-dB compression
point has been reached. At this point, for example, the measured power level would
be 1-dB less than it would be if the receiver was erroneously considered to follow
a linear characteristic. The careful intercept operator frequently inserts calibration
signals of known amplitude and stores the receiver output to allow later comparison
of the actual signal amplitudes to the known calibration levels.
Figure 3.3 1-dB compression point.

3.4.1 Dynamic Range Requirements

Given that there are problems of specifying and measuring dynamic range, the real
driving force behind our interest in the subject is the radar signal environment
itself and the wide dynamic range that can be encountered. First, consider that the
transmitters of interest can have power outputs ranging from a few watts (beacons,
altimeters, IFF, and so on) to a few megawatts (long-range search radars). Thus,
a 60-dB dynamic range might be expected due to the variety of transmitters alone.
Next, the antennas used range from omni-directional types with less than 10-dBi
gain to large apertures with gains approaching 40 dBi—another 30 dB of possible
dynamic range. Indeed, even for a single scanning radar, the received signal level
may be of interest to the interceptor over a 40- to 50-dB dynamic range to allow
exploration of the antenna pattern.
Another factor that greatly affects the received signal level is the range from
the emitter to the interceptor. The variation in this parameter depends largely on the
height of the interceptor (for ground-based emitters). As shown in Figure 3.4, the
distance to the horizon for a point at an altitude, h, above the Earth is given by
√(h + R Earth )
2
d= − R e2 ≅ √2R Earth h (3.1)
(This is derived in Chapter 7.) The minimum range (h) is when the emitter
and interceptor are on a line normal to the Earth’s surface. The received power
Figure 3.4 Computing the distance to the horizon.

varies as the reciprocal of the square of the range. Therefore, the path loss dynamic
range (PLDR) is
maximum range 2R Earth

PLDR dB = 20 log = 10 log (3.2)
minimum range h
where R Earth is radius of Earth, and h is the height of interceptor above the Earth.
The Earth’s radius is about 4,000 miles. Therefore, the PLDR values for a 4/3
Earth-propagation model are as shown in Table 3.5.
The important fact is that the PLDR decreases with altitude so that at high
altitudes the PLDR is greatly reduced. This is not surprising since in space, at a
large distance from the Earth, all signals, regardless of their terrestrial location,
would travel about the same distance. For aircraft operations and typical altitudes
of 1 to 10 km, the PLDR is on the order of 32 to 42 dB. For ground stations, the
PLDR can be decreased by locating the station at a high elevation point and in an
area free from nearby emitters.
Combining all of these dynamic range terms gives a total dynamic range
expected at the intercept receiver of
60 dB + 30 dB + PLDR or 132 dB to 122 dB
for the 1- to 10-km altitude typical of an airborne intercept receiver. Regardless

of the altitude, it is clear that a very wide dynamic range will be encountered in
the signal environment.
The instantaneous dynamic range of a linear receiver is typically on the order
of 50 dB, while that of log-video and limiting receivers (such as IFM) may be
80 dB.
The environment’s total dynamic range need not be handled instantaneously
in all cases. Several receiver channels can be used to cover different portions of
the dynamic range if necessary. Also, a gain control can be used to position the
available instantaneous dynamic range at any given portion of the total range to
examine a particular signal. Likewise, the gain control may be used to bring different
portions of a signal within the linear portion of the receiver’s dynamic range at
different times. For example, one setting may capture only the main beam and a
few sidelobes of radar’s scan. Another setting causes saturation when the main
beam passes, but the sidelobe structure may be easily observed. Joining separate
amplitude versus time plots from both gain settings can provide a total of, let us
say, 80 dB of dynamic range for analysis (50 dB from each, setting, with 10 dB
of overlap to allow for proper alignment).
Table 3.5 Path Loss Dynamic

Range as a Function of Altitude
Altitude h PLDR (dB)
100m 52.3
1 km 42.3
10 km 32.3
100 km 22.3
1,000 km 12.3
3.5 Sensitivity 39
3.5 Sensitivity
The sensitivity of an intercept receiver is defined as the minimum signal power

required at the receiver input. ‘‘Required’’ is a tricky word here, for it is necessary
to describe the purposes to be fulfilled before the minimum signal level can be
determined. At the outset, it should be stated that the reason for the existence of
a sensitivity floor (below which the interceptor’s purposes cannot be fulfilled) is
noise. Generally, it is random, thermal noise. Man-made noise or interference can
also be a problem, but it must be considered on a case-by-case basis.
Nearly all noise at radar frequencies is generated in the receiver itself (internal
receiver noise). Noise also arrives via the antenna, and is basically a consequence
of the thermal radiation reaching the antenna from either terrestrial sources or
space. The level of this radiation is a function of the temperature in the direction
the antenna is pointing. (The dark night sky has a lower noise level than if the
antenna were pointed toward the Earth or the Sun.) This is the principle used in
radio astronomy to determine maps of the radiation intensity across the sky at
various wavelengths.3 If the antenna is directed toward a quiet region of the night
sky, it might observe an equivalent temperature of only a few degrees Kelvin or,
if looking at the Earth’s horizon, the observed temperature might be closer to 290
Kelvin (290K is used as the standard or nominal noise temperature expected in an
ordinary, noncryogenic environment).
The noise power at the receiver terminals is given by kTB, where k is Boltz-
mann’s constant which is 1.2 × 10−23 watts per hertz per Kelvin, B is the effective
noise bandwidth (in hertz) through which the noise is received, and T is the
temperature of the noise source (K). For intercept receivers, we may assume an
input (antenna terminal) temperature T = 290K and express the noise power as
3.7 × 10−15 W/MHz or −114 dBm per MHz of bandwidth. The ratio of the actual
receiver output noise to that which would result from this amount of noise applied
to its input is called the noise figure:
Noise out of receiver

(NF) =
Gain of receiver × Thermal noise at input with 290K term
Clearly, (NF) can be no less than 1. Notice that even if the noise figure were to
be reduced to 1, the noise at the input terminals would still depend on what the
antenna was observing and in many cases could not be reduced below −114 dBm/
MHz. Often, (NF) is expressed in decibels, in which case the equivalent noise level
referred to the receiver input terminals is −114 dBm/MHz plus 10 log (NF). It is
this noise background against which the intercept receiver attempts to detect signals
and measure their characteristic parameters.
Low-noise receivers may have a noise temperature specified rather than a
noise figure. For low-noise applications—such as communications satellite ground
stations—the receiver may be specified as having a certain noise temperature (e.g.,
80°). This means that the amplifier adds 80° to the input noise temperature from
3. The fascinating personal story of John Kraus’ pioneering efforts in radio astronomy is contained in The
Big Ear (Cygnus-Quasar Books, 1976).
the antenna, which may be low since it is directed towards a satellite. Noise figure
is defined for an input noise temperature of 290°. The noise figure is, therefore,
290 + 80
or 1 dB
290
Typical broadband low-noise amplifiers have noise figures in the 2- to 8-dB

range at common microwave frequencies.
At the low end of the frequency spectrum in which radars are found, the noise
limitation is not receiver noise, but atmospheric noise and man-made interference.
This is the limit in the 3- to 30-MHz band, where over-the-horizon (OTH) radars
operate. These make use of ionospheric reflections to detect targets along great
circle paths around the world. Naturally, the variations in the ionospheric path
length and its characteristics make radar range and velocity measurements less
precise. Also, highly directive antennas are difficult to build, making the bearing
to the target more difficult to determine.
3.5.1 Noise Figure Measurement4

Historically, the discussion of noise figure measurement and the many different
definitions was important [2–4]. The IEEE definition of noise figure clarified this
situation. In this book, the temperature presented by the antenna and feed network
to the receiver input is assumed 290K unless otherwise noted. There are essentially
two ways to measure the noise figure of a network (amplifier, receiver): (1) using
a noise generator, or (2) using a signal generator.
Generally speaking, the noise generator method is more accurate and is used
mainly to measure small noise figures on the order of 30 dB or less. (The limitation
is the amount of power that the noise source must supply above and beyond that
generated internally in the network.) Limitations in the amount of noise power
available from noise diodes limit the measurable noise figure using a noise generator.
The two basic methods available to measure the noise figure of a network
using a noise generator (source) are: (1) increase the external noise by 3 dB, and
(2) measure the Y-factor.
In the 3-dB power-increase method, the device under test and the external
amplifier (if one is used) must be capable of supplying a power output 3 dB above
the network’s internal noise level and remain in linear operation (not saturated).
Figure 3.5(a, b) shows typical equipment configurations. Basically, the procedure
is to determine a noise level on the power meter or true rms voltmeter [Figure
3.5(a)] or on the detector/voltmeter combination [Figure 3.5(b)] with the 3-dB pad
out of the signal path. For Figure 3.5(a), the noise source is turned on and the
noise output level is increased until the power indicator reads 3 dB more than with
the noise source off. For Figure 3.5(b), switch in the 3-dB pad and adjust the noise
source to obtain the same reading on the voltmeter as when the pad was not in
the circuit. The noise figure is then determined from the calibration on the noise
source for the noise power required to obtain the 3-dB increase.
4. The author is grateful for the help of Mr. Grover M. Boose in preparing this section.
3.5 Sensitivity 41
Figure 3.5 Noise figure measurement configuration: (a) 3-dB increase; and (b) 3-dB increase with
attenuation substitution.
3.5.2 Y-Factor Measurement

The Y-factor is defined as the ratio of the network’s output power with the noise
source on to that when the noise source is off. The noise figure in decibels is then
Th
−1
To
NF dB = 10 log = 10 log ENR − 10 log (Y − 1)
Y−1
where Th is the temperature of the noise source when turned on, and To is the off
noise temperature. The numerator, Th /To − 1, is the available excess noise ratio
(ENR), which is given as part of the noise source calibration. The equation above
assumes that the temperature of the noise source when it is off is the standard
temperature, 290K. If more accurate measurements are required, the actual hot
and cold temperatures can be measured along with the hot and cold output noise
powers. Some precautions to be taken when connecting the equipment for Figure
3.5(a, b) are as follows:
1. The meter in Figure 3.5(a) must either be a power meter or a true rms
voltmeter. The meter must also be connected in the circuit prior to any
detector (i.e., at the IF output in a receiver).
2. Calibrated and precision attenuators should be used in the attenuator substi-
tution method.
3. The meter in Figure 3.5(b) is only used as an indicator. Accuracy is a
function of the calibrated attenuator.
4. Attention must be paid to impedance-matching the noise source to the device
under test. In fact, impedance should be matched at all connections (auxiliary
amplifier, attenuator) to ensure maximum power transfer and correct noise
figure measurements.
Using a noise generator provides a simple way to measure noise figure that
does not require a measurement of the noise bandwidth.
The signal-generator method accuracy is determined by the calibration of the

generator, and the requirement that the noise bandwidth (not to be confused with
the 3-dB bandwidth) must be known to complete the determination of the noise
figure. Additionally, it is generally assumed that the input and output noise tempera-
tures are equal to 290K and that the gain of the network is the same for noise as
for the signal (i.e., that the system is linear). This method is most applicable to the
measurement of the ‘‘radar’’ noise figure, which is defined as the noise figure of a
receiver when the signal spectrum is contained in one narrow band. If there are
two bands in which a signal can enter a receiver, such as in a superheterodyne
receiver with no preselector, the actual noise figure is 3 dB less than that measured
by the small-signal method. The small-signal method has won wide acceptance as
a means of checking network performance, even if the accuracy is ±2.0 dB or
poorer, since the equipment needed to perform the measurement is readily available
in most laboratories, namely, a signal generator and some indicating device.
The measurement of a network noise figure using a signal generator follows
this procedure:
1. Connect equipment as in Figure 3.5 except that a signal generator is substi-

tuted for the noise source.
2. With signal generator turned off, obtain a reading of the noise output power.
3. Turn on the signal generator and tune it to the center of the band of the
amplifier or receiver under tests.
4. For Figure 3.5(a), adjust the signal generator power output to obtain a
reading exactly twice what it was in step 2 (3 dB higher on a decibel scale).
For Figure 3.5(b), insert the attenuator and adjust the signal generator to
obtain the same output indication as before.
Under the above conditions and assuming that the input termination tempera-
ture is 290K, the noise figure can be determined from
Input signal level

NF =
k (290)B N
where B N is the noise bandwidth. The noise bandwidth of a receiver or amplifier

can be determined by:
1. Measuring the 3-dB bandwidth and knowing the type and number of cou-
pling circuits comprising the amplifier stages. Table 3.6 is then used to
determine the noise bandwidth. If there are more than five stages, using the
3-dB bandwidth as the noise bandwidth results in an error of approximately
10% (or approximately 0.5 dB).
2. The most straightforward way of measuring the noise bandwidth of a circuit
(or receiver) is to measure its frequency response using a variable-frequency
signal generator, and integrate the area under the response normalized to
unity at the point of maximum response.
In most instances, the 3-dB bandwidth is known or easily measured; then the
noise bandwidth is determined from Table 3.6. Unless extreme accuracy is required
3.5 Sensitivity 43
Table 3.6 Comparison of Noise Bandwidth and 3-dB Bandwidth

Type of Receiver Coupling, Number Ratio of Noise Bandwidth
Ckt of Stages to 3-dB Bandwidth
Single-tuned 1 1.57
2 1.22
3 1.16
4 1.14
5 1.12
Double-tuned* 1 1.1
2 1.04
Stagger-tuned triple 1 1.048
Stagger-tuned quadrature 1 1.019
Stagger-tuned quintuple 1 1.01
Gaussian 1 1.065
*Applies to a transitionally coupled double-tuned circuit or to a stagger-tuned circuit
with two tuned circuits.
Source: [5].
for the noise figure, this procedure, with a calibrated signal generator, will be
accurate to ±2 dB. Care must be taken in selecting the meter to ensure that it has
sufficient bandwidth. A broadband power meter is adequate for applications where
the center frequency is above 10 MHz. A true rms meter is useful for frequencies
below 10 MHz. If the noise is impulsive in nature, the meter crest factor becomes
important; the higher, the better [5]. Impulsive noise is most prevalent at HF (2 to
32 MHz) where thunderstorm activity around the world contributed to atmospheric
noise levels.
3.5.3 Some Sensitivity Measures

Receiver sensitivity is measured using several criteria. Figure 3.5(a) shows one
measurement scheme for determining the receiver noise. The power meter is like
a square-law detector in that the output (dial reading) is proportional to the input
power. The integration or averaging of the metering circuit and the tendency of
an observer to read the average value of the meter pointer along the dial corresponds
to postdetection filtering. If there were little or no filtering, the ideal power meter
reading would rapidly fluctuate over a wide range of power values as the noise
level changed. Suppose we wanted to detect the presence of one signal against the
thermal noise background by observing the power meter. By using very long averag-
ing times, the power meter reading would become quite steady (very small fluctua-
tions). Then a small amount of additional power (which might be several times
larger than the amount of fluctuation) could be reliably detected. Under these
conditions, it is clear that while the receiver noise figure and bandwidth set the
noise level reaching the power meter, our ability to detect a signal in the presence
of that noise is greatly affected by the averaging done after the RF signal has been
converted to a voltage proportional to the power. This postdetection averaging is
characterized by a video bandwidth, B v , which is the 3-dB cutoff frequency of the
low pass filter placed after the detector. The important point is that our ability to
detect the signal in the presence of noise depends not on the average value to which
the power meter needle points, but on how much it is fluctuating. In concept, we
may simply offset the meter scale to place the average value at zero and then detect
the signal in the presence of the fluctuations only.
Tangential signal sensitivity is a commonly used receiver sensitivity term. In
the light of this model, it is defined as follows:
When the signal is present, the lower limit of the power meter reading fluctuation
is just equal to the upper limit of its fluctuation when the noise alone is present.
In practice, an oscilloscope is used to display the receiver output and a pulsed
signal is used. The tangential condition is when the ‘‘top’’ of the noise alone is at
the same level as the ‘‘bottom’’ of the noise present on the top of the pulse.
Making laboratory measurements of the signal level required to meet this

condition is somewhat subjective and depends on the judgment of the observer.
Usually, the variation from observer to observer is on the order of 2 dB. Tangential
signal sensitivity is not a good measure of receiver sensitivity because it is subjective
and no receiver function or performance is specified. However, it is useful as a
tool for rapidly determining approximate receiver sensitivity.
While the concept of tangential signal sensitivity is traditionally used and easy
to measure in the laboratory, receiver sensitivity must be measured in terms of the
signal power needed in order to perform some function. The detection function
has been analyzed in great detail by radar designers, and is formulated in terms
of the probability of detection, PD (i.e., the probability that the signal plus noise
crosses a threshold), and the probability of false alarm, PFA (i.e., the probability
that noise alone crosses the same threshold). As the SNR increases, a threshold
can be selected such that PD → −1 and PFA → 0.
For the tangential signal condition, one can imagine that a threshold is selected
so that the probability of noise alone crossing the threshold is the same as the
probability that the signal plus noise fails to cross the threshold; that is,
1 − PD = PFA
This condition alone does not define a SNR, since for any SNR a threshold
can be selected for which the above condition is satisfied.
3.5.4 Output SNR and Receiver Applications

The signal level at the receiver input determines its output SNR. The output SNR
is related to the input SNR by the processing gain:
SNR out = SNR in ⭈ (Processing Gain) (3.3)
The SNR at the input is the ratio of the signal power at the input to the noise
level referred to the input. The ELINT processing gain is basically related to the
postdetection filtering done prior to any measurement or detection decision; for
example, the averaging done by the power meter in Figure 3.5(a). The processing
gain in this case is [6]
3.5 Sensitivity 45
GP = 冋册
B RF ␥
2B V
where:
B RF = predetection or RF bandwidth
B V = postdetection or (one-sided) video bandwidth
␥ = integration efficiency (typically 0.5 to 0.8)
In this situation, we have
Signal power
R in =
Noise power referred to input
Thus,
SNR out = SNR in 冋册

2B V
=
kTB RF 2B V 冋册
B RF ␥ Signal input B RF ␥
For ␥ = 0.5, we have
Signal input
SNR out =
kT (2B V B RF )1/2
The above result explains the use of the term ‘‘effective bandwidth’’ for
√2B V B RF
Consider estimating the pulse power. The fluctuation in the measurement
depends on the output noise level; in fact, the fractional standard deviation in the
measurement will be the reciprocal of the output SNR:
√
kT (2B V B RF )1/2 1
␴P = =
Signal power (SNR out )1/2
This form is characteristic of signal parameter measurements of all types; that

is, the error in the measurement is proportional to the reciprocal of the square
root of output SNR. Examples of parameters so affected include amplitude, angle
of arrival, time of arrival, and carrier frequency. (This assumes an output SNR
larger than 10 to 15 dB.)
3.5.5 Threshold Detection

Threshold detection plays a central role in automated ELINT receivers, since it is
necessary to first make a decision that a signal is present before any parameter
measurements can be made. This problem has been considered extensively in the
radar literature (e.g., Barton [7]). The only difference is that for radar receivers
detecting target echoes, the receiver postdetection bandwidth and predetection
bandwidths are matched to the transmitted signal and to each other (i.e., B V − 0.5
B RF ). In an ELINT receiver, particularly the wideband types that most often use
automatic detection, the ratio of the RF bandwidth to the video bandwidth can
be quite large, which affects the statistics of the output noise. This aspect is analyzed
by Tsui [8]. Figures 3.6 and 3.7 show the changes in the statistics at the square
law detector output for various values of B RF /2B V for no signal and for an input
SNR of 2 (or 3 dB). As the postdetection bandwidth narrows, the output distribution
peaks near the expected (or dc) value of unity (no signal) or 3 (signal = 2 units,
noise = 1 unit). The resulting detection probability is shown in Figure 3.8 for
several values of B RF /2B V for a false alarm probability of 10−8.
As the ratio B RF /2B V becomes large, the output noise and signal plus noise
distributions become normally distributed [9, 10]. Suppose the noise at the input
to the detector consists of in-phase and quadrature components each having a
mean of zero and variance ␴ 2; that is,
2
1 −v in /2␴ 2
P (v in ) = e (3.4)
√2␲␴
Figure 3.6 Output distribution for the noise voltage for several values of B RF /2B V . (From: [9],
p. 22.  1967 IEEE. Reprinted with permission.)
3.5 Sensitivity 47
Figure 3.7 Output distribution for SNR = dB for several values of B RF /2B V . (From: [9], p. 23.
 1967 IEEE. Reprinted with permission.)
Then the noise voltage after the linear envelope detector has the Rayleigh
distribution, namely,
v −vd2 /2␴ 2
P (vd ) = d e , vd > 0 (3.5)
␴
Note that ␴ 2 = N 0 B = noise power in bandwidth, B, at the detector input.

The distribution of the square of the detector output voltage is
2
−vd /2␴ 2
P 冠vd 冡 =
2 1 2
e , vd > 0 (3.6)
2␴ 2
2
=0 , vd < 0
Then the means and variances for each of these distributions can be computed
as:
Figure 3.8 Probability of detection for several values of B RF /2B V .
Mean of envelope, vd = ␴ √␲ /2
Variance of envelope = (2 − ␲ /2)␴ 2

2
Mean of square envelope, vd = 2␴ 2
Variance of squared envelope = 4␴ 4 = 冠vd 冡

2 2
The action of the video low pass filter is to compute the sum of a number of
independent samples from the detector output, the number being B RF /2B V . The
distribution of the sum of a large number of identically distributed random variables
approaches a Gaussian distribution whose mean is the sum of the mean values of
the individual samples and whose variance is the sum of the individual variances.
For ␴ = 1, and B RF /2B V large, the noise distribution at the output of the low pass
filter is, therefore,
[v out − (B RF /2B V )]2
1 4(B RF /B V )
P (v out ) = e , v out > 0 (3.7)
√2␲ √ RF /B V
2B
That is, the mean value is B RF /2B V times the mean value of the squared
envelope and the variance is B RF /2B V times the variance of the squared envelope.
3.5 Sensitivity 49
When the signal is present, the means and variance of the squared signal pulse
noise are given by [9–11]:
Mean of squared signal plus noise: (1 + SNR)␴ 2

Variance of squared signal plus noise: 4(1 + SNR)␴ 4
Using the same reasoning, the output distribution when the signal is present
is (again taking ␴ = 1)
[v out − (B RF /B V )(1 + SNR)]2
1 4(1 + 2 SNR)(B RF /2B V )
p (v out ) = e (3.8)
√2␲ √2B RF /B V √1 + 2 SNR
Note that for ␴ = 1, v out is expressed in units of ␴ 2.
For the case where B RF /B V is large (say, more than 100), (3.7) and (3.8) can
be used to compute the probability of detection and false alarm using the widely
tabulated normal distribution and its integral. For the case B RF /2B V = 1, the curves
of the radar detection probability can be used [7]. For intermediate values, the
published results of Emerson [12], Urkowitz [9], and Tsui [8] are valuable.
An example will serve to illustrate the use of (3.7) and (3.9). Suppose
B RF = 1 GHz and B V = 5 MHz. Then B RF /2B V = 100. Suppose the predetection
SNR is 0 dB. For a false alarm probability of 10−5, the threshold must be set at
the mean plus about four times the standard deviation. From (3.7), the mean is
200 and the standard deviation is 14.1. Hence, the threshold should be set at 256.4
and the probability of detection can now be found. Using (3.8), the mean of the
signal plus noise distribution is at about 400 and the standard deviation is about
34.6. The threshold at 256.4 is located about 4.1 standard deviations below the
mean. Therefore, the probability of detection will be about 0.99995. Using the
curves from Barton [7] provides the same results. If the number of samples integrated
is 100, the integration efficiency (␥ ) at PFA = 10−5 is approximately 0.75. The
processing gain is then (100)0.75 = 31.62, or 15 dB. Since the input SNR was
0 dB, the output SNR is 15 dB. Referring to the probability of detection curve for
a 15 dB SNR shows that the probability of detection should be greater than 0.9999
at PFA = 5 × 10−5. What is not clear from this approach is the close approximation
of the output distributions to the normal distribution and the rather easy calcula-
tions which result.
Next, consider what would be seen on an oscilloscope if a pulsed signal at
0 dB input SNR were applied to such a receiver, as shown in Figure 3.9(a). In the
normal case, the mean value of the noise would be blocked, possibly by using an
ac-coupled video amplifier. In that case, with no signal present, normally distributed
noise would be observed, with a standard deviation of 14.1 (measured in units of
␴ 2 ). Then, when the signal pulse appears, the mean value shifts upward from the
nominal baseline of noise by 200 (measured in units of ␴ 2 ). The observer might
adjust the ‘‘peak-to-peak’’ noise to occupy one division on the oscilloscope screen.
If ±2 standard deviations is chosen as the peak-to-peak noise swing, the noise
occupying one screen division represents 56.7 units of ␴ 2. The pulse height on the
screen will then be 200/56.7, or 3.5 oscilloscope divisions above the center of the
noise. The noise on the pulse has a standard deviation of 34.6 (units of ␴ 2 ). Its
Figure 3.9 (a) Video amplitude versus time (0 dB input SNR for B RF /2B V = 100). (b) Distribution
of noise alone and signal plus noise for 0 dB input SNR and B RF /2B V = 100.
peak-to-peak (±2 standard deviations) range will be 138.4/56.7, or 2.4 oscilloscope

divisions. Figure 3.9(b) is a sketch of the output signal plus noise and noise distri-
butions with the horizontal scale marked in the same units as the oscilloscope
calibration.
3.5 Sensitivity 51
3.5.6 Sensitivity and the Received Pulse Density

Once a threshold power level is established for a receiving system, signal pulses
that exceed this level are processed by the receiver while those below that level are
ignored. For a given emitter and receiver, the maximum range at which that emitter
can be received can be determined. For an airborne receiver, suppose that the
maximum range is less than the altitude at which the aircraft is flying; then none
of that emitter’s pulses will activate the receiver. As the aircraft altitude is decreased,
a region is defined on the Earth’s surface in which all emitters of that type can be
received, and the number of received pulses increases as the area of the region
increases. As the aircraft altitude continues to decrease, at some point, the distance
to the horizon will be less than the maximum range at which the emitter can be
received. As the altitude decreases further, the number of pulses which can be
received again decreases. Thus, the maximum number of received pulses occurs at
that altitude for which the distance to the horizon is equal to the maximum distance
at which the emitter can be received. The maximum range at which a signal can
be received is [13]
RE = 冋 PT G TE GE ␭ 2 1/2
(4␲ )2 S E L T L E
册
where:
R E = range to emitter from ELINT receiver

G TE = gain of transmitter in the direction of the ELINT receiver
GE = gain of the ELINT receiver in the direction of the emitter
␭ = wavelength
S E = signal power required at ELINT receiver
PT T = emitter transmit power
L T = transmitter losses
L E = ELINT receiver losses
Recall that the distance to the horizon is given by
d= √2R Earth h
Thus, the altitude (h) at which the maximum coverage is achieved is when
PT G TE GE ␭ 2
2R Earth h o =
(4␲ )2 S E L T L E
or
PT G TE GE ␭ 2
ho =
2R Earth (4␲ )2 S E L T L E
For sidelobe intercepts, G TE ∼ 1. (For new radar designs, the average sidelobe
level is more typically 0.1 relative to isotropic.) For an omni-directional ELINT
antenna E E ∼ 1. Assuming L T = L E = 1, and ␭ = 0.1m, the altitude at which the
maximum number of pulses may be received as a function of receiver sensitivity
is shown in Figure 3.10. To fly at 10,000m and receive sidelobe signals from a
1-kW transmitter at the horizon requires a sensitivity of −83 dBm. (Sensitivity in
this case means that power level which permits the receiver to perform those
functions required of it.)
3.6 The Ultimate Limits to ELINT Parameter Measurements
In a way, threshold detection only determines the presence of a signal. To provide

useful intelligence, signal parameters must be measured. Often, just as in radar,
threshold detection provides more information than mere existence of a signal.
With varying degrees of precision, the threshold crossing gives some information
about the following:
Figure 3.10 Optimum altitude as a function of receiver sensitivity.

1. When the signal begins and ends;

2. The minimum signal amplitude;
3. The bearing angle, if a directional antenna is used;
4. The carrier frequency—at least that it is in the band covered by the receiver
at the time the signal occurred.
As Woodward [14] points out, radar systems cannot divorce simple target
existence information from target location information. The radar receiver works
best when separate range and velocity resolution cells are provided and when a
directional antenna is used. The existence threshold is then the same as the resolution
threshold.
An ELINT receiver has somewhat similar properties. A receiver that sought to
intercept only one signal in an otherwise empty environment might find excess
resolution of time, frequency, angle, and amplitude to be a burden. In practice, in
an environment containing many signals, the interceptor needs parameter resolution
to make sense of the energy being received.
In signal reception, the word ‘‘information’’ has been given a precise meaning
by information theory. In particular, if the initial uncertainty about some signal is
U 1 , and after some energy has been received the uncertainty is U 2 < U 1 , we say
that an amount of information, I, has been gained, where
I ≡ log 2 (U 1 /U 2 ) bits (3.9)
A consequence of information theory is that to transfer I bits of information

from one place to another requires at the least an energy exchange of [14]
N0 I
E min = (3.10)
log 2 e
where:
E min = minimum energy required at receiver

N 0 = kT = noise power density at receiver input
I = information gain (bits)
Of course, this minimum energy can be achieved only in when transmission

of the information is intentional and planned and the reception of the information
makes use of that knowledge (as in communication systems). Consider the idealized
situation where the purpose of the radar’s transmission is to transmit information
about its signal to the ELINT station. For example, instead of measuring the carrier
frequency of an intercepted radar pulse, suppose the radar transmitted a message
that said, in effect, ‘‘My carrier frequency is now 3167.4285 MHz.’’ According to
information theory, information could be received at the ELINT site with transmis-
sion of the minimum energy by the radar. It is interesting to think about this
admittedly unrealistic situation and to explore the amount of additional energy
that is needed to make similar determinations of signal parameters by the traditional
methods of radar and ELINT.
The fact that the energy needed at the receiver is determined by the final
measurement uncertainty, U 2 , is not surprising. However, the required energy
also depends on the initial parameter uncertainty, U 1 . Consider carrier frequency
measurement. If the initial uncertainty is large, a wideband receiving system is
needed. Even if a channelized receiver is used, the large number of channels makes
the system more susceptible to false alarms and the threshold must be raised in all
channels to avoid being distracted from the channel with the real signal by a noise
threshold crossing in any one of the other channels. Similarly, if the signal can
occur at any time over a long period, the threshold must be raised to provide the
very low false alarm probability needed to avoid a false threshold crossing during
the time we are searching for the real signal. Thus, the initial uncertainty also
determines the signal energy needed at the receiver.
Notice that the energy is the important parameter. For one radar pulse, the
energy is the product of the signal power and the pulse duration. Should this energy
be too small to achieve the needed resolution, then more than one pulse may be
needed and the receiver must provide a means for adding the received energy in a
sensible manner.
A most interesting aspect of the information theoretic view of reception is that
true resolution can be achieved only by destroying part of the received information.
Consider an ideal recording of the signal as applied at the receiver input (or its
equivalent after frequency translation). While the recording preserves the signal,
we remain ignorant of its parameters. The recording contains all the information
which was received about both the signal and the noise—in fact, the received signal
could be exactly reproduced upon playback (and possibly frequency translation)
and the situation would be just as it was at the time the signal was originally
recorded. However, nothing would have been learned. This characteristic applies
to all reversible signal processing operations. Information can be gained only from
irreversible operations. Since the original signal cannot be recreated after an irrevers-
ible operation, some of the original information has been destroyed or lost. The
important thing is to destroy the unwanted information (i.e., that which describes
the noise) and retain the signal descriptors or interest. For example, an envelope
detector destroys the phase information. A threshold detector destroys most of the
amplitude information (except for 1 bit), but the resulting signal provides a good
time of arrival estimate.
A radar measurement can provide a fractional error of [15, p. 401]
␦M L
= (3.11)
M √2E /N 0
where:
␦ M = error in measurement M
M = resolution in measured coordinate
E = received signal energy
N 0 = thermal noise power density at the receiver input
L = a constant on the order of unity which is dependent on the
measurement parameter and signal processing used
Consider a carrier frequency measurement. In radar, the transmitted frequency

is known, so the initial uncertainty is the Doppler shift, which can easily be deter-
mined to within the reciprocal of the pulse duration, ␶ . The final frequency uncer-
tainty is [15, p. 407]:
√3
␦F = (Rectangular pulse) (3.12)
␲␶ √2E /N 0
The performance projected by (3.11) should certainly be poorer than that

predicted by the information theoretic approach. The error in the measurement
(␦M ) is associated with the final uncertainty U 2 . The initial uncertainty, U 1 , is
certainly no smaller than ␦M (and probably much larger for an ELINT receiver).
Solving (3.11) for E and dividing by E min from (3.10) yields
2 2
E U2 U2
= 2 2 log 2 (U 1 /U 2 ) ≤ 2 log 2 (U 1 /U 2 ) (3.13)
E min ␶ M U1
The term on the right has the form x −2 log 2 x. The minimum of this function
occurs at x = e −1/2, and the value at this point is 0.265, which indeed is less than
1. Thus, the typical radar and ELINT measurements are at least 5.7 dB [5.7 dB =
10 log (0.265)] worse than the theoretical limits established by information theory—
and frequently much worse, as shown in Figure 3.11.
For practical threshold circuits, discriminators, and direction finders, the mea-
surement error is proportional to the reciprocal of the square root of the signal-
to-noise energy ratio. On a single-pulse measurement, E = S␶ and N 0 = N /B, where
␶ is the pulse duration and B is the noise bandwidth. Thus, the measurement error
on a single rectangular pulse is
␦M 1
= (3.14)
M √B␶ √2 SNR
Figure 3.11 Radar compared to information theory limits.

where M is the initial measurement resolution (bandwidth) which may be approxi-

mately the reciprocal of the pulse width.
3.7 ECM and ELINT Receivers
In this discussion of ELINT receiver performance there has been an unstated

assumption that these receivers are not subject to either noise jamming or deception.
Since ELINT, as discussed here, is a peacetime activity,5 jamming is rarely a problem.
Even during a conflict, it is unlikely that noise jamming would be used to hide a
radar from an ELINT receiver since that would generally also interfere with the
radar’s function. Jamming initiated by the forces that were also using ELINT would
presumably be done consciously and with knowledge of the consequences.
Spoofing (or deception) of ELINT receivers is another matter. Certainly, during
peacetime, spoofing could take place at any of several levels. Radars could simply
be operated with restricted sets of parameters that are changed during wartime.
Radar-like transmitters could emit large numbers of pulses across a wide band of
frequencies, making the ELINT process much more difficult and costly. Such count-
ers to ELINT have simply not been judged cost-effective in times past. The radar
designers have concentrated on making the radar operate to meet certain perfor-
mance specifications. Spending resources on spoofing ELINT receivers has not been
worthwhile. In tactical situations, such may not be the case. Clearly, there are
benefits to deceiving a radar warning receiver, and such activities may become a
problem.
The peacetime, strategic, manpower-intensive character of ELINT has meant
that interference and deception against it are rare. While one can speculate about
the existence of these activities in the future, it is well to remember that until now
ELINT has always reacted to the directions established by the radar community.
To reverse that and have radar designs be a reaction to ELINT collection activities
seems unlikely; however, the efforts to design and deploy LPI radar indicate that
some activity along these lines is taking place.
3.8 Crystal Video Receivers
The simplest ELINT system is the crystal video type consisting of an antenna, a
detector, and a video amplifier, as shown in Figure 3.12(a). Adding a low-noise
RF preamplifier, as in Figure 3.12(b), provides additional sensitivity. In practical
receivers, there is always some bandpass filtering action ahead of the detector to
restrict the receiver output to some band of interest. The band-limiting function
may, in some simple receivers, be that provided by the antenna alone. In the low
sensitivity receiver, the sensitivity is determined by the detector characteristics and
the noise generated in the video amplifier. In the high sensitivity receiver, it is
determined by the noise figure of the RF preamplifier. In both cases, the RF and
video bandwidths place fundamental limits on the receiver’s sensitivity. In some
5. It has been said that ELINT presumes peace on Earth but not goodwill towards men.
Figure 3.12 Crystal video receivers: (a) low sensitivity receiver; and (b) high sensitivity receiver.
applications, high sensitivity may be undesirable—as for a radar warning applica-

tion in which an aircraft is to be warned of being tracked at relatively short range
by radar associated with a surface-to-air missile system.
The sensitivity of a crystal video receiver has been studied extensively [16].
The sensitivity can be characterized as either gain-limited (when it can be improved
by increasing the gain of the RF preamplifier) or noise-limited (when improvement
is possible only by reducing the noise figure of the RF preamplifier or the receiver
bandwidths). The discussion of sensitivity above was based on a noise-limited
receiver in which the detector characteristics are inconsequential. In this case, the
sensitivity is determined solely by the RF and video bandwidths and the receiver
noise figure. For ELINT applications this is generally the case. In some applications,
however, small size and low cost are important, and the sensitivity of the configura-
tion in Figure 3.12(a) [or Figure 3.12(b) with only a small amount of gain ahead
of the detector] may be used. In this case, the sensitivity varies in proportion to
the square root of the video bandwidth, the proportionality constant being a
characteristic of the detector. The tangential sensitivity of the crystal video receiver
is [16]
TSS = kTF 6.31B V + 2.5 √2B RF B V − B V + AB V /(GF ) 

2 2
(3.15)
where:
TSS = tangential sensitivity (W)
k = Boltzman’s constant 1.38 × 10−23 Ws/K

T = room temperature (290K)
F = noise figure of amplifiers ahead of detector
B V = video bandwidth (Hz)
B RF = RF bandwidth (Hz) ≥ 2B V
A = diode parameter [see (3.16)]
G = gain of amplifiers ahead of the detector
The diode parameter, A, is given by
4FV 4FV R
A+ 2
= (3.16)
kTM kTC 2
where:
FV = video amplifier noise figure

M = detector figure of merit = C /√R
R = dynamic impedance of detector (⍀)

C = detector sensitivity (V/W)
For the gain-limited crystal video receiver, the term containing A is very large,
and (3.15) reduces to
kT
TSS GL ≅ 2.5 √AB V (gain-limited) (3.17)
G
If there is no preamplifier, G = 1, and we have
TSS GL ≅ 2.5kT √AB V (no preamplifier) (3.18)
For TSS in dBm and B V in megahertz, (3.18) becomes
TSS No Preamp (dBm) ≅ −110 + 5 log A + 5 log B V
If A = 1014 and B V = 1 MHz, the tangential sensitivity of the receiver is

−40 dBm. In an actual receiver, there will be an RF filter ahead of the detector,
and the insertion loss of the filter is equivalent to a gain (G) of less than one (and
a noise figure, F, equal to the reciprocal of the insertion loss). This must be taken
into account using (3.15). The example used in [15] has a filter loss of 2.1 dB,
providing a sensitivity of −38 dBm. These values are typical of a crystal video
receiver with no preamplifier and a 1-MHz postdetection bandwidth.
Suppose a radar with 100-kW peak transmit power at 3 GHz and 30-dB
antenna gain is of interest to the crystal video receiver. The one-way range equation
shows that a −32-dBm signal from the radar main beam will be available at the
output of an omni-directional ELINT antenna at a distance 100 km away from
the radar. Thus, a simple crystal video receiver is adequate to detect such a radar
and could perform in the role of radar warning receiver.
As RF gain (G) is added ahead of the detector, the noise-limited condition is
reached and the term containing A can be neglected in (3.15). The tangential
sensitivity then becomes
TSS NL ≅ kTF 冠6.31B V + 2.5 √2B RF B V − B V 冡 (noise-limited)

2
(3.19)
If B RF Ⰷ 2B V in megahertz, (3.19) becomes
TSS NL ≅ kTF 冠2.5 √2B RF B V 冡 (3.20)
This is essentially the expression used above for effective noise bandwidth. In
(3.20) the factor 2.5 (4 dB) is the predetection SNR empirically determined to
produce the TSS condition. If the receiver bandwidth is taken as B RF , then the
receiver can be said to exhibit a processing gain of [B RF /2B V ], which reduces the
signal level needed to produce the TSS condition compared to a receiver in which
2B V = B RF . Based on the examples from [16], a receiver with B RF = 2,000 MHz
and B V = 1 MHz has the TSS shown in Table 3.7 under several conditions for
A = 9.55 × 1013.
It should be noted that the SNR at the tangential condition is not sufficient
for most ELINT measurements; thus, practical considerations raise the required
signal level about 10 dB above the TSS value.
It should also be noted that the sensitivity varies as √B V in both the gain-
limited case when √AB V /G 2 F 2 is the predominant term in (3.15), and in the noise-
limited case with B RF Ⰷ B V when √B RF B V is the predominant term in (3.15). It
is clear that the use of pulse-to-pulse frequency hopping to force reduced sensitivity
on intercept receivers has no effect on a gain-limited crystal video receiver with an
RF bandwidth larger than the likely agility band of the radar (only the postdetection
or video bandwidth affects its sensitivity).
3.8.1 Crystal Video Applications

The simple crystal video receiver has very poor frequency selectivity—it responds
to the total power entering the RF filter. Therefore, applications center on those
Table 3.7 Effect of Front-End Gain on Tangential Sensitivity of a Crystal Video Receiver
Gain-Limited Intermediate Noise-Limited
Gain ahead of detector (dB) −2.1 10.0 20.0 30.0 40.0 > 50.0
Noise figure ahead of
detector (dB) 2.1 8.6 8.6 8.6 8.6 8.6
TSS (dB) −38.0 −50.2 −60.2 70.2 −79.2 −83.4
in which the envelope of the RF energy provides sufficient information. For pulsed
signals, if the number of pulses present is not too large, the envelope can be quite
informative since it allows examination of the radar PRI. If the sensitivity is low,
the number of received pulses is reduced. Also, directional antennas can be used
to restrict the angular field of view and hence reduce the number of pulse trains
present in any one detection channel. Since a study of the pulse time and amplitude
history gives the PRI and pulse duration as well as the scanning characteristics
of the radar, it is possible to identify radar types and functions with surprising
effectiveness, provided that the interleaving of the pulse trains is not so great that
a large fraction of the pulses from any radar are overlapped by other pulses.
Excessive overlap can occur either by having a large number (more than 5 to 10)
of similar pulse trains present at once, or by having a very short PRI (high PRF)
signal present along with a very long PRI (low RPF) signal.
While most ELINT applications would not be well served by a crystal video
receiver alone, such receivers may be used in conjunction with other receivers to
provide general monitoring or ‘‘tip-off’’ capabilities to bring other receivers into
action.
Clearly, there are limits to the number of signals that can be successfully
distinguished using PRI alone. Once these limits are reached, additional processing,
no matter how sophisticated, can provide only a minor improvement in capability.
For radar warning receiver applications, the crystal video receiver’s problem
of too many signals is often approached through the addition of multiple channels
(e.g., four to eight) to provide angle of arrival information. In other cases, the
system design may make use of either an IFM unit (which is basically a discriminator
covering the full band) or a narrowband tunable receiver whose selectivity in
frequency can resolve complex situations.
3.8.2 Postdetection Signal Recording and Sorting

The crystal video ELINT receiver output can be preserved by digitizing the video
signal. In this case video bandwidths on the order of 5 to 10 MHz are required.
Should there be more than one signal present (which is quite likely in view of the
wide band covered by most crystal video receivers), the amplitude and time of
arrival information allows separating the signals (if there are not too many) using
the simple techniques of time-gating based on predetermined PRI values coupled
with amplitude tests designed to prevent neighboring pulses with widely differing
amplitude from being assigned to the same pulse train. The PRI-gating requires
wider and wider gates into which the pulse must fall if the signal is intentionally
PRI-modulated in some way (e.g., random jitter, periodic stagger, dwell and switch,
pulsed Doppler types). This is discussed in Chapters 12 and 13. Even if large PRI
variations are present, the amplitude variations due to the radar’s antenna scanning
may allow the skillful signal analyst to visually separate the interleaved signals.
3.8.3 CV System Design Considerations

The system designer’s choice of RF amplifier obviously determines the sensitivity
of the crystal video receiver, and the sensitivity required is a function of the system
application. The low end of the dynamic range is determined by the sensitivity.
The high end of the dynamic range depends on the level at which the RF amplifier,
the detector, and the video amplifier saturate. For a single detector, the total receiver
dynamic range can be no greater than the dynamic range of the detector. This is
obvious in a gain-limited (no RF amplifier) system. If noise-limited operation is
chosen, then the noise at the detector output is set by the RF amplifier. The dynamic
range will be somewhat reduced from that of the detector alone, since the maximum
detector output remains the same, but the noise level with no signal present has
increased. Increased total dynamic range can be obtained by using several channels
with differing amounts of RF gain and summing the outputs (with some loss of
sensitivity).
The video amplifier is often ac-coupled to reduce the effects of low frequency
noise (which increases as the reciprocal of the frequency). However, this means
that CW signals will not be detectable and may desensitize the receiver (e.g., by
saturating the receiver, a single strong CW signal would effectively mask all weaker
pulsed signals across the entire band).
Crystal video ELINT receivers often include an RF switch [Figure 3.13(a)] to
chop the input as selected by the operator. This allows any CW signals to be
detected at the receiver output. A sweeping bandpass filter can achieve the same
result (that is, chop a CW signal) with the added advantage of allowing the frequency
Figure 3.13 CW detection schemes: (a) RF switch to chop the input; and (b) swept bandpass filter
as a chopper.
of the CW signal to be determined. A notch filter adds the capability of reducing

the CW signal at the detector to possibly allow pulsed signals to pass to the output.
The addition of a tunable narrowband filter, as in Figure 3.13(a), creates a receiver
of the tuned radio frequency (TRF) type which was extensively used in the early
days of radio to provide selectivity. Clearly, a pulsed signal’s RF frequency also
can be measured with the aid of the filter.
Another system problem associated with crystal video receivers is that of simul-
taneous signals. Since only the total RF energy present is measured, difficult prob-
lems arise if two (or more) pulses are present at the same time. For example, in
pulse-gating signal sorting, pulses may be dropped from one or more pulse trains.
Likewise, improper pulse amplitudes may be measured. One way to detect simulta-
neous signals is to make use of the difference frequency generated by the detector.
Consider the system shown in Figure 3.14. The RF band covered is from RF LOW
to RF HIGH . Any two signals simultaneously present will produce signals at the
detector output whose frequency is the difference between the two input frequencies
and which must be between zero and (RF HIGH − RF LOW ). The normal video
amplifier looks for the signal envelope and passes frequencies from almost zero to
B v . The upper amplifier covers the frequency band from somewhat above B V to
(RF HIGH − RF LOW ). The detector then measures the envelope of the energy at the
difference frequency at its output and can thereby warn of the presence of multiple
signals.
The largest difference frequency considered must be lower than the lower band
edge, RF LOW , otherwise signals near RF LOW will pass directly through the multiple-
signal channel. In addition, one of the multiple signals must be strong enough to
produce a difference frequency component above the noise. Thus, the multiple-
signal channel will generally be unable to indicate the presence of two weak signals,
if both are near the threshold.
Crystal video systems sometimes use amplitude comparison of the signals from
multiple antennas for angle of arrival measurement. In these systems, the antenna
patterns must be carefully matched over the band covered. Since there is no fre-
Figure 3.14 Multiple signal detector.

quency information, it is not possible to correct for antenna-to-antenna gain varia-

tions as a function of RF. Even if the antennas match perfectly, it is clear that the
receiver channels must also match well to obtain good angle-of-arrival performance.
This means matching the gain versus frequency characteristics of the receiver chan-
nels. One way around errors of this type is to rapidly switch a single receiver
channel from one antenna to another. This reduces receiver sensitivity since the
switching must be done within the time of the shortest pulse of interest. This
increases the video bandwidth (B v ) required by an order of magnitude. Another
approach is to switch antennas between pulse arrivals. However, this means that
the angle-of-arrival determination cannot be done on a single pulse basis.
In summary, crystal video receivers are useful in low duty cycle pulsed-signal
environments. Intercept receiving systems often use these receivers in conjunction
with other receivers (e.g., IFMs or superheterodynes). The advantages, disadvan-
tages, and typical characteristics are summarized in Table 3.8.
3.9 Superheterodyne Receivers
The superheterodyne receiver (referred to hereafter as a superhet) is the most widely

used receiver design for nearly all uses; ELINT applications included. A basic block
diagram is shown in Figure 3.15. The idea is to use a local oscillator to convert
the incoming signal to a fixed intermediate frequency by the mixing process (hetero-
dyning). Then the IF amplifier need operate at only one frequency and its character-
istics can be precisely controlled (gain, center frequency, bandwidth, band edge
roll-off, group delay). The bandwidths for typical narrowband ELINT receivers
are B RF ∼ B IF ∼ 20 MHz and B V ∼ 10 MHz. This allows for good pulse fidelity
for normal radar pulses (pulse width ∼1 ␮ s). Such receivers were the mainstays of
ELINT operations during World War II. More recent designs have additional
capabilities. Rapidly sweeping the local oscillator in frequency produces an ampli-
tude versus frequency display, which is created by displaying the amplitude detector
output versus the tuning voltage, thus creating an RF spectrum analyzer. Very
rapid sweeping creates a rapid sweep superhet, described in Chapter 9, which is
useful for detecting high duty cycle, low power signals in the presence of low duty
cycle interfering signals. Still more rapid sweeping is the basis of the microscan
and metascan compressive receivers described later. Using very wide receiver band-
widths (e.g., 500 MHz to 2 GHz), and using a microwave IF (e.g., 2 GHz) creates
Table 3.8 Crystal Video Advantages and Disadvantages

Advantages Disadvantages
Low cost No frequency information
Mature technology broadband Multiple signals cause errors
Simple
Typical Characteristics
Sensitivity: −70 dBm (noise-limited) to −30 dBm (gain-limited)
Dynamic range: 50 dB single channel; > 60 dB multiple channel
RF bandwidth: Multiple octaves
Video bandwidth: Dependent on application, 10 MHz typical
Figure 3.15 Superhet block diagram.
a frequency conversion system which can serve as a front-end to an IFM receiver

(IFMs are discussed later). Combining microsecond tuning capabilities with com-
puter control creates a receiver that can tune to a given frequency just in time to
collect a given pulse with a known PRI. This kind of ‘‘time-shared’’ operation
allows a single receiver to intercept many pulses from a number of different emitters
in spite of its single narrowband channel design.
The wide variety of superhet designs shows that there are still many ways to
make use of this old but powerful concept. As shown in Appendix C, a slow sweep,
narrowband superhet even has some capability against a random frequency hopped
radar signal if it is properly used.
Historical Note. The superheterodyne was invented by Edwin H. Armstrong.

RCA owned the patent and produced superhet radios for sale to the public in 1924
[17]. These were very popular because they allowed tuning via a single knob.
To get similar performance, other manufacturers had multiple-tuned stages that
required separate adjustment (this was the tuned radio frequency receiver, or TRF).
The superhet required only one RF filter ahead of the mixer and a tuning control
for the local oscillator. The remainder of the stages could be fixed-tuned at the IF
to provide very good selectivity and sensitivity compared to the competing TRF
designs, which had up to five tuned stages at RF. The TRF manufacturers tried a
variety of mechanical linkages using complex cams to attempt to tune all of the
stages with a single knob. The advantages of the superhet design nearly forced
RCA’s competitors out of the receiver field until, threatened with antitrust action,
RCA granted licenses to the others in 1930. The ELINT receivers of the 1940s
were superhets with gang-tuning of the RF preselector and the local oscillator.
Complex mechanical adjustments were necessary to maintain tracking between the
two circuits, much like the earlier generation of broadcast receivers.
3.9.1 Superhet Performance

The sensitivity of a narrowband superhet is the power required at the input to
produce a given SNR at the IF amplifier output. It is given by
RSL = (SNR)(NF)(kTB IF ) (3.21)
where:
RSL = required signal level

SNR = signal-to-noise power ratio required at IF to perform specified functions
NF = noise figure
k = Boltzmann’s constant
T = equivalent noise temperature at the input (K)
B IF = IF bandwidth (Hz)
It is common to express this in decibels referred to a milliwatt and to use a

temperature of 290K, in which case (3.21) becomes
RSL (dBm) = SNR (dB) + NF (dB) − 114 dBm/MHz + 10 log B IF (MHz)
A wideband superhet may exhibit processing gain. It would be rare to make

the video bandwidth much wider than 10 to 20 MHz, whereas the RF and IF
bandwidths may be several hundred megahertz. In this case the SNR to be used
in (3.21) is the output SNR, which is the IF SNR increased by the processing gain
of
PG ≅ √B IF /2B V
The required signal level at the input to the wideband receiver, therefore, is
increased in proportion to the square root of the factor by which the IF bandwidth
is increased beyond 2B V . For a 14-dB IF SNR, a receiver having B IF = 20 MHz
and B V = 10 MHz has a required signal level of
RSL (dBm) = −114 dBm/MHz + 10 log 20 MHz + 14 dB

= −87 dBm
Increasing the RF bandwidth to 200 MHz increases the required signal level
to about −82 dBm since
√200/20 = √10, or 5 dB
The dynamic range of the superhet is set by the saturation of the RF preamplifier
at the high power end and by its output noise level at the low end. For maximum
dynamic range and nearly the best sensitivity, the noise level at the detector output
must increase slightly when the RF amplifier is placed in the circuit.
The probability of intercept depends on the tuning speed and the RF and IF
bandwidth. Generally, for CW or high duty cycle signals, the probability of intercept
is quite high. For pulsed signals, the dwell time at a given frequency should be
longer than several PRIs in the conventional design approach using slow-speed
tuning or sweeping. Wide bandwidth superhets with programmed scan strategies
can have fairly high probability of intercept even for low duty cycle signals. Fre-
quency hopped signals are a special case. Generally low probability of intercept is
expected; however, proper use of the superhet still provides useful data in some
situations, as shown in Appendix C. For low duty cycle scanning emitters, the high
sensitivity of a narrowband superhet often allows detection of the radar’s sidelobes,
thus enhancing the probability of intercept. The superhet has good performance
in many applications, and it provides this performance with low cost and modest
size.
The ability to measure frequency depends on the IF filter bandwidth and
whether a discriminator or IF spectrum analyzer is available to aid in tuning. The
resolution of signals in frequency depends on the local oscillator tuning steps (if
any) and on the IF filter skirts.
For applications in which the coherence or RF stability of the incoming signal
is to be measured, the stability of all of the local oscillators in the receiver must
be better than the stability of the signal being received.
The conversion of the incoming signal to an intermediate frequency can be
accomplished by heterodyning (mixing) the signal with a local oscillator whose
frequency is either above or below that of the signal; that is,
IF = Signal Frequency − Local Oscillator Frequency
or
IF = Local Oscillator Frequency − Signal Frequency
If the local oscillator frequency is below that of the signal, the signal spectrum
is shifted directly to be centered at the IF. If the local oscillator frequency is above
that of the signal, the signal spectrum is inverted and shifted to be centered at the
IF. Spectrum inversion may be undesired. If a second frequency conversion takes
place, a second spectrum inversion may be used to recreate the original signal
spectrum.
Dual (or triple, or multiple) conversion superhets are useful to achieve transla-
tion to a low IF, particularly if a narrow IF bandwidth is necessary. The ratio of
the 3-dB IF bandwidth to the IF center frequency is proportional to the ‘‘Q’’ of
the IF filter, hence very narrow band filters are more easily constructed at a low
IF. On the other hand, tuning a local oscillator at a microwave frequency with
enough precision to center the incoming signal in a narrow IF may be quite difficult.
For this reason, a dual conversion receiver with the first IF relatively high (e.g.,
160 MHz) and with a wide bandwidth (e.g., 20 MHz) may be used to initially
locate a signal. Conversion to a second IF (e.g., 21.4 MHz) allows for narrowband
filtering and demodulation for those signals for which it is appropriate. For step-
tuned local oscillators, this arrangement allows coarse frequency steps by the first
LO, and fine frequency steps by the second LO.
In some ELINT applications, the signal is digitized at IF before detection
(predetection digitizing). In this case, it may be necessary to translate a relatively
wide spectrum down to a low center frequency. In translating to a low IF, it is
important to make sure that filtering at each step prevents a part of the spectrum
from being translated to negative frequencies. This portion folds back into the
positive frequency region and may cause distortion, as shown in Figure 3.16. The
ideal filter indicated in Figure 3.16(b) is impossible to realize. The actual filter
skirts must be considered in determining how to place the IF to maintain any folded
components at sufficiently low amplitude to be less than the quantizing noise of
the A/D converter.
3.9.2 Sweeping Superhet Receivers

The concept of a receiver which sequentially tunes through the band to search for
signals has been important for ELINT applications since the beginning. The early
receivers used motorized tuning in which electric motors were used to repeatedly
Figure 3.16 Spectrum fold-over problem: (a) insufficient filtering prior to conversion to low IF;
and (b) filtering prior to final conversion to eliminate fold-over.
tune through the band when searching. Later, electronically tuned filters and synthe-
sized local oscillators allowed for much more rapid tuning. The natural question
to ask is how fast should the tuning occur for various applications and what limits
are imposed by the receiver bandwidth.
The analysis of the behavior of a sweeping superhet as the sweep speed increases
is accomplished most easily for the assumption of a linear frequency sweep and
an IF filter having a Gaussian amplitude and linear phase response [18, 19]. For
a CW signal, it is easy to see that for a slow sweep, the envelope of the IF
signal will trace out the IF bandpass forming a pulse-like signal whose duration is
determined by the filter bandwidth divided by the sweep speed of the local oscillator;
the sweep speed being the extent of the band swept divided by the time required
to sweep through it. As the sweep speed increases, the duration of the IF pulse
decreases until at some point the amplitude of the pulse is reduced and its duration
remains fixed. At very fast sweep speeds, the envelope of the IF pulse becomes
essentially the impulse response of the filter and it is basically unchanged by the
sweep speed. For the Gaussian IF bandpass, the amplitude of the envelope and
the frequency resolution as a function of the sweep speed is shown in Figure 3.17.
The amplitude response is relatively unaffected by the sweeping as long as
⌬F
< B2 (3.22)
T
where:
Figure 3.17 Sensitivity and resolution loss as a function of sweep speed. (From: [19].  1981
Horizon House, Inc. Reprinted with permission.)
⌬F = the extent of the band being swept (Hz)

T = the time to sweep the band (sec)
B = the IF 3-dB bandwidth (Hz)
For a step-tuned system, the dwell time at a frequency should be longer than
the reciprocal of the IF bandwidth. If steep filter skirts are needed and a Gaussian
IF filter is not suitable, then reducing the sweep speed to a fraction of B 2 may be
necessary.
For an ELINT application in which a 1-GHz band is to be searched using a
10-MHz IF filter bandwidth, the time to sweep the band should be longer than
109
= 10 ␮ s
(107)2
If step tuning were used, the dwell time at any frequency should be longer than
1/107 = 0.1 ␮ s. Thus, a step-tuned LO which increased its frequency in 10-MHz
steps could also tune across a 1-GHz band in 10 ␮ s.
ELINT receivers generally take much longer times to sweep across the band
since it is important to dwell at least several radar pulse repetition intervals at
each frequency. This improves the probability of intercept and also allows PRI
measurement to occur. (The PRI is an important parameter used to help decide
whether the signal is of interest.) At 10 ms per frequency, it would require about
1 second to search a 1-GHz band in 100 steps of 10 MHz each. As a result, the
‘‘speed limit’’ imposed by the IF bandwidth has generally not been an important
factor in the design of ELINT receivers. Of course, the receiver described in Chapter
9 for detection of high duty cycle signals while ignoring pulsed signals is an exception
to this rule. For that receiver, the sweeping should take place at close to the
speed limit. [In order to sweep even faster, it is necessary to use the microscan
(compressive) approach described later.]
It must be noted that as the sweep speed approaches the limit of (3.22), the
IF filter characteristics become an important part of the receiver design. Clearly,
if the impulse response of the IF filter does not die away in a time comparable to
the reciprocal of its bandwidth, a strong signal can obscure a weak signal nearby.
While traditional slow-scan receiver designs can choose IF bandpass characteristics
with steep skirts for good selectivity, as the sweep speed increases, it is necessary
to reduce the slope of the filter skirts to approximate a Gaussian shape. The
Gaussian IF bandpass that provides the best resolution is [18]
B o = 0.66 √⌬F /T (3.23)
For this bandwidth, the resolution is 0.94 √⌬F /T and the amplitude of the
envelope at IF is reduced by a factor of 0.84 from the nonswept receiver. This
amounts to a sensitivity loss due to sweeping of about 1.5 dB. Returning to the
previous example, to sweep a 1-GHz band in 10 ␮ s, the optimum IF bandwidth
is 6.6 MHz. This can resolve signals of equal power separated by about 9.33 MHz.
For the optimum resolution case, if a wider IF bandwidth is used, the frequency
resolution is degraded by virtue of the larger bandwidth. If a narrower IF bandwidth
is used, the frequency is also degraded because at this sweep rate the impulse
response of the filter is important. Making the bandwidth narrower increases the
duration of the ringing time of the filter. As it rapidly sweeps through a CW signal,
the output, in the time domain, approaches the filter impulse response as the
bandwidth is reduced. The ringing reduces the ability to resolve signals in frequency.
The optimum IF bandwidth balances these two opposing forces in the way that
gives the best frequency resolution.
3.9.3 Tuning Considerations
Successful use of a narrowband, slowly tuned superhet as an intercept receiver

requires attention to several important details. One is the tuning process itself. As
the operator tunes across the band and encounters a signal, proper tuning to center
the signal spectrum within the receiver’s passband is essential for proper signal
analysis and recording.
Tuning aids in the form of a discriminator and spectrum analyzer (pan display)
at IF are very helpful in tuning the signal to the center of the IF filter. Of course,
the task is much easier if the signal spectrum has a narrower bandwidth than the
receiver’s IF bandpass. In a well-designed receiver, there may be a choice of pre-
and postdetection filters to help tailor the receiver to the signal characteristics. In
most ELINT applications, the IF bandwidth should be a multiple on the order of
1 to 10 times the reciprocal of the pulse duration. If there is intrapulse modulation,
then the IF (predetection) bandwidth should be somewhat larger than the peak-
to-peak frequency deviation on pulse or the reciprocal of the phase modulation
‘‘chip’’ duration.
It is also important to consider the problems of image responses in tuning the
receiver. A very strong signal may enter the receiver’s image response. The apparent
RF will then be in error by twice the IF frequency, even when carefully tuned to
the center frequency. This condition can be detected using either the discriminator
or pan display, since tuning the receiver LO causes the frequency within the IF
passband to change in the opposite direction from that expected. (Image rejection
is provided by a tuned preselector, which ‘‘tracks’’ the frequency of the LO or by
an image rejection mixer.) In some systems, there is no image rejection scheme.
Both responses are processed as valid signals and the proper RF is determined by
slightly changing, (dithering) the LO, and noting the resulting phase relationship
of the frequency changes on the IF signal.
Inaccurate tuning of a conventional constant RF pulsed signal can severely
distort the pulse envelope as observed at the AM detector output. To understand
why, recall that differentiation in the time domain corresponds to multiplication
of the frequency spectrum by 2␲ f ; that is,
d n [s(t)]
= (2␲ if )n S ( f ) (3.24)
dt n
Therefore, if the signal is tuned so that the center of its spectrum is located on
the skirts of the IF filter, differentiation occurs and the output of the envelope
detector will develop ‘‘rabbit ears’’ as sketched in Figure 3.18.
Such rabbit ears may also occur in some systems if the received signal is so
strong that severe saturation occurs. For example, receivers with low noise traveling
wave tubes as preamplifiers have the characteristic that the output of the tube
actually decreases when it is strongly driven. This causes the rise and fall of the
pulse to appear at the output at higher amplitudes than the center part of the pulse.
Such effects are more serious in automated receiving systems since their thresh-
old circuits may erroneously report a group of two short duration pulses in place
of one conventional pulse.
Accurate tuning is especially difficult when fine resolution spectrum analysis
of the portion of the spectrum near the centerline is necessary, as in some coherence
measurements and measurement of very small RF modulations. (The later can be
found in some pulsed Doppler radar systems as part of a range ambiguity resolution
scheme. RF variations on the order of 0.01% may be used in some cases.) Discerning
such frequency variations may require conversion to a low frequency (with its
attendant aliasing or spectrum fold over problems) followed by high-resolution
spectral analysis.
For a scanning signal, one may be limited to using the center portion of the
main beam, since the sidelobes may have an insufficient SNR for such fine-grain
analysis (recall that the error due to noise is proportional to the square root of the
reciprocal of the SNR). As a result, only a few pulses may be available for measure-
ment of the carrier frequency. The error in the frequency measurement is also
proportional to the reciprocal of the square root of the number of pulses available.
In such situations, it is best to test the receiver and analysis system using a signal
with known properties. In this way, the absolute limits to the ability to measure
Figure 3.18 Effect of inaccurate tuning on pulse envelope: (a) center-tuned; and (b) inaccurately
tuned.
the RF characteristics of short bursts of pulses can be clearly established before it

is necessary to deal with the fleeting signals of unknown character so often encoun-
tered in the world of ELINT.
Extremely accurate tuning of an oscillator to the carrier frequency of a pulsed
signal may arise if the frequency stability (coherence) or precise average value of
the RF is to be determined. The basic dependence of frequency measurement
accuracy was given above as
√3 √3
␴f = = (3.25)
␲␶ √2E /N 0 ␲␶ √2 SNR √B␶
where:
␴ f = rms frequency error due to noise

␶ = duration of signal or measurement time
E = signal energy = S␶
N 0 = noise spectral density = kT
B = receiver noise bandwidth
SNR = signal-to-noise ratio
Note that in many situations, B␶ may be approximately 1. For a single pulse

measurement, ␶ is the pulse duration, and for a 20-dB SNR, and a 1-␮ s pulse
duration and a 1-MHz bandwidth, this relationship gives an rms frequency error
due to noise of 100 kHz. If the carrier frequency error was required to be 100 Hz,
simple postmeasurement averaging of the single pulse values would require 106
pulses! In most ELINT applications, the signal can be stored for later analysis.
Then the frequency estimation can take place off-line, using multiple pulses and a
variety of spectrum analysis techniques. If coherent processing can be performed
on multiple pulses, the rms frequency error can be reduced in proportion to the
number of pulses instead of the square root of the number of pulses. If coherent
processing can be used, only 1,000 pulses would need to be processed in the
previous example—but note that the signal must be coherent from pulse to pulse
for 1,000 received pulses.
3.9.4 Other Heterodyne Receivers

A zero-frequency IF is sometimes used. In this type of receiver, the local oscillator
is tuned exactly (or nearly so) to the center frequency of the signal being received.
Unlike the envelope detector, the phase information is retained in the baseband
signal when the zero-IF technique is used.
Of course, precisely zero-beating a pulsed signal is quite difficult. Manually
tuned zero-IF receivers often produce a nonzero IF of hundreds of kilohertz. One
receiver configuration is shown in Figure 3.19. The in-phase and quadrature outputs
Figure 3.19 Zero IF receiver.
are particularly convenient for predetection digitizing. The envelope can be recon-
structed from
√I
2
A(t) = (t) + Q 2(t) (3.26)
and the instantaneous phase can be reconstructed from
Q (t) Q (t)
␾ (t) = tan−1 = sin−1 (3.27)
I (t) A (t)
A kind of ‘‘wide open’’ receiver that coverts any input signal to a fixed IF
signal (sometimes called a homodyne) is shown in Figure 3.20 [20]. Any frequency
modulation on the input signal will be removed in the mixing processes. The IF
signal, therefore, contains the frequency instabilities of the local oscillator. Note,
however, that any differential delay through the two channels means that the input
signal at one time is being mixed with the input signal (offset by the IF) from a
slightly different time. This effect can be used to ascertain the carrier frequency by
measuring the phase of the output signal with respect to the internal LO phase
much like an IFM (see below). This receiver has few applications since the broad-
band mixing processes are difficult to implement and the functions of the receiver
can be better performed by crystal video or IFM systems. With low noise preamp-
lifiers, there is no sensitivity advantage to the homodyne over a crystal video
receiver.
Figure 3.20 Wide-open homodyne receiver.
3.10 Instantaneous Frequency Measurement Receivers
The need to measure the frequency on each pulse motivated the development of
the IFM receiver. The IFM is basically a crystal video receiver with the addition
of a frequency sensing method. The usual frequency sensing method is to divide
the signal into two paths with a short delay inserted in one path. The phase of the
signals is compared and the phase shift is proportional to the carrier frequency
␪ = 2␲ f␶ (3.28)
where:
␪ = phase shift (radians)

f = carrier frequency (Hz)
␶ = differential delay (sec)
The phase difference is proportional to the frequency, which means that for a
given delay, the unambiguous band covered is that for which the phase change is
1/␶ . (For an unambiguous bandwidth covered equal to 1 GHz, the required delay
is 1 ns.)
There are several problems in applying the IFM. Some are the same as for CV
receivers. Perhaps the most pressing problem is caused by simultaneous signals.
When more than one signal is present at the same instant, the phase shift above
is not a meaningful measure of the frequency of any of them.
3.10.1 Limiters Applied to IFMs

The IFM discriminator output is proportional to the signal amplitude. In view of
its use to measure frequency, a natural addition (ahead of the discriminator) is a
limiting amplifier. Adding a true limiter makes the discriminator output indepen-
dent of the input signal level (which means that, in the absence of any signal, noise
alone creates a full scale output). In such a system, noise limited sensitivity is
obtained. However, a separate crystal video amplitude channel is required to oper-
ate a threshold detector, which determines the times at which there is sufficient
signal level for the discriminator output to be valid. With a limiter, accurate
frequency measurement over a dynamic range of 70 dB or more is practical.
A limiter impacts the operation of the IFM in ways other than increasing its
dynamic range. Consider frequency accuracy. Because the microwave components
in the IFM are not perfectly matched, a constant signal amplitude at the input
makes it possible to calibrate the signal frequency only as a function of the angle
of the output vector. The same calibration is valid over a broad range of input
signal amplitudes. Without the limiter, separate frequency versus angle calibration
tables would be needed at each input amplitude level.
Another effect of the limiter concerns operation of the IFM in the presence of
simultaneous input signals. The familiar capture effect of the FM broadcast receiver
applies in a similar way to the IFM: if more than one signal is present, the strongest
one tends to suppress the others.
It should be noted that by making multiple frequency measurements on a single
pulse, intrapulse frequency modulation can be detected and characterized using an
IFM. Phase reversal signals may also be detected by this procedure since large
positive and negative frequency excursions will appear at the IFM output. The
phase detection circuitry will indicate the proper frequency during the time the
phase is constant. When a 180° phase reversal occurs, the phase detector will
momentarily indicate a frequency that is 180° away from the actual one. This
disturbance will last as long as the differential delay within the IFM.
3.10.2 The Simultaneous Signal Problem

An IFM system can have many advantages, such as:
• 100% probability of intercept;

• Reasonably good sensitivity;
• Frequency accuracy to 0.1%, or better;
• Digital output;
• Rapid response time;
• Wide dynamic range.
In an environment of pulsed signals in which there is usually one pulse present

at a time, the IFM is nearly an ideal solution to many intercept problems. As the
environmental pulse density increases, and if high duty cycle or CW signals are
present, the problem of how to process simultaneous signals becomes very
important. Because the simultaneous signal problem is usually cited as a major
reason for this lack of IFM deployment, this problem requires careful examination.
There are two aspects to the problem. Consider the static case with two signals
simultaneously present, each at a constant amplitude. (In the dynamic case, one
signal pulse begins or ends during the critical time when the frequency of the other
is being measured.) In both cases, the error depends on the relative amplitudes of
the two signals. Of course, the relative amplitudes at the discriminator input are
affected by the action of the limiter (if used).
Static Case Without Limiter

In the case where there are two constant amplitude signals present during the
frequency measurement and no limiter is used, the situation is easily analyzed
graphically, as shown in Figure 3.21. The signals are represented as phasors.
Without loss of generality, the strongest signal, S 1 , is drawn along the vertical axis.
The weaker signal S 2 , is shown adding vectorially to S 1 to produce a resultant
signal S displaced by an angle ⌬␾ from the strongest signal S 1 . Clearly, ⌬␾ is zero
if S 2 is at the same angle as S 1 and has a maximum value if the angle of S 2 is such
that the resultant S is just tangent to the circle formed by rotating S 2 . This worst-
case angle occurs when
2 2 2
S = S1 − S2 (3.29)
in which case
⌬␾ = arcsin (S 2 /S 1 ) (3.30)
Figure 3.21 Error due to static simultaneous signals.

If S 2 /S 1 is small, ⌬␾ ∼ S 2 /S 1 . In general, the worst-case error as a function of

S 1 /S 2 is shown in Figure 3.22. To put these results in perspective, suppose that
the finest resolution used in a digital IFM is 1/64 of 360°, or 5.6°. Thus, if the
amplitude ratio S 1 /S 2 is 10 dB, the presence of the weaker signal has little effect
on the measured frequency value.
Static Case with Limiter

When a limiter is placed ahead of the discriminator, the capture effect of the limiter
acts to decrease the amplitude of the smaller signal relative to the larger one. This
helps to make a nearly correct measurement. The highly nonlinear limiter also
generates various intermodulation products at frequencies which are, for example,
the differences between twice the frequency of the stronger signal and the frequency
of the weaker signal.
When such intermodulation products are within the passband of the filter
following the RF limiter, they act like additional simultaneous signals. However,
because the total power out of the limiter is constant, these intermodulation prod-
ucts only reduce the power of the weaker signal and hence reduce any frequency
measurement error. Thus, a limiter always acts to decrease the frequency measure-
ment error for the static simultaneous signal case. How much the error is reduced
depends on the SNR of the stronger signal at the input to the limiter. According
to [21], the output signal ratio (S 2 /S 1 )out relative to the input signal ratio (S 2 /S 1 )in
for a hard limiter and a high SNR (30 dB) varies, as shown in Figure 3.23. This
Figure 3.22 Maximum angular error due to a simultaneous signal (static case, no limiting).
Figure 3.23 Limiter capture effect.
means that the worst case angular error can be obtained from Figure 3.22 using
the value of (S 1 /S 2 )out obtained from Figure 3.23.
Dynamic Case
The problem caused by exactly coincident pulses is not too serious, as such a case
is rare in a low density environment. The real problem is the case when a second
signal pulse arrives shortly after the leading edge of the first, as shown in Figure
3.24. This can cause large frequency errors due to transients in the IFM. Digital
IFMs consist of several discriminators with differing delay line lengths. To see why
there may be a large error, notice that at any instant the delayed arm of the
discriminator may have only the first signal, while the undelayed arm has the
combined signal at its output. With different delay lines in use, one module may
have a different condition than another. In tests of an IFM with no circuitry to
protect against this condition, significant probability of error (e.g., 20%) occurred
even if S 1 /S 2 was in excess of 20 dB for a 60-ns delay between the two pulses.
These errors can be greatly reduced by circuitry that detects the presence of such
a transient and inhibits the frequency measurement during those times.
Several methods for detecting simultaneous signals are available. One of the
earliest was to detect the difference frequency between the two signals produced
in the detection process. This operates by filtering the detector output from above
the video band of interest (e.g., 20 MHz) to some maximum difference frequency
Figure 3.24 Overlapping pulsed signals.
of interest (e.g., 500 MHz), then detecting and low pass filtering the output of this
bandpass filter, and comparing the result to a threshold. If two or more strong
signals are present, their simultaneous presence is detected. (These circuits were
incorporated in the IFMs of the early 1960s.) Unfortunately, such circuits are not
useful for simultaneous signal detection when one signal is weak relative to the
other, which can still result in transient frequency errors.
A more effective method is to require the video voltages out of the detectors
to remain constant during the measurement process. If it does not remain constant,
then a ‘‘no-read’’ signal can be generated to scrap that particular measurement.
This method can reduce the probability of error to less than a few percent if
| S 1 /S 2 | exceeds 2 dB and can maintain probability of error of less than 20%, even
for S 2 = S 1 for ␶ = 30 ns. There are three cases of interest:
1. For S 1 > S 2 there are no transient problems and the frequency of S 1 will
be correctly measured if S 1 /S 2 exceeds a few decibels.
2. If S 2 > S 1 , the frequency of S 2 will be correctly determined if the delay is
less than that required by the IFM video bandwidth to give a stable video
voltage (e.g., < 100 ns typical). The circuitry to process the video can detect
the transients and simply delays the reading until the video voltage stabilizes
[provided (S 2 /S 1 ) > a few decibels].
3. If the delay between the pulses exceeds that for producing a stabilized
video voltage after the second pulse arrives, the frequency of S 1 is correctly
determined, and the frequency of S 2 is correctly determined if S 2 /S 1 , exceeds
a few decibels.
In many other cases, the frequency reading is inhibited. Most of the effort to
surmount the simultaneous signal problem has been devoted to analyzing the
detected video signals at the IFM output ports, after the limiter. If the signal
envelope ahead of the limiter were used, it seems possible to do additional processing
(e.g., it would be possible to actually measure the ratio of S 1 /S 2 in cases where the
delay allows such a measurement). This information would contribute to correctly
interpreting the digitized RF values. Of course, adding wide dynamic range AM
detection, as well as the necessary high speed amplitude measuring and processing
circuitry, would increase the cost, size, weight, and power consumption of the
IFM.
The probability of occurrence of overlapped pulses is a key question. The
methods of Chapter 4 can be used to compute the probability that pulses from
any pair of radar are coincident. Also, one can use a probabilistic model to estimate
that probability. Given that a pulse has begun from some radar, it is of interest to
compute the probability that a pulse from none of the other radars begins within
some time ␶ after the start of the first pulse. Let the PRIs be denoted T1 , T2 , T3 ,
. . . , TN for the N radars in the environment. The probability that a pulse from
anyone of them begins within time ␶ is approximately ␶ /Tk . The desired case is
for none of the other radars to produce a pulse in time ␶ . Assuming statistical
independence and random pulsing at average PRFs of 1/Tk , this is
写
N
(1 − ␶ /Ti )
i =1
Probability of no other pulses occurring = 1 − (3.31)
1 − ␶ /Tk
If all the pulse trains have the same PRI, this reduces to 1 − (␶ /T)N − 1 and
some simple examples can be explored. Figure 3.25 is a plot of 1 − (1 − ␶ /T)N − 1
versus the number of emitters N for ␶ /T equal to 10−4. For instance, if ␶ = 100 ns,
which is typical of the time required to make a valid frequency measurement by
a digital IFM, and if the average PRI (T) is 1 ms, ␶ /T = 10−4. Note that each
emitter is assumed to produce 1,000 pps, so that the pulse density at the receiver
is simply the number of emitters multiplied by 1,000 pps. When the probability
reaches a level of about 0.01 to 0.1, the problem of overlapped pulses becomes
significant. For this model of the environment, the overlapped pulse problem
becomes significant when between 100 and 1,000 emitters are being received by
the IFM (or 100,000 to 1,000,000 pps). Clearly, there is always a limit to how
dense the environment can be and have the IFM remain effective. Roughly speaking,
␶ /T times the number of pulses per second to which the IFM responds should be
less than 0.1. If the environment is too dense, then the sensitivity of the IFM can
be decreased; otherwise, its bandwidth can be restricted by using bandpass filters
at the input to the IFM. This probabilistic model clearly is inexact for radars with
nearly constant PRIs. The synchronization problems discussed in Chapter 4 may
cause additional pulse overlaps.
Even for very dense environments, IFMs should remain useful to cover RF
bandwidths of 0.5 to 1 GHz. However, methods of detecting simultaneous signals
will be even more important as the pulse density increases. Reducing the time (␶ )
required to make a frequency measurement would also be helpful. However, this
can be reduced only to a certain point established by the longest delay line used
in the IFM and by the video bandwidths it uses. Another limit is imposed by the
Figure 3.25 Probability of pulse coincidence versus number of emitters.
rise time of typical radar pulses. Therefore, it is unlikely that reductions in ␶ can
be achieved while retaining good performance for all types of pulses.
3.10.3 CW Signals and IFMs

In a signal environment consisting entirely of pulsed signals, the preceding discus-
sion of IFM performance in the case of simultaneous signals would describe the
situation, and the pulse density in the band of interest would determine the expected
faction of coincident pulses. However, if there is even one CW signal present, then
all pulses are coincident with that signal. In the usual case, the CW signal is
significantly weaker than the pulsed signals, and therefore, the frequency of the
pulsed signals can be accurately determined. Indeed, ac-coupled video circuitry can
remove any residual bias created by a weak CW signal. Of course, a strong CW
signal would change the situation drastically insofar as any pulsed signals weaker
than the CW signal will be masked by it.
There are several methods of coping with the CW signal problem. Knowing
about the presence of the CW signal is the first step. One way is to insert an RF
chopper ahead of the limiter. Then the frequency of the CW signal can be measured
in the same way as the RF of pulsed signals. Once the CW frequency has been
measured, a tunable notch filter can be used (ahead of the limiter) to remove it
from the receiver. Of course, multiple CW signals are still a problem. Generally,
the frequency of the strongest one can be measured; then, when it is removed by
a notch filter, the frequency of the next strongest can be measured, and another
notch filter can be used to remove it. This process can continue until either all of
the CW signals are removed or all of the system’s notch filters have been brought
into action.
Another method of dealing with multiple CW signals is to create a narrowband
channel with a swept bandpass filter. As this filter sweeps across the band, all of
the CW signals can be detected and their frequencies measured. The sweeping
bandpass filter output can be compared with the signals in the full bandwidth to
determine when the filter output is actually a CW signal, or if it is just a pulse that
happened to occur at the frequency to which the filter is tuned.
Other methods of detecting strong CW signals include measuring the average
dc level from a video detector (which can be caused by either CW signals or a
dense environment.). Another way is to combine the IFM with a superhet receiver,
which then performs the CW detection and measurement function while also
controlling a notch filter.
Clearly, an IFM cannot perform well in a system application without taking
into account the CW problem. As noted, in the usual situation the power of the
CW signals is weak compared to that of the peak pulse power, and therefore, the
IFM continues to operate even though there are some weak CW signals present.
3.10.4 Digitizing the IFM Output

Reading the frequency using a multiple delay line IFM is similar to reading a watt-
hour meter with several dials rotating at different speeds as power is consumed.
The slower-rotating, most significant dial is used to resolve the ambiguity of the
next most significant and so on. The IFM is different because, instead of precise
gear ratios, the phase indicated by one channel varies as a function of signal
amplitude and temperature as well as frequency. Furthermore, the phase versus
frequency curve is not precisely linear due to variations in the response of the
discriminator across the band. A typical phase (or frequency) versus phase plot for
two IFM modules with a 4:1 delay line ratio is shown in Figure 3.26. Also shown
are the ‘‘handoff corridors’’ used to unambiguously separate the four sets of curves.
While the delay line ratio is 4:1, it is necessary to resolve the phase from the coarse
discriminator to better than one part in four so as to avoid ambiguities near the
frequencies at which the fine-discriminator phase crosses the 360°/0° boundary.
In general, the resolution required for the coarse phase measurement must be
about as fine as the variations due to power and temperature changes, which in
this case is about 1/16 of 360°—that is, the coarse discriminator must be digitized
to 4 bits, even though, in the final frequency word, it furnishes only about 2 bits
of real information. This excess resolution is needed to provide digitizing free of
ambiguity. Consider a three-channel IFM with 4:1 delay line ratios and a 32-cell
resolution on the fine resolution channel. The fine channel resolution will be
approximately (1/32)(1/4)(1/4), or one part in 512 (9 bits). However, the two
coarse channels each require digitizing to one part in 16, for a total of 5 + 4 + 4
= 13 bits in the raw digital word. Read-only memory (ROM) is generally used to
translate from the raw digital word to a frequency value. Many of the 213 − 1
possible combinations of the raw data bits never occur.
Figure 3.26 ×1 − ×4 handoff corridors. (Courtesy of Anaren Microwave.)
The net result can be thought of as handoff corridors. The coarse (×1) IFM
along the horizontal axis is digitized to one part in 16. For ×1 segments 0, the fine
(×4) discriminator must be between 0 and 9, and so on, as shown in Table 3.9.
This information is sufficient to resolve the RF band (i.e., the band over which
the ×1 phase detector has a phase change of 360°) to one part in 64. There is one
part in 16 resolution from ×4 digitizer and it is used four times across the band.
To see this, assume that the input frequency is such that the ×4 discriminator is
in frequency cell #5. According to Table 3.9, if the ×1 discriminator is in cells 0,
1, 2, or 15, the ×4 discriminator is on the first of the four 360° rotations across
the band. If the ×1 discriminator output is 3, 4, 5, or 6, ×4 discriminator is on the
third rotation. Finally, if ×1 is 11, 12, 13, or 14, then ×4 is on the fourth rotation.
Thus, while ×1 is digitized to one part in 16, it serves only to identify unambiguously
which of the four fine discriminator tracks applies, thereby providing 2 bits of
information.
The linearity of the phase measuring circuitry determines the maximum delay
line ratio that can be used. As the slope and the number of tracks of the fine
discriminator increases, Figure 3.26 shows that it would be more and more difficult
to find corridors to separate them. In the limit, curves from one track would cross
those of another and no amount of increased resolution in the ×1 measurement
Table 3.9 Handoff Corridors for Coarse (×1) and Fine (×4) Discriminators
Fine (×4)
Coarse (×1) First Rotation Second Rotation Third Rotation Fourth Rotation
0 0–9
1 0–13
2 2–15 0–1
3 6–15 0–5
4 14–15 0–13
5 0–15
6 2–16 0–1
7 6–15 0–5
8 10–15 0–9
9 14–15 0–13
10 2–15 0–1
11 6–15 0–5
12 10–15 0–9
13 14–15 0–13
14 0–1 2–13
15 0–5 6–15
would resolve the ambiguity. IFM manufacturers choose different ratios, typically
in the range of 2:1 to 8:1. A small ratio (2:1) requires more discriminator modules,
but coarser resolution can be used in digitizing each one; whereas a large ratio
reduces the number of discriminators, but each requires a higher resolution digitizer.
The optimum configuration depends somewhat upon the application.
Analysis of the problem of ambiguity resolution in such a multiple vector
system results in the following relationship for discriminators with a delay line
ratio of p :n [21]:
| n ⌬␪ p + ⌬ ␪ n | < ␲
where:
⌬␪ p = phase error at output of discriminator with delay line p

⌬␪ n = phase error at output of discriminator with delay line n
As an example, the formula can be applied to the hands of a clock. In this

case p = 1, n = 12. If we assume no error in the minute hand (⌬␪ n = 0), then
maximum hour hand error is
12⌬␪ p ␲ , or ⌬␪ p < 15°
which amounts to half an hour, or half of one of the usual 5-minute major divisions
on the minute scale.
For typical IFM delay line ratios, the maximum allowable tolerance is given
in Table 3.10, assuming that each discriminator contributes an equal phase error
(i.e., ⌬␪ p = ⌬␪ n = ⌬␪ ).
The single pulse frequency measuring capability of the IFM must be combined
with other features to make a useful ELINT receiving system. Clearly, a system
Table 3.10 The Maximum Allowable Tolerances for Delay Line Ratios
Delay line ratio 2 3 4 5 8
Phase tolerance (A8) 60° 45° 36° 30° 20°
that measures RF, time of arrival, angle of arrival, pulse duration, pulse amplitude,
and that gives an indication of intrapulse FM has indeed rather completely charac-
terized a single radar pulse. Such a system, with recording capability to preserve
these single pulse values, could record (for later analysis) a complete description
of the signal environment in its band. This would be especially useful in characteriz-
ing radar systems with pulse-to-pulse RF agility and varying pulse intervals. Such
a system, which captures 50 to 60 bits of information per pulse, creates a difficult
recording problem in an environment of perhaps 100,000 to 1,000,000 pulses per
second (5 to 50 megabits per second). For this reason, it may be very important
to reduce the recorded data to bursts of pulses for those radars having constant
RF and pulse intervals. This reserves the recording and analysis capabilities for the
unusual (agile) signals.
3.11 Other Receivers
The narrowband superhet receiver has high sensitivity and wide dynamic range.
It is not troubled by simultaneous signals. However, it does not cover a broad
frequency band, and thus has low probability of intercept in some situations and
cannot adequately handle wideband signals (e.g., frequency hop) in others. The
IFM provides 100% probability of intercept over broad bands, but it is troubled
by time overlapped pulsed signals and CW signals. The receivers described here
attempt to overcome these deficiencies.
3.11.1 Channelized Receivers

A straightforward approach to utilizing the properties of a narrowband superhet,
while also providing wide frequency coverage, is to use a number of parallel
channels, each one tuned to a slightly different RF center frequency. The frequency
separation between the channels can be roughly equal to the 3-dB IF bandwidth
used. Then, no matter what the carrier frequency, it will be detectable with only
a small loss in sensitivity compared to a properly tuned narrowband receiver.
Actually, the sensitivity of each channel must be degraded slightly by raising the
detection threshold to maintain the same total false alarm rate as a single-channel
receiver.
Consider an example in which there are 100 parallel receiver channels. If the
tolerable false alarm probability is PFA , 1 in a single channel, the probability of
false alarm probability for N independent channels is
PFA , N = 1 − [1 − PFA , 1 ]N (3.32)
If PFA , N is small, such that NPFA , N Ⰶ 1, this is approximately

PFA , 1 ≅ NPFA , 1
For a single channel, for 90% probability of detection and 10−6 probability of
false alarm, the required SNR is 13 dB in the IF bandwidth. If 100 channels are
to provide the same overall performance, a 10−8 false alarm probability is required
on each channel. This raises the threshold about 1.3 dB, making the SNR for 90%
probability of detection about 14.3 dB at IF [22]. This is true at the center of each
IF band. If the signal RF is exactly halfway between two of the channels, the
sensitivity may be reduced by the straddling loss, typically an additional 3 or
4.3 dB total. As can be seen, in practice there is a sensitivity loss connected with
channelization. The sensitivity loss is less than the penalty incurred by a wideband
IFM or crystal video receiver, but there is nonetheless a significant cost and complex-
ity penalty to be paid for channelization.
There is also a severe signal processing problem associated with channelization.
If the signal bandwidth is large compared to the channel bandwidth, a number of
adjacent channels will be excited simultaneously. Even if the signal bandwidth is
small compared to the channel bandwidth, a strong signal will excite a number
of channels whose outputs will have distorted pulse shape envelopes because of
inaccurate tuning. This makes correct interpretation of the channelized receiver
outputs difficult, especially in a dense environment.
These problems are addressed in the design of a channelized receiver by various
thresholding schemes [23, Ch. 5] and careful filter skirt design.
A difficult problem is to detect simultaneous signals of widely differing ampli-
tudes but with small carrier frequency separation. The detection of such signals
depends, in part, on the character of the signal spectra. (Some pulses may have
very sharp rise times, and thus the signal spectrum falls off as the reciprocal of the
frequency separation from the carrier. If the pulse has a long rise time, the signal
spectrum falls off as the reciprocal of the square of the frequency separation from
the carrier.) In this discussion, rectangular pulses are assumed. If a 1-␮ s pulse is
received, its spectrum will be down only 33.7 dB, 15.5 MHz away from the carrier.
Therefore, it is not possible to detect another simultaneous signal more than about
35 dB below the first, if its carrier frequency is less than 15 MHz away.
The main lobe spectrum width of a rectangular pulse (which contains about
90% of the energy) is twice the reciprocal of the pulse duration. Thus, good pulse
fidelity requires that the channel processing bandwidth be no smaller than twice
the reciprocal of the minimum pulse duration. In practice, it is advisable to make
the channelizer filter bandwidth about 3.2 times the reciprocal of the pulse duration
[24].
A filter bank (such as used by a channelized receiver) provides, in effect, an
instantaneous spectrum analysis capability in which the channel bandwidth is a
measure of the frequency resolution and the impulse response of the processing filter
is a measure of the time resolving power. The product of the channel bandwidth and
the impulse response is no less than 2. This gives rise to a view of the channelized
receiver as an attempt to measure the frequency-time-amplitude-characteristics of
the signal environment, as shown in Figure 3.27. Other receivers may be character-
ized by their coverage of this frequency-time plane (as indicated, a narrowband
receiver covers a single frequency channel for a long period of time). A swept
Figure 3.27 Time-frequency-amplitude signal space.
receiver covers the entire frequency band, but spends only a short time at each
frequency. The crystal video covers the entire frequency band, but it provides no
frequency resolution. The IFM also covers the entire frequency band, and provides
good frequency resolution, but the measured value is a function of the amplitude
as well as the frequency of signals which overlap along the time axis.
Channelization Schemes
The use of contiguous parallel receiver channels is clearly a very expensive approach
if wideband coverage is needed using narrowband channels. In practice, the contigu-
ous filters are constructed at an intermediate frequency, for example, 10 channels,
each 20 MHz wide, centered at about 300 MHz. Then, 200-MHz bands are selected
in the microwave region for mixing down to the broadband IF filter bank. If a
wider instantaneous band is to be covered, several portions of the microwave
spectrum can be simultaneously mixed into the same IF band. The true frequency
can be determined by detecting the energy at RF prior to conversion, as shown in
Figure 3.28. Here, only 15 filters are needed to obtain 20-MHz channel spacing
across a 1-GHz band. This would take 50 filters using the brute force approach.
There are two problems. First, the indication of which RF band contains the signal
is accomplished via broadband detection (200-MHz bandwidth in this example),
which reduces the sensitivity. Second, this approach introduces a simultaneous
signal problem. If two signals occur at the same time in different RF bands, the
output of the narrowband filters cannot properly indicate the frequency. Even if
different filters in the contiguous filter bank are excited, there can still be confusion
as to which response was the result of a particular input pulse. Also, a CW signal
can render one of the narrowband channels useless, which, in effect, causes five
different RF channels to be lost.
Many of the problems associated with channelizing equipment are greatly
reduced if the channel bandwidth is significantly larger than the bandwidth of the
signal of interest. For example, a channel bandwidth of 500 MHz greatly reduces
88
Figure 3.28 Channelized receiver with 1-GHz band coverage.

Characteristics of ELINT Interception Systems
the pulse density compared to that of a wideband receiver covering, say, 1 to

10 GHz. The 500-MHz channel can be fed to a variety of additional receiving
and processing devices such as IFMs, narrowband sweeping receivers, and even
additional (narrowband) channelizing equipment. Hence, a wideband channelizer
in conjunction with other types of receivers is a very useful system design approach.
The use of high-speed A/D converters followed by a fast Fourier transform (FFT)
or other digital filter bank is a modern approach to channelizer design.
3.11.2 Acousto-Optic (Bragg Cell) Receivers

The cost and complexity of wideband channelizers led to the development of
acousto-optic receivers. These use the angle of deflection of a laser beam as it
passes through a Bragg cell to provide a measure of frequency. The amplitude of
the signal is determined by the light level striking an array of photocells, as shown
in Figure 3.29. Because multiple signals produce multiple output deflections, con-
ceptually, this receiver is a replacement for the channelized receiver. The frequency
resolution is a function of the angular spacing of the photocell array that measures
the output light intensity as a function of deflection angle. The time resolution is
a function of the bandwidth of the photocell detectors. Narrowband photocells
(in the range of a few kilohertz) have been used as wide-open high-frequency
resolution spectrum analyzers. These have poor ability to reproduce the envelope
of pulsed signals for ELINT applications. Wideband photocells (−10 MHz) with
individual outputs are needed if this technique is to replace the channelized
approach. In terms of the frequency-time plane of Figure 3.27, narrowband photo-
cells provide good frequency resolution, but they sum the spectra of the events
along the time axis during their millisecond impulse response time.
The principles of Bragg cell operation are well understood; however, there are
many design problems associated with component shortcomings as well as the
inherent problems of channelization. One promising application [23, Ch. 7] is to
use a two-dimensional Bragg cell (along with a photocell planar array) to resolve
both the angle of arrival and the frequency of pulse. In this approach, the Bragg
cell is modulated by the signals from separate antennas, and the interferometry
effects provide an (x, y) deflection of the beam, which is a function of both frequency
Figure 3.29 Acousto-optic receiver.

and angle of arrival. To achieve the same function with a conventional channelized
receiver requires multiple phase-matched channelizer banks with phase comparison
circuitry at the outputs of the channel filters.
All of the problems associated with processing the multiple outputs of channel-
ized receivers are present in the acousto-optic implementation as well, with the
added problems associated with the less mature technology of the components.
The significant advantage of the single Bragg cell replacing the filter bank of the
channelized receiver provided an incentive for development. This incentive decreases
as high-speed A/D converters and real-time DSP developments become better and
cheaper.
3.11.3 Microscan Receivers

The microscan receiver is a fast sweeping superhet that makes use of a pulse
compression or chirp filter at the intermediate frequency. This makes it possible
to sweep the band at sweep speeds greater than the square of the frequency resolving
power of the receiver. As noted in Section 3.10.2, in a conventional swept superhet
with IF bandwidth (and, hence, frequency resolving power) B, the sweep speed
limit is repeated from (3.22)
⌬F
< B2
T
where:
⌬F = band swept (Hz)

T = time to sweep the band (sec)
B = IF bandwidth (Hz)
As one sweeps the conventional receiver local oscillator at a faster and faster
rate, the duration of the IF filter output due to a CW signal in the band becomes
shorter and shorter until it reaches the minimum value, which is that due to the
impulse response of the IF filter. To permit more rapid sweeping, it is necessary
to make the impulse response shorter (i.e., to widen the IF bandwidth). Unfortu-
nately, this degrades the frequency resolving ability of the receiver. The microscan
solves this problem by using the equivalent of pulse compression in radar.
The sweeping local oscillator creates a chirped signal at IF as a result of mixing
with an incoming signal of constant frequency, as shown in Figure 3.30. (In a
conventional swept receiver, this chirping also occurs, but it is small enough to
ignore.) These chirped IF signals can be fed into a dispersive delay line, which has
the opposite slope as the sweep rate of the local oscillator. The total time of the
delay line should be at least equal to the time the signal is present at the IF amplifier
output, while the bandwidth must be equal to the selected IF bandwidth. Under
these conditions, the duration of the output pulse can be compressed by an amount
approximately equal to the product of the IF bandwidth and the duration of the
dispersive delay.
Figure 3.30 ‘‘Chirped’’ IF signal generation in microscan receiver.
Within certain constraints, the microscan receiver can be thought of as a swept

superhet receiver that is able to sweep the band as fast as desired and yet achieve
good frequency resolution. The design of such receivers is no simple matter [23,
Ch. 6; and 25, pp. 161–166]. However, the main topic of interest here is the novel
ways such a receiver can be used to assist in the interception of radar signals.
Conceptually, its output is like that of a channelized receiver with the output of
each channel being sampled sequentially at the local oscillator’s sweep repetition
rate. In this situation, examining the envelope of a pulsed signal can be done, but
the fidelity achieved is limited by the sampling rate (i.e., a single pulse duration
and PRI can be determined only to the resolution provided by the sweep rate
sampling interval). A design using a 1-GHz band swept in 1 ␮ s achieves 100%
probability of intercept for all pulses 1 ␮ s or greater in duration with PRI and PD
resolution of 1 ␮ s.
Although the microscan can process simultaneous signals (at the output of the
delay line, they are separated in time), precise determination of their PRIs and PRI
variations is degraded when compared to the capabilities of the superhet or IFM.
A microscan receiver can achieve frequency resolution of approximately the
reciprocal of the duration of the signal stored in the delay line (i.e., a 1-␮ s delay
line provides approximately 1 MHz of resolution). Suppose the full band is to be
swept in the same time as the duration of the delay line. It is clear that the output
of the delay line will not be completed until one sweep time later. Under these
conditions, the microscan receiver can only examine signals 50% of the time.
Coverage for 100% of the time requires two channels with interlaced scans. Such
a two-channel receiver with interlaced scans can provide samples of each signal
spaced by the duration of the dispersive delay line.
A major problem with microscan receivers is processing their high-speed output
signals. A 1-GHz sweep using a 1-␮ s dispersive delay for 1-MHz resolution provides
1,000 frequency channels in 1 ␮ s of time, or 1 ns per frequency resolution cell.
Mathematically, the principle of the microscan receiver is to compute the
Fourier transform of the input signal by means of chirp (linear FM) waveforms.
The usual Fourier transform notation is
∞
F (␻ ) =
1
2␲ 冕
∞
f (t) e j␻ t dt (3.33)
where F (␻ ) is the Fourier transform of the input signal f (t). In a chirped system
the frequency ␻ is a linear function of time ␶ (i.e., ␻ = 2 ␮␶ , where ␮ is the slope
of the chirp and ␮ is a constant). Thus, (3.33) can be rewritten
∞
F (␶ ) = 冕
∞
f (t) e −j(2␮␶ )t dt (3.34)
The exponent can be formally written as
−2(␮␶ )t = ␮ (␶ − t)2 − ␮␶ 2 − ␮ t 2 (3.35)
so that (3.34) becomes
∞
F (␶ ) =
冤冕
∞
f (t) e j␮ t e j␮ (t − ␶ ) dt e j␮␶
2 2
冥
2
(3.36)
The terms exp ( j␮ t 2 ) and exp ( j␮␶ 2 ) are recognized as premultiplication and
postmultiplication by a chirped local oscillator. The entire process to implement
(3.36) is as follows:
• Step 1: Multiply (mix) the input signal f (t) with a chirped LO signal
exp ( j␮ t)2.
• Step 2: Convolve the result with a second chirp exp ( j␮ − ␶ )2 (done in the
dispersive delay line).
• Step 3: Multiply the result by a third chirp exp ( j␮␶ )2.
Practical implementations require finite rather than infinite limits on the pro-
cess. Finally, the phase information is usually not needed. Thus the final (post)
multiplication via a chirp can be left out. The output of the dispersive delay line
can be processed by an envelope detector as indicated in Figure 3.30.
Multiple channel microscan receivers called metascan receivers were described

by Ready [26] and some design considerations are given in [27]. The use of high
temperature superconductive materials to construct wideband compressive receivers
is described in [28]. These units are able to achieve bandwidths of over 1 GHz
and dispersive delay times of 40 ns (time-bandwidth product in excess of 100).
Projections of future developments able to cover 10 GHz of bandwidth and
3.3-MHz frequency resolution were included. Such receivers must be cooled to
about 77K.
3.12 System Considerations
Various receivers have different approaches to monitoring the signal space of time,
frequency, and amplitude, as summarized in Table 3.11. It is clear that preserving
the amplitude at every time and frequency resolution cell results in a much higher
data rate than the simpler, traditional approaches, which either monitor a single
narrowband channel (tuned superhet), or which monitor a broad band but can
only measure the frequency of the strongest signal present in any time cell (IFM).
For this reason, channelization must be introduced in a deliberate and thoughtful
way. The brute force approach of preserving the entire time-frequency-amplitude
space for later analysis will result in massive recording, playback, and analysis
problems but very little additional knowledge will result.
Table 3.11 Summary of Receiver Advantages and Disadvantages

Type Advantages Disadvantages
Crystal video Cheap No frequency measurement
Small Degraded sensitivity
Proven Poor in dense environment
Saturation by one strong signal
Superhet High sensitivity No frequency agility
Good ECCM Slow search
Proven Poor narrow pulse response
Useful in interferometer and Two or more channels for DF
precision measurements
Channelized High sensitivity Limited frequency agility
High POI Limited resolution
Good ECCM Poor to moderate narrow pulse
No DF without parallel units
Cost, weight, size, power
IFM High POI Same as crystal video
Frequency agile No simultaneous signals
Fair frequency accuracy No DF
Proven
Bragg High POI Limited dynamic range
High sensitivity Poor pulse response
High selectivity No frequency agility
Narrow pulse Two for DF
Unproven
Microscan/Compressive High frequency accuracy Poor narrow pulse
Simultaneous signals Sweep-scan alignment
Handles frequency agility Wide IF required
Delay lines state of the art
True knowledge is obtained only by destroying unwanted parts of the original

signals. It is important to destroy unwanted information as soon as possible in the
intercept-record-analyze process to avoid overloading the entire system. In an
abstract sense, a high-speed A/D converter (say, a 20-GHz sampling rate with
14-bit resolution; that is, 84-dB dynamic range) coupled to a mass memory provides
an ideal ELINT receiver. It has 100% probability of intercept and preserves every-
thing in a 10-GHz bandwidth. However, recording a simple circular scan radar
for a few scans (e.g., 1 minute of data) would mean preserving and processing
1.6 × 1013 bits of data. Clearly, this is an ideal system only if a high-speed processor
coupled to the A/D converter can eliminate most of those bits and provide more
concise outputs that are of real value, that is:
• Direction of arrival;
• Peak amplitude (for estimating ERP);
• Antenna patterns (beam width, sidelobe levels);
• Pulse shape (duration, rise and fall time);
• Pulse intervals (interval variations, average PRI);
• Radio frequency (center frequency, pulse-to-pulse and on-pulse RF varia-
tions);
• Associated signals.
All of the information is present in the 1.6 × 1013 bps data stream from the
A/D converter. However, the bit stream has little value unless it can economically
produce the data items above. In designing intercept receivers of the future, those
that are most successful will focus on providing only that information that is really
needed at the output, given the signal environment in which the receiver must
operate.
References
[1] Frank, R. L., ‘‘Phase Shift Codes with Good Periodic Correlation Properties,’’ IEEE Trans.
on Information Theory, Vol. IT-8, October 1962, pp. 381–382.
[2] Friis, H. T., ‘‘Noise Figures of Radio Receivers,’’ IRE Trans., July 1944, p. 419.
[3] North, D. O., ‘‘The Absolute Sensitivity Radio Receivers,’’ RCA Review, Vol. 6, January
1942, pp. 332–344.
[4] Skolnik, M. I., Introduction to Radar Systems, New York: McGraw-Hill, 1962, p. 24.
[5] Folsom, J. B., ‘‘That ‘True-rms’ Meter: Will It Be True to You?’’ EDN, November 5,
1975, pp. 91–95.
[6] Mumford W. W., and E. H. Scheibe, Noise Performance Factors in Communications
Systems, Dedham, MA: Artech House, 1968.
[7] Barton, D. K., Radar Systems Analysis, Dedham, MA: Artech House, 1979.
[8] Tsui, J. B., Microwave Receivers and Related Components, Springfield, VA: NTIS, 1983.
[9] Urkowitz, H., ‘‘Energy Detection of Unknown Deterministic Signals,’’ Proc. IEEE,
Vol. 55, No. 4, April 1967, pp. 523–531.
[10] Urkowitz, H., Signal Theory and Random Processes, Dedham, MA: Artech House, 1983,
pp. 488–465.
[11] Torrieri, D. J., Principles of Military Communications Systems, Dedham, MA: Artech
House, 1981, Ch. 4.
[12] Emerson, R. C., ‘‘First Probability Densities with Square Law Detectors,’’ Journal of
Applied Physics, Vol. 24, No. 9, September 1953, pp. 1168–1176.
[13] Wiley, R. G., Electronic Intelligence: The Analysis of Radar Signals, Dedham, MA: Artech
House, 1982.
[14] Woodward, P. M., Probability and Information Theory, with Applications to Radar,
Dedham, MA: Artech House, 1980, Ch. 6. (Originally published by Pergammon Press,
London, 1953.) See also J. R. Pierce, Symbols, Signals and Noise, New York: Harper and
Row, 1961, Chapter X.
[15] Skolnik, M. I., Introduction to Radar Systems, New York: McGraw-Hill, 1980.
[16] Tsui, J., ‘‘Tangential Sensitivity of EW Receivers,’’ Microwave Journal, October 1981,
p. 99.
[17] Harrison, A. P., Jr., ‘‘The World Versus RCA: Circumventing the Superhet,’’ IEEE Spec-
trum, Vol. 20, No. 2, February 1983, p. 67.
[18] Engelson, M., and F. Telewski, Spectrum Analyzer Theory and Applications, Dedham,
MA: Artech House, 1974, Ch. 5.
[19] Torrieri, D., J., Principles of Military Communication Systems, Dedham, MA: Artech
House, 1981, pp. 155–160.
[20] Tsui, J. B., Microwave Receivers and Related Components, Springfield, VA: NTIS, 1983,
p. 59.
[21] East, P. W., ‘‘Design Techniques and Performance of Digital IFM,’’ IEE Proc., Part F,
No. 3, June 1982.
[22] Barton, D. K., Radar System Analysis, Dedham, MA: Artech House, 1976, p.17.
[23] Tsui, J. B., Microwave Receivers and Related Components, Springfield, VA: NTIS, 1983.
[24] McCune, E., ‘‘SAW Filters Improve Channelized Receivers,’’ Microwaves and RF,
January 1984, pp. 103–110.
[25] Torrierri, D. J., Principles of Military Communication Systems, Dedham, MA: Artech
House, 1981.
[26] Ready, P. J., ‘‘Multichannel Coherent Receiver,’’ U.S. Patent No. 4,204,165; granted
May 20, 1980.
[27] Moule, G. L., ‘‘SA W Compressive Receivers for Radar Intercept,’’ IEEE Proc., Vol. 129,
Part F, No. 3, June 1982, p. 180.
[28] Lyons, W. G., et al., ‘‘High-Tc Superconductive Wideband Compressive Receivers,’’ The
Lincoln Laboratory Journal, Vol. 9, No. 1, 1998, pp. 33–64.
CHAPTER 4
Probability of Intercept
Arthur G. Self
4.1 Background
Probability of intercept (POI) is a key performance feature of EW surveillance and

reconnaissance systems; it relates to the probability of time coincidence of two or
more parametric ‘‘window functions,’’ such as scanning antennas, sweeping or
stepping receivers, and frequency agile emitters.
Many publications and examples that examine POI and its associated intercept
time statistics invariably quote examples involving crystal video (wide open) receiv-
ers (CVRs), scanning superheterodyne (SSH) receivers, and the implications of
scanning, directional antennae. However, over the last 20 years, significant develop-
ments in technology have occurred as have several armed conflicts. These have
dictated radical changes in requirements, force modus operandi, and other issues.
In previous chapters, the impact of receiver developments, the impact of more
recent conflicts (such as the Gulf War), and the emergence of a number of new
threats such as the low probability of intercept (LPI) emitter have been examined.
Clearly, there is a significant POI issue associated with such emitters in the presence
of the littoral electromagnetic environment. Finally, as shown in the USS Cole
incident, the new threat (now referred to as the ‘‘asymmetric threat’’) from fast
attack craft (FAC), for example, is significant and must be accounted for in tasks,
combat strategies, weapons, sensors, and so on, for fast reaction and adequate
firepower.
So today’s EW systems have to deal with increasing complexity and density of
the electromagnetic environment; comprising multimode emitters, high and low
ERPs, increased signal agilities (RF, PRI, and so forth), complex scans, and fleeting
emissions.
EW system architectures are now more complex and, with more capable weapon
systems (hard kill and soft kill), result in a more sophisticated modus operandi
involving integration not only within their own platform or service but with other
platforms and force structures. EW receiving systems have become EW suites
which provide situational awareness, protection, and ELINT (fine grain analysis)
capabilities. Note that some of the latest ESM systems are offering a specific
emitter identification (SEI) capability—typically through amplitude and/or phase
digitization on a per-pulse basis. Given the complex mission environments, and
thus the required EW system architectures, ESM is now a likely, and necessary,
precursor to ELINT.
97
98 Probability of Intercept
In summary, there is now far less distinction between ESM and ELINT. Empha-
sis is turning to EW systems designs with both a tactical (ESM) and strategic
(ELINT) capability. Systems designers need new tools to assist them, now that POI
is now a more complex problem.
4.2 Developments in the Theory Behind POI
4.2.1 Intercept Description

Table 4.1 summarizes a range of typical kinds of intercept [1]; these can be subdi-
vided into spatial, frequency, and time domains.
• Spatial domain: The beam-on-beam example where both transmitting (Tx)

and receiving (Rx) systems use rotating, directional antennas and where the
type of intercept is related to detectability (main beam or a percent of
sidelobes).
• Frequency domain: Typically a scanning or sweeping receiver (e.g., a SSH,
against a frequency agile transmitter).
• Time domain: Here, a number of distinct types of intercept are relevant.
For example, the concept of a minimum number of pulses (or intercept
duration) is important since it is required in ESM systems in order to differen-
tiate between real signals and noise, as well as for subsequent signal sorting
processes. Emission control policies will dictate when a transmitter is on
and when it is not—thus introducing the scenario element to the intercept
time arena. The time for an ESM system to process initially detected pulses
will, in certain processing architectures, dictate whether or not subsequent
pulses from the same emitter are tracked continuously or lost due to insuffi-
cient processing bandwidth/capability.
• Others: Propagation fluctuation aspects can also impact the time to intercept.
These can be subdivided into two principal effects, namely, coherent (specu-
lar) scattering and tropospheric scattering. Coherent scattering is important
for within-the-radio-horizon scenarios and low elevation threats (particularly
sea-skimming missiles or high altitude platforms near the radio horizon).
Tropospheric scattering relates to scattering of electromagnetic waves off
the Earth’s troposphere as witnessed by rapidly varying signal amplitude
and fading. The intercept probability of the ESM system will be determined
by the mean period of simultaneous overlap between the random pulse
train representing such propagation fluctuations and the regular pulse trains
representing the cyclic properties of the emitter and ESM systems.
4.2.2 Implications of Today’s Environments/Operations on Intercept Time

The following are the types of intercepts shown in Table 4.1:
• In the spatial domain, phased array technology allows scanning space in

a pseudo-random manner and such radar systems often perform multiple
Table 4.1 Types of Intercepts

Receiver Activity Transmitter Activity Comments
Spatial Domain Scanning antenna Scanning antenna Beam-on-beam
intercept
Main beam to main
beam
Main beam to sidelobes
(variable)
Frequency Domain Stepping or scanning Fixed RF, random Narrow instantaneous
Rx bandwidth across jumping, or some bandwidth (IBW)
a specific operating other varying receiver which is
bandwidth transmitted RF stepping or sliding
across a defined RF
bandwidth looking for
either a fixed or agile
threat emitter RF. One
window function if
fixed RF else two.
Time Domain
1. Minimum Main beam MB or SL intercept Given a spatial
intercept duration intercept has occurred,
how long is the
coincidence; valid
intercept if the number
of received pulses is ≥
threshold
2. Emitter EMCON Main beam MB or SL intercept For ECCM purposes,
the emitter switches on
for only short periods
of time
3. Missing pulses Environmental and
emitter transmit effects
could give rise to
missing pulses in the
received pulse train
4. ESM processing Receive and For nonparallel
time processing architectures and/or
architecture nonmultiple buffering
data, the ESM could
be blind to new data
while it is processing
the last received data
Others
Tropospheric Scanning Scanning Rayleigh distributed
scattering effects amplitude versus time
Window function
description: t = time
above detection
threshold; (T − t) =
time below detection
threshold (nulls)
Source: [1].
functions. The electronic scanning antenna allows eliminating mechanical

inertia as a design limitation; however, radar functions performed often
impose a structure on the motion of the beam that can be exploited by
would-be interceptors. Radar systems using mechanically scanned antennae
have provided the scan characteristics of the intercept time problem. A

change of mode (e.g., surveillance to tracking) or the switching on of a
specific tracking radar afforded the EW designer further useful intelligence.
Multifunction phased array radars reduce such possibilities, thus making
intercept much more difficult.
• The time domain, including such operational aspects as emission control
(EMCON), further increases the time to intercept. Operations in the littoral
(i.e., close to shore) raise additional multipath and possibly missing pulse
issues.
• In the frequency domain, the littoral electromagnetic environment makes
signal search and identification that much more difficult in comparison to
deep ocean environments. Clearly, more agility in the Tx signal parameters
(be it RF or PRI) again directly impacts time to signal intercept. Additionally,
low power frequency modulated continuous wave (FMCW) signals force
the EW designer to long integration times, and thus, directly impacting time
to intercept.
In summary, EW system designers are being pushed more and more towards
architectures that deliver single pulse or pulse burst intercept capabilities.
4.2.3 Mathematical Models

The methodology of Self and Smith [2, 3] has been widely used due to its simplicity
and ease of use, its applicability to scenarios involving more than two window
functions, and its ability to predict intercept time accurately in a wide range of
situations.
This section includes both the key formulae as well as several enhancements
to the original work, including some specific, realistic examples with practical
application. This is followed by a number of new and exciting theoretical develop-
ments involving some elegant mathematical theories on the subject of synchronism.
From [3], a convenient way of investigating the intercept problem is to represent
the activities of the receiving and transmitting systems by window functions, as
shown schematically in Figure 4.1. A key assumption is that the window functions
are independent (or of ‘‘random-phase’’). Such ‘window functions’ can also be
described or represented as pulse trains—in this chapter, these descriptions are
synonymous and interchangeable.
Key formulae are as summarized next.
For a number of coincidences, M, required, the equation for the mean period
between coincidences, T0 , has the following explicit form:
写
M
(Tj /␶ j )
j =1
T0 = (4.1)
M
∑ (1/␶ j )
j =1
and for M = 1, 2, and 3, this becomes:

Figure 4.1 Basic window functions and their coincidences.
M=1 T0 = T1 (4.2)
T1 T2
M=2 T0 = (4.3)
(␶ 1 + ␶ 2 )
T1 T2 T3
M=3 T0 = (4.4)
(␶ 1 ␶ 2 + ␶ 1 ␶ 3 + ␶ 2 ␶ 3 )
where T1 , T2 , T3 are the periods of the individual window functions and ␶ 1 , ␶ 2 ,

␶ 3 are the durations of the individual window openings.
The probability of at least one intercept in time T is
P (T ) = 1 − Ke −(T /T0 ) (4.5)
where K = 1 − P (0), and P (0), the probability of an intercept occurring in the first
instant, is given by
写
M
␶j
P (0) = (4.6)
T
j =1 j
The mean duration ␶ 0 of the simultaneous overlaps of all the M window

functions is given by
M
1 1
␶0
= ∑ ␶j
(4.7)
j =1
which, for M = 1, 2 and 3, becomes

M=1 ␶0 = ␶1 (4.8)
1
M=2 ␶0 = (4.9)
1 1
+
␶1 ␶2
1
M=3 ␶0 = (4.10)
1 1 1
+ +
␶1 ␶2 ␶3
Turning to the issue of intercept duration, Figure 4.2 shows a comparison [3]
of computer simulation results for examples of three-window intercept scenarios.
Results are expressed as cumulative probability versus intercept duration. Analytic
forms for these results show the following (in each case ␶ 1 denotes the smallest
window and T denotes time into the actual intercept) (private communication with
N. R. Burke, April 1983).
For the case of two windows:
Figure 4.2 Intercept duration.

2T
P (T ) = 0 ≤ T < ␶1 (4.11)
(␶ 1 + ␶ 2 )
=1 T ≥ ␶1
For the case of three windows:
2T (␶ 1 + ␶ 2 + ␶ 3 ) − 3T 2
P (T ) = 0 ≤ T < ␶1 (4.12)
(␶ 1 ␶ 2 + ␶ 2 ␶ 3 + ␶ 3 ␶ 1 )
=1 T ≥ ␶1
A number of developments of the theory of window functions with random

phase have been made, and these are summarized here.
Interception of at least m pulses Declaration of an adequate intercept may require

at least m pulses. Thus, if the PRI of the received signal is T2 , then
d = mT2 (4.13)
写
M
Tj
j =1
(␶ j − d )
T0 = where ␶ j > mT2
M
1
∑ (␶ j − d )
j =1
Using (4.3), Table 4.2 demonstrates the effect on T0 of increasing the minimum
intercept duration d from 0 to 25 ms for the cases of T1 = 0.7 and 1.3 seconds
(two pulse trains are ␶ 1 = 0.058, ␶ 2 = 0.078, T1 , T2 = 6.0 seconds). If a minimum
of 10 pulses is required for a valid signal detection, then a minimum intercept
duration of 20 ms is required for a 600-Hz PRF radar (less than 5 ms for a 4-kHz
radar), which, in turn, implies a mean time between intercepts of at least 40 seconds
for a receiver scan period of 0.7 second; this is in comparison to a mean time
between intercepts of only 31 seconds for the purely theoretical example where
receiver integration aspects are not included.
Effect of Auto Pause Mechanism For a sweeping receiver, such a mechanism

compromises intercept time. This can be deduced as an extension of the theory. If
Table 4.2 Effect on T0 of Increasing d

T0 Number of Pulses
d (ms) T1 = 0.7 T1 = 1.3 PRF = 600 Hz PRF = 4 kHz
0 31 57
5 33 62 3 20
10 36 67 6 40
15 40 74 9 60
20 44 81 12 80
25 49 91 15 100
Source: [3].
the receiver interrupts its frequency sweep for a known amount of time every time
a signal is encountered, then the effective sweep period of the receiver will be
extended proportionately and thus the mean period between coincidences will
depend both on the auto pause time and the average number of signals transmitting
at any one time.
Propagation Fluctuations The effects of propagation can be added in a simple

manner to this theory [1]. Propagation fluctuations are assumed to be able to be
subdivided into two types:
1. Long-term fading effects (affects the way in which average power received
varies and hence will show variations in terms of hourly, daily changes);
2. Short-term fading effects (affects POI more directly).
Intercept of the radar signal from a distant transmitter will be possible when
the strength of the fading signal rises above the threshold level of the receiver. The
occurrence (in time) of bursts of the signal above the threshold may be represented
by a random rectangular pulse train. The mean period of the pulse train <T > is
the reciprocal of the mean rate of upward crossings of the threshold level by the
fluctuating signal amplitude. The mean duration of the pulse train <␶ > is the mean
duration of the signal amplitude above the threshold level.
The intercept probability will be determined by the overlaps between the ran-
dom pulse train representing propagation fluctuations and the regular pulse trains
representing the cyclic properties of the transmit and receive systems:
<T > = 冉冊冉冊
1 1/2 1
␲a 2f
⭈ ea (4.14)
<␶ > = 冉冊冉冊
1 1/2 1
␲a 2f
(4.15)
where:
a = L /b 0 is a dimensionless threshold parameter

b 0 = mean power level of the fading signal
f = rms bandwidth of amplitude fluctuations
L = threshold level
For example, if f = 1 Hz at S-band and a = 1, then substituting into (4.14) and

(4.15) gives <T > = 0.77 second and <␶ > = 0.28 second.
A Practical Example of Three Pulse Trains

Looking at a more realistic radar type example, let us derive the intercept time
statistics [1]. Table 4.3 shows the assumed pulse train values (fading is neglected).
Table 4.3 Realistic Receiver/Transmitter Scenario

Parameters for Example POI Calculation
j Mean Duration, j (sec) Mean Period, T (sec) Comments
1 ␶ 1 = 0.03 T1 = 3.0 Tx antenna scan
2 ␶ 2 = 2 × 10−6 T2 = 5 × 10−4 Pulse modulation
3 ␶ 3 = T3 /100 10−4 ≤ T3 ≤ 104 Receiver frequency sweep
The receiver’s sweep is chosen to have an instantaneous bandwidth of 1%,

corresponding to 20 MHz in a total sweep range of 2 GHz. The mean time between
intercepts, T0 , is calculated as a function of the sweep period, T3 , which varies
over a wide range, say, 10−4 seconds to 10+4 seconds.
The correct expression for T0 is to be chosen for any particular value of T3 ,
from among the equations that represent the mean period of overlaps of different
combinations of the three overlapping pulse trains:
T0 = Tj where j = 1, 2, 3 (4.16)
T j Tk
T0 = where j = 1, 2, 3 and k ≠ j (4.17)
(␶ j + ␶ k )
T1 T2 T3
T0 = (4.18)
(␶ 1 ␶ 2 + ␶ 2 ␶ 3 + ␶ 1 ␶ 3 )
In general, the above formula for mean time between intercepts does not
discriminate against the possibility of multiple overlaps, and thus, it may, in certain
circumstances, underestimate the mean time. Smith analyzes such a possibility and
concludes that the correct formula is given by the one which has the greater
numerical value, based on there being 2 M − 1 formulae in the case of M overlapping
pulse trains (in the above case this gives seven distinct formulae) (private communi-
cation with B. G. Smith, 1983).
The curve of T0 versus T3 divides into four regions as shown in Figure 4.3.
• Region #1: Very large values of T 3 . The duration of the sweep is much
larger than the scan period; the pulse modulation period of the mean period
of overlaps between these two trains. Therefore, for large T 3 , the mean time
between intercepts equals the sweep period [(4.16) with j = 3].
• Region #2: As T3 decreases, the duration of the frequency sweep falls until
it is less than the period of the scan cycle. Interception in a single sweep is
no longer certain. The mean duration of the overlap between the scan and
sweep cycles is long compared with the period of the pulse modulation, and
so the mean time between intercepts is determined by the scan and sweep
parameters in (4.17) with j = 3 and k = 1.
• Region #3: As T3 falls further, the mean overlap between the scan and sweep
cycles falls below the period of the pulse modulation and the mean period
between intercepts is (4.18) (dependent on parameters of the three interacting
pulse trains).
Figure 4.3 Intercept time. (From: [4].  1996 IEEE. Reprinted with permission.)
• Region #4: Eventually, T3 falls sufficiently that the mean period of overlaps
of the sweep and pulse modulation cycles falls below the duration of the
scan cycle. Interception within a single scan is certain, and the mean time
between intercepts equals (4.16) with j = 1.
Specific Example of a Fast Step-Scan Receiver

From [2], consider the following example: a broad-beam receiving antenna so
that it is not necessary to search in azimuth, but add the use of a fast step-scan
receiver which searches a 2,000-MHz frequency band in 200 steps, each step being
10 MHz. (The receiver bandwidth should be about 20 MHz to provide overlapping
coverage from one step to the next.) See Table 4.4.
Hence, from (4.3), the mean time between intercepts is now 17.74 seconds,
and from (4.9) the mean duration of the intercepts is 4.93 × 10−4 seconds, and the
Table 4.4 Case 1: The Result Using a Receiver

Scan Time of 100 ms
␶1 ␶2 T1 T2
0.0005 sec 0.0333 sec 0.1 sec 6 sec
average coincidence fraction is 2.78 × 10−5. This leads to a time for 90% probability
of intercept of 40.85 seconds. See Table 4.5.
Now the mean time between intercepts is increased to 277.13 seconds, their
mean duration to 7.69 ms, and the average coincidence fraction remains at
2.78 × 10−5. The time required for a 90% probability of intercept is increased to
638.12 seconds or 10.64 minutes. Clearly, increasing the receiver sweep time results
in a much longer time to intercept the signal, with an attendant increase in the
expected duration of the intercept.
Next, consider three pulse trains by adding a rotating antenna. The receiver
is sweeping in frequency as well as searching in angle, and the transmitter is also
circularly scanning. See Table 4.6.
From (4.4), the mean time between intercepts is increased drastically to 5,423
seconds (about 1.5 hours), and from (4.10) the average duration of an intercept
is 1.266 × 10−6 seconds and to reach a probability of intercept of 90% requires
3.47 hours. See Table 4.7.
This increases the mean time between intercepts to 7.35 hours with an average
intercept duration of 2.02 × 10−5 seconds and a 90% probability of intercept time
of 16.92 hours.
Table 4.8 summarizes the calculations for T0 in the above examples from [3],
and shows their relationship to POI over the interval 30% to 90%.
Cases 3 and 4 illustrate why it is usually necessary to eliminate one or more
of the pulse trains. A highly sensitive receiver (or strong signal level at the receiver)
allows us to drop the transmitter antenna scan pulse train from the analysis by
making it possible to receive the signal of interest through the emitter antenna’s
side or back lobes. The best sensitivity is achieved economically with a narrowband
swept receiver and a narrowbeam receiving antenna. Thus, by introducing two
search processes at the receiver, it may be possible to eliminate the transmitter’s
scan from the probability of intercept calculation.
A limitation of this approach is that it does not address detailed impacts of
synchronism effects between pulse trains. The above examples clearly identify such
Table 4.5 Case 2: The Same Situation as for

Case 1, with the Receiver Sweep Slowed to 2
Seconds
␶1 ␶2 T1 T2
0.01 sec 0.0333 sec 2 sec 6 sec
Table 4.6 Case 3: The Result of the Faster Receiver Sweep

␶1 ␶2 ␶3 T1 T2 T3
0.00278 sec 0.0333 sec 0.0005 sec 1.0 sec 6.0 sec 0.1 sec
Table 4.7 Case 4: The Result Using the Slower Receiver Sweep
␶1 ␶2 ␶3 T1 T2 T3
0.00278 sec 0.0333 sec 0.01 sec 1.0 sec 6.0 sec 2.0 sec
Table 4.8 Predicted Results for POI and T0 for Cases 1 to 4

Mean Period
of Simultaneous Probability of Intercept P(T)
Case Parameters (sec) Overlaps, T0 (sec) 0.3 0.5 0.7 0.9
1 ␶ 1 = 0.005 T1 = 0.1 18 6 12 21 41
␶ 2 = 0.0333 T2 = 6.0
2 ␶ 1 = 0.01 T1 = 2.0 277 99 192 333 638
␶ 2 = 0.0333 T2 = 6.0
3 ␶ 1 = 0.00278 T1 = 1.0 5,423 1,934 3,759 6,529 12,486
␶ 2 = 0.0333 T2 = 6.0
␶ 3 = 0.0005 T3 = 0.1
4 ␶ 1 = 0.00278 T1 = 1.0 26,450 9,434 18,334 31,846 60,905
␶ 2 = 0.0333 T2 = 6.0
␶ 3 = 0.01 T3 = 2.0
Source: [3].
relationships between the values of T[1,2,3] ; thus, for a two pulse train example, if
the period of one pulse train is harmonically related to the period of the other,
then the two pulse trains will either be in synchronism or they will not. If the
latter, then an intercept will not occur (note that ␶ i Ⰶ Ti in the radar case); if the
former, then there will be synchronism and intercepts will occur at known, discrete
times that can be deduced. Note that for regular pulse trains, there will always be
a fixed relationship between the pulse trains once the first intercept has occurred.
In such instances, an ESM system operating with a directional antenna may improve
its intercept probability if its rotation rate is variable.
4.2.4 Recent Developments on POI

More recent work on intercept time and POI has focused on the underlying theory
of pulse train coincidence where the pulse train phases are nonrandom. In other
words, earlier references [1–3, 5], and private communications with N. R. Burke
(April 1983), A. G. Self (1986), and B. G. Self (1983), examined noncorrelated
pulse trains (herein now referred to as the random-phase model). However, there
are instances (for example, with a stepping/scanning receiver or scanning transmit
and receive antennae) where the resultant intercept pulse train becomes highly
correlated and dependent on the input pulse train attributes. In this section, the
latest results on synchronism effects [6] are summarized as well as the elegant
application of number theory [4, 7] to the intercept problem domain.
4.2.4.1 Synchronization Effects on POI

Historically, mechanical scanning transmit systems (especially long range surveil-
lance systems) had rotation rates of typically 3, 6, 9, 10 (and beyond) rpm. Clearly,
a receiving system with the same or sub/multiples of such rotation rates (such as
4:3 or 7:5) encounters synchronism issues, expressed as ‘‘the ratio of the periods.’’
In the worst case, coincidence (intercept) may never occur, or occur in such long
time periods that fast interception requirements could not be met. Reference [6]
examines the problem of pulse synchronization with pairs of pulse trains, and
derives a theoretical approach supported by Monte Carlo simulations.
The theory [6] is constrained to the intercept probability of two pulse trains
with specific reference to a swept SSH Rx and a pulsed emitter. The approach is
based on the concept of pulse clusters overlapping in phase space (i.e., based on
the relative start times or phases between the various pulse trains). Based on this,
exact expressions are developed for the POI after a given number of sweeps,
averaged over all relative start phases of the transmitter’s pulse train and the
receiver’s pulse (or sweep) train.
Given the pulse train characteristics are [T1 , ␶ 1 , T2 , ␶ 2 ], the analysis [6] starts
by non-dimensionalizing this set by dividing by T2 , which yields (R, ␶ 1 /T2 , 1, D 2 )
where R = T1 /T2 is the PRI ratio and D 2 = ␶ 2 /T2 . When considering the POI, it
is only the distance from one pulse to the nearest pulse of the other train that is
of interest; the TOA of a pulse can then be considered as a dimensionless quantity
that can be viewed as being equivalent to wrapping or folding the TOA sequence
around a cylinder of circumference 1. The phase is chosen such that the initial
pulse from train 1 falls at zero; all pulses from train 2 will fall at the same point
given a fixed PRI (analysis considers this as a series of spikes) while pulses from
train 1 are distributed on the interval [0, 1]. After n pulses from train 1, the region
covered by pulses represents the set of positions of the train 2 spike, which leads
to interception within n pulses. The POI is the probability that the spike generated
by train 2 intersects a region covered by pulses from train 1. Regarding the folded
interval concept, then, this probability is equivalent to the ratio of the length
covered by pulses from train 1 to the length of the interval. As shown in [6], the
POI can be considered within four distinct regions:
(j)
Pn = Pn j = 1, 2, 3, 4 (4.19)
where the magnitude of j defines the region of interest dependent on the magnitude
of the sweep number n relative to certain critical values which may vary with the
PRI ratio and P1 [T1 /T0 ].
These four regions can be expressed as follows, where n c = number of clusters:
(1)
Pn = nP1 n ≤ nc (4.20)
(2) (1)
Pn = Pn + (n − n c )⌬ nc ≤ n ≤ n ′ (4.21)
(n − n ′ )
(3) (2)
Pn = Pn ′ +
nc
冠1 − Pn(2)′ 冡 n ′ ≤ n ≤ n ′ + nc (4.22)
(4)
Pn = 1 n > n ′ + nc (4.23)
and ⌬ represents an amount that each new pulse adds to the phase length of its
cluster; for an exact coincidence, ⌬ = 1/2 − | n cR mod1 − 1/2 | . From [6], Figure
4.4 shows the POI plotted for P1 = 0.1, with sweep numbers ranging from n = 1
to n = 12 as a function of the PRI ratio R. It shows a number of interesting features:
• A number of plateau regions of extended maxima—these are obtained when

n ≤ n c and where (4.20) holds.
Figure 4.4 POI between uniform (periodic) pulse trains as a function of PRI ratio, R, and pulse
number (or sweep number) from pulse train 1. (From: [6].  1996 IEEE. Reprinted with
permission.)
• The extended maxima are separated by broad peaked minima which occur
when n > n c and (4.21) holds, or if n > n ′ and (4.22) is valid.
• The width of each minima varies in proportion to P1 .
• When n = 1, Pn is given by the random phase result; for n > 1, the random-
phase theory generally underestimates the POI when it is in a plateau and
overestimates the POI when it is approaching a minimum. As n increases,
the density of the minima also increases.
• Other minima of Pn (R) associated with Pm (R) for m < n are also present.
Figure 4.4 can be useful in selecting the optimal search parameters against a
given transmitted PRI in order to maximize the POI.
However, this approach has limitations against unknown pulse train character-
istics when using a uniform pulse train on the receiving system. Adding jitter to
the search pulse train is suggested in [6] in order to smooth out regions of destructive
synchronization (i.e., where no intercept will happen or not for very long periods).
Adding a small amount of jitter in a cumulative fashion on each pulse is also
suggested in [6]. Based on random walk theory, the position of each pulse then
performs a random walk about a mean TOA position with a step distribution
given by the distribution of jitter on each pulse. From [6], for a uniform step
distribution ranging from −J/2 to +J/2, where J is defined as the jitter amplitude,
the variance of the mean PRI of the pulse train is given by
␴ 2 = nJ 2/3 (4.24)
Since jitter is assumed cumulative, the variance on the TOA of the nth pulse
increases linearly with n, and so the pair of pulse trains are more likely to satisfy
the requirements of the random phase assumption as the sweep number increases.
From [6] and using Monte Carlo simulations, Figure 4.5 shows POI as a function
of the mean PRI ratio [R] and the jitter amplitude on T1 , measured in units of T 2 ,
for n = 4 sweeps and P1 = 0.1. Jitter was added to T so that the PRI varied over
the range T1 (1 − J/2) to T1 (1 + J/2). Each point on the graph shows the mean
POI in 4 sweeps. When the jitter amplitude is zero, it is seen that Figure 4.4 is
derived; as the jitter amplitude increases, Pn (R) approaches the value given by the
random-phase theory earlier. As [6] points out, quite small values of J can move
the POI as derived from this cluster theory to that of the random-phase model as
a function of P1 . Also, as expected, the required value for J decreases as the number
of sweeps, n, increases.
In summary, there now exists a more complete and accurate picture of POI
for cases where the phases of pulse trains are nonrandom. As shown in [6], such
occurrences give rise to sharp maxima and minima in POI space, and this can
aid an EW system designer in optimizing search strategies. However, in real life,
synchronization effects can work against the performance of receiving systems,
and the authors propose introducing a jitter mechanism to smooth out such cases.
This, in turn, is shown to link directly back to the random-phase case of earlier
POI publications.
4.2.4.2 POI and Number Theory Application

In [4, 7] the mathematical background is developed even further for the two periodic
pulse train cases and beyond. The applicability of number theory to a range of
POI problems involving pulse trains is demonstrated in [7], where (1) phases are
known and equal; (2) known and unequal; and (3) one or both are random variables.
The problem becomes finding an algorithm for computing when the first intercept
occurs and then when subsequent intercepts will occur. Key in this analysis is that
the ratio of the PRIs (principal focus is for two pulse trains only) is not restricted
to rational numbers. Through formulating the problem as a Diophantine approxi-
mation (i.e., based on only integer solutions), [7] demonstrates how number theory
can directly assist in the intercept problem.
Figure 4.5 Effect of adding jitter. POI as a function of PRI ratio and jitter amplitude for sweep
number n = 4 on pulse train 1. (From: [6].  1996 IEEE. Reprinted with permission.)
While mathematically intensive, [7] contains an elegant theoretical basis for

pulse train synchronism. Indeed, a recursive algorithm is outlined showing how
POI is related to the Farey point notation, where the Farey series F N is the set of
all fractions in lowest terms between 0 and 1 whose denominators do not exceed
N, arranged in order of magnitude. For example, F 6 is given by [8]:
0 1 1 1 1 2 1 3 2 3 4 5 6
, , , , , , , , , , , ,
6 6 5 4 3 5 2 5 3 4 5 6 6
The mean time to intercept for the discrete time case can be expressed as [7]:
E (N ) = {T2 + q(␶ ) A(␶ )2 + a(␶ ) Q (␶ )2 + 2[a(␶ ) + q(␶ ) − ␶ ] A(␶ ) Q (␶ )}/2T2

(4.25)
Using the above, plus a range of other theorems and definitions contained in
detail in [4, 7], the POI can be determined over the interval
T1 ∈ [ p′T2 /p , q′T2 /q ] (4.26)
other parameters being held constant. Based on being given two adjacent Farey
points and their left/right parents, Figure 4.6 shows the POI as a function of the
number of pulses N from the first pulse train, its PRI T1 with T2 = 1 and ␶ = 0.1.
The diagram shows the rise of POI from small values of N to unity; also, at various
PRIs, there are deep troughs which are centered around the Farey points (conversely,
this is where the mean time to intercept approaches infinity). Figure 4.7 shows
how Figure 4.6 can be interpreted as regions [7]. For any given N and T1 , it shows
Figure 4.6 Probability of intercept as a function of PRI ratio. (From: [7].  2004 Association of the
Old Crows. Reprinted with permission.)
Figure 4.7 POI regions as a function of PRI ratio. (From: [4].  1996 IEEE. Reprinted with permission.)
how the POI lies in the region of initial growth, single or double overlap or
probability 1.
Taking a specific example [4], Figure 4.8 shows the intercept time between
two pulse trains—the duty cycles for pulse trains 1 and 2 are 0.13/0.26, respectively.
The PRI of train 1, T1 , is held constant with T1 = 1 and T2 allowed to vary between
0.1 and 4. Intercept time is shown as the solid line, and it can be seen how it goes
to infinity at the synchronization ratios. The theoretical lower limit of intercept
time given by
T1 T2
T0 =
(␶ 1 + ␶ 2 )
Figure 4.8 Intercept time as a function of PRI 2. (From: [4].  1996 IEEE. Reprinted with permission.)
is also shown as the dotted line where the predicted intercept time approaches T0
at several points. Using such a formalism, it is possible to assist in deriving receiver
characteristics such as sweep time that will minimize T0 against known radar
transmitter scan characteristics. Ongoing work into three window intercept cases
is suggested at the conclusion of [4].
4.3 Summary
Since [2], there have been some significant developments on the important subject
of POI. Several new publications have appeared indicating both developments of
the (random-phase) window functions model and its application across a variety
of Rx architectures. Tools now exist to support EW system designers beyond the
earlier two or three window function scenarios such as scan-on-scan and stepping
or scanning Rx. This wider understanding of POI has been critical not just to
improve system designs but also to acknowledge that certain conventional window
functions such as scan may diminish in today’s POI analyses (due to the proliferation
of phased array radars), thus requiring more in-depth review and appreciation.
More recently, there have been a number of new publications addressing the
real-life issue of synchronization; this has been described through some elegant
mathematics. Current theory appears well matched to describing the two window
case (such as with a scanning receiver system against a rotating transmitter antenna);
future work is already examining three windows and beyond, together with a
variety of techniques to aid EW system designers. EW practitioners are already
considering practical solutions to correlation; for example, [8] looks at POI issues
for a naval EW system against radar threats incorporating a wide instantaneous
field of view (IFOV) receiver behind a high rotation speed, directional antenna.
Detailed calculations are shown for a scan-on-scan scenario where synchronism
between receive and transmit window functions is removed (thus enabling intercept)
through a variable receiver scan rotation rate.
Overall, it is clear that there now exists strong and clear linkage between
uncorrelated and correlated aspects of POI.
A scenario of a frequency agile emitter and a sweeping narrowband receiver
is analyzed in some detail in Appendix C. The random frequency agility allows
the use of conventional probability concepts to characterize the data provided at
the receiver output.
References
[1] Self, A. G., EW Design Engineers’ Handbook, Norwood, MA: Horizon House, 1990.
Artech House, 1985.
[3] Self, A. G., and B. G. Smith, ‘‘Intercept Time and Its Prediction,’’ IEE Proceedings,
Vol. 132, Pt. F, No. 4, July 1985.
[4] Vaughn, I., et al., ‘‘Number Theoretic Solutions to Intercept Time Problems,’’ IEEE Trans.
on Information Theory, Vol. 42, No. 3, May 1996.
4.3 Summary 115
[5] Self, A. G., ‘‘Intercept Time and Its Prediction,’’ Journal of Electronic Defense,
August 1983.
[6] Kelly, S. W., G. P. Noone, and J. E. Perkins, ‘‘Synchronization Effects on Probability of
Pulse Train Interception,’’ IEEE Trans. on Aerospace and Electronic Systems, Vol. 32,
No. 1, January 1996.
[7] Vaughn, I., and L. Clarkson, ‘‘The Farey Series in Synchronization and Intercept Time
Analysis for Electronic Support,’’ Trans. of the AOC, Vol. 1, No. 1, October 2004.
[8] Sullivan, W. B., ‘‘A Simple Tune: Single-Channel Naval ESM Receivers Remain Popular
and Effective,’’ Journal of Electronic Defense, November 2003.
CHAPTER 5
Antennas and Direction Finders
The antenna is a vital link in the reception process. It is the transducer to convert
electromagnetic radiation propagating through space into a microwave signal that
can be processed by a receiver. It is also a signal processor ahead of the receiver.
The antenna can be used to discriminate against some signals and enhance others
based on angle of arrival and polarization. It is also the key element in the process
of measuring angle of arrival or polarization.
The window functions described in Chapter 4 are often a result of directional
antennas being used to search for signals over an angular region. Furthermore, the
power available at the intercept receiver input is a function of the antenna gain.
5.1 Omni-Directional Antennas
An idealized isotropic antenna responds equally well to a signal from any direction
of arrival over a full sphere of coverage. Such an antenna does not actually exist,
but the performance of real antennas is often measured with respect to such an
idealized isotropic antenna. The gain of an antenna is the ratio of the signal power
output of the antenna to the signal power from an ideal isotropic antenna. This
quantity is expressed in decibels with respect to an isotropic antenna as dBi.
Since no actual isotropic antenna exists, standard antennas with calibrated
gain are used to provide the reference levels. ‘‘Omni-directional’’ antennas are often
used in signal interception applications. Omni-directional in this context means
approximately equal response to signals arriving from any direction in one plane.
For example, a vertical monopole antenna above a ground plane will respond
equally to signals from any azimuth angle at a fixed elevation angle, with the peak
response at a 0° elevation angle (i.e., toward the horizon) and no response at all
to a signal arriving from straight above the antenna. If the isotropic antenna’s
response is thought of as a sphere, then any other antenna pattern can be thought
of as deforming the sphere. A high gain antenna pattern can be thought of as being
formed by squeezing a spherical isotropic pattern until the entire volume of the
unit sphere extends outward only in a narrow angular region.
5.1.1 Omni-Directional Antenna Applications

While the low gain of omni-directional antennas reduces the received power, this
is not a significant drawback in some applications. For radar warning receiver
117
118 Antennas and Direction Finders
applications, the signal intensity from the radar is generally quite high, and the
characteristic of most interest is the ability to provide a warning regardless of the
direction from which the signal arrives, and also to give approximate indication
of the direction of the threat. A typical example system uses four broad beam
(∼90°) antennas. By comparing the signal amplitudes, the system can determine
direction of arrival.1
Of more direct interest for a nontactical application of broad beam antennas
is to cover a large angular sector so that no signals are missed (i.e., to achieve
100% probability of intercept). A circular array of broad beam elements with
appropriate RF signal processing can determine the direction of arrival of a single
radar pulse over a broad frequency band. (See Section 5.4.3 for more details.)
Finally, there is the combination of an omni-directional antenna with a directional
antenna for purposes of eliminating those radar pulses intercepted through the
sidelobes of the directional antenna. In such a system, the amplitude of the received
signal from both the omni-directional antenna and the high gain direction-finding
antenna are compared. If the signal from the omni-directional antenna is larger,
the conclusion is that the pulse was received from other than the direction in which
the directional antenna is pointed. This is illustrated in Figure 5.1.
Yet another application of broad beam antennas is to use them on the same
antenna pedestal with a narrow beam antenna. The narrow beam antenna can be
used to track a moving target via radar, while the associated intercept antennas
with broad beams receives any signals from the general vicinity of the target; as,
for example, when an aircraft fires a missile, the guidance signals (if any) can be
intercepted for only a short time. The tracking antenna ensures that the missile is
in the field of view of the intercept antennas at the time of launch.
5.1.2 Parameters for Omni-Directional Antennas

The significant parameters that define the performance of broad beam antennas
are the gain and the capture area. The directive gain of an antenna is defined as
the ratio of the maximum radiation intensity to the average radiation intensity
Figure 5.1 Omni-directional antenna used to control the field of view.
1. The AN/ALR-56 is an example of a warning receiver with such an arrangement of antennas [1].
5.1 Omni-Directional Antennas 119
over a full sphere. The power gain is defined as the ratio of the maximum radiation
intensity to the radiation intensity of an ideal (no loss) isotropic antenna. If the
antenna has no loss, the directional gain is equal to the power gain. Antenna theory
makes use of the concept of a current element, which consists of a very short
(relative to one wavelength), very thin conductor carrying a uniform current over
its short length. The gain of such an antenna (which cannot be realized in practice)
is 1.5 [2, p. 20]. Since neither the isotropic radiator nor the current element is
realizable, it is convenient to use another type of antenna as a standard. One such
standard is a small dipole (or doublet), which also has a gain of 1.5. If the dipole
is one-half wavelength long, the gain is 1.64. Using the method of images, a
monopole above a ground plane has the same pattern (in one hemisphere) as a
dipole but no radiation in the other hemisphere. Hence, its gain is twice that of
a dipole. These gains and other parameters are given in Table 5.1 [3].
The radiation resistance is the real part of the ratio of the voltage to the current
at the antenna terminals. The impedance of a short dipole is capacitively reactive
(the current leads the voltage) and its radiation resistance decreases as the square
of the length. As the length increases to ␭ /2, the capacitive reactance vanishes and
the radiation resistance becomes 73⍀.
The capture area of an antenna is easier to visualize when the antenna consists
of a large reflector or array (dimensions are large compared to a wavelength), when
the electrical area and physical area are approximately the same. The capture area
is found by measuring the power density (watts/meter squared) at the antenna and
Table 5.1 Omni-Directional Antenna Parameters

Gain Radiation Capture
Antenna Type a Numeric dB Resistance (⍀) Area (␭ 2 )
Isotropic 1.0 0.0 — 0.08
Current element 1.5 1.76 80␲ 2 (l /␭ )2 0.12
Short dipole 1.5 1.76 20␲ 2 (l /␭ )2 0.12
c
Short dipole using local ground 6.0 7.78 0.48
reflectionb
␭ /2 dipole 1.64 2.15 73 0.13
␭ /2 dipole using local ground 6.56 8.17 ≈ 73° 0.52d
reflection
Short monopole: 3.0 4.77 40␲ 2 (h /␭ )2 0.24d
Transmitting—all cases
Receiving—space wave
Short monopole receiving a single 0.75 −1.25 80␲ 2 (h /␭ )2 0.06d
ray (surface wave)
␭ /4 monopole 3.28 5.16 36.5 0.26d
Transmitting—all cases
Receiving—space waves
␭ /4 monopole receiving a single ray 0.82 −0.86 36.5 0.07d
(surface wave)
a
Unless noted otherwise, all antennas, except monopoles, are assumed situated in free space. Monopoles must be
located on a ground plane of substantial area.
b
Gain values for antennas that employ a ground reflection apply to the maximum of the resulting elevation
angle radiation pattern.
c
Depends on height above ground and ground characteristics [2, p. 303].
d
These values of capture area and effective height apply to the space wave impinging on the antenna-ground
combination on which gain is based.
Source: [3].
then measuring the received power delivered to a matched load. Since the total
power received is equal to the power density multiplied by the capture area, the
capture area is defined as the ratio of the received power to the incident power
density.2 Using this idea, the capture area for any antenna can be computed. The
capture area of an isotropic antenna is
␭2
Ai = ≅ 0.08␭ 2 (5.1)
4␲
and the capture area of the other antennas in Table 5.1 is simply the gain times
the capture area of the isotropic radiator.
Effective height (h eff ) is a concept that is applied at lower frequencies (< 10
MHz). In this case, the incident power density is often expressed in terms of the
electric field strength (volts/meter). Then the voltage at the antenna terminals
(assuming a matched load) will be
V = E ⭈ h eff (5.2)
The incident power density in a plane wave is given by the cross product of ExH,
which in free space is | E | 2/120␲ in the MKS system of units. The voltage in (5.2)
appears across the radiation resistance if the antenna is connected to a matched
load. Since half the received power is delivered to the load, the total received power
is 2V 2/R,
2(Eh eff )2 E 2A
= (5.3)
R 120␲
or
2
240␲ h eff
A= (5.4)
R
Substituting the radiation resistance for the short monopole from Table 5.1 of
40␲ 2 (h/␭ )2 verifies that the effective area of the short monopole is
␭2
A = 1.5 (5.5)
4␲
A surface wave is one traveling from one monopole to another (i.e., both
monopoles are above the same ground plane, possibly the Earth). Hence the distinc-
tion in Table 5.1 between ‘‘space waves’’ and ‘‘surface waves.’’ The presence of
the infinite ground plane assumed for the monopole makes it impossible to also
assume free-space conditions of propagation [3].
In the development of intercept systems, especially for aircraft, it has been
important to obtain broadband performance over 360° in azimuth with small
2. This assumes the polarization of the antenna is matched to that of the incident wave.
antennas, preferably flush-mounted to avoid drag. Such antennas can achieve multi-
ple octave bandwidth, 60° to 90° of elevation coverage, and nominal gains of −3
to +5 dBi using a cavity-backed spiral design. Other flush antennas are of the
microstrip type [4].
A typical nonflush mounting omni-directional antenna is the conical spiral.
The patterns for such an antenna at two frequencies, 2 and 12 GHz, are shown
in Figure 5.2.
5.2 Directional Intercept Antennas
In many microwave applications, directional antennas are the general rule. Certainly
most of the radar signals to be intercepted are transmitted by high gain, very
directional antennas for at least two reasons: (1) the radar needs the gain to increase
the target return, and (2) the directionality is needed to find the bearing to the
radar’s target. Electronic intelligence operations may need high gain antennas
for the same reasons. From the interceptor’s viewpoint, another reason to use a
directional antenna is to eliminate signals arriving from directions other than the
sector of interest. In an active radar environment, a directional antenna may be a
very important signal processing aid to reduce interference. Suppose the weakest
radar signal at the horizon is well above the threshold level of the intercept receiver.
Then an antenna beamwidth of 3.6° can reduce the number of signal pulses being
received by a factor of approximately 100 compared to an omni-directional
antenna.
Compared to a radar system, an intercept system will generally use a lower
antenna gain. While the radar designer may choose antenna gain in excess of 30 dB,
Figure 5.2 Antenna pattern for a broadband conical spiral.

the typical intercept antenna ‘‘high gain’’ may be 15 to 30 dB, depending on the
frequency of interest. Since the interceptor designer is concerned with frequencies
over a much wider band than is a radar designer, the intercept antenna may cover
several octaves (e.g., 1 to 12 GHz). Broadband directional antennas can be made
using a broadband omni-directional antenna as the feed for a parabolic dish. This
means that the gain increases in proportion to the square of the RF since the
aperture area is constant. Hence, a 10 to 1 frequency band implies a 20-dB gain
variation and a 10 to 1 beamwidth change.
Broadband coverage can also be achieved using horn antennas with internal
ridges or a dielectric lens over the mouth of the horn to provide phase correction
across the aperture. A typical horn of this type, covering 1 to 11 GHz and having
an aperture of about 5 × 7 inches, has gain ranging from 2 to 20 dBi. Figure 5.3
shows the pattern expected from such a horn antenna.
These directional antennas, when used for angular search, can be rotated at
much higher rates than their search radar counterparts. First, their broader beam-
width means that a given dwell time in a given direction can be achieved with a
higher rotation rate. Second, the dwell time in a given direction can probably be
less than that used by the radar for the simple reason that there is often a much
higher power density at the intercept antenna compared to the weak target echo
at the radar antenna. For this reason, antenna scan rates of hundreds of revolutions
per minute may be used by the intercept system designer, whereas the radar designer
(for search systems) typically chooses rotation rates on the order of 10 rpm (or
even less).
The natural increase of antenna gain with frequency serves a useful purpose
in matching the intercept receiver to the signal environment. The one-way range
equation gives the signal power available at the ELINT receiver as
Figure 5.3 Antenna patterns for a broadband horn antenna.

PT G TE GE ␭ 2
SE = 2 2
(5.6)
(4␲ ) R E L T L E
where:
S E = signal level at ELINT receiver (W)

G TE = gain of radar transmitting antenna in the direction of the ELINT receiver
GE = gain of the ELINT antenna
R E = range from radar to ELINT receiver
L E = loss from ELINT antenna to receiver
PT = radar transmitter power (W)
␭ = wavelength
L R = radar transmission losses
If the aperture of the ELINT antenna is fixed, GE is given by
4␲ A E
GE = (5.7)
␭2
Combining (5.6) and (5.7) gives
PT G TE A E
S= 2
(5.8)
(4␲ )R E L T L E
Thus, the frequency (wavelength) dependence of the received power level is

effectively eliminated. There is still a tendency for a reduced received power level
at higher frequencies because these frequencies are generally used for shorter range
(and hence lower ERP) radar systems.
The higher gain of broadband antenna systems at higher frequencies affects
the probability of intercept since the window duration at a given search (scan) rate
is reduced in proportion to the reduction in the beamwidth.
The discussion of antenna beam patterns in Chapters 8 and 10 concerning
radar antennas applies equally well for ELINT antennas, as does the polarization
discussion in Chapter 8. In particular, ELINT antennas should be equipped to
mechanically rotate or to switch polarizations, both to aid in determining the
polarization of the intercepted signals and maximize the received signal level.
Sidelobe levels are not as critical to intercept system designers as they are to
radar system designers. The ELINT station can be sure that a particular signal is
entering only the main beam by attenuating the received signal until it can be
received only at a particular angle. Likewise, the typical ELINT system is not
subject to jamming, spoofing, or angle deception signals, all of which cause radar
designers to try reducing the sidelobe levels. The combination of an omni-directional
antenna whose gain is above the sidelobe levels of the directional antenna can be
used to eliminate errors due to sidelobe reception. Of course, the ELINT antenna
should be located so that there are no severe nulls or lobes caused by the antenna’s
surroundings (e.g., adjacent antennas, masts, buildings, and so on).
The use of radomes to cover the ELINT antenna apertures can significantly
affect performance. The fact that broadband radomes are needed for ELINT systems
means that radome loss can vary significantly across the band. Such losses and
gain variations must be thoroughly documented if accurate measurements of the
received power level are to be made. As noted in Chapter 7, estimating the radar’s
ERP is a part of the interceptor’s task, which requires very good calibration of the
antenna gain versus frequency, as well as accurate pointing of the antenna toward
the emitter of interest. As shown by (5.6), if S E is measured, and if the ELINT
antenna gain and system losses are known, and if the range to the emitter is known,
the radar’s ERP can be determined.
5.3 Direction Finding
Determining the direction or angle of arrival of a radar signal is fundamental to

electronic intelligence. The AOA is particularly important in an era of radar parame-
ter modulation and random variation: the AOA is not subject to control by the
radar designer. Direction finding is an extensive subject in its own right. The brief
discussion here points out the fundamental ideas and principles involved. Also,
some examples are provided.
Historically, direction finding goes back to the early 1900s when long-range
communication in the HF band was already possible. Even earlier, Marconi had
used copper parabolic mirrors to increase range to 2 miles. Before 1893, Hertz
used cylindrical parabolic mirrors to focus energy from his transmitter. (These
early experiments actually used a frequency of about 200 MHz. Later, the HF
band was used because of the long ranges made possible by ionospheric propagation
[5].) Since the majority of ELINT signals are in the microwave region of the
spectrum, the large body of knowledge concerning direction finding for HF signals
is of limited interest. The most common technique from the ELINT viewpoint uses
a rapidly spinning reflector, with ‘‘instantaneous’’ direction finding techniques used
in more modern systems.
A third approach, using time difference of arrival over a relatively long baseline,
may also be used (for emitter location purposes). This is included in Chapter 6
because the role of the antenna in this process is not crucial.
The traditional direction finding unit used in the past on U.S. ships is the AS-
899F, a part of the WLR-1 receiving system [6]. The key specifications of this
antenna are given in Table 5.2. Figure 5.4 is a photograph of the antenna system
with the radome removed. Note that only the reflector rotates. It is mounted on
a slant above the antenna feed, which is aimed vertically at the reflector. This
avoids the need for rotary joints. The window functions (see Chapter 4) that such
a system can achieve at the highest scan rate of 300-rpm range from a window
duration of 22 ms at the low end of the band to 2.2 ms at the high end of the
band (40° beamwidth at 1 GHz, 4° beamwidth at 20 GHz), with a period of 200
ms. The 200 ms required for 360° rotation is quite fast compared to the scan of
5.3 Direction Finding 125
Table 5.2 Specifications of a Typical Direction Finding Antenna for Electronic Intelligence
Applications
ANTENNA AS-899F /SLR
TYPE/USE: The AS-899F is a rotating, very broadband microwave direction finding antenna
designed for use with such counter measures receiving equipment as the AN/WLR-I where it is
responsive to signals in bands 6, 7, 8, 9, and 10 (1 to 20 GHz). The antenna elements consist of a
single, stationary spiral feed, a multiplexer, and a rotating, shaped reflector. (Earlier models of the
AS-8OO employed a more complex system of four feed horns, two mounted coaxially at the top
center and two coaxially centered at the bottom, with a dual-surface parabolic reflector oriented at
45° in between. These antennas, which were limited in frequency response to bands 6, 7, 8, and 9
of the WLR-I, were subsequently modified to be equivalent to the extended range AS-899F.) The
antenna is capable of being operated in an automatic, continuous rotation mode with speeds
variable from 0 to 250 rpm, or manually positioned to any desired bearing. To protect it from a
sea environment, the antenna is enclosed within a radome.
DIMENSIONS: Height: 82 inches (with radome); Diameter: 34 inches (radome); Weight: 268
pounds (total).
FREQUENCY RANGE: 1 to 20 GHz.
INPUTIMPEDANCEIVSWR: 50 ohms. VSWR not greater than 4:1 over entire frequency range.
RF POWER RATING: N/A (receive only).
POLARIZATION: Right-hand circular or any sense of linear polarization.
TYPE OF FEED: Unbalanced, three coaxial cables and one waveguide for RF.
PRIMARY POWER REQUIRED: Servo power and control voltages supplied from control unit;
115 vac, 60 Hz, one-phase required for space heaters.
INPUT CONNECTOR: Coaxial Type N. (Waveguide WR-62 for highest band.)
BEAMWIDTH: 40° on lowest band to 4° on highest band.
AZIMUTH SCAN: 360° mechanical.
SCAN RATE: 300 rpm maximum, CW or CCW.
INSTALLATION REQUIREMENTS: The antenna assembly should be mounted as high as possible
on the ship’s superstructure in order to obtain maximum omni-directional coverage. The assembly
base casting has four 1-inch bolt holes on 19.25-inch centers for mounting purposes. A clearance of
12 inches is required beneath the base for cover removal.
SPECIAL CONSIDERATIONS: In order to reduce the possibility of receiver overload damage, the
AS-899F /SLR should not be located in the strong fields of transmitting antennas.
REFERENCES: Technical Manual for Antenna Assembly AS899E/SLR and AS-899F/SLR,
NAVELEX 0967-550.3010.
Source: [6].
most radars. Thus, the DF antenna rotation essentially measures the radiation from
such a radar as if the radar antenna had stopped its scan. As a result, if the
receiver sensitivity allows receiving the radar’s sidelobes, an operator can determine
direction of arrival at all times by noting the angle at which the strongest signal
occurs, whether or not the main beam of the radar scan coincides with the scanning
of the DF antenna. (Of course, the idea of a snapshot of the radar signal while its
antenna has stopped scanning is only approximately true. Since the radar’s own
antenna pattern will have a complex structure of sidelobes and nulls, the signal
level will also be changing during the 200-ms rotation period of the DF antenna.)
The problem of signal sorting is handled in the normal ways at the receiver
(i.e., either a narrowband receiver is used or time domain PRI gating is used, or
the two may be combined). Also, an omni-directional antenna can be used to
eliminate pulses arriving through the sidelobes of the DF antenna. The DF display
Figure 5.4 Photograph of rotating DF antenna (radome removed).
can consist of a polar plot of the signal amplitude versus the bearing angle to which
the spinning reflector is aimed at each instant. A particular signal may be studied
by stopping the antenna scan. Its beam can then be pointed toward a particular
bearing. Also, a sector search mode can be used to make it easier to read the
absolute bearing angle from an expanded angular scale.
Notice that the specifications in Table 5.2 include the capability for right-hand
circular polarization and any linear polarization. Clearly, the polarization of each
signal will need to be investigated along with its direction.
5.4 Instantaneous Direction Finding
Rapid rotation of high gain antenna apertures is possible in ground and ship-borne
receiving systems. However, this type of direction finder is not suited for use on
aircraft since it occupies too much space. In addition, the complexity of the radar
environment may make it very important to be able to measure the AOA of a
single radar pulse.3
The AOA of a single pulse can be measured in several ways. Multilobe amplitude
comparison makes use of antennas with their beam peaks (boresights) aimed in
different directions (squinted). The amplitude is measured independently in each
antenna, and the AOA is calculated as being between the boresight angles of the
3. The main drawback to all instantaneous AOA techniques is the error caused if more than one signal is
present during the measurement.
pair of antennas having the largest amplitudes. The actual angle is computed based
on the ratio of the amplitudes.
Another technique is to use phase comparison. This is called an interferometer.
One advantage to the interferometer is that the use of nondirectional antennas
means that off-axis AOA measurements do not suffer from a reduced SNR as they
do in amplitude comparison system. In practice, similar broad beam antennas are
often used for either amplitude comparison or an interferometer, and thus the
interferometer may also suffer an SNR reduction for off-axis AOAs.
5.4.1 Amplitude Comparison AOA Measurement

The multilobe amplitude comparison system concept is shown in Figure 5.5(a) for
four receiver channels. (Any number of antennas can be used, but four or eight
are common choices for 360° azimuth coverage.) The individual channel amplitudes
are measured, as are the difference shown in Figure 5.5(a) to allow selection of
the channel pair having the largest and next-to-largest amplitudes. Figure 5.5(b)
Figure 5.5 (a) Four-channel multilobe AOA system concept: (b) antenna patterns superimposed;
and (c) amplitude comparison system angles defined.
Figure 5.5 (Continued.)
shows the type of antenna patterns needed, with the boresight direction of each
antenna offset by 90°. Figure 5.5(c) shows one pair of patterns with definitions
for the angles needed in the subsequent analysis.
The angular error in the AOA is a function of the slope of the ratio of the
antenna patterns to the AOA. Assuming channels 1 and 2 are selected, we have
G 1 (␪ )
R (␪ ) = (5.9)
G 2 (␪ )
or, in decibels,
R dB = G 1 dB − G 2 dB (5.10)
The variation in the measured AOA is related to the variation in the ratio
measurement by
␴ R dB
␴␪ = (5.11)
∂R dB /∂␪
where:
␴␪ = rms bearing or AOA error (degrees)

␴ R , dB = rms error in measuring R dB (decibels)
∂R dB /∂␪ = rate of change of R dB with respect to the AOA (decibel/degree)
To relate bearing errors to amplitude errors, it is convenient to assume (voltage)

antenna patterns that are Gaussian in shape [7, 8]; that is,
2
冋
G 1 (␪ ) = A 1 exp −
k 2(␪ − ␣ )2
␪ B2
册
2
冋
G 2 (␪ ) = A 2 exp −
k 2(␪ + ␣ )2
␪ B2
册
where:
G 1 and G 2 = offset antenna gain patterns

A 2 and A 1 = square root of antenna gain at peak or boresight (relative to
isotropic)
␣ = squint angle, (i.e., one-half of the angle between the boresight
directions of the two antennas)
␪ B = 3-dB beamwidth
k 2 = (2 ln 4), so that G(␪ B ) = 0.5 (so that the crossover point of the
patterns occurs at the −3-dB point)
While the assumptions that the antenna patterns are Gaussian is clearly not
realistic for the sidelobe regions of the patterns, the interest in this case is in the
main beam patterns not many decibels down from the peak. In this case, the
Gaussian assumption is not too far from reality.
Then, R dB is given by
G 1 (␪ )
R dB = 10 log (5.12)
G 2 (␪ )
A k2
冉冊
= 20 log 1 − 10 (log e) 2 [(␪ + ␣ )2 − (␪ + ␣ )2 ]
A2 ␪B
The partial derivative is then
∂R dB
∂␪
= 10 (log e) 冉冊
4k 2␣
␪ B2
= 10 (log e) (8 ln 4) 冉冊
␣
␪ B2
For the case where ␣ = 0.5␪ B , we have
∂R dB
= +10 (log e) (4 ln 4)/␪ B
∂␪
≈ 24 dB/beamwidth
The rms bearing can now be related to the SNR using (5.1). Suppose the
amplitude in each channel is subject to a measurement error ⌬A due to noise. In
(5.12), the term log (A1 /A 2 ) does not affect the partial derivative. However, it
contains the errors due to noise. Even with perfectly matched channels (or a
technique to correct for any mismatch), thermal noise causes the measured values
of the channel outputs to fluctuate. The change in the ratio caused by amplitude
variations due to noise is
⌬R dB = (20 log e) 冉 ⌬A1 ⌬A 2

A1
−
A2 冊
Since A1 and A 2 are the signal voltages at the peaks of the antenna patterns
in channels 1 and 2, respectively, the error in R dB is related to the SNR in each
channel.
⌬A1 1 ⌬A 2 1
= and = (5.13)
A1 √2 SNR 1 A2 √2 SNR 2
Therefore, assuming the noise in the two channels is uncorrelated,
␴ R dB = 20 (log e) 冉√ 1
2 SNR 1
+
1
√2 SNR 2 冊 (5.14)
Combining (5.14) with (5.11) gives
冉冊
1/2
1 1
20 log e +
2 SNR 1 2 SNR 2
␴␪ = (5.15)
∂R dB /∂␪
For the Gaussian antenna patterns, the SNR is a function of the angle ␪ . At
the crossover point where G 2 (␪ ) = G1 (␪ ) = 0.5, the SNR in each channel is the
same. If the SNR at the peak of either beam is SNR 0 , then
冉冊
1/2
2
20 (log e) ␪B
SNR 0
␴␣ = degrees
10 (log e) [8 ln (4)]
or
␴␣ = 0.510 冉 ␪B
SNR 0
1/2 冊 degrees
For ␪ B = 90° and SNR 0 = 13 dB, ␴␣ = 10.26°.

Substituting the expression for the SNR, as a function of angle into (5.15)
along with the value of ∂R dB /∂␪ from (5.13), allows a calculation of the rms bearing
error as a function of the squint angle. For Gaussian patterns, the optimum squint
angle is
␪B
␣= = 0.6␪ B
2 √ ln 2
For this case, to allow ␣ = 45° (90° between antenna boresights), the 3-dB
beamwidth can be approximately 75°. For an eight-antenna system, the beamwidth
can be cut in half, which would also reduce the angular error due to noise by a
factor of two at the cost of twice as many antennas (and possibly twice as many
receiver channels).
The performance degrades very slowly for squint angle values other than 0.6
times the beamwidth. Squint angles from 0.4 to 0.8 times the beamwidth give
nearly optimum performance for Gaussian antenna patterns.
Errors are also caused by variation in the antenna boresight as a function of
frequency. This can be a significant problem in wideband systems.
5.4.2 Phase Interferometers

In phase comparison systems (interferometers), the phase difference in two channels
is measured. First consider the interferometer as shown in Figure 5.6. The phase
front of the plane wave arriving at an angle ␪ from the boresight direction must
travel the additional distance 1 = d sin ␪, where d is the distance separating the
two antennas. The resulting phase shift can be expressed as [7]
Figure 5.6 Two-channel interferometer.
2␲ d 2␲ d
␾= sin ␪ = f sin ␪ (5.16)
␭ c
The maximum separation for a total phase change of no more than 2␲ (for
unambiguous angle measurement) is half of the wavelength, for which a phase
change of ±␲ occurs over the angular region ±90°. Equation (5.16) shows that
errors in the phase measurement are directly proportional to errors in the baseline,
d, to errors in the measurement of the carrier frequency, f, as well as to thermal
noise.
To take advantage of baselines longer then ␭ /2, it is possible to use several
baselines, with the shorter ones resolving the angular ambiguities of the longer
ones. For example, a four-antenna configuration using baselines of ␭ /2, ␭ , and 2␭
is a possibility. The phase shift over the longest baseline can then be up to ±4␲
and the angular resolution has been improved by a factor of 4. The size of the
longest baseline is restricted by the resolving power of the shorter ones. For example,
if the (−␲ to +␲ ) range of the ␭ /2 baseline pair can be resolved into ␲ /8 sectors,
the next longer baseline can be up to 8 times as long as the first. In such a system
the problems of matched phase shifts over broad bands in the components becomes
a critical problem.
The effect of thermal noise on interferometric bearing measurement can be
found with the aid of (5.16) and a derivation of phase variation due to noise. First,
note that
∂␪
␦␪ = ⌬␾
∂␾
Using (5.16), the rate of change of the bearing measurement with respect to a
change in phase is
∂␪
∂␾
=
d␪冉冊冉
∂␾ −1
=
2␲ d cos ␪ −1
␭ 冊
A calculation of the rms bearing error ␴␪ in terms of the rms phase error yields
␴␪ = ␴␾ 冉 ␭
2␲ d cos (␪ ) 冊 (5.17)
The rms phase variation due to noise can be computed using the phasor diagram
in Figure 5.7. The quantity of interest is the incremental phase change, ⌬␾ , resulting
from a noise disturbance, ⌬A. Using the law of cosines,
| A | 2 + | A1 | 2 − | ⌬A | 2
cos ⌬␾ =
2 | A | | A1 |
For relatively high signal-to-noise ratios, ⌬␾ will be small so that cos ⌬␾ ≅

⌬␾ /2 and | A1 | ≅ | A | . This yields
Figure 5.7 Phasor diagram to compute phase variation due to noise.

⌬␾ 2 1 | ⌬A | 2
1− ≅1−
2 2 |A|2
or
| ⌬A | 2
⌬␾ 2 ≅ (5.18)
|A|2
Since ⌬A is the instantaneous noise phasor and A is the signal phasor, the
right-hand side of (5.18) averaged over time is the reciprocal of the SNR and the
left-hand side is the variance of the phase measurement:
1
␴␾2 =
SNR
Combining this result with (5.17) yields the rms variation in AOA measurement
due to noise as
␭
␴␪ = (5.19)
2␲ d cos ␪ √SNR
The implementation of a wideband phase interferometer presents the very

difficult task of maintaining knowledge of the phase shift at any frequency of
interest in two receiving channels. In practice, this requires injection of test signals
at the frequency of interest for calibration purposes or some other technique (such
as multiplexing a single receiver among two or more antennas). The variation of
phase with component aging, temperature changes, and maintenance operations
means that mere calibration across the band for looking up in a table will not
allow for AOA measurements of sufficient accuracy in microwave applications.
The need for bearing measurements on a single-pulse basis leads to the consider-
ation of the monopulse angle measurement antenna configurations used in radar
systems. Such systems can be used to obtain accurate bearing information while
remaining open to reception over broad sectors [9]. In these systems, two or four
broadband antennas (e.g., spiral or log periodic) are used, as well as two or
three matched receiving channels. A beam-forming network processes the antenna
outputs to form one sum beam and a difference beam in azimuth and a difference
beam in elevation. The sum beam is used as a reference to which the other beams
are compared, and the AOA in azimuth and elevation is computed by comparing
the signals from the difference channels to that in the sum channel. Either the
signal amplitude or phase (or both) can be used. Achieving broadband operation
means that the same phase or amplitude difference must occur at the same angle
of arrival. This means that the phase centers of the antenna elements must be the
same number of wavelengths apart. This can be accomplished with conical log
spiral antennas by placing the outer (high frequency) end closer together than the
base, as shown in Figure 5.8.
Figure 5.8 A two-element array of conical log spirals (frequency independent).
The signal processing needed is similar to that used in monopulse radar systems,
as shown in Figure 5.9. Typical antenna patterns at the sum and difference ports
are shown in Figure 5.10. The antenna spacing is 0.6␭ in one case shown and 0.8␭
in the other. To achieve frequency independent performance, it is necessary to
maintain antenna spacing that is a fixed fraction of a wavelength at over the
frequency band to be covered. The variation in the indicated bearing angle with
frequency is one of several sources of error in monopulse angular measurement
systems. The error in the phase and amplitude of the microwave beam-forming
hybrids, the imbalance in the receiver channels, and the misalignment of the mechan-
ical components contribute to the AOA error, as does the available SNR. These
effects are summarized in Figure 5.11 [8]. According to the calculations summarized
in Figure 5.11, a phase comparison system with a baseline of three wavelengths
can achieve an rms accuracy of 0.63°, which is generally sufficient for ELINT
purposes. Note that linear arrays of antennas generally cover a sector of less than
90°.
An rms bearing accuracy of 2° means that 99.7% of the measurements will
fall inside a 12° sector. This is sufficiently small to allow a jammer to be directed
towards the emitter or for launching an antiradiation missile using only the informa-
tion provided by a single receiver. ELINT, which is generally not concerned with
responding in real time, can make use of less accurate AOA measurements if there
are a large number of bearing measurements that can collectively narrow the
geographic region in which the emitter is located.
136
Figure 5.9 Block diagram of three-channel monopulse system. (From: [8].  1971 IEEE. Reprinted with permission.)
Figure 5.10 Sum and difference beams from DF system for two-element spacing. (From: [8].
 1971 IEEE. Reprinted with permission.)
Multiple Channel Interferometers 4

Some of the systems described in Figure 5.11 use antennas spaced too far apart
for unambiguous measurement of angles by an interferometer. The desire for greater
accuracy leads to the use of longer baselines. However, the angle indication becomes
ambiguous when the phase shift exceeds ±180°. This leads to the use of additional
antennas and receiving channels. A closely spaced pair of antennas can indicate
the approximate angle, and a larger spacing can indicate the more accurate but
4. The author is grateful to Brian Moore for providing this material.

Figure 5.11 Comparison of rms bearing accuracy (1␴ ) for a ±30° field of view. (From: [8].  1971
IEEE. Reprinted with permission.)
ambiguous angle. The two measurements together resolve the ambiguity. A block
diagram for a dual baseline system (three channels) is shown in Figure 5.12.
Performance is expressed in terms of the ratio of the wide baseline, d w , to the
narrow baseline, d n . This ratio is set to be an integer.
m/n = d w /d n m and n are integers
The unambiguous angular field of view for such a system is given by
Figure 5.12 Dual baseline interferometer.

FOVSYS = 2 arcsin 冉冊
m␭
2d w
, for ␭ < 2d w /m (5.20)
= 180° for ␭ ≥ 2d w /m
or, equivalently,
FOVSYS = 2 arcsin 冉冊n␭

2d n
, for ␭ < 2d n /n (5.21)
= 180° for ␭ ≥ 2d n /n
Of course, the accuracy is determined by the wide baseline, d w :
␴␸ ␭
␴␪ = (5.22)
2␲ d w cos ␪
The maximum phase error allowed to properly resolve the ambiguity is given
by
␲ 1
| ⌬␸ | < radians (5.23)
√2 √(n 2 + m 2 )
Some deployed systems make use of four antennas with three different baselines
and further divide the band covered into high, medium, and low frequency ranges
for a total of 12 antennas.
5.4.3 Bearing Discriminators

The need for 360° azimuth coverage for a direction finding system combined with
the need for instantaneous (monopulse) AOA measurement leads to the idea of a
circular array of antennas. This is a natural extension of the four-antenna system
shown in Figure 5.5(b). We have seen that comparing the phase of the signals from
two separated antennas can be used to indicate the AOA of a signal (Figure 5.6)
and that the phase difference can be increased in the signal processing (Figure 5.7)
to provide an increased electrical phase change for a given angular change if the
SNR is sufficient. We have monopulse angle measurements. This sets the stage for a
discussion of a circular array of antennas, which can achieve accurate, instantaneous
bearing measurements over wide bandwidths for a full 360°. This bearing discrimi-
nator makes use of high-resolution phase measurements using the same kind of
phase discriminator, which is also the heart of the instantaneous frequency measur-
ing receivers (IFMs). The only drawback of this approach (as with all of the
instantaneous AOA techniques) is the error caused by the presence of more than
one signal at the time the angle is measured. In this situation, if one signal is much
stronger (e.g., > 6 dB) than any other, the angle of the strongest signal will be
measured; however, with increase error due to the presence of the interfering signal.
To start, consider a phase comparison version of a four-antenna system (similar
to Figure 5.6). As shown in Figure 5.13, the output port (labeled n = 0) is the
Figure 5.13 Four-antenna bearing discriminator block diagram.
reference signal. The north antenna output passes through the two left 180° hybrids
with no phase shift to port zero. The east antenna output passes through the +90°
shifter, is shifted −90° through the 90° hybrid, and passes through the bottom left
hybrid with no phase shift for a net total of 0°. The south antenna’s signal passes
through both the left hybrids with no phase shift. The west antenna’s signal reaches
output port zero by passing through both the 90° hybrid and the bottom left hybrid
with no phase shift. Thus, no matter which antenna receives a signal, it reaches
the n = 0 port with 0° of phase shift. Next consider the phase at the n = 1 output
port. The signal from the north antenna reaches this port with no phase shift,
passing directly through the two left hybrids. The signal from the east antenna
reaches the n = 1 output port by passing through the 90° phase shift and then
directly through the two right hybrids for a net +90° phase shift. The south antenna’s
signal reaches the n = 1 port via 180° phase shift in the top left hybrid and then
directly through the bottom right hybrid for a net 180° phase shift. The west
antenna’s signal reaches the n = 1 output port with 90° phase shift through the
90° hybrid and 180° phase shift through the bottom right hybrid for a total of
270°. The results are shown in Table 5.3.
As shown, the phase difference between the n = 0 and n = 1 ports is equal to
the spatial angle around the circular array. Note that only one-fourth of the power
from one antenna is available at a given output port due to the power dividing
properties of the hybrids. Also note that the phase difference between the n = 0
and the n = 2 port is exactly twice the phase difference between the n = 0 and
n = 1 ports (modulo 360°).
By measuring the n = 0 to n = 1 phase difference, the bearing angle is obtained.
This is true even if the signal is, say, at a 45° bearing angle. Then the north and
east antennas have equal signals and the result is a 45° phase difference at the
output (i.e., equal components at 0° and 90° produce a 45° phase shift). As long
as the antennas produce equal amplitudes, the proper bearing angle can be found
regardless of AOA. If the hybrids and antenna elements are broadband, the AOA
measurement can be made over a broad band.
A network with the properties of the one shown in Figure 5.13 is called a
Butler matrix [10]. The power at any of N input ports divides equally among N
output ports. Furthermore, a signal at one input port results in a signal at each
output with a constant phase difference between them. The output ports are num-
bered 0, ±1, ±2, . . . , up to ±(N − 1) and N. For a 32-port Butler matrix, the
output ports would be numbered 0, ±1, ±2, ±3, . . . , ±14, ±15, ±16. The phase
progression at any output would be its number times the spatial angle (bearing)
of the signal arriving at a circular array of N antennas [9]. This concept is the
heart of a digital bearing discriminator using a circular array of 32 broadband
antennas to achieve a broadband bearing accuracy of 2° rms over a full 360° in
azimuth and −10° to +40° in elevation [11].
A block diagram of this receiver is shown in Figure 5.14. The Butler matrix
outputs 1, 2, 4, and 8 are compared to output 0 using digital phase comparators
to achieve a bearing resolution of 8 bits (or 1.4°). This means the net phase
resolution required from the port 8 measurement is 11.25 electrical degrees since
the phase changes 8 times as much as the spatial angle changes. The bearing
accuracy is maintained over the 2- to 18-GHz band down to −55 dBm relative to
an isotropic antenna and for pulse duration as short as 200 ns.
The antenna array is a stripline configuration with conical reflectors mounted
above and below to shape the beam in elevation. The circular array is shown in
Figure 5.15 and the entire unit is shown in Figure 5.16. This unit has two antenna
arrays. The upper one covers the 7.5- to 18-GHz band and the lower one covers
the 2- to 7.5-GHz band.
Table 5.3 Bearing Discriminator Using Four Antennas

Electrical Phase Shift at Port
Antenna Spatial Angle n=0 n=1 n=2 n = −1
N 0° 0° 0° 0° 0°
E 90° 0° 90° 180° 270° (−90°)
S 180° 0° 180° 0° (360°) 180° (−180°)
W 270° 0° 270° 180° (540°) 90° (−270°)
142
Figure 5.14 Block diagram of digital ESM receiver incorporating a bearing discriminator. (From: [9]. Courtesy of Anaren Microwave.)
Figure 5.15 Circular antenna array (7.5 to 18 GHz). (From: [9]. Courtesy of Anaren Microwave.)
Figure 5.16 Antenna unit (2 to 18 GHz) (radome removed). (From: [9]. Courtesy of Anaren Micro-
wave.)
5.5 Arrays, Lenses, and Subspace DF Methods
As the electromagnetic environment grows in complexity, the need for high quality
AOA measurements on a single-pulse basis increases. Certainly, single-pulse AOA
is critical to automated systems. As the likelihood of multiple received signals
overlapping in time increases, the need for finding the AOA for multiple signals
also grows. One method for doing this is analogous to the channelized receiver.
Instead of multiple channels across the frequency band of interest, multiple antenna
beams can be formed to cover the angular region of interest. Then, independent
receivers and processors can be used to deal with the signals arriving in each beam.
Such a system can be created using a microwave lens. An example is shown in
Figure 5.17.
The use of an array of antenna elements includes the amplitude comparison
and phase comparison (interferometer) systems described above. In the general
case, there are many antenna elements and their outputs are combined in various
ways. One way is to form beams indifferent directions, as is done in the system
shown in Figure 5.17. A general way of looking at this is shown in Figure 5.18
[13]. Here the array outputs (arrayports) are shown feeding into a Lens Transfer
Matrix to produce the lens outputs (lensports). These are processed by CVRs to
provide an output of the power from each lensport. These outputs are the beam-
ports. The Butler matrix is one example of a lens transfer matrix. Another is the
beamformer used to form the beams in the multiple beam array above.
The use coherent processing of the arrayport outputs is the basis of subspace
methods such as the Multiple Signal Classification (MUSIC) algorithm. In this
algorithm, it is necessary to calibrate the system using a variety of angles of arrival
to form what is called the array manifold. Using the eigenvalues and eigenvectors
obtained from the data correlation matrix, one can express the signals in terms of
the larger eigenvalues and their eigenvectors, as indicated in Figure 5.19.
An example of the performance of the MUSIC algorithm is shown in Figure
5.20. In [13] a method is derived for using the outputs of the CVRs in Figure 5.17
to obtain the correlation matrix and hence accurate AOA values. The method
depends on ergodic signals (time average the same as the ensemble average).
Figure 5.17 Lens fed constant beamwidth array antenna. (From: [12].  1984 Horizon House, Inc.
Reprinted with permission.)
Figure 5.18 Lens-based array system with CVRs. (After: [12, 13].)
5.6 Short Baseline TDOA for AOA
Although time difference of arrival (TDOA) methods are usually thought of as

long baseline systems with independent platforms (see Chapter 6), it is also possible
to obtain AOA information by measuring the time difference instead of the phase
difference. (Refer to Figure 5.6.) The extra distance traveled by the signal is d sin
␪ . At the speed of light, this produces a time difference of ␶ = d (sin ␪ )/c. Then
the AOA can be found from
␪ = arcsin (␶ ⭈ c /d ) radians (5.24)
The error in AOA due to an error in time difference is given by the derivative
c[1 − sin2(␪ )]−1/2

⌬␪ = ⌬␶ radians (5.25)
d
For a time difference error of 1 ns and for a 30° angle of arrival and d = 3m,
the AOA error is 6.6°. It is clear that very precise time measurements are required,
which in turn, requires a high SNR and/or a pulse with a steep rise time. Of course,
increasing the baseline improves the accuracy directly. For 30-m baseline, the AOA
accuracy is 0.66°. There is no need to measure the frequency or wavelength with
this technique.
Figure 5.19 MUSIC block diagram. (From: [14].  1986 IEEE. Reprinted with permission.)
Figure 5.20 Example of MUSIC algorithm performance.
References
[1] Braun, A. E., ‘‘RWRs Face New Threats,’’ MSN, November 1983, p. 53.
[2] Kraus, J. D., Antennas, New York: McGraw-Hill, 1950.
[3] Ames, J. W., and W. A. Edson, ‘‘Gain, Capture Area, and Transmission Loss for Grounded
Monopulse and Elevated Dipoles,’’ RF Design, November/December 1983.
[4] Bahl, I. J., and P. Bhartia, Microstrip Antennas, Dedham, MA: Artech House, 1980.
[5] Boyd, J. A., et al., Electronic Countermeasures, Los Altos, CA: Peninsula Publishing,
1978, Chapter 5; originally published in 1961 as a classified book.
[6] Law, P. E., Shipboard Antennas, Dedham, MA: Artech House, 1983, p. 426.
[7] Tsui, J. B., Microwave Receivers and Related Components, Springfield, VA: National
Technical Information Service, 1983, (PB84-1O8711), pp. 33–37.
[8] Bullock, L. G., G. R. Och, and J. J. Sparagna, ‘‘An Analysis of Wideband Microwave
Monopulse Direction Finding Techniques,’’ IEEE Trans. on Aerospace and Electronic
Systems, Vol. AES-7, No. 1, January 1971, pp. 188–203.
[9] Anaren Microwave, Inc., Publication MI826-I8, East Syracuse, New York.
[10] Moody, H. J., ‘‘The Systematic Design of the Butler Matrix,’’ IEEE Trans. on Antennas
and Propagation, November 1964, pp. 786–788.
[11] Rehnmark, S., ‘‘2–18 GHz Digital ESM Receivers with 2 Degree RMS Bearing Accuracy,’’
Military Microwaves ’82, London, 1982.
[12] Archer, D. H., ‘‘Lens Fed Multiple Beam Arrays,’’ Microwave Journal, September 1984,
p. 194.
[13] Diggavi, D., J. Shynk, and A. Laub, ‘‘Direction-of-Arrival Estimation for a Lens Based
Array,’’ IEEE Trans. on Antennas and Propagation, Vol. 42, No. 5, May 1994.
[14] Schmidt, R. O., ‘‘Multiple Emitter Location and Signal Parameter Estimation,’’ Proc.
RADC Spectrum Estimation Workshop, Griffiss Air Force Base, New York, 1979,
pp. 243–258. Reprinted in IEEE Trans. Antennas and Propagation, Vol. AP-34,
March 1986, pp. 276–280.
CHAPTER 6
Emitter Location
Charles Estrella
6.1 Introduction
The capability of performing accurate emitter location has become a primary

requirement for today’s reconnaissance platforms. In practice, one or more plat-
forms can be used to perform geolocation using emitter bearing, time, and/or
frequency information. A single platform can use pulse angle of arrival information
to form lines of bearing (LOBs) along which the emitter lies. The intersections of
the LOBs provide a location estimate, as shown in Figure 6.1.
With multiple platforms, measurements of an emitter’s time of arrival from
one platform can be used with TOA measurements from another to compute a
time difference of arrival. Each TDOA forms a hyperbola, or isochrone, upon the
surface of the Earth. An intersection of TDOA isochrones provides a potential
location of the emitter. An example of isochrones for various values of TDOA
between three aircraft is shown in Figure 6.2.
Figure 6.1 Single aircraft AOA using emitter bearing measurements.
149
150 Emitter Location
Figure 6.2 Various isochrone lines; three aircraft TDOA.
Frequency difference of arrival (FDOA) measurements can be made in conjunc-

tion with TDOA measurements by determining the peak of the complex ambiguity
function (CAF) between two intercepts from moving platforms [1]. In this case,
the intersection of isochrones and iso-Doppler lines as shown in Figure 6.3 provide
location estimates of the emitter.
6.2 Emitter Location Estimation
Because of uncertainties in TDOA, FDOA, and AOA measurements, uncertainties

exist in the estimated emitter location. An estimate of uncertainties in emitter
location (x Em , y Em ) and the associated location covariance matrix C x Em y Em are
ultimately desired in geolocation analysis. The location covariance matrix
C x Em y Em provides uncertainties in terms of the variance ␴ x2Em , ␴ y2Em and correlation
␳ x Em , ␳ y Em between emitter x-y location coordinates, and is given as
C x Em y Em = 冋 ␴ x2Em
␳ y Em x Em ␴ y Em ␴ x Em
␳ x Em y Em ␴ x Em ␴ y Em
␴ y2Em
册 (6.1)
For a given emitter location, the resulting eigenvalues (␭ 1 , ␭ 2 ) of the computed

location covariance matrix are used to compute the semimajor/minor axes (a, b)
of an uncertainty ellipse:
6.2 Emitter Location Estimation 151
Figure 6.3 Isochrone/iso-Doppler lines; two aircraft time/frequency difference of arrival.
a= √␭ 1
b = √␭ 2
冎 ␭1 > ␭2
(6.2)
The uncertainty ellipse has an orientation angle ␪ x Em y Em obtained from the

associated location covariance matrix computed as
␪ x Em y Em
1
= ⭈ tan
2
−1
冉 ␴ x2Em − ␴ y2Em
2 ⭈ ␳ x Em y Em ␴ x Em ␴ y Em 冊 (6.3)
For two dimensions, the circular error probable (CEP) is often used as a measure
of accuracy, defined to contain a percentage of the values within the error ellipse.
A CEP calculation containing 50% of values within the location ellipse can be
estimated from [2] using
CEP = 0.75 ⭈ √␭ 1 + ␭ 2 (6.4)
An example error ellipse and associated CEP is shown in Figure 6.4.

This ellipse is typically centered at the emitter’s true location (x Em , y Em ) when
uncertainties in TDOA, FDOA, and AOA measurements are unbiased; however,
the relative position of the error ellipse can incur a biased total location error
(TLE) even when the individual non-signal-dependent errors are unbiased. A
Figure 6.4 Location ellipse example.
significant TLE can occur because of the nonlinear relationship between the error
sources and the resulting location covariance matrix. Examples for the AOA case
are given in Section 6.4.
6.3 Deriving the Location Covariance Matrix
The analysis presented herein derives theoretical expressions for the covariance
matrix C x Em y Em in terms of various errors (e.g., collector position, heading, velocity).
These errors are treated as zero-mean random variables. Given two functions f 1
and f 2 consisting of zero-mean random variables X and Y, having variances
2 2
[␴ X , ␴ Y ] and correlation ␳ XY , the resulting uncertainty in f 1 and f 2 can be estimated
using
Cov f 1 f 2 = J ⭈ Covxy ⭈ JfTf (6.5)

f1 f2 1 2
xy xy
where
Cov f 1 f 2 = 冋 ␴f21
␳ f2 f1 ␴ f2 ␴ f1
␳ f1 f2 ␴ f1 ␴ f2
␴f22
册
冤冥
∂f 1 ∂f 1
∂x ∂y
J =
f1 f2 ∂f 2 ∂f 2
xy ∂x ∂y
Covxy = 冋 ␴x2
␳ yx ␴ y ␴ x
␳ xy ␴ x ␴ y
␴y2
册
6.4 Angle of Arrival Location Analysis 153
In employing (6.5), the relationship between the uncertainties in collector posi-

tion, heading, velocity, and emitter AOA and the resulting overall uncertainty in
AOA, TDOA, and FDOA uncertainty is assumed to be linear. Because in practice
these relationships are not linear, the expressions can only provide an approxima-
tion of the overall location accuracy. To validate the derived expressions using the
approach above, Monte Carlo simulation is used to generate results for comparison
and verification. The resulting analysis can then be used to define platform geodetic
requirements and used to estimate overall location performance when combined
with signal dependent errors. Derivations for the associated location covariance
matrix using AOA, TDOA, and TDOA/FDOA location methods are presented in
the following sections.
6.4 Angle of Arrival Location Analysis
To derive the associated location covariance matrix C x Em y Em for single aircraft AOA
location in terms of nonsignal-dependent errors in aircraft position and estimated
emitter AOA, the expression for AOA for a stationary emitter at (x Em , y Em ) and
an aircraft located at (x n , y n ) is used:
AOA n = tan−1 冉 x Em − x n
y Em − y n 冊 (6.6)
The emitter position (x Em , y Em ) is computed from two AOA measurements

(also known as LOBs) from aircraft positions (x 1 , y 1 ), (x 2 , y 2 ) as
tan (AOA1 )
x1 − x2 ⭈ − y 1 ⭈ tan (AOA1 ) + y 2 ⭈ tan (AOA1 )
tan (AOA2 )
x Em = (6.7)
tan (AOA1 )
1−
tan (AOA2 )
tan (AOA1 )
x1 − x2 ⭈ − y 1 ⭈ tan (AOA1 ) + y 2 ⭈ tan (AOA1 )
tan (AOA2 ) x2
y Em = − + y2
tan (AOA2 ) − tan (AOA1 ) tan (AOA2 )
(6.8)
Treating each uncertainty in aircraft position as independent random variables

⌬x n , ⌬y n having zero mean with standard deviation ␴ x n , ␴ y n , the associated aircraft
position covariance matrix can be written as
64748 0
冤冥
␴ x2n ␳ xn yn ␴ xn ␴ yn
C xn yn = (6.9)
64748 0
␳ yn xn ␴ yn ␴ xn ␴ y2n
As with aircraft position, uncertainty in measured AOA is treated as indepen-

dent random variables ⌬AOA n having zero mean with standard deviation
␴ AOA n . The associated AOA covariance matrix C AOA 12 can be written as
644474448
0
冤冥
2
␴ AOA 1
␳ AOA 1 AOA 2 ␴ AOA 1 ␴ AOA 2
C AOA 12 = (6.10)
644474448
0
2
␳ AOA 2 AOA 1 ␴ AOA 2 ␴ AOA 1 ␴ AOA 2
The emitter position covariance matrix C x Em y Em due to uncertainties in aircraft

positions (x 1 , y 1 ), (x 2 , y 2 ) and respective emitter angles of arrival (AOA 1 , AOA 2 )
can be computed as
C x Em y Em = Jx Em y Em | x n y n C x n y n JxTEm y Em | x n y n + Jx Em y Em | AOA 1 AOA 2 C AOA 1 AOA 2 JxTEm y Em AOA 1 AOA 2

(6.11)
The variables Jx Em y Em | x1 x 2 y 1 y 2 and Jx Em y Em | AOA 1 AOA 2 are the partial derivatives of

the location estimate with respect to the aircraft position and AOA, given as
冤冥
∂x Em ∂x Em ∂x Em ∂x Em
∂x1 ∂x 2 ∂y 1 ∂y 2
Jx Em y Em | x1 x 2 y 1 y 2 = (6.12)
∂y Em ∂y Em ∂y Em ∂y Em
∂x1 ∂x 2 ∂y 1 ∂y 2
冤冥
∂x Em ∂x Em
∂AOA 1 ∂AOA 2
Jx Em y Em | AOA 1 AOA 2 = (6.13)
∂y Em ∂y Em
∂AOA 1 ∂AOA 2
An example is given in Figure 6.5 to compare AOA error ellipse parameters

generated via Monte Carlo simulation to those computed using (6.11) for an emitter
located at [0, 150]. An AOA and position error of 1° and 1m, respectively, were
used for aircraft positioned at [−100, 0] and [100, 0].
The results for this case have negligible bias; however, when the AOA error
is increased to 10°, results given in Figure 6.6 show that the bias becomes significant.
The resulting bias in the error estimate illustrates the limitations in assuming
a linear relationship between the nonsignal-dependent error sources and their
resulting impact on the geolocation estimate. A more thorough investigation of the
bias in the estimate as a function of collection geometry is provided in [3], using
multiple LOBs computed and combined over time. Work done by Brown provides
three algorithms for the combining of LOBs along a flight track using distance
least squares, quadratic, and asymptotic algorithms described in [4]. A comparison
of location performance for these algorithms is shown in Figure 6.7 for several
AOA standard deviation values using 200 LOBs collected as a collector moves
from [−100, 0] to [100, 0].
Figure 6.5 AOA location ellipse example, 1° AOA, 1-m position error.
Figure 6.7 shows significant improvement in the reduction of the TLE when
employing either the quadratic or asymptotic algorithm over the distance least
squares for larger AOA errors. The Stansfield algorithm is similar to the Brown
least squares algorithm except for a weighting matrix. The bias of the Stansfield
algorithm was analyzed extensively in [5].
6.5 Time Difference of Arrival Location Analysis
The derivation of the associated location covariance matrix C x Em y Em for three

aircraft TDOA location is now derived in terms of errors in collector position. The
TOA for a stationary emitter at (x Em , y Em ) and an aircraft located at (x n , y n ) can
be computed as
1
c √ Em
TOA n = (x − x n )2 + (y Em − y n )2 (6.14)
Associated errors in each TOA measurement will occur due to uncertainties

in collector position:
Figure 6.6 AOA location ellipse example, 10° AOA error.
Figure 6.7 Comparison of performance: Brown distance least squares, quadratic, and asymptotic
algorithms.
1
c √ Em
TOA n + ⌬TOA n = [x − (x n + ⌬x n )]2 + [y Em − (y n + ⌬y n )]2 (6.15)
Treating each uncertainty in collector position as independent random variables

⌬x n , ⌬y n having zero mean with standard deviation ␴ x n , ␴ y n , the associated collec-
tor position covariance matrix can be written as
64748 0
冤冥
␴ x2n ␳ xn yn ␴ xn ␴ yn
C xn yn = (6.16)
64748 0
␳ yn xn ␴ yn ␴ xn ␴ y2n
The associated errors in individual TOA measurements can be written in terms

of the TOA covariance matrix C TOA , derived from known errors in collector
position as
T
C TOA = JTOA | x n y n ⭈ C x n y n ⭈ JTOA | x n y n (6.17)
where
冤冥
2
␴ TOA 1
␳ TOA 1 TOA 2 ␴ TOA 1 ␴ TOA 2 ␳ TOA 1 TOA 3 ␴ TOA 1 ␴ TOA 3
2
C TOA = ␳ TOA 2 TOA 1 ␴ TOA 2 ␴ TOA 1 ␴ TOA 2
␳ TOA 2 TOA 3 ␴ TOA 2 ␴ TOA 3
2
␳ TOA 3 TOA 1 ␴ TOA 3 ␴ TOA 1 ␳ TOA 3 TOA 2 ␴ TOA 3 ␴ TOA 2 ␴ TOA 3
(6.18)
The variable JTOA | x n y n is a 3 × 2 matrix of partial derivatives of the TOA

measurements with respect to collector position, written as
∂TOA 1 ∂TOA 1
∂x n ∂y n
∂TOA 2 ∂TOA 2
JTOA | x n y n =
∂x n ∂y n (6.19)
∂TOA 3 ∂TOA 3
∂x n ∂y n
For TDOA position estimation in two dimensions, three TOA measurements

are required. Using three collectors with one as a common reference, the two TDOA
measurements resulting from three independent TOA measurements can be written
as
冋 TDOA 12
TDOA 32
册冋=
1 ⭈ TOA 1
0 ⭈ TOA 1
−1 ⭈ TOA 2
−1 ⭈ TOA 2
+0 ⭈ TOA 3
+1 ⭈ TOA 3
册 (6.20)
Associated errors in each TDOA measurement will occur due to resulting

uncertainties in TOA:
冋 TDOA 12 + ⌬TDOA 12
TDOA 32 + ⌬TDOA 32
册 (6.21)
= 冋 1 ⭈ (TOA 1 + ⌬TOA 1 )
0 ⭈ (TOA 1 + ⌬TOA 1 )
−1 ⭈ (TOA 2 + ⌬TOA 2 )
−1 ⭈ (TOA 2 + ⌬TOA 2 )
+0 ⭈ (TOA 3 + ⌬TOA 3 )
+1 ⭈ (TOA 3 + ⌬TOA 3 )
册
The associated errors in TDOA measurement pairs can be written in the form
of the TDOA covariance matrix C TDOA , derived from known errors in TOA
measurements as
T
C TDOA = JTDOA | TOA ⭈ C TOA ⭈ JTDOA | TOA (6.22)
冋册
2
␴ TDOA 12
␳ TDOA 12 TDOA 32 ␴ TDOA 12 ␴ TDOA 32
C TDOA = 2
␳ TDOA 32 TDOA 12 ␴ TDOA 32 ␴ TDOA 12 ␴ TDOA 32
(6.23)
The variable JTDOA | TOA is a 2 × 3 matrix of partial derivatives of the TDOA

measurement pairs with respect to the individual platform measurements, written
as
冤冥
∂TDOA 12 ∂TDOA 12 ∂TDOA 12
JTDOA | TOA =
∂TOA 1
∂TDOA 32
∂TOA 2
∂TDOA 32
∂TOA 3
∂TDOA 32
= 冋
0
1 −1
−1
0
1
册
∂TOA 1 ∂TOA 2 ∂TOA 3
(6.24)
The expressions for TDOA in terms of collector position uncertainty can be

written as
TDOA 12 + ⌬TDOA 12 =
1
c
再 √[x Em − (x 1 + ⌬x 1 )]2 + [y Em − (y 1 + ⌬y 1 )]2 (6.25)
− √[x Em − (x 2 + ⌬x 2 )]
2
+ [y Em − (y 2 + ⌬y 2 )]2 冎
TDOA 32 + ⌬TDOA 13 =
1
c
再 √[x Em − (x 3 + ⌬x 3 )]2 + [y Em − (y 3 + ⌬y 3 )]2 (6.26)
− √[x Em − (x 2 + ⌬x 2 )]
2
+ [y Em − (y 2 + ⌬y 2 )]2 冎
With uncertainties in the TDOA measurement pairs known, the resulting loca-
tion covariance matrix C x Em y Em can be written in terms of TDOA uncertainties as
C x Em y Em = Jx Em y Em | TDOA 12 TDOA 32 ⭈ C TDOA ⭈ JxTEm y Em | TDOA 12 TDOA 32 (6.27)
Because the expressions for emitter location in terms of (TDOA 12 , TDOA 32 )

are not readily found, the following properties are used:
J ⭈ C ⭈ J T = {(J T )−1 ⭈ C −1 ⭈ J −1 }−1 (6.28)
JTDOA 12 TDOA 32 | x Em y Em = ( Jx Em y Em | TDOA 12 TDOA 32 )−1 (6.29)
The variable Jx Em y Em | TDOA 12 TDOA 32 is the Jacobian matrix of the partial derivatives
of (x Em , y Em ) with respect to (TDOA 12 , TDOA 32 ), written as
冤冥
∂TDOA 12 ∂TDOA 12
∂x Em ∂y Em
JTDOA 12 TDOA 32 | x Em y Em = (6.30)
∂x Em ∂y Em
Using (6.28) and (6.29), the three aircraft TDOA location covariance matrix
can then be written as
C x Em y Em = 再 JTDOA 12 TDOA 32 | x Em y Em ⭈ C TDOA ⭈ JTDOA 12 TDOA 32 | x Em y Em 冎

T −1 −1
(6.31)
C x Em y Em = 再 JTDOA 12 TDOA 32 | x Em y Em
T
⭈ 再 JTDOA 12 TDOA 32 | TOA ⭈ JTOA | x n y n ⭈ C x n y n ⭈ JTOA | x n y n ⭈ JTDOA 12 TDOA 32 | TOA 冎

T T −1
⭈ JTDOA 12 TDOA 32 | x Em y Em 冎
−1
(6.32)
An example given in Figure 6.8 compares expected TDOA error ellipse parame-
ters to those computed using (6.32) for an emitter located at [0, 150]. A position
error sigma of 1m was used for three collector positioned at [−100, 0], [0, 0], and
[100, 0].
6.6 Time/Frequency Difference of Arrival Location Analysis
The derivation of the associated location covariance matrix C x Em y Em for two collec-
tor TDOA/FDOA location is derived in terms of errors in aircraft position, heading,
and velocity. The TOA for a stationary emitter at (x Em , y Em ) and a collector
located at (x n , y n ) as given in (6.14) is used with a corresponding FOA measure-
ment, given as
1 V ⭈ cos ␪ n ⭈ (x Em − x n ) + Vn ⭈ sin ␪ n ⭈ (y Em − y n )
FOA n = ⭈ n (6.33)
␭ Em
√(x Em − x n ) + (y Em − y n )
2 2
Figure 6.8 Comparison of three collector TDOA ellipse parameters, simulated versus theoretical.
Associated errors in each FOA measurement will occur due to uncertainties in

collector position, heading, and velocity:
1
FOA n + ⌬FOA n = ⭈ (6.34)
␭ Em
(Vn + ⌬Vn ) ⭈ {cos (␪ n + ⌬␪ n ) ⭈ [x Em − (x n + ⌬x n )] + sin (␪ n + ⌬␪ n ) ⭈ [y Em − (y n + ⌬y n )]}
√[x Em − (x n + ⌬x n )]
2
+ [y Em − (y n + ⌬y n )]2
Treating each uncertainty in aircraft position, heading, and velocity as indepen-

dent random variables ⌬x n , ⌬y n , ⌬␪ n , and ⌬Vn having zero mean with standard
deviation ␴ x n , ␴ y n , ␴ ␪ n , and ␴ Vn , the associated collector position, heading, and
velocity covariance matrix C PosHeadVel for two collectors results in an 8 × 8 matrix,
written as
C PosHeadVel =
0
6447448
0
6447448
0
6447448
0
6447448
0
6447448
0
6447448
0
6447448
␴ x21 ␳ x 1 y 1 ␴x 1 ␴y 1 ␳ x 1 ␪ 1 ␴x 1 ␴␪ 1 ␳ x 1 V 1 ␴x 1 ␴V 1 ␳ x 1 x 2 ␴x 1 ␴x 2 ␳ x 1 y 2 ␴x 1 ␴y 2 ␳ x 1 ␪ 2 ␴x 1 ␴␪ 2 ␳ x 1 V 2 ␴x 1 ␴V 2
0 0 0 0 0 0 0
6447448 6447448 6447448 6447448 6447448 6447448 6447448
␳ y 1 x 1 ␴y 1 ␴x 1 ␴ y21 ␳ y 1 ␪ 1 ␴y 1 ␴␪ 1 ␳ y 1 V 1 ␴y 1 ␴V 1 ␳ y 1 x 2 ␴y 1 ␴x 2 ␳ y 1 y 2 ␴y 1 ␴y 2 ␳ y 1 ␪ 2 ␴y 1 ␴␪ 2 ␳ y 1 V 2 ␴y 1 ␴V 2
0
6447448
0
6447448
0
6447448
0
6447448
0
6447448
0
6447448
0
6447448
␳ ␪1 x1 ␴ ␪1 ␴ x1 ␳ ␪1 y1 ␴ ␪1 ␴ y1 ␴ ␪21 ␳ ␪1 V1 ␴ ␪1 ␴ V1 ␳ ␪1 x2 ␴ ␪1 ␴ x2 ␳ ␪1 y2 ␴ ␪1 ␴ y2 ␳ ␪1 ␪2 ␴ ␪1 ␴ ␪2 ␳ ␪1 V2 ␴ ␪1 ␴ V2
0 0 0 0 0 0 0
6447448 6447448 6447448 6447448 6447448 6447448 6447448
␳ V1 x1 ␴ V1 ␴ x1 ␳ V1 y1 ␴ V1 ␴ y1 ␳ V1 ␪1 ␴ V1 ␴ ␪1 ␴ V2 1 ␳ V1 x2 ␴ V1 ␴ x2 ␳ V1 y2 ␴ V1 ␴ y2 ␳ V1 ␪2 ␴ V1 ␴ ␪2 ␳ V1 V2 ␴ V1 ␴ V2
6447448
0 6447448
0 6447448
0 0
6447448
0
6447448
0
6447448
0
6447448
␳ x2 x1 ␴ x2 ␴ x1 ␳ x2 y1 ␴ x2 ␴ y1 ␳ x2 ␪1 ␴ x2 ␴ ␪1 ␳ x2 V1 ␴ x2 ␴ V1 ␴ x22 ␳ x2 y2 ␴ x2 ␴ y2 ␳ x2 ␪2 ␴ x2 ␴ ␪2 ␳ x2 V2 ␴ x2 ␴ V2
0 0 0 0 0 0 0
6447448 6447448 6447448 6447448 6447448 6447448 6447448
␳ y2 x1 ␴ y2 ␴ x1 ␳ y2 y1 ␴ y2 ␴ y1 ␳ y2 ␪1 ␴ y2 ␴ ␪1 ␳ y2 V1 ␴ y2 ␴ V1 ␳ y2 x2 ␴ y2 ␴ x2 ␴ y22 ␳ y2 ␪2 ␴ y2 ␴ ␪2 ␳ y2 V1 ␴ y2 ␴ V2
0
6447448
0
6447448 6447448
0 0
6447448
0
6447448
0
6447448
0
6447448
␳ ␪2 x1 ␴ ␪2 ␴ x1 ␳ ␪2 y1 ␴ ␪2 ␴ y1 ␳ ␪2 ␪1 ␴ ␪2 ␴ ␪1 ␳ ␪2 V1 ␴ ␪2 ␴ V1 ␳ ␪2 x2 ␴ ␪2 ␴ x2 ␳ ␪2 y2 ␴ ␪2 ␴ y2 ␴ ␪22 ␳ ␪2 V2 ␴ ␪2 ␴ V2
0 0 0 0 0 0 0
6447448 6447448 6447448 6447448 6447448 6447448 6447448
␳ V2 x1 ␴ V2 ␴ x1 ␳ V2 y1 ␴ V2 ␴ y1 ␳ V2 ␪1 ␴ V2 ␴ ␪1 ␳ V2 V1 ␴ V2 ␴ V1 ␳ V2 x2 ␴ V2 ␴ x2 ␳ V2 y2 ␴ V2 ␴ y2 ␳ V2 ␪2 ␴ V2 ␴ ␪2 ␴ V2 2
(6.35)
The associated errors in individual TOA/FOA measurements can be written in

the form of the TOA/FOA covariance matrix C TOAFOA , and derived from known
errors in collector position, heading, and velocity as
T
C TOAFOA = JTOAFOA | PosHeadVel ⭈ C PosHeadVel ⭈ JTOA | PosHeadVel
2
冤冥
␴ TOA ␳ TOA 1 TOA 2 ␴ TOA 1 ␴ TOA 2 ␳ TOA 1 FOA 1 ␴ TOA 1 ␴ FOA 1 ␳ TOA 1 FOA 2 ␴ TOA 1 ␴ FOA 2
1
2
␳ TOA 2 TOA 1 ␴ TOA 2 ␴ TOA 1 ␴ TOA ␳ TOA 2 FOA 1 ␴ TOA 2 ␴ FOA 1 ␳ TOA 2 FOA 2 ␴ TOA 2 ␴ FOA 2
2
= 2
␳ FOA 1 TOA 1 ␴ FOA 1 ␴ TOA 1 ␳ FOA 1 TOA 2 ␴ FOA 1 ␴ TOA 2 ␴ FOA ␳ FOA 1 FOA 2 ␴ FOA 1 ␴ FOA 2
1
2
␳ FOA 2 TOA 1 ␴ FOA 2 ␴ TOA 1 ␳ FOA 2 TOA 2 ␴ FOA 2 ␴ TOA 2 ␳ FOA 2 FOA 1 ␴ FOA 2 ␴ FOA 1 ␴ FOA
2
(6.36)
The variable JTOAFOA | PosHeadVel is a 4 × 8 matrix of partial derivatives of the

TOA/FOA measurements with respect to aircraft position, heading, and velocity,
written as
JTOAFOA | PosHeadVel =
∂TOA 1 ∂TOA 1 ∂TOA 1 ∂TOA 1 ∂TOA 1 ∂TOA 1 ∂TOA 1 ∂TOA 1

∂x1 ∂y1 ∂␪ 1 ∂V1 ∂x 2 ∂y 2 ∂␪ 2 ∂V2
∂TOA 2 ∂TOA 2 ∂TOA 2 ∂TOA 2 ∂TOA 2 ∂TOA 2 ∂TOA 2 ∂TOA 2
∂x1 ∂y1 ∂␪ 1 ∂V1 ∂x 2 ∂y 2 ∂␪ 2 ∂V2
∂FOA 1 ∂FOA 1 ∂FOA 1 ∂FOA 1 ∂FOA 1 ∂FOA 1 ∂FOA 1 ∂FOA 1
∂x1 ∂y1 ∂␪ 1 ∂V1 ∂x 2 ∂y 2 ∂␪ 2 ∂V2
∂FOA 2 ∂FOA 2 ∂FOA 2 ∂FOA 2 ∂FOA 2 ∂FOA 2 ∂FOA 2 ∂FOA 2
∂x1 ∂y1 ∂␪ 1 ∂V1 ∂x 2 ∂y 2 ∂␪ 2 ∂V 2
(6.37)
For position estimation in two dimensions, two TOA and two FOA measure-
ments are required. The TDOA/FDOA measurements resulting from two TOA/
FOA measurements can be written as
冋 TDOA 12
FDOA 12
册冋=
1 ⭈ TOA 1
0 ⭈ TOA 1
−1 ⭈ TOA 2
+0 ⭈ TOA 2
+0 ⭈ FOA 1
+1 ⭈ FOA 1
+0 ⭈ FOA 2
−1 ⭈ FOA 2
册
(6.38)
Substituting associated errors in collector position, heading, and velocity into

(6.38) yields
TDOA 12 + ⌬TOA 12 =
1
c
再 √[x Em − (x1 + ⌬x1 )]2 + [y Em − (y 1 + ⌬y 1 )]2
− √[x Em − (x 2 + ⌬x 2 )]
2
+ [y Em − (y 2 + ⌬y 2 )]2 冎
(6.39)
1
FDOA 12 + ⌬FOA 12 = ⭈
␭ Em
冦冧
(V1 + ⌬V1 ) ⭈ {cos (␪ 1 + ⌬␪ 1 ) ⭈ [x Em − (x 1 + ⌬x 1 )] + sin (␪ 1 + ⌬␪ 1 ) ⭈ [y Em − (y 1 + ⌬y 1 )]}
√[x Em − (x 1 + ⌬x 1 )]
2
+ [y Em − (y 1 + ⌬y 1 )]2
(V2 + ⌬V2 ) ⭈ {cos (␪ 2 + ⌬␪ 2 ) ⭈ [x Em − (x 2 + ⌬x 2 )] + sin (␪ 2 + ⌬␪ 2 ) ⭈ [y Em − (y 2 + ⌬y 2 )]}
−
√[x Em − (x 2 + ⌬x 2 )] + [y Em − (y 2 + ⌬y 2 )]
2 2
The TOA/FOA covariance matrix C TOAFOA can then be used to compute

uncertainty in TDOA/FDOA measurement pairs, given in the TDOA/FDOA covari-
ance matrix C TDOAFDOA as
T
C TDOAFDOA = JTDOAFDOA | TOAFOA ⭈ C TOAFOA ⭈ JTDOAFDOA | TOAFOA
冋册
2
␴ TDOA 12
␳ TDOA 12 FDOA 12 ␴ TDOA 12 ␴ FDOA 12
= 2
␳ FDOA 12 TDOA 12 ␴ FDOA 12 ␴ TDOA 12 ␴ FDOA 12
(6.40)
The variable JTDOAFDOA | TOAFOA is the partial derivative of TDOA/FDOA mea-

surements with respect to individual TOA/FOA measurements, derived from (6.38)
as
JTDOAFDOA | TOAFOA = 冋 0
1 −1
0
0
1 −1
0
册 (6.41)
With uncertainties in the TDOA/FDOA measurement pairs known, the resulting

location covariance matrix C x Em y Em can be written in terms of TDOA/FDOA uncer-
tainties as
6.7 Geometric Dilution of Precision 163
T
C x Em y Em = Jx Em y Em | TDOA 12 FDOA 12 ⭈ C TDOAFDOA ⭈ Jx Em y Em | TDOA 12 FDOA 12
(6.42)
The variable Jx Em y Em | TDOA 12 FDOA 12 is the Jacobian matrix of the partial derivatives
of (x Em , y Em ) with respect to (TDOA 12 , FDOA 12 ), written as
冤冥
∂x Em ∂y Em
JTDOA 12 FDOA 12 | x Em y Em = (6.43)
∂FDOA 12 ∂FDOA 12
∂x Em ∂y Em
Because the expressions for emitter location in terms of (TDOA 12 , FDOA 12 )

are not readily found (as the case with three aircraft TDOA), the following relation-
ship is again employed:
JTDOA 12 FDOA 12 | x Em y Em = ( Jx Em y Em | TDOA 12 FDOA 12 )−1 (6.44)
The two aircraft TDOA/FDOA location covariance matrix can then be written
as
C x Em y Em = 再 JTDOA 12 FDOA 12 | x Em y Em ⭈ C TDOAFDOA ⭈ JTDOA 12 FDOA 12 | x Em y Em 冎

T −1 −1
(6.45)
An example is given in Figure 6.9 to compare expected TDOA/FDOA error

ellipse parameters to those computed using (6.45). An emitter located at [1, 150]
with a frequency of 900 MHz was used. A position error sigma of 1m was used
for two collectors positioned at [−100, 0] and [100, 0], traveling in an easterly
direction at 200 m/s with heading and velocity uncertainties of 1° and 1 m/s,
respectively.
6.7 Geometric Dilution of Precision
Other errors occur with geometry. Poorer location performance is observed when
the angle between intersecting LOBs, TDOA, and/or TDOA/FDOA lines
approaches zero. This phenomenon is known as geometric dilution of precision
(GDOP) and is typically observed at cross- and down-range regions outside of the
baseline, as shown in Figure 6.10 for TDOA using three collectors.
The effects of GDOP can be readily seen in observing CEP values for emitters
at various cross- and down-range locations. CEP values for the three collector
TDOA scenario with 1° position error are given in Figure 6.11 to illustrate the
poorer performance away from the center of the baseline.
To minimize the unwanted effects of GDOP, different collector paths can be
used. Typically, nearly perpendicular intersecting lines of bearing and constant
Figure 6.9 Comparison of two collector TDOA/FDOA ellipse parameters, simulated versus
theoretical.
TDOA or FDOA are desired, especially when errors at each platform are equal.
In such a case with two collectors measuring AOA, the minimum CEP occurs when
the emitter is down-range from the midpoint of the line joining the collectors at a
distance of R /√2 when the collectors are separated by a distance of R, as shown
in Figure 6.12.
6.8 Incorporation of Measurement Error
The derived AOA, TDOA, and TDOA/FDOA covariance matrices for non-signal-
dependent errors can be summed with associated covariance matrices computed
from signal-dependent error sources to provide an overall estimate of system perfor-
mance. Signal-dependent error is largely determined by the SNR. As indicated in
Chapter 5, the standard deviation of the AOA error function of SNR for an
interferometer is given by
6.8 Incorporation of Measurement Error
Figure 6.10 Effects of geometry on location performance; three aircraft TDOA.

165
Figure 6.11 Effects of geometry on geolocation performance; three collector TDOA.
Figure 6.12 Minimum CEP; two aircraft AOA.

6.9 Summary 167
␭ Emitter
␴ AOA = (deg) (6.46)
2␲ ⭈ d ⭈ cos (AOA) ⭈ √SNR
The variables d and ␭ Emitter are the antenna separation and emitter wavelength,
respectively. The AOA value is given with respect to the perpendicular off boresight;
SNR is given in as a magnitude ratio (not in decibels). Errors in TOA as a function
of SNR are typically computed as a function of the pulses’ leading edge rise time,
t R , given in Chapter 12 as
1.25t R
␦T = (sec) (6.47)
√2 SNR
When TDOA/FDOA measurements are made with complex ambiguity function
processing, errors in terms of signal bandwidth B s , noise bandwidth B, integration
time T, and receiver SNR values ␥ 1 and ␥ 2 can be estimated using [1]
0.55
␴ TDOA ≈ (sec) (6.48)
B s √BT␥
0.55
␴ FDOA ≈ (Hz) (6.49)
T √BT␥
The variable ␥ is the effective SNR computed from the SNR at each platform
as
␥= 冋
1 1
+
1
+
1
2 ␥1 ␥2 ␥1␥2 册 (6.50)
6.9 Summary
In this chapter the effect of various errors on estimates of emitter location was
addressed. Theoretical expressions for location uncertainty in terms of errors in
collection platform position, heading, and velocity were derived and used to predict
error ellipse parameters for single collector AOA and multiple collector TDOA and
TDOA/FDOA techniques. The error ellipse parameters predicted using theoretical
expressions were compared with computer simulation results. Geometric dilution
of precision was also discussed. Several expressions for errors in AOA, TDOA,
and FDOA measurements due to noise were also mentioned. Combining all of
these expressions provides estimates for total location system performance. The
detailed location analysis in [6] is a comprehensive summary of the subject for the
interested reader.
Instantaneous phase or frequency tracking using a single moving platform can
also be used to locate an emitter. If the emitter is coherent, one measures the pulse-
to-pulse phase change as the platform moves. These phase changes may be converted
to instantaneous frequency by differentiating the instantaneous phase. By assuming
a location for the emitter, one can plot the instantaneous frequency that would
have been observed by a collector moving along the known path. This is compared
to the observed frequency variation and the assumed location is adjusted to mini-
mize the difference and thus estimate the emitter location. This process can only
be used if the emitter is coherent. If the emitter is not coherent, two antennas and
receivers are placed on the collector separated by some distance. The phase differ-
ence now depends on the angle of arrival of the pulses from the emitter. (The two
antennas form an interferometer.) The observed phase variations along the path
are compared to what would have been observed from an emitter at a trial location,
and then the differences are minimized to find the best location estimate. Generally
the best location performance is obtained using an S-shaped collector path. A
similar location process is possible using very accurate TOA measurements if the
emitter has a stable PRI. Then as the collector moves, the TOA changes slightly
and the emitter location is estimated by determining which location gives the
observed TOA variations.
References
[1] Stein, S., ‘‘Algorithms for Ambiguity Processing,’’ IEEE Trans. on Acoustics, Speech, and
Signal Processing, June 1981.
[2] Brown, R., Emitter Location Using Bearing Measurements from a Moving Platform, NRL
Report 8483, June 1981.
[3] Torrieri, D. J., ‘‘Statistical Theory of Passive Location Systems,’’ IEEE Trans. on Aerospace
and Electronic Systems, March 1984.
Artech House, 1985.
[5] Gavish, M., and A. Weiss, ‘‘Performance Analysis of Bearing-Only Target Location Algo-
rithms,’’ IEEE Trans. on Aerospace and Electronic Systems, July 1992.
[6] Poisel, R., Electronic Warfare Target Location Methods, Norwood, MA: Artech House,
2005.
CHAPTER 7
Estimating Power at the Transmitter
7.1 Power Estimation Through ELINT
The radar’s transmitted power is a basic quantity that defines its range performance.
Unfortunately, transmit power is difficult to estimate from the ELINT point of
view. Historically, transmitter power was one of the last aspects of a signal to be
examined by signal analysis. In ELINT work, the quantity that can be most readily
estimated is the peak effective radiated power (ERP), which is the product of the
peak transmit power, the antenna gain, and any loss between the transmitter and
transmit antenna. (The radar’s antenna gain can be estimated from its beam pattern,
and then the peak transmit power can also be estimated once the ERP is determined.)
Consider the one-way or ELINT range equation. Solving this equation for the ERP
gives the following result:
2 2
P G (4␲ ) S E R E L E
ERP = T TE = (7.1)
LT GE ␭
where:
PT = transmitted power
G TE = transmit antenna gain in the direction of the ELINT site
L T = losses at the transmitter
S E = received signal power
R E = distance from the transmitter to the ELINT site
L E = losses at the ELINT receiver
G E = gain of the ELINT site antenna in the direction of the transmitter
␭ = wavelength = speed of light divided by the RF
In radar usage, the term ‘‘peak power’’ means the transmitted power averaged
over the duration of one pulse. The term ‘‘average power’’ means the power
averaged over one PRI. The average power is related to the peak power by
P ␶
PAV = T (7.2)
T
169
170 Estimating Power at the Transmitter
where:
PAV = average transmitted power

PT = peak transmitter power
␶ = pulse duration
T = pulse repetition interval
The ratio ␶ /T is referred to as the duty cycle of the radar transmitter.

Many radar transmitters are designed for large peak power output. They are
typically operated as saturated amplifiers or oscillators. At present, it is still unusual
for a radar transmitter to have the capability of varying its power level; however,
some systems can select from among several discrete power-output levels on a
pulse-by-pulse basis.
One of the key terms in (7.1) is the distance from the transmitter to the ELINT
site. Generally speaking the transmitter location is not known. Emitter location is
a difficult task. It is the subject of Chapter 6, but could the subject of an entire
book. Some of the basic ideas useful for estimation of transmit power are included
below.
7.2 Distance to the Horizon
Generally speaking, the emitter must be located within line of sight of the ELINT
station. If the interceptor is in an aircraft and the emitter is on the ground (or vice
versa), the line-of-sight range can be rather large—several hundreds of kilometers.
Key parameters for determining the maximum line of sight are the height of the
transmit and receive antennas above the Earth. The line of sight to the horizon,
assuming a spherical Earth, can be calculated from plane geometry using Figure
7.1.
d 2 = h 2 + 2rh (7.3)
Because usually r Ⰷ h, this is often approximated by
Figure 7.1 Distance to the horizon.

d≈ √2rh
With propagation at microwave frequencies, the effective radio horizon is
observed to be greater than the geometrical horizon. This is caused by refraction
by the atmosphere. Refractive effects vary widely as a function of the water vapor
in the atmosphere; however, it is common practice to approximate this by multi-
plying the Earth’s radius by a factor greater than 1. The usual factor is 4/3, making
the effective Earth radius, reff = 4r /3. This yields the distance to the radio horizon
as
d ≈ (2reff h)1/2 ≈ 冉冊
8rh 1/2
3
For Earth radius r = 3,900 miles, where height (h) is in feet, this reduces to
d miles ≈ √2h ft
For r = 3,900 miles, where height (h) is in meters and distance (d) is in kilometers
d km ≈ √17.3h m
Figure 7.2 illustrates the distance to the horizon as a function of altitude.
One of the methods to find an upper bound of the ERP is to insert the distance
to the horizon into (7.1). This gives the maximum ERP value possible.
P G (4␲ )2S E reff hL E

ERPmax = T TE = (7.4)
LT GE ␭
As (7.4) shows, for a given received signal level S E , the ERP estimate increases
in proportion to the collector altitude for an emitter at the horizon.
If the height of the ELINT antenna and the height of the transmitter are known,
simply use Figure 7.2 to compute the distance to the horizon for each antenna
height and add the values together. For example, if the ELINT antenna is on an
aircraft flying at an altitude of 10,000m, the distance to the horizon is 416 km.
If the emitter antenna is on a hilltop 1,000m high, the distance to its horizon is
131 km. The total distance possible for line-of-sight propagation is then 547 km.
In much of what follows, the discussion is concerned with determining what
conditions must be met to limit the ERP error caused by the individual factors in
(7.1) to 0.5 dB.
7.3 ERP Errors Due to Antenna Pointing Errors
In principle, the gain of the ELINT antenna GE is known. But, it may be difficult
to accurately point the ELINT antenna toward the radar. The power received is
fluctuating due to the scanning of the radar beam and the collection platform
Figure 7.2 Distance to the horizon (for R multiplied by 4/3).
may be moving. Thus, maximizing the signal amplitude at the ELINT site means
performing scan-to-scan amplitude comparisons. The ELINT site would prefer a
relatively broad beam for its antenna so that pointing accuracy is not a problem
for power measurement. On the other hand, the ELINT station may need a narrow
beam so that the bearing (AOA) to the radar can be determined. This bearing
should be accurately determined in order to find the range to the radar by means
of triangulation (this requires the cooperation of at least one other ELINT site or
else requires that the ELINT station be moving). (Triangulation is discussed in
Chapter 6.) These contradictory requirements mean that some stations may use
one antenna for direction finding and another for power measurement. Another
approach is to use monopulse techniques to aid in pointing the antenna to within
a fraction of the 3-dB beamwidth. The pointing accuracy required to achieve a
given ERP accuracy can be computed if the antenna pattern is known. Consider
the case of a uniformly illuminated rectangular aperture. This produces one-way
power gain that varies as [1]
G (␪ , ␾ )
=
sin2冉 ␲L
␭ 冊冉
sin ␪ cos ␾ ⭈ sin2
␲W
␭
sin ␾ cos ␪ 冊 (7.5)
冉冊冉冊
G (0, 0) ␲ L sin ␪ cos ␾ 2
␲ W sin ␾ cos ␪ 2
⭈
␭ ␭
where:
G (␪ , ␾ ) = gain in direction (␪ , ␾ )
L = length of antenna in azimuth direction
W = width of antenna in elevation direction
␪ = azimuth angle relative to normal to the aperture
␾ = elevation angle relative to normal to the aperture
G (0, 0) = boresight for (0, 0)
␭ = wavelength
Suppose the receive antenna gain used to compute the ERP is to be accurate
to within 0.5 dB. Then the antenna must be pointed toward the transmitter to
within about 20% of the 3-dB beamwidth if the pointing error is only in one
dimension (azimuth or elevation). If the antenna must be pointed in both dimensions
(azimuth and elevation), the pointing error components must each be less than
about 15% of the corresponding beamwidth, corresponding to a vector error of
21% if the beamwidths are equal. (See Figure 7.3.)
It is unlikely that the antenna gain of the ELINT antenna will be known to
within 0.5 dB. In addition to the pointing of the antenna, other factors include the
fact that the antenna pattern is known only at certain frequencies and for specific
Figure 7.3 Antenna gain as a function of pointing angle error.

polarization conditions. It is likely that the polarization of the received signal will
be matched to the polarization of the receiving antenna or that the frequency of
the received signal will be the same as the frequencies at which the antenna gain
has been measured.
Polarization errors also make determining the received power difficult. It is
necessary to try different receiving antenna polarizations at the ELINT site to
determine which polarization results in maximum received power when the center
of the main beam is directed toward the ELINT site. For a linearly polarized
radar transmit antenna, the loss at the ELINT site due to improper polarization
orientation is given by [cos (⑀ )]−2, where ⑀ is the error in polarization [2]. Achieving
an error of less than 0.5 dB requires that the error in matching the polarization
of the receiver to that of the transmitting antenna be less than 19.25°.
Another problem is that when the collector is moving, it will not be at the
peak of the main beam indefinitely. Consider a situation where the collector is flying
toward a ground-based radar at constant altitude. As the collector approaches, the
elevation angle relative to the transmitter increases. Therefore, both the receive
antenna gain and the transmit antenna gain change slightly as the collector
approaches.
7.4 Estimating the Distance to the Radar
Consider the effect of range errors on determining the radar power. Equation (7.1)
shows that the received signal level is proportional to (␭ /R E )2. Measuring the
wavelength (or radio frequency) can be accomplished quite accurately; the main
problem is finding the range, R E . The power is related to the range by
1
P␣ 2
(7.6a)
RE
Differentiating and rearranging terms gives
⌬P −2⌬R E
= (7.6b)
P RE
This means that a 10% range error produces a power error of about 20%. To
maintain power accuracy to 0.5 dB requires
⌬P
= 1 − 10−0.05 = 0.109
P
or
⌬R E ⌬P
= = 0.054
RE 2P
7.4 Estimating the Distance to the Radar 175
This means that the range must be known to within about 5%. Using triangula-
tion, the range error can be related to the pointing error. As shown in Chapter 6,
the error in location is a function of the ratio of the baseline between the two
interception points to the distance to the emitter. Consider the simple case with
two stations and an accurately known baseline distance, D, between them, as
shown in Figure 7.4.
The following relationships are clear:
x y
R E1 = = (7.7)
cos ␪ 1 sin ␪ 1
D−x y
R E2 = = (7.8)
cos ␪ 2 sin ␪ 2
Equation (7.7) may be solved to give
D sin ␪ 2
R E1 = (7.9)
sin (␪ 1 + ␪ 2 )
Differentiating gives
⌬R E1 ⌬␪ 2 ⌬␪ 1 + ⌬␪ 2
= − (7.10)
R E1 tan ␪ 2 tan (␪ 1 + ␪ 2 )
To choose a simple case, take ␪ 1 = ␪ 2 = 45°. Then,
⌬R E1
= ⌬␪ 2
R E1
For a power accuracy of less than 0.5 dB, the range error must be less than
0.054, so that the direction-finding accuracy of station 2 must be 0.054 rad or
3.1°. For this example, the pointing error due to station 1 does not appear because,
for the angles chosen, the vector from station 2 to the emitter is perpendicular to
the vector from station 1 to the emitter.
Figure 7.4 Emitter location using two station triangulation.

Consider the case ␪ 1 = 90°, ␪ 2 = 45°. If the two ELINT stations have the same
angular accuracy (⌬␪ = ⌬␪ 1 = ⌬␪ 2 ), then
⌬R E1
= 3⌬␪
R E1
For ⌬R E1 /R E1 = 0.054, ⌬␪ = 0.018 rad = 1.04°. In other words, for a power

measurement accuracy of 0.5 dB, the angular bearing error (for triangulation) must
be no more than 1.04° in this case.
There are other methods of range estimation, including TDOA and FDOA as
discussed in Chapter 6.
7.5 Multiple Signal and Multipath Problems
Significant errors in measuring the received signal power may occur due to the
inadvertent reception of more than one signal at one time or through multipath.
In the case of a pulsed signal, it is usually possible to isolate the desired signals
from others in the receiver pass band through the time separation of the pulses
even if the carrier frequencies of the two signals are quite close. Usually, the two
signals will differ considerably in power, and it is possible to trace the amplitude
envelope of one signal as it scans by in spite of interference from another pulsed
(or CW) signal. Figure 7.5 shows the pulse amplitude variations over time of two
overlapping radar scans; separation of these signal pulses, at least manually, is
clearly possible using their amplitude versus time functions.
Figure 7.5 Pulse amplitude versus time for two scanning radars.
7.5 Multiple Signal and Multipath Problems 177
In the case of CW signals, it is necessary to rely on frequency separation.

However, it may be possible to use much narrower receiver bandwidths, depending
on the modulation in use.
Multipath presents a difficult problem in that the interfering signal is a precise
replica of the signal of interest, except that it is delayed in time, its amplitude is
different, and its phase relative to that of the direct signal is unpredictable. The
best protection in this situation is to emphasize the leading edge of pulsed signals.
If the rise time of the pulse is larger than the delay introduced by the greater length
of the indirect path, this method also fails. The path length difference produces a
delay given by
⌬L
⌬␶ =
c
where:
⌬␶ = multipath time delay

c = speed of light (3 × 108 m/s)
⌬L = path length difference (meters)
Many rise times are less than 300 ns, so that multipath differences of 100m
or more produce multipath interference that can be discriminated from the direct
path by time differences. Very large path length differences reduce the amplitude
of the multipath. Also, if the arrival angle of the multipath differs from that of the
direct path (as when reflections occur from objects near the ELINT station), the
gain of the ELINT antenna may be used to help reduce their amplitude.
Because multipath signals are exact coherent replicas of the direct path signal
(except for an amplitude change and a phase shift), the interference signal can
either add or subtract from the direct path signal. If the multipath and direct signals
were equal in amplitude, the amplitude of their vector sum could increase by as
much as 6 dB or decrease to 0 depending on the phase shift. The length squared
of the sum of two vectors is
› ›
| A 1 + A 2 | 2 = A12 + A22 + 2A1 A2 cos ␪ (7.11)
where A1 and A2 are the two signal amplitudes, and ␪ is the phase angle between
the two signals.
2 2
If the ratio of the multipath power to the direct power, A2 /A1 is denoted as
PR, the error in the power estimate is
› ›
|A1 + A 2|2
= 1 + PR + 2 cos (␪ ) √PR (7.12)
| A1 | 2
Figure 7.6 shows the power measurement error due to the multipath as a
function of the power ratio (PR) of the indirect signal to the direct signal for in-
phase multipath and for 45°, 90°, and 180° phase differences. For 180° phase
Figure 7.6 Power measurement error due to multipath.
difference, there is cancellation and the power estimate is too low. The multipath
power ratio range is from −25 dB for a 0° phase difference to about −9.9 dB for
a 90° phase difference if the error in the power estimate is to be held to 0.5 dB.
For a 180° phase difference, the power error is −0.5 dB for a power ratio of
−25 dB. It turns out that the rms error in ERP for uniformly distributed ␪ exceeds
0.5 dB when the multipath reflection exceeds −21.8 dB. Because the phase is
unpredictable, maintaining an ERP error of 0.5 dB rms requires that the multipath
power be 21.8 dB lower than the direct path power.
7.6 Summary of Power Measurement Requirements
This chapter has shown what is necessary to measure the ERP of the radar to
within 0.5 dB for each of the following factors:
1. Pointing the ELINT antenna (within 15% of its 3-dB beamwidth);

2. Polarization mismatch (for linear polarization must be less than 19.25°);
3. Bearing accuracy for triangulation (better than 1.04° or 3.1° for the two
cases considered).
4. Ratio of multipath power to direct path power (must be less than −21.8
dB).
If all four of these errors are maintained to produce a peak error of less than
0.5 dB, combining them on an rms basis gives the total expected ERP measurement
error as about ±1 dB and a worst case error of ±2 dB. Considering the many other
sources of error (e.g., calibration of the antenna, receiver gain, and uncertainty
about propagation path attenuation), ERP measurements more accurate than
±2 dB are rarely possible.
7.7 Sample ERP Calculations
To illustrate the points of this chapter, it is helpful to consider a sample ERP

calculation. Table 7.1 shows the results of two ELINT stations located 100 km
apart intercepting the same signal. To find the distance from each emitter, (7.9)
can be used, or plane trigonometry may be applied to yield the results shown in
Table 7.2.
The other parameter needed is the wavelength. This is found by dividing the
speed of light by the RF, which gives a wavelength of 0.1m. Then, using the values
from station 1, the ERP in dBm is found using Table 7.3.
It is essential to make sure the units are correct, namely, that the range and
wavelength have the same units. The next step is to decide if this value is reasonable
for the type of signal reported. If the antenna gain is about 25 dB (a typical value),
then the transmit power would be in the range of 51.2 to 65 dBW—about 130 to
3,000 kW. If this represents the peak of the main beam, then it is clear that this
high power radar could be expected to have a long PRI. The uncertainty in the
ERP of nearly 14 dB is rather typical of the type of errors inherent in ERP measure-
ments using only a single pair of reports. Improved accuracy can be achieved,
especially by a more exact emitter location or by averaging many reports from
many stations. Nevertheless ERP measurements are subject to many uncertainties.
Another way to check the reasonableness of the distance to the emitter is to
compare the difference in the received power levels (in decibels) to that expected
Table 7.1 Two ELINT Stations Intercepting the Same Signal

Station 1 Station 2
ELINT antenna gain (dBi) 10 ± 2 9±1
Signal bearing (degrees) 90 ± 4 60 ± 5
Peak received power (dBm) −24 ± 1 −26 ± 1
RF (MHz) 3,000 ± 1 3,000 ± 1
Table 7.2 Determining the Distance from Each Emitter

Angle A Angle B Angle C Side a Side b Side c
30 90 60 100 200 173.2
21 94 65 100 278.4 252.9
39 86 55 100 158.5 130.2
Table 7.3 Values from Station 1

Parameter Name Parameter Value Parameter dB
Constant 4␲ 22
Received power −24 ±2 dBm
Maximum range 253 km +108 dB (m2)
Minimum range 130 km +102.3 dB (m2)
ELINT antenna gain −10 ± 2 dBi
Wavelength 0.1m +20 dB(m2)
Total (ERP) 120 dBm maximum
106.2 dBm minimum
from the difference in range to the two stations. In this case, if the longest ranges
are used, the power level difference would be 0.83 dB. If the shortest ranges are
used, the difference would be 1.71 dB. From the reported power and antenna gain
values, the difference could be anywhere in the range of −2 to +8 dB. In this case,
there is no inconsistency with the difference expected from geometrical considera-
tions.
It can be shown that if the receive sites are separated by a unit distance, then
the contours of the constant received power ratio, K, are circles centered on the
baseline extended a distance of 1/(K − 1) times the distance between the receive
sites, with radius of the square root of K divided by (K − 1). In this formulation,
K is larger than 1 (i.e., it is the ratio of the power from the stronger station divided
by the power from the weaker station after compensating for any differences in
receive antenna gain). For a power ratio of 3 dB or 2:1, K = 2, the radius of the
circle is 1.414 times the distance between the stations, and the center of the circle
is along the baseline extended a distance equal to the distance between the stations.
These circles are illustrated in Figure 7.7.
Suppose the TOA of each pulse can be measured accurately at each station.
The time difference is given by the difference in the distance to the emitter from
each station divided by the speed of light. In this example, the maximum difference
in distance is 28.3 km and the minimum difference is 25.5 km. These correspond
to time differences of 94.3 to 85 ␮ s. Because time differences can be easily measured
to an accuracy of 0.1 ␮ s, the difference in the distance to the two stations can
sometimes be measured to an accuracy of 30m. Determining the complete emitter
location requires a pair of stations for each dimension to be found—three-dimen-
Figure 7.7 Circles of constant power ratio.

Figure 7.8 Lines of constant time difference. Station 1 is at (225,0) and station 2 is at (275,0)
(TDOA in microseconds; location coordinates in kilometers).
sional emitter locations require at least four stations and they must not all be in
a single geometric plane. Note that there is a problem deciding which pulse at one
station is the same pulse at another station. This is difficult if the environment is
so dense that many pulses occur in a time comparable to the time difference. Also,
if the time difference exceeds the radar PRI, an ambiguous situation exists that
must be resolved. Lines of constant time difference are illustrated in Figure 7.8 for
two stations 50 km apart. The vertical line is the line of zero time difference and
is the perpendicular bisector of the line between the stations or baseline. The
hyperbolic curves are lines of 50, 100 and 150 ␮ s of time difference.
References
[1] Skolnik, M. I., Radar Handbook, New York: McGraw-Hill, 1970.

[2] Skolnik, M. I., Radar Handbook, New York: McGraw-Hill, 1970, p. 2–61.
CHAPTER 8
Antenna Parameters
The antenna is an extremely important part of radar. It must couple the energy
from the radar to and from space; it must radiate in the desired direction and
discriminate against receiving radiation from other directions. The ELINT-observ-
able antenna parameters are chiefly its polarization characteristics, the shape of
its main beam, and its sidelobe levels. This chapter begins by considering aspects
of polarization and continues with consideration of beam shape and sidelobe levels.
8.1 Polarization Defined1
The polarization of an electromagnetic wave is described by the geometric figure

traced by the electric field vector as the wave travels through space.2 The polariza-
tion of the transmitted radar signal can be determined at an ELINT station by
specialized measurement equipment or simply by manually determining which
polarization of the ELINT antenna provides the greatest signal strength from
the transmitter. For electronic warfare purposes, it is important to determine the
polarization because the effectiveness of a jammer having improper polarization
is reduced.
8.2 Elliptical Polarization [1]3
The geometric figure traced by the electric field vector is, in general, an ellipse.
Under some conditions the ellipse may collapse into a straight line, in which case
the polarization is called linear. In the other extreme, the ellipse may become
circular. Thus, linear and circular polarizations are special cases of elliptical polar-
ization. The parameters used to describe polarization are either those appropriate
to describing an ellipse or, for a more general description, the Stokes’ parameters
may be used. Elliptical polarization can be thought of as the result of two linearly
polarized waves at the same frequency. Polarization is determined by the electric
field vector, E. If the electric field vector lies in a vertical plane, the wave is vertically
polarized. For elliptically polarized waves, the tip of the electric field vector, as a
1. The author is grateful for the help of Mr. Grover M. Boose in preparing the polarization sections.
2. This is the convention in electrical engineering. In physics and optics, polarization refers to the magnetic
field vector.
3. Some of this material is a condensation of Section 15.10 in [1].
183
184 Antenna Parameters
function of time, traces out an ellipse. Circular polarization is a special case of an

ellipse whose major and minor axes are equal.
Following Kraus [1, 2], let the components of the instantaneous electric field
of the horizontally and vertically polarized wave be designated E x and E y . Then,
as a function of time, t, and distance, z,
E x = E 1 sin (␻ t − ␤ z) (8.1)
E y = E 2 sin (␻ t − ␤ z + ␦ ) (8.2)
where ␦ is the phase angle between E x and E y . In general, the tip of E describes
a locus that is an ellipse. This may be shown by proving that (8.1) and (8.2) with
z = 0 are the parametric equations of an ellipse. With z = 0,
E x = E 1 sin (␻ t) (8.3)
and
E y = E 2 sin (␻ t + ␦ ) (8.4)
Expanding (8.4), eliminating ␻ t, and rearranging terms, leads to
aE x2 − bE x E y + cE y2 = 1 (8.5)
where:
2
a = 1/E 1 sin2 ␦
b = 2 cos ␦ /E 1 E 2 sin2 ␦
c = 1/E 2 sin2 ␦
Equation (8.5) is the equation for an ellipse in its most general form. Figure
8.1 shows an illustration of a polarization ellipse where the axes of the ellipse do
Figure 8.1 Polarization ellipse.

8.3 Stokes’ Parameters 185
not correspond to the x- and y-axes. The ratio of the semimajor axis, OA, to the
semiminor axis, OB, is called the axial ratio, ar:
OA
ar = (8.6)
OB
The axial ratio can also be determined from the amplitudes of the horizontally
and vertically polarized waves, E 1 and E 2 , and the phase angle, ␦ , between them:
√E 1 + E 2 + 2E 1 E 2 sin ␦ + √E 1 + E 2 − 2E 1 E 2 sin ␦
2 2 2 2
ar = (8.7)
√E 1 + E 2 + 2E 1 E 2 sin ␦ − √E 1 + E 2 − 2E 1 E 2 sin ␦
2 2 2 2
Kraus [1] derives an expression for the tilt angle, ␶ , where:
2E 1 E 2 cos ␦
␶ = 0.5 tan−1 2 2
(8.8)
E1 E2
The angle, ␶ , as given by (8.8) is the angle between the x-axis and either the
major or minor axis of the ellipse. The same result is obtained from (8.8) if the
ellipse is as shown in Figure 8.1 or if the ellipse is rotated 90° so that the major
axis is in the y-direction.
Elliptical polarization is completely characterized by the axial ratio, ar, of the
polarization ellipse; the sense of rotation; and the spatial orientation of the ellipse
as determined by the tilt angle, ␶ . Linear polarization has a tilt angle of 0° because
either E 1 or E 2 is zero. Circular polarization has an axial ratio of 1.
The sense of rotation of E that describes either a circle or an ellipse in the
plane of polarization (perpendicular to the direction of propagation) is called the
sense of polarization or handedness [3]. For an observer looking in the direction
of propagation, the sense is called right handed if the direction of rotation is
clockwise and left handed if the direction is counterclockwise. In ELINT applica-
tions, the sense of the polarization of the radar transmitter is determined when
facing the ELINT antenna. Figure 8.2 illustrates the polarization for different
horizontal and vertical component amplitudes and phase angles.
8.3 Stokes’ Parameters [2]
Another way to describe polarization is through the use of Stokes’ parameters.

These are often used to describe radiation from celestial radio sources. Such radia-
tion extends over a wide frequency range and within any bandwidth, ⌬ f, consists
of the superposition of a large number of statistically independent waves of a
variety of polarizations. The resultant wave is said to be unpolarized. For such a
wave, (8.3) and (8.4) are still useful, but E 1 , E 2 , and ␦ are now random variables.
A wave of this type could be generated by connecting one noise generator to a
vertically polarized antenna and another noise generator to a horizontally polarized
Figure 8.2 Polarization ellipse as a function of E 1 /E 2 and the phase angle. (From: [1].  1950
McGraw-Hill, Inc. Used with permission of McGraw-Hill.)
antenna having the same phase center. If the waves appear at an observation point
having the same average power, the total wave will be completely unpolarized. A
partially polarized wave is one where one part may be considered completely
polarized and the other part completely unpolarized.
Stokes’ parameters may be used to characterize all waves, including partially
polarized ones. Stokes’ parameters are defined as follows:
〈E 1 〉 〈E 2 〉
2 2
I = S = Sx + Sy = + (8.9)
Z Z
〈E 1 〉 〈E 2 〉
2 2
Q = S = Sx − Sy = − (8.10)
Z Z
2
U= 〈E E cos ␦ 〉 (8.11)
Z 1 2
2
V= 〈E E sin s␦ 〉 (8.12)
Z 1 2
where S is the Poynting vector; S x and S y are the x and y components of the
Poynting vector; Z is the intrinsic impedance of the medium; E 1 , E 2 , and ␦ are as
defined previously; and the angle brackets indicate the time average.
Normalized Stokes’ parameters are often used and are obtained by dividing
(8.9) through (8.12) by S, where S 0 = I /S, S 1 = Q /S, S 2 = U /S, and S 3 = V /S. If
any of the parameters, Q, U, or V (or S 1 , S 2 , or S 3 ), have a nonzero value, it indicates
the presence of a polarized component in the wave. The degree of polarization, d,
is defined as the ratio of completely polarized power to the total power, or
2 2 2
d = S1 + S2 + S3 (8.13)
2 2 2
If S 1 + S 2 + S 3 = 1, the wave is completely polarized.
For a completely polarized wave, Stokes’ parameters do not require time aver-
ages and the brackets denoting time average can be removed from (8.9) through
(8.12).
Substituting into (8.7) and (8.8), the axial ratio and tilt angle in terms of Stokes’
parameters are
√S 0 + S 3 + √S 0 − S 3
ar = (8.14)
√S 0 + S 3 − √S 0 − S 3
and
S2
␶ = 0.5 tan−1 (8.15)
S1
Table 8.1 gives the values of the normalized Stokes’ parameters for seven
polarized wave states. Stokes’ parameters are measurable using an antenna system
and some amplitude signal processing.
Jamming signals may require Stokes’ parameters to be measured to describe
their polarization.
8.4 Measuring Polarization

There are a number of techniques to measure the polarization of a wave, as can
be seen by the listing shown in Table 8.2. IEEE STD 149-1979 lists three broad
categories and four measurement techniques. Kraus [1] listed three techniques, but
two are essentially the same phase-amplitude techniques using different antennas.
Basically there are three techniques that could be used at a collection site:
1. Polarization pattern method;

2. Phase-amplitude method;
3. Multiple antenna (amplitude-only measurement) method.
For wideband operation, the polarization pattern method is not as easily imple-
mented as the other two methods. Modern receiver systems make phase-amplitude
Table 8.1 Normalized Stokes’ Parameters for Seven Polarization Wave States
Normalized Completely Completely Polarized Waves
Stokes’ Unpolarized Linearly Polarized Circularly Polarized
Parameter Wave ␶ = 0° ␶ = 90° ␶ = 45° ␶ = 135° Left Hand Right Hand
S0 1 1 1 1 1 1 1
S1 0 1 −1 0 0 0 0
S2 0 0 0 1 −1 0 0
S3 0 0 0 0 0 1 −1
Source: [2].
Table 8.2 Polarization Measurement Techniques

Kraus [1] IEEE STD 149-1979 Kanareikin [4] Hollis et al. [5]
1. Polarization pattern Broad Categories 1. Polarization diagram 1. Phase-amplitude of
method using 1. Techniques that give method linear polarized
rotating linear partial information 2. Compensation components
antenna and includes about antenna’s method 2. Phase-amplitude of
a measurement of polarization 3. Orthogonally circular polarized
sense 2. Techniques that give polarized components
2. Linear component complete polarization components phase 3. Multiple antenna
method: measure information but and amplitude amplitude
amplitude of two require comparison 4. Method of several or measurements
orthogonal linear to a polarization one antenna with 4. Polarization pattern
polarized components standard (comparison variable polarization to get tilt angle and
and phase angle techniques) 5. Modulation method axial ratio; use
between them 3. Techniques that separate antenna for
3. Circular component require no sense
method: same as #2, comparisons and
only using circular provide complete
polarized (with polarization
opposite senses) information on an
components. antenna (absolute
methods)
Measurement
Techniques
1. Polarization pattern
method
2. Rotating source
method
3. Multiple-amplitude
component method
4. Phase-amplitude
method
measurements for fine-grain analysis. The multiple antenna, or amplitude-only,

method makes use of Stokes’ parameters and amplitude-only measurements from
four antennas simultaneously or from one antenna capable of variable polarization.
8.4.1 Polarization Pattern Method

The polarization pattern method makes use of a rotating linearly polarized antenna.
The received signal level is recorded at each position of the rotating antenna and
is a measure of the component of the E field in the direction of the rotating antenna
at any instant. If the radar is also scanning, then the rotation of the receiving
antenna’s polarization must be fast enough to complete several revolutions within
the time the main beam is directed toward the intercept site.
These measurements provide a polarization pattern from which the polarization
ellipse can be determined. From the polarization pattern, the axial ratio and tilt
angle can be determined (see Figure 8.3).
To determine the polarization sense, an additional measurement is required.
Generally, the outputs of two circularly polarized antennas with opposite senses
(one right handed and one left handed) are compared in order to determine the
polarization sense. Thus, by specifying the axial ratio, tilt angle, and polarization
Figure 8.3 Polarization patterns and polarization ellipse.
sense, the polarization characteristics of a wave are completely described. Figure 8.4
shows photographs of the polarization pattern for linear, elliptical, and circularly
polarized waves [6].
8.4.2 Phase-Amplitude Method

The phase-amplitude method requires that either two orthogonal, linearly polarized
antennas or two circularly polarized antennas of an opposite polarization sense be
connected to a receiver that can measure the amplitude of each component and
the phase angle between them.
From a knowledge of E 1 and E 2 , along with the phase angle, the axial ratio
can be determined from (8.7), the tilt angle from (8.8), and the direction of rotation
from (8.3) and (8.4). Kraus [1] has created a chart of polarization ellipses as a
function of the ratio E 2 /E 1 and the phase angle ␦ . In this chart, reproduced above
Figure 8.4 Photographs of polarization patterns. (From: [6].  1951 Horizon House, Inc. Reprinted
with permission.)
as Figure 8.2, the wave is approaching the viewer; therefore, a clockwise wave
approaching the viewer is a left-hand circular polarization and a counterclockwise
wave is a right-hand circular polarization.
To make polarization measurements over a broad frequency range, one could
use a broadband horn antenna as the feed for a parabolic dish with one probe for
horizontal polarization and another for vertical. Crossed log periodic antennas
could also be used as the feed to a dish. A receiver that measures the phase difference
between two pulsed signals permits evaluation of the polarization in subsequent
processing. Errors result if there is more than one signal in the receiver’s pass band.
One method to display polarization is to mix the pulsed RF signals to a
frequency within the bandwidth of an x-y oscilloscope. Monitoring the Lissajous
figure gives an indication of the polarization. As an antenna sweeps (as in a circular
scan) through the beam of a receiving antenna, the indicated polarization can be
expected to vary, especially if multipath is present. When the peaks of the two
beams are ‘‘looking’’ at each other, the polarization can best be measured (beam-
on beam data).
8.4.3 Multiple Antenna Method

It is possible to measure polarization by amplitude measurements from multiple
antennas without making phase measurement [7, 8]. The procedure outlined here
based on Kraus [7] makes use of Stokes’ parameters. As shown by (8.7) and (8.8)
and (8.14) and (8.15), the parameters of a polarization ellipse can be derived from
either the Stokes’ parameters or from phase-amplitude measurements.
Let the antenna system consist of four linearly polarized antennas and two
circularly polarized antennas with an opposite polarization sense, as shown in
Figure 8.5. The received wave is incident on each antenna in a normal direction.
The output power (signal level) from each antenna is measured either by separate
receivers or one receiver sampling each antenna. The power received from each
antenna may then be designated by:
Px = horizontal antenna
Py = vertical antenna
Px′ = slant 45° antenna
Py′ = slant 135° antenna
P L = left-hand circular antenna
P R = right-hand circular antenna
Figure 8.5 Six antennas used for polarization measurements of an unknown wave. Note that for
ELINT uses, the antennas should have the same phase center. (From: [7].  1982 John
Kraus. Reproduced by permission.)
All six antennas should provide an equal signal level to a completely unpolarized
wave (i.e., have the same effective aperture). From (8.9) through (8.12),
I P x + Py
S0 = = (8.16)
I P x + Py
I P x + Py
S1 = = (8.17)
I P x + Py
U Px′ − Py′
S2 = = (8.18)
I P x + Py
V P L + Py
S3 = = (8.19)
I Px + P y
Assuming the sums of each of the power pairs (Px + Py ), (Px′ + Py′ ), and (P L + P R )
are unity, the normalized Stokes’ parameters S 1 , S 2 , S 3 are then related as shown
in Table 8.3.
From Table 8.3, it is clear that the outputs of pairs of antennas are either
summed (S 0 ) or differenced (S 1 , S 2 , and S 3 ) to obtain the normalized Stokes’
parameters. Although the same data (normalized Stokes’ parameters) could be
obtained from four antennas versus the six used here, the use of four antennas
complicates the calculations beyond the simple sum and differences of this method
[7]. Six antennas are recommended and have proven useful in data collection
systems [8]. Kanareikin [5] presents, in considerable detail, polarization measure-
ment techniques and equipment implementations.
8.5 Cross-Polarization
Practical antennas radiate some energy with polarization orthogonal to that

intended. This is referred to as cross-polarized radiation. The cross-polarized signal
from a reflector antenna increases as the curvature increases. This is shown in
Figure 8.6, which gives the cross-polarization level in terms of the power loss in
the main pattern. Note that although the loss seems small, 0.1 dB represents 2.3%
of the power, which may be observed by ELINT. In this figure, the half angle refers
to the angle between a line representing the boresight direction of the reflector and
a line drawn from the focus to the edge of the reflector.
Table 8.3 Relation of Normalized Stokes’ Parameters to Antenna Pairs

Stokes’ Vertical and
Parameters Horizontal Antennas Slant Antennas Circular Antennas
S0 Sum Sum Sum
S1 Difference
S2 Difference
S3 Difference
Figure 8.6 Cross-polarization loss for a reflector antenna.
The cross-polarized signal level may actually exceed that of the proper or co-
polarized signal when sidelobes are received. If the ELINT site is unknowingly
using a cross-polarized antenna, and then finds the peak signal level, this will not
represent the true ERP of the radar. Because it is common to not be in the main
beam of the radar, the effect of cross-polarized energy can be more significant for
the ELINT observer than it is for the radar designer. Consider the measurements
shown in Figures 8.7 and 8.8 (furnished by Advent Systems, Inc.). Figure 8.7 is
the power received from an ASR-3 radar when using a vertically polarized antenna
to match the vertical polarization of the ASR-3. The peak of the radar beam in
elevation is 3° above the ELINT site; therefore, this pattern is taken 7.4 dB below
the peak of the beam. Figure 8.8 shows the cross-polarized pattern taken with a
horizontally polarized receiving antenna. Although the peak of the beam is about
10 dB lower on the cross-polarized pattern, the cross-polarized pattern is actually
above the copolarized pattern at many azimuths. Note the dip in the center of the
main beam of the cross-polarized pattern. Clearly, the ELINT site must examine the
received signal using several polarizations on receive to learn about the transmitted
polarization.
The polarization tilt angle is shown in Figure 8.9 for 2.6° on each side of the
peak of the beam in azimuth. Three elevation cuts are shown: at the peak of the
beam in elevation and 3° above and below the elevation boresight. Inside the 3-dB
points of the azimuth beam, the axial ratio basically indicates vertical polarization.
Figure 8.7 Copolarized pattern of an ASR-3 radar antenna.
Figure 8.8 Cross-polarized pattern of an ASR-3 radar antenna.
However, the axial ratio indicates polarization varying from about 45° to 150°
over just 5.2° of azimuth. This shows the importance of measuring the polarization
in the main beam. Complete reversals of the polarization occur in the sidelobes of
the pattern.
Figure 8.9 Tilt angle of the ASR-3 antenna polarization near boresight.
The consequences of unknowingly using a cross-polarized antenna on receive

can be severe. Not only is the ERP greatly underestimated, but the bearing to the
emitter, as determined by the peak signal when the receive antenna is rotated, can
be in error. This happens because the peak of the response of the receive antenna
in the cross-polarized mode is generally offset from the boresight of the copolarized
beam. Likewise, the pulse shape of the weaker cross-polarized signal may be cor-
rupted by multipath and noise.
8.6 Propagation Effects
As a wave propagates through the atmosphere, the polarization is rotated. This is

due to the Faraday effect and is caused by free electrons. Figure 8.10 shows the
amount of rotation for a two-way trip through the entire atmosphere at several
elevation angles during the day and during the night. There is less rotation during
the night because the Sun’s energy is largely responsible for creating the free
electrons. There is less rotation at the higher elevation angles because the path of
the wave traverses less atmosphere vertically (90°) than at the horizon (0°). At
higher frequencies, the rotation is quite small—less than 0.1 radian above 5 GHz
in the worst case: of two-way propagation through the entire atmosphere at the
horizon in the daytime. At 1 GHz, under the same conditions, there may be about
5 radians of rotation. At short range, and for one-way ELINT propagation, this
effect is often negligible. At lower frequencies and long range, however, Faraday
rotation must be considered and taken into account.
8.7 System Aspects of Polarization 195
Figure 8.10 Two-way polarization rotation for a target at a height of 1,000 km.
8.7 System Aspects of Polarization
The choice of polarization for a radar design may be made on the basis of the
function of the radar. Ground or sea reflections are much greater for horizontal
than for vertical polarization. A long-range search radar can take advantage of
this by obtaining twice the detection range at certain elevation angles and no
coverage at other elevation angles [9]. For an airborne radar (or ground radar) to
detect targets in or behind heavy rain cells, circular polarization is generally selected.
There is a polarization effect on a radar cross-section that depends on the target
area (e.g., a long thin wire produces little reflection when the incident field is
linearly polarized at right angles to the wire axis and produces maximum reflection
when the wave polarization is parallel to the axis of the wire). For any given target
and aspect angle, there is a polarization of the incident field (wave) that gives a
maximum echo (received signal). Conversely, there is a polarization that also gives
a minimum signal. Therefore, use of polarization diversity by the radar could
maximize the signal return. Some immunity from jamming is also obtained (unless
a noise jammer uses two orthogonal linear antennas with noise generators to
produce an unpolarized wave).
The selection of a particular polarization for the same function may be different
from one location (country) to another. In the United States, for instance, television
band antennas are horizontally polarized. This choice was made to maximize the
SNR because most man-made noise sources are vertically polarized. However, in
some other places, the same television band uses vertical polarization because it
provides the maximum signal strength from a wave-propagation point of view.
Propagation at various frequencies can sometimes dictate a particular polarization,
as shown in Table 8.4 [10].
Table 8.4 Polarization Choice Versus Frequency

Frequency Polarization Comment
VLF Vertical Horizontal polarization will be cancelled by
image antenna
HF Mostly vertical
UHF/VHF Vertical or horizontal Some horizontal may be used for specific
application
Microwave Choice may be made Vertical provides maximum signal.
(above 1,000 MHz) based on function Horizontal discriminates against man-made
noise:
1. Circular for detection of targets in rain or
for transmission through the ionosphere at
any frequency due to Faraday rotation
2. Horizontal for clutter discrimination
3. Vertical to minimize lobing due to the
ground
4. Systems design (antenna and radome
design may affect polarization choice)
Polarization measurements can aid in understanding the radar function and in

analysis of the ELINT data. Detection of a rotating linear polarization may lead
to the conclusion that the radar has a conical scan mode.
Table 8.5 gives the polarization loss (take 10 log P R /Pmax for the loss in
decibels) for various transmit and receive antenna configurations. For two linearly
polarized antennas, the polarization power loss is inversely proportional to the
cosine of the angle between their polarizations.
Knowledge of basic information regarding wave polarization is essential. Some-
times polarization measurement is overlooked due to the need for special equipment.
Advances in antennas, microwave components, and signal processing gear may
mean that polarization measurements will receive more attention in the future. The
interested reader is directed to [1] for basic information and to [3] for an excellent
bibliography on all subjects related to antenna measurements.
8.8 Antenna Beam Shape

The usefulness of radar depends heavily on the ability to determine the direction
to targets of interest. Indeed, even in the absence of target range, the direction
Table 8.5 Polarization Loss for Various Transmit/Receive Antenna Combinations

Transmit Antenna Receive Antenna Ratio of Power Received
Polarization Polarization to Maximum Power
Vertical Vertical 1
Vertical Slant (45° or 135°) 1/2
Vertical Horizontal 0
Vertical Circular (right hand or left hand) 1/2
Horizontal Horizontal 1
Horizontal Slant (45° or 135°) 1/2
Horizontal Circular (right hand or left hand) 1/2
Circular (right hand) Circular (right hand) 1
Circular (right hand) Circular (left hand) 0
Circular (right hand or left hand) Slant (45° or 135°) 1/2
8.9 Basic Antenna Pattern Relationships 197
to the target may be sufficient information to engage it. Therefore, the angular
measurement capability of the radar is a key aspect of its design.
This being the case, the ELINT analyst is charged with determining (as much
as possible) all of the radar’s directional properties. Many radar systems make use
of a highly directional antenna that is used for both transmitting and receiving.
The antenna pattern of this type of radar usually can be determined rather com-
pletely from ELINT. Systems that use separate transmit and receive antennas are
less amenable to ELINT analysis because little can be learned about the receive
antenna. The assessment of radar antenna capabilities and characteristics is much
more successfully completed if the radar antenna can be seen or photographed in
addition to the reception of the radar signal by an ELINT station. Photographs
are often available from sources such as trade publications, export sales versions,
parades, or photoreconnaissance. The fundamental difficulty is that there is often
no way to associate a photograph of a particular antenna configuration to the radar
signal received at an ELINT station. Analysis approaches that require knowledge of
the antenna’s physical configuration as well as the ELINT information are often
used.
Except in the case of radar tracking a target, the radar may search for targets
by moving its directional beam in azimuth and elevation—either mechanically or
electronically. This motion of the beam past an ELINT station provides an indica-
tion of the antenna’s radiation pattern. Determining the properties of the directional
beam pattern is called beam analysis. Determining the types, rates, and angular
extent of the beam motion is called scan analysis. Scan analysis is the subject of
Chapter 10. However, insight into the beam pattern generally requires knowledge
of the scanning action.
8.9 Basic Antenna Pattern Relationships
The antenna pattern can be characterized by parameters such as its 3-dB azimuth
and elevation beamwidths, gain, and sidelobe levels. It can also be described by
plots of the field strength (at a fixed distance from the antenna) as a function of
angle. Historically, the antenna theory is based on the theory of the diffraction of
light. As the antenna size increases with respect to the wavelength, the approxima-
tions improve.
Basically, the far field of the antenna pattern can be thought of as the sum of
the contributions from various parts of the radiating4 area. These contributions
add vectorially, and the various phases create peaks and valleys in the radiation
pattern. The radar antenna is usually designed to concentrate the radiation in a
single direction,5 both to locate the target directionally and to concentrate the
available power for increased detection range. The availability of a sharply defined
beam alone does not provide the azimuth and elevation directions to the target,
because a means of accurately pointing the beam is also needed.
4. While ‘‘radiation’’ is the term used, antenna structures for our purposes are reciprocal and can be used
for reception as well as transmission.
5. An important exception is the difference pattern of monopulse radar receivers.
The far field of the antenna pattern refers to the region beyond [11]
2D 2
␭
where D is the maximum antenna dimension and ␭ is the wavelength. Also, the
distance must exceed eight wavelengths. Obviously, ELINT observations are in the
far field.
Radar antennas are characterized by an aperture that is illuminated with electro-
magnetic energy. For example, a parabolic reflector combined with a feed horn
produces a constant phase in a front of the reflector. The illuminated aperture,
therefore, is thought of as a plane area directly in front of the reflector and feed
assembly and normal to the conic axis or line drawn through the center of the
reflector and its focal point. The illumination function describes the variation of
the excitation across the aperture. The far field for many aperture illumination
functions can be described as the Fourier transform of the illumination function.
This means that narrow beams in the far field require large apertures. Likewise,
illumination functions that are constant across the aperture and drop suddenly at
the aperture edge produce high sidelobes, while those illumination functions that
taper off gradually toward the edge of the aperture produce lower sidelobes but
less sharply defined beams. Antenna theory shows that the sharpest beam (and
highest gain) is obtained when the illumination is constant over the aperture. The
beamwidth is defined as the angular width of the main beam between the points
of the far-field pattern that are 3 dB below the peak. The directive gain of the
antenna is the ratio of the peak of the far-field pattern to the total pattern averaged
over the entire spherical angle of 4␲ . Thus, losses in the antenna are ignored in
determining the directive gain. While the ELINT analyst may be able to determine
the directive gain, the actual power gain (directive gain reduced by antenna losses)
cannot be determined from ELINT alone.
For a given aperture, uniform illumination provides the maximum directive
gain. The actual directive gain of an actual aperture therefore is less than this value:
4␲ A 4␲ LW
GD ≤ 2
= (8.20)
␭ ␭2
where:
G D = directive gain of aperture

A = area of aperture (m2)
L = length of rectangular aperture
W = width of rectangular aperture
␭ = wavelength (m)
For uniform aperture illumination, the directive gain can be related to the 3-dB
beamwidths of the pattern by the approximation:
4␲
GD ≈ (8.21)
␪ AZ ␪ EL
where:
G D = directive gain
␪ AZ = azimuth bandwidth (rad)
␪ EL = elevation beamwidth (rad)
Equation (8.21) is the ratio of the area of a unit sphere to that part of the
area covered by a narrow beam of angular extent ␪ AZ × ␪ EL . Converting this to
beamwidths measured in degrees requires multiplication by (180/␲ )2. Combining
(8.20) and (8.21) yields
␭2
LW ≈ (8.22)
␪ AZ ␪ EL
Thus, estimating the beamwidths provides insight into the aperture’s dimensions
and vice versa. This is a valuable aid in deciding whether a photograph of an
antenna is associated with a particular signal. Likewise, the sidelobe levels can be
helpful in determining the character of the illumination function and the extent to
which the aperture size exceeds the value given by (8.22).
8.10 Beam Patterns from ELINT
As a simple example, consider a search radar with an elevation fan beam scanning
circularly in azimuth. The ELINT signal level at an ELINT station varies with the
angular position and reaches a peak when the radar antenna is directed toward
the ELINT station. This variation is repeated, and one main beam peak is separated
in time from the next by the rotation time of the circular scan. If the signal level
is accurately recorded over a wide dynamic range, the ELINT analyst obtains the
azimuth antenna pattern. From this, the angular position from the main beam peak
can be related to the time base of the plot because the separation of the main beam
peaks corresponds to 360°.
It is difficult to maintain an accurate amplitude scale from the receiver antenna
over a wide dynamic range. The sidelobe structure of interest may be obscured by
receiver noise or multipath effects. Saturation of the receiver may make the 3-dB
beamwidth difficult to determine directly; therefore, the time locations of the peaks
and nulls can often be discerned more precisely than the −3-dB points. For most
practical antennas, the locations of peaks and/or nulls can be related, at least
approximately, to the 3-dB beamwidth.
For a rectangular aperture, the antenna gain as a function of angle was given
in (7.5). From this the 3-dB azimuth beamwidth is given approximately as
2.78␭
␪ 3 dB = (8.23a)
␲L
Assuming the elevation angle is zero, the first null in azimuth is located at an
angle such that
␲L
sin ␪ 1 = ␲ (8.23b)
␭
or
␭
sin ␪ 1 = (8.23c)
L
For small values of ␪ 1 , the ratio of the 3-dB beamwidth to the width between
the first nulls is
␪ 3 dB 2.78
≈ = 0.44 (8.24)
2␪ 1 2␲
For a circular aperture, a similar calculation yields ␪ 3 dB /2␪ 1 ≅ 0.42, as indicated

by the graph in Figure 8.11. Similar ratios can be computed for comparing the
3-dB beam width to the separation of the sidelobe peaks.
The 3-dB beamwidth is broadened by tapering of the illumination function to
reduce the sidelobes. In fact, a radar having the −13-dB first sidelobes that character-
Figure 8.11 Far-field antenna patterns for circular and rectangular apertures (uniform illumination).
(From: [11].  1970 McGraw-Hill, Inc. Used with permission of McGraw-Hill.)
ize a uniformly illuminated rectangular aperture would not be very useful if present
on both the transmit and receive antenna patterns. Radar cross-sections of common
targets can differ by many decibels, so that a large target return in such sidelobes
could obscure a small target return in the main lobe. Likewise, strong clutter returns
could arise due to the sidelobe energy striking the ground. Generally speaking, the
peak sidelobe level for the two-way antenna pattern must be more than 40 dB
below the main lobe for a practical radar system. In radar systems that use active
transmit arrays, it is currently not practical to taper the illumination on transmit
to reduce the sidelobes. Uniformly illuminated antennas can be termed ‘‘ultra high
sidelobe antennas.’’ In such systems, sidelobe reduction is achieved on receive by
tapering the received signals. Note that in airborne radar systems, it is necessary
to keep the aperture as small as possible. The need to avoid returns from the
ground through the sidelobes means that the two-way sidelobes are maintained
∼40 dB below the main lobe by tapering the aperture illumination function. In
ground and shipboard applications, aperture size is not quite as limited. However,
the ground or sea is much closer in range. These radars usually have two-way
sidelobes more than 40 dB below the main beam, resulting in even greater widening
of the beam relative to that of a uniformly illuminated aperture. Very few radars
have antennas with one-way sidelobes more than 40 dB below the main beam.
Those that do are termed ‘‘ultra low sidelobe antennas.’’ As the technology of
microwave transmit amplifiers improves and amplitude programming is available,
tapering the transmit illumination function may occur on active arrays, which, at
present, have high sidelobes on transmit. This is a favorable situation for ELINT
work because it allows observation of the transmitted signal’s sidelobes at higher
levels; however, the sidelobe level on receive is not observable from ELINT. Never-
theless, knowledge of radar system requirements for good performance combined
with the observation of the transmit sidelobe level gives insight into the likely range
of sidelobe levels on receive.
The ELINT analyst can have a reasonable idea of the probable relationship
between the aperture size, 3-dB beamwidth, and sidelobe level on transmit for
various classes of radar systems. Uniform illumination produces the highest gain
and the highest sidelobe levels. The natural question to ask next is what aperture
distribution gives the highest gain and narrowest beamwidth for any given level
of peak sidelobes? The answer is the Taylor distribution. To specify this distribution,
it is necessary to choose the maximum level of the sidelobes and the parameter n,
which specifies the number of sidelobes having the maximum level. Figure 8.12
illustrates several such distributions, and Figure 8.13 illustrates the beam pattern
for sidelobes 25 dB down and n, equal to 5. From the viewpoint of radar design, this
illumination function is difficult to realize for reflector antennas fed by conventional
feeds. In fact, the Taylor distribution is generally associated with array antennas.
Nevertheless, from the ELINT viewpoint, it provides an interesting limiting case.
For an observed sidelobe level, the Taylor distribution gives the maximum possible
gain and narrowest beamwidth for a specified antenna dimension.
For rectangular apertures, the Taylor pattern can be thought of as the superpo-
sition of sin (x)/x beams. For circular apertures, the patterns are very similar and
can be thought of as a superposition of the J 1(x)/x Bessel function beams. Figure
8.14 shows the beam broadening and gain reduction for the Taylor distribution
Figure 8.12 Taylor n aperture distributions.
Figure 8.13 Taylor line source pattern for sidelobe = 25 dB, n = 5.
Figure 8.14 Taylor broadening and aperture efficiency.

in two dimensions. Here n has been selected to be as large as possible while still
maintaining a monotonically decreasing aperture distribution. The gain reduction
is computed as if the same aperture distribution and sidelobe levels are specified
for both the azimuth and elevation directions. Note that the gain reduction is
related to the aperture efficiency: gain reduction (dB) = −10 log (aperture efficiency).
Aperture efficiency is a measure of the extent to which the aperture is fully illumi-
nated—uniform illumination represents 100% aperture efficiency.
From the ELINT viewpoint, if the sidelobe level is measured, Figure 8.14 can
be used to estimate the beam broadening factor (BBF) and gain reduction relative
to a uniformly illuminated aperture of the same size. Then, if the azimuth and
elevation beamwidths are also measured, one can estimate the antenna dimension
using (8.23a) plus the knowledge of the BBF and the wavelength:
Antenna dimension = (0.88 BBF)(wavelength)/3-dB beamwidth)
The relationship of the width between the first nulls and the 3-dB beamwidth
for the Taylor distribution is illustrated in Figure 8.15. For a uniformly illuminated
antenna, this ratio is 0.44. If the sidelobe level is estimated and the width between
the nulls is measured, Figure 8.15 provides a way to estimate the 3-dB beamwidth
from the null-to-null beamwidth.
Although the theoretical basis for a low-sidelobe antenna design is available,
practical considerations limit the extent to which the sidelobes can be suppressed.
For reflector antennas, the feed at the focus of the reflector blocks part of the
aperture and this raises the level of the sidelobes. Figure 8.16 illustrates this effect
for a circular reflector illuminated with a Hansen one-parameter distribution [13]
for a center-fed reflector. Suppose the design calls for a −40-dB sidelobe level,
assuming no aperture blockage. (This is the point at the left of the lowest curve
in Figure 8.16.) If the diameter of the feed is 0.1 times the diameter of the reflector,
Figure 8.15 Taylor illumination: ratio of 3-dB beamwidth to null-to-null beamwidth.

Figure 8.16 Sidelobe level versus the blockage diameter ratio for Hansen one-parameter circular
distribution.
the sidelobe level will rise to about −28 dB. Of course, this blockage also reduces
antenna efficiency. Many search systems use reflector antennas with offset feed.
This greatly reduces aperture blockage compared to that for a center-fed reflector.
The sidelobe level rises and the gain is reduced by inaccuracies in the reflector
surface. These effects are shown in Figures 8.17 and 8.18. Figure 8.17 shows how
the sidelobe level is affected, and Figure 8.18 shows how the gain is affected. In
Figure 8.17 Mean sidelobe level versus universal factor.

Figure 8.18 Reflector directivity loss versus rms tolerance.
these figures, the reflector imperfections can be related by noting that the rms
tolerance (⑀ ) in Figure 8.18 is 0.5 times the tolerance (␴ ) in Figure 8.17.
8.11 Beam Patterns of Array Antennas
The basic idea of any array is to create an illuminated aperture, just as any other
antenna. These arrays use discrete elementary antennas spread over an area and
control the signal to each one in order to form and steer the beam. The major
difference to the radar designer (and to an ELINT observer) is the speed with which
the beam can be moved—the problem of the inertia of a large structure is overcome.
Such a beam motion (scanning) is considered in Chapter 10. The beam pattern of
array antennas is affected by the discrete spacing of the elementary antennas (and
their individual beam patterns) as well as by the methods for controlling the
signal to each element. The discrete elements typically are equally spaced across
an aperture with a spacing of one-half the wavelength. The directive gain is related
to the aperture dimension [see (8.20)]; therefore, if the elements are spaced ␭ /2
apart, the gain is related to the number of elements by
GD ≤ ␲ NL NW (8.25)
where:
N W = 2W /␭ = number of elements along the width of the aperture

N L = 2L /␭ = number of elements along the length of the aperture
Also, because the 3-dB beamwidths are related to the aperture dimensions
(approximately) by
␭
␪ AZ = (8.26a)
L
and
␭
␪ EL = (8.26b)
W
it can be said that the approximate number of elements in an array is
␭
NL ≈ (8.27a)
␪ AZ
and
␭
NW ≈ (8.27b)
␪ EL
Of course, these estimates should be adjusted for the nonuniform aperture

illumination functions used for sidelobe control. This can be done using Figures
8.14 and 8.15.
The foregoing applies for beams directed normal to the aperture or broadside.
As the beam is steered away from broadside, the beamwidth (and gain) changes
because the effective aperture is smaller when viewed from a direction other than
normal. The gain is reduced by a factor of cos ␪ , where ␪ is the beam pointing
angle as measured from normal to the aperture.
The spacing of the elements need not be ␭ /2; however, it is economical to use
the fewest number of elements possible. The maximum spacing is related to the
appearance of multiple lobes (called grating lobes) in the antenna pattern. The
element spacing required to prevent the appearance of grating lobes depends on
how far from broadside the beam is to be pointed. The spacing limit is [11]
␭
S≤ (8.28)
(1 + sin ␪ )
where:
S = element spacing
␭ = wavelength
␪ = angle away from broadside
Because | sin ␪ | ≤ 1, a spacing of ␭ /2 or less guarantees that only one main

lobe appears. The maximum element spacing as a function of the maximum scan
angle is shown in Figure 8.19.
The number of independent beam positions in a sphere is approximately equal
to the antenna gain if the gain is constant. Away from broadside, the gain is reduced
(and the beam broadened) so that the number of independent beam positions is
reduced to approximately (␲ /2) (N L ⭈ N W ) [1].
Figure 8.19 Maximum element spacing as a function of the maximum scan angle.
The control of the beam pointing angle is critical if electronic scanning is to

be used. However, some radar systems use array antennas with fixed illumination
functions that are pointed mechanically. The advantages are that good control of
the aperture illumination is possible and the peak power required of each element
is reduced by the number of elements used.
If the antenna illumination is controlled by digital techniques (e.g., digital phase
shifters), the quantization errors act to reduce the gain and increase the sidelobes.
The gain reduction [11] is shown in Table 8.6. The rms sidelobe level due to
quantizing is approximately at an isotropic level (i.e., below the main beam by an
amount equal to the main beam gain), increased by the gain loss shown in Table 8.6.
Clearly, four to eight states in the phase shifter are sufficient to reduce quantization
sidelobes below the sidelobe levels due to the illumination of the finite aperture
[12]. This situation is to be expected by the ELINT analyst.
The energy lost from the main beam appears in the sidelobes, and the rms
sidelobe level due to the quantization in the phase shifters is approximately given
by the ratio of the loss in gain to the average main beam gain.
An illustration of the degradation that can occur is shown in Figure 8.20 [14].
The antenna pattern as designed is illustrated on the left. On the right is the pattern
with ±2-dB amplitude errors, ±10° random phase errors, and 5% random element
failures.
Table 8.6 Number of Phase Shift States Versus the Loss in Gain
Due to Quantization
Number of Phase Shift States Loss in Gain Due to Quantization
4 1.0 dB
8 0.23 dB
16 0.06 dB
Figure 8.20 Effects of phase and amplitude errors. (From: [14].  1987 Horizon House, Inc.
Reprinted with permission.)
8.12 Antenna Beam Summary
ELINT measurements of beam patterns are often made possible by the motion of
the radar antenna beam due to its scanning (i.e., angular searching); therefore, the
principles and techniques described in Chapter 10 are often used in conjunction
with those described here. The importance of beam pattern information is clear.
The directive gain is one of the three most important factors that determine the
capabilities of a radar (the other two being the power transmitted and integration
time). The 3-dB beamwidth largely determines the ability of the radar to resolve
targets in the angular coordinates in space. This, in turn, heavily influences the
type of search pattern the radar must employ. (A radar with a wide elevation beam
and a narrow azimuth beam will search by moving the beam in azimuth, and vice
versa.) Likewise, the sidelobe levels and angular positions can be important in
developing effective jammers to work against the radar. The sidelobe levels also
determine the extent to which the radar is affected by ambiguous angular responses.
For example, a strong point target can appear at many angles as long as the sidelobe
levels exceed the minimum detectable signal level for that radar’s receiver.
An example helps illustrate the principles of beam analysis discussed here.
Suppose an ELINT measurement shows a circular scan that repeats every 10
seconds. This means that the antenna rotates 360° in 10 seconds. The measurement
also shows that the first nulls are separated by 0.1 second, which represents 3.6°.
If uniform illumination were used, the 3-dB beamwidth would be 0.44 times the
width between the nulls, or 1.58°. Because acceptable radar performance requires
sidelobes that are lower than that of a uniformly illuminated aperture, Figure 8.14
can be used to find the ratio between the 3-dB beamwidth and the null-to-null
beamwidth for any measured or estimated sidelobe level for the Taylor illumination
function. If the peak sidelobe level is 30 dB down, a ratio of about 0.40 would be
expected. This means the 3-dB beamwidth would be at least 1.44° based on the
ELINT data and an assumed 30-dB sidelobe level.
8.12 Antenna Beam Summary 209
Next, suppose that it is necessary to decide whether or not an observed ELINT

signal could be radiated by a radar sold internationally, but whose specifications
are not fully known. The dimensions of the reflector antenna are known from
advertising to be 4.5m wide by 1m high. Suppose the frequency of operation from
ELINT is 3 GHz. This means the wavelength is 0.1m. Using (8.23a), this means
that the azimuth beamwidth (radians) is 0.88 times the ratio of the wavelength
(0.1m) to the horizontal dimension of the antenna (4.5m)—assuming uniform
aperture illumination. This works out to 0.0196 rad or 1.12°. Because the observed
beam width from ELINT is 1.44°, this means that the beam broadening ratio is
1.44/1.12 = 1.29. Figure 8.14 shows that this beam broadening ratio is consistent
with a −30-dB sidelobe level. From this evidence, it may be concluded that the
observed beam is consistent with a reflector of this size and that the gain reduction
expected due to a nonuniform aperture illumination in both azimuth and elevation
is about 1.3 dB. This implies that −30-dB sidelobes in elevation are assumed. The
corresponding elevation beamwidth based on the 1m height of the reflector and
the 0.1 wavelength from ELINT is (0.88)(1.29)(0.1/1) = 0.114 rad or 6.5°.
To estimate the gain, either the dimensions or the beamwidths can be used.
Using (8.21), the gain from the dimensions is
4␲ (LW )/(wavelength)2 = 5,564.8 or 37.5 dB
This assumes uniform illumination; therefore, the gain should be reduced by

1.3 dB due to the beam broadening caused by the nonuniform aperture illumination,
resulting in 36.2 dB for the estimated antenna gain.
References
[1] Kraus, J. D., Antennas, New York: McGraw-Hill, 1950, pp. 464–484.
[2] Kraus, J. D., and K. R. Carver, Electromagnetics, New York: McGraw-Hill, 1973,
Ch. 11 and 14.
[3] ‘‘Test Procedures for Antennas,’’ Section II, IEEE STD 149-1979, December 19, 1979.
[4] Kanaraeikin, D., et al., Polyarizatsiya Radiolokatsionnykh Signalev [Polarization of Radar
Signals], Moscow: Sovetskoye Radio Press, 1966, English Translation of Chapters 4 and
11 are available from NTIS as AD-783 761, May 1974.
[5] Hollis, J. S., T. J. Lyon, and L. Clayton, Jr., Microwave Antenna Measurements, Atlanta,
GA: Scientific Atlanta, 1970.
[6] Rumsey, V. H., et al., ‘‘Techniques for Handling Elliptically Polarized Waves with Special
Reference to Antennas,’’ Proc. IRE, May 1951, p. 533.
[7] Kraus, J. D., Radio Astronomy, New York: McGraw-Hill, 1966.
[8] Clayton, L., and S. Hollis, ‘‘Antenna Polarization Analysis by Amplitude Measurements
of Multiple Components,’’ Microwave Journal, Vol. 8, January 1965, pp. 35–41.
[9] Skolnik, M. I., Radar Handbook, New York: McGraw-Hill, 1970, pp. 236–238.
[10] Blake, L. V., Antennas, New York: John Wiley & Sons, 1966.
[11] Skolnik, M. I., Radar Handbook, New York: McGraw-Hill, 1970, Ch. 9–13.
[12] Stark, L., ‘‘Microwave Theory of Phased-Array Antennas—A Review,’’ Proc. of IEEE,
Vol. 82, No. 12, December 1974.
[13] Jordan, E. C., (ed.), Reference Data for Engineers: Radio, Electronics, Computer, and
Communications, 7th ed., Indianapolis, IN: Howard W. Sams & Co., 1985,
pp. 32-44–32-55.
[14] Geideman, W. A., et al., ‘‘An All GaAs Signal Processing Architecture,’’ Microwave
Journal, Vol. 30, No. 9, September 1987, p. 132.
CHAPTER 9
LPI Radar and the Future of ELINT
9.1 What Is LPI Radar?
The meaning of ‘‘low probability of intercept’’ is not precise. The general idea is
to make a radar system whose signal is below the level of threshold of detection
of opposing ELINT receivers while still being able to detect targets at useful ranges.
Proponents of LPI radar have said that the purpose is ‘‘to see without being seen.’’
Consider the particular case of a warning receiver carried by the radar’s target.
Then the range from the radar to the intercept receiver is equal to the range from
the radar to the target, and the receiver and the target are in the same part of the
radar’s antenna beam. The goal of the LPI radar is to detect the target at a range
greater than the range at which the receiver can detect the radar’s signal. Can this
be achieved?
This is analogous to the situation where a person with a flashlight (the radar)
is trying to find another person hiding in the dark. In this analogy, both the target
and interceptor use their eyes as their receiver. If both have the same eyesight (or
ability to detect light), the interceptor usually has the advantage and can see the
flashlight beam (and even its sidelobes) from great distance—much greater than
the distance at which the person using the flashlight can see the light reflected from
the target. In this situation, the person with the flashlight can achieve LPI operation
if the interceptor (or target) has very poor eyesight—equivalent to wearing dark
glasses. This is analogous to the approach taken to designing LPI radar—the radar
receiver can be nearly matched to the transmitted signal but the intercept receiver
generally is not. Many details concerning LPI radar design as well as approaches
to intercepting such signals are summarized in [1]. Earlier work on LPI radar and
detecting its signals is contained in [2].
The signal strength available to the radar receiver varies as the negative fourth
power of the range, and that available to the ELINT receiver varies as the negative
second power of the range. Therefore, there is always a range short enough to
cause the power available to the radar receiver to exceed the power available to
the ELINT receiver. This is illustrated by the example in Figure 9.1, which shows
the power available from a 1W transmission to the radar receiver (as reflected
from a 1 m2 target) and to an ELINT receiver in the radar’s main beam. As can
be seen, the range at which the two received powers are equal is at approximately
200m. The power received at a range of 10 km is about −89 dBm for the ELINT
receiver and about −123 dBm for the radar receiver. To hide its main beam from
detection at 10 km requires that the radar receiver respond to a signal level of
211
212 LPI Radar and the Future of ELINT
Figure 9.1 Comparison of received signal strength at radar received (solid line) and ELINT receiver
(dotted line) in the radar’s main beam. Transmit power of 1W; radar antenna gain of
30 dBi on both transmit and receive; target cross-section of 1 m2; wavelength of
10 cm; ELINT receiver antenna gain of 0 dBi.
−123 dBm but also that the ELINT receiver fails to respond to a signal level of
−89 dBm—in other words, a difference in sensitivity of 34 dB is required.
9.2 Radar and ELINT Detection of Signals
Radar detection range depends on the total energy returned from the target during
the observation time. If the power returning from the target is constant over the
observation time, the energy is the product of the average power and the observation
time. If it is not constant, then the energy is the product of the average power
times the observation time. For pulsed radar, the average power is the product of
the peak power times the duty factor. The radar duty factor is the ratio of the pulse
duration to the pulse repetition interval (PRI). However, unless pulse compression is
used, radar range resolution is the pulse duration times the speed of light divided
by 2. This means that narrow pulses provide fine range resolution whether you
want it or not.
In addition, narrow pulses reduce the duty factor when the PRI is determined
according to the unambiguous range and velocity requirements. The use of short
duration pulses makes it difficult to get sufficient energy back from the target.
Pulse compression is used to increase the average power while retaining the maxi-
mum range capability of long pulses and the range resolution of short pulses. As
illustrated in Figure 9.2, a CW signal can have much lower peak power than a
pulsed signal but have the same energy returned from the target.
Most of today’s intercept receivers are designed to detect single pulses and then
to form pulse trains and further analyze these pulses, whereas radar receivers are
9.4 One Interception Strategy: Noncoherent Integration 213
Figure 9.2 Illustration of pulsed and CW radar signals with the same average power.
designed to make use of all of the energy returned from the target prior to making
a decision.
9.3 Matched Filter Theory

The maximum output SNR is obtained from a matched filter. The matched filter,
when the noise is stationary white and Gaussian, is a filter whose impulse response
is the same shape as the signal but reversed in time. For such a filter the maximum
output SNR is equal to the input signal energy–to–noise spectral density ratio. This
is the best performance the radar can achieve to perform its function. Of course,
a truly matched filter may not be achieved in practice, and usually an ‘‘integration
loss’’ is incorporated into the radar performance calculation. For a single pulse
signal with a rectangular pulse shape processed by a receiver with a rectangular
bandpass with the optimum bandwidth, there is a loss of about 0.85 dB relative
to a matched filter [3]. The input SNR is
SNR in = S /(kTo NFB) (9.1)
where S is the input signal power to the matched filter, k is Boltzmann’s constant,
To is 290° Kelvin, NF is the receiver noise figure, and B is the receiver noise
bandwidth (assumed wide enough to pass the signal spectrum).
The output SNR from a matched filter is
SNR mf = S␶ /(kTo NF) (9.2)
where ␶ is the integration time (equal to the duration of the signal).

The integration gain is the ratio of the output SNR to the input SNR, or
Gain mf = B␶ (9.3)
9.4 One Interception Strategy: Noncoherent Integration

For ELINT systems, the use of a matched filter is almost never practical. First, the
waveform of interest is not known exactly, and second, the ELINT receiver usually
attempts to preserve the signal’s time, amplitude, and frequency characteristics so

that it can be identified. In this case, the receiver of choice may consist of an
envelope detector followed by a noncoherent integrator. (For example, a low pass
filter.) This scheme is also often used in radar, for example, when a train of pulses
is transmitted with random phase and returned from the target. Each pulse is
processed by a matched filter, and then the individual pulse amplitudes are added
without regard to phase in a noncoherent fashion. In the same way, multiple
samples of the envelope of a radar signal at an ELINT system can be added
noncoherently to enhance sensitivity. For example, suppose the signal-of-interest
(SOI) is known to be present in some frequency band for some duration of time.
The envelope of the signal plus noise in that band can be sampled, and these
samples can be added noncoherently for an appropriate period of time and then
compared to a threshold. The performance of such noncoherent integration is
mathematically difficult to determine exactly; however, convenient approximations
are available. For a large number of samples averaged, the distribution of the
average approaches a normal distribution (consider the law of large numbers). In
this case, the integration loss relative to coherent integration approaches the square
root of the number of samples averaged. In the coherent (or matched filter case),
the integration ‘‘gain’’ is proportional to the number (N) of independent samples
integrated, or, in dB, 10 log (N). This gain is reduced in the case of noncoherent
integration by approximately 5 log (N) − 5.5 dB [2]; that is, for values of probability
of detection and false alarm generally used in practical systems, noncoherent inte-
gration gain is approximately 5 log (N) + 5.5 dB for large N. For example, for
N = 100, the coherent or matched filter gives an output SNR, which is 10 log (100)
= 20 dB greater than the input SNR, whereas the noncoherent integrator gives an
output SNR that is approximately 5 log (100) + 5.5 = 15.5 dB greater than the
input SNR [4]. In this example, the loss relative to a matched filter is approximately
4.5 dB.
Gain nc ≅ 3.55 √(B␶ ) or 5.5 + 5 log (B␶ ) dB (9.4)
In the absence of knowledge of waveform specifics, one strategy is for the

ELINT equipment to integrate noncoherently for as long as the radar integrates.
In this case, the loss relative to coherent radar is approximately
Loss ESM ≅ 5 log (B␶ ) − 5.5 dB (9.5)
9.5 ESM and Radar Range Compared
From the one-way and two-way free space range equations (see Chapter 2), ranges
at which the minimum signal strength is available at the ELINT receiver and at
the radar receiver can be compared:
R ELINT
R Radar 冋S G G
= R Radar Radar TE ELINT
S ELINT G T GR ␴
1 0.5
册 (9.6)
9.5 ESM and Radar Range Compared 215
where:
R ELINT = signal detection range for the ESM receiver

R Radar = target detection range of radar
S ELINT = sensitivity of ELINT receiver
S Radar = sensitivity of radar receiver
␴ = radar target cross-section
G TE = gain of radar transmit antenna in direction of ELINT receiver
G T = gain of radar transmit antenna in direction of target
GR = gain of radar receive antenna in direction of target
G ELINT = gain of ELINT antenna in direction of radar
If the radar receiver uses matched filtering and if the ELINT receiver uses
noncoherent integration over the same time interval, and if the receivers have the
same noise figure, the ratio of ELINT receiver sensitivity to radar receiver sensitivity
is given approximately by (9.5) for large values of B␶ . (Because radars do not
achieve ideal matched filtering in practice, it is typical to reduce the performance
of the radar from that predicted by matched filter theory by some amount; here,
however, ideal matched filter performance is assumed.)
If the terms on the right of (9.6) are known, then the ratio of radar range to
ELINT range can be calculated. Alternately, the minimum value of the ELINT
antenna gain required could be calculated for given radar and ELINT ranges. If
the main beam-to-sidelobe ratio of the radar antenna is known, as well as the
radar antenna gain on receive, and also if the target’s radar cross section is known,
then the minimum antenna gain required for the ELINT equipment can be deter-
mined.
冉冊冉冊 √B␶
2 2
GT R ELINT ␴ S ELINT GT R ELINT ␴
G ELINT = GR 4
≅ GR 4
(9.7)
G TE R Radar S Radar G TE R Radar 3.55
(Note that G TE /G T = SLR or ratio of sidelobe to main beam gain of the radar’s
transmit antenna.)
This framework can be used to investigate the feasibility of ELINT systems
using noncoherent integration of signal energy to detect LPI radar signals—without
knowledge of the details of the signal modulation. If the signal modulation is
known, that added knowledge could improve the sensitivity of the ELINT receiver.
Of course, there is a need to do more than just detect the presence of the radar
signal. Information to determine the type of threat is generally required, and for
this, some general waveform information is usually needed. Another key aspect is
to obtain information about where the emitter is located using angle of arrival or
time difference or frequency difference techniques.
If the transmit power of the radar is reduced, the radar target detection range
decreases, but the ELINT detection range decreases even more. For example, if the
transmit power is reduced by a factor of 16, the radar target detection range is
reduced to one-half or its original value (the fourth root of 16), whereas the ELINT
detection range is reduced to one-fourth of its original value (the square root of
16).
If the transmit power is reduced sufficiently, a point is reached where the radar
range and the ELINT range are equal. If the ELINT intercept range is to be equal
to the radar’s target detection range, this condition requires setting (9.7) equal to 1
and solving for the range. The result is given in (9.8), assuming that the ELINT
receiving station has an isotropic antenna (GE = 1). This has been termed the
‘‘quiet’’ radar condition because its signal is detectable only at ranges shorter than
the range at which it detects its target.
RR = RE = 冋 S Radar 4␲ 册
S ELINT ␴ (SLR) GR 0.5
(9.8)
In this equation, R R is the radar range in meters, R E is the ELINT range in

meters, ␴ is the radar target cross-section in square meters, SLR is the ratio of
radar antenna main beam gain to sidelobe gain (assuming the ELINT observer is
in the radar sidelobes), and GR is the radar antenna gain on receive. The ratio of
the ELINT receiver sensitivity to the radar receiver sensitivity is generally larger
than 1 because the radar receiver is nearly matched to the transmitted signal and
the ELINT receiver is not matched. The ELINT receiver is not matched partly
because it must process a variety of signals and partly because it tries to preserve
the characteristics of the transmitted signal. Preserving such characteristics as rise
time, pulse width, and intra pulse modulation requires a larger bandwidth compared
to the bandwidth of a matched receiver.
What values can be expected for the ratio of the sensitivity of the ELINT
receiver to that of the radar receiver? For an energy detection wideband ELINT
receiver, this is approximately the value given by (9.5) when the ELINT receiver’s
predetection bandwidth is matched to the radar’s signal modulation bandwidth
and when the ELINT receiver’s noncoherent integration time is matched to the
radar’s integration time.
Assuming that the SNR required is the same at both the radar receiver output
and the ELINT receiver output, the conclusion is that the approximate value of
the receiver sensitivity ratio is the square root of the radar’s time-bandwidth product
divided by 3.55. The time bandwidth product is limited by target motion to c/2V
in order for the target to remain in one range cell for one coherent integration
period. For target velocity of 300 m/s, this value is about 500,000. Therefore, the
ratio of ELINT receiver sensitivity to radar receiver sensitivity is limited to about
23 dB for energy detection receivers adapted to the radar’s waveform.
Consider a ‘‘quiet’’ radar capable of detecting a 10-m2 target at 10 km.
According to Figure 9.3 and (9.8), it must find a way to make the intercept receiver
about 50 dB less sensitive than the radar receiver when the ELINT system’s antenna
gain is 1. Because of the limitation above of 26-dB sensitivity loss for a noncoherent
receiver using the same integration time as a matched filter radar receiver, energy
detection receivers should be sufficient to detect the main beams of radars designed
9.6 Some Pulse Compression Modulation Constraints 217
Figure 9.3 Quiet radar range as a function of receiver sensitivity ratio (␦ ). (ELINT receiver in the
main beam of the radar uses an omni-directional antenna.)
to detect small targets at ranges in excess of 10 km. Designing a radar whose main
beam is barely detectable at the same range at which the radar can detect its target
is very difficult at militarily useful ranges and for the cross-sections typical of
military targets. Of course, existing ELINT receivers may not be designed to cope
with a signal from a ‘‘quiet’’ radar. In these circumstances, quiet radar operation
may be achieved at greater ranges.
9.6 Some Pulse Compression Modulation Constraints
The use of energy detection receivers is necessary if the radar waveform is noise-
like to the interceptor. While such waveforms can be used, their time sidelobes
and Doppler properties may be far from ideal in any specific time interval; it is
only over a long-term average that their ambiguity function approaches the ideal
‘‘thumbtack’’ shape. Generally, at present, radar designers prefer to control their
ambiguity function by using specially designed, deterministic pulse compression
waveforms such as linear FM (LFM) or binary phase shift keyed (BPSK). Polyphase
shift keyed and nonlinear FM waveforms are also described in the literature but
have not been deployed extensively. For example, the ‘‘best’’ BPSK waveforms are
the Barker codes. These have time sidelobes at zero Doppler of maximum amplitude
of 1. However, the longest known binary Barker code is only 13 bits long. It has
been shown that odd length Barker codes with length greater than 13 do not
exist; however, there are longer sequences of polyphase Barker codes. When pulse
compression modulation other than random (or pseudo-random) noise is used, it
is possible to process the intercepted signal in ways that outperform energy detec-
tion. This kind of processing permits the interception of known deterministic
waveforms at greater range than wideband energy detection. Guosui et al. describe
development of random signal radars [5]. Figure 9.4 shows the current concept of
Figure 9.4 Challenging ELINT signal: frequency agile CW.
an LPI signal in terms of its frequency as a function of time. The individual ‘‘blocks’’
represent radar coherent processing intervals.
9.7 Interception Techniques Using the Envelope of the Received

Signal
The use of noncoherent integration techniques in the interception process allows

the receiver to operate without specific knowledge of the details of the transmitted
waveform. For example, the particular pulse compression modulation in use may
be unknown, but the approximate range resolution (and hence the signal’s band-
width) may be inferred from the function the radar performs. Some examples make
clearer the processing required. An early example of using noncoherent integration
to detect LPI signals is the use of rapid scanning of the RF band of interest using
a swept narrowband receiver. Each time the receiver sweeps past the LPI signal, a
sample of the envelope of the signal plus noise is obtained. Various types of
noncoherent integration can be applied to these samples. Similarly, a channelized
receiver can also be used to sample the energy in each of its frequency channels,
and these samples can be combined noncoherently to achieve improved detection
of the LPI signal. These two concepts are very similar. In the case of the swept
receiver, the envelope samples are obtained sequentially at the different frequencies
across the band covered, whereas in the case of the channelized receiver the envelope
samples are obtained simultaneously from each channel. Hence the sample rate
(for a given frequency channel) for the swept receiver is limited to the reciprocal
of the sweep time but it is at the Nyquist rate for a channelized receiver. For a
given signal duration, the channelized receiver provides more samples than a swept
receiver.
One Rapid Sweep Superhet Receiver covered a 2-GHz band and swept that
band in 256 steps of 8 MHz in 20 ␮ s. Its sampling rate was 50,000 samples per
channel per second. A similar channelized receiver with 256 channels each 8 MHz
wide could provide 16 million samples per channel per second. The bandwidth of
8 MHz was selected to be wide enough to pass the pulse compression modulations
then in use. (Many current threat radar systems use bandwidths of less than
10 MHz.) A common design is to test the sample against a threshold and declare
the presence of a pulse if one sample exceeds the threshold. To provide 90%
probability of detection and a 10−6 probability of false alarm requires a SNR of
13.2 dB. The computation of the average of the envelope over a number of samples
and then comparing the average to a threshold allows the use of a much lower
threshold to achieve the same probability of detection and false alarm. The approxi-
mate improvement in sensitivity is found from (9.4). If the waveform dwells at one
carrier frequency for 2.5 ms and the bandwidth of the channels is 8 MHz, then
BT = 20,000 and the sensitivity improvement is approximately 27 dB for the
channelized receiver. The swept receiver provides a sample every 20 ␮ s or 125
samples in 2.5 ms. Equation (9.4) gives the improvement in sensitivity as approxi-
mately 15.5 dB for the swept receiver. There is an added benefit of the swept
receiver in eliminating interference from the pulses from ordinary radar signals.
Sampling the signal for 80 ns every 20 ␮ s means that most pulsed signals will not
contribute much to the average over 2.5 ms. The probability of coincidence between
the sampling gate of the sweeping receiver and the pulsing of the interfering radar
is very low. Of course, this could also be seen as a drawback if one is interested
in detecting both pulsed signals and modulated CW signals.
Another way of processing the samples of the envelope is to test each sample
against a threshold and then require that a certain minimum number of samples
cross the threshold out of a given number of samples tested—often called M of N
detection or binary integration [2, 6]. Here M samples out of N must cross the
threshold (N ≥ M). This process is not quite as effective as computing the average
of the samples. It has the effect of suppressing very strong signals of short duration—
which is sometimes an advantage. The computation requires use of the binary
probability distribution. The first step is to determine the probability of noise alone
crossing the threshold at least M times out of N for a given probability that a single
sample of noise crosses the threshold. The notation Pfa, 1 denotes the probability that
one sample of noise alone crosses the threshold, and Pfa, N is the probability that
at least M of N samples of noise alone cross the threshold. These are related by
the binary probability distribution for N trials and for Pfa, 11 the probability of
‘‘success’’ on one trial. Likewise, the probability that signal plus noise crossed the
threshold on one trial is Pd, 1 and the probability that signal plus noise crosses
the threshold at least M times out of N tries is Pd, N . These are also related by the
binary probability distribution. Some examples of the results are given in Table
9.1. There is an optimum value of M for any specific case. Values of M near N/2
or 1.5N 0.5 have been suggested in the literature [6, 7]. The latter is used in Table
9.1. The required SNR has a broad minimum [6], and so choosing the exact
optimum value of M is not critical.
Although there is no theoretical limit to the improvement as the value of N
increases, there are important practical considerations; namely, as the threshold
Table 9.1 Examples of Integration Gain Using M of N Integration

N = 8, M = 4 N = 16, M = 6 N = 32, M = 8 N = 64, M = 12
SNR = 7.3 dB SNR = 5.4 dB SNR = 3.4 dB SNR = 1.7 dB
Gain = 5.9 dB Gain = 7.8 dB Gain = 9.8 dB Gain = 11.5 dB
M is selected as 1.5N 0.5; gain compared to Pd,1 = 0.9, Pfa,1 = 10−6 (single pulse SNR =
13.2 dB). N determined by signal duration during its coherent processing interval for Pfa,N
= 10−6, Pd,N = 0.9.
decreases, the value of Pfa, 1 becomes larger and so does the value of Pd, 1 . Eventually
the threshold is so low that there is not much difference between these two values.
Then a slight change in the noise level could drastically affect the final values of
Pd, N and Pfa, N . This is illustrated in Figure 9.5, which shows the probability of
both false alarm and detection at −1.5-dB SNR. If the threshold is selected at 2.65
normalized units, then Pd, 1 = 0.12 and Pfa, 1 = 0.03. With a relatively small difference
between probability of detection and false alarm, a small change in the noise level
could cause a drastic change in the performance of the system. After the M of N
process (for M = 24, N = 256), the probability of detection and false alarm at SNR
= −1.5 dB is shown in Figure 9.6. Now the probability of detection at the threshold
of 2.65 is 90% and the probability of false alarm is 10−6. Detection using a single
sample would require the SNR to be 13.2 dB to give this same performance;
therefore, the M = 24, N = 256 process provides 13.2 − (−1.5) = 14.7 dB of
processing gain. Coherent processing of 256 samples provides 24.1 dB of processing
gain; hence the loss of the M of N process is 9.6 dB relative to coherent integration.
The approximate gain expected from a noncoherent process as given by (9.2) is
Figure 9.5 Probability of detection (solid) and false alarm (dotted). SNR = −1.5 dB. Threshold set
at 2.65 yields Pfa, 1 = 0.03 and Pd, 1 = 0.12.
Figure 9.6 Probability of detection (solid) and false alarm (dotted) after M of N processing; M = 24,
N = 256. Threshold of 2.65 yields Pfa, 256 = 10−6 and Pd, 256 = 0.90 at SNR = −1.5 dB.
17.5 dB. Hence the loss of the M = 24, N = 256 process relative to ideal noncoherent
integration is about 2.8 dB.
The case of M = N is of interest for detecting long duration (or CW) signals.
This means that the signal plus noise must be above the threshold for N consecutive
samples. As shown in [2], the SNR required to provide a specified level of perfor-
mance is
冋册
1/N
1 − Pd
SNR = −ln 1/N
1 − Pfa
For a large number of samples, this value approaches
−ln 冋
ln (Pd )
ln (Pfa ) 册
as N approaches infinity.
For 0.9 probability of detection and 10−6 probability of false alarm, the required
SNR approaches 6.9 dB as N approaches infinity. This illustrates how requiring
consecutive samples to cross the threshold limits the processing gain. For this
example, 20 consecutive samples provide a gain of about 6 dB—only 0.9 dB less
than if the number of consecutive samples approaches infinity.
9.8 Narrowband Channels and Frequency Modulated Signals
Although linear FM is a common modulation used in communications and radar

systems, any sort of FM can be thought of as linear over short time intervals. For
FM signals, the question is what is the best intercept receiver bandwidth to use to
intercept the signal and warn a user of its presence? Described below is the funda-
mental principal for designing a receiver for linear FM; and for nonlinear FM, the
same approach can be used over portions of the signal when the frequency versus
time variation is approximately linear. A technique for pulse compression of linear
FM signals is described in [8]. The technique can accommodate any pulse compres-
sion ratio less than the square of the number of bandpass filters in a filter bank
(generally implemented using an FFT). In radar applications a programmable pulse
compressor allows the transmission of pulses with a variety of linear FM pulse
compression characteristics. Of course, in radar, the frequency deviation and
pulse duration are known and the receiver is programmed properly to receive the
waveform transmitted. In intercepting unknown signals of the linear FM type, this
information is not available to program the signal processor. Furthermore, the
purpose of the intercept receiver is not pulse compression per se, but rather to
detect the presence of the signal and measure its frequency deviation and time
duration. An approach similar to the method of [8] can be applied to the design
of an intercept receiver, although here the channel outputs are converted to ampli-
tudes and no use is made of the phase information. The basic design of a digital
channelized receiver is shown in Figure 9.7. Whether the channels are implemented
in analog form or digital form is a matter of convenience.
For a signal whose frequency is linearly changing with time, the output of a
bandpass filter whose bandwidth is considerably less than the frequency excursion
is approximately a pulse whose duration is the channel bandwidth divided by the
rate of change of the RF at its center frequency. For example, if the channel
bandwidth is 1 MHz and the RF is changing linearly at the rate of 0.01 MHz/␮ s
over a band of 100 MHz, the filter output will be a pulse of duration 1/0.01 =
100 ␮ s, which occurs at the approximate time the RF signal passes through that
channel’s bandpass. Compare this to a wideband receiver covering the entire
100-MHz band in one channel. The noise level in one channel is increased from
1 to 100 MHz (20 dB), but the signal lasts 10,000 ␮ s, not 100 ␮ s. For a wideband
receiver, the best sensitivity occurs when the bandpass filter is followed by an
Figure 9.7 Channelized receiver configuration.

envelope detector and a video (lowpass) filter whose bandwidth is comparable to

the reciprocal of the pulse length. The effective noise bandwidth of such a wideband
receiver is approximately
ENB = √2B V B RF
where:
ENB = effective noise bandwidth

B V = video (lowpass) bandwidth after the envelope detector
B RF = RF bandwidth before the envelope detector
Other things being equal, the sensitivity of two receivers is given by the ratio
of their effective noise bandwidths. The ratio of noise bandwidths is
ENB1/ENB2 = √B V1 B RF1 /(B V2 B RF2 )
For our example, B RF1 = 100 MHz, B V1 ∼ 1/1,000 ␮ s. Also B RF2 = 1 MHz,
B V2 ∼ 1/100 ␮ s. We see that the ENB values for these two receivers are the same,
and their ratio is 1. In general, the ENB for a receiver designed in this way for
linear FM is given as
ENBlfm = √⌬ f /⌬t
Here, ⌬ f is the frequency excursion of the linear FM and ⌬t is the time over
which the excursion occurs. Their ratio, R, is the rate of change of the frequency
(slope) of the linear FM. Notice that if the channel bandwidth is the square root
of the LFM slope, then the number of channels is
Number of channels = ⌬ f /√⌬ f /⌬t = √⌬ f ⌬t
Put another way, this says that the square of the number of channels is equal
to the time-bandwidth product—which is precisely the basis of [8].
How narrow can the channel RF bandwidth be? If we reduce the RF bandwidth
so that the signal is in the channel for a time less than the reciprocal of the channel
bandwidth, we begin to lose signal amplitude. R is the rate of change of the RF
and B is the channel bandwidth; we want B /R to be greater than 1/B or
B> √R
For R = 0.01 MHz/␮ s, B must be wider than 100 kHz. Consider the following
example. A 100-MHz linear FM frequency excursion occurs over 10 ms (BT =
106 ). The receiver may have the same effective noise bandwidth for any choice of
receiver RF bandwidth from 100 MHz down to 0.1 MHz; provided that the video
filter bandwidth is approximately R/B—the reciprocal of the ‘‘pulse duration’’

present at the channel outputs. For 0.1-MHz channel bandwidth, there are such
pulses at the outputs of 1,000 channels to test for the presence of signals. To keep
the total false alarm rate the same as in a single wideband channel system, the
threshold of each narrowband channel would necessarily be set a bit higher than
that for a single channel.
A receiver designed for linear FM has an effective noise bandwidth of the
square root of the FM slope over a wide range of choices for RF bandwidth,
provided appropriate video filtering is used. The sensitivity of a single channel of
this receiver depends on the FM slopes expected, not on the total frequency deviation
or the time duration of the FM waveform. Figure 9.8 illustrates how this principle
may be applied to a channelized receiver design.
There are several advantages of using channel bandwidths smaller than the
frequency excursion of the linear FM. First, the character of the FM (namely the
slope) can be determined by examining the sequences of pulses at the outputs of
the narrowband channels. Second, by considering the outputs of many channels
before making a decision as to whether a signal is present or not, sensitivity can
be improved compared to the sensitivity of a single channel. In principle, this is
the same concept as in pulse compression; however, for interception purposes, one
may ignore phase and other waveform details but achieve less than optimum
sensitivity. For example, one way to improve sensitivity is to declare a signal is
present if some number of channels exceeds a threshold in a certain length of time,
regardless of their order. Suppose we had a system with 256 channels and required
that at least 60 of them exceed a threshold in any 10-ms window of time. The
threshold on the individual channels can be lowered and the result is a sensitivity
improvement of about 18 dB for 256 narrowband channels followed by M out of
N detection logic. (In this example, M = 60 and N = 256.)
Logic can be used to combine the signals from all of the channels in other ways.
For example, one could add their power spectra (ignoring their phase differences).
Another way is to use frequency versus time templates. This is similar to the Hough
(or Radon) transform, which sums the channel outputs along straight lines of all
possible slopes and starting points. The slope is a key parameter of the signal.
Accurate determination of the slope can help sort and identify several linear FM
signals present in the same band at the same time; in other words, postdetection
processing of the outputs of a bank of narrowband channels can enhance sensitivity
Figure 9.8 Processing after the channelizer.

and characterize the slope of linear FM signals. In a modern design, the channels
could be realized by predetection digitizing followed by an FFT and magnitude
computation, and then possibly followed by averaging the FFT amplitudes. Com-
parison to a threshold can also be used, and the threshold crossings counted in an
M of N (or similar) detection process. Slope measurement is accomplished by
examining the sequence of the narrowband channel threshold crossings in a manner
similar to the Hough transform, or similar to the method of [8].
If the signal uses nonlinear FM, the instantaneous frequency could be approxi-
mated using linear segments. The largest slope determines the minimum channel
bandwidth. The same techniques can then be applied for signal detection. To find
the character of the FM, various templates of appropriate shape could be tried, as
is done using the Hough transform to look for arbitrary shapes.
Many FMCW systems use multiple FM slopes (including zero slope) to make
range measurements with differing resolution and ambiguity and to deal with
Doppler shifts. The waveforms of interest may also switch slopes in different modes.
To deal with these signals with a receiver having a single value of channel band-
width, the channel bandwidth is selected as the minimum to accommodate the
steepest slope. Then there would need to be different video bandwidths or postdetec-
tion processes (e.g., differing values of M and N) to accommodate the different
pulse durations caused by the different slopes. In the situation where the steepest
slope requires a channel bandwidth so wide that the character of the smallest slope
cannot be determined, it would be necessary to use two or more channel bandwidths
(i.e., multiple FFT lengths).
The use of channel bandwidth equal to the square root of the slope is optimum
in the sense that it is the narrowest bandwidth for which the output of the channel
is not reduced by the rapid sweeping of the signal through the channel. For linear
FM, once the channel bandwidth becomes less than the square root of the FM
slope, there is a reduction in the signal amplitude (as well as the noise level).
Because the time spent sweeping through one channel is less than the impulse
response of that channel, the signal is not present for the entire time that the filter
requires to build up to its full output amplitude. The signal energy input to the
filter is the signal power multiplied by the time the signal is present in the band
of that filter, which is (channel bandwidth/slope). The noise energy at the input is
the noise power kTo (NF)(channel bandwidth) multiplied by the time the noise is
present, which is 1/(channel bandwidth). Therefore, the signal-to-noise energy ratio
is S(channel bandwidth/slope)/[kTo (NF)]. This is the maximum SNR present at
the channel output when the channel bandwidth is less than the square root of the
slope. When the channel bandwidth is greater than the square root of the slope,
the output SNR becomes equal to the input SNR or S/kTo (NF)(channel bandwidth).
When the channel bandwidth is less than the square root of the slope, the
signal frequency sweeps through several channels during the impulse response time
of the channel filters, and there will be several adjacent channels excited during
that time. For example, if the slope is 1 MHz/␮ s, in 32 ␮ s the signal will sweep
though 32 MHz. The square root of the slope is 1 MHz. Suppose the channel
bandwidth is set at 31 kHz instead of 1 MHz. This reduces the energy into each
filter by 1/32 or about 15 dB. But notice that 1,000 of the 31-kHz channels will
be excited in 1 ␮ s instead of 32 1-MHz channels. If it were possible to use M of
N or other detection logic on the 1,000 31-kHz frequency channels excited by the
sweeping signal, the overall detection performance would be improved by about
18 dB, so the net gain is about 3 dB relative to 1-MHz channels. (But note that
the slope is obscured.) During the next 32 ␮ s, a different set of 1,000 channels
(centered 32 MHz higher in frequency) will be excited, and so on, until the end
of the sweep occurs. If the sweep lasts longer than 32 ␮ s, there will be additional
channel outputs available to combine. Suppose the sweep lasts 256 ␮ s. Then there
will be 8 sample periods of 32 ␮ s available during each LFM sweep, and these
could be combined (for example, by using 4 of 8 detection logic) to gain another
6 dB—but the proper way to combine these is not known unless the slope is known.
Carrying this to the limit, suppose the channel bandwidth is so narrow that
the impulse response time of the channels is as long as the sweep time; for example,
if the sweep time is 256 ␮ s, the channel bandwidth could be reduced to about
4 KHz. Now there is only one sample available from each of the 64,000 4-kHz
channels during one sweep, and the detection logic must combine the outputs from
64,000 frequency channels. Now there is no information about the slope contained
in the timing of the filter outputs, and knowledge of the slope is not needed to
combine the channel outputs for detection of the presence of the signal. Giving up
the knowledge of the slope and using very narrowband channels provides about
the same detection performance. These three cases are summarized in Table 9.2.
9.9 Predetection Processing Methods to Detect Linear FM and Other

LPI Signals
Many of the LPI radars advertised to date make use of linear FM as the pulse
compression waveform—not the random signals mentioned above. Linear FM is
also commonly used in altimeters. One method of ranging is to mix the received
signal with the transmitted signal. Then, for a stationary target, the beat frequency
is equal to the product of the FM slope and the time delay due to the range. For
this type of signal, the matched filter in the radar becomes a mixer (to derive the
beat frequency) followed by a filter bank to determine the value of the beat frequency
and hence the range. The bandwidth of the filters can be as narrow as the reciprocal
of the sweep time of the FM. For moving targets, the effects of the Doppler shift
must be taken into account [9]. For example, the Wigner-Hough Transform (WHT)
or Radon-Wigner Transform (RWT) and the Radon Ambiguity Transform (RAT)
Table 9.2 Comparison of Three Bandwidth Channels

Detection Performance for a Linear FM Segment
Integration Straddle
Channel BW Noise Signal M/N Gain Loss Net Gain
1.0 MHz Ref Ref* 60/256 15 dB 1.5 dB 0
0.031 MHz −15 dB −15 dB 1,024 & 4/8 21 dB 2.1 dB 3.9 dB**
0.004 MHz −24 dB −24 dB 64,000 24 dB 1.5 dB 7.5 dB**
*The 1-MHz channels and the 0.031-MHz channels require knowledge of the correct slope to achieve the M of N
processing gain.
**Requires raising the threshold to achieve the same false alarm rate. This reduces the net gain by ∼2 dB.
9.9 Predetection Processing Methods to Detect Linear FM and Other LPI Signals 227
can achieve nearly coherent detection performance of linear FM waveforms

[10, 11].
Pace [1, Ch. 9] explores the application of the Wigner-Ville Distribution (WVD)
to detect the presence of linear FM signals and other signals in noise. The WVD
has the greatest concentration of energy of any two-dimensional time-frequency
distribution for linear FM. The WVD of a signal x(t) is given by
Wx (t, f ) = 冕 x (t + ␶ /2) x* (t − ␶ /2) e −j2␲ f␶ d␶ (9.9)
The ambiguity function is given by
A x (␯ , ␶ ) = 冕 x (t + ␶ /2) x* (t − ␶ /2) e j2␲␯ t dt (9.10)
or
A x (␯ , ␶ ) = 冕冕 Wx (t, f ) e j2␲ (␯ t + ␶ f ) dt df (9.11)
Equation (9.11) shows that the ambiguity function in (9.10) is the two-dimen-
sional Fourier transform of the WVD in (9.9). The radon transform of the WVD
is found by computing the integral of the time-frequency distribution along straight
lines at different angles. Hence, for linear FM, when one uses the correct angle (or
FM sweep rate) and starting time, a large value will be obtained. At other angles
and/or starting times the contribution to the integral will be largely due to noise.
Figure 9.9 shows a contour plot of the WVD of a linear FM signal 1 ␮ s long with
an FM deviation of 20 MHz, starting at 30 MHz and extending to 50 MHz. Figure
9.10 shows the radon (or Hough) transform of the WVD in Figure 9.8. The peak
in Figure 9.10 occurs at the slope and starting frequency of the linear FM. Figure
9.11 shows the same thing as Figure 9.10 but with a 0-dB SNR. It shows the
enhanced detectability of the signal using the WVD and radon transforms. Detection
would be achieved by establishing a threshold value for the amplitude of the radon-
WVD process. Note that the original data window must include the starting time
and ending times of the signal and the bandwidth must include the entire FM
sweep. The minimum number of input data points required is determined by the
time-bandwidth product of the signals of interest. The radon transform of the
WVD results in a two-dimensional space of FM slope versus starting frequency
(or starting time).
Consider the ambiguity function (AF) instead of the WVD. Because the AF is
a correlation function, its peak is always at zero delay and zero frequency shift.
For linear FM, there is a ridge in the AF extending from the origin at an angle
corresponding to the slope of the linear FM. If the radon transform of the AF is
computed, the result is a one-dimensional plot of amplitude versus slope. This
replaces the two-dimensional plot of amplitude versus slope and starting frequency
(or time) obtained from the radon transform of the WVD.
Figure 9.9 Contour plot of a Wigner-Ville Transform of linear FM signal (1-␮ s duration; 20-MHz
linear FM deviation; BT = 20).
Figure 9.10 Radon (or Hough) transform of WVD in Figure 9.9. Linear FM sweep starts at 30 MHz
and extends to 50 MHz.
Figure 9.12 shows a contour plot of the AF of an FMCW signal with up and
down slopes equal to 1 MHz/␮ s. The original signal consisted of two up 32-MHz
and down 32-MHz cycles of linear FM lasting a total of 128 ␮ s. Figure 9.13 shows
the RAT, which is the radon transform of the AF shown in Figure 9.12. The peaks
due to the linear FM up-slope and down-slope are clearly visible at an input SNR
of −10 dB. The one-dimensional nature of the plot of the magnitude versus slope
should be compared to the two-dimensional plot of magnitude versus slope and
starting frequency of Figure 9.10. There is more information in the plot of Figure
Figure 9.11 Same as Figure 9.10 but with 0-dB SNR. (Signal peak is clearly visible.)
Figure 9.12 Contour plot of ambiguity function of FMCW signal (up-slope = down-slope = 1 MHz/
␮ s). Horizontal scale: max. time lag = 800 ns. Vertical scale: zero frequency shift =
50; max frequency shift = ±1.6 MHz.
9.10; however, if the task is to detect the presence of linear FM and find its slope,
then the RAT of Figure 9.13 is sufficient and requires less processing.
The performance of the RAT relative to a matched filter is shown in Figure
9.14 [11]. If the input signal-to-noise energy ratio is above 0 dB, the loss approaches
3 dB. Achieving signal energy greater than the noise spectral density determines
the length of time the interceptor must integrate. If this time is less that the duration
of the LFM sweep, then Figure 9.14 can be used to determine the loss relative to
a matched filter when using the RAT. If the time required exceeds the duration of
the FM sweep, then detection may still be possible; however, Figure 9.14 does not
apply directly. Instead, a way must be found to combine the energy of two or more
sweeps. This may cause additional losses, as indicated in Figure 9.15. Here the
integration time was increased sufficiently to allow reliable detection of the presence
Figure 9.13 Radon transform of ambiguity function shown in Figure 9.12. SNR = −10 dB. (Duration
of waveform: 128 ␮ s; BT = 4,096.)
Figure 9.14 Loss of RAT relative to a matched filter (maximum lag assumed to be less than the
LFM sweep time).
of the weak signal—that is so that the output SNR was above about 13 dB. As
can be seen, worst-case performance is about 10 dB below that predicted for a
sweep time longer than the duration of the integration.
Another type of signal processing often suggested for detecting LPI signals is
to compute the cyclostationary spectral density (CSD). This is defined as the Fourier
transform of the cyclic autocorrelation function [1, Eq. (11.4)]:
R ␣x (␶ ) = 冕 x (t + ␶ /2) x* (t − ␶ /2) e −j2␲␣ t dt (9.12)

Figure 9.15 Modeled performance of RAT relative to a matched filter. (Low input SNR requires
multiple cycles of the LFM waveform, resulting in more loss than predicted.)
In (9.12), the variable ␣ is called the cyclic frequency. Now notice that (9.12)
is actually the same as (9.10) if the symbol ␣ is changed to ␯ —meaning that the
CSD is the same as the ambiguity function. The spectral correlation density is the
one-dimensional Fourier transform of the cyclic autocorrelation function:
S x␣ ( f ) = 冕 R ␣x (␶ ) e −j2␲ f␶ d␶ (9.13)
The CSD is a two-dimensional density function of cyclic frequency and Fourier

frequency. For example, if the signal had a triangular frequency versus time varia-
tion, and if the variable ␶ extended over a large number of cycles of the triangular
variation, then there would be peaks when ␣ is a multiple of the repetition rate of
the triangular modulation and f is at the center or carrier frequency of the signal.
The emphasis of the CSD is to find periodicities in the cyclic correlation function
(or AF)—for example, the chip rate of a BPSK waveform or the repetition rate of
a periodic FM waveform. For example, a noise-like binary sequence using a fixed
chip rate would be detectable by this type of processing.
However, consider again the concept of the programmable pulse compressor
[8]. In this method of pulse compression, the channel (or FFT) outputs are combined
using a parabolic phase weighting to provide pulse compression for any pulse
compression ratio equal to or less than the square of the number of channels.
While in radar the pulse compression ratio is known and selected in advance, the
same technique can be used to create simultaneously pulse compression processing
for a variety of time-bandwidth products. In a manner similar to the WHT or
RAT, the pulse compression ratio that provides the highest output would also
provide the means to detect the presence of a LFM signal.
9.10 ELINT Receiver Requirements for Interception of Low Peak

Power Signals
The implications of using the various interception techniques described above for
ELINT are that those wideband signals must be digitized and then processed in
parallel in a variety of ways. For example, there could be one path for high peak
power pulsed signals, another for energy detection, another for LFM detection,
another for BPSK detection, and so on. Note that LPI signals are below the noise
in a wideband receiver. While they can be detected using a variety of processing
algorithms, it will be necessary to set the LSB of the A/D converter far below the
receiver noise to preserve these signals. This implies a reduction in dynamic range
or the need for more resolution (bits) from the A/D converter. A major requirement
for LPI radar detection by a general purpose ELINT receiver is more bits and
sampling rates in excess of 1 GHz. The high sampling rates are required to search
wide bands, such as the radar’s frequency agility band, and/or to provide good
probability of interception in a short time, not so much because the radar’s coherent
(or instantaneous) bandwidth is wide but because the center frequency is unknown
and/or the signal uses frequency agility. Wideband communications signals using
very short pulses may be used to transmit missile guidance data or otherwise be a
part of the ELINT signal environment. These could be very difficult to detect if
random pulse spacing known only to the receiver but not to the interceptor is
superimposed on the data. Wideband radar signals using short pulses are less of
a problem to the ELINT community. In addition to the problem of excessively fine
range resolution as the pulse duration becomes short, there is the need to get the
same amount of energy back from the target as when a longer duration pulse is
used. This means that except for very short range radars, the transmitted pulses
are of higher amplitude and can be more easily detected. LPI radar trends are to
reduce peak power (and increase the duty factor) to avoid detection by current
ELINT receivers designed for single pulse detection (i.e., those which use peak
power for single pulse detection).
Because LPI radar is a qualitative term, quiet radar is suggested here as a
quantifiable alternative. Low sidelobes are important to avoid detection of the
radar signal except in the main beam—especially as a defense against anti-radiation
missiles. Power management and atmospheric attenuation shielding are often con-
sidered as LPI techniques but are of very limited benefit in search radar. Power
management is useful in altimeters and some tracking systems. When used, the
transmit power is reduced to the level needed to maintain track. Atmospheric
shielding is of limited use due to the two-way path length of the radar. In a
homogeneous atmosphere, the radar two-way path has more attenuation than the
9.10 ELINT Receiver Requirements for Interception of Low Peak Power Signals 233
one-way path to the intercept receiver unless the interceptor is located more than
twice the range from the radar’s target. Of course, if high altitude ELINT collectors
are used, the atmospheric effects are largely eliminated as the density of the atmo-
sphere decreases.
The ELINT challenge is frequency agile, noise-like signals of modest bandwidth
transmitted with high duty factor (as shown in Figure 9.4). The instantaneous
bandwidth is not likely to exceed that of conventional radars having the same
function and in every application will probably be less than 500 MHz. Randomness
in the radar’s pulse compression waveform makes it necessary for the ELINT
receiver to use energy detection techniques. Today’s radars typically transmit the
same pulse compression code or waveform in every coherent integration interval
during one mode. Tomorrow’s radars may choose to transmit different waveforms
from one coherent processing interval to the next. In this case, the ratio of the
ELINT receiver’s sensitivity compared to that of the radar’s receiver need not be
greater than the square root of the radar’s time-bandwidth product
Wideband and ultra-wideband radar are loosely defined in terms of the frac-
tional bandwidth used by the radar. F high and F low are the upper and lower 3-dB
points of the radar’s bandwidth. The fractional bandwidth, FB, is given by
F high − F low
FB = 2 (9.14)
F high + F low
Wideband radar has a fractional bandwidth greater than 10%, and ultra-
wideband greater than 25%. Recently, a new definition of UWB radar has been
suggested, as follows [12]:
Ultrawideband radar is a radar set having range resolution (⌬r) much smaller than
the target length, L, along the radiated direction.
Ultra-wideband applications include underground probing (for example, mine

detection), very short range (a few meters) such as proximity sensors and automatic
braking devices, and short range (< 100m) such as intrusion alarms and through-
the-wall detection. Due to limited energy on target, it is not likely that long range
UWB radar will be deployed. The use of ELINT receivers to detect such signals
must be weighed against the cost of doing so relative to the threat posed by these
signals. Due to their current short range, monitoring these signals is also best done
at short range. In this case, the need to sort them from among a myriad of other
signals is reduced and their high peak power may make detection relatively easy
at short range.
References
[1] Pace, P. E., Low Probability of Intercept Radar, Norwood, MA: Artech House, 2004.
Artech House, 1985, Chapter 2.
[3] DiFranco, J. V., and W. L. Rubin, Radar Detection, Englewood Cliffs, NJ: Prentice Hall,
1968, p. 184.
[4] DiFranco, J. V., and W. L. Rubin, Radar Detection, Englewood Cliffs, NJ: Prentice Hall,
1968, p. 389.
[5] Guosui, L. et al., ‘‘Development of Random Signal Radars,’’ IEEE Trans. on Aerospace
and Electronic Systems, Vol. 35, No. 3, July 1999, p. 770.
[6] Tsui, J. B. Y., Digital Techniques for Wideband Receivers, Norwood, MA: Artech House,
1995 (Section 9.12).
[7] Skolnik, M., Radar Handbook, New York: McGraw Hill, 1970.
[8] Thor, R., ‘‘Programmable LFM Signal Processor,’’ Patent Number 4,591,857, May 27,
1987.
[9] Stimson, G. W., Introduction to Airborne Radar, 2nd ed., Raleigh, NC: Scitech Publishing,
1998, Chapter 13.
[10] Ozdemir, A. K., and O. Arikan, ‘‘Fast Computation of the Ambiguity Function and the
Wigner Distribution on Arbitrary Line Segments,’’ IEEE Trans. on Signal Processing,
Vol. 49, No. 2, February 2001, p. 381.
[11] Jennison, B. K., ‘‘Detection of Polyphase Pulse Compression Waveforms Using the Radon-
Ambiguity Transform (RAT),’’ IEEE Trans. on Aerospace and Electronic Systems, January
2003, p. 335.
[12] Immoreev, I. Y., and J. D. Taylor, ‘‘Ultrawideband Radar Special Features and Terminol-
ogy,’’ IEEE A&E Systems Magazine, May 2005.
CHAPTER 10
Antenna Scan Analysis
10.1 Introduction
The radar antenna beam pattern typically covers only a small portion of the angular
region of interest. The antenna must couple the energy between the transmitter
and space. To observe targets over a portion of space larger than covered by the
main beam, the antenna beam is moved to observe different angles at different
times. Often the spatial coverage is thought of as requiring a certain number of
beam positions. For hemispheric coverage, there are approximately N bp indepen-
dent beam positions required:
2␲
N bp = (10.1)
␪ AZ ␪ EL
where:
␪ AZ = 3-dB azimuth beamwidth (rad)

␪ EL = 3-dB elevation beamwidth (rad)
The motion of the beam (angular position versus time) is called scanning. The
scanning pattern is designed to cover the entire volume of interest to the radar
system. The size of this volume can vary depending on the purpose of the radar
at the time. If searching a hemisphere is necessary to acquire a target, a smaller
search volume can subsequently be used to refine its location. If the radar can
track the target, scanning may cease altogether. Such constant illumination of the
target by the radar can be a warning to the target that tracking is occurring and
an attack may soon occur. In the case of an electronically steered array (ESA),
tracking may be indicated by repeated bursts occurring at the same beam position
(or having the same amplitude when observed by an ELINT receiver).
In mechanically scanned radar systems, the scanning pattern is typically peri-
odic. In electronically scanned systems, the search pattern may be periodic; however,
it can change rapidly in response to events in the environment. In addition, multiple
targets can be tracked by assigning a certain fraction of the available pulses to
track them. This lengthens the time required to complete one search pattern and
also makes the beam motion appear to be partially erratic or random. Mechanically
scanned systems make up the bulk of all existing radar systems. However, the
deployment of ESAs is proceeding rapidly due to:
235
236 Antenna Scan Analysis
1. Rapidly decreasing costs of digital memory and computational power;

2. Development of precision microwave integrated circuits;
3. Development of solid-state power sources at microwave frequencies.
Some radar systems may use manual (operator) control to position the antenna.
Perhaps, when the radar is acquiring a target to be tracked, the nonperiodic beam
motion may confuse an ELINT observer. The usual case, however, is that the radar
performs a periodic beam motion pattern that is amenable to ELINT analysis
almost immediately. Of course, thousands of different ELINT observations may
be required to catalog all of the combinations of angular coverage, scan speeds,
and patterns available to the radar.
The scanning function’s overall character is summarized by the following
parameters:
1. Plane of scanning;
2. Slew rates of the beam motion (angular change per unit of time);
3. Dwell time at one beam position;
4. Period of time required to complete the scan (if periodic).
Although these general characteristics are important, it is usually necessary to

describe the scanning pattern in more detail. The common types have been given
names (such as circular, sector, and raster) and different parameters are important
in describing each of these types.
10.2 Some Principles of Searching
Theoretically, it makes no difference whether the transmitted radar signal is formed

into a narrow beam or not. The key to the search process is the energy received
from a target. A single, broad-transmitted beam covering the whole search sector
combined with multiple narrow receive beams can achieve the same result. The
received echo power is then integrated over the same total time as would have
been taken to point a narrow transmit beam sequentially to cover the sector. This
approach is rarely followed because:
1. It is costly to implement many parallel receive channels.

2. The long integration times required are difficult to implement, especially
for moving targets with significant Doppler variations.
3. Radially moving targets may fly through a particular range cell before the
integration is completed.
Many radars, therefore, use the same antenna and beam shape to both receive
and transmit.
Beam motion is typically in the same direction as the longest dimension of the
antenna aperture because that is the direction of the narrowest antenna beam. For
example, a cut paraboloid antenna wider than it is high has a narrow azimuth
10.3 Relationships Among Scan Rate, Maximum Unambiguous Range, and Energy on Target 237
beamwidth and a broad elevation beamwidth. This antenna would typically be

rotated (scanned) in azimuth. This search radar could provide target azimuth and
range, but target elevation information is missing unless the radar includes multiple
beams in elevation or some type of elevation scanning along with azimuth scanning.
In a traditional radar installation, such a two-dimensional search radar would be
used together with a height finder having a tall but narrow antenna aperture that
searches in elevation at a particular azimuth (to which it is directed by the search
radar). The height finder then provides elevation and range information. Combining
the data from the two radars locates the target in three dimensions. This illustrates
a fundamental principle of search: it is quicker to search a large volume coarsely
and reserve high-resolution searching for those regions of the volume where a
target has been coarsely located. In this example, the search proceeded sequentially
by dimension—first the azimuth dimension was searched, then the elevation dimen-
sion.
Suppose a given resolution cell requires illumination for a certain time, Tcell ,
to determine if a target is in a cell. If the number of such cells in a given volume
is (NAZ N EL ), then examining all of the cells requires searching each cell and this
requires a time equal to (NAZ N EL )Tcell . If the dimensions are searched sequentially,
the time required is (NAZ + N EL )Tcell —a dramatic reduction. If the number of cells
is the same in each dimension, the reduction is a factor of NAZ /2. This fundamental
idea is the driving force behind most of the search schemes now in use.
Usually, the radar scans a given angular region so that the same amount of
time is devoted to each part of the region and the search is begun at some convenient
portion of the region. In some cases, the time per cell is reduced if there is less
need to detect at long range in that region. For example, an air search radar may
need to detect targets at long range (say, 300 km) near the horizon (at low elevation
angles) but only need to detect targets at short range (say, 30 km) at high elevation
angles because aircraft are limited to flying at some known maximum altitude.
Clearly, better results can be obtained if the search takes into account any a priori
knowledge of where the target may be located. For example, it would be useful
to search along the predicted trajectory of a projectile when the launch time and
velocity vector are known. If the probability of detection is related to the time
spent searching a particular cell, and if the a priori probabilities of the target being
in each cell can be assigned, an optimum search strategy can be formulated [1].
While present radar designs generally do not use such a strategy, electronically
scanned, computer-controlled arrays have the potential for such optimal search.
10.3 Relationships Among Scan Rate, Maximum Unambiguous

Range, and Energy on Target
As described in Chapter 2, the radar designer must carefully choose the waveform
parameters when designing the radar. The ELINT analyst can make use of these
same design criteria to understand the signal and determine the function of the
radar system. The total time to cover the search region must be selected to meet
operational criteria. Obviously, if the region is searched only once a day, this is
not acceptable for an air-defense radar. For repetitive scans, one can immediately
know that the radar needs to obtain target information at least once per scan cycle
to do its job. The maximum unambiguous range of the radar is determined by the
PRI (see Chapter 2). It is given by
R u = (c) PRI/2
The time spent illuminating one target is important because the ability to detect
a target depends on the amount of energy that hits it. The energy is proportional
to the number of pulses sent in the direction of the target. Clearly, the faster the
beam moves in angle, the fewer the number of pulses that strike a particular target.
Suppose the number of pulses received as the beam moves past the bearing of the
ELINT site is determined. This is referred to as the number of pulses per beamwidth
(PPBW). Then the time spent by the radar ‘‘looking’’ in the direction of one target
is
Tcell = (PRI)(PPBW)
The number of cells to be searched in some manner is determined by the

antenna beamwidths and the size of the angular region to be searched. For an
antenna having certain beamwidths, the number of cells is
AZ sector EL sector
NAZ N EL = ⭈
AZ beamwidth EL beamwidth
Now it is clear that the total scan time is
Tscan = Tcell (N AZ N EL )
Consider the following example. Suppose a 100° × 50° (AZ)(EL) region is to

be searched with a 1° beamwidth in both azimuth and elevation. There are then
5,000 beam positions. Suppose the radar designer wished to have 16 pulses strike the
target to take advantage of integration to reduce the peak power of the transmitter.
Suppose the maximum range to be searched is 300 km. This requires a 2-ms PRI.
The total scan time is then
Tscan = (5,000)(16)(0.002) = 160 sec
Clearly, this design is not acceptable for an aircraft detection system because
the targets can move an appreciable distance during that time. At 200 m/s, an
aircraft would move 32 km between radar scans. Because the scan time, beamwidth,
PPBW, and PRI are observable and because the extent of the scan can often be
determined, these relationships can be used to check for consistent values among
these ELINT parameters. If a fan beam is used in elevation so that all elevations
of interest are illuminated in one azimuth beam position, the number of beam
positions to be searched is reduced to 100 and the scan time required is reduced
to 3.2 seconds—a much more practical value obtained by giving up knowledge of
the target’s elevation.
10.4 Fan Beam Scanning: Circular and Sector
An antenna beam that is wide in one direction and narrow in the perpendicular
direction is called a fan beam. Such a beam is used for scanning in one direction
only. As noted, the classic scan pattern for a long-range search radar is circular
scanning in azimuth. This type of radar is expected to provide azimuth bearing
and range information, but not elevation. To achieve coverage of all elevations, a
wide elevation beam is needed. To provide good azimuth resolution, a narrow
azimuth beam is needed, as shown in Figure 10.1(a). The antenna aperture (as
noted in Chapter 8) is therefore wide in the horizontal direction and narrow in
the vertical direction. The elevation beam should extend to the maximum altitude
to be searched at the minimum range of interest.
The direction of rotation could be either clockwise or counterclockwise without
affecting the radar performance. If the ELINT analyst wished to determine the
direction of rotation, two intercept antennas and receivers could be used to deter-
mine which antenna first intercepts the main beam of the radar, provided that it
can be determined on which side of a line connecting the interceptors the radar is
located.
The speed of the azimuth scanning motion depends on the performance required
of the radar. The detectability of targets is basically dependent on the energy that
strikes the target, which for a pulsed radar is proportional to the number of pulses
Figure 10.1 Typical circular scan intercept pattern: (a) circular scan region; and (b) amplitude
versus time pattern.
used to illuminate the target as the beam sweeps past. An important parameter of
the radar is, therefore, the number of PPBW. For a circular scan this is given by
(Scan Time)␪ AZ
PPBW = (10.2)
(360)(PRI)
where:
PPBW = pulses per beamwidth

Scan Time = time required to scan 360° (one revolution)
␪ AZ = 3-dB azimuth beamwidth (degrees)
PRI = radar pulse repetition interval (average)
The expected pulse amplitude versus time for such a radar appears as shown
in Figure 10.1(b).
For long-range search radars, more energy is needed on target; therefore, more
PPBW are to be expected. However, a longer pulse interval is also needed so that
a large unambiguous range can be achieved. Both of these require a longer scan
time. Typical long-range search radars (with a range capability of hundreds of
kilometers) have scan times in the 10- to 30-second range. The long rotation time
does not hamper the performance of long-range search because, at such ranges,
targets need not be immediately tracked or engaged. Shipboard surface search
or ground-controlled intercept radars might have somewhat less detection range
capability, but they could make use of the more frequent scan capability. Circular
scan times up to 10 seconds might be typical, with correspondingly shorter pulse
intervals. It would be rare to find a 360° circular scan completed in as little as
1 second.
A circular scan is only one of a number of commonly used periodic scans. If
only a portion of the full azimuth circle is of interest, the radar may scan only a
sector. A sector scan can occur in two forms: bidirectional and unidirectional. It can
also be performed in azimuth or elevation (or any other orientation). A bidirectional
mechanical scan is simply a reversal of the direction of the scan when an angular
limit is reached. This is often used in height-finding radars that have a tall, narrow
aperture to produce a narrow elevation beam. Such ‘‘nodding’’ height finders are
commonly used along with circular scanning search radars to obtain elevation
angle measurements on targets at azimuths specified by the search radar. Therefore,
such elevation sector scanning is often accompanied by nonperiodic slewing in
azimuth. The height finder should have range capabilities similar to that of the
search radar with which it is associated. (The display of a height finder is typically
an elevation versus range map.) Targets are associated with those detected by the
search radar by noting the range as measured by both the search radar and the
height finder. Often, the pulsing of the two radars is synchronized.
The intercepted scan pattern for a bidirectional sector scan depends on the
position of the intercept receiver, as shown in Figure 10.2. A horizontal bidirectional
sector scanning in azimuth is most often observed in aircraft search radar modes
Figure 10.2 Intercepted bidirectional sector scan patterns at four different receiver locations.
or shipboard radars. This is the situation when the scan patterns indicated in Figure
10.2 are encountered at ground or shipboard intercept sites located at different
azimuth positions with respect to the radar. With height finders, the situation for
ground-level ELINT sites is generally that of position 3 or 4 in Figure 10.2. The other
situations in Figure 10.2 can be encountered from an airborne ELINT platform if
a height-finding radar set is being intercepted. Unfortunately, positions 3 and 4
present a difficult situation from which to determine the time between the 3-dB
points (or even between the nulls). Position 3 can cause an estimate of up to twice
the actual time between the 3-dB points. Position 4 reveals only sidelobe levels.
(The scan period can be estimated from position 4, however.)
The sector width cannot be directly determined from the amplitude-versus-
time data alone; only the sector-to-beamwidth ratio can be determined. This ratio
is given by the ratio of the sector period to the time between the 3-dB points.
Determining the angular width of both the sector and the beam is usually done by
estimating the beamwidth from the antenna dimensions. Then the sector width is
determined from the sector-to-beamwidth ratio. For height finders, the sector width
can be estimated from the function. If the maximum range is known (it could be
estimated from the PRI) and the maximum altitude of the targets of interest is
known, then the maximum elevation limit is known at the maximum range. Usually,
the height finder must cover high elevations only to ranges much closer than the
maximum range. Typical sector limits are a few degrees below the horizon to 30°
or more above the horizon, as indicated in Figure 10.3. If the range from the
interceptor to the height finder is known and the elevation of the ELINT receiver
is changed, the scan pattern changes as shown in Figure 10.2. The sector width
could then be estimated from the altitude change needed to produce a given change
in the scan pattern.
Unidirectional sector scanning is used when the radar function requires rapid
scanning (such as rapidly changing, short-range situations). Reversing the motion
of the aperture takes time and results in spending too much valuable search time
at the sector edges. Unidirectional scanning can be implemented by moving the
antenna feed, not the whole reflector. The idea is shown in Figure 10.4. RF energy
enters the rotating feed at the center. Several feed horns are spaced around the
rotating mechanism, but only one at a time couples to the center waveguide and
faces the reflector. No matter where the intercept site is located within the sector,
the same scan pattern is observed. The effect is that of a circular scan at the feed,
made to cover a sector by the geometry of the feed and reflector mechanism.
Figure 10.3 Height finder with a bidirectional elevation scan.

Figure 10.4 Unidirectional sector scan mechanism.
Usually, the high speed is an indication that circular scanning is not being observed.
Sometimes, slight variations in the effectiveness of the various feed horns cause an
amplitude variation from main beam to main beam that allows the ELINT analyst
to infer the number of horns in use.
Circular scanning radars also can include multiple beam systems. The beams
are pointed at different elevation angles and thus provide a rough indication of
the target altitude. Each beam is often transmitted at a separate RF, and multiple
transmitters and receivers are used for each beam. Typically, four to six separate
beams are used with narrower beam patterns near the horizon. The ELINT station
must be equipped with multiple receivers to demonstrate conclusively the simultane-
ous use of more than one beam. Of course, the signal power of the beams whose
sidelobes are directed toward the ELINT site will be lower than that of one whose
main beam is received.
Another type of circular scan system designed to measure elevation is the
V-beam radar. This is an obsolescent approach, but it is still in use. The V-beam
uses two reflectors with one mounted at an angle to the other (typically slanted
45°). The two beams are transmitted at different frequencies and form a V pattern
in space. The time interval between the detection in one beam and the other depends
on both the target range and altitude. The V-beam can also include multiple beams
in each leg of the V, as shown in Figure 10.5. An ELINT intercept of the V-beam
radar using a single narrowband receiver channel appears exactly the same as an
ordinary circular scan radar. Two receiver channels are needed to demonstrate the
V-beam property. The antenna scan patterns from the two receiver channels will
appear as shown in Figure 10.6. The beam offset shown in Figure 10.6 depends
on the beam-to-beam angular separation at the horizon plus the angular separation
due to the V-beam pattern if the ELINT receiver is at a higher elevation.
Figure 10.5 V-beam and multiple-beam spatial coverage.
Figure 10.6 Intercepted scan patterns from (a) multiple-beam and (b) V-beam radars.
10.5 Pencil Beam Scanning: Raster, Helical, and Spiral 245
10.5 Pencil Beam Scanning: Raster, Helical, and Spiral
A radar beam with nearly equal azimuth and elevation beamwidths is called a
pencil beam. To search effectively with such a beam requires motion in both
azimuth and elevation. One common way to do this is through a raster scan, where
a line is scanned in one direction (e.g., azimuth) over a certain sector, and then a
second line is scanned in azimuth but at a different elevation angle. Figure 10.7
shows a four-line raster pattern with the lines horizontal (they could also be vertical
or at any other angle).
The important parameters needed to describe a raster are its overall angular
dimensions, the number and orientation of the lines, and the time required to
complete one raster pattern. As illustrated by the intercepted scan patterns in Figure
10.7, the number of lines and the time period of one raster are usually determined
rather easily. The time per line can then be estimated. The angular length of one
line can also be estimated by multiplying the beamwidth by the ratio of the time
per line to the time for one beamwidth. Because a raster is usually designed so that
there are no gaps between the lines, the angular extent of the raster perpendicular
to the line direction is approximately the beamwidth in that direction multiplied
by the number of lines.
Figure 10.7 Scan pattern from a four-line raster.

A helical scan is a search pattern raster with horizontal lines but where each
line covers 360°. Actually, the helical scan consists of a pencil beam circularly
scanning with the elevation of the beam increasing by about one beamwidth during
each azimuth revolution, as shown in Figure 10.8.
A spiral scan is used for searching a restricted region of space, like a raster
scan; but the beam starts at a central point and spirals out from there, as shown
in Figure 10.9. The signal amplitude versus time can be quite complex and varies
with the intercept point. The spiral shown rapidly returns to the center and is
generally outward. However, radars are also designed to spiral out and then spiral
back in. The intercepted scan pattern reveals the basic period required to complete
the scan. Some airborne intercept radars use a spiral scan for searching. A ground-
based ELINT site, therefore, may be in the antenna sidelobes and must deal with
a moving emitter.
10.6 Tracking Scans and Monopulse
A radar system is said to be performing tracking if it has located a target and is

following its movement in range, elevation, and azimuth. One technique for doing
this is conical scanning. When conical scan is used for tracking a target, the peak
of the main beam never quite strikes the target. Instead, the peak of the main beam
rotates circularly about the target, with the target being the center of the circle.
The line from the radar through the peak of the main beam is called the beam
axis. The motion of the beam axis forms a cone in space, as shown in Figure 10.10
[2]. The motion of the target causes the center of the cone to move as the radar
follows the motion. The ELINT antenna will not be in the main beam of the radar
unless it is on the target being tracked. For all other intercept points, the received
Figure 10.8 Helical scan pattern.

Figure 10.9 Spiral scan pattern.
Figure 10.10 Conical scan. (From: [2].  1970 by McGraw-Hill, Inc. Used with permission of
McGraw-Hill.)
pulse amplitude will vary approximately sinusoidally in step with the conical scan.
This variation will be much less if the ELINT antenna is on the target, because
the conical scan system tracks by feedback loops that attempt to maintain the
target position at the center of the cone, where equal amplitude from the main
beam exists.
Usually, the ELINT analyst can easily determine the period of the conical scan
rotation (or its reciprocal, the conical scan rate). Typically, the conical scan rate
is in the range of 25 to 80 Hz (revolutions per second). The conical scan rate is
selected so that the target movement during one rotation period is limited to a
small fraction of the beamwidth. The other parameters of interest for such a tracking
radar are the beam axis slewing rates that can be achieved and the squint angle
(the angle between the beam axis and the target axis of the cone formed by the
conical scanning of the beam axis).
The slewing rate can be determined only from observing the radar performance
in situations where the geometry is known. For example, if the ELINT antenna is
on the target that the radar is tracking, abrupt aircraft maneuvers can test the
limitations of the tracking system.
The squint angle is difficult to determine through ELINT. Clearly, the squint
angle must be on the order of the 3-dB beamwidth because the radar loses target-
detection capability if the cone angle is too large, that is, the tracking point is too
many decibels down on the antenna pattern.
The key conical scan parameter is the rotation rate. If a jammer transmits a
pulsed signal with amplitude variations at the same rate as the conical scan rate,
the tracking circuitry can be disrupted.
An ELINT operator can easily identify a conical scan by the sound alone. The
25- to 80-Hz amplitude modulation of the basic pulse repetition frequency creates
a warbling sound. Naturally, the tracking condition also causes a continuous signal
with no noticeable amplitude modulation when the ELINT receiver is on the target
being tracked.
A conical scan can also be implemented by providing scanning of only the
receive beam. This is called conical scan on receive only (COSRO). One common
implementation is to transmit through the sum port of a monopulse-type horn feed
cluster. On reception, the azimuth and elevation difference ports are sampled with
weights varying sinusoidally at the scan frequency ␻ s to form a composite ⌬ channel:
⌬ = ␦ az cos ␻ s t + ␦ el sin ␻ s t
This is combined in a hybrid to create ⌺ + ⌬, which represents a receiving

beam offset from the ⌺-channel axis by a fraction of the beamwidth and rotating
about the axis at ␻ s . This implementation avoids mechanical vibration and rotation
of the polarization vector, and permits rapid and variable scan rates.
It is difficult for an ELINT analyst to determine the conical scan rate (or even
the existence) of COSRO. Although the transmitted signal is not scanned, there is
usually enough mismatch and reflection in the microwave system to impose small
AM on the transmitted signal, disclosing the scan rate to ELINT. A tracking system
that does not exhibit any other type of tracking scan technique may be presumed
to use some type of receiver technique for tracking purposes. In this case, a high
SNR, possibly combined with a high-resolution spectrum analysis, may reveal the
COSRO rate.
After locating a target, the conical scan tracking system must make the transition
from the search scan to the tracking scan. To hasten this process, the conical scan
may be left on during the search process. The search scan may be of the raster,
helical, or spiral variety. A helical plus conical scan is shown in Figure 10.11. (Such
scans are sometimes called Palmer helical or Palmer raster.) The main difference
observed at the ELINT site is the relatively fast variation in pulse amplitude at the
conical scan rate superimposed on the normal received scan patterns.
Figure 10.11 Helical plus conical scan.
Historically, the conical scan systems were preceded by lobe switching systems
where the beam position could be rapidly switched slightly from right to left (and
up and down). At first this switching was done manually and the operator adjusted
the beam axis to make the returns from the various beam positions equal. Such
lobing was done either with the combined receive/transmit beam or on receive only
[dubbed lobe on receive only (LORO)]. ELINT analysis can reveal lobing on
transmit by the discrete steps in the received signal strength. The resultant signal
amplitude would have a square-wave-like modulation. LORO is more difficult to
detect, but high SNR intercepts may reveal slight variations in the transmitted
beam caused by the switching of the LORO antenna system components.
The disadvantages of all of the conical scan and lobing techniques include the
susceptibility to tracking errors due to fluctuations in the echo amplitude and
problems of tracking at long range. The conical scan or lobing rate should be
relatively fast for good tracking, but at long range, the pulse interval becomes so
long that only a few pulses occur in time to complete one conical scan or lobing
cycle. Furthermore, when the echo from a pulse transmitted in one direction returns
from a target far away, the beam will have moved to another direction.
These disadvantages were overcome by the development of the monopulse.
(This term does not mean that only one pulse is used, but rather, in essence, lobing
is performed on receive only using a single pulse.) Because monopulse is essentially
a receiver technique, its use cannot be directly revealed through ELINT analysis.
A tracking radar that does not exhibit any other scanning technique may be pre-
sumed to use monopulse, especially if it is of recent design. The relative performance
of monopulse, compared to the other techniques, is summarized in Table 10.1.
Another important type of tracking radar is track-while-scan (TWS). One TWS
method uses two fan beams and orthogonal sector scans to cover two overlapping
Table 10.1 Relative Performance of Some Tracking Systems

Parameter Monopulse Conical Scan Sequential Lobing
Effective beamwidth 1.0 1.4 1.3
Normalized to monopulse
Beam offset loss (two-way) 0 −3 dB −6 dB
Maximum update rate PRF PRF/4 PRF/4
Typical update rate PRF PRF/10 PRF/4
Source: [2].
sectors using the orthogonal scans, as shown in Figure 10.12. The target being
tracked is kept in the center of the overlapping scanned areas by feedback systems to
the antenna drive motors. TWS is used in virtually all two-dimensional surveillance
radars to provide azimuth and range data. It can also be used for three-dimensional
data [as in most ground controlled approach (GCA) precision approach radars]
by time-sharing a single transmitter between azimuth and elevation sector-scanning
antennas. Other TWS systems can be implemented in receive-only configurations
by using the sector scanning antennas to feed the receivers and a third nonscanning
antenna for transmit. In this case, the transmit beam must be broad enough to
cover the entire sector scanned by the receive antennas. In the receive-only system,
a single transmit frequency is used; whereas in the TWS system using two transmit-
ters, two frequencies must be used.
Like the conical scan and lobing systems, TWS systems are susceptible to
jamming based on amplitude modulation at the scan rates. TWS on receive-only
is a remedy for this. TWS radars (and monopulse radars) are used in missile systems
for target and missile tracking. As such, they are of high interest to ELINT analysts.
The parameters of interest for the TWS system include the beamwidths, sec-
torwidths, and sector scan directions. (For example, both of the sector scan direc-
tions could be tipped 45° to create a diamond-shaped overlap area. The important
factor is that the two scans are orthogonal.) The receive-only versions make determi-
nation of such parameters very difficult using ELINT techniques alone. If two
transmit beams are scanned, the problems are the same as described for sector
scanning in Section 10.3. In the scan-on-receive-only systems, the transmit beam
is relatively broad and yet the nonscanning character of a tracking system is evident.
Figure 10.12 Track-while-scan mechanisms.

10.7 Electronic Scanning 251
10.7 Electronic Scanning
Because electronic scan systems are often computer controlled, analysis of their
scanning patterns really amounts to determining what computer programs exist in
the software library of such a multiple-function array radar. The functions of the
various programs as well as the beam-steering technique and its limitations are of
interest. In addition, ELINT observations can define many signal parameters
whether or not their functions are apparent. These parameters include pulses per
beam position, the period for a repeated scan pattern, and possibly the number of
discrete beam positions. A commonly used search scan is the raster type, which
has many of the same properties as mechanically scanned systems.
Note that electronic scanning in one dimension (e.g., elevation) is often com-
bined with mechanical scanning in another (e.g., azimuth). The ELINT observer
can easily distinguish electronic scanning and multiple functions by the large variety
of periods and patterns observed.
Periodic programs are often used for searching. They may be interrupted when
targets are detected for verification routines—typically, a pause at a particular
beam position to allow more pulses to hit the target. Verification may be done
more than once (e.g., within several successive search patterns). After verification,
target tracking is often performed by directing the beam toward each target on
some schedule. These dwells also interrupt the search function so that the more
targets there are, the longer it takes to search the volume of space for more targets.
Ultimately, a limit is reached so that either the searching stops or a maximum
number of tracks is reached. The number of pulses used during each track beam
dwell, as well as the period of time between the track dwells, should be determined
by the ELINT analyst. This can sometimes be done because the ELINT analyst
can associate a given pulse amplitude with fixed radar beam angles (for a constant
intercept geometry). Thus, while the actual beam angle remains unknown, the
return to the same beam position can be determined by analysis of the pulse
amplitude pattern. The ELINT analyst may also discover that the RF, pulse interval,
and pulse duration may be programmed differently. Also, the analyst may find
that the radar performs different functions. These variations may correlate with
beam position. A good example is a frequency-steered array with the beam elevation
a function of the RF. The pulse interval may be longer at low elevation angles near
the horizon, where a longer unambiguous range is required. As the elevation
increases, the pulse interval can be shortened so that the maximum altitude searched
remains fixed. This reduces the time required to complete the search pattern. To
maintain the same average power, the pulse duration may be reduced in proportion
to the pulse interval. Finally, because the maximum detection range is reduced at
high elevation angles, less energy is required for target detection. Therefore, the
number of pulses per beam position may also be reduced. The energy reduction is
partially accomplished by reducing the pulse duration. Because the PRI reduction
is in direct proportion to the range reduction, but the energy required varies as
the fourth power of the range, a significant reduction in the number of pulses per
beam position is possible.
The key parameter characterizing the multifunction ESA is the beam dwell
time. This is often the same as the CPIs of the radar, or the beam dwell time may
be as long as several CPIs. The CPI reveals what the coherent integration time is
and this helps the ELINT analyst estimate the detection capabilities of the radar.
10.8 Scan Measurement and Analysis Techniques
Historically, scan analysis was first performed by using headphones and a stop-
watch. The received signal pulse repetition frequency is usually in the audio range
(especially on older radars, although not on pulse Doppler systems). When the
main beam was directed toward the ELINT site, the strong tone burst was easily
identified and the stopwatch was started. On the next burst (or some number of
bursts later) the watch was stopped, and hence, the scan period was determined.
Sector, conical, and other scans can also be determined in this way: a sector scan
by the syncopation due to an intercept point away from the sector midpoint; a
conical scan by the warbling amplitude modulation; and a spiral scan by the fading
and then strengthening of the main beam tone bursts. The time resolution was in
the 0.1- to 0.01-sec range. As wider bandwidth chart recorders became available,
this became the most prevalent method for scan analysis. A problem common to
both audio analysis and chart recorder analysis is that the duty cycle of most radar
signals is quite low—on the order of 10−3. The amplitude of such narrow pulses
is greatly reduced by the lowpass filtering action of the audio system or chart
recorder. It is therefore necessary to incorporate pulse stretch circuitry, such as a
sample-and-hold circuit that is discharged after a fixed time (e.g., 100 ␮ s), ahead
of the recorder or audio system.
In systems that digitize and store pulse arrival time, amplitude, and duration,
many options for analysis are available. Time-domain analysis techniques are the
most useful because the scan action of a radar is usually designed from the view
of accomplishing sequential search or track functions, which are best conceived
(at present) in the time domain. Scan analysis techniques tend to be intuitive and
heuristic rather than analytical. This is partly a reflection of the fact that radar
scanning schemes themselves are not strongly based in any theory, but historically
have been developed in response to the need to search in ways that could be
accommodated by the available hardware.
Bidirectional sector scanning can be distinguished by the sidelobe structure.
Real antenna patterns are not symmetrical, and because sector scanning involves
reversing the direction of the beam motion, these reversals can be discovered
easily by the mirror-image appearance of the scans. This allows distinguishing
bidirectional sector scans from unidirectional sector scans. Conical scanning can
be implemented by moving the feed in a circular pattern while maintaining a fixed
orientation of the feed, as shown in Figure 10.13(a). Then the polarization remains
fixed. This same effect can be obtained by keeping the feed fixed and rotating the
reflector dish by one-half of the conical scan cone angle. The rotating feed of Figure
10.13(b) produces rotating polarization.
Multipath can cause problems in determining scan patterns. It may cause large
signal amplitudes to occur that may be interpreted erroneously as additional main
beam illuminations. For fixed ground ELINT sites, these are caused by reflections
from local objects (e.g., water tanks and mountains), and they can be anticipated.
10.9 A Three-Dimensional Search Example 253
Figure 10.13 Two conical scan mechanisms: (a) nutating feed; and (b) rotating feed.
For intercepts from airborne radars, or especially airborne ELINT stations, reflec-
tions from aircraft propellers can be confusing because the propeller can cause
amplitude modulations similar in appearance to conical scanning at rates of 40 to
60 Hz.
These hints for scan analysis indicate that experience is very important in
properly interpreting the pulse amplitude versus time variations: ELINT scan analy-
sis is more an art than a science.
10.9 A Three-Dimensional Search Example
A unique radar that electronically searches in elevation while circularly searching

in azimuth illustrates a number of points about scanning in particular and search
in general. (If you are unfamiliar with pulse compression, you may wish to read
Chapter 11 before continuing with this example.) The combination of electronic
scanning in elevation with mechanical scan in azimuth is a design frequently used
by three-dimensional search radars. The electronic scan in elevation can be achieved
using a constant delay in the feed to each row of an array antenna. Then changing
the carrier frequency changes the phase of the signal to each row, which, in turn,
changes the elevation pointing angle of the main beam. However, the changing
carrier frequency, if done rapidly within the pulse, can also be used for pulse
compression [3].
From the ELINT point of view, the S-band signal exhibits a circular scan with
a period of, say, 10 seconds. Conventional analysis provides the information that
the azimuth beamwidth is 1.5°. The pulse duration is 30 ␮ s, and the pulse amplitude
varies many decibels about a peak, whose position within the pulse at a ground
site is near the leading edge of the pulse. When intercepted from an airborne
receiver, the peak is later in the pulse. There is a large amount of frequency
modulation within the pulse—with a frequency deviation of 120 MHz. The
frequency deviation within the region of the peak pulse amplitude is always about
8 MHz regardless of the location of the peak within the pulse. The PRI is long,
like that of a search radar. Suppose a photograph were available of a square array
antenna, which is thought to be the source of this signal.
The task at hand is to explain the capabilities of this radar in light of what is
known about the signal. The frequency deviation of 120 MHz would provide range
resolution of about c/(2 bandwidth) or 1.25m, where c is the speed of light. This
is too fine for general search applications, which are indicated by the 10-second
scan interval. If a photograph were available of the antenna showing a square
array, that would indicate that the elevation beam width is the same as the azimuth
beamwidth, which is 1.5°. Such a beam would necessarily be scanned in elevation
to make an effective 3-D search radar. Furthermore, the array means that electronic
scanning is possible.
A simple type of electronic scan is elevation scanning by frequency, which in
this case must occur within a single pulse. If that were the case, the peculiar
amplitude peak would correspond to that portion of the 30-␮ s pulse that was most
directly aimed toward the interceptor’s elevation. In this case, the ratio of the
120-MHz frequency deviation to the 8-MHz deviation in the amplitude peak of
the pulse indicates that there are about 15 beamwidths of 1.5° scanned in eleva-
tion—possibly 0° to 22.5°. Furthermore, a pulse-compression technique could be
used to provide 1/8 MHz or 0.125 ␮ s of range resolution or 18.75m. The elevation
angle of the target would be indicated by the frequency at which the peak echo
amplitude occurred. The pulse-compression ratio would be the time bandwidth
product represented by the duration of the peak amplitude portion of the pulse
times its bandwidth of 8 MHz. The 8-MHz portion represents 1/15 of the total
pulse duration of 30 ␮ s or 2 ␮ s. Hence, the pulse-compression ratio would be on
the order of 2 × 8 = 16. (This example is based on a three-dimensional search
radar originally made by Plessey.)
References
[1] Stone, L. D., Theory of Optimal Search, New York: Academic Press, 1975.
[2] Skolnik, M. I., Radar Handbook, New York: McGraw-Hill, 1970, Chapter 21.
[3] Radford, M. F., and R. Greenwood, ‘‘A Within-Pulse Scanning Height Finder,’’ IEE Int.
Conf. Radar-73, London, October 23–25, 1973, pp. 50–55.
CHAPTER 11
Intrapulse Analysis
11.1 Introduction
Many of the ELINT signals of interest consist of pulses—particularly radar signals.

Determining the character of the intrapulse modulation gives valuable insight into
the radar’s function and design. The term ‘‘intrapulse’’ refers to the shape of the
pulse envelope (or amplitude modulation function) and also to the frequency or
phase variations within the pulse. If the carrier is not frequency modulated, the
pulse is sometimes referred to as a CW pulse. (This term is sometimes used in
radar literature and simply means a pulse with no intentional frequency or phase
modulation.) If there is no intentional FM or PM within the pulse, the envelope
of the pulse determines the range resolution of the radar. If there is intentional
FM or PM, demodulating it (or determining its bandwidth) gives insight into the
range resolution or other functions of the radar. Examining the unintentional AM
and FM or PM can sometimes provide insight into the transmitter type and may
help to identify the signal.
The radar waveform designer, of course, does not consider the individual pulses
as an isolated parameter but must consider the entire modulating function, including
the pulse intervals and phase or frequency modulation. The radar waveform deter-
mines its ambiguity function,1 which is essentially the autocorrelation of the wave-
form with itself as it would appear if reflected from a point target having range
delay ␶ and Doppler shift fd . This function is plotted in three dimensions as
magnitude versus range delay and radial velocity (Doppler shift).
If the signal at the ELINT site is carefully analyzed in terms of the amplitude
and frequency modulation (including pulse-to-pulse phase variations), an ambiguity
function for the radar can be estimated. The actual radar ambiguity diagram also
depends on the processing performed by the radar’s receiver (which cannot be
determined through ELINT) and the antenna scan, which may determine the length
of time the signal is directed toward a target.
It is customary in ELINT analysis to examine the pulse envelope in some
detail by itself because so many radars use waveforms that consist of a repeated
transmission of identical (or nearly identical) pulses. This helps to identify the
signal type or at least its likely function. The pulse envelope shape must be observed
at the output of a relatively wideband detector. In the past, pulse shape could be
preserved using oscilloscope photographs or wideband analog tape recorders (e.g.,
5- to 10-MHz bandwidth). At present, high-speed A/D converters are used. Some
1. See [1], Chapter 3, for a discussion of the ambiguity function.
255
256 Intrapulse Analysis
typical A/D specifications are listed in Figure 11.1. At this time, A/D converters
having 8 bits per sample operate up to 4 gigasamples per second (GSPS). Sample
rates of about 400 megasamples per second (MSPS) at 12 bits per sample are also
available.
The results of using a high-speed A/D converter to capture a single pulse are
shown in Figure 11.2. A 200-MHz digitizing rate has been used to capture the
predetection IF pulse (centered at about 50 MHz), along with the output of an
AM detector (labeled ‘‘SQR DET’’) and the FM within the pulse from the output
of a frequency discriminator (labeled ‘‘DISCRIM’’). The predetection (IF) pulse
Figure 11.1 Some high-speed A/D converters.
Figure 11.2 IF, AM, and FM digitized data (512 samples at 5 ns = 2.56 ␮ s).
11.2 Pulse Envelope Parameters 257
looks as if there is some modulation present; however, the AM and FM outputs

show an ordinary unmodulated pulse. The apparent modulation on the IF signal
is a beat between the 200-MHz digitizing rate and the fourth harmonic of the
50-MHz IF signal. (This is a consequence of having only four samples per IF cycle.)
To capture the intrapulse AM and FM simultaneously requires multiple high-speed
digitizers and the associated memory. It is common to have two channels available.
Three or more channels is less common but sometimes useful. While the AM and
FM waveforms may be obtained through digital demodulation of the predetection
IF samples, analog detection methods may offer the advantage of wider dynamic
range (especially in the case of an FM limiter-discriminator combination).
The actual pulsing of the radar is performed in a variety of ways [1, Ch. 7].
The details of the envelope shape concern the radar designer mainly for controlling
the frequency spectrum outside the radar’s band of interest. Interference with other
radar or communications systems can result if poorly shaped pulses are transmitted.
The ELINT analyst can gain insight into both the radar’s capabilities and the
modulation design by examining the details of the pulse envelope.
11.2 Pulse Envelope Parameters
In describing the pulse envelope it is convenient to use time-domain descriptors

that are easy to measure and can be related to radar performance. The usual choices
are the pulse duration, rise and fall times, and intentional or unintentional variations
in pulse duration (or even pulse amplitude). The pulse duration is usually defined
as the time between the half-power points of the pulse. For a linear detector, these
are the points at 0.707 of the peak pulse voltage. For a square law detector, the
voltage point of interest is at 0.5 of the peak pulse voltage. For any particular
receiver detector, a calibration curve should be made to determine the half-power
point voltage as a function of the peak pulse voltage.
Likewise, the rise time and fall time are the time between the 10% and 90%
power points on the rise of the pulse and between the 90% and 10% power points
on the fall of the pulse. In all cases, the percentages apply to the average amplitude
of the pulse. A photograph of three successive radar pulses is shown in Figure
11.3. The photograph, taken from a moving platform, shows pulses from a circular
scan search radar. The pulse amplitude, therefore, changes slightly. Because the
signal is very powerful, receiver noise is not significant. However, multipath can
cause variations in the fine structure along the top of the pulse.
Such a photograph of individual pulses is quite different from that of numerous
pulses overlaid, which results when an ordinary oscilloscope setup is used. Single
trace photographs (or at most a few traces) are important for an analysis of the
pulse envelope details. If the amplitude is changing drastically due to scanning, the
successive pulse envelopes will appear to have different pulse durations and rise
and fall times, as indicated in Figure 11.4.
In the past, the difficulty in recording the pulse envelope over the wide dynamic
ranges encountered meant an emphasis on oscilloscope and photographic tech-
niques. Photographing successive pulses (as shown in Figure 11.3) requires a simple
device such as a counter and a digital-to-analog (D/A) converter to offset the trace
Figure 11.3 Envelopes of successive radar pulses.
Figure 11.4 Multiple trace photograph effect on pulse envelope.
before (or after) each pulse. The scheme is shown in Figure 11.5. Each pulse triggers
the sweep. After a delay expires, the counter advances, increasing the voltage from
the D/A converter. This offsets the traces vertically on the screen in preparation
for the next pulse. The camera shutter remains open during the entire process.
11.3 Envelope Parameter Measurements
High-speed A/D converters with sufficient resolution and sampling rate allow the
digital capture of pulse envelopes. Precise measurements often require interpolation
between the samples.
Figure 11.5 Raster technique for photographing successive pulses.
Automatic digital measurements of the pulse duration and rise and fall times
must be interpreted carefully because changing irregularities in the pulse shape due
to multipath and noise can cause wide variations in the readings. This is particularly
true of time interval measurements. The problem becomes one of accurate threshold
detection at, say, the 50% points, which must be properly located in spite of
amplitude variation.
One type of threshold can be constructed using a delay and a sample and hold,
as shown in Figure 11.6.
Rise and fall time measurements are difficult to automate due to the short
times involved (often less than 50 ns). The digital approach can be used if the times
are not too short relative to the time between the samples of the A/D converter.
The bandwidth of the ELINT receiver and A/D converter is very important in
making the rise and fall time measurements. For good pulse fidelity, oscilloscopes
incorporate amplifiers having an amplitude response that rolls off very slowly with
Figure 11.6 Half-power threshold scheme.

frequency. The pass band limit is often specified so that the 12th harmonic of the
specified pass band limit is attenuated by no more than 75% [2].
The key point to be determined is the effect of the receiving system on the
observed rise time. Figure 11.7 illustrates the effect of the various filters on the
shape of the output pulse (private communication with D. K. Barton, February
2005).
11.3.1 Rise and Fall Times2

The rise time ␶ r of a pulse is defined between the 10% and 90% voltage levels.
When normalized to the bandwidth, B␶ r varies as shown in Figure 11.8, as a
function of the noise bandwidth B n and half-power bandwidth B 3 dB (which are
identical for the rectangular filter). The curves for that filter contain ripples,
resulting from the interaction of the (sin x)/x spectrum of the pulse with the edges
of the rectangular bandpass. However, for B␶ Ⰷ 1 the normalized rise time is
essentially constant at 0.9/B for the rectangular filter, and 0.7/B to 1.1/B for the
other filters. For all filters ␶ r is reduced for B␶ < 2. This results from the fact that
output peak amplitude is reduced more rapidly by the narrow bandwidth than is
the slope of the leading edge. For B␶ Ⰶ 1, ␶ r < 0.7/B, a result of the reduced peak
outputs for the Gaussian and single-pole filters. For the more usual filters that are
approximately matched or wider than matched, the well-known approximations
given in [3–5], are applicable:
• From [3, p. 40]: ␶ r = 0.7/B, where B is the IF bandwidth;

• From [4, p. 413]: ␶ r = (0.7 to 0.9)/B3, where B3 is the half-power IF
bandwidth;
• From [5, p. 464]: ␶ r Ⰷ 1/B, where the bandwidth definition is unspecified.
Note that Terman [4, p. 289] gives a value for video amplifiers as ␶ r = (0.35
to 0.45)/B v , where B v is the (lowpass) video bandwidth measured from zero. This
agrees with the factor (0.7 to 0.9)B 3 dB for IF bandwidth that is measured on both
sides of the center frequency.
A rule that has been used states that the rise time that results from cascading
n stages (each of which is free of overshoot) is given by [6]
␶ = 冠␶ 12 + ␶ 22 + . . . ␶ n2 冡
0.5
(11.1)
If the actual rise time of the transmitted radar pulse is ␶ T and the rise time of
the ELINT receiver and display system is taken as ␶ R , the measured rise time, ␶ M ,
is
␶ M = 冠␶ T2 + ␶ R2 冡
0.5
(11.2)
or
␶ T = 冠␶ M − ␶R冡
2 2 0.5
(11.3)
2. The author thanks Mr. David Barton for portions of the material in this section.
Figure 11.7 Output waveform versus filter bandwidth (additional delay applies in the cases of
realizable Gaussian and rectangular filters).
Figure 11.8 Normalized rise times versus filter bandwidth.
Equation (11.3) provides a means for estimating the transmitter rise time given
the measured rise time and the inherent rise time of the receiving system. Figure
11.9 shows a plot of the ratio of the transmitter rise time to the measured rise
time as a function of the ratio of the measured rise time to the receiver rise time.
As can be seen, good estimates of the transmitter rise time require that the observed
rise time be more than 1.5 times the receiver rise time.
Consider an example to illustrate the use of (11.3). The first step is to apply
a laboratory pulse generator with a very short rise time to the input of the ELINT
equipment and measure the rise time of the equipment. Suppose this is 50 ns. Next,
suppose an ELINT signal is measured and found to have a rise time of 70 ns. Then,
(11.3) gives the estimated rise time of the transmitted signal as about 49 ns—
considerably less than was observed at the receiver output and slightly less than
the rise time of the equipment itself. If this seems contradictory, remember that
the laboratory measurement was made with essentially a zero rise time pulse applied
and that the observed output was 50 ns. Increasing this to 70 ns requires an input
rise time of about 49 ns.
Figure 11.9 Rise time of transmitted signal related to measured rise time.
The receiver rise time should be measured prior to deployment. However, for
planning purposes, the approximate relationship of this rise time to the receiver
bandwidth may be used:
0.7
␶R ≈ (11.4)
BR
where:
B R = receiver bandwidth (IF bandwidth or, for most systems, twice the video
bandwidth)
␶ R = receiver 10% to 90% rise time
Equation (11.4) applies exactly for a simple resistor-capacitor (R-C) filter. It

also closely approximates the response of a cascade of R-C filters or of any filter
whose step response has less than a 5% overshoot [6].
It should be noted that the maximum slope of the leading edge of the pulse
occurs at the half amplitude point and is approximately equal to the amplitude
divided by the rise time. For the usual case of an ELINT receiver with bandwidth
considerably larger than the reciprocal of the pulse duration, the slope/amplitude
ratio ranges from about 0.9 to 1.1 times the reciprocal of the rise time [private
communication with D. K. Barton, February 2005].
Historical Note. A pulse analysis technique designed to ease the burden of the
ELINT analyst was the multiple gun oscilloscope [3]. By displaying several traces
with different sweep speeds on a single screen, the pulse duration and pulse interval
could be estimated at a glance without adjusting the controls. Full-screen sweep

times of 5, 50, 1,000, 5,000, and 50,000 ␮s were used on one model. The first
two covered most pulse durations of interest, while the last three covered the most
PRIs of interest. The PRI scales were sometimes calibrated in pulses per second
rather than units of time. Some of these pulse analyzers used traces with nonlinear
sweeps (log or exponential) to cover wide signal parameter ranges.
The accurate estimation of pulse shape parameters is one of the more difficult
tasks faced by the ELINT analyst. It is essential to have available calibration signals
of known rise time, pulse duration, and fall time to calibrate the entire receiver,
detector, digitizing, display, and measurement chain.
11.4 Some Radar Performance Limits Related to Pulse Envelope
The elementary radar equation that relates range resolution to pulse duration is
range resolution ∼ (speed of light)(pulse duration)/2
In addition, for moving target indication (MTI) systems using delay line cancel-
lers, the MTI improvement factor is limited by the presence of pulse duration jitter
[1, p. 1748]:
I (dB) = 20 log (pulse duration/pulse duration jitter)
The envelope of the RF spectrum is related to the pulse shape. Figure 11.10
shows the approximate relationship between the RF spectrum envelope and the
rise (or fall) time of a trapezoidal pulse as a function of the pulse duration. Figure
11.11 shows the same relationship for several other pulse shapes [1, p. 2920]. The
electromagnetic interference to other signals that may be caused by a powerful radar
Figure 11.10 RF spectrum envelope for trapezoidal pulses (for rise time ␦ 1 = fall time ␦ 2 = ␦ ).
(From: [1].  1970 by McGraw-Hill, Inc. Used with permission of McGraw-Hill.)
11.4 Some Radar Performance Limits Related to Pulse Envelope 265
Figure 11.11 RF spectrum envelope for several pulse shapes. (From: [1].  1970 by McGraw-Hill,
Inc. Used with permission of McGraw-Hill.)
makes it necessary for the transmitter designer to carefully control the spectrum roll-
off. This often determines the pulse rise and fall times observed through ELINT.
To see how important this can be, consider the situation where two radars
operate in close proximity (e.g., at the same airport). It is not unusual for the main
beam power to be above +100 dBm and for the receiver to be sensitive to −100-dBm
signals. Therefore, if the two systems are not to interfere with each other, there
must be 200 dB of isolation between the transmitter of one and the receiver of the
other. Suppose 100 dB is obtained by putting one on one side of the airport and
one on the other side. The remaining 100 dB of isolation is to be obtained by
operating the radars at different frequencies.
If a radar were to transmit rectangular pulses, the upper curve in Figure 11.10
gives the attenuation as a function of the frequency separation. Suppose the pulse
duration is 1 ␮ s. Then the horizontal scale of Figure 11.10 is expressed in megahertz.
To obtain 100 dB of isolation when transmitting rectangular 1-␮ s pulses requires
about 20,000 MHz of frequency separation. If radar 1 is at 2 GHz, radar 2 would
be at 22 GHz! However, most radar systems have band-limiting elements in the
transmission path, and hence, the actual pulse becomes more nearly trapezoidal,
with rise and fall times set by this bandwidth. If a trapezoidal pulse with a rise
time of 10% of the pulse duration is used, the spectrum falls 20 dB per decade
instead of 10 dB per decade and the required frequency separation is reduced to
about 300 MHz—a much more practical situation.
Because radar transmitters are generally high-power saturated amplifiers, the
more rounded pulse shapes of Figure 11.11 are not generally used. (A Gaussian-
shaped pulse has a very narrow spectrum, but it is infinitely long in time and
impractical to generate.)
By carefully observing the pulse envelope, the ELINT analyst can infer some-
thing about the ability of the radars to operate in close proximity and also perhaps
the minimum frequency spacing of RF channels if these are used.
Sometimes, the spectrum width of the pulse is of interest. For example, this
might be useful for determining the spectrum width of a noise jammer to counter
the emitter. For an ideal trapezoidal pulse with the rise time equal to the fall time,
the 3-dB bandwidth of the spectrum in terms of the pulse duration, as measured
between the 50% points, is shown in Figure 11.12. For a rectangular pulse, the
3-dB bandwidth is 0.88 divided by the pulse duration, while for a triangular pulse
(the rise time and fall time equal to the pulse duration) the 3-dB bandwidth is
about 0.64/pulse duration. The first spectrum nulls for a trapezoidal pulse are
always separated by 2, divided by the pulse duration, regardless of the ratio of the
rise and fall times to the pulse duration. Often the fall time is longer than the rise
time. The 3-dB bandwidth of a trapezoidal pulse having a fall time three times the
rise time is also indicated in Figure 11.12. In this case, the spectrum does not have
a null near the reciprocal of the pulse duration; however, a deep minimum occurs
there (see Appendix A).
11.5 Multipath Effects
Multipath may drastically alter the pulse envelope and also the apparent frequency
variations within the pulse. Multipath can be modeled simply as a summation of
Figure 11.12 Bandwidth of trapezoidal pulses.

11.5 Multipath Effects 267
delayed replicas of the direct path signal with various amplitudes and phases. In
any given situation the multipath effect can vary significantly from pulse to pulse.
If the radar or the intercept platform is moving, movement of a significant fraction
of a wavelength during the interpulse period is sufficient. For example, at 3 GHz
the wavelength is 0.1m and a typical PRI is 1 ms. Moving a quarter-wavelength
(0.025m) in 1 ms corresponds to 25 m/s or 90 km/hr. Even if both the radar and
interceptor are not moving, the rotation of the antenna beam causes the amplitude
of the signal reflected from different objects to change. This is especially true if
the intercepted signal is from the radar’s sidelobes.
Although the effects of multipath are most often associated with the observed
pulse envelope (AM), the observed FM is also affected. Both effects can be demon-
strated in the laboratory with the help of a microwave delay. Figure 11.13 shows
the effect of adding equal amplitude signals at S-band—the original signal and the
same signal delayed 300 ns. An adjustable line was used to vary the phase of the
delayed signal, as shown in the Figure 11.13. On the AM pulse shape, the multipath
may add or subtract, depending on the phasing of the delayed signal. To understand
the intrapulse FM, consider that the onset of the multipath causes a discontinuity
in the phase. The FM waveform is the derivative of the phase. Therefore, the
discriminator output contains a transient at the beginning and end of the multipath.
The transient may be positive or negative depending on the sign of the phase
change.
The multipath shown in Figure 11.13 is quite extreme: the delayed signal and
direct signal are of equal strength. Nevertheless, it serves as a warning to those
Figure 11.13 Laboratory simulation of the multipath effect on AM and FM.

who wish to make use of the pulse shape: both the envelope and the phase can be
significantly affected by the multipath. Even the most careful analysis of the inter-
cepted waveform may fail to reveal the transmitted pulse shape under these condi-
tions.
The strongest multipath comes from reflecting objects near the transmitter and
receiver and along the path between them. Normally, the radar site will be on a
hill or otherwise away from large objects if at all possible. Likewise, the ELINT
receiver will also be located away from nearby large objects if possible. The path
between these sites may have hills or cities nearby that are unavoidable. The
location of reflecting objects determines the delay of the reflected signal. If the
range difference is 300m, the arrival time will be delayed by 1 ␮ s. (The speed of
light is about 300 m/␮ s.) If the delay is less than the pulse duration, there will be
a multipath distortion of the pulse shape. If the delay is more than the pulse
duration, two separate pulses will be observed and neither will be affected by the
multipath. It is clear that the leading edge of the pulse is the best hope for observing
at least part of the pulse free of the effects of multipath and that the trailing edge
may consist entirely of multipath effects.
11.6 Intrapulse Frequency and Phase Modulation
Intentional intrapulse phase and/or frequency modulation is quite common in

modern radar. Perhaps most radars of recent design use such modulation for pulse
compression purposes. (This does not mean that half of the radars manufactured
recently have pulse compression. Many radars are designed but never manufactured,
and many more inexpensive than expensive radars are manufactured.) But it is a
common technique used to increase radar range while maintaining good range
resolution without increasing the peak power of the radar. The frequency or phase
may also exhibit unintentional modulation and it may be combined with intentional
modulation.
The radar’s pulse compression ratio is the ratio of the transmitted pulse duration
to that of the pulse duration at the receiver output after compression. The pulse
compression ratio can be estimated from ELINT because it is always less than or
equal to the time-bandwidth product of the radar pulse. By measuring the pulse
duration (using the pulse envelope) and by measuring the bandwidth of a single
pulse (by examining the output of a frequency discriminator or phase detector or
by spectral analysis), the ELINT analyst can determine the time bandwidth product.
The time bandwidth product of conventional radar that does not use pulse
compression is about unity (when the pulse duration is measured at the −3-dB
level). The bandwidth is about equal to the reciprocal of the pulse duration. Radar
that uses pulse compression is likely to have a time bandwidth product of at least
10. It is seldom worth adding the cost and complexity of pulse compression to a
radar design unless at least this much compression is used. (As a counter example,
the modified APY-2 AWACS radar uses a five-element Barker code to improve its
ability to resolve range ambiguities caused by the high-PRF waveform.) Typical
radars use time bandwidth products between 10 and 1,000. Some special-purpose
very long-range and/or very fine-range resolution radars have time bandwidth
products in excess of 100,000. There is an upper limit for the radar’s time bandwidth
product due to target motion. If the target moves radially with velocity, v, and if
the range resolution of the radar (range cell extent) is distance, d, the amount of
time the target will be in that range cell is T = d /v. Achieving this range resolution
requires a bandwidth of B = c /2d, where B is the bandwidth and c is the speed of
light. Thus, if the target is to stay in the range cell for the entire pulse duration,
the time bandwidth product is limited to BT < c /2v. If the velocity is 500 m/s, the
upper limit for BT is 300,000.
The most common types of modulation are phase reversals and linear FM. For
a phase reversal signal, the range resolution is easily determined as being equivalent
to that of conventional radar with the pulse duration equal to the minimum time
between the phase reversals. The waveform for a 13-bit Barker code is shown in
Figure 11.14. The predetection IF signal shows some drop in amplitude at the
times of the phase reversals. This is usually due to a restricted bandwidth in the
transmitting path, which causes the envelope to drop to zero as the phase is reversed.
The waveforms marked Phase I and Phase Q are a result of mixing a limited version
of the predetection signal with local oscillator signals that are 90° out of phase
and nearly at the same frequency as the IF. It is easy to see the phase reversals in these
waveforms. For linear FM (chirp), there would be a quadratic phase progression that
would show up as an increasing frequency of the Phase I and Phase Q signals. But
the more usual approach would be to use a frequency discriminator to observe the
modulating signal directly. The predetection signal is often digitally demodulated.
For many methods of digital FM demodulation, the first step is to generate in-
phase (i) and quadrature (q) samples from the predetection IF samples. This can
be done either in the time domain by convolving the IF signal with the impulse
Figure 11.14 Part of a pulse with 180° phase reversals (5-ns sampling interval).
response of the Hilbert transform, or in the frequency domain by multiplying the

fast Fourier transform (FFT) of the IF samples by −( j) sgn (2␲ f ) and then taking
the inverse FFT. (This function is the Fourier transform of the impulse response
of the Hilbert transform.) Once the i and q samples are available, the phase samples
can be found by using the following equation:
phase = arctan(q /i) (11.5)
⌬␾ = arctan (q 2 /i 2 ) − arctan (q 1 /i 1 ) (11.6)
The frequency is the phase rate of change; the average frequency over the
sample interval is the change in phase over the sampling interval divided by the
sampling interval. Because the phase is determined over an interval of 360°, it is
necessary to unwrap the phase (i.e., add 360° when the phase change from one
sample to the next is too great). This can be avoided by using the expression for
the difference of two arctan values:
1 d␾ ⌬␾ q1i 2 − q2 i1
f= ≈ = (11.7)
2␲ dt 2␲ ⌬t 2␲ (t 2 − t 1 ) (i 1 i 2 + q 1 q 2 )
The frequency value computed using (11.7) is actually the average frequency
over the time interval between the two sets of (i, q) values rather than the true
instantaneous frequency. To approximate the instantaneous frequency, the fre-
quency should not change significantly during the time between the samples.
To see the precise relationships between the amplitude, phase and in-phase,
and quadrature functions, it is useful to represent a phase-modulated signal as
s(t) = a(t) cos [2␲ fo t + ␾ (t)] (11.8)
where a(t) is the amplitude or envelope, fo is the carrier frequency, and ␾ (t) is the
phase modulation; that is, the integral of the frequency modulation. Expanding
the cosine function gives
s(t) = a(t) cos ␾ (t) cos (2␲ fo t) − a(t) sin ␾ (t) sin (2␲ fo t) (11.9)
s(t) = i(t) cos (2␲ fo t) − q(t) sin (2␲ fo t) (11.10)
from which:
i(t) = a(t) cos ␾ (t)

q(t) = a(t) sin ␾ (t)
√i
2
a(t) = (t) + q 2(t)
␾ (t) = arctan [q(t)/i(t)]
One of the main problems with digital FM demodulation is the limited dynamic
range when A/D converters of only a few bits of resolution are used. Analog
discriminators, when preceded by a limiter, provide a much wider dynamic range

over which accurate FM demodulation is accomplished. One drawback is the
nonideal discriminator curve (frequency-to-voltage transfer function), which varies
from system to system. This means that the same signal, when processed by two
different receiving systems, produces slightly different outputs. Of course, this may
also occur with digital receiving systems, where equalization may be required to
make two systems produce the same output.
11.6.1 Choosing the Receiver Bandwidth

Even though the modulating signal for a phase or frequency modulated signal is
band-limited, the modulated signal is not band-limited.
The bandwidth required to preserve the fine structure of the modulation is
about twice the deviation plus about 10 times the rate of modulation. This is shown
in Appendix B, Section B.2. (This bandwidth requirement is sufficient to retain
sidebands in the sinusoidal FM spectrum that are no more than 40 dB down.) If
the pulse duration is on the order of 1 ␮ s and there are five cycles of modulation
across the pulse, a discriminator bandwidth of 50 MHz is needed. Of course, the
wider the bandwidth, the more noise in the output (which may then obscure the
FM).
Whether or not analog discrimination or digital demodulation is used, noise
will be present in the FM output. If the signal is large compared to the noise, the
phase angle variation due to noise is given (to close approximation) by the quadra-
ture component of the noise divided by the signal amplitude. In the case of a
limiter-discriminator combination, the weaker noise is reduced relative to the signal
so that the output SNR increases by about 3 dB. Therefore, the performance of
this type of discriminator is enhanced somewhat compared to linear demodulation
methods. Figure 11.15 shows the amount of frequency noise at the output of a
Figure 11.15 Discriminator rms noise output as a function of SNR input.

digital or analog discriminator with no limiter as a function of the input SNR. In

this calculation, the bandwidth after the discriminator is matched to the bandwidth
at the input. Because the output noise rises in proportion to the square of the
frequency, lowpass filtering the discriminator output can significantly reduce the
output noise at the cost of obscuring the FM waveform. For a given input SNR,
there is actually more signal level present in a wide input bandwidth than in a
narrow one. The video filter is able to eliminate so much of the noise that the
output SNR is greater than the input SNR for wide input bandwidths. (This is
apparent from (B.53) and (B.50) in Appendix B.)
As can be seen, the SNR required for rms discriminator output noise to be less
than 100 kHz is quite large. This situation can be improved by post discrimination
video filtering if the modulation rate allows.
11.7 Intentionally Modulated Pulses
It is customary to include the measurements of intentional phase or frequency

modulation-on-pulse with other measurements of the RF, such as pulse-to-pulse
frequency agility and the carrier frequency itself. For this reason, a discussion of
some of the techniques used to analyze these characteristics is included in Chapter
14. Of course, many of the same techniques are used regardless of whether the
modulation is intentional or not.
Intentional intrapulse modulation is nearly always for the purpose of pulse
compression. The modulations most often used are linear FM (chirp) and binary
phase reversals, sometimes in the form of Barker codes. An example of a linear
FM waveform is shown in Figure 11.16, and an example of a 13-bit Barker code
BPSK signal is shown in Figure 11.17. Other modulation types include stepped
frequency modulation (or FSK), a discrete form of linear FM and phase shift keying
(PSK) using phase steps of other than 180°. These are sometimes called polyphase
codes to distinguish them from the more common binary codes. Both FM and PSK
must deal with the problem of time sidelobes—these are ‘‘false’’ peaks in the output
of the pulse compression processor caused by partial correlation of the received
echoes with the transmitted waveform at time shifts (range delays) at other than
the true range. For linear FM, it is common to use weighting networks at the
output of the pulse compression filter to decrease these undesired outputs in a
manner similar to the reduction of sidelobes in antenna patterns by tapering the
aperture illumination. In the case of BPSK, the Barker codes have time sidelobes
of maximum amplitude of 1, with the correlation peak amplitude equal to the
length of the code. However the longest known Barker code is 13, which limits
pulse compression using these codes to a maximum of 13:1. Longer BPSK sequences
can be used at the price of higher time sidelobes. For example, random signals can
be used with time sidelobes on average near zero but on any particular random
code transmitted the time sidelobes are expected to be approximately the square root
of the code length. Polyphase codes that use small phase steps often approximate the
performance of linear FM. Note that linear FM has a quadratic phase progression.
Nonlinear FM may be used, often for the purpose of reducing time sidelobes. The
smallest frequency steps that result in orthogonal segments are such that there is
11.7 Intentionally Modulated Pulses 273
Figure 11.16 Modeled linear FM. Pulse width = 1 ␮ s; ⌬ f = 20 MHz (BT = 20).
a ±90° phase accumulation over the duration of the constant frequency step. Thus,
if the step duration is T, the minimum frequency shift is such that FT = ±90°.
These frequencies are transmitted in a manner to provide minimum shift keying
(MSK) waveforms, sometimes called Taylor codes. The N-length sequence Taylor
coded waveform consists of phase modulated chips in the I and Q components of
the signal at increments of ±90°. Each chip is comprised of a one-half cycle sine
wave envelope. Listed in Table 11.1 are the known Taylor coded waveforms as
given in [7]. Figure 11.18 illustrates the waveform for the 28A Taylor code.
From the ELINT viewpoint, it is important to determine the time bandwidth
product of the signal and also to determine what kind of modulation is used. This
information helps identify it and also helps determine its function.
Most radars have pulse compression ratios in the range of 10 to 1,000, with
a few specialized radar systems for missile defense or for mapping having very
large time bandwidth products up to 100,000 or more.
Once the intentional modulation has been analyzed, additional information
can be obtained from the unintentional modulation. For example, it may be possible
to tell one radar from another by the amount of nonlinearity found on its LFM
waveform or by the amplitude ripple observed on the top of the pulse due to
bandpass filtering a BPSK signal prior to transmission.
There is a need for a measure of distortion and/or how well two waveforms
match. Another way of looking at this is to ask the question how much bandwidth
is needed in an ELINT system to avoid distortion of the signals of interest. Several
Figure 11.17 Modeled Pulse with 13-bit Barker code.
Table 11.1 The Known Taylor Codes

Code Length, N Phase Sequence
13 1 j −1 −j 1 −j 1 −j 1 −j −1 j 1
25 1 j −1 j −1 j −1 −j −1 −j 1 j −1 −j −1 j 1 −j −1 j −1 j 1 j 1
27 1 −j −1 −j −1 j 1 j 1 −j 1 j 1 −j 1 j 1 −j 1 j 1 j −1 −j −1 −j 1
28A 1 j 1 −j 1 −j −1 j −1 j 1 j −1 j 1 j −1 j 1 j −1 j −1 −j 1 −j 1 j
28B 1 j −1 j −1 j −1 −j −1 −j 1 j −1 −j −1 j 1 −j −1 j −1 j 1 j 1 −j 1 −j
answers have been given already, based on the forms of Carson’s rule developed
in Appendix B and, for CFM, the distortion as measured in the peak and rms
frequency deviation for the distorted and undistorted signal have been investigated
(see Table B.2). A more general method of comparing signals for distortion measure-
ment of other purposes is to use the normalized Euclidean distance (NED). For
two signals s 1 (t) and s 2 (t), this is defined as
√
t2
冕
t1
[s 1 (t) − s 2 (t)]2 dt
NED = (11.11)
√ √
t2 t2
冕t1
[s 1 (t)]2 dt + 冕
t1
[s 2 (t)]2 dt
Figure 11.18 Modeled 28A Taylor code pulse.
For comparing waveforms, it is then convenient to adopt some maximum value

of NED as the threshold as the criterion for deciding two signals are the same (or
matched). This NED threshold value can also be used to determine what bandwidth
is needed for the ELINT system in order that the output matches the input. For
CFM, the relationship between modulation index and NED when only the first
pair of sidebands is preserved is shown in Figure 11.19.
11.8 Incidental Intrapulse Shape—Uses and Causes

The traditional ELINT measurements include pulse duration; rise and fall times;
and time bandwidth product, intrapulse phase, or frequency modulation for pulse
Figure 11.19 NED for CFM when the bandwidth is limited to the first sideband.
compression systems. Also included are the variations of these values (e.g., pulse
duration jitter). There is additional information contained in the fine structure of
the AM and PM wave shape, but the nature and use of this information depends
on many factors. One of the most important of these is the SNR. To discover the
fine structure of the intrapulse modulation through ELINT may require SNR values
in excess of 40 dB. Once the ‘‘true’’ shapes are unmasked from the effects of noise
and multipath, it may be possible to recognize which type of transmitter is in use
(e.g., magnetron, klystron, and solid-state). Some have thought that finely resolved
pulse shape data can be used at the front end of EW systems to deinterleave pulse
trains even if they come from the same emitter type. Unfortunately, this would
require high SNR values for single pulses. Bringing the pulse shapes out of the
noise usually requires averaging together many pulses, which implies that they
have already been deinterleaved.
Such averaging involves taking many individual wave shapes, aligning them in
time and frequency, and then averaging them point by point across the waveform
ensemble. The alignment process is critical. The simplest technique is to align the
3- or 6-dB points on the envelope and then average both the AM and FM waves
based on that alignment. A more sophisticated method is to start with that alignment
and then shift each waveform right or left a few samples to determine the alignment
that gives the largest cross-correlation (or smallest NED). This alignment of pulse
envelopes and averaging the envelopes is a noncoherent ensemble average of pulses
having high individual SNR values (e.g., 20 dB). It is expected that the SNR
improves at least as the square root of the number of pulses averaged—better than
that if the SNR is high to begin with. If the individual pulses have a 20-dB SNR
and a 40-dB SNR is desired, it may be necessary to average 1,000 pulses. Collecting
that number of pulses may or may not be practical. A circularly scanning search
radar may provide on the order of 25 to 50 main beam pulses. Hence, 20 to 40
scans could be required to collect 1,000 main beam pulses. If the SNR is high to
begin with, the averaging may improve the SNR by a factor equal to the number
of pulses averaged. Note that, for a scanning signal with varying amplitude, it is
not very useful to include pulses that are too far below the peak. Consider the
coherent ensemble average of two pulses of amplitudes a 1 and a 2 in the presence
of noise of power N. Coherent here means that the peaks of the cycles of the
predetection (or IF) sine waves for the two pulses are aligned in time. Adding these
signals corrupted by noise directly gives a sine wave whose peak is now the sum
of the peaks of the two signals and whose noise power is now 2N. The SNR of
the sum is then
(a 1 + a 2 )2
SNR ENS = (11.12)
4N
With no weighting of the amplitudes, it is easy to see that if the amplitude of

the weaker pulse is small, the SNR of the ensemble may be less than the SNR of
the stronger pulse alone. That is, for SNR improvement, it is necessary that
2
(a + a 2 )2 a 1
SNR ENS = 1 ≥ (11.13)
4N 2N
This implies that the weaker pulse amplitude must be at least 0.414 times the
amplitude of the stronger pulse (or no more than 7.66 dB down) in order to
improve the SNR of the ensemble. If a weighting factor is used, this limitation can
be removed. Consider using a weighting factor of k times the amplitude of the
weaker signal:
(a 1 + ka 2 )2
SNR ENS =
2(1 + k 2 )N
Differentiating the SNR with respect to k and setting the derivative equal to
zero shows that the value of k for maximum SNR of the ensemble average is
a
k= 2
a1
This means that the best ensemble average is formed by scaling the amplitudes
of the weaker signals by the ratio of the amplitude of the weaker signals to the
amplitude of the strongest signal. Clearly, signals that are (say) 20 dB below the
amplitude of the strongest one will not contribute much to the overall SNR of
the ensemble. Also, misalignment of the cycles of signals that are weak can easily
result in SNR degradation instead of enhancement. Thus, practical considerations
limit the ensemble average to the relatively strong pulses. Such ensemble averaging
is often needed to bring out the character of the unintentional intrapulse frequency
or phase modulation—or the unintentional variations of intentional modulation
from the ideal—for example, incidental nonlinearity of a chirp signal.
The causes of unintentional intrapulse modulation are often inherent in the
modulation sensitivity of high-power transmit tubes [6]. The major effects are:
(1) pushing; (2) pulling; and (3) temperature, aging, and poor maintenance.
Pushing is caused primarily by variations in the output of the modulator and
high-voltage power supply. Pulling is caused by variations in the load impedance
seen by the tube primarily for pulsed oscillators (magnetrons). An example is a
poor rotating joint in the waveguide to the antenna. The other effects are more
slowly changing; for example, changes in the intrapulse wave shape as the tube
warms up (changes over several hours) or as it ages over a period of weeks, months,
or years. An example of poor maintenance, which may be seen in the wave shape,
would be if the high-voltage power supply output decreases and no action is taken
to adjust it.
The effects of different types of modulators on incidental intrapulse modulation
are summarized in Table 11.2. There is not much incentive to reduce such effects
if there is no significant improvement in radar performance. The greatest variations
are caused by the modulator rate of rise voltage, ringing, and droop. Although the
rate of rise is usually thought of in terms of the envelope, the frequency variations
of a magnetron can be significant—it is even a strong determinant of the mode in
which magnetron oscillations occur. In fact, many FM artifacts may also be seen
in the envelope as well (i.e., there is a definite correlation between the FM and
AM wave shapes). Both the amplitude and phase change as a function of the
Table 11.2 Modulator Interpulse Effects by Type

Intrapulse
Type Switch Application Pulse Shape Expected*
Line Thyratron silicon Most common Ripples 5
controlled rectifier
(SCR)
Hard tube Capacitor coupled Fairly common Good 3
Transformer coupled Not often used Fair 4
Modulator anode Usually high power Excellent 2
(floating deck)
Grid Widely used Excellent 2
SCR SCR and magnetic Becoming more Ripples 5
cores common
*5 = significant; 1 = minimal.
Source: [8].
applied voltage. For example, a 1% ripple in the high-voltage modulation of a

10-GHz magnetron would cause a frequency deviation of 200 to 600 kHz at the
same rate as the ripple [8]. For klystrons and traveling wave tube (TWT) amplifiers,
the FM is more commonly specified in terms of phase variation per unit change
in voltage. For a TWT, a 1% ripple in the cathode to anode voltage may cause a
phase change of 5° or an amplitude change of 0.1 dB [9].
11.9 Comparing Wave Shapes
For FM and AM, the NED can be a useful measure of difference between two
waveforms. These differences can be obtained to compare each pair of waveforms.
Once the differences are available, a histogram of the differences can reveal the
ability of the waveform to separate the data into classes. Figure 11.20 illustrates
such a histogram of differences. In this example, to properly associate the pairs
that belong to the same class (intraclass), it would be necessary to set a NED
threshold of about 0.4 units. Unfortunately, this would also include about 24%
of the pairs that do not belong together (interclass). If a threshold of 0.2 is used,
all of the interclass pairs will be rejected, but only about half of the intraclass pairs
will be accepted. This is typical of such classification schemes. There is a trade-off
between correctly grouping those that belong together and falsely including those
that do not belong together. Hence, two thresholds may be needed. For example,
if the NED is less than 0.2, the pulses may be said to match. If the NED is between
0.2 and 0.4, the pulses may or may not match. If the NED is greater than 0.4, the
waveforms are said to not match.
The receivers used to compare waveforms must be well matched in amplitude
and group delay across their pass bands, and the signals must be tuned to near
the center of the channel. Equalization is generally needed to match receivers and
digitizers well enough to achieve small NED values when the same test signals are
applied to each receiver. Figure 11.21 shows the phase error of the demodulated
FM signal as a function of time alignment error between the upper and lower
sidebands of a circular FM signal. The phase shifts allowed for NED values of 0.1
and 0.2 are indicated. Of course, as the modulation rate of the FM increases, the
11.9 Comparing Wave Shapes 279
Figure 11.20 Histogram of waveform differences.
Figure 11.21 Phase shift of the demodulated FM related to sideband time alignment error.
more critical is differential delay for one sideband relative to the other. For example,
if the receiver is to account for less than NED = 0.1 of mismatch for an FM signal
with a 15-MHz modulation rate, the differential delay between the sidebands
(30 MHz apart in frequency) must be less than 2 ns.
References
[1] Skolnik, M. I., Radar Handbook, New York: McGraw-Hill, 1970.

[2] Shackill, A., ‘‘Digital Storage Oscilloscopes,’’ IEEE Spectrum, July 1980, pp. 22–25.
[3] Chance, B., et al., Electronic Time Measurements, Vol. 20 of MIT Radiation Laboratory
Series, New York: McGraw-Hill, 1949. Also, ‘‘Radiation Lab Series: CD-ROM Edition,’’
Norwood, MA: Artech House, 1999.
[4] Terman, F. E., Electronic and Radio Engineering, 4th ed., New York: McGraw-Hill, 1955.
[5] Skolnik, M. I., Introduction to Radar Systems, New York: McGraw-Hill, 1961.
[6] Valley, G. E., Jr., and H. Wallman, Vacuum Tube Amplifiers, Vol. 18 of MIT Radiation
Laboratory Series, New York: McGraw-Hill, 1949, p. 77. Also, ‘‘Radiation Laboratory
Series: CD-ROM Edition,’’ Norwood, MA: Artech House, 1999.
[7] Blinchikoff, H. J., and J. W. Taylor, ‘‘Quadriphase Code—A Radar Pulse Compression
Signal with Unique Characteristics,’’ IEEE Trans. on Aerospace and Electronic Systems,
Vol. 24, No. 2, March 1988.
[8] Boyd, J. A., et al., Electronic Countermeasures, Los Altos, CA: Peninsula Publishing, 1978,
pp. 233–234.
[9] E. M. Galey, ‘‘Intrapulse and ELINT System Design,’’ Journal of Electronic Defense,
October 1989, pp. 87–106.
CHAPTER 12
Pulse Repetition Interval Analysis
12.1 Introduction
Pulse repetition interval timing is one of the major aspects of ELINT signal analysis.
The radar’s performance is affected significantly by the PRI timing used. In addition,
many warning receivers and jammers make use of properties of the PRI. Finally,
PRI data can generally be recorded easily for off-line analysis in the laboratory.
Most PRI values are such that modest bandwidths can be used. Generally, the
video or envelope bandwidth is of interest for recording or digitizing the PRI
(and pulse duration) data. It is normally the case that the video bandwidth is
approximately one-half of the two-sided predetection or IF bandwidth; however,
as noted in Chapter 3, in the case of a wideband receiver, the video bandwidth
may be small compared to the RF bandwidth of the receiving system.
The radar pulse is usually repeated; however, the interval between pulses may
change. It is common to speak of the PRI used by radar even though the interval
itself may not be repeated. In this discussion, the term PRI is used to mean the
time from the leading edge of one radar pulse to the leading edge of the next radar
pulse. For radar systems that use groups (or bursts) of several closely spaced pulses,
the PRI refers to the time from the leading edge of the first pulse in one group to
the first pulse in the next group. This may be referred to as the pulse group repetition
interval (PGRI).
Just as close scrutiny of the pulse envelope shape gives insight into some aspects
of the radar’s performance and the mechanization of the modulator, close scrutiny
of the pulse intervals gives insight into other aspects of the radar’s performance
and the mechanization of the pulse timing circuitry.
A subject that arises frequently in connection with PRI analysis is the use of
countdown circuitry so that the PRI is a multiple of some shorter interval. To
provide for accurate range calibration, the radar designer may find it convenient
to synchronize the transmission of the radar pulse with the oscillator generating
the timing marks. The PRI is therefore some multiple of the period of this oscillator.
Early countdown circuitry involved using analog multivibrator circuits to establish
a minimum pulse interval. The radar’s transmission occurred on the first pulse
from the timing oscillator after the expiration of the analog delay. Depending on
the stability of the delay, one of a number of different multiples of the timing
oscillator period might be selected as the PRI. Digital countdown circuitry is much
less prone to counting errors.
Another facet of PRI analysis is dropped or weak pulses. This is more a problem
of magnetron radars than of other types. A typical manufacturer’s specification
281
282 Pulse Repetition Interval Analysis
may be for 0.01% up to 1% missing pulses [1]. While this may make no noticeable
difference in the performance of the radar, a few missing pulses in a burst subjected
to ELINT analysis can cause confusion. Of course, weak pulses near the detection
threshold may also be dropped with probability equal to (1 − probability of detec-
tion). If the threshold is set to achieve a probability of detection equal to 90% at
some amplitude, then for that amplitude there is a missed pulse probability of
10%. A good rule of thumb is to set the threshold so that the probability of
detection is greater than (1 − missing pulse fraction) to avoid missing significantly
more pulses than those inherently missed by the transmitter.
12.2 Common PRI Categories

The variety of PRI schemes is seemingly infinite. There are, however, a number of
schemes used sufficiently often that these categories have been given names. Gener-
ally, the ELINT analyst explores the PRI variations and attempts to place the radar
emitter into one of the categories listed in Table 12.1. For each of these categories,
the ELINT analyst emphasizes different aspects of the PRI variations. As indicated
in Table 12.1, each PRI category is associated with different radar functions.
12.2.1 Constant PRI

The radar has a nearly constant PRI if the peak variations are less than about 1%
of the mean PRI. Such variations are considered incidental in that they generally
Table 12.1 Common PRI Variations

Type Typical Function Remarks
Constant Common search or track radars Variations typically less than 1%
of the average PRI value. Very
stable constant PRIs are
associated with MTI and pulse
Doppler systems
Jittered Reduces the effects of some types Large variations—up to about
of jamming 30% of the average PRI
Dwell and switch Resolve velocity (or range) Bursts of pulses with several
ambiguities especially in pulse stable PRIs switched from one
Doppler radars burst to the next
Stagger Eliminate blind speeds in MTI Several stable PRIs switched on a
systems pulse-to-pulse basis in a periodic
pattern
Sliding Provide constant altitude Maximum PRI (at minimum
coverage during elevation elevation) usually less than six
scanning or avoid eclipsing times minimum PRI at maximum
elevation
Scheduled Used in electronic scan, multiple Numerous complex patterns
function (search and track) (some periodic) may adapt to
computer-controlled systems target situation
Periodic variations Missiles guidance; avoidance of Nearly sinusoidal variations up
eclipsing, ranging to 5% of average PRI. Rates up
to 50 Hz or more
Pulse groups May improve range or velocity Also used for IFF and beacon
resolution interrogation
serve no useful purpose. If the peak PRI variations exceed 1% of the mean PRI, the
variations may serve some particular function. Occasionally, circuit malfunctions or
design peculiarities may cause large PRI variations that serve no useful purpose.
These may also be placed in the ‘‘constant’’ category. For constant PRI radar
signals, the parameters of interest include the mean PRI, the peak-to-peak incidental
PRI variations, the amount of PRI drift, and the time period over which the drift
was measured.
If the average PRI value is to be used to estimate the radar’s maximum unambig-
uous range, determining the PRI to an accuracy of about 1% is clearly sufficient.
On the other hand, very high accuracy may be needed if it is desired to investigate,
for example, the variation in the mean frequency of the crystal oscillator controlling
the PRI.
The incidental variations in the pulse interval are useful in determining the
characteristics of the radar’s trigger generating circuitry. For example, MTI systems
that use a delay line for trigger generation and cancellation of returns from station-
ary targets require precisely controlled pulse intervals. The limit on the MTI
improvement factor due to PRI jitter is
PD
I jitter (dB) = 20 log (12.1)
2 ⭈ PRIJ
where:
I jitter = limitation on the MTI improvement factor due to jitter

PRIJ = RMS PRI jitter
PD = pulse duration
This means that for 30-dB MTI improvement, a radar system using 1-␮s pulse
duration can tolerate an rms PRI jitter of no more than 15.8 ns. Measuring individ-
ual PRIs to such accuracy is difficult. It requires both a high SNR and a scheme
for avoiding apparent PRI changes due to changes in pulse amplitude as the radar
scans. A good rule to follow is to try to obtain jitter measurements to about 1%
of the pulse duration. However, practical limitations prevent meaningful jitter
measurements of smaller than 10 to 20 ns. Clearly, whatever the SNR and other
limitations faced by the analyst, the maximum jitter due to the radar can be
estimated from the received signal. The radar obviously has incidental jitter of no
more than the observed value. For that jitter, if (12.1) yields a useful MTI improve-
ment factor, the radar could have MTI capability.
The PRI drift characteristic is also of interest because it can indicate the type
of likely PRI stabilization in use. (For example, a trigger generator in an oven
compared to one with no oven or even a proportional oven compared to an on/
off thermostatically controlled oven.) Drift is expressed as a fraction of the average
PRI over some time period. The procedure is to determine the average PRI over a
short time interval (such as the number of pulses received during one scan of the
radar main beam past the ELINT receiver antenna). These short time averages are
then compared to the average of all the observed PRIs. An example is shown in
Figure 12.1. The maximum change in the short-term PRI average is divided by
the long-term average PRI to obtain a fraction. In Figure 12.1, the maximum
change is 12 ␮ s in a 3,000-␮ s PRI. The fractional change is, therefore, 6 × 10−3
in 3 minutes. The time interval is usually at least several minutes but may extend
over hours or even days.
It is normal for constant PRI radars to have several different PRIs that may
be selected manually or by software control. This may be done to resolve range
and/or Doppler ambiguities or to avoid eclipsing, as described in Chapter 2. In
this case, the ELINT analyst must look for transmissions using all of the PRIs and
perform the same jitter and drift analysis for each one.
12.2.2 Jittered PRIs

Intentional PRI variations are used for a variety of radar purposes, as indicated in
Table 12.1. The use of such variations implies radar capabilities lacking in constant
PRI radar. Intentional jitter (random PRI variations of large amounts) is used for
electronic protection (EP) from certain types of jamming and would therefore be
of interest to the ELINT analyst. The amount and type of jitter can also aid in
identifying the type of radar transmission being received. The parameters of interest
are the same as those for constant PRI radars, but analysis of the jitter waveform
is emphasized. Generally, individual intervals should be measured to an accuracy
of a fraction of a percent to give a good representation of the jitter waveform and
its statistics.
Figure 12.1 Typical PRI drift for a circuit in a thermostatically controlled oven.
Some radar sets may use jitter consisting of random selections among discrete
interval values. Others may have intervals that can take on any value within a
certain range. In the case of discrete intervals, the ELINT analyst should determine
what values are used. For either discrete or continuous PRI variations, the shape
of the distribution as well as the overall range over which the PRI varies must be
determined.
12.2.3 Dwell and Switch PRI

In this type of radar, several (perhaps many) different PRIs may be selected for
use. However, switching among them is done automatically and rapidly to perform
certain radar functions. The parameters of interest include all of those used to
characterize nominally constant PRIs, as well as additional information about how
many PRIs are used, what switching patterns are observed, and how long the
emitter transmits a fixed PRI (i.e., the dwell time). This technique can be used to
resolve range ambiguities (in short PRI pulse Doppler radars) and velocity ambigu-
ities or to eliminate target eclipsing (blind ranges) or blind speeds. Some range-
tracking radars with short PRIs may adjust the PRI to maintain the target return
at a more or less fixed time after the transmitted pulse. PRIs less than about 100
to 125 ␮ s are switched among several values to eliminate eclipsing or resolve range
ambiguities. PRIs in this range are intended to minimize Doppler ambiguities in
pulsed Doppler radar. Having selected such a PRI, it then becomes necessary to
use dwell and switch to eliminate eclipsing and resolve range ambiguities. In other
words, that is the result, not the objective of using short-PRI (usually called high-
PRF) radar. Once the ELINT analyst determines the set of PRIs used, the unambigu-
ous range can be determined using methods described in the radar literature [2].
If the sequence of PRIs used is periodic, that tends to indicate that range ambiguity
elimination or similar purposes are the reasons for the switching. A PRI sequence
that is not periodic but has a pattern tends to indicate a process that adapts to the
target situation. An apparently random sequence could be related to EP.
It is useful to determine the time required to switch PRIs. Any time should be
characterized along with the dwell time for each PRI.
12.2.4 PRI Stagger

PRI stagger is the use of two or more PRIs selected in a fixed sequence. The
sequence may contain more than one of the several intervals before it repeats. The
sequence is described by the number of ‘‘positions’’ used to make up one period
of the sequence and the number of different intervals used to make up one period
of the sequence. A common stagger sequence consists of alternating long and short
PRIs. This has two intervals and two positions. A sequence using the same two
intervals but having a sequence of long, long, short has two intervals and three
positions. Generally, stagger is used to eliminate blind speeds in MTI radar systems
[3]. The use of stagger is a strong indication that MTI capability exists and that
highly accurate PRI measurements are necessary to determine performance limita-
tions due to incidental jitter. Of course, the intervals must be measured with
sufficient accuracy to notice the presence of the stagger. This normally requires
accuracies in the 0.1% range or better.
In the past, manually selected stagger was used to enable a radar operator to
check to see if a target was beyond the unambiguous range. A target ‘‘blip’’ becomes
two blips if stagger is used and if the target is beyond the maximum unambiguous
range. Stable PRIs are not required for this technique. Incidental jitter measurements
usually can be used to distinguish whether this or MTI is the reason for the stagger.
12.2.5 Sliding PRIs

A sliding PRI is characterized by monotonic increases or decreases in the PRI
followed by a rapid switch upon reaching one extreme limit to return to the other
extreme limit. This may be done to eliminate eclipsing (blind ranges) as in some
dwell-switch PRI schemes. But, in this case, the variation is continuous. A sliding
PRI may also be used to optimize elevation scanning by providing constant altitude
coverage. In this application the sliding PRI pattern is usually periodic. The idea
is to decrease the maximum unambiguous range as the elevation angle increases
because targets of interest fly no higher than a certain altitude. Another application
is to maintain a constant SNR in a mapping radar system.
The ratio of the minimum PRI to the maximum PRI is roughly the ratio of the
minimum range to the maximum range in elevation scanning systems. For example,
if the elevation scan covers about 0° to 30° and the altitude of interest is 15 km,
then the minimum range is 30 km; whereas the horizon-limited distance at an
elevation angle of 0° is about 500 km—a range ratio of nearly 17:1. A typical
height-finder range limit is 150 to 200 km for a range ratio of 5 or 6 to 1. The
ELINT analyst should determine the shape of the PRI ‘‘sweep’’ waveform, as well
as the upper and lower limits and the time to complete the sweep. Short PRIs (100
to 125 ␮ s) and more limited PRI variations in an unpredictable pattern may indicate
target tracking in which the PRI is varied to eliminate eclipsing.
12.2.6 Scheduled PRIs

Scheduled PRIs occur in computer-controlled electronic scan radars that interleave
search and track functions, often in three dimensions. The PRI variations can
include some features of all of the other types as determined by the controlling
program. While considerable variability and responsiveness to the environment is
the rule, the complexity of the real-time software is such that wholesale changes
rarely occur. The number of pulse interval patterns seen by an ELINT receiver
depends on the number of targets being tracked and their locations. If no targets
are present, a periodic raster-type search may occur with constant altitude coverage.
It is important to discover the number of tracking pulses used (i.e., how many
targets can be tracked at once) and how frequently the radar illuminates a tracked
target. The ELINT analyst should determine all of the intervals used and some of
the typical sequences. It is very useful to correlate pulse interval values with the
beam angle, target range, or target velocity. Generally, interval accuracies of 0.1%
are adequate. This type of PRI control is the rule for newer multifunction and/or
solid-state radar systems.
12.2.7 Periodic PRI Variations

Periodic PRI modulation is a nearly sinusoidal variation over a more limited range
than sliding PRI. It can also be used to avoid eclipsing or for ranging, but is more
often used in conjunction with conical scan tracking systems as a missile guidance
technique. When conical scan radar tracks a target, its boresight along the center
of the cone is aimed at the target. If a missile is fired along the beam, the conical
scan amplitude modulation will vanish if the missile is headed toward the target.
To make course corrections, the missile needs a reference signal from the conical
scanner. One method to do this is to sinusoidally modulate the radar PRI in
synchronization with the conical scanning.
A deception-type electronic attack or countermeasure to defeat this system
needs information on the sinusoidal PRI modulation. PRI modulation amounts of
5% at rates up to 50 Hz can be expected. Correlation with the conical scan envelope
is a key analysis technique. The average PRI and maximum and minimum values
should be determined to about 0.1%.
12.2.8 Pulse Groups

Radars may transmit groups of closely spaced pulses, possibly as part of a ‘‘sched-
uled’’ sequence, separated by longer times between pulse groups. These PGRIs can
obviously be categorized just as any PRI can be. Pulse groups can be used for
basic radar functions such as increased range and velocity resolution. (The range
resolution is set by the duration of a single pulse within the group, whereas the
Doppler resolution is set by the duration of the whole group.) Pulse groups can
also be used to eliminate blind speeds in MTI systems. (Two pulses per group are
used in the latter application.) Radar applications of pulse groups generally make
use of a fixed pulse group. In other applications (e.g., where telemetry data are to
be transmitted) pulse position modulation can be expected. IFF systems may also
make use of pulse groups that can be changed from time to time, but the pattern
may remain fixed for hours or days. The ELINT analyst should determine the pulse
durations and intervals within the group if the group is fixed. If the pulse patterns
within the group change, the analyst should determine the number of pulses (maxi-
mum and minimum) that can be included and any repeated synchronizing pattern
that may be present. Such patterns are often used for data transmission. Pulse
code modulation and pulse position modulation are representative of typical data
transmission schemes.
12.3 Time Interval Measurements
Measurement of the radar pulse’s TOA at an ELINT receiver is similar to the

problem faced by the radar receiver in determining the range to a target. However,
there are some differences.
1. The radar receiver is matched (more or less) to the pulse being received.
The ELINT receiver is not usually matched to the pulses it receives.
2. The radar range delay uncertainty for a single pulse usually need not be
much smaller than the pulse duration. Also, the target range can be deter-
mined based on the integration of a number of pulses. Some ELINT analyses
may require a very small uncertainty in measuring the TOA of a single
pulse, possibly a fraction of the pulse rise time (e.g., emitter location via
TDOA).
3. In the radar, the time interval of interest is from a known time of the pulse
transmission to the unknown time of the target echo return. The ELINT
parameter of interest is usually the time between two pulses. There is uncer-
tainty about the time of occurrence of the pulse at each end of the interval.
The effect of the ELINT receiver being poorly matched to the radar signal
usually means that the ELINT receiver bandwidth is wider than the reciprocal of
the pulse duration. This reduces the available SNR but also permits the pulse shape
and rise time of the transmitted pulse to be the limiting factor in the timing
measurement. A matched ELINT receiver would produce (approximately) a triangu-
lar video waveform of twice the pulse duration at the base. The ‘‘excessive’’ band-
width of the ELINT receiver is the means by which the TOA of a single pulse can
be determined to a fraction of its rise time. The other important factor is a high SNR
in ELINT, which is sometimes available because of the R −2 versus R −4 propagation
advantage of ELINT.
12.3.1 SNR Limitations

The problem of the TOA uncertainty due to noise has been analyzed thoroughly
for the high SNR situation. If a fixed amplitude threshold is used and a pulse with
a linear leading edge is applied, the rms variation of the measured time of arrival
is [4]
1.25t R
␦T = (12.2)
√2 SNR
where:
␦ T = rms variation of measured TOA due to noise

t R = pulse time from the 10% to 90% points on the leading edge of the
video pulse
SNR = signal-to-noise ratio at the receiver output, assumed much greater than 1
Notice that the average slope of the leading edge of a pulse of amplitude (A)
is 0.8 A /t R . The factor of 1.25 = 1/0.8 is required because the rise time t R is
measured from the 10% point to the 90% point. The amplitude change in time
t R is then 0.8A. Note that the actual slope changes during the rise of the pulse and
is generally steepest at the 50% point. Thus, the actual disturbance of the TOA
due to noise depends on the threshold setting relative to the peak amplitude of the
pulse. (The effects of bandwidth on rise time are discussed in Chapter 11.)
The SNR to be used in (12.4) is the SNR at the receiver output when a strong
signal is present. In other words, the analysis is valid only when the pulse is clearly
distinguishable above the noise. Next, consider that the PRI is the difference between
the times of arrival of two pulses:
PRI i, j = Ti − Tj (12.3)
where:
PRI i, j = pulse repetition interval between pulse i and pulse j

Ti = time of arrival of pulse i
Tj = time of arrival of pulse j
The probability distribution of the values of PRI i, j is obtained by convolving the

distributions for Ti and Tj . For a high SNR, the noise when the signal is present
is a Gaussian random variable. Therefore, the distributions for Ti and Tj are
Gaussian. Because Ti and Tj are separated in time by many times the reciprocal of
the video bandwidth, the receiver noise at time Ti is independent of that at time
Tj . Under these conditions, the probability distribution of PRI i, j is also Gaussian,
with the standard deviation given by
2 2 2
␴ PRI = ␴ Ti + ␴ Tj (12.4)
If the rise times and amplitudes (i.e., SNR) for the two pulses are the same,
then
␴ PRI = √2t R /0.8 √2 SNR = t r /0.8 √SNR (12.5)
Figure 12.2 shows a graph of this relationship between the rms pulse interval
variation and the SNR for various rise times. As shown in Chapter 11, the measured
rise time can be no less than that of the receiver system. For example, if a 2-MHz
video bandwidth is used (typically the predetection bandwidth is then 4 MHz),
the minimum pulse rise time is about 175 ns. If the available SNR is 30 dB, then
the rms variation of the PRI values due to noise would be at least 13.7 ns according
to (12.5). Note that, if the SNR is less than about 12 dB (shown by the shaded
area), Figure 12.2 is not valid.
12.3.2 Limitations Due to Pulse Amplitude Changes

In the usual ELINT situation, the radar beam is moving past the ELINT position
causing pulse-to-pulse amplitude variations. If the TOA of a pulse is determined
by the crossing of a fixed threshold, the nonzero rise time combined with the
amplitude variation causes TOA errors. If the amplitude is increasing, each succes-
sive pulse crosses the threshold a little earlier. If the amplitude is decreasing, the
opposite is true. For a trapezoidal pulse the situation is shown in Figure 12.3. The
change in pulse interval due to a change in pulse amplitude is
Figure 12.2 Pulse interval variation due to noise.
Figure 12.3 Triggering error due to amplitude change when a fixed threshold is used.
1.25A T (t R ) (1 − A1 /A 2 )
t⑀ = (12.6)
A1
where:
t⑀ = pulse interval change due to fixed amplitude threshold

A T = threshold amplitude
A1 , A 2 = amplitude of pulses at beginning and ending of the pulse interval
t R = rise time of pulse (10% to 90% points)
Figure 12.4 shows a graph of t⑀ /t R for various pulse-to-pulse amplitude ratios

in decibels and for several values of A T /A1 . If the first pulse just crosses the
threshold, A T /A1 = 1. If the first pulse is two or five times the threshold,
A T /A1 = 0.5 or 0.2, respectively. For pulses near the threshold, even 1 dB of pulse-
to-pulse amplitude variation produces an apparent PRI variation of more than
15% of the rise time. Reducing this variation in PRI due to pulse amplitude
changes can be accomplished through the use of a more sophisticated thresholding
technique. For example, if the threshold is placed at the 50% amplitude point, the
TOA is not affected by pulse amplitude variations when the leading edge rises
linearly with time. Of course, finding the 50% point implies inserting a delay at
least as long as the rise time of the pulse. If the pulse is digitized, the 50% points
can be found in non-real time. An additional advantage of setting the threshold
to the 50% point is that the slope of the leading edge is increased to (1.0 to
1.5)A/t r , compared to the 0.8A /t r averaged over the leading edge, improving the
accuracy of measurement for a given SNR.
Figure 12.4 Pulse interval change due to pulse amplitude change for a fixed threshold.
A good test of the effectiveness of various thresholding schemes is to first

measure the PRI using a constant amplitude pulse test signal to determine the
residual jitter and then use alternating high- and low-amplitude pulses with a
known rise time and amplitude ratio to measure the jitter again. The extent to
which the measured jitter for the amplitude modulated signal is less than that
expected when using a constant amplitude threshold indicates the effectiveness of
the thresholding technique. Even if the threshold technique is effective in choosing
a fixed amplitude point on the rise of the pulse, pulse-to-pulse shape variations
due to multipath and transmitter anomalies contribute jitter that is not due to the
radar’s timing generator.
12.3.3 Improving Interval Measurements

The jitter due to noise of the output pulse is proportional to the rise time, as shown
in (12.5). Likewise, the jitter due to amplitude variations and imperfect thresholding
is proportional to the rise time, as shown by (12.6). However, the SNR is improved
by reducing the receiver bandwidth, which in turn reduces the jitter but may increase
the rise time. If the SNR is large, the jitter due to noise will be negligible and a
bandwidth sufficient to pass the transmitted pulse without increasing its rise time
should be used. If the bandwidth is narrowed in subsequent processing, TOA
variation due to amplitude changes and noise may become important, even though
the pulses at the receiver output had a very short rise time. The contribution of
the receiver to the rise time of the output pulse is inversely proportional to the
bandwidth, and the noise is directly proportional to the bandwidth. Jitter due to
noise is reduced only in proportion to the square root of the SNR; therefore,
reducing the bandwidth will not always reduce the jitter due to noise. Once the
bandwidth is narrowed to the point at which the rise time becomes inversely
proportional to the bandwidth, further bandwidth reductions actually increase the
jitter due to noise. Using the expression for rise time from Chapter 11 gives
√ 冉冊√
1 0.7 2 kT (NF)B
␴ PRI /t ri ≈ 1+ (12.7)
0.8 B × t ri √S
where t ri is the rise time of the input pulse.
A plot of (12.7) is shown in Figure 12.5. The optimum bandwidth occurs at
B = 0.7/t ri (or B v = 0.35/t ri ).
In processing signals for PRI analysis, frequent use is made of lowpass, notch,
and highpass video filters. As noted above, the lowpass video filter is set for a
minimum bandwidth so that the pulse rise time is not appreciably affected. Highpass
and notch filters are used to eliminate portions of the frequency band where
interference may be found. These include power line frequencies and interference
from other analysis equipment.
Nonlinear processing such as amplitude clipping or limiting and slicing is used
to remove unwanted portions of the signal’s dynamic range For example, if the
threshold is set well above the noise, the amplitude below that could be sliced off
for a better view of the signal of interest. Likewise, if the scanning of the antenna
Figure 12.5 Jitter due to noise as a function of receiver bandwidth × input rise time.
produces very high amplitude pulses as well as other lower amplitude pulses of
interest, the amplitude may be limited or clipped or compressed logarithmically to
more easily view the sidelobes.
At the heart of digital time interval measurement systems is a timing reference
oscillator used to generate a discrete time scale against which the time between
the radar pulses is measured. There are two kinds of limitations. The first is due
to the instabilities of the reference oscillator, which introduces its inaccuracies and
drift into the measured PRI values. The second is the quantization error due to
the noncoherence between the pulse arrival times and the reference oscillator.
Reference oscillators used in ELINT applications are typically the quartz type,
although rubidium gas cell and cesium atomic beam oscillators are sometimes used.
The latter is a primary standard requiring no calibration, whereas quartz and
rubidium oscillators require periodic calibration to a primary standard to achieve
the best long-term accuracy. Quartz crystal oscillators have excellent short-term
stability and are usually used within the other types of frequency references. When
an atomic standard is used, the quartz oscillator is usually phase locked to the
atomic standard. Then the atomic standard determines the long-term stability while
the short-term stability is determined by the quartz oscillator.
The long-term stability of the reference affects drift measurements and the
ability to compare measurements at different times (e.g., hours or days apart). The
short-term stability of the reference oscillator affects the observed pulse interval–to–
pulse interval variations that make up the incidental jitter. If the reference oscillator
can be characterized by its rms fractional frequency variation (or its Allan variance)
over a time interval of one pulse interval,
␴ y (PRI) = (Allan Variance)0.5 (12.8)

then the rms variation of pulse interval measurements of an ideal stable pulse train
would be
␴ PRI = ␴ y (PRI) ⭈ (PRI) (12.9)
Quite often the more important error is the quantization that results when
counting the number of cycles of the reference oscillator between the radar pulses.
PRIs from an ideally stable pulse train measured by counting cycles of a reference
oscillator differ by one count, depending on the relative phase of the radar pulses
and the cycles of the reference oscillator. For a given stable pulse interval and
stable reference, the count can change by only one count for different phases;
however, if a single pulse interval is measured, it is not known whether the count
might increase by one or decrease by one for a different phase relationship. For
this reason, PRI values have an ambiguity range from plus one count to minus one
count even though the quantizing error is at most only one count. Because the
counting rate of digital circuits exceeds hundreds of megahertz, the limitations due
to the available SNR and the amplitude fluctuations are generally the limiting
factors in making PRI measurements for ELINT purposes.
12.3.4 Digital Thresholding

The usual method to determine the TOA of a pulse is by the use of thresholds.
However, high-speed A/D converters make it possible to use various digital signal
processing algorithms to define the TOA. The A/D conversion rate ( fs ) and quanti-
zation error are important factors in determining the accuracy with which the TOA
can be determined. The minimum conversion rate required for reconstructing the
signal is determined by the video bandwidth (or signal rise time). To simplify the
threshold determining process, it would be useful to have a number of samples,
h, within the rise time of the pulse:
fs ≈ h /t r (12.10)
On the other hand, a large number of samples along the rising edge of the
pulse may be beyond the capability of available A/D converters. As long as the
sampling rate is greater than twice the highest frequency present in the video signal,
interpolation between samples (signal reconstruction) can be used to reconstruct
the pulse to interpolate to find the TOA.
The effect of the quantizing error of the A/D converter on the measured TOA
must also be considered. Generally, the number of bits of amplitude accurately
obtained from the converter decreases as the signal bandwidth (and conversion
rate) increases. The reason is that there is a finite duration sampling window during
which the A/D converter acquires the signal level.
The maximum SNR from an A/D converter with a given number of bits of
resolution and for a full scale sine wave input is given by [5]
SNR max = 6 (# bits) + 1.76 dB (12.11)

The significance of (12.11) is that for a given SNR capability, the A/D converter
must have at least that number of bits of resolution. For example, for preserving
a 40-dB SNR through the conversion process, the A/D converter resolution required
is in excess of 6.3 bits. Note that a high-speed low-resolution A/D converter can
achieve a higher SNR by digitally lowpass-filtering the output. If the sampling rate
is higher than required for the signal of interest, some of the noise can be eliminated
by filtering. The improvement in SNR is given by
10 log (F Nyquist /B Filter ) dB
As can be seen, oversampling by a factor of four yields a 6-dB SNR improve-

ment, equivalent to one additional bit of resolution.
Once the signal is in digital form, elaborate thresholding schemes can be envi-
sioned. However, a high-speed resolution A/D converter provides a high data rate.
(For example, a 1-GHz sampling rate at 8 bits per sample produces 109 bytes/sec.)
Either a very large storage area is required or real-time thresholding must be done
to reduce the quantity of data stored. In the latter case, only simple threshold
algorithms are practical. If the computational resources are available, the calcula-
tion of the pulse centroid location can be considered as a definition of the TOA.
A more precise location of the 50% amplitude point could be performed or a
number of other schemes could be implemented. One scheme is as follows:
1. Find the peak amplitude of the pulse.

2. Define the amplitude as the median of the samples above 50% of the peak
amplitude.
3. Interpolate to find the time at which the pulse rise crosses the 50% amplitude
point.
12.4 PRI Analysis Techniques
PRI analysis is concerned with quantitatively describing the variations observed in

the PRI, relating these variations to other aspects of the radar’s signal, and permit-
ting intelligent speculation about their functional use by the radar and the technique
used in generating the PRI.
Historically, a frequently used manual technique for accurately determining
the average PRI is to match the PRI of a synthetically generated pulse train to that
of the unknown signal. The synthesized signal triggers an oscilloscope, and the
signal to be analyzed is applied to the oscilloscope’s vertical input. The sweep
duration is set to be slightly less than the PRI so that the position of a single pulse
on the screen is determined by the difference in time between occurrence of the
trigger and the TOA of the pulse. If the PRI of the synthesized signal is different
from the signal of interest, the horizontal position of the pulse on the screen will
move—to the right if the synthesized PRI is too long and to the left if the synthesized
PRI is too short. The synthesizer is adjusted until the pulse position is fixed. The
synthesized PRI is then accurately measured using a counter. (It is easy to measure
because it is not affected by scanning or by thermal noise.) This same idea is used
in the raster displays discussed below.
The accuracy with which the signal PRI can be determined using this approach
is limited by various factors. Of course, the synthesized PRI can be measured to
very high accuracy, and this was sometimes mistakenly reported as the accuracy
with which the unknown PRI was measured. The actual error involved in matching
the reference PRI to the unknown PRI is greatly reduced compared to that involved
in making a single PRI measurement because a large number of pulses are, in effect,
averaged during the measurement process. First, consider the effects of noise. If
the synthesized PRI exactly matched the unknown PRI, the observer would see a
large number of overlaid pulses with noise and varying amplitudes. The noise
would create some uncertainty about the precise location of the pulse. Usually
there is a limited time available to make the measurement. The observer would be
uncertain as to whether the pulse location moved due to noise disturbing the edges
or whether the synthesizer was improperly adjusted. Thus, the measurement time
and SNR both affect the accuracy of this method. Assume that the noise is the
limiting factor in judging the position of the pulses on the screen. The rms variation
in the location of the pulse due to noise would be approximately given by (12.4).
The observer can make use of more information than the time at which the pulse
crosses a threshold, in which case the error predicted by (12.4) could be the error
over the entire averaging time.
Time-domain analysis techniques (originally based on the oscilloscope) are
among the most useful and widely used. Incidental jitter and some stagger types
can be analyzed with an oscilloscope having a dual time base (A delayed by B).
The delay is needed because the jitter is usually very small compared to the PRI.
A normal oscilloscope display showing a single PRI would permit rough peak-to-
peak jitter measurements down to, for example, 1% of the PRI. With the delayed
time base in use, the full width of the screen can be adjusted to display a small
fraction of the PRI, making it much easier to analyze the extent of the jitter. The
analyst should adjust the delay successively through several multiples of the PRI
to examine the jitter of these intervals. For a random disturbance of a nominally
constant PRI, the jitter should slowly increase as the delay increases. If the jitter
is independent of the delay, the determining factor of the jitter may be related to
threshold crossing variations due to noise rather than the properties of the radar’s
trigger generator.
The same procedure is also useful in analyzing stagger. One common problem
is to distinguish staggered PRIs from discrete random jitter having only a few
possible interval values. Because stagger is periodic, it is easy to tell when the delay
selected is equal to that period [6].
12.4.1 Raster Displays

The raster display is an oscilloscope technique that uses z-axis modulation to
indicate the presence of a pulse. The x-axis is a conventional time sweep. The
y-axis is also time related such that each x-sweep is displaced slightly. The scheme
in the most common use starts with the trace at the top of the screen and each
sweep is displaced downward; hence the name ‘‘falling raster.’’ The raster display
makes it convenient to observe the variations of the PRI over a much larger number
of intervals compared to the conventional oscilloscope display. This type of display
is very useful and permits the analyst to observe at a glance the PRI variations of
hundreds or thousands of pulses at once. With the advent of time base digitizing,
it became possible to generate this type of display digitally. This resulted in the
development of specialized processors dedicated to pulse sorting and deinterleaving
as well as PRI measurements. These became available in the United States in the
late 1970s. These units provided for time adjustment across the screen in steps as
small as 1 ns and storage on the order of 16,000 events. Of course, if a separate
time base digitizer is available to capture the data, a general-purpose computer
and software can be used to create the raster-type displays for analysis. (Software
was available that simulates common PRI variations and interleaving and also
models the functions of these raster display units on the PC [7].)
Nearly all raster display units designed for pulse analysis also display pulse
amplitude versus time. Simultaneously displaying time versus time and amplitude
versus time shows at a glance correlation between the scan and the PRI. Often,
cursors or markers are included to allow selecting those pulses that make up a
portion of the display for further analysis.
There are several ways to control the triggering of raster displays. In one mode,
the signal controls the start of the trace. After a trace is completed, the next trace
occurs only after another pulse is received. In the free-running mode, the next trace
starts immediately after the preceding one. Then trace duration must be adjusted
to be equal to the average PRI to form a vertical line of pulses. In this mode the
average PRI is determined by PRI synthesis. A zoom effect can be created by using
a delay between the end of one trace and the start of the next to magnify the
horizontal distance between pulses.
12.4.2 PRI Sounds

Listening to the sound of the pulse train using a loudspeaker or headphones is the
oldest technique of PRI analysis. It is still useful today. The low duty cycle of radar
signals makes pulse stretch circuitry an important aid. Also, because widely varying
amplitudes are confusing to the listener, constant amplitude pulses may be used.
The simplest technique is to listen to an audio oscillator simultaneously with
the radar pulse train. The analyst matches the tone of the generator to that of the
pulse train by listening for beats as in tuning a musical instrument. An inexperienced
analyst may set the audio oscillator to a harmonic or subharmonic of the PRI, but
this error is rarely made after some practice. The analyst increases the sound level
until the beat note is heard. The frequency of the beat note is equal to the difference
between the frequency of the audio oscillator and the PRF = 1/PRI. The analyst
tunes the audio oscillator until the beat note frequency is zero (i.e., the beat
disappears). The error under the best of circumstances is about ±20 Hz, because
this is the lower limit of human hearing. Additional error can be caused due to
scanning, which makes the beat note more difficult to hear.
Modern ELINT systems are designed to produce audible sounds even when
the PRF is well above the range of human hearing. This is done by nonlinearly
mapping the true PRF to a synthetic PRF sound. For example, PRFs up to, say,
1 kHz are reproduced as they are. Then PRFs from 1 to 200 kHz could be mapped
into the range from 1 to 20 kHz.
12.5 PRI Analysis Theory and Practice
Hardware is available to measure the TOA of individual pulses (or, equivalently,

the time from the previous pulse) on a pulse-to-pulse basis. This may be done using
counters and a high-quality quartz crystal time base or using other digital records
of the signals of interest. The resulting digital TOA values can be analyzed (in non-
real time) using a variety of computer-based techniques. As an example of the kind
of analysis that can be done, consider the problem of estimating the PRI of an
observed sequence of N pulses. The average PRI is usually estimated by dividing
the time between the initial pulse (t 1 ) and the last pulse (t N ) by the number of
pulse intervals:
t − t1
PRI ave = N (12.12)
N−1
(This can also be done using a counter in the period averaging mode.) However,
this calculation makes no use of the TOAs of the intermediate pulses. It might be
better to find the pulse interval of the ideal constant PRI pulse train that most
closely fits the observed TOA values. One scheme would be to find the pulse
interval that minimizes the squared error between the observed TOAs and those
of the idealized pulse train TOAs. The arrival times for the ideal pulse train would
be
to , n = n ⭈ I + ␾ (12.13)
where:
t o , n = arrival time of the nth pulse from the idealized pulse train
n = an integer between 1 and N, the number of pulses observed
␾ = time between the first pulse and the first pulse of the idealized pulse train
I = interval between pulses of idealized pulse train
The squared error is
⑀2 = ∑ (t n − t o , n )2 (12.14)
Substitute (12.13) into (12.14), differentiate ⑀ 2 with respect to I and ␾ . Then

set the results equal to zero, and solve for the value of I, the estimated PRI, and
␾ . With little loss of generality, the first pulse of the idealized pulse train can be
made to coincide with the first actual pulse. Then ␾ is zero and the PRI estimate
is found to be
冤∑ 冥
N
6
I= n ⭈ tn (12.15)
(N − 1)N (N + 1)
n =1
The question of whether (12.12) or (12.15) is better for estimating the PRI
was investigated in [8]. The best estimator depends on the type of disturbance
perturbing the pulse TOAs. In [8], two types of jitter were considered. For noncumu-
lative jitter (NCJ), the TOA of each pulse is perturbed independently from any
other pulse. In other words, one adds a random number to the TOAs in (12.13)
to get the perturbed TOAs. The second case is cumulative jitter (CJ). In the CJ
model, the TOA of the current pulse is based on the TOA of the previous pulse
plus the PRI plus a random number.
For NCJ the equation for modeling the TOA is
to , n = n ⭈ I + ␾ + vn (12.16)
where v n is the jitter due to measurement.

For CJ, the equation for modeling the TOAs is
to , n = to , n − 1 + I + ␾ + un (12.17)
or
n
to , n = n ⭈ I + ␾ + ∑ ui (12.18)
i =0
In [8], the maximum likelihood estimators are shown to be PRI ave for CJ and
I for NCJ. Furthermore, the Cramer-Rao bounds (CRB) for estimating the PRI for
each type of jitter are found and are the same as (12.12) and (12.15). Of course,
it is unlikely that the type of jitter would be known in advance for an ELINT
signal. This leads to the question of what happens when the nonoptimum estimator
is used. The results are summarized in Table 12.2. The mean squared error (MSE)
is the minimum for the estimators given in (12.12) and (12.15).
If the incorrect estimator is chosen, the MSE penalty is of interest. Comparing
the MSE to the CRB gives insight into the best strategy. If the estimator I is used
when CJ is present, the penalty is
Table 12.2 Estimator Performance for Both Jitter

Models
Estimator MSE for NCJ MSE for CJ
2
12␴ 6(N 2 + 1)␴ 2
I
N (N 2 − 1) 5N (N 2 − 1)
2␴ 2 ␴2
PRI ave
2
(N − 2) (N − 1)
12␴ 2 ␴2
CRB
2
N (N − 1) (N − 1)
6(N 2 + 1)
5N (N + 1)
This approaches 6/5 when N is large. On the other hand, if the PRI ave is
computed when there is NCJ present, the penalty is
N
6
This grows in proportion to N. It is easy to see that if there are many pulses
(say, N ∼ 8), it is best to use the NCJ estimator if the type of jitter is not known.
In practice, both CJ and NCJ will be present to some extent.
There are many procedures that could be used to improve the PRI estimates.
For instance, a weighting function based on pulse amplitude or SNR could be
incorporated into the procedure so that those TOAs with better SNRs influence
the estimate more than those with poorer SNRs. This type of analysis falls into
the general category of fitting functional models to the observed data. Similar
techniques can be developed for other functions (such as computing drift rates).
In doing this, the analyst first assumes a functional model for the data and then
uses the computer to calculate the best fit possible to that function and the resulting
error. The analyst’s judgment is still the key ingredient in the process. The dataset
to be used must also be selected carefully. This requires looking at the raw data
through a variety of display techniques.
12.5.1 Statistical Techniques

The TOA sequence can be analyzed through the computation of histograms and
by analyzing the mean and standard deviations. A key choice faced by the analyst
is in determining how much data to include in such computation—the window
through which the data is to be viewed. Histograms and statistics, like the mean
and standard deviations, are not affected by the order of the data. This means that
such statistics are not useful for analyzing PRI variations in which the sequence
of the intervals is important. The variation of statistics with time can be useful but
this depends on computing them for different segments of the data, including
overlapping segments. For example, to analyze drift, the mean PRI could be com-
puted for several segments of the data separated in time.
Histograms are helpful in determining the overall statistics of a PRI sequence.
A histogram is obtained by first dividing the expected parameter range into intervals
(called bins). Then a count is made of the number of occurrences of the parameter
value within each bin. For a random process, as the bin size approaches zero
and as the number of samples approaches infinity, the histogram approaches the
probability distribution of the random process. If the number of samples is very
large but the bin size is fixed, the histogram level in a particular bin is proportional
to the integral of the probability density function over the parameter range included
in that bin. There is a delicate compromise between the amount of data available
and the bin size. If the bin size is too small, then the average number of occurrences
per bin will be very low and the histogram will consist of a large number of empty
bins interspersed with bins with one or two counts. This is not useful to the analyst.
Likewise, if the bin size is too large, all of the data samples fall into one or two
bins, and once again the analyst is unable to see the shape of the probability
distribution. The usual situation requires the generation of several histograms with
different bin sizes. The analyst then decides which bin size is most appropriate for
the quality and quantity of data available.
Figure 12.6 shows a histogram with many points per bin. The probability
distribution is shown as well. The effect caused by reducing the number of pulses
by a factor of 10 and 100 is shown in Figures 12.7 and 12.8. When there is limited
data, wider bins improve reliability but sacrifice resolution, as illustrated in Figure
12.9.
The histogram of a PRI with sinusoidal jitter plus a small amount of random
jitter is shown in Figure 12.10. The typical bucket shape is caused by the sine wave
spending more time near its peak values than it does at its average value. Figures
12.11 and 12.12 are the same signal but the number of pulses is 1/10 and 1/100
of that used in Figure 12.10.
Figure 12.6 Histogram of PRI with random jitter (50,000 pulses).
Figure 12.7 Histogram of PRI with random jitter (5,000 pulses).

Figure 12.8 Histogram of PRI with random jitter (500 pulses).
Figure 12.9 Histogram of PRI with random jitter—bin width increased (500 pulses).
Figure 12.10 Histogram of sinusoidal PRI with random jitter (50,000 pulses).
Figure 12.11 Histogram of sinusoidal PRI with random jitter (5,000 pulses).
Figure 12.12 Histogram of sinusoidal PRI with random jitter (500 pulses).
The histogram for a sliding PRI is shown in Figure 12.13. Note that the top
is not quite flat. This is because at shorter PRI values there are more pulses per
second than at the longer PRI values. Hence, there are more short PRI values than
long ones in a fixed length record. Figure 12.14 shows the effect of adding 100 ns
of random jitter.
This obscures the slope of the histogram. Figure 12.15 shows how 10 times
as much data helps to show the slope of the histogram in spite of the random
jitter. Table 12.3 summarizes the histogram shape characteristics of common PRI
variations.
12.5.2 Delta-T Histogram

The delta-T histogram1 is a histogram of pulse intervals made by considering the
intervals between all pulse pairs (usually some maximum interval is specified). In
1. The author believes that this histogram was first conceived by Dr. Peter J. Knoke in 1964.
Figure 12.13 Histogram of a sliding PRI.
Figure 12.14 Histogram of a sliding PRI (random jitter added).
other words, the intervals from pulse 1 to pulses 2, 3, 4, and so forth are considered,
then the intervals from pulse 2 to pulses 3, 4, 5, and so forth are considered. The
total number of intervals is given by the number of pulse pairs in a set of N pulses,
which is
冉冊N
2
=
N (N − 1)
2
(12.19)
The value of the delta-T histogram is clear if the effects of noise or interference
pulses are considered. By examining the time between all pulse pairs, the true PRI
(and its multiples) will be found and these interval values will accumulate in peaks
in the histogram, whereas other pulse pairs will have intervals distributed across
a number of histogram bins. Formally, the delta-T histogram can be shown to be
equal to the integral over each histogram bin of the autocorrelation of a function
Figure 12.15 Histogram of sliding PRI (10 times as many pulses as in Figure 12.16).
Table 12.3 Typical Pulse Interval Histogram Shapes

Distribution
Signal Type Spikes Flat Bell Convex Trapezoid Ramp
Discrete PRI jitter or stagger X
Scheduled X
Switched X
Gaussian random X
Uniform random X
Nonuniform random X
Sinusoidal X
Sliding X
that consists of impulses located at the pulse TOAs. If the arrival times are denoted
by t n , this function is
f (t) = ∑ ␦ (t − t n ) (12.20)
where ␦ (t) is the unit impulse.

The autocorrelation of f (t) is
∞
h (␶ ) = 冕
−∞
f (t) f (t − ␶ ) dt (12.21)
∞
= 冕冋∑
−∞
n
␦ (t − t n ) 册冋∑k
册
␦ (t − t k − ␶ ) dt (12.22)
The integrand of (12.22) has a value only if t − t n = 0 and t − tk − ␶ = 0

simultaneously. This occurs when t n = t k − ␶ . Thus,
h (␶ ) = ∑ ∑ ␦ (t n − t k − ␶ ) dt (12.23)
n k
The integral of h (␶ ) from ␶ 1 to ␶ 2 is
␶2 ␶2
冕
␶1
h ( ␶ ) d␶ = 冕∑∑
␶1
n k
␦ (t n − t k − ␶ ) d␶ (12.24)
The integrand of (12.24) has a value for any pair of arrival times such that
␶ 1 < tn − tk ≤ ␶ 2 (12.25)
Thus, a count of the number of pulse pairs, such that their arrival times
are separated by an amount between ␶ 1 and ␶ 2 , is equal to the integral of the
autocorrelation function h (␶ ) over the same range of delay values—exactly the
value of the delta-T histogram for the bin from ␶ 1 to ␶ 2 .
As an autocorrelation function, the delta-T histogram emphasizes periodicities.
Figure 12.16 shows the delta-T histogram from a three-position stagger. The inter-
vals are 95, 100, and 105 ␮ s.
For stagger, the Delta-T histogram peak occurs at the so-called stable sum. If
one adds adjacent PRIs, eventually the period of the stagger is reached. It makes
no difference on which pulse the sum starts; the stable sum is reached when the
number of adjacent PRIs added is equal to the number of positions in the stagger
sequence. Because the stable sum and the number of intervals and positions are
often used to recognize radar threats, it is important for the ELINT analyst to
correctly determine the stagger period and sequence.
The delta-T histogram is also useful in analyzing interleaved pulse trains.
Consider a simple case of two signals having nearly the same PRI. Clearly, when
an interval begins on a pulse of one pulse train and ends on a pulse of the other
pulse train, the interval is somewhat random. When the interval begins and ends
on pulses of the same pulse train, the interval is a multiple of one of the true PRIs.
Figure 12.16 Delta-T histogram for three-interval three-position stagger.

Figure 12.17 shows a histogram for interleaved pulses from two emitters that have
5-␮ s (rms) NCJ and average PRIs of 100 and 110 ␮ s. This is a puzzling situation
for most analysis techniques. The histogram shown has a bin size of 1 ␮ s. The
short intervals near the left are clearly noise intervals consisting of those that begin
on a pulse of one pulse train and end shortly thereafter on a pulse of the other
pulse train. Near 100 to 110 ␮ s, the large peak represents the jittered values of
the both true PRIs. There is only one peak instead of two because the mean PRIs
are separated by only the rms jitter times two. Near the 200- to 220-␮ s region,
there is another large peak representing twice the true PRIs. Near the 300- to
330-␮ s region, there are what looks like two peaks. Near the 400- to 440-␮ s
region, there are clearly two peaks. This leads to the conclusion that there are, in
fact, two pulse trains present. Furthermore, the mean PRIs can be found by noting
the positions of the fourth peak and dividing by four. The peaks of the multiple
PRIs separate by more than the increase in the width of the peaks. This illustrates
the fact that the jitter of the multiple intervals tends to increase no faster than the
square root of the number of intervals added, while the peaks separate directly in
proportion to the number of intervals added. If there is only one signal and the
SNR is high (virtually no noise pulses), the delta-T histogram is no more useful
than a regular histogram or time-domain displays of PRI versus time. When more
than one signal is present or many noise pulses are present, the delta-T histogram
can be a valuable aid to the analyst. Using the delta-T histogram as an aid in
deinterleaving is covered in some detail in Chapter 13.
12.6 Interpreting the Results
While the major function of the ELINT analyst is to characterize the PRI waveform
in detail, there is also the need to explain the waveforms in terms of rational
radar design and operation. In terms of radar operation, it is common practice to
synchronize the pulsing of different radar sets that are functionally related. A good
example is a search radar (range and azimuth) controlling a height finder (range
and elevation). When data from one radar is to be displayed with that from another,
it is especially convenient to pulse the radars at the same time. To show the
Figure 12.17 Delta-T histogram for two interleaved and jittered pulse trains (bin size = 1 ␮ s).
synchronization, the analyst can use an oscilloscope triggered by one signal to

observe the other (or a raster display can be used). Sometimes an overall synchroni-
zation exists but with one set operating at a multiple or submultiple of the control-
ling PRI. Sometimes the jitter characteristics of one transmitter’s pulses will differ
from that of the controlling PRI, especially if the controlled set transmits at the
shorter PRI. Sometimes different radar PRIs will be simultaneously collected, possi-
bly at different sites, for later analysis of the timing relationship (if any) among
the signals.
Another relationship of great interest is the correlation of PRI changes to
changes in other parameters (e.g., with pulse duration to maintain a constant duty
cycle or with scan to maintain a constant number of pulses per beamwidth). Of
still greater interest is the correlation of parameter changes with external events.
In addition to such relationships of the PRI to other radars, other parameters,
and events, there is the fundamental relationship of the observed PRI to the circuit
that generated it. After a microscopic look at PRI variations (drift, jitter, and the
waveform used), the ELINT analyst is in a good position to infer what type of
circuitry was the likely source of the PRI. The PRI generators in use as viewed
from the ELINT perspective are described in the following sections.
12.6.1 Delay Line PRI Generators

Delay line PRI control has been used in conjunction with MTI systems that subtract
the target returns from successive pulses to distinguish moving targets (whose echo
changes) from stationary targets (whose echo is fixed). The delay line is used to
store the return from the previous PRI as well as to control the pulse timing. A
pilot pulse enters the delay line at the time the transmitter is pulsed. It emerges
one PRI later; it is used to pulse the transmitter; and it is fed back through the
delay line as the next pilot pulse. The PRI will be as stable as the delay line (usually
very good). The delay line may use an oven to ensure stability. Close examination
of the PRI drift can reveal whether or not an oven is used and whether a proportional
or on/off-type thermostat is used. The fractional PRI stability expected from this
type of generator is in the range of 1 part in 103 to 105. Use can be made of
multiple delay lines to generate stagger.
The analog delay line techniques used were generally acoustic because delays
of 1 ms or longer are often needed for search radars. Mercury columns, quartz
blocks with multiple faces, and surface acoustic wave (SAW) devices are typical
analog delay elements. The base band signals are generally impressed on an IF
before being passed through the delay line. Of course, digital memories are also
used to delay the echoes, followed by digital subtraction in real time.
12.6.2 Crystal Oscillators and Countdown Circuits

Quartz-crystal controlled oscillators have been used for many years to control radar
PRI, and their use is increasing (the venerable SCR-584 used such an oscillator). One
reason is that radar sets need an accurate timing reference for range measurement
so that the oscillator is already available. Because crystal oscillators usually operate
at frequencies well above the normal radar PRF values, digital or other types of
frequency division or ‘‘countdown’’ are needed. Multiple position stagger, pseudo-

random jitter, and PRI switching can be generated easily by programming the
countdown circuitry.
While crystal oscillators are usually thought of as being fixed in frequency,
small frequency changes (less than 0.1%) can be made by varying an adjustable
capacitor in the resonant circuit of the oscillator. Temperature changes also cause
changes in the mean PRI, but long-term drift is usually less than 1 part in 105. An
accurate measurement of the mean PRI of such radar sets would show some
variation from set to set because any group of crystal oscillators exhibits small
differences in the mean frequency among its members.
The use of a crystal oscillator as a timing reference for the range delay often
takes the form of range marks on the radar display. The time between the range
marks spaced by a range interval ⌬R is
⌬T = 2⌬R /c (12.26)
where:
⌬T = time between the range marks (second)

⌬R = range spacing of range marks (meters)
c = speed of light (meters/second)
If each cycle of the crystal oscillator generates one range mark on the display,
the crystal frequency required is
fxtal = c /2⌬R (12.27)
The speed of light in a vacuum, c v is 299.7925 × 106 m/s. In the atmosphere,

the speed of light is reduced slightly. The speed depends on the density of the
atmosphere, which is a function of altitude. A useful approximation to the speed
of light in the atmosphere at an an altitude h is2:
cv
c= (12.28)
1 + 313 ⭈ 10−6 ⭈ e −h/h 0
where:
c = speed of light at altitude h

c v = speed of light in vacuum
h = altitude (meters)
h 0 = reference altitude (≈ 7,000m)
At sea level, h = 0 and the speed of light is approximately
2. The author is grateful to Mr. David Barton for pointing out this relationship.
c(0) ≈ c v /1.000313 (12.29)
For radar set with 10-km range marks, the proper crystal frequency is 14,985
Hz.
The SCR-584 radar had 2,000-yd range marks. The proper crystal frequency
(at sea level) is 81,939 Hz at sea level. The frequency for operation in a vacuum
would be 81,964 Hz. The radar actually was specified to use a crystal frequency
of 81,950 Hz, which would correspond to an assumed average altitude of about
4,110m throughout the radar beam’s propagation path. When the SCR-584 radar
was supplied to the Soviet Union during World War II, a crystal frequency
of 74,919 Hz was substituted to make the range marks 2,000m apart instead of
2,000 yd apart. The crystal frequencies that correspond to common range marker
spacings are given in Table 12.4.
Of course, range marks cannot be directly observed by the ELINT analyst.
However, inside the radar circuitry, the PRI is sometimes obtained by selecting
every Nth range marker pulse. This can be done in several ways, but the result is
that the average pulse interval(s) multiplied by the crystal frequency (hertz) is the
countdown integer N:
N = (PRI) ( fxtal ) (12.30)
When measured values for the PRI are available, and considering the possible
range of crystal frequencies, the product is never exactly an integer. If N is not
too large, it should be possible to determine N exactly. Obviously, if the measure-
ment error of the PRI times the crystal frequency exceeds 0.5, there will be some
uncertainty as to the value of N in use. Also, more than one common range mark
crystal frequency may produce an integer value. It is usual to assume that the
lowest crystal frequency that gives an integer result to (12.30) is correct.
For many radars, several PRIs are used. In this case, assuming that there is
one crystal frequency, it is necessary to satisfy the relationship in (12.30) for all
of the PRIs. This gives a set of countdown integers that are used by the PRI
generator.
Countdown circuitry can be implemented in several ways. Some older sets used
an analog timing circuit to select one of the range marks. Due to variations in the
analog timing, any of several countdown integers might result. In this case, a
Table 12.4 Crystal Frequency for Common Range Marks

Crystal Frequency, Hz
Range Marker Spacing Sea Level Atmosphere Vacuum
0.25 nmi 323,649 323,750
1.0 nmi 80,912 80,938
125m 1,198,795 1,199,170
500m 299,699 299,793
1,000m 149,849 149,896
2,000m 74,925 74,948
2,500m 59,940 59,959
2,000 yards 81,938 81,964
random discrete jitter is observed with the countdown values being a set of succes-
sive integers.
The more precise method is to use digital flip-flop networks. In times when
such circuits were costly, it was common to make up the total countdown from a
set of circuits, each of which divided by a relatively small (often prime) integer. If
the total countdown can be factored into a set of smaller integers, these may
represent the type of circuits in use. For example, a countdown value of 240 might
be factored as 8 × 3 × 10 or 2 × 2 × 2 × 2 × 3 × 5. A model of the countdown
circuitry then consists of frequency division stages corresponding to each factor.
Of course, this is purely speculation on the part of the ELINT analyst. However,
if a PRI is observed (possibly a malfunction) that can be explained by a change in
only one of the factors, more credence can be placed in the speculative model.
Suppose the countdown integer changes from 240 to 216 but no other integer is
observed. This could be explained by a divide-by-10 circuit malfunctioning to
become a divide-by-9 circuit.
For the SCR-584 radar, a countdown of 48 is used, giving a PRI of 585.72 ␮ s
(a PRF of 1,707 pps). The metric range marker crystal of 74,919 Hz would have
had a PRI of 640.69 ␮ s (a PRF of 1,561 pps) if the countdown integer had remained
48. However, the actual countdown value was changed from 48 to 40 to give a
PRI of 533.91 ␮ s (a PRF of 1,873 pps). Countdown integers of 60 and 80 were
also sometimes used in the SCR-584 radar.
The ELINT analyst should be aware that crystal-based PRI generators are very
stable and any shift in PRI is usually in integral multiples of some basic interval.
Finding this basic interval is equivalent to finding the greatest common divisor of
all the observed PRIs. This value may represent the reciprocal of the frequency of
the crystal oscillator that controls the PRI generation.
To see how to find the countdown integers from ELINT data, consider the
following example. A radar pulse train is analyzed and the following intervals are
found:
1. 2,440.8 ␮ s;
2. 2,428.7 ␮ s;
3. 2,465.3 ␮ s;
4. 2,453.1 ␮ s;
5. 2,562.9 ␮ s.
The intervals occur in an unpredictable sequence; therefore, the conclusion can

be drawn that this is discrete random jitter. The next step is to put these values
in order from the smallest to the largest and find the differences as shown in
Table 12.5.
The smallest difference is a candidate for the basic oscillator period from which
all PRIs are derived. All other differences must be integer multiples of the basic
period. In this example, the first three differences are nearly the same value. Bearing
in mind that the resolution of the measurements is 0.1 ␮ s, the best choice is to
average the differences of 12.1, 12.2, and 12.3 ␮ s to get a starting value of 12.2
␮ s. Note that the other difference value is eight times 12.2, which also points to
using this value. Next, find the ratio of the observed intervals to the smallest
Table 12.5 Ordered Intervals

and Their Differences
Ordered Intervals Difference
2,428.7 ␮s
2,440.8 ␮s 12.1 ␮s
2,453.1 ␮s 12.3 ␮s
2,465.3 ␮s 12.2 ␮s
2,562.9 ␮s 97.6 ␮s
difference and determine if this ratio is close to an integer, as shown in Table 12.6.
The average of the PRI values divided by the selected integers in the last column
is a good estimate of the basic clock period; this is 12.20439 ␮ s.
The reciprocal of this is 81.9377 kHz, which is close to the value for a 2,000-
yd range mark from Table 12.4. Of course, it often takes many trial values before
all of the ratios are close to an integer. If the value of 12.1 was selected in this
example, the ratios become the values shown in Table 12.7.
The only problem is that the first column is not very close to an integer, and
the resulting frequency is not very close to any of the common range mark values.
In this example, the measurement resolution (0.1 ␮ s) multiplied by the integers
involved (about 200) is larger than the smallest PRI difference (about 12 ␮ s). This
means that it may be difficult to find the proper basic period even if it is concluded
that a countdown sequence is in use by the radar. It is almost better to assume as
a starting point the standard range mark period closest to the smallest PRI difference
observed. When there is a large amount of data, it is useful to make histograms
of the ratio of the PRI to the assumed basic period. This histogram can show at
a glance how close the ratios are to various integers even if thousands of PRIs are
being analyzed. The histogram also reveals if there are values between the integers
in a pattern, which indicates that the basic oscillator period is actually the assumed
Table 12.6 Ratio of Observed Intervals to the

Smallest Difference Equal to 12.2 ␮ s
PRI/12.2 Nearest Integer PRI/Integer
199.073 199 12.20452 ␮s
200.065 200 12.20400 ␮s
201.073 201 12.20448 ␮s
202.074 202 12.20446 ␮s
210.074 210 12.20429 ␮s
Table 12.7 Ratio of Observed Intervals Integer

to the Smallest Difference of 12.1 ␮ s
PRI/12.1 Nearest Integer PRI/Integer
200.719 201 12.0831 ␮ s
201.719 202 12.0831 ␮ s
202.735 203 12.0842 ␮ s
203.745 204 12.0848 ␮ s
211.810 212 12.0892 ␮ s
12.0849 ␮ s
Average frequency 82.748 kHz
12.8 MTI Radar Blind Speeds 313
value divided by some integer. For example, if the histogram entries cluster near
the integers and halfway between the integers, then divide the assumed basic period
by 2.
It should be noted that radar designers may choose to use two crystal oscillators,
one to control the PRI and the other to create the range marks. In this case, there
is no need for the PRI increments to be related to a common range mark.
12.7 PRI and Range Velocity Ambiguities

As discussed in Chapter 2, in a modern pulse Doppler radar system, once the PRI
sequences have been observed and categorized, it is natural to try to determine
what purpose the sequence may have. One way to do this is to make a plot of the
eclipsed ranges and blind velocities for each PRI and then overlay them to see how
well the set of PRIs covers the expected ranges and velocities. If the PRIs do not
cover the ranges and velocities well, it may be that the sequence has some other
purpose.
The same basic method is used to resolve both range and velocity ambiguities
through the use of different PRIs. An example showing how to resolve range
ambiguity actually illustrates the principles for both. Suppose a radar has a PRI
of 40 ␮ s and there is a target echo with a delay of 12 ␮ s from the last transmitted
pulse. The maximum unambiguous range is 6 km. The indicated range is 1.8 km.
After a few pulses, the PRI is switched to 30 ␮ s and the target echo is observed
at 2 ␮ s. The maximum unambiguous range is now 4.5 km and the indicated range
is 300m. The fact that the indicated range changed when the PRI was switched
shows that the true range is different from the indicated range. A simple trial and
error approach can be used to find the true range. The true range delay is some
integer times 40 ␮ s plus 12 ␮ s. It is also some other integer times 30 ␮ s plus
2 ␮ s. Consider the information in Table 12.8.
The condition is satisfied when both values in the last two columns are equal.
Therefore, the true range delay is 92 ␮ s (or a multiple of 92). For this kind of radar,
the maximum unambiguous range is determined by the least common multiple of
the PRI values—in this case, 120 ␮ s. It should be noted that the use of multiple
PRIs is complicated considerably if there are multiple targets in the same angular
cell.
12.8 MTI Radar Blind Speeds

MTI radar makes no attempt to measure target speed. Instead, the MTI feature
eliminates large stationary echoes from the display. This is done by subtracting the
Table 12.8 Integers Versus Range Delay

Integer 1 Integer 2 Range Delay 1 (␮ s) Range Delay 2 (␮ s)
1 1 52 32
1 2 52 62
2 2 92 62
2 3 92 92
echo phase from one pulse from that of the next pulse. If the phase does not change,
the result is zero and the stationary targets produce nearly the same phase in each
echo. However, if the echo phase changes by 360° (or a multiple thereof) during
one PRI, this also produces nearly complete cancellation. Therefore, the blind
speeds are those for which the target moves radially toward the radar any integer
number of wavelengths during the PRI:
Vb = n ⭈ c /[2(PRI)(RF)] (12.31)
where:
Vb = blind speed (meters/second)

n = an integer
c = speed of light (3 × 108 m/s)

PRI = pulse repetition interval (seconds)
RF = radio frequency (hertz)
For example, a radar operating at 6 GHz with a 2,500-␮ s PRI has blind speeds
at multiples of 10 m/s or 36 km/h. If the radar adds stagger so that a 2,500-␮ s
PRI alternates with a 3,000-␮ s PRI, the blind speeds for this second interval if
used alone would be multiples of 8.333 m/s. The blind speed of a radar with the
average PRI of 2,750 ␮ s would be 9.0909 m/s. The blind speed of the staggered
PRIs is the lowest speed that is some integer times each blind speed—in this case,
50 m/s, which is 5 × 10 and 6 × 8.333. Another way to look at stagger is to reduce
the PRIs to a ratio of integers (i.e., 3,000:2,500 = 6:5). These are the integers that
multiply the blind speeds of the individual PRIs to obtain the blind speeds with
stagger. Alternately, the blind speed for the average PRI can be found and multiplied
by the average of the two integers (in this case 9.0909 × 5.5 = 50 m/s). If the PRIs
are close together, the true blind speed can be made quite large; however, there
will still be deep nulls in the MTI response (dim speeds).
12.9 Moving Target Detection
Moving target detection (MTD) is a combination of MTI and Doppler filtering.

The velocity itself is not measured; additional clutter rejection is achieved by adding
narrowband filters to further bring the signals out of the noise. From the ELINT
viewpoint, the radar has a low PRF but the intervals have a pattern similar to
dwell switch. To permit filtering after the canceller, the PRI must remain constant
for enough intervals to ‘‘fill’’ the canceller and FFT processor. Typically, this may
require 10 PRIs—two for the canceller and eight for the FFT. This allows reducing
the bandwidth of the final filters to the PRF/8. In the FAA ASR radar, a three-
pulse canceller alone achieves a 25-dB improvement factor. When the eight-pulse
FFT is added, the improvement factor increases to 45 dB [9]. For this type of radar,
the number of PRIs with each value should be on the order of 10; however, the
12.9 Moving Target Detection 315
problem of blind speeds is still the same as before. Therefore, there must still be
a sequence of interval values to make sure that all targets are detectable regardless
of their speed.
References
[1] Vyse, B., and H. Levinson, ‘‘The Stability of Magnetrons Under Short Pulse Conditions,’’
IEEE Trans. on Microwave Theory and Techniques, Vol. MTT-29, No. 7, July 1991.
[4] Skolnik, M. I., Introduction to Radar Systems, New York: McGraw-Hill, 1980,
pp. 401–402.
[5] Tsui, J. B.Y., Digital Techniques for Microwave Receivers, Norwood, MA: Artech House,
1995, p. 167.
[6] Wiley, R. G., Electronic Intelligence: The Analysis of Radar Signals, 2nd ed., Norwood,
MA: Artech House, 1993, pp. 179–180.
[7] Szymanski, M. B., and R. G. Wiley, Pulse Analysis Using Personal Computers, Norwood,
MA: Artech House, 1986.
[8] Gray, D. A., B. J. Slocomb, and S. D. Elton, ‘‘Parameter Estimation for Periodic Discrete
Event Processes,’’ IEEE Proc. ICASSP-’94, 1994.
[9] Skolnik, M. I., Introduction to Radar Systems, New York: McGraw-Hill, 1980,
pp. 127–128.
CHAPTER 13
Deinterleaving Pulse Trains
ELINT analysis often involves interleaved pulse trains. This means that several
pulsed signals are present and that successive pulses may not come from the same
emitter. The first problem the analyst faces, then, is to separate the interleaved
pulses into groups of pulses from the same emitter. This process is called deinterleav-
ing. The first step is usually to make use of information derived form individual
pulses, often referred to as pulse descriptor words (PDWs). PDWs consist of para-
metric information such as pulse duration (PD), pulse amplitude (PA), pulse radio
frequency (RF), angle of arrival (AOA), and time of arrival (TOA). Other on-pulse
parameters such as descriptions of intrapulse modulation and polarization may
also be available. If fine frequency measurements are made, the RF parameter may
be thought of as the frequency of arrival (FOA) because the RF originally trans-
mitted has been altered by the Doppler shift due to relative motion of the ELINT
collector and the emitter.
13.1 Pulse Sorting
In many cases, deinterleaving can be accomplished using single pulse parameters

as reported in the PDWs. This is also called pulse sorting. If there is no frequency
agility, RF is a very useful first sort. AOA is a very useful sorting parameter because
it is determined by the emitter’s location and not by its signal design. If there are
no reflected signals to cause confusion, a constant AOA will be present over rather
long periods of time even when the platforms are moving. To have the same AOA,
the emitters must be on the same line of bearing. Pulse duration can also be used;
however, it is a less reliable sorting parameter due to multipath and thresholding
problems associated with changing pulse amplitudes. When PDWs from more than
one collector are available, emitter location also produces fixed values for the
difference of the TOA and/or the difference of the FOA. These can also be used
for pulse sorting once it is determined that the PDW from one collector is describing
the same pulse as the PDW from another collector. Pulse sorting is often done
using histogram techniques; that is, one might choose to determine the distribution
of the PDWs in terms of, say, RF and AOA in a two-dimensional histogram. In
the absence of frequency agility and multipath, the histogram will contain a
single peak at the AOA and RF of each emitter. If the RF and AOA cells are each
1% of the range of values covered by intercept system, there will be ‘‘space’’ for
100 × 100 = 10,000 cells in the histogram. Suppose a space of one or two cells in
317
318 Deinterleaving Pulse Trains
all directions is allowed in order to separate the peaks (a 3 × 3 or 5 × 5 cell box).

This means that the histogram could separate about 10,000/25 = 400 to 10,000/9
= 1,111 emitters—which is probably sufficient for most situations. However, when
several emitters of the same type are received from the same direction with nearly
the same frequency, and/or when frequency agility is present, when RF or AOA
data is missing, or when multipath is present, this kind of sorting can be inadequate.
In the sorting/clustering operation, it may be that the RF, PW, AOA windows
for the emitters overlap. It may be that the overlapping of the clusters causes
problems in the deinterleaver. For example, the cluster boundaries may overlap,
but the actual pulse data may be separated. Consider two overlapping circles. The
pulses from emitter 1 may all lie in circle 1 and those from emitter 2 in circle 2,
but none of them is in the overlap region. Or it may be that all of the pulses from
the two emitters lie in the overlap region of the two circles. If the data can be
actually separated by using smaller parameter windows for RF, PW, and AOA,
the deinterleaving process can be aided by doing this ahead of the deinterleaver.
Use of smaller parameter windows may erroneously break up a cluster into
smaller clusters. This might increase the variance of the parameter estimates later
in the process. Another problem is that by imposing smaller limits on the cluster
boundaries, artificial structure might be introduced that the deinterleaver subse-
quently detects and reports.
By definition, all of the pulses in the cluster have RF values within some
‘‘window.’’ Clearly, the actual RF values can be examined and their spread deter-
mined. If the spread is significantly smaller than the original RF window, the
window width can be reduced to see if the overlap condition is removed. The same
thing can be done with the other parameters of pulse width and AOA. If the overlap
conditions can be removed in the sorting/clustering, then deinterleaving is already
accomplished. Even if the sorting is only partial, deinterleaving should be easier.
Of course, the sorting process has a downside as well. Suppose a cluster is
erroneously divided into subclusters, and as a result two emitters appear where
there should be only one. For example, consider a frequency agile signal where
only a limited number of pulses was intercepted. The underlying distribution might
be uniform, but a few samples might appear to cluster.
The deinterleaver may also create reports of more emitters than are actually
present. For example: a stagger could erroneously be called two or three constant
PRI pulse trains instead of one.
Generally, the clustering window widths would be chosen a priori to get good
clustering results, and these widths would be based on variations expected in the
measured parameter values for actual emitters. The values reported for the cluster
to a subsequent deinterleaver are usually the means of the parameters for those
pulses in the cluster. Going back to the original data values and examining the
distribution of the measured values relative to the original window widths used to
define the cluster is one way to possibly improve the sorting of the pulses.
These sorting processes are key to the operation of Radar Warning Receivers
(RWRs) and Electronic Warfare Support Receivers (ES). These generally make use
of relatively short snapshots of the environment, whereas ELINT operations often
make use of much longer collection periods. Nevertheless, today’s RWRs often
13.2 PRI-Based Gating 319
can make the same quality of measurements as ELINT receivers. The difference is
in the length of time allowed for both collection and signal analysis.
13.2 PRI-Based Gating
One of the earliest (and still useful) deinterleaving techniques is a gating device. A
block diagram is shown in Figure 13.1.
These devices operate by having the first pulse arrive at the input to start a
delay. At the end of the delay, an acceptance interval begins. If the next pulse
arrives during the acceptance interval, it restarts the delay and the process is
repeated. If no pulse is found in the acceptance interval, the next pulse to arrive
restarts the delay and the process is repeated. In this way, if there is a signal present
having a PRI value between the delay and the delay plus the acceptance interval,
the device ‘‘locks on’’ to that pulse train and only those pulses appear at the output.
This process was usually done in real time; however, it may also be applied to
digitized or recorded signals. These devices can be used in series in the reject mode.
(This is similar to the ideas in [1].) In this way, more than one interfering pulse
train can be rejected. In a series connection, only the last device in the chain can
be in the accept mode. The number of rejection devices that can be successfully
used in a series connection is limited. If the fraction of time that the rejection gates
taken together block the input is significant, then the probability increases that
one of the desired pulses will also be blocked. When this happens the deinterleaving
device loses lock and the process must start over. In the worst case, so many of
the desired pulses will be missing that the final deinterleaver cannot lock on to the
desired pulse train at all. Deinterleavers are also available to continue when one
or more pulses are missing.
The acceptance gate width can be selected to allow for uncertainty in the PRI
value or to allow for PRI modulation such as jitter or stagger. The delay interval
Figure 13.1 Simple PRI accept/reject device.

should be selected to allow passing the shortest PRI of the signal of interest, and
the acceptance gate width plus the delay length should be equal to the longest PRI
in the signal of interest. Choosing these values usually requires a preliminary
analysis of some kind. Various start-up methods are also in use in computerized
deinterleaving. More sophisticated tracking devices adapt the values of the accep-
tance gate width and the delay length to match the observed signal values seen at
the output. Such tracking can follow the pulse train of interest even if its PRI is
drifting.
It is important to note that if the sorter is set to any multiple of the PRI, the
sorter will lock on to that multiple. Therefore, it is best to try short PRI values
first.
13.3 Deinterleaving Algorithms
If pulse train data is reduced to digital form (such as PDWs), a variety of computer
algorithms can be used to assist in deinterleaving. At the outset, the analyst should
realize that, in principle, there is insufficient information in the pulse arrival times
alone to be certain about how the pulses should be sorted into groups representing
different emitters. In practice, it is possible to do a good job with TOA data alone
if there are not too many emitters present at once.
Given that such pulse sorting will not be 100% correct and that the actual
emitter configuration that produced a dataset is unknown, how is the analyst to
judge the results of deinterleaving? Suppose that the analyst can try a number of
sorting algorithms, each having a number of variable parameters (e.g., acceptance
gate widths, the number of pulses allowed to be missing, and the range of acceptable
pulse intervals, and so on). Each of these choices of algorithms and parameters
provides a slightly different sorting result. How can the analyst decide which is
the best? This calls for the analyst’s judgment, based on experience. In general,
though, a deinterleaving algorithm should:
1. Assign as many pulses as possible to some emitter;

2. Result in as few different emitters as possible;
3. Have the least PRI variation associated with each emitter.
The end result of the deinterleaving process can be thought of as a model of

the environment that produced the interleaved pulse train. In other words, it is a
theory that is thought to explain the observed data. Generally, the best theory is
one that accounts for (predicts) the most pulses correctly with the least complexity.
For example, it could be said that each pulse came from a separate, independent
transmitter that transmitted one pulse (i.e., the number of pulse trains is equal to
the number of pulses). This would require explaining any periodicities observed
as being due to chance. This model should be rejected as having too much complex-
ity and requiring improbable assumptions. At the other extreme, it could be said
that all of the pulses came from a single emitter that has a very complex PRI, pulse
duration, and pulse amplitude generation sequence. This model accounts for all
of the pulses and has as few emitters as possible, but the PRI, pulse duration, and
13.3 Deinterleaving Algorithms 321
amplitude variations that must be assumed for that emitter are too great and this
model must be rejected. These extreme cases illustrate the need for the three
criteria listed. The key to successful deinterleaving is the analyst’s judgment, both
in selecting algorithms and the parameter limits used, as well as in determining
when the results are realistic. The analyst must also judge when, in spite of the
application of all of the best techniques, the pulses cannot be successfully deinter-
leaved. In other words, it is necessary to know when to quit and move on to a
new analysis task!
In an environment of signals from emitters that scan in angle, the pulses from
an emitter tend to arrive in bursts. Frequently there will be at least a few pulses
that are not interleaved with those from another signal. A good approach is to try
to work with those pulse trains that are at least partially deinterleaved to begin
with. The following algorithm works well in this case.
Step 1. Look for a few pulses separated by equal intervals. The number of
pulses must be at least three, but four or five may be better choices. Based on
typical radar target illumination times, 10 pulses would probably be too many.
The tolerance to which the intervals are required to match must be larger than the
digitizers quantizing noise and also larger than the jitter expected from the radar’s
pulse interval generator’s design.
Step 2. Upon finding such a set of pulses equally spaced pulses, try to extend
the pulse train both forward and backward in time to account for more pulses.
Extension techniques are discussed below. The extension is terminated when a
certain number of ‘‘missing pulses’’ occurs. Those pulses associated into a group
from a single emitter are deleted from the interleaved pulse train and the process
is repeated. Because removal of a set of pulses may result in another pulse train
for which four or five pulses are no longer interleaved with any other pulse train,
the process should be repeated by returning to Step 1. This algorithm is very reliable
in that it rarely locates pulse trains not actually present. However, it may be
ineffective in very dense environments because it is unable to get started.
The method of extending the group of associated pulses can range from very
simple to very complex. Some simple techniques are to compute the average pulse
interval and hypothesize pulses at times spaced by that amount. A window of
tolerance is used that can be based either on the observed variability of the few
available intervals or on the tolerance used initially. Another simple method is to
base the next pulse interval on the immediately adjacent pulse interval. This allows
for the tracking of a drifting pulse interval (but also can cause locking on to the
wrong pulse train). One procedure for estimating the ‘‘correct’’ pulse interval for
a group of pulses is that for the least mean square error. (This technique is discussed
in Chapter 12.) Whatever method is selected, when the algorithm can extend the
associated group of pulses no further, it is fruitful to take all those pulses defined
as being associated, recalculate the predicted pulse locations based on all the data,
and search again to see if any additional pulses are located. The loop terminates
when no pulses are added to the group. Note that successful algorithms of this
type allow for some fraction of pulses to be missing so that the associated groups
are not broken into small segments by occasional missing pulses. In computing the
radar’s pulse interval, the missing pulses must be considered as present.
There will be situations where there are no segments of any pulse train free of
interleaving and the technique just discussed cannot get started. For example,
perhaps 75% of the pulses are successfully deinterleaved, but a residue of 25%
remains. Another start-up technique should then be applied. A useful one is to
select one pulse (possibly near the center of the data being analyzed) and assume
that it is paired with the pulse next to it. The interval between those two pulses
is used in an attempt to project forward and backward in time to see if that is the
interval of a pulse train. If it is, the same pulse association extension techniques
can be employed. If that interval does not work, the same starting pulse is paired
with the next closest pulse and that interval is tried. This process continues until
the trial interval reaches some maximum or until a pulse train burst is found. If
no pulses are associated, another pulse is selected as the starting point. The process
continues until every pulse is either included in a burst or has been tried as a
starting point. By considering every pair of pulses as defining a potential PRI, the
process resembles the delta-T histogram described in Chapter 12. Various methods
of using histogram techniques to obtain PRI values to start the deinterleaving are
described in [2, 3].
Staggered pulse intervals cause two kinds of problems for these pulse-sorting
algorithms. In one case, the acceptance window is so wide that all intervals of the
stagger are accepted into a burst. Now, however, the average pulse interval in the
burst will not be related to any of the stagger values. A technique that may overcome
this difficulty is the following. For each interval in the stagger sequence a different
delay function is available that attempts to locate a pulse within an acceptance
gate (i.e., one staggered PRI interval plus or minus a specified tolerance). If a pulse
is found, the next different delay in the stagger sequence is started. This chaining
process continues until a pulse cannot be found within an acceptance gate. A second
problem occurs if the tolerance window is so small that several constant PRI pulse
trains with a multiple of the average pulse interval are found instead of one staggered
pulse train. This is a case of erroneous deinterleaving. If two bursts overlap in time
and have the same average pulse interval, they may be pulses from the same pulse
train. If the pulses are truly from more than one emitter, their relative phase will
change slowly. One way to detect this is to compute the mean and variance of the
time between the pulses of one pulse train and the pulses of another pulse train.
If they are from a staggered emitter or from emitters that are synchronized in their
pulsing, the variance will be small. Signals having multiple pulse characteristics can
cause similar kinds of problems and require extra attention. Usually, a preliminary
analysis of the data by the analyst can reveal the multiple pulse groups.
13.4 Delta-T Histogram Applied to Deinterleaving
The delta-T histogram described in Chapter 12 can be used to help find intervals
that may be PRIs. These intervals may then be used as starting values for other
algorithms. The more interleaved pulse trains, the more difficult the problem
becomes.
The delta-T histogram can give gives some insight into how many pulse trains
can be successfully deinterleaved. One of the outstanding problems of ELINT is
determining how many pulse trains can be deinterleaved under various conditions.
The following gives one way to make an estimate of the number of pulse trains
that can be deinterleaved. For the purposes of this analysis, assume that all of the
pulse trains have nominally constant PRIs. For a segment of time, T, there are a
total number of pulses, N, present. These are made up of n k pulses from each of
k pulse trains. Each pulse train has a nominal PRF.
The following relationships are needed for this analysis:
K K
N= ∑ n k = T ∑ PRF k
k =1 k =1
T
n k = PRF k T =
PRI k
Next consider that some of the intervals in the histogram are made up of the
difference in the TOA between two pulses from the same pulse train, while others
are made up of the difference in the TOA of pulses from different pulse trains.
Denote the former as pure intervals and the latter as impure intervals. Then the
ratio of the number of pure to impure intervals is given by the purity ratio, PR:
K
∑ n k(n k − 1)
k =1
PR =
冢 ∑ 冣冢 ∑ 冣
K K
nk nk − 1
k =1 k =1
If all of the pulse trains have about the same PRF, the number of pulses from
each will be about the same. For this special case, the purity ratio becomes
n−1 1
PR = ≈ (if Kn Ⰷ 1)
Kn − 1 K
This makes clear the fact that the ratio of pure to impure intervals decreases
as the number of interleaved pulse trains increases.
Next, consider the heights of the histogram bins that result. Suppose the impure
intervals are spread uniformly among all of the histogram bins and the pure intervals
fall into sets of adjacent bins spaced by the PRI. (Usually, the size of the bins in
the histogram is on the order of the incidental jitter of the individual pulse trains.)
The number of bins, B, will be T /b, where b is the width of the individual bins.
The average height of the impure intervals is given by
冤冢 ∑ 冣冥
K K
1 b
2
nk − ∑ n k2 T
k =1 k =1
The pure intervals are placed in sets of a few adjacent bins near multiples of
the PRI. There are n = T /(PRI) sets of such bins. In each set, the number of bins
into which intervals are placed is approximately J/b, where J is the PRI jitter. This
makes the average height of the bins due to the pure intervals equal to
(n k − 1) 冉冊
b
j
Note the fact that the bin height at multiples of the PRI decreases slightly has
been ignored.
For the special (but interesting) case where all the PRIs are equal, the ratio of
the height of the pure to impure intervals in a bin is given by
n−1 T PRI
R= ≈
Kn(Kn − n) 2J 2K 2J
Consider the case illustrated in Figure 13.2. The average height of the first 400
bins (all impure intervals) is 27.7. The height of the tallest peak is 94, and the
observed ratio, R, is then 2.4. Applying the equation above gives a theoretical
value of R of 2.9 when the mean PRI is 115.4 ␮ s and the total jitter is 200 ns.
The bin heights due to the pure intervals should be larger than those of the
impure intervals in order to attach much significance to that bin. It is clear that
the number of pulse trains that can be reasonably deinterleaved using interval-only
techniques is on the order of
K= 冉冊
PRI 1/2
2RJ
In this equation, R is the ratio of pure to impure intervals required in a bin to

achieve sufficient height for the analyst to believe it is a potential PRI value.
Figure 13.2 Delta-T histogram for 10 interleaved pulse trains (bin size = 400 ns).
This relationship is a quantitative expression of the intuitive idea that the ability
to deinterleave pulse trains improves if they have quite stable PRIs. In a typical
application, the typical PRI is 1 ms and there may be jitter on the order of 1 ␮ s.
This analysis indicates that for a ratio R = 3, the number of pulse trains that could
theoretically be distinguished is about 13. On the other hand, for 100 ␮ s of jitter
and R = 2, the number of distinguishable pulse trains may be less than two. Clearly,
the presence of a number of intentionally jittered signals can cause problems for
interval-only deinterleavers.
The use of cumulative difference (CDIF) histograms and sequential difference
(SDIF) histograms in deinterleaving has been described [2, 3]. The first CDIF
consists of a histogram of the first differences of the TOAs. This is examined for
peaks. Peaks will occur if there are regions in the data record where there is no
interleaving present. If no peaks are found, a histogram of the second differences
of the TOAs is added to the histogram of the first differences and the cumulative
histogram is examined for peaks, and so on. (This creates the delta-T histogram
as it accumulates.) The advantage is that a noninterleaved pulse train can be found
immediately. This greatly reduces the number of differences needed to get started
(approximately n instead of n 2 ) and may also reduce the confusion caused by
multiples of the PRIs. The SDIF histogram approach makes use of the first differ-
ences of the TOAs in the same way as the first step of the CDIF approach. However,
the second differences of the TOAs are used to form a separate histogram, rather
than being added to the histogram of the first differences. Each higher order
difference is used to form its own histogram, whose peaks are then used to infer
interval values in conjunction with the results from the other lower order difference
histograms.
The threshold function to be used to determine which histogram bins represent
possible PRIs is considered in [3]. Clearly, if the number of pulses available is
fixed, the number of intervals present is inversely proportional to the length of the
interval in the SDIF histogram. The impure intervals, which represent the time
between the pulses of different pulse trains, are represented as random Poisson
points [3]. This leads to a threshold function of the form,
threshold (PRI) = x(N − d ) e −(PRI)/gB
where N is the total number of pulses, B is the total number of bins, d is the
difference level, and the constants x and g are experimentally determined. Using
simulated data, the exponential threshold is shown to be better than the reciprocal
of the interval or its square root [3]. (The parameter x depends on the percentage
of missed pulses in the data.) In the full delta-r histogram (or CDIF histogram with
all differences considered), the bin height falls slowly as a function of the length
of the time differences—at least in the region of greatest interest, which is a relatively
small integer multiple of the typical PRI.
Another algorithm based on finding the period of any complex pulse train has
been developed that is capable of finding the period of any single pulse train of
the dwell-switch or staggered type [private communication with R. B. Millette,
Delata Consultants, Baie O’urfe, Quebec, Canada, July 1992]. Once this period is
found, one can then predict when the next pulse will occur regardless of the
complexity of the PRI sequence. This algorithm is not designed for deinterleaving
and (so far) does not specify the type of PRI sequence present. However, it can be
used to drive an interval-based deinterleaver once the true period has been found.
One of the major problems with interpreting the delta-T histogram is the peaks
at multiples of the true PRIs. It would be quite helpful to have a method for making
histograms without peaks at multiples of the PRI. One method for doing this is
to add a complex value to the histogram bin for each interval found rather than
to add one count. Dr. Doug Nelson described this technique to the author around
1983. The technique was conceived independently by K. Nishiguchi in Japan at
about the same time [4]. The excellent analysis in [4] is covers nearly all aspects
of this type of histogram. Note that in [4], the complex delta-T histogram is called
the PRI transform.
The complex value to be added to the appropriate histogram bin can be thought
of as the phase of the TOA of the pulse at one end of the interval with respect to
a constant PRI pulse train. The PRI is equal to the particular bin into which that
TOA difference falls. This means that for each pair of pulses, one computes the
difference of their TOAs as usual. However, instead of increasing the count of the
proper histogram bin by unity, it is increased by a complex number given by
exp [ j(phase kn )]
where:
I kn = Tn − Tn − k
phase kn = 2␲ (Tn mod I kn )/I kn
After this complex histogram has been formed, the absolute value of the con-
tents of the bins is examined. Consider the bin contents for a constant PRI. If the
interval is a true PRI and not a multiple, then
Tn = nI + To
phase kn = 2␲ (To )/I = Constant
For intervals that are multiples of the PRI, the phase values are spaced around
the unit circle like spokes in a wheel, with the number of spokes equal to the
integer multiple of the PRI under consideration. When these complex quantities
are added, the result is nearly zero. Therefore, the only bin with a significant
accumulation is that bin corresponding to the true PRI. Figure 13.3 compares the
delta-T histogram with the complex delta-T histogram. The suppression of the
peaks at multiples of the PRIs is clearly shown when there is a small amount of
jitter. Note that the same 10 PRIs were used in Figure 13.3 as were used in Figure
13.2. The dots show the true PRIs.
Next, consider the effect of jitter on this kind of histogram. Once the number
of PRIs considered exceeds I /J, where I is the PRI and J is the jitter, the phase
value will begin taking on random values around the unit circle. Therefore, the
height of the bin will be somewhat limited by the jitter. Without jitter, the height
Figure 13.3 Comparison of the delta-T and complex delta-T histograms.
of the bin representing the fundamental PRI is equal to the number of pulse intervals
which is one less than the number of pulses in the record:
n = (T /I )
With jitter, the bin height is approximately limited to
(I /J)
Consider an example with T = 0.1 second, I = 1 ms, and J = 1 ␮ s. There will

be about 1,000 pulses. With no jitter, the bin height will be 999. With 1 ␮ s of
jitter, the bin height is about the same. If the jitter is increased to 100 ␮ s, the bin
height decreases to less than 10.
The situation modeled in Figure 13.3 resulted an observed noise level of about
13 for the delta-T histogram and about 3.2 for the complex delta-T histogram.
Because the phases of the impure intervals are random, the resulting amplitude of
the sum might be expected to be approximately the square root of the number of
intervals in that bin. The height of the bins where PRIs occur remains about the
same because in the absence of jitter, the phase remains constant. In [5], a Poisson
model is used to compute the expected noise level of the complex histogram. The
result obtained is that the expected noise level (ENL) is given by
√n
2
ENL = b /T
where:
n = the number of pulses in the record

b = histogram bin size
T = duration on the data record
Applying this result to the situation in Figure 13.3 gives a calculated noise level
of 3.7, which compares well to the observed noise level of 3.2.
Figures 13.4 to 13.6 show the effect of increasing jitter on the regular and
complex delta-T histogram. The noise increases and the ability to resolve the peaks
is degraded because the bin width was increased to 0.5 ␮ s. With 5 ␮ s of jitter and
10 emitters near 100 ␮ s PRI, the utility of the histograms is limited.
In [4] an improved complex histogram is described and analyzed in detail. In
it, the fact that jitter causes larger and larger phase errors for intervals far from
the time origin is partially overcome by shifting the time origin according to simple
rules. The second improvement is to use histogram bins that are wide enough to
accommodate the jitter but whose spacing is determined by the desired PRI resolu-
tion. As shown in Figure 13.7, this provides good resolution of PRIs along with
maintaining the harmonic suppression properties of the complex delta-T histogram.
When these kinds of histograms have sufficient data and the peaks are clearly
formed, automated techniques can be used to find PRI values. The following method
can be considered as an example [5].
Step 1. Arrange the histogram bins in order of amplitude from largest to

smallest.
Step 2. Find those bins larger than a threshold. The threshold can be based on
the observed average histogram level.
Step 3. Arrange the PRI values corresponding to the bins above the threshold
from shortest to longest.
Figure 13.4 Effect of jitter on delta-T histograms (jitter = 1 ␮ s).

Step 4. Eliminate adjacent bins.

Step 5. Eliminate PRIs that are more than twice the smallest (these could be
PRI multiples).
These values may be used to start a deinterleaving process or may be considered

directly as the PRIs present in the data. For the example shown in Figure 13.3.
The results can be very good, as shown in Figure 13.8.
Figure 13.7 Complex delta-T histogram: original and improved (PRI values 1, √2, and √5). (From:
[5].  2000 IEEE. Reprinted with permission.)
Figure 13.8 Example of automated peak processing algorithm results.

13.5 The Pulse Train Spectrum 331
13.5 The Pulse Train Spectrum
Another technique available for finding the PRFs in interleaved data is the pulse
train spectrum [6]. This provides a frequency domain view of the interleaved data
and the peaks occur at the PRF = 1/PRI of the pulse trains. One motivation for
this technique is to reduce the computational load of the delta-T histogram type
of auto correlation calculations The number operations required is approximately
equal to the number of pairs of pulses or is approximately equal to the square of
the number of pulses, while the FFT can be computed with n (log n) operations.
In this calculation, the pulse TOAs are converted to phase values equal to
Phase = 2␲ 冉冊
TOA
T
These complex values spaced around the unit circle. Their FFT is computed
and its absolute value is the pulse train spectrum. Figure 13.9 shows the pulse train
spectrum for the same example of 10 interleaved pulse trains used above.
The number of pulses used in this pulse train spectrum is 8,705—about 10
times the number used in the delta-T histograms above. This provides 10-Hz
resolution. For a PRI of 100 ␮ s, this corresponds to a PRI resolution of 100 ns.
The equation below relates PRI resolution to PRF resolution for a constant PRI
and when PRI = 1/PRF.
−d PRF
d PRI =
PRF 2
Figure 13.9 Pulse train spectrum of 10 interleaved pulse trains (PRF resolution = 10 Hz).
To achieve the same resolution as the histograms, it is necessary to use a

relatively long record. Such a long record is unlikely to be available from radars
in a tactical environment. When the record is shortened to the same number of pulses
(870) used in the delta-T histograms, the resolution of the pulse train spectrum is
insufficient. This is shown in Figure 13.10.
13.6 Combining Pulse Bursts
After deinterleaving, many algorithms produce bursts of pulses with the same PRI
but separated in time. One example is pulses from a scanning emitter. The bursts
may consist of the main beam and the stronger sidelobes as the antenna bema
moves past the intercept receiver. It is necessary to group these bursts together into
one emitter pulse train. This can be done using the burst parameters of average
PD, PRI, AOA, and so on. Another example is that of a coherent pulse Doppler
radar with high PRF bursts with different PRIs transmitted during a single coherent
processing interval. Many airborne multimode radars are of this type. In this case,
the PRI is not the same from one CPI to the next. But the bursts are close in time
and probably have approximately the same CPI duration. This is explored in [7].
A typical signal of this type is described in Table 13.1.
The deinterleaver proposed in [7] consists of two stages. The first extracts the
PRI bursts and the second extracts sequences of packets that share a coherent
timing structure. A coherent structure is one in which the event times (in this case,
the packet times) have the same phase with respect to a periodic reference signal.
It is this use of coherence that sets this approach apart from the other in the
references. Clearly, any information known about the structure of the signal should
be used to help deinterleave and identify it, including the coherence of the signal
from CPI to CPI. In [7], the method is used to separate two emitters having the
same PRI sequences and overlapping CPIs.
Figure 13.10 Pulse train spectrum of a shorter record.

13.8 Measuring Deinterleaver Performance 333
Table 13.1 Typical Pulse Doppler Signal

CPI PRI(␮ s) Number of Pulses CPI Duration
1 25.8 121 3.4848
2 45.6 76 3.456
3 32.0 109 3.4880
4 50.4 69 3.4776
5 35.2 99 3.4848
6 48.8 71 3.4648
7 33.6 104 3.4944
8 47.2 74 3.4928
9 30.4 115 3.4960
10 44.0 79 3.4760
Source: [7].
13.7 Raster Displays and Deinterleaving
The raster displays described in Chapter 12 could also be used for deinterleaving.
When the time across the raster horizontally is not equal to the PRI of any signal
present, constant PRI signals form diagonal patterns of straight lines at various
angles in the raster. The use of the Hough transform is suggested in [8] to detect
these straight lines and the angles at which they occur. These angles are related to
the radar PRI. Then the pulses making up each linear pattern can be grouped
together and hence are deinterleaved. Another way to process such data may be
to use the fractional Fourier transform [8].
13.8 Measuring Deinterleaver Performance
There are at least two aspects to deinterleaver performance. One is the processing
complexity: how much processing power is needed to apply the deinterleaver to a
batch of data? This aspect is often addressed analytically in the references. The
second is the accuracy of the deinterleaving results. This aspect is often addressed
by showing the accuracy when the deinterleaver is applied to specific data sets.
Generally, simulated data must be used because in that way the accuracy of the
results can be verified by comparing the known sources of the pulses in a burst
formed by the deinterleaver. Consider the effects of jitter on accuracy. Performance
could be determined by examining each burst it formed. Then one can call correct
those pulses having the most prevalent source in that burst. The erroneous pulses
are the rest in that burst. Now a simple ratio of incorrect pulses to correct pulses
gives the error rate. Several techniques can be applied to the same data and the
technique with the lowest error rate can be determined.
In a fixed gate deinterleaver, such as that described in Section 13.1, the time
tolerance is an important parameter entered to start the process. The score is a
function of the choice of the time tolerance. Clearly, if the time tolerance is less
than the time jitter of the PRI, many pulses will be missed. If the time tolerance
is much larger than the jitter, incorrect pulses may be included in the bursts formed.
For a given scenario, there will be a minimum error rate for some particular value
of the time tolerance. Improvement of performance when additional parameters
are added to deinterleaving process can also be explored in this way.
References
[1] Costas, J. B., ‘‘Residual Spectrum Analysis—A Search and Destroy Approach to Signal
Processing,’’ Technical Information Series No. R80EMH8, Syracuse, NY: General Electric
Company, Military Electronic Systems Operation, Box 4840 CSP, 4-18.
[2] Mardia, H. K., ‘‘New Techniques for Deinterleaving of Repetitive Sequences,’’ IEE Proc.
F, Communication, Radar, and Signal Processing, Vol. 136, No. 4, 1989, pp. 149–154.
[3] Milojevec, D. J., and B. M. Popovic, ‘‘Improved Algorithm for the Deinterleaving of Radar
Pulses,’’ IEE Proc. F, Communications, Radar, and Signal Processing, Vol. 139, No. 1,
February 1992, pp. 98–104.
[4] Nishiguchi, K., and M. Korbyashi, ‘‘Improved Algorithm for Estimating Pulse Repetition
Intervals,’’ IEEE Trans. on Aerospace and Electronic Systems, Vol. 36, No. 2, April 2000.
[5] Frankpitt, B., J. Baras, and A. Tse, ‘‘A New Approach to Deinterleaving for Radar Intercept
Receivers,’’ Proc. of SPIE, Vol. 5077, 2003, pp. 175–186.
[6] Orsi, R., J. Moore, and R. Mahony, ‘‘Interleaved Pulse Train Spectrum Estimation,’’ Intl.
Symp. on Signal Processing and Its Applications, ISSPA, Gold Coast, Australia, August
25–30, 1996.
[7] Perkins, J., and I. Coat, ‘‘Pulse Train Deinterleaving Via the Hough Transform,’’ Proc.
IEEE 1994 Intl. Conf. on Acoustics, Speech, and Signal Processing, Vol. 3, April 1994,
pp. III-197–III-200.
[8] Ray, P. S., ‘‘Defining Signal Descriptors by Fractional Fourier Transform,’’ 5th Intl. Symp.
on Signal Processing and Its Applications ISSPA ’99, Brisbane Australia, August 22–25,
1999, pp. 247–249.
CHAPTER 14
Measurement and Analysis of Carrier

Frequency
14.1 Pulsed Signal Carrier Frequency
The determination of the carrier frequency of a particular pulsed signal is largely

a function of the receiving system. The typical manual receiver has a calibrated
frequency display and a tuning aid (e.g., a spectrum analyzer display or discrimina-
tor), which are used to center the signal in the IF bandpass of the receiver. The
typical automatic system searches the band with certain center frequencies and
receiver bandwidths specified in advance. In that case, the signal may not be
centered in the receiver bandwidth and the tuning steps may be small enough so
that if the signal is near the band edge on one frequency setting, it will be near
the center of the band on the next setting. Then the processing must recognize the
fact that the same signal (with distortion) has been received on adjacent tuning
positions. The receiver reports the center frequency indicated by the local oscillator
setting. The actual RF for a signal requires determining the offset from the center
of the band using frequency discrimination techniques. A manual process may
make use of a signal generator and frequency counter. The operator attempts to
match the frequency of the signal generator to that of the incoming signal. This
procedure can give rise to reported frequency values with many significant digits
(e.g., 1-Hz resolution). However, the accuracy of the reported value is subject to
many variables, including:
1. The accuracy of the available frequency standard (e.g., the counter time
base or synthesizer reference oscillator calibration);
2. The available SNR for signal pulses and the ability of the receiver system
to integrate a number of pulses;
3. Doppler shifts due to the motion of the emitter or the receiver.
14.1.1 Frequency Measurement Accuracies

Counter or synthesizer-based measurement systems now in common use are gener-
ally accurate to less than 1 part in 106 to 107 per month. At 10 GHz, such a frequency
measurement would be accurate to 1 to 10 kHz. (In the past, the calibration of
mechanically tuned receivers was accurate to less than 1% or 2% of the band
covered, or 10 to 20 MHz at 1 GHz.) Note that accurately determining the exact
335
336 Measurement and Analysis of Carrier Frequency
center of the receiver bandpass is only the first step in measuring the signal fre-
quency. Analysis of predetection data or discriminator outputs is required to find
the true frequency.
Excellent measurements of pulsed signal frequencies (if the SNR is sufficient)
are achieved by accumulating pulses until the total duration of the pulses is equal
to a selected time. For example, if 1-␮ s pulses occur at a PRI of 1,000 ␮ s and the
time selected is 1 ms, it will take 1,000 pulses to fill the 1-ms time requirement,
this will take 1 second of real time if every radar pulse is received. The accuracy
of RF measurements made by counting the total number of cycles in the accumulated
pulses over the selected time is approximately:
1
␴ pul sin g = (14.1)
PD √NP
where:
PD = pulse duration
NP = number of pulses or selected time per PD
For this example,
1
␴ pulse = = 3.16 × 104 Hz (14.2)
−6
√10
3
10
For a scanning radar at 12 seconds per revolution and 40 pulses per beamwidth,
accumulating 1,000 pulses requires 25 scans or 5 minutes of data. Of course, if
such accuracy is not needed, a shorter time may be selected. Often, the effective
pulse duration must be reduced by 30 to 40 ns from its observed value to account
for cycles lost at the leading and trailing edge. This becomes an important consider-
ation for short pulses.
Achieving this accuracy also requires a reasonably good SNR so that correct
cycle counting can occur. The number of independent noise samples per second is
approximately twice the receiver bandwidth. The time the receiver is susceptible
to noise triggering is the time required to accumulate the number of pulses needed.
If the probability that each pulse triggers the counter is PD and the probability
that a single noise sample triggers the counter is P FA , then the probability of
detecting that number NP of consecutive pulses is
PD , NP = (PD )NP (14.3)
Note that this requires constant signal amplitude over NP pulses.

Likewise, the probability of detecting no false triggers in the time required for
NP pulses is
P FA , BT = 1 − (1 − P FA )BT (14.4)
where B is the noise bandwidth and T is the time over which the pulses occur.
14.1 Pulsed Signal Carrier Frequency 337
For example, the required single pulse SNR can be estimated from PD and P FA
for NP = 1,000, B = 20 MHz, T = 1 second, PP, D = 0.9, and P FA, BT = 10−4; then,
PD = 0.99989
and
P FA = 5(10−12 )
Then using tabulated functions for PD , P FA , and SNR, it is clear that the
required single pulse SNR is about 18 dB for this example.
14.1.2 Doppler Shifts

All measured frequency values are affected slightly by the relative motion between
the emitter and the intercept receiver. The Doppler shift in the RF is given by
VR
fD = f (14.5)
c C
where:
fe = carrier frequency
VR = radial velocity (positive for the movement of the transmitter and receiver
toward each other; negative if they are moving apart)
c = speed of light
f D = Doppler shift
At 300-m/s radial velocity, the Doppler shift is about 1 part in 106 or 1,000 Hz/
GHz. Obviously, this is not important for nominal RF carrier frequency reporting
purposes. However, it becomes an important parameter in keeping a reference
oscillator phase locked to the receiver signal. This is done when using the differential
Doppler shift (or FDOA) between receivers at different locations and different
motions to help determine flight paths and trajectories of moving emitters, as
discussed in Chapter 6. Doppler shifts also limit efforts to measure the coherence
or short-term stability of signals.
14.1.3 Drift Measurement

Any RF measurements must be made over some time period. Frequency changes
that occur over time periods that are considerably longer than the measurement
time are considered drift. The analyst may infer drift from reported RF values. For
example, if an RF value is reported for the same emitter on successive hours, the
drift may be inferred as the difference in RF values divided by the elapsed time
between the intercepts. Generally, drifts over time periods of minutes to hours are
of most interest in helping to determine the transmitter type.
Notice that very small drifts over very short intervals may be impossible to
measure. Determining a small frequency change requires very accurate frequency
measurements. For a pulsed signal from a scanning emitter, measurements to
accuracies of tens of kilohertz may require several minutes of data. Obviously, a
drift of a few kilohertz per minute could not be measured unless it continued for
a long time in the same direction.
Finally, it may not be possible to distinguish small drifts from Doppler shifts
without information about the transmitter’s motion.
14.1.4 FM Ranging in Radar

Intentionally changing the frequency slightly from pulse to pulse is a common
technique used in high PRF radar to determine range. The frequency shifts used
are such that the difference frequency between the transmitted frequency and the
received echoes is less than the PRF. The radar determines range by using knowledge
of the frequency versus time slope and by measuring this difference frequency. The
FM ranging computation must be competed during the time the radar beam looks
at the target—generally a few milliseconds. This type of pulse-to-pulse frequency
change can be observed by ELINT using the same techniques as for drift and
Doppler shift.
Note that the SNR may be too low to allow measurement of the frequency of
each and every pulse with sufficient accuracy to determine the characteristics of
the FM ranging waveform. It may be necessary to process in a manner similar to
the radar by multiplying delayed and undelayed versions of the signal together to
obtain the difference frequency and then determining the difference frequency by
spectral analysis. The delay used is nominally a multiple of the PRI so that the
delayed and undelayed pulses are aligned. Obviously, a delay of a multiple of the
PRI plus half of the PRI would not normally result in any overlapping pulses in
the product. This is not as good as the processing in the radar, where the undelayed
version can be noise free. The ELINT processor must multiply two noisy signals
together.
14.2 Intrapulse Frequency or Phase Modulation
As discussed in Chapter 11, intentional frequency modulation on pulse can indicate

the pulse-compression capabilities of that type of radar. Unintentional FMOP
can indicate the type of radar transmitter in use. Wideband radar signals can be
demodulated at the collection site or, if high-speed digitization is used, demodula-
tion may occur during subsequent analysis.
Historical Note. In the past, the analyst obtained this type of data in the form
of wideband tape recordings of the output of a discriminator at the intercept site.
Signals that did not have such wide bandwidths were recorded using predetection
techniques. These two kinds of recordings are quite different. If the intercept site
recorded the output of a frequency discriminator, the analyst must deal with a
postdetection recording. In this case, at least three tracks were generally recorded:
1. Track 1, with an AM detector or video signal;

2. Track 2, with a precision time reference tone;
3. Track 3, with a discriminator output (FM recording process required).
Ideally, the tape included a calibration signal for both the AM detector (e.g.,
the amplitude versus the intercept signal level in 10-dB steps) and the FM detector
(e.g., the discriminator output versus the frequency shift in convenient frequency
steps). To preserve the precise amplitude of the discriminator output, an FM tape
recording process was used if sufficient FM recording bandwidth was available.
The AM track was necessary to determine the times at which the discriminator
output is valid. The calibration run indicated the time delay between the AM signal
and the discriminator output. Because a limiter-discriminator has an output even
if there is no signal, the AM video signal was used to gate the FM video signal
prior to analysis. This was often done visually using a dual-trace oscilloscope or
an actual gating device was used. The AM signal was also used to control the
oscilloscope triggering and the z-axis intensification input of the oscilloscope.
For determining pulse-to-pulse variation such as drift, the average discriminator
amplitude for each pulse was traced on the oscilloscope in much the same way an
antenna scan pattern was made. The discriminator output was observed using a
very slow sweep speed and the AM output was used for z-axis intensification.
As noted in Chapter 11 and in Appendix B, the ability of a frequency discrimina-

tor to indicate an instantaneous frequency is a function of the RF band covered,
the output FM video bandwidth, and the available SNR. The standard deviation
of a frequency discriminator’s noise output (measured in hertz) is (B.50):
√
1 BV
␴ f (rms) = B V (14.6)
3 SNR B R
where:
␴ F = standard deviation of the discriminator output (Hz) = discriminator

rms frequency noise, fn
B V = discriminator video output bandwidth (Hz) = maximum video
frequency, f V
B R = discriminator input RF bandwidth (Hz) = input noise bandwidth, B n
SNR = available input signal-to-noise ratio
As an example, consider a 40-MHz discriminator input bandwidth and a

5-MHz video or recording bandwidth. At an SNR of 20 dB (100 to 1), the discrimi-
nator noise will have a standard deviation of at least 105 kHz (see Appendix B).
For determining FMOP parameters, the same limitations apply. Intentional
FMOP would have a large frequency deviation and the limitation due to noise
would be less important. The extent of the frequency variation on the pulse is
usually a good indicator of the bandwidth of the RF spectrum. Some care must
be exercised in making this bandwidth estimate because the bandwidth of an FM

signal also depends on the waveform of the modulating signal. For instance, if a
carrier is frequency modulated by a sine wave, the bandwidth of the spectrum is
approximately the peak-to-peak frequency deviation or twice the frequency of the
modulating sine wave, whichever is larger. (Also see the discussion of Carson’s
rule in Appendix B.) For FMOP used for pulse compression, a linear FM or
some similar waveform is used. In this case, the linear modulating function has a
bandwidth comparable to the reciprocal of the pulse duration; however, the band-
width of the resulting linear FM will be much larger—tens or hundreds of times
the reciprocal of the pulse duration. Achieving a large time bandwidth product
requires that the frequency deviation be the dominant factor in determining the
overall width of the radar spectrum. Woodward’s theorem [1] then requires that
the RF spectrum shape be approximately the probability density function of the
modulating waveform. For wideband linear FM, the RF spectrum is approximately
uniform between the limits set by the peak deviation. For FMOP, the analyst tries
to determine the shape of the modulating function and the extent of the FM.
Phase reversal signals or other phase modulation on pulse waveforms can also
be analyzed using the output of a frequency discriminator. The instantaneous
frequency of a signal is the derivative (with respect to time) of the instantaneous
phase. Therefore, a phase reversal type of signal will produce a series of positive
and negative impulses at the output of a frequency discriminator. The impulses
will represent the impulse response of the frequency discriminator and their duration
will be approximately the reciprocal of the video bandwidth. The chip rate can be
determined by noting the separation of those impulses closest together in time.
The 3-dB bandwidth of a BPSK-coded signal is approximately equal to the chip
rate. This is illustrated in Figure 14.1. If the phase shifts are other than 180°, the
amplitudes of the impulses may show this. A quadraphase signal should show four
discrete, equally separated impulse amplitudes.
Figure 14.1 Discriminator response to a biphase modulation signal.

An integrator may be beneficial in analyzing the output of a frequency discrimi-

nator when dealing with phase-modulated signals. The integrator should be reset
after the end of each pulse.
Historical Note. Analog predetection recordings required heterodyning the signal

down to a low intermediate frequency so that it could be recorded directly onto
tape. This was done using a single tape channel with the IF centered in the tape
recorder bandpass. For example, if the tape recorder could handle a 5-MHz band-
width, the selected IF would be 2.5 MHz and the signal band could extend
±2.5 MHz on each side of the IF. Practical limitations prevented the use of the full
bandwidth of the tape recorder. For a tape recorder bandwidth of B, a good rule
of thumb was that the center frequency of the predetection recording should be at
0.6B instead of 0.5B and that the signal band should extend from 0.2B to B for
a total bandwidth of 0.8B instead of B. The tapes had the precise center frequency
used in making the recording marked on the accompanying log sheet. It was
generally thought that about five cycles or more of the IF should be available from
each pulse for predetection recording. This meant, for example, that for a 5-␮ s
pulse, at least a 1-MHz predetection center frequency should be used. Another
predetection recording process was to use a zero frequency IF and provide two
channels of recording: one for the in-phase (I) component and the other for the
quadrature (Q) component. These processes are shown in Figure 14.2.
The single-channel approach is simpler compared to the I and Q approach,

but the total bandwidth required of the single channel is twice that required for
Figure 14.2 Predetection processing: (a) low IF process, and (b) zero-IF I and Q process.
each of the I and Q channels. Both processes can make use of digital recording.
An A/D conversion rate of at least 2B is required for each channel. For pulsed signals,
a high-speed A/D conversion can be followed by a low-speed digital recording as
long as the average digital recording rate is in excess of the peak A/D conversion
rate of all channels added together and then multiplied by the duty factor (PD/
PRI) of the pulsed signal.
14.2.1 Analysis of Predetection Data

If single-channel predetection recording data is available, it must be demodulated
to proceed with the analysis. An envelope detector is used to produce the AM
video pulses. This AM detection can be done using a conventional envelope detector
and a lowpass filter or by squaring the digital samples1 and then digitally smoothing
them using a lowpass filter. The lowpass filter must have a cutoff frequency below
B/2.
Frequency demodulation of the channel data can be accomplished using any
of several frequency discriminator approaches. One good discriminator makes use
of the zero crossings of the signal. A standard amplitude and duration pulse is
generated at each zero crossing and then this pulse train is lowpass filtered at a
frequency of less than B/2. The result is a voltage proportional to the frequency.
Accurate demodulation over the full band of zero to B is possible using the zero
crossings. The time between the zero crossings is carefully measured. The reciprocal
of twice this time represents the average frequency during that interval. This stair-
case-like waveform is often a sufficiently accurate demodulation for most purposes.
If large changes in the average frequency occur from one pair of zero crossings to the
next, reconstructing the true instantaneous frequency may require more processing
using an iteration [2]. Another FM demodulation approach is to convert the single-
channel data to I and Q form using the process shown in Figure 14.2(b), with
f1 = B/2 and the filter cutoff frequency equal to B/2.
If two-channel I and Q data is available, the AM and FM can be found using
the techniques discussed in Chapter 12.
Note that the instantaneous phase can be observed on an x-y display on which
the I signal drives the x-axis and the Q signal drives the y-axis. The angle of the
displayed spot at each instant represents the instantaneous phase of the signal.
14.3 Coherence (Short-Term RF Stability)
The term ‘‘coherence’’ implies an ability to predict the future waveform of a signal
based on a past observation of it. In other words, if the RF and phase of a carrier
frequency can be determined during one period of time, an ideal coherent waveform
would be completely predictable for all times in the future. In practice, one cannot
1. Care must be exercised in performing nonlinear processing on digitized data to avoid aliasing. Squaring
the samples will produce frequency components at twice the frequencies present before the squaring
operation. If these components exceed the Nyquist rate, they will be aliased into the band of interest. The
sampling rate must be increased prior to the nonlinear operation and can be decreased after filtering has
ensured that the Nyquist criterion has once again been met at the original sampling rate.
determine the frequency and phase with complete accuracy. In addition, these
quantities vary with time and ambient conditions. Therefore, complete coherence
is an abstraction that never exists in practice. Curry [3] was one of the first to
suggest a quantitative (and elegant) measure for short-term stability.
Coherence is a very important property of some radar signals. This is true for
radars designed to measure or to discriminate among targets on the basis of target
velocity. The reflected signal (as Doppler shifted by the target’s motion) must be
analyzed by the radar. If the transmitted signal is not coherent (stable), its spectrum
will have sidebands of energy near the carrier frequency. These sidebands, when
reflected by large stationary targets can produce strong returns that appear in the
same part of the frequency spectrum as that of the Doppler-shifted target echo
signal.
As an example, consider a radar observing an aircraft flying near a mountain.
This mountain has a huge radar cross-section and reflects a large signal back to
the receiver. The aircraft has a small radar cross-section, but its radial motion
produces a frequency shift of the reflected signal. If the transmitted spectrum is
down 40 dB (at the frequency separated from the carrier that is equal to the Doppler
shift), then the smallest radar cross-section moving target that can be detected by
that radar will be one whose radar cross section is 1/10,000 that of the mountain
range. In practice, it is larger than this by the required output SNR.
Coherence is another way of saying that the signal frequency must be like an
ideal sinusoid with a constant frequency (or linear phase). Therefore, to determine
the degree of coherence, one must determine the degree to which the signal frequency
fluctuates over the time periods of importance to the operation of the system
transmitting the signal. For a radar, this time is that for which the target echoes
are integrated coherently—typically no more than the time that the beam dwells
on (or illuminates) a single target. Because these times are on the order of millisec-
onds, the frequency stability to be measured is called short-term stability. (Long-
term stability could extend over days, months, or years.) Those using frequency
stability measures realized that there were two kinds of measurements needed for
different applications and situations. On the one hand, frequency stability could
be determined by examining the narrowness of the Fourier frequency spectrum
and its sidebands. On the other hand, it could be expressed in terms of the variation
of the instantaneous frequency as a function of time. Note that the word ‘‘fre-
quency’’ is used in two different ways: one is the independent variable in the
frequency domain, which is independent of time (the Fourier frequency); and the
other is the time-dependent frequency of the oscillator whose stability is to be
measured. Since the 1960s, these stability measures have been developed in theory
and applied in practice [4].
These two frequency stability measures are referred to as frequency-domain
measures and time-domain measures. Attempts were made to use the variance of
the instantaneous frequency as the time-domain measure of stability; however, for
real oscillators, as the measurement time decreases, the estimate of the variance
increases without a limit. Therefore, it was necessary to introduce measurements
that are based on the changes in the average of the instantaneous frequency over
some averaging time.
One of the earliest ways to measure frequency stability was spectrum analysis.
To see how the frequency spectrum relates to the fractional frequency stability,
note that (from FM theory) for a small sinusoidal frequency modulation, the ratio
of the carrier level to the sidelobe level is related to the index of modulation by
⌬ f VSB
␤= = (14.7)
fm VC
where:
␤ = index of modulation
⌬ f = peak frequency deviation
fm = modulation frequency
VSB = spectrum sidelobe level (V)
VC = carrier level in spectrum (V)
If the fractional frequency variation is of interest, one can obtain a high-

resolution spectrum of the signal and measure the sidelobe to carrier ratio at a
given frequency fm away from the carrier. Then the fractional frequency stability
due to disturbances in a bandwidth equal to the spectrum analyzer resolution at
a frequency fm away from the carrier is
⌬ f VSB fm
= (14.8)
fc Vc f c
The fractional stability is used because it is independent of frequency multiplica-

tion or division.
For example, if the spectrum level of a sideband 400 Hz away from the carrier
is 40 dB below that of a 4-GHz carrier, then the fractional frequency stability is
⌬ f 0.01 400
= = 10−8 (14.9)
fc 1 4 ⭈ 109
There are several problems when trying to apply this to pulsed signals. Because
of the short duration of the pulses (typically 1 to 10 ␮ s), the spectrum is rather
wide (the order of megahertz). The separation of the PRF lines is typically less
than 1 kHz. Thus, there may be several hundred spectral lines in the vicinity of
the carrier line and all of them have nearly the same amplitude. Spectral lines other
than the carrier line are affected by both PRI jitter and RF instability; therefore,
it is necessary to locate the carrier line if there is significant PRI jitter. The portion
of the spectrum of interest extends on each side of the carrier line out to one-half
of the PRF. RF disturbances that occur more rapidly than one-half of the PRF
cannot be distinguished from slower ones due to aliasing. This is caused by the
fact that the RF carrier is, in effect, sampled at the rate of the PRF.
For a pulsed signal, disturbances of interest all have modulating frequencies

in the range 0 < fm < PRF/2. Regardless of what measurement processes are used,
disturbances that occur more rapidly will be identified as disturbances at rates less
than one-half of the PRF.
This discussion of spectral analysis shows that the character of the power
spectrum is related to the time-varying frequency of the sinusoid. In the discussion
that follows, this relationship is explored in detail for three time-domain measures
of frequency stability:
1. The rms differences of the phase values measured at intervals ␶ apart;

2. The rms differences of the frequency averaged over intervals of duration D
spaced ␶ apart2;
3. The average fraction of the signal remaining when one segment is subtracted
from another segment separated by an interval ␶ .3
While these measures offer different views of the frequency stability, they are
related mathematically to each other and to the frequency-domain measure, which
is the spectral density of the fractional frequency variations. Note that the desired
spectrum is that of the variations in the fractional frequency. This is not the same
as the power spectrum, which also includes the effects of changes in the signal
amplitude. If the variations in signal amplitude are negligible or somehow elimi-
nated, then the power spectrum and the spectrum of the frequency variations are
related.
Although the recommended stability measures are the Allan variance in the
time domain and the spectrum of the fractional frequency variations in the frequency
domain, other measures (as previously listed) plus the frequency-domain measure
of the carrier to the sideband ratio, are sometimes used in ELINT. This is the
reason for including them here.
In the discussion that follows, the relationships between the measures are
obtained by using the autocorrelation function of the instantaneous phase. This is
a mathematical convenience and is only a means to allow the relationships to be
derived. It is not necessary to actually measure the autocorrelation of the phase.
The following important symbols are used:
␾ (t) = instantaneous phase variation in excess of the ideal phase, which is

2␲ times the carrier frequency times the elapsed time
R␾ (␶ ) = autocorrelation of the phase variation ␾ (␶ ) at a lag of ␶
␴ x2 = variance of the subscripted variable

S x ( f ) = spectral density of the subscripted variable
f = Fourier frequency
2. When D = ␶ , this is the Allan variance, named for D. W. Allan of the National Bureau of Standards [5].
This is the recommended time-domain measure.
3. The subtraction of one pulse from another and comparing the residue energy to the energy of one pulse
has been called the cancellation ratio. This ELINT usage is not the same as the use of this term in MTI
radar.
For example,
␴␾2 = variance of ␾ (t)

␴⌬2␾ (␶ ) = variance of ␾ (t) − ␾ (t − ␶ ) = ⌬␾
␴⌬22␾ (␶ ) = variance of ⌬␾ (t) − ⌬␾ (t − ␶ )
␴⌬2 f (D, ␶ 0 ) = variance of the difference between the average frequencies
measured for a duration D and separated by To
␴y2 (␶ 0 ) = Allan variance = ␴⌬2 f (␶ 0 , ␶ 0 ) divided by the average or carrier

frequency.
14.3.1 RMS Phase Fluctuation

If a measurement of a sinusoid phase at specific times is made, measures of the
stability of the sinusoid can be derived. For example, the differences in the phases
at times separated by T can be measured and the variance of these differences can
be used as a stability measurement.
Let the total phase of the sinusoid be represented by
␪ (t) = ␻ t + ␾ (t) (14.10)
where ␻ is 2␲ times the nominal carrier frequency and ␾ (t) is the variation of the
phase.
The autocorrelation of the phase disturbance ␾ (t) is given by
∞
R ␾ (␶ ) ≡ 冕
−∞
␾ (t) ␾ (t − ␶ ) dt (14.11)
Note that
∞
R␾ (0) ≡ 冕
−∞
␾ 2(t) dt = ␴␾2 (14.12)
Useful stability measures can be obtained by sampling the phase variation

periodically. Suppose the phase is sampled at particular times separated by ␶ 0 .
Then sequence of values obtained is
␾ (0), ␾ (␶ 0 ), ␾ (2␶ 0 ), . . . , ␾ (n␶ 0 ) (14.13)
The first differences of this sequence are
␾ (␶ 0 ) − ␾ (0), ␾ (2␶ 0 ) − ␾ (␶ 0 ), . . . , ␾ (n␶ 0 ) − ␾ [(n − 1)␶ 0 ], . . . (14.14)

The variance of this sequence expressed in terms of the autocorrelation function

is
N
1
␴⌬2␾ (␶ 0 ) = lim
N →∞ N ∑ {␾ (n␶ 0 ) − ␾ [(n − 1)␶ 0 ]}2
n =1
N
1
= lim
N →∞ N ∑ {␾ 2(n␶ 0 ) − 2␾ (n␶ 0 ) ␾ [(n − 1)␶ 0 ␾ 2[(n − 1)␶ 0 ]} (14.15)
n =1
= 2R␾ (0) − 2R␾ (␶ 0 )
Similarly, higher order differences can be computed. The variance of the second
phase differences is obtained by taking the differences of the first phase differences
and averaging their squares:
N
1
␴⌬22␾ (␶ 0 ) = lim
N →∞ N ∑ ({␾ [(n + 1)␶ 0 ] − ␾ (n␶ 0 )} − {␾ (n␶ 0 ) − ␾ [(n − 1)␶ 0 ]})2
n =1
(14.16)
In terms of the autocorrelation function, this works out to
␴⌬22␾ (␶ 0 ) = 6R␾ (0) − 8R␾ (␶ 0 ) + 2R␾ (2␶ 0 ) (14.17)
These measurements can be performed on both pulsed and CW signals. In the

case of the pulsed signal, the time separating the phase measurements must be a
multiple of the PRI. For a pulsed signal, the first phase differences are a measure
of the pulse-to-pulse phase variation. For such a signal, one would be interested
in the phase change over the pulses separated in time by more than one PRI because,
in a coherent radar, coherent integration would take place over a number of pulses.
2 2
Therefore, one would measure not only ␴⌬␾ (PRI), but also ␴⌬␾ (K × PRI) where
K is an integer up to the order of the number of pulses to be coherently integrated.
Similarly, for a CW radar signal, one would want to know the phase variation
over time intervals that are comparable to the reciprocal of the bandwidth of the
Doppler filter being used for velocity measurements.
14.3.2 RMS Frequency Fluctuations

If the average frequency of a signal is measured over a time duration D, this is
equivalent to measuring the phase change during that time and dividing by the
time (see Figure 14.3).
Suppose the average frequency is measured during the two pulses of duration
D and the frequency change over the interval ␶ 0 is obtained by subtracting
␾ (␶ 0 + D) − ␾ (␶ 0 ) ␾ (D) − ␾ (0)
⌬f = − (14.18)
2␲ D 2␲ D
Figure 14.3 Measurement time and time between measurements.
The variance of this frequency difference is of interest because it can be related

to the Allan variance as follows.
The variance of this frequency difference can be related to the autocorrelation
of the phase disturbance to obtain
2
2(2␲ D)2 ␴⌬ f (␶ 0 , D) = 4R␾ (0) − 4R␾ (D) + 2R␾ (␶ 0 − D) (14.19)
+ 4R␾ (␶ 0 ) + 2R␾ (␶ 0 + D)
In this expression, D is the measurement or averaging time for the individual

frequency measurements (in Figure 14.3, it is the pulse duration) and ␶ 0 is the time
between the frequency measurements.
For pulsed signals, if a single phase value is measured for each pulse, the average
frequency over the entire pulse interval can be computed. In this case, D = ␶ 0 and
using (14.19) gives
2 2
2(2␲␶ 0 )2 ␴⌬ f (␶ 0 , ␶ 0 ) = 6R␾ (0) − 8R␾ (␶ 0 ) + 2R␾ (2␲ 0 ) = ␴⌬2␾ (␶ 0 )
(14.20)
In other words, this variance of the frequency differences is equal to the variance
of the second phase differences divided by 2(2␲␶ 0 )2. The frequency variance for
the case where the measurement interval is equal to the interval between the
measurements is called the Allan variance when it is expressed as a fraction of the
average (or carrier) frequency. The Allan variance is denoted in the literature by
␴y2 (␶ ) and is related to the variance of the second phase difference by
␴⌬22␾ (␶ )
␴y2 (␶ ) = (14.21)
2(2␲␶ fc )2
where:
␴␾2 ␶ = Allan variance

␶ = frequency measurement time and time between measurements
fc = average or carrier frequency
␴⌬22␾ (␶ ) = variance of second phase differences
For stable pulsed signals, the Allan variance can be measured for ␶ equal to a
multiple of the PRI if the phase can be sampled once per pulse. The rms fractional
frequency variation can also be measured and reported when the measurement time
is less than the time between measurements. This is of value when the instabilities are
too great to allow meaningful phase difference measurements from pulse to pulse.
In this case, a single-pulse frequency measurement can be made (by means of a
discriminator, a counter, spectral analysis, or other techniques) and the variance
of these frequency differences can be computed.
Consider the case where the phase ␾ (t) from one pulse to the next is a uniformly
distributed random variable over the range (−␲ , ␲ ). Then the variance of the phase
measurements is given by
␲
␴␾2 =
1
2␲ 冕
−␲
␾ 2 d␾ =
␲2
3
(14.22)
␲
␴␾ = = 103.9 deg = √R ␾ (0) (14.23)
√3
Therefore, if the standard deviation of the phase measurements approaches
100°, the phase change from pulse to pulse is essentially random. In this case, the
standard deviation of the first phase differences would be about 147°, while that
of the second phase differences would be about 254°.
␲
␴␾ = √R␾ (0) = (14.24)
√3
␴⌬␾ = √2R␾ (0) = √

2␲
(14.25)
√3
␴⌬2␾ = √6R␾ (0) = √2␲ (14.26)
In other words, this technique to obtain the Allan variance through phase
differences is valid only for Allan variance values less than
1
␴ y (␶ )max = (14.27)
√2␶ fc
For example, a 300-MHz signal with a 3,333.33-␮ s PRI would be required to
have a square root of the Allan variance no larger than
1
␴ y (␶ )max = 6
= 0.707 ⭈ 10−6 = 0.707 ppm (14.28)
√2 ⭈ 10
to make use of the phase measurement technique.
14.3.3 Signal Repeatability

Another method of measuring signal stability is to take a waveform section and
overlay it on top of another waveform section and note the differences. This can
be called signal repeatability measurement. The smaller the difference between the
two waveform samples, the more repeatable it is. Again, this can be done with
pulsed CW signals, but it is most appropriate for a pulsed signal. In this situation,
a pulse is overlaid with a nearby pulse and the rms difference is measured. Indeed,
the procedure would be to adjust the amplitude and phase of one of the pulses to
minimize their difference. This is the procedure used in Chapter 11 to compare
waveforms, which was defined there as the normalized Euclidean distance (NED).
In Chapter 11, the application was to compare the demodulated envelopes of the
AM and FM waveforms of signals received at different locations or times. Here
the NED is used to compare the predetection waveforms of successive pulses from
the same radar.
Consider the NED defined as
D
冕 [s 1 (t) − s 2 (t)]2 dt
0
NED = (14.29)
D D
冕 2
s 1 (t) dt + 冕 2
s 2 (t) dt
0 0
where
NED = normalized Euclidean distance

s 1 (t) = signal 1
s 2 (t) = signal 2
D = pulse duration
Usually, for coherence measurements,
s 1 (t) = one pulse of a radar signal

s 2 (t) = next pulse of a radar signal
Then s 2 (t) = s 1 (t − ␶ 0 ) where ␶ 0 is the separation of the pulses in time and

s 1 (t) is assumed to be the radar’s output signal for all time. Rewriting (14.29) this
way gives
D
冕 [s 1 (t) − s 1 (t − ␶ 0 )]2 dt
0
NED = (14.30)
D D
冕 2
s 1 (t) dt + 冕 2
s 1 (t − ␶ 0 ) dt
0 0
The denominator is simply the total energy of the two waveform samples
(separated in time) that are being compared in the numerator. Usually, the denomi-
nator will be a constant and serve as a normalizing factor. The numerator is of
more interest. Expanding it gives
D D D
冕 2
s 1 (t) dt + 冕 2
s 1 (t 冕 2
− ␶ 0 ) dt − 2 s 1 (t) s 1 (t − ␶ 0 ) dt (14.31)
0 0 0
so that the NED can be written as

D
冕
2 s 1 (t) s 1 (t − ␶ 0 ) dt
0
NED (␶ 0 ) = 1 − (14.32)
D D
冕 2
s 1 (t) dt 冕 2
s 1 (t − ␶ 0 ) dt
0 0
The numerator is the correlation coefficient at ␶ = ␶ 0 . The denominator is

actually the total energy in the signal over the interval from zero to ␶ 0 + D, as is
shown in Figure 14.4.
Suppose
s 1 = sin (␻ t + ␾ 1 ) (14.33)
s 2 = sin (␻ t + ␾ 2 )
Then the NED becomes
NED (␶ 0 ) = 1 − cos (␾ 2 − ␾ 1 ) (14.34)
Assume ␾ 2 − ␾ 1 is a small angle (i.e., that there is not much pulse-to-pulse

phase change); then the NED is approximately
Figure 14.4 Timing relationships.

(␾ 2 − ␾ 1 )2
NED (␶ 0 ) ≅ (14.35)
2
The expected value of the NED can then be related to the autocorrelation
function of the phase. Let ␾ 1 = ␾ (t) and ␾ 2 = ␾ (t + ␶ 0 ). Then,
1 2
NED (␶ 0 ) ≅ R␾ (0) − R␾ (␶ 0 ) = ␴⌬␾ (14.36)
2
All of these time-domain stability measures are now seen to be related. For
example, the Allan variance can be expressed in terms of the NED ratio as
〈4NED (␶ 0 )〉 − 〈NED (2␶ 0 )〉

␴y2 (␶ 0 ) = (14.37)
(2␲ fc ␶ 0 )2
To derive this, note that
␴⌬22␾ = 4␴⌬2␾ (␶ 0 ) − ␴⌬2␾ (2␶ 0 ) (14.38)
These relationships are summarized in Table 14.1.
14.3.4 Effects of Variations in ␶ 0

When making these measurements, the value of ␶ 0 may fluctuate. As long as this
fluctuation is small (10% or so), the effect will be to obtain the average of the
stability measure over the range of the measurement separation times represented
by the fluctuation of ␶ 0 . Because, typically, the stability measure changes roughly
in proportion to the measurement separation time, small variations in ␶ 0 are not
Table 14.1 Relationship Among Three Time-Domain Frequency Stability Measures

RMS Phase
Allan Variance NED Difference
In terms of — 〈NED (␶ 0 )〉 = (1/2) (2␲␶ c fc )2
2
␴⌬␾ (␶ 0 ) = (2␲␶ 0 fc )2
Allan variance
∞
∑ (1/2)n␴y2 (2n␶ 0 )
n =8
In terms of 2 4〈NED (␶ 0 )〉 − 〈NED (2␶ 0 )〉 —

NED ␴ y (␶ 0 ) = 2
2
␴⌬␾ = 2〈NED (␶ 0 )〉
(2␲ fc ␶ 0 )
2
In terms of
2 ␴⌬␾ (␶ 0 ) 2
〈NED〉 = (1/2)[␴⌬␾ (␶ 0 )] —
rms phase ␴y = 2
difference 2(2␲␶ 0 fc )
2 2
2␴⌬␾ (␶ 0 ) − ␴⌬␾ (2␶ 0 )
=
2(2␲␶ 0 fc )2
troublesome. Of course, such variation may make the measurement process more
difficult to carry out.
14.3.5 Frequency-Domain Stability Measures

The spectral density of the phase fluctuation is given by the Fourier transform of
the autocorrelation function of the phase disturbance, or
冕
s␾ ( f ) = 4 R ␾ (␶ ) cos 2␲ f␶ d␶ (14.39)
0
Because the instantaneous frequency fluctuation is the derivative of the instanta-

neous phase fluctuation divided by 2␲ , the spectral density of the frequency fluctua-
tion is related to the spectral density of the phase fluctuation by
s f ( f ) = f 2s␾ ( f ) (14.40)
When divided by the square of the carrier frequency (or average frequency),
the spectral density of the fractional frequency variations is obtained:
sy ( f ) = 冉冊
f 2
s (f)
fc ␾
(14.41)
Any or all of these spectra may be estimated to specify frequency stability.

Barnes et al. [5] have defined s y ( f ) as the general frequency-domain measure of
stability. It is clearly easy to obtain from s␾ ( f ). In practice, the ratio of the single-
sideband phase noise level to the carrier level is often measured. Assuming that
the phase modulation index is small, this is equal to s␾ ( f )/2. The phase noise is
usually expressed in decibels below the carrier level per hertz of analysis filter
bandwidth (dB/Hz). Typically, oscillator stabilities are specified by giving the maxi-
mum levels of the phase noise spectral density at different frequency separations
from the carrier, as shown in Table 14.2. This type of frequency-domain stability
measure is also valuable and in wide use in the radar community for specifying
oscillators to be used in Doppler systems.
When measuring the phase spectral density stability measure for pulsed signals,
it is first necessary to use high-quality filtering and down-conversion so that a
Table 14.2 Maximum Levels of Phase Noise Spectral

Density at Various Frequency Separations
Offset from Carrier (Hz) Maximum Phase Noise (dB/Hz)
0.1 to 1 −70
1 to 10 −100
10 to 100 −110
10 and up −120
single line of the pulsed signal spectrum can be isolated for analysis through CW
techniques. If the pulsing function itself is stable, any of the lines near the center
of the spectrum will be acceptable. However, if the pulse interval jitters, only the
center line of the spectrum will be acceptable. This is because the pulsing instabilities
contribute to the spectral width of each line in proportion to their separation from
the carrier frequency. The practical problems with measuring the phase noise
spectra of pulsed microwave signals are first to mix down to the base band coher-
ently, without any aliasing or spectrum fold-over, and second to locate the center
line of the spectrum. Once this is done, the resulting single line can be analyzed
in the same way as a CW signal.
The question of how well the frequency stability can be determined from noisy,
time-limited, or band-limited data must be answered. For this reason, it is important
to specify if the stability estimate is limited by the SNR or the measurement
bandwidth. The next section gives these limitations for the Allan variance. Based
on the relationships between the various measures, it is clear that the available
SNR and bandwidth are important parameters no matter how the stability is
measured.
In the general case, the Allan variance is determined by measuring the average
of the instantaneous frequency over intervals of T sec and then computing the
mean square of the difference of successive readings:
␴y2 (␶ ) = 〈(y n +1
2
− y n )2
〉 (14.42)
where:
(n + 1)␶
y=
1
␶ 冕 y(t) dt
n␶
〈x〉 = infinite time average of x
y(t) = instantaneous fractional frequency
In the usual case, there are limitations in the measurement process due to
bandwidth restrictions and thermal noise.
14.3.6 Bandwidth Limitations on Allan Variance Measurements

If the signal is processed through a narrowband filter before the Allan variance
estimate is made, the variance cannot be computed over an infinite frequency range.
In practice, there is always a restricted measurement bandwidth. The effect of this
on the Allan variance estimate depends on the value of ␶ and on the spectral density,
s y ( f ), of the fractional frequency variation.
Barnes [5] shows that the Allan variance may also be obtained from s y ( f )
through the relation
∞
␴y2 (␶ ) 冕冋
= 2 sy ( f )
sin2 ␲␶ f
(␲␶ f )2
−
sin2 2␲␶ f
(2␲␶ f )2
册 df (14.43)
0
∞
冕冋
= 2 sy ( f )
sin2 ␲␶ f
(␲␶ f )2
册 df
0
Note that this relates the Allan variance and all the other time-domain measures
in Table 14.1 to the frequency-domain measure of stability. Suppose the fractional
frequency variation spectrum is white so that
sy ( f ) = s0
then the true value of the Allan variance is found to be
␴y2 (t) = 2s 0 ␶
However, if the signal from the oscillator is passed through a filter of bandwidth
± f␤ from the carrier, then the measured value of ␴y2 (␶ ) will be less than the true
value. Substituting f␤ for the upper limit of integration in (14.43) and integrating
numerically gives the result shown in Figure 14.5. Here, the vertical scale is the
Figure 14.5 Attenuation of Allan variance as a function of measurement bandwidth.

ratio of the estimated Allan variance to the true Allan variance, and the horizontal
scale is the measurement bandwidth, f␤ , expressed in terms of 1/␶ . As can be seen,
a measurement bandwidth of 2/␶ provides better than 90% accuracy.
For short values of ␶ , it is important to ensure that sufficient measurement
bandwidth is used.
14.3.7 Noise Limitations on Allan Variance Measurements

Suppose it is necessary to estimate the Allan variance in the presence of white
noise. The rms frequency error, due to thermal noise, in making frequency measure-
ments over a signal of duration ␶ is [6]
1
␴f = (14.44)
2␲␶ √SNR
The fractional rms error is then
1
␴ y (␶ ) = (14.45)
2␲ f␶ √SNR
where:
SNR = signal-to-noise power ratio

f = frequency being measured
␶ = time interval over which measurements are made
Notice that ( f␶ ) is the number of cycles available within the pulse. Equation
(14.45) is easily seen to be applicable to a device that measures the time required
for ( f␶ ) cycles.
If it is assumed that successive average frequency measurements are independent
Gaussian-distributed random variables, as would be expected for noise-induced
errors, then the Allan variance as given by (14.43) is equivalent to (14.45) because
the variance of the difference of two identically independent Gaussian random
variables is twice that of one of them. Equation (14.45) is plotted in Figure 14.6
for various SNRs as a function of the product of averaging time and frequency.
Of course, the measurement bandwidth has a direct effect on the available
SNR. The SNR is given by
s
SNR = (14.46)
kT (NF)f␤
where:
Figure 14.6 Noise floor for ␴ y (␶ ).
s = signal power
k = Boltzmann’s constant
T = absolute temperature
(NF) = system noise figure
f␤ = measurement bandwidth
The optimum measurement process would use the smallest bandwidth commen-
surate with the averaging time being used. If this strategy is adopted, then
2
f␤ ∼ (14.47)
␶
s␶
SNR = (14.48)
2kT (NF)
␴ y (␶ ) = √ 3/2
2kT (NF)
(14.49)
2␲ f␶ √s
14.3.8 Frequency Stability Measures for Power Law Spectra

Figure 14.7 shows the spectral density of frequency fluctuations for five common
types of instability observed in practice. These fluctuations are given names
according to the exponent in
Figure 14.7 Spectral density of common frequency instabilities.
+2
sy ( f ) = ∑ An f n (14.50)
n = −2
These names and the corresponding theoretical Allan variance are shown in
Table 14.3. Here, f␤ is the upper bandpass limit of the stability measuring system.
Because the Allan variance (as a function of the averaging time for the white
phase noise and flicker phase noise) is virtually identical among these noise types,
it cannot be used to distinguish among them. For this reason, a modified Allan
variance was defined [7].
Table 14.3 Common Power Law Frequency Instabilities and Their Allan Variances
Spectral Allan Variance
Density Single Pole
Noise Sy ( f ) Exponent Rectangular Lowpass Lowpass
3f␤ A 2 3f␤ A 2
White phase A2 f 2 −2 (−20 dB per decade)
2 2
(2␲ ) ␶ (2␲ )2␶ 2
[1.038 + 3 ln ( f␤ ␶ )]A 1 3 ln ( f␤ ␶ )A 1
Flicker phase A2 f −1 (−10 dB per decade)
(2␲ )2␶ 2 (2␲ )2␶ 2
White frequency A0 0 A 0 /2␶ A 0 /2␶
Flicker frequency A −1 f −1 (+10 dB per decade) 2 [ln (2)]A −1 2 [ln (2)]A −1
Random walk 2␲ 2 2␲ 2
frequency A −2 f −2 (+20 dB per decade) ␶A −2 ␶ A −2
3 3
14.3.9 Sinusoidal FM and Linear Frequency Drift

Often the frequency disturbance is periodic. In this common case, it is important
to realize that the Allan variance approaches zero when the averaging time is equal
to a multiple of the disturbance period. If it is not known in advance that the
disturbance is periodic, different measurements of the Allan variance may take on
widely differing values if slightly different averaging times are used by different
observers. The Allan variance and spectral density for sinusoidal disturbances are
given by [7]
Sy ( f ) = 冉冊
1 ⌬f
2 fc
␦ ( f − fm ) (14.51)
␴ y (␶ ) = 冉冊
⌬ f 2 sin4 (␲ fm ␶ )
fc (␲ f m ␶ )2
(14.52)
If there is linear drift of frequency with time, this should be removed before
attempting to make frequency-domain (spectral) stability measurements because
there is no useful model for the spectrum in this case. If the fractional frequency
is given by the product of the drift, d, and the time separation, ␶ , then
Y = d (␶ ) (14.53)
and direct calculation in the time domain gives the Allan variance as [7]
d⭈␶
␴ y (␶ ) = (14.54)
√2
Note that the Allan variance contribution from a linear frequency drift has the
same functional dependence on the averaging time as the contribution from random
walk frequency noise (n = 2; see Table 14.3). For ELINT purposes, it is usually
not fruitful to attempt to determine which type of frequency process is present
(i.e., to find the value of the exponent) from the dependence of the Allan variance
estimate on the averaging time.
14.3.10 Short-Look Problem

Most of the literature relating to short-term stability measurement assumes that
the signal is available for an indefinitely long time. In ELINT, the signal may be
observable at high SNR only during the relatively short times that the radar beam
passes. This means that instead of an idealized line spectrum of a pulsed signal,
the observable spectrum (as viewed through the duration ‘‘window’’ of the radar
beam illumination time) appears as a series of sin (x)/x functions, each located at
a multiple of the PRF and having a width approximately equal to the reciprocal
of the illumination time. Clearly, if the sidebands due to frequency variations are
hidden beneath those sidebands of the spectrum caused by the limited duration of
the pulse train, it will not be possible to measure the true carrier to sideband ratio.
Generally speaking, if there are more than 30 pulses in a group available for
analysis, the spectrum contributions due to the short-look problem will be small
enough so that one can assume that the components located between the PRF lines
are due to either thermal noise or frequency fluctuations. If there are fewer than
20 pulses, the spectral components due to the short-look problem will probably
exceed those of the frequency fluctuations. Of course, the Allan variance can still
be estimated, but again the limited number of samples will cause a greater uncer-
tainty in the result.
Usually, there are many such short looks available—one is received each time
the radar beam scans past the ELINT site. These additional short looks can help
improve the estimates by averaging together more frequency values or more spectra.
However, this will not solve the problem completely because the successive looks
are so far apart in time that they are essentially independent insofar as their phase
is concerned. Figure 14.8 shows this effect for 50, 30, 20, and 10 pulses received
in a single scan. The resolution of these plots is approximately 0.2% of the PRF.
As some consolation, the ELINT analyst should note that the radar is limited
in its ability to detect targets in clutter by this same problem.
Figure 14.8 Short-look spectra.

14.4 Frequency Character of CW Signals

All of the techniques relating to the measurement of the carrier frequency character-
istics of pulsed signals generally apply to CW signals. For CW signals, the spectrum
analysis problems caused by the convolution of the carrier line spectrum with the
video pulse spectrum are eliminated. On the other hand, there is usually some type
of modulation present that may obscure the properties of the carrier. The modula-
tion may include:
1. Antenna scanning;
2. Frequency sweeping;
3. Multiple channel subcarrier modulation using AM or FM;
4. Discrete modulation such as frequency shift keying (AM or FM) or phase
reversals.
Measuring the carrier frequency of modulated CW signals may require long-

time averaging in the case of FM sweeps. In this case, the average frequency (if it
is of interest) can be found by counting over an integral numbers of complete
sweep repetitions.
In the case of AM, a limiter can precede the frequency-measuring counter, and
the carrier or center frequency can be properly measured. In the case of FM, the
long-term average can be investigated through counting. With FM-FSK, the average
frequency may never actually be transmitted.
The usual technique in analyzing the modulation details is to try various types
of demodulators and to examine their outputs. These can include discriminators,
filter banks, and envelope detectors. Once the type of modulation is determined,
demodulation can usually be achieved. For example, if 180° phase reversals are
involved, and, if the SNR is high, a frequency doubler followed by a divide-by-
two circuit provides a means for recovering the carrier. This can be analyzed using
any of the techniques described for pulsed signals. In addition, this recovered carrier
can be used as one input to a phase detector to determine the code sequence being
used.
In the case of many ECM signals that are CW in nature, a look-through interval
is often present. The signal may have a duty cycle approaching one. However, the
pulsing effects of the on-off look-through modulation must be considered. For
frequency-multiplexed signals, spectrum analysis can usually locate the subcarriers.
If FM multiplexing is used, the failure of a limiter to remove the modulation can
indicate that FM is in use.
Tracking filters of narrow bandwidths can be used with CW signals to follow
weak signals in the noise. These may be of the phase-locked loop type. In this case,
the frequency variations of the incoming signals are represented by the variations
of the voltage-controlled oscillator tuning the voltage within the phase-locked loop.
14.5 Pulsed Signal Example

Digitized predetection data is normally used for measuring intrapulse AM and PM.
It can also be used for measuring pulse-to-pulse RF coherence if proper digitizing
techniques are used. The key is to measure the instantaneous phase of many pulses
at precisely known points in time. The collection system needed for interpulse
coherence measurement differs from that needed for intrapulse unintentional modu-
lation on pulse (UMOP) measurement by the addition of a counter that measures
the number of clock pulses that elapse between the end of one pulse and the
beginning of the next. In addition, the digitizing clock and all local oscillators
must have good stability over the time duration of the coherence measurement
(e.g., 100 ms).
Consider the following example:
nominal pulse duration = 1 × 10−6 sec
nominal PRI = 1,000 × 10−6 sec
digitizing clock period, Ts = 5 × 10−9 sec
The UMOP data will be contained in approximately 200 samples that are
obtained during a single pulse. Then, for about 200,000 clock pulses, the signal
is absent and the A/D clock pulses are simply counted. Then, the next pulse rises
above the threshold and digitizing of about 200 more samples occurs, and so on.
This data stream is sufficient to measure coherence, as is shown here:
Phase of pulse 1, 1st sample = ␾ 1, 1

Phase of pulse 1, 2nd sample = ␾ 1, 1 ⌬␾ , ⌬␾ = 2␲ f IF TS
Phase of pulse 1, 3rd sample = ␾ 1, 1 + 2⌬␾
⯗
Phase of pulse 1, 200th sample = ␾ 1, 1 + 199⌬␾
The number of missing samples between pulse 1 and pulse 2 is 199,800.
Phase of pulse 2, 1st sample = ␾ 2, 1 = ␾ 1, 1 + 200,000⌬␾

Phase of pulse 2, 2nd sample = ␾ 2, 1 = ␾ 1, 1 + 200,000⌬␾
and so on.
If the sampling rate is well above the Nyquist rate, ⌬␾ will be less than 180°.
Suppose the signal of interest is mixed down to an IF of 60 MHz and that the
bandwidth is about 20 MHz (e.g., 50 to 70 MHz). Suppose the actual center
frequency is 59.444 MHz. If the digitizing rate is 200 MHz, then ⌬␾ is 107°.
Arbitrarily assume ␾ 1, 1 = 0 and the phase data become as shown in Table 14.4.
Note that the modulo-360° phase of the first sample of the nth pulse is given
by [(n − 1)(200,000)(107)] modulo 360 and is shown in Table 14.5. Coherence is
indicated by the linear progression of the unwrapped phase, at least over the number
of pulses used by the radar for coherent integration. Any deviation from a linear
phase progression indicates instabilities of the signal relative to the A/D converter
clock and system local oscillators.
Table 14.4 Phase Samples Versus Unwrapped Phase

Phase Sample (modulo 360°) Unwrapped Phase
(First sample of Pulse 1) ␾ 1,1 = 0 0
␾ 1,2 = 107 107
␾ 1,3 = 214 214
␾ 1,4 = 321 321
␾ 1,5 = 68 428
␾ 1,6 = 175 535
␾ 1,7 = 282 642
␾ 1,8 = 29 749
␾ 1,9 = 135 856
␾ 1,10 = 243 963
␾ 1,11 = 350 1,070
␾ 1,12 = 97 1,177
(Last sample of Pulse 1) ␾ 1,200 = 53 21,293
(First sample of Pulse 2) ␾ 2,1 = 160 21,400,000
(Second sample of Pulse 2) ␾ 2,2 = 267 21,400,107
(Third sample of Pulse 2) ␾ 2,3 = 14 21,400,214
␾ 2,4 = 121 21,400,321
␾ 2,5 = 228 21,400,428
␾ 2,6 = 335 21,400,535
␾ 2,7 = 82 21,400,642
Table 14.5 First Phase Sample Versus Unwrapped Phase

Pulse Number (n) Phase of First Sample (°) Unwrapped Phase
1 ␾ 1,1 = 0 0
2 ␾ 2,1 = 160 160
3 ␾ 3,1 = 320 320
4 ␾ 4,1 = 120 480
5 ␾ 5,1 = 280 640
6 ␾ 6,1 = 80 800
7 ␾ 7,1 = 240 940
8 ␾ 8,1 = 0 1,100
9 ␾ 9,1 = 160 1,260
10 ␾ 10,1 = 320 1,580
11 ␾ 11,1 = 120 1,700
etc. etc. etc.
Note that if the PRI is constant and the threshold is insensitive to pulse-to-
pulse amplitude changes, the first samples of consecutive pulses may be separated
by a fixed (constant) number of A/D clock periods. In this case, it is not necessary
to count the number of A/D clock periods between the pulses, and the coherence
of the signal can still be estimated.
So far, only the first phase sample of each pulse has been used to explore the
pulse-to-pulse phase progression. However, when there are additional samples
within the pulse, it is reasonable to also make use of these. Because the first sample
of pulse (n), ␾ n, 1′ and the first sample of pulse (n + 1), ␾ n + 1, 1 are separated by
exactly the same time as the second sample of pulse (n), ␾ n, 2 and the second sample
of pulse (n + 1), ␾ n + 1, 2 , it is clear that the phase change will be the same if the
phases of the second samples of each pulse are subtracted—provided that there is
the same FMOP or PMOP on each pulse. Notice that in this example:
␾ 2, 1 − ␾ 1, 1 = 160 − 0 = 160
␾ 2, 2 − ␾ 1, 2 = 267 − 107 = 160
␾ 2, 3 − ␾ 1, 3 = 14 − 214 = −200 (or +160)
␾ 2, 4 − ␾ 1, 4 = 124 − 324 = −200 (or +160)
␾ 2, 5 − ␾ 1, 5 = 228 − 68 = 160
and so on
This indicates one way to use these additional intrapulse phase samples; namely,
compute the difference in the phase between like numbered samples on successive
pulses and find the average of the differences. This has the effect of reducing the
effects of quantizing noise and thermal noise.
14.6 Measuring Coherence
Once the pulse-to-pulse phase change is determined, any of the methods described
in this chapter may be used to determine a coherence estimate. For example, the
best straight line may be fitted to the unwrapped phase data and then subtracted.
The residues are then Fourier transformed to obtain the phase noise spectrum. The
better the coherence, the lower the phase noise spectral components. Another
method is to examine the changes in the average frequency to compute the Allan
variance. The average frequency during a particular pulse interval is
␾n − ␾n −1
fn =
2 ␲ Tn
where:
fn = average frequency between pulse n − 1 and n

␾ n = phase of the nth pulse (rad)
␾ n − 1 = phase of the (n − 1)th pulse (rad)
Tn = time between pulse (n − 1) and pulse n
In this example,
160° 1
fn = × = 444.44 Hz
360° 1 × 10−3 sec
To compute the Allan variance, it is necessary to find the average of the sum
of the squares of the frequency changes over many pulses:
N
1
⌬f 2 =
N−1 ∑ ( fn + 1 − fn )2
n =1
14.7 Effects of Drift 365
In this example, fn + 1 = fn , and the Allan variance is zero because the signal is
completely stable. Then this value is divided by the square of the carrier or RF:
⌬f 2
Allan variance =
(RF)2
Notice that for a completely noncoherent signal, the individual phase values
will be random numbers uniformly distributed over 360°. The difference of two
such random numbers will also be random, but with a triangular distribution
extending from −360° to +360°. An Allan variance value can be computed, but
there is no real coherence present.
14.7 Effects of Drift
If the frequency of any of the local oscillators (including the digitizing clock) drifts,
the effect is to create a quadratic (rather than linear) phase versus time plot. Then
unwrapping the phase may become difficult and the pulse-to-pulse phase changes
may also appear to be random. While in principle it is possible to remove the drift
by fitting a quadratic to the unwrapped phase, this is of limited value unless it is
possible to determine whether the drift is due to the intercept equipment or to the
incoming signal. Making the local oscillators very stable is a prerequisite for a
useful coherence measuring system.
References
[1] Blachman, N. M., and G. A. McAlpine, ‘‘The Spectrum of a High Index FM Waveform:
Woodward’s Theorem Revisited,’’ IEEE Trans. on Communications Technology,
Vol. COM 17, No. 2, April 1969, pp. 201–207.
[2] Wiley, R. G., et al., ‘‘Demodulation Procedure for Very Wideband FM,’’ IEEE Trans. on
Communications, Vol. COM 25, No. 3, March 1977.
[3] Curry, T. F., ‘‘Specification of Short-Term Frequency Stability by Maximum Likelihood
Estimates,’’ IEEE-NASA Symposium on Short-Term Stability Proc., NASA Rep. Sp-80,
November 1964, pp. 101–109.
[4] Allan, D. W., ‘‘Statistics of Atomic Frequency Standards,’’ Proc. IEEE, Vol. 54,
February 1966, pp. 221–230.
[5] Barnes, J. A., et al., ‘‘Characterization of Frequency Stability,’’ IEEE Trans. on Instrumenta-
tion and Measurement, Vol. IM-20, May 1971, pp. 105–120.
[6] Skolnik, M. I., Radar Handbook, New York: McGraw-Hill, 1970, pp. 4–7.
[7] Rutman, J., and F. L. Walls, ‘‘Characterization of Frequency Stability in Precision Frequency
Sources,’’ Proc. IEEE, Vol. 79, June 1991, pp. 952–960.
CHAPTER 15
Determining ELINT Parameter Limits
15.1 Introduction
When faced with a number of parameter values measured by various methods in

different locations at different times, the analyst must determine the likely range
of the parameter involved for a given type of radar. One simple method is to take
the largest and smallest observed values as the parametric limits (e.g., the RF range
could be from the lowest reported RF value to the highest reported RF value).
Unfortunately, this can place too much emphasis on extreme values that may
actually represent errors.
One common practice is to report ELINT parameter limits as being those that
include 95% of the data values. Selection of the data to be used requires care and
judgment on the part of the analyst. It may be useful to eliminate values having a
poor accuracy or that are too old. Generally, it is necessary to look at the range
determined by the overall set of data values and then compare this to the ranges
indicated when various subsets of the data are analyzed. For example, the parameter
range indicated by 5 years of data values may be compared to those indicated by
data from each year individually to see if changes with time are occurring.
15.2 Histograms Used to Determine Parameter Limits
A histogram gives the number of occurrences of a parameter value within a range

of values. For example, a table showing the number of people from 0 to 10 years
of age, from 10 to 20 years, from 20 to 40 years, from 40 to 60 years, and 60
years and older would be a histogram showing the distribution of the population
by age groups.
Figure 15.1 shows a histogram of pulse duration values. The horizontal axis
consists of 0.1-␮ s bins, and the vertical axis is the number of reported occurrences
of each pulse duration value observed. This histogram was made from tens of
thousands of pulse duration values, but it gives a very compact and convenient
representation of this large body of data.
A more challenging histogram of pulse duration is shown in Figure 15.2. The
amount of data available to build this histogram is smaller by a factor of about
400, which may account for the gaps and its generally ragged appearance. In this
case, each individual intercept may require special scrutiny. For example, all of
the points above 5 ␮ s are accounted for by only nine intercepts. Examining these
367
368 Determining ELINT Parameter Limits
Figure 15.1 Histogram of pulse duration values.
Figure 15.2 Another pulse duration histogram.
to determine where and when they took place, and whether or not they have been
properly identified, may allow some or all of them to be deleted. Also, the gaps
may represent different pulsewidth modes, in which case separate histograms for
each mode’s pulse durations are called for. Discovering radar modes often requires
the use of conditional histograms, which are described in Section 15.3.
15.2 Histograms Used to Determine Parameter Limits 369
The histogram approximates the shape of the probability density function

(PDF) of a random process. The nature of the approximation is that the histogram
more closely resembles the PDF as (1) the bin size becomes very small and (2) the
number of data values or occurrences becomes very large.
In practice, the number of data values is often fixed, and the smallest bin size
that is practical depends on the resolution of the measuring device. For example,
in Figures 15.1 and 15.2, if the pulse duration measuring system uses a 0.1-␮ s
clock interval, the histogram bin size cannot be reduced.
To use histograms, it is frequently necessary to make several, each having
different bin sizes. The analyst chooses the bin size range that seems to give the
best approximation to the underlying distribution. This judgment is somewhat
subjective; however, one looks for a range of bin sizes over which the essential
shape or character of the histogram is the same. There are two extreme cases.
When the bin size is very large, all of the values fall into the same bin. For example,
if Figure 15.1 were redrawn with 5-␮ s bins (e.g., 0 to 5 ␮ s and 5 to 10 ␮ s), all
of the data values would fall into the first bin. When the bin size is too small, there
is insufficient data to provide a statistically significant number of points per bin.
The histogram would typically have a single point in some bins and the others
would be empty. Generally, the ratio of the number of points to the number of
bins, which is the average number of points per bin, should be more than 10 and
preferably more than 100.
The effect of insufficient data on the histogram is shown in Figure 15.3, which
shows histograms of data values drawn from a uniform distribution and from a
triangular distribution. With an average of only one point per bin, the character
of the distribution is not apparent, whereas with 100 points per bin it is.
Figure 15.3 Effects of insufficient data.

Another property of histograms is that the bin size can be increased to any
multiple of a smaller bin size by simply adding the number of occurrences in a
number of adjacent bins. This is convenient when dealing with very large quantities
of data because one can begin with a relatively small bin size, compute the histo-
gram, and then create larger bin size histograms from the first one without examin-
ing the individual data values.
A second additive property is that several histograms with the same bins can
be added bin by bin to create a composite. For example, a parameter might be
studied for changes in time by making histograms of the data for each year for the
five previous years. A composite (5-year histogram) could then be created by simply
adding the occurrences in each bin for each of the five 1-year histograms.
15.3 Types of Histograms
In addition to the conventional histogram giving an approximation to the PDF,

histograms also often show the cumulative distribution function (CDF). This is
done by simply tabulating a total for the number of occurrences in all the bins
below a given bin. The CDF makes it very easy to determine important parameters
of interest. For example, the median is that bin for which half of the occurrences
are above and half below. Using the CDF, the median can be easily located.
Similarly, one can easily find the range of values within which some fraction (e.g.,
95%) of the occurrences lies. The CDF is the integral of the PDF. A good histogram
program provides both of these.
A second histogram type is the multiple-dimension histogram. For example,
one might plot the number of intercepts reporting both a pulse duration in a certain
bin and a PRI in a certain bin. This would be a three-dimensional plot above the
PRI by pulse duration plane. Adding more dimensions makes the plotting problem
more difficult (e.g., using the PRI, pulse duration and RF as coordinates). To avoid
such problems, it is usual to make multiple histograms of, say, the pulse duration
with the added condition that the PRI reported was also within some specified
range. This is necessary to determine what modes of operation are possible for the
emitter. For example, the overall pulse duration histogram may reveal two distinct
values, and the overall PRI histogram may also reveal two distinct values, as shown
in Table 15.1.
To find out whether (and how frequently) PRI 1 is observed with PD 1 , condi-
tional histograms are needed. Therefore, we count the occurrences of the pulse
duration values for which the radar’s PRI was in the range typical for PRI 1 and
also the pulse duration values for which the radar’s PRI was in the range typical
of PRI 2 . From the results shown in Figure 15.4, we see that PRI 1 is usually seen
Table 15.1 Pulse Duration Values

Versus PRI Values
Pulse Duration Values PRI Values
PD 1 (near 3 ␮ s) PRI 1
PD 2 (near 5 ␮ s) PRI 1 , PRI 2
15.3 Types of Histograms 371
Figure 15.4 Conditional histograms.
with PD 1 , but is sometimes observed with PRI 2 . On the other hand, PD 2 is never
observed with PRI 1 .
The conditional histograms can be added (if the bin sizes are the same) to
obtain a composite. Adding annual histograms over a 5-year period is a form of
conditional histogram, where the condition is placed on the date of the reported
intercept rather than on other parameters of the radar.
Note that simple histograms are completely independent of the order in which
the intercepts are analyzed. To take a simple example, shuffling a deck of cards
will have no effect on a histogram of the number of aces, twos, and threes. There
will always be four of each.
A special histogram that is dependent on the order of the data values is the
delta-␶ histogram, which is described in Chapter 12. This histogram consists of
the time intervals between all pairs of pulses and is related to the autocorrelation
of the TOAs.
There are two parameters to be selected before making a histogram: the parame-
ter range to be covered and the number of parameter subranges, or bins, into which
the overall range of interest is to be divided. If the size of the parameter bins is
too large, the histogram gives little information because all of the data are in one
or two bins. If the bin size is too small and there is insufficient data, the histogram
again fails to be useful because each bin contains only one or two measurements.
Analyst judgment and experience must be used to select the proper number of bins
or bin size so that there is enough data in enough bins to give a useful result. In
this way, the 95% limits of the available data can be directly determined. The
analyst simply notes the parameter range as being those bins that exclude the lower
2.5% and upper 2.5% of the data. For example, if there are 400 data values, the
analyst would eliminate those bins containing the lowest 10 data points and the
highest 10 data points. Then, one estimate of the 95% limits would be the limits
of the parameter values represented by the remaining histogram bins. This can
always be done regardless of the number of data values or the shape of the distribu-
tion. There is no doubt that 95% of the observed values are contained within limits
found in this way. It is not certain that 95% of all of the measurements that might
have been made would also fall in this range. Our confidence about this depends
on the amount of data available.
To account for this lack of certainty about the limits of the parameter values
that are possible as opposed to those values that have actually been observed, it
is necessary to examine what is called the confidence level of the limits as determined
by the analyst from the data. This can be done using the methods of statistics
known as ‘‘parametric’’ [1]. It is possible to state that the limits obtained in the
previous example of 400 data points by deleting the 10 lowest and 10 highest
include 93.5% of the parameter values at a confidence level of 0.9. In other words,
9 times out of 10, 93.5% of a set of 400 data values would fall within these limits.
We could also state that 99 times out of 100, 92% of the data points would fall
within those same limits.
The amount of data to be deleted at each end of the histogram in order to
specify parameter limits so that 95% of the data falls between the limits 19 times
out of 20 (i.e., in 95% of all tests using that much data) varies with the amount
of data, as shown in Figure 15.5. For fewer than 100 data values, none of the data
should be deleted. The parameter range should be given as that covered by the
extremes of the available data and then a notation made that the quantity of data
Figure 15.5 Percentage of data to be deleted for 95% parameter limits.

15.4 Two-Sigma Parameter Limits 373
available is insufficient to specify the 95% limits to a 95% confidence level. Above
10,000 data points, 2.5% of the data at each end of the histogram can be deleted
with virtually no loss of confidence. (If only 1,000 data points are available and
2.5% are deleted at each end, then for 9 out of 10 groups of 1,000 measurements,
95% of the data, will fall between the specified limits; that is, the confidence level
drops from 95% to 90%.)
It should be noted that this technique requires no assumptions about the
underlying distribution of the data, and the limits can be easily determined without
computation. If the measurements are ordered from the smallest to the largest, the
95% range is determined as the extreme values of the data when the fraction of
points given in Figure 15.5 is deleted from each end of the list. A histogram can
also be used and the bins containing that fraction of the data are eliminated from
each end to obtain the 95% limits of the parameter.
Note that the analyst may wish to break the parameter range up into several
modes if the histogram shows the data points clustered around several peaks. The
number of data values in each peak can be noted and individual intervals or
parameter ranges about each peak can be specified using the same method. Note
also that statistically it does not matter if equal numbers of data points are deleted
from each end of the range of values. The total percentage of the data to be deleted
is twice that shown in Figure 15.5. Different numbers of points can be deleted at
each end of the range to achieve the desired total amount of data to be deleted.
15.4 Two-Sigma Parameter Limits
In addition to the histogram, another frequently used statistical approach involves

computation of statistical parameters based on the data values. The most common
ones are the sample mean and the sample standard deviation. The formulas are as
follows:
N
∑ xi
i =1
ms = (15.1)
N
冤冥
N 1/2
∑ (x i − m s )2
␴s = i =1 (15.2)
N−1
where:
m s = sample mean
␴s = sample standard deviation
N = number of data values
x i = the ith data value
The sample mean is at the center of gravity of the histogram of the data. The
sample standard deviation is a measure of the spread or dispersion of the data. In
many cases, it is common to assume that the parameter values being observed are
distributed according to the normal distribution. This is a bell-shaped curve given
by
1
e −(x − m) /2␴
2 2
␾ (x) = (15.3)
√2␲ ␴
where ␾ (x) dx is the probability a data value will be between the values x and
(x + dx), ␴ is the standard deviation, and m is the mean value.
If the data values being observed are distributed according to the normal
distribution, then 95% of them will lie between the mean minus twice the standard
deviation and the mean plus the standard deviation:
m + 2␴
冕 ␾ (x) dx ≅ 0.95 (15.4)

m − 2␴
If the analyst chooses, these so-called ‘‘two-sigma’’ limits may be specified for
a parameter. This means that the lower value is given as the sample mean minus
twice the sample standard deviation, and the upper value is given as the sample
mean plus twice the sample standard deviation. There are two problems with this
approach. First, as with the histogram approach, there is no assurance that the
data available is typical and that the sample mean and standard deviation are the
same as the true mean and true standard deviation. Second, we cannot be certain
that the data values are distributed normally.
The problems of the differences between the sample mean and true mean and
the sample standard deviation and true standard deviation can be handled by
the methods of statistics using the student’s t distribution and the chi-squared
distribution. Using these distributions and assuming that the true distribution of
the data is normal, we can specify an interval in which the true mean or standard
deviation will lie, 9 times out of 10 (a 90% confidence level). For example, 9 times
out of 10 the true mean will fall within the range
␴s
ms ± t (n , 90) (15.5)
√N
where t (n , 90) is given by Table 15.2.
Any standard text or handbook [2] on statistical methods as applied to experi-
ments can be consulted to learn more about these methods.
Quite often the true distribution of data values is not described by the normal
distribution. For this reason, it is preferable to use the histogram approach to
specify a parameter range. This is done very simply by deleting a few points at
each end of the range and is valid for any underlying distribution of the data.
For example, if the data distribution is actually uniform between two limits,
L min and L max (as in a radar set operating equally probably at any frequency within
Table 15.2 Student’s t Multiplier Values

Number of Number of
Data Points t (n, 90) Data Points t (n, 90)
2 6.314 12 1.796
3 2.920 15 1.761
4 2.353 20 1.729
5 2.132 30 1.699
6 2.015 40 1.684
8 1.895 60 1.670
10 1.833 ∞ 1.645
an assigned band), the mean is at the center of the range between two limits and
the standard deviation is the range divided by √3. Then the estimated two-sigma
limits include about 15% more than the actual range:
4␴ 2
= ≅ 1.15 (15.6)
L max − L min √3
Of course, even if the parameter values came from a uniformly distributed

process, the inaccuracies present in measurement of the values would add additional
variations, so that the distribution estimated from measured data would appear
as shown in Figure 15.6. Mathematically, Figure 15.6 is obtained by convolving
the distributions of the measurement errors and the population’s radar signal values.
These measurement errors would increase the estimated standard deviation
and push the two-sigma limits even farther out. In fact, the sample standard
deviation can be significantly affected by just a few measurements that are grossly
in error. For this reason, in order to get reasonable two-sigma limits, outlying
points must be discarded before the sample standard deviation is calculated.
While statistical techniques can be useful to the analyst, judgment and experi-
ence must be combined with knowledge of statistical techniques to specify parameter
limits.
15.5 Histogram Analysis Techniques

15.5.1 Parameter Limits Example
In many cases, the important results to be obtained from the analysis process are the
parameter limits of the emitter (e.g., the RF range over which the radar transmits).
Figure 15.6 Uniform distribution and the effects of measurement errors.

The following hypothetical situation illustrates the complications caused by the

use of different ELINT receivers at different sites.
Suppose there are 100 ELINT stations and that 90 of these are equipped with
old receivers capable of ±50-MHz accuracy RF measurements while 10 are equipped
with newer receivers capable of a ±5-MHz accuracy. Suppose each site reports 100
intercepts of a particular radar and a histogram of the RF values from 10,000
intercept reports. The result is shown in Table 15.3.
For this histogram, the mean ±2␴ limits include nearly all of the data. (These
limits would include 95% of the data if it followed a normal distribution.) The
technique shown in Figure 15.5 does not require the assumption of a normal
distribution and would simply delete 250 data values (2.5%) from each end of the
histogram and give the limits as the extreme values of the remaining 9,500 points.
This would give limits of 2,945 to 3,085 MHz. For this bin size, the closest we
can come to deleting 250 points is to delete one bin at the lower end and one bin
at the upper end of the distribution. Note that the median is near 3,010 MHz, as
indicated by the CDF.
A significant problem is to determine the accuracy with which we know the
parameter limits. In a sense, the accuracy is ±50 MHz because that is the worst
accuracy in the raw intercept report data. Certainly the accuracy is no worse than
that; however, it does not take into account the benefits of having 10,000 intercepts
to consider. Nevertheless, when parameter limits from histograms are being entered
into a database, it seems more descriptive to the eventual user to enter the accuracy
of the measurements themselves rather than try to compute a confidence interval
for the lower and upper limits. One should comment on the histogram (e.g., give
the bin size, number of data values, and overall shape) to assist the end user.
Sometimes the accuracy entered by the analyst may be the histogram bin size
because this is one factor that determines his ability to determine where the 95%
Table 15.3 Histogram of RF Values from 10,000 Intercept Reports

Cumulative Number
Bin (MHz) Number of Reports of Reports
2,940* (and below) 179 179
2,950** 268 447
2,960 453 900
2,970 1,057 1,927
2,980 723 2,680
2,990 631 3,311
3,000 1,059 4,370
3,010 716 5,086
3,020 627 5,713
3,030 1,067 6,780
3,040 731 7,511
3,050 619 8,130
3,060 1,066 9.196
3,070 444 9,640
3,080 175 9,815
3,090 (and above) 185 10,000
Note: mean = 3,015 MHz; ␴ = 38.7 MHz; mean −2␴ = 2,937.5;
mean +2␴ = 3,092.5.
*RF values below 2,945.
**RF values from 2,945 to 2,954.99.
limits are located. However, because the bin size can be selected arbitrarily, it is
not as meaningful as the accuracy value associated with the measurement.
15.5.2 Intercepts Separated into Accuracy Classes

In the previous example, not all of the intercepts have a ±50-MHz accuracy. Ten
percent (1,000) of the reports have a ±5-MHz accuracy. If separate histograms for
the two types of intercepts are made (i.e., conditional histograms, with accuracy
as the conditioning factor), the results are as shown in Table 15.4.
Using only the data from the newer receivers seems to indicate that the radar
operates on four RF channels separated by about 30 MHz. The analyst needs
to make four finer resolution histograms centered at 2,970, 3,000, 3,030, and
3,060 MHz to determine the distribution of the values for each channel.
The data from the older receivers is still of value. The older receivers may be
the only ones available in many geographic areas. It is useful to compare frequency
operation by geographic area and the data from the less accurate receivers would
allow that. Figure 15.7 shows the data of Table 15.4 graphically.
This example also points out the different information needs of different users.
Suppose the user was an intercept operator with one of the older receivers. One
of his most important jobs is to identify the signal type. If he were told that this
radar operates on one of only four specific frequencies, then he possibly could not
identify the radar because his receiver might indicate some different frequency. On
the other hand, a designer of a radar warning receiver that uses an IFM would be
very interested in the channelization statistics to help identify the radar. The analyst
should recognize these different information needs and provide as complete a
picture of the data as possible.
Table 15.4 Results of Conditional Histograms

New Receivers Only Old Receivers Only
Bin (MHz) PDF CDF PDF CDF
2,940* (and below) 0 179 179
2,950** 0 268 447
2,960 0 453 900
2,970 243 814 1,714
2,980 243 723 2,437
2,990 243 631 3,068
3,000 243 495 807 3,875
3,010 495 716 4,597
3,020 495 627 5,218
3,030 252 752 810 6,028
3,040 752 731 6,759
3,050 752 619 7,378
3,060 257 1,000 818 8,196
3,070 1,000 444 8,640
3,080 1,000 175 8,815
3,090 (and above) 248 1,000 185 9,000
Mean = 3,015 ␴ = 33 Mean = 3,014 ␴ = 39
Mean −2␴ = 2,948 Mean −2␴ = 2,938
Mean +2␴ = 3,082 Mean +2␴ = 3,090
*RF values below 2,945.
**RF values from 2,945 to 2,954.99.
Figure 15.7 RF histogram example.
A better example of the effects of RF channelization is shown in Figure 15.8.

Here each histogram peak should be examined in detail and their individual ranges
and an overall range reported. A discussion of the apparently unused channels
(e.g., near 2,140, 2,300, 2,360 MHz) would also be appropriate.
15.5.3 Most Probable Values

When the histogram of a parameter indicates a well-defined peak, the parameter
value that is most frequently reported is called the most probable value. While
there is nearly always a histogram bin with a greater number of intercepts than
Figure 15.8 RF histogram showing channelization.

15.6 Analysis Problems 379
any other bin, the term ‘‘most probable value’’ is reserved for a well-defined peak
that is well separated from other peaks (i.e., a broad distribution where the bin
with the largest number of intercepts has a number of intercepts only slightly larger
than several adjacent bins does not have a most probable value). Naturally, the
shape of the histogram depends on the bin size selected, and the ability to locate
a peak is affected by the bin size. Generally, one chooses a bin size comparable to
the accuracy of the individual intercepts because smaller bins are not very meaning-
ful (i.e., the variations shown begin to reflect measurement errors rather than the
behavior of the signal of interest).
15.6 Analysis Problems
15.6.1 Signal Identification Errors

Perhaps the most difficult problem with analyzing intercept reports is dealing with
those that have mistakenly identified the signal type. A few errors, even among
thousands of reports, can significantly affect statistical calculations. Outlying data
values must be carefully checked to see that the identification is proper. Beyond
that, it is useful to make as many checks as possible to look for mistakes in
identification among all the intercepts. For example, a different parameter value
(as reported by only one station) should be examined for possible misidentification.
Clearly, there is no guarantee that identification errors can be found. Feedback
from the analyst to the interceptors can be very helpful in minimizing the problem.
For example, the analyst may suspect a new mode has appeared. By asking for
certain special measurements, the analyst may be able to confirm or deny that
suspicion.
15.6.2 Transforming Parameters and Their Accuracies

Consider changing a parameter from one form (in which the interceptor reported
it) to another form that is required for a report or user application. A simple
example is changing PRF to PRI. A method that can always be used is to transform
each intercept report of the parameter to the new parameter and find the histogram
and limits of the new parameter directly. However, if a histogram of the old
parameter exists, transforming it to the new parameter is possible. In general, if
the relationship between the old parameter, x, and the new parameter, y, is single
valued and monotonic, the distribution of the new parameter is given by
P(x)
P ′(y) = (15.7)
| dy /dx |
where P ′(y) is the probability distribution of y , P(x) is the probability distribution

of x, and dy /dx is the derivative of y with respect to x. For example,
1 −1
y= and dy /dx = 2
x x
implies
P ′(y) = x 2P(x) (15.8)
If the PRF is uniformly distributed between 1,000 and 1,100 pulses per second,
P(x) is 0.01/Hz for 1,000 < x < 1,100 Hz. The distribution of the PRI is then
P ′(y) = x 2P(x) or (0.01)(PRF)2 sec −1 (15.9)
These two distributions are shown in Figure 15.9. The mean PRF is 1,050 Hz,
whose reciprocal is 952.3 ␮ s. The mean PRI, however, is 953.1 ␮ s, a difference
of 0.8 ␮ s. In other words, the statistical mean of the PRI is not the reciprocal of
the mean PRF: the relationship depends on the distribution. The calculation is as
follows:
1,100
冕冋册
1,100
0.01x 2
x= 0.01x dx = = 1,050 (15.10a)
2
1,000
1,000
Figure 15.9 A PRF histogram transformed to a PRI histogram.

15.7 Histogram Analysis Summary 381
1 1
= ≅ 952.381 ␮ s (15.10b)
x 1,050
1/1,000
y= 冕 0.01
y 2
y dy = 0.01 冋log (y)册1/1,100
1/1,000
(15.10c)
1/1,100
y = 0.01(−6.90776 + 7.003063) = 953.102 ␮ s (15.10d)
Note that 1/953.102 ␮ s = 1,049.206 Hz.

If the raw intercept data is expressed as PRFs and if the analyst wishes to use
PRI in his report, the question of accuracy arises. If the accuracy of the PRF is
±⌬F (Hz), then the accuracy of the PRI will be
⌬F
± ⌬T = ⫿ (15.11)
F2
(This is obtained by differentiating T = I /F.) If the PRF reports are accurate to

±1 Hz and if the average PRF is 1,000 Hz, then the PRI will be accurate to ±1 ␮ s:
⫿1 Hz
± ⌬T = = 10−6 sec (15.12)
1,000 Hz2
The PRI accuracy is reported as 1 ␮ s even though the reported PRF accuracy
is 1 Hz.
Notice that a given fractional PRF error results in the same fractional PRI
error:
⌬T ⌬F
=− (15.13)
T F
15.7 Histogram Analysis Summary
The following rules apply to histogram analysis:
1. Histograms require 10 to 100 points per bin for reliability (the more the
better).
2. Several bin sizes must be tried to find a good representation of the PDF
(keeping in mind rule 1 as the number of bins increase).
3. Conditional histograms are needed to investigate modes of operation (i.e.,
which PRIs go with which pulse durations?). Other uses include examining
only high-accuracy reports or reports from a given geographic area or
investigating parameter changes over time.
4. The accuracy values associated with the intercept data should be reported
to users rather than the statistical quantities or histogram bin sizes.
5. Comments on the number of intercepts, the shape of their histogram, and

any significant peaks or valleys should be provided for the user.
6. Because different users have different needs, the more information provided,
the more likely that the users’ needs will be met.
7. Try to verify that intercept reports have properly identified the signal. A
few bad values can have a significant effect (see rule 8).
8. Statistical values (particularly the standard deviation) are significantly
affected by outlying values. The nonparametric approach to finding limits
avoids this problem.
9. Transforming reported parameter values to another form (e.g., PRI =
I /PRF) must be done correctly.
References
[1] Mood, A. M., Introduction to the Theory of Statistics, New York: McGraw-Hill, 1950,
Chapter 16.
[2] Jordan, E. C., (ed.), Reference Data for Radio Engineers, Indianapolis, IN: Howard W.
Sams and Company, 1977, Chapter 42.
CHAPTER 16
ELINT Data Files
16.1 Introduction
The users of ELINT are primarily concerned with emitter identification (as in radar
warning receivers), and with countering the emitter once it is recognized. Other
uses include driving simulation systems used in training. For identification, the
interceptor makes use of the ELINT parameters and the parameter limits expected
for each type of emitter as discussed in Chapter 15. For designing an electronic
attack to counter an emitter, it is necessary to have more detailed information
about how the radar operates and especially how the receiver processes signals.
This is because only receivers are jammed—not transmitters. For this, detailed
technical information is needed and not all of it is available from ELINT. This
gives rise to dividing the information into observed and assessed categories.
Observed data is largely from ELINT. Assessed data is an analyst’s best estimate
of how the system performs based on all information available at the time and
including engineering calculations and the limitations due to target motion.
ELINT can be divided into technical information (TECHELINT) and opera-
tional information (OPELINT) as shown in Figure 16.1. This book is mainly
concerned with technical ELINT. But note that emitter location, an OPELINT
parameter, makes use of highly technical signal processing techniques. Of course,
data files of non-ELINT data are also maintained and used. SIGINT other than
ELINT includes the data listed in Figure 16.2. Non-SIGINT intelligence files are
listed in Figure 16.3.
Figure 16.1 Technical and operational ELINT.
383
384 ELINT Data Files
Figure 16.2 Types of non-ELINT SIGINT data.
Figure 16.3 Types of non-SIGINT intelligence data.
OPELINT is concerned with answering questions such as:
• What? This information is contained in the ELINT notation and this identi-
fies the type of emitter.
• When? This information is contained in the time of intercept.
• Where? This information is obtained by emitter location.
• Who? This information is obtained by correlating the what, when, and where
information with the ‘‘Order of Battle’’ (OB) databases. The electronic order
of battle (EOB) lists emitters by type (or ELINT notation) and their locations.
The ground emitters are contained in the ground EOB. It tells which military
units have a particular type of emitter. For ship-borne and airborne emitters,
the naval and air orders of battle contain information about the types of
ships and their home ports and the types of aircraft and their units and
home airfields. The electronic fit provides information about which electronic
equipment is carried by each type of aircraft, ship, or ground unit.
Uses of OPELINT include adjusting ESM/RWR emitter identification logic,

planning penetration routes, monitoring the readiness of enemy forces, and discern-
ing enemy tactics.
Both TECHELINT and OPELINT require databases that provide a repository
of the parameters that describe each signal. There are so many parameters that
may be used to describe a signal, and so many different ELINT signals to describe,
that it becomes a problem to maintain up-to-date and accessible files suitable for
16.2 Signal Identification 385
the several users of such data. To help manage this situation, it is natural to try
to use computerized databases. Such databases can also help point out gaps and
shortcomings in the data—in effect, blanks in the data file that should be filled or
parameter values with considerable uncertainty about them. Each chapter in this
book that deals with an ELINT parameter might be thought of as an outline for
a list of parameter names that should be accompanied by measured values that
describe the signal
For each parameter to be included in the file, there is a need to indicate the
lower and upper limits that cover 95% of the emitter population for that signal.
There should also be an indication of the accuracy to which the parameter is
measured. For each value entered, there should be a reference to a technical report
or other data containing the information so that the interested user can learn more
details about the parameter value, measurement methods, and circumstances. In
addition, there is a need to key the data entered to a textual commentary that is
generally used to further explain or qualify the data.
To be useful, an ELINT data file must also provide a way to show how the
various parameters operate together—which PRIs go with which pulse durations,
or which RF channels are used for different functions. It is also necessary to indicate
any relationships that may exist between this signal and other emitters in the file
(e.g., a height finder that is always collocated and synchronized with a certain
search radar).
Another important requirement is that the data file must contain different kinds
of parameters for different kinds of signals. Airborne pulse Doppler radars require
different parameters than over-the-horizon search radars; data links require differ-
ent parameters than an air traffic control radar, and so on.
Finally, the data file must be useful for people interested in different levels of
detail. A jammer designer has different information needs than those of a radar
warning receiver designer, who has different information needs than those of the
intercept receiver operator, who has different information needs than those of the
mission planner, and so on.
16.2 Signal Identification
One of the critical uses of the data files is to provide a guide to the interceptor to
allow him to name the signal. The interceptor makes basic field measurements of
the fundamental ELINT parameters such as RF, PRI, pulse duration, and scan time
or rate.
From these and other indications (such as PRI modulation or mode switching
or an associated signal), the interceptor tries to determine which known signal this
might be. This is done by comparing the observed parameters from the intercept
with those listed in the interceptor’s identification guide. Usually, the starting point
is a list in RF order of all known signals. The observed RF may fall between the
lower and upper limits of the RFs listed for a number of signals. For each of these
signals, the other parameters are compared. If the interceptor finds a set of param-
eter limits that includes all those he observed, he concludes that the signal
being intercepted can be so identified. At this stage, the interceptor attaches to his
measured values of ELINT parameters the alphanumeric code from the identifica-
tion guide that identifies that signal and then submits the report to the ELINT
analyst responsible for analyzing all of the reported intercepts of that signal. The
identification step is critical Identification must be verified by analysts in order to
keep errors out of the estimates of the parameter limits (which, of course, will be
used for creating the next edition of the identification guide).
Clearly, the identification guide parameter limits may be different from the
true parameter limits for any particular radar. This is because the limits must
include not only the variations observed across the emitter population but also the
variations in the calibration errors across the population of intercept receivers.
Also, the identification guide may omit a number of parameters that are not
observable at the intercept sites. For instance, RF coherence or stability measures
may not help identify a signal and may not be measurable except under certain
conditions. As intercept equipment became more accurate and could be automati-
cally calibrated, these differences between the measured and true population param-
eter values became less pronounced than when the equipment was more primitive.
Proper signal identification is perhaps the most critical step in the ELINT
process.
16.3 ELINT Data for Radar Warning Receiver Design
The function of an RWR is to warn of a threat (usually to the aircraft on which

it is mounted) from some radar-based weapon. In some ways this is similar to
the signal identification problem faced by the intercept operator, yet there are
fundamental differences in the information needed from ELINT data files. One of
the main differences is that the RWR must reach a decision automatically and in
a very short time. Also, because it is generally mounted in military aircraft (designed
for combat or weapon delivery), it must be much smaller than the typical intercept
gear used in reconnaissance aircraft. On the other hand, it need only be concerned
with threats to that particular aircraft and only those that may be found in its
theater of operation.
Modern RWRs are generally reprogrammable so that their threat libraries and
the parameter limits contained in them can be altered as the threat changes or as
the aircraft mission changes. Clearly, the parameters of most interest depend on
the particular design of the RWR to be reprogrammed. If it is based on a wide-
open (crystal-video) receiver, there will be little interest in RF limits except to
ensure that a threat is in one of the bands covered by the RWR. On the other
hand, the PRI and PRI modulation parameters will be crucial to the threat identifica-
tion process. For a staggered PRI, for example, the period of the stagger (or stable
sum) can be used to distinguish a threat from a nonthreat radar. There have been
dozens of RWR systems deployed and tens of thousands manufactured. Therefore,
they represent one of the major uses of ELINT products.
The following example helps to show how ELINT data might be used to
program an RWR to identify threats. First, a crystal-video receiver provides the
pulse train envelope to a processor that examines the pulse duration and PRI. The
threat identification process first requires that the pulse duration be tested against
16.3 ELINT Data for Radar Warning Receiver Design 387
a threshold to divide the signals into long and short pulse durations. For example,
all pulse durations above 1 ␮ s would be in one class and all those below 1 ␮ s
would be in the other class. For each of these pulse duration classes, a PRI table
similar to Figure 16.4 would be constructed. This is constructed by using the PRI
upper and lower limits from ELINT and arranging them in order of the lower
limits. The x-axis is the PRI value and the y-axis is an arbitrary scale representing
the emitter identification number. For this example, emitter 1 has a conventional
(nominally constant) PRI with a lower limit of 40 ␮ s and an upper limit of 130 ␮ s.
The processor also checks for stagger (emitter 3) and jitter (emitter 6) and notes
the presence of these modulations. It might also measure the stagger period.
If a PRI of 100 ␮ s is observed, the RWR identifies it as emitter 1. If it finds
a PRI between 130 and 180 ␮ s, the signal is ignored as a nonthreat. Between 180
and 220 ␮ s, emitter 2 is identified. Between 220 and 250 ␮ s there is confusion
between emitters 2 and 3, but this can be resolved by noting whether or not stagger
is present. Between 250 and 280 ␮ s, there is confusion among emitters 2, 3, and
4. However, the presence of stagger identifies the signal as emitter 3.
To distinguish between emitters 2 and 4, some other parameter must be used.
One possibility is to add a tunable narrowband superheterodyne receiver that can
measure RF. This is too slow to search for signals. But because the possibilities
have been narrowed to just two emitters, the superheterodyne receiver can be
directed to search only the RF range used by one of the emitters. If it is found,
then the RF limits of the other emitter are also checked in case both signals are
present. Then the identification process is complete—unless the RF limits also
overlap and the signal RF is found to be inside the overlap range. This implies the
existence of an RF sort plot similar to Figure 16.4 within the identification proces-
sor. If the RF search fails to break the tie and identify the emitter, another parameter
such as a scan characteristic or the simultaneous operation of an associated emitter
may be used. The logic and associated sorts by parameter value are often called
the emitter identification (EID) table.
Figure 16.4 Emitter identification through PRI sort.

The construction of a good EID table is like solving a puzzle whose clues are
embedded in the ELINT data files. Because the ELINT analyst never knows what
little quirk of the emitter might be used to identify it, it is important to try to put
as much information as possible in the database concerning everything that can
be observed about the emitter. Likewise, different RWRs will make use of different
ELINT parameters, depending on how each RWR operates. Note that the PRI sort
is similar to inverting the ELINT data file so that instead of entering the emitter
identification and finding the PRI limits, one enters the PRI limits and determines
the possible emitter identifications for that PRI.
16.4 ELINT Data for Simulation and Training
Using ELINT data for simulation and training is different in that the parameter
limits themselves are not used. Instead, to simulate a specific emitter requires that
particular values within the parameter limits be selected. For example, one might
use a random number generator to select an operating radio frequency from within
the allowable range of the emitter type being simulated. In this case, simulation
of a realistic radar environment containing a number of emitters of the same type
requires that the distribution of the RF assignments among this class of emitters
be known.
16.5 Adding Non-ELINT Data
To refine the data available for applications like threat warning and jamming
applications, the observed ELINT data are augmented by any other information
that may be available from sources such as photographs, manuals, and interviews
with designers or operators. This might result in wider limits than are observed
by ELINT and the addition of other parameters not available from ELINT. For
example, jammers operate against the radar’s receiver, not its transmitter. The
receiver parameters are not directly observable from ELINT. Some important exam-
ples include the receiver bandwidth and the servo characteristics of range, angle,
and velocity tracking loops [1]. Such non-ELINT data can be added in the same
format and database organization as the ELINT data. A file that includes this non-
ELINT data and also includes the assessments of the ELINT observable parameters
by systems experts is a better source of the information needed for many applications
than an ELINT-only file.
While it would be helpful if all of this information could be stored, retrieved,
and manipulated by computers, it is usually necessary for the user to examine the
data manually, at least to some extent. This happens because there are always
important details that do not quite fit into the established data file entries, or else
it is necessary to explain some important aspect of the system operation in sentence
form. So far, the user of the file has no option but to read and understand these
sentences. All of these databases are computerized and automated, but interpreting
the data in the files requires human intellect.
16.6 Summary 389
16.6 Summary
Computerized databases are used to create many ELINT products. These include:
1. An identification guide for interceptors;

2. A comprehensive technical parameter file;
3. An augmented file that includes non-ELINT data, especially data concerning
receivers.
In addition to these products, there are data files to retain the intercept reports.
These require a standardized reporting format so that any intercept report can be
handled, regardless of who generates it. These files would be used to create the
histograms (described in Chapter 15) used to define the parameter limits.
Clearly as radar evolves, the data files used to describe their signal must also
evolve. The new emphasis on multifunction radar and on electronically steered
arrays means that the data files must include more capabilities to describe and
catalog the signals from such radar systems.
These data files keep the ELINT data cycle working. Because this book is
concerned with ELINT analysis, the starting point to describe the cycle is the
analyst. The analyst reports results to various users. The users, as well as the
analyst, determine the information that is lacking. These information needs are
passed on to those who establish ELINT interception requirements (i.e., those who
let the intercept operators know what signals to search for, what parameters to
measure, and how accurately the measurements should be). These requirements
direct the intercept site operators to try to obtain certain specific intercept data,
which results in intercept reports arriving back to the analyst. This completes the
ELINT cycle—and this book.
Reference
[1] Lothes, R. N., M. B. Szymanski, and R.G. Wiley, Radar Vulnerability to Jamming,
Norwood, MA: Artech House, 1990.
APPENDIX A
Spectrum Widths: 3-dB and First Nulls

for Trapezoidal Pulses
A.1 Introduction
The width of the spectrum of a simple pulsed radar signal is largely determined
by the pulse shape. ‘‘Simple’’ means that there is no intrapulse FM or phase
modulation for pulse compression purposes. Spectrum width is conventionally
measured at the half power points, while pulse duration is measured at the half
voltage points.
A.2 Rectangular Pulses
For a rectangular pulse, the power spectrum has a sin2 x /x 2 shape. In this case,
the 3-dB points are 2.78 units apart, while the first nulls (which occur at x = ±␲ )
are 2␲ units apart. Therefore, the ratio of the width between the first nulls to the
3-dB spectrum width is 2.27. This means that the null-to-null spectrum width is
2.27 times the 3-dB bandwidth.
It should be noted that the peaks of the spectrum sidebands decrease in propor-
tion to 1/f 2 for frequencies separated by f from the carrier. The first nulls occur
at a frequency of plus and minus the reciprocal of the pulse duration away from
the carrier frequency. Therefore, the null-to-null spectrum width = 2/PD and the
3-dB spectrum width = 0.88/PD.
A.3 Trapezoidal Pulses: Equal Rise and Fall Times
A trapezoidal pulse with equal rise and fall times can be thought of as a narrow
pulse of width ␶ convolved with a wider pulse of width T. This produces a total
rise time of the composite pulse of ␶ (this is also the fall time), as shown in Figure
A.1.
The spectrum of the trapezoidal pulse is simply the product of the spectra of
each individual pulse, one of width ␶ and one of width T. Therefore, the combined
spectrum is
␶ 2 sin2 (␲ f␶ ) T 2 sin2 (␲ fT )
| S 2( f ) | = ⭈ (A.1)
(␲ f␶ )2 (␲ fT )2
391
392 Spectrum Widths: 3-dB and First Nulls for Trapezoidal Pulses
Figure A.1 Trapezoidal pulse produced by convolution of two rectangular pulses (␶ < T ).
If ␶ Ⰶ T, then the first factor on the right is near 1 at the frequency for which
the second factor is 0.5 (i.e., the 3-dB point). For example, if ␶ = 0.1T, then the
frequency at which the product on the right is 0.5 is still approximately 0.88/T
and the null-to-null spectrum width is still approximately 2.27 times this. Note
that the frequency at which the first null occurs is still the frequency at which the
second factor on the right goes to zero (i.e., f = 1/T ). This holds for all rise time
values such that ␶ < T. For ␶ = T, the pulse is triangular and the spectrum becomes
T 4 sin2 (␲ fT )
| S 2( f ) | = (A.2)
(␲ fT )4
For this case, the 3-dB bandwidth is 0.64/T and the null-to-null spectrum width
is still at f = 2/T, but this is now 3.13 times the 3-dB bandwidth.
For the case ␶ = T /2, the 3-dB bandwidth is determined by the frequency for
which
(T /2)2 sin2 (␲ fT /2) T 2 sin2 (␲ fT )

| S 2( f ) | = ⭈ = 0.5 (A.3)
(␲ fT /2)2 (␲ fT )2
This occurs approximately at f = 0.4/T, so the 3-dB bandwidth is 0.8/T and

the null-to-null frequency separation is 2.5 times the 3-dB bandwidth. This is
summarized in Table A.1 for pulse width T.
A.4 Trapezoidal Pulses: Unequal Rise and Fall Times 393
Table A.1 Rise and Fall Times Versus the 3-dB Bandwidth and
Null-to-Null Separation
Rise/Fall Time 3-dB Bandwidth (B) Null-to-Null Separation
0 0.88/T 2.27B
0.1T 0.88/T 2.27B
0.2T 0.87/T 2.3B
0.3T 0.848/T 2.36B
0.4T 0.832/T 2.40B
0.5T 0.8/T 2.5B
0.6T 0.77/T 2.6B
0.7T 0.74/T 2.7B
0.8T 0.7/T 2.86B
0.9T 0.67/T 2.99B
1.0T 0.64/T 3.13B
A.4 Trapezoidal Pulses: Unequal Rise and Fall Times
Often the fall time is longer than the rise time (typically three times longer). In
this case, the spectrum is the spectrum of the central rectangular pulse plus the
spectra of the triangular pulses representing the rise and fall times. The triangular
pulses must be shifted in time to the beginning and end of the rectangular portion.
Also, the time axis must be inverted for the triangle representing the rising edge.
After manipulating the complex spectra to obtain the real and imaginary parts of
their sum, the absolute value can be computed to locate the 3-dB points and first
nulls. The complex spectrum of the sum is
T sin (␲ fT )
s( f ) =
(␲ fT )
t0
+ [cos (␲ Tf ) − 2␲ t 0 f sin d(␲ Tf ) − cos ␲ (T + 2␲ t 0 ) f ]
4(␲ t 0 f )2
jt 0
+ [−sin (␲ Tf ) − 2␲ t 0 f sin (␲ Tf ) + sin ␲ (T + t 0 ) f ] (A.4)
4(␲ t 0 f )2
t1
+ [cos (␲ fT ) − 2␲ t 1 f sin (␲ Tf ) − cos ␲ (T + 2t 1 ) f ]
4(␲ t 1 f )2
jt 1 f
+ [sin (␲ Tf ) − 2␲ t 1 f sin (␲ Tf ) − sin ␲ (T + 2t 1 ) f ]
4(␲ t 1 f )2
Here T is the duration of the rectangular portion, t 0 is the total rise time, and
t 1 is the total fall time. Note that the time between the 50% points of the pulse is
PD(50%) = T + 0.5 (t 0 + t 1 ) (A.5)
This result can be compared with that obtained above for equal rise and fall
times. For t 1 = t 0 , the imaginary portions cancel and the real portions add, resulting
in
394 Spectrum Widths: 3-dB and First Nulls for Trapezoidal Pulses
T sin ␲ TF t0
s( f ) = + [cos (␲ Tf ) − 2␲ t 0 f sin ␲ Tf − cos ␲ (T + 2t 0 ) f ]
␲ TF 2(␲ t 0 f )2
(A.6)
Combining terms yields
sin [␲ f (T + t 0 )] sin (␲ t 0 f )
s( f ) = (A.7)
␲f ␲f
Notice that this has the same form as the equation for equal rise and fall times,
which was derived using convolution in Section A.3. The 50% pulse duration in
this case is the width of the rectangular portion plus half of the rise and fall times.
In other words, in this analysis T + t 0 is the 50% pulse duration, whereas in Section
A.3, T is the 50% pulse duration.
For unequal rise and fall times, the first minimum of the spectrum occurs very
close to the reciprocal of the 50% pulse duration on each side of the center
frequency. However, this minimum is not strictly a null. In other words, the
spectrum does not actually drop to zero. The situation is summarized in Table A.2
for a 50% pulse duration T.
The true spectral nulls are quite far removed from the center of the spectrum—
on the order of 12 to 16 times the 3-dB bandwidth. For practical trapezoidal pulses,
the separation of the first minima is a more realistic measure of the spectral width
than the separation of the first nulls. As the rise and fall times become short and/
or as the pulse shape becomes more symmetrical, the depth of the minima
approaches that of a true null.
Table A.2 Unequal Rise and Fall Times for a 50% Pulse Duration
3-dB Bandwidth Separation of Depth of
Rise Time Fall Time (B) First Minima First Minima
0.1T 0.3T 0.87/T 2.3B −34.3 dB
0.2T 0.6T 0.82/T 2.44B −23.2 dB
0.3T 0.9T 0.75/T 2.67B −18.5 dB
0.01T 0.5T 0.85/T 2.38B −24.7 dB
APPENDIX B
Some ELINT Considerations of FM

Signals
John F. Green, Ph.D.
B.1 Introduction
In some ELINT applications, there is a need to collect and process signals in a

manner that enables their high-fidelity demodulation. One of the most critical
parameters is the minimum acceptable RF collection bandwidth. It is shown here
that Carson’s rule, used in communications signal engineering, can be modified
appropriately to provide a relevant ELINT specification.
This modification of Carson’s rule [1] is derived from an analysis of the spec-
trum of circular (sinusoidal) frequency-modulated signals and from an assessment of
the influence of individual spectral features on demodulated signals. A quantitative
evaluation is made of the degradations imposed on such signals by excessive RF
spectral band limiting.
Finally, a brief derivation is presented, the results of which describe the effects
of weak and moderate Gaussian noise on frequency-demodulated signals.
B.2 Effects of Sinusoidal Interference on Phase- and Frequency-

Demodulated Signals
The intent of this section is to establish a quantitative basis for estimating the
effects of an individual sinusoidal interference on phase- and frequency-demodu-
lated signals.
If a sinusoidal interference is added to an arbitrary signal, the result can be
expressed in the form:
g(t) = cos [␾ s (t)] + ␥ cos (2␲ fo t) (B.1)
where fo is the frequency of the sinusoidal interference and ␾ s (t) is the total phase
of the signal, including the carrier. In (B.1), ␥ is taken to be a (local) constant. The
analytic signal components of the signal in (B.1) are
g I (t) = cos [␾ s (t)] + ␥ cos (2␲ fo t) (B.2)

g Q (t) = sin [␾ s (t)] + ␥ sin (2␲ fo t)
395
396 Some ELINT Considerations of FM Signals
The instantaneous phase, ␾ (t), of the composite signal defined by (B.2) is
g I (t)
tan [␾ i (t)] = (B.3)
g Q (t)
If (B.2) is introduced into (B.3), and ␥ is assumed to be small, then
tan [␾ i (t)] = 1 +冋 sin ␾ s cos ␾ s 册

␥ sin (2␲ fo t − ␾ s )
tan ␾ s (B.4)
For a small ␦␾ ,
tan (␾ s + ␦␾ ) ≈ 1 + 冋 ␦␾
sin ␾ s cos ␾ s 册
tan ␾ s (B.5)
By comparison of (B.4) with (B.5):
␾ i (t) ≈ ␾ s (t) + ␥ sin [2␲ fo t − ␾ s (t)] (B.6)
Equation (B.6) indicates that the effect of the weak additive tone on the instanta-
neous phase is to introduce a sinusoidal ripple, with its frequency equal to the
difference frequency between the two signals and amplitude equal to the ratio of
their amplitudes.
If the signal-to-interference power ratio, SIR, is introduced,
1
␾ i (t) ≈ ␾ s (t) + sin [2␲ fo t − ␾ s (t)] (B.7)
√SIR
It is necessary to emphasize that derivation of this description of additive tone
effects on the instantaneous phase has assumed that the tone has low power,
compared with the signal of interest. This description increasingly underestimates
the effects of this additive tone as its relative power level grows. However, for SIR
levels on the order of 20 dB or higher, (B.7) is a good approximation. If the SIR
is 20 dB, the total range of the phase ripple is about 11.5°.
The instantaneous frequency, fi (t), of the weak additive tonal composite signal
can be derived easily by direct time differentiation of (B.7).
1
fi (t) ≈ f s (t) + [ fo − fs (t)] cos [2␲ fo t − ␾ s (t)] (B.8)
√SIR
In (B.8), f s (t) is the instantaneous frequency associated with ␾ s (t). This result
shows that the effect of the weak additive tone is to cause a sinusoidal ripple of
the instantaneous frequency (the rate of which is the frequency difference between
the two signals), the same rate as affects the instantaneous composite phase.
Of greater importance, in some ELINT applications, is the amplitude of the
frequency ripple. Its magnitude is the product of the frequency difference times
B.2 Effects of Sinusoidal Interference on Phase- and Frequency-Demodulated Signals 397
the signal amplitude ratio. In effect, the phase ripple is amplified by the fre-
quency difference. For example, if the SIR is 20 dB and the frequency difference
is 20 MHz, the total range of the frequency ripple is about 4 MHz. If the SIR is
40 dB and the frequency difference is 10 MHz, the total range of the ripple is
about 200 kHz.
Figure B.1(a, b) shows several examples of the demodulated phase and fre-
quency signals, derived without the approximation from (B.3). This illustrates the
effects of sinusoidal interference. Close examination shows that the demodulated
signals, at an SIR level of 20 dB, have diverged only slightly from pure sinusoids.
Figure B.1 (a) Phase error caused by sinusoidal interference. (b) Frequency error caused by sinusoi-
dal interference.
Note that the phase of the 0-dB SIR output signal is linear and the frequency is
constant, yielding only a beat frequency amplitude modulation.
These results indicate that the instantaneous frequency can be perceptibly
altered by signal components having relative power levels as low as −40 dB within
typical ELINT RF and video signal processing bandwidths. The phase signal is
also degraded, but the degree of degradation is not sensitive to the frequency offset
of the interference.
B.3 Signals with Sinusoidal Frequency Modulation
The intent of this section is to describe the spectra of sinusoidally frequency modu-
lated signals. Such signals are of interest because their spectral analysis facilitates
derivation of Carson’s rule and its ELINT modification, which is presented in the
subsequent section.
The effects of extreme spectral band limiting of FM signals, with respect to
the degradation of their frequency-demodulated output, are evaluated in Section
B.6.
A sinusoidal FM signal, g(t), can be defined as
g(t) = cos [␻ o t + m sin (␻ m t)] (B.9)
where
␻ o = constant carrier frequency (B.10)

␻ m = constant modulation frequency
The modulation index, m, is defined as
␻d
m= (B.11)
␻m
where
␻ d ≡ constant peak radian frequency deviation (B.12)
The instantaneous radian frequency, ␻ i (t), associated with the signal, g(t), is
␻ i (t) = ␻ o + ␻ d cos(␻ m t) (B.13)
Equation (B.13) shows that the signal, g(t), is indeed frequency modulated in
sinusoidal fashion. By expanding the composite cosine, (B.9) can be rewritten as
g(t) = cos (␻ o t) cos [m sin (␻ m t)] − sin (␻ o t) sin [m sin (␻ m t)] (B.14)
From [2],
B.3 Signals with Sinusoidal Frequency Modulation 399
∞
cos [m sin (␻ m t)] = J0 (m) + ∑ J 2k (m) cos (2k␻ m t) (B.15)
k =1
and
∞
sin [m sin (␻ m t)] = 2 ∑ J 2k + 1 (m) sin [(2k + 1)␻ m t] (B.16)
k =0
where Jn (z) are Bessel functions of the first kind of order, n. After using (B.15)
and (B.16) in (B.13), one finds
g(t) = +J 0 (m) cos (␻ o t) − 2J1 (m) sin (␻ o t) sin (␻ m t) (B.17)

+ 2J 2 (m) cos (␻ o t) cos (2␻ m t) − 2J 3 (m) sin (␻ o t) sin (3␻ m t) + . . .
By multiple angle expansions, (B.17) can be rewritten as
∞
g(t) = J 0 (m) cos (␻ o t) + ∑ J k (m) cos [(␻ o + k␻ m )t] + (−1)k cos [(␻ o − k␻ m )t]
k =1
(B.18)
Equation (B.18) demonstrates that the spectrum of a sinusoidally modulated

FM signal is a discrete set of components, equally spaced at an interval of ␻ m , the
modulation frequency. The spectrum is centered at, and is symmetrical about, the
carrier frequency, ␻ o . The amplitude of a given spectral component, located at
the frequency offset (k␻ m ) from ␻ o is the Bessel function, J k (m).
In the case of weakly modulated signals, m is small and only the lowest order
Bessel functions have significant amplitudes. For such a case
g(t) ≈ cos (␻ o t) + 0.5m cos [(␻ o + ␻ m )t] − 0.5m cos [(␻ o + ␻ m )t] (B.19)
Equation (B.19) shows that the spectrum contains only the closest sideband,
and most of the signal power is concentrated at the carrier frequency, for weakly
modulated signals.
For more highly modulated signals, the spectrum becomes wideband, with
significant power occurring at considerable displacements from the carrier fre-
quency. This is demonstrated explicitly in the subsequent section. For certain values
of the modulation index, m, J 0 (m) = 0, causing the carrier frequency spectral region
to exhibit no signal energy.
Spreading the spectrum for comparatively highly modulated FM signals can
present a substantial problem to their high-fidelity ELINT capture and demodula-
tion. Ordinarily, it is advantageous to limit the processing bandwidth of ELINT
analysis systems to minimize inclusion of thermal and other noise and to minimize
the probability of cochannel interference. This, in turn, improves the detectability
of lower power signals.
However, the utilization of narrow bandwidths can seriously degrade the fidel-
ity of highly modulated FM signals. This degradation can be demonstrated easily
either by direct laboratory experimentation or by computer emulation. Examples

of the latter are provided in Section B.6. A highly practical and efficient criterion
for determining the critical minimal bandwidth necessary to avoid gross degradation
of FM signals is Carson’s rule, which is discussed in the next section.
B.4 Carson’s Rule
Power spectra of virtually all types of FM signals generally decrease in amplitude

with an increasing displacement in frequency from the carrier frequency. While
details of these spectra may be diverse and complicated, as demonstrated in the
subsequent section, observation shows that each spectrum tends to exhibit a more
or less well-defined break point, beyond which the spectrum decreases rapidly.
This break point, in the case of highly modulated CFM signals, occurs at a frequency
displacement equal to the deviation frequency.
For small amounts of frequency deviation, this break point frequency is at the
first sideband, a displacement equal to the modulation frequency. Both circum-
stances can be encompassed approximately by the specification that the minimum
RF bandwidth necessary to contain the bulk of the signal energy is equal to twice
the sum of the modulation and the deviation frequencies. This specification is
known as Carson’s rule, which has been found to be generally quite useful for
communications signal applications.
Carson’s rule can be derived from an examination of the spectra properties of
CFM signals. It was previously shown that the spectral components of CFM signals
consist of Bessel functions.
Under certain conditions, these functions can be approximated by the expres-
sion [2]
J k (m) ≈ 冉冊
m k 1
2 k!
(B.20)
In (B.20), m is the modulation index and k is the order of the spectral compo-
nent. Stirling’s approximation for k! is
k! ≈ 冉冊
√2␲ k e
k k
(B.21)
Combining (B.20) with (B.20) gives
J1 (m) ≈
1
冉冊
em k
√2␲ k 2k
(B.22)
The term on the right-hand side of (B.22), raised to the kth power, exhibits a
dramatic decrease in magnitude as k, the spectral component order (i.e., frequency
displacement), is increased somewhat beyond m, the modulation index. Although
B.5 Modification of Carson’s Rule for ELINT Applications 401
the Bessel function approximation is not ideal for m ≈ k, this roll-off in power is
described well by (B.22) and is shown in Figure B.2.
This implies that the spectral components, J k (m), should decrease rapidly in
amplitude if k > m. Because the spectral component, of order k, is displaced from
the RF carrier frequency by the frequency, k␻ m (where ␻ m is the FM modulation
frequency), the bulk of the spectral power should be confined within a (two-sided)
linear RF bandwidth, B, where
B ≈ 2m 冉冊
␻m
2␲
(B.23)
Equation (B.23) is deficient (in that it excludes the first order FM sidebands)
at the frequency displacement ␻ m , in the case of weakly modulated signals. Thus,
(B.23) is commonly modified to
B ≈ (2m + 1) 冉冊
␻m
2␲
= 2( fd + fm ) (B.24)
in which fd is the frequency deviation and fm is the modulation frequency.

Equation (B.24) is known as Carson’s rule.
B.5 Modification of Carson’s Rule for ELINT Applications
In Section B.2, it was shown that relatively very weak sinusoidal interference could
cause a perceptible degradation in frequency-demodulated signals. The magnitude
Figure B.2 Bessel function power as a function of modulation index.

of this degradation was found to be the SIR times the frequency offset of the
component from the carrier frequency. This theoretical result serves as a model
for modifying Carson’s rule by supplying rough estimates of the contribution to
the FM demodulation fidelity attributable to wideband spectral components (and
to their potential inadvertent removal by underestimating acquisition bandwidths).
For example, a demodulated signal degradation with a magnitude 200 kHz
was found for an additive tonal signal component having relative power of −40 dB
and frequency offset of 10 MHz. If 200 kHz is the maximum level of allowable
degradation, the constraint on the argument of the term raised to the kth power,
in (B.22) should be stronger than that used to derive Carson’s rule.
In the present case, if the component in question must be about −40 dB or
smaller, the argument should be not much greater than 1/4 (compared with 1 for
Carson’s rule). If this modified constraint is adopted, then
em 1
≤ → k ≥ 2em ≈ 5m (B.25)
2k 4
If the weak modulation (i.e., the first sideband only) case is accommodated,
as in (B.24), then the modification of Carson’s rule, for this ELINT application,
becomes
B = 5m (2fm ) + 2fm = 10fd + 2fm (B.26)
Such a modification of Carson’s rule should not be taken as a requirement to

be observed religiously and without contemplation, because it has been developed
for a specific case, namely 200-kHz degradation. In addition, it is potentially a lot
shakier than Carson’s rule, given its high sensitivity to modest variations in the
specifically selected cutoff ratio.
To place this speculation on a stronger foundation, Table B.1 shows, without
approximation, the Bessel function order for various power thresholds as a function
of the modulation index. For this range of parameters, it appears that a safe guess
would be to multiply the Carson’s rule–derived bandwidth by a factor of about 3,
rather than 5, as indicated by (B.26). This modification of Carson’s rule, for certain
high-fidelity ELINT applications, has been motivated by an evaluation of the
influence of small spectral components on demodulated frequency signals. Often,
it is the demodulated phase, rather than the demodulated frequency, signal that is
of primary ELINT interest. The question arises as to the suitability of Carson’s
rule, and its modification, for use with phase-demodulated signals.
Table B.1 Bessel Function Order Required for a Given Threshold Below the Peak of the
Spectrum
Modulation Index
Threshold 0.2 0.4 0.6 0.8 1.0 2.0 4.0 6.0 8.0 10.0
−6 dB 1* 1* 1* 1* 1 4 5 8 — 11
−40 dB 2 3 3 4 4 6 8 12 — 16
−60 dB 3 4 4 5 5 7 10 14 — 18
* = required to maintain first sidebands.
It is easily shown that the time differentiation transfer function is 2␲ jf, where
f is spectral frequency. This transfer function can be used on the phase spectrum
to generate the frequency spectrum because they are related directly by the time
derivative. Because this type of transfer function increasingly emphasizes those
phase spectral components, with an increasing displacement from the carrier, it is
clear that the phase spectrum must, in general, be more localized than its corre-
sponding frequency spectrum. Thus, Carson’s rule is quite suitable for use with
phase-demodulated signals. Indeed, it represents an overly severe criterion for phase
demodulation.
Another related effect that must be considered in the selection of ELINT signal
minimum collection and processing bandwidths for FM signals is that of asymmetric
pass band filtering. For example, assume that the spectrum of a weakly modulated
circular FM signal contains only the carrier component and the nearest sidebands.
If this signal is asymmetrically filtered in such a manner as to eliminate one of the
two sidebands, the resultant signal has its modulation index reduced by a factor
of 2. The corresponding demodulated FM signal would be decreased, in deviation,
by a factor of 2.
Generally, asymmetric filtering of the spectrum of a frequency-modulated signal
tends to reduce and distort the original FM signature. The magnitude of the conse-
quent degradation can often be estimated theoretically by variations of the approxi-
mation techniques presented here.
To accommodate all of the effects discussed concerning circular and other
frequency-modulated signals, it is reasonable to select collection and processing
bandwidths that are:
1. At least three times that implied by Carson’s rule;

2. At least three times any linear frequency sweep;
3. Wide enough to avoid asymmetric signal filtering.
B.6 FM Demodulation Degradation by RF Band Limiting
This section develops a quantitative evaluation of the degradation caused by RF

spectral band limiting to the nearest sideband of signals having sinusoidal frequency
modulation. In the case of weakly modulated signals, the index of modulation m
is small and only the lowest order Bessel functions have significant amplitudes.
Equation (B.19) shows that the spectrum contains only the closest sideband
for weakly modulated signals and that most of the signal power is concentrated
at the carrier frequency. If the higher order sidebands have very low relative power,
RF band limiting of this signal, to include only the nearest sideband, results in
only a little degradation of the resultant demodulated frequency signal. However,
the derivations in the previous sections have demonstrated that even weakly fre-
quency modulated signals deposit enough energy in the higher order sidebands to
perceptibly affect the fidelity of the resultant demodulated signals.
To gain an initial theoretical estimate of the degradation caused by bandpass
filtering to retain only the RF carrier and first order sidebands, assume such filtering
of a sinusoidal FM signal having an arbitrary modulation index, m. From (B.18),
1 ␭ ␭
g(t) = cos (␻ o t) + cos [(␻ o + ␻ m )t] − cos [(␻ o − ␻ m )t] (B.27)
J 0 (m) 2 2
where
J1 (m)
␭ (m) = 2
J 0 (m)
Equation (B.27) reduces to the form
1
g(t) = cos (␻ o t) − ␭ sin (␻ o t) sin (␻ m t) (B.28)
J 0 (m)
An examination of (B.28) reveals that the quadrature signal has the form
1
g(t) = sin (␻ o t) − ␭ cos (␻ o t) sin (␻ m t) (B.29)
J 0 (m)
After using (B.28) and (B.29) to extract the demodulated frequency signal,
cos (␻ m t)
␻ i (t) = ␻ o + ␭␻ m (B.30)
1 + ␭ 2 sin2(␻ m t)
where ␻ i (t) is the instantaneous frequency. This is to be compared with the actual
(unfiltered) instantaneous frequency, described by (B.13).
For a small modulation index, m, ␭ (m) = m, so that the narrowband-filtered
signal frequency and the actual instantaneous frequency are nearly identical. For
larger values of the modulation index, a considerable degradation in the frequency
deviation can result from the narrowband filtering. As was noted in Section B.3,
the zero-th order Bessel function is zero for certain (physically achievable) values
of the modulation index. In these cases, ␭ (m) becomes infinite, demonstrating that
narrowband-filtered CFM signals can generate uncontained demodulated frequen-
cies.
The mean square expectation of the sinusoidal instantaneous frequency devia-
tion about the carrier frequency, as computed from the actual signal, (B.13), is
1 2 2
〈[␻ i (t) − ␻ o ]2〉 = m ␻m (B.31)
2
In the case of the narrowband-filtered signal, the mean square expectation is

found, from (B.30), to be
1 ␭2
〈[␻ i (t) − ␻ o ]2〉 = 2
␻m (B.32)
√1 + ␭
2 2
A comparison of (B.30) with (B.13) for the demodulated frequency, and (B.32)
with (B.31) for the mean square demodulated frequency, indicates that narrowband
filtering of CFM signals can indeed cause serious degradation.
Figures B.3 and B.4 show examples of CFM signal magnitude spectra and
demodulated frequency signals for a variety of modulation indexes. Provided in
each plot of demodulated frequency are results obtained by bandpass filtering to
the first sideband only and the results with no bandpass filtering. The induced
distortions can be seen to be extreme where the modulation index is in the neighbor-
hood of zeroes of Jo (m), the carrier frequency spectral contribution.
Table B.2 lists the ratios of the maximum deviation frequency and rms frequency
deviation for the bandpass limited to unfiltered CFM signals for a range of modula-
tion indexes.
Figure B.5(a, b) shows graphs of the ratios of the maximum deviation frequency
and RMS frequency amplitude for bandpass limited to unfiltered sinusoidal FM
Figure B.3 (a) Effects of RF band limiting on demodulated sinusoidal FM signal (m = 0.5).
(b) Spectrum of sinusoidal FM signal (m = 0.5). (c) Effects of RF band limiting on
demodulated sinusoidal FM signal (m = 1.0). (d) Spectrum of sinusoidal FM signal
(m = 1.0).
Figure B.3 (continued.)
Figure B.4 (a) Effects of RF band limiting on demodulated sinusoidal FM signal (m = 1.5).
(b) Spectrum of CFM signal (m = 1.5). (c) Effects of RF band limiting on demodulated
sinusoidal FM signal (m = 2.0). (d) Spectrum of sinusoidal FM signal (m = 2.0).
Figure B.4 (continued.)
signals for a range of modulation indexes. These theoretical results can be used to
specify the required minimum ELINT signal collection and analysis bandwidths
necessary to maintain a desired degree of demodulated signal fidelity. In circum-
stances where detailed spectral information is not available before signal collection,
it is recommended that the collection be made using the widest bandwidth feasible.
Table B.2 Ratios of the Maximum Frequency Deviation and RMS

Frequency Deviation for Signals that Are Band Limited to the
First Sideband, Compared to Unfiltered Sinusoidal FM Signals
Modulation Index Frequency Peak RMS Frequency Deviation
0.0 1.0 1.0
0.2 1.01 1.0
0.4 1.02 0.98
0.6 1.05 0.96
0.8 1.09 0.95
1.0 1.15 0.93
1.2 1.24 0.92
1.4 1.37 0.93
1.6 1.56 0.95
1.8 1.90 1.01
2.0 2.58 1.12
2.2 4.58 1.44
2.4 172.85 8.49
2.6 3.74 1.20
2.8 1.58 0.74
3.0 0.87 0.52
3.5 0.21 0.19
4.0 0.08 0.08
4.5 0.32 0.24
5.0 0.74 0.38
5.5 18.14 1.82
6.0 0.61 0.31
6.5 0.18 0.15
7.0 0.00 0.00
7.5 0.14 0.11
8.0 0.34 0.20
8.6 1.53 0.42
9.0 0.60 0.26
9.5 0.18 0.13
10.0 0.04 0.03
B.7 Effects of Noise on FM Demodulated Signals
Assume that a signal, A cos [␾ s (t)], is degraded by additive, band limited, white
Gaussian noise, n(t). The composite of signal plus noise can then be denoted as
g(t), where
g(t) = A cos [␾ s (t)] + n(t) (B.33)
The total noise (circular) bandwidth is denoted as ␻ N . The noise power spectral
density is denoted as ␮ so that the total noise power, N, is
N = ␮␻ N (B.34)
The noise can be resolved into two mutually orthogonal components, n I (t)
and n Q (t), which are instantaneously in phase and in quadrature, respectively,
with the signal. Equation (B.34) can then be rewritten as
g(t) = [A + n I (t)] cos [␾ s (t)] + n Q (t) sin [␾ s (t)] (B.35)

Figure B.5 (a) Ratio of the maximum frequency deviation for signals that are band limited to the first
sideband, compared to unfiltered sinusoidal FM signals. (b) Ratio of the rms frequency
deviation for signals that are band limited to the first sideband, compared to unfiltered
sinusoidal FM signals.
Both n I and n Q are also Gaussian distributed, having the spectral power density,
␮ . Equation (B.35) can be expressed in the form
g(t) = a(t) 再冉 A + n I (t)

a(t) 冊cos [␾ s (t)] +
n Q (t)
a(t)
sin [␾ s (t)] 冎 (B.36)
where
√[A + n I (t)]
2 2
a(t) = + n Q (t) (B.37)
A phase angle, ␾ n (t), can be defined, where
A + n I (t)
cos [␾ n (t)] ≡ (B.38)
a(t)
n Q (t)
sin [␾ n (t)] ≡ (B.39)
a(t)
Equations (B.38) and (B.39) are used in (B.36); one finds
g(t) = a(t) {cos [␾ s (t) + ␾ N (t)]} (B.40)
If the noise is weak, compared with the signal amplitude, A, then (B.37), (B.38),
and (B.39) can be combined to yield
n Q (t)
␾ n (t) ≈ sin [n Q (t)] = (B.41)
A
The instantaneous frequency, ␻ i , is defined as
d␾
␻ i (t) ≡ (B.42)
dt
If (B.4l) is introduced into (B.42), then
1 dn Q (t)
␻ i (t) = ␻ s (t) − (B.43)
A dt
in which ␻ s (t) is the instantaneous frequency of the signal, A cos [␾ s (t)].

In the frequency domain, the transfer function for the differentiation, d( )/dt
= j␻ . Consequently, the noise term in (B.43) possesses a spectral noise power
density, p(␻ ), where
␮␻ 2 ␻
p(␻ ) = 2
for | ␻ N | ≤ N (B.44)
A 2
Equation (B.44) reveals that the noise is preferentially concentrated (i.e.,

increases quadratically) in the higher frequency regions of the spectrum of the
demodulated (video) signal. Consequently, if the demodulated FM (video) signal
is lowpass filtered (without significant information loss), much of the noise can be
eliminated. However, any such lowpass filtering process must be designed and
used with some caution, especially if the (video) signal of interest is wideband.
Suppression of signal information by the lowpass filtering process can substantially
degrade the content, appearance, and characteristics of the FM video signal.
An estimate can be obtained for the rms amplitude of the PM noise described
by (B.44). If the frequency noise amplitude is denoted as ␻ N , then
␻ N /2
冕冉冊
3
1 2␮ ␻N
〈␻ n2 〉 = ␮␻ 2 d␻ = (B.45)
A2 3A 2 2
−␻ N /2
Because the signal power S = A 2/2 and the noise power [from (B.34)]
N = ␮␻ N , (B.45) can be rewritten as
2
␻N
〈␻ n2 〉 = (B.46)
24 SNR
in which SNR = S /N = A 2/(2N). Thus,
␻ ␻
√〈␻ n2 〉 = √24 SNR
N
≈ 0.2
√
N
SNR
(B.47)
Converting (B.47) to frequency gives
BN
fn (rms) ≈ 0.2 (B.48)
√SNR
where
␻n
fn = (B.49)
2␲
␻N
BN =
2␲
Equation (B.48) shows that the effect of white noise on the frequency-demodu-
lated (strong) signal is directly proportional to the input RF bandwidth and inversely
proportional to the square root of the total input RF SNR. Note that the output
video bandwidth of the demodulated signal is exactly equal to one-half of the input
noise bandwidth because no video filtering has been imposed. The effects of lowpass
filtering of the demodulated video signal are considered next.
If the FM-demodulated video signal is lowpass filtered such that the resultant
maximum video (circular) frequency, ␻ N = 2␲ fv is less than ␻ N /2, then the rms
frequency noise is found to be
√
1 fv
fn (rms) = fv (B.50)
3 SNR B N
where
␻v
fv = (B.51)
2␲
Equation (B.50) demonstrates that there is a nonlinear gain in the output

demodulated signal video SNR by lowpass filtering, as long as no signal contribu-
tions are eliminated.
If the noise is moderate, rather than weak, the first order approximations of
(B.38), (B.39), and (B.40) must be extended to the higher order. If this is done,
the phase noise is found to be
nQ nQ nI
␾ n (t) ≈ − + (B.52)
A A2
After introducing (B.52), it is found that
fn (rms) = fv
√
1 fv
3 SNR B N 冋 1+
3
4 SNR
1−冉 3fv 3f 2
+ v2
2B N 5B
N
冊册 (B.53)
Equation (B.53) is reasonably valid for essentially all ELINT purposes for all
input SNR levels at or above 10 dB.
References
[1] Carson, J. R., ‘‘Notes on the Theory of Modulation,’’ Proc. of IRE, Vol. 10, February
1922, pp. 57–64.
[2] Abramowitz, M., and I. Stegun, (eds.), Handbook of Mathematical Functions, New York:
Dover Publications, 1972.
APPENDIX C
A Frequency Hop Radar Example
A hypothetical interception situation can serve to illustrate a probability of intercept

calculation without using the window function approach. Assume that there is a
radar whose radio frequency (RF) is changed randomly to one of N frequencies
on a pulse-to-pulse basis, and that we wish to intercept this signal and learn as
much as possible using a relatively slowly sweeping narrowband receiver. The
receiver sweeps a band, B, corresponding to the band over which the radar’s
frequency hops. The receiver instantaneous bandwidth ␤ is equal to the spacing
of the N radar frequencies (i.e., N ∼ B /␤ ). Suppose further that the receiver’s sweep
time is such that it stays open to receive a specified frequency for a time during
which R pulses are transmitted by the radar.
The frequency scanning receiver can receive signals at essentially one frequency
at a given time. If such a receiver is operating in the frequency range of a random
frequency hopping radar, certain information concerning this radar will be available
to the receiver. The calculations below predict the type and amounts of data that
would be expected.
The only pulses of interest to the scanning receiver are those occurring at the
frequency to which the receiver is tuned. Suppose the radar transmits on N discrete
frequencies (or the receiver resolves N frequencies). Assuming the receiver is tuned
to one of the N frequencies, then the probability that a particular pulse occurs at
the frequency to which the receiver is tuned is
p = 1/N (C.1a)
Of course, the probability that the pulse occurs at some other frequency is
q = 1 − p = (N − 1)/N (C.1b)
Utilizing these probabilities, some interesting calculations can be made. In the

examples that follow, it is assumed that the receiver scans through a particular
frequency in a time necessary for the transmitter to emit R pulses.
C.1 Probability That One or More Pulses Occur at the Desired

Frequency
One of the simplest and yet most interesting requirements necessary for an intercep-
tion is that at least one of the R pulses transmitted must occur at the frequency
413
414 A Frequency Hop Radar Example
(bin) to which the receiver is tuned. The easiest way to make a computation of
this probability is to calculate the probability that the observer receives zero pulses
and then subtract this probability from 1. Since the frequency of each pulse is
independent of the others, the probability that all R pulses go into some other bin
is
Po = q R (C.1c)
Then the probability of getting at least one pulse in the desired bin is
P = 1 − P0 = 1 − q r (C.2)
A plot of this function is shown in Figure C.1.

Equation (C.2) may be interpreted in an interesting manner. If the probability
of success on a trial is p, then a certain number of trials, R, will be needed to have
P probability of obtaining at least one success. Thus, if P and p are specified, the
number of trials needed may be calculated using (C.2).
C.2 Probability That Exactly One Pulse Occurs at the Desired

Frequency
If the event of interest was that ‘‘exactly one pulse occurred in the desired bin out
of the R pulses transmitted,’’ the probability of this event would be the probability
of obtaining one pulse in the desired bin times the probability of getting all the
other R − 1 pulses in some other bin times the number of ways (orders) in which
this can happen. (Since the first pulse could be of the desired frequency, and the
Figure C.1 Probability of receiving one or more pulses out of R at a particular frequency (N is the
number of discrete radar frequencies).
C.4 Probability That Several Pulses Occur at the Desired Frequency Less Than G Pulses Apart 415
rest of some other frequency, or the second could be of the desired frequency, and
so on, there are R ways to get exactly one pulse of the desired frequency.) Thus,
the probability of getting exactly one pulse of R in the desired bin is
P 1 = Rp(1 − p)R − 1 (C.3)
C.3 Probability That Exactly k Pulses Occur at the Desired Frequency
This reasoning can be extended to determine the probability of k pulses out of R

occurring in the desired frequency bin. The result is the familiar formula for the
binomial distribution:
P1 = 冉冊
R k
k
p (1 − p)R − k (C.4)
where
冉冊 R
k
=
k!
R!
(R − k)
Note that the expression for P k is recognized as the kth term of the expansion
(p + q)R, hence the name ‘‘binominal distribution.’’
C.4 Probability That Several Pulses Occur at the Desired Frequency

Less Than G Pulses Apart
For some scanning receivers, it might be useful to require that a minimum of two
pulses occur at the desired frequency before deciding that a valid interception had
occurred so that PRI measurements could be made.
In this case, a certain maximum time will be allowed between the two pulses.
This maximum allowable PRI will usually be some fraction of the total time the
receiver is open to a particular frequency. The number of pulses transmitted during
the maximum allowable PRI will be taken as G. Clearly, G is less than or equal
to R. In this case, the event of interest can be defined as the case when two or
more pulses out of R occur at the desired frequency and the pulses are separated
by fewer than G pulse intervals.
To illustrate the requirements for an event in this case, suppose that the pulses
are occurring at a steady rate and are numbered 1, 2, 3, . . . , R. In a total of R
pulses, an event occurs when two or more pulses arrive in the frequency bin to
which the receiver is tuned and when their numbers differ by less than G.
The probability of an event can be found by subtracting from unity the probabil-
ities of all the results of the R trials which do not cause an event.
The event cannot occur if the trial of R pulses ends with zero pulses in the
desired bin, and it cannot occur if the trial ends with only one pulse in the desired
bin. Thus, two terms that must be subtracted are P 0 and P 1 from the binomial
distribution, (C.4).
However, if two pulses occur at the desired frequency, there will be some cases
that will cause an event and some that will not. It is necessary to count the ways
in which two pulses may occur at the desired frequency and not cause an event.
The pulses transmitted are numbered 1, 2, . . . , G, . . . , R. Let A mean that a
pulse has occurred at the desired frequency and let B mean that a second pulse
has occurred at the desired frequency. How many arrangements are there such
that B occurs G or more pulses later than A?
If A occurs on the first pulse, then B may occur on the (G + 1)st pulse, or
(G + 2)nd, and so on up to R. Hence, there are R − G positions in which B may
occur and still be G or more pulses from A, provided A occurs on pulse one. If A
occurs on pulse two, then B may occur on pulse number (G + 2), (G + 3), and so
on up to R. Hence, there are R − G − I positions in which B may occur if A occurs
at pulse two. This process can be continued until A occurs on the (R − G)th pulse.
Then there is only one position for B which will not cause an event: namely, B
must occur on the Rth pulse.
Evidently, the total number of ways in which two pulses may occur in the
same frequency bin and not cause an event is the sum (R − G) + (R − G − 1) +
(R − G − 2) + . . . + 1. The sum of the integers from 1 to k is given by
k(k + 1)/2 (C.5)
Thus, the total number of ways to receive two pulses out of R at the desired
frequency and have them be more than G pulses apart is
(R − G) (R − G + 1)/2
Similar methods can be used to calculate the probability that three pulses occur
at the desired frequency and do not cause an event. Let A mean that the first of
the three pulses has occurred at the desired frequency. Similarly, let B and C mean
that the second and third of the three pulses have occurred at the desired frequency.
The event will not occur if A occurs on the first pulse, B occurs on the
(G + 1)st pulse, and C occurs on the (2G + 1)st pulse. In fact, if A occurs on the
first pulse, then B and C may occur in any manner among the last R − G pulses
such that they are at least G apart. However, this number has already been calcu-
lated above (for R pulses). Hence, the number of ways three pulses cannot cause
an event if A occurs on pulse one can be obtained by substituting R − 2G for k
in (C.5):
(R − 2G) (R − 2G + 1)/2
In like manner, if A occurs on pulse two, then B and C may be arranged among
the last R − G − 1 pulses in any way so as not to cause an event. This number
would be
(R − 2G − 1) (R − 2G)/2
C.4 Probability That Several Pulses Occur at the Desired Frequency Less Than G Pulses Apart 417
If A occurs on the jth pulse, then the number of arrangements is
(R − 2G − j + 1) (R − 2G − j + 2)
2
Then this counting process can be continued until A occurs on the (R − 2G)th
pulse. Then B may occur on the (R − G)th pulse and C may occur on the Rth
pulse, not causing an event. Thus, the total number of ways not to get an event
when three pulses occur at the desired frequency is given by
冉冊
R − 2G + 1
S
∑ 2
(C.6)
S =2
The relation for the sum of the binomial coefficients is
冉冊冉冊
N
S N+1
∑ =
M+1
(C.7)
S =M M
Thus, the total number of ways three pulses out of R may occur at the desired
frequency and not cause an event is
冉 R − 2G + 2
3 冊
=
(R − 2G + 2) (R − 2G + 1) (R − 2G)
6
(C.8)
Then, to account for those situations in which three pulses occur but do not
cause an event, the probability is
p 3q R − 3(R − 2G + 2) (R − 2G + 2) (R − 2G)/6 (C.9)
This process can be generalized to give the probability of getting no event if

k pulses out of R occur at the desired frequency as follows.
Let P(k) be the probability that k pulses out of R occur at the desired frequency
and do not cause an event. Then,
P(k) = p kq R − k 冉
R − (k − 1)G + k − 1
k 冊 (C.10)
It should be observed that P(k) is equal to the binomial term P k for k = 0 or 1.

Thus, the complete expression for P(E) is
冉冊
L
R − (k − 1) (G − 1)
P(E) = 1 ∑ p kq R − k (C.11)
k −0 k
where
L= 冋 R+G−1
G 册 (C.12)
The brackets around the above expression designate L as the integer value of
the expression. In other words, the value of (R + G − 1)/G will be of the form
A.B, where A is some integer and B is some decimal fraction; then L = A. For
example, if R = 9, G = 3, then (R + G − 1)/G = 3.667. The corresponding value
of L is, therefore, 3.
As a check, in the limiting case where G = R, L = 2 − 1/R, or L = 1, the
formula reduces to P(E) = I − P 0 − P 1 as is expected.
A simple method of summing (C.11) into a closed form has not been found.
In the particular cases of most interest, however, R is not very much larger than
G (R = 2G, 3G, 4G, 5G, for example). For this reason, calculations using the
above formula are not overly cumbersome.
For R = 3G, the formula reduces to
P(E) = 1 − q R − Rpq R − 1 − p 2q R − 2(R − G) (R − G + 1)/2 (C.13)
− p 3q R − 3(R − 2G + 2) (R − 2G + 1) (R − 2G)/6
This formula for P(E) with R = 3G has been numerically evaluated and the
results are shown in Figure C.2.
A computer simulation has been made to check the validity of (C.13). It was
comprised of 450 trials of 39 pulses per trial. There were 50 frequency bins among
which the pulses were divided (R = 39, G = 13, N = 50). As a result, 47 events
were counted. This leads to an empirical probability of 47/450, or 0.1044. The
formula given by (C.13) gives 0.1150, a favorable comparison. The most probable
(expected) number of events in 450 trials, according to the calculated probability,
would be 51, or four more than the results of the simulation. However, since the
standard deviation of the number of events in this case is 7, and the probability
that between 44 and 58 intercepts will be obtained from the simulation is about
0.85, the result of 47 is quite reasonable.
C.5 Probability Distribution of the Interval Between Two Pulses
In view of the importance of the PRI in signal identification, the distribution of

the interval between the two pulses making up such an intercept is of interest. This
distribution is derived below, and the results for the case where the maximum
pulse interval is one-third of the receiver dwell time are shown in Figures C.3 and
C.4. The expected value of the PRI observed is also given. (For G = 12, the
maximum allowed PRI is 12 of the radar’s PRIs.)
The probability that two consecutive radar pulses make up the intercept is of
interest, since in that case the observed PRI is the actual PRI. This function is
shown in Figure C.5.
For a receiver requiring two pulses from a frequency hop radar before a valid
intercept is reported, it is useful to calculate the distribution of the spacing between
Figure C.2 Probability of getting an intercept at a given frequency.

Figure C.3 Distribution of PRI values (G = 12).
the two pulses which combine to make up an intercept at a given frequency. The
present problem is then to find out how many pulse intervals occur between the
pulses making up such an event. Assume that the radar transmits R pulses during
the time the receiver dwells at one frequency. If the R pulses are numbered 1, 2,
3, . . . . , R and occur in order, let the first pulse of the two necessary to make up
an event be numbered k 2 and the second k 2 , Then the number of pulse intervals
between the two pulses is I = k 2 − k 1 . Clearly, I must be an integer from 1 to
(G − I); G being the maximum allowable PRI divided by the PRI of the radar.
Figure C.4 Distribution of PRI values (G = 6).
The methods used lead to a formulation that gives the probability distribution
for the number of pulse intervals between the pulses that make up an event for
any values of R and G. However, inasmuch as the specific case of R = 3G is typical,
numerical calculations have been confined to values of R that are three times the
values of G.
The distribution of I over the values 1, 2, 3, . . . , (G − 1), can be found by
first counting all the ways in which two successive pulses can cause an event
Figure C.5 Probability that an intercept has the correct PRI.
(I = 1), then by counting all the ways in which the second pulse of the event is
one pulse removed from the first pulse of the event (I = 2), and continuing to
(G − 1). Suppose, for example, G = 3 and R = 9. What is the probability of an
event with I = 1? This could happen if the first two pulses both occurred at the
correct frequency. The rest of the R pulses would not matter.
Thus, an event can occur with I = 1 if the first two pulses occur at the correct
frequency with probability p 2. It could also happen if the second and third pulses
occurred at the correct frequency and the first did not occur at the correct frequency.
The probability of this happening is pq 2. Similarly, the third and fourth pulses
could occur at the desired frequency, and the first and second at some other
frequency. The probability of this happening is q 2p 2. Notice that the case where
the first pulse occurs at the desired frequency, the second at some other frequency,
and the third and fourth at the correct frequency (pqp 2 ) cannot be counted because
it would cause an event with I = 2 after the third pulse and, once an event occurs
in the R trials, none after it is considered.
The next case to be considered is when the fourth and fifth pulses occur at the
desired frequency. Then the second and third pulses must occur at some other
frequency (since G = 3) in order to not cause an event. However, the first pulse
may or may not occur at the desired frequency since it is G pulses away from the
fourth. Thus, for this case, the probability expression contains two terms: q 3p 2 +
q 2p 3.
Table C.1 shows this situation (‘‘*’’ indicates a pulse at the desired frequency;
‘‘—’’ indicates a pulse at a different frequency; and ‘‘?’’ indicates a pulse that may
or may not be at the desired frequency).
The probabilities numbered 1, 2, 3, and 4 have been discussed. Number 5
comes about in the following way. There are now two ‘‘question’’ pulses, numbers
1 and 2. One of them may occur at the desired frequency, or both may occur at
some other frequency. However, the case in which both occur at the desired
frequency is not permitted. (This would repeat the situation in case 1.) Thus, the
probability expression comprises the following terms: qqq 2p 2 + pqq 2p 2 + pqp 2p 2
= q 4p 2 + 2q 2p 3. This counting process may be continued until the eight probabilities
are obtained. The total probability of obtaining an event with I = 1 is, then, the
sum of all these probabilities. Notice that the first G probabilities have one term,
the next G probabilities have two terms, and the next (G − I) have three terms.
Also, there are (R − I) probabilities in all.
For the case G = 3, there will also be events with I = 2. These cases are illustrated
in Table C.2.
The total probability that I = 2 is the sum of all the above terms. Notice that
the first G terms have only one term, the next G terms have two, and the remaining
(G − I) terms have three.
Consideration of the mechanics of the counting process, with the help of some
further examples, has led to the general form for the probabilities. Let k be an
index that numbers the rows, and let j be an index that numbers the columns.
There are (R − I) rows and L columns; L having been found to be
L= 冋 R+G−I−1
G 册 (C.14)
Table C.1 Ways to Get an Intercept with I = 1

Pulses 1 2 3 4 5 6 7 8 9 Probability
* * 1. p 2
— * * 2. qp 2
— — * * 3. q 2p 2
? — — * * 4. q 3p 2 + q 2p 3
? ? — — * * 5. q 4p 2 + 2q 3p 3
? ? ? — — * * 6. q 5p 2 + 3q 4p 3
? ? ? ? — — * * 7. q 6p 2 + 4q 5p 3 + q 4p 4
? ? ? ? ? — — * * 8. q 7p 2 + 5q 6p 3 + 3q 5p 4
Table C.2 Ways to Get an Intercept with I = 2

Pulses 1 2 3 4 5 6 7 8 9 Probability
* — * 1. qp 2
— * — * 2. q 2p 2
— — * — * 3. q 3p 2
? — — * — * 4. q 4p 2 + q 3p 3
? ? — — * — * 5. q 5p2 + 2q 4p 3
? ? ? — — * — * 6. q 5p 2 + 3q 5p 3
? ? ? ? — — * — * 7. q 7p 2 + 4q 6p 3 + q 5p 4
The brackets here mean to take L as the integer value of the expression. The
determination of L is discussed later.
The probabilities in this matrix are given by the general term
P(I; j, k) = 冉 j−1
p 冊
k − (G − 1) ( j − 1) − 1 j + 1 k − j + 1 − 1
q (C.15)
in which k ranges between I and (R − I), and j ranges between I and L.

In order to derive an expression for the probability of the occurrence of an
event with a specified I, it is necessary to sum all the possibilities in this array:
L R −1
P(I) = ∑ ∑ P(I, j, k) (C.16)
j =1 k −1
Before performing this sum, it is helpful to discuss the derivation of the expres-
sion for L. In the P(I, j, k) terms, the binomial coefficient that appears has a value
only for values below a certain j, after which it becomes zero. This value of j occurs
when the ‘‘upper’’ number becomes equal to the ‘‘lower’’ number, or when
K − (G − 1) ( j − 1) − 1 = j − 1 (C.17)
Thus, the largest value of j to be assigned in the calculation is given by
L= 冋 k+G−1
G 册 (C.18)
However, in the above expression for P(I), the summation eliminating the
value k has already been performed; hence, the largest j value is given when k
assumes its largest value, namely (R − I). Thus,
L= 冋 R−k+G−1
G 册 (C.19)
Notice that j must become at least as large as L, but since the binomial term
becomes a zero for j values larger than L, any limit for j that is larger than L will
give the same answer.
The initial step is to perform the inner sum:
冉冊
R −1
k − (G − 1) ( j − 1) − 1 j = 1 k − j + I − 1
∑ j−1
p q (C.20)
k −1
For purposes of this sum, the values of j, G, and I are constants. The terms
from k = 1 to k = ( j − 1) G + I are zero, since for these values of k the upper
number in the binomial coefficient is smaller than the lower number. Thus, the
sum can just as well be performed from k = ( j − 1) G + 1 to k = R − I:
冉冊
R −1
k − (G − 1) ( j − 1) − 1 j + 1 k − j + I − 1
∑ j−1
p q (C.21)
k = ( j − 1)G + 1
The usefulness of this last step can be seen by making a change of variable.
Let
K − (G − 1) ( j − 1) − 1 = h (C.22)
Then the sum is given by
冉冊
R − I − (G − 1) ( j − 1) − 1
h
∑ j−1
p j + 1 q h + (G − 1) ( j − 1) − j + 1 (C.23)
h =j −1
If the constant factors are moved outside the summation sign, the simplified
sum to be performed is
冉冊
R − I − (G − 1) ( j − 1) − 1
h
p j + 1 q 1 + (G − 1) ( j − 1) − j ∑ j−1
qh (C.24)
h =j −1
In order to simplify the notation, it is useful to make the following definitions:
冉冊
␤
h h
S= ∑ q (C.25)
h =␣ a
␣=j−1 (C.26)
␤ = R − I − (G − 1) ( j − 1) − 1 (C.27)
The summation may be performed by using a generating function. One defines

a function A(t) as
A(t) = 1 + qt + q 2t 2 + q 3t 3 + . . . + q ␤ t ␤ (C.28)
By applying the formula for the sum of a geometric series, A(t) can be expressed
in closed form:
␤
1 − (qt) ␤ − 1
A(t) =
1 − qt
+ ∑ q ht h (C.29)
h =0
The ␣ derivative of A(t) is
␤
d ␣A(t)
dt ␣
= ∑ (h)␣ q tt −␣ (C.30)
h =␣
in which (h)␣ = h(h − 1) (h − 2) . . . (h − ␣ + 1). This sum, with t = 1, is identical

to S above with the exception of the ␣ ! factor, which should be in the denominator.
Thus,
S=
1 d ␣A(t)
␣ ! dt ␣ t = 1 | (C.31)
In terms of the original problem, the inner sum in the expression for P(I) can
now be replaced. Equation (C.16) becomes
p ( j + 1) (G − 1) ( j − 1) − j d j − 1A(t)
|
L
P(I) = ∑ ( j − 1)! dt j − 1 t = 1
(C.32)
j =1
in which
A(t) = [1 − (qt)R − I − (G − 1) ( j − 1)]/(1 − qt) (C.33)
Although this formula is somewhat cumbersome, the value of L is practically

never more than 5 or 6. The specific case for R = 3G was selected as an example,
hence the formula for P(I) was obtained in closed form for this relationship between
R and G.
If R = 3G, then L is found from
L= 冋 3G + G − I − 1
G 册 (C.34)
L=4 冋册 I+1
G
(C.35)
Since I must be between 1 and (G − 1), L must always be an integer equal to

or greater than 3 and less than or equal to (4 − 2/G). This means that L is always
3 for any value of I if R = 3G. In this case, P(I) is given by
J
P(I) = ∑ p(I : j) (C.36)
j =1
C.6 Determining an Optimum Receiver Sweep Rate 427
or
冉
P(I) = pq I − j 1 − q 3G − 1 + q G − 1 [1 − q G − 1 (2G − I + 1) − (2G − I) q]
再
+ q 2G − 2 1 +
qG − 1
2
[2(G − I) (G − I + 2) q − (G − I + 1) (G − I + 2)
− q 2(G − I) (G − I + 1)] 冎冊 (C.37)
Figures 3.3 and 3.4 show several characteristic ‘‘I-distributions’’ for some
typical values of G and N (for R = 3G). The distributions have been normalized
so that the values of P(I) obtained from the curves are given by
P(I)
G −1
∑ P(I)
j =1
The summation of the P(I) terms over all possible values of I must, of course,
equal the probability of getting an event, as given by P(E).
C.6 Determining an Optimum Receiver Sweep Rate
The similar shapes of the curves in Figures C.3 and C.4 lead one to consider
combining them and expressing the probability of intercept in terms of the ratio
G /N. Recall that G is the maximum number of radar PRIs that the receiver allows
for a valid intercept, and N is the number frequencies in the radar’s frequency hop
sequence. The result is shown in Figure C.6.
Recall that we have also required that the receiver dwell time be three times
the maximum permitted PRI, and therefore R = 3G. Therefore, the faster the
receiver sweep speed, the smaller are R and G. The best strategy may be to choose
the receiver sweep speed to maximize the expected number of intercepts. Figure
C.7 shows the expected number of intercepts divided by G versus G /N, which is
based on the empirical curve in Figure C.6. The optimum value for G /N is near
1.0, for which the expected number of intercepts is 0.7G. Thus, the correct strategy
is to set the sweep speed so that the dwell time at any given frequency is the product
of the number of frequency slots used by the radar (N) times 3 times the radar’s
PRI. Similar results could presumably be obtained for ratios of R /G other than 3.
Figure C.6 G /N versus probability of intercept.
Figure C.7 Expected fractions of frequencies having intercepts versus G /N.

APPENDIX D
History and Fundamentals of the IFM
The use of crystal video receivers for the interception of radar signals was a natural
development because of the high power of radar transmitters combined with the
low cost and small size of crystal video receivers. Their main shortcoming was that
they did not indicate the carrier frequency. After all, a small antenna, detector,
video-audio amplifier, and recorder could preserve the PRI and scan information
and be packaged in a pocket-sized unit. The missing parameters were pulse duration
and the carrier frequency. Techniques to encode the pulse duration in a way that
could be stored on an audio recorder were of interest, as were similar methods for
encoding the carrier frequency.
During the late 1950s, work on instantaneous frequency indication (IFM)
techniques was in progress in at least three locations: Syracuse Research Corpora-
tion (SRC),1 Stanford Electronic Laboratory (SEL), and in the United Kingdom at
Mullard Research Laboratories. Mullard began such research in 1954 [1]. During
1957 and 1958, work at SRC emphasized a switched stub technique [E.M. Williams,
private communication with E. M. Williams, 1960]. In this technique, after each
pulse, an RF switch was opened or closed to either insert a shorted transmission
line stub into the RF circuit or not. This affected the voltage standing wave ratio
(VSWR) and hence the amplitude. Variations could then be related to the carrier
frequency by means of a calibration chart. Clearly, this technique was affected by
pulse-to-pulse amplitude changes from other causes (e.g., scanning), and pulse-to-
pulse frequency agility could not be handled. Similar work at SEL was also in
progress [2].
If audio recording of the RF was not necessary, then an oscilloscope could be
used to display the pulsed signal data. Otherwise, pulse width modulation or pulse
amplitude modulation was used to encode the RF information within an audio
bandwidth signal. The next step was to use two channels, one with a frequency
sensitive element (e.g., shorted stub) and one without, as shown in Figure D.1.
SRC delivered such a device in 1962: a 5.25″ × 8.25″ box covering the 50- to
250-MHz band [3]. Dynamic range was 15 dB and the frequency accuracy was
13%. The RF output was encoded as pulse duration, as shown by the calibration
curves in Figure D.2.
Events were moving in what turned out to be a more fruitful direction in
the United Kingdom. In 1957, S. J. Robinson of Mullard Research Laboratories
conceived the basic microwave phase discriminator circuit, which became the basis
of the modern IFM [3]. By 1959, Mullard delivered a device called Pendant [4].
1. Then called Syracuse University Research Corporation.
429
430 History and Fundamentals of the IFM
Figure D.1 An early two-channel IFM approach.
Figure D.2 Early IFM calibration curve.
This used the polar type of CRT display, which had been demonstrated in 1958.
Each pulse was displayed as a vector drawn from the center of the CRT. The length
of the vector was proportional to the received signal strength and the angle was
proportional to the RF. A mechanical cursor could be rotated to lie along any
observed radial strobes, and the RF frequency could be read from a scale around
the outer edge of the CRT.2 In contrast to the miniature IFM described above,
Pendant was very large (a full 19-inch rack of equipment) and was based on vacuum
tube circuitry (by contrast, much of the energy of U.S. electrical engineers at the
time was spent transistorizing circuitry of all kinds). Nonetheless, to those who
saw it operate, Pendant was clearly a very useful device which performed admirably.
Dr. Thomas F. Curry of SRC recognized the potential of this device and, no
doubt influenced by SRC’s requirements at the time for compact instantaneous
frequency measurement gear, assigned Mr. Grover M. Boose the task of converting
2. Pendant arrived at SRC for evaluation purposes at about the same time as the author.
History and Fundamentals of the IFM 431
the S-band waveguide microwave circuitry contained in the Mullard receiver into
stripline form. By 1963, an S-band receiving unit had been completed using a
stripline discriminator module. The calibration curve for this module is shown in
Figure D.3 [5]. (This was the first stripline IFM. The nonlinearity of the curve was
Figure D.3 Calibration curve for the first stripline IFM.

partially due to the use of a frequency sensitive phase-shift element. Later this
element was replaced by a broadband 90° hybrid.) The next IFM design by the
workers at SRC included a tunnel-diode limiting preamplifier on the front end for
wide dynamic range and an analog-to-digital converter connected to the video
outputs. This unit, completed in late 1964, was a U.S. ancestor of modern digital
IFMs.3
The work done in 1964 considered nearly every IFM problem: simultaneous
signal effects [6], CW bias problems, the need for limiting, and the utility of a
digital output of the RF on a single pulse. The analog IFMs to follow were concerned
with increasing the dynamic range and sensitivity, providing frequency coverage
to 12 GHz (or to 18 GHz), warning the operator of the presence of simultaneous
signals, CW signals, and other system aspects of the IFM.4 The two IFM systems
produced at SRC in 1965 and 1967 are shown in Figures D.4 and D.5. Frequency
coverage from 0.5 to 18 GHz with high sensitivity was provided. TWT preamplifiers
with leveling were used because low-noise solid-state amplifiers were very costly
or unavailable at the higher frequencies. The 12.4- to 18-GHz band discriminators
were made from waveguide components. The large ‘‘black box’’ on top of the CRT
in Figure D.4 is a mechanical device that converted the angular position of the
cursor to an RF value. The cursor was manually tuned to coincide with the signal
Figure D.4 An analog IFM receiving system produced in 1965. (Courtesy of SRC.)
3. The issue of ambiguity elimination was a hot topic of conversation at SRC during the summer of 1964.
Dr. Peter Knoke argued forcefully for the inclusion of ambiguity elimination circuitry to avoid the problem
of ambiguous frequency indication due to IFM nonlinearity when the outputs of a number of IFMs with
different frequency sensitivities are combined to provide a digital word. He was overruled at the time,
partly due to the cost of logic circuits in 1964 and partly due to the belief that sufficient discriminator
linearity could be achieved to avoid the ambiguity problem. The digital discriminator without ambiguity
elimination was successfully built, but endless hours of delay line cutting and calibration were required.
Today’s digital IFMs all incorporate ambiguity elimination logic, which is indispensable to their successful
operation.
4. It was believed by many at SRC that the golden age of IFMs was at hand in 1965, the author included;
but it was several years before the AN/FSQ-65(I) was even tested in the field.
Figure D.5 AN/FSQ-65(V) analog IFM receiving system produced in 1967. (Courtesy of SRC.)
of interest using a knob. This also moved a 35-mm film calibrated with the RF
frequency behind the rectangular ‘‘windows’’ cut in the panel of the black box.
The appropriate window was illuminated, depending on which band was selected.
The same calibration technique was used in the AN/FSQ-65(V) shown in Figure
D.5 with considerable improvement in the implementation. The AN/FSQ-65(V)
included a matrix of alarm indicators. Each band had an activity indicator, as well
as indicators for CW and for simultaneous signals in the same band (intra) and in
different bands (inter). The unit included a chopper to make CW signals visible
on the display and an expansion scale (longer delay line), which could be used in
each band as desired to show frequency hopping or FMOP more clearly. Nar-
rowband tunable filters could also be inserted in each band as needed, along with
calibrated marker signals.
The 1967 stripline design shown in Figure D.6 is believed to be the first such
stripline unit to operate above 12 GHz [7].5
D.1 The Broadband Microwave Frequency Discriminator
The basic building block in the instantaneous frequency discriminating (IFD)

receiver is shown in Figure D.7. Consider the operation of the circuit when the
input is a pulsed signal of the form
E 2 (t)
e2 = cos (␻ t − ␤ L 1 ) (D.1)
√2
5. In 1964, several SRC engineers started another company (then called Curry, McLaughlin, and Len, and
later called Microwave Systems, Inc.). One of the engineers was Robert L. McLaughlin who developed
the 12- to 18-GHz stripiine IFM shown in 1967. Another engineer, James E. Secord, built in the late
1960s a waveguide IFM unit that operated to 40 GHz.
Figure D.6 12- to 18-GHz discriminator stripline layout.
Figure D.7 Illustrating the basic principle of the operation of the microwave discriminator IFD.
in which E 0 (t) is the pulse modulating function. This signal is divided by a power
splitter into two channels, commonly comprising two lengths of transmission line
of different electrical lengths, ␤L 1 and ␤L 2 (␤ represents the phase constant of the
lines). The signal at the outputs of the two channels are
E 0 (t)
e2 = cos (␻ t − ␤ L 1 ) (D.2)
√2
and
E 0 (t)
e3 = cos (␻ t − ␤ L 2 ) (D.3)
√2
These two signals are added at the input to the detector. Using the identity
1 1
cos ␣ + cos b = 2 cos (a + b) ⭈ cos (a − b) (D.4)
2 2
the input to the detector may be expressed as
␤⌬ ␤ (L 1 + L 2 )
e 4 = 2E 0 (t) cos cos ␻ t − (D.5)
2 2
in which ⌬L is L 2 − L 1 .
The detector output, assuming a square-law detection characteristic and neglect-
ing the detector figure or merit, is
冋
e S = 2 E 0 (t) cos 冉冊册
␤ ⌬L
2
(D.6)
This output is always positive and a function of frequency because of the

frequency dependence of ␤ . A more convenient bipolar frequency-dependent output
may be obtained by combining the circuit of Figure D.7 with a second circuit in
which an additional delay of 180° has been introduced between the two channels.
The resultant circuit is shown in Figure D.8. The 180° delay (␭ /2 length at the
design center frequency) in the L 2 arm of the second channel combination results
in a sine-squared output at the second channel detector output unlike the cosine-
squared output of the first channel. A video output e 6 is then formed from the
difference of the two channel outputs:
Figure D.8 Combination of two units to obtain a bipolar video output as a function of frequency.
冋冉冊册冋冉冊册
2 2
␤ ⌬L ␤ ⌬L
e 6 = 2 E 0 (t) cos − 2 E 0 (t) sin (D.7a)
2 2
or
2
e 6 = E 0 (t) cos (␤ ⌬L) (D.7b)
A second bipolar output e 7 can be formed by a circuit identical to that of

Figure D.8, except that the first split is obtained by the use of a backward wave
coupler, so that a phase shift of 90° appears at the coupled arm output to line L 2 .
For this case, with the same signal input, the video output, e 7 , is
2
e 7 = 2E 0 (t) sin ␤ L (D.8)
The two outputs, e 6 and e 7 , may be used to form a frequency-indicating display

in a number of ways. The most obvious approach is to feed e 6 and e 7 to the
horizontal and vertical plates of a CRT. The resultant display is a polar strobe
with an angle ␪ dependent upon frequency and amplitude dependent upon signal
amplitude. The angle ␪ of the strobe is actually
␪ = ␤ ⌬L (D.9)
The variation of ␪ cannot exceed 360° without ambiguity. The frequency range,
␻ 2 − ␻ 1 , then, is limited by the requirement that
␻2 ␻
⌬L − 1 ⌬L = 2␲ (D.10a)
V0 V0
or
2␲ V0
␻2 − ␻1 = (D.10b)
⌬L
in which V0 is the velocity in the transmission lines. By appropriate choice of the

delay differential, ⌬L /V0 , either a wide or narrow range of frequencies can be
covered without ambiguity.
The entire RF assembly, as implied earlier, is sometimes termed a microwave
discriminator; the term IFD is less ambiguous and will be used hereafter.
A wide variety of designs for IFD units have been successfully implemented.
Perhaps the most favorable results are obtained with variations of an all-quadrature
hybrid configuration, such as shown in Figure D.9. The couplers (quadrature
hybrids) are connected so that the power at each detector has the proper phase
and amplitude relationship. The four detectors receive signals as follows:
E0 E
E1 = ∠ −180° − ␤ L + 0 ∠ −270° (D.11a)
2√2 2√2
Figure D.9 Seven-coupler IFD.
E0 E
E2 = ∠ −180° + 0 ∠ −270° − ␤ L (D.11b)
2√2 2√2
E0 E
E3 = ∠ −␤ L + 0 ∠ −180° (D.11c)
2√2 2√2
E0 E
E4 = ∠ −90° + 0 ∠ − ␤ L − 90° (D.11d)
2√2 2√2
in which E 0 is the input signal and ␤ L the differential delay. The square-law
detection process results in
2
E
V1 = 0 (1 − sin ␤ L) (D.12a)
4
2
E0
V3 = (1 − cos ␤ L) (D.12b)
4
2
E0
V2 = (1 + sin ␤ L) (D.12c)
4
2
E
V4 = 0 (1 + cos ␤ L) (D.12d)
4
The signals for the display are obtained by combining these detection output
voltages to form
2
E0
V y = V2 − V1 = sin ␤ L (D.13a)
4
2
E0
Vx = V4 − V3 = sin ␤ L (D.13b)
4
As shown in Figure D.9, couplers C and D are connected in tandem to produce

a 90° shift in one leg of the IFD. This is done in order to set the power output of
coupler 0 equal to the power passing from B to G. Couplers C and D can be
replaced by a single coupler with shorted or open terminations, as shown in Figure
D.10. Because it is difficult to obtain a perfect open or short circuit termination,
a phase shifter (Schiffman) is added to trim the phase. The couplers B, C, and 0
can be replaced with a single 180° coupler or hybrid, as shown in Figure D.11. In
this case, the diamond hybrid shown is limited to about an octave in bandwidth.
The proper RF processing can also be achieved with four 180° couplers and one
90° phase shifter in place of couplers B through G. In the original U.K. units and
earliest stripline versions, a quarter-wavelength line length difference at the design
center frequency was used to achieve the 90° shift. Because of the resulting depen-
dence of phase shift upon frequency, this design is limited to less than one octave
in bandwidth, and the ␪ versus frequency characteristic is nonlinear.
Figure D.10 Techniques for achieving a 90° phase shift with no loss in power: (a) schematic of
90° hybrid; (b) two hybrid phase shifter; (c) shorted hybrid as 90° phase shifter;
(d) open hybrid as 90° phase shifter; and (e) Schiffman phase shifter.
Figure D.11 IFD using a diamond hybrid.
References
[1] Goddard, N. E., ‘‘Instantaneous Frequency-Measuring Receivers,’’ IEEE Trans. on Micro-

wave Theory and Techniques, April 1972, pp. 292–293.
[2] May, B. B., ‘‘A Wide Band Frequency Discriminator Using Open and Shorted Stubs,’’
Technical Report 791-2, Stanford Electronic Lab, November 1960 (SEL Project 791 K).
[3] Boose, G. M., E-l/O Instantaneous Frequency Indicating Receiver, Final Report, Task
Order 7, Subtask 1, Contract 582RR2, October 31, 1962.
[4] Robinson, S. J., ‘‘Comment on Broadband Microwave Discriminator,’’ IEEE Transactions
Professional Group on Microwave Theory and Techniques, March 1964.
[5] Boose, G. M., ‘‘Wide Open Frequency Indicating Receiver,’’ paper presented to a joint
meeting of the Syracuse Chapters of the IEEE Professional Groups Microwave Theory and
Techniques and Antennas and Propagation, April 1964.
[6] Boose, G. M., ‘‘Wide Open Frequency Indicating Receiver,’’ Syracuse University Research
Corporation, Electronics Research Laboratory, Technical Memo No. 24, August 1964.
[7] Wiley, R. G., and E. M. Williams, ‘‘Spectrum Display Is Instantaneous,’’ Microwave Journal,
August 1970, pp. 44–47.
APPENDIX E
Emitter Location Partial Derivatives
This appendix includes expressions for partial derivatives of TOA and FOA mea-
surements due to position, heading, and velocity in the notation used in the Math-
Cad software tool. The subscripts i and k denote indices for a cross and downrange
x-y emitter position, respectively.
Partial Derivative of TOA with Respect to Aircraft Position
∂TOA N −1 (x i − X M )
= , M=N
∂X M i,k
√(x i − X M ) + (y k − YM )
c 2 2 (E.1)
0 otherwise
∂TOA N −1 (y k − Y M )
= , M=N
∂YM i,k
√(x i − X M )
c 2
+ (y k − YM )2 (E.2)
0 otherwise
Partial Derivative of FOA with Respect to Aircraft Position
=
∂X M i,k ␭ 再
∂FOA N −1 −(y k − YM )2 ⭈ (VM cos ␪ M ) + (x i − X M ) ⭈ (y k − YM ) ⭈ (VM sin ␪ M )
[(x i − X M )2 + (y k − YM )2]3/2
冎
, M=N
0 otherwise
(E.3)
=
∂YM i,k ␭ 再
∂FOA N 1 (x i − X M ) ⭈ (y k − YM ) ⭈ (VM cos ␪ M ) + (x i − X M )2 ⭈ (VM sin ␪ M )
[(x i − X M )2 + (y k − YM )2]3/2
冎
, M=N
0 otherwise
(E.4)
Partial Derivative of FOA with Respect to Aircraft Heading
∂␪ M i,k
=
␭ 再
∂FOA N 1 −(x i − X M ) ⭈ (VM sin ␪ M ) + (y k − YM ) ⭈ (VM cos ␪ M )
√ (x i − X M ) 2
+ (y k − Y M ) 2
, 冎 M=N
0 otherwise
(E.5)
441
442 Emitter Location Partial Derivatives
Partial Derivative of FOA with Respect to Aircraft Velocity
∂VM i,k ␭ 再
∂FOA N 1 (x i − X M ) ⭈ sin ␪ M + (y k − YM ) ⭈ cos ␪ M
=
√(x i − X M ) + (y k − YM )
2 2
, 冎 M=N
0 otherwise
(E.6)
The following are the expressions for partial derivatives of emitter x-y location
coordinates due to position and AOA measurements. The AOA measurement is
computed as
AOA M i,k = tan−1 冉 xi − XM

y k − YM 冊 (E.7)
Partial Derivative of Emitter x-y Location with Respect to Aircraft Position
∂x 1
= (E.8)
∂X1 i , k tan (AOA1 i , k )
1−
tan (AOA2 i , k )
∂y 1
= (E.9)
∂X1 i , k tan (AOA2 i , k ) − tan (AOA1 i , k )
∂x −tan (AOA1 i , k )
= (E.10)
∂Y1 i , k tan (AOA1 i , k )
1−
tan (AOA2 i , k )
∂y −tan (AOA1 i , k )
= (E.11)
∂Y1 i , k tan (AOA2 i , k ) − tan (AOA1 i , k )
tan (AOA1 i , k )
−
∂x tan (AOA2 i , k )
= (E.12)
∂X2 i , k tan (AOA1 i , k )
1−
tan (AOA2 i , k )
−tan (AOA1 i , k )
∂y tan (AOA2 i , k ) 1
= − (E.13)
∂X2 i , k tan (AOA2 i , k ) − tan (AOA1 i , k ) tan (AOA2 i , k )
∂x tan (AOA1 i , k )
= (E.14)
∂Y2 i , k tan (AOA1 i , k )
1−
tan (AOA2 i , k )
∂y tan (AOA1 i , k )
= +1 (E.15)
∂Y2 i , k tan (AOA2 i , k ) − tan (AOA1 i , k )
Emitter Location Partial Derivatives 443
Partial Derivative of Emitter x-y Location with Respect to AOA
冦冧
tan (AOA2 i,k )
∂x cos (AOA1 i,k )2
= ⭈ X1
∂AOA1 i,k [tan (AOA2 i,k ) − tan (AOA1 i,k )]2
冦冧
tan (AOA2 i,k )
cos (AOA1 i,k )2
+ ⭈ X2
[tan (AOA2 i,k ) − tan (AOA1 i,k )]2
冦冧
−tan (AOA2 i,k ) ⭈ [tan (AOA2 i,k ) − tan (AOA1 i,k )] − tan (AOA1 i,k ) ⭈ tan (AOA2 i,k )
cos (AOA1 i,k )2
+ ⭈ Y1
冦冧
−tan (AOA2 i,k ) ⭈ [tan (AOA2 i,k ) − tan (AOA1 i,k )] + tan (AOA1 i,k ) ⭈ tan (AOA2 i,k )
cos (AOA1 i,k )2
+ ⭈ Y2
(E.16)
冦冧
−tan (AOA1 i,k )
∂x cos (AOA2 i,k )2
= ⭈ X1
∂AOA2 i,k [tan (AOA2 i,k ) − tan (AOA1 i,k )]2
冦冧
tan (AOA1 i,k )
cos (AOA2 i,k )2
+ ⭈ X2
冦冧
−tan (AOA1 i,k ) ⭈ [tan (AOA2 i,k ) − tan (AOA1 i,k )] + tan (AOA1 i,k ) ⭈ tan (AOA2 i,k )
cos (AOA2 i,k )2
− ⭈ Y1
冦冧
−tan (AOA1 i,k ) ⭈ [tan (AOA2 i,k ) − tan (AOA1 i,k )] − tan (AOA1 i,k ) ⭈ tan (AOA2 i,k )
cos (AOA2 i,k )2
+ ⭈ Y2
(E.17)
444 Emitter Location Partial Derivatives
冦冧
1
∂y cos (AOA1 i , k )2
= ⭈ X1
∂AOA1 i , k [tan (AOA2 i , k ) − tan (AOA1 i , k )]2
冦冧
−tan (AOA2 i , k ) ⭈ [tan (AOA2 i , k ) − tan (AOA1 i , k )] tan (AOA1 i , k )
2 2
−
cos (AOA1 i , k ) ⭈ tan (AOA2 i , k ) cos (AOA1 i , k )2 ⭈ tan (AOA2 i , k )
+ ⭈ X2
[tan (AOA2 i , k ) − tan (AOA1 i , k )]2
冦冧
cos (AOA1 i , k )2
+ ⭈ Y1
冦冧
[tan (AOA2 i , k ) − tan (AOA1 i , k )] + tan (AOA1 i , k )
cos (AOA1 i , k )2
+ ⭈ Y2
(E.18)
冦冧
−1
∂y cos (AOA2 i , k )2
= ⭈ X1
∂AOA2 i , k [tan (AOA2 i , k ) − tan (AOA1 i , k )]2
冦冧
tan (AOA1 i , k ) ⭈ [tan (AOA2 i , k ) − tan (AOA1 i , k )] tan (AOA1 i , k )
+
cos (AOA2 i , k )2 ⭈ tan (AOA2 i , k )2 cos (AOA2 i , k )2 ⭈ tan (AOA2 i , k )
+ ⭈ X2
+ 再 1
cos (AOA2 i , k ) ⭈ tan (AOA2 i , k )2
2 冎 ⭈ X2
冦冧
tan (AOA1 i , k )
cos (AOA2 i , k )2
+ ⭈ Y1
冦冧
cos (AOA2 i , k )2
+ ⭈ Y2
(E.19)
About the Author
Richard G. Wiley received a B.S. in electrical engineering in 1959 and an M.S.

in electrical engineering in 1960, both from Carnegie Mellon University. After
employment at Syracuse University Research Corporation (SURC), he served in
the U.S. Army from 1961 to 1963. Dr. Wiley continued employment at SURC
from 1963 to 1967. From 1967 to 1974, he was employed at Microwave Systems,
Inc. He continued his education during this period at Syracuse University and
received a Ph.D. in electrical engineering in 1975. He was employed by Syracuse
Research Corporation from 1975 to 1985 as a staff consulting engineer and then
became a cofounder and chief scientist of Research Associates of Syracuse, Inc., a
position he still holds. He was elected a Fellow of the IEEE and is a member of
the Association of Old Crows (AOC). In 2001 Dr. Wiley was named chairman of
the AOC’s Professional Development Committee and is in charge of the AOC’s
Continuing Education Program.
445
Index
A C
Acousto-optic receivers, 89–90 Capture area, 119
Adding non-ELINT data, 388 Carson’s rule
Allan variance, 348 for FM spectra, 400–1
Ambiguity function, 227 modified for ELINT, 401–3
Amplitude comparison D/F, 127–31 CEP related to error ellipse, 151
Analysis bandwidth, 31 Chaff, 2
Analysis of predetection data, 342 Channelized receivers, 85–89
Angle of arrival location analysis, Circular array, 139–43
153–55 Circular error probable. See CEP
Antenna Circular scan, 239–40
beam shape, 196–97 Collector location uncertainty, 160
gain error due to pointing error, Combining pulse bursts, 332–33
173 COMINT, 4
patterns, 197–99 Complex delta-T histogram, 326–28
Aperature blockage, 204–7 effect of jitter on, 327–29
Array antenna patterns, 205 improved, 328–29
Array element spacing, 206 Conditional histogram
Automated peak processing for accuracy class example, 375–78
histograms, 329–30 Conical scan, 247
Conical scan on receive only (COSRO),
248
B
Constant beamwidth, 121
Bandwidth effects on Allan variance, Constant power ratio
354–56 circles of, 180
Beam broadening related to sidelobe Constant PRI, 282–83
level, 202 Constant TDOA
Beam patterns from ELINT, 199–205 lines of, 181
Bearing accuracy comparisons, 138 Constraints on pulse compression,
Bias in AOA location, 154–55 217–18
Biphase modulation effect on Covariance matrix
discriminator, 339–40 emitter location, 150, 152–53
Bistatic radar, 24 Cross polarization, 191–94
Bragg cell receivers, 89–90 Crystal video receivers, 57–63
Brown’s AOA location algorithms, 156 CW signals and IFMs, 81
447
448 Index
D Emitter ID example, 387

D/F, 5 Emitter location, 149–68
amplitude comparison, 127–31 partial derivatives, 441–44
error due to noise, 133–34 Ensemble averaging, 276–77
instantaneous, 126–48 ERP
Deinterleaving errors due to antenna pointing,
algorithms, 320–22 171–74
delta-T histogram applied to, 321–25 sample calculations, 179–81
via PRI-based, gating, 319–20 Error ellipse
pulse trains, 317–34 emitter location, 151
purity ratio, 322 ESM and radar range compared, 214–17
Delta-T histogram and 10 interleaved
pulse trains, 324 F
Digital IFM, 82–85
Fall time, 257, 260–64
Digitized pulse data, 256–57
Fan beams
Direction finding, 5, 124–48. See also
circular and sector scan, 239–44
D/F
Far-field patterns, 200
Directional antennas, 121
Fixed aperture, 122
Distance to the emitter
FM degradation by bandlimiting, 403–8
ERP error, 174–76
Frequency
Distribution
agility, 19
free statistics, 371–373
character of CW signals, 361
of the interval between pulses, 418–22
coverage/bands, 30
Doppler resolution, 17–18
domain stability measures, 353–54
Dynamic range, 35–38
drift measurement, 337–38
E frequency hop POI example, 413–28
ECM receivers, 56 measurement, 335–65
Effect of sine wave interference, 396–98 measurement accuracy, 335–37
Effective noise bandwidth, 44–45 stability and power law spectra,
Effect of noise on FM demodulation, 357–58
408–12
Effect of variation of measurement G
separation, 352–53 Geometric dilution of precision, 163–64
Electronic intelligence. See ELINT Geometric effects on location error,
Electronic scan, 251–52 164–67
Electronic warfare, 3
ELINT
H
considerations of FM signals, 395–412
data files, 383–89 High-power microwave weapons, 27–28
data for RWR design, 386–88 Histograms
data for simulation and training, 388 conditional, 370
defined, 1 effects of insufficient data on, 369
parameter limits, 367–82 and ELINT parameter limits, 367–72
parameters, 6–7 Homodyne receivers, 72–74
range equation, 10 Horizon
Elliptical polarization, 183–85 distance to, 170–71
Index 449
I Multipath effect on pulse shape, 266–68

Image rejection, 66 Multiple channel interferometers, 137–39
Incidental intrapulse modulation causes Multistatic radar, 24
and uses, 275 Multipath effect on ERP estimation,
Instantaneous frequency measurement 176–78
(IFM), 74–85 MUSIC algorithm for AOA, 146
calibration curve, first stripline, 431
N
history and fundamentals, 429–39
Intentional pulse modulation, 272–75 Narrow band channels and FM signals,
Intercept duration, 102 222–26
Intercept receivers, 6 Noise effects on Allan variance, 356–57
Intercept system characteristics, 29–51 Noise figure measurements, 40–42
Intercept time, 98–100 Noncoherent integration in ELINT,
Interferometers, 131–39 213–14
Intrapulse Normalized Euclidean distance (NED),
analysis, 255–80 273
FM, 338–39 Null-to-null beamwidth related to 3-dB
phase and frequency modulation, beamwidth, 203
268–79
O
J Omnidirectional antennas, 117–19
Jacobian in location analysis, 157 OPELINT, 384
Optimum receiver sweep rate and
L bandwidth, 427–28
L-band, 2–3 Overlapping pulse model, 79
Lenses for D/F, 144–45
Limiters and IFMs, 74–75 P
Linear FMOP radars, 34–35 Path loss dynamic range, 37
Linear frequency drift frequency stability, Pencil beam scans, 245
359 Periodic PRI variations, 287
LPI interception using the envelope, Phase comparison D/F, 131–39
218–21 Phase quantization and loss of gain, 207
POI
M examples, 104–8
M of N integration, 220–21 mathematical models, 100–8
Matched filter theory, 213 number theory, 111
Measuring coherence, 364–66 PRI ratio, 111
Measuring deinterleaver performance, theory, developments in, 98–115
333 Polarization
Microscan receivers, 90–93 of antennas, 183–96
Monopulse tracking, 249 pattern method, 187–88
Most probable values, 378–79 phase amplitude method, 189–90
Moving targets and integration (MTI) multiple antenna method, 190
time, 15–18 system aspects, 195–96
MTI improvement factor and PRI jitter, Postdetection sorting, 60
283 Power estimation via ELINT, 169–81
Multipath effect on AM, FM, 267–68 Probability of interception. See POI
450 Index
Predetection processing to detect LPI Pulse

signals, 226–32 compression, 16–17
PRI density received, 51–52
agility, 19 envelope parameters, 257–58
analysis, 281–315 sorting/clustering, 317–19
analysis techniques, 295–307 train spectrum, 321–22
average, 298–300 width, 257–59
common range marks, 310 Pulsed signal example frequency stability
categories of PRI, 282–87 calculation, 361–64
countdown example, 311–13 Pulses per beamwidth, 240
cumulative jitter, 298–300
deinterleaving and delta-T histogram, R
306
digital thresholding, 294–95 Radar
drift, 284 capabilities inferred, 5
dwell and switch, 285 and ELINT in World War II, 1–3
generators, countdown, 308–13 FM ranging in, 338
generators, delay line, 308 LPI and ELINT, 7–8
histogram examples, 301–4 LPI radar and the future of ELINT,
histogram shapes, typical, 304 211–34
histogram-delta-T, 303–7 performance, related to pulse shape,
jitter, 284–85 264–66
least mean squared error, 298–300 range equation, 9
measurement, SNR limitations to, trends, 22–26
288–89 Radon ambiguity transform (RAT), 229
measurements, pulse amplitude effects loss relative to matched filter, 230
on, 289–92 Range ratio, 9–12
measurement, bandwidth effect on, Range resolution, 13–14
292 Raster displays and deinterleaving, 333
for MTD radars, 314 Raster scan, 245–46
MTI blind speeds, 313–14 Raster scan technique, 258
noncumulative jitter, 298–300 Receiver bandwidth and discriminator
range/velocity ambiguities, 313 noise, 271–72
raster displays for, 296–297 Receiver effect on rise time, 262–63
scheduled PRIs, 286 Relationships among three stability
sounds, 297–98 measures, 352
stagger, 285–86 Rise time, 257, 260–64
stagger and delta-T histogram, 306 RMS
transform of PRI, 326 frequency fluctuations, 347–49
Probability of an intercept, 422–27 phase fluctuations, 346–47
Probability of exactly ‘‘k’’ pulses, 415
Probability of intercept, 4, 97–115. See
S
also POI
Probability of one or more pulses, S-band, 2–3
413–14 Scan analysis, 235–54
Propagation effects on polarization, techniques, 252
194–95 Scan and energy on target, 237–38
Index 451
Search Tech-ELINT, 383

example, 253–54 Threshold detection, 45
principles, 236–37 Time and frequency location analysis,
Sector scan, 240–243 159–63
Sensitivity (of receivers), 39–51 Time interval measurements, 287–95
Short look problem in frequency stability Time-bandwidth product, 16–17
measurements, 359–60 Track while scan, 250
Short-term RF stability, 342–60 Tracking scans, 246–50
Sidelobe level and array errors, 208 Transforming parameters, 379–81
Signal identification Trapezoidal pulses
errors, 379 spectrum of, 391–94
guide, 385–386 unequal rise and fall times, 393–94
Signal repeatability and NED, 350–352 Tuning requirements/distortion, 70–73
Simultaneous signal problem, 75–81 Two sigma limits, 373–75
Sinusoidal FM
frequency stability, 359 U
signals, 398–400 U-boat, 2–3
Sliding PRI, 286 Ultimate limits (to ELINT
Spiral scan, 247 measurements), 52–55
Spread spectrum, 14–15 Unambiguous range, 20–23
Stokes’ parameters Unambiguous velocity, 20–23
for polarization, 185–87 Unidirectional sector scan, 241
Superheterodyne receivers, 63–72
Sweeping superhets, 67–68 V
Synchronization effects on POI, 108–14 V-beam scan, 243–244
T W
TDOA location, 149, 155–59 Waveshapes compared, 278–79
TDOA for AOA, 145 Wideband signal trends, 33
TDOA/FDOA location, 151, 159–63 Wigner-Ville transform, 227
Recent Titles in the Artech House
Radar Library
David K. Barton, Series Editor
Advanced Techniques for Digital Receivers, Phillip E. Pace
Advances in Direction-of-Arrival Estimation, Sathish Chandran, editor
Airborne Pulsed Doppler Radar, Second Edition, Guy V. Morris and
Linda Harkness, editors
Bayesian Multiple Target Tracking, Lawrence D. Stone, Carl A. Barlow, and
Thomas L. Corwin
Beyond the Kalman Filter: Particle Filters for Tracking Applications, Branko Ristic,
Sanjeev Arulampalam, and Neil Gordon
Computer Simulation of Aerial Target Radar Scattering, Recognition, Detection,
and Tracking, Yakov D. Shirman, editor
Design and Analysis of Modern Tracking Systems, Samuel Blackman and
Robert Popoli
Detecting and Classifying Low Probability of Intercept Radar, Phillip E. Pace
Digital Techniques for Wideband Receivers, Second Edition, James Tsui
Electronic Intelligence: The Analysis of Radar Signals, Second Edition,
Richard G. Wiley
Electronic Warfare in the Information Age, D. Curtis Schleher
ELINT: The Interception and Analysis of Radar Signals, Richard G. Wiley
EW 101: A First Course in Electronic Warfare, David Adamy
EW 102: A Second Course in Electronic Warfare, David L. Adamy
Fourier Transforms in Radar and Signal Processing, David Brandwood
Fundamentals of Electronic Warfare, Sergei A. Vakin, Lev N. Shustov, and
Robert H. Dunwell
Fundamentals of Short-Range FM Radar, Igor V. Komarov and Sergey M. Smolskiy
Handbook of Computer Simulation in Radio Engineering, Communications, and
Radar, Sergey A. Leonov and Alexander I. Leonov
High-Resolution Radar, Second Edition, Donald R. Wehner
Introduction to Electronic Defense Systems, Second Edition, Filippo Neri
Introduction to Electronic Warfare, D. Curtis Schleher
Introduction to Electronic Warfare Modeling and Simulation, David L. Adamy
Introduction to RF Equipment and System Design, Pekka Eskelinen
Microwave Radar: Imaging and Advanced Concepts, Roger J. Sullivan
Millimeter-Wave Radar Targets and Clutter, Gennadiy P. Kulemin
Modern Radar System Analysis, David K. Barton
Multitarget-Multisensor Tracking: Applications and Advances Volume III,
Yaakov Bar-Shalom and William Dale Blair, editors
Principles of High-Resolution Radar, August W. Rihaczek
Principles of Radar and Sonar Signal Processing, François Le Chevalier
Radar Cross Section, Second Edition, Eugene F. Knott et al.
Radar Evaluation Handbook, David K. Barton et al.
Radar Meteorology, Henri Sauvageot
Radar Reflectivity of Land and Sea, Third Edition, Maurice W. Long
Radar Resolution and Complex-Image Analysis, August W. Rihaczek and
Stephen J. Hershkowitz
Radar Signal Processing and Adaptive Systems, Ramon Nitzberg
Radar System Analysis and Modeling, David K. Barton
Radar System Performance Modeling, Second Edition, G. Richard Curry
Radar Technology Encyclopedia, David K. Barton and Sergey A. Leonov, editors
Range-Doppler Radar Imaging and Motion Compensation, Jae Sok Son et al.
Signal Detection and Estimation, Second Edition, Mourad Barkat
Space-Time Adaptive Processing for Radar, J. R. Guerci
Theory and Practice of Radar Target Identification, August W. Rihaczek and
Stephen J. Hershkowitz
Time-Frequency Transforms for Radar Imaging and Signal Analysis, Victor C. Chen
and Hao Ling
For further information on these and other Artech House titles, including previously
considered out-of-print books now available through our In-Print-Forever® (IPF®)
program, contact:
Artech House Artech House

685 Canton Street 46 Gillingham Street
Norwood, MA 02062 London SW1V 1AH UK
Phone: 781-769-9750 Phone: +44 (0)20 7596-8750
Fax: 781-769-6334 Fax: +44 (0)20 7630-0166
e-mail: artech@artechhouse.com e-mail: artech-uk@artechhouse.com
Find us on the World Wide Web at: www.artechhouse.com

ELINT The Interception and Analysis of Radar Signals The Artech

Uploaded by

Copyright:

Available Formats

ELINT The Interception and Analysis of Radar Signals The Artech

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ELINT The Interception and Analysis of Radar Signals The Artech

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What are some major ELINT signal parameters discussed in the book?

What are some major ELINT signal parameters discussed in the book?

ELINT

The Interception and Analysis of Radar Signals

British Library Cataloguing in Publication Data

Cover design by Igor Valdman

 2006 ARTECH HOUSE, INC.

International Standard Book Number: 1-58053-925-4

4.2.3 Mathematical Models 100

10.9 A Three-Dimensional Search Example 253

12.9 Moving Target Detection 314

15.4 Two-Sigma Parameter Limits 373

About the Author 445

1.1 Electronic Intelligence Defined

1.2 The Importance of Intercepting and Analyzing Radar Signals

an extensive reconnaissance program at the time, it is probable that they would

The military effectiveness of electronic countermeasures, even with emphasis on

1.3 Limitations Due to Noise

1.4 Probability of Intercept Problems

1.5 Direction Finding (DF) and Emitter Location

Of special importance to operational ELINT (OPELINT) is the location of emitters.

1.6 Inferring Radar Capabilities from Observed Signal Parameters

1.7 Receivers for Radar Interception

1.8 Major ELINT Signal Parameters

1.9 The Impact of LPI Radar on ELINT

ELINT Implications of Range Equations

2.1 Range Equations

S R = signal power at radar receiver (W)

S E = signal level at ELINT receiver (W)

The other symbols are defined under (2.1).

In (2.5), R R is in meters because the radar target cross section of 1 m2 is

2.2 Radar Constraints

Figure 2.1 ELINT to radar range ratio.

gigahertz bandwidths and be undetectable by ELINT receivers. Should ELINT

2.2.1 Range Resolution Related to Bandwidth

Figure 2.2 Range resolution related to radar coherent bandwidth.

2.2.2 Spread Spectrum: Radar Versus Communications

Table 2.1 Instantaneous Bandwidth Required for Selected Applications

2.2.3 Moving Targets and Integration Time Constraints

v − [v 2 + 2a(␦ R)]0.5 v − [v 2 + 2a(⌬R)]0.5

Figure 2.3 Range versus time.

2.2.4 Constraints on Time-Bandwidth Product or Pulse Compression Ratio

2.2.5 Constraints on Doppler Resolution

Figure 2.4 Limit on time-bandwidth product.

To avoid large straddling losses, practical radar designs allow a considerable

Clearly, the dispersion of Doppler frequencies due to acceleration must be no

Figure 2.5 Doppler spread and maximum signal bandwidth.

target velocities and accelerations by search radars requires huge processor

2.2.6 Frequency Agility

2.2.7 PRI Agility

PRI 1,000 ␮ s, Vu = 15 m/s and R u = 150 km

As can be seen, the product of unambiguous range and velocity is a constant.

Figure 2.6 Range/velocity related.

PRI 0 = (PD/B e )1/2

A typical value for B e at X-band is 3 kHz, and a typical pulse duration is

2.2.8 Power Constraints

2.2.9 Pulse Compression Modulation Constraints

SNR out = SNR in 冋册