Understanding Radar Systems

UNDERSTANDING RADAR SYSTEMS
UNDERST ANDING
RADAR SYSTEMS
Simon Kingsley
DeparlftJenl of Electronic ond Electriea/ Engineering
University of Sheffield
Shaun Quegan
Deparlmelfl of Applied and Campu.lalional Mathematics
UniwrsitJ' of Sheffield
SciTech Publishing, Inc

Raleigh, NC
w"w.sci{cchpub, com
This is a reprinting of the 1992 edition originally
published by McGraw-Hill Book Company Europe.
© 1999 by SciTech Publishing, Inc.

Raleigh, NC 27613
© 1992 by McGraw-Hill International (UK) Limited.
All rights reserved. No part of this book may be reproduced

or used in any form whatsoever without written permission from
the publisher.
Printed in the United States of America
10 9876543
ISBN 1-891121-05-7
ISBN 13: 978-1-891121-05-0
SciTech books may be purchased at quantity discounts for educational,

business, or sales promotional use. For information, contact the publisher:
SciTech Publishing
911 Paverstone Drive, Suite B
Raleigh, NC 27615
Phone: 919-847-2434
Fax: 919-847-2568
www.scitechpub.com
CONTENTS
Preface xi
Acknowledgements xiii
t Fundamentals t
1.1 What is radar? 1
1.2 A simple radar explained 3
1.3 Overview of radar frequencies 4
1.4 Antenna gain 6
1.5 The radar equation 11
1.6 Accuracy and resolution 15
1.7 Integration time and the doppler shift 18
1.8 Summary 22
1.9 References 23
1.10 Further reading 23
1.11 Problems 23
2 Designing a surveillance radar 25

2.1 Radar and surveillance 25
2.2 Antenna beam width considerations 26
2.3 Pulse repetition frequency and unambiguous range
and velocities 28
2.4 Pulse length and sampling 31
2.5 Radar cross-section 33
2.6 Clutter 38
2.7 Noise 41
2.8 Final design 43
2.9 Summary 45
2.10 References 46
2.11 Problems 46
vi CONTENTS
3 Tracking radar 48
3.1 Introduction 48
3.2 Sequential lobing 49
3.3 Conical scanning 49
3.4 Monopulse radar 50
3.5 Tracking accuracy 54
3.6 Frequency agility 55
3.7 The tracking process 56
3.8 Radar guidance 58
3.9 Summary 60
3.10 References 60
3.12 Problems 60
4 Radar detection theory 62
4.1 Introduction 62
4.2 The basis for decision making-probability theory 66
4.3 The effects of the receiver on the noise distribution 73
4.4 The distribution of signal plus noise 78
4.5 The signal-to-noise ratio 79
4.6 Detection and false-alarm probabilities 81
4.7 The correlation receiver 84
4.8 The matched filter 90
4.9 Key elements of signal detection 91
4.10 Detection using multiple observations 91
4.11 Modifications for intermediate-frequency input
to the receiver 94
4.12 Target fluctuations-the Sweriing cases 95
4.13 Summary 99
4.14 References 101
4.15 Problems 101
5 Signal and data processing 104
5.1 Introduction 104
5.2 Properties of clutter 106
5.3 Moving-target indicator processing 107
5.4 Fast Fourier transform processing 112
5.5 Thresholding 115
5.6 Plot extraction 118
5.7 Plot - track association 119
5.8 Track initiation 120
5.9 Tracking 122
5.10 Summary 125
5.11 References 126
5.12 Problems 127
CONTENTS vii
6 Designing radar waveforms 128

6.2 Bandwidth and pulse duration 129
6.3 Range and doppler accuracy-the uncertainty relation 131
6.4 Resolution 134
6.5 The ambiguity function 138
6.6 Examples of the ambiguity function 142
6.7 Pulse compression 145
6.8 Chirp 145
6.9 Phase coding 150
6.10 Summary 154
6.11 References 156
6.12 Problems 156
7 Secondary surveillance radar 159

7.2 Basic principles 162
7.3 Problems with secondary surveillance radar 164
7.4 Multipath 166
7.5 Mode S and the future 168
7.6 Summary 169
7.7 References 169
7.8 Problems 169
8 Propagation aspects 171

8.2 The radar horizon 172
8.3 Atmospheric effects 173
8.4 Diffraction by the terrain 178
8.5 Battlefield radar systems 179
8.6 Ionospheric effects 181
8.7 Summary 187
8.8 References 189
8.9 Problems 189
9 Radar studies of the atmosphere 190

9.2 Scattering mechanisms 191
9.3 Mesosphere-stratosphere-troposphere radar 197
9.4 Meteor wind radar 200
9.5 Other radar studies of the atmosphere 203
9.6 Summary 203
9.7 References 204
9.8 Problems 205
viii CONTENTS
10 Over-the-horizon radar 206

10.2 Surface-wave radar 207
10.3 Skywave radar 212
10.4 Skywave propagation and frequency management 215
10.5 The over-the-horizon radar equation 220
10.6 Problems with high-frequency radar 223
10.7 Summary 223
10.8 References 224
10.10 Problems 225
11 Radar remote sensing 226

11.1 High-frequency radar scattering from the sea 226
11.2 Measuring ocean currents 230
11.3 Measuring waves 231
11.4 The future of high-frequency remote sensing 233
11.5 Microwave scatterometry 234
11.6 Radar altimetry 237
11.7 Synthetic aperture radar 240
11.8 Summary 252
11.9 References 254
11.1 0 Further reading 254
11.11 Problems 254
12 Ground-probing radar 257

12.2 Designing ground-probing radar systems 258
12.3 Carrier-free radar 260
12.4 Antenna designs 264
12.5 Data processing 267
12.6 Summary 269
12.7 References 269
12.8 Problems 270
13 Multistatic radar 271

13.2 Multistatic concepts 272
13.3 The bistatic radar equation 275
13.4 Multistatic target location 277
13.5 Bistatic doppler 280
13.6 Applications 283
13.7 Summary 284
13.8 References 285
CONTENTS ix

13.10 Problems 285
14 Electronic warfare 286

14.1 Objectives and definitions 286
14.2 Noise jamming and the radar equation 288
14.3 Types of electronic countermeasures and electronic
counter-countermeasures 290
14.4 Stealth applications 292
14.5 Summary 294
14.6 References 294
14.7 Problems 294
15 Recent developments 296

15.2 Phased arrays 297
15.3 Digital beamforming 308
15.4 Active arrays 310
15.5 Multifunction radar 311
15.6 Multihead radar 312
15.7 High-resolution radar techniques 314
15.8 Summary 316
15.9 References 317
15.11 Problems 317
16 The future of radar 319

16.2 Developing the concept of bandwidth 320
16.3 Adaptivity 322
16.4 Summary 325
16.5 Reference 325
Appendices 326
I Symbols, their meaning and SI units 326
II Acronyms and abbreviations 330
III Useful conversion factors 333
IV Using decibels 335
V Solutions to problems 337
Bibliography 364
Index 367
PREFACE
Understanding Radar Systems is a book to convey facts and figures and also
explain why things are the way they are. It is written for students and young
engineers in industry, already competent in electronic engineering or physics,
who need to understand modern radar principles, applications and some of
the jargon used.
There are three parts to the book. The first three chapters form an easy
(and deterministic) guide to the basics of what radar is all about, and what
it can do. Chapters 4-6 are possibly more difficult because probability is
introduced, but these chapters will be necessary for those who wish to begin
calculating the performance of radar systems. The remaining chapters
describe different types of modern radar, the problems and the research areas.
These applications are used to introduce some more new ideas, so that their
relevance is apparent.
This book can be used in several ways: as a 'streetwise' introduction
to radar; as a rough guide to calculating the performance of most types of
radar; or as a handy reference when starting off on a new radar topic, before
going on to the more detailed texts in that area. The most important
thing is to understand the subject and enjoy your career in radar.
Simon Kingsley and Shaun Quegan

ACKNOWLEDGEMENTS
We are grateful to those companies who supplied us with details of

their systems and especially to Marconi who also supplied the images
on the front cover.
The meteor scattering theory in Chapter 9 derives from information
supplied by Dr John Milsom (GEC-Marconi Research Centre) who
continues to be a fount of knowledge concerning propagation and
communications.
Other researchers, past and present, at GEC-MRC who deserve spe-
cial mentions are Matthew Radford, who provided helpful and infor-
mative comments on the draft manuscript and Trevor Blake and John
Dawson for their contributions on over-the-horizon radar.
We would like to thank the UK Ministry of Defence for their support
of various research programmes that led to some of the original mate-
rial included in the contents of this book.
Many of the diagrams in Chapter 6 were produced with the expert
help of Linda Wilkinson of Sheffield University, to whom we express
our thanks.
Finally, we wish to thank all those who offered comments and ad-
vice, and also our wives and children for putting up with us while we
were writing.
Simon Kingsley and Shaun Quegan

CHAPTER
ONE
FUNDAMENTALS
• What is radar?
• What can it do?
• How well can it do it?
A rough-and-ready guide to radar and its capabilities.
1.1 WHAT IS RADAR?
Radar is all about using radio waves to detect the presence of objects and
to find their position. The word radar, first used by the US Navy in 1940,
is derived from radio detection and ranging, thus conveying these two purposes
of detection and location. Modern radar goes further and is being developed
to classify or identify targets, and even to produce images of objects, for
example mapping the ground from a satellite.
The principle of radar is that a transmitter sends out a radio signal,
which will scatter off anything that it encounters (land, sea, ships, aircraft),
and a small amount of the energy is scattered back to a radio receiver, which
is usually, but not always, located near the transmitter. After amplification
in the receiver, the signals are processed to sort out the required echoes from
the 'clutter' of unwanted echoes by a combination of both electronic signal
processing and computer software (data processing).
There are many applications for radar. on scale sizes that vary from a
few centimetres, such as the measurement of the thickness of furnace walls,
to long-range systems probing planets across the solar system. Table 1.1
2 UNDERSTANDING RADAR SYSTEMS
Table t.t Applications of radar
Civil
Ground-based Air traffic control
Sea traffic control
Weather forecasting
Speed traps
Intruder alarms
Radar astronomy
Ground probing
Industrial measurement
Sea-borne Navigation
Collision avoidance
Air-borne Altimeters
Navigation
Weather
Space-borne Studying Earth resources
Sea sensing
Manipulating spacecraft
Mapping planets and minor bodies
Military
Detection of own forces or enemy forces
Tracking of air, sea, land or space targets
Guidance of own weapons systems
gives some idea of the variety of applications, and the list is growing as radar
systems find their way into industry, homes and even the motor car.
There are many systems similar to radar, such as sonar (which uses
sound instead of radio waves), medical ultrasonics and passive detection
systems t. Knowledge of radar principles is a good starting point for
understanding these other subjects and would help a physicist or electronic
engineer, for example, to transfer into medical electronics and body scanning.
Just as in the movie industry, where the apparently immense task of
making a film can be broken down into identifying 60 or 70 key scenes, so
other problems can be broken down. If you find the prospect of learning all
about radar a bit daunting, remember that there are less than 100 key things
you really need to know to understand the basics of radar, and these are
summarized for you at the end of each chapter, as they occur.
t An example of a passive (no transmitter) system is the technique for locating thunderstorms
using several receivers to obtain bearings on the radio signals emitted by lightning flashes; the
lightning is, in effect, acting as the transmitter for the rest of the system. Biologists tracking
animals tagged with tiny radio transmitters have a similar interest in passive detection and
location.
FUNDAMENTALS 3
1.2 A SIMPLE RADAR EXPLAINED
Figure 1.1 shows a basic radar system. For this example we have chosen a
pulse radar, although we shall be looking at other forms of radio transmission
later. The principal radio frequency (RF) of the radar, the carrier, is set by
the frequency synthesizer. This continuous signal is pulsed on and off (usually
spending much more time off than on) by the modulator. The short bursts
of radio energy that result are amplified by the transmitter and sent to the
antenna via a switch called variously a transmit -receive switch, a T - R switch
or a duplexer. There are different designs for these switches, but they all
have the same two functions: to connect the antenna to either the transmitter
or the receiver at the appropriate times, and to protect the receiver from the
full force of the transmitted pulse.
When a pulse is transmitted, the radar clock begins to count time. The
radio pulse travels away from the radar at the speed of light, is scattered
from a target and returns to the radar. The distance to the target, R (called
the range of the target), can then be calculated from the time delay.
Remembering that velocity = distance/time, we can rearrange this as
[m] (I.l )
Stable local Coherent local

oscillator (STALO) oscillator (COLO)
'Baseband'
RF
(e.g. S-band)L-------'
mixer
I
/' "'
{ Raw \
,
\ displa\ I
'- ./
/
Figure 1.1 Block diagram of an elementary radar system, and some abbreviations commonly used.
where R = range [m], C = velocity oflight = 3 x 108 m s - 1 and td = signal

propagation delay [s]. Strictly speaking, an equation such as Eq. ( 1.1 ) should
not have units attached, because the identity is true whatever units are used.
However, it is sometimes clearer if dimensions and units are identified, and
so the appropriate SI units are given. Where units do not aid understanding,
for example in the probability theory in Chapter 4, we have not included
them. When a quantity has no dimensions, we represent this by empty
brackets, thus [ ]. In radar practice, units other than SI are often used,
especially aircraft height in feet, distance in nautical miles and speed in knots.
Conversion factors are given in Appendix III.
The factor 2 in Eq. (1.1) arises because the distance travelled by the radio
pulse is 2R, i.e. to the target and back. A convenient rule to remember is
, 150 m per microsecond', meaning that, after transmission, each extra
microsecond that elapses before the echo pulse is received implies that the
target is 150 m further away. When discussing range, it is common practice
to find it being used synonymously with time delay, since they are related
by Eq. (1.1). Also, it is quite usual for the duration of radar pulses to be
described as lengths, as in 'a pulse length of 1 flS'.
When the echo pulse returns to the radar, it is absorbed by a receiver,
which has been carefully designed to ensure that none of the energy in these
very weak signals is wasted. After amplification and conversion to an
intermediate frequency (IF) where the electronics is easier to engineer, the
signal is detected and then processed ready for display to a human operator.
The plan position indicator (PPI) is one type of display that is familiar to
many people; a rotating trace produces a circular map in which the radar is
represented as being at the centre and range is represented as the distance
towards the edge of the display. This display can be 'raw' information
containing all the echoes received by the radar, but nowadays it is more
common for the display to be synthesized such that only the desired
information on targets of interest is displayed.
1.3 OVERVIEW OF RADAR FREQUENCIES
Radio waves are part of the electromagnetic spectrum and so they could be
described by either quantum theory or wave theory. It turns out that, at the
frequencies used by radar, it is much more useful to think in terms of waves,
and we do not refer to quanta in the rest of this book. The frequency f
[Hz = s - 1] t of an electromagnetic wave is related to the wavelength A. by
C = fA [m s - 1 ] ( 1.2)
t The hertz [Hz] is named after Heinrich Rudolf Hertz, a professor at Karlsruhe, who is
widely credited as having founded radar because of his work in the late 1880s, which included
the reflection of radio waves generated by a spark gap generator.
FUNDAMENTALS 5
where c is the velocity of light, which is about 3 x 108 m s - 1 in air or space

but may be quite a lot less in other materials such as coaxial cables.
The lowest frequency useful for radar purposes is probably about 1 MHz.
Such low frequencies propagate very long distances, but the problems of
using them get worse as the frequency gets lower. Some of these problems
are as follows:
1. Antennas become very large at low frequencies; for example, at 1 MHz,

a quarter-wavelength monopole would be 75 m high, and it becomes
impracticable to build high-gain antennas.
2. The ionosphere is a strong scatterer of low frequencies and gives rise
to unwanted echoes, which, because the ionosphere is not a static
phenomenon, can be confused with moving targets. Given an antenna
with a very narrow beam, it might be possible to arrange search patterns
that avoided looking in the direction of the ionosphere, but because of
the difficulties described in item 1, this is not a practical option.
3. Long radio wavelengths mean that only small changes in frequency occur
when signals are scattered from moving targets (the doppler shift), and
such small changes take a long time to resolve. Often, targets can only
be distinguished from the background of stationary echoes by means of
these changes in frequency, and long-wavelength radars can become
ineffective at detecting slowly moving targets.
4. At low frequencies there are lots of logistical problems, such as the
difficulty of obtaining a transmitting licence and the problem of finding
a suitable radio channel free of other users. Also, background noise levels
are generally high and only narrow radar bandwidths are available,
which, as we shall soon see, restrict the resolution of systems.
Despite all these problems, the rewards for long-range over-the-horizon

detection of targets are so attractive that radars do operate at low freq uencies,
especially in the high-frequency (HF) band, which is from 3 to 30 MHz.
Over-the-horizon radar is explained more fully in Chapter 10.
At the higher end of the spectrum used by radar systems, most problems
encountered are connected with atmospheric absorption and the limits of
available technology. The very highest-frequency radars are laser systems
called lidars (derived from light detection and ranging), which are usually
used for accurate distance measurement, atmospheric studies and in some
defence applications. Lidars have a practical upper frequency limit of about
3 x 10 14 Hz or 111m, set by atmospheric absorption.
The highest operating frequencies used by 'true' radar (using radio
frequencies rather than light) are confined to various atmospheric 'windows',
which are frequency bands where the atmospheric absorption is not too
severe, such as at 35 GHz and near 94 GHz. There are not a great
many electronic components available at these frequencies, although this is
High Very high Ultra high Super high Extremely high

frequency frequency frequency frequency frequency
(HF) (VHF) (UHF) (SHF) (EHF)
Frequency(f) 3 MHz 30 MHz 300 MHz 3GHz 30 GHz 300 GHz
:>
X~~ ~millimetre
Commonly used US
hand names L S C XJK )VO
, UK lEE
- N "T oc~;S~ ~~~GHz
A B C D E FGHI
, J K LM
NATO system
2 - N ' 'T\Ooco 0 0 - xGHz
I/)
N _ "T~
- N
::;;
Wavelength (Aj 100 m 10m

1
1m IOcm Icm Imm
Figure 1.2 Radar frequency band names. The code letters L, S, C, X and K were used for
security reasons during World War II and have been widely adopted by radar engineers ever
since, despite the introduction of more rational systems.
improving with the development of gallium arsenide technology. The

wavelength at 35 GHz is less than a centimetre, and so the term millimetric
radar is sometimes used to describe these systems. The range of millimetric
radars is quite short, because of the absorption problems, but they tend to
be compact devices that can have applications such as in the guidance system
on a missile. The possibility exists for using optical-type technology at
millimetre wavelengths, with the development of devices such as radio lenses
and imaging systems, which can be useful when trying to identify targets.
In between the two frequency extremes, the spectrum has been divided
up into bands having names that have become part of the jargon of radar.
Figure 1.2 explains some of these. The World War II code letters L, S, C,
etc., are part of the everyday language of radar engineers, so it is necessary
to learn them if you wish to appear knowledgeable! More recently, the
alternative NATO band names are increasingly being used.
1.4 ANTENNA GAIN
The antenna is the device for radiating and receiving electromagnetic energy
and it has three main functions:
1. To 'beam' power in a given direction in order to increase the radar

sensitivity in that direction.
FUNDAMENTALS 7
2. To provide for beam steering such that some area of coverage can be
provided.
3. To permit the measurement of angular information so that the direction
of a target can be determined.
When considering radar antennas it turns out that much of the theory
developed in physics courses for optics and optical astronomy is very useful.
The concept of reciprocity is especially helpful in allowing us to say that, if
transmitter power is fed into an antenna and it spreads out in a certain
angular pattern, then, when the same antenna is used for receiving, it has a
sensitivity to incoming signals that follows the same pattern. It is common
practice to mix these two cases, and you can hear antenna patterns
and polar diagrams being discussed one moment as if the antenna were
transmitting and the next moment as if it were receiving.
If an antenna were omnidirectional, it would radiate power uniformly
over the whole 4n steradians of a sphere. (Do not worry if you have not
come across steradians. They are the unit of 'solid' or three-dimensional
angle, and are not crucial to what follows.) Such an antenna is called an
isotropic radiator, but it would not be useful to us because, first, it would
not help with the three functions we require of it and, secondly, it can be
demonstrated mathematically that it cannot exist in practice. It is more
common to build antennas designed to focus or concentrate power in a
certain direction. This property of beaming power is known as the directive
gain D(O, <jJ) or the power gain G(O, <jJ). These terms are often abbreviated
to directivity D or gain G.
While the directive gain describes the radiation pattern of the antenna,
it does not give any indication of how lossy the antenna is. The power gain
includes the concept of losses, which occur through heating of the antenna
itself, the ground plane and any matching devices, as well as power radiated
into sidelobes. Power gain may be defined as the ratio of the radiation
intensity in the main lobe of the antenna to the radiation intensity from a
100 per cent efficient isotropic antenna having the same power input. The
gain of a lossless antenna is given by
[ ] ( l.3a)
where 110 = width of the beam in the azimuth direction [radians] and
11<jJ = width of the beam in the elevation direction [radians].
The azimuth angle 0 gives directional or bearing information and, as on
a magnetic compass, it is measured clockwise from the north. The elevation
angle <jJ is measured from the horizon upwards. Any point in the sky can be
specified by quoting its azimuth and elevation (see Fig. 1.3). If 110 and 11<jJ
/ G a i n (6. d»
--~
Horizontal
Relative gain
Elevation d>
Figure 1.3 Azimuth and elevation angles and beamwidths.
are expressed in degrees, then Eq. (1.3a) may be rewritten as

41253
G~-- [ ] (1.3b)
MrA¢o
(the constant 41253 is the number of square degrees on the surface of a
sphere). This formula takes no account of the beamshape and yields a result
that is about 25 per cent or so too high. For rule-of-thumb calculations, a
better constant to choose is 32000, and when typical losses are included the
power gain is usually about
25000
G~--- [ ] (1.3c)
MrA¢o
Occasionally Eq. ( l.3a) is given as G = n 2 / MJ A¢, which assumes a gaussian
beamshape with M) and A¢ as the half-power beam widths (see Skolnik!).
If you feel comfortable with solid angles, you can also use
G ~ 4n/0 [ ] (1.4 )
where 0 = solid angle of the beam [steradians].
Antenna patterns are directional in three dimensions, and producing one
is a bit like squeezing a spherical balloon to produce a protrusion; if you
want G > 1 in some particular direction, then you have to crush the rest of
it and accept G < 1 somewhere else. The important point is that the gain
integrated all the way round an antenna adds up to unity (or less than unity
if efficiency factors are included). The term 'power gain' is most commonly
FUNDAMENTALS 9
used to denote the maximum gain of an antenna, but beware of the frequent
use of the term to describe how the antenna's radiation pattern changes
with angle-'the gain falls off rapidly with elevation angle ... ', etc.
The maximum gain of an antenna can also be calculated from its size:
[ ] (t.5 )
where Ae = effective area of the antenna. The effective area is usually less
than the real area A (area of a parabolic dish, for example) and is related
to it by
Ae = eA (1.6 )
Here e [ ] is an efficiency factor, usually falling in the range 0.4-0.9 for
parabolic dishes, which arises because it is difficult to illuminate a dish
perfectly from the radio source at the centre. Some antennas, such as those
used by domestic televisions, can have effective areas greater than the physical
cross-section of the device.
Where does Eq. (1.5) come from? Any source of waves having a linear
dimension d, which is large compared to the radio wavelength, tends to have
a characteristic beam width A() in the appropriate plane given by
[radians] (1.7 )
We can explain Eq. (1.7) using an optics analogy. Wavelets coming from
many small (Huygens) sources along the aperture d interfere at some distant
point, as in Fig. 1.4a. Directly in front of the source, we find that the
wavefronts add constructively and a beam is formed; but as we begin to
move off-axis, they begin to interfere with each other because of the differing
path lengths. The actual signal received at a point off-axis depends on the
(a) (b)
+-------d-------+
Figure 1.4 (a) Wavelets from small Huygens sources add coherently in front of the aperture.
(b) At an angle sin () = A./ d they interfere destructively and the antenna gain falls to zero, thus
giving an approximate characteristic beamwidth to the first nulls of 22/d and A./d to the -3 dB
(half-power) points.
summation of the contributions from all the Huygens sources. We can

estimate that, at angles where the contribution from the midpoint of the
source has an extra path length of )./2 (Fig. l.4b) over a source near
the edge, the process becomes destructive and the beam degrades (the
contributions from either side of the centre have smaller and greater
differences, which tend to cancel). The off-axis angle at which this occurs is
given by
e '" sin e = )./2 == ~ [radians]
d/2 d
As destructive interference occurs at the same angle each side of bore-sight,
we can say that the characteristic beamwidth to the first nulls is given by
2)./d. (Optics books have more details; see for example Chapter 19 on
Fraunhofer diffraction in Pedrotti and Pedrotti 2 .) It is more common,
however, to find radar beam widths described in terms of the width of the
beam at the half-gain points (also called the half-power points or - 3 dB
points). Because of the roughly triangular shape of the beam, the beamwidth
at this point approximates to ).jd as in Eq. (1.7). The identity in Eq. (1.7) is
well known and is a much used approximation.
Imagine, now, a rectangular aperture having dimensions d 1 and d 2 as
shown in Fig. 1.5. The azimuth and elevation beamwidths are given by
L1e '" i./d 1 [radians]
From the definition of G
G~~= 4nd 1 d2 = 4nA []

L1e L1<jJ },2 ;. 2
Worked example The Siemens- Plessey type 46C weather radar is a

C-band system optimized for 5625 MHz and using a 2.44 m diameter
---
Figure 1.5 The azimuth and elevation beamwidths arising from a rectangular aperture.
FUNDAMENTALS 11
parabolic reflector antenna. If the antenna efficiency factor e is 0.5, what

gain and beamwidth would you expect the system to have?
SOLUTION The physical area of the antenna aperture A is given by

nd 2 /4 = 4.68 m 2, so the effective area Ae = 2.34 m 2. The antenna gain
can now be calculated from Eq. (1.5) as
G = 4nAe = 4n x 2.34 == 10350 == 40.1 dB [J

).2 (0.0533)2
At first glance the beamwidth might be expected to be given by Eq. (1.7)
as
). 0.0533 .
A() = A</> - - = - - == 0.02 radIans == 1.25°
d 2.44
but in practice the effective value of d is less than 2.44 m because of the
efficiency factor, and the beam width might be expected to be slightly
greater thaT} 1.25°.
Siemens- Plessey Radar Limited quote the gain as 40 dB and
the beamwidth as IS (to the half-power points) in their technical
specifications for this radar 3 .
All these arguments involve approximations. More rigorous calculations

of antenna gain require a more careful definition of what is meant by
beamwidth. There are also many types of radar antenna other than parabolic
dishes, and we will be describing some of them later in this book as we look
at the different applications of radar.
1.5 THE RADAR EQUATION
The ability of a radar to detect the presence of a target is expressed in terms

of the radar equation, which is worth deriving, rather than just quoting,
because of the insight it gives into the way radars work.
We begin with the transmitter, which has a peak power output PI [W].
If this power is radiated isotropically by the antenna, then the power flux
(the power density per unit area) at a range R is given by
Power flux at distance R = PI 2 (1.8 )

4nR
because 4nR2 is the area of a sphere of radius R through which all the power
must pass.
If the transmitting antenna is not isotropic and concentrates the power
towards the target, then we modify Eq. (1.8) by introducing the gain factor
Gt • The power flux in the direction of the beam is now
Power flux at the target = -PtG

-2
t
( 1.9)
4nR
The target intercepts a portion of this incident power and re-radiates it. The
measure of the incident power intercepted by the target and radiated back
towards the radar is called the radar cross-section, which is often abbreviated
to ReS and is given the symbol Cf. The ReS of a target has units of area
and indicates how large the target appears to be as viewed by the radar.
Res is defined as the power re-radiated towards the radar per unit solid
angle divided by the incident power flux/4n steradians.
In reality, the target may be physically much larger in area than the
ReS but trying to keep a low radar profile (such as a Stealth aircraft), or
it may be very small but trying to make itself appear visible on a radar screen
(such as the corner reflecting antennas used on buoys and yachts). There is
no fixed ReS for a target, no number that can be painted on the side to
say how big it appears to a radar set. The ReS of a target depends on the
angle of incidence at which it is viewed, the radar frequency and the
polarization used. The ReS also fluctuates with time, as we shall see later.
The power re-radiated by the target is now
. PtGtCf
Power re-radlated = --2 [W] (1.10)
4nR
On the return path this power again spreads out over the sphere of area
4nR2. Although it does not usually spread out uniformly, the 'gain' of the
target is automatically included in the concept of the ReS. The power density
at the radar thus becomes
(1.11)
The amount of this returning power that is intercepted by the antenna is

determined by its effective area Ae. The mean power received by the radar
Pr is thus
p = PtGtCfA e [W] (1.12)
r (4nR2)2
The next move is to substitute for Ae by using Eq. "( 1.5) :

Gr = 4nAe/ ;.2 []
where Gr = gain of the receiving antenna. Finally, the inevitable inefficiencies
in a radar system must somehow be introduced and, for now, this is best
done by lumping them all together as a system loss factor Ls [ ]. Loss
factors may be arranged to appear on either the top or bottom of an equation,
FUNDAMENTALS 13
but we will adopt the convention that Ls is always less than 1, and therefore
appears on the top. Using this definition, the power received by the radar,
from the target, is given (with appropriate units also shown) by
__ Pt [W]Gt [ ]Gr [ ]a [m 2 ]).2 [m 2 ]Ls [ ]
P [W] (1.13)
r (41r)3 [ ]R4 [m4]
Although Eq. (1.13) is a complete description of the power received, it is still

not useful because it does not indicate whether this power is larger or smaller
than the background noise level. Unfortunately, noise is always present,
either as internal noise from the electronics, or as external noise from such
sources as the galaxy, the atmosphere, man-made interference or even
deliberate jamming signals. All these noise sources are wideband compared
to the radar signal, and one of the functions of a radar receiver is to tailor
the bandwidth to accept the signal, without permitting any unnecessary
further noise to enter. If we were to examine the analogue front end of a
radar system using a fast oscilloscope, we would be able to see the noisy
signals and perhaps be able to pick out the shape of the echo pulses, if they
were large enough. After the analogue-to-digital (A/D) converter, however,
there is only one sample for each range gate, which represents the sum of
the signal and noise at that point in time.
We will spend more time looking at the properties of noise later, but
for now we will simply say that there is an average noise power present in
the system, to which we will give the symbol N (for typical values see Sec.
2.7). We can now compare the power received from the target with the noise
power, in what is variously known as the signal-to-noise ratio, SNR or S / N:
[ ] (1.l4 )
This is the all-important radar equation, which is much used in one form
or another. It is perhaps a little too complicated to learn, and the best
strategy is probably to remember how to derive it quickly from the basics.
Often, the radar equation is used to solve for one unknown. For example,
supposing a particular SNR is required for reliable target detection (a typical
figure might be 13 dB, which is 20 times noise). The maximum detection
range Rmax of a given radar can be calculated from
[m] (1.15)
Worked example A short-range surveillance radar operates at 3 GHz

and uses aIm diameter dish for both transmitting and receiving. If the
mean transmitter power is 10 kW and the noise level is -140 dB W (see
Table 1.2 A worked example using linear values or decibels
Multiply these Add these

OR
terms terms
Symbol Linear value dB Comments
1'. 10000 [W] 40 [dB W] Beware mistaking k W for watts

G, 1000 [ ] 30 [dB]
Gr 1000 [ ] 30 [dB]
(T I [m2] o [dB m 2]
22 0.01 [m 2] -20 [dB m2 l
L, 0.32 [ ] -5.0 [dB]
1/(411:)3 1/2000 [ ] -33.0 [dB] (411:)3 - 2000 is a useful
approximation to remember
I/N I x 1Q14 [W- 1 ] + 140.0 [dB W- 1 ]
I/(SNR) 1/20 [ ] -13.0 [dB]
R!ax 8.0 x 10 16 [m4] 169.0 [dB m 4]

Rmax -16.8 kIn -16.8 kin
Appendix IV), calculate the maximum range at which a small aircraft

of radar cross-section 1 m 2 could reliably be detected t . Assume 5 dB
losses and a SNR of 13 dB.
SOLUTION The only difficulty in solving this type of problem is in making

the appropriate rearrangement of the radar equation. In this case, it has
already been done in Eq. (1.15). AIm diameter dish has an area of
nr2 = 0.25n m 2 . At 3 GHz,A. = 0.1 m,giving the gain of the antenna as
G =G = 4nA = 4n x 0.25 x n _ 1000 == 30 dB []
t r A.2 (0.1 )2
In practice, the gain would be about half this value because of the
efficiency factor of the dish. Calculations can now be carried out either
linearly or in decibels, whichever you find most convenient. Decibels are
explained in Appendix IV, and the advantage of using them is that you
have only to add and subtract reasonably sized numbers instead of
multiplying and dividing by mind-bogglingly large figures. You can use
a table to calculate the answer (see Table 1.2). Those of you familiar
with the use of spreadsheets will find this a convenient way to handle
such a radar calculation.
t What is meant by reliable detection? We have to consider both the probability of detecting
the target Pd and the probability that we might create a false alarm Pfa in a sensitive system
by declaring a target to be present when in reality only noise is present. In Chapter 4 we discuss
how to calculate these probabilities, but as a working rule if Pfa is set at 10- 6 and we require
an 85 per cent probability of detection on a single observation. this requires a SNR of around
13 dB.
FUNDAMENTALS 15
COMMENTS This type of problem is used quite often as an exercise in

familiarization with the radar equation and the size of the numbers
involved. The object is not usually to get exact answers but rather to
compare one system with another or to find out what happens when
one of the parameters is changed. Suppose we wanted to rework this
example to find Rmax for a parabolic dish with twice the diameter (and
thus four times the area). We could do this in the dB column by changing
both 30 dB values to 36 dB and the total from 169 tol181 dB, giving
Rmax == 33.5 km.lWe could also get the same answer by using the linear
column, but it would mean having to do all the multiplying again. Note
that doubling the size of the dish doubles the range; why is this?
In the example above, the two really big factors are the R4 propagation
losses and the noise. Very little can be done about either of these and the
overall detection range of radars is thus fairly predictable; it is the cleverness
and the adaptability of radars that improves with time, rather than the
absolute detection range. Also, the radar cross-section used in Eq. (1.14)
fluctuates continuously for most real targets and we must use statistical
means to describe it. This infers that the radar equation itself is a statistical
method of detecting targets and measuring range, rather than being a
deterministic calculation; we say more on this in Chapter 4.
1.6 ACCURACY AND RESOLUTION
How well can a radar measure range? To answer this, we must be a bit more
careful with our question and define what we mean by the words 'how well'.
The range accuracy indicates the uncertainty in a measurement of the absolute
distance to an object, whereas the range resolution tells us how far apart two
targets have to be before we can see that there are indeed two targets rather
than one large one.
The range resolution of a radar system is fairly straightforward. If the
time delay between the echoes from two objects is greater than the pulse
duration r, then two separate echoes are seen (see Fig. 1.6a). If the targets
are closer than r, the echoes will merge (Fig. 1.6b). When the echoes are
separated in time by an amount similar to the pulse duration, then they
become resolvable (Fig. 1.6c; see also Fig. 6.1). This is like the Rayleigh
criterion used in optics, and combining it with Eq. (1.1 ) gives an expression
for the range resolution llR:
llR = crf2 [m] (1.16)
Usually the radar receiving system samples the receiver output every
r seconds, and each sample represents a distance llR called a 'range gate'
or a 'range bin'. A radar using a pulse duration r = 1 JlS would thus employ
(a)
Target I Target :1
(0)
(e)
Figure 1.6 Radar range resolution. (a) Two targets are easily resolved when they are more
than a pulse length apart, (b) unresolvable when they are much closer than a pulse length, and
(c) just resolvable when separated by a pulse length.
a 1 MHz AID converter to sample the receiver output every 150 m in range,
out to the practical maximum range of the system.
Unfortunately, understanding the range accuracy of a system is not as
simple as the resolution. Intuition tells us that the accuracy of a range
measurement should depend on the 'sharpness' of the pulseshape, and it is
therefore rather surprising to discover that the crucial factor determining the
range accuracy is the bandwidth occupied by the radar. Figure 1.7 gives some
insight into why there is this dependence on bandwidth. The system shown
in Fig. 1.7a emits a single continuous tone and measures the phase of the
echo to find a rough position for the target, but it is ambiguous every
wavelength. In Fig. 1.7b a second frequency is added to the transmission to
reduce ambiguities and sharpen the position of the target. Adding further
frequencies (more bandwidth) eliminates the ambiguities and gives even
greater accuracy to the range measurement (Fig. 1. 7c).
In practice, pulseshape and bandwidth are related in simple pulse radars.
Short pulses take up more bandwidth B of the radio spectrum than long
pulses, a result well known to those familiar with Fourier transforms and
FUNDAMENTALS 17
A phase measurement of 90° indicates that the target

(a) / is i/ne of these loea~ns ~
E
(b)
(e)
E
Target
Figure 1.7 Radar range accuracy dependence on bandwidth. (a) With a single frequency,
accuracy is poor and the target position is ambiguous. (b) With two frequencies, accuracy and
ambiguity are improved. (c) Using more bandwidth, ambiguities are removed and accuracy
improves further.
frequency-domain thinking. For those who are neit so familiar, it is perhaps

sufficient to say that the pulseshape can be built up by adding together a
set of sinusoidal waves and that, the shorter the pulse, the higher the frequency
of the sinusoids needed to reproduce its shape. For a pulse duration of! it
is a reasonable approximation to say that
B", II! (1.17)
To make a more precise statement would require careful definitions of what

is meant by 'bandwidth' and 'pulse duration '. We cover these aspects later
in the book but, for quick calculations, take ! as the time between the
half-power points of the pulse (the 3 dB points) and measure the bandwidth
between the half-power points of the spectrum.
Using Eq. ( 1.17) we can say that a radar with a 1 MHz bandwidth might
be expected to be transmitting 1 JJ.s long pulses. In turn, this corresponds to
a basic range measurement step size of 150m using Eq. (1.16). To achieve
15 cm resolution would require a shorter pulse with T = 1 ns, so that a

bandwidth of roughly 1 GHz would be needed.
It is important to be aware that the bandwidth of a radar does not have
to be limited by Eq. (1.17). As an example, suppose we developed a system
that transmitted long pulses during which we swept the frequency of the
oscillator deliberately to increase the bandwidth; could we achieve greater
range accuracy this way? Radars using such modulation schemes are
common, and are known as chirp systems when the frequency sweep during
a pulse is linear. By careful processing chirp radars do indeed achieve high
range accuracy.
The other factor determining the accuracy of the range measurement is
the signal-to-noise ratio, because of the effect that noise has on corrupting
the shape of the pulse. Noise is always present and degrades measurements
in any system. For example, the length of a line drawn on a piece of paper
can be estimated with a centimetre ruler to a lot less than a centimetre under
a bright light (SNR > 1). In a dimly lit room (SNR - 1) however, it is hard
even to estimate it to the nearest line on the ruler and the measurement
cannot be made better than the basic step size of one centimetre.
In general, any measurement made with a basic resolution of M will
have a root-mean-square (RMS) error bM given by
bM- M [units of M] (1.18)

- )(2 x SNR)
For the range step size ~R given by Eq. (1.16), substituting for Tusing
Eq. (1.17) gives the range error as
c
bR - - - , - - - - - [m] (1.19)
- 2BJ(2 x SNR)
Again, more precise statements could be made if we were to be more careful
over definitions, especially the shape and effective bandwidth of the pulse
being used. We explore this further in Chapter 6.
As an example, a radar using 1 IlS pulses with associated bandwidth of
about 1 MHz might be expected to be capable of determining the range of
a target observed with a SNR of 20: 1 (13 dB) with an error of about 24 m.
For a target detected at poor signal-to-noise ratio (SNR - 1), not only
would the detection be uncertain but we could do no better than to say in
which range bin the target occurred, i.e. bR - ~R.
1.7 INTEGRATION TIME AND THE DOPPLER SHIFT
Imagine that you are out walking and, glancing briefly over your shoulder,
you spot a helicopter flying in the distance. With such a short glimpse it is
FUNDAMENTALS 19
impossible to tell whether the helicopter is moving or not, and you stop to
look more carefully. It soon becomes obvious that the helicopter is not
moving quickly, but is it hovering or moving slowly? You stare at it for a
long time and decide that it really is hovering. What you have discovered
is that measurements of the position of a target can be made quickly, but it
takes time to estimate velocities and to distinguish differences in velocities.
The smaller the velocity difference, the longer the time needed to estimate it.
The length of time taken to make an observation with a radar set is
called the integration time, because all the data on a target are integrated or
added up until the measurements are sufficiently accurate. It is exactly like
photographing moving objects: a night-time photograph of the sky showing
a meteor burning up, taken with an exposure of 1/1 OOOth of a second,
will show only a faint object, which appears almost frozen in the sky;
another photograph taken by leaving the shutter open for half a second will
show a much brighter object, whose speed can be estimated from the length
of the trail.
How can the speed of an object be measured by radar? There are two
methods. The simpler method is similar to watching the helicopter or taking
a photograph, in that the speed of a target is estimated from the way its
position changes with time (except that optical systems measure transverse
velocities and position, whereas radar range information gives the radial
component). Observing changes in range is not a very accurate method for
a radar to measure speed, although tracking the target for a long time can
improve the estimate. This technique is used by incoherent radar systems, in
which the receiver is tuned to the same frequency as the transmitter but is
not phase-locked to it, and so is unaware of any small drifts in frequency
between them.
A more accurate way of measuring target speed is to make use of the
doppler shift, which is the change in the frequency of the radio signal caused
by the motion of the target. The doppler t shift is named after Christian
Johann Doppler (1803-1853), who pointed out that the colour of a luminous
body and the pitch of a sounding body are changed by the relative motions
of the body and the observer. In order to detect small changes in frequency,
a coherent system is needed in which the transmitter and receiver oscillators
are phase-locked to reveal any difference in the echo frequency.
Figure 1.8 shows a radar observing a jump-jet. If the jet is hovering,
each radio signal sent out by the radar will return with the same phase (as
measured with respect to the transmitted signal). If the jump-jet approaches
the radar, then each radio signal has to travel a shorter distance and the
phase of the echo will change continuously with time. Every A/2 through
t Although it seems disrespectful to previous generations of scientists, the modern convention
is that capital letters are not used to describe the effects named after them (e.g. gaussian noise,
doppler shift). The exception is when their name is used as a unit, for example electrical currents
are measured in amperes (or amps), but the unit is A.
Figure 1.8 The motion of a target causes a change of phase in the radar signal, equivalent to
a frequency shift.
which the target moves means that the path length has been shortened by
A. and the phase of the echo will have changed by 2n or, in other words, will
have rotated by one complete cycle. If the jet were to fly at ;./2 [m s - 1 ]
towards the receiver, the radar would detect a 1 Hz change in the radio
frequency. If the jet flies at Vr [m s -1], then we must work out what this
represents in multiples of A.12 in order to calculate the corresponding doppler
shift Id; this leads directly to the formula (another one worth learning)
[Hz] (1.20 )
where Vr = radial component of the target speed towards the radar.
Transverse components of velocity do not contribute to the doppler shift
because the target neither approaches the radar nor recedes from it. Because
radars measure only radial components and optical systems, such as television
and infrared cameras, measure only transverse information, it is not
uncommon to find the two types of system integrated together in 'point' or
'local' defence arrangements.
The expression in Eq. (1.20) is actually an approximation to the full
relativistic formula, but it is valid for all cases where Vr « c, which is almost
always the case-very few radar targets approach the speed of light!
Very roughly, the integration time t needed to resolve two doppler
frequencies separated by Aid is given by
[s] ( 1.21 )
and using Eq. (1.18) again we can say that the error bid in the measurement
of doppler frequency is given by
lJJ. 1 [Hz] (1.22 )

d'" tJ(2 x SNR)
FUNDAMENTALS 21
Combining this with Eq. (1.21) means that the error c5vr in the velocity
measurement is
A.
c5v ~ --,---- (1.23 )
r 2tJ(2 x SNR)
The formulae above are not exact, because it all depends on how the
measurements are made. For example, if the doppler measurement is made
during a single long pulse, then the exact formula depends on the shape of
the pulse; if the measurements are made by integrating many pulses, then
the accuracy of the result depends on the 'weighting' or how the information
is put together. More details can be found in Chapter 4 and in Skolnik I.
Worked example A ship sailing at a radial speed of 5 m S-1 is observed

entering a harbour by a high-frequency (HF) radar operating at 3 MHz
and an S-band radar operating at 3 GHz. What observation times would
the two systems need to distinguish the ship from the background echoes
of the land? If the signal-to-noise ratio is 20 dB in both cases, how
accurately could the velocity be measured?
SOLUTION At 3 MHz the radio wavelength is 100 m and the doppler

shift of the ship echo can be calculated from
This frequency must be distinguished from the land echoes, for which
Id = 0, and so the frequency difference !lId = 0.1 Hz. From Eq. ( 1.21 ) we
can say that the observations must be made over to s for the ship to be
distinguished. A pulsed HF radar usually sends out several hundred
pulses per second, and so a few thousand echo pulses would be collected
together for doppler processing when the radar was in ship detection
mode.
The microwave radar uses A. = 0.1 m, and when this is inserted in
Eq. (1.20) the doppler shift for the ship comes out to be 100Hz. An
observation period of only 0.01 s or to ms would therefore be needed
to separate the ship and land echoes, and typically this would be achieved
using a short burst of about 16 pulses.
A SNR of 20 dB means that the signal power is 100 times the noise
power, and putting this into Eq. (1.23) gives velocity errors of 0.5/fi == 0.35
ms- I for both radar systems. Try it!
1.8 SUMMARY
Radar is used to detect the presence of an object and measure its position
and speed. It is possible to make a swift estimate of the performance of a
radar system using the formulae given below.
Key equations
• The range of a target:

R = c'd/2 [m]
• Relation between frequency and wavelength of an electromagnetic wave:
c = fA
• The power gain of a typical antenna:
• The maximum gain of an antenna:

G = 4nA.IA2 []
• Characteristic beam width :
[radians]
• Signal-to-noise ratio:
[ ]
• Range resolution:
AR = c,/2 [m]
• Bandwidth:
B", II, [Hz]
• Range error:
c
bR ~ 2BJ(2 x SNR) [m]
• Doppler shift:
fd = 2vr l A [Hz]
FUNDAMENTALS 23
• Error in measuring doppler frequency:
<>/. 1 [Hz]
d - tJ(2 x SNR)
• Error in measuring radial velocity:
A
<>v ----,--------
r 2tJ(2 x SNR)
1.9 REFERENCES
1. Introduction to Radar Systems, M. I. Skolnik, McGraw-Hill, New York, 1985. [This is the
classic paperback on radar systems, but is more of a reference work than a good read.]
2. Introduction to Optics, F. L. and L. S. Pedrotti, Prentice-Hall, Englewood Cliffs, NJ, 1987.
3. The Plessey Type 46C Weather Radar, RSL 1479, Issue 8, 1983.
4. Racal-Decca 2690 BT Marine Radar Series, Racal Marine Electronics publication reference
RMEjOOO8j1187 JAD.
5. Plessey ACR 430 Airfield Control Radar, RSL, Issue 2,1987.
1.10 FURTHER READING
There are many specialist technical books on different aspects of radar

including a whole 'library' published by Artech House. These are not for
the uninitiated, however, and below is a selection of alternative texts that
the beginner might consider reading to get different perspectives on the way
radars work.
Radar Principles for the Non-specialist, J. C. Toomay, Van Nostrand Reinhold, New York,
1989. [Patchy in coverage, but contains some useful explanations.]
Radar Systems, P. A. Lynn, Macmillan, London, 1987. [Concentrates principally on air traffic
control applications.]
Understanding Radar, H. W. Cole, BSP Professional Books, Oxford, 1985. [Readable,
concentrating mainly on primary and secondary surveillance radar.]
Radar Principles, N. Levanon, Wiley, New York, 1990.
Technical History of the Beginnings of RADAR, S. S. Swords, Peter Peregrinus for the IEEE,
Stevenage, Herts, 1986. [A mine of historical information.]
Radar Design Principles, F. E. Nathanson, McGraw-Hili, New York, 1969. [For the 'advanced'
beginner.]
1.11 PROBLEMS
1.1 The Racal-Decca 2690 BT Marine Radar Series4 offers radar options that include (a) a
2.7 m wide antenna operating at a wavelength of 3 cm and (b) a 3.6 m antenna operating at
10 cm. Calculate the horizontal beam width of the two antennas, assuming each has a linear
efficiency factor of 80 per cent.
1.2 The vertical beamwidth (-3 dB) is quoted as 23" for the 3 cm version of the 2690 BT
system. What would you expect the antenna gain to be?
1.3 One of the 2690 BT radar bandwidth settings is 4 MHz. What range resolution would this
give?
1.4 The Siemens- Plessey ACR 430 5 is an X-band airfield control radar for the close control
of aircraft approach to land in poor weather conditions. The effective aperture size is 3.4 m
horizontally by 0.75 m vertically. What would be the beamwidths at an operating frequency of
9.40Hz?
1.5 What antenna gain would you expect for the ACR 430 antenna?
1.6 During evaluation of the ACR 430, aircraft were positioned for approach by controllers
using the radar for the first time. Averaged over 41 approaches to a runway, the worst
mean offset from the centreline approach was less than 25 m at a range of 4 n. mile (see
Appendix III). How good is this performance, and would it be sufficient to position an aircraft
actually on a runway in poor visibility?
1.7 Compare the detection ranges of the two 3 cm radars above for a 10 m 2 target using a
single pulse. Assume both radars have a mean noise level of -'131 dB W, have losses of - 5 dB
and require a SNR of 13 dB for reliable detection. The marine radar has a peak transmitter
powerof25 kW, but the airfield control radar is more powerful with a peak power of 55 kW.
CHAPTER
TWO
DESIGNING A SURVEILLANCE RADAR
• Surveillance
• Choosing parameters
• Radar cross-sections of targets and clutter
• Final design
An attempt to design a surveillance radar reveals how difficult it is to watch

the whole sky.
2.1 RADAR AND SURVEILLANCE
Marine radar, air traffic control, air-borne radar defence systems, ground-
based radar arrays searching for satellites in space-these are all well known
examples of radar surveillance systems that work in all weather conditions,
and at all times of day, testifying to the versatility of the radar technique.
Most modern surveillance radars are 'multifunction', jargon for their
capacity to carry out other activities at the same time as searching for new
targets. One of these functions is to keep track of existing targets until they
are within a certain distance of the radar, when an entirely separate system
takes over: at a civil airport, for example, a Terminal Area radar guides in
the planes; a ship under attack, on the other hand, might engage a close
support weapons radar to lock on to the target. In this chapter we are going
to concentrate on one of the main applications of surveillance radar, which
is to search the sky continuously for new targets.
2S
2.2 ANTENNA BEAMWIDTH CONSIDERATIONS
Surveying the sky presents the radar design engineer with something of a
dilemma; the radar needs to scan the whole sky and get back to the starting
point as quickly as possible in order not to miss any new developments.
However, the radar also needs to spend as much time as possible staring at
each part of the sky in order to obtain good results. Perhaps the best way
to illustrate the difficulties is for us to follow a typical design process and
develop the problem for ourselves.
A common choice of frequency for long-range surveillance is L-band
( '" 1.3 GHz) because this avoids the bad-weather problems that can affect
higher frequencies. S-band ('" 3 GHz), for example, is often used only for
medium-range surveillance up to about 60 nautical miles or 111 km. In
practice, many factors would determine the final choice of frequency,
including the type of target to be detected and the coverage required. We
will assume that a frequency of 1300 MHz ( = 1.3 GHz) has been chosen,
When considering the antenna design, there are two conflicting require-
ments. First, there are several good reasons for choosing narrow beams:
1. The angular position of the target can be measured with good precision.
2. The number of unwanted echoes cluttering up the picture at anyone
time is reduced.
3. The number of interfering signals that can get into the beam at anyone
time is also reduced.
4. The antenna gain factors Gt and Gr in the radar equation are increased
(i.e. the transmitter power is more concentrated in the direction of the
target) and, consequently, the signal-to-noise ratio is improved, making
the target easier to detect.
Secondly, set against these advantages, is a rather serious drawback. If

a given area of sky is to be searched, narrower beams imply that there are
more beam positions within that area to be investigated. If we imagine that
the whole hemisphere of sky is to be surveyed, this equals 2n steradians of
solid angle (4n being the entire sphere). For a beamwidth n = AO A¢, the
number of beam positions required to fill the hemisphere is
.. 2n 2n
N urn ber 0 f bearn posItIons = - = ~~ [ ] (2.1 )
n 11011¢
But remembering Eq. (1.4), we can write this as

Number of beam positions = G /2 [ ] (2.2)
This leads to the general rule that an antenna with a gain of G must probe
DESIGNING A SURVEILLANCE RADAR 27
G directions to survey an entire sphere; in fact, this is another way ofthinking

about the meaning of antenna gain.
Worked example If our L-band radar operates with a 6 m diameter dish

antenna, how many positions must it search to ensure coverage of the
whole sky? Assume an antenna efficiency factor of 0.6.
SOLUTION The physical area of the dish is 28.3 m 2 but, because of the
efficiency factor, this reduces to an effective area of 17 m 2 • The wavelength
at 1300 MHz is 23.1 cm and thus the gain of the antenna is given by
G = 4n:Ae/ A.2 ~ 4000 []
There are therefore 2000 beam positions in the sky to be searched.
The width of the radar beam is given roughly by Eq. (1.7) as
A 0.231 .
110 = 11</> '" - = - - =: 0.039 radians =: 2.2°
d 6
but again effidency considerations mean that in practice the dish would
not be uniformly illuminated by the source antenna at the focus and the
effective beam would be broader, perhaps 3° wide.
COMMENT Two thousand beam positions may seem a lot to search, but
a 3° beamwidth is wide by today's standards, and radars with narrower
beams have even more positions to search.
Each beam position must be inspected at least every to s because data

at this rate are required to follow the path of targets accurately (a process
known as 'tracking', which is carried out by the data processing software).
There is a danger that, if the repeat search time is made any longer than
to s, then not only will the tracking accuracy be poor but the target may
manoeuvre during the interval between scans and will be lost by the tracker.
If our radar had only a single beam, then the search time per beam position
would be (10s)/(2000 positions) giving 5 ms per position. This presents us
with a problem-physically it would be difficult to scan a dish at the speed
needed to achieve 5 ms per beam width and in any case, as we shall see later,
more time than this is needed in each beam position to get good results.
Also, while the repeat scan time of to s is probably adequate to spot the
manoeuvring of civil airliners, there are military aircraft, such as jump-jets,
that can make sudden changes in speed and direction capable of confusing
radar tracking algorithms. These targets require even more frequent inspection.
We will return to this problem of the search time after examining a few more
of the surveillance radar system parameters.
2.3 PULSE REPETITION FREQUENCY AND

UNAMBIGUOUS RANGE AND VELOCITIES
The greater the rate at which pulses are transmitted by a radar system, the
greater the mean power radiated, and it is interesting to ask the question:
'In the design of a radar system, what is the constraint on the maximum
rate at which pulses can be transmitted?' The limit occurs when pulses are
transmitted so frequently that one pulse is transmitted before the previous
pulse has completed the round trip to the target and back. In this situation,
it is unclear which transmitted pulse originated which echo pulse, and the
target range becomes ambiguous.
The rate at which pulses are transmitted is called the pulse repetition
frequency, often abbreviated to PRF. Using Eq. (1.1) and the' 150 m per
microsecond' rule we can say that, if the time interval T between infinitesimally
short transmitter pulses were 1 J1S (a PRF of 1 MHz), then ranges up to a
maximum of 150 m could be surveyed before the receiver had to be turned
off to allow the next transmitter pulse to be sent. More practically, for a
PRF of 1 kHz, range ambiguities are 150 km apart (although the range that
can be surveyed is a little less than this because of the pulse duration and
T - R switching times). The maximum PRF that can be used for an
unambiguous range Rmax is given by
PRF ~ c /2Rmax [Hz] (2.3 )
What would happen if we tried to survey ranges up to 150 km with a
1.5 kHz PRF (which has an unambiguous range of only 100 km)?The answer
is shown in Fig. 2.1, in which a continuous sequence of transmitter pulses
and the associated echoes from a target are shown. The most obvious
assumption is that echo pulses 1 and 2 come from the transmitted pulses A
and B respectively, meaning that the target is within the 'first-time-around'
range interval of 0-100 km. In the case shown in Fig. 2.1, the first-time-
around range is 30 km. However, it is also possible that echo 2 came from
Unambiguous range
of 100 km UOkm
I' l
i i i i i i
Transmitter Echo Transmitter Echo Transmitter Echo
pulse pulse pulse pulse pulse
A I B ;
Figure 2.1 A radar system with an unambiguous 'first-time-around' range of 100 km. The
target range appears to be 30 km, but it could be 130 km.
transmitter pulse A and echo 1 came from the previous transmission. In this
case, the target lies in the' second-time-around ' range interval of 100-200 km
and the target range is 30 + 100 = 130 km. Likewise it is possible to have
third-time-around echoes in the range 200 to 300 km, and so on.
Worked example Our surveillance radar is required to have an unam-

biguous range of 450 km; what is the maximum PRF that may be used?
If the pulse length is 3 jJ.s, what is the 'duty cycle' or on/off ratio for
the transmitter?
SOLUTION The maximum PRF can be found directly from Eq. (2.3) as
333 pulses/second. Another way to look at the problem is to rearrange
Eq. (1.1) as
T= 2R/c [s]
which tells us that when R = 450 X 103 m the round-trip time for a radar
signal is T = 3 ms. For a pulse emission every 3 ms, the PRF must be
l/T = 333 pulses/second, often referred to as a PRF of 333 Hz.
The duty cycle is the ratio of the pulse length r to the inter-pulse
period T. In this case
r 3 x 10- 6
Duty cycle = - = == 10- 3 []
T 3 X 10- 3
COMMENT Often, this PRF is too low to adequately sample the doppler
frequency. Also, the duty cycle may be too low for certain types of
transmitters that need to be worked harder than this if sufficient power
is to be developed to give a good performance over 450 km.
There are several ways in which the PRF can be increased without
reducing the unambiguous range. One method is to 'colour' or label each
pulse in some way so that it can be distinguished from its neighbours.
Methods of labelling pulses include transmitting them with different fre-
quencies, phases, polarizations or pulseshapes. There are difficulties with
these methods, however, when the target radar cross-section (RCS) fluctuates
from pulse to pulse. Labelling pulses also interferes with doppler processing
and moving-target indication (MTI)-see Chapter 5-and so has only
limited usefulness.
Another method of increasing the PRF is to use bursts of pulses on
different PRFs. Each one of these independently may be ambiguous over
the range interval, but when used in combination with the others the
ambiguities can be eliminated. Two staggered PRFs are demonstrated in
Fig. 2.2; when PRF (a) is used the target range is measured as 30, 180, or
330 km, but when PRF (b) is used the range is measured as 105 or 330 km.
The true range of the target must therefore be 330 km.
(a)
Target!
~~V/N'''-''''''''_~......J!(,-___~____~_. ~
(0) () 105 225 330 km
Figure 2.2 The two PRFs (a) and (b) give ambiguous range information when used
independently. In combination, only one range is possible.
Historically, low-PRFradars, designed to avoid range ambiguity problems

and often using MTI, have become known as MTI systems to distinguish
them from high PRF radars. The main difficulty with MTI systems is that
the doppler sampling rate (the PRF) is too low for the speeds of modern
aircraft. This is essentially an aliasing problem and leads to ambiguous
velocity estimates and, worse, to the occurrence of blind speeds at which a
target appears stationary and cannot be resolved against the clutter back-
ground. A common example of aliasing occurs in old cowboy movies when
the wagon wheels appear stationary because the camera shutter (the sampling
device) operates over the same time interval as the spokes take to move
from one position to the next. A system designed to detect moving parts of
the image would not detect the spokes.
It is possible to overcome the doppler ambiguities with a high-PRF
radar; in this case the system is often known as a pulse doppler radar.
High-PRF pulse doppler systems generally suffer from range ambiguities
and 'blind' ranges where transmission of other pulses disables the receiver.
It is not usually possible to find a compromise PRF that avoids both range
and doppler ambiguities. For example, the US AWACS (air-borne warning
and control system) uses one mode in which it is ambiguous in range, and
a separate mode in which the doppler information is ambiguous. With pulse
doppler systems the solution to the ambiguity problem is also to use multiple
PRFs, so that the blind ranges occur at different places. Many air-borne
radars are medium-PRF systems, possessing both range and doppler ambi-
guities, but using several PRFs to remove these ambiguities.
Note that some older radar systems use staggered PRFs in which the
pulse intervals vary from pulse to pulse according to some carefully chosen
rule. In this case, delays are introduced before processing to permit stationary
clutter to be removed by MTI. However, targets beyond the first ambiguity
are not properly integrated and second-time-around clutter is not properly
cancelled.
In practice, our surveillance radar would probably use multiple PRFs
near 333 Hz to resolve doppler ambiguities and blind speeds and also to
eliminate any spurious second-time-around echoes. Because it uses a low
PRF, our system would be classed as an MTI radar.
2.4 PULSE LENGTH AND SAMPLING
In this section we wish to show that short transmitter pulses imply fast
sampling of the receiver output, which can create difficulties for the digital
signal processing. Even with the recent rapid growth in computing power
we must still be careful not to design a radar that produces digits so quickly
that we cannot afford to buy computers large enough to process them on-line.
The pulse length of 3 p.s chosen above is fairly typical for a long-range
air surveillance radar, and from Eq. (1.16) we know this gives a nominal
range resolution of about 450 m. In practice, the resolution would be worse
than this because of pulse shaping and other losses in the systems, and the
final figure might be nearer 750 m. On the other hand, Eq. (1.19) tells us
that the range can be measured more accurately than the nominal 450 m if
we have a good signal-to-noise ratio. These improvements in the estimates
of range come from two processes. The first is known as 'plot extraction'
and involves interpolating between adjacent range samples on each scan.
The second process is tracking, which is the smoothing of the apparent path
of the target through many observations. The overall range accuracy of the
system would be about 50 m after all these processes had been undertaken,
and this would normally be adequate for long-range air surveillance.
The output of our radar receiver should be sampled every 3 p.s because,
as Fig. 2.3a shows, the samples are then separated by a time/distance equal
to the transmitted pulse length (as measured at the 3 dB or half-power
points). Because the echo pulses must be at least the same length as the
transmitted pulses, this sampling rate ensures that no information can be
missed. If the sampling were carried out less frequently than every 3 p.s,
information would be lost and a small target might escape detection by
falling between two samples, as shown in Fig. 2.3b. There are some benefits
in sampling more frequently than every 3 p.s (known as 'oversampling')
because it improves the SNR on a target that straddles two range bins
(Fig. 2.3c ) but it involves a lot more signal processing for only a small reward.
By 'sampling' we usually mean using an analogue-to-digital (AID)
converter to represent the voltages coming out of the receiver by binary
numbers that digital and computer hardware can process. If a sample must
be taken every 3 p.s, this implies an AID converter speed of 333 kHz.
Radar systems need a high 'dynamic range', meaning that they must be
able to process large echoes from nearby objects (often clutter) at the same
time as echoes from distant objects that are very faint; this is a natural
consequence of the 1I R4 term in the radar equation. The bigger this range
of voltages in the receiver, the greater the number of bits needed in the A/D
converter to represent them and it is roughly true to say that one more bit
is needed every time the voltage range is doubled ( = 6 dB increase in dynamic
range). However, the dynamic range requirement can be eased in practice
by using range-dependent gain or swept gain. During the transmission of each
Transmitter Echo
pulse pulse
(a) Power
Target
"-'-"'---- Time or range
t t t t t AID samples
Target detected
(b) Power
I "'--"'~"--- Time or range
t t t t AID samples
Target missed
(c) power, ( \
l/ ~~_TimeOrrange
t t t t t t iii t t t t AID samples
Target detected
SNR improved
I I I I I I I I • Range Iml
0 450 900 1350 1800 2250 2700 3150
I
0
I
3
I
6
I
9
I
12 15
I I
18
I
21
. .
Time ljJ.s]
Figure 2.3 (a) The receiver output sampled correctly; (b) undersampled, which may cause a
target to be missed; and (c) oversampled, improving the SNR for a target straddling two range
bins.
pulse, the receiver is held at zero gain; at short ranges the gain is gradually
increased and at long ranges the full receiver gain is used. Beyond 40-50 km,
clutter decreases owing to the curvature ofthe Earth and shadowing effects.
In our surveillance radar we will assume that an 8-bit word is insufficient
for each A/D sample and that 2 bytes per sample must be used.
Our receiver output is now being sampled at 666000 bytes per second,
and worse is to come; we need two receiver channels in order to measure
the doppler information. A single coherent receiver can reveal the speed of a
target but it cannot tell whether it is moving towards, or away from, the
radar. A second receiving channel is used to resolve this ambiguity by shifting
it 90° in phase from the first channel; these are known as I and Q channels,
which stands for In-phase and Quadrature. This second channel must also
be sampled at 666000 bytes per second and the overall data rate produced
by our radar is thus about 1.3 Mbytes per second. This sampling rate is not
too great for modern computers to store or move around in memory, but
it becomes more challenging if complex floating-point calculations have to
be carried out. Short-range radars working with much higher bandwidths
and shorter pulse lengths than our surveillance system can present the design
engineer with some tough and expensive data processing problems to solve.
2.5 RADAR CROSS-SECTION
To make further progress in our design of a surveillance radar we must

investigate the radar cross-section (ReS) of typical targets likely to be
encountered. We have seen (Sec. 1.4) that the ReS describes the apparent
area of the target as perceived by the radar and that it is a measure of how
much power flux is intercepted by the target and re-radiated to the radar
receiver.
For simple objects, such as a perfectly conducting copper sphere, it is
possible to use electromagnetic theory to calculate a single value for the ReS
(and this turns out to be quite useful because copper spheres are used to
calibrate radar systems, especially on indoor test ranges). However, real
targets such as aeroplanes have many reflecting surfaces, which radiate in
and out of phase with each other and cause large fluctuations in the ReS.
These fluctuations mean that some form of statistical description must be
used for the ReS, such as an average value and the extent of the variations
about this average. First of all we will look at what influences the average
value of the ReS before going on to describe the fluctuations.
The ReS (J of an object is partly dependent on the radar wavelength,
and for simple shapes it is possible to give the following guidelines:
I. For target sizes »)., the ReS is roughly the same size as the real area
of the target. This is known as the optical region because the ReS
approaches the optical value.
2. For target sizes ~ }., the ReS varies wildly with changes in wavelength,
and it may be greater or smaller than the optical value. This is known
as the resonance or Mie region.
3. For target sizes <d, the ReS ex ). -4. This is known as the Rayleigh
region after Lord Rayleigh, who discovered that the scattering of light
by particles in the atmosphere varies as ). -4 (Tyndall, Mie and Oebye
also contributed to these studies). This wavelength-dependent scattering
explains why the sky is blue and the sun appears yellow. When white
light from the sun arrives at the Earth, the relatively long-wavelength
yellow Ired components of the spectrum pass more or less straight
through the atmosphere compared with the shorter-wavelength blue
light, which is scattered over the sky. Much the same is true on a larger
scale size where low-frequency radar signals are undisturbed by water
droplets in rain and clouds, but millimetric radar signals suffer significant
scattering. This reasoning lay behind our choice of the relatively
low-frequency L-band for surveillance at the beginning of the chapter.
The ReS of a few simple shapes are given in Table 2.1, for the case
when the object size is large compared with a wavelength. These simple ReS
values turn out to be quite useful for several reasons. First, it is sometimes
possible to get a feeling for the Res of an object by building it up out
of a few simple shapes; for example, see the chapter by J.D. Olin in
reference 1. Secondly, at long wavelengths, targets often behave as uncompli-
cated structures because the scattering is not from many tiny surfaces but
Table 2.1 The ReS of some simple shapes when the size of the object is
large compared with a wavelength
Target Aspect t RCS
Sphere
G
+
More general large
curved surface ~/r2
\
'(I /
/ \
\
/
\ I \ /
\ I \ I
'./ \,1
3
16n r4 2 (Jd4nrSin ()/).))2
--cos ()
).2 4nr sin ()/).
Circular flat plate
broadside
4
4nr 2 (sin(2nrSin()/).))2
-cos2
()
). 2nr sin () / ).
Square flat plate
broadside
Table 2.1 Continued
Target Aspect t RCS
21trU 2 (sin (21tL sin (J/i.»)2

--cos (J
i. 21tLsin (J/i.
Circular cylinder 21trL2
broadside
I.
Circular cone
Dihedral corner
reflector
r
• ~,
Trihedral corner
reflector
'I \t 121tr4
'2
I.
Short-circuit half-wave 0.86i. 2

dipole
~2
Chaff
N
-
,,12
-0.lSNi.2
t Aspect is considered in the direction of the arrow (where appropriate).

instead involves induced electric currents flowing throughout the target.

There are also times during radar systems testing when it is advantageous
to have an antenna of known RCS as a calibration target, and so a simple
resonant half-wave dipole has been included in Table 2.1.
Although the concept of RCS implicitly includes the contingency that
the target of scattering area A has gain G in the direction of the radar, this
may be explicitly stated as
(J = GA (2.4 )
Combining Eq. (2.4) with G = 41tAI).2 gives

(J = 4nA 2/ A. 2 [ m 2] (2.5 )
This can be a useful way of thinking about RCS when the target is an
antenna, for which the concept of gain is well defined.
When the target is not simple, and cannot be synthesized from simple
shapes, there are various well established computer programs that can be
applied to the problem of calculating the RCS. If all else fails, a scaled copper
model of the target can be built and measured in an anechoic chamber at
an appropriate scaled frequency. A good summary of the state of the art was
presented in 1989 when an entire issue of the Proceedings of the IEEE was
devoted to the radar cross-sections of complex objects2.
In the case of our L-band radar, the radar cross-section of an aircraft
will lie in the optical region because the wavelength of 23 cm is much less
than the size of the target. We cannot assign a single value to the RCS
because it will depend on the aspect angle at which the target is viewed,
both in azimuth and in elevation, and also on the polarization angle of the
radar. These factors, combined with interference from different scattering
surfaces on the target, mean that as we observe the aircraft the RCS will
fluctuate. These fluctuations can be treated by finding the mean value of the
RCS (Jay and a probability density function (PDF) p«J) to describe the
variations about the mean. A well known density function for random
variables is the chi-squared variable (X 2 ) with two, or four, degrees of
freedom. The forms are
p«(j) = _1 ex p ( _~) [ ] (2.6)

(Jay (Jay
4(J
p«(j) = -Zexp (2(j)
-- [ ] (2.7)
(jav (jay
In 1960, P. Swerling 3 used these expressions as the basis for four proposed
mathematical models describing different types of RCS fluctuation. These
four Swerling cases are discussed in Chapter 4 when we examine target
detection theory, and they are still in widespread use today, even though
more sophisticated models have been developed since. With low-frequency
systems, such as over-the-horizon radar, the less complicated scattering
mechanisms mean that quite often the RCS of a target remains constant for
a considerable period of time; such non-fluctuating targets are sometimes
referred to as Swerling case 5 targets.
The time-averaged (RMS) value of the RCS (Jay is sometimes used in
the radar equation because this is the only information available. However,
the fluctuations mean that the predicted performance will only be achieved
for that part of the time when (J ~ (Jay' The Swerling models predict RCS
values lower than average to occur more frequently than above-average
values because the amplitudes and phases of two interfering signals vary
independently (see Chapter 4). Amplitude and phases must both be nearly
equal for above-average values of (J to occur. The PDF for the first case is
shown in Fig. 2.4, and it can be seen that values of (J significantly above (Jay
rarely occur. The use of (Jay in the radar equation thus gives an over-optimistic
performance estimate, and it becomes important to know which model best
describes the RCS fluctuations of the target. Appropriate adjustments can
then be made when calculating the probability with which a target can be
detected by a particular radar system, or what coherent gain may be assumed
when pulses are added together to improve the signal-to-noise ratio.
For our surveillance radar, typical values of RCS that might be expected
are given in Table 2.2. Further information may be found in Maffett 7 •
1.0
Probability density function
0.5
Figure 2.4 The probability density function for a Swerling case I target.
Table 2.2 Typical RCS values for some common

targets
RCSon RCS on
linear log
Target scale scale
Bird 0.001 m 2 -30dB m 2

2
Cruise missile 0.01 m -20dBm 2
Person
Small boat 1 m2 OdBm 2
Small aircraft
Cabin cruiser 10m 2 IOdB m 2
Fighter-bomber aircraft
Road traffic 100m 2 20dBm 2
Large aircraft
Tankers 1000 m 2 30dBm 2
Large passenger ships
2.6 CLUTTER
Clutter is a single word used to describe all the unwanted echoes that clutter
up the radar picture. Clutter is nearly always present and usually widespread,
with echoes arising from hills, buildings, the sea, birds and insects, meteors,
the aurora and many other sources. Of course, what is clutter in one
application may not be so in another. An example of this is HF radar, which
is often used to observe waves on the sea, in which case echoes from ships
(ship clutter) confuse the picture; but when HF radar is used for over-the-
horizon ship tracking, it is the echoes from waves (sea clutter) that cause
the problems.
Clutter may occur as distributed clutter, which increases with the
resolution cell size, or as point clutter, which does not. Point clutter arises
from discrete scatterers such as electricity pylons. The term surface clutter
is used to describe land or sea echoes from the area illuminated by the radar.
The radar cross-section of the clutter is best described by calculating the
average RCS density <To, the RCS per unit area, which is given by the ratio
[ ] (2.8 )
where <Tc = RCS of the area Ac. The symbol <To is sometimes referred to as
sigma zero, and it can be thought of as representing the radar reflectivity of
the terrain.
Volume clutter is similarly used to describe echoes from the atmosphere,

where it is the volume illuminated that determines the RCS of the clutter
observed. An average RCS per unit volume '1 is defined as
(2.9)
where (1c = RCS of the volume v.,. We will investigate '1 and its usefulness
for atmospheric studies in Chapter 9.
The value of (10 varies with many factors: the type of terrain observed,
the direction it is observed from, the weather, the radar wavelength, the
polarization used, etc. Detailed studies have been made and databases are
available to help the radar designer in any particular location 4 • Typical
values for (10 are 1/100 (-20 dB) for land, meaning that the RCS observed
by the radar is about 1/ l00th the area actually illuminated, whereas for sea,
a reflectivity of 1/1000 (-30 dB) is an average figure.
The echoes from surface clutter are large, often much larger than from
the targets of interest. For a given system we can calculate how large they
are with the aid of Fig. 2.5. First we must calculate the physical area
illuminated, which is a strip of ground one range bin AR wide and with a
length given roughly by
Length of arc = R M} [m] (2.10 )
where R = range [m] and AO = beamwidth [radians]. The total area
illuminated is
Area illuminated = R AO AR [m 2 ] (2.11 )
For a radar looking down to the ground at a depression angle a, the area illu-
minated is that given by (2.11) multiplied by a factor sec a (= l/Cos a).
We can calculate typical clutter values for our surveillance radar using
the beamwidth of 3° (remembering to convert to radians) and the range
bin size of 450 m. At a range of 50 km, the area of ground illuminated
RMI1R
Radar
Figure 2.5 The area of ground illuminated by a radar beam at grazing incidence.
(provided the radar was sited high enough to see the ground at this range)
would be roughly
Area = 50 x 103 x 450 x 0.05 == 1.2 x 10 6 [m 2 ]
A reflectivity of 1/100 suggests that the clutter from the ground would be
12000 m 2 • Assuming the radar is searching for an aircraft of RCS 1 m 2 , the
signal-to-clutter ratio would be 1/12000. In decibels, this figure corresponds
to 41 dB, and is one of the factors that determine the dynamic range
requirement for the receiver. In practice, at least another 13 dB of dynamic
range is needed, so that an adequate SNR is available for target detection,
plus some extra allowance for clutter variations around the mean value.
Worked example An HF radar transmits 100 J1S pulses at 10 MHz with

a 100 m antenna array looking out over the sea. What are the dynamic
range requirements for the receiver if a - 25 dB m 2 cruise missile must
be detected at a range of 30 km and 60 km?
SOLUTION The wavelength is 30 m and }./d for the array gives a

beamwidth of roughly 0.3 radians. The pulse length implies a range gate
size of 15 km. At 30 km range, the physical area of sea illuminated is
therefore roughly
Area illuminated = 30 x 103 x 15 X 103 x 0.3 [m 2 ]
= 135000000 [m 2 ]
Taking the reflectivity of the sea as 1j1 000 implies that the clutter has
an RCS of 135000 m 2 or 51 dB m 2 • The radar receiver must be able to
detect noise signals 13 dB lower than the target (for 13 dB SNR
detection), and so it must be capable of detecting - 38 dB m 2 signals.
The dynamic range required by the receiver, as shown in Fig. 2.6, must
therefore be at least 89 dB or 28000: 1 as a voltage ratio. At 60 km range,
the clutter will be 3 dB (two times) greater, the target size remains the
same and so the dynamic range requirement rises to 92 dB.
COMMENT With increasing range, the signal strength from a target decays
as R4 because of the radar equation but the clutter power falls off as R3
because the area illuminated has a linear dependence on R. Signal-to-
clutter problems therefore get worse at longer ranges.
If target echoes are so much smaller than clutter, how can they be
detected at all? The answer is that most targets of interest are moving and
can be distinguished from stationary clutter through the use of the doppler
effect. Doppler processing to give sub-clutter visibility (SCV) is an important
part of radar processing and is discussed in Chapter 5. However, the
Power [dB]
proportional to RCS [dB m C]
+50
Sea echo-- 76dB

89 dB dynamic
o
13 d-B-f----f"---;.rn'g' f'";rem".
-20 -10 0 10 / 20
Cruise missile Doppler shift [Hz)1
Figure 2.6 The dynamic range requirement of an HF radar receiver.
implication for our surveillance radar design is that we must dwell for long
enough in each beam position to gain sufficient doppler resolution to separate
target and clutter echoes.
Most surveillance radars also maintain clutter maps, which can be static
(loaded prior to operation) or dynamically modified during operation.
Separate maps are maintained for ground clutter, rain, chaff, etc. The radar
antenna beam and the signal processing can then be adaptively modified to
suit the environment in a process known as clutter fixing.
2.7 NOISE
Before we can proceed with our final design, we need to know what noise
level to expect. At L-band the noise is likely to be dominated by internal
noise produced by the random motion of thermally excited electrons. Below
about 6000 GHz (i.e. for all practical radars), thermal noise can be considered
to be white noise having a flat power spectral density S(f) given by
S(f) = kTo (2.12 )
where k = Boltzmann's constant = 1.38 x 10- 23 J K -1 and To = system
temperature [K]. To is generally assumed to be 290 K, giving a noise density
of - 204 dB W Hz - 1, a figure engraved on the hearts of most radar engineers.
This noise spectrum extends far beyond the bandwidths of the radar system
and we are only concerned with band-limited white noise, which has been
restricted by the passband of the receiver. The mean noise power in the
receiver N is given by
[W] (2.13 )
where B = bandwidth [Hz]. Note that this figure is independent of the radar
operating frequency; an S-band and an L-band radar, both operating with
a bandwidth of 1 MHz, would have identical mean noise levels of -144 dB W
(= -204 dB W Hz- 1 + 60 dB Hz bandwidth).
In practical receivers, the noise level is found to be worse than in
Eq. (2.13 ) by a factor known as the noise figure F. The noise figure is measured
over the linear part of the receiver operating range as
F = noise power out of actual receiver
(2.14 )
noise power out of ideal receiver
In the case of our L-band surveillance radar, the bandwidth can be found
from Ij(pulse length) = 333.3 kHz = 55.2 dB Hz. Assuming a noise figure
for the receiver of 3 dB, the mean noise level is
N = kToFB [dBW]
-204 + 3 + 55.2 [dBW] (2.15 )
= -145.8 dB W
Table 2.3 Losses in radar systems

Loss Typical value
(a) Losses encountered in many surveillance radar systems
'Plumbing' loss (in waveguides, T -R switch, rotating joint,

phaseshifters, feed) 3.5 dB
Beamshape loss (target not in centre of the beam for duration of the
scan time) 2.0 dB
Pulse compression and filter weighting loss 1.0 dB
Sampling loss (target not in centre of the range rate) 0.5 dB
Fast Fourier transform weighting loss 2.0dB
(b) Components causing losses in some radar systems
Tx filter Beamformer
Isolator Rx protection circuit
Connectors Rx front-end filter
Radome (protective antenna coating)
(c) Other losses
Mismatches to antenna impedance

Eclipsing loss at short ranges (system not switched back to full receiver sensitivity after
transmitting)
Integration losses (i.e. coherent integration not perfect)
Constant false-alarm rate processing losses (discussed in Chapter 5)
The instantaneous noise power in the receiver will fluctuate about this mean
value. In Chapter 4 we discuss the statistical properties of noise in more
detail, and the way in which this affects the probability of target detection.
2.8 FINAL DESIGN
Having decided most system parameters, we can now work out the
transmitter power needed to undertake the task of surveillance. We will
assume that the targets will be no smaller than 1 m 2. Rewriting the radar
equation (Eq. (1.14» gives
(SNR)N (4n)3 R~ax
Pt = - - - - - - : : - - - - [W] (2.16 )
Gt Gr}·2 a L s
Next we insert the values chosen during this chapter, i.e.
SNR 20 [ ] 13.0 [dB]

Noise N 2.6 x 10- 15 [W] - 145.8 [dB W]
(4n )3 2000 [ ] 33.0 [dB]
R~ax 4.1 X 10+ 22 [m4] 226.1 [dB m 4] (maximum)
I/G t Gr 6.3 x 10- 8 [ ] -72.0 [dB]
1/,1,2 18.7 [m- 2] 12.7 [dBm- 2]
lla 1 [m-2] 0 [dB m- 2]
IlLs 3.2 [ ] +5 [dB] (typical minimum)
Pt 16 [MW] 72 [dB W]
A transmitter providing 16 MW of peak power would be very expensive.

However, we have so far only applied the radar equation to a single pulse,
which would give no sub-clutter visibility. We need to look at the phase
difference of the echo from pulse to pulse to see if the target is moving.
The radar equation can be modified to include the integration gain of
several pulses added together by including a factor nLn on the top of the
equation; n is the number of pulses and Ln is the loss over perfect integration
(i.e. Ln is less than unity). In modern radars, the integration is usually
coherent (taking into account phase, as well as amplitude) and Ln can be
quite close to unity (near zero loss). Some older systems used incoherent
integration by adding up the signals after the detector, where all phase
information has been lost. Losses of several dBs were incurred. In fact, there
are many potential sources of loss in a radar system, and a summary is
shown in Table 2.3 opposite, where suggested typical values for the more
regularly incurred losses are also given. Further details of how to calculate
losses are to be found in Skolnik 5 and Rohan 6 •
One final modification to the radar equation that might be mentioned

here is to replace the peak transmitter power ~ by the average power ~
using the duty cycle tiT:
[W] (2.17 )
i.e.
p= ~ [W] (2.18 )
t (PRF)t
The radar equation can now be rewritten as

~GIGr(1), 2nLnLs
SNR = - - - - - - - - [ ] (2.19 )
(41t)3kTaF(Br) (PRF)R4
The terms Band t are collected together because their product is approxi-
mately unity. This 'pulse radar equation' is easy to convert to an energy
equation by replacing the SNR ratio with the energy Inoise ratio (E IN) and
the transmitted power ~/(PRF) by the transmitted energy Et • Thinking of
the radar equation in terms of energy can be a useful concept when there is
a fixed time interval for data collection, as is the case with our surveillance
radar.
In our final design, we would probably integrate 32 pulses in each beam
position for the following reasons:
1. At 333 pulses per second, 32 pulses gives an integration time of 0.096 s,

which, applying Eq. (1.21), corresponds to a useful basic doppler
resolution of about 1 m s - 1.
2. The 32 pulses give a convenient number on which to carry out fast
Fourier transform processing to extract doppler information and achieve
sub-clutter visibility.
3. If the antenna were of the revolving dish design with a fixed beam, then
with a beamwidth of 3° and a rotation time of 10 s, a target would
remain in the main beam for a duration similar to the integration time.
4. Using 32 pulses for target detection means that the peak transmitter
power can be reduced to 500 kW with a consequent considerable saving
in cost.
All these reasons dictate the need for an observation time of nearly
100 ms in each beam position for successful target detection. At the beginning
of the chapter we worked out that only 5 ms per beam position is available
if the whole sky is to be scanned. This reveals one of the truisms of
surveillance-there is never enough time to do everything properly and
compromises have to be made. There are several ways out of the dilemma;
perhaps the most obvious method is to use several beams simultaneously,
for this reduces the rate at which a dish must be scanned. An option available
to the most modern radars, where the beam can be steered electronical1y
(Chapter 15), is to make intelligent decisions about how the search pattern
should be adapted to the situation, according to a set of priorities. In some
cases, the simplest solution is simply to abandon searching certain parts of
the sky, usual\y those at high elevation angles.
In real-life radar design there are many more sophistications and
problems to be dealt with. There would be integration losses during the
32 pulses because ofRCS fluctuations, a variety of modulation schemes would
be used, and so on. But these are all complications that will be easily mastered
if you understand the basic design process and the physics of what is
happening in radar survei11ance.
2.9 SUMMARY
Radar surveillance can be improved through the use of narrow beams, but
this may lead to there being more beam positions to be searched than is
possible in the time available. Multibeam systems can help with this problem,
but they put more pressure on the data processing activities, which are often
already stretched, even with today's technology.
The RCS of real targets fluctuates and its statistical nature must be taken
into consideration if the radar detection performance is not to be over-
estimated. The problem of clutter and the need for sub-clutter visibility is
often severe and leads to a need for doppler processing. Careful design
of the transmitted waveform is needed to avoid range and/or doppler
ambiguities.
Key equations
• Number of beam positions to fill the sky (hemisphere):
Number of beam positions = G /2 []
• Maximum PRF for an unambiguous range Rmax:
PRF ~ c/2R max
• Radar cross-section of a target:
a = 4nA 2 / i. 2 [ m2 ]
• Radar cross-section per unit area:
aO=ac!Ac []
• Radar cross-section per unit volume:
'1=ac/~
• Total area of ground illuminated at grazing incidence:

Area illuminated = R ,1,.0 AR
• Mean noise level:
N = kTaFB [W]
• Average transmitter power:
~ = Pt't/T= Pt't(PRF) [W]
• Radar equation:
P = (SNR)N(4n)3R~ax [W]
t 2
Gt GrA 0Ls
• Radar equation in alternative form:
~GtGr(JA 2nLnLs
SNR = -~~-=----"-----''---- [ ]
(41t)3kTaF(B-r)(PRF)R4
2.10 REFERENCES
1. Modern Radar Techniques, M. J. B. Scanlan, Collins, Glasgow, 1987. [Chapter 3 on target

characteristics is a useful summary.]
2. IEEE special issue on radar cross sections of complex objects, Proc. IEEE, 77, 5, 1989.
3. Probability of detection for fluctuating targets, P. Swerling, IRE Trans., IT-6,269- 308, 1960.
4. Radar Design Principles, F. E. Nathanson, McGraw-Hili, New York, 1969. [Contains good
chapters on RCS and c1utter.]
5. Radar Handbook, Ed. M. E. Skolnik, McGraw-Hili, New York, 1990. [Chapter II is
concerned with the RCS targets.]
6. Surveillance Radar Performance Prediction, P. Rohan, Peter Peregrinus for the lEE,
Stevenage, Herts, 1983.
7. Topics for a Statistical Description of Radar Cross Section, A. L. Maffett, Wiley, New York,
1989. [A detailed study of radar cross section and statistical aspects.]
8. Technical History of the Beginnings of RADAR, S. S. Swords, Peter Peregrinus for the lEE,
2.11 PROBLEMS
2.1 The Italian company Alenia manufactures the ATCR-44K, an L-band medium-range
primary radar designed for use in modern automated air traffic control (ATC) systems. The
antenna is a rotating reflector and if a rotation rate of 6 revolutions per minute (RPM) is
selected then typically a pulse repetition frequency (PRF) of 480 Hz and a pulse duration of
1.5 JlS is used. Alternatively, at 12 RPM the PRF is typically 691 Hz and the pulse duration is
1 JlS. Given that the peak transmitter power is 1.2 MW, what is the duty cycle and the mean
power for the two modes of operation?
2.2 Derive the expression, given by Alenia, that the number of hits n on a target each time the
antenna scans past is given by
M(PRF)
n=
(RPM) x 6
where !l(} = azimuth beamwidth [degrees]. If !l(} = 1.2°, how many hits per scan are there
when the antenna rotation rate is 6 RPM?
2.3 The documentation for the Alenia ATCR-44K radar gives the maximum unambiguous
range as
1000000
R =----- [nautical miles]
max 12.4 X (PRF)
What do you think the number 12.4 represents?
2.4 What is the detection range of the ATCR-44K radar for a 2 m 2 target? Assume 5 dB system
losses, a further 2 dB for atmospheric absorption, and that a 10 dB SNR is required. Take the
elevation beamwidth as 4.7".
2.5 Does the ATCR-44K radar require staggered PRFs?
2.6 What AjD converter requirement might the ATCR-44K radar be expected to have?
2.7 During World War II the German Navy used a 125 MHz medium-range early-warning
radar known as Freya, The rotatable antenna consisted of 12 dipoles mounted in front of a
rectangular wire mesh reflector (see Swords 8 ). If the peak transmitter power was 20 kW and
the pulse duration 2 ps, at what range could the radar detect a 2 m 2 target? Assume an antenna
gain of 16 dB and losses of 5 dB. How does this compare with the modern Alenia radar?
CHAPTER
THREE
TRACKING RADAR
• Measuring angle
• Monopulse radar
• Tracking accuracy
• Radar guidance systems
Tracking radars are dedicated to a target and are designed to measure angular
information accurately.
3.1 INTRODUCTION
A tracking radar continuously measures the coordinates of a moving target

in order to determine its path and to predict where it is going. Tracking can
be carried out using range, angle or doppler information, but it is the tracking
in angle that forms the characteristic feature of tracking radars.
To some extent, all types of surveillance radar, civil or military, could
be considering as tracking systems since they form an estimate of the target
position each time the scanner returns to look in that particular direction-a
process known as track-while-scan. Surveillance radars can keep track of
many targets simultaneously but the positional accuracy they provide,
especially in angle, is not adequate for some purposes.
In contrast, a tracking radar (often a military system) is dedicated to a
target and observes it continuously and with great precision, usually with
a view to engaging a weapons system. Some tracking radars have their own
search facilities, but the more usual mode of operation is for a main
48
TRACKING RADAR 49
surveillance radar to warn of any targets posing a particular threat and to

download the target coordinates to the tracking system, which then searches
the immediate area to acquire the target before initiating the tracking process.
The antenna of a tracking radar usually faces the target all the time in
order to keep it in the centre of the beam and so maximize the signal-to-noise
ratio. If the target moves away from the centre of the beam, this produces
an error voltag~, which is amplified and fed to servo-motors to drive the
antenna back onto the target. The azimuth and elevation of the target are
then read from angle transducers, such as synchros, mounted on the antenna
axes. The methods of producing error signals from the antenna patterns form
one of the main aspects of tracking radar design.
3.2 SEQUENTIAL LOBING
Switching between several beam positions or sequential lobing was one of

the earliest methods used to derive angular information about a target. An
antenna would have two beams displaced slightly left and right of its
centreline and the receiver switched rapidly between them. If the target
appeared to grow larger in the left-hand beam, this would drive the antenna
to the left until the signal was equal in both beams again. Similarly, two
beams displaced vertically could be used to give elevation information.
A common method of producing several beams from one antenna was
to use a parabolic dish with a block of four microwave horns feeding it at
the focus; a fifth horn could be used for transmitting at the centre of the
block. At least four pulses had to be transmitted, one for each quadrant,
and this is the weakness of the method because rapidly fading targets
(Swerling cases 2 or 4) can vary in amplitude from pulse to pulse, thus
making comparisons between beams less valid.
3.3 CONICAL SCANNING
An alternative to stepping the antenna beam around the direction of the

target is to rotate it continuously. You can simulate the effect by pointing
your arm straight out at a distant object and then begin moving your arm
in small circles around the object (the angle between the object and the circle
you are describing is called the squint angle). If you then deliberately 'miss'
the object with the centre of your scan, you will see that once per revolution
your arm comes nearer to the object; if this were a radio system, you would
get the biggest signal in this position and the information could be used to
correct the pointing of the antenna.
One of the first successful applications of conical scanning was Telefunken's
600 MHz 'Wurzburg' radar used by the German Luftwaffe in World War II
for gun laying and ground-controlled interception of allied aircraft; see

details in Swords!. The technique is still used today as a reliable and low-cost
method of tracking in non-critical situations; for an example see problem 3.1.
Conical scan radars usually process 10 or more pulses per revolution,
but even so they suffer from the same handicap as sequentiallobing in that
the RCS of the target can change between pulses. Also, the repetitive nature
of the scanning makes the system vulnerable to electronic countermeasures.
3.4 MONOPULSE RADAR
The problems of pulse-to-pulse varIatIOns in echo amplitudes can be

overcome by using more than one beam simultaneously to measure the
angular position of the target on a single pulse. This technique is known as
monopulse tracking, a name suggested by Bell Telephone Laboratories in
1946. However, it was originally known as simultaneous lobing, which may
be a better description, and it was experimented with as early as 1928 (see
Rhodes 2 ). Monopulse tracking can make use of amplitude information from
the antennas, phase information or even both together to give much better
precision than the earlier sequential lobing and conical scan systems. Another
advantage of monopulse radar is that, in principle, a target can be located
from a single pulse measurement, which may be useful when the radar is
being jammed and the target is only viewed in glimpses.
The simplest mono pulse systems, and in many ways the most reliable,
are the amplitude-comparison monopulse radars. Two antenna beams are set
at an angle to each other and the outputs are connected to a hybrid, which
forms sum 1: and difference L\ signals from them. These signals are fed to a
pair of matched receivers, mixed to a lower frequency and amplified (see
Fig. 3.1).
The sum channel forms a beam that is a combination of the signal power
of two individual beams and so has an improved signal-to-noise ratio. This
combined beam is used for target detection and to measure the range and
doppler information. The gain of the sum channel beam, which directly faces
the target, gives monopulse a SNR advantage over the earlier techniques in
which the target was viewed off the bore-sight. However, the sum beam is
wider than the individual beams and so it is not used to measure angle.
The difference channel produces an error voltage that is roughly
proportional to the angular deviation of the target from the bore-sight (see
Fig. 3.2), and no output is obtained when the target echo amplitude is the
same in both antenna beams. Fading of the echo occurs equally in both
beams and so it does not affect this comparison, except for the usual
inaccuracies at low signal-to-noise ratio.
The phase-sensitive detector shown in Fig. 3.1 is used to determine the
sign of the error voltage so that the servo-motors know which way to drive
TRACKING RADAR 51
1---,...--+ Target detection. range and

L -_ _--' doppler estimation
S;gn ,,( ""I< """1

1-_--'-_ _+ Angle • Angl~
error estimation I
Difference channel
Figure 3.1 Block diagram of an amplitude-comparison monopulse radar system.
(a) Polar (h) Rectangular
G
I
-----
--
"
'-
'-
..... ,
\
I
Two heams
...
"0
.E
Q..
E
'-
' ..... _--/
I
<
Sum ~
Difference .l
Figure 3.2 Monopulse antenna beam patterns expressed (a) in polar form and (b) in
rectangular coordinates.
the antenna to get back on target. It should be stressed that the presence
of the phase-sensitive detector does not mean that the system exploits the
phase information contained within the radio echo as a means of deriving
angular information.
Amplitude monopulse can be improved by taking the ratio M which
normalizes the difference channel by the sum channel. This gives a quotient
that is independent of the signal strength and linear against the angle error
over a wide range of angles. The function of dividing signals can be under-
taken digitally, but in the past it has been performed by processing the signals
with logarithmic amplifiers and then taking differences.
Perhaps the best way to get to grips with amplitude monopulse is to
draw out the antenna patterns in Fig. 3.1 for yourself; try using a pocket
calculator to generate a (sin e/ e) 2 pattern, plot two versions of it with one
slightly displaced from the other, and then add them together to form the
sum channel and subtract them to form the difference channel. Chapter 7
of Hirsch and Grove 3 describes in a particularly helpful way how to use a
computer to carry out a more sophisticated simulation of monopulse antenna
patterns and differences.
Phase-comparison monopulse is also possible and was tested early on in
radar history; in fact, the technique was patented in 1943 (see Rhodes 2 ).
Two antennas are fixed adjacent and parallel to each other and, by comparing
the phase difference of the two outputs, it is possible to derive angular
information. An echo arriving along the bore-sight of the antennas will arrive
at both of them at the same time (Fig. 3.3a), but a signal arriving at an
angle e to the bore-sight will arrive at one antenna later than at the other
because it has had to travel an extra distance x given by
x = d sin e [m] (3.1 )
where d = separation of the antennas Em]. The distance x can be expressed
as a fraction of the radar wavelength A to give the difference in phase 111/1
between the two signals as
111/1 = 2nd(sin e)/A [radians] (3.2 )
The factor 2n in Eq. (3.2) arises because the phase difference increases by
2n radians for every complete wavelength A. travelled by the signal. Note
that for small angles sin e "" e, leading to the approximation
111/1 "" 2nde / A [radians] (3.3 )
which is a roughly linear relationship between the angular deviation from
the bore-sight direction and the phase-difference error signal.
Phase-comparison monopulse is identical to the Young's slit experiment
of physical optics and the interferometry techniques used by radio astronomers.
However, the usefulness of phase-comparison methods in tracking radar is
somewhat limited because of problems created by multiple signals from
TRACKING RADAR 53
(a)
t tE
'.~oU ~----1.---- Bore-sight
(0)
~011 = 21Td sin e

A
Figure 3.3 Phase-comparison monopulse. (a) When the echo arrives along the bore-sight,
there is no phase difference between the signals. (b) When the echo arrives at an angle 0, this
gives rise to a phase difference 1'11/1.
a single target (a problem known as muitipath, described further in

Chapter 7). Another problem is concerned with the ambiguities in angular
position, called grating lobes, that occur when only two widely spaced
antennas are used (see Chapter 15).
The best strategy to use when designing tracking radars is to make use
of all the information available (amplitude, phase, range and doppler) and
to use many antenna elements to reduce grating lobes; this leads on to the
design of phased array antennas, which are described in more depth in
Chapter 15. Modern phased array antennas can overcome one of the major
drawbacks of older tracking radars by making use of electronic beam steering
to track more than one target simultaneously. Mechanically driven dishes
cannot move fast enough to watch two widely separated areas of sky at the
same time, and a weapons system on a ship, for example, could be defeated
by a simultaneous attack by two aircraft from different quarters.
3.5 TRACKING ACCURACY
At long range, a target can usually be considered as a point source and the
angular accuracy of a tracking radar is determined by both electromechanical
factors associated with the control of the antenna turning gear and by the
SNR of the radar measurements. Whichever tracking method is used, from
Eq. (1.18) we would expect the angular accuracy be to be related to the
beamwidth and to be fundamentally limited by
be,.... AeIJ(2 x SNR) [radians] (3.4)
However, the use of sum and difference channels allows this to be improved
upon such that be I k may be achievable, where k is the slope of the A/~
curve near e = 0, see Levanon 4 •
The use of sequential lobing, conical scanning or monopulse is merely
the means by which the precision of Eq. (3.4) is achieved, but note that the
first two techniques will be further degraded by amplitude fluctuations
whereas monopulse is not. In practice, RCS fluctuations, multipath and
changes in the atmospheric propagation all add together to increase the
tracking noise. These errors have been elegantly discussed and summarized
in Chapter 7 of Berkowitz s, for those who wish to delve deeper.
Worked example A large ground-based tracking radar has a beamwidth

of 1 and detects a target with a SNR of 17 dB on a single pulse. What
0
angular accuracy might be expected if 100 pulses are coherently summed

without loss? If the radar bandwidth is 1 MHz and the target were at
a range of 10 km, would the transverse error in the target location be
greater or smaller than the radial error?
SOLUTION Twice the final signal-to-noise ratio would be 10 000: 1 and

so the best accuracy obtainable would be 0.01 (this is close to the best
0
achievable in practice).
At a range of 10 km the transverse error R be corresponds to a
distance of 1.7 m. A 1 MHz bandwidth implies a basic range resolution
of 150 m, which would improve to 1.5 m for this SNR, so the two
accuracies are similar. It would be easier to improve the range accuracy
(using more bandwidth) than the angular accuracy, provided the
scattering size of the target was small enough to justify this.
At short ranges, the finite size of the target begins to limit the accuracy
of the system. Tracking radars need narrow 'pencil' beams and so tend to
operate at relatively short wavelengths, which puts the target RCS into the
optical region. At these short wavelengths the target behaves as many
independent scatterers, each of which contributes in a complex way (i.e. in
both amplitude and phase) to the overall RCS and causes an effect similar
TRACKING RADAR 55
to optical glinting. We can define a 'centre of gravity' of the target and

attempt to track this, but at any given instant the strongest scattering facet
may lie elsewhere on the object and the radar may start to wander off the
central point. Skolnik 6 shows how certain phase relationships between the
scattering elements can even cause the apparent centre of gravity to lie outside
the object. The wandering of the apparent target direction, and the attempts
to control this effect by increasing the time constant of the antenna drive
feedback loop, lead to a form of angle noise, which dominates receiver noise
tracking errors at short ranges.
3.6 FREQUENCY AGILITY

Frequency agility (similar to frequency diversity in communications) is the
process of changing the radar frequency from pulse to pulse. There are
considerable advantages to this technique, despite the additional complexity
of the radar system (for more details, see Nathanson 7 , for example). These
benefits include:
1. A reduction in angle noise, because the complex sum of the contributions

from the individual target scattering surfaces changes from one frequency
to another, and a more accurate mean estimate of the target centre of
gravity can be made.
2. A reduction in multi path effects, because the change in wavelength
changes the position at which destructive interference occurs.
3. Greater resistance to electronic countermeasures than radars operating
on a single frequency.
The frequencies chosen must be sufficiently far apart to give uncorrelated

measurements, which means that they must be at least a radar bandwidth
apart. The frequency channels must also be a 'target decorrelation bandwidth'
apart, defined in a similar way to Eq. (1.16). If the characteristic dimension
ofthe target in range is I, then the separation offrequencies!!.f must satisfy
N ~ cj(2/) [Hz] (3.5 )
One of the most critical applications of tracking radar is in the defence
of ships against incoming missiles, such as Exocets, which have small
cross-sections and fly just above the surface of the sea. Compensation for
the motion of the ship increases angle noise, the sea creates a strong clutter
background and, because salt water is a good conductor, there is a strong
multipath image of the missile. In addition to these troubles, the angle noise
can be made worse if the missile is programmed to weave (usually to evade
gatling gunfire) and so present an ever-changing aspect to the radar. Under
these circumstances, frequency agility can be invaluable, as can the contri-
bution from other sensors such as TV systems, lidars and infrared trackers.
3.7 THE TRACKING PROCESS
The first task of a tracking radar is to acquire the target allocated for
engagement. Tracking radars usually have a search mode to survey a
restricted area of sky around the expected target position. Different search
patterns are used, but it is common to find options of a horizontal raster
scan, similar to a television, a vertical variant of this called a nodding scan,
and a low-elevation scan round the entire horizon, as shown in Fig. 3.4.
After the target has been detected, the next step is to estimate its position
in a process known as plot extraction. We have already seen how the angular
information is extracted, but on each pulse the target range and velocity
must also be measured.
A rough estimate of the range can be found by locating the range cell
in which the echo is largest, but much more precision can be obtained by
Nodding scan
~])'. '.
Figure 3.4 Various methods of scanning for a tracking radar to acquire a target.
TRACKING RADAR 57
interpolating between range cells. If two adjacent cells show equal echo
power, then clearly the target lies exactly half-way between them; but if one
is larger than the other, then some interpolation formula is required. Usually
the interpolation is carried out by fitting the pulse shape or a quadratic curve
to the data samples and then differentiating to find the peak. The error in us-
ing a simple quadratic rather than the actual pulse shape after matched filter-
ing is quite small. As an example, a quadratic curve can be fitted to the AID
samples shown in Fig. 3.5 using the equation Y =ax2 + bx + c. For the central
sample, x =0, so that c = yo; at x =+1 range cell, Y+ = a + b + yo and al- x =
-1 range cell Y- = a - b + yo. Solving these equations gives
2a = v+ + Y- - 2Yo and b =(Y+ - Y-)/2.
The maximum of the quadratic occurs at

-b
x =-
max 2a
The range of the target is then that of the central range cell + Xmax '
Amplitude [volts
or AID samples
representing volts]
L-------~~------~-r----L------L-- _________ _ + x
-1 +1
I
Xmax
(target position)
Figure 3.5 Interpolation using a cosine-squared pulse to improve the range accuracy.
Again, the accuracy of the interpolation is limited by the SNR as in Eq.

(1.18). Other techniques include hardware range trackers based on the
early/late gate synchronizer familiar to most digital communications engi-
neers.
Velocity information can be acquired in a similar manner to range:
interpolation between adjacent doppler cells in the fast Fourier transform
(or other type of filter) is used to refine the estimate of the target speed.
There are, however, complications with velocity measurements; the doppler
shift of the target can be spread by weaving and other manoeuvres, and also
contaminated by rotating devices such as turbines or helicopter blades. It is
sometimes necessary to define a 'centre of gravity' of the doppler measure-
ment and use this in the tracking algorithm. When frequency agility is used,
and some of the channels are lost through either jamming or RCS fading,
the process of extracting velocity information can become quite interesting!
As the antenna continues to face the target and the radar evaluates its
position plot by plot, mathematical algorithms built into the radar software
work out the most probable true path of the target through the noisy
measurements. One of the most important aspects of the tracking software
is to predict where the target is going, the objective being to improve the
radar tracking process and to update the aim of any engaging defensive
weapons system. Tracking algorithms are common to many types of radar
and are covered in Chapter 5.
3.8 RADAR GUIDANCE
A common application of tracking radar is in the defence of local or 'point'

targets from incoming air attack. One of the simplest methods (still used)
is for the tracking radar to lock on to the target, fire an intercepting missile
and then track both objects with a view to narrowing the gap between them
by sending control signals up to the intercepting missile on a communication
link. This technique is known as radar command to line of sight (RCLOS)
and is used, for example, by the Rapier missile, which distinguished itself
during the Falklands crisis. The drawback of this system is that the missile
must stay within line of sight from the radar to the target if it is to be tracked
accurately, and this may not be the optimum trajectory. For crossing targets,
for example, high lateral acceleorations are needed by the intercepting missile
just before impact-it would be preferable for the missile to travel outside
the beam.
Once a tracking radar has locked onto a target, there is an invisible
beam between the two, and in the late 1940s it was realized that it should
be possible to fly a missile up this beam; these weapons have since become
known as beamriding missiles. The missile is equipped with a rearward-facing
antenna to provide bearing information, which the autopilot uses to keep
TRACKING RADAR 59
the missile in the centre of the radar beam. The attraction of this approach
is that it is simple, and a strong SNR is received by the missile because the
signal travels only on a one-way path. The disadvantages are that the target
must be kept accurately in the centre of the beam and, as with RCLOS, the
path up the beam may not be the ideal trajectory for a missile to follow 8 .
The disadvantages of beam riding can be overcome to some extent
through the use of semi-active homing in which the tracking radar acts as a
target illuminator. On board the missile a receiver detects the scattered energy
from the target, tracks its position and works out the best trajectory for an
intercept. The.missile does not have to fly within the illuminating beam and
the tracking radar is not required to keep the bore-sight exactly on the target,
merely to keep the target illuminated.
Although semi-active homing requires more complicated electronics to
be carried by missiles, it has proved to be an effective system. During the
Gulf War, for example, the Lynx helicopters of the UK Royal Navy
illuminated Iraqi ships with their radars and attacked successfully with their
semi-active homing Sea Skua missiles. The Patriot missiles, used with such
telling effect in the interception of Scud missiles during the same crisis, also
use semi-active homing, with the passive radar receivers on board the missiles
relaying information back to the main radar processor on the ground, which
then feeds commands back up a link to the missile.
The most advanced weapons employ active homing radar guidance
systems in which the entire tracking radar is carried on board the intercepting
missile. Initially the missile is locked on to a target by its host, but once
fired it has a high degree of autonomy to pursue its target. Developments
in miniaturized analogue and digital electronics, target image processing and
the evolution of 35 and 94 GHz radar seeker heads mean that some very
sophisticated radar missile guidance technology should be available soon.
The first of this new generation of missiles is likely to be the ERINTs
(extended-range interceptors), which use active homing and are so small
that 16 of them will fit into a launcher that at present holds four Patriots.
Worked example The new generation of anti-armour submunitions are
expected to be tiny missiles (0.1 m by 0.6 m long), which are fired in
clusters but which contain their own 94 GHz guidance radars for
independent targeting 8 • What diameter would an older X-band guided
missile have needed in order to obtain the same angular accuracy?
SOLUTION A 10 cm dish at 94 GHz has a nominal angular resolution
of A.ld = 1.8°, although, in practice, this would probably be a little
over 2°.
If, for convenience, we assume an X-band frequency of 9.4 GHz, we
can see immediately that the older-type missile would have needed a
diameter of the order of 1 m to have a resolution comparable with the
proposed new submunitions.
3.9 SUMMARY
Tracking radars are distinguished by their dedication to a target and the

precision of their angle measurements. Angular estimation is achieved by
comparing the echo in two adjacent beams; the problem of pulse-to-pulse
variation in the echo amplitude (or phase) is overcome in mono pulse radar
by making the comparison simultaneously on each pulse. One of the main
applications of tracking radar is the guidance of weapons systems.
Key equation
• The decorrelation bandwidth formula:
Af~ c/(21) [Hz]
3.10 REFERENCES
1. Technical History of the Beginnings of RADAR, S. S. Swords, Peter Peregrinus for the lEE,
2. Introduction to Monopulse, D. R. Rhodes, McGraw-Hili, New York, 1959.
3. Practical Simulation of Radar Antennas and Radomes, H. L. Hirsch and D. C. Grove, Artech
House, Norwood, MA, 1988. [A useful book, including software listings.]
4. Radar Principles, N. Levanon, Wiley, New York, 1988.
5. Modern Radar, Ed. R. S. Berkowitz, Wiley, New York, 1965.
6. Introduction to Radar Systems, M. I. Skolnik, McGraw-Hili, New York, 1985.
7. Radar Design Principles, F. E. Nathanson, McGraw-Hill, New York, 1969.
8. The evolution of radar guidance, D. A. Ramsay, GEe J. Res., 3(2), 92-103, 1985. [The
whole of this special issue is worth reading.]
Artech House produce several books on monopulse radar including:

Monopulse Principles and Techniques, S. M. Sherman, Artech House, Norwood, MA, 1984.
Monopulse Radar, A. I. Leonov and K. J. Fomichev, transl. W. F. Barton and D. K. Barton,
Artech House, Norwood, MA, 1986.
Secondary Surveillance Radar, M. C. Stevens, Artech House, Norwood, MA, 1988. [See the
second half of Chapter 5.]
3.12 PROBLEMS
3.1 The Siemens-Plessey WF3 Wind finder radar is an X-band tracking radar operating at
9375 MHz that automatically tracks a corner reflector attached to a meteorological balloon.
Conical scanning of a 2° beam at 1500 RPM is used and the PRF is 625 Hz; how many pulses
per scan are there?
TRACKING RADAR 61
3.2 The radar in problem 3.1 has a peak transmitter power of 55 kW, an antenna gain of
35 dB, a receiver bandwidth of 2 MHz and a noise figure of 10 dB. Assuming 6 dB losses and
given the performance indication that a reflector of RCS 120 m 2 can be tracked at a range of
100 km, estimate the angular accuracy of a standard I minute observation.
3.3 A weapon control radar system on a modern fighter aircraft operates at Ill-band using a
0.75 m x 0.5 m planar antenna. In dog-fight mode, the radar scans a 20° x 20° field ahead of
the aircraft at a medium PRF of 10 kHz. If three pulse bursts of 32 pulses are required to
confirm a target detection, what is the repeat time to search the field?
3.4 The Marconi Radar Systems ST802 is a lightweight monopulse naval tracking radar with
a 2.4° beamwidth. If a 30 dB SNR echo were received from an attacking missile, what tracking
accuracy would you expect? Would this be sufficiently accurate to bring the missile down by
radar-controlled gunfire?
3.5 One method of improving azimuth information is to use doppler beam sharpening (similar
to synthetic aperture radar described in Chapter II). Here the motion of an air-borne radar
can be used to subdivide the antenna beam because the relative velocity of ground clutter varies
across the beam. Try these calculations:
(a) For a frequency of 10 GHz (l/J-band), a beamwidth of 4° and an aircraft speed of
300 m s - 1, what is the doppler spreading of the clutter across the beam when the azimuth angle
is 45° away from the direction of motion and the look-down angle is 10°?
(b) If the doppler resolution of the processing is 10 Hz, to what azimuth resolution would
this correspond?
(c) Would doppler beam sharpening be effective looking directly ahead of the aircraft?
CHAPTER
FOUR
RADAR DETECTION THEORY
• How likely are we to detect a target?

• How often will we make mistakes?
• The correlation receiver and matched filter
• Detecting fluctuating targets
• Using multiple measurements
Finding targets and making measurements in a noisy, cluttered environment

is the essential purpose of radar.
4.1 INTRODUCTION
In constructing a complete radar system, the designer must bear in mind

not just the production of signals by the radar receiver, but the interpretation
of those signals. Though radar is now put to many diverse uses, its original
purpose of detecting objects in some volume of space still constitutes a major
part of all its applications. In this case, the user is interested in distinguishing
'targets' in the illuminated volume from the clutter and noise that tend to
obscure it. Once a target has been detected, properties such as its range and
velocity are likely to be of interest.
For a variety of reasons, the voltage supplied by the receiver is never
steady, even if the receiving antenna is fixed. Thermal noise is one source of
fluctuation for which there is no cure; the laws of physics are always with
us. Other fluctuations may be due to variation within the illuminated volume
of space, as can happen if a sidelobe of the main beam is illuminating the
62
RADAR DETECTION THEORY 63
ocean or wind-blown vegetation. There may be random emitters contributing

to the received signal, such as radio waves from space ('cosmic noise'). For
a comparatively small part of the time, under most circumstances, part of
the fluctuating signal will be due to the presence of targets.
By examining. this fluctuating signal, the operator (human or machine)
attempts to find 'events' corresponding to objects of interest. The absence
of events may also be important, as can occur when an apparent detection
appears in one scan, but fails to appear in the next. Probability theory
provides us with criteria for locating events. The approach described here
is not the only one possible (see, for example, Shafer's 'theory of evidence ,1 ),
but it provides one way of answering the most basic question in detection
theory: 'Does my signal indicate the presence of a target?'
To answer this question, it is necessary to examine the various stages
in the formation of the signal on which we base our decision about the
possible presence of a target. In a very simplified form, these are illustrated
in Fig. 4.1. After transmitting a known signal u(t), the radar is switched to
'receive' mode. A fluctuating voltage will appear at the front end of the
receiver, made up of clutter c(t), noise n(t) and possibly signal, if targets
are present. The clutter is the contribution to the return from all extraneous
scatterers. The noise is dominated by the contribution from the receiver itself,
except at very low frequencies. For the purposes of this chapter, we assume
that both types of contribution that tend to cloak the presence of a signal
have the same statistical behaviour, and can be treated in the single term
n(t), In Chapter 5 we discuss the behaviour of clutter further. If a single
scatterer is present in the illuminated volume, embedded in this unwanted
voltage will be, in the simplest case, a signal component Au(t - 1:d ) (for the
moment we are ignoring any doppler shift caused by target motion). The
u(t)
h(t) Decision
--'-------=:"--.c +
H(w) y(t) = I'(t) + m(t) maker
Receiver
c(t) lI(t)
Figure 4.1 The basic system elements affecting target detection.

value A represents the scaling of the signal due to propagation, losses and
the RCS of the scatterer (see Eq. (1.13». For simple hard scatterers, A will
be a constant, but in many cases the target RCS will fluctuate (see Sec. 4.12).
The value rd represents the delay introduced by the propagation of the signal
to and from the target, and provides the measure of range.
At this stage, the presence of a signal will normally be difficult to detect,
because the energy in the signal is very small compared to the noise power,
and to a large extent the voltage trace is dominated by high-frequency noise.
It is the task of the receiver to extract all possible information about the
presence of a target before any decision is made. In order to make progress
with analysing how it should be designed in order to do this, we need to
know some basic facts about linear systems. These are covered in any basic
text on communications or signal processing, such as Schwartz2 or Stremler 3 ,
and you may wish to consult one of these texts to remind yourself of their
derivation. For the moment, all we need to know is that the behaviour of
a linear system is completely described if we know the way it reacts to an
input signal that is a short, sharp spike (an impulse or delta function). This
is the impulse response function h (t), which tells us the response to a delta
function b (t) at time t = o. Since the response cannot occur before the
stimulus, h(t) is causal, i.e. t < 0 implies h(t) = O. For a general input signal
I (t), the output 9 (t) is given by a convolution operation
g(t) = I*h(t) = f:", I(s)h(t - s) ds (4.1 )
In the integral s is treated as a time variable. If f (t) were the sum of two
signals, i.e. 11 (t) + 12 (t), then we can see straight away by substituting into
the integral that the response of the system to the sum is the same as the
sum of the individual responses. This is what we mean by linearity; it is also
called the superposition property. The other vital property of convolution
can be obtained by replacing I (t) by I( t - r d ), a delayed version of I (t).
If you do this, you find that the response is also delayed by rd.
The relations between the input and output signals take on a more
attractive form when we take Fourier transforms. This has the remarkable
property of converting convolution into multiplication, so that Eq. (4.1)
becomes
G(w) = F(w)H(w) (4.2)
Here the use of lower- and upper-case letters indicates a Fourier transform
pair, i.e.
I(t)-F(w)
etc. The Fourier transform of the impulse response function H (w ) provides

a complete description ofthe system, and is called the system transfer function.
The receiver will be treated as a linear filter with impulse function h (t)
and system transfer function H (OJ), so that the output from the receiver,
y (t), can be written as
y(t) = v(t) + m(t) = Au*h(t - r d) + n*h(t) (4.3)
where v(t) and m(t) are the responses to the signal and noise inputs
respectively, i.e.
v(t) = A f:oo u(t - rd - s)h(s) ds (4.4 )
and
m(t) = toooo n(t - s)h(s) ds (4.5 )
Since, in practice, both u (t) and h (t) are of finite duration and h (t) will be
causal, we can change the integration limits and write the signal term as
v(t) = A IT u(t - rd - s)h(s)ds (4.6)
where we have assumed that h(t) is non-zero only in the range 0 ~ t ~ T.

The noise term m (t) can be written similarly.
Both u (t ) and h(t ) are under the control of the system designer, so that
u*h(t) is deterministic and known, but the values of A and rd are unknown.
However, we expect them to remain approximately constant during the
interval over which u (t) is significantly different from 0, i.e. during the
illumination of the target by a single pulse. Hence the signal component of
y (t) is of known shape, but scaled by A and shifted in time by rd. Effectively,
there are two degrees of freedom associated with the signal (ignoring
frequency shifts for the moment). By contrast, n*h(t) is a random waveform
with many more degrees of freedom (which are determined by the bandwidth
of the receiver). The detection problem is to tell whether the output y (t )
contains noise only or includes a signal term.
In order to assist in this, a major part of the task of the receiver is to
bandpass the input. Typically, the input noise has a much wider bandwidth
than the signal term, so that bandpass filtering can be performed without
losing any information. The bandwidth of the receiver must be matched to
that of the signal. If the bandwidth used is too broad, unnecessary noise will
corrupt the radar measurement (Fig. 4.2a); if it is too narrow, a significant
amount of signal may be lost (Fig. 4.2b). Just having the correct bandwidth
is not enough, however. It turns out that the best possible way of extracting
information is to shape the system transfer function H (OJ) of the receiver so
that the SNR on output is maximized. This optimum spectral shape is called
the matched filter, and its time-domain implementation is known as a
Signal Spectral response Signal Spectral response

spectrum of receiver spectrum of receiver
I
L__ _ \ L
I \ \
I \ \
I \
I \
( \
I \
w w
(a) (b)
Figure 4.2 The effects of a bandpass filter on target detection. In (a) the passband is too wide,
allowing excessive noise to affect detection. In (b) the passband is too narrow, and signal energy
is lost.
correlation receiver. These are derived in Secs. 4.7 and 4.8. (The equivalence
of optimizing information extraction and maximizing SNR is not as obvious
as it may appear; Woodward 4 provides a classic analysis of the problem.)
Both before and after matched filtering, the trace is at the intermediate
frequency (IF), i.e. it is a narrow-band signal centred on the IF carrier
frequency, and it is normal to perform envelope detection to remove the
carrier before making any decision. However, since all the essential ideas are
unaltered if we ignore the carrier frequency and assume that the input to
the filter is at baseband, we have elected to make this simplification. This
keeps the mathematical complexity to a minimum. For the reader who prefers
a more precise treatment, in Sec. 4.11 we indicate the modifications necessary
to deal with the case that occurs in practice.
4.2 THE BASIS FOR DECISION MAKING-PROBABILITY

THEORY
The trace y (t) available to the decision maker might appear as shown in
Fig. 4.3, and at each instant (corresponding to each range), it is necessary
to decide whether the value of y(t) indicates the presence of a target. For
example, should the spikes at t1 and t2 in Fig. 4.3 be attributed to anything
other than noise? This involves a judgement as to whether this value was
more likely to have arisen from noise alone or from signal plus noise. The
basis for this decision must be a knowledge of the frequency with which
different values of y (t) will occur in the two circumstances.
A very useful way to describe how frequently y(t) takes different values
is through its probability density function (PDF) Py (y). This function is
defined by the property that, for small dy, values of y(t) in the range
y ~ y(t) ~ y + dy will occur with frequency given approximately by py(y) dy.
6
4
-2
-4t
-6
Figure 4.3 A voltage trace on which detection decisions might be made.
Hence, as ~y tends to 0,
p{a:::;y(t):::;b} = I b
py(y)dy (4.7)
(Read the left-hand-side of this equation as 'the probability that the measured
value y(t) lies between a and b'. The right-hand side is the area under the
graph of Py (y) between a and b.) Since Py (y) is measuring relative frequencies,
py(y) ~ 0 (4.8 )
and
(4.9)
The second condition simply states that, when a measurement y(t) is made,
the value obtained is certain to be a real number.
We have already seen the exponential PDF (Eq. (2.6) and Fig. 2.4). It
is easy to show that this PDF meets the two conditions. It occurs in a number
of important places in radar applications. Of these, we shall meet it as a
model for clutter in synthetic aperture radar images, and as a model for
targets with fluctuating ReS (Sec. 4.12).
Three particularly useful parameters that can be extracted from the PDF
are the mean p., the variance (12 and the standard deviation (1. The mean is
one type of representative value for the random function y (t). The other
two measure the spread of values taken by y (t). We shall see that they also
have very simple physical interpretations for electrical signals.
In order to explain how these parameters are found from the PDF, it
is easier initially to do this for a digital signal. Suppose that y( t) can take
only a finite set of values {Yl' Y2' .""' YM} (which in modern radars is normally
the case, since the signal will have passed through an A/D converter). In a
long series of measurements, Yi will occur ni times. Then the total number
of measurements is N = n 1 + n2 + ... + nM' and the average value of Y is

n 1Yl + n2Y2 + ... + nMYM " ()
N = ,--YiP Yi
Here we have written P(Yi) = nd N for the relative frequency of Yi. So

the average value of Y is found by multiplying each possible value of Y
by its relative frequency or probability. By the same reasoning, the average
value of y2 would be Lyrp(Yi)' and similarly for any other average value
involving y.
When we deal with a continuous PDF, the only modification needed to
these expressions is to replace the sum by an integral. The mean or average
value is then defined by
(4.10)
The average square deviation from the mean is known as the variance, and
is given by
(12 = f:ro (y - Jl)2py(y)dy
f:ro y2Py(y)dy - 2Jl Lroro ypy(y)dy + Jl2 f:oo py(y) dy
= Loooo y2py(y) dy - Jl2 (4.11)
Here we have used Eqs (4.9) and (4.10) to get from the second to the third
line. There are thus two ways to find the variance. The first is to subtract
the mean and then average the square differences. The second is to find the
average square value, then subtract the square of the mean. Either way gives
the same answer. The RMS deviation from the mean is the square root of
the variance and is known as the standard deviation (1. In these expressions
we are implicitly assuming that these parameters do not depend on the time
of measurement (the system is stable).
We need to take averages at several points in this chapter, and we do
not want to have to write the full integral with the PDF every time. Hence
we use the notation E[ ] (for expectation or expected value) to stand for
the average value of whatever appears between the square brackets. As
examples, we could write
Jl = E[y]
It is worth remembering that taking the expectation is a linear operation,
so that E[y + z] = E[y] + E[z] and E[cy] = cE[y] if c is a constant.
Worked example In a synthetic aperture radar (SAR) intensity image,

an extended target such as an agricultural field may occupy many pixels
(picture elements). Each of these pixels is a separate measurement of
the energy returned from the target. Because of the coherent nature
of SAR processing, each such pixel is affected by a type of noise known
as speckle. This causes the measurements to have values that are
exponentially distributed,
for y ~ 0
Is a single pixel measurement likely to be a good estimate of the mean
value of the target?
SOLUTION The mean value is given by
p. = L'" (yja) e- y/a dy = a

It is easy to show that the standard deviation is also given by a (check
it!). For this asymmetrical distribution (shown in Fig. 2.4), the proba-
bility of finding a value of y below the mean is
f: (lja)e-y/ady= l-e- 1 ;:::;0.63
So the probability of finding a value of y above the mean is e - 1 ;:::; 0.37.

Hence from a single measurement we are nearly twice as likely to
under-estimate the true ReS of the target as to over-estimate it. Because
the standard deviation is equal to the mean, single measurements give
very unreliable estimates of the true mean value. This creates major
difficulties in obtaining precise measurements of the ReS of extended
targets in SAR images.
The mean, variance and standard deviation have very direct interpretations
in circuit terms. During the time T, y(t) will be expected to have values in
the range y ~ y(t) ~ y + Ay for approximately At = Tpy(y) Ay seconds.
Hence, if we regard the voltage y(t) as an input to a 1 ohm resistance, then
y(t) has a De value and a mean power given by
1
Devalue;:::; - ~> At = LYPy(y) Ay [V] (4.12 )
T
and
1
Power;:::; - Ly2 At = Ly2py(y) Ay [W] (4.13 )
T
Allowing Ay to tend to 0, the sums on the right-hand side become integrals,
which are E[y(t)] and E[y2(t)] respectively. Hence we can see that
• DC value = J.l [V]

• Mean signal power = u 2 + J.l2 [W]
• AC signal power = u 2 [W]
• RMS AC signal variation = u [V]
We will make extensive use of the gaussian or normal distribution, since

the voltage or current due to thermal noise is known on both theoretical
and empirical grounds to conform to this distribution. Normally distributed
noise n (t) with mean J.l and standard deviation u has a PDF given by
1
Pn(n) = J 2 exp[ -(n - J.l)2/2u 2 ] (4.14 )
(2nu )
so that if we know its mean and standard deviation we know everything
about it. (To check that J.l and u really are the mean and standard deviation
of the gaussian distribution, we should carry out the integrations of
Eqs (4.10) and (4.11); they give the right answers!) Thermal noise has zero
mean (J.l = 0), so that thermal noise with power u 2 has PDF
1
Pn(n)=J 2 exp(-n 2/2u 2 ) (4.15 )
(2nu )
For detection problems, we are often concerned with the probability
that our measurement exceeds some threshold. If n (t) has a gaussian
distribution, the probability that a single measurement n (t) exceeds k
standard deviations above its mean is given by the area of the shaded region
in Fig. 4.4. This area has the value
p{n(t) > J.l + ku} = J 1 2 feo exp[ -(n - J.l)2/2u 2] dn

(2nu ) II +ka
1 foo exp( -z2/2)dz

=~
V (2n) k
1
= ~ feo exp( _Z2 /2) dz
v (2n) 0
_ ~ [k exp( -z2/2)dz
V (2n)]0
= t- <1>(k) (4.16 )
To get from line 1 to line 2 in this derivation we made the substitution
z=(n-J.l)/u
0.5
IL
Figure 4.4 The gaussian PDF for mean J.I and standard deviation u; the shaded area
corresponds to the probability of observing a noise value k standard deviations above the mean.
In the last line we introduced the function
(t) = ~ I' exp( -x 2 /2)dx

v' (2n) J0
which is commonly used in texts on probability and statistics. Values of (t )

are given in Table 4.1; the area corresponding to (t) is shown as shaded
in the Fig. with this Table. The last line of the derivation of Eq. (4.16) uses
the fact that <1>( 00 ) = 1/2. Another helpful fact is that
<1>( -t) = -(t)
which is clear from the Fig. with Table 4.1 (if you think about it).
Many engineering texts prefer to use a close relation of (t) known as
the error junction, defined by
2
erf(t) = ~ JoIt exp( _x 2
) dx
The two functions are related by

erf(t) = 2<1>(t}2)
(Be careful if you look up either of these functions that the definition being
used coincides with that above, since there are several variations on the
definition of both functions.)
Table 4.1 Values of the function (J) (k), which correspond to the area shown
in the diagram for a mean-zero gaussian variable with standard deviation 1.
Extreme values of (J) (k) corresponding to false-alarm probabilities of 10 - 5,
10- 6 ,10- 7 and 10- 8 are given in Table 4.1(b).
(a)Values of (k)
K 0 2 3 4 5 6 7 8 9
0.0 0.0000 0.0040 0.0080 0.0120 0.0160 0.0199 0.0239 0.0279 0.0319 0.0359
0.1 0.0398 0.0438 0.0478 0.0517 0.0557 0.0596 0.0636 0.0675 0.0714 0.0754
0.2 0.0793 0.0832 0.0871 0.0910 0.0948 0.0987 0.1026 0.1064 0.1103 0.1141
0.3 0.1179 0.1217 0.1255 0.1293 0.1331 0.1368 0.1406 0.1443 0.1480 0.1517
0.4 0.1554 0.1591 0.1628 0.1664 0.1700 0.1736 0.1772 0.1808 0.1844 0.1879
0.5 0.1915 0.1950 0.1985 0.2019 0.2054 0.2088 0.2123 0.2157 0.2190 0.2224
0.6 0.2258 0.2291 0.2324 0.2357 0.2389 0.2422 0.2454 0.2486 0.2518 0.2549
0.7 0.2580 0.2612 0.2642 0.2673 0.2704 0.2734 0.2764 0.2794 0.2823 0.2852
0.8 0.2881 0.2910 0.2939 0.2967 0.2996 0.3023 0.3051 0.3078 0.3106 0.3133
0.9 0.3159 0.3186 0.3212 0.3238 0.3264 0.3289 0.3315 0.3340 0.3365 0.3389
1.0 0.3413 0.3438 0.3461 0.3485 0.3508 0.3531 0.3554 0.3577 0.3599 0.3621
1.1 0.3643 0.3665 0.3686 0.3708 0.3729 0.3749 0.3770 0.3790 0.3810 0.3830
1.2 0.3849 0.3869 0.3888 0.3907 0.3925 0.3944 0.3962 0.3980 0.3997 0.4015
1.3 0.4032 0.4049 0.4066 0.4082 0.4099 0.4115 0.4131 0.4147 0.4162 0.4177
1.4 0.4192 0.4207 0.4222 0.4236 0.4251 0.4265 0.4279 0.4292 0.4306 0.4319
1.5 0.4332 0.4345 0.4357 0.4370 0.4382 0.4394 0.4406 0.4418 0.4429 0.4441
1.6 0.4452 0.4463 0.4474 0.4484 0.4495 0.4505 0.4515 0.4525 0.4535 0.4545
1.7 0.4554 0.4564 0.4573 0.4582 0.4591 0.4599 0.4608 0.4616 0.4625 0.4633
1.8 0.4641 0.4649 0.4656 0.4664 0.4671 0.4678 0.4686 0.4693 0.4699 0.4706
1.9 0.4713 0.4719 0.4726 0.4732 0.4738 0.4744 0.4750 0.4756 0.4761 0.4767
2.0 0.4772 0.4778 0.4783 0.4788 0.4793 0.4798 0.4803 0.4808 0.4812 0.4817
2.1 0.4821 0.4826 0.4830 0.4834 0.4838 0.4842 0.4846 0.4850 0.4854 0.4857
2.2 0.4861 0.4864 0.4868 0.4871 0.4875 0.4878 0.4881 0.4884 0.4887 0.4890
2.3 0.4893 0.4896 0.4898 0.4901 0.4904 0.4906 0.4909 0.4911 0.4913 0.4916
2.4 0.4918 0.4920 0.4922 0.4925 0.4927 0.4929 0.4931 0.4932 0.4934 0.4936
2.5 0.4938 0.4940 0.4941 0.4943 0.4945 0.4946 0.4948 0.4949 0.4951 0.4952
2.6 0.4953 0.4955 0.4956 0.4957 0.4959 0.4960 0.4961 0.4962 0.4963 0.4964
2.7 0.4965 0.4966 0.4967 0.4968 0.4969 0.4970 0.4971 0.4972 0.4973 0.4974
2.8 0.4974 0.4975 0.4976 0.4977 0.4977 0.4978 0.4979 0.4979 0.4980 0.4981
2.9 0.4981 0.4982 0.4982 0.4983 0.4984 0.4984 0.4985 0.4985 0.4986 0.4986
Table 4.1 Continued
K 0 2 3 4 5 6 7 8 9
3.0 0.4987 0.4987 0.4987 0.4988 0.4988 0.4989 0.4989 0.4989 0.4990 0.4990
3.1 0.4990 0.4991 0.4991 0.4991 0.4992 0.4992 0.4992 0.4992 0.4993 0.4993
3.2 0.4994 0.4993 0.4994 0.4994 0.4994 0.4994 0.4994 0.4995 0.4995 0.4995
3.3 0.4995 0.4995 0.4995 0.4996 0.4996 0.4996 0.4996 0.4996 0.4996 0.4997
3.4 0.4997 0.4997 0.4997 0.4997 0.4997 0.4997 0.4997 0.4997 0.4997 0.4998
3.5 0.4998 0.4998 0.4998 0.4998 0.4998 0.4998 0.4998 0.4998 0.4998 0.4998
3.6 0.4998 0.4998 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999
3.7 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999
3.8 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999 0.4999
3.9 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000
(b) Extreme values
k 4.26 4.75 5.20 5.61
<1>(k) 0.49999 0.499999 0.4999999 0.49999999
4.3 THE EFFECTS OF THE RECEIVER ON THE NOISE

DISTRIBUTION
We have already remarked that the thermal noise n(t) entering the receiver
has a zero-mean gaussian PDF. This is not an adequate description of the
noise at the point where the decision is made, because at that point the noise
(and any signal component) has passed through the receiver. Hence we need
to consider the noise and signal-plus-noise distributions at the output of the
receiver. Again, we have to use results from linear system theory (see
references 2 and 3). Initially we deal with the case where we have noise n (t)
alone on input, giving a noise term m (t) at the output from the receiver.
Though for a general input distribution the calculation of the output
PDF can present great difficulties, the situation for gaussian inputs turns
out to be remarkably simple. In fact, gaussian inputs give rise to gaussian
outputs. Hence all we need to know to describe the output distribution are
the output mean value and standard deviation. The mean is easily taken
care of, since it is related to the mean of the input noise by
(4.17 )
The input noise has mean 0 (fln = 0), so flm = O.

We could calculate the standard deviation of m{ t ) directly, but a proper

understanding of the effects of filtering is gained by a more indirect approach.
This involves examining the relationship between values of the output
separated in time. To do this we use the autocorrelation function (ACF),
defined as
Rm{'r) = E[m{t + e)m{t)] (4.18)
which is assumed to depend only on the separation in time r (normally called
the lag) between the measurements. The significance of Rm ( r) is perhaps
best understood by considering the average square difference between the
values ofm{t) at two times separated by an intervale. This can be written as
E[{m{t + e) - m{t)}2] = E[m 2 {t + e) + m2 {t) - 2m{t + e)m{t)]
= 2[0"~ - Rm{e)] (4.l9)
where we have used the fact that Jlm = 0, and the assumption that the average
value of m2 (t) does not depend on t. Since 0" m is independent of the lag (it
depends only on the PDF), all the essential information on the rate of change
is contained in the ACF. We can obtain the standard deviation from the
ACFby
(4.20)
Rate of change is intimately connected with bandwidth, and the ACF
has a corresponding frequency-domain expression in the power spectrum
Sm{w) of the trace, which is obtained by taking the Fourier transform t of
the ACF
(4.21 )
From this relation we can write down the ACF if we are given the power
spectrum, since, using the definition of the inverse Fourier transform,
Rm{r) = - 1 foo Sm{w)ejO), dw (4.22)

2n -00
We can then also write down the variance, since
O"~ = Rm{O) = -1 foo Sm{w) dw (4.23 )

2n - 00
The power spectrum can be physically interpreted as giving the distribution
of noise power as a function of frequency, in the sense that the contribution
t For consistency with most modern books on radar and communications, alJ the sections
of this book dealing with Fourier transform ideas use radian frequency w rather than cyclic
frequency f. By contrast, most engineers in their normal discourse work in Hz, and most of
our discussions do the same. Since f = w /2n, conversions are easy enough; translating between
the two is just part of the radar game (but still annoying).
to the total noise power from a narrow band I1w of frequencies centred on
w is well approximated by Sm (w) 11m/ (21t).
Let us first apply these ideas to the input noise. An important idealization
assumes that the input noise can fluctuate infinitely fast, so that knowing
the value at any instant of time t gives no information about the value of
the noise at any other time, no matter how close it is to t. In correlation
terms, this means that the ACF is a delta function,
(4.24 )
with corresponding power spectrum

(4.25)
The power spectrum is flat, and contains all frequencies with equal power.
Noise with this type of power spectrum is known as white noise. White noise
has infinite bandwidth, which implies infinite power and infinitely fast rate
of change. More realistically, band-limited white noise has a spectrum that
is flat over some finite bandwidth B [Hz] and zero elsewhere, i.e.
ifw ~ 2nB
ifw> 2nB
We can relate the constant N /2 to the noise power per unit bandwidth and
to the system temperature. To do this, note that in either case the total power
in the bandwidth B is
-
1 f27tH N
-dw=NB (4.26)
2n _ 27tH 2
Hence the noise power per unit bandwidth is given by the constant N. We
have already seen in Chapter 2 that the power in bandwidth B is kToB, where
k is Boltzmann's constant and To is the effective noise temperature of the
receiver. Hence
N =kTo (4.27)
For a temperature of 290 K, this gives N ~ 4 x 10 - 21 W per unit bandwidth,

i.e. - 204 dB W per unit bandwidth. In essence, the factor 2 on the right-hand
side of Eqs (4.24) and (4.25) arises because the bandwidth is defined using
the positive frequencies only, while the spectrum is defined for positive and
negative frequencies.
To find the power spectrum and ACF (and hence the standard deviation)
of the output m(t), we need the following key result from linear system
theory. The input and output power spectra from a filter with frequency
transfer function H (w ) are related by
(4.28)
So, if the input is white noise, then

Sm(w) = (N/2)IH(wW (4.29)
Notice that this relation applies whether the input noise is band-limited or
not, as long as the noise bandwidth is equal to or greater than the bandwidth
of the receiver. For. most practical applications, this will always be the case.
The receiver therefore moulds the spectrum of the noise into the shape of
its own transfer function, and forces it to occupy the same bandwidth. In
Fig. 4.5 we see the effect of this. As input we have white noise (Fig. 4.5a),
and as output we have correlated noise (Fig. 4.5b). The rate of change of
the output is now controlled by the bandwidth of the receiver, since using
Eq. (4.22) we can write the output ACF as
Rm("C) = -
N f 2
7[B
4n - 21EB
IH(w)j2ejw< dw (4.30)
N f21tB
CT~ = - IH(wW dw (4.31 )
4n - 21EB
This has a very simple interpretation, since the energy (or integrated gain)
(a) 6
4
~
~
::
-4
-6
(b) 6
4
-6
Figure 4.5 The effect of filtering white noise: (a) shows white noise, while (b) shows the output
when this is passed through a filter with a cosine-squared impulse response function.
of the receiver is defined by

x 1 f 2
1[B
Eh = f -:yo
Ih(tW dt =-
2n -27<8
IH(wW dw (4.32 )
(The equality of the two integrals is a very important result known as

Parseval's theorem. It is essentially a statement about energy conservation,
since it tells us that the total energy can be found by summing the power
at each frequency. Calculating the energy in the time or frequency domain
will yield the same result.) So
Output noise power = O'~ = N Eh/2 (4.33 )
This equation illustrates some important properties:
1. The output noise power is independent of the shape of the impulse

response of the receiver, but depends simply on its integrated gain.
2. For a fixed receiver gain, we can therefore alter the impulse response to
meet any conditions on signal detection we choose, without affecting the
noise variance.
3. The output noise power is independent of the total input noise power
(which is very large; for the white noise model, it is infinite) but instead
depends on the input noise power per unit bandwidth.
Some of the more general results of this section are needed elsewhere in
this book (especially Chapter 5), but if the input is white or band-limited
white zero-mean gaussian noise with noise power N W per unit bandwidth,
they can be summarized very simply. In fact, the output is also zero-mean
gaussian with variance O'~ = N Eh/2, so has PDF
1
Pm(m) =J 2 exp( _m 2 /20'~) (4.34 )
(2nO'm)
It is then easy to find the probability that the output noise m (t) exceeds any
threshold.
Worked example What is the noise power of a radar operating on

100 MHz, using a 1 MHz bandwidth?
SOLUTION If we assume that the receiver passes all frequencies in this

band with a constant gain factor A, then H (w) = A if Iw - Wo I ~ n x 106
and is zero otherwise; here Wo is the centre frequency (see Fig. 4.6a).
Therefore
H(w)
I MHz
A- 4 •
-wo wo
I A
y y
J
t w
100 MHz
H(w)
2 MHz
A
i w
100 MHz
Figure 4.6 Two possible receiver system transfer functions.
For N = 4 X 10-21 and A = 1 this gives a noise power of approximately

4 x 10-15 W. Note that the operating frequency of the radar is irrelevant to
this calculation; all that matters is the bandwidth.
If the system transfer function of the receiver had a triangular shape
(see Fig. 4.6b) with the same bandwidth (defined here as the bandwidth
between the two points at which the transfer function drops to half its
peak value), then the noise power would be two-thirds of this value (see
Problem 4.4).
4.4 THE DISTRIBUTION OF SIGNAL PLUS NOISE
Suppose now that a signal component is present at the input to the receiver.
We know that if a delayed signal Au(t - 't d ) enters the receiver together
with additive noise n(t), then the output is given by (see Eq. (4.3»
y(t) = v(t) + m(t) (4.35 )
where v (t) and m (t) are the signal and noise terms in the output. Since at
the point of making a decision the user does not know 'td or, indeed, whether
there is a signal present at all, all that can be said is that, if there is a signal
present, the output distribution at time t will be gaussian, but with an
unknown mean v (t). Hence the decision maker is trying to separate, on the
I'
I \
I \ Signal plus
I \ noise
I \
I \
I \
I \
\
' .....
Figure 4.7 The PDFs of noise and signal plus noise at the output of the receiver.
basis of an observation at time t, two possible cases:
1. Noise only: the PDF of a single sample is zero-mean gaussian, with

variance CT~ = N Eh /2.
2. Signal plus noise: the PDF of a single sample is gaussian, with unknown
mean v(t) and variance CT~.
In case 2, we can write
PY(Y)=J(21tCT~)exp -
1 (y 2CT~
- V)2) (4.36)
where V is the value of v (t) at the moment of observation. The expression

for case 1 is given by setting V = O. The PDFs in the two cases are as shown
in Fig. 4.7. Two things seem clear from this figure:
• A simple threshold is the best way of separating the two cases.

• The greater the value of V(t)/CT m , the better the chance of making the
right choice t.
Though we have simplified things by working at baseband, these conclusions

also apply at IF (see Sec. 4.11).
4.5 THE SIGNAL-TO-NOISE RATIO

A slight digression is needed at this point, in order to clarify the concept of
signal-to-noise ratio. There are two definitions widely used in the literature,
t The perceptive reader will realize that for a voltage trace, both the cases of positive and
negative V should be considered. This leads to a threshold on ly(t)1 and the probability of
making the right choice increases as Iv(t)/uml increases. This refinement complicates the
mathematics without clarifying the essential principles, so that both here and in Sec. 4.6 we
treat only the case where vet) is positive.
often without any clear distinction being made between them. For a voltage
trace y(t) = f(t) + n(t) consisting of a signal component together with a
noise component, the first definition is the ratio of the energy in the signal
part (integrated over some time T) to the expected energy in the noise over
the same time,
SNR = energy in signal [ ] (4.37 )
energy in noise
For mean-zero noise of average power (12, the energy received in time Tis
T(12, so that this definition can be written
SNR = energy in signal power in signal
[ ] (4.38)
T power in noise (12
This definition is of most use at IF, so that the integration can be carried
out over many cycles of the carrier frequency, while the modulation remains
essentially unchanged. In this case, the received signal v(t) effectively has
the form V cos (wot), where Wo is the carrier frequency and V is a constant.
Hence the signal power is V 2 /2, giving SNR = V 2 /2(12. This is the SNR
calculated using the radar equation in the preceding chapters.
The second definition compares the instantaneous value ofthe signal with
the RMS noise power,
SNR' = IV(tW/(12 [ ] (4.39)
(some authors use Iv(t)I/(1). This definition is most appropriate when
discussing baseband signals. The two definitions are related, since, in practice,
the baseband signals will actually be the envelope of an IF signal. Over
intervals short compared with the rate of change of the envelope (i.e. much
shorter than 1/ B, where B is the bandwidth of the baseband modulation),
the IF signal can be written V cos (wot), where the baseband envelope has
instantaneous value V. Hence
[ ] (4.40 )
and we can see that

SNR' = 2 x SNR [ ] (4.41 )
For consistency, throughout this book we adopt the first definition, which
is the one used in the radar equation. Care must therefore be taken in
comparing our results with those of other authors, particularly in the sections
on the correlation receiver and matched filter (Secs 4.7 and 4.8).
Notice that neither of these definitions is particularly useful to describe
the signal-to-noise ratio prior to the receiver, because at this stage we regard
the noise as having a very large bandwidth, and hence very large total power
(for white noise, this power would formally be considered infinite). It is only
after passing through the receiver ( or the noise bandwidth is reduced in some
way) that the SNR becomes meaningful, since the noise power is then simply
NEh/2. Figure 4.7 indicates that, as the SNR increases, the whole of the
signal plus noise PDF moves to the right. This implies that the chance of
making the right decision when the measured signal is thresholded also
increases.
4.6 DETECTION AND FALSE-ALARM PROBABILITIES
A threshold decision may be the right way to separate signal from noise,
but how do we choose the threshold? For any threshold Y, four types of
events can occur. These are:
• Target present; y(t) > Y; correct detection.

• Target present; y(t) < Y; missed detection.
• Target not present; y(t) > Y; false alarm.
• Target not present; y(t) < Y; no action.
The frequency of occurrence of these events can be described using two

probabilities. The first is the probability of deciding that a signal was present
when in fact there was only noise. This is the probability of a false alarm,
given by
PCa = too Pn(Y) dy ( 4.42)
The second is the probability of a correct detection, given by
Pd = too Psn(Y) dy (4.43 )
The subscripts indicate the PDFs of noise (n) and signal plus noise (sn).
Using the PDFs of Fig. 4.7, the areas defining PCa and Pd are shown in Fig. 4.8.
It is clear that, whatever the value of Y, increasing it will cause both PCa
and Pd to decrease. This illustrates that there is always a trade-off between
improving detections and reducing false alarms. In practice, most radars are
operated at a threshold level giving very low false-alarm rates (and hence
long times between false alarms), for reasons discussed in Chapter 5. In most
places in this book we have adopted a false-alarm probability of 10- 6 as
our working criterion. We now show how this fixes the threshold level, and
hence the reliability with which we can detect targets.
Let us calculate the probabilities of a correct detection and a false alarm
for a threshold set at k times the RMS noise value. The constant k is known
as the (multiplication) margin; when expressed in dBs it is additive. If there
(a)
(b)
Signal
plus
noise
Figure 4.8 The areas in the PDFs of (a) noise and (b) signal plus noise corresponding to the
false-alarm and detection probabilities for a threshold Y.
is a signal value V present, then

Pd = p{y(t) > kt1m }
= p{m(t) + V> kt1 m }
= p{m(t) > kt1 m - V}
=t-CI>(k- Vjt1 m ) (4.44 )
using the same type of calculation as in Eq. (4.16). Since SNR = V j2t1~, 2
we can write this as

Pd = t- CI>(k - J(2 x SNR» (4.45 )
In Fig. 4.9, we plot Pd as a function of SNR for a range of values of k; this

figure shows that, as the SNR rises, the detection probability also increases,
as expected.
For fixed k, the probability of a false alarm is fixed, and has the value
Pfa = p{m(t) > kt1m} = t- CI>(k) (4.46 )
Note that, as V becomes close to 0, Pd = Pfa' i.e. we are as likely to decide

that there is signal present as we are to detect a false alarm.
SNR
Figure 4.9 A plot of P d against SNR for a range of values of Pra.
Worked example What would be the false-alarm rate for a threshold set
at three times the noise power; what would the single-pulse detection
probability be if the SNR for a single pulse was 10 dB?
SOLUTION Since k = 3, we can immediately write

Pfa = t - <1>(3) ~ 0.0013
Since SNR in power units (as well as in dB) is 10,
Pd = t- <1>(3 - j20) ~ 0.9292

For a false-alarm rate of 10- 6 we find from Eq. (4.46) th;tt k must
have the value 4.75, i.e. the threshold must be set at 4.75 times the noise
power. Hence
Pd = t - <1>(4.75 - j(2 x SNR»
If we work with a detection rate of 80 per cent, this implies that
0.8 = t- <1>(4.75 - j(2 x SNR»
or
$(4.75 - J(2 x SNR» = 0.3
Hence our working criterion needs a SNR of 15.6 (11.94 dB) for reliable
detection. 'Reliable' here means Pd = 0.8.
COMMENTS As we have already said, practical receivers do not use the

output ofthe receiver directly, but would normally envelope-detect before
making a decision about the presence of a target. This is the subject of
Sec. 4.11. In practice, this does not greatly change the value of the
detectable SNR. Under the same conditions of 80 per cent detection rate
and false-alarm rate of 10- 6 , we find that a threshold value k of 5.26
times noise power is needed, and the required SNR for detection is 17.9
(12.53 dB).
4.7 THE CORRELATION RECEIVER
Up to now, we have ignored the form of the signal term v(t) coming from
the receiver, and have simply examined the implications of thresholding the
combined signal plus noise. In practice, the shape of the signal coming from
the receiver is largely under the control of the system designer, and its
properties are chosen to meet such specified criteria as desirable time or
range sidelobe levels, resolution, etc. The most fundamental criterion, against
which all other criteria must be balanced, is the ability to detect targets,
which, as we have seen, is a function of the SNR. Hence a natural priority
is to design the receiver to maximize SNR. By Eq. (4.33), we know that the
noise power (Jm is dependent only on the gain of the receiver, not on the
shape of its impulse response function. Hence, for fixed gain, the best SNR
is obtained by maximizing the response to the signal term. This is achieved
by an elegant, conceptually simple and readily implemented processing
scheme, known as the correlation receiver, whose frequency-domain imple-
mentation is the matched filter.
The scheme relies on a fundamental result known as the Cauchy-Schwartz
inequality. This states that, given two functions f (s) and 9 (s) of finite energy,
r
then
lIb f(S)g*(S) dS I
2
~ If(sWds Ib Ig(sWds (4.47)
with equality if and only if

f(s) = cg(s) (4.48 )
for some constant c. Here the asterisk denotes complex conjugate, and we
allow complex functions so that we can, if desired, use complex repre-

sentations of real signals or we can use the result in the frequency domain,
where the signals will normally be complex. For time-domain signals, taking
a = - 00 and b = 00, the inequality can be interpreted directly as
If:oo f(s)g*(s) dsl2 ~ EfE g (4.49 )
where
Ef = J:oo If(sW ds (4.50)
is the energy in f(s), and Eg is defined similarly.

Applying this to the signal component of the output of the receiver
(Eq. (4.3»,
v(t) = toooo Au(t - rd - s)h(s) ds (4.51 )
we must replace f(s) by Au(t - rd - s) and g(s) by h(s) (assumed real).

Since Au (t - rd - s) is simply a scaled, reversed and shifted version of u (s),
its energy is 1A 12 Eu , and we can write
(4.52)
Note that this only depends on the energy in the transmitted signal u(t),
not its shape. Since the noise variance is N Eh/2, the instantaneous SNR
satisfies the inequality
(4.53 )
The maximum possible SNR at the output of the receiver is equal to the
ratio of the total energy of the input signal and the input noise per unit
bandwidth. This is independent of the integrated gain of the filter, E h , since
both the maximum value of the signal term and the noise power are
proportional to the gain.
How do we obtain this maximal value of SNR ? The Cauchy-Schwartz
inequality tells us that we must choose h(s) to be proportional to u(t - rd - s),
so that h (s) is a reversed, scaled and shifted copy of u (s). The time at which
the maximum is attained depends on the delay chosen in the filter. Assuming
that u( t) is non-zero only in the range 0 ~ t ~ T, let us set the delay equal
to T, so that
h(t) = u(T - t) (4.54 )
(Here we have set the scaling to 1; since the choice of scaling only affects
the integrated gain of the receiver, it will not affect the SNR or the detection
and false-alarm probabilities. In order to convince yourself of this, carry out
the following calculations with h (t) = Cu ( T - t), where C is a constant.)
This implies that:
• h(t) is causal
• Eh = Ell
• (J; = NEIl/2
To find when the peak SNR will occur, substitute this expression for h(s)
in Eq. (4.51). Then
v(t)=A tT u(t-'rd-s)u(T-s)ds
= A tT u(t - 'rd - T+ s)u(s)ds (4.55 )
(using a change of variables). At time t = T + 'rd , V (t) therefore takes the

value
(4.56)
and
SNR = Iv(tW = A2E; = A2 Ell (4.57)

2(J; NEh N
As Eq. (4.53) shows, this is the maximum possible value of SNR. This key
result means that the maximum response of the correlation receiver gives a
signal-to-noise ratio
SNR = ____s_ig_n_a_l_e_n_er_g_y_ _ __ E
(4.58 )
noise power per unit bandwidth N
where E is the signal energy. It is important to realize that both E and N
are calculated at the input to the receiver, so that the details of the receiver
impulse response are irrelevant as far as detection is concerned. Useful
pulseshapes are affected by other considerations, however, such as resolution;
these are discussed in Chapter 6.
'r
The maximum response occurs at the time T + d , owing to the delay
of T seconds selected in the receiver. We can therefore find 'rd,
as long as
the signal is detectable. But we can easily write down the probability of
detection at this time, using Eq. (4.45). For a threshold k(Jm'
Pd =! - ~(k - Aj(2EIl /N)) =! - ~(k - j(2E/N)) (4.59)
Worked example Assume the radar is operating at a wavelength of 1 cm,

and has a gain G = (50)2 = 2500. The individual pulses are of mean
power 10kWand the pulse length is 1 ms. Let us find the single-pulse
detection probability of a target with ReS 10 m 2 at a range of 10 km,

if the false-alarm rate is 10- 6 .
SOLUTION Since the energy per pulse is Eu = 104 X 10 - 3 = 10 W, and

we have already seen that N ~ 4 x 10 - 21, we simply need to find the
value of A. Using the range equation, we need to work in power units,
so that
Then
AJ(2Eu IN) ~ 1.77
We already know that we must use a value for k of 4.75, which gives a
detection probability
Pd ~ t- $(2.98) ~ 0.001
Such very low single-pulse detection rates are not uncommon for radars
using mUltiple-pulse detection techniques. These are discussed in Sec. 4.10.
There we show that, for coherent processing, using M pulses gives
a SNR improvement by a factor M. In this example, that means that
using 10 pulses would bring the detection rate up to 80 per cent.
The operation being carried out by the receiver is perhaps most clearly
explained by making the change of variables s = t' - t + T in Eq. (4.55).
This gives
v(t) = A II
I-T
u(t' - Ld)U(t' - t + T) dt' (4.60 )
Here u(t' - Ld) is the incoming signal, and u(t' - t + T) is a copy of u,

shifted to t - T; their product is integrated over the range of values for which
the filter takes non-zero values. As t varies, the copy of u (t) moves into
alignment with the incoming signal, then out of alignment again (see
Fig. 4.10). When it is fully aligned, the maximum possible signal response
occurs. At this time, since the receiver energy is fixed at the value Eu , the
maximum SNR also occurs. In Fig. 4.11 is shown the response of the
correlation receiver to a signal contaminated by white noise (cf. Fig. 4.5).
The top two plots show the signal and noise traces; their superposition and
the output from the receiver are shown below.
The operation of integrating the product of one function by a shifted
version of another is known as correlation, and the basis of the operation
described above is to correlate the input with a copy of the transmitted signal
u (t). This is easily implemented by recycling a copy of the transmitted pulse,
and is the principle of the correlation receiver.
Input signal
f-----'-::----~------ - - - - - - - --+
f--------L.,.----'------- - - - - - -
Output signal
f - - - - - - - - - - l ' - - - - - - ' - - - - - --+
f - - - - - - - - - - - - - ' - - = - - - - - ' - - - - - - -+ T d+ T
t - T
Figure 4.10 The response of the correlation receiver to a returned signal. The top plot shows
the return (delayed by 1'd) and the plots below show a succession of positions of the correlating
function. On the right is shown the complete time response of the receiver.
The signal term in the output from the correlation receiver is given by
(see Eq. (4.55»
v(t) = A foT u(t - 1'd - T + s)u(s) ds (4.61 )
which is the correlation of the signal with itself. This is known as auto-
correlation, and appears so often in radar processing that it deserves its own
notation. Since in general we need to consider complex signals f (t), we
define the deterministic autocorrelation function of a signal f (t) as
Pf(t) = L'"oo f(t + s)f*(s)ds = toooo f(s)f*(s - t)ds (4.62)

(a)
O~-------L---- __ L-~ ______- L_ _ _ _ _ _ ~~ --+

_ _ _ _ _ _~ _ _ _ _ _ _~ _ _ _ _
-ol
-6
(b)
-4
-6
(e)
-4
-6
30
(d)
25
20
15
Figure 4.11 The response of the correlation receiver to a signal contaminated with white noise:
(a) shows the signal (delayed by 'd); (b) shows the noise-alone trace; (c) is their summation;
and (d) is the squared response of a receiver matched to the signal shape of (a) when the
signal-plus-noise voltage trace (c) is input to it.
Though the deterministic ACF is related to the ACF of random signals

discussed in Sec. 4.3, they are in practice used in quite different ways, and
it is better to use different notation to describe the two concepts.
We can therefore write the signal term as
v(t) = Pu(t - 'rd - T) (4.63 )
The output for a simple pulse u (t) is shown in Fig. 4.10. It is clear that the
shape of the input has been drastically changed, but this is irrelevant as far
as target (and range) detection is concerned. The form of this change is
usually irreversible, as we now show by considering the frequency-domain
implementation of the correlation receiver.
4.8 THE MATCHED FILTER
The correlation receiver has an impulse response h (t) = u (T - t), where the
constant T simply ensures causality. In the frequency domain, this gives rise
to a transfer function
H (w) = f:", u ( T - t) e - jwt dt
= f:", u(s)e-j,.,(T-S)ds
= e- jroT f:", u(s) ej,.,s ds
=e-jroTU*(w) (4.64 )
where we have assumed that u(t) is real t . Since e- jwT is simply a delay term,
we see that the essential part of the transfer function is the complex conjugate
of the Fourier transform of the transmitted pulse. This receiver transfer
function is known as the matched filter.
The response of the receiver to the signal term is given by
v(t) = u*h(t - 'rd ) (4.65 )
which in the frequency domain becomes
V(w) = e-i"'Td U(w)H(w) = e-jw(Td + 7')1 U(w)12 (4.66 )
t For simplicity, Sees 4.7 and 4.8 have assumed a real transmitted pulse u(t). When u(t) is
phase-modulated, it is normally more convenient to treat it as a complex signal. In this case,
the impulse response function giving maximum SNR is
hit) = u*(T- t)
The form of the frequency domain filter (Eq. (4.64» is unchanged.
This normally implies a loss of information about the shape of u (t), since
all the phase relations between the various frequencies have been lost. The
signal term in the output from the matched filter has a Fourier transform
that is the energy spectrum of the transmitted pulse, multiplied by a linear
phase term corresponding to range and filter delay.
4.9 KEY ELEMENTS OF SIGNAL DETECTION
Essentially, there are four steps in the treatment of signal detection given
above:
1. Formation of the PDFs of noise and signal plus noise.

2. Given a threshold, determination of Pd and Pfa.
3. Checking that, for a fixed threshold, Pd depends only on the SNR, and
that, as the SNR increases, so does Pd.
4. Hence, verifying that a matched filter is appropriate.
We have carried out this sequence of steps for the simplified case where
the incoming signals are at baseband. In practice, they will be at IF. This
more realistic case is dealt with in Sec. 4.11.
4.10 DETECTION USING MULTIPLE OBSERVATIONS
Surveillance radars usually combine multiple observations per scan before

thresholding to detect the presence of targets. Essentially, this can be done
in two ways, either by coherent processing of a sequence of M pulses, or by
processing each separate pulse and then combining the separate observations
in some way (normally simply by averaging). The latter case is called
incoherent processing. The use of the word 'coherent' here refers to whether
the phase relationships (at the carrier frequency) between successive pulses
are known and stable.
If they are, then we can regard a sequence of M pulses as one single
known waveform. Hence we can construct a matched filter for this waveform.
For a non-fluctuating target, the energy returned from a sequence of M
pulses will be M times greater than from a single pulse, while the noise power
is unchanged. Hence the SNR and peak response of the matched filter will
increase by a factor M. The effect that this has on detectability is given by
Eq. (4.45).
Worked example A radar system combines eight pulses coherently before

making a decision about the presence of a target. We have already seen
that this requires a SNR (after coherent processing) of 15.6 if Pfa = 10- 6
and Pd = 0.8. Hence the SNR on a single pulse is 1.95. This gives a
detection probability of only 0.003 for a single pulse.
For the radar to work on a higher single-pulse detection rate would
be a waste of power, given that coherent integration is possible. Multiple
pulses are an essential part in designing radar systems to meet specified
detection probabilities, while operating at reasonable powers.
It is useful to realize that coherent processing can be achieved by

averaging the successive outputs from the receiver. Suppose the radar has
available M measurements Yl' Y2"'" YM per scan, each of which contains
a signal term and a noise term, so that
Yi = Vi + mi
We will make the assumption that the noise terms m i are zero-mean gaussian
and independent of each other and of the signal terms Vi' As above, we
consider the simplest case of a non-fluctuating target so that we can write
Vi = Vo for each i. This is where coherence enters, since without it each Vi
would be of the form Vo cos <Pi' where <Pi is an unknown phase term.
Averaging the M measurements gives
1 1 1
Y = M L Yi = M L (vo + m i ) = Vo + M L mi (4.67 )
We can see that y is made up of a signal term Vo and a noise term L md M.

Without coherence, the signal terms would not have added up in phase, and
no advantage would be gained by averaging. The average value of M
zero-mean gaussian random variables is also zero-mean gaussian, with
variance (J!/ M. The effect on the PDFs of noise and signal plus noise for
M = 8 is shown in Fig. 4.12. The mean values of signal and signal plus noise
are unaltered, but the distributions become more sharply peaked about these
means.
Averaging leaves the signal value Vo unchanged, but reduces the noise
power by a factor M. Writing SNR M for the SNR of the average of M
observations, we therefore have
V02 V2
SNR M = 2 = M -% = M x (SNR 1 ) (4.68 )
2«(Jm/M) 2(Jm
and the effective SNR is increased by a factor M. This is the conclusion we

reached above on energy considerations.
In the incoherent case, nothing is gained by coherent averaging of the
receiver outputs, because the signal terms will be out of phase. We must
envelope-detect first. Envelope detection for single pulses is discussed in the
next section. We find a more complicated expression for Pd (though Pfa is
(a) 0.5
,-,
I \
I \
I \
\
\
\
\
\
\
\
\
\
\
\
\
\ ,
o
(b) O.5,j8
O.4,j8
,I,,
1\
, I
, I
O.3,j8 I I
I I
I I
I I
O.2j8 I I
I I
I I
I
O.l,j8 I
I
I
I
I
,
I
\
\
2
Figure 4.12 The PDFs of noise and signal plus noise for (a) a single pulse and (b) the coherent
average of eight pulses.
simpler) for single-pulse detection. Evaluating the effective SNR after

averaging an envelope-detected pulse is more complicated than for the
gaussian case, but is fully discussed by Marcum 5 • Another useful reference
is Levanon 6 • A convenient rule of thumb is that averaging M square-Iaw-
detected pulses gives an SNR improvement of the order M, if M is large. ofJ
4.11 MODIFICATIONS FOR INTERMEDIATE-FREQUENCY

INPUT TO THE RECEIVER
As we have already indicated, a more realistic model for the radar system
and receiver takes account of the fact that radars use a modulated carrier
wave of frequency Wo on transmission, and that the signal is narrowband-
filtered on reception. After reception, the carrier is removed by an envelope
(or square-law) detector. The target detection operation is carried out on
the ensuing baseband signal. In this case, after reception, the return from a
hard (non-distributed, non-fluctuating) target would have the form
y(t) = x(t) + m(t) (4.69)
Here x (t) = v (t) cos (wot), where v (t) is a baseband modulation term
(corresponding to the baseband signal used in Sec. 4.4 et seq.), and m(t) is
narrowband noise. Such noise can be represented as
m(t) = r(t) cos [wot - tJ>(t)] (4.70)
where r(t) and tJ>(t) represent the fluctuating amplitude and phase of the
noise, both of which have bandwidths comparable to the bandwidth of the
receiver and small compared with Wo. Equation (4.70) can be expanded as
m ( t) = X ( t ) cos (wot) + Y ( t ) sin (wot ) (4.71 )
whereX(t) = r(t) cos tJ>(t) and Y(t) = r(t) sin tJ>(t) are uncorrelated gaussian
zero-mean random coefficients, with the properties
E[X2(t)] = E[y2(t)] = E[m 2(t)] = u! (4.72)
(see e.g. Schwartz2 ). Then
y(t) = [v(t) + X(t)] cos(wot) + Y(t) sin(wot) (4.73)
The detection operation uses the envelope of this signal,
r(t) = J{[v(t) + X(t)J2 + y2(t)} (4.74)
The distribution of this envelope was first derived by S.O. Rice of Bell
Telephone Laboratories 7 • It has since been discussed by many authors (e.g.
Schwartz2, Papoulis8), and has the form
Pr(r) r exp( - (r2 + V 2 )/2u! )/0

= 2"
Um
(rv)
2"
Um
(4.75)
where we have written V for the instantaneous value of the slowly varying
signal v(t). In Eq. (4.75), lo(z) is the zero-order modified Bessel function
of the first kind, given by
lo(z) = - 1 f2" eZ cos 9 dO (4.76)

21t 0
Tabulated values of this function will be found in most books of mathematical

tables (e.g. Abramowitz and Stegun 9 ).
°
Setting V == and using 10(0) = 1 gives the distribution of noise alone
(4.77)
This is the Rayleigh distribution. A threshold kU m gives false-alarm and

detection probabilities
(4.78 )
and
Pd = OO
p,(r) dr = foo exp (_V2) e- 10 (V- J2u )du
-2-
u
(4.79)
f
tam k 2 [2 2um Um
The last expression was obtained from Eq. (4.75) by the substitution
u = r2 /2u~. Since SNR = V 2/2u~, this can be written
Pd = e-(SNR) (00 e-U 10 (2J!(u x SNR)) (4.80)

Jk 2 [2
so that Pd depends only on the SNR and k. Using Eq. (4.78), it is easy to
express k in terms of Pfa. This has been used in Fig. 4.13 on page 96 to plot
Pd as a function ofSNR for a variety of values of Pfa. We see that, as the SNR
increases, so does Pd , which indicates that matched filtering is required to
maximize Pd.
In Levanon 6 , a convenient approximation is quoted that effectively
eliminates k from Eqs (4.78) and (4.80). This yields the following direct and
easy-to-use relation between Pd , Pfa and SNR
SNR = A + 0.12AB + 1.7B (4.81 )
where
A = In (0.62)
Pea.
B=ln(~)
1- P d
and SNR in Eq. (4.81) is not in dB. This relation is fairly accurate if
10- 7 < Pfa < 10- 3 and 0.1 < Pd < 0.9
4.12 TARGET FLUCTUATIONS-THE SWERLING CASES

We have shown that the probability of detection of a target from a single
observation depends on the peak response of the matched filter, which is
directly proportional to the target's RCS at the moment of observation. In
Chapter 2 we noted that this RCS fluctuates for many targets. This comes
1.0
O.S
0.6
1-+-t---Pf " = 10- 5

I - f - - - P f " = 10- 0
0.4
0.2
Figure 4.13 Plots of Pd against SNR for a range of values of Pra' after envelope detection.
about if the target is in fact made up of several scattering centres whose

orientations (and hence phase relations) relative to the radar change with
time. Such changes in phase relationships will give rise to interference effects,
so that the target ReS will exhibit fading characteristics. Swerling 10 suggested
four types of target model for these complex targets, consisting of two types
of variation. The first variation is in the distribution of the observed Res
values, and he identified two distributions as of particular interest. For a
target of average ReS (Jav' these are
p,,((J)=_l exp(-~) (J~O (4.82)

(Jay (Jay
and
(4.83 )
These are both forms of the chi-squared family of distributions. The first is
also known as the exponential PDF, and applies to a target comprising many
independent scattering centres of approximately equal strength. This distri-
bution has a standard deviation (Jav that is equal to the mean. The second
PDF is appropriate to targets that have a dominant scatterer together with
a number of other smaller scattering centres. In this case, the standard
J2.
deviation is (Jav/
The other type of variation is in the rate of change of the RCS. Again,
Swerling suggested two idealizations. The first of these is when the target
RCS changes on a timescale comparable to or faster than the PRF of the
radar, so that different pulses provide independent samples of the fluctuating
RCS. The second is when the target RCS is effectively unchanging between
pulses, but fluctuates on the timescale of the radar's scanning pattern. These
different timescales affect the probability of detection when several pulses
are combined (see Sec. 4.10). Combining the two types of PDF with the two
rates of variation yields the four Swerling models. These are normally referred
to as Sweriing cases 1-4, as set out in Table 4.2. We include the non-
fluctuating case as case 5.
Other types of fluctuation have been considered (see, e.g. Berkowitz l l ),
but the original Swerling cases still form the basis of most target detection
studies.
In order to find out how Swerling fluctuations affect the detection
probability Pd , we must follow the steps set out in Sec. 4.9. The basic problem
is the derivation of the PDF of signal plus noise, and its subsequent
integration. Swerling 10 carried out the necessary calculations for square-law
detection. In this case, if V= v(t) is the signal voltage, the variable V 2 /2
(which measures signal power) will have a PDF of the same form as the
Swerling cases, with a mean value that we will write as V~. For single pulses,
there is no difference between pulse-to-pulse and scan-to-scan variation.
Using a normalized variable
y2(t)
Z=--
2(J;'
Table 4.2 The Swerling cases
PDF Rate of variation Case
4.82 Slow I
4.82 Fast 2
4.83 Slow 3
4.83 Fast 4
Delta function None 5
which expresses the square-law-detected output in terms of the output noise

power u;" the single-pulse PDF of signal plus noise is
pz(z)
1 (-z)
= 1 + 2So exp 1 + 2So (4.84 )
for cases 1 and 2, and
pz(z) = (1 +lSo)2 ( 1 + 1 + zl/SJ expC :~J (4.85)
in cases 3 and 4. Here So = V5;(2u;,) can be regarded as the mean SNRl2.

The corresponding detection probabilities for a fixed threshold Yare
(4.86)
1.0
0.8
0.6
/-H---P la = 10-'
H - - - - P la = 10-<>
# - - - - P la = 10- 7
0.4
0.2
SNR
Figure 4.14 Plots of Pd for single pulses after square-law detection, as a function of SNR, for
Swerling cases 1 and 2 and a range of values of Pr••
for cases 1 and 2, and
Pd = C sJ (
+\ / 1 + 1 : So + ;J C~ ~J
exp
(4.87)
in cases 3 and 4. The PDF of noise alone (normalized to unit standard

deviation) can be obtained by setting So = 0 in Eqs (4.84) or (4.85). In
Fig. 4.14 we plot Pd for SwerIing cases 1 and 2 as a function of SNR, for a
range of values of PCa.
SwerIing lO also provides expressions for the PDFs of averages of
square-law-detected pulses, and of the associated detection probabilities (in
some cases the expressions are too cumbersome to be described exactly, and
approximations are supplied). These expressions are not reproduced here
(SwerIing's paper is very clear and readable, and is recommended if you need
to make detailed detection performance calculations).
4.13 SUMMARY
Target detection is a probabilistic idea; noise and clutter prevent us from

being certain to find the targets we are looking for, and will normally present
us with plenty of 'targets' we are not looking for. We can only define the
probabilities of detection and of false alarm that we are prepared to live
with. These determine the signal-to-noise ratio that is required for detection.
Optimal detection performance is achieved by maximizing the SNR at the
output of the receiver. The receiver that does this uses the correlation or
matched filter principle. As long as a matched filter is being used, the form
of the transmitted pulse is irrelevant for detection purposes. All that matters
is the ratio of the signal energy to the noise power per unit bandwidth on
input to the receiver. By developing expressions for the PDFs of signal plus
noise and noise only, relatively straightforward calculations of detection and
false-alarm probabilities can be carried out for single pulses. These become
more complicated when fluctuating targets and multiple pulses need to be
considered, so that graphical or numerical techniques are needed.
Key equations
• For mean-zero gaussian noise:
p{n(t) > ka} = t- (k)
• Output noise power from receiver with integrated gain Eh :
a; = NEh/2
• Noise power:
N = kTo = -204 dB W /unit bandwidth
• Parseval 's relation:
• Probability density function of signal plus noise (gaussian case):

1 (_(Y_V)2)
Py (y) = J (21t(T~) exp 2(J~
• Probabilities of detection and false alarm (gaussian case):
Pd = t- (k - j(2 x SNR»
Pfa = t - (k)
• SNR at output of matched filter, where E is signal energy on input:
SNR = E/N
• Probability of detection after matched filtering (gaussian case):
Pd = t- (k - j(2E/N»
• The signal term in the output of the correlation receiver is the auto-
correlation function of the transmitted pulse:
Pu(t) = L')oo u(t + s)u*(s) ds = f:oo u(s)u*(s - t)ds
• The matched filter has a system transfer function that is the complex
conjugate of the Fourier transfer of the pulse (ignoring the delay for
causality):
H(w) = U*(w)
• Swerling PDFs:
p,,«(J) = _1 ex p ( _~)
(Jay (Jay
• Coherent averaging of M pulses improves the SNR by a factor M:

SNR M = M X (SNR 1 )
• Detection and false-alarm probabilities after envelope detection:
Pfa = exp( -P/2)
Pd = e-(SNR) roo e- U 10(2j (u x SNR» du

Jk 2 /2
• Relation between Pd' Pfa and SNR after envelope detection:

SNR = A + O.l2AB + 1.7B
where
-1 (0.62)
A- n -
Pfa
B=ln(~)
1- P d
which can be used if

10- 7 < Pfa < 10- 3 and 0.1 < Pd < 0.9
4.14 REFERENCES
1. A Mathematical Theory of Evidence. G. Shafer, Princeton University Press, Princeton. NJ,

1976.
2. Information Transmission. Modulation and Noise, M. Schwartz, McGraw-Hill, New York.
1980.
3. Introduction to Communication Systems. F. G. Stremler, Addison-Wesley. Reading, MA.
1982.
4. Probability and Information Theory. with Applications to Radar. P. M. Woodward, Pergamon
Press, Oxford, 1953.
5. A statistical theory of target detection by pulsed radar, J. Marcum, IRE Trails .. IT-6.
145-267, 1960 [This is the original reference, but the books by Berkowitz" and
Levanon 6 give useful treatments of this problem.]
6. Radar Principles, N. Levanon, Wiley, New York, 1988.
7. Mathematical analysis of random noise, S. O. Rice, Bell Syst. Tech. J .. 23 and 24. 1944;
reprinted in Selected Papers on Noise and Stochastic Processes. Ed. N. Wax, Dover. New
York, 1954.
8. Probability, Random Variables and Stochastic Processes. A. Papoulis, McGraw-Hili, New
York, 1985.
9. Handbook of Mathematical Functions, M. Abramowitz and I. A. Stegun, Dover, New York,
1964.
10. Detection of fluctuating pulsed signals in the presence of noise, P. Swerling, IRE Trans .•
IT-3, 175-178, 1957. [Again, this is the original reference, but the Swerling models are
described in many places, such as the Introduction to Radar Systems. M. I. Skolnik,
McGraw-Hill, New York, 1985.]
11. Modern Radar: Analysis, Evaluation and System Design. R. S. Berkowitz. Wiley, New York,
1966.
4.15 PROBLEMS
4.1 A synthetic aperture radar image consisting mainly of grassland is of dimensions 512 x 512
pixels. The image is being scanned to find stationary vehicles thought to have an Res 10 dB
above the mean clutter level. Assuming exponential speckle, find the probability of a false alarm;
how many false alarms would be likely to occur in the image? To stress the great difference
between exponential and gaussian probabilities, find the signal-to-c1utter ratio needed to give
the same false-alarm rate if the background could be considered gaussian. (The signal-to-c1utter
ratio is the ratio of target brightness above the mean and the clutter standard deviation).
4.2 (a) Show that the signal output from a correlation receiver matched to a rectangular pulse
is a triangular pulse.
(b) Two possible waveforms to be used for detection purposes are shown as (i) and (ii); (ii)
is a binary pulse generated by phase modulation. Each has its own matched filter. Are their
detection performances different? Find the outputs from their respective matched filters, and
comment on any possible consequences of the differences between them.
(i) (ii)
v
v
1 ms
-I ms I ms -1 ms
-v
4.3 What would be the expected thermal noise power at the output of a receiver whose impulse
response function is
2
COS (ttt/T) ifltl";; T/2
h(t) ={
o otherwise
when T = 10- 3 s? This receiver is matched to pulses of the same shape (because it is
symmetrical). Find the SNR at the output of the receiver if the returned signal on input is
10- 8 cos 2 (tt/T)(t - 10- 3 )
Sketch the form of the signal part of the output.
4.4 Confirm the statement in the worked example following Eq. (4.34) that a receiver with a
triangular system transfer function generates only two-thirds of the noise power of a receiver
with a rectangular passband of the same bandwidth.
How would this result change if we defined the bandwidth for the triangular system transfer
function as the bandwidth between the two points at which the system transfer function falls
to zero?
4.5 A ground-based surveillance radar with effective aperture 10 m 2 operates at a wavelength
of 25 cm, using 1 ps pulses and a peak power of 0.1 MW. Assuming a matched filter and a noise
figure of 10 dB, what would be the single-pulse detection probability from an aircraft with RCS
5 m 2 at a range of (a) 50 km and (b) 100 km, if the false-alarm rate was set at 1O- 6 ?
4.6 The signal-to-noise ratio defined in Chapter I involves bandwidth, but this does not appear
in the definition of Sec. 4.5. Explain why there is no inconsistency here.
4.7 What SNR would be needed for a 90 per cent single-pulse detection rate and a false-alarm
rate of 10- 6 using (a) the gaussian treatment and (b) the equations for an envelope-<ietected
signal?
RADAR DETECTION THEORY t03
4.8 If 16 pulses can be integrated coherently. what single-pulse detection rate is allowable in order to
pennit 80 per cent detection rate with a false-alann probability of 1(}-'5?
4.9 We will meet the chirped pulse
exp( -jat2) ifltl..: T
u(t) ={
o otherwise
(where a is a constant) in Chapter 6. Find its matched filter. (You will need the footnote
following Eq. (4.64).) •
4.10 For 80 per cent detection rate and 10- 6 false-alarm rate, what peak power would be
acceptable in a radar coherently averaging thirty-two 2 JIS pulses, if it is operating on a 24 ern
wavelength with an effective aperture of 15 ml, has a loss factor of 50 per cent and needs to
locate aircraft with RCS down to 20 m l within a range of 100 km?
CHAPTER
FIVE
SIGNAL AND DATA PROCESSING
• Doppler information aids target detection

• Why do we choose low false-alarm rates?
• How do we follow a target once it has been detected?
Separating signals from clutter, and tracking targets have become sophisticated
arts.
5.1 INTRODUCTION
The reliability and low cost of modern digital electronics have revolutionized
radar engineering. Over the years, the AID converter has moved progressively
up the receiver chain towards the antenna, increasingly replacing analogue
electronics by digital hardware and computer software. This engineering
activity has two objectives: to remove unwanted clutter as far as possible,
and to detect targets against the residual clutter and noise.
Radar engineers make a distinction between signal processing, the fast
hardware processing developed mostly by electronic engineers, and data
processing, which is concerned mainly with target-detection software and is
very much the preserve of mathematicians. The dividing line between the
two is neither clear-cut nor stationary, as it gradually moves up the receiver
chain with the increasing ability of modern computers to replace hard-wired
circuits. The current position for a typical radar system is shown in Fig. 5.1.
The objective of this chapter is to take a guided tour through the signal
processing and data processing shown in the block diagram of Fig. 5.1, but
104
~ r---,
r-l MTI f-..,
I L _ _ ..J I Track initiation
I ,--------, I
Plot extraction
Plot-track
association
Signal processing
Data processing
Figure 5.1 Signal and data processing in a modern radar system.
~
we cannot begin until we know more about the properties of the clutter we
are trying to filter out. The simplistic case might be the problem of detecting
an aircraft flying against a bare mountainside in the background. In this
case the clutter is stationary. But suppose the mountainside is covered in
trees blowing about in the wind, and the aircraft is a helicopter nearly
hovering; our problem becomes more difficult. Perhaps the ultimate challenge
is presented by an air-borne radar in look-down mode trying to distinguish
a tank from wind-blown vegetation and moving clouds of rain. Clearly it
becomes important to study the properties of ground clutter before we begin
signal processing.
5.2 PROPERTIES OF CLUTTER
Typically, clutter arises from area- or volume-extensive regions illuminated

by the main beam or one of the sidelobes of the antenna pattern. In Chapter 4
we made the assumption that such clutter has the same (gaussian) statistics
as thermal noise at the pre-detection stage. This means that after envelope
detection the clutter values have a Rayleigh distribution
Pe(c)
C
= 2"exp
Ue
(_c
--2
2ue
2
)
(5.1 )
The expressions for detection and false-alarm probabilities are exactly the
same as those derived in Sec. 4.11, with clutter taking the place of noise.
This clutter model is valid as long as, within the illuminated region, the
clutter can be regarded as consisting of many independent random scatterers
of comparable strength. Such conditions normally hold in weather clutter
or chaff, where the scattering is from very many droplets or particles. They
are also encountered in other common situations, such as low-resolution
marine radars operating in calm sea conditions, or radars illuminating desert
or agricultural areas. When the radar is operating under conditions where
the return may be dominated by a few brighter scatterers within each
resolution cell (such as can happen when a high-resolution radar illuminates
a patch of ocean whose dimensions are comparable with the water wave-
length), then this model no longer holds. Under such conditions, larger
values of clutter occur more often than expected from the Rayleigh distri-
bution, so that false alarms will occur more frequently unless the detection
threshold is raised.
Analysis of the detection/false-alarm rates for non-gaussian clutter
requires a model for the clutter and signal-plus-clutter PDFs at the detection
stage. In practice, this normally relies on empirical distributions, supple-
mented by physical reasoning. Several types of model are in use for different
circumstances. For sea clutter, log-normal, Weibull and K distributions have
been used inter alia. With different parameters, the same models also have
SIGNAL AND DATA PROCESSING 107
Table 5.1 Values of Vrms and doppler spread for 1 GHz

transmitter
Target type Doppler spread [Hz] = ¥v rms
Sea clutter, windy day 0.9 6

Forest, moderate wind 0.2 1.3
Chaff 1 6.7
Rain clouds 1.9 12.7
their place in describing land clutter from a variety of terrain types. (See
reference 1 for a description of these distributions and for further references. )
When the targets being sought by the radar are moving, doppler
information can also be used to separate them from the clutter. This is
complicated by the fact that many forms of clutter, such as the sea surface,
wind-blown vegetation, chaff and rain clouds, are themselves in motion, and
hence also give rise to doppler shifts. As a result, the clutter spectrum is
normally spread about zero doppler frequency. For most purposes, the
spectrum C (w ) can be treated as having a gaussian shape, i.e.
C(w) = C(o)ex p [ - ~ (:oYJ (5.2)
where Wo [rad s -1] is the transmitter frequency and a is a dimensionless

constant controlling the width of the spectrum. For a gaussian function
exp( -w 2/2(12), the RMS spread has the value (1, so that the RMS
spread of doppler frequencies corresponding to Eq. (5.2) is Wo I a. Since
doppler frequency is related to radial velocity by fd = 2v,I). [Hz] or
Wd = 2v,wo/e [rad S-1], the spread in doppler frequencies will be related to
the apparent RMS velocity spread (v rms ) of the clutter by the relation
(5.3 )
so that a = e/(2vrms). It can be seen that this clutter spectrum model predicts
that the spectrum width should be proportional to the operating frequency,
which is approximately true in practice. Typical values of V rms and the
corresponding doppler spread for a 1 GHz transmitter are given in Table 5.1.
5.3 MOVING-TARGET INDICATOR PROCESSING
Most radar systems need some form of doppler processing to filter out clutter
and thereby reveal faster-moving targets. These days, such filters are
implemented digitally, either as some form of fast Fourier transform (FFT)
algorithm or as a set of transversal filters.
In digital signal processing, the data I (t ) are sampled at discrete instants

separated by a fixed timestep At. This gives rise to a digital signal {I,}, where
I, = I (l .1t). Transversal filters simply take a weighted sum of blocks of
length M of the sampled data to give output
M-l
gp = L a,jp-I
1=0
Here the a, are the weighting constants. Digital approximations to, for
example, the matched filter described in Chapter 4 can be constructed in
this way. (There are many good books dealing with this topic; Stremler 2 is
an example.)
A digital approximation to the Fourier transform can also be constructed
in a similar way, using complex weights. This is the discrete Fourier transform
(DFT).1t is related to the sequence {I,}~(/ by the equation
M-l
Fp = L I, e - 21!jlp/M for 0:::::; p:::::; M - 1 (5.4 )
1=0
where we have followed the convention of using upper-case letters to represent

the Fourier transform. The importance of the DFT as an approximation to
the continuous Fourier transform stems largely from the fact that, if M is a
power of 2, there is a very efficient way to perform the M weighted
summations implied by Eq. (5.4). This implementation of the DFT is known
as the fast Fourier transform (FFT). (See, for example, Brigham 3 for a very
good treatment ofthe FFT and its use in signal processing.) It makes possible
efficient calculations of, for example, convolution or correlation, which can
be prohibitively expensive on computing time if carried out in the time
domain. The FFT was known to Gauss, but came to prominence when it
was rediscovered by J.W. Cooley and J.W. Tukey in 1965, and has since
revolutionized signal processing. There are now many ways of implementing
it, and many other algorithms that may be used to compute the doppler
spectrum (such as the fast Hartley transform), but the FFT remains the
backbone of modern doppler processing.
With these techniques, the modern radar design engineer can arrange
the signal processing to produce filters of the desired characteristics and
speed of operation. In the past, doppler processing was not easy to achieve.
Some systems made use of analogue bandpass filters to separate those signals
arriving at the carrier frequency (no doppler shift) from those displaced in
the spectrum by the target motion. By far the most common technique was
the moving-target indicator (MTI), still in use in many radars round the
world.
The principle of MTI is shown in Fig. 5.2. Each echo from a given range
gate is subtracted coherently from a delayed version of the previous echo
from that range gate. If nothing has changed, cancellation occurs, which
74 Tx I
Delay = 1 pulse
interval T
Delay = I pulse
interval T
Figure 5.2 Block diagram of MTI processing.
$
,---Simple canceller
~---->'...--- Improved canceller
~-------------=-"';::"'-Doppler frequency
Slow targets Fast targets
Figure 5.3 Clutter cancellation by MTI processing.
would be complete in the absence of noise. If the echo has changed phase
slightly (and amplitude strictly, but limiting amplifiers can be used to remove
the amplitude dependence) because of its motion, then cancellation will be
less complete. For a target in uniform motion there is a constant change in
phase from pulse to pulse and cancellation does not occur. The two-pulse
cancelling MTI is therefore acting as a high-pass filter, as shown in Fig. 5.3.
The impact of two-pulse cancellation on the clutter can be quantified
by the clutter attenuation factor
CA = input clutter power (5.5 )
output clutter power
We saw in Chapter 4 that noise power and noise variance are the same thing,
and that we can calculate noise power by integrating the power spectrum
(see Eq. (4.23». We also saw how to calculate the output spectrum given
the input spectrum and the system transfer function (see Eq. (4.28 Exactly ».
the same principles can be applied to clutter. This gives
CA = --:------=-J_C_(£0_)d_£O---=--_ (5.6 )
Jc(£O)IH(£OW d£O
where H (£0) is the system transfer function of the canceller. Since the output
of the canceller from an input f (t) is given by
g(t) = f(t) - f(t - T) = f(t)*[b(t) - b(t - T)] (5.7)
where Tis the pulse separation, its impulse response function is b(t) - b(t - T),
and hence
H(£O) = 1 - e- jwT = 2j e- jwT/ 2 sin(£OTj2) (5.8)
Using Eqs (5.2) and (5.8) in Eq. (5.6) gives, after simplification,
CA = _ _ _ _0_.5_ _ __ (5.9)
SIGNAL AND DATA PROCESSING III
(a)
~~ -r I pol~ del., ~ I po", d,I., C=JMTIOO'PO"
(b)
From
PSD
+-- Multiply by
coefficients
)---+ MTI output
Figure 5.4 (a) A three-pulse cancelling system. (b) A more general multipulse cancelling
scheme.
Normally (1)0 « C, so that the exponential term can be approximated by the

first two terms in its power series. This gives
2
CA _ C (5.10)
( )
2OJoTvrms
For a radar operating at 1 GHz, with a PRF of 1 kHz, the power in stable
clutter with Vrms = 1 m S-1 would be attenuated by a factor in excess of 570
(27.6 dB).
The performance of the two-pulse canceller can be improved by extending
the principle to a three-pulse canceller as shown in Fig. 5.4a, or to the more
general scheme of Fig. 5.4b. The attenuation of low-velocity clutter is
improved, as shown in Fig. 5.3. As Fig. 5.4b shows, multipulse cancelling
schemes are well on the way to becoming a modern digital transversal filter.
Historically, the time delay was implemented using an ultrasonic delay
line. A delay line is really only an analogue first-in, first-out memory device,
and, when digital electronics arrived, digital delays were introduced. Digital
delays permit increased flexibility in the PRF, which can be used to remove
blind speeds.
Blind speeds are caused by the same aliasing effect discussed in
Chapter 2. If the target speed is such that it moves a distance of ),./2 (or
n;.j2) between pulses, then it will appear to have stationary phase and will
be cancelled in the same way as zero-velocity clutter. Small changes in the
PRF alter the blind speed to prevent a target remaining undetected, but
require the MTI delay to be changed to keep in step.
The other problem of MTI, known as tangential fading, has no remedy

because it is something of an own-goal. When the path of the target carries
it tangentially through the radar beam, the radial component of its speed
goes through zero and the echo is filtered out by the signal processor.
On the whole MTI, and the modern transversal filter equivalents, are
very successful because the calculations are simple and sequential, and the
answer is available as soon as the last digit arrives. In contrast, FFT-based
processing req uires blocks of data to be collected before processing can begin.
Despite its simplicity, MTI processing needs to be carefully engineered.
Local oscillators must have a frequency stability and phase noise better than
the frequency separations that the MTI is seeking to impose. The system
must have sufficient mechanical stability to cause no detectable phase
changes; this may sound easy, but with a dish antenna rotating in a strong
wind it can be difficult to avoid detrimental vibrations. Rotating antennas
have the additional problem that, at angles off-bore-sight, there is con-
tamination of the doppler spectrum caused by the motion of the dish itself.
In many cases, the dish motion is the factor limiting MTI performance, and
is a further reason why modern radars are more and more tending to use
flat antenna arrays with electronic beam steering (see Chapter 15).
5.4 FAST FOURIER TRANSFORM PROCESSING
Transversal filters and MTIs are often modified to null out moving clutter,
such as rain clouds, with a clutter notch that can be controlled adaptively.
The whole of MTI thinking is geared towards removing clutter in this way,
so that a target is left competing only with system noise.
A different approach is to concentrate on target detection by developing
a special filter that allows only the target through. The target would then
be detected in the narrowest possible bandwidth, so minimizing the noise.
Unfortunately, until the target has been detected, we do not know its speed
and we cannot construct the filter. The solution to this dilemma is to use a
bank of filters, just overlapping, such that the target must appear in one of
them. The slow-speed filters will be contaminated with clutter, but we can
search through the remainder to find the target. The output of this filter bank
is known as the doppler spectrum (see Fig. 5.5).
There are many ways of constructing a filter bank, including analogue
methods, but these days the quickest way is to use a FFT algorithm. To
do this, a set of samples {u/} from a fixed range gate are gathered, using the
multiple pulses available within a single scan. The effect of the FFT is most
easily explained if we assume a non-fluctuating target with doppler frequency
Wd' We can therefore write the signal sequence as
(5.11 )
Frequency [Hz)
Figure 5.5 The output of the FFT can be thought of as a bank of bandpass filters.
where T is the time spacing of the samples. The amplitude of each sample
is the constant V, and for simplicity we have ignored the carrier frequency.
The DFT of this sequence (using Eq. (5.4)) is
M-l M-l
Up =V L exp(jwd/T) exp( -2nj/p/ M) = V L exp(j(wdT - 2np/ M )/]
1=0 1=0
(5.12)
Writing
z = exp[j(wdT- 2np/M)] (5.13 )
we can see that this is a geometric progression. Hence we can write down
the sum as
1- ZM
U =V-- (5.14 )
p 1- z
Since ZM = exp(jwdMT), the quantity on the top line of Eq. (5.14) is
constant; it does not depend on p. Hence as p varies, Up is maximized when
we make the bottom line of Eq. (5.14) as small as possible, i.e. set z = 1.
Using Eq. (5.13), this means that the maximum term in the FFToccurs when
(5.15)
From this, we can find the doppler frequency Wd'
Equation (5.15) implies that the possible solutions for the doppler
frequency are also digital (they are all multiples of 2n / MT). Hence there
may be some error introduced in the estimate of the doppler frequency, with
the correct value lying between the maximum and one of its neighbours. At
this point, you may object that this error is needed to stop the expression
in Eq. (5.14) becoming infinite when z is exactly 1. This does not happen,
because if z is exactly 1, the summation in Eq. (5.12) gives the value MV.
(You should check this.) So the maximum value in the FFT is MV. (It is
also worth noting, if the argument above makes you uneasy, that the limiting
value of Eq. (5.14) as p gets close to wdMT /2n is MV.)
To find the SNR improvement from this process, we can write the input as
I, = u, + n,
where u, and n, are the signal and noise terms in the input sequence. The
signal terms have the same form as above, and we assume that the noise
terms are all mean-zero, uncorrelated and with variance <1;. Taking the DFT
gives
M-l M-l
Fp= L u,exp(-2njlp/M)+ L n,exp(-2njlp/M) (5.16 )
'=0 '=0
The first summation is deterministic, and is given by Eq. (5.14), but the noise
summation is random. We need the average noise power, where the noise
term is given by the sum in Eq. (5.16), i.e. by
M-l
Np = L n,exp( -2njlp/M) (5.17 )
'=0
This requires a slight modification of the methods used in Chapter 4, since
the noise term here is complex. In this case, the noise pow,er is found by
averaging the squared modulus of N p , so that
Noise power = E[INpI2] = E[NpN;] (5.18)
Therefore
M-l M-l ]
Noise power = E [ ,~o n, exp (2njlp / M) k~o nk exp ( - 2njkp / M)
M-l ]
= E [ Ii.to n,nk exp[ -2njp(l- k)/ M]
M-l
= L E[n,nk] exp[ -2njp(l- k)/ M] (5.19)
k.'=O
Since the noise is uncorrelated,

if I = k
otherwise
Hence
M-l
Noise power = L <1; = M<1; (5.20)
'=0
The noise power does not depend on p, so is the same for each term in
the DFT sequence.
If we now look at the SNR before and after the DFT, we find that the
signal energy for each sample is V 2 , and the SNR is therefore given by
SNR (before) = V2/20';
At the maximum response of the OFT, however, the signal gives a value
of M V. Hence the SNR is
SNR(after) = M 2V 2/2MO'; = MV 2 /20';
Thus there is a gain in SNR by a factor M at the peak response of the 0 FT.
The analysis has only considered the case of uncorrelated noise. For
clutter, we would need to take account of the correlation of the clutter
samples (including the phase correlation, which gives rise to the doppler
spectrum of the clutter). This complicates the analysis, and is not carried
out here. However, in the range of doppler frequencies where the clutter
contribution is negligible and noise is the dominant factor, the analysis carried
out above is valid.
5.5 THRESHOLDING
Radar data processing has been described as 'the ruthless abandonment of

data '. Somehow the overwhelming radar data rate, which we calculated in
Chapter 2, must be reduced to a few numbers representing observations of
known and new targets. The key operation to achieve this data reduction
is the thresholding process, where the data are compared with a reference
level, and only those few signals exceeding this level are investigated further.
Early radars made use of voltage comparator devices in which the
reference voltage was under the control of the operator. When the voltage
exceeded the threshold, the data were displayed or recorded. A system using
an externally imposed reference is only as good as the speed and experience
of the operator, and it has the limitation that this threshold cannot be
adjusted continuously, as the antenna rotates, to compensate for variations
in the clutter and noise levels in different directions.
The introduction of digital processing permitted the reference level to
be generated internally from the observations themselves, thereby permitting
more sensitive and faster reacting thresholds to be used. There are two basic
steps to setting the threshold. First, the mean value of the noise or clutter
is found, when no targets are thought to be present. As described in Chapter
4, the threshold is then normally set at some mUltiple of this noise power.
This multiple is the factor k introduced in Eq. (4.44), and is known as the
(multiplicative) margin. Usually the margin is added to the mean in dB, i.e.
Threshold [dB W] = mean [dB W] + margin [dB] (5.21 )
(see Fig. 5.6). For envelope detection, a false-alarm probability of 10- 6
requires a value of k equal to 5.26 (see Sec. 4.6), so that the margin would
Signal
~
Q:l
~ Noise
t I-----~...------_f__+_---------_r- Threshold
~
~
- Mean
L -_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ~~--~~__+
Time/range/doppler
Figure S.6 The mean and margin used in a thresholding system.
be 7.2 dB. This addition of a margin in dB implies that, as the mean noise
level varies, the threshold also varies in such a way as to keep a constant
false-alarm rate, as long as the noise model described in Sec. 4.11 is valid.
In the case of some rarer noise models, a more sensitive detection system
can be designed (for the same false-alarm rate) by using other ways of creating
a threshold, such as the linear addition of a margin to the mean noise power.
There are only two main decisions to be made in thresholding: the first
is the choice of the basis on which to calculate the mean power, and the
second is the setting of the margin to achieve the correct sensitivity.
The mean power is calculated by finding the average power in reference
cells, which are near the test cell, but which are not so close that they are
contaminated by the presence of the target. The usual procedure is to leave
guard cells either side of the test cell, which are not included in the mean,
as shown in Fig. 5.7. The number of cells contributing to the mean, and the
width of the guard band, depend on how quickly the background is changing
and how much the target is spread across the cells. The method must therefore
be optimized for each radar system, but typical schemes are shown in Fig. 5.8.
So far we have not explained what we mean by a cell adjacent to the
test cell. It could be a radar resolution cell close to the test cell in range,
doppler, azimuth, elevation or even time (i.e. what happened in that
particular cell in previous scans). All of these parameters have been used at
one time or another, but one-dimensional thresholding in the range dimension,
or two-dimensional thresholding in the doppler/range plane are probably
the most commonly used schemes.
Having found a method of determining the mean, the next task is to
add a suitable margin. This is the problem we analysed in Chapter 4, where
we derived the relationship between SNR, the margin k and the false-alarm
and detection probabilities. We showed that, if the margin is set very low,
r---L----, Decision
T = test cell When clutter and other targets contaminate the

G = guard cell reference cells. some selection logic can be
R = reference cell introduced here to determine the true noise level
Figure 5.7 Guard cells G are used to prevent large signals in the test cell T from contaminating
the mean of the reference cells R.
Time
---+
Ranger
RR RR R RR RR R
RR R T RR R
RR RR R RR RR R
---+
Doppler
Figure 5.8 Various schemes for finding local reference cells to determine the mean level around
the test cell.
the probability of detection will be good (even faint target echoes are likely
to be detected) but the probability of false alarm will also be high because
a significant number of noise spikes will trigger the system (see Fig. 4.8).
Frequent false alarms can be very inconvenient if they trigger a defence
system, but there is an even greater danger. Too many false alarms will
swamp the tracker and cause the whole data processing computer to crash;
this could temporarily paralyse a defence system completely. As a result, low
false-alarm rates are used; the value for PCa of 10- 6 that we have adopted
throughout this book is comparatively large for many systems.
The solution to the problem of setting the margin correctly is to leave
the job to the computer. If there are too few threshold crossings, the margin
can be lowered to make the system more sensitive, so that the system is
triggered more frequently. If the number of threshold crossings increases too
far, exceeding a set limit, the margin is automatically increased to control
it. To the beginner this seems a rather empirical approach to the problem,
which leads to the probability of detection varying with the level of
clutter/noise/jamming. In practice, there really is no other way to proceed,
except possibly in the most advanced computer systems where more
processing power can be allocated to those areas where higher clutter levels
are experienced. Keeping the number of threshold crossings constant is
effectively the same as keeping the false-alarm rate constant (assuming that
very few of the crossings are due to real targets). Hence this technique has
become known as constant false-alarm rate (CF AR) processing and is widely
used in radar practice. If the false-alarm rate is kept constant and the clutter
statistics change, it is inevitable that this leads to changes in the probability
of detection. This is because changes in the clutter power are equivalent to
changes in SNR; Eqs (4.45) and (4.80) describe quantitatively how this
affects Pd.
The CF AR process can be improved to cope with contamination
of reference cells by clutter or other targets. Sometimes it is considered
worthwhile to make a second pass through the data after the largest signals
have been identified and removed by the first pass. This double thresholding
improves the sensitivity of the system, but at the cost of increased pro-
cessing effort.
5.6 PLOT EXTRACTION
Once a potential target, or plot, has been identified by a signal crossing the
threshold, the next step is to extract all the available information about it.
Azimuth and elevation information are usually extracted by monopulse
techniques, and the range by interpolation between range gates, as described
in Chapter 3. Similarly, the target speed is found by interpolation between
doppler cells.
The purpose of collecting this information at this stage is because a true
target will behave in a fairly predictable manner in each of the range/angle/
speed dimensions and will be said to 'track' correctly. Plots arising from
noise or clutter spikes (false alarms) usually behave randomly and do not
track; for example, the rate of change in range will not agree with the velocity
estimate obtained from the doppler information. However, this is not always
the case; on occasion, sea clutter spikes due to persistent waves may recur
sufficiently often for false tracks to be initiated.
5.7 PLOT-TRACK ASSOCIATION
When a radar returns to search a particular area, a number of plots will be

detected. These plots may form part of a known track in the track list, part
of a new emerging track, or be random false alarms. The first part of the
tracking operation is to sort the plots into these three categories, in a process
known as plot-track association, or association-correlation.
Plot-track association can be thought of in terms of the matrix shown
in Fig. 5.9. Each of the new plots is considered as a possible candidate for
association with each of the existing tracks. The matrix may be thinned
somewhat because, as Fig. 5.9 shows, some plots cannot be associated with
existing tracks because they are too far away. The remaining possibilities
are then considered by minimizing the function:
L (forecastposition)m - (actualposition)n
(5.22)
tracking error variance
where the sum is over all tracks m and all plots n. Some plots will associate
with known tracks and the remaining free plots are tested as potential new
tracks (see next section). Any remaining free plots, failing to form tracks,
are regarded as false alarms and are discarded at this stage.
There are more complications in plot-track association than might be
imagined. Figure 5.10a shows two tracks that are about to cross; does
Existing track 1
Association gates
Plot A Plot B Plot C •••

Track 1 ./ ,/
Existing track 2 X
Track 2 ./ ./ X
Figure 5.9 Associating plots to tracks.

(a) (b) (c)

Track 1
Track 1
~
~
Track 2 Track 2
Track 2
Figure S.10 There are potential complications when tracks appear to cross.
plot A go with track 1 and B with track 2? The answer depends on our
track model. If the two targets were ships, we would assume this association
to be correct, as they might be expected to diverge, as shown in Fig. 5.lOb;
but if the targets were aircraft at different heights, our belief might be that
they continued as straight tracks, as in Fig. 5.1Oc.
Other information is also contained within the track model. For example,
the plots should be distributed randomly either side of a straight track,
assuming the observations are corrupted by zero-mean gaussian noise. The
run sequence of '1 's (above the track) and 'O's (below) can be tested for
randomness and used to help decide which plot best fits with which track.
5.8 TRACK INITIATION
When a new sequence of plots appears on a radar display and begins to

form a straight line, or other recognizable trajectory, an operator is likely to
become increasingly convinced that a true target is present. How could this
process be automated? The answer is that there are almost as many schemes
as there are radar systems, and that every engineer has a favourite, but most
methods are based on simple up/down counting logic.
Imagine a surveillance radar with a rotating scanner. On the first
revolution a free plot is identified that could form part of a track and so an
up/down counter is incremented. On the next revolution the plot is there
again and the counter is incremented further. On the next scan the plot is
missing and the counter is decremented. This process continues until either
the count reaches zero and the track is deleted from the track list, or the
count reaches some predetermined value and the track is declared to be
genuine and the target existence is reported to the radar operators.
Simple track initiation schemes might use + 3 for a hit, - 2 for a miss
and declare the target to be present when the count exceeds 7. A weak, fading
target might therefore take some time to be detected as it approaches the
radar, as the following sequence shows:
Scan Hit/Miss Count Total
1 H +3 3
2 M -2 1
3 H +3 4
4 M -2 2
5 H +3 5
6 M -2 3
7 M -2 1
8 H +3 4
9 H +3 7
10 H +3 10 Track declared present
Some of the limitations of such simple schemes can be avoided by the Boolean
combination of rules such as 'seven out of ten scans must be hits AND two
of the first three must be hits'.
Track initiation schemes tend to be empirical and they turn out to be
quite hard to compare analytically, because of the Bayesian nature of the
problem. In the long term they may be replaced by artifical intelligence
methods of searching for patterns in the data (see Chapter 15).
How should you decide whether a plot is near enough to a track to be
included as a possible candidate for that track? The answer is shown in
Fig. 5.11. The first plot defines the position of the target, a civil airliner for
example, at a moment in time. Ten seconds later the radar returns to scan
the same sector, and the aircraft could be anywhere within a circle of radius
equal to the maximum possible speed of the aircraft times the 10 s interval,
giving a typical radius of 3 km. After the second plot has been identified
Figure 5.11 The plot-track association gate decreases with increased observations.
within this circle, there is some indication of the aircraft heading, so the next
association gate need not be so large. As Fig. 5.11 shows, the association
gates can eventually be reduced to a size determined by the accuracy of the
tracker. Usually, this gate size is about one standard deviation for a straight
track, but nearer 3 standard deviations for a manoeuvring target.
5.9 TRACKING
The actual process of tracking a target is composed of a track-smoothing

algorithm, identical to a conventional digital filter, together with track-
maintenance algorithms to cope with missing data or apparent branching
caused by multiple choices of plots.
The simplest smoothing algorithm is the (1.-fJ filter shown in Fig. 5.12.
The object of the algorithm is to find a compromise between the two extremes
of a track composed of measurements joined together (no smoothing, in
which case the target appears to move in a random zig-zag manner) and a
track drawn as a straight line that ignores the data points completely.
Radar plots are assumed to be related to the true target position by
Measured position = true position + plot noise (5.23 )
where the plot noise is assumed to have a zero mean and a constant noise
power of (12.
When tracking in cartesian coordinates, suppose the predicted next x
value of the target is xp and the measured position is x,,; then we need to
take some account of the difference by using (1. (x" - xp), where (1. is a
Measurement
~.
• • • ---_--~ Predicted position xp
Smoothed track Xn xp •
• •
• • Measured position x"
New smoothed position x, = xp + a(x" - x p )
a = 1 ~ Joins measurements
a=O •• • •• • •• • Ignores data
Figure S.12 The rx-p tracker.

number between 0 and 1. When a = 1 we are joining the dots, and when (X = 0
we are ignoring the data. In this way we can use (X to define the new smoothed
position of the target Xs to be
(5.24)
Likewise, the new velocity estimate can be found from
(5.25 )
where v = speed [m S-1] and T= time interval between observations [s].

Once the track has been updated by the addition of the new smoothed
positions, the new predictions are made for the next scan using
(5.26)
For a straight-line track, a tedious but uncomplicated proof (based on

minimizing the sum of squares of the differences between the track model
and the measurements) can be used to show that the optimal values of (X
and f3 are given by
2(2n - 1)
(X=---- (5.27)
n(n + 1)
f3 = 6 (5.28)
n(n + 1)
where n = number of observations.
Putting a few values of n into Eqs (5.27) and (5.28) shows that there is
almost no smoothing of the initial data, but the damping factor is pro-
gressively increased as the observations continue. This increased confidence
in the true track of the target means that the tracking error is reduced with
increasing time.
The error variance of the smoothed positions, which tells us the tracking
error, is given by
2 2(2n - 1) 2
(1 = (1 (5.29)
n n(n + 1)
The error variance of the predicted position, used to set the size of the
association gate, is given by
2(2n - 1)
(1;I(n- J) = (12 (5.30 )
(n-1)(n-2)
Two independent trackers are used in the x and y coordinates and the
algorithm can also be extended to track in acceleration in an (X - f3 - y
tracker. The algorithm may also be developed to take advantage of doppler

measurements of velocity.
The advantage of the a-f3 tracker is that it is very easy to implement.
The disadvantages are that it provides optimal smoothing only for linear
tracks, evenly spaced observations and gaussian noise.
A more general-purpose tracker is the Kalman-Bucy filter. An analogy
can be formed between tracking algorithms and the suspension on a motor
car, which attempts to smooth out the irregularities of a nominally straight
road. The a-f3 tracker learns how bumpy the road is and adjusts the damping
factor of the suspension accordingly. The Kalman-Bucy filter is more
sophisticated; it attempts to learn the 'pattern' in the sequence of bumps in
order to predict what will happen next. It then adjusts the suspension for
the predicted event.
Like the a-f3 tracker, the Kalman-Bucy filter relies on minimizing the
square difference between the track model and the measurements. At each
step, the best estimate of the whole track up to that time is made, using all
previous observations and the evolving track model. The prediction of the
parameters (position, velocity and perhaps acceleration) of the next position
along the track uses all the available information, with more recent
measurements weighted more heavily. This leads to a superficially compli-
cated algorithm (detailed descriptions of Kalman filtering can be depressing).
However, the beauty of this algorithm is that it can be applied recursively,
i.e. the solution to the minimization problem at each step makes as much
use as possible of the solution at the previous step. This continual updating
provides a numerically stable and computationally efficient means of pre-
dicting the evolving track, in a manner consistent with all the available
information.
Even after tracking has been successfully initiated, there are still a number
of things that can go wrong, and track-maintenance algorithms are required
to ensure that the target is followed successfully. One of the commonest
problems is when an aircraft, which has previously been flying in a straight
line, suddenly makes a manoeuvre. This must be both detected and allowed
for in the tracker.
One method of detecting manoeuvres is the integrated bias detector,
shown in Fig. 5.13. If the straight-track model is correct, the measurements
will fall equally on either side, and so a running summation of the errors will
be roughly zero (when the sign of the error is included). If the target begins
to turn, then the measurement errors will quickly build up on one side of
the track, and when the total exceeds some predefined level, a manoeuvre
is declared to have been detected. In an a-f3 filter the values of a and f3 can
be increased to improve the ability of the tracker to follow the target (but
with less accuracy). With a Kalman - Bucy filter the procedure is usually to
add plant noise to the filter (increasing elements in the predicted error
covariance matrix) to achieve the same effect. This reflects the increased
Actual
target track
Measurements
/~
...L.._..,...._.%...."""T'"_...L.._~_"T'"""_~:........:L---L---L...J.._ _ Forecast track
Sliding average Average begins to become

near zero significantly greater than zero
Figure 5.13 The integrated bias detector.
uncertainty (and hence error variance) in the track model when the larget
is manoeuvring.
Another problem is to decide what to do if the track appears to branch.
Track branching occurs when there are so many false plots that the true one
cannot be identified with certainty. Some systems attempt to cope with this
problem by tracking both branches for a while and then trying to select the
most promising looking branch; this is not unlike the Viterbi algorithm used
in digital communications to find the decoding route through the trellises
used in convolution encoding. Another approach is to use probabilistic data
association (PDA), formulated originally by Bar-Shalom and Tse 4 and
extended by others to take account of multiple targets, manoeuvres, track
initiation, etc. (see e.g. Colgrave et al. 5 ). PDA builds on the Kalman- Bucy
filter to use an average of all the validated plots, weighted by the probability
of their having originated from the target. Although PDA introduces
inaccuracies from the (weighted) inclusion of incorrect plots, it avoids the
risks of an increasing number of track branches eventually jamming the
computer and causing a system crash.
If the target track enters a partic~larly dense area of clutter or jamming,
it may not be possible for the software to maintain the track and operator
intervention may be necessary. Part of the effectiveness of radar systems is
determined by the skill, intelligence and experience of operators. In future,
it may be possible to use expert systems to replicate the local knowledge of
experienced operators, but it is unlikely that high-level decisions in radar
(e.g. whether to fire on a target) will be taken without .human control, or at
least a human right of intervention.
5.10 SUMMARY
The target detections that initiate a track must take advantage of whatever
gains are possible in SNR, because viable radar systems must operate at
very low false-alarm rates. Making use of doppler information is a crucial

way of increasing the SNR. Once a track has been initiated, a variety of
methods of increasing sophistication are possible to retain the track. All of
them run into trouble as tracks become more complicated and the radar
environment becomes more congested.
Key equations
• The Rayleigh clutter model:

C
pAc) = -exp - -)
(1;
(_c2(1;
2
• The gaussian clutter spectrum:
C(w) = C(O) ex p [ - ~ (:oY ]

where a = 2c/vrms •
• The clutter attenuation factor:
CA _ ( 2c )2
woTvrms
• The equations for a.-fJ tracking:

Xs = xp + a.(xn - xp)
fJ
Vn = Vn - 1 + T (xn - xp)
2(2n - 1)
a.=----
n(n + 1)
fJ = 6
n(n + 1)
where n = number of observations.
5.11 REFERENCES
1. Introduction to Radar Systems, M. I. Skolnik, McGraw-HilI, New York, 1981.

2. Introduction to Communication Systems, F. G. Stremler, Addison-Wesley, Reading, MA, 1982.
3. The Fast Fourier Transform, E. O. Brigham, Prentice-Hall, Englewood Cliffs, NJ, 1974.
4. Tracking in a cluttered environment with probabilistic data association, Y. Bar-Shalom and
E. Tse, Automatica, 11,451-461,1975.
5. Track initiation and nearest neighbour incorporated into probabilistic data association,
S. B. Colgrave, A. W. Davis and J. K. Ayliffe, IE Aust. IREE Aust., 6(3),191-198, 1986.
5.12 PROBLEMS
5.1 If the signal and noise were roughly equal at the output of a radar receiver, what SNR
would you expect after averaging four 32-point FFT blocks of observations? Assume the target
RCS and speed to be constant. What would happen to the SNR if the target began to manoeuvre?
5.2 A surface-wave HF radar uses a PRF of 275 Hz and observes an aircraft for 3.72 s. What
is the potential coherent gain and how would this improve the range estimate? These radars
use vertical polarization (see Chapter 10). Would the target remain visible through a manoeuvre?
5.3 A radar system is required to have a high probability of detection of95 per cent on envelope
detection of a single pulse. This Pd is to be achieved at the expense of a high false-alarm rate of
10- 5 • What margin should be chosen for the thresholding and what.SNR is required?
5.4 The following series of x-coordinate position measurements of a target were made at the
rate of one per second:
1 2 3 4 5
x. o 35 88 118 158
Tabulate the values of IX, p, Xs and v. for the IX-P tracker and predict the x value for the next
position.
5.5 In problem 5.4, what improvement in the target position results from the track smoothing?
Suggest a size for the association gate in the x dimension for the sixth ·observation.
CHAPTER
SIX
DESIGNING RADAR WAVEFORMS
• How accurately can we measure range and velocity?

• Can we distinguish closely spaced targets?
• How much freedom do we have to improve radar performance?
How do we choose waveforms that tell us as much as possible about the

properties of the target?
6.1 INTRODUCTION
In Chapter 4, we established that the correlation receiver (or matched filter)

provides the optimum method of detecting a stationary target. Detectability
is unaffected by the form of the transmitted pulse. All that matters is the
ratio of the signal energy E to noise power per unit bandwidth N, both
calculated at the input to the receiver. The signal-to-noise ratio at the output
of the correlation receiver is then given by E/ N. However, there are other
important aspects of system performance for which the signal shape does
become important. These include the resolution, ambiguity and accuracyt of
the measurements made by the radar.
The resolution of a radar system is a measure of its ability to separate
closely spaced targets in range or velocity. These can be treated separately,
but simultaneous measurement of both range and velocity is often required.
In this case there is an unavoidable uncertainty, so that two targets at different
ranges and velocities may in principle and practice be indistinguishable.
t Throughout we assume no bias in the measurement, so that accuracy and precision will
be equivalent.
128
DESIGNING RADAR WAVEFORMS 129
Ambiguity occurs if the output of the receiver from a single target contains
multiple peaks that can be mistaken for other targets. Such peaks may be
caused by noise, but may also be produced by the shape of the transmitted
waveform. An obvious example is if there are significant sidelobes in the
radar antenna pattern. Objects illuminated by the sidelobes will be interpreted
as though they were in the main lobe. This will lead to angular positioning
errors or mUltiple detections generated by a single target. It can also cause
clutter to be interpreted as target.
Accuracy refers to the expected spread of measurements about the true
value. We have seen that detection and ranging are essentially the same
problem. The range of a target depends on the time delay at a peak in the
output from the matched filter, as long as this peak exceeds some threshold.
Noise in the output will cause the peak to be displaced randomly from its
true position. The standard deviation of this variation will be our adopted
measure of range accuracy. We adopt a similar definition for the accuracy
of measurements of doppler frequency.
6.2 BANDWIDTH AND PULSE DURATION
All these aspects of system behaviour are affected by the radar bandwidth,
whose definition we need to make more precise. Several different definitions
of this important concept are in use. The simplest definition is applicable if
the signal u (t) is band-limited, i.e. it has a Fourier transform U (w) for which
U (w) = 0 if w > Q rad s - 1. Then it is natural to take the bandwidth as
Q rad s - 1, or Qj2n Hz. None of the commonly used pulse modulations satisfy
this relation exactly, but it is of considerable theoretical value because of its
relation to the Shannon-Whittaker sampling theorem t (band-limited signals
can be reconstructed exactly from their samples as long as the sampling
frequency exceeds the Nyquist rate of Qjn samples per second.) In practice,
many real signals are deliberately band-limited by the use of anti-aliasing
filters.
The most commonly used engineering definition of bandwidth is the 3 dB
width, which is the separation in frequency of the half-power points in the
energy spectrum of u (t), i.e. it is obtained by solving
IU(wW = !IU(OW (6.1 )
Clearly this definition is only applicable if 1U (w)1 has a maximum at w = 0,

which is the case for most commonly used pulseshapes.
From a theoretical point of view, perhaps the most useful definition is
t This chapter uses a number of results from Fourier transform theory. A good reference is
Bracewell l .
the effective bandwidth a ()) used by Woodward 2 • It is given by the formula
2 f ~oo (Ill U (w Wdw toooo w 2

1 U (w Wdw
a = -----,,---------- (6.2 )
21tE
()) tOOooIU(WWdW
J
This definition assumes that rolU( ro )l2dro == 0, which can always be
arranged. (For real signals, it will always be the case, since then IU (w)1 is
an even function of w; more generally, since u (t) is in fact the modulation
of a carrier frequency, it will arise by defining a suitable centre frequency
for the modulated signal.) This means that a;
is the second moment of the
J
Fourier transform of the unit energy signal u (t )/ E about its mean, 0. It
is therefore a measure of the spread of energy in the spectrum of the signal,
and is analogous to the variance in probability theory or moment of inertia
in mechanics. For calculation purposes, it can be very convenient to use the
Fourier transform pair
u' (t) = du/dt +-+ jwU (w)
Then using Parseval's theorem, we can write
2 toooo \u'(tW dt energy in u' ( t )

a ()) =------
E (6.3 )
energy in u( t)
There are analogous definitions of pulse duration. For time-limited
pulses, such as a rectangular or raised cosine pulse (see Table 6.1 band d),
we can take the actual pulse length. The 3 dB duration is frequently used,
and for theoretical purposes we can use the effective pulse duration a" defined
by
f~oo IU'(wW dw
(6.4 )
21tE
In Eq. (6.4) we have assumed that the centroid of the pulse is at 0, i.e.
Jtlu(tW dt = 0, and we have used the Fourier transform pair
tu(t)+-+jU'(w)
Though the definition of effective bandwidth and pulse duration may
seem unnecessarily complicated, they are of considerable value because they
can be handled analytically. From them, we can derive a number of
fundamental relations between accuracy, resolution and ambiguity that
clarify our understanding of the inherent limitations of any radar system,
even though we may in practice use simpler definitions. Hence we adopt

these definitions in the following sections.
The value of E, U W and U t for some basic pulseshapes are given in
Table 6.1. In this table the values for the rectangular and triangular pulses
(b and c) are found from the values of the trapezoidal pulse (a) by setting
a = T and a = 0 respectively. Notice that, as the trapezoidal pulse becomes
more like the rectangular pulse, the required effective bandwidth increases
without limit. This reflects the fact that instantaneous jumps in signal value
are not physically possible. However, because of the great value of idealized
rectangular pulses in discussing system performance, it is common to assign
a less stringent definition of bandwidth to this pulse. Often this is taken as
the distance to the first null in its Fourier transform. For the rectangular
pulse shown in Table 6.1, this gives a bandwidth of 11: IT rad s - 1.
A fundamental property of the bandwidth and duration of a pulse arises
from the scaling relation of Fourier transforms. This states that, for any
Fourier transform pair and non-zero constant A, f (t ) +-> F (0)) implies
[f(At)+-> (l/IAI )F(O)I A)]
Since f(At) is a pulse of the same shape as f(t), but reduced in width
by a factor 1I A, the scaling relation says that, for any sensible definition of
bandwidth and pulse duration, decreasing the duration of a pulse by a factor
A increases its bandwidth by the same factor, and vice versa. So, for pulses
of the same shape, the time x bandwidth product is constant. For many
simple pulses, this product is of the order 1 (see items c, d and e in
Table 6.1). This is one of the many rules of thumb employed by radar
engineers, but must be used cautiously, as it depends on the definition
of bandwidth. Often a definition is used in which it becomes true, as we
have just seen for the rectangular pulse. However, for more complex
pulses, this relation is far from the case. Indeed, pulses using frequency
modulation are designed to give a large time x bandwidth product, for
reasons explained below.
6.3 RANGE AND DOPPLER ACCURACY-THE

UNCERTAINTY RELATION
We saw in Chapter 4 that the output of the matched filter consists of two
terms, one due to the correlation of the signal with itself, and a noise term
due to the correlation of the signal with the noise. In the absence of noise,
the peak of the signal term would give the true time delay corresponding to
the target's range. However, the noise term will cause this peak to move
around. In fact, if the gradient of the noise term is not zero at the correct
time delay, the peak will be displaced. As a result, the measured value of
the time delay of the peak can take a range of values. Woodward and Davies 3
c: Table 6.1 Fourier transform pairs and values of E, (1", and (1t for some basic pulseshapes
N
Pulseshape Fourier transform pairs E (12 ., (1'f
(a)
3 T3 + 2aT2 + 3a 2T + 4a 3
(T+a) sa(T+a)~) sa(T-a)~) t(2a + T)
(2a + T)(T - a) IO(2a + T)
(b)
2T Sa(Tw) 2T 00 T2/3
(c)
T Sa 2 (Tw/2) 2T/3 3/T 2 T 2 /IO
T
Cd)
cos 2 (Trtl2 n
T 3T n2 2T2 (n2/2 _ 3)
1 _ (Tw/n)2 Sa(Tw) 4 3T 2 3n 2
(e)
exp (-t 2I2T 2 )
J(2n)Texp[ - (TW)2/2] TJn 1/(2T2) T2/2
~
-
succeeded in showing that for large SNR the PDF of the measured delay
would be approximately gaussian with a mean value occurring at the true
delay and with standard deviation
1
J'td=
21turo
J (2E/ N) [s] (6.5)
A related result, due to Manasse\ gives the standard deviation of measure-

ments of doppler frequency as
1
[Hz] (6.6)
JWd = 21tut J(2E/N)
Hence the accuracy of either type of measurement increases with increasing
values of the SNR, E/ Nt. Good range accuracy is not dependent on short
pulse lengths, but needs large bandwidth; good doppler accuracy needs long
pulses.
An immediate question is whether we can simultaneously obtain good
range and doppler accuracy. Any discussion of this must take account of a
fundamental property of Fourier transform pairs: if u (t ) +-+ U (w ) then
(6.7)
This relation leads to the famous Heisenberg uncertainty principle of
quantum mechanics. Applying it to the measure of accuracy given by
Eqs (6.5) and (6.6), we find that
1 1 1 1 1
J'td JWd = - - - - - ~ - - - (6.8)
81t 2 E/ N UroU t 41t2 E/ N
Hence the product of the accuracies in range and velocity is inversely
proportional to the signal-to-noise ratio. This has the important interpre-
tation that both accuracies can be improved simultaneously and without
limit by increasing the SNR. For fixed SNR, the product of the accuracies
can be decreased by increasing the time x bandwidth product UwU t • This can
be achieved by appropriate choice of waveform (see Sec. 6.8).
6.4 RESOLUTION
Accuracy tells us how reliably we can measure the parameters of a single

target, but it tells us nothing about our ability to recognize that there are
two targets present if their velocities and/or ranges are similar. For this we
need the notion of resolution. We have seen in Sec. 4.7 that, after matched
t These two equations have already been encountered as the range error and doppler
frequency error (in Hz) of Eqs (1.19) and (1.22).
filtering, a single stationary target would give rise to an output

Aop,,(t - '0) + m(t)
where p,,(t) is the ACF of the transmitted waveform u(t), and Ao and '0
describe the amplitude and delay effects due to propagation and the target's
RCS. If there were two targets present, the linearity of the system would
cause the output to be
Aop,,(t - '0) + A 1P,,(t - ,d + m(t)
Figure 6.1 shows the output for two targets with the same RCS for various
values ofro - '1' and a signal of the form u (t) = exp ( - t 2 ). It is clear that
the effect of having closely spaced returns is to smear the peak response. For
'0 close to '1, it is difficult to decide whether there is a single or a double
peak present in the noisy output. (In fact, for closely spaced targets there
will only be a single peak since, near its maximum, each of the two signal
terms can be approximated as a quadratic curve. The sum of quadratic curves
is again quadratic.) The resolution of the system is a measure of how large
the time difference '0 - '1 must be before the signal terms give rise to two
distinguishable peaks, or meet some other condition by which we can decide
if there is more than one target present.
Just as in the case of accuracy, there is no unique criterion by which we
can define resolution, since it hinges on what is meant by distinguishability.
Many different measures of resolution are therefore in use. Unlike measures
of accuracy, they are independent of the noise, but are all defined in terms
of properties of the waveform for which we are trying to tell whether there
is a single copy present, or two closely spaced copies.
This waveform, after matched filtering, is the ACF p,,(t). The ACFs of
radar signals have certain general properties, which are useful in analysing
the properties of resolution. These include:
1. E = Ip,,(O)1 ~ Ip,,(t)1 for all t, i.e. the ACF is maximum at zero lag, and
its value there is the signal energy.
2. For finite energy pulses, p" (t ) will tend to 0 as t increases, though there
may be subpeaks (ambiguity peaks) in addition to the peak at O.
3. p" (t) +-+ IV (w W, i.e. the Fourier transform of the ACF is the energy
spectrum of the waveform.
These properties of the ACF have led to such definitions of time resolution
as the 3 dB width of p" (t), or the time over which the ACF drops to a value
(1 Ie )p,,(O), or the time to the first zero of the ACF. These have their uses
for particular types of ACF. A measure of resolution that is more general
and is better for analysis is
A, = JIp,,(tW dt [s] (6.9)
d p;(O)
1.0
0.8 TO = 0
T[ = 0
0 1.5 2.0
1.0
0.8 TO =0
T[ = 0.25
0.2
0 2.0
1.0
0.8 TO =0
0.6 T[ = 0.50
1.0
0.8 TO = 0
0.6
T[ = 0.75
0 1.5 2.0
1.0
0.8 TO =0
0.6 T[ = 1.0
Figure 6.1 The output from the matched filter when two targets with range delays '0 and, \
are present, for different values of '0 - '\. Here '0 is set to 0 and, 1 is marked on each successive
plot.
-l
I
I
I
I
I
I
I
I
I
I
I
I
I
0.4
-4 4
Figure 6.2 The relation of the equivalent-rectangle resolution to the area under the graph of
Ip.(tW and its value at O.
This definition has a very simple interpretation (see Fig. 6.2). It is the width
of a rectangle of height p;(O) that has the same area as is under the curve
IPu(t W. Hence it is sometimes called the equivalent-rectangle resolution.
Using properties 1 and 3 of ACFs and Parseval's theorem, Eq. (6.9) may
also be written as
[s] (6.10)
This form of the definition also has a useful and instructive interpretation.
If U (co) took only the values 1 or 0, and the total length of the intervals
in which it took the value 1 was F, then F could be considered as a measure
of occupied bandwidth, or what Woodward 2 calls the frequency span. The
energy of this signal would be F12n, and hence for this signal d'l'd = 2n /F.
This expresses the reciprocal relationship between resolution and bandwidth.
Similar definitions for the frequency resolution are possible. We first
need the frequency-domain A CF
pu(co) = f U(v+co)U*(v)dv (6.11)

Then the frequency resolution is

L\r =JIPu(wWdw =2nJlu(tW dt (6.12 )
w pt(O) E2
6.5 THE AMBIGUITY FUNCTION
Equations (6.9) and (6.10) apply when the targets to be resolved are known
to be stationary, while Eq. (6.12) applies when targets are at the same range.
However, if we do not have such prior knowledge of the target character-
istics then we need to worry about the combined effect of a shift in range
and frequency. This can be analysed by considering the behaviour of the
output of the correlation receiver when the input is doppler-shifted. (This
means that the filter is not properly matched to the incoming signal, since
it does not replicate its frequency behaviour.) If the transmitted signal is a
complex modulation u (t) of a carrier frequency Wo rad s - 1, and the target
velocity causes a doppler shift in frequency of Wd rad s - 1, then the correlation
operation will give as output at time t
fOoo u(s) exp[j(wo + wd)s]u*(s - t) exp[ -jwo(s - t)] ds (6.13)
which can be rearranged as
exp(jwot) too", u(s)u*(s - t) exp(jwds) ds (6.14)
The essential information about resolution is carried by the integral term in

Eq. (6.14). Normalizing to a unit energy waveform u(t)/JE, this gives the
quantity
X(t, wd) = -1 foo u(s)u*(s - t) exp(jwdS) ds (6.15 )

E -00
which is known as the ambiguity function of the transmitted waveform. It has

an equivalent frequency-domain expression
X(t,w d ) = - 1 foo
U*(v)U(v - wd)exp(jvt)dv (6.16 )
2nE - 00
The cuts across the ambiguity function along the delay (t) and doppler (Wd)
axes are directly related to the time-domain and frequency-domain ACFs (and
hence to the time and doppler resolutions), since
x(t,O) = Pu(t)/E (6.17 )
and
(6.18 )
To understand the role of the ambiguity function in a discussion of

resolution, we need to consider the combined signal when there are multiple
targets present. A target whose response (including doppler shift) is perfectly
matched to the receiver will give rise to the output signal X(t,O). (For
simplicity, we ignore amplitude effects on the pulse, and assume that the
range delay of the returned pulse corresponds to time 0.) Another target of
the same ReS but whose range delay and doppler frequency differ from the
first target by 'td and Wd will give an output X(t - 'td , Wd). Unless X(t, 0) and
X(t - r d , w d ) are significantly different, the two targets will be difficult to
separate. In particular, if X(O, 0) is close to X( - rd, Wd), it will be hard to
recognize the presence of more than one target.
The properties of the ambiguity function will be clearer if we discuss an
example. A single rectangular pulse of width T (see Table 6.1h) has an
ambiguity function for which
if It I > T
and
1
2 SIn
Ix(t, wd)1 = Wd 2 (T -Itl) )1 T
. (Wd
1
if It I ~ T (6.19 )
where Sa(x) = sin(x)/x. A sketch of Ix(t,wd)1 is given in Fig. 6.3, and

Fig. 6.4 shows the cuts along the delay and doppler frequency axes. The
important features of this ambiguity function are a central peak of height 1
at (0,0), which falls off linearly as It I increases and behaves as a sin (x)/x
function as IWdl increases, surrounded by numerous subsidiary peaks. The
way to interpret this figure is that two targets that differ in delay by t and
in doppler frequency by Wd will give nearly the same response and hence
will be hard to resolve if Ix(t, wd)1 is nearly equal to Ix(O, 0)1 = 1.
Note that there are two conceptually different aspects of resolution
involved here. Small values of t or Wd correspond to points in the central
peak that are not far removed from the true values for range and velocity.
However, pairs (t, Wd) lying on one of the subsidiary peaks may be far
removed from the correct values of delay and doppler frequency while still
giving a comparable response. These correspond to genuine ambiguities in
interpretation of the response. Both aspects of resolution are wrapped up in
the measures of resolution given by Eqs (6.9) and (6.12).
~
6
J: 0
:=
-3
-6 ~
Figure 6.3 The modulus of the ambiguity function of a rectangular pulse (T = 1.0).
(a)
1.0
X 0.5
(b)
whr
-0.5
Figure 6.4 Cuts along (a) the delay and (b) the doppler axes of the ambiguity function plot
of Fig. 6.3.
These general comments can be made more specific when we investigate

the analytic properties of the ambiguity function. It is easy to see that
X(O, 0) = 1 (6.20)
and (using a change of variables in the integral) that the ambiguity function
has a rotational symmetry, expressed as
\X(t, Wd)\ = \X( -t, -wd)1 (6.21)
Using the Cauchy-Schwartz inequality (Eq. (4.47» we find that the modulus
of the ambiguity function has its maximum at (0,0), since
Ix(t, wdW = ;21 f:oo u(s)u*(s - t) exp(jWdS) dsl2
1 foo
~E2 _oolu(sWds foo
_oolu(s-t)exp(-jwdsWds
1 foo 2 foo
= E2 -00 lu(s)1 ds -00 lu(s - tW ds (6.22)
Each of the integrals on the right-hand side is equal to the energy of the
signal, so that
(6.23 )
Perhaps the most remarkable fact about the ambiguity function is that, if
we view \X(t, wd)1 2 as a surface, then the volume under this surface is always
1, irrespective of the shape of the waveform, i.e.
(6.24 )
This equation has very important consequences. It tells us that, while we

are free to design the waveform u (t) to give a very sharp peak in the ambiguity
function at (0, 0), the fixed height of this peak (Eq. (6.20» implies that the
volume of the central peak will then be only a small fraction of 1. Hence
more ambiguity must appear away from the central peak, and targets well
separated in range and velocity may become indistinguishable.
The squared ambiguity Ix(t, WdW may be thought of as a quantity of
sand of total volume 1. The system designer is free to distribute the sand as
he or she sees fit in the (t, Wd) plane as long as the height at (0,0) takes the
value 1 and the symmetry condition of Eq. (6.21) is met. However, all the
sand must be used, so that a narrow peak near (0,0) can only be achieved
by greater ambiguity further from the origin. As long as the ambiguities can
be moved to regions of the (t, Wd) plane where there is reason to believe
that targets cannot be present, the limitations imposed by the ambiguity
diagram can be escaped. As always, prior knowledge of what is expected
can be used to improve performance.
6.6 EXAMPLES OF THE AMBIGUITY FUNCTION
As other examples of types of ambiguities that can occur, we now consider

two waveforms, the gaussian pulse and a repeated rectangular pulse. More
examples will be found in the later sections of this chapter.
The gaussian pulse
has ambiguity function

X(t, Wd) = eXPUwdt/2) exp( -t 2/4T 2 ) exp( - T 2 wV4) (6.25 )
so that 1X(t, W d ) 12 is constant on the elliptical curves
2
- t + -W~- = constant (6.26)
4T2 4/T2
A contour plot of this ambiguity function is shown in Fig. 6.5. There are no
subpeaks, and all the 'ambiguity' arises from the spread ofthe central peak.
We now deal with a coherent pulsetrain containing M identical rect-
angular pulses of width T each separated by 1'., where 1'. > 2T (see
Fig. 6.6). The ambiguity function for this waveform is comparatively
2.5
2.0
1.5
1.0
0.5
-; 0.0
-0.5
-1.0
-1.5
-2.0
-2.5
-10 10
Figure 6.5 A contour plot of the ambiguity function of a gaussian pulse (T = 2.0).
M
, -_______________________ A~ ______________________ ~
o 0 0 0····0
Tr
Figure 6.6 A coherent pulsetrain with PRF T,. containing M pulses of width T.
complicated, and has the form
- -
IXM (t ,Wd )1 - 1 M~l ISin[T.(M -IPI)W d /2]\1 Xl (t
L...
- T.
P r,Wd
)1
M p=-(M-I) sm(T.wd/ 2 )
(6.27)
where XI (t, w d ) is the ambiguity function of a single pulse, given by
Eq. (6.19). A contour plot for the case M = 3 is given as Fig. 6.7, and has the
'bed of nails' structure characteristic of the ambiguity diagrams of pulsetrains.
The system designer can control both T and T. in order to ensure that targets
of interest only occur near the central peak of this ambiguity diagram, giving
enhanced range and doppler accuracy and effective resolution.
The cuts along the delay and doppler axes are given in Fig. 6.8. Along
the delay axis Eq. (6.27) takes the form
IXM(t,O)1 = II (l_lfl)(l_lt - PT.I)

p=-(M-l) M T
(6.28)
if It - pT.1 < T, and is zero everywhere else. The term in the second
parentheses corresponds to a triangular peak of base width 2T, with its centre
at PT.. The term in the first parentheses corresponds to a triangular weighting.
This is shown in Fig. 6.8a. The triangles are spaced at intervals T., and the
condition T. > 2T is to prevent these triangles overlapping; if they do so,
the expression in Eq. (6.28) becomes more complicated.
Along the doppler axis, Eq. (6.27) has the form
IXM(O, wd)1 = -1 I Sa ( TWd/ 2 )1 ISin(MT.W d /2)1

- .~--=---=:'--..:. (6.29)
M sm(T.wd /2)
Since MT. > T, the Sa term varies less rapidly than the second term (which
is periodic with period 27r1T;), so that we can regard Fig. 6.8b as a modulation
of the second term by Sa(Twd/2).
3r-----------rn~.----.~~----_rrrrTr----------_,
0 10 000 0 10
2
00 88 00
CJ CJ
(0) CgJ
C@) C@)
) 0
~ @j)
C@) C@)
-1 (Q) (Q)
C) o
-2 00 00
000 000
Figure 6.7 A contour plot of the ambiguity function for the pulsetrain shown in Fig. 6.6
(M = 3, T = 1.0, T, = 4.0).
(a)
(b)
wl-rr
of Fig. 6.7.
6.7 PULSE COMPRESSION
The radar designer's ideal waveform would give good performance as regards
the following:
• Target detection
• Range and doppler accuracy
• Range and doppler resolution
These requirements appear to be incompatible. Detectability of targets

is dependent on the total energy of the illuminating waveform, which is
processed by the matched filter. Good range accuracy and resolution require
large energy and high bandwidth, which, for simple pulses, implies short
pulses. Short-duration and large-energy pulses require very large peak
powers, which may not be available, and short pulses imply poor doppler
resolution. Fortunately, there is an escape from these apparently contra-
dictory requirements and the need for large peak powers, by using more
sophisticated waveforms.
An essential ingredient of such waveforms is the use of long pulses
requiring a reasonable peak power to obtain good doppler resolution and
the high energy needed for good detection performance. Range resolution
is obtained by designing the waveform shape to give high bandwidth,
normally by using frequency modulation. Because the matched filtering
'compresses' the long pulse to an ACF of short duration at the output of
the receiver, this type of radar processing design is known as pulse
compression. Both range and doppler accuracies are good for this form of
processing, since the requirements of large bandwidth and pulse duration
are both met. The question of resolution is more complicated. If targets are
known to be at the same range, then there is good velocity resolution.
Similarly, if targets are known to have the same velocity, then there is good
range discrimination. If there is no such a priori information, and we need
to separate targets that are of unknown relative range and velocity, then we
cannot escape the mandates of the ambiguity diagram. In order to illustrate
this, we consider the effect of linear frequency modulation (FM).
6.8 CHIRP
The simplest form offrequency modulation is linear FM, i.e. a pulse described
by
u(t) = a(t)exp(jnkt 2 ) (6.30)
where a (t) is a pure amplitude modulation. The instantaneous frequency is

found by differentiating the phase, to give

w(t) = 2nkt [rad S-I] (6.31 )
which is clearly linear. The total frequency deviation during the pulse (which
we can consider as a reasonable approximation to the bandwidth) is therefore
2nkT. Plots of the in-phase part of the signal and frequency as a function
of time for a rectangular pulse with linear FM are shown in Fig. 6.9. This
form of pulse is popularly known as chirp because of the sound made by a
signal of this type at audio frequencies. Lightning strikes also give rise to
chirp signals at radio frequencies. Emissions from a lightning strike begin
as a very compressed pulse. Different frequencies propagate at slightly
different speeds (a phenomenon known as dispersion), so that at great
distances a long falling tone is heard on a radio receiver. Pulse compression
is analogous to receiving this signal and correcting for the dispersion, in
order to recover the pulse generated by the lightning strike.
A chirp signal has an ambiguity function of the form
Ix(t, wd)l = (
1 _!!l)
T
sa(T-ltl)(Wd +
2
2nkt))1 ifltl ~ T
(6.32 )
1
1o otherwise
The shape of the ambiguity function is not apparent from this equation, but
contours are plotted in Fig. 6.10, and cuts along the delay and doppler axes
are shown in Fig. 6.11. The most obvious feature of the ambiguity diagram
(a)
v
~~
{\ .75-
.50r
0.25r
n
H~:25 Vi
I I J J
4 - f--
2 3 4
t
0.50 r
-0.75 t-
V 1.00 ~ V
(b)
w
-4 4
Figure 6.9 (a) The in-phase component of a linearFM signal and (b) the instantaneous
frequency of this signal as a function of time.
16
12
) 0
-4
-8
-12
Figure 6.10 A contour plot of the ambiguity function of a linear FM pulse (T = 1.0, k = 5).
(a)
1.0
0.5
0.0 -1.0 -0.8
-0.5
(b)
whr
-0.5
of Fig. 6.10.
is that the principal ambiguities are distributed around the line

Wd + 2nkt = 0 (6.33 )
This implies that a misinterpretation of the range of a target will also lead
to a misinterpretation of its velocity, since the peak in the ambiguity diagram
at a given range has an associated velocity that depends on the range.
The cuts through the ambiguity diagram parallel to the two axes convey
no idea of this structure, but confirm our original expectations about the
range and doppler resolutions taken in isolation. In fact, it can be shown
that for a large time x bandwidth product, i.e. kT» 1, the first null in the
cut along the time axis occurs at approximately 1/ kT, which is the reciprocal
of the pulse bandwidth. The gain in range resolution by using frequency
modulation is clear when Fig. 6.l1a is compared with Fig. 6.4a. For the
rectangular pulse without frequency modulation, the cut along the delay axis
extended to T before becoming o. If we regard the distance to the first zero
along the delay axis as a measure of range resolution, then the resolution
has been improved by a factor kT 2 , which is known as the compression ratio.
As we would expect, the compression ratio is directly proportional to k.
There is no improvement in the doppler resolution.
A striking example of the ambiguity inherent in a chirp pulse often occurs
in synthetic aperture radar (SAR) images. SAR uses a side-looking pulsed
radar carried on a moving platform (normally an aircraft or satellite) to
generate images of the earth's surface (see Chapter 11). In the simplified
configuration of Fig. 6.12, we can see that a stationary scatterer has a doppler
frequency relative to the platform given by
fd = 2 V sin 0 :::::: _ 2 Vx [Hz] (6.34 )
2 2R
for small O. The minus sign is present because, in our system of coordinates,
if 0> 0, the distance x is negative (see Fig. 6.12). Since x = Vt (taking the
scatterer at the origin of the x axis, and measuring time from the instant
when the scatterer is broadside to the platform)
2V 2
fd = - - t [Hz] (6.35)
2R
This is a linear FM signal like that described by Eq. (6.31), for which
k = - 2 V 2 /2R. The SAR processor compresses this signal, and for a
stationary scatterer the peak response will occur in the right place in the
image relative to other stationary scatterers. If the scatterer is moving and
has a velocity component along the line of sight to the radar, the associated
doppler shift causes the response from the scatterer to be a cut across the
ambiguity diagram parallel to the delay axis, but moved up or down. As we
can see from Fig. 6.10, this moves the maximum response to a later or earlier
-----r4h, [ \
~
~ F
[
[
Flight track
•
I6\ I
I \ I
I \ I
[\ I
I \ I
[ \ I
I \ I
I \ [
[ \ :
\ [
\ I
\6 1
\i
, I
\ I
\ [
------------------~,~----------------. x=O x
Figure 6.12 A simplified view of the geometry of a SAR system.
time. The time error, using Eq. (6.33), is

O)d
t
e
= -27rk
-= [s] (6.36 )
where we have used the notation fdu to indicate the excess doppler frequency
uncompensated for in the SAR processing. Since the maximum response
occurs at the wrong time, the scatterer will be misplaced in the image. The
positional error corresponding to the time error of Eq. (6.36) is
RA
Vte = 2V fdU [m] (6.37)
Because of this, SAR images often show effects such as ships displaced from
their wakes, or cars apparently in the middle of fields instead of on the road
along which they are travelling. The magnitude of the displacement (which
is in the along-track direction, even though caused by cross-track motion)
can sometimes be used to estimate the velocities of moving scatterers.
Worked example A ship with speed 10 km h - I is travelling on a bearing

95°. It is imaged by an air-borne X-band SAR travelling due west at
200 km h - I at a range of 40 km. How far will the ship apparently be
displaced from its wake in the image?
ISO UNDERSTANDING RADAR SYSTEMS
SOLUTION Since fdu = 2Y.1 A, where y. is the radial velocity, the dis-
placement is given by
d=R(Y.IV) [m]
Here Y. = - 20 cos 85° ~ - 1.74 km h - 1, so that the displacement is
approximately 348 ffi.
COMMENTS It is clear that quite small doppler shifts can cause large
apparent displacements in the image. In fact, the calculation is compli-
cated by the fact that SAR is sampling the signal. Since the bandwidth
of the FM signal used in the along-track processing is typically only a
few hundred hertz for an air-borne SAR (see Chapter 11), even modest
cross-track velocities can move the scatterer out of the frequency band
used in the processing. The scatterer may then be aliased (or signal may
be lost). Calculating the ensuing displacement effects requires detailed
examination of the way the SAR processing is being carried out.
6.9 PHASE CODING
Another form of pulse compression is one in which a long pulse of duration

T is made up of M contiguous subpulses each of length TIM. The subpulses
each have their own modulation, which can be in frequency (e.g. Costas S )
or in phase. We only discuss the latter. A general formula for the ambiguity
function of these types of waveforms is given in Skolnik 6 (equation 175 of
Chapter 3).
There are numerous schemes for such phase coding, but the simplest are
the Barker codes 7 • In these codes, the phase is either 0 or 1t, so that the
transmitted signal is effectively a binary sequence taking the values 1 and
- 1. The choice of such a sequence is constrained by ambiguity requirements,
which means that the ACF of the sequence should have low sidelobes. The
Barker codes are those sequences for which the sidelobes at zero doppler do
not exceed 11 N. Only nine such sequences are known, the longest being of
length N = 13. These are given in Table 6.2, where + represents 1 and -
represents - 1. A contour plot of the ambiguity function for the Barker code
oflength N = 7 is shown in Fig. 6.13, and the corresponding cut along the
axis of zero doppler is shown in Fig. 6.14a. We also give in Table 6.3 the
sequence of outputs along the delay axis when the code and the replica it is
being correlated with are displaced by an exact multiple of TIM. (In
Fig. 6.l4a all values of the lag are considered, not just the multiples of TI M.)
Though the sidelobes are larger than would be desired, the form of the
ambiguity plot is approaching the ideal 'thumb-tack' ('drawing-pin') form,
i.e. a sharp central peak surrounded by a comparatively flat plateau. (The
DESIGNING RADAR WAVEFORMS lSI
Table 6.2 The Barker codes (here + corresponds to 1 and - to -1)
2 + +
2 +
3 + +
4 + + +
4 + + +
5 + + + +
7 + + + +
11 + + + + +
13 + + + + + + + + +
2~-------------------.rrnnTInr-------------------
t:
'3 0
-1
-2L-__~L-____L -____~~Ull~~_ _~_ _ _ _~_ _ _ _~_ _ _ _~

-8 8
Figure 6.13 A contour plot of the ambiguity function of the Barker code of length 7 ( T = 1.0,
N = 7).
plateau must be there, to meet the requirements of the total ambiguity having
unit volume.) The cut along the doppler axis is shown in Fig. 6.14b.
A form of phase coding that does not suffer from the restricted length
constraints (and associated sidelobe levels) of the Barker codes is known as
Frank coding 8 • These codes are of length M2, and can be thought of as M
sub-sequences each of length M. Each sub-sequence starts with zero phase.
(a)
(b)
wIn
of Fig. 6.\3.
Table 6.3 The output sequence for a Barker code of order 7
Code o 0 0 \ I 1 -\ -\ -I o 0 0
'd -6 -5 -4 -3 -2 -\ 0 \ 2 3 4 5 6
Output -\ 0 -\ 0-1 o 7 0 -\ 0 -\ 0-\
The output is obtained by placing a copy of the code, displaced 'd places to the right,
under the code (zero displacement is when the code and its copy are exactly aligned), multiplying
component wise and summing the products.
In the first sub-sequence, all the phases are 0; in the second, the phase of
successive pulses increases by 2n/ M; in the third, it increases by 4n/ M; and
so on (see Fig. 6.15 for the phases of the Frank code of length 16). The
average phase change in the pth sub-sequence is 2n(p - 1)/ M radians. Since
average rate of phase change is a measure of frequency, this means that the
frequency of the coded pulse increases linearly with p. Hence we might expect
the signal to display some of the properties of a linear FM signal. This is
borne out by the contour plot of the ambiguity function of the Frank code
of length 16 shown in Fig. 6.16. The central feature of this plot is a diagonal
;---
I- - r--
I- r-
- ;---
'IT I- r- ;---
T/4
I I Tl2 3T/4 T
Figure 6.15 The phase changes associated with the Frank code of length 16.
1.51-
1.01-
0.5r-
~ 0.01-
()
-0.51-
-1.01-
D
-1.51-
-2.0 -16
I I I oI I I I I
-12 -8 -4 o 4 8 12 16
Figure 6.16 A contour plot of the ambiguity function of the Frank code of length 16 (T = 1.0.
M=4).
(a)
1.0
0.5
(b)
whr
of Fig. 6.16.
ridge similar to that in the ambiguity plot of linear FM (Fig. 6.10). There
are also parallel ridges (such ridges also occur for linear FM, but at a lower
level, and hence are lost by the contour levels used to generate Fig. 6.10).
The cut along the delay axis corresR-0nding to Fig. 6.16 is shown in
Fig. 6.17a. The peak sidelobe level is ../2/16. This demonstrates that the
sidelobe levels of the Frank codes do not fall as rapidly as 1/ M, the rate
achieved by the Barker codes. However, there is in principle no restriction
on the length of a Frank code. For large Frank codes, it can be shown that
the peak sidelobe level declines as 1/(nJ M), so that arbitrarily low sidelobe
levels can be obtained. The cut along the doppler axis is shown in Fig. 6.17b.
6.10 SUMMARY
Detection performance is optimized and range and doppler accuracy are

both improved by maximizing the SNR. Since this needs the transmitted
pulse to have high energy, long pulses are required unless a very high-power
transmitter is available. Long pulses also permit good doppler accuracy, but,
for simple pulses, give poor range accuracy. The system designer can avoid
this apparent dilemma by using phase-modulated pulses. These can be
constructed to give the large bandwidth that, after pulse compression, leads
to good range accuracy. Whatever transmitted waveform is chosen, the
resolution constraints imposed by the ambiguity function cannot be escaped.
All that the designer can do is to attempt to move the significant areas of
ambiguity into regions of the range-doppler plane where targets are unlikely
to be present.
Key equations
• Effective bandwidth:
2 f:oo eo21U(eoW deo toooo IU'(tW dt

(f", 21[E E
• Effective pulse duration:
2 f:oo t 2lu(tW dt f:oo IU'(wW dw

~ = [~J
E 21[E
• Accuracy of delay measurement:
1
[sJ
<5'd = 21[(f",J (2EI N)
• Accuracy of doppler frequency measurement:
<5Wd =
21[(ft
J (2EIN)
• Product of range delay and doppler accuracies:
1 1
<5'd <5eod ~ - -- [J
41[2 EIN
• Equivalent-rectangle time resolution:
Md = JIPu(tW dt = JIU(wWdeo
2
[sJ
p~(O) 21[E
• Equivalent-rectangle frequency resolution:
L\, =JIPu(wWdw =21[Jlu(tW dt
'" p~(O) E2
• Ambiguity function:
x(t, w d ) = -1 fro u(s)u*(s - t) exp(jwdS) ds

E -ro
= - 1 fro U*(v)U(v - w exp(jvt) dvd)

2nE - 00
• Properties of the ambiguity function:
X(O,O)=l
Ix(t, wd)1 = Ix( -t, -wd)1
Ix(t, wd)l ~ 1
6.11 REFERENCES
I. The Fourier Transform and its Applications, R. N. Bracewell, McGraw-Hili, New York, 1986.
2. Probability and Information Theory, with Applications to Radar. P.M. Woodward, Pergamon
3. A theory of radar information, P.M. Woodward and I. L. Davies, Phil. Mag., 41, 1001,
1950. [Most of this material is covered in reference 2, but not the details leading to the
relation between SNR and accuracy given as Eq. (6.5).]
4. Range and velocity accuracy from radar measurements, R. Manasse, MIT Lincoln Lab.
Report, 312-326, 1955. [This is the original reference, but there is some discussion in
reference 6.]
5. A study of a class of detection waveforms having nearly ideal range-doppler ambiguity
properties, J.P. Costas, Proc.IEEE. 72, 996-1009,1984.
6. Radar Handbook, M. Skolnik, McGraw-Hill, New York, 1970.
7. Group synchronizing of binary digital systems, Barker, R.H., in Communication Theory,
Ed. W. Jackson, Academic Press, New York, pp. 273-287, 1953.
8. Polyphase codes with good nonperiodic correlation properties, R.L. Frank, IEEE Trans.
l'lformation Theory, IT-9, 43-45, 1963.
9. Pulse compression techniques with application to HF probing of the mesosphere, CA.
Gonzales and R.F. Woodman, Radio Sci., 19, 871-877,1984.
6.12 PROBLEMS
6.1 What is the equivalent rectangle resolution of the output of a receiver matched to (a) a
rectangular and (b) a gaussian pulse? What is the frequency resolution of a gaussian pulse?
6.2 Sketch the output of a receiver matched to rectangular pulses of length T [s], if there are
two equal-amplitude targets present separated by range delays of (a) T/4, (b) T/2, (c) T,
(d) 3T/2 and (e) 2T[s].
6:3 Sketch the output from a receiver matched to zero doppler and a rectangular pulse of length
T [s] if there are two targets simultaneously present, both with the same RCS and range but
one of which has zero doppler and one has doppler frequency 2n/T (for simplicity, not reality,
use T= 1).
6.4 Compare the resolutions of the two pulses of problem 4.2.
6.5 A book on radar makes the following statement: 'M pulses, each with duration t p ' can be
viewed as a single pulse of duration Mtp. Since bandwidth B is related to pulselength t by
B = l/t, this combined pulse will have a bandwidth fBI M, where fB is the bandwidth of a
single pulse.'
Consider a radar transmitting a coherent amplitude-modulated pulsetrain u(t) as shown
below.
II (I)
aT
How does the effective bandwidth of this pulsetrain compare with the same quantity for a single
pulse? Show that the amplitude of the Fourier transform of this pulsetrain satisfies
sin(2Tw) I
I
IU(oo)l= sin(Tw/2) IP(w)1
where P(w) is the Fourier transform of a single pulse. Sketch JP(w)l, Isin(2Tw)/sin(Tw/2J1
and 1U (00) I. Do you agree with the statement that heads this problem? Are there any hidden
assumptions in it?
6.6 An air-borne C-band (6 cm) SARtravelling at 300 km h - I carries a I m antenna. It
produces an image in which a ship is apparently displaced by a distance RP/4 m in the
along-track direction at range R, where P is the beam width of the real antenna. What can we
infer about the ship's velocity?
6.7 Find the output from the binary codes:
(a) I, I, I, -I;
(b) 1, I, -I, 1;
( c) I, I, I, - I, 0, 0, 0, 0, I, I, - I, I.
This is an example of a complementary code. Sequence (c) is constructed from a pair of shorter
binary codes whose sidelobes cancel each other, and which are separated by (at least) the
length of the shorter codes. Complementary codes give complete cancellation of the range
sidelobes (in the absence of noise) in the vicinity of the main lobe. They are used when very
low sidelobes at zero doppler are important, and are widely exploited in mesosphere-strato-
sphere-troposphere (MST) radars (e.g. Gonzales and Woodman 9 ).
Find all the complementary pairs of length 2, and show that there are no complementary
pairs of length 3.
CHAPTER
SEVEN
SECONDARY SURVEILLANCE RADAR
• What secondary surveillance radar is, and what it does

• Advantages over primary radar
• Problems with secondary surveillance radar
• The future
Secondary surveillance radar is one of the main tools used in air traffic
control.
7.1 INTRODUCTION
Secondary surveillance radar (SSR) is not a true radar system at all but a
two-way communication system between an interrogator on the ground and
transponders fitted to aircraft, which reply automatically. We include SSR
here because the system is very similar to radar in the way it operates, suffers
from many of the classical radar problems and is widely used throughout
the world, often in conjunction with primary surveillance radar.
The origins of SSR lie in the 'identify, friend or foe' (IFF) systems of
World War II; a signal was transmitted from the ground towards a suspect
aircraft, which was required to reply with the appropriate code or be treated
as a foe. Modem SSR works in a similar way; interrogation messages are
transmitted on a one-way uplink frequency of 1030 MHz and cooperating
aircraft reply on a one-way downlink frequency of 1090 MHz. The replies
are fed to a plot extractor, which decodes the aircraft identity and height
IS9
and passes them on to the air traffic controllers together with the measured
range and bearing. A classic work on secondary surveillance radar is Stevens!.
Secondary surveillance radar cleverly avoids problems with clutter
through the use of two frequencies, because the receivers at either end of the
link are not in tune with the adjacent transmitter and they do not pick up
unwanted echoes. Another advantage of SSR is that the transmitter and
antenna gain requirements for the uplink and downlink are much more
modest than for a primary surveillance radar operating over the same
range-this is because the combined R2 propagation losses of each one-way
link are much lower than the R4 losses of the two-way radar signal path.
Worked example In the past, the International Civil Aviation Organization

(ICAO) specification for SSR has limited the peak effective radiated
power (ERP = transmitter power x aerial gain) to 52.5 dB W. What is
the uplink range of the system if the transponder needs to receive a signal
of -101 dB W for adequate probability of detection of a single pulse?
Assume that the transponder antenna on the aircraft has no gain, that
propagation losses amount to 2 dB and that a further allowance of 3 dB
must be made because the aircraft does not remain at the peak of the
azimuth antenna pattern for the entire interrogation time.
What ER P would the primary surveillance radar designed in
Chapter 2 need in order to detect the same aircraft if its RCS were
20 dB m 2 ?
SOLUTION Using the same notation as in Chapter 1, the power density

per unit area at the range R of the aircraft is given by
PtGt
Power flux = --2
4nR
This is intercepted by an antenna with an effective area of

G,,t2
A = _r_ [m 2 ]
• 4n
So the power received by the transponder is
2
P = IWt Gr J.. L [W]
r (4nR)2
We can rewrite this to give the maximum operating range Rmax as

PGGJ.. 2L)1/2
R = t t r
Em]
max ( P (4n)2
r
SECONDARY SURVEILLANCE RADAR 161
Putting in values:
IWt 52.5 [dBW]

Gr 0 [dB] Nominal omnidirectional antenna
A,2 -10.7 [dB m 2 ] =0.291 m
L -5 [dB] Total losses
II Pr 101 [dB W- 1 ]
1/(4n)2 -22 [dB]
R~ax 115.8 [dB m 2 ]

Rmax 616km (300 nautical miles)
In practice, the maximum range would probably be less than this because
the aircraft would not necessarily be in the centre ofthe antenna elevation
pattern and receiving the full 52.5 dB W illumination.
In comparison, the power needed by the primary radar can be
worked out using Eq. (2.11) as follows:
(SNR) x N x (4n)3 x R4
PtGt = "2 [W]
Gr x Ii. X (J x Ls
= + 13 [dB] - 145.8 [dB W] + 33 [dB] + 231.6 [dB m4 ]
- (+36 [dB] - 12.7 [dB m 2 ] + 20 [dB m 2 ] - 5 [dB])
= 93.5 dB W
Assuming 36 dB as the antenna contribution to the ERP, then the
transmitter power would need to be over 500 k W.
COMMENTS The range performance of the SSR that we have just

calculated is quite sufficient, for two reasons. First, most aircraft are
below the horizon when they are 600 km away; and secondly, they would
almost certainly be in a different air traffic control zone.
Primary and secondary radar are often used together (sometimes the
antennas are even attached and rotate together); the primary system is used
to provide air traffic controllers (A TC) with a 'map' on a plan position
indicator (PPI) display of everything moving in the sky in their region. The
SSR interrogates each target, usually with the request: 'Who are you and
what height are you at?' Cooperative targets, such as all civil airliners and
most private and military planes, reply with the information requested, which
is then displayed in alphanumeric form at the appropriate place on the PPI
display. In today's crowded airlanes, this information is valuable, for after
an aircrew have filed their flight plans, including the flight identification
number, the aircraft can be tracked automatically along its route without
the need for ATC requests for the plane to identify itself. Such requests would
be necessary if the only information were from primary radar.
7.2 BASIC PRINCIPLES
The interrogation of an aircraft by SSR involves a pair of pulses modulating

the 1030 MHz carrier frequency. The pulses are labelled PI and P 3 and the
spacing between them determines the information requested from the aircraft
(see Fig. 7.1). Different requests are known as 'modes' of operation and,
although there are quite a few, mostly military, there are two modes used
more frequently than the others because they are common to both civil and
military aeroplanes; these are modes 3/A and C:
P 1 -P 3 spacing Request Mode
Identify 3/A
Height? C
Aircraft reply to this interrogation with a train of pulses that are 0.45 its
wide and spaced 1.45 its apart and which are used to modulate the 1090 MHz
downlink carrier. The first and last pulses in this train are always present
and are known as framing pulses F 1 and F 2 (Fig. 7.2). In between F 1 and
F 2 are 12 pulses, which mayor may not be present, depending on the message
AmPlitudetJ
F1LJP4'--________________---I1F1L
Time
Figure 7.1 The interrogation message is determined by the separation of the pulses PI and
Pl' Pulse P2 is used for sidelobe suppression.
Amplitude
i F,
Figure 7.2 The form of the reply from the transponder on the aircraft.
being transmitted, and an extra centre pulse, which is not currently used.
The 12 pulses are used as a 12-bit code, which has 4096 possible combinations
-enough to give the aircraft identity when in mode 31 A. In mode C, one
of the pulses is not used, but the remaining 2048 codes are sufficient to give
the aircraft height in steps of 100 feet. When both height and identity are
required, the requests are made alternately, which is known as mode
interlacing. There is plenty of time to interlace modes A and C because, as
the SSR antenna rotates, the beam illuminates an aircraft for about 30 ms
and during this time its transponder is interrogated about 15 times.
Normally, an airline pilot will select mode A/C and the transponder is
programmed to respond to requests from the ground without crew inter-
vention. The aircraft altimeter automatically feeds height information to the
transponder. There are also four other switches the aircrew can set, which
send out additional messages for unusual situations such as a hijacking,
radio communications failure or if ATC are having problems identifying the
aircraft.
Secondary surveillance radar is used to locate aeroplanes in a similar
way to primary radar. Ranges are measured by the round-trip time of a
pulse travelling up to the aircraft and returning, with appropriate allowances
for the delays in the equipment on the aircraft and on the ground. The
accuracy of the range measurement can be improved by using correlation
methods on the entire received pulsetrain, rather than using just one pulse.
The aircraft height is known from the information given by its own altimeter,
and so the remaining piece of information required to fix its position is the
azimuth, as measured at the SSR.
Azimuths are now measured using the monopulse technique described
in Chapter 3, and this provides much greater accuracy than earlier methods.
The importance of monopulse for SSR is that it provides a measurement of
angle on every pulse, so that when a train of pulses is received, these
measurements can be combined to form an improved estimate of the azimuth
(in general, n independent measurements can be combined to improve the
J
accuracy by a factor of n). A typical SSR antenna is about 8 m wide,
giving an azimuth beam width of about 2.5°, but after monopulse processing
the error in the azimuth of the aircraft is as low as a few minutes of arc.
Because the antenna is not required to measure elevation angles (the aircraft
height being already known), the avertical size of older SSR antennas
has often been only about 0.4 m, giving vertical beamwidth approaching
50°. These long, thin parabolas have given rise to the descriptive term
'hogtroughs '.
Modern SSR antennas are usually flat phased arrays having vertical
dimensions of the order of 1.6 m to give improved control over the elevation
pattern, and these have become known as LV As, large vertical apertures.
However, they remain several times wider than they are tall, and to some
extent still retain the long, thin appearance of the old hogtroughs.
7.3 PROBLEMS WITH SECONDARY SURVEILLANCE

RADAR
The advantages of secondary surveillance radar over primary radar are

perhaps fairly obvious; less transmitter power is required, all aircraft give
the same amplitude response independent of their size and there are no ReS
fluctuations to worry about. It is also less expensive and conveniently
identifies the aircraft. But SSR is not without problems of its own, although
fortunately most of these now have engineering solutions.
One problem is caused by the sidelobes of the ground antenna initiating
the transponder response, as shown in Fig. 7.3. The radar system does not
know that the echo has come from a sidelobe transmission, and so incorrectly
plots the aircraft position as being in the centre of wherever the main beam
happens to be pointing. This side/obe interrogation can be resolved by
transmitting an extra interrogation pulse P2 on a separate antenna (this is
the mystery pulse shown on Fig. 7.1 and not referred to thus far). This new
'control' antenna is simpler than the main antenna and is designed to have
an omnidirectional pattern and lower gain; however, it still has more gain
than the sidelobes of the main antenna (Fig. 7.3) because of the rule that
the gain integrated all the way round an antenna must come to unity.
On the aircraft, the transponder is programmed to reply to main-beam
interrogations, in which case the amplitudes of PI and P 3 should be equal
and larger than P 2 , which has come from the control antenna. However, if
~AZimuth
rotation
Pattern of
main SSR antenna
/ '" '"
4::i::
/ '"
/ '"
/
/ '"
/
/
/
",-
" , '" /
/ 'K
/ /"'\
I \
I \
I ~ Pattern of omnidirectional
\ I
\ antenna
\ I
\ /
\ /
"- /
/
" '- _/
Figure 7.3 SSR main-beam omnidirectional antenna patterns. Aircraft at close range may be
interrogated by a sidelobe of the antenna as well as by the main beam.
P 2 exceeds PI or P 3' then the request must have come from a sidelobe of
the main antenna and the transponder makes no reply.
The idea of using a control beam is an example of good engineering.
Rather than spending a great deal of money trying to improve sidelobe
performance of the main antenna, the problem can be solved by adding an
extra low-budget antenna. Most modern SSR systems do not in fact have a
separate control antenna-the control beam is synthesized by the main
antenna, so that both main beam and control beam suffer equal multipath
fading (see next section), and amplitude comparisons between the two remain
valid.
Other problems and solutions involving SSR systems are as follows:
• Fruit (false replies unsynchronized in time) occurs when the transponder

on an aircraft is triggered by one SSR and the signal is received by
another SSR that was not expecting a reply. The fact that fruit replies
arrive at the receiver at arbitrary times means that they have random
ranges and are relatively easy to filter out.
• Garbling is a good descriptive term for overlapping replies from two
aircraft. Garbling can occur because a fruit reply happens to occur
at the same range as a true reply, in which case the problem quickly
resolves itself as the fruit range changes. However, garbling can also
occur because two aircraft are at similar ranges (this does not mean that
they are in danger of collision, for they may be at different altitudes and
azimuths but are at the same radial distance from the radar). Before
monopulse, garbling was a serious problem, but now the azimuthal
accuracy of monopulse can be used to assign a bearing to each pulse
received, thus enabling the two intermixed pulsetrains from the aircraft
to be resolved.
• Other difficulties with SSR include co-channel interference (because all
systems work on the same frequency and may interfere with each other),
capture when one system monopolizes an aircraft's transponder and
causes another system to lose data, and false replies in which a
transponder may be accidentally interrogated twice because the trans-
mitted signal reaches the aircraft by an additional path such as by
reflection from a large building, see reference 2 for more details. A familiar
example of radio signals being reflected from buildings is the 'ghost'
image sometimes seen on a television picture; the main image is the
direct signal, and a weaker signal arrives slighter later after bouncing
off a nearby block of flats or other large structure.
The problem of radio signals arriving at the receiver by more than one route
and corrupting the information carried is common to most forms of radar
and to radio communications in general. It is usually known as multipath
and frequently involves reflections from the ground as well as buildings.
7.4 MULTIPATH
Figure 7.4 shows a radio signal travelling from a radar system on a tower
up to an aircraft; some of the transmitted energy travels by the shortest path
and some is reflected from the ground and appears to come from a source
beneath the ground. These two radio sources act in the same way as Young's
slits in optics, and give sum and difference interference patterns. In fact, a
version of the Young's slit experiment using a single source of light and a
mirror to generate an apparent second source is called Lloyd's mirror, and
is an even better analogy of the multipath case.
When reflected from the ground, the radio signal undergoes a phase
change, which is nominally 180 If the difference in path length between the
0
•
direct and reflected signal is ),,/2, an interference maximum is formed and

the signal arriving at the aircraft is larger than expected. Moments later,
however, the path difference might change to .A. and almost complete
cancellation of the signal occurs. This results in wild fluctuations of amplitude
and phase, which are worse at low elevation angles. The problem is the same
for systems receiving radio signals and for those transmitting, and it can
cause severe problems for tracking radars and SSR systems.
One way of thinking about these interference fringes is to consider them
as breaking up the vertical antenna pattern into a series of narrow lobes, as
shown in Fig. 7.5. A simple way to demonstrate the reality of these lobes is
to use a radar in receive-only mode to watch the sun setting (the sun is quite
a powerful radio source). At high elevations, the sun appears as a constant-
amplitude radio source, but as it nears the horizon, amplitude fluctuations
become apparent, and then finally become severe as the sun sets (the sun is
sometimes used in this way to measure the vertical radiation pattern of a
radar antenna). An aircraft flying horizontally at low elevation passes
through the lobes of the antenna, just as the sun does, and the signal level
varies, causing the radar system to keep losing track of the target.
Phase change of ' , , , :

nominally 1T radians '..J Image
Figure 7.4 Propagation over a plane reflecting surface results in two signal paths that interfere
with one another.
Free-space antenna pattern ..-_,,_____- ___ Lobes resulting from the

presence of reflecting ground
\
"-
"-
"-
"-
"-
"-
j
""- "-,
,,
,,
,,
\
Figure 7.5 The presence of a reflecting surface causes the antenna pattern to break up into
lobes at low elevation angles.
Perhaps the first attempt at solving the multipath problem might be to

try to make use of the time delay between the direct and the reflected signals,
but the time difference is usually too small. This can be shown with a
simplified calculation using the geometry of Fig. 7.4 with a few assumptions,
such as the Earth being flat and the range R of the aircraft being large
compared with the height of the radar antenna hr. A little elementary
geometry can be used to show that the difference in range AR between the
direct and indirect wave is given by
[m] (7.1 )
where ha = height of the aircraft. Putting some typical numbers into
Eq. (7.1) for example hr = 10 m, ha = 1000 m ('" 3000 feet altitude) and
R = 100000 m (100 km range), gives AR = 0.2 m. This is equivalent to a
difference in time of 0.7 ns between the reception of the direct and reflected
signals, and is far too small in comparison with SSR pulse lengths to form
a practical method of separating them. This same difficulty is true of digital
communications in general.
Other attempts to deal with multi path include placing scattering structures
on the ground to break up the reflected wave and even painting offending
structures with radar-absorbing paint, as developed for Stealth aircraft
applications. The most promising methods of reducing the effect of multipath
probably lie in improved antenna design and more advanced signal and data
processing. The challenge is to design an antenna with high gain at low
elevation angles but which avoids 'looking' at the ground at zero elevation.
The process of achieving this has already begun with the new LV A antennas,
which have a much better control over the vertical beam pattern. Other
techniques that might be considered are:
• Locking the antenna with a slight positive elevation to reduce the

sensitivity to ground reflections, but with the disadvantage that it also
reduces the signal-to-noise ratio for low-elevation aircraft.
• Developing an antenna pattern with a deep null that is directed towards
the ground-the problem with this method is that the elevation of the
ground changes with azimuth as the antenna rotates.
• Arranging for the greatest rate of change of beamshape to lie along the
direction of the ground, rather than a null, to give the largest possible
difference in output between low-elevation signals.
• Using two antennas, one above the other, and correlating their outputs
to separate the muItipath signals.
• To use existing antennas and put more effort into rejecting data showing
unreasonable behaviour, such as wild variations in range or azimuth,
and trying to form tracks with the remaining data. Information from
other sensors such as radar, infrared systems and optical tracking systems
might be added into the tracking process at this point, especially on
military systems.
7.5 MODE S AND THE FUTURE
For all its association with the world of radar, SSR is essentially a digital
communications system, and a new mode of operation known as mode S is
coming into service to exploit this aspect more fully, see Scanlan 3 . Longer
pulsetrains are transmitted to increase the information transfer, and the
interrogation of aircraft is selective in that the ground base addresses them
one at a time.
The first task of a new SSR system is to find out which of the aircraft
it is interrogating are equipped with mode S transponders. This identification
is achieved by the use of a fourth interrogation pulse P4, which can be given
a duration of either 0.8 or 1.6 J.LS. Because pulses PI to P 3 are transmitted
as before, the older mode Ale transponders continue to reply as normal. A
mode S transponder examines the duration of P 4; if it is set to 0.8 J.LS the
transponder makes no reply, but if P 4 = 1.6 J.LS then it replies by giving its
own unique address. These addresses are noted by the SSR system, which
then schedules the mode S transponders for special interrogation.
The mode S interrogation proper is contained within a long P 6 pulse,
which is modulated by differential phaseshift keying (DPSK-a standard
digital communications technique) to convey either a 56- or 112-bit word
depending on whether the length of P 6 has been set to 16.25 or 30.25 J.LS.
These bits are used for selectively addressing the mode S transponders.
Aircraft fitted with the older transponders do not reply to this new type of
transmission because, cleverly, the first two pulses PI and P 2 are set to the
same amplitude, which older transponders treat as a sidelobe interrogation
and .therefore ignore.
Mode S transponders reply in a similar manner to the interrogation; a
four-pulse preamble is followed by a 56- or 112-bit binary word. Several of
these words may be strung together in successive transmissions to enable
quite complex messages to flow between aircraft and the ground.
7.6 SUMMARY
Secondary surveillance radar is partly a communication system between

aircraft and air traffic controllers on the ground; a limited amount of
information (aircraft height and flight identification number) is requested
by an interrogator on the ground and automatically supplied by a transponder
on the aircraft. In the future, this flow of information will increase.
Secondary surveillance radar also acts as a radar system because the
position of the aircraft is found by measuring the range (from the time delay
between interrogation and reply) and the azimuth, as measured by an antenna
on the ground. Many of the early problems with SSR have now been solved,
and the system is in widespread use throughout the world.
7.7 REFERENCES
1. Secondary Surveillance Radar, M.C. Stevens, Artech House, Norwood, MA, 1988. [The
classic book on modern SSR; it is clear and full of detail.]
2. Understanding Radar, H.W. Cole, BSP Professional Books, Oxford, 1985. [Contains more
than 60 pages of information on SSR in a very readable form.]
3. Modern Radar Techniques, Ed. MJ.B. Scanlan, Collins, Glasgow, 1987. [Chapter 6, also
written by H.W. Cole, is dedicated to modern SSR.]
7.8 PROBLEMS
7.1 The Cossor Condor 9600 is a complete ATC system that includes full monopulse SSR. One
of the SSR antenna options is the Condor 9642 large vertical aperture, which has a peak gain
of 27 dB and a beam width of 2.45 at the - 3 dB points.
0
If the power of the Cossor interrogation transmitter is adjusted so that the ERP conforms
to the ICAO specification of 52.5 dB W when it is used in conjunction with the 9642 antenna,
would you expect an aircraft transponder to receive an adequate signal at the maximum
instrumented range of 256 nautical miles? Assume additional propagation losses of 2 dB and
an antenna pattern loss of 3.5 dB because the elevation angle of the aircraft places it above the
angle of maximum antenna gain.
7.2 For the downlink the ICAO defines the transponder output power as 24 dB W. For an
aircraft at a range of 100 nautical miles, what signal strength would the Condor 9600 system
receive? Assume 1 dB loss for atmospheric attenuation and a further 7 dB for system and vertical
antenna pattern losses. Remember that the downlink frequency is not the same as the uplink
frequency used in the previous question.
7.3 If the minimum working signal level of the interrogation system on the ground is
- 110 dB W, what is the theoretical maximum downlink range of the Condor system? Allow
3 dB in your calculations for azimuth beam width loss.
7.4 Why does the downlink apparently give better performance than the uplink if the transmitter
is less powerful?
7.5 Estimate roughly the angular uncertainty that might be expected when the Condor system
tracks an aircraft with a range of 100 nautical miles. Assume a receiver noise level of - 130 dB W.
CHAPTER
EIGHT
PROPAGATION ASPECTS
• The radar horizon

• The effect of the atmosphere
• The effect of the ionosphere
• Diffraction effects
How to cope with radio waves not travelling In straight lines in the
atmosphere.
8.1 INTRODUCTION
Radars frequently operate through the atmosphere of the earth, often at low
elevation angles where most targets occur, but where there are most problems
for the radar. Aircraft flying at constant altitude towards a surveillance radar
appear over the curvature of the earth and are difficult to detect and track.
Similar problems occur with ship surveillance when using maritime radar.
When close to the horizontal, radar beams have their greatest path length
through the atmosphere, which is itself at its most dense and turbulent.
Low-elevation radar beams can also encounter obstacles, such as hills, and
become diffracted into the shadow regions behind.
In this chapter we are going to investigate the radio horizon and discover
whether the atmosphere, the ionosphere and the terrain significantly affect
the propagation of radio waves. Again, elementary optics is useful to describe
what is happening.
171
8.2 THE RADAR HORIZON
Assuming for the time being that radio waves travel in straight lines through
the atmosphere, the radio horizon is defined by a line tangent to the surface
of the earth, as in Fig. 8.1. If R is the range of a point on the radio horizon
(or of a target appearing at zero elevation), h is the height of the point and
Re the radius of the earth, then
R2 = (Re + h)2 - R;
= 2Reh - h 2 [m 2 ]
It is reasonable to assume that h 2 « 2Reh and therefore
R ~ j(2Reh) Em] (8.1 )
Although the earth is not truly spherical, being flattened at the poles and
bulging slightly at the equator, there is little error in assuming it to be a
Centre of
Earth
Figure 8.1 Geometry for calculating the height of the radar horizon as a function of range.
PROPAGATION ASPECTS 173
sphere with an average value of Re = 6378 km. Many radar engineers have
to work with aircraft height in feet and range in nautical miles. (A nautical
mile is the distance around the earth's surface corresponding to an angle
of one minute measured at the centre of the earth. It equals about 6080 feet
or 1.85 km; see Appendix III.) Restating Eq. (8.1) in nautical miles
[no mile], but with h in feet, gives
R ~ )(2 x 3444h/6080) [no mile]
and a rough approximation that can prove useful is
Rn. mile -- ) hreet [no mile] (8.2 )
If the radar is at an altitude hradar> Eq. (8.1) can be modified as
R ~ ) (2Rehradar) + ) (2Rehtarget ) em] (8.3 )
Worked example A radar situated at sea level is approached by a missile

flying at a height of 10 m, a low-flying aircraft at a height of 100 m and
a high-flying bomber at an altitude of 15000 m. At what ranges will
these targets appear over the radar horizon and be detected by the radar,
again assuming that radio waves travel in straight lines?
SOLUTION Using Eq. (8.1) and Re = 6378 km, we get the following
answers: Sea-skimming missile is detected at a range of 11.3 km.
Low-flying fighter is detected at a range of 35.7 km. High-flying bomber
is detected at a range of 437 km.
COMMENTS This quick calculation reveals several interesting aspects of

radar; low-altitude missiles are very difficult to detect until the last
minute or so of their incoming flight. Even then their RCS is often so
small that they are hard to identify and track, especially if they are
weaving. At the higher altitudes, few civil aircraft are likely to fly above
15000 m, so there is little need to design air traffic control radars for
ranges greater than about 450 km.
Beware, however-the calculations above are pessimistic because radio

waves tend to be refracted over the horizon by the earth's atmosphere,
thereby extending detection ranges. We will examine the effects of the
atmosphere next.
8.3 ATMOSPHERIC EFFECTS
The atmosphere can cause radio waves to be dispersed, attenuated, refracted

and retarded. Dispersion is small enough to be ignored, except in a few
specialist wide band systems.
...'" 100
'"0-
..c Water vapour resonance
<J> - 22 GHz
0
E
co
'"
.::
C
10
'"'"
-=
..c
bI)
::l
2
-=c
.2 resonance
co::l - 60GHz
c
~
co
>.
<II
~
C
~
f- 0.1
0.1 10 100
Radar frequency [GHz)
Figure 8.2 Two-way atmospheric absorption through the entire troposphere as a function of
radio frequency for two elevation angles.
Attenuation by the atmosphere is a function of frequency and, as

Fig. 8.2 shows, it can become quite significant above about 10 GHz, and even
below, if long ranges are surveyed at low elevation angles. Atmospheric
attenuation decreases with air pressure (and therefore with altitude), and
most of the absorption occurs in the troposphere-the lowest, turbulent
region of the atmosphere where the weather is found. The attenuation also
increases fairly linearly with increasing rainfall rate (usually measured
in millimetres per hour, mm h -1) or increasing snowfall, and radars
susceptible to significant atmospheric losses must make use of local weather
statistics in their performance predictions.
Refraction by the atmosphere is a serious problem for most radar systems
and occurs because of the refractive index profile of the atmosphere. The
refractive index of air depends on temperature, pressure and the water vapour
content. Of these factors, it is the partial pressure of the water vapour that
has the largest influence and the total air pressure that is least criticaL
The refractive index of air n is always close to unity; it is about 1.0003
at the earth's surface and falls even nearer to unity with increasing height.
The small differences of n from unity are important, so they are deliberately
magnified by introducing the radio refractivity N defined by
N = (n - 1 ) x 106 [ ] (8.4 )
Hence N has a value of about 300 at the surface. The radio refractivity is
usually expressed in N units, and can be calculated from:
N = 77.6(~) + 3.73 X105 ( ;2) [ ] (8.5 )
where p = air pressure [millibars], T = temperature [K] and e = partial

pressure of water vapour [millibars].
The troposphere changes more quickly in the vertical dimension than
horizontally, and the important parameter is therefore the change of refractive
index with height, known as the refractive gradient dN /dh. Under normal
atmospheric conditions this gradient is negative, roughly linear below about
1 km, and causes radio signals to be bent downwards with a radius of
curvature p given by (see e.g. Meeks l ):
p= ( ----.!2cos¢
. ndh
1d )-1 Em]
at low elevation (8.6)
where ¢ = elevation angle of radar beam.

As Fig. 8.3 shows, the effect of the refractive index decreasing with
altitude is to give the target an apparent height greater than its true height.
The angle Q( through which the beam bends can be calculated from the
following equation, and is positive for a downward-bending beam 2 :
dn
f
largel
Q( = cot ¢ - [radians] (8.7 )
radar n
When radio waves travel in curves, much of the simple geometry used
in radar becomes very difficult. A useful trick, much used by radar engineers,
is to replace the radius of the earth Re by a larger number kR e • This increased
Figure 8.3 Atmospheric refraction causes the apparent height of a target to be greater than
its true height.
radius of the earth has the effect of bending the radio waves back up into
a straight line over this artificial earth, and allows us to return to our simple
geometry provided we remember to replace Re by kRe every time it occurs
in the calculations. But what value should we choose for k? Geometrically
k is given by
k = p/(p - Re) [ ] (8.8 )
Combining this transform with Eqs (8.4) and (8.6), and assuming cos 4> == 1
for low elevation angles, gives
k = ( I + 10- 9 Re : : ) - 1 [ ] (8.9)
using dN / dh in units of N per kilometre (the conventional units) and Re in

m. You might see Eq. (8.9) used with a factor of 10 - 6 instead of 10 - 9; this
happens when Re is expressed in kilometres.
For a standard atmosphere, dN /dh = -40 N /km, giving k = 1.255, but
engineering practice has been to use the approximation k = 4/3, and the
expression 'four-thirds earth' is commonplace. When calculating radar
performance there is special paper available for calculating the vertical
coverage that is drawn on a four-thirds earth (see Fig. 8.4). If you follow
Typical coverage for a surveillance radar:

1 m2 target detectable in clear
conditions. P d - 80%. P fa - 10-1>
140000~ 40 r-----..j_1
\ Larger targets
detectable
120000~
_ 100000
v E 30 r--+--ic....LJ
~ ::!.
80000
~
'<; 20 r---ftf+-I-LI
:I:
40000
10
oL~~~~~~~i:=~~:~;:t:=~joo
20000
o o 200 400 500

Range [km]
o 50 100 ISO 200 250

Range In. mile]
Figure 8.4 A typical graph for plotting radar vertical coverage, using the four-thirds earth
model.
the curve of a target flying at an altitude of 10 km, it crosses the horizon at

a range of over 400 km; without refraction and the four-thirds earth
correction, Eq. (S.l) gives a range of about 357 km.
Worked example Assuming four-thirds earth, at what ranges could the

targets described in the worked example in Section S.2 be detected?
SOLUTION Using Eq. (S.1), but with Re replaced by 1Re we now have:
Sea-skimming missile is detected at a range of 13.0 km. Low-flying fighter
is detected at a range of 41.2 km. High-flying bomber is detected at a
range of 505 km.
For accurate tracking of radar targets, the four-thirds approximation is

insufficient and an exponential formula is used for the variation of N with
height (see Rotherham in reference 3, for example):
N = Ns exp( -hi H) [ ] (S.10)
where Ns = surface value of N [ ] and H = scale height of the atmosphere
[m, if h is in m]. The CCIR reference atmosphere 4 gives the values Ns = 315
and H = 7.35 km, and worldwide maps of Ns are to be found in Bean and
Dutton 2 •
In practice, there can be large day-to-day variations in the refractive
index structure of the atmosphere. Sub- or super-refractive layers can occur
in which the refractive gradients are respectively less or greater than the
standard atmosphere. At low elevation angles (0-2°) radar beams can
become trapped in super-refracting ducts close to the earth's surface. When
dN Idh = -157 N Ikm, the radius of curvature becomes equal to Re, and
greater values cause radio signals to bend back to earth (Fig. S.5).
/'
/'
/'
/'
,-
,-
,-
,-
,-
,-
/'
~
Figure 8.S Super-refracting ducts can cause radar signals to bend back to earth, creating
anomalous clutter problems.
Ducting is usually caused by temperature inversions on land, or

by evaporation at sea, and it can permit substantial over-the-horizon
propagation when conditions are favourable. This is exploited in com-
munications, but is too lossy and intermittent to form the basis of a reliable
over-the-horizon radar system. It does, however, cause some radars to
observe ground clutter at greater ranges than usual. Some types of tempera-
ture inversion can also cause sub-refraction, so that the radio wave is bent
towards the earth less strongly than usual. Sub-refraction can result in
various forms of anomalous propagation (anoprop ), including the possibility
of the radio wave entering an elevated duct formed between layers in the
atmosphere having different temperatures.
Atmospheric refraction effects occur at all radar frequencies, but in
general they begin to affect systems operating above 500 MHz before affecting
those working at longer wavelengths.
Retardation is the effect by which radio waves travel more slowly in air
than in free space. Usually the range error that this causes is small enough
to be ignored, but there are cases, such as the battlefield radar described
later in this chapter, where it becomes necessary to make corrections. The
range error depends on the value of n integrated over the path length R, but
if we can define some average value ii then
!1R = (ii - I)R em] (8.11 )
Assuming a linear refractivity profile gives
em] (8.12 )
where No = value of N at the radar site [ ], hr = height of the radar em]

and hI = height of the target em]. (This formula is sometimes expressed with
h in kilometres, in which case the divisor 2000 is reduced to 2.)
8.4 DIFFRACTION BY THE TERRAIN
The phenomenon of diffraction was discovered by the Jesuit, Francesco

Maria Grimaldi, who described the experiments leading to its discovery in
his book De Lumine, published in Bologna in 1665. It was Grimaldi
who gave the phenomenon the name 'diffraction' from the Latin verb
diffringere = dis + frangere, meaning 'to break in different directions'. The
name is a good description of the effect, which is important at radio
wavelengths, because it can sometimes make radar detection possible in the
shadow regions behind hills.
There are well established mathematical models and computer programs
that describe diffraction over simple obstacles such as knife-edges and
cylinders, and the usual engineering approach is then to try to describe real
-----------4.~----------d2----------~.1
Figure 8.6 The geometry used to evaluate diffraction effects.
terrain in terms of these obstacles. Fortunately, most hills and ridges can be
described quite well by single and multiple diffraction edges or by cylinders
with knife-edges on top of them. Procedures exist for determining the outcome
when there are several different types of these obstacles in the path of the
radio wave.
The simplest and most rigorously analysed diffraction obstacle is the
knife-edge, shown in Fig. 8.6. Various texts describe the mathematics; see
for example Griffiths 5 or Meeks 1 (which includes useful program listings).
Although a full mathematical analysis is complicated, the results may be
summarized, using the notation shown in Fig. 8.6, as follows:
• Diffraction losses generally increase with frequency f, but for the

important case of grazing incidence the diffraction loss is independent
of frequency and is about 6 dB. At large diffraction angles, the loss has
a to x log(.f) dependence.
• The minimum loss occurs when the obstacle is midway between the
radar and the target, rather than near either end of the radio path.
• Increasing the diffraction angle a + increases the diffraction losses, so
increasing the distance between the radar and the obstacle helps to reduce
the losses by lowering a + .
• Rounded hills, modelled as cylinders, behave in a similar manner to
knife-edges but with additional losses. Often, though, a hill that is
geologically smooth is broken by trees, scrub and terrain irregularities,
so that the diffraction edge model remains a good assumption.
• Diffraction from a knife-edge upwards can modify the free-space field
by interference, but as a- increases, this loss quickly falls to zero.
8.5 BATTLEFIELD RADAR SYSTEMS
Radars used on the battlefield for the detection or guidance of weapons are
interesting in many respects, not least of which is the sophisticated command
ISO UNDERSTANDING RADAR SYSTEMS
and control communication networks that link them together. However,

weapon location radars are particularly interesting because of their sensitivity
to variations in the propagation conditions.
One of the most lethal weapons of World War II was the mortar, and
radar systems for locating them were soon developed. The task is not so
difficult, because mortar bombs are usually fired at a high elevation; they
may be detected against a sky background and their flight path can be
measured and predicted backwards to find the source from which they came.
Unfortunately, modern artillery shells, and rocket barrages fired from mobile
launch vehicles, follow flatter trajectories, which makes them harder to detect
because of ground clutter. The low trajectory also forces radars to use
low-elevation beams, and the geometry is such that small errors in the
measurement of the ballistic flight path can cause large errors in the
calculation of the weapon site (Fig. 8.7).
Modern artillery duels are deadly affairs fought at ranges of up to a
maximum of about 35 km. The objectives of weapon-locating radars are to
detect enemy shells or rockets as soon as they rise from the ground, calculate
the position of the source and then call for fire by friendly forces on that
location. Generally speaking, a gun battery has insufficient fire-power to
saturate an area much larger than a football pitch, and an enemy weapon
must therefore be located with an accuracy of about 100 m on the ground.
Because of the flat trajectories, this ground error can translate to an accuracy
of a few metres along the flight path of the shells.
For a microwave weapon-location radar operating at a maximum range
of 35 km, the correction in the height of the target due to refraction at low
elevation angles can be calculated by integrating Eq. (8.7) from the radar to
the target to give the vertical displacement from straight-line propagation.
For a standard atmosphere with dN Idh = -40 N Ikm, the height correction
is about 25 m and is roughly independent of elevation. However, if super-
refraction occurred with dN Idh = -120 N Ikm, then the correction needed
Retardation
Measurement of
Target
errors
Actual position True Estimated

of target at weapon weapon
moment of location location
measurement
Figure 8.7 Small errors in the measurement of a ballistic trajectory can cause large errors in
the location of the source.
would be nearer 75 m and, if a standard atmosphere had been assumed, an

error of up to 50 m in the calculated height of the missile would be introduced.
The correction to the range due to atmospheric retardation can be calculated
from Eq. (8.12) and would be around 10 m in the worst cases. Although
this is smaller than the height correction, it remains sufficiently large to cause
problems if the wrong atmospheric model is assumed. Failure to locate enemy
gun positions accurately can have fatal consequences, not only for the friendly
forces but for the radar itself, which will quickly be located (by passive radio
monitoring devices) and placed high on the enemy list of target priorities.
There is a need for good meteorological information on a battlefield
in order that the flight of projectiles can be predicted accurately and to
compensate for the effects of atmospheric refraction on the radar per-
formance. At present, atmospheric data are best supplied by radiosonde
balloon systems similar to those used by the meteorological office. Often,
though, only surface observations are available, and a standard atmospheric
profile is assumed. Even radiosonde data are oflimited value because balloons
can usually be flown only over friendly territory. Also, balloon instru-
mentation does not respond to very sharp refractive index changes. New
laser, radar and passive sounding systems are being developed to measure
temperature, pressure and humidity, and in the future these may be found
on the battlefield in support of radar location systems.
8.6 IONOSPHERIC EFFECTS
So far, we have discussed the effects of the neutral atmosphere on radar

systems operating at low elevation angles. When radar systems look out into
space, or from space down to Earth, the free electrons in the ionosphere-the
ionized part of the atmosphere-may playa significant role in radio wave
propagation. This can be viewed as a problem if the radar is looking for an
undisturbed propagation channel between itself and the target. It can also
be turned to advantage as a means of studying the ionosphere from the
ground. In fact, ground-based radars play a major role in ionospheric
research. Incoherent scatter radar measures the minute accounts of power
back scattered by electrons in the ionosphere when a high-power electro-
magnetic (EM) wave passes through it. Systems such as the European
Incoherent Scatter Radar Facility (EISCAT) in Tromso, Sweden, have
revolutionized our understanding of ionospheric processes in the last decade.
Other systems, such as the PACE radar in Antarctica, and the SABRE and
STARE systems in northern Europe, measure the backscatter from irregu-
larities in the ionosphere, and use these to study electric fields and plasma
processes.
The ionosphere begins at a height of 60 km and continues up to about
1000 km. It was one of the first targets to be detected by early pulsed radar,
I
~r
300
E
.=..,
"0
oE 200
<:
Electron density [m-']
Figure 8.8 The variation of electron density with altitude in the day-side ionosphere.
which, in 1924, revealed the presence of several ionized layers, as shown in

Fig. 8.8. The electron content in these layers depends on the solar energy
falling on the atmosphere, so has latitudinal, diurnal, seasonal and long-term
variations, as well as sudden changes due to outbursts of solar activity.
(The structure of these layers is discussed further in Chapter 10.) There
are also belts surrounding the magnetic poles in both the northern and
southern hemispheres where energetic particles (mainly electrons and protons)
bombard the atmosphere. This particle precipitation (and other related
processes) gives rise to the visible aurora, so these regions are known as the
auroral ovals. It also causes significant modification of ionospheric structure
and electron content. These regions can be particularly troublesome for radar
systems. The electron content normally varies irregularly, with blobs and
holes giving rise to backscatter that could be mistaken for targets. Normally
there are also strong electric fields. These cause the irregularities to have
velocities and associated doppler shifts that may be comparable to those
expected from targets. Hence, extraction of targets from auroral clutter can be
a difficult problem.
The effect of the ionosphere on radio waves passing through it is to
cause attenuation, refraction and a rotation of the plane of polarization
known as Faraday rotation. Reflections from the ionosphere form a special
case of refraction, which plays a vital role in skywave over-the-honzon radar.
This is discussed separately in Chapter 9. Diffraction effects may also
be important when the ionosphere is spatially irregular, such as occurs most

of the time at auroral latitudes, and in the post-sunset equatorial region.
Attenuation by the ionosphere increases with increasing ionization, but
is inversely proportional to frequency squared. Above a few hundred
megahertz, the effect is small. For frequencies exceeding 2 MHz, the attenua-
tion can be estimated very roughly by6
Two-wayloss=A/j2 [J (8.13)
2
where j is the radar frequency [HzJ and A [s- J is a constant, which is of
the order of 4 x lO-8 in the daytime and 2 x lO-10 at night. More detailed
theory can be found in Hall and Barclay 3, Picquenard 7 and Scanlan 8 •
Refraction is the slowing of the wave velocity as it passes through a
medium, which causes the direction of travel of a plane wave to alter. The
bending of the rays gives rise to the problem you encounter when you try
to pick up a coin at the bottom of a pool. If you do not correct for the
tilting of the wave-normal to a steeper angle (see Fig. 8.9a), you will touch
the pool bottom too far away. For the ionosphere, it is more like having a
slab of material that displaces the line of sight to the object (Fig. 8.9b). This
aspect of refraction leads to positional errors in the horizontal plane.
The slowing of the wave causes errors in the apparent distance to the
target. The refractive index n is given by
[ J (8.14 )
where Ne is the electron density [electrons m - 3 J and j is given in hertz. The
minus sign is used when we are considering phase velocity, and the plus sign
when we are interested in the velocity of propagation of signals. The refractive
index tells us by what factor the wave is slowed compared to the speed of
light. To find the increase in time for a signal propagating through the
ionosphere, we must integrate the refractive index along the ray path. For
frequencies in excess of 30 MHz, the ensuing one-way range error is given by
III =
f
s
n ds ~ 40.3
-2
j
f
s
Ne ds [mJ (8.15 )
where s is distance along the path S from the radar to the target. The integral
of electron density along the path is called the total electron content (TEe),
for obvious reasons. It varies considerably with position and time, but
generally does not exceed 5 x 10 17 electrons. This value of TEe would give
rise to a one-way range error of 900 m at 150 MHz, which is the frequency
used, for example, by the TRANSIT positioning system. More problem-
atically, variations in TEe would cause this error to vary between a few
metres and hundreds of metres unless they were properly corrected.
Worked example The ray path from a space-based radar operating at

1 GHz must pass through an ionosphere for which the TEe is equally
(a)
True Apparent
position position
(b)
Refracting
slab
""
""
""
""
"
True " Apparent
position position
Figure 8.9 Positional errors caused by refraction.
likely to be anywhere in the range 1 x 10 16 and 5 x 10 17 electrons. What

errors would be possible in calculating the true range to a target 600 km
away, and what would be the average error?
SOLUTION Iff is in gigahertz and TEe is given in units of 10 16 electrons,

the range error equation becomes
Al = 0.403 x (TEe)/ j2 [m]
In these units f = 1 and TEe varies between 1 and 50, so the possible
range errors are between 0.4 and 20 m. If all errors are equally likely,
the average error is 10.2 m.
Since the refractive index of the ionosphere increases with increasing

electron density, a ray entering the ionosphere from the ground is pro-
gressively bent further away from its original direction. In the extreme case
(which occurs at high frequencies (HF)), the ray path can be bent so much
that the ray returns to the ground (if the ray was launched from space, it
would return into space). Only rays that are of a high enough frequency to
penetrate the peak ionospheric electron density can pass through the
ionosphere. (This peak typically occurs at a height of 300-400 km.) This
•reflection' from the ionosphere is used for ionospheric sounding from either
the ground or space (the type of radar that carries out this sounding is called
an ionosonde, and it is discussed in Sec. 10.4). Reflection from the ionosphere
is also the principle used by over-the-horizon radar to obtain radar
backscatter from the earth's surface at very long ranges (see Chapter 10).
Faraday rotation is caused by a combination of the earth's magnetic
field, the electron density and the path length through the ionosphere. The
plane of polarization is rotated by an angle
Q = M x (TEC)/j2 [radians] (8.16 )
where M is dependent on the geometry of the earth's magnetic field and

the wave-normal. A reasonable order of magnitude for M is 0.5 s - 2, which
indicates that at gigahertz frequencies rotations of at most a few degrees
occur (though extreme cases in excess of 80° have been recorded). Hence
Faraday rotation will cause little problem to, for example, space-borne
synthetic aperture radars operating at gigahertz frequencies. However, like
attenuation, Faraday rotation has a II f2 dependence, and will become
increasingly important as frequency decreases. Severe effects can be caused
at HF.
Attenuation, refraction and Faraday rotation can be thought of as acting
on individual ray paths passing through the ionosphere. However, except at
the highest frequencies, we need to take account of diffraction effects. These
occur when the electromagnetic wave passes through an irregular ionosphere.
As we have already noted, this is the normal state of affairs in the auroral
zones and in the post-sunset equatorial region. Extreme disturbances could
be caused by chemical releases into the ionosphere (rocket releases of barium
oxide are often used in ionospheric research), or by high-altitude nuclear
explosions. If a space-based radar is well above the ionosphere, the
transmitted wavefronts will be nearly planar by the time they reach the
ionosphere. As the wave passes through an irregular slab of ionosphere, local
variation in the refractive index causes phase perturbations. On exit from
the ionosphere, the contours of constant phase are no longer planar (see
Fig. 8.10). While the wave propagates downwards, interference effects set
in, which cause amplitude and phase fluctuations at the ground. An exactly
similar process is what causes the stars to twinkle, and the fluctuations are
i i o
Planar
phase fronts
Irregular
ionosophere
Perturbed
phase
front
Propagation
to
ground
Amplitude
scintillations
Figure 8.10 The mechanism by which phase perturbations in the ionosphere give rise to
scintillations observed on the ground. As the perturbed wave propagates to the ground,
interference effects develop, and give rise to large fluctuations in phase and amplitude.
hence known as scintillations. On the return path to the satellite, the same
type of disturbance will occur.
The amplitude scintillations can be thought of as fluctuations in the ReS
of the target, and hence are treated in the same way as for the Swerling cases
(though there is still no universal agreement on the correct PDF to describe
these scintillations). Whether the received signal varies on a pulse-to-pulse
or a scan-to-scan basis depends on the relative motion of the satellite, the
target and the ionosphere. Phase scintillations can also have important effects
on systems that rely on phase coherence, such as space-borne SARs. If the
pulse-to-pulse signal contains a random phase element that changes signifi-
cantly as the synthetic aperture is being formed, the radar performance may
be degraded. Progressively more serious effects include displacement of the
beam (leading to geometric errors), increased sidelobe levels (leading to loss
of contrast in the image) and destruction of the focus (leading to complete
loss of the image). Figure 8.11 shows all these effects for simulated SAR data
as the operating wavelength increases. At a wavelength of 0.83 cm,lthe synthetic
Distance [m I Distance [m I
5.0 5.0
-50~.0~~~~~O~~~~~5~0.0 -50.0 o 50.0
A =0.83cm
-45.0 ~ -45.0
I a:l ~
~ ~
c c
'<;j '<;j
o o
Distance [m I Distance [m)
5.0 5.0
-50.0 o -50.0 o 50.0
A = 6.00cm A = 24.00cm
-45.0 ~ Cil
~ ~
c -45.0 c
'<;j '<;j
0 0
Figure 8.11 Simulated distortions of the synthetic aperture gain pattern caused by ionospheric
irregularities. The different patterns correspond to increasing wavelength (and hence increasing
synthetic aperture length; see Chapter 11), as marked on each pattern. The ionosphere was
the same in all cases.
beam is hardly disturbed. By the time the wavelength has increased to 24 cm,
the beam has been almost completely destroyed.
8.7 SUMMARY
At low elevation angles, the propagation of radio waves is degraded by both

atmospheric effects and scattering by the terrain. For many general-purpose
surveillance requirements, the four-thirds earth approximation is a suffi-
ciently good correction for atmospheric refraction (the strongest effect) and
the other problems are ignored. When precision target tracking is required,
all atmospheric effects have to be considered, and appropriate corrections

made, based on local climatic information.
The electron content of the ionosphere also causes attenuation and
refraction effects, and a rotation of the plane of polarization. Radio waves
passing through the ionosphere may suffer a form of scintillation that can
affect space-borne earth-imaging radars.
Key equations
• Range of a point on radio horizon:
R ~ J(2R eh) [mJ
which can be restated in nautical miles as:
[no mile]
• Radio refractivity:
N = (n - 1) x 106 [J
which can be calculated from:
N = 77.6(~) + 3.73 x 105(;2) [ ]
• Effective earth radius factor:
k = ( 1 + 10 - 9 Re ~~) - 1 [ ]
• Range error due to atmospheric effects:
~R = 1O- 6 R (N.o + (hI2000

- hr ) dN)
dh
[m]
• Attenuation by the ionosphere:

Two-way loss = Alf2 [ ]
• Refractive index of the ionosphere:

n2 ~ 1 ± 80.6N / P e [ ]
• One-way range error through the ionosphere:

~l = 0.403 x (TEC)/ P [m]
• Faraday rotation:
n=M x (TEC)/P [radians]
8.8 REFERENCES
There are many good books on the subject of propagation because this is a
subject of importance in communications as well as in radar. The following
selection is suggested as a useful introduction to the subject.
I. Radar Propagation at Low Altitudes, M. L. Meeks, Artech House, MA, 1982. [A very
readable monograph complete with program listings and a good bibliography.]
2. Radio Meteorology, B. R. Bean and E. J. Dutton, Dover Publications, New York, 1968.
[This is the standard text on large-scale atmospheric variations in radio refractive index.]
3. Radiowave Propagation, Ed. M. P. M. Hall and L. W. Barclay, Peter Peregrinus for the
IEE, Stevenage, Herts, 1989. [Originally derived from course notes accompanying an lEE
school on radio wave propagation, this book forms an indispensable guide to the whole
subject.]
4. Reference Atmosphere for Refraction, Recommendations and Reports of the CCIR, CCIR
Rec369-3, V, lTV, Geneva, 1986.
5. Radio Wave Propagation and Antennas, J. Griffiths, Prentice-Hall, Englewood Cliffs, NJ,
1987. [A particularly easy to read guide to propagation.]
6. Radar Handbook, Ed. M. I. Skolnik, McGraw-Hili, New York, 1970.
7. Radio Wave Propagation, A. Picquenard, Macmillan, London, 1974.
8. Modern Radar Techniques, Ed. M. J. B. Scanlan, Collins, Glasgow, 1987.
8.9 PROBLEMS
8.1 At what range could a target flying at a height of I km be detected by a radar at sea level,
assuming (a) no refraction, (b) four-thirds Earth, (c) a refractive gradient of - 20 N jkm and
(d) a refractive gradient of - 120 N jkm ?
8.2 In the cases (c) and (d) in problem 8.1, what range corrections would be needed to allow
for atmospheric retardation? Assume No = 300 at the radar site.
8.3 (a) In the next chapter it is shown that the power received by a weather radar, when
precipitation fills the beam, is proportional to 1j R2. If the radar were badly sited such that
many of the low-elevation observations were made at grazing incidence to the local horizon,
how would the performance be impaired when scanning for light rain nominally detectable at
4.5 km in the clear?
(b) How would the detection range of a point target be affected at the same elevation?
8.4 A space-based radar operating on the two frequencies 200 and 400 MHz measures the range
delay to a target as 44 and 14 ms respectively. What is the true range to the target?
CHAPTER
NINE
RADAR STUDIES OF THE ATMOSPHERE
• Why atmospheric radars are useful

• Angels and echoes from clear air
• Special-purpose research radars
Our atmosphere is a major environmental issue; radar is one of the key

research tools.
9.1 INTRODUCTION
The atmosphere would not appear to be a very promising radar target at

first sight, and it is perhaps surprising to discover that there are a considerable
number of radars dedicated to observing and investigating the atmosphere
at all altitudes. There are three main areas in which atmospheric radars are
used, and these are now outlined.
Hazard avoidance Most airliners carry a weather radar in the nose to look
ahead and give a warning of severe weather on a display inside the cockpit.
Pilots may then take avoiding action within the limits set by their flight
lane. Airports prone to severe storms are now also installing ground-based
weather radars to give warnings of severe down-draughts, which are a
danger to aircraft taking off and landing. At present, strong down-draughts
during thunderstorms cause about one aircraft crash every 18 months in
the USA alone.
190
RADAR STUDIES OF THE ATMOSPHERE 191
Forecasting Radar maps of rainfall are a familiar sight on televised weather

forecasts, but the radars supplying this information also provide other data
useful for forecasting. The main data source for weather forecasts remains
satellite images, aircraft measurements and the radiosonde balloon flights,
which are equipped with instruments to measure temperature, pressure and
humidity. However, radar continues to make increased contributions to
weather forecasting, and there are plans to replace balloons with a network
of stratospheric-tropospheric (ST) radars, which are described in Sec. 9.3.
(The balloons themselves are tracked by the type of radar described in
problem 3.1.)
Research Balloons rarely reach altitudes in excess of 25-30 km and most of

what we know about the upper atmosphere comes from radar and lidar
observations. Radar is important in lower-atmosphere research, too. Although
balloon networks can provide a synoptic picture of the weather, they cannot
observe the evolution of small-scale atmospheric features revealed by
continuous radar surveillance.
9.2 SCATIERING MECHANISMS
In this section we look at the different types of scattering mechanisms that

cause radars to receive echoes from apparently clear air, and at how these
mechanisms can be exploited to investigate the structure and dynamics of
the atmosphere.
Radar echoes from the clear atmosphere have interested scientists since
the mid-1930s, although during World War II they were regarded as
something of a nuisance (see Atlas!). After the war, the increasing use of
radars, and the strength of some types of echo, led to serious research
into understanding the possible scattering mechanisms. By the 1960s, a
new generation of high-power, high-frequency radars had been developed
especially for the detailed study of weather processes in the lower atmosphere.
These systems were often X-band (9.4 GHz), but sometimes S-band (2.8 GHz)
and UHF (near 400 MHz). The 1970s saw a major shift to VHF frequencies
and the use of large, vertically pointing, phased array antenna systems.
The turbulent lower atmosphere in which we live is called the troposphere,
and extends from ground level up to roughly 11 km. Within this region there
are many scattering sources including rain, snow, hail, clouds, birds, insect
clouds, debris raised by fires, steps in temperature, clear-air turbulence and
fog (at millimetre wavelengths). Understanding these scatterers is important
to radar engineers for two reasons: first, they are a source of clutter for
conventional radar, and secondly, they form a mechanism by which the
atmosphere may be studied. The more prevalent scatterers, and their ReS,
are now considered.
Hydrometeors
This is the meteorological term for scattering particles (rain, ice particles,
etc.). They cause the Rayleigh scattering that we encountered in Chapter 2
because the radar wavelength A is usually greater than the diameter D of the
particles. The ReS of a single droplet is
= C n 5D /-A.
6 4
(J particle (9.1 )
where C is a dimensionless constant that depends on the dielectric constant
of the particles. For water, C is near 1, whereas for ice particles it is about
0.2. To calculate the RCS of a rain cloud, we must sum over the volume
resolution cell of the radar, to give
(9.2 )
where N = number of scatterers in the volume illuminated [ ]. The

sixth-power dependence on the particle diameter means that heavy rain,
which usually contains large droplets, gives a much stronger backscatter
echo than light rain. This effect is easily observed on the weather radar
colour intensity maps shown on television weather forecasts when storms
are present.
In practice, Eq. (9.2) is evaluated by defining L: Dr
per unit volume as
a radar reflectivity factor Z (see Battan 2 , Nathanson 3 or Skolnik4). This
volume reflectivity is empirically related to the precipitation rate r [mm h - 1 ]
by
N
Z = L: Dr /unit volume
1
= arb (9.3 )
The millimetre units creep in because of the meteorological conventions of
describing drop size in mm; do not blame the radar engineers! There have
been many different experimental determinations of a and b, but some
commonly accepted values are presented in Table 9.1. Note that, at low
Table 9.1 Empirically determined constants

relating the rainfall rate r to the radar reflectivity
factor Z
Precipitation a b
Rain 200 1.60

Thunderstorm 486 1.37
Ice crystal 500 1.66
Wet snowflakes (t > O°C) 2000 2.0
Dry snowflakes (t < O°C) 1050 2.0
frequencies (in the Rayleigh region), Z is not frequency-dependent, and by

building Z into the radar equation we can develop a formula that will tell
us the power received from precipitation by radars with operating frequencies
up to X-band. If the volume of the resolution cell v'es is given by
v'es = (R AO)(R AcfJ)( C7: /2) (9.4 )
where 7: = pulse duration [s], then the RCS of the resolution cell is given by
(T = v'es L (Tpartic'e/unit volume
(9.5 )
and the radar equation becomes (with any appropriate units also shown)
[W] [ ][ ][ ][ ] [m 3 ] [rad][rad][m S-I ][s][ ) [ )
r-A--.
[W] PI GI Gr C n2 (to- 1H Z) de drj> c r L,ys Latmos
[W]
Pr 64 ).2 R2 2
[ ][m 2 ][m 2 ] [ ]
(9.6 )
with Z in mm 6 m- 3 • The propagation loss Latmos may have to include the

effects of attenuation through the rain cloud itself. Collecting the constants
together (in units ofm S-I), and assuming the constant C to be unity, gives
p = 2.3 X 10- 11 PtGtGrZ AO AcfJ 7:LsysLatmos
[W] (9.7)
r .PR 2
The constant 2.3 x 10- 11 has dimensions of speed because it includes the
velocity of light c. Occasionally, variations of this formula are used to take
into account the elliptical beamshape. Sometimes v'es is reduced by a factor of
n/4, and Skolnik 4 suggests a further reduction of 1/(2In 2) because the
effective two-way illuminated volume is less than the one-way value suggested
by the half-power beamwidths used in Eq. (9.4). At long ranges, radar
resolution cells are large and are unlikely to be completely filled by
precipitation. In this case, a beam-filling factor must also be introduced.
Radars designed specifically to investigate cloud physics sometimes
operate at frequencies higher than X-band to increase the RCS of the
hydrometers. To surveillance radars, rain represents a form of clutter that
is particularly unwelcome because it has a doppler shift imparted by the
motion of the cloud, as shown in Fig. 9.1. For this reason, surveillance is
carried out at relatively low frequencies, usually L-band, to achieve long
ranges.
Worked example In the UK, the Rutherford Appleton Laboratory

operates an S-band p. = 9.75 cm) high-resolution radar for the study of
Horizon
r\----+-+--- Backscatter from the ground
0--'-/-+--- Ship
Aircraft travelling
' \ - - - Backscatter from the atmosphere radially towards
the radar
Figure 9.1 Backscatter from rain and other atmospheric features amount to a form of moving
clutter that can mask slowly moving targets.
cloud physics. The antenna is a 25 m diameter parabolic dish giving a

beamwidth of 0.25 (see reference 1). If the peak power is 560 kW and
0
the pulse duration is 1.0 p.s, what echo power would you expect to receive
from heavy rain (r = 4 mm h - 1), if the rain entirely filled a resolution
cell, at a range of 50 km?
SOLUTION We have Z = 200(4)1.6 = 32.6 [dB mm 6 m- 3 ] and then

using Eq. (9.7)
Constant = - 106.4 dB m s - 1 == 2.3 X 10- 11 m S-1
PI = 57.5 dB W
GIGr = 116.4dB
Z= 32.6dB mm 6 m- 3
AO A<b = -47.2 dB rad 2
r= -60.0dB s
l/A? = 20.2 dB m- 2
1/R2 = -94.0 dB m- 2 (Range = 50 km)
Pr = -80.9dB W
The received power is well above the expected noise level for a I MHz
bandwidth, even allowing for additional system and propagation losses.
Research radars, such as the Rutherford Appleton instrument, make

provision for measuring the polarization of the echo, which gives information
on the shape of the hydrometeors. A surveillance radar, on the other hand,
would almost certainly select circular polarization during heavy rain; this
is because a simple scattering object like a raindrop reverses the direction
of polarization on reflection and if left-hand circular is transmitted, then the
echo returns as right-hand circular and is ignored by the system. Complex
scattering objects such as aircraft, where mUltiple reflections take place,
generally return a mixture of polarizations and so may be detected through
the rain.
As well as acting as a radar target, precipitation also absorbs radio
waves and causes an attenuation that rises rapidly with frequency. For
example at I GHz, rain falling at 25 mm/h causes almost no attenuation
but at 10 GHz the two-way attenuation is around 0.5 dB/km. More details
can be found in Meeks 5 .
Angels
This intriguing name was given to the echoes of discrete targets that returned
from an apparently empty sky to trouble early radar operators. Birds are
probably the main cause, but clouds of insects can also cause angels and,
because they drift in the wind, they can cause clutter problems similar to
rain clouds.
The RCS of a bird generally lies in the resonance region and large
variations are observed, but a rough rule is that a bird has the same echoing
area as a plastic bag filled with water having the same weight. Typical RCS
values are -40 dB m 2 for a small bird and -20 dB m 2 for a larger bird such
as a gull. Flocks of birds appear as much larger, moving targets. Radar can
be used to study bird migration, and to identify birds by wing beat frequency,
altitude and speed; see for example Eastwood 6 .
Insects have an RCS of - 50 dB m 2 or lower, but, again, concentrations
of insects within the volume resolution cell of the radar can cause sufficient
scattering to create the angel effect.
Clear-Air Turbulence
Echoes from genuinely clear air were originally quite unexpected and difficult
to explain, although it was guessed that weak fluctuations in the radio
refractive index of the air must be responsible. It is now generally understood
that echoes from clear air arise from two main types of scattering mechanism:
one is volume scattering from turbulence and the other is specular (mirror-
like) reflections from thin layers.
The atmosphere is everywhere turbulent but there exist regions of greater
intensity such as convective cells or 'thermals'. Volume scattering from these
regions (and also from the whole atmosphere, but fainter) is due to
constructive interference from fluctuations in refractive index. Those scatterers
with a scale size equal to half the radar wavelength are mainly responsible
for the backscattering, in a process similar to the Bragg scattering discussed
in more detail in Chapter 11.
One of the hazards of flying is to encounter the sudden turbulence that
forms in thin horizontally stratified layers and which sometimes has a mean
vertical gradient, or refractive index, much greater than in the surrounding
atmosphere. These layers may be only a few tens of metres in thickness and
yet they can extend horizontally for tens of kilometres. The radio scattering
from these layers causes strong radar echoes in both the troposphere and
the lower stratosphere. The stratosphere is the horizontally stable layer above
the troposphere and extends in altitude from roughly 12 to 50 km.
The region above the stratosphere is called the mesosphere, and thin
layers are found here too. The early Soviet cosmonauts produced some
excellent free-hand sketches of mesopheric layers that are visible when viewed
edge-ways on against the bright limb of the earth. They have since been
confirmed by radar observation. Above 100 km, the atmosphere can no
longer be considered entirely neutral, and ionospheric scattering processes
dominate.
The refractive index of the neutral atmosphere was described by
Eq. (8.5), but in the mesosphere there are sometimes free electrons present,
which cause an ionospheric term to be added:
n= 1 +77.6 x 1O-6(~)+3.73 x 10- 1 (;2)-40.3(;;) []

(9.8 )
where Ne = number of free electrons per m [ ] and f = radar frequency
3
[Hz]. Equation (9.8) describes the refractive index n, from ground level up
to 1()() km (to get the radio refractivity N, subtract 1 and multiply by 106 ).
Using reasonable values of T, e, p and Ne we find that the wet term (containing
e, the partial pressure of water vapour) dominates in the troposphere but
above 50 km the ionospheric term begins to determine the refractive index.
The usual practice in atmospheric physics is to incorporate refractive index
changes into the radar equation by adopting yet another version, based on
Eq. (9.6):
p = P)2LSYs(C; + C;) [W] (9.9 )
r 16n 2 R2
where C; = contribution due to volume scattering by turbulence [ ] and

C; = contribution due to reflection from layers [ ].
The volume scattering term is given by
C; = ARGI] [ ] (9.10)
where AR = range resolution Em], G = antenna gain and I] = radar reflect-

ivity of the scattering volume, or ReS per cubic volume [m 2 m -3 = m -1].
The reflectivity of the turbulence is related to the structure constant C~ (a
measure of turbulence) by
I] = 0.38C~;. - 1/3 [m - 2/3 m - 1/3 = m - 1] (9.11 )
C~ is related to the scale size of turbulence in the atmosphere, which increases
with height. Formally C~ is calculated from
C; = 5.26 x An 2 La 2/3 (9.12)
where An 2 = mean-square variation in the refractive index [ ] and Lo = scale
size of the turbulent eddies Em].
These rather complicated, but well established, formulae mean that we
can relate turbulent refractive index changes to the power received by a radar
observing them, for all altitudes in the neutral atmosphere.
Although clear-air echoes may be a source of clutter for surveillance
radars, they are increasingly being exploited as a means of investigating
waves, winds and turbulence in the Earth's atmosphere, as the next section
shows.
9.3 MESOSPHERE-STRATOSPHERE-TROPOSPHERE
RADAR
During the 1970s it was realized that VHF radars operating at wavelengths
of 1-10 m had considerable potential for studying atmospheric dynamics
because these wavelengths are well matched to turbulence scale sizes. These
radars, known as mesosphere-stratosphere-troposphere (MST) radars,
have several advantages over rocket, balloon and aircraft measurements,
including:
• Very good time and height resolution.

• The capability of making observations simultaneously over a wide range
of heights.
• The possibility of continuous operation.

• The capability of measuring vertical air motions.
A typical MST radar operates at 50 MHz and uses a high-power
transmitting system driving a large antenna array to form several narrow
beams. One beam is pointed vertically and the others are tilted a few degrees
off-vertical in the compass directions N, E, Sand W, as shown in Fig. 9.2.
The same antenna array is used for reception in order to gain the maximum
possible power x aperture product. The need for such a large radar system
lies in the volume scattering mechanism for turbulence, which, despite the
careful choice of wavelength, produces relatively weak echoes compared with
most other types of radar target.
100m
Figure 9.2 Typical MST radar arrangement using several fixed beam positions to measure
horizontal and vertical air motions.
The structure constant C~ typically has a value of to- 17 [m- 2 / 3 ] at

II km, falling by roughly I dB per kilometre increase in altitude. This leads
to a value for fJ of 2.1 x 10 - 18 [m - 1], falling by I dB per kilometre, giving
a very small RCS even when quite large volumes are illuminated.
Worked example A 50 MHz MST radar uses a pulse duration of IllS,

a peak power of 125 kW and an array of dipoles 100 m x 100 m. Would
you expect volume scattering from 11 km altitude to be detected?
SOLUTION If the antenna aperture of 104 m 2 has an efficiency factor of

0.5 (mainly due to weighting to reduce the sidelobes) then from
G = 41£A./ ;. 2 we estimate the antenna gain to be 1745 (32.4 dB ).
Assuming !1R = ISO m, we can calculate C; to be 5.5 x 10 - 13. The
power received can be calculated from Eq. (9.9), with C; = 0:
PI = 51.0 dB W
),2 = 15.6 dB m 2
Lsys = -7.0 dB
C; = -122.6 dB
1/(161£2) = -22.0dB
I/R2 = - 80.8dBm- 2
Pr = -165.8 dB W
If the system were internally noise-limited, the level would be
-204 dB W Hz- I + 60dB (1 MHz bandwidth) plus the noise factor
of the receiver. In this case the volume scattering echo from a single
pulse would be below noise level. In practice, a coherent integration
time of about 1 s would be used and the final bandwidth would
therefore be I Hz. However, external noise dominates internal noise by
about 19 dB at this frequency, and so the final noise level is about
-204 + 19 = 185 dB W, and volume scattering would be detected with
a 19dB SNR.
COMMENTS Roughly speaking, at 11 km the signal strength falls by about
1 dB per kilometre because of the 1/ R2 factor, and a further 1 dB per
kilometre because of the change in C~, giving a total of 2 dB per
kilometre. If no useful atmospheric science can be carried out with a
SNR of less than 10 dB, these results imply that the radar would be a
useful instrument only up to an altitude of about IS or 16 km (longer
coherent integration cannot be used owing to the short timescales of
turbulent processes). The presence of any scattering layers, adding a C;
term into the radar equation, would increase the SNR at lower altitudes,
but by 20 km, they too are beginning to disappear. There is no escape
from the conclusion that higher-altitude atmospheric research requires
more powerful transmitters, larger antenna farms and more money.
A MST radar will measure returns from the troposphere and the lower
stratosphere, and may sometimes receive echoes from the mesosphere when
free-electron scattering occurs, but there is no easy way of investigating the
region of the middle atmosphere between the two. For this reason, some
systems were not designed to try to see beyond the lower stratosphere and,
as a result, two distinct types of radar have evolved: the true MST system,
and the smaller stratosphere-troposphere (ST) radars.
An example of a powerful modern MST radar is the Japanese MU radar,
mentioned in problem 9.6 at the end of the chapter. The MU system uses
an active antenna array (see Chapter 15) consisting of 475 solid-state
transmitter modules, and Vagi antennas, to develop a total peak transmitted
power of 1 MW. At the other end ofthe scale, there is an increasing use ofST
radars to study turbulence and severe wind shears in the lower atmosphere,
and in the USA, operational ST systems are to be installed at 47 of the
country's busiest airports to warn pilots of potentially dangerous wind
conditions. Some of the lower-atmosphere radars are more compact UHF
wind profilers, which operate in much the same way, but at frequencies around
400 or 900 MHz.
New methods of operating ST and MST radars, and of carrying out the
signal and data analysis, are continually being developed, and this field of
radar is expanding rapidly.
9.4 METEOR WIND RADAR
Winds in the upper atmosphere between about 70 and 110 km have been
studied since the 1950s by meteor wind radar. When tiny meteoritic particles
enter the earth's atmosphere, they heat up due to friction (like the Space
Shuttle during re-entry). The heating causes a meteor to boil away in a
process known as ablation, which leaves behind a long, slightly cone-shaped,
column of ionization, which is a strong scatterer of lower-frequency (HF
and VHF) radio waves (see Fig. 9.3). The ionized trails drift in the neutral
wind with a velocity that is easily measured by pulse doppler radar. There
are always sufficient numbers of sporadic meteor echoes per hour for even
a radar of modest power to be able to measure the larger-scale atmospheric
features. At regular times each year, the earth passes through comet debris
causing meteor showers that give an increased echo rate for a few days.
Although meteor wind radars do not have the resolution of the newer
MST radars, they remain of considerable interest because the meteor trail
formation process gives some clues about upper-atmosphere chemistry
(recent work suggests that this could also include ozone concentrations).
There is also an interest in over-the-horizon communications by forward
scatter from meteor trails, and astronomers are interested in learning more
about the distribution and sizes of meteoritic material within the solar system.
Figure 9.3 Radar scattering from meteor trails can be used to measure winds at high altitudes.
A typical meteor radar operates at about 40 MHz with a 15 dB gain

Yagi-Udah or phase array transmitting antenna. The transmitter pulse has
a 10 f.1S duration and a peak power of 200 k W in the larger systems, although
useful results can be obtained with 20 kW. The receiving antenna often
consists of several low-gain elements arranged as an interferometer to measure
the echo angles of arrival, which, combined with the range, are used to
determine the height of the echo and wind measurement 7 .
Although a full treatment of meteor trail scattering theory is complicated,
an estimate of the RCS of a meteor trail can be made in the following way.
Although a few of the brightest meteor trails are said to be overdense (the
electron density is so high that the radio wave cannot penetrate the trail,
rather like a tube of copper), most meteor trails are underdense, with an
RCS given by
(9.l3)
Taking each of the terms in turn, the first is the principal echoing area
given by
(9.l4 )
where J. = radar wavelength, typically 7.S m; q = electron line density,

which for an averagely bright underdense meteor would be about
1013 electrons m -I; re = effective radius of the electron = 2.8 x 10- 15 m;
and R = range to closest point on meteor trail, typically 90 km. These values
give a typical ReS of 3300 m 2 •
The second term is a loss factor to account for interference between the
illuminating radiation and re-radiation from the electrons when the trail
diameter is of the same order as the radio wavelength. It is given by
[ ] (9.1S)
where rj = initial radius of the trail em], which has been found empirically
to be about 0.22 m at 80 km altitude, O.S m at 90 km and 1.1 m at 100 km.
Using 0.5 m for the initial trail radius gives (12 = 0.7.
The third term (13' and the exponential factor, allow for the expansion
of the trail by diffusion; (13 is given by
[ ] (9.16)
The constant K [m] is related to the way the trail diffuses and the velocity
of the meteoritic particle. The value of K is about 0.09 m at 80 km, 0.05 m
at 90 km and 0.1 m at 100 km. A typical value for (13 is therefore 0.6.
The exponential term gives a characteristic decay to meteor echoes,
which is used in some systems as a method of determining the echo height.
The time constant for the received power to decay by a factor e 2 is given by
[s] (9.17)
The rate of diffusion D (known as the ambipolar diffusion constant) can be

found from a formula by Greenhow and Neufeld 8 :
logloD=6.7xlO- sh-S.6 [m 2 s- l ] (9.18)
where It = echo height em].

Putting these factors together gives the ReS of a typical meteor as
1400 m 2 , but decaying with time in an exponential fashion. More details can
be found in Sugar 9, who also gives an expression for the ReS of overdense
meteors.
The number of sporadic (background) meteors encountered by a radar
system can be estimated from
160
Number of meteors with q > qo =- (9.19)
q
During meteor showers, the meteor echo rate can increase dramatically. The
best known, and most spectacular, meteor showers are the Perseids, around
12 August each year, and the Geminids, around 13 December.
Besides their use in atmospheric physics, these formulae describing
meteor rates and ReS can be used to estimate the extent of meteor clutter
detected by various types of radar (for example, HF radars, described in the
next chapter, experience meteor clutter problems).
9.5 OTHER RADAR STUDIES OF THE ATMOSPHERE

Another method of studying the atmosphere between about 5 and 100 km
is lidar (light detection and ranging). This technique involves the emission
oflight pulses in a vertical beam, which are then back scattered by atmospheric
gases and particles. The scattering height is determined from the time delay,
as with conventional radar. The receiver usually involves some form of
frequency selection followed by a photomultiplier detector 1o . The scattering
process may be Rayleigh scattering from molecules, resonant scattering from
specific atomic species to which the laser has been tuned, or Mie scattering
from aerosols and ice crystals in the stratosphere.
Lidar has found applications in measuring stratospheric densities,
temperature, composition and the distribution of pollutants. The recent
development of coherent lidar means that troposphere and lower-stratosphere
wind measurements should soon be possible, perhaps as part of a satellite-
borne global wind field mapping system.
Among the variety of other experimental radar soundings of the
atmosphere is the radar acoustic sounding system (RASS). This ingenious
technique involves the vertical emission of a pulse of sound, which travels
upwards at a velocity that depends on the air temperature and density.
The wavefront is tracked by radar backscatter from the refractive index
discontinuity induced by the sound pulse. The technique can provide
useful information at low altitudes, provided the wavefront is not distorted
by wind shear.
9.6 SUMMARY
From initially being a nuisance, the scattering of radar signals from the
atmosphere has been turned into a useful, and expanding, research technique
for the study of atmospheric physics and meteorology. Scattering occurs
from discrete sources (rain, birds, etc.) and also from changes in the refractive
index of the air, mainly caused by turbulence.
Weather radars, investigating lower-atmosphere cloud physics and
precipitation, operate at frequencies in S-band and above, but new VHF
phased array radars have emerged as a tool for probing the atmosphere from
1 to 100 km altitude.
Key equations
• Radar reflectivity factor:
Z = arb
• Radar cross-section of resolution cell:
s
(J = Cn ZR 2 /1() /1</> CT
4
..1. 2
• Radar equation for hydrometeor scattering:
p = 2.3 x 10- 11 PtGtGrZ /1() /1</> rLsysLatmos
r A2R2 [W]
• Radar equation for clear-air turbulence:

p = PtA2Lsys(C; + Cn [W]
r 16n2 R2
• Volume scattering factor due to turbulence:

C; = /1R G x O.38C~;. - 1/3 []
9.7 REFERENCES
I. Radar in Meteorology, Ed. D. Atlas, Battan Memorial and 40th Anniversary Radar
Meteorology Conference, American Meteorology Society, 1990. [A large but fascinating
book full of the history of meteorological radars, as well as current thinking.]
2. Radar Meteorology, L. J. Battan, University of Chicago Press, Chicago, 1959.
3. Radar Design Principles, F. E. Nathanson, McGraw-Hill, New York, 1969.
4. Introduction to Radar Systems, M. I. Skolnik, McGraw-Hill, New York, 1985.
5. Radar Propagation at Low Altitudes, M. L. Meeks, Artech House, MA, 1982.
6. Radar Ornithology, Sir. E. Eastwood, Methuen, London, 1967.
7. Meteor Science and Engineering, D. W. R. McKinley, McGraw-Hill, New York, 1961.
8. Turbulence at altitudes of (80-100) km and its effects on long duration meteor echoes,
J. S. Greenhow and E. L. Neufeld, J. Atmos. Terr. Ph),s., 384-392, 1959.
9. Radio propagation by reflection from meteor trails, G. R. Sugar, Proc. IEEE,52, 116-136,
1964.
10. Laser Monitoring of the Atmosphere, Ed. E. D. Hinkley, Springer-Verlag, Berlin, 1976.
Other titles that may prove useful are:

II. Radar Ohservation of the Atmosphere, L. J. Battan, University of Chicago Press, Chicago,
1973. [A pre-MST radar book.]
12. Doppler Radar and Weather Ohsen'ations, R. J. Doviak and D. S. Zrnic, Academic Press,
New York, 1984. [A bit mathematical.]
13. Radar Observations of Clear Air alld Cloud~, E. E. Gossard and R. J. Strauch, Elsevier.
Amsterdam, 1983.
9.8 PROBLEMS
9.1 The UK Meteorological Office weather radar network has used Siemens-Plessey S-band
and C-band radar sensors to measure rainfall. The Siemens-Plessey type 45S sensor features
a 3.66 m diameter parabolic antenna giving a 2° pencil beam and a gain of 37 dB at 2.9 GHz.
The peak transmitter power is 650 kW and the pulse duration is 21ls. If the noise level is
-137 dB W, what is the SNR of the echo from a rain cloud filling the beam at a range of
20 km, if the rainfall rate is 1 mm h - 1 ?
9.2 What would be the effect of increasing the rainfall rate to 100 mm h - 1, in the above problem?
9.3 What would be the dynamic range requirement of a receiver covering rainfall rates of 1;8
to 64 mm h - 1 and ranges from I to 200 km? Could this be reduced by sweeping the gain of
the receiver to increase the sensitivity with increasing range? If so, what power law should be
used for the swept gain?
9.4 The German SOUSY MST radar system located in the Harz Mountains operates at
53.3 MHz and has an antenna array 70 m in diameter giving a beamwidth of 5' and a gain of
31 dB. The beamformer uses 4-bit phaseshifters (see Chapter 15) to steer the beam anywhere
within a 30' cone centred on the vertical. At 7' off-vertical, it is found that the echo power
received from a thin layer at an altitude of II km is 10 dB lower than the measurement in the
vertical beam. Could this be explained by the geometry and the longer range to the layer in
the off-vertical beam?
9.5 The peak transmitter power of the SOUSY radar is 600 kW. What is the strength of an
echo scattered from turbulence at II km, assuming a 21ls pulse duration and 7 dB system losses?
9.6 The Japanese MU MST radar system operates at 46.5 MHz, uses a circular antenna array
103 m in diameter and a peak transmitter power of I MW. How much larger would the SNR be
than that of the SOUSY system, assuming the other parameters are identical?
9.7 If the UK MST radar (near Aberystwyth, Wales) uses a coherent integration of 512 pulses.
and is set to have an unambiguous range of 24 km and a total observation time of 10.5 s, what
is the width of the doppler spectrum in the fast Fourier transform?
9.8 If the UK MST radar uses a vertical beam, and beams steered to a maximum of 12'
off-vertical, would an aircraft travelling at 300 m S-1 cause any problems?
9.9 Can an MST radar be used as a meteor radar? Would you expect to detect many meteors
in the vertical beam?
CHAPTER
TEN
OVER-THE-HORIZON RADAR
• The two types of over-the-horizon radar

• Surface-wave radar
• Skywave radar
• Over-the-horizon radar equation
Over-the-horizon radars are even less perfect than microwave systems, but
the rewards for seeing over the horizon are worth pursuing.
10.1 INTRODUCTION
It has been known since the experiments of Guglielmo Marconi in 1901 that
radio waves could propagate beyond the horizon because of the signals he
successfully transmitted from Poldhu in Cornwall, UK, to St John's,
Newfoundland. Investigations into the cause of this propagation soon
revealed that the solution to Maxwell's equations for a wave at a plane
interface between two media gives a space wave (free-space propagation)
and a surface wave (a wave guided along the interface). With the discovery
of the earth's ionosphere in the 1920s, it was realized that there was also a
third possible mode of propagation, the ionospheric wave, which turned out
to be the explanation of Marconi's transatlantic communications.
The propagation of HF (3-30 MHz) radio waves over great distances
has always been exploited in communications, and sometimes frequencies
lower than HF are used, as listeners to long-wave radio will know. During
World War II the UK air defence radar 'Chain Home', operating on
20-30 MHz, was occasionally troubled by 'nth-time-around' clutter created
206
OVER-THE-HORIZON RADAR 207
when the radio signal was scattered by the ionosphere and travelled unusually
long distances. Under these conditions the normal operating PRF of 25 Hz
was reduced to t 2.5 Hz (details in Neall). In other countries, similar
discoveries were made, and many radar engineers began thinking of turning
this unwanted propagation to advantage.
From the early experiments with long-range propagation, two kinds of
HF radar or 'over-the-horizon' (OTH) radar have been developed, known
as surface-wave (or groundwave) radar and skywave radar, making use of
the surface-wave and the ionospheric modes of propagation respectively.
Surface-wave systems were first operated in the early t 950s and effective
skywave systems a little later. The reason for the slow development of OTH
radar compared to more conventional systems is not a reflection on the
ability or the imagination of the engineers involved but rather the constraints
of the technology available at the time.
Surface-wave radar uses the surface-wave propagation mode to look
over the immediate horizon, and it may be used to survey ranges up to a
maximum of certainly no more than 400 km. It is most useful as a local area
defence system and as a method of collecting good-quality wave and tidal
information over a restricted area of ocean. Although bistatic systems have
been operated successfully, surface-wave radar is regarded as being a
predominantly monostatic technique with a relatively low capital cost.
Skywave radars, on the other hand, are almost always large, bistatic
and very expensive. These radars make use of the ionosphere to scatter radio
waves very long distances beyond the horizon, sometimes in forward scatter
mode to a receiver beyond the target. The minimum range is about 1000 km
and the maximum useful range is around 4000 km. Skywave radar is thus
more suited to the defence and remote ocean sensing needs of countries of
such continental proportions as the USA, the former USSR and Australia.
10.2 SURFACE-WAVE RADAR
The principle of surface-wave or groundwave (gw) radar is that a surface

radio propagation mode can be utilized to make a radar signal follow the
surface of the sea as it disappears over the horizon. The method works only
for vertically polarized antennas in contact with salty conducting water; it
cannot be used over land, on freshwater lakes, or where fresh water dilutes
the sea, such as in the Baltic or the Nile Delta.
The propagation of radio waves along surfaces has been analysed
extensively in the past; the theory is complex and is, for all practical purposes,
impossible to solve manually2. However, intuitively we would expect to find
such propagation because, although the sea is a good conductor, it is
not perfect and it supports a small horizontal electric field induced by
Figure 10.1 Surface-wave propagation.
the transmitted signal (see Fig. 10.1). The vector sum of the large vertical
transmitted signal and a small induced horizontal field causes the resultant
wavefront to tilt over such that the Poynting vector (the direction of energy
flow) enters the sea at the Brewster angle t. One way to imagine this is to
picture the lower part of the wavefront having a slightly lower velocity due
to the water, dragging behind, and so bending the beam downwards. This
propagation mode has various names, but it is possibly best known as the
Norton surface wave.
The block diagram of a typical gw radar is shown in Fig. 10.2.
'Floodlight' illumination of a sector of sea is provided by a log-periodic
transmitting antenna, which has vertically radiating elements of varying
length to enable it to operate over a wide bandwidth. The receiving antenna
is a 100 m array of monopoles parallel to the coast, which are used to form
a number of beams simultaneously. If the system were installed in a confined
area such as on an oil-rig or ship, the transmitting antenna would almost
certainly have to be a monopole as well.
The transmitter is usually a wideband linear amplifier chosen from a
range of commercial HF communication equipment. Unlike pulse trans-
mitters, which can only deliver their peak power in short bursts, com-
munications transmitters are continuously rated, meaning that they can
deliver their specified power output continuously. The SNR of an HF radar
system can thus be improved by finding waveforms that extend the percentage
of time that the transmitter is on.
t The Brewster angle, named after a distinguished Scottish physicist of the last century. Sir
David Brewster, is best known in elementary optics. A beam of light striking a glass block at
a steep angle of incidence will enter and be refracted, whereas a beam striking at a very oblique
angle will be reflected from the surface. In between these two cases lies the Brewster angle, at
which the light runs along the interface between the glass and the air. In the case of gw radar,
the air/sea boundary acts in a similar way to the air/glass interface, and the radio wave travels
akmg the sea surface.
Low-phase
noise synthesizer
FIT AID Comms.

and Converter and Rx Beamformer t : = - - - - -
display pre-summation
~-----~y~------~
Similar to a low-
frequency spectrum anlayser'
Figure 10.2 A typical surface-wave radar installation.
The modulation used in HF radar may be a pulse, but frequency-

modulated interrupted continuous wave (FMICW) is also used, as this
makes communications transmitters work harder. Continuous-wave (CW)
modulation gets the highest mean power out of a transmitter, but it
gives no range information because it occupies an almost infinitely
narrow bandwidth. Frequency-modulated continuous wave (FMCW)
involves sweeping the frequency of the CW tone to gain bandwidth and
therefore range resolution. FMICW interrupts the transmission process (with
a roughly half-on, half-off ratio) so that a co-sited receiver can receive echoes
during the 'off' time. Bistatic systems, with a receiver some distance from
the transmitter, can make use of FMCW operation to transmit and receive
continuously.
Often, the receiver is also a wide band communications receiver, suitably
modified to accept the particular modulation chosen. The two outputs of
the receiver (one is needed to find the doppler shift and the other to determine
its sign) are digitized by an AID converter and stored in the memory of a
computer. For each range gate, the samples are added together coherently
for a period that might vary from a few seconds (in the case of missile or
aircraft detection) up to about 200 s for oceanographic observations. Beyond
200 s integration, the echoes from the sea add together less coherently.
The number of samples averaged in each range gate, before further processing,
is known as the coherent integration, predetection integration or pre-
summation figure n. After this integration, the reduced number of samples is
analysed by a fast Fourier transform (FFT) to produce the doppler spectrum.
If the FFT has M points (usually a power of 2), we can relate the number
of pulses transmitted during the integration time tj to the number processed
by the receiver as
(PRF) x tj = nM [ ] ( 10.1 )
Worked example An HF gw radar is required to operate up to a

maximum range of 300 km and must be able to integrate data up to
200 s. If the array processor can handle FFTs up to a maximum size of
1024 points, what pre-summation is needed?
SOLUTION The maximum range of 300 km suggests that a PRF of 500 Hz

be used. The maximum integration n that may be needed is given by
500 x 200
n = = 97.7
1024
COMMENTS In practice, these values would not be chosen. It is not only

easier to carry out pre-summation in powers of 2, but also preferable,
in order that n can be adjusted to compensate for changes in the FFT
length M without requiring changes to the PRF or integration time.
Also, 500 Hz would be an unwise choice of PRF because of harmonic
interference from 50 Hz European powerlines, and the unambiguous
range of 300 km could give rise to second-time-around echoes from the
F region of the ionosphere. An HF radar engineer would probably choose
325 Hz x 201.6 s = 64 x 1024 point FFT
In this way the integration time is easily changed by powers of 2 while
keeping the unambiguous range constant at 461 km, which is beyond E
and F region ionospheric problems.
In the case of FMCW or FMICW modulation, there is a primary FFT,

to sort the echoes into the appropriate range bins on the basis of their
frequency, followed by a second FFT for each range bin to develop the
doppler spectrum.
The extensive reliance of HF radar on digital processing, often carried
out in array processors, is the reason for its late development, although some
early systems used analogue filters (and analogue integration) to produce a
crude form of doppler spectrum.
The range of frequencies used by HF gw radar varies from about 3 to
15 MHz. At the lower end of the band, propagation over the sea is really
very good (at ISO km range the extra loss in addition to R4 is about 7 dB)
and the state of the sea makes little difference to this loss. However, there
are problems with the size of antennas, with low doppler shifts and with
unwanted reflections from the ionosphere. Propagation at higher frequencies
Range [km]
Figure 10.3 The additional two-way loss for propagation over a smooth surface: four-thirds
Earth, 40hm- 1 m-I conductivity. Frequency [MHz] is shown next to each curve.
deteriorates rapidly; at 15 MHz the additional loss for a range of 150 km is

42 dB and a rough sea increases these losses. Typical values for the two-way
loss over a smooth sea are shown in Fig. 10.3. Computer programs have
been developed that predict these losses for any range, frequency, sea state
and radar or target height.
One of the main problems with HF radar is the difficulty in finding a
sufficiently wide bandwidth clear of interference and communications traffic.
Channels are often allocated 3 kHz at a time and it is arduous trying to
operate with a bandwidth greater than about 10 kHz. Pulse durations are
thus of the order of 100 JlS and range cells are 15 km wide; this is larger
than is desirable for both military and oceanographic applications.
Dynamic range is also a problem for surface-wave radar. As the example
in Chapter 2 showed, dynamic ranges of up to 90 dB can be required if small
targets are to be detected against a background of sea clutter. The huge echo
from the sea at a low doppler shift, and the small echo from an aircraft or
ship at a doppler frequency only a few hertz away, is an extreme test of the
linearity ofa receiver. Indeed, one form of test used in receiver design involves
the injection of two closely spaced tones. The receiver must be linear over
a wide dynamic range, otherwise sum and difference frequencies would appear
in the spectrum as spurious signals, which would cause false alarms. The
linearity, dynamic range and wideband requiredments of OTH radar is the
reason why many systems are based around good-quality HF communications
equipment.
Despite all the problems of HF gw radar, the rewards for being able to
stay in the safety and comfort of the shore, and yet monitor the sea and sky
beyond the horizon, are such that there remains considerable interest in this
form of radar.
10.3 SKYWAVE RADAR
Skywave (sw) radar makes use of scattering from the ionosphere to look
down on a 'footprint' well beyond the horizon. The main applications lie
in detecting ballistic missile launches and tracking military and civilian air
targets. Because wind direction is relatively easy to extract from sea clutter,
storm tracking is also possible. Most skywave radars also have a significant
capability for detecting surface targets and for sea sensing. Cruise missiles
are not easily detected because they have a small RCS when viewed at long
wavelengths. The possible exception to this is submarine-launched cruise
missiles, which leave the water vertically and for a few moments may present
a large RCS before they settle into horizontal flight.
Unlike gw radar, skywave systems do not need to be near the sea and
are perhaps best located inland, where they are safe from storms, salt spray
and enemy attack. Skywave radars also tend to use up a lot of ground space,
and it is often easier and cheaper to find suitable sites inland, especially since
the receiver needs a radio-quiet location. The receivers and transmitters are
almost always separated, often by as much as 100 km, thus permitting the
use of FMCW modulation to maximize the mean power output of the
transmitters.
A typical arrangement for a sw radar is shown in Fig. 10.4. A number
of continuously rated high-power transmitters amplify an FMCW waveform
and drive a total power of about 500 k W into a 200 m long array. The type
of transmitting antenna is chosen to provide a good impedance match to
the power amplifiers over a wide bandwidth and to have a beamwidth
somewhat wider than the receiving antenna. There are two reasons for using
a relatively wide transmitting beam: first, the dwell time may be increased
if several narrow receive beams are used simultaneously to survey the area
illuminated by the transmitter array (Fig. 10.5); secondly, for a given antenna
gain product, the clutter cell size is smaller when the two antenna gains are
unequal.
Continuous scanning is not usually possible with sw radar because a
finite amount of time is needed to change the beamformer of the transmitting
antenna to a new position. It may also be necessary to retune to a new
frequency to illuminate a different range, in the same way that HF
communications engineers have to. For surveillance of a wide area, a step
scan technique is used, shown in Fig. 10.5, in which the radar illuminates
each sector for several seconds before moving on to the next. As is usual
scatter
receiving
station
Frequency FMCW
management waveform
system generation
"
"
"
"
"
'I
"
N "
~
- Figure 10.4 Typical skywave radar installation.
2
Sectors illuminated
by transmitting ,
•
------------ I
antenna \ ---------- ____ l ~
, ------------- I
t :::::::::::..-___---- J _ Sectors.s~rveyed
• _____----- __ ----, by receIVIng
-
, ------=-::.-.:.:: - antenna
Figure 10.S Step scanning is used to survey a wide area. Each sector is illuminated for a few
seconds and surveyed by four narrow receiving beams, before moving on to the next sector.
with surveillance, the objective is to scan a wide area and return to the
starting sector before a target can pass through it undetected.
The transmitting array elements are typically vertically polarized log-
periodic, although the North American CONUS-B system uses short, fat
dipoles canted 45° off-vertical ('fattening' a dipole increases its bandwidth).
The elevation pattern of the array depends on the impedance of the ground,
which in turn can be affected by the weather, so usually the ground is levelled
and some form of conducting wire mesh is laid down to stabilize the
impedance. The improved ground conductivity also helps to predict and
control the antenna sidelobes.
The receiving antenna is usually a long (greater than 1 km) array of
monopoles, sometimes with a backscreen to reduce the backlobe and
occasionally with the monopoles fattened to improve the match at low
frequencies. Often, antenna losses at low frequency are not important because
the system is externally noise-limited, meaning that the external noise
(galactic, atmospheric, man-made) arriving with the signal is greater than
the internal noise of the receiver. In these circumstances, some losses at the
antenna are permissible because noise and signal are attenuated equally and
the SNR remains unaffected. But the losses must not be allowed to become too
severe-in HF engineering it is regarded as something of a crime to permit
losses so large that a system becomes internally noise-limited.
The receiving array is connected to a beamformer. This may be an
analogue device, but more often these days this function is performed as a
digital process using one receiver plus A/D converter for each antenna
element, or for each sub-array group of elements. Analogue beamformers
steer the beam by introducing a frequency-independent time delay (see
Chapter 15); digital beamforming usually involves a phase delay that is
equivalent to the time delay at one specific frequency f (the operating
frequency). For this reason, phase delay beam steering has a restricted
bandwidth B, given roughly by
B / f [%] ~ ~8 [ degrees] (10.2)
where ~8 = antenna beamwidth. This bandwidth restriction also implies a
limit to the range resolution ~R and a similar working rule of thumb is given
by
~R ~ D sin 8 Em] (10.3 )
where D = length of the antenna array Em] and 8 = beam angle off-bore-
sigh t [degrees].
After the beamforming and receivers, the signal and data processing is
based extensively on the mapping of the doppler spectrum into range/azimuth
resolution cells. These cells are searched using an adaptive thresholding
process, and any targets detected are tracked making use of the doppler
measurements to compensate for the relatively poor range and azimuth
information. The azimuth resolution is poor because even a 1 beam diverges
0
to 35 km at a range of 2000 km. The range resolution is determined by the

FM frequency sweep, which in practice is limited to about 100 kHz,
corresponding to a range resolution of 1.5 km. Higher bandwidths would
run into problems with the array (see Eq. (10.2)) and the frequency-
dependent nature of ionospheric propagation, and it is not uncommon to
find lower bandwidths being used.
Ionospheric propagation over long ranges is a lossy process and large
power x aperture products are needed by sw radars. One method of
comparing these radars is to examine their power x aperture products PtGt
or PtGtGr [dB W], but sometimes the product of the transmitter power, the
total antenna gain and the coherent integration time is used; this is known
as the figure of merit or FOM [dB J]. Table 10.1 sets out these parameters
for the best known systems, as far as they can be determined from published
information. Descriptions of current military systems are to be found in
Jane's Radar 3 •
10.4 SKYWAVE PROPAGATION AND FREQUENCY

MANAGEMENT
In Chapter 8 we discussed the effect of the ionosphere on the propagation

of radio waves. The skywave radar engineer seeks to capitalize on these
N
0\
-
Table 10.1 Relative performance of the best known skywave radars
Resolution
Radar Power (typical)
Name or type P, Gl G, PlGlG, t FOM flO flR

Country of system [dBW] [dB] [dB] [dB W] [dB s] [dB1] [degrees] [km]
Australia 1indalee-B 52 21 32 105 17 122 0.5 20

Canada/USA RADC/SARA 42 10 22 74 17 91 3 3
China OTHB 50 18 26 94 6 100 5 15
France Valensole 24 20 26 70 25 95 22.5
UK Monostatic 30 21 21 72 17 89 4 75
Bistatic 40 14 22 76 4 80 5 75
USA CONUS-B 61 23 28 112 3 115 3 20
Madre 27 28 22 77 20 97 1 7.5
ROTH 53 21 34 108 1l? 119 0.5 ?
Sea-echo 51 24 24 99 14 113 2 15
WARF 43 20 30 93 11 104 0.5 0.8
USSR 'Woodpecker' 61 ? ? ? ? ? ? ?
effects, particularly refraction, to look beyond the horizon. Of the three

ionized regions in the ionosphere, D, E and F, only the E and F regions
turn out to be useful for sw radar. The free-electron content of the D region
(70-90 km altitude) is insufficient to scatter HF radio waves and the layer
acts as an unwanted absorber of radio energy. This absorption occurs because
the electrons, while trying to oscillate with the incident radio wave, experience
many collisions with neutral air molecules.
The E region is a narrow layer of ionization at about 110 km altitude.
The ionization is uneven, but can be very intense at times (owing to auroral
effects and intense patches known as sporadic E), and it is essentially a
daytime phenomenon. The maximum electron density of approximately 10 11
[electrons m - 3] occurs near midday and corresponds to a critical frequency
(called the foE) of 2.8 MHz, meaning that this is the highest frequency that
will be reflected at vertical incidence. Frequencies higher than foE pass straight
through the E layer if travelling vertically but can be scattered if they are
transmitted at oblique incidence.
The F layer often divides into two layers, Fl and F2, during daytime
(see Fig. 10.6). The lower Fllayer lies between 130 and 210 km altitude and
has a maximum ionization of about 2 x 1011 [electrons m - 3] at noon, giving
foF 1 = 4 MHz. The F2 layer usually has the strongest electron density at
about 10 12 [electronsm- 3 ] during the day (foF2=9MHz), and this
F2 region
E region
L-------f~o~E~~fu~F~1--------~fo~F~2~----+
Electron density or Critical frequency
Figure 10.6 The E and F regions of the ionosphere.

layer persists during the night with the ionization falling to around
5 x 10 10 [electrons m- 3 ] [foF2 = 2 MHz).
The bottom of the ionosphere is not smooth, but has a roughness
comparable with the scale sizes of the hills and valleys on the earth's surface,
with the added complication that the roughness is moving and changing
with the background neutral wind and electric fields. All three ionospheric
layers are subject to various disturbances, irregularities and anomalies, and
the electron densities may vary with time of day, season, location and solar
activity. The ionosphere is thus an uncertain and ever-changing medium of
propagation, and there may be many paths by which a signal can pass from
the transmitter to the target and back to the receiver-this can cause a single
target to appear at many apparent ranges (see Fig. 10.7). There is also a
limited band of frequencies that can be used to illuminate a target at any
given range and, as with HF communications, the band required may be
congested with radio traffic and high background noise levels.
'--__> Possible signal path -::::::::>_:::::::::::::::____~
F2 region
Fl region
iii' 50
x
:2.
c.:::
zVl x x
0
."
'"
>
.;;
-50
X
x x
u
'"
c.:::
4000
Apparent range [km]
Figure 10.7 Radio signals may follow any combination of E and F region outward and return
paths to give several apparent ranges and values of SNR for a signal target.
The answer to the problems of using the ionosphere for OTH radar lies
in a process known as frequency management. Extensive mathematical
modelling and continuous observations of the ionosphere are used to select
the best frequency band and then wide band 'look-ahead' or channel occupancy
monitoring receivers are used to find a channel free of interference. ,
Over the years, quite realistic mathematical models of the ionosphere
have been constructed (as computer software) to include both local data-
bases of observations built up over several solar cycles and maps of the
mean ionosphere produced by the CCIR (International Radio Consultative
Committee ). Statistically, these models provide a good prediction of expected
conditions, but on any given day the vagaries of the ionosphere are such
that the prediction can be substantially in error. These errors can be
minimized by ionospheric sounding, which measures the current state of the
ionosphere as an aid to selecting the most appropriate model. Occasionally,
part of the radar system itself is used for oblique sounding, but it is more
usual to use a vertical sounder, or ionosonde, a small pulsed radar system
that transmits upwards and sweeps in frequency to locate the critical
frequencies of the E and F layers.
The most common form of display for sounding data is an ionogram,
a plot of virtual height (not always equal to the true height because of the
slowing of radio waves in the ionosphere) against frequency, as shown in
Fig. 10.8. After some calculation, Fig. 10.8 can be replotted as true height
against electron density to produce a form suitable for ray tracing programs
to predict where signals transmitted on any given frequency will end up.
This is the information needed by sw radars to choose the best operating
frequency.
There are practical methods of ensuring that the frequency management
program is working as predicted. The radar operator can ensure that easily
Frequency [MHz)
Figure 10.8 A typical ionogram.

recognized targets (cities, coastlines, islands) are detected and appear in the
correct locations. Also known targets and transponding devices can be placed
within the coverage area to give confidence in the terrain illumination and
some feedback on the ionospheric absorption.
Ionospheric modelling, frequency management and calibration form a
major part of skywave radar operation and contribute to the expense of
these large surveillance radars.
10.5 THE OVER-THE-HORIZON RADAR EQUATION
The skywave radar equation is generally expressed (see Kolosov 4 , for

example) as
SNR = PtGtGr(J'),2{Js/it/ir
[ ] (10.4 )
(4n) 3 R~ R; FakTo
where Pt = mean transmitted power [W]; GtGr = antenna gains relative to
an isotropic radiator in free space [ j; (J' = target cross-section, not
free-space but as measured, i.e. including any surface reflection effects [m2];
A. = radar wavelength Em]; ti = coherent integration time [s]; Is = system
loss factor [ ]; lit, lir = additional path losses on outward and return paths
due to D and E region absorption [ ]; R t , Rr = outward and return
distances to target Em]; and FakTo = apparent external noise level [W Hz-I],
with Fa the effective antenna noise factor, usually expressed in dB above kTo,
i.e. above - 204 dB W Hz - 1. The bandwidth is assumed to be 1/ t i •
Except during auroral absorption the loss factors lit and lir may amount
to only a few decibels during the day and are near zero at night (details of
how to calculate these losses may be found in reference 4 ). System losses are
more severe and are due to a variety of causes related to the antenna design
and signal processing. Typical system losses are listed in Table 10.2 and,
while they may vary from one system to another, they seldom fall below 10 dB.
The effective antenna noise factor Fa (the result of external noise)
increases with increasing wavelength, and over the range 3-15 MHz a
realistic value to choose for back-of-the-envelope performance estimates is
Fa ~ 60 - 2fMHz [dB] (10.5 )
so that the noise level in a bandwidth B [dB Hz] is given by
Next = 60 - 4fMHz - 204 +B [dBW] (10.6 )
More detailed formulae, representing different noise environments in the
USA, are to be found in Skolnik 5 . The classic work in this field is the CCIR
Report 322 6 , which, although published in 1963, remains a very useful
database of global atmospheric noise levels.
Table 10.2 Typical over-the-horizon radar system losses
Loss Typical value
Transmitter feeder and voltage standing-wave ratio losses 1.0 dB

Transmit antenna ground and resistive losses 0.5dB
Transmit antenna weighting loss 0.8 dB
Target not in centre of the transmit beam
Azimuth 0.9 dB
Elevation 0.5dB
Target not in centre of the receive beam
Azimuth 0.9 dB
Elevation 0.5 dB
Receive antenna weighting loss 1.1 dB
Receive antenna ground loss 1.0 dB
Degradation by internal noise sources and intermodulation products 0.5 dB
Eclipsing losses due de-ramping (only correctly timed at one range)
-pulse radars use protection gating giving a similar eclipsing loss 0.9 dB
Receiver filter and digitization losses 0.2 dB
Range weighting loss over one frequency modulation sweep 1.2 dB
Target not in centre of range bin 0.5 dB
Fast Fourier transform weighting loss in doppler processing 2.2 dB
Target not in centre of fast Fourier transform doppler bin 0.4 dB
Range and doppler smearing due to target movement during dwell time 0.6 dB
Loss due to constant false-alarm rate thresholding process 1.3 dB
Total system loss Ls for over-the-horizon radar 15.0 dB
Worked example Could the CONUS-B system detect an aeroplane with

a 15 m wingspan at a range of 2000 km ?
SOLUTION The size of the aircraft places it in the resonance region at
HF wavelengths and the greatest RCS would be observed when the
operating frequency was such that the wings formed a short-circuited
half-wave dipole. In this case the frequency would be 10 MHz and the
RCS is 0.86..1 2 (Table 2.2), giving (J = 28.9 dB m 2 . The radar equation
would now look something like:
PIGIGrt i = 115.0dBJ FOM, see Table 10.1
(J = 28.9 dB m 2
..1 2 = 29.5 dB m 2
ls= -15.0dB Typical, see Table 10.2
IiI' lir = - 6.0 dB May be worse at times
1/(4n)3 = -33.0dB
1/(R~Rn = -252.0dB m- 4
1/(FakTo) = + 164.0dB W- 1 Hz
SNR = 31.4dB
Under these conditions the aircraft would be detected. The CONUS-B

data processing software can probably detect air targets with a SNR as
low as 6 dB..
COMMENTS In practice, the interference between signals on different

ionospheric paths would cause the apparent RCS to fluctuate, and at
times the aircraft track might be lost. At night, with only weak F layer
propagation paths available, sw radars move to lower frequencies
(together with much of the HF communications traffic) and the RCS
of the aircraft would fall, making target detection more difficult.
The ground-wave variant of the radar equation differs from the sw version
because the presence of the conducting ground modifies the antenna gains
and creates the ground-wave effect (see Hall and Barclay2 and Shearman in
Scanlan 7). A frequently used version of the gw radar equation is
SNR = ~G;G;a'tJs161/
[ ] (10.7)
(4,.)3R4F.kTo
where G;G; = equivalent free-space gain of the antenna relative to isotropic

(the actual gain of the antenna is height-dependent, but for an antenna
at ground level the gain would be measured as 4G'); (J' = free-space RCS
of the target [m2]; and lp = additional loss over R4, including sea-state losses.
It is desirable to produce a common gw and sw form of the radar equation
to avoid the type of confusion that can occur when a mixed mode of operation
is used, such as a target illuminated by a distant sw radar but detected by
a nearby gw receiver. The propagation factor (4nRj..1Y can be replaced by
the one-way transmission factor f (usually provided by computer modelling
of the gw or sw path) to derive the following form of the radar equation for
use in all OTH situations;
SNR = 4nPt Gt Gr (Jmt;ls

[ ] (10.8 )
FakToUh)' )2
where Pt = mean transmitted power [W]; GtG r = antenna gains as measured

in situ, relative to an isotropic antenna in free space [ ]; (J = free-space
RCS [m2]; m = RCS modification factor [ ] (for gw, m = 1; but for sw,
m can lie between 0 and 16 depending on the propagation paths, the mean
sw value being 4 ); f = one-way propagation parameter, reducing to ( 4nR j ).) 2
at short ranges [ ]; and h = target height factor [ ] (for targets on the
ground; h = 1; for elevated targets, h =. 2). In this definition of the radar
equation, the backscatter RCS of the fully developed sea surface should be
taken as -23 dB m 2/m2 •
10.6 PROBLEMS WITH HIGH-FREQUENCY RADAR
Besides the propagation difficulties, there are four other problem areas
encountered when operating both gw and sw HF radar. First, there are
problems associated with licensing powerful HF transmitters and with getting
the frequency allocations clear of other traffic that are wide enough to give
good range resolution. These difficulties should not be underestimated.
Secondly, there are often siting difficulties. It can be difficult to find
sufficiently large and level radar sites that meet the requirements of being
radio-quiet at the receiving end and of presenting no sensitive environmental
issues at the transmitter location.
Thirdly, both gw and sw radar experience ionospheric clutter (unwanted
backscatter from the moving ionosphere) and meteor clutter, which can at
times swamp the presence of wanted echoes. The extent of the meteor echo
problem can be evaluated using the equations given in Chapter 9. Skywave
radars can also suffer significant polarization losses and focusing/defocusing
due to ionospheric effects.
Lastly, military HF radars are susceptible to deliberate jamming and
only modest powers are required to disable them, even from relatively long
ranges. However, OTH radars are usually classed as early warning devices,
rather than as accurate tracking radars, and so they probably fulfil their
warning role to some extent if they report that electronic warfare has been
initiated.
10.7 SUMMARY
There are two forms ofOTH radar, surface-wave and skywave. Surface-wave
systems are relatively inexpensive and have found applications in sea sensing
(see next chapter), for the defence oflocalized areas against low-flying missiles
and to some extent for monitoring ship traffic.
Skywave radars are used to monitor very large areas of land and sea to
search for air targets, ballistic missiles during launch phase and some types
of surface target. They also have remote sensing capabilities, especially storm
tracking and ocean wave monitoring. These radars are large, powerful,
expensive and require sophisticated frequency management systems in order
to operate via the ever-changing ionosphere.
Key equations
• For pre-summation and fast Fourier transformation:
(PRF) X tj = nM []
• Restriction for bandwidth in phase delay beam steering:

B/ f [%] ~ dO [degrees]
• Approximation for effective antenna noise factor:
[dB]
• Maximum radar cross-section of a short-circuited dipole (such as aircraft
wings) or a vertical monopole shorted to the surface (such as a missile
rising out of the sea):
Res = O.86A,2
• The radar equation in a form suitable for use in all over-the-horizon
situations:
10.8 REFERENCES
I. CH-the first operational radar, B. T. Neal, GEe J. Res., 3(2), 73 - 83, 1986.
2. Radiowave Propagation, Ed. M. P. M. Hall and L. W. Barclay, Peter Peregrinus for the
lEE, Stevenage, Herts, 1989. [There are chapters on both gw and sw propagation.]
3. Jane's Radar and Electronic Warfare Systems 1989-90, Jane's Information Group, London,
1989.
4. Over-the-Horizon Radar, A. A. Kolosov et al., Artech House, MA, 1987. [This book attempts
to layout the foundations of OTH radar engineering, but the coverage is patchy and biased
towards single-hop skywave backscatter systems. The WARF radar is extensively cited.]
5. Radar Handbook, Ed. M. I. Skolnik, McGraw-Hili, New York, 1990. [Chapter 24 is dedicated
to HF OTH radar.]
6. World Distribution and Characteristics ofAtmospheric Radio Noise, CCIR Report 322, 1963.
7. Modern Radar Techniques, Ed. M. J. B. Scanlan, Collins, Glasgow, 1987. [Chapter 5 is a
useful review of skywave propagation, clutter problems and HF remote sensing by Professor
E. D. R. Shearman.]
8. Real-time sea-state surveillance with skywave radar, T. M. Georges, J. W. Maresca, J. P.
Riley and C. T. Carlson, IEEE J. pcean Eng., OE-8(2), 97-103,1983.
Over-the-Horizon Radar, Y. A. Mischenko, Zagorizontnaya Radiolokatsiya, Military Publishing

House, Ministry of Defence, Moscow, 1972. [The principles of skywave radar are
introduced, a number of operational systems examined and some interesting military
applications discussed.]
Introduction to Radar Systems, M. 1. Skolnik, McGraw-Hili, New York, 1985. [Headrick and
Skolnik made one of the first skywave detections of ships in 1974. Consequently, the section
in Chapter 14 is an authoritative review.]
10.10 PROBLEMS
10.1 SRI International have used the Wide Aperture Research Facility (WARF) in California
to explore the benefits of high range and azimuth resolution for skywave OTH radars. A unique
feature of WARF is a 2.5 km long receiving aperture. The system operates over the range
6-30 MHz but, for all the problems below, assume a typical operating frequency of 15 MHz:
(a) What is the azimuth beamwidth?
(b) The elevation beam width is 36° between half-power points. What is the antenna gain?
(c)Two 10 kW transmitters feed an antenna of gain 20 dB. What is the total power x aperture
product p'G,G,?
(d) What is the figure of merit when an integration time of 12.8 s is used?
10.2 (a) If the transmitter in problem 10.1 is of the FMCW type and sweeps from 15.000 to
15.050 MHz, what is the area of water surveyed by the radar at a range of 1500 km?
(b) Using the definition in Eq. (10.8) of the reflectivity of sea water as -23 dB m 2 m- 2 ,
what would be the RCS of the sea echo?
10.3 What noise level would you expect to find in the problem above?
10.4 What SNR should be received for the sea echo in the problems above? Assume the
propagation term (fh)') is 10 dB worse than (4nR)2 /)..
10.5 An aircraft is detected in the same resolution cell as the sea echo in the problem above.
If the tracker reveals the aircraft to be approaching the radar with a doppler shift of 15 Hz and
a SNR of 23 dB, what can you deduce about the type of aircraft?
CHAPTER
ELEVEN
RADAR REMOTE SENSING
• Measuring waves and currents

• Microwave scatterometry
• Microwave altimetry
• Synthetic aperture radar and its applications
The way radio waves scatter from the earth's surface can be used to
investigate large-scale features of land and sea.
11.1 HIGH-FREQUENCY RADAR SCATTERING FROM

THE SEA
Random, noise-like processes are often interesting, but the surface of the sea
has a particular fascination. People stare at it for hours. For a radar engineer,
the challenge is first to find out how radio waves are scattered from the sea
and then to turn the problem round and ask, 'How can we determine what
is happening at sea from our radar observations?' In some ways, this type
of inverse problem is like being given an answer and having to find the
question.
What is an ocean wave? We must answer this question before we can
make any progress with the problem. Wave motion is carried by particles
of water exhibiting circular motion as the wave travels past. This is easily
demonstrated by floating a cork, or small piece of wood, on the sea and
watching it move as a wave travels by. At the surface of the sea these circular
motions can have quite large amplitudes, but they quickly die away with
226
RADAR REMOTE SENSING 227
Wave propagation
•
Figure 11.1 The trochoidal shape of sea waves generated by the circular motion of water.
depth (see Fig. 11.1). Submarines soon avoid the effects of wave motion
when they submerge.
Figure 11.1 shows why the shape of the wave itself is not a sinusoid but
is a rather flat-bottomed/pointed-crest type of shape known as a trochoid.
The height of a sea wave is small compared to its length; a typical wave
steepness is 1/18 for open ocean sites, with 1/10 rarely exceeded.
Waves are generated by the wind either locally or by distant storms.
These latter waves, known as swell, have long wavelengths, which enable
them to travel great distances with low attenuation. Locally driven waves
are more sophisticated. The wind blows over the water and begins to form
short-wavelength waves, but as these build up, and processes become
non-linear, the energy is transferred into longer waves with larger amplitudes.
For a constant wind speed, an equilibrium is reached with the largest waves
having a speed close to that of the wind. If the wind strengthens, then longer
and larger waves will be generated, giving the type of spectrum shown in
Fig. 11.2.
30 knots = Beaufort scale wind I).

giving rise to sea state I)
20 knots=sea state 4
\0 knots
0.2
Frequency [Hz)
Figure 11.2 A typical waveheight spectrum. This is a non·directional spectrum because it gives
no information about wave direction. In fact, the largest waves tend to be aligned with the
driving wind force but the smaller waves are omnidirectional, giving rise to the complex 'choppy'
look of the sea surface.
- - - - - - - - ~----- -- - - - -------
--- -------- --~----- - -- ~ -------=-----: -----~--------- --- -------------=-
Figure 11.3 When the sea wavelength equals half the radar wavelength, the backscatter from
one crest has a path length shorter by a whole radio wavelength than the backscatter from the
next crest.
High-frequency (HF) radar, with its capability for making over-the-

horizon measurements of a wide area, would seem to be an ideal way of
investigating ocean waves from the safety of dry land; but how does the
chaotic spectrum of sea waves appear when viewed with the long-wavelength
'eyes' of an HF radar?
The spectrum of Fig. 11.2 shows that all waveheights (and wavelengths)
are present up to some maximum value. Any of these waves which are travel-
ling directly towards, or away from, the radar will give rise to backscatter, but
those having a length equal to half the radar wavelength (AJ2) will give a
much stronger echo because the reflected radio waves add together coherently
as illustrated in Fig. 11.3. This selectivity in the echoing process is known as
Bragg resonant scatter, after W.L. Bragg, who first proposed this mechanism
to explain the scattering of X-rays from crystals. Bragg scattering is equally
important for microwave systems.
Worked example A narrow-beam groundwave HF radar, operating at

10 MHz, has a 15 km range cell size. How much stronger is the echo
arising from Bragg resonant scattering than the contribution from waves
of similar wavelengths?
SOLUTION At 10 MHz, A/2 = 15 m and therefore 1000 sea waves will

be present in each range cell. The contribution from n coherent terms
is proportional to n 2 , whereas incoherent terms add up proportionally to
n. The echo power from the Bragg resonant waves will therefore be
(1000)2/1000 = 1000 times, or 30 dB, greater than from non-resonant
waves of similar amplitude.
COMMENT Since most radar systems use range cells many times larger
than the radar wavelength, this enhancement due to Bragg scattering
becomes the dominant mechanism irrespective of the radar frequency
used.
/
/
/ e
~-------------
L"
--
--- -- -- --"'------ ---------- --=---::::- -- -- ----- -- -
-
Figure 11.4 Bragg resonance occurs when the path difference between the echo from two wave
crests is a multiple of }./2.
We now work out the doppler shift of the Bragg resonant sea waves, as
observed by the radar. To do this, three pieces of information are needed.
The first of these is the Bragg formula, which should be slightly more general
than the case shown in Fig. 11.3 because the radio wave could be incident
on the sea surface at an oblique angle. The full case is shown in Fig. 11.4
and the following equation:
Ln = nA/(2 sin 0) [m] (ILl)
and for groundwave radar sin 0 = 1. Note the possibility that signals
back scattered from the sea waves of wavelength 2 x )./2, 3 x )./2, etc., will
also add coherently at the receiver.
The second piece of information needed is the dispersion relation for sea
waves, which relates the phase velocity Vn to the sea wavelength Ln:
v" = J (g L n/ 2n ) ( 11.2)
where g is the acceleration due to gravity.
Finally we need the doppler formula of Eq. ( 1.20), derived in Chapter 1:
in = ±(2v" sin 0)/). [Hz]
Putting these together (an exercise left to the reader!) gives
in = ± J(ngfra:~ sin 0) [Hz] (11.3 )
At first Eq. ( 11.3 ) appears to be a surprising result, because the doppler

shift from the waves has a square-root dependence on the radar frequency
iradar = c/ )., whereas for a 'hard' target (such as a ship or aircraft) In is
directly related to iradar' On further consideration, it is apparent that changing
the radar frequency causes a different set of sea waves to be selected by the
Bragg process and these waves have a velocity with a square-root dependence
in the dispersion relation. This distinction between waves and hard targets
o Approaching Bragg line
-10
a:l
/Yz X Bragg
~
Receding Bragg line /21 x Bragg
t -20
~
oQ.
<l)
.~ -30
.,
(;j
0::
-40
o 2
Doppler/Bragg frequency
Figure 11.5 A typical doppler power spectrum recorded at 8 MHz.
is a useful property when tracking ships with HF radar. If a ship is travelling

at the same speed as the dominant Bragg waves, it usually cannot be detected
because the sea echo is much larger than that of the ship. However, a change
in radar frequency will cause the two echoes to be separated in the doppler
spectrum.
When n = 1 in Eq. (11.3) we call this the 'first-order' doppler spectrum
and it is the most conspicuous of the signals in the radar output. The features
of a doppler spectrum can be observed by attaching a spectrum analyser to
an HF radar and using an integration time of around 100 s to display the
spectral output. A typical example is shown in Fig. 11.5, where the first-order
spectrum consists of the two prominent 'Bragg lines' that represent the
velocities of the advancing and receding Bragg resonant waves.
11.2 MEASURING OCEAN CURRENTS
A surprising amount of information can be derived from an inspection of

the Bragg lines in a doppler spectrum such as that shown in Fig. 11.5. The
two lines are not generally of equal amplitude and the difference is related
to the wind direction, such that a wind blowing predominantly towards the
radar will cause the approaching Bragg line to be greater than the receding
line. The ratio of the Bragg lines is sometimes exploited by skywave radars
in order to plot wind maps over large areas of ocean, especialiy when there
is danger from a hurricane.
The effect of a tidal current in the sea is to shift the whole of the doppler
spectrum by an amount equal to the radial component of the drift. In practice,
current measurements are often made by measuring the displacement of just
the two Bragg lines. A single radar can only measure the radial component
of the current and it is common practice to use two separate radars, with
overlapping beams, in order to derive the full current vector information.
There are now several commercial HF radar systems for measuring tidal
currents. Even a short-range system, measuring only coastal currents, can
have many applications because of the need to understand more about such
problems as coastal erosion, sandbank movements, sewerage distribution,
warm effluent distribution (from power stations), docking large container
ships and stress on dykes during tidal surges.
The measurement of the Bragg line displacement represents a weighted
average of the current over the top layers of the ocean. The radio wave itself
does not propagate far into the water (the attenuation with depth being
about 60 dB per metre at 3 MHz and 189 dB per metre at 30 MHz). However,
the orbital motion of the water particles, shown in Fig. 11.1, means that the
radar is sensitive to subsurface currents at depths up to 4 m at 3 MHz and
0.4 m at 30 MHz.
Surface currents measured remotely by HF radar agree well with in situ
measurements made by current meters and drift measurements and are
accepted as being reliable. HF radar also has the advantages of being able
to survey a wide area of sea simultaneously and of being unaffected by bad
weather and rough seas. It is sometimes possible to gain further insight into
small-scale surface current structures by examining any broadening of the
Bragg lines or by using several frequencies simultaneously.
11.3 MEASURING WAVES
The extraction of wave information from a doppler power spectrum is quite

difficult and depends upon the mechanisms that generate the remainder of
the spectrum after the Bragg lines have been explained. There are essentially
two classes of second-order processes at work.
The first mechanism arises because the waves are not sinusoids but have
the trochoidal form shown in Fig. 11.1. This shape may be represented as
a fundamental sine wave plus harmonics all travelling at the same speed.
Bragg resonant scatter can take place not only with the principal wave (n = 1
in Eq. (11.3)) but also with the harmonics of longer waves (n = 2, 3, ... ).
J J
The speed of these waves corresponds to 2, 3, ... times the velocity of
the first-order Bragg line waves. In practice, the first of these harmonics is
quite frequently observed in the doppler power spectrum; for example, such
harmonics can be seen in Fig. 11.5.
The second type of process is an electromagnetic effect in which radio

waves are scattered back to the radar from two sea waves that are travelling
at right angles to each other, reminiscent of the way a corner reflector works.
This echo is strongest when the radar bore-sight bisects the 'corner' and
gives rise to a feature in the doppler spectrum that is at 2 3 / 4 times the Bragg
line velocity. This effect can just be discerned on the positive side of
Fig. 11.5. The remainder of this figure is a continuum of echoes that fall away
either side of both Bragg lines in a way reminiscent of a sea wave spectrum
such as that shown in Fig. 11.2. Indeed, the continuum is generated by sea
waves of all sizes, present in the proportions given by the sea spectrum.
The forward problem is not too difficult; given a knowledge of the wind
and sea-state conditions, it is possible to predict fairly reliably the doppler
power spectrum that an HF radar will record. The inverse problem is much
harder because the formula relating an observed doppler spectrum to an
unknown spectrum of waves on the sea surface is a non-linear integral
equation. This type of equation is notoriously difficult to invert. In some
ways it is like trying to evaluate a polynomial such as y = a + bx + cx 2 + ....
It is easy to calculate y, given the values on the right-hand side, but try
going the other way and solving for x, a, b, c, ... for a given value of y. There
are obviously many solutions that will add up to the same value of y.
For the HF radar operator there are some added difficulties:
1. The sea-wave spectrum is two-dimensional, whereas observations from
a single radar are one-dimensional.
2. Radar echoes are contaminated by noise and interference.
3. The sea spectrum is an average description of the sea surface, but any
given observation will see a random surface that will not be exactly
represented by the mean.
4. The whole equation is ill-posed in the sense that it contains singularities
and ambiguities and because large changes in the sea conditions may
cause only small changes in the observations.
Despite all these problems, mathematicians have worked towards finding
methods of extracting wave information from HF radar observations and
useful techniques are now emerging. There are plenty of incentives to succeed
because of the interest in accurate long-term wave measurements. For
example, the offshore oil industry would like to be able to predict the largest
wave likely to occur in a 50- or 100-year period. Oil-rigs having platforms
lower than this extreme wave height are in danger of being swamped, but
building rigs excessively high increases the cost by millions of pounds for
each extra metre of height added.
Although waves have been measured for many years by tethered buoys
and ship-borne systems, these methods are not without their own problems,
and they do not provide the very wide instantaneous coverage and long-term
prospects offered by radar remote sensing.
11.4 THE FUTURE OF HIGH-FREQUENCY REMOTE

SENSING
Bistatic Operation
Most HF sea sensing systems now use at least two radars to observe a single
patch of water in order to increase the amount of information on the
two-dimensional sea wave spectrum. If two radars are operated together,
further information can be derived by also operating them bistatically such
that transmissions by one are received by the other. This arrangement gains
a third view of the sea spectrum along a line that bisects the angle between
the two radar beams. In future, this may help to increase the accuracy with
which wave information can be obtained from HF radar observations.
Sea Ice
Detecting the presence and movement of sea ice is possible using HF radar,
and to some extent the thickness of ice may also be estimated. The penetration
of radio waves into sea ice is somewhat variable because ice is a complicated
substance whose dielectric properties depend on such things as the tempera-
ture, the age of the ice and the amount of brine trapped within it.
Skywave mapping of the Greenland ice cap shows that thin ice has a
reflectivity similar to that of sea water, and the contour of reflectivity 10 dB
lower corresponds to an ice thickness of 1000 m. Groundwave observations
show pack ice reflectivities between 2 and 10 dB greater than the reflectivity
of the sea, and icebergs appear to be 12 or 13 dB greater.
Ice is a 'hard' target, and drift speeds may be inferred directly from the
doppler shift-just as for a ship or an aircraft-although long integration
times may be necessary to detect slow ice movements. An alternative
approach is to plant transponders on the ice to increase the accuracy of the
velocity measurements by imposing a small calibrated frequency shift on the
echo. This frequency shift makes it easier to distinguish the ice movement
from echoes due to 'stationary' targets such as land, stationary ice,
second-order sea clutter and transmitter-receiver breakthrough.
Worked example If the doppler power spectrum shown in Fig. 11.5 were
obtained from an HF groundwave radar operating at ! MHz and an
echo from a transponder antenna were just visible as a spike at the centre
of the spectrum, what would be the improvement in signal/sea clutter
ratio if the antenna impedance were modulated such that half the signal
appeared at a doppler shift of 0.85 Hz?
SOLUTION Substituting 8 MHz into Eq. (11.3) shows that the Bragg
lines in Fig. 11.5 must have a doppler shift of 11 = 0.29 Hz (sin (J = 1
for groundwave radar). A doppler shift of 0.85 Hz is close to three times

the Bragg line frequency and it would move the transponder echo from
the centre of the spectrum to near the edge. At the centre the second-order
sea echo is at a power (relative to the peak of the advancing Bragg line)
of about - 27 dB. Near the edge of the spectrum, the sea clutter
has fallen to about -46 dB, which is 19 dB lower. Only half the
transponder echo appears in this sideband, giving a signal/clutter gain
of 19 - 3 = 16 dB.
At high latitudes, the mapping of ice movement and the tracking of

icebergs by HF radar is likely to remain a subject of considerable scientific,
and possibly commercial, interest.
Ship Tracking
A necessary part of wave sensing is the removal of 'ship clutter' appearing
as spikes in the doppler spectrum and which may be confused with
second-order sea echo features. Conversely, when tracking ships, it is necessary
to remove sea clutter. There seems to be a good case for developing integrated
systems with two-stage data processing. First, recorded data would be
scanned for the presence of ship echoes, which would then be tracked through
the recordings, removed and the information transferred to the appropriate
users. Secondly, the ocean wave and current information could then be
derived from the cleaned-up data, and distributed to interested parties.
Besides the obvious military interest in ship detection and tracking, there
are also civil applications for the control of shipping. This already takes
place in small, densely populated shipping areas using microwave radar
systems, but there is no long-range control of shipping equivalent to air
traffic surveillance.
With the wide range of applications for HF radar remote sensing and
the possibility of combining many of these functions into a single system, it
is likely that this will remain an active and interesting field of radar
research for many years.
11.5 MICROWAVE SCATTEROMETRY
High-frequency (HF) radar may offer continuous monitoring of sea conditions

in a local area, but it can never hope to provide the global coverage available
from satellite-borne systems. Satellite- and air-borne remote sensing frequently
involves the use of microwave radars, and one of the simplest and yet most
useful of these instruments is the scatterometer.
Microwave scatterometers are radars that look down at the earth's
surface from above, normally at oblique angles of incidence between 20° and
60°. Their purpose is to make accurate and precise measurements of the

radar cross-section (Jo of extended targets. The targets may be natural, such
as the ocean surface, snow, vegetation, etc., or man-made, such as tarmac
or concrete runways. Such targets are of interest in their own right for the
purposes of remote sensing, or can be considered as the clutter background
against which hard target detection is carried out. Scatterometers typically
operate in the gigahertz part of the radio spectrum, and can be mounted on
ground-based platforms, such as towers or cranes, or on air-borne or
space-borne platforms. They have been used for extensive studies of the
variation of (Jo with frequency, polarization and incidence for many types
of target!.
In order to make accurate measurements of (J0, scatterometers need to
be carefully calibrated. Methods of doing this are discussed by Ulaby et al. 2 •
In order to achieve precision, they need to counteract the severe fading that
is characteristic of observations of the RCS of clutter. Single measurements
may give a very imprecise estimate of the mean RCS. In order to get round
this problem, scatterometers must be designed to average or integrate many
independent measurements. The independent samples may be achieved using
temporal, spatial or frequency sampling.
For a scatterometer carried on a moving platform, the doppler spectrum
is normally used to obtain independent samples. This is a comparatively
complicated topic, and the reader interested in a full treatment would be
advised to consult volume 2 of Ulaby et aU. However, we can easily explain
the basic idea. Let us assume that, as a result of platform motion, doppler
frequencies of bandwidth B are returned to the scatterometer antenna (we
have already discussed how this would occur in Chapter 6, in the context
of SAR). To obtain the greatest possible resolution from the system, the full
bandwidth would be used (this is what a SAR does). Another option is to
divide the available bandwidth into M non-overlapping frequency bands of
equal width. Each sub-band can be used to perform a measurement of the
RCS. This gives M independent (because the frequency bands do not overlap)
measurements of the RCS, which can be added to achieve greater precision
in the estimate. Since the bandwidth used per measurement is decreased by
a factor M, the resolution is degraded by the same factor.
Worked example When we are using square-law detection, and the

clutter is due to many independent scatterers per resolution cell, the
power returned in a single measurement will be exponentially distributed,
I.e.
for 1 ~ 0
Here T is the average power we want to estimate. We saw in Chapter 4
that, for this distribution, the standard deviation and mean both have
the value T, and that single measurements are likely to give an unreliable
estimate of T.
If we average M independent measurements from this distribution,

the mean is unchanged, but the standard deviation is reduced to II M. J
The PDF of the average is a chi-squared distribution with 2M degrees
of freedom, which is tabulated in most statistics books. For M of the
order of 10 or greater, this distribution is approximately gaussian with
J
mean I and standard deviation Tj M. How many samples must we
average in order to have a 95 per cent probability of estimating the mean
power T to within TjIO?
To answer this, we need to write the spread of values about the
mean in terms of multiples of the standard deviation. Clearly 1110 is
equivalent to .JM
JIO standard deviations from the mean. From
Fig. 11.6, we can see that we need each of the shaded areas in the gaussian
2.00
1.50
1.00
0.50
-1.0 -0.5 o 0.5 1.0 1.5 2.0 2.5
Figure 11.6 The approximate gaussian PDF obtained by averaging measurements from
exponential clutter; the shaded areas must have combined area less than 5 per cent to give 95
per cent chance of estimating the mean within a tenth of its value.
PDF to be of area 0.025. Using Table 4.1 and the associated figure, we
can see that this occurs when (f)( t) = 0.475, so that the distance from
the mean is 1.96 standard deviations. Hence JM = 19.6, or we need
384 independent samples. Such large sample sizes are normal if very
precise measurements are needed in exponential fading.
Scatterometry is likely to play a major role in current efforts to

understand the ocean-atmosphere interactions that playa large part in
determining our weather and climate. This is because the RCS of the ocean
is dependent on the surface wind velocity (as well as the incidence angle and
polarization of the incident wave). Several different formulae have been
derived to describe this relationship, all of them empirically derived from
observations that are themselves usually hampered by an inadequately
detailed description of the wind over the area of water illuminated by
the radar.
The relationship between wind speed and measured RCS was first
demonstrated using air-borne scatterometers operated by NASA and the US
Naval Research Laboratory. Its operational potential was established by
observations from the Skylab and Seasat satellites (see Ulaby et al. 2 for
references and a general account of these developments). Wind vector
scatterometery requires radar measurements made in two or more directions
to determine the wind vector. Seasat carried a four-beam scatterometer with
the beams oriented fore and aft looking at an angle of 45° to the track on
either side of the track. This geometry led to a 180° ambiguity in the estimated
wind direction. Nonetheless, it was possible to produce global wind field
maps over the earth's oceans using measurements from the Seasat scatter-
ometer. A three-beam scatterometer operating at 5.3 GHz is one of the main
instruments carried on the European Space Agency's ERS-l satellite, which
was launched on 17 July 1991. The beams are oriented broadside and at 45°
to the track (see Fig. 11.7 overleaf) in order to avoid ambiguities. This will
provide surface wind measurements in a continuous strip 500 km wide
parallel to the satellite track with 50 km spatial resolution. Such measure-
ments can be incorporated into models of atmospheric circulation, and will
form an important part of the World Ocean Circulation Experiment.
11.6 RADAR ALTIMETRY
Another source of information on ocean dynamics from satellites is provided

by altimeters. These are radar systems designed to produce very precise
measurements of the two-way delay suffered by a radar pulse as it propagates
vertically to and from the earth's surface. Combined with a knowledge of
the satellite orbit, this permits very precise measurements (of the order of
i 0 cm or less) of surface heights, as long as the surface can be regarded as
,/
\ /
\ /
/
225km
~~~~rometer ~\
/
/
/ /
/
antennas \ /
/ \ /
/ 500km
/ \ /
// \ /
/ \ /
/
/ \
// /
k----"'-----'( Aft beam
2}/
~'1i'/
x,~ /
0/
Figure 11.7 The geometry of the space-borne scatterometer carried on the ERS-l satellite.
approximately planar within the radar footprint. This is the case over the
ocean and the earth's great ice shelves. One such device was carried on the
Seasat satellite. It provided remarkable images showing that the ocean surface
has a topography reflecting the trenches and submerged mountain ranges
in the deep oceans. Large-scale currents such as the Gulf Stream can also
be extracted from the data because they give rise to deviations from the
mean height.
An altimeter operates by emitting a pulse vertically downwards. A
pulse-limited altimeter uses a very short pulse, and the sequence of events
that follow is as outlined below:
1. The leading edge of the pulse strikes the surface, and a return signal
begins to propagate back to the receiver.
2. The illuminated area grows like an expanding disc, until the trailing edge
of the pulse reaches the ground. As the illuminated area grows rapidly
in size, so does the returned power.
3. The illuminated area now assumes the form of an annulus, which

propagates outwards along the surface, while preserving its area and
hence the returned power. The power received at the earth's surface
begins to diminish when the leading edge of this spreading annulus
reaches the edge of the main beam, defined by the antenna pattern of
the transmit antenna. This in turn leads to a decline in the returned power.
This sequence and the associated returned power is illustrated in Fig. 11.8.
The ERS-l satellite carries an altimeter that operates at 13.8 GHz, using
an antenna of 1.2 m diameter. In order to avoid the generation of very short
pulses and the associated rapid processing, a slightly different operating
strategy is used. A chirped pulse is transmitted, and the analysis of timing
and shape is carried out in the frequency domain. The essential principle is
unchanged. Because the altimeter operates on a single frequency, the
correction for atmospheric delay (see Chapter 8) must make use of iono-
spheric models or ancillary data.
The returned waveform in fact yields more information than just the
time delay to the surface. Over the ocean, which is anything but a flat surface
Radius of annulus
Trailing edge grows, but
reaches surface - Illuminated area is constant.
Leading edge Illuminated area disc has region becomes until edge
strikes surface grows maximum area an annulus of beam reached
Figure 11.8 The sequence of events that occurs for a single pulse of a pulse-limited altimeter
system, and the expected form of the return at the satellite.
on the length scales the altimeter is measuring, the midpoint of the slope of
the leading edge of the return corresponds to the mean distance to the sea
surface. This is effectively an indication of the geoid, on which the earth's
gravitational potential is constant. The slope of the leading edge of the return
is determined by the spread of surface heights in the altimeter footprint. This
gives information on the variance of surface height, i.e. waveheight. The total
returned power depends on the small-scale roughness of the ocean surface,
which depends on the wind speed; this is the dependence that a scatterometer
exploits.
Altimetry can also be used over the ice shelves. This aspect of altimeter
operation is less well understood, because of the possibility of penetration
of the surface by the radar wave and because we do not have a simple
description of the surface, as we do for ocean waves. Nonetheless, this is an
active area of research and experiment usirtg the ERS-l altimeter, because
the dynamics of the ice shelves are a critical indicator of the effect of global
warming. They also play an important part in the amount or energy absorbed
by the earth from the sun, and in earth-atmosphere interactions.
11.7 SYNTHETIC APERTURE RADAR
Perhaps the most sophisticated radar technique (certainly the most tech-
nically demanding) used in remote sensing is SAR. The major attractions of
SAR are that it can provide high-resolution images of extensive areas of the
earth's surface from a platform operating at long ranges, irrespective of
weather conditions or darkness. The resistance to weather conditions derives
from the use of wavelengths of the order of centimetres, with the X-band
(3 cm), C-band (6 cm) and L-band (24 cm) being favoured. Some systems
use shorter wavelengths (down into the millimetre bands); these will be
adversely affected by precipitation and cloud, for the reasons discussed in
Chapter 8. P-band (68 cm) SAR is also in operation. A few reservations are
needed here. For air-borne SAR operating at longer ranges (see Fig. 11.9),
the propagation path may have to traverse extended regions of rain in the
troposphere. The effects discussed in Chapter 8 can then affect performance.
For space-borne SAR, the ray paths through the troposphere are much
shorter (see Fig. 11.9), and only the most extreme tropical rainfall could
have any effect, and even then only at the shorter wavelengths. However,
depending on the orbit height, the path lengths through the ionosphere may
be long. In the post-sunset equatorial regions and in the auroral zones,
ionospheric irregularities can destroy the phase coherence essential to SAR
performance. Examples of the consequences for the antenna pattern of a
SAR were shown in Fig. 8.11.
This all-weather capability makes SAR a most attractive tool for
environmental monitoring in regions affected by clouds or darkness and for
_----t ~----------_______ _
-- ---- Satellite altitude
(up to several
thousand kilometres)
--.------
-- --
----------, , Ionospheric
peak electron
density (- 300 km)
Aircraft
"-
"-
"-
" '- Tropopause
(- 15 km)
Figure 11.9 A comparison of the propagation paths of air-borne and satellite-borne SARs.
military reconnaissance. An essential element of monitoring changes in the

earth's environment is the ability to observe them on a regular and reliable
basis. Over many parts of the globe this is not possible using optical and
infrared radiation. For large parts of the year the polar regions are in
darkness, but observing ice formation and break-up is an important element
of safe operation at these latitudes. SAR not only has the ability to operate
at these times of the year, but also has sufficient resolution to detect ice leads
and edges and icebergs. Ice monitoring is a major aim of the Canadian
Radarsat project, which will place an orbital SAR in space in 1995. Other
parts of the globe of critical concern are often inaccessible to optical remote
sensing. An outstanding example is provided by the tropical rainforests,
which are normally affected by cloud, or by smoke if significant amounts of
forest burning are taking place. Natural disasters such as floods rarely occur
during good weather. The ERS-l satellite carries a C-band SAR, which will
be used to investigate the utility of SAR in monitoring such events.
The principle by which SAR obtains high spatial resolution was first
pointed out by Carl Wiley of Goodyear in 1957, and is in fact very
straightforward. Consider a moving platform (aircraft or satellite) that
carries a pulsed radar pointing sideways to its motion. At a given range, the
antenna illuminates a strip of scatterers; as it moves forwards, new scatterers
enter the beam, and others leave it (see Fig. 11.10). For a single scatterer,
on each pulse of the radar, the two-way propagation to and from the scatterer
causes the phase to change by
[rad] (11.4 )
Here S, is the distance between the radar and the scatterer when the lth pulse
is emitted.
Buried in the sequence of returns during which the scatterer is illuminated
by the antenna is this phase history, which is unique to a scatterer in that
position. By storing the sequence of returns and correcting for this particular
phase history (by subtracting V't from the phase of the lth pulse), the returns
from the scatterer can all be brought into phase. Adding all the corrected
,,
,,
Scatterer entering
,, " the beam
Illumination pattern -""'-:--

Strip of scatterers ----"'-""""---,1-/
at rangeR
Scatterer leaving
the beam
Figure 11.10 The area illuminated by the SAR antenna.
pulses together will cause the returns from the scatterer to add up con-
structively (see Fig. 11.11 ). For scatterers with different phase histories, this
process will be destructive, and they will contribute far less to the summation
than the scatterer to which the phase correction is 'tuned'. Scatterers that
are close together in the along-track direction will have similar phase
histories, and they will have comparable weightings in the summation. The
separation that causes one of them to suffer significant destructive interference
when the processing is tuned to the position of the other is a measure of the
system resolution. From Fig. 11.11 we can see immediately that the SAR
simply operates like a big lens or antenna array. The only real difference is
that the individual rays through the lens are gathered sequentially rather
than simultaneously. In doing this, a large-aperture antenna is synthesized.
Large apertures lead to good angular resolution, and this improvement
Sum coherently
Correction of
phase
Pulses emitted
Phase and amplitude

of returned signals
are recorded
d<Vk = phaseshift
along two-way
propagation path
Synthetic antenna
gain pattern
Figure 11.11 The SAR principle from the viewpoint of correction of a sequence of phase delays.
The beam pattern of the synthesized antenna is indicated at the bottom of the diagram.
in resolution is the whole purpose of SAR processing. Though the principle

is simple, its implementation is very demanding. The treatment below avoids
many of the complexities, but contains the central ideas of the imple-
mentation. A fuller treatment will be found in Ulaby et aU. We can work
out the gain in resolution using ideas we have already met. First, let us find
the length of the synthesized antenna, using the simplified geometry of
Fig. 11.l2a. At range R[m], the synthetic antenna has length D[mJ given
by the distance travelled by the radar platform while a stationary target
remains within the main lobe of the real antenna. Thinking of the real beam as
sweeping over the target, this means that
D = RfJ [m] (11.5 )
where fJ is the beam width of the real antenna. Since this is given by
fJ = Aid [rad] (11.6 )
where Ais the radar wavelength and d is the real antenna length, we can write
D = RAid [m] (11.7)
The key point of this relation is that the synthetic antenna is proportional
to range. Longer ranges automatically have longer synthetic antennas. We
would now like to use again the relation in Eq. (11.6) between wavelength
and antenna length to find the beamwidth of the synthetic antenna. We need
(a) (b)
Flight track
~
---+
D
Figure 11.12 A simplified form of the geometry of a SAR system.

to be a little careful, however, because this relation is valid for one-way

propagation, whereas in SAR we have two-way propagation before the
synthetic antenna is formed t. Taking this into account improves the angular
resolution of the SAR by a factor 2, so that it becomes
[rad] (11.8 )
The angular beam width of the SAR is inversely proportional to range. But
the spatial resolution on the ground will be the product of the angular
resolution and the range. Hence the along-track resolution of the SAR is
ra=R{3s=d/2 Em] (11.9)
This is a key equation. It tells us that the ground resolution of the SAR in
the along-track direction is independent of the range and also of the
wavelength. It also illustrates the curious property that, in order to obtain
better resolution we should shorten the antenna, in complete contrast to
what we expect for real antennas. This property arises because shorter real
antennas give rise to longer synthetic antennas (see Eq. ( 11.7». In fact, other
operating characteristics, such as the width of the usable swath, impose
constraints on the minimum length of the antenna 2 (see problem 11.5) and
hence on the along-track resolution. Equations (11.8) and ( 11.9) should be
compared with what happens in a conventional radar for which {3 is given by
Eq. (11.6), so that the alo~g-track resolution at range R is R{3 = RA/d.
Worked example Compare the azimuthal resolution of a conventional
radar antenna and a focused SAR for a radar frequency of 3 GHz and
an antenna size of 10 m. Calculate the results for ranges of 10, 100 and
1000 km.
SOLUTION At 3 GHz, the wavelength is 0.1 m. For the focused SAR, the
resolution is independent of range at d/2 = 5 m. The resolution of an
ordinary antenna varies from 100 mat 10 km range to 10 km at a range
of 1000 km.
The discussion above only relates to the along-track direction. Cross-
track resolution is obtained by transmitting a chirped pulse and pulse
compression. In fact, the along-track processing can also be thought of as
linear FM with pulse compression. To see this, consider how the range to
a stationary scatterer varies as it passes through the beam. Using Fig. 11.12b
we can write
(11.10)
t If you are familiar with antenna theory, you will know that Eq. (11.6) is valid in the lill"
field of the antenna. For SAR, scatterers are in the far field of the real antenna but in the near
field of the synthetic antenna. That causes the quadratic phase variation discussed below.
Since x « R in all practical applications of SAR,

S = R(1 + x 2/R2)1/2 ~ R + x 2/(2R) [m] (11.11)
The phase change along the two-way propagation path is then
2n 4nR 2nx 2
2x-S=-+-- [rad] (11.12 )
A. A. A.R
The first term on the right is a phase delay, which is constant for all scatterers
at a fixed range and can be ignored. The second is a phase delay, which is
quadratic with distance through the beam. If the platform is travelling with
uniform velocity V then x = Vt, and the time dependence of phase is
,1,( ) _
'I'
2nV2 2
t - ---t [rad] (11.13)
A.R
(The minus sign occurs because we must subtract the phase delay to get the
phase; the absolute phase reference is unimportant, and we have set it to
0.) Frequency is the time derivative of phase, so this is a linear FM signal;
the instantaneous frequency is
4nV2
w(t) = - - - t (11.14 )
AR
We arrived at exactly the same result from a doppler frequency point of view
in Chapter 6 (see Eq. (6.35».
From Eq. (11.9), we can see that SARs should have spatial resolutions
of the order of metres using practical antennas. (For example, the C-band
antenna on ERS-l is of length 10 m, giving a theoretical resolution of 5 m.
When a more precise defiinition of resolution is used, and the antenna and
processing weighting is taken into account, the quoted resolution is 6 m.)
What price is being paid for this resolution? There are two principal costs.
The first of these is that each pixel represents only a single sample from
what is, for many types of surface, an extended clutter-type target giving
exponential fading. This is the speckle phenomenon, which causes major
problems in SAR image interpretation. Many systems choose to sacrifice the
highest resolution possible from the system in order to reduce the effect of
speckle. They do this by averaging the detected powers from several
neighbouring pixels (either along- or cross-track, or both). This causes the
resolution to be degraded; averaging M pixels in either of the directions
degrades the resolution by the same factor in that direction. The gain in
accuracy (and reduced variability) of the measured RCS comes about because
the pixel distribution changes from an exponential to a chi-squared distri-
bution with 2M degrees of freedom, where M is the total number of pixels
being averaged for each measurement. As we already remarked when
discussing scatterometry, this distribution tends to normality when M
becomes larger; the mean Jl. is preserved, and the standard deviation becomes
Jl.IJM. This process of averaging powers (or, in some cases, amplitudes) is
known as multi-looking. Note that the averaging of pixels can be built into
the processing; the formation of the full-resolution pixels may never occur.
This is, in fact, what a scatterometer does; a scatterometer and a SAR are
very closely related instruments. (On ERS-l they are integrated in what is
called the Active Microwave Instrument; while sharing electronics, they use
different antennas because of the different operational requirements of the
scatterometer. )
The second cost is hidden in Eq. (l1.5). To obtain the best possible
resolution at range R, the SAR must save and process all the returns in the
synthetic aperture of length D. The number of these returns is determined
by the PRF. This is, in turn, determined by the Shannon-Whittaker sampling
theorem: the sample rate must be at least twice the bandwidth (in Hz). From
Eq. (l1.l4), we can see that the frequency sweep due to a scatterer is
determined by the time it is illuminated by the real antenna, i.e. the time
during which the platform travels a distance D. This is simply DIV. Hence
the total frequency sweep is given by
[Hz] (11.15)
For a broadside beam the frequency sweep is symmetrical about 0 Hz, so

that the frequency changes from VI d to - VI d Hz. This gives a bandwidth
VI d, which forces the condition
PRF ~ 2Vld (11.16)
Since the time between samples is I/PRF, this is equivalent to the condition
that the distance moved by the platform between samples must not exceed
d12, i.e. the radar must emit at least two pulses in the time the platform
moves a distance equal to the real antenna length.
This means that, to carry out the processing at range R, we must save
the returns from 2Dld pulses for each pixel we generate, i.e. 2R).ld 2 pulses
are needed. Everyone of these pulses must be phase-corrected. This implies
a complex multiplication (four real multiplications) for each pulse, using the
I and Q channels of the returned signal. Therefore, each image pixel at range
R requires 8R)'; d 2 real multiplications. Since new samples are produced at
the rate 2 V / d per second at the minimum sampling rate, a real-time processor
would require 16RV).ld 3 real multiplications per second at range R. The total
number of operations must also take into account the number of range gates
MR' If we ignore the variation with range in order to get an order-of-
magnitude estimate of the task, we arrive at the need for MRR V ).1 d3
multiplications per second for real-time imaging, even without the processing
corrections that are normal in most modern processors. Note the dependence
on A; longer wavelengths yield longer synthetic antennas, which give a greater

processing load. Note also that the process is described in the time domain,
but can also be carried out in the frequency domain.
Worked example The UK Royal Signals and Radar Establishment has

operated an X-band SAR for many years. Taking the antenna length as
I m, let us find the processing requirement for real-time operation if the
aircraft is flying at 200 km h - 1 and producing SAR images at an average
range of 60 km, with 1000 range gates.
We have all the information needed; the processing must be able
to carry out
1000 x 60000 x 20?_()()0 x 0.03 :::::; 108 multiplications per second

3600
The discussion above has assumed that the processing in the cross-track
direction does not add significantly to the processing cost. This is true for
many air-borne systems, which use surface acoustic wave (SAW) devices to
perform the compression of the chirped pulse on reception. For satellite
systems and the modern generation of air-borne systems, the whole process
is carried out digitally. This increases the estimate of the processing required
by three orders of magnitude or more. As a result, the processing requirement
derived in the worked example is modest by the standards of current systems.
(As an example, the data rate of the ERS-I SAR is 105 x 106 bits per second.
This amount of data cannot be stored on board, and the SAR can only
operate if it is within line of sight of a receiving station. All the processing
is carried out on the ground.) Consequently, most SAR processors work
several orders of magnitude slower than real time, and for satellite SAR
real-time processing is considered very much state of the art. Even most
air-borne systems do not operate in real time. This occurs because high image
quality relies on a number of corrections being applied in the processing.
Uncorrected images are often produced as a quick-look product, with a
precision product being produced later (and slower). In fact, real-time
processing was normal in most early systems, using optical processing
elements (see Kovaly3 for a survey of the early development of SAR ). However,
these do not offer the flexibility and control of image quality possible with
digital methods, and almost all modern systems use digital technology.
Despite these drawbacks, SAR offers our only possibility of reliable high
resolution monitoring of many parts of the earth's surface. This has caused
major efforts to understand the interaction of the transmitted waves with
the surface, which gives rise to the observed backscatter. Nowhere is this
more true than over the ocean. The nature of the imaging mechanisms by
which SAR reveals the presence of ocean waves has been a subject of heated
debate ever since the L-band SAR carried by Seasat revealed the immense
wealth of structure in images of the ocean surface. This led to questions

about how SAR systems can be optimized for oceanographic purposes. There
is also military interest in SAR for applications such as ship tracking and
the detection of submarines by locating the way they disturb large-scale
underwater wave structures.
Large-scale ocean waves do not give rise to direct backscatter at the
centimetre wavelengths used by SAR. The observed variations in backscatter
caused by such waves are thought to be due to a modulation they impose
on the shorter wavelengths in the sea surface spectrum. SAR is sensitive to
these shorter wavelengths by, for example, the Bragg scatter mechanism.
Three processes seem to be important here, although the relative
contributions of each may depend on the circumstances of both the sea and
the radar:
1. The slope at different parts of a large wave may be important in altering

the backscatter from ripple waves. This tilt modulation mechanism is
shown in Fig. 11.13.
2. The short ripple waves observed by the SAR are modulated by a
hydrodynamic interaction with the longer waves. When the phase
velocity of a ripple wave is in the same direction as the drift velocity,
the ripple wavelength increases and the amplitude decreases. When the
phase and drift velocities are opposed, the ripple wavelength is reduced
but the amplitude increases. The orbital motion of the water in large
wavelengths causes a periodic modulation of the ripple waves, which
may be detected by the SAR (see Fig. 11.14).
3. The velocity perturbations of the Bragg ripple waves imposed by the
orbital motion of the water in large waves may cause the SAR image to
go in and out of focus, thus revealing the presence of the larger wave.
This is known as velocity bunching and is shown in Fig. 11.15.
A difficulty in reconstructing an accurate representation of the ocean

surface from its SAR image arises from the sensitivity of SAR to the doppler
Considerable
backscatter to
satellite
I
To satellite
Figure 11.13 Tilt modulation of the backscattered signal.

Direction of motion
of long wave
c= :> I
To satellite
Figure 11.14 Long-wave modulation of Bragg waves by hydrodynamic interaction.
I
To satellite
Positive
doppler shift
imposed
Figure 11.15 Velocity modulation of small waves by the circulation of water in larger waves.
frequency of the scatterers and is related to this third form of modulation.

As we saw in Chapter 6, scatterers with a velocity along the line of sight to
the radar will be misplaced in the image, unless corrective steps are taken.
Since the ocean surface is continuously in motion, this effect must be
accounted for in interpreting the data. There is also the problem of speckle.
Nonetheless, both theoretical and experimental studies indicate that SAR
data can be inverted to recover the ocean wave spectrum, as long as the sea
state is not too high. Many interesting questions about the way SAR images
waves remain. It is hoped that the major experimental and calibration efforts
associated with the ERS-l SAR will help to resolve some of these questions,
and will allow us to assess the viability of SAR as an ocean imaging tool.
The use of SAR over land presents different problems. Much effort has
been devoted to investigating the possible utility of SAR for monitoring
vegetation, such as agricultural crops. To do this successfully, the SAR needs
to operate at resolutions that are significantly less than the typical sizes of
fields. This is to allow averaging to remove the speckle. Very accurate
estimates of RCS are required because of the comparatively small dynamic
range (of the order of 10 dB) of Res values encountered across different
crop types. It has been realized that effective use of SAR for this application
is likely to require multifrequency and/multipolarization systems. The ability
to gather and relate measurements gathered at different times also seems
essential. Most of the work in this field has been carried out using air-borne
radars. The use of SAR in forestry has also attracted much attention. This
has been accelerated by the focusing of the world's attention on the
importance of the great forests for the long-term habitability of the earth.
The effects of acid rain, felling and burning are having drastic consequences
on the extent and structure of forests throughout the world. An important
task for the SAR carried on ERS-l is to provide information on these
processes.
Operating SAR in mountainous regions gives rise to another peculiarity
of SAR imaging, known as lay-over. Because the SAR positions objects in
the cross-track direction by the time of propagation to them, the tops of
mountains are apparently moved towards the radar compared to their
bottoms, in the two-dimensional image (see Fig. 11.16). In the extreme, as
the slope steepens, the top may be closer to the radar than the bottom. This
will cause their respective positions in the image to be reversed compared
to the normal two-dimensional map projection. The slopes on the lee-side
of mountains compared to the radar are also likely to be in radar shadow
(Fig. 11.16). Such effects hamper the use of SAR in studies of snow
mapping, but can be mitigated by the use of digital terrain models in
combination with the SAR data.
The role of SAR in ice mapping has already been mentioned. It seems
clear that SAR has a lot to offer in this application. This will include the
gathering of information on ice age and thickness, ice type and ice dynamics.
Together with location of ice leads and icebergs, this information should
help to make operations at higher latitudes safer.
SAR techniques have also been dramatically successful at imaging the
planets from orbiting spacecraft. First among these are the images of Venus
taken by the Pioneer and Venera 15 and 16 spacecraft. Venus is entirely
covered by cloud and nothing can be seen of the surface using ordinary
optical methods-indeed, the speed and direction of rotation of the planet
were not known until radar revealed it to be spinning in a retrograde fashion.
The Pioneer images showed that the surface topography is not unlike that
of earth, but with any oceans now gone and with what appears to be an
out-of-control greenhouse effect, giving high temperatures and very strong
winds. The Magellan spacecraft has now produced much higher-resolution
images of Venus, which show a remarkable amount of detail, and should
tell us more about the differences between the two planets.
SAR-type processing can also be carried out when the radar is stationary
but the target is moving. This technique, known as inverse SAR, is useful
for imaging objects, and has been used in radar astronomy to map planets
by making use of the differential doppler shifts across the planetary surface,
Radar
E shadowing
Hilly terrain _ _ _ _ _ _ _ _-...!.A~

B
C
',1 , G
F
Image
positioning A' 8' C' E' D' shadow G'
'----y----'
lay-over
I
Map I II
positioning A" B" C" D" E" G"
Figure 11.16 Effects that occur when SAR is used in hilly terrain. The map positioning of
points has been displayed so that point C coincides in the image and map. The distortion due
to terrain effects is obvious, as well as the more extreme effects of lay-over and shadowing.
as viewed from earth. Inverse SAR is a powerful tool in the development

of target recognition algorithms because the doppler stability of an object
(such as the way a ship rolls) may help to identify it.
The survey of uses for SAR set out above does not come close to
summarizing the activity and potentials in this field. We indicate some of
the more recent developments in Chapter 15, but much more will be found
in Ulaby et af.2 and Elachi 4 .
11.8 SUMMARY
The scattering behaviour of the earth's surface at radar wavelengths can
provide useful information about many natural processes. Since most
applications of remote sensing require large areas to be surveyed, HF radars

provide the only useful ground-based systems. These are used for sea sensing.
Extensive coverage can be provided by air-borne platforms. Scatterometers
mounted on helicopters or aircraft, and air-borne side-looking radars and
SARs are employed around the world for a wide range of applications.
Perhaps the most exciting prospect is the new generation of space-borne
radar instruments, which promise to be a major soutce of information on
global-scale processes. Weather conditions have little effect on these instru-
ments at the range of wavelengths employed (though ionospheric effects have
to be allowed for in altimetry and SAR operation). This will allow reliable
gathering of surface information, which cannot be guaranteed at optical
wavelengths. Radar instruments can also provide information, such as the
global wind field over the oceans, that is not available by other means. Much
of this information, such as the global wind field or mean sea height, can
be gathered at comparatively low spatial resolutions by scatterometry or
altimetry (though high-resolution altimetry is attracting much current
interest, and we can expect significant progress in the range of problems to
which altimetry can be applied). For applications requiring high spatial
resolution, SAR provides a possible answer, at the expense of increased
system complexity (and cost!) and problems of data interpretation in the
presence of speckle. With the launch of ERS-t and other space-based radars
throughout the 1990s, this decade should see major advances in under-
standing these techniques and their application to monitoring the earth's
environment.
Key equations
• The Bragg scattering condition:
Ln = n).I(2 sin 0) [m]
• The length of the synthesized antenna:
D = R).ld [m]
• Along-track resolution of a SAR:
ra = dl2 [m]
• Along-track phase history of a scatterer in the SAR beam:
./,( ) _ 2nV2 2
'l't - - - - t [radians]
)'R
• Associated linear frequency modulation:

4nV2
w(t)= - - - t [radians s - 1 ]
).R
• Total frequency sweep:
llf= 2VD = 2V [Hz]

AR d
• Sampling condition:
PRF ~ 2Vjd [Hz]
11.9 REFERENCES
1. Radar Reflectivity ofLand and Sea, M. W. Long, Lexington Books, Lexington, Mass., 1975.
2. Microwave Remote Sensing: Active and Passive, F. T. Ulaby, R. K. Moore and A. K. Fung,
vols 1-3, Addison-Wesley, Reading, Mass., 1981.
3. Synthetic Aperture Radar, J. J. Kovaly, Artech House, MA, 1976.
4. Spaceborne Radar Remote Sensing: Applications and Techniques, C. Elachi, IEEE Press,
New York, 1987.

Remote sensing of sea state by radar, D. E. Barrick, chapter 12 in Remote Sensing (Jf the
Troposphere, Ed. V. E. Derr, NOAA Environmental Research Laboratories, Boulder,
Colorado, 1972.
Radio science and oceanography, E. D. R. Shearman, Radio Sci .• 18(3), 299-320, 1983. [A
good starling place for beginners.]
Extraction of sea state from HF radar: mathematical theory and modelling, B. J. Lipa and
D. E. Barrick, Radio Sci., 21(1), 81-100,1986.
Theory ofSAR ocean wave imaging, S. Rotheram, in Satellite Microwave Remote Sensing, Ellis
Horwood, Chichester, 1983.
11.10 PROBLEMS
11.1 A 15 MHz HF radar is arranged to measure tidal currents across the entrance to a harbour.
The dominant Bragg line is observed to have a doppler shift of 0.7 Hz. just as a large ship is
seen approaching the harbour. Do you think it is important to convey the result of your
observation to the ship?
11.2 A scatterometer mounted on a tower, as shown in the diagram, uses frequency modulation
of bandwidth B [Hz] to give a maximum possible time resolution 1/ B [s]. Show that this gives
a slant range resolution
r, = c/(2B)
Each of these resolution cells corresponds to an independent measurement of the RCS of the
ground (see diagram). Show that the number of independent samples available for averaging is
2B
N ~ - hfJ sec (J tan (}
c
where fJ is the beamwidth and () is as shown in the diagram. Plot the variation of N as ()
increases from 10° to 70°.
11.3 Assuming a beam width of at most a few degrees, show that the annulus generated by a
pulse-limited altimeter has approximate area 2nRcr, where the satellite is at altitude R, c is the
speed of light and r is the pulse length. Show also that the annulus forms while the leading
edge of the pulse is still in the main beam if the condition
r < RfJ2 /(8c)
is met.
The ERS-l altimeter operates with a 1.2 m diameter dish at a frequency of 13.8 GHz; the
satellite has a nominal altitude of 785 km. If it were operating in pulse-limited form, show that
the pulse length would need to be less than 0.11 JlS. For a pulse length of 0.1 Jls, a loss factor
of 50 per cent and a transmitted power of 500 W, what would the signal-to-noise ratio be at
the satellite if the surface has an RCS of 5 dB m 2 m - 2?
11.4 Compare the azimuth processing requirements of (a) an air-borne SAR travelling at
200 km h - I with aim antenna, operating at a range of 50 km, and (b) a space-borne SAR
travelling at 7.5 km s - I with a 10 m antenna, operating at a range of 1000 km, if real-time
operation is required. Do the comparison for the cases where both platforms are carrying a
C-band (6cm) and an L-band (24cm) radar.
11.5 The returns from each pulse of a SAR come from a swath defined by the maximum and
minimum slant ranges (see diagram). Returns from successive pulses must not overlap on
reception. Show that, for a SAR platform whose speed is V and whose antenna has an along-track
dimension d, this leads to the two conditions (using Eq. (11.16)
(a) PRF,;:; c/(2R,)
(b) R,';:; cd/(4V)
What is the widest possible swath that can be imaged by an air-borne SAR if it is carrying a
2 m antenna and travelling at 300 km h - I? What is the best possible resolution in the along-track
direction available to a space-borne SAR travelling at 7.5 km s - I if it is designed to produce
images from a slant swath 100 km wide?
11.6 The ERS-I satellite operates at incidence angles varying from 20.\" to 25.9°, with a
mid-swath incidence angle of 23". Are lay-over and shadowing likely to be major problems?
Compare this with the effects likely in SAR images gathered by a long-range air-borne SAR
operating at an altitude of 10 km and with coverage from 50 to 100 km ground range.
CHAPTER
TWELVE
GROUND-PROBING RADAR
• Applications
• Propagation in the ground
• Carrier-free radar
• Wideband antennas
Short-range radar is proving to have many applications for probing the

ground and other structures, such as walls.
12.1 INTRODUCTION
Ground-probing radar, also known as subsurface radar, is becoming an

important subject. The number of applications for ground-probing radar is
growing and the technology is beginning to be employed in other radar
fields; for example, impulsive or carrier-free radar, which is in widespread
use for ground probing, is now being developed by the military for its
anti-Stealth capabilities (see Chapter 14).
The development of ground-probing radar has been driven by its
applications, and these are so widespread that the subject has become a
fertile area for the creative electronics engineer. Many applications now lie
outside ground probing, as the following list of uses indicates:
• Buried objects: detecting landmines; locating pipes and cables.

• Civil engineering: finding voids and structural flaws; inspection of
reinforcing bars; measurement of wall thickness; tunnel and mines haft
location.
257
• Security: locating objects hidden in walls or floors.

• Archaeology: grave and burial-mound investigation; locating foundations
of buried buildings; investigating interiors of pyramids.
• Geophysics: examining structure and strata; surveying subsurface of the
Moon and Mars.
• Earth resources: coal and peat reserves mapping; locating water tables
in dry climates.
• Ice mapping: of ice caps and glaciers.
There are many other ways of imaging the subsurface and the interior
of structures; these include seismology, ultrasound, magnetometry, impedance
imaging and low-frequency induction methods. However, ground-probing
radar systems have the advantage of being relatively lightweight, mobile,
well focused and easily configured to detect particular types of target. The
disadvantage of ground-probing radar is that good radio propagation is only
achieved in dry or low-conductivity materials, and the penetration into wet
rock or soil may be restricted to a few metres.
Over the years, a great many articles have been published on ground-
probing radar in a wide variety of journals and conference proceedings.
Fortunately, in 1988, a special issue of the lEE Proceedings-F was devoted
to this topic! , and the introductory article presents a comprehensive review
that is a useful starting place for those wishing to discover more.
12.2 DESIGNING GROUND-PROBING RADAR SYSTEMS
Subsurface radar design is constrained by several factors; the propagation

conditions, the depth (range) resolution, the need for any near-surface (very
short-range) measurements and the antenna requirements.
Propagation in the ground is limited by the dielectric losses, which are
well documented for different types of ground. The attenuation in most types
of ground rises with frequency, and also with water content, because water
has a relative permittivity er of about 80 whereas most dry rocks and soils
have values of er in the range 2 to 6.
The solution of Maxwell's equations for propagation in the ground
shows that radio waves are exponentially damped with depth. The attenuation
expressed in dB therefore becomes a linear function with depth, and
propagation losses can conveniently be expressed in dB m - !. Typical values
for one-way attenuation in different materials are shown in Fig. 12.1.
We learned in Sec. 1.5 that the two dominating factors in the radar
equation are the R 4 10sses and the noise level; few radar systems can tolerate
R4 losses of more than 200 dB, and so the right-hand scale of Fig. 12.1
instantly gives us a feeling for the sort of depths that can be plumbed using
this technique. Geological ground probing is restricted to frequencies below
GROUND-PROBING RADAR 259
10
Ice
Sea ice
E
-
I g
e 2 X 102 iii
CQ
~
~
c e
.2 8
~
C
::l ....
2 '"Co
~ '"
'"
;>, Ice ..Q
'"~ 10- 1
Sea ice 20 ;...
'"
~
V
c Permafrost
6
0 ~
f-
lO-31~M--H-z-------l~M~H~Z--------1~OO~M~H~z------~1-G~H~z--~
Frequency
Figure 12.1 Propagation losses in various materials.
100 MHz and buried object location to below 1 GHz. Few systems operate
above 1 GHz and antenna constraints effectively prevent operation below
1 MHz.
The range/depth resolution is determined by the bandwidth, but this is
not always the bandwidth of the transmitter (which in the case of impulsive
radar may be inherently very wide ). One limiting factor is the transfer function
of the antenna and, although wideband antenna designs are used, the
transmitter pulseshape has usually suffered some distortion before it enters
the ground. A second limitation is the nature of the ground, which acts like
a low-pass filter by attenuating the higher frequencies. The phase velocity v

of radio waves in the ground is usually well below the speed of light c and
values of vic are found in the range 0.2-0.4 for many types of rock.
In defining the resolution, these effects can be avoided if the effective
bandwidth Berr of the system is defined in terms of the received power
spectrum. In this case the resolution is given by
[mJ ( 12.1 )
The range resolution (before any compression) effectively determines the
short-range performance and the ability of the radar to locate objects or
structures immediately below the surface. A short transmitted pulse means
the system can be switched very quickly into receive mode for close-range
observations. A long pulse (or other long-duration waveform) means that
either the radar must be elevated above the surface or some form of eclipsing
loss must be suffered-this is the signal loss that occurs when part of an
echo has already returned before the receiver can be switched back to full
sensitivity.
The requir~ment for good range resolution and near-surface performance
means that short-pulse/high-Berr systems are needed. This, in turn, implies
a high carrier frequency, because conventional radio engineering requires
the carrier frequency to be ~ 100 x Berr . Unfortunately, as Fig. 12.1 shows,
high frequencies do not propagate well underground, and the dilemma of
needing high frequencies for good resolution and low frequencies for good
propagation has led to the development of carrier-free radar in order to
enjoy the benefits of both.
12.3 CARRIER-FREE RADAR
A sudden electrical impulse, such as a flash of lightning, generates a wide

range of radio frequencies and, indeed, thunderstorms cause interference up
to the FM radio band and beyond. A radar based on the principle of
connecting a discharge device to an antenna, rather than of modulating a
carrier frequency, would therefore have an intrinsically very wide bandwidth,
which could be exploited to give exceptional range resolution.
The ideal of using signals with a large relative bandwidth has been
around for some time, and several books have been published on the subject;
see for example Harmuth 2 • Relative bandwidth 11 has been defined by
Harmuth as
fH -fL
11 = - - - = [ ] (12.2)
fH +fL
where fH' fL = highest and lowest frequency of interest. For a pure sinusoid
1] = 0; for a typical radar (or communication system) 1] ~ 0.01; but for

wideband impulses 1] can be close to unity. In general, carrier-free radar is
regarded as having 1] ~ 0.8. A well known example of a carrier-free system
is a time-domain reflectometer, a device for transmitting short pulses down
a cable to locate faults by detecting signals reflected from the discontinuities.
In conventional signals and systems analysis, any signal can be repre-
sented by a Fourier series of sinusoids, and for smoothly varying signals the
series converges to the signal in a few terms. However, this not a good
representation of signals that are rectangular, or contain straight edges, and
it may take 40 or 50 terms in a series to get a reasonable approximation to
the original signal shape. Signals can also be represented as the sum of a
series of other orthogonal expressions, such as exponentials or Legendre
polynomials, but the obvious representation of rectangular-type signals is
to use a series of rectangular functions, the most common of which are known
as Walsh functions. In fact, rather than thinking of signal analysis in terms
of the Fourier series and frequency, it can be considered in terms of Walsh
series and sequency (the analogue to frequency, which is a sinewave concept ).
Various experiments have been carried out to transmit data using Walsh
transmitters and receivers, and it has been suggested by Harmuth 3 that parts
of the radio spectrum could be simultaneously shared by all users adopting
orthogonal Walsh functions. In general, these ideas have not been widely
adopted because of practical engineering difficulties, but the Walsh transform
approach remains a useful and widely used concept for the analysis of
carrier-free radar waveforms. It may be useful to read up about them if you
are going to do any signal or data processing in this field.
A block diagram of a typical carrier-free radar system is shown in
Fig. 12.2. There is no RF oscillator and the transmitter is usually a simple
device such as an avalanche semiconductor, connected across a high-voltage
( ~ 250 V DC) power supply and triggered into conduction by a pulse from
the PRF generator. Because ranges are short, high PRFs are used, sometimes
as high as 1 MHz. The transmitter waveform is a pulse with risetime of 1 ns
or less, and a decay of perhaps 25 ns as shown in Fig. 12.3. Other types
of transmitter can be constructed using pulse-sharpening diodes.
It is difficult and expensive to build T - R switches fast enough for
carrier-free radar, and consequently most practical systems use separate
antennas for transmitting and receiving. The antennas may be either widely
separated to reduce mutual coupling, or co-located using some geometrical
arrangement (such as crossed dipoles) to reduce the coupling. Fast T - R
switches have been built, but special-purpose Schottky barrier diodes are
needed, which must have a low forward impedance, a low reverse capacitance
and the capability to withstand the high reverse voltages imposed by the
transmitted impUlse.
The receiver must have a wide bandwidth, a good dynamic range, and
must also be non-dispersive in order to have a fast risetime. Generally
Volts
~Tim'
Figure 12.2 Typical carrier-free radar system.
100
Exponential decay
25 ns
26
Time [ns]
Figure 12.3 Carrier-free radar transmitter impulse waveform.

speaking, below 1 GHz, this does not present too many engineering diffi-
culties and commercial units are available. Sampling the receiver output is
much more of a problem. With spectral components up to 1 GHz, direct
AID conversion would be prohibitively expensive with today's technology,
and some method of slowing down the data rate is needed. One method
commonly used amounts to a 'stroboscopic' or aliasing technique; the echo
pulse from a given range gate is successively sampled a few nanoseconds
later on each PRF sweep so that the shape of the echo pulse is gradually
revealed, as shown in Fig. 12.4. Some time-domain refiectometers work in
this way. There is no inherent integration gain in the stroboscopic process,
and the SNR is less than for direct sampling. However, since subsurface
targets are stationary, there is plenty of time to integrate after stroboscopic
sampling, provided the radar is not on a fast-moving platform.
An example of a recording made by a relatively simple carrier-free radar
with a stroboscopic receiver is shown in Fig. 12.5a. Physical investigation
of the ground confirms that the major features have been revealed correctly,
as shown in Fig. 12.5b.
There has been some military interest in the use of carrier-free radar
technology. Irregular wideband impulses are not easily detected and located
by ESM (electronic support measures) receivers (see Chapter 14), and so
these radars have some immunity to countermeasures. It is also difficult to
Transmit Echo
pulse
... -\-_ _ _ _ _ _ _ _ _ _ _-::::~Sampling window slides along

• and traces out the waveform
.. iILn
100 ns
Figure 12.4 Stroboscopic or aliasing receiver to slow down sampling rate.

(a) x[m]
(b) xlm]
100 200 300 400
2
table
E 4
.c Sand
.,
15..
""'"+-- Permafrost Lithological
o 6 boundary
Figure 12.5 (a) Carrier-free radar profile of the ground made from a helicopter at a height of
50 m and travelling at 50 km h - I (after Finkelstein at the Riga Institute of Civil Aviation,
Latvia). (b) Geological investigation confirms the observations.
'Stealth' targets against a wideband signal because the common types of

radar-absorbing materials (see Chapter 14), such as dielectric and magnetic
absorbers, offer significant absorption only over a limited frequency range.
12.4 ANTENNA DESIGNS
The antenna is the weak point in a carrier-free radar system. If a simple

dipole antenna were used, the shock excitation of the impulse would
cause it to resonate at its fundamental frequency (plus harmonics) and
the transmitted pulse would be reduced to a sinusoidal 'ringing' waveform.
A further problem concerns the impulse response of the antenna
in a more general sense. The time-domain form of the echo received
from the ground is a convolution of the transmitter pulseshape with
all of the following:
• The impulse response of the antenna.

• The transmission properties of the ground.
• The shape of the target.

• The ground and antenna on the return path.
• The impUlse response of the receiver.
This is shown diagrammatically, in both the time and frequency domains,

in Fig. 12.6. If the shape of the target is to be recovered from the shape of
the received waveform, there is a need to deconvolve these effects at the data
processing stage. This requires the various system impulse responses to be
very well characterized.
While the impulse response of the receiver (and T - R switch if fitted)
can be measured in the laboratory, the impulse response of the antenna
depends upon its reactive coupling with the ground, and will change with
height about the ground or if the terrain changes.
The best solution to the ringing and the impulse response problems is
to develop very wideband antennas, known as frequency-independent antennas.
Although truly frequency-independent antennas cannot be made, there are
several common design approaches that are used to approximate them (see
for example Rumsey4).
Resistively loaded dipoles In a conventional dipole, the resonance occurs

because the transmitted current travels down the wire, is reflected from the
end with a phase reversal, and arrives back at the terminals in phase with
the transmitted signal. This resonance can be avoided by loading the dipole
resistively along its length, by resistive end-loading or by using radar-
absorbing materials along its length. Basically, these techniques are designed
so that the impulse travels down the antenna wire and disappears.
Resistively loaded dipoles have been found to be the most useful type
of antenna for ground-probing radar, and the inevitable losses associated with
the technique have been minimized by applying an increasing exponential
taper to the resistive loading along the dipole arms.
Biconical antennas An infinitely long biconical antenna behaves as a perfect

transmission line with a real impedance and so becomes an ideal radiator.
Practical biconicals do suffer standing waves, but the bandwidth is better
than a conventional dipole and improves as the cone angle increases. The
two-dimensional version of a biconical, often called a bow-tie or butterfly
antenna, is particularly useful in some applications, such as on the wings of
an air-borne probing system.
Angle principle If the dimensions of a simple antenna are doubled and the
wavelength is also doubled, the performance remains the same. This means
that the impedance, polarization, antenna pattern, etc., are invariant to a
change of scale in proportion to wavelength 4. If the shape of the antenna is
determined entirely by angles, the performance has to be independent of
Time domain Frequency domain
Time III
o x
Impulse response of Transfer function

transmit antenna of transmit antenna
o x
Shape of target Spatial transform

of target
o x

receive antenna of receive antenna
o x

receiver of receiver
Figure 12.6 The signal engineering in a carrier-free impulse radar.

frequency. In practice, there is an upper and lower limit, but between them
a very large bandwidth can be achieved. The most commonly used designs
are planar and helical spirals, and log-periodic structures.
Horns At the higher frequencies, horn antennas can be used, and the
bandwidth may be extended by careful design of the shape, including internal
ridges, and by dielectric loading.
With some of these antenna designs there is an effect in which the active
region of the antenna moves to different parts of the antenna as the frequency
is changed, causing a frequency-dependent delay. Further details can be
found in many antenna books, and for an interesting history of the
development of frequency-independent antennas try Chapter 7 of Weeks 5 •
12.5 DATA PROCESSING
Some types of modulation, such as frequency-modulated continuous wave

(FMCW), Barker or pseudo-noise coding and chirp, may require some initial
signal processing to compress the waveform and extract range information.
After this, all ground-probing radar systems need to carry out a lot of data
processing, often more than with conventional radar, in order to reduce
clutter and remove defocusing effects that arise from the propagation. Also,
many ground-probing radars are aimed at producing a three-dimensional
map of the subsurface, and some form of imaging software is needed.
Clutter from discrete surface scatterers and discontinuities in the surface
are a problem for ground-probing radar because propagation in free space
is much better than in the ground. Surface clutter effects can be minimized
by placing the antenna close to or on the ground, because reactive coupling
tends to focus the beam into the ground and reduce the stray horizontal
radiation. Any motion of the radar can be used to help identify clutter
because the phase of front and rear surface echoes changes linearly with
distance while subsurface echoes have non-linear phase variations. One
disadvantage of the carrier-free impulse approach is that, while there is time
delay information, there is no equivalent phase information available (in the
time domain) because phase is usually considered to be a single-frequency
concept. It may, however, be possible to make some progress with this
problem by using the phase of the spectral components of the signal in the
frequency domain.
Other receiving problems include subsurface clutter (i.e. underground
targets not of interest), transmitter breakthrough into the receiver (usually
due to ringing in the antenna), which causes close-range problems similar
to clutter, and radio interference, often from the television bands.
A point scatterer located below the surface will be detected at several
(a)
~r x
(b)
Air
.1".';
Figure 12.7 Response of a buried scatterer: (a) geometry showing apparent increase of depth
with horizontal distance; (b) characteristic hyperbolic shape observed. In practice, the picture
is more complicated, especially for elevated antennas, because of diffraction at the surface and
interaction between space-wave and surface-wave effects.
different ranges by a moving radar, because of the finite width of the antenna
pattern (see Fig. 12.7a). The actual range is given by
2
R = -J(x 2 + d 2 ) Em] (12.3)
V
where d = depth Em], v = propagation velocity and x = horizontal distance

from scattering object Em]. Range is normally assumed to be depth, and
this causes the plot of a scattering object to have a characteristic hyperbolic
shape, as shown in Fig. 12.7b, when plotted on an 'echo-sounder' type of
display without further processing. The exponential increase in propagation
loss with range means that the resolution of the radar is somewhat better
than might be expected from the beamwidth; moving to higher frequencies
increases the attenuation further and improves the resolution.
The hyperbolic defocusing of an object can be corrected in the data
processing software. The focusing depends on a knowledge of the propagation
velocity v and the rate of attenuation, but methods of estimating these
parameters from the echo have been evolved. Most systems use some form
of linear focusing algorithm, usually performed in the frequency domain,

and the software is sometimes extended to include full synthetic aperture
processing and high-resolution mapping. It has been found that the length
of a synthetic aperture need only be a quarter of the target depth or less.
Radio holographic techniques have also been used to image the subsurface;
more details and appropriate references can be found in reference 1. It may
also be possible to use super-resolution techniques to increase the effective
bandwidth of ground-probing radar and improve resolution.
12.6 SUMMARY
Ground-probing radar is one of the few ways of inspecting geological features

and locating hidden objects and structural flaws. The technique is not new,
but it is only since the advent of fast digital processing and microprocessors
that sufficient low-cost signal and data processing has been available to solve
many of the problems raised by this type of radar.
It is these very problems that make ground probing such an interesting
field of radar. The technique is relatively inexpensive, has many and varied
applications and raises abundant deconvolution and mapping problems for
signal and data processing engineers to work on. There are further interesting
issues to study such as novel antenna design, autonomous control of radar
mapping vehicles and methods of three-dimensional data display. There is
also plenty of scope for adapting the radar modulation to the type of target
expected, and one form of modulation, carrier-free impulses, is becoming a
specialized field in its own right, with potential military applications.
An interesting future development concerns a proposed unmanned
mission to Mars, which includes a ground-probing radar to explore the
geophysics and search for subsurface water-ice. The radar is to be built into
the guide-rope of a balloon; during the Martian day the balloon will float
through the Martian atmosphere, but during the night it will trail the
guide-rope along the ground. The guide-rope itself will form the single
antenna of the system. The development of this radar system has led to a
lightweight, low-power, single-antenna system that has many applications
for air-borne Earth resources mapping, including the search for water in arid
regions.
12.7 REFERENCES
I. Special issue on 'subsurface radar', lEE Proc.-F, 135(4), 1988. [A very useful collection
of papers.]
2. Nonsillusoidai Ware/or Radar and Radio Communication, H. F. Harmuth, Academic Press,
New York, 1981.
3. Transmission of Information by Orthogonal Functions, H. F. Harmuth, Springer-Verlag,

Berlin, 1972.
4. Frequency Independent Antennas, V. H. Rumsey, Academic Press, New York, 1966.
5. Antenna Engineering, W. L. Weeks, McGraw-Hili, New York, 1968.
6. Underground mapping of utility lines using impulse radar, R. Caldecott, M. Poirier,
D. Scofea, D. E. Svoboda and A. J. Terzuoli, lEE Proc.-F, 135(4),343-353, 1988.
12.8 PROBLEMS
12.1 An article in the special issue of the lEE Proceedings":'F on subsurface radar describes an
impulse radar for mapping underground utility lines 6 . The 3 ns transmitter pulse has an
amplitude of 200 V and the receiving system uses a preamplifier and a sampling oscilloscope.
If the noise level at the sampling head is 10 m V, calculate the minimum gain of the preamplifier
if objects are to be detected to a depth of 4.3 m. Assume a round-trip attenuation of 23 dB m - I
and 17 dB losses, mainly in the antenna system. Remember that voltage ratios are converted
to decibels by 20 log ( VI / Vz ).
12.2 What is the maximum unambiguous PRF that could be used if the radar in problem 12.1
were to be instrumented for a maximum range of 4.3 m? Assume the phase velocity of radio
waves in the ground to be c/3.
12.3 If the system in problem 12.1 were to pass over a metal grate in the road, what do you
think would happen?
12.4 If the radar in problem 12.1 were to be used for probing very dry rock, such as on Mars,
how much greater might the penetration into the ground be? Use Fig. 12.1 to estimate the
attenuation.
12.5 If a low-frequency version of the radar in problem 12.1 were designed specifically for
geological applications and operated at a frequency near 10 MHz, what depths might be probed?
12.6 Calculate the maximum unambiguous PRF and the likely antenna size for the radar in
problem 12.5.
CHAPTER
THIRTEEN
MULTISTATIC RADAR
• The earliest form of radar

• Basic system configurations
• Ways of locating targets
• Applications
Multistatic radars have some special uses, particularly as a counter to

jamming and anti-radar munitions.
13.1 INTRODUCTION
In a monostatic radar system the transmitter and receiver are located at the
same place, sometimes sharing a single antenna. If the transmitter and receiver
are widely separated, by a baseline typically one-third of the distance to the
target, then the system is said to be bistatic. If there are several widely
distributed receivers associated with a single transmitter, or (more rarely)
several transmitters, then the system is multistatic. Bistatic radar is thus a
subset of multi static radar, but is by far the most common form used.
The earliest radars had to be bistatic since it was not practicable to build
pulse waveform radars and T -R switching. An early example of a bistatic
radar was built by Dr Albert Taylor and his assistant, Leo Young, at the
US Naval Research Laboratory in 1922. Using a 500 Hz CW modulated
transmitter, and a receiver placed in a car, Taylor and Young drove across
the Potomac River and detected a variety of targets, including a wooden
ship (for more details see Swords!).
271
It is not always necessary to have your own cooperative transmitter in

a bistatic radar. During World War II the Germans made use of the British
'Chain Home' radar as a transmitter of opportunity for their 'Kleine-
Heidelberg' receiving system, and were able to give warning of Allied
bombing raids over the channel without giving away the position of their
ground sites. Since then, bistatic and multistatic techniques have evolved
into a sophisticated branch of radar engineering.
13.2 MULTISTATIC CONCEPTS
Figure 13.1 shows the basic arrangement of a muItistatic radar system. The
transmitter illuminates a sector either by a scanning pencil beam or by
floodlighting the whole area. Usually the transmitter plays no further part
in the radar, although it could be part of a monostatic radar in its own right
(as was Chain Home during World War II).
The first signal detected by the receiver is the direct pulse, which has
travelled along the baseline from the transmitter. This signal initiates the
receiver timing, from which the target range will be determined. The receiving
antenna system uses either a floodlight beam, a staring array of many fixed
beams, or a scanning pencil beam carefully synchronized to a scanning
transmitter beam.
Floodlight illumination - - - f - -
-'\-_--+-- Scanning beam
Scanning beam - - - ' - f - - - :
·,......\-~)tarml!. array
Direct
Transmitter pulse Receiver
Figure 13.1 Basic antenna arrangements used by bistatic and multistatic radar systems.
MULTISTATIC RADAR 273
Most technical aspects of multistatic systems are the same as for

monostatic radar, but there are certain advantages in using bistatic systems,
which make them suitable for particular tasks:
• The receiving site cannot be located by electronic warfare receivers (see
Chapter 14) and is safe from attack by anti-radiation missiles and highly
directional jamming.
• The receiver can be sited in a favourable radio-quiet location, perhaps
somewhere where no transmitters are allowed, such as near ftammable-
liquids stores, gas terminals, etc.
• The transmitter can be placed in a radio-noisy location, for example in
a city.
• The receiver requires little or no protection from the transmitter pulse
and, because there are no large-amplitude, close-range echoes, the
dynamic range requirement is less than for a monostatic radar.
• No T -R switch or duplexer is required; these devices are lossy, expensive
and heavy (for air-borne radar applications).
• With certain configurations, less transmitter power is needed to detect
targets bistatically than mono statically .
• High PRFs can be used because a bistatic system does not suffer the
same range blindness as the equivalent monostatic system.
• Several receivers in a multistatic system increase the probability of
detecting a given target since it is unlikely that RCS fading and jamming
will affect all the receivers at the same time.
• If the target angle can be measured at both sites, as well as the bistatic
range, data can be checked for self-consistency to remove false alarms.
• Bistatic radar measures a doppler component toward neither the
transmitter nor the receiver, but along a line between them (actually,
along a tangent to the confocal hyperbola); in Sec. 13.5 we discuss
bistatic doppler further. If the bistatic doppler component is combined
with a monostatic measurement, the full two-dimensional velocity vector
can be derived, e.g. the direction and speed of a ship. If a third site is
added into the system, true three-dimensional velocity vectors can be
derived.
There are, of course, disadvantages to using multistatic radar, which

prevent it from being found in more widespread use:
• The geometry of target location is more complicated.

• The synchronization between transmitter and receiver, which is trivial
in a monostatic radar, may be quite complicated.
• Multibeam receivers are usually required; these are expensive.
• Two radar sites are required. In some ways this is an advantage because
each site may be smaller than for a complete radar and better adapted
to the functions of transmitting or receiving. On the other hand, two
radar sites (two aircraft, two pieces of coastline, etc.) genedllly cost more
than a slightly larger single site.
• A further complication in using two sites is that both must be able to
see the target. For low-flying targets this increases the problem of terrain
obscuration.
• There is usually a 'dead zone' between the sites in which it is difficult
to detect targets; see Sec. 13.5.
(a)
r - - I - - - - - - Surveillance area
Two-way sidelobe
protection
Tx.Rx
(b)
->r~'----'c-----Area under
surveillance
.x:=.d,----Arleas of one-way
side lobe protection
oo......---"''''':--''r--\-----Area of
two-way side lobe
protection
Tx Rx
Figure 13.2 (a) The clutter in the side10bes of a monostatic radar antenna is attenuated on
both transmit and receive. (b) With a bistatic system, there are clutter patches illuminated by
the sidelobe of one antenna and the mainlobe of the other.
• As a rule, there is only one-way sidelobe protection against clutter. This

is illustrated in Fig. 13.2; in Fig. 13.2a the clutter in the sidelobes of a
monostatic radar antenna is attenuated on both transmit and receive;
but in the case of a bistatic system Fig. 13b, there are clutter patches
illuminated by the sidelobes of one antenna and the mainlobe of the other.
13.3 THE BISTATIC RADAR EQUATION
Using the same notation as in Chapter 1, the power flux density arriving at
the target is
. PtGtL
Flux densIty = ----t ( 13.1 )
4nR~
where L t = losses in the transmitting system [ ] and Rt = distance from

the transmitter to the target Em]. The power re-radiated towards the receiver
by a target of bistatic RCS CT b is
. PtGtLtCT b
Power re-radlated =----- [W] (\3.2 )
4nR~
The power flux density at the receiving site is therefore
. PtGtLtCT b \
Flux densIty = 2 X -- --- ( 13.3 )
4nR t 4nR;
where Rr = distance from the target to the receiver Em]. The power received
by an antenna of effective aperture (Gr )"2 /4n) [m 2 ] at the receiving site is
PtGtGrCT b)" 2 LtL r
p = -----:----:----:-- [W] ( \3.4 )
r (4n)3 R~ R;
where Lr = receiver losses [ ].
When R t and Rr are about the same as R for a monostatic radar, there
is little difference between the two techniques, especially since experimentally
the RCS of many targets is found to have similar bistatic and monostatic
values (in fact, the bistatic RCS is often close to the monostatic value
measured along a line bisecting the bistatic angle). However, there are large
gains to be made if either the transmitter or the receiver can be moved closer
to a target.
Worked example Compare the power received by a monostatic radar

at a range of tOO km from a target, and by a covert receiver, tuned to
the same transmitter, at a range of to km from the target.
SOLUTION Assuming all other system factors are equal, the R4 losses
are the important factor. For the monostatic case
R4 = 10 20 [m4] == 200 [dB m4]
and for the bistatic case
R~R; = 10 18 [m4] == 180 [dB m4]
In the second case, a transmitter of only 1/1 OOth the power of the
monostatic radar is required for the same SNR to be received. The
bistatic receiver is relatively safe from attack because it is electronically
silent, and the transmitter is safer because the transmissions are weaker
and consequently harder to detect.
The contour of equal signal strength for a monostatic system, which

determines the detection range of the radar, is set by the R4 factor and is
just a sphere centred on the radar. For a bistatic system, the R4 factor must
be replaced by (R t x Rr )2, and this leads to the dog's bone type of shape
shown in Fig. 13.3a. The intersection of this shape with a plane gives the well
known oval of Cassini (Fig. 13.3b). A target moving round this Cassini oval
would give a constant SNR at the radar.
(a)
(b)
e>l
e»l
Figure 13.3 (a) Surface of constant SNR for a bistatic system. (b) The intersection of the
surface with a plane gives the ovals of Cassini.
The shape ofbistatic constant signal strength ellipsoids can be compared

by normalizing them to the baseline 2a and defining an ellipticity factor e
given by
J(R\ X Rr)
e= -~------~--- [ ] (13.5 )
2a
If e » I, then a is small compared to the target distance and the oval
of Cassini is produced. At even larger values of e, these ovals become nearly
circular as the radar approaches a monostatic configuration (see Fig. 13.3b).
If e = I, a figure-of-eight shape is created, known as the lemniscate of
Bernoulli.
If e« I, then small ovals are generated about the transmitter and
receiver. As e gets smaller, the target is constrained to lie somewhere on a
tiny near-circle close to one or other site. This would be the case in the
worked example above, where e would be of the order of 0.16.
13.4 MULTISTATIC TARGET LOCATION
The first piece of information used for target location is the time delay between
the transmission of a pulse and the reception of an echo (analogous to range
in a monostatic radar). After finding out how to use this, we go on to examine
the different possible geometries that can be used for target location.
The range measured by a monostatic radar defines a circle (an iso-range
contour) centred on the radar, upon which the target must lie (see Fig. 13.4a).
In the case of a bistatic system, the total path length (R\ + Rr) defines a
three-dimensional elliptical shape shown in Fig. 13.4b. On any given flat
plane you can construct the appropriate ellipse by sticking two pins into a
sheet of cardboard, tying a piece of cotton between them of a suitable length
to represent (R\ + Rr ), and drawing out the shape with a pencil while keeping
the cotton tight, as in Fig. 13.4c.
The importance of the elliptical path shown in Fig. 13.4b is that usually
Rl and Rr cannot be measured independently because the transmitter~receiver
timing information only gives the combined distance 2r = (R 1 + Rr)' Thus
we know only the total length of the cotton, and not the two separate
distances from the pins to the pencil. Note that these iso-range ellipses are
not coincident with the constant SNR ovals of Cassini. A target flying around
a constant-range ellipse would thus have a different SNR at each place on
the contour; this is not true monostatically.
If only time-delay information were available, we could draw the
appropriate ellipse, but we would not know where the target lay on that
ellipse (in the same way, knowing only the range in a monostatic radar puts
the target somewhere on a circle). Angular information is needed to make
further progress, and there are several ways in which this may be introduced.
(a) (b)
(c)
Figure 13.4 (a) Monostatic iso-range, or constant-range, contour. (b) Three-dimensional

bistatic iso-range ellipsoid. (c) How to draw a two-dimensional bistatic range contour.
Elliptical Geometry
This is the obvious, and most common, method of target location (see
Fig. 13.5a). It is necessary to know: 2a, the baseline; 2r = (R( + Rr ) the
round-trip distance; and Or' the target azimuth, measured at the receiver.
The range Rr of the target from the receiver can be found by using the cosine
formula to solve the triangle in Fig. 13.5a, giving
r2 _ a2
R =---- Em] (13.6 )
r r - a cos Or
Knowing the range and the azimuth from the receiver, the target position
is found in the same way as for a monostatic radar. The target height can
be determined by introducing the elevation angle in a similar way.
(a) Target
Tx 2a Rx
(b)
(c)
Tx Tx
Figure U.S Methods of target location: (a) elliptical geometry; (b) theta-theta geometry;
(c) hyperbolic geometry.
Theta - Theta Geometry

This is used when the target azimuth is also measured at the transmitter site.
The target can be located by the intersection of two beams (see Fig. 13.5b)
in much the same way as a radio 'fix' is obtained. The technique is useful
when the transmitter is also a monostatic radar in its own right and is
measuring the transmitter azimuth. This method can also be employed if the
target jams the radar; in this case the radar transmitter is turned off and
the target is tracked passively from the two angular measurements made on
the jamming signal.
Theta-theta target location requires: 2a, the baseline; and et , en the
target azimuth measured at the transmitter and receiver.
Note that, as Fig. 13.5b shows, there are places near the baseline where
the accuracy is poor. It is generally true to say that the advantages of
multistatic radar are maximized when the target is at intermediate range and
near-broadside to the baseline.
Hyperbolic Geometry
This is used with two transmitters and one receiver. Here the target is located
from the time d(fference between the two transmitter signals arriving at the
receiver, which places the transmitter on a hyperbola. The receiver beam is
then used to fix the position on the hyperbola, as shown in Fig. 13.Sc. The
information needed is: 2a, the baseline; 2r, the difference in the round-trip
distances; and en the azimuth of the target, measured at the receiver.
From these measurements, the target can be located from the azimuth
and the distance to the receiver given by
a 2 _ r2
R =---- Em] (13.7)
r r + a cos er
A comprehensive table of target location equations, extended to three
dimensions, is given by Skolnik 2.
13.5 BISTATIC DOPPLER
Understanding the meaning of bistatic doppler frequency shifts is more

difficult than for the monostatic case. The change in the radio frequency
arises from the rate of change of the total path length travelled by the signal.
For monostatic radar this change in path length equals a change in range,
but for a bistatic system the doppler shift id is given by
- +
1 d (R t Rr)
[Hz] (13.8 )
.f
d - - .. - - - - - -
). dt
The negative sign indicates a decrease in doppler frequency for an increasing
path length.
An interesting case is that of the zero doppler contour, or isodop.
Figure 13.6a shows an aircraft flying in a circle around a monostatic radar.
(a) (b)
Tx.Rx
Figure 13.6 Zero isodop contours for (a) monostatic case and (b) bistatic case.
Target
Tx Rx
Figure 13.7 Geometry for evaluating the doppler shift.
Since the range does not change, there is no doppler shift and the aircraft
would be hidden by ground clutter (although it would not pose much of a
threat). In the case of a bistatic system, the iso-range ellipses represent the
general zero isodops, but there is also the particular case of an aircraft flying
along the baseline from the transmitter to the receiver; as R t increases, Rr
decreases by the same amount, and (as there is no net change in the total
path) we see zero doppler again. In general, the doppler frequency represents
a component of the target motion along a line roughly bisecting the bistatic
angle 21/1 as shown in Fig. 13.7.
When the transmitter and receiver platforms are moving, as in the case
of air-borne or ship-borne radar, the doppler frequency is modified by the
component of platform velocity along R t and R r • The best way to express
this mathematically is to define unit vectors U t and U r along R t and Rr
respectively, such that the doppler shift becomes
[Hz] (13.9)
where rs , r t , rr = position vectors of the target, transmitter and receiver (see

Fig. 13.7).
Monostatically, Eq. (13.9) reduces to the normal formula
2 2v
id = - (it - is)·ut = - - [Hz] (13.10)
A. A.
There is no distinction between target and system platform speed contri-

butions to the doppler shift.
In general, the doppler shift of a target can be evaluated from Eq. (13.9),
although the mathematics can get quite complicated for air-borne bistatic
radar when transmitter, receiver and target are all moving in different
directions at different heights. Bistatic clutter is more difficult to determine,
especially when wide beamwidths cause clutter spreading in the doppler
spectrum. One approach to predicting the masking of targets by clutter is
to plot clutter isodops and calculate the position on the locus of a target
track where the clutter and target speeds will merge.
A further bistatic complication concerns the synchronization of the
receiver with the transmitter in both time (in order to measure range
information) and frequency (needed to determine the doppler shift). There
are three ways of accomplishing the synchronization, although all have their
drawbacks:
• Locking the receiver local oscillator to the direct transmitter signal

received along the baseline. Platform motion at either end will cause
errors because this reference signal will be doppler-shifted when it arrives.
• Locking both transmitter and receiver oscillators to an external time
signal or beacon. This renders the radar dependent on an external signal,
which might fail, or which might be attacked in wartime.
• Using a rubidium frequency standard at both ends. This is expensive,
but for many applications the frequency stability is adequate for doppler
processing requirements. Relativistic effects due to platform motion are
rarely a problem.
One must not become too gloomy about the complications posed by
bistatic doppler and the problems of doppler spreading of clutter. First, this
is a subject amenable to computer simulation and prediction. Secondly, the
transmitter and receiver platform speeds can be arranged for the benefit of
the radar user. An example of this is clutter tuning, which concerns air-borne
detection of surface vehicles such as tanks; normally this is a difficult task
for monostatic radar because the high platform speed of the aircraft and the
finite width of the antenna beam causes a significant clutter spreading
problem (see problem 3.5 at the end of Chapter 3). Using a bistatic
arrangement, the speeds of the transmitter and receiver aircraft can be
MULTISTATlC RADAR 283
adjusted to give zero platform doppler shift over the area of interest and the
capability of being very sensitive to slowly moving land targets. This
procedure is known as engaging complementary motion.
13.6 APPLICATIONS
Perhaps the classic application of multistatic radar is in air defence, because

of the invulnerability of the receivers. A common design approach is to have
a monostatic radar with a narrow scanning beam illuminating a given sector.
A second (possibly concealed) receiver is used to increase the coverage area
and to evade jamming and attack. This second receiver needs to be equipped
with a staring array of antenna beams and to switch rapidly through them
at the appropriate rate in order to follow the region of sky illuminated by
the transmitter pulse-a process known as pulse chasing (see Fig. 13.8).
More receivers may be used to increase further the coverage and the resistance
to jamming.
The additional doppler information provided by a multistatic system
can be valuable for sea sensing. The problems of inverting the doppler
spectrum to find the sea state, discussed in Chapter 11, are eased somewhat
by using two radars to probe the sea surface from different directions. If
the two radars are also operated as a bistatic pair, then a third view is
generated crossing the bistatic angle, which provides more information for
very little extra cost.
Bistatic techniques are used extensively for targeting missiles; typically,
a monostatic radar illuminates a target and the missile carries an additional
Pulse --~-----'~~'-,,~
--·Staring· array
of beams
Tx Rx
Figure 13.8 Pulse chasing.

receiver so that it can keep watch on the target while following an optimal
trajectory that carries it outside the radar beam. Further details are given
in Chapter 3.
All these applications require bistatic and multistatic tracking algorithms
to be developed. Transferring the measurements onto cartesian coordinates
means that non-linear filtering is necessary, which in turn means more
computer processing power and complexity. Ifthe target is detected by more
than one receiver simultaneously, then tracking accuracy can be improved
by integrating all the data into the filter.
13.7 SUMMARY
Bistatic and multistatic radars differ in a number of ways from equivalent
monostatic systems. Some of the differences are unwanted and merely add to
the system cost and engineering complexity, but others can be exploited to
produce significant operational advantages.
Perhaps the biggest single advantage of multistatic operation is the
reduced vulnerability of the receiver to jamming or physical attack by
anti-radiation missiles. Another big advantage, in the case of mobile systems,
is that clutter tuning may be employed to increase the sensitivity of the
system to slowly moving targets. A higher PRF can be used with bistatic
radar than with monostatic, which also aids sub-clutter visibility.
The original reason for the application ofbistatic techniques in the 1920s,
which was to isolate the receiver from the transmitter when CW waveforms
were used, remains valid today and most large HF over-the-horizon radars
have transmitters and receivers separated by over 100 km (see Chapter 10).
These radars also follow the typical bistatic format of a single transmitter
beam and several narrow receiver beams. The cost of multi beam receiving
antennas, and other system complexities, means that multistatic radar will
never replace monostatic radar in general usage, but in certain applications
it remains a powerful technique.
Key equations
• The bistatic radar equation:
p = PtGtGrO"b}·2LtLr [W]
r (4n)3R?R;
• Distance from target to receiver for elliptical geometry:

r2 _ a 2
R = Em]
r - a cos Or
r
• The bistatic doppler frequency shift:

fd = -~ ~~R~_~r1 [Hz]
). dt
MVLTISTATIC RADAR 285
13.8 REFERENCES
I. Technical History of/he Beyinnin!Js 0/ RADAR, S. S. Swords, Peter Peregrinus for the lEE,
2. Radar Handhook, Ed. M. E. Skolnik, McGraw-Hili, New York, 1990.
Bistatic Radar, N. J. Willis, Artech House, Norwood, MA, 1991.

The geometry ofbistatic radar systems, M. C. Jackson, lEE Proc.-F, 133(7),604-612,1986.
[This is a particularly useful summary of bistatic geometry, which includes coverage,
bistatic resolution cells and PRF.]
13.10 PROBLEMS
13.1 A bistatic short-range air surveillance radar operates at a frequency of 3 GHz. An aircraft
flies directly towards the transmitter at 100m s - 1 •
(a) At the peint where its path brings it at right angles to the receiving beam, what
doppler shift is recorded?
(b) If the transmitter is a monostatic radar in its own right, what doppler shift would it
detect?
(c) Explain the difference.
13.2 Assume that the radar in problem 13.1 has a baseline of 30 km. An echo from a target
arrives at the receiver at an angle of 30" to the baseline and 200 lIS after the direct pulse from
the transmitter arrives.
(a) What is the round-trip distance 2r?
(b) Calculate the range of the aircraft from the receiver.
(c) Calculate the range of the aircraft from the transmitter.
Neglect the curvature of the Earth and the aircraft height and treat this as a two-
dimensional problem.
13.3 (a) In the problem above, if the target were travelling parallel to the baseline at 100 m s - "
what doppler shift would be recorded?
(b) Calculate the bistatic angle.
13.4 A I GHz, 1 W transmitting buoy is dropped in a shipping channel s!lch that ships must
pass within I km of it. Overhead, an aircraft is listening for the echoes with an antenna of gain
20 dB. Assume that the buoy has no useful antenna gain (assume a gain of unity) and the
aircraft receiver requires a signal strength of - 130 dB W for detection.
(a) What is the maximum height at which the aircraft could detect a ship of bistatic RCS
30 dB m 2 (neglect losses).
(b) Woald the air-borne receiver have a dynamic range problem with the direct signal
arriving only a few microseconds before the echo?
(c) If the aircraft had its own monostatic radar on board, would this need more, or less,
power than the transmitter on the buoy?
CHAPTER
FOURTEEN
ELECTRONIC WARFARE
• The jargon of electronic warfare

• The effectiveness of electronic warfare
• Measures and countermeasures
• Stealth
The jamming of radar systems has been going on since radar was invented,
and is now very sophisticated.
14.1 OBJECTIVES AND DEFINITIONS
A modern war is won by the side that best exploits the electromagnetic (EM)
spectrum. During the Gulf War in 1991, Iraqi radars were blinded by
electronic countermeasures and the opposing Allied forces were able to gain
control of the air. Once air superiority has been achieved, victory is often
just a matter of time. One of the key elements in a modern battle is how
well the attacking and defending radars of each side stand up to the
electromagnetic onslaught from the other.
The objectives of electronic warfare (EW) systems are simple:
• To deny the enemy the effective use of the EM spectrum.

• To protect friendly EM systems against EW attack.
286
ELECTRONIC WARFARE 287
There are three basic divisions of electronic surveillance:

1. Electronic support measures (ESM). This cryptic name describes the
whole field of passive electronic eavesdropping using special-purpose
intercept receivers. Within ESM, there are two subclasses, COMINT
(gathering communications intelligence) and ELINT (gathering elec-
tronic intelligence), with which we are concerned. ESM has quite a long
history, and during World War I, the interception of army field telephone
and naval radio traffic was exploited successfully by both sides.
2. Electronic countermeasures (ECM). This is the active disruption of enemy
EM communications and surveillance, and is often known as jamming.
ECM also dates back to World War I, when wireless communications
were deliberately interfered with by jammers. The difficulties with ECM
are that it not only presents a possible interruption to your own EM
signals, but it gives away your hostile intentions and can act as a beacon
for enemy munitions to home in on.
3. Electronic counter-countermeasures (ECCM or anti-ECM). For every
measure there is a countermeasure, and indefinite counter-counter-
measures. We will discuss some ECCM techniques in Sec. 14.3, but the
subject can be summarized as the art of making radars and other
electronic systems ECM-resistant by trying to null out the effects of
jammers or decoys.
A typical radar EW scenario is shown in Fig. 14.1. A coastal air defence
radar illuminates a flight of incoming enemy bombers, which protect
~
Escort jammer Stand-off
jammer
Self-screening jammers
Expendable
jammer
Figure 14.1 A typical radar electronic warfare scenario.

themselves by launching an anti-radiation missile (ARM) at the radar and

by turning on their self-screening jammers. ARMs are sometimes designed
to be fired at the sidelobes of a radar transmitting antenna to make the
detection of the tiny missile itself even more difficult. The defence against
ARMs includes separate radars to detect them, decoy transmitters, or even
the use of bistatic systems using relatively inexpensive transmitters that are
easily replaced.
The attacking flight is accompanied by an additional aircraft, which
carries no bombs at all; this escort jammer (EJ) is packed solely with EW
equipment to provide even more sophisticated confusion to the radar
operator.
The radar uses its ECCM capabilities to evade these jammers and may
fire home-on jammer (HOJ) missiles at the aircraft. HOJs use a similar
technology to the ARMs. There is one weapon the defensive system cannot
reach; at long range, within the safety of its own territory, a large enemy
aircraft loiters at high altitude providing refined high-power jamming and
other EW support; this is known as a stand-off jammer (SOJ).
One last type of ECM that may be used is an expendable jammer, which
can be dropped by aircraft, carried by a remotely piloted vehicle (RPV),
fired by artillery or even delivered by enemy agents, who may park an
innocuous looking car, full of ECM electronics, near to a radar installation.
Although expendables are short-term devices, they can provide enough time
for a target to escape.
Unfortunately, the whole of this branch of electronic engineering is
riddled with the type of abbreviations and acronyms we have seen above.
You really do need to know the meaning of many of them if you wish to
converse knowledgeably about military radar systems. A more complete
classification of EW is given in Johnston in reference 1, and a catalogue of
current EW systems may be found in Jane' s Radar2.
The effectiveness of all these EW techniques and weapons depends upon
the ingenuity of their design and the performance of the jammer versus the
echo power from the radar signal.
14.2 NOISE JAMMING AND THE RADAR EQUATION
Swamping radar or communication signals by transmitting high-power noise

is one of the earliest and most widely used forms of jamming. In a typical
application, an air-borne ESM receiver is used to detect hostile EM emissions
and the signals are analysed to identify the radar. If the radar is thought to
pose a threat, the pilot is warned by a radar warning receiver (R WR ) display
in the cockpit. The RWR display can give a simple indication of the target
threat sector on the head-up display, a range-bearing display (as on the F15
Eagle) or an alphanumeric display (as used on the Tornado F2). The usual
response to a threat is for the aircraft ECM system to begin automatically

jamming the radar by transmitting noise on the same frequency.
When a target is detected clear of clutter problems, receiver noise becomes
the factor limiting the radar performance. The effect of the jamming signal
is to raise the noise floor and reduce the operating range. Since the radar
signal suffers R4 losses and the jamming signal only has to travel one way
and experience R2 losses, the advantage is usually held by the jammer.
Worked example If an L-band surveillance radar of the type designed

in Chapter 2 used a peak transmitter power of 200 kW and an antenna
gain of 36 dB, at what range could it detect an aircraft with an RCS of
10 dB m 2? What increase in noise power would reduce the radar
detection range to one-third? If the jammer antenna on the aircraft had
no useful gain, how powerful does the ECM transmitter need to be?
SOLUTION The maximum detection range is given by

R _ PGG -2Lsys )1/4
t t raA.
max - ( (4n)3N(SNR)
Putting in values:
Pt = 53 dB W
GtGr = 72 dB
a = 10 dB m 2
),2 = -12.7 dB m 2 ), = 0.231 m
Lsys = -7.0 dB Typical
1/(4n)3 = -33.0dB
I/N= +145.8dBW- 1 3 fJ.s pulse, 3 dB noise figure
I/(SNR) = -13.0dB
R!ax = 215.1 dB m 4
Rmax = 238.5 km
If Rmax is to be reduced by a factor 3, the noise power must be increased

by a factor of 81 because of the (noise) - 1/4 dependence above. The noise
floor must therefore be raised by 19.1 dB to -126.7 dB W- 1 • Try putting
this noise level into the radar equation to convince yourself that the
detection range drops to just under 80 km.
The one-way jamming signal received by the radar is given by
2
P = PtGtLt x Gr A. Lr
r 4nR2 4n
and so
The jammer transmitter therefore needs a power of
Pr = -126.7 dB W
(411:)2 = 22.0 dB
R2 = 107.5 dB m 2 Aircraft range of 238.5 km
I/GI = 0 dB
I/Gr = -36 dB
1/..1? = 12.7 dB m- 2
llLlotal = 7.0dB
PI = -13.5 dBW =45mW
COMMENT It is quite staggering to realize that a surveillance radar with

such a large power x aperture product can be so effectively disabled by
so little jamming power. As the aircraft moves nearer to the radar, it
can still keep the detection range below its own range because the jammer
R2 term always dominates the radar R4 term.
14.3 TYPES OF ELECTRONIC COUNTERMEASURES AND

ELECTRONIC COUNTER-COUNTERMEASURES
The objectives of an ECM system are either to deny the radar an opportunity
to detect a target, as in the worked example above, or to deceive the radar
into following either the wrong target (such as a chaff cloud) or a
non-existent target that has been generated electronically.
The simplest method of denial is spot jamming, in which the noise power
is concentrated into the radar receiver bandwidth. The radar can counter
this by hopping in frequency from pulse to pulse (which also increases the
probability of detection because it decreases target fading effects).
The counter to frequency 'hopping is for the ECM system to use multiple
spot jamming or to begin barrage jamming over an entire radar band. There
are at least two ways the radar can counter barrage jamming: by using a
look-ahead receiver to find the weakest point in the barrage (because jamming
antennas are not perfectly frequency-independent) or by adopting frequency
diversity and changing to an entirely different band (sometimes to a different
radar system in fact). Frequency diversity is more effective with higher-
frequency radars because the antenna problems are easier to solve. One
possible ploy is to operate in one radar band and mode (e.g. pulse) during
peacetime and to switch suddenly to an entirely different frequency band
and mode (e.g. FMCW) in the event of war.
Dispensing chaff has been a successful ECM denial technique since

World War II, although it is now only effective for slow-moving targets
because of the development of doppler processing. If a radar has locked on
to a slow-speed target, such as a ship or tank, it may still be possible to
eject a cloud of chaff along the line of sight such that the radar initially
tracks both target and chaff, but remains locked on to the latter as it drifts
away in the wind. Possibly the best ECCM measures for dealing with chaff
are improved doppler processing and the use of polarimetry to desensitize
the radar system to the long, thin shape of chaff while maximizing the echo
from the expected target shape. Polarimetry is not without problems,
however, since different polarizations rarely propagate equally.
Sometimes it is not the radar target that undertakes electronic counter-
measures, but a separate stand-off jammer, which attacks the sidelobes of
the radar antenna. The ways of countering this threat are:
• To build antennas with very low sidelobes.

• To use a separate omnidirectional antenna to pick up the jamming signal
for use in a coherent cancelling device.
• To use adaptive nulling, and effectively put 'holes' in the antenna pattern
in the direction of the jammer.
Other general defences against denial jamming including firing HOl

missiles at the jammer, trying to burn through the jamming (by increasing
the power x aperture x integration time product in the direction of the
target) and turning the radar off and trying to follow the jamming signal
using passive tracking techniques.
Noise jamming is relatively easy to detect and, as we have seen, there
are a variety of countermeasures, some of which can be dangerous for the
jammer. For this reason, deception jamming (to fool the radar or its operator
with 'spoof' transmissions) has become more popular.
An active method of deception jamming is a repeater jammer, which
generates false echoes by receiving a radar pulse, delaying it in a digital
memory and retransmitting it to appear as targets at other ranges. A
transponder jammer uses the incoming pulse to trigger a suitably modulated
reply to the radar, which may be delayed until after the radar beam has
swept past so that it appears at the wrong azimuth.
One defence that a target can use, when it has been detected, is a range
gate stealer and deliberately strengthen its own echo with a larger radiated
pulse. The timing of this artificially enhanced echo is then slowly shifted to
change the apparent range; hopefully the radar tracker continues to follow
the larger 'target'. The range gate stealer is often countered by jittering the
PRF in a quasi-random manner. Similarly, a velocity gate stealer can be
engineered to falsify the target speed by tuning the reply frequency.
There are many other ECM tricks designed to break the tracking
performance of radars. Provided one knows exactly how the enemy radar
works, it is relatively easy to defeat it. A good guide on the use of ELINT
to analyse enemy radar is given by Wiley 3. The key to ECCM is for a radar
not to be detected in the first place and to have a low probability of intercept
(LPl). There are many ways of achieving LPI, but they are all based on
spectral spreading of the radar signal by such techniques as FMCW, long
pseudo-random codes, etc. Radar systems can accurately compress these
signals on reception to a narrow final bandwidth. The intercepting ECM
receiver, not knowing the compression algorithm, must receive these signals
over the full swept bandwidth and so suffers a large noise penalty. Another
defence against ELINT and ECM is to use separate peacetime and wartime
operational modes, and to transmit deliberately misleading surveillance-type
signals during peacetime to try to get the enemy to develop the wrong type
of jammer.
One final deception system that should be mentioned here is the use of
decoy RPV aircraft. These are relatively inexpensive and can be flown in large
numbers at the enemy until their air defence weapons are exhausted. The
possibility that the enemy radar can recognize the decoys and ignore them
can be countered by placing a nuclear or biological weapon on the occasional
one to ensure that they must all be treated seriously.
The ECM-ECCM game can be played indefinitely. There are many
sophisticated modifications that can be made to radar transmitter waveforms,
antenna patterns and receiver techniques to defeat denial and deception
jamming. However, it is probably also true to say that the countermeasures
to a radar are always less expensive to build than the radar itself. The final
arbiters in electronic surveillance are probably, therefore, espionage (to know
how the countermeasure will work in wartime) and money (to fund the
development of a counter to it). All of this expensive game can be invalidated
if targets are not visible to radar in the first place; this has been the driving
force behind the Stealth programme.
14.4 STEALTH APPLICATIONS
Stealth is the art of concealing targets from detectors. For submarines, it is

important to be acoustically Stealthy; ships and aircraft probably need to
balance acoustic, visual, infrared and radio Stealth in proportion to the
threat posed by the sensors in each of these spectra. The threat of radar
detection is usually a major concern to an attacking force and a lot of effort
is put into trying to reduce radar signatures.
The subject of Stealth arose from the analysis of Allied bomber losses
during World War II in which the tiny, but very fast, wooden Mosquitoes
(carrying a bombload almost as large as a B17) fared very much better than
contemporary larger and slower metal bombers. After the war, the Northrop
'Flying Wing' and the Avro Vulcan bomber showed that the aircraft shape,
as well as the construction material, was important in the reduction of the
RCS.
A change in tactics also occurred after the war; the large defensive
bomber formations, and escorting fighters, were replaced by solo nuclear
missions in which single bombers were required to navigate through enemy
defences alone. Escaping enemy ground fire by flying at very high altitudes
was eventually abandoned after the U2 of Francis Gary Powers was shot
down in 1960. Flying at low altitudes was dangerous because of the accuracy
of radar-controlled anti-aircraft artillery, unless the aircraft flew very low to
get beneath the radar defences. The solution lay in adopting an idea tried
in World War I, to make aircraft 'invisible' by covering the airframes in
translucent materials, and then extending the concept to include the radar
spectrum (see Sweetman 4 , for example).
The maximum radar detection range (Eq. ( 1.15 )) can be summarized as
[m] (14.1 )
The RCS u must be changed dramatically if the radar performance is to be
seriously affected. Computer modelling of airframes and ship structures
shows that most of the scattering comes from flat surfaces and corner
reflectors, formed where different parts of the structure join together.
Table 2.1 gave the RCS of flat plates and corner reflectors; try putting a few
numbers in to get a feel for the values that can occur. The RCS of most
structures can be dramatically reduced by choosing an inherently low RCS
design and then using smooth, continuously curved shapes and avoiding any
right angles. In the case of an aircraft, this avoidance of right angles may
mean that a single fin cannot be used and the aircraft must be designed
without one (the flying wing approach used in the B2 bomber) or with two
inclined fins (as used on the F 117 fighter).
After shaping, any remaining 'hot spots', such as the reflections from
engine air intakes on aircraft, can be covered with radar-absorbing material
(RAM). RAM is generally some form of lossy dielectric material built into
a wafer, which is tuned for maximum absorption in a particular radar band.
Care also needs to be taken to ensure that the radar antennas on ships and
aircraft do not act as RCS hot spots.
In a battIe scenario, Stealth is a very effective counter to radar, especially
when other ECM activities are going on to reduce the radar efficiency. There
are, however, a few antidotes to Stealth. Wideband carrier-free impulse radar
may be used to try to exploit RCS weaknesses in the radio spectrum (see
Chapter 12) and, in general, radar needs to move to lower frequencies to
combat Stealth. Over-the-horizon radar (Chapter 10) may be effective
because the scattering mechanism is more related to radar-induced currents
flowing throughout the structure of the target than from the individual
scattering facets on the surface that dominate the RCS at microwave
frequencies. Skywave OTH radar is particularly useful because it looks down

on air targets, the aspect from which they present the largest scattering
cross-section. Bistatic radar has also been cited as a possible countermeasure4
because it is hard to design Stealth shapes against an unknown radar
geometry.
14.5 SUMMARY
The very success of radar at detecting and tracking military targets means
that it has become the target of electronic attack itself. As soon as a military
radar system is turned on, its EM emissions will be detected unless it uses
spread spectrum LPI techniques to evade detection. Any radar signals
detected will be analysed by ESM receivers and an ECM strategy formulated;
this may include the use of Stealth to reduce the RCS of the target. While
electronic countermeasures are effective, and expensive to defend against
using ECCM techniques, the balance at present probably resides on the side
of radar, provided the waveforms are controlled intelligently.
14.6 REFERENCES
1. Modern Radar Techniques, Ed. M. J. B. Scanlan, Collins, Glasgow, 1987. [Chapter 4 is an

organized guide to ECM and ECCM.]
198.9.
3. Electronic Intelligence: The Analysis of Radar Signals, R. G. Wiley, Artech House, Dedham,
MA,1982.
4. Stealth Bomber, B. Sweetman, Airlife Publishing, Shrewsbury, 1989.
14.7 PROBLEMS
14.1 Consider a radar speed trap based on a 10 GHz 'unity' radar system (I W, unity antenna
gain, zero losses, I JlS pulse). What is the SNR for a single-pulse detection of a car at a range
of 100 m if the car has a radar cross-section of 10 m 2 ?
14.2 If the motor car in problem 14.1 contained a spot jammer, how much would this need to
increase the noise level in the radar receiver to reduce the range to 10 m for the same received
SNR?
14.3 What noise power would the jammer have to radiate to achieve the result in problem 14.2,
assuming the noise was all concentrated in the same bandwidth as the radar signal?
14.4 If the radar trap above were to use random frequency hopping over the range 10- 30 GHz,
and the jammer were to counter by barrage jamming the entire band, by how much would the
total jammer output noise power have to be raised?
14.5 If. as an alternative. the radar in problem 14.1 were to attempt to burn through the spot
jamming using (perfect) coherent integration. how many pulses would need to be integrated?
14.6 Suggest a suitable PRF for the radar and estimate whether the integration time determined
by problem 14.5 would give adequate doppler resolution to book a motorist for speeding.
14.7 What would seem to be the most economical strategy to avoid a fine for speeding:
(a) to build an in-car radar warning receiver;
(b) to develop a radar jammer;
(c) to cover the car in Stealth materials?
CHAPTER
FIFTEEN
RECENT DEVELOPMENTS
• Electronic beam steering

• Transmitting arrays
• Combining radar ouptuts
• Improving resolution
Electronic methods of steering beams, communication between radars and

advanced processing give modern radar systems considerable flexibility.
15.1 INTRODUCTION
In this chapter we look at current radar technology and find out how it is
being used. We will try to predict the future of radar in the next chapter.
The antenna has been the subject of some of the most interesting radar
developments in recent years. Domestic satellite dishes are representative of
the old level of antenna technology because they must be mechanically
pointed at their target. Radar versions of these parabolic reflectors are rotated
by heavy and expensive (but reliable) turning gear, and a target can only
be observed once every few seconds when the dish faces that direction.
At the centre of a reflecting antenna, such as a satellite dish, is the device
that illuminates it, called the feed. Feeds can be dipoles, microwave horns
or other devices for radiating or receiving electromagnetic waves. The purpose
of the reflector is merely to increase the collecting area, and hence the gain,
of the antenna. However, a large collecting area can also be achieved in an
entirely different way. It is possible to arrange many dipoles on a flat plate
296
RECENT DEVELOPMENTS 297
and connect them together in a phase-coherent manner to form a phased

array. In the UK, one domestic system for receiving satellite broadcasts was
constructed in this way.
Phased arrays are often considerably more expensive than reflector
antennas but they have the advantage that the beam can be steered
electronically rather than mechanically. If the phasing of a receiving antenna
is carried out digitally by computer calculation, rather than by analogue
devices (which may be controlled by a computer), the process is known as
digital beamforming, and there is considerable flexibility for beamshaping
and the formation of multiple beams.
Phased arrays are now in widespread use for receiving, but transmitting
versions are only just coming into service because of the difficulties of
building, and controlling, large arrays of small transmitting modules. These
active arrays represent the current level oftechnology for single radar systems.
An entirely separate development area has been concerned with netting
independent radars (and other sensors) together to improve target detection
and tracking accuracy. We briefly examine this field, which is sometimes
known as multihead radar.
Finally, we take a brief overview of the recent advances made in signal
and data processing, which have enabled radars to go beyond their original
function of radio detection and ranging.
15.2 PHASED ARRAYS

An array of radiating elements can be one-dimensional (linear array) or
two-dimensional (planar array). Arrays work the same way on both transmit
and receive, but it is often easier to follow the theory for the transmitting
case because of the analogy in optics with Young's double slit experiment
and diffraction gratings.
A simple linear array, with element spacing d, is shown in Fig. 15.1. If
all the elements are excited in phase, then a wavefront is radiated broadside
to the array. If a successive time delay is introduced across the array, causing
a phase change tjJ between elements, then the wavefront will be radiated at
an angle () given by
tjJ d sin ()
[ ] (15.1 )
2n },
The field Ea «()) from N elements can be found by summing across the array.
Using a little trigonometry the summation comes out as
E «()) = E1 «()) (sin[ Nn (dj 2) sin ()])
(15.2)
a 7N sin[n(dj2)sin()]
where E 1«()) = field due to a single element [V m -1].
Broadside
lr
N
Figure 15.1 A linear array of dipole elements.
In optics, the field arising from a diffraction grating has two components;
the diffraction pattern from a single slit and the interference pattern created
by the interaction of coherent signals from different slits. It is exactly the
same with antenna arrays. Each element of the array acts as a source with
its own diffraction pattern E 1 (0), and the radiation from the sources interacts
to form an interference pattern, which is usually called the array factor (AF)
given by
- Ea((J) -
AF-~-- J N (sin[Nn(djJ.)SinO J) [ J (15.3 )
Ed8) N sin[n(dj}.) sin OJ
The radiating elements are usually simple devices such as dipoles, slots,
patches or sometimes small horns. All these elements have fairly wide
diffraction patterns and the antenna pattern is determined by the much
narrower array factor. Even so, care must be taken not to steer the array
beam outside the pattern of the individual radiating elements. A normalized
array pattern is shown in Fig. 15.2 in comparison with the diffraction pattern
of a single half-wave dipole.
Part of the purpose in rearranging Eq. (15.2) to get Eq. (15.3) is because
the term in large parentheses in Eq. (15.3) can be shown to have a maximum
value of unity. This means that the field strength has been increased by N J
times the field from a single element, and therefore the power gain of the
array has been increased by N. Some other properties of array patterns are
now outlined.
The half-power beanmidth of the pattern steered off-broadside through
an angle 8 can be found from Eq. (15.3) by setting the expression in large
Normalized array factor (= AF/yN)
cos [(1T/2) sin e]

- - Seven-element array /'
/' -
1.0
", cos e
- - - Single element /
/
"" for a half-wavelength dipole
/
/
/
'-'-'-/
,
/
/
,,
/
/
,,
/
/
/
, \
/ \
/ \
/ \
/ 0.4 \
/ \
/ \
/ 0.3
\
/ \
/ \
/
e [degrees]
I I [ I I [ [ I [ I I •
o 0.2 0.4 0.6 0.8 0.9 0.99 1.0
sin e
Figure 15.2 The normalized array factor for a seven-element array, of spacing d = i./2,
compared with the diffraction pattern of a single element.
parentheses to JO.5. For large N, this leads to

~e = O.886i. [radians] ( 15.4)
Nd cos e
For the broadside pattern this reduces to
~e = O.886i./ D [radians] (15.5 )
where D = overall length of the array = N d. This, of course, corresponds to
the i./d approximation for antenna beamwidth that we have seen earlier in
the book. The width increases by 1/cos e as the beam is steered away from
broadside because the effective length of the array is reduced by cos e when
viewed from an angle e off-bore-sight.
A convenient approximation of Eq. (15.5) in degrees, to remember for
rough calculations, is
[degrees] ( 15.6)
The beamwidth factor (the aperture in wavelengths needed for a one-degree
beam) is about 51 for the uniform illumination assumed in Eq. (15.6), but
is between 60 and 70 for tapered distributions, according to the sidelobe
level required. The beamwidth factor is also higher for apertures that are
round, rather than rectangular.
Nulls occur in the pattern at angles 0nuJl given by
sin 0nuJl = ±kA/Nd k = 1,2, ... [ ] (15.7)
Often, element spacings near )./2 are used, giving only one mainlobe (plus
one to the rear, if the array is' not mounted above a reflecting plane). For
wider element spacings, grating lobes (other principal maxima, or 'secondary
beams') occur when
sin Ogl = ±H/d k = 1,2, ... [ J (15.8 )
With careful design, the individual element pattern can sometimes be used
to null out grating lobes near 90°, as shown in Fig. 15.2.
Side lobes (secondary maxima) occur at
. 0 (2k + 1»),
sm 51 = + -----'- k = 1,2, ... [ J (15.9)
2Nd
The first sidelobe has a power gain of - 13.4 dB with respect to the peak
gain of the array. This is a well known result arising from a rectangular
illumination function. Those of you familiar with Fourier transform (FT)
theory will realize that the FT of a rectangular function (in this case the
aperture) gives rise to a (sin x )/x function of a similar form to the array
factor. Equation (15.3) can be approximated, for small values of 0, by an
expression of the form
E.(O) ~ IN sin[n(D/).) sin OJ

[n(D/).) sin OJ
( 15.10)
Thus the far-field pattern of the array can be represented by the FT of the
spatial illumination function, provided that the element spacing is close
enough to avoid grating lobes. For this reason, FTs are used extensively in
antenna theory.
Some of you may be confused at this point because you have only used
the FT to transform between the time domain and the frequency domain.
Table 15.1 shows you how to relate this new way of using the FT to the
more usual time/frequency form. Bracewell! is particularly good at covering
this use of FTs. These days, there are some very readable books on
antenna theory, and on arrays in general. If you wish to explore further, try
references 2-4, especially 4.
Table 15.1 Use of Fourier transforms in antenna theory
Waveforms Antennas
Time. t Aperture distance in wavelengths. Nnd/;.

Waveform, x(t) Aperture field distribution, E(Nnd/),)
~ ~
Angular frequency, w Direction, sin II
Spectrum, X (lOn) Field radiation pattern. E(sin II)
x (t) [V] E(NTrd/)..) [Vm-I]
----J I L---. t [s] ----J L--. • Array length in

TOT NTrd NTrd wavelengths I 0 I
2 2 ~ ~
~
*
X(w): T sin (wT/2) E(sin Il) '" E1(sin Il)sin INTr(d/)..) sin III
(wT/2) NTr(d/)..) sin H
7""" I ~ \ 7" 'x tw /' ' , J \ , r "- • Sin ij

~
If all elements are fed equally, an array has a rectangular illumination

creating sidelobes only 13.4 dB smaller than the mainlobe, as discussed above.
Sidelobes this large are generally fatal for a radar system, owing to the
difficulty of separating small targets in the mainlobe from larger targets in
a sidelobe.
Low sidelobes are achieved through accurate element and array design
and the application of a weighting or window function, such as triangular,
raised cosine, Dolph-Chebyshev polynomials, etc., to the array. The effect
of weighting is such that the elements at the centre of the array are fully
utilized but those towards the edge are increasingly attenuated, thereby
causing a wider beam and a loss of gain and SNR; unfortunately, these are
necessary afflictions.
Worked example Consider the air-borne surveillance radar, shown in
Fig. 15.3, operating in look-down mode to detect cruise missiles flying
across the sea surface below. If the doppler shift of main-beam echoes,
caused by the velocity V of the aircraft, are compensated in the signal
processing, what would be the apparent speed of a large oil-rig in the
forward sidelobe? How can the problem be corrected?
Target
v
Oil-rig
Figure 15.3 An air-borne surveillance radar and antenna pattern showing sidelobe structure.
SOLUTION Stationary targets in the main beam will have an apparent

motion of V cos e towards the aircraft, which must be removed in the
processing. The echo from an oil-rig appearing in the forward lobe will
have a relative velocity of V. After processing, this velocity will appear
as V - V cos e. The oil-rig will therefore appear to be moving towards
the radar and will be detected as a possible hostile target.
There are two possible solutions to the problem. One is to use the
tracker to show that the apparent doppler shift of a target does not
correspond to its rate of change of position when plotted over many
scans. The problem with this approach is that, when flying near land,
an antenna with large sidelobes may detect hundreds of 'moving' targets,
which completely swamp the tracker, thereby allowing the smaller, real
target to escape detection. The other solution is to use antenna arrays
with very low sidelobe levels so that sidelobe echoes occur infrequently
enough for the tracker to cope.
COMMENT Perhaps the best known air-borne surveillance radar is the

Westinghouse E-3 air-borne warning and control system (A WACS).
This uses a large phased array mounted inside a saucer shaped radome
mounted above the fuselage of the aircraft and rotating once every 10 s.
The S-band array employs over 4000 elements and has exceptionally
low sidelobes. AWACS has been so successful that an improvement
programme is under way that will take until 1998 to complete (more
details can be found in Jane's Radars).
The processes of beam/orming and beamsteering are identical. A pro-

gressive time delay is applied to the elements, creating the tilted wavefront
shown in Fig. 15.4. A beam steered through an angle es has an array factor of
AF = jN {sin[Nn(d j ). )(sin e - sin es )] } [ ] (15.11)
N sin[n(dj).)(sin e- sin es )]
As an alternative to using a time delay, at a single frequency the steering
can be applied to incrementing the phase between elements by
t/J = 2n(dj).) sin es [radians] (15.12)
Worked example Calculate the beam pattern for a five-element array

with an element spacing of one wavelength. Is this a useful design? Hint:
try calculating the pattern for a beam steered 50°, before deciding.
SOLUTION The normalized array factor is shown in Fig. 15.5a. Grating

lobes appear near ± 90°. It might be thought that the individual element
diffraction pattern would largely eliminate these grating lobes, allowing
the wide element spacing to be used to give an extra-narrow mainlobe.
Normal
Beam
r
Figure 15.4 Beam steering by the introduction of a progressive time delay between array
elements.
Unfortunately, if Eq. (15.11 ) is used to calculate the steered array factor

the resulting grating lobe, shown in Fig. 15.5b, is fatally large.
COMMENT The difficulties of steering array patterns to wide angles

effectively limits most systems to ± 60 The angle scanned can be
0
•
increased by using two arrays, for example on adjacent sides of a building.

This arrangement is used by the huge US Pave Paws early-warning
radar for detecting ballistic missiles. Each array uses 2000 elements
arranged on a circular plane 30 m in diameter, and the two systems
0
cover a 240 sectors. The UK version of Pave Paws, on Fylingdales
Moor, uses three arrays giving a 360 capability.
0
Other solutions to the problem of limited scan include circular arrays

and conformal arrays. A circular array of elements can be used to give good
0
360 azimuth coverage, but it has the difficulty that amplitude as well as
phase has to be controlled as the beam scans if a reasonable sidelobe level
is to be obtained. Also, it becomes defocused at high elevations unless the
phasing is changed. Curved arrays, conformal to the fuselage of an aircraft,
are also an attractive idea in principle, but difficult to engineer in practice.
Grating lobes can be avoided, even when steering, if the element spacing
is restricted. Rearranging Eq. (15.8) gives the spacing limit for a steering
angle es :
d<i.( 1 + sin1 e _~)

s N
Em] (15.13 )
(a) Normalized array factor
e [degrees]
(b) Normalized array factor
e[degrees]
l
Grating lobe
l
Figure IS.s The beamshape for a five-element array with d = A.: (a) the broadside pattern;
(b) steered 50° off-boresight.
For the problem above, this gives a spacing of near ), for a broadside beam,
near ), / 2 if the beam is steered to ± 30° and near ), / 4 if the beam is to be
steered to ± 60°. When weighting is used, a slightly smaller element spacing
is needed and a suitable approximation is given by Radford 6 :
d
<
.
I.
(I
1+ sin 8s -
1.5)
N Em] (15.14 )
Both time-delay and phase-delay beamforming and steering are used.

Analogue time-delay beamformers are somewhat slow and clumsy, involving
the use of relays to switch lengths of cable in and out, but there are no
frequency restrictions and simultaneous operation over a wide band is
possible. Phase delays can be introduced by a variety of means, illustrated
in Fig. 15.6, which include:
• Diodes to switch between alternative path lengths or to change the

reactance of loading stubs.
• Ferrite phaseshifters; these use a cylindrical ferrite rod, inserted in a
waveguide, which can be magnetized.
• An array of hybrids, known as a Butler matrix, arranged in a similar way
to an FFT algorithm.
• A preset resistor matrix combining signals of different phases to give
the appropriate value.
Surprisingly, it is not necessary to maintain precise phase and amplitude

values at each element on the array. The summation of the field from many
elements tends to cancel out small errors, and consequently phases are
approximated in binary steps, often with as few as three bits giving 45° steps
in phase. The loss in main-beam gain from a 3-bit phaseshifter is only 0.23 dB,
falling to 0.01 dB for a 5-bit device 6 .
The problem with steering by phase, rather than by time delay, is that
changing the radar wavelength alters the steering angle. This frequency
dependence has been exploited in the past as a method of electronic scanning
(a given frequency corresponds to a given direction), but with modern
electronic warfare the radar designer needs the liberty to use frequency in
a much freer way. The upshot of this wavelength dependence is that most
modern phased array radars are restricted in instantaneous bandwidth, with
a limit (as a percentage of the carrier frequency) roughly equal to the
beamwidth in degrees (see Eq. (10.2)). This limit equates to a pulse length
restriction related to the aperture size, as in Eq. (10.3).
So far we have only discussed linear arrays, while most radars use
two-dimensional planar arrays. In many respects, planar arrays are similar
to linear arrays. The simplest way to develop a planar array is to form a
stack of linear array planks and vary the phase between the planks to
Diodes
Ferrites
Signai--+-.......-l
Control
Waveguide
Butler matrix Resistor matrix

Array
Beam ports
Figure 15.6 Methods of phaseshifting between array elements.
beamform in the second dimension. Other designs use triangular or rect-

angular lattice layouts of elements on the plane. The increased complexity
of these arrangements requires larger computational resources to control the
phaseshifters and creates something of a wiring distribution problem.
The wiring problem has been tackled in recent years by using fibre optics
to control the phaseshifters and, in a further development, optical beam-
formers have been constructed. Optical fibres avoid most of the problems
ofRF cables, being lightweight, lost-cost, low-loss and immune to interference

and crosstalk. These attractive properties of optical transmission in man-
ipulating microwave signals have given rise to an emerging technology known
as microwave optics.
One of the additional problems of planar arrays is that moving away
from the normal causes a form of conical distortion because azimuth and
elevation measurements are no longer independent. Corrections have to be
made when tracking off-axis targets, which again adds to the complexity of
the system and restricts performance at large angles off-bore-sight.
15.3 DIGITAL BEAMFORMING
In general, the increase in flexibility of radar antennas is matched by an

increase in cost. The early mechanically rotating designs are the least
expensive, followed by mechanically rotated arrays that beamform electroni-
cally in elevation. A full three-dimensional radar (a two-dimensional array
plus range) using an analogue beamformer is very expensive and the beams
are relatively difficult to set up. But there is still one further degree of
adaptability (and cost) to be gained from using arrays. The ultimate flexibility
is to have one receiver and AID converter per element and to form beams
by performing calculations in a computer on the AID samples. This is digital
beamforming. Sometimes, the amount of hardware needed can be reduced
by combining local groups of elements to form sub-arrays, and using a
receiver for each of these.
There are tremendous advantages to digital beamforming:
• Many independent beams can be formed simultaneously.

• The beams can be made very agile.
• Long dwells are possible.
• Sharp nulls can be created in the beam pattern to suppress jamming
signals.
• The beam pattern can be adapted to enhance angular resolution.
• If the AID samples from each array element are stored, beams can be
formed off-line, at a later time.
The disadvantages of digital beamforming are high cost and performance

limitations. The latter are determined mainly by the technology of the AID
converters. Slow converters with few bits make an array narrowband with
a limited dynamic range. As technology improves, so does the performance
of this type of array.
A digital beamforming system usually has a relatively simple receiver
for each element, which down-converts the frequency into I and Q (in-phase
and quadrature) channels for the AID converter (see Fig. 15.7). Real-time
Array element
I
Low-noise pre-amp I
I
I
Image rejection mixer ·~FirstLO
I
I
IF filter I
I
IF amp I
I
-'
T
Mixer Second
LO
Video amps
Sample and
hold
T f
Figure 15.7 Block diagram of a typical digital beamforming system.

$
beamforming takes place by multiplying these complex pairs of samples by

appropriate weights in multiply/accumulate integrated circuits (ICs). The
array output is formed from
N-l
Array output = L v"w,. e-j21tn(dl).) sin 8 Cn [V] (15.15)
n;O
where v" = complex signal from nth channel, w,. = weighting coefficient,
e - j21tn(dl).) sin 8 = steering phaseshift and Cn = correction factor. Corrections
are necessary for several reasons. These include errors in the position of the
element, temperature effects and the difference in behaviour between those
elements embedded in the array and those near the edge, caused by coupling
effects with neighbouring elements.
Digital beamforming is now used extensively for receiving arrays, where
sufficient funding is available, and it can be found in use from small HF
sea-sensing systems up to large-scale surveillance radars. An example of a
system in service is the Patriot radar system, described in Chapter 3.
15.4 ACTIVE ARRAYS
When an array is used for transmitting, as well as receiving, it is said to be

an active array, and represents the most up-to-date (and highest-cost) form
of antenna. An active array has a number of separate modules, each with
its own amplifier and phase control. There are considerable advantages to
the use of an array of small transmitting elements (rather than one large
transmitter), including the flexibility common to receiving arrays plus the
possibility of very large power x aperture products. The disadvantages are
those of yet further increase in cost and complexity.
One of the main difficulties of active arrays is that not all phaseshifters
(ferrites, for example) are reciprocal, and a different phase is received from
that transmitted. Furthermore, the transmitting beam may need to be wider
than the receiving beams to encompass a block of four mono pulse beams.
The end result is that transmitting and receiving phaseshifters need to be set
up independently, leading to designs similar to that shown in Fig. 15.8. These
modules require solid-state gallium arsenide transmitter ICs as well as the
receiving chip set. The power output per module is generally of the order of a
few watts.
An example of an active array is the UK multifunction electronically
scanned adaptive radar (MESAR) developed by the UK Admiralty D RA (M )
and Siemens-Plessey Radar. This is an S-band array of 918 elements using
4-bit phaseshifters. Each element uses GaAs ICs for the microwave circuitry
and silicon ICs for control. Array thinning is used to improve the transmit
sidelobe level.
PIN diode
Power amp receiver
- few watts protection
Receiver
Driver pre-amp
Phaseshifter Phaseshifter
r, r------:l
I/\-./..----'''DIA ~ A~dress for I
L -,~L!~~ ele~~~J
To other
elements
Signal
in/out
to beamformer
Figure 15.8 Block diagram of an active array element.
The MESAR design demonstrates the current level of technology.

Generally speaking, very-high-speed silicon ICs (VHSICs) are needed to
carry out all the calculations and gallium arsenide microwave monolithic
ICs (GaAs MMIC) are needed for the RF components and fast AID
converters. The development of these technologies, and of fibre optic control,
will be crucial to the future of both digital beamforming and active arrays.
15.5 MULTIFUNCTION RADAR
Given the flexibility of phased array radar, the traditionally separate roles of
surveillance, tracking and even radio communications can be taken over by
a single multifunction radar system. The system's resources (beams, receivers,
computers) must be allocated efficiently to carry out all the required
functions, which might include:
• Continually searching the horizon for new targets.

• Taking confirmatory looks at potentially new targets.
• Tracking many existing targets, with a scan rate adjusted to meet the
target characteristics.
• Tracking out-going missiles.
• Providing high-power, narrow-beam communications.
Of course, the problem is the same as with every surveillance radar-there

is never enough time to do everything, and a priority list must be drawn up
for the beam scheduling. Making such complex, weighted decisions in a
computer is quite difficult and is generally held to be within the area of
artificial intelligence (AI). Fortunately, modern software languages, such as
ADA, are structured such that antenna array operations can be written as
software modules, which can be called and run by the scheduler with a
predetermined priority list.
15.6 MULTI HEAD RADAR
When several monostatic radars can see the same target, as shown in
Fig. 15.9, their outputs can be combined to improve the tracking. This method
Flight
........... path
Figure lS.9 When the coverage of surveillance radars overlaps, their outputs may be combined
to improve the overall tracking performance.
of netting independent radars together is sometimes known as multihead

radar and has become popular since the development of digital com-
munications enabled reliable low-cost data links to be formed.
Multihead radar is especially applicable for air defence and ATe radars
with overlapping cover, but other situations where it may be applied include
naval systems where several radars, and other sensors, may be combined
within the area of a fleet.
The outputs from several sensors must be sent to a data processing
centre for combining. There are then three ways in which the data may be
combined:
1. Individual plots from each radar can be transmitted to the data

processing centre and the tracker can make use of all the data simul-
taneously. This gives the best results, but requires high-bandwidth data
links and a lot of computer processing power.
2. Individual plots can be sent to the centre and combined together before
trackforming. This requires a wideband data link but it reduces the
computational load and still gives quite good results.
3. Tracks can be formed locally at each radar and sent to the centre for
combining; this requires only a low-capacity data link but the improve-
ment in tracking accuracy is modest.
There are other advantages to multihead radar in addition to improved

tracking accuracy. Low-flying aircraft in mountainous country will only be
detected by radars in short glimpses, but netting several radars together, to
view the target across different terrain, increases the probability of detection
and of forming some sort of assessment of the threat. A multihead radar
system is also more jammer-resistant because each radar views the target
from a different angle.
In some ways, multihead radar networks are more attractive than
multistatic arrangements. Each radar operates independently and causes only
a relatively small degradation to the system if it should fail or be jammed.
Multihead is also much simpler than multistatic because no synchronization
is required and different radars, from different manufacturers, can readily be
netted together.
Merging data in multihead systems is not without difficulties. For
example, data are usually combined by giving each measurement a weight
that is inversely proportional to its variance. In this way the tracker takes
more account of good data than bad. A problem occurs when one radar
correctly detects a target manoeuvre and uses a filter with a larger variance
in order to keep track. The data from this radar will receive less weight than
those radars that have not yet spotted the manoeuvre. This is typical of the
type of quandary that can be so interesting for a radar engineer to work on.
15.7 HIGH RESOLUTION RADAR TECHNIQUES
Modern radar has gone beyond 'radio detection and ranging' to incorporate
target classification and recognition algorithms. These facilities require
improvements in resolution to identify distinguishing features on targets. We
have seen throughout this book how the resolution depends on the system
parameters, but we have not yet mentioned the contributions that can be
made by developing the hardware, the signal processing and the data
processing software.
There are many methods of increasing the resolution of radar systems.
Often, techniques were developed in other fields of research and adapted for
use with radar. Some of the current approaches involving changes in hard-
ware, or data collection strategies, are discussed below.
Wideband radar systems High range resolution can sometimes be used to

identify features on a target, if a sufficiently large bandwidth can be obtained.
Wideband techniques include carrier-free radar (Chapter 12), pulse to pulse
frequency modulation, noise modulation, wideband chirp, etc. The main
disadvantage is that the increased noise bandwidth usually restricts this
technique to short range applications.
Multifrequency illumination Rather than having a single radar channel of

wide bandwidth, it is possible to select a number of narrower bandwidth
channels and correlate the outputs to look for certain spatial properties of
the target. This subject has been extensively investigated by Gjessing
(reference 7) and can be quite powerful when a priori information on the
target shape is available, although it is somewhat sensitive to aspect angle.
Multipolarization techniques It has been realized that more information on

many types of target can be improved if multipolarization techniques are
used. This arises because the polarization signature of targets has, in principle,
five degrees of freedom (three amplitudes and two relative phases). In
contrast, frequency diversity yields only a single extra degree of freedom.
Evaluation of the polarization characteristics of targets is an active area of
research, and radars to exploit these polarization signatures are under
development.
Quadpolarized systems Use of several frequencies and full polarization

information is a characteristic of the latest generation of airborne remote
sensing radars. These quad polarized systems are capable of recognizing the
different scattering mechanisms contributing to the backscatter. This can
be utilized in target recognition and classification. Some of the most
significant progress in understanding the scattering properties of extended
targets and in the application of SAR are likely to stem from this advance.
Interferometry A further advance in the use of SAR is in an interferometric

mode. In this mode, the phase difference between SAR images gathered on
closely spaced orbits can be used to construct a map of the topography of
the imaged area. Tests on Seasat and ERS-l data have indicated that it
should be a practicable technique. Very curious results were also obtained
over flat agricultural land, and there is clearly more investigation needed.
These results have led to a flurry of development of multi-antenna interfero-
metric systems for aircraft operation, and SAR processors giving high phase
accuracy.
Below we describe some of the software and mathematical techniques that

can be used to increase radar resolution.
Inverse synthetic aperture radar ISAR is used to increase angular resolution

in the same way as SAR (see Chapter 11) but with the difference that the
radar remains stationary and advantage is taken of the target motion.
Improved doppler spectrum analysis There are many modern spectral esti-
mators, such as the maximum likelihood method, maximum entropy method,
minimum eigenvector method, etc. These can be used to improve the doppler
resolution and to identify targets by resolving such features as turbine blade
rotation, or the pitch, roll and yaw of an aircraft or ship. Improved spectral
estimators are also applied to phased arrays to improve the angular
information.
Radar cross-section fluctuation analysis When a target is composed of a

limited number of individual scatterers that characterize it, the target may
be identified by statistical or spectral analysis of the RCS amplitude
fluctuations.
Super-resolution Even when the shape of a target is smaller than the range
resolution, some information may be recovered from the echo by carrying
out a deconvolution process known as super-resolution, provided that the
'point spread function' (= impulse response) of the system is known.
Essentially, the radar bandwidth is artificially increased by restricting the
possible choice of target shapes.
Tomography The structure of a given target maps into a particular shape

in the ambiguity diagram (the convolution of the target structure with the
pulse shape in range and doppler). Different chirp rates correspond to
different angles on the ambiguity plane and so may be used to gain
information on the target shape, in a similar way to tomographic imaging
used in medicine.
Many of these techniques provide good results in computer simulations, but

in practice the effects of noise and phase errors due to unresolved motions
may lead to degradation of performance. The best strategies often involve
the combination of several of the techniques above. If the objective is to
identify a small target, such as an aircraft or ship, it can be useful to include
a 'library' or 'catalogue' of possible target shapes. Some means of relating
high resolution observations to the library is then required and these days
there is a lot of interest in neural nets which can be trained to recognize
features in the data that relate to certain types of target.
15.8 SUMMARY
This brief review of recent developments in radar is intended to help you

identify the key areas to study if you wish to pursue a career in modern
radar. Phased array theory and high-resolution techniques are particularly
important because of their use in other fields such as sonar. Some of the
useful phased array formulae to remember are given below.
Key equations
• Phase change between elements in a phased array:

'" = 2n (d / A. ) sin 0 [radians]
• Array factor:
AF = JN ( sin [N n ) (d / A. ) sin () ) [J
N sin [n(d/ A.) sin 0]
• Half-power beam width (for large N):
i10 = 0.886,1. [radians]
Nd cos 0
• Approximate half-power beamwidth for broadside pattern:
i10,..., 5U/D [degrees]
• Positions of nulls in pattern:
sin 0null = ± kA./ Nd [J
where k = 1, 2, ....
• Positions of grating lobes:
sin Ogl = ±kA./d [J
• Positions of sidelobes:
(2k + 1»),
sin 8s1 =± 2Nd []
• Spacing limit to avoid grating lobes for steering angle 8s:
d<). (I + 1sin 8s 1.5)--

N
Em]
• Array factor for beam steered through angle 8s:

AF = j N(sin [Nn(d/)' )(sin 8 - sin 8s)]) []
N sin [n(d/).)(sin 8 - sin 8s)]
• Array output for digital beamforming:

N-l
Array output = L v"w"e-j21tn(dl).)sin9Cn [V]
n=O
15.9 REFERENCES
I. The Fourier Transform and its Applications, R. N. Bracewell, McGraw-Hili, New York, 1986.
2. Antennas, F. R. Connor, Edward Arnold, London, 1972 [Short and very readable.]
3. The Handbook of Antenna Design, vol. 2, Eds A. W. Rudge, K. Milne, A. D. Olver and
P. Knight, lEE Electromagnetic Wave series, Peter Peregrinus, Hitchin, Herts, 1983. [Long
but still readable.]
4. Radio Wave Propagation and Antennas, J. Griffiths, Prentice-Hall, Englewood Cliffs, NJ, 1987.
1989.
6. Electronically scanned antenna systems, M. F. Radford, Pro£". lEE, 12S( II R), 1100-1112,
1978. [A particularly lucid review.]
7. Target Adaptive Matched Illumination Radar, D. T. Gjessing, lEE Electromagnetic Waves
series 22, Peter Peregrinus, Hitchin, Herts, 1986.
8. The MU radar with an active phased array system: I. Antenna and power amplifiers,
S. Fukao, T. Sato, T. Tsuda, S. Kato, K .. Wakasugi and T. Makihira, Radio Sci., 20(6),
1155-1168, 1985.

Practical Phased-Array Antenna Systems, Ed. E. Brookner, Artech House, 1991. [A promising
new book, not yet released at the time of writing.]
15.11 PROBLEMS
15.1 The Japanese MU (middle and upper atmospheres) MST radar uses an active antenna
array composed of 475 crossed Yagi antennas (each having a gain of 7.24 dB) arranged on a
circular plane of diameter 103 m. Each antenna has its own solid-state transmit-receive module,
with a peak output power of 2.4 kW (see Fukao et al. S ). The operating frequency is 46.5 MHz.
What is the antenna gain and beam width ?
15.2 The beam in thl? above can be steered electronically up to 30° off-vertical. If the elements
were laid out on a rectangular grid, what would be a suitable array spacing? (The authors 9 tell
us that 'neither spatial nor electrical tapering is incorporated '.)
15.3 The triangular lattice arrangement of antenna elements (see solution to 15.2) has a
maximum separation of 1.2), in some directions. Where might the first grating lobes be expected
to appear?
15.4 We are told that it is possible to steer the beam in each inter-pulse period, which may be
as short as 400 itS. Would you expect the necessary phaseshifts to be calculated each time?
CHAPTER
SIXTEEN
THE FUTURE OF RADAR
• Where is radar going?

• Increasing the bandwidth
• More complex systems
• More complex processing
The increasing sophistication of computers and software will give rise to a

new generation of intelligent radar systems.
16.1 INTRODUCTION
Has radar been developed as far as it is going to go, or does it have a bright
future with many further applications and capabilities to be discovered? As
with any branch of science or engineering, the way to answer this type of
question is to consider whether the system is limited by physical laws, by
cost or by technology.
The laws of physics applying to radar are well understood, but do not
yet limit our systems. We cannot escape the range resolution being limited
by the effective bandwidth of the system, for example, but we can apply
processing power to such techmques as super-resolution to extend the
effective bandwidth.
Cost is a serious limitation to what can be achieved; there are many
cases where phased array radars are desirable but mechanical turning gear
is used instead because it is less expensive. However, the influence of large
military research programmes has continued radar development, and the
319
falling cost of components, especially computing power, continues to help

progress.
Technology is therefore the limiting factor. Faster AID converters,
gallium arsenide transmit and receive modules, better simulation software,
more advanced mathematics for data processing algorithms-these are the
tools needed to build better radars in the future. As we show below, the
overriding need is to increase the bandwidth of radar systems. The other
benefits will then follow as a matter of course.
16.2 DEVELOPING THE CONCEPT OF BANDWIDTH
There are three principal performance requirements for a radar sensor:

sensitivity, resolution and data rate. A military radar has a fourth requirement
of survivability (see Radford 1). The fate of radar will be determined by how
well these requirements can be fulfilled.
In future, radars will be required to be more sensitive so that smaller
targets can be detected in more hostile environments. Microlight aircraft
may need tracking near civil airlanes and military radars will be required to
detect increasingly Stealthy targets. Meanwhile, the radio environment
deteriorates as radio interference and the sophistication of jammers continue
to increase.
The sensitivity of a radar can be thought of in the following way. If a
solid angle of sky 0 is required to be searched in time T, then
T time to search the whole sector
dwell time in each beam position
= whole sector)
( [ ] (16.1 )
beamwidth
where t = the dwell time in each position [s]. Rearranging Eq. ( 16.1 ) gives
t = (T 6.0 6.</> )/0 [s] (16.2 )
The final radar bandwidth after processing is 1It and the noise power in the
receiver is therefore
N=kToF -
t (1) FkToO
=---
T 6.0 6.</>
[W] ( 16.3 )
Putting Eq. (16.3) into the radar equation, and replacing Gt by 4n I (6.0 6.</> )
and Gr by 4nAe/..1. 2 gives
[ ] (16.4)
THE FUTURE OF RADAR 321
which is the formula developed by Radford 1 • Note that the expression is

frequency-independent, although some terms, such as the RCS, will vary
with frequency. The importance of expressing the radar equation as in
Eq. ( 16.4 ) is that only the power x aperture product P,Ae is under the control
of the radar designer, the others being set by the requirement or the
environment.
Survivability and cost mitigate against larger power x aperture products,
so where is the improved performance to be found? The answer lies in
improved resolution and data rate, and both of these require an increase in
the fundamental limiting factor of radar performance, the bandwidth.
Bandwidth can be thought of in terms of the rate of searching radar
resolution cells. The total number of beam positions n 1 to be searched is
Q
1 [ ] (16.5)
n = dO dl/>
The number of range bins n 2 is given by

interval between pulses I/PRF
n2 = ----
pulse duration
[ ] (16.6 )
(PRF) x r
and the maximum number of doppler filter channels n3 is set by the number
of pulses in the dwell time as
n3 = (PRF) x t [] (16.7)
Combining Eqs (16.5) and (16.7) with Eq. (16.2) gives the total number of
resolution cells n to be searched as
Q 1
= -- x x (PRF) x t
dO dl/> (PRF) x r
= Tit [ ] (16.8)
Remembering that 1Iris the bandwidth B of the receiver, we can rearrange
Eq. (16.8) as
nlT= B [Hz] (16.9 )
Thus the maximum number of resolution cells that can be searched every
second is set by the bandwidth of the system.
The simplest way to increase the bandwidth of a radar system is to use
multiple antenna beams and receivers; the effective bandwidth Beee is then
Beee = mB [Hz] (16.10)
where m = number of simultaneous beams and receivers.
Worked example The Marconi Radar System Martello S723 is a

long-range L-band surveillance radar using a single transmit beam, but
eight receive beams stacked in elevation. If the pulse length (after
compression) is 0.25 JlS, and the antenna rotation rate is 10 RPM, how
many resolution cells could be searched?
SOLUTION If the bandwidth of a single channel is 4 MHz, the effective

bandwidth of the system is 32 MHz; if this represents the number of
cells inspected every second, and one antenna revolution takes 6 s, the
total number of cells to be searched is about 2 x 108 .
COMMENT Although 2 x 108 seems a very large number of cells to be

inspected, it looks more realistic if you work out the values of n 1, n2 and
n3 separately. This large number of resolution cells reinforces the point
made in Chapter 2 that front-end radar data rates are high.
Increasing the bandwidth of a radar brings several benefits:
• Increasing the receiver bandwidth decreases the range resolution cell size
and cuts down the clutter.
• The use of more simultaneous beams implies that longer integration
times can be used in each beam position. This increases the radar
sensitivity.
• Longer integration times mean that the doppler resolution and clutter
rejection can be improved.
• The data processing rate for each target can be increased.
In this way, all our requirements can be met without increasing the
power x aperture product and exposing the radar to an increased risk of
ECM attack. Generally speaking, the analogue parts of a radar system are
not the factor limiting the bandwidth, but the computational power available.
This situation is changing with the arrival of parallel processing, very
large-scale integration (VLSI) signal processing integrated circuits and
optical methods of solving integrated circuit interconnection problems. Soon,
large amounts of low-cost computing power will be available, and we must
decide how to make best use of it in the design of the next generation of
radars. One of our main aims must be to make radars more adaptable to
the tasks they have to carry out.
16.3 ADAPTIVITY
In the past, radars have been rather unintelligent devices, perhaps more so
than was necessary. Early radars often did not know if they were being
jammed and yet the noise level in the receiver, measured continuously as
the antenna rotated, could have been used to draw a shape on a PPI display
representing the maximum detection range for targets of various Res. For a
radar that is internally noise-limited, these contours would be perfect circles
on a PPI display, but if external noise was received from any particular
direction, this would cause a dint in the circle and warn the operator that
the radar was not working well in that direction.
Imagine, now, a phased array radar in which all aspects of the system
are controlled by the data processing computer, as shown in Fig. 16.1. The
beam scheduling is controlled adaptively and well behaved targets are
inspected less frequently than those which have been found to be manoeuvring.
If the transmitted power is variable, this may also be controlled adaptively
to illuminate small long-range targets more strongly. It may be more
important to control the transmitter waveform adaptively and reserve long
pulse and wide swept bandwidths for difficult situations. When allocating
air defence resources, for example, it may be important to find out whether
a single echo represents one attacking aircraft or a very tight formation of
several planes; this may be possible by allocating extra integration time,
bandwidth and computing resources to the problem until it is resolved, while
other continuing searches are maintained using shorter pulses.
Adaptive doppler filtering and tracking algorithms may be one method
by which other information can be integrated into the system, in a process
known as data fusion. A modern warship, for example, has many sensors
physically close to each other and it is relatively easy (electrically) to
connect them together. If a target is identified by other means (e.g. visual
identification), the radar data processing can be allocated on the basis of
the known characteristics of the target. A modern fly-by-wire aircraft would
Figure 16.1 Adaptive phased array radar, in which all aspects of the system are controlled by
the data processing.
ll~ Time/range
Figure 16.2 Searching for targets as a pattern among a series of scans can be more productive
than scanning each threshold independently. These scans were generated artificially using white
noise and a double pulse generator, set to give a low SNR. To find the targets, try looking at
the diagram with your eyes nearly level with the bottom of the page.
need more frequent inspection than a 852 bomber, because of the fighter's
greater manoeuvrability.
Data fusion is one of the areas where artificial intelligence (AI) techniques
can be used to make radars more intelligent. There is no magic in AI, but
there are some very useful programming ideas. The process of track initiation
could be improved by moving away from simple 'hit or miss' thresholding
to the utilization of' soft thresholding' and fuzzy logic processing. In addition,
pattern recognition algorithms could be used to determine whether a series
of poor-quality hits represents the presence of a real track or not. This process
is illustrated in Fig. 16.2; try looking at these range sweeps with your eye
almost level with the page and the two targets (both at very low SNR) should
become obvious. In fact, historical displays, which progressively reveal the
last few hits in a sector, have been used in the past, but it was the human
operator who made the decisions; AI will allow us either to give extra aid
to radar operators or to replace them by software, which is more reliable
and can detect targets at lower SNR for the same probability of detection
and false alarm as at present.
16.4 SUMMARY
The future of radar does not lie in larger and more powerful systems, but
rather in slightly smaller systems that are more agile, intelligent and difficult
to detect because of the larger bandwidths that will be used. The resolution
of radars, and the number of targets that can be tracked, can be expected
to increase as large amounts oflow-cost computer power become available.
The factor limiting radar performance is likely to remain technology
(rather than any law of physics) for some time to come. Until recently, it
was the ability to process the large amounts of data that created an upper
limit to performance, but with the advent of parallel processing the weakest
link in the chain is almost certain to become the A/D converter because of
the problem of increasing the sampling rate while maintaining a high dynamic
range. The recent trend of moving the point at which the A/D conversion
takes place further and further up the receiving chain towards the antenna
only exacerbates this problem.
Radar seems certain to provide challenging engineering, mathematical
and computational problems to be solved for years to come. We hope that
this book has conveyed the main ideas and helped you to understand the
underlying principles of radar. We hope also that you have gained an
appreciation of the importance of radar in many diverse areas, and sensed
some of the excitement of working in this field.
16.5 REFERENCE
I. Whither radar. M. F. Radford. GEe J. Res .. 3(2),137-143.1985.
APPENDIX
I
SYMBOLS, THEIR MEANING AND SI UNITS
These are the basic symbols used. Some have several subscripts, e.g. L, L.,
L.y " etc.
A signal scaling factor [ ]

A antenna area [m 2 ]
Ac clutter area [m 2 ]
2a bistatic baseline Em]
B bandwidth [Hz or rad s - I ]
C array element correction coefficient [ ]
Cn2 structure constant [m 2/3 ]
Cr2 layer scattering factor [ ]
Cs2 volume scattering factor [ ]
c speed of light [m S-I]
c(t) clutter voltage [V]
D antenna directivity [ ]
D diameter of a water droplet Em]
D ambipolar diffusion coefficient [m 2 s-l]
D length of an antenna array Em]
d linear antenna dimension Em]
d array element spacing Em]
d distance between synthetic aperture
radar and scatterer Em]
E energy [J]
E[ ] expectation value [units]
E field from N elements of an array [V m- I ]
326
SYMBOLS, THEIR MEANING AND SI UNITS 327
e partial pressure of water vapour [millibars J

e bistatic ellipticity factor [ J
F receiver noise figure [ J
Fa effective antenna noise factor [ J
I frequency [HzJ
I over-the-horizon propagation factor [aJ
Id doppler frequency shift [Hz or rad s - 1 J
AId doppler resolution [Hz or rad s - 1 J
bid doppler uncertainty [Hz or rad s - 1 J
AI frequency separations or swept
frequency [Hz or rad s - 1 J
G antenna gain [ J
9 acceleration due to gravity [m s-IJ
H atmospheric scale height [mJ
H(w) system transfer function [ J
h height [mJ
h over-the-horizon height factor [ J
h(t) system impulse response function [ J
k Boltzmann's constant [J K -IJ
k threshold multiplicative factor [ J
k effective Earth radius factor [ J
L loss factor [ J
L length of a sea wave [mJ
Lo turbulence scale size [mJ
1 characteristic dimension of a target [mJ
M number of observations to be
averaged { J
M fast Fourier transform length [ J
m over-the-horizon radar cross-
section modification factor [ J
m(t) system response to noise input [VJ
N noise power [W per unit bandwidthJ
N radio refractivity [ J
N. electron density [electrons m - 3 J
n refractive index [ J
n number of observations [ J
n(t) noise voltage [VJ
P power [WJ
Pd probability of detection [ J
Pfa probability of false alarm [ J
Pr transmitted power [wJ
PI received power [wJ
p(u) radar cross-section probability
density function [ J
P air pressure [millibars]

q electron line density [electrons m - 1 ]
R range [m]
~R range resolution [m]
bR range uncertainty [m]
Re radius of the Earth [m]
Rm (r) autocorrelation function [ ]
r precipitation rate [mm h- i ]
2r bista tic transmitter-target-recei ver
distance [m]
ra synthetic aperture radar along-
track resolution [m]
re effective radius of the electron [m]
rj meteor trail initial radius [m]
S(f or w) power spectral density [W per unit bandwidth]
Sm(w) power spectrum of m (t ) [W rad- i ]
s convolution integration variable [s]
T inter-pulse period (reciprocal of
pulse repetition frequency) [s]
T interval between observations [s]
T temperature [K]
To system noise temperature [K]
I;. underdense echo delay time [s]
t time [s]
t integration time [s]
U(w) Fourier transform of u (t ) [Vsrad- i ]
u(t) radar signal voltage [V]
V voltage [V]
~ clutter volume [m 2 ]
v'es volume of a resolution cell [m 3 ]
v velocity [ms-i]
ov velocity uncertainty [ms- i ]
v(t) system response to signal input [V]
W array element weighting coefficient [ ]
x target position Em]
y(t) receiver output voltage [V]
Z radar reflectivity factor [mm 6 m- 3 ]
IX track position smoothing factor [ ]
IX angle of refraction [radians or degrees]
p track speed smoothing factor [ ]
p synthetic aperture radar antenna
real beamwidth [radians]
Ps synthetic aperture radar synthetic
beamwidth [radians]
SYMBOLS, THEIR MEANING AND SI UNITS 329
<5 (t) delta function [ ]

e antenna efficiency factor [ ]
er relative permittivity [ ]
'1 average radar cross-section per unit
volume [m- 1 ]
'1 relative bandwidth [ ]
e azimuth angle [radians or degrees]
Ae azimuth beamwidth [radians or degrees]
<5e azimuth uncertainty [radians or degrees]
A. wavelength Em]
p radius of curvature Em]
Pu(t) autocorrelation function of u (t ) [V]
(J radar cross section [m 2 ]
(J standard deviation [units]
(J2 variance [units 2 ]
T pulse duration [s]
Td two-way propagation delay to target [s]
</J elevation angle [radians or degrees]
A</J elevation beamwidth [radians or degrees]
<5</J elevation uncertainty [radians or degrees]
X(t,Wd) ambiguity function [ ]
tjJ signal phase [radians]
<5tjJ phase difference between signals [radians]
n solid angle [ steradians ]
Wd doppler frequency [rad S-l]
APPENDIX
II
ACRONYMS AND ABBREVIATIONS
AC alternating current
ACF autocorrelation function
A/D analogue-to-digital (converter)
AI artificial intelligence
ARM anti-radiation missile
ATC air traffic control
CAT clear-air turbulence
CCIR International Radio Consultative Committee
CFAR constant false-alarm rate
chirp linear frequency sweep during a pulse
COMINT communications intelligence
CW continuous wave
DC direct current
ECCM electronic counter-countermeasures
ECM electronic countermeasures
EJ escort jammer
ELINT electronic intelligence
EM electromagnetic
ERP effective radiated power
ERS-J European remote sensing satellite
ESM electronic support measures
EW electronic warfare
FFT fast Fourier transform
330
ACRo.NYMS AND ABBREVIATIo.NS 331
FIR finite impulse respo.nse

FM frequency mo.dulatio.n
FMCW frequency-mo.dulated co.ntinuo.us wave
FMICW frequency-mo.dulated interrupted co.ntinuo.us wave
FOM figure o.f merit
fruit false replies unsynchronized in time
FT Fo.urier transform
GHz gigahertz
gw groundwave
HF high frequency
HOJ ho.me-o.n jammer
IC integrated circuit
ICAO International Civil Aviation Organizatio.n
IF intermediate frequency
IFF identify, friend or foe
ISAR inverse synthetic aperture radar
kHz kilo.hertz
lidar light detectio.n and ranging
LPI lo.w probability of intercept
LVA large vertical aperture
MHz megahertz
MST meso.sphere-stra tosphere - troposphere
MTI moving-target indicator
OTH o.ver-the-horizon
PDA probabilistic data asso.ciation
PDF probability density functio.n
PPI plan position indicator
PRF pulse repetition frequency
radar radio. detection and ranging
RAM radar-absorbing material
RCS radar cro.ss-section
RF radio frequency
RMS root mean square
RPV remo.tely piloted vehicle
RWR radar warning receiver
Rx receiver
SAR synthetic aperture radar
SCV sub-clutter visibility
SNR signal-to.-no.ise ratio
SOl stand-off jammer

SSR secondary surveillance radar
ST stratosphere-troposphere
sw skywave
T-R transmit - receive
Tx transmitter
UHF ultra high frequency
VHF very high frequency
VLSI very large-scale integration
APPENDIX
III
USEFUL CONVERSION FACTORS
Nautical mile (n. mile) = 1/60 of a degree of latitude

= 6080 ft
= 1852 m
= 1.852 km
This is the International nautical mile; the UK nautical mile is 1.85318 km.
We have adopted the symbol n. mile to avoid confusion with nanometres
(nm).
Knot (kt) = 1 n. mile per hour
~ 0.51444 m S-l
= 1.852 km/h - 1
Foot (ft) = 0.3048 m exactly
A foot is the unit of length in the old British system of units.
1000 ft = 0.3048 km
Speed of light (c) = 299792458 m s - 1 exactly
Boltzmann's constant (k) = 1.380658 x 10 - 23 J K - 1
This has 8.5 ppm uncertainty. When a standard temperatureof290 K is used:

kTo = 4.003908 x 10- 21 dB J
= -204dBJ
= -204dB W Hz- 1
Radius of the Earth Re = 6378.160 km
333
The Earth's shape is an oblate spheroid with an equatorial bulge caused by

its rotation. The figure above is the equatorial radius or semi-major axis.
The polar radius, or semi-minor axis is 6356.775 km.
Some SI prefixes commonly used in radar are:
Prefix Symbol Power Example
pico p 10- 12 picosecond

micro Jl 10- 6 microsecond
milli m 10- 3 millimetre
centi c 10- 2 centimetre
kilo k 103 kilowatt
mega M 106 megahertz
giga G 109 gigahertz
Reference source: The McGraw-Hill Dictionary of Science and Technical

Terms, 4th edn, Ed. S. P. Parker, McGraw-Hill, New York, 1989.
APPENDIX
IV
USING DECIBELS
Decibels can be one of those little things that you can get a mental block
about. It is not that there is anything difficult about the maths, but rather
the concept of handling voltage and power ratios in this way. However, it
is worth learning to use decibels, because they make life so much easier for
the radar engineer.
The idea behind decibels is that the numbers used in radar calculations
are often very large (e.g. R4) or very small (e.g. Pr ), and they must usually
be multiplied and divided. This makes life difficult and can easily lead to
mistakes. Converting these numbers to logarithms means that, first, they
become a sensible size and, secondly, they only have to be added and
subtracted. This greatly facilitates repeating calculations, when one of the
parameters is varied.
The definition of decibels is that, if PI and P2 are two amounts of power
then the first is said to be n decibels greater than the second, where
n = IOlog lo (PdP2 )
The ratio of two voltages VI and V2 can be converted to decibels by
n = 2010glO(VI/V2)
In radar, it is more common to find the 10 loglo power version used in
calculations, rather than voltage ratios, but this is not necessarily so in other
subjects. To escape from decibels and return to linear numbers, divide n by
10 (or 20 for voltage ratios) and then use the IOXfunction on your calculator.
It can sometimes be difficult to understand negative decibels. These
represent ratios less than unity. A figure of - 10 dB means a ratio of 1/10,
335
- 20 dB = 1/100, - 30 dB = 10 - 3, etc. Inverting a ratio just changes the sign

in decibels. For example, (4n)3 = 33 dB and 1/(4n)3 = -33 dB. Note that
if you have a noise power of -144 dB Wand you increase the noise level
by 10 dB, the new noise power is -134 dB W.
Although decibels are often used to denote straightforward ratios, they
can also be used to express how large a number is compared to some reference
level. Thus if we say a transmitter power is 30 dB W, we mean that it is
30 dB above a watt, i.e. it is a kilowatt. One reference level commonly used
is the milliwatt, and powers are referred to this level using the dB mW. Thus
a received power of -100 dB mW is the same as -130 dB W,or 10- 13 W.
Some commonly used values are listed below.
Ratio dB Ratio dB
(4rrj-3 -33.0 1 0
10- 3 -30.0 2 3.0
(4rr j-2 -22.0 4 6.0
10- 2 -20.0 5 7.0
10- 1 -10.0 10 10.0
1/4 -6.0 102 20.0
1/2 -3.0 103 30.0
1 MHz 60 dB Hz
1 GHz 90dB Hz
Boltzmann's constant k = 1.38 x 10 - 23 W Hz - 1 K - 1

= 228.6 dB W Hz- 1 K- 1
Speed of light c = 3 X 108 m s - 1 = 84.8 dB m s - 1
Historical note: The decibel was invented by American telephone
engineers in the 1920s. The unit was named after Sir Alexander Graham
Bell, who is credited with inventing the telephone on 2 June 1875 (although
this was after a long and competitive research programme). The original
unit, the bel, was too large to be useful.
Reference source: The McGraw-Hill Dictionary of Science and Technical
Terms, 4th edn, Ed. S. P. Parker, McGraw-Hill, New York, 1989.
APPENDIX
v
SOLUTIONS TO PROBLEMS
Chapter 1
1.1 (a) J./d = 0.03/(2.7 x 0.8) = 0.014 [rad] = 0.8°
(b) J.jd = 0.10/(3.6 x 0.8) = 0.035 [rad] = 2°
(Racal- Decca quote O.~o and 2° respectively in their specifications.)
1.2 G = 32000/(.Mo A</J°) = 32000/(0.8° x 23°) = 32.4 dB

(Racal-Decca quote 32 dB.)
1.3 AR '" c/2B '" 37.5 m

(Racal-Decca quote 35 m.)
1.4 AO = 0.032/3.4 = 0.54°

A</J = 0.032/0.75 = 2.4°
(Siemens- Plessey quote 0.55° and 2S respectively in their data summary.)
1.5 G = 4nA/J. 2 = 4n(3.4

x 0.75)/(0.032)2 = 45 dB
(Siemens- Plessey quote a figure of 'not less than 41.5 dB' for this radar.)
1.6 The transverse resolution of the system R AO = 7400 x 0.032/3.4 = 70 m.

This could be improved on considerably because of good SNR at such short
ranges. For an airfield controller to use these measurements to position an
aircraft with a mean error of 25 m appears to be very good. An error of
25 m at 4 n. mile (= 7.4 km) reduces to 6.3 m at 1 n. mile and in the
337
evaluation the error did not in fact exceed 6 m. Assuming a runway length
of 1 n. mile, this should be sufficiently accurate for a bad-weather landing.
1.7
2690 BT ACR 430

Parameter marine radar airfield control radar
P, 44 dBW 47.4 dB W
G,G, 64 dB 83.0dB
(J 10 dBm 2 10 dBm 2
•2
I. -30.5 dB m 2 -30.5 dB m 2
L, -5 dB -5 dB
Ij(4n)3 -33 dB -33 dB
IjN +131 dBW +131 dBW
Ij(SNR) -13 dB -13 dB
R~ax 167.5 dB m4 189.9 dB m 4

15.4 km 55.9 km
Rmax.
8.3 n. mile 30.2 n. mile
A calculation worksheet included in the Siemens- Plessey reportS gives

the prediction of radar range as 30 n. mile. In practice, these detection ranges
would be increased by the integration of many pulses; such integration would
be necessary for the doppler processing, especially in the case of the marine
radar.
Chapter 2
2.1 Duty cycle = T/T = T x (PRF)
Mode 1: Duty cycle = 1.5 x 10- 6 x 480 = 1/1389
Mode 2: Duty cycle = 1.0 x 10- 6 x 691 = 1/1447
The mean power = peak power x duty cycle.
Mode 1: Mean power = 864 W
Mode 2: Mean power = 829 W
(Alenia quote 0.87 and 0.83 k W, respectively.)
2.2 We have
n = AO x (no. pulses/s) x (time [s] of one revolution)

3600
AO (PRF) AO x (PRF)
=-x----
360 (RPM )/60 6 x (RPM)
SOLUTIONS TO PROBLEMS 339
1.2 x 480 .
n = = 16 hIts/scan
6x6
2.3 Equating
R = C with
max 2 X (PRF) 12.4 X (PRF)
and remembering that 1 nautical mile = 1853 m, tells us that the number 12.4
is a nautical mile expressed in microseconds. This is another way of expressing
the '150 m per microsecond' rule as '12.4 J.lS for every nautical mile'.
2.4
PI = 60.8 dB W 1.2MW
GIGr = 73.0 dB G = 25000/1.2 X 4.7
(J = 3.0dB m 2 2 m 2 target
).2 = -12.7 dB m 2
L= -7.0dB
1/(4n)3 = -33.0dB
SNR = -10.0 dB
(Noise) -1 = + 145.8 dB W Assuming bandwidth = 1:- 1
= (1.5 X to- 6 )-1
R~ax = 219.9 dB m 4
Rmax = 55.0 dB m ==314km-170n.mile
The Alenia engineers' radar calculation worksheets use a more detailed
procedure, but arrive at a similar conclusion; 164 n. mile for this case.
8
2.5 c = 1.5 X to = 312.5 km
Runambig = 2 x (PRF) 480
This is similar to the maximum detection range and so the system is nominally
unambiguous in range. However, Alenia provides a 5 per cent PRF stagger
to remove long-range ground clutter effects caused by tropospheric ducting.
2.6 Clutter cell size = R 11() I1R

= 300000 Em] x 2.1 x to- 2 [rad] x 225 m (== 1.5 J.ls)
Using a land reflectivity of 1/100, this implies an RCS for the clutter in each
resolution cell of 14000 m 2 == 41.5 dB m 2 . If the target RCS is 3 dB m 2 and
the system noise is 13 dB lower, the minimium dynamic range requirement
is 51.5 dB m 2 • An A/D converter nominally gives 6 dB of dynamic range
per bit, suggesting that a minimum of 9 bits is required. In practice 6 dB/bit
is not achieved because the sensitivity must be set such that the quantization
noise is small compared to the system noise.
(The ATCR-44K radar uses a 12 bit A/D converter sampling every
1.5 liS.)
2.7
Pt = 43.0dB W 20kW
GtGr = 32.0 dB
(1 = 3.0dB m 2
A. 2 = 7.6 dB m 2
L = -5.0dB
1/(4n)3 = -33.0dB
(SNR)-1 = -13.0dB
(Noise) -1 = + 147.0 dB W
181.6 dB m4
45.4dBm == 34.5 km == 18.6 n. mile
This is much lower than the modern Alenia radar.
To be fair to the Freya system, there would probably have been
considerable incoherent gain involved in the display process, which would
have improved the detection range over this single-pulse calculation. None-
theless, the calculation serves to illustrate that the improvement in the A. 2
factor over a modern L-band radar is not sufficient to compensate for the
much smaller power x aperture product.
Chapter 3
3.1 (PRF) x 60/(RPM) = 25 pulses/scan
3.2 The radar equation indicates a SNR of 0.3 dB for a 120 m 2 target at
100 km. In one minute, 1500 angular measurements are made, which would
increase the SNR of the measurement by up to 31.8 dB. The best that might
be achieved is 2° / j (2 x SNR) = 0.04°, although the spinning ofthe reflector
as it dangled below the balloon would cause RCS variations that would
degrade this figure. (Siemens- Plessey quote better than 0.15° for this system.)
3.3 The beamwidth is roughly 2S x 4°, making 40 positions to be searched.

Each requires 9.6 ms and the search could nominally be completed within
0.4 s. In practice, additional time would be required to change beam position,
but it would still be sufficient to maintain target contact during a dog-fight.
Some of the parameters used in this problem are similar to those used by
the Westinghouse APG-66 radar fitted to the F-16 aircraft-the 'fighting
falcon'.
3.4 We would expect an accuracy - 3 minutes of arc. At 1 km range this

corresponds to a tangential error -1 m (accuracies of 3 minutes RMS in
angle and 5 m in range are claimed for this radar). With these accuracies it
would be dangerous for an air target to approach nearer than 1 km, and the
ST802 can in fact be used for controlling guns of calibres from 20 to 76 mm.
3.5 (a) v,. (beam centre) = 300 cos 10° cos 45° = 208.9 m S-1
v,. (beam edge) = 300 cos 10° cos 47° = 201.5 m S-1
The difference of 7.4 m s - 1 corresponds to a doppler shift of ± 493 Hz across
the beam.
(b) 10 Hz = 0.15 m S-1 - 0.04°. These parameters are similar to those
used on the GEC-Ferranti Blue Vixen radar recently fitted to the UK Sea
Harrier aircraft. The Blue Vixen is a multimode fire control radar, but with
the capability of doppler beam sharpening. (GEC-Ferranti suggest that a
resolution of 1/20° or better is achievable with this technique.)
(c) v,. (ahead) = 300 cos 10° cos 0° = 295.4 m S-1
v,. (beam edge) = 300 cos 10° cos 2° = 295.3 m S-1
The difference in velocity is insufficient to give good resolution and doppler
beam sharpening is not normally used within ± 15° of the aircraft track.
Chapter 4
4.1 If mean RCS = s, PDF is
p(x) = (lIs) e-x!s x~0
Thresholding 10 dB above the mean gives
Pfa = foo (l/s)e- x!S ds = e- 10 '" 4.54 x 10- 5

10 s
So expected number of false alarms is 512 2 X Pfa - 12. For gaussian PDF,
we must solve
t- cI>(k) = 4.54 x 10- 5
which gives k = 3.91.
In the exponential case we are nine (additive) standard deviations above
the mean; in the gaussian case only 3.91.
4.2 (a) Output = p(t) = f:oo u(s)u(s - t)ds
This is the product integrated in the overlap (see diagram):
It I > 2T; no overlap p(t) =0

o ~ t ~ 2T(asshown) p(t) = V2(2T - t)
1/(5) V u(s - t)
O - : - - - - - t - - - r - - r - - - - - - - - - ----,
I
I
I
-T t- T T t+ T 5
By symmetry
It I > 2T
It I ~ 2T
(b) Energies are the same, therefore there is the same detection per-
formance after matched filtering. Output in case (i) is as above with T = 10- 3.
Output in case (ii), using diagrams as above, is:
Both outputs have the same maximum value at 0; this is the energy 2V2T.
The second waveform gives a sharper main lobe; this has implications for
system resolution, as discussed in Chapter 6.
4.3 By Eq. (4.33), noise power = kToEh/2, and
Eh =fT12 cOS 4 (nt)dt=3T

-T12 T 8
(use cos 2 x = HI + cos(2x)] twice). Therefore,
Noise power = 7.5 x 10- 25 W ...., -241 dB W
Energy in signal = E = 10- 16 x 3T/8. Hence
SNR = E/N
= 10- 16 X i X 10- 3 X ----
21
4X 10-
= 9.375
'" 9.7 dB
4.4 1 MHz
14 ~I
A 1 1
1 1 H(w)
1 1
100 MHz
Use
Eh = - I fX [H(wW dw
210 - x,
Since the integral of each triangular function squared will not change if it is
moved to the origin, we can use the diagram below:
A
The function is even, so energy in each triangular function is
2 x -1
2n
Ie
0
A2(l - OJ/C)2 dOJ =A-
C
3n
2
Using both sidebands gives
So
Output noise power = ~ x 106 X A2 N
If we use the second definition of bandwidth, we must use C = n x 106 ,
giving half the output noise power above.
4.5 Energy in returned pulse = r:Pr • So

E = r:PIA;qL s
4nA.2R4
Therefore,
E E 5 X 1021
SNR = - = ---::-:-
N 4 X 10- 21 nR 4
(a) R = 50km:
(b)R=I00km:
4.6 In Chapter 1,
SNR = signal power
noise power
signal power
= x--------------------
B noise power / unit bandwidth
For simple pulses, 1/ B '" r: where r: = pulse duration. Therefore, first term
is signal energy.
4.7 (a) k = 4.75, and so,

0.9 = 0.5 - ~(4.75 - j(2 x SNR»
~(-4.75 + j(2 x SNR» = 0.4
SNR = 18.2 (12.6 dB)
(b) A = In(0.62 x 106 )
B= In9
SNR = 20.6 (13.1 dB)
4.8 SNR for integrated pulses = 15.6 (Sec. 4.6). Therefore,

SNR 1 = 15.6/16 = 0.975
and
Pd = 1- <1>(4.75 - J1.95) = 0.0004
4.9 Matched filter = u* ( - t) = exp (jat 2 ) It I :::; T
Pu(t) = f:T u(s + t)u*(s)ds

T-
f
t
= -T exp[ -ja(s + t)2] expUas 2 ) ds 2T ~ t ~ 0; see diagram
r---------<r----~--~------~
[
1
I I
-t - T -t -T -t +T T
= exp( -jat 2 ) f T-
-T
t exp( -2jast) ds
exp( _jat 2 ) . •
. {exp[ -2Ja(T - t )t] - exp(2JaTt)}
2Jta
= ~ sin[(2T - t)at]
at
When t = 0, this gives the value 2 T.

A similar calculation for - 2T :::; t :::; 0 gives
Pu(t) =
{ ~ sin[(2T -Itl)at]
at
o
It I :::; 2T
It I ~ 2T
This is plotted in the diagram.
Pu (t)
T=1
a=5
-1
-2
4.10 After averaging, an SNR of 15.6 is needed. Therefore,

. 15.6
Smgle-pulse SNR = -
32
Hence PI ~ 31.4 W.
Chapter 5
5.1 FFT gain is 15 dB plus a maximum incoherent averaging gain of 3 dB.
Manoeuvring not only alters the RCS, causing a loss of integration gain,
but also spreads the echo across several doppler cells, thereby making
detection difficult.
5.2 30.1 dB gain; position uncertainty reduced by a factor of 32(=--J1023).

When an aircraft turns, its wings have a much larger vertical component,
which increases the RCS. Despite doppler spreading, it is often possible for a
surface-wave HF radar to follow an aircraft as it turns.
5.3 Equation (4.78) gives k = 4.8, i.e. a margin of 6.8 dB must be added to
the mean. Equation (4.81) gives a SNR of 19.9 or 13.0 dB.
5.4
Observation/time 2 3 4 5 6
:x 1.0 1.0 0.83 0.7 0.6

fl 1.0 1.0 0.5 0.3 0.2
c. 0 35 88 118 158
.'C p 0 35 70 128.9 162 199.5
x~ 35 84.9 121.3 159.6
lin 35 44 40.7 39.9
5.5 Track error is 0.6 times measurement error after five observations. The
standard deviation is about 4 m, and Eq. (5.29) suggests an association gate
size of 6 m.
Chapter 6
6.1 (a)
V u(t)
p,,(t)
-TI2 TI2
Output from matched filter

(b) u (t) = V exp ( - t 2j T2 )
Pu(t) = f"oo V 2
exp[ -(s + t)2jT2] exp( -s2jT2)ds
= V 2 exp(-t 2j2T 2 ) fX:oo exp[-2(s+tj2)2jT2]ds

= V 2T J(1tj2)exp( -t 2j2T 2 )
Using
0-
J I(21t) f oc'
-00
exp( -s2j20-2)ds = I
drd = too", exp(-t 2jT 2 )dt= TJ1t

21tJV 4 exp(-4t 2 jT 2 )dt 21tJ(1tT 2j4) 2J1t
.1.. = = =--
W [JV2exp(-2t2jT2)dtF 1tT2j2 T
6.2 The outputs are as shown:
2
Td = 3T12
6.3 Use Eq. (6.19). Zero doppler:
Output = X(t,O) = { °
I -ltllT It I ~ T
It I > T
Doppler frequency = 2n1T:
ejlttlT (lIn) sin(nltl/T) It I ~ T
Output = X(t, 2n1T) = {
° It I > T
Total output = I - ~I + 1 eilttlT sin(nltl) It I ~ T

TnT
Square law detecting this output yields
(I -liY + :2 sin 2(n~l) +-~! (I _I~) cos( i) sin( n~l)

= (I - ~Y + :2 sin2(~) + ~(l _1~)sin(2n;tl) It I ~ T
In practice, the values in this question are not too realistic for aircraft. What
is the doppler velocity corresponding to T = I ms?
6.4 Using the notation of problem 4.2, and the answer to problem 6.1, gives
~!d = 4 T 13 (replace T by 2 T) for first pulse. For the second (see diagram)
1.2
~rd = ~(fT p2(t)dt + f2T p2(t)dt

~rd 4~ T(fT (02T _ 3t)2 dt :
= 2 f2T (t _ 2T)2 dt) = ~!.
2T ° T 3
This confirms our remark in the solution to problem 4.2 about the resolution
of the second pulse.
6.5 If there are M pulses, the energy both in u ( t) and in u' ( t) are M times
greater than for a single pulse. Hence using Eq. (6.3 ), the effective bandwidth
does not change.
u(t) = p(t) + p(t - T) + p(t - 2T) + p(t - 3T)
Therefore,
U(w) = (1 + e- jcoT + e- 2j(t)T + e- 3j(t)T)p(w)
1- e- 4j (t)T
= P(w)
1- e-j(t)T
e- 2j (t)T (e 2j (t)T _ e- 2j (t)T)
= e- j (t)T/2 (ej (t)T/2 _ e- j (t)T/2) P(w)
-3;.,T!2 sin(2Tw) ( )
= e ' P w
sin(Tw/2)
Plots (shown below) of IP(w)1 for a = 1 and a = t, of Isin(2Tw )/sin(Tw/2)1
and of their product show that the statement does not hold for any of the
other commonly used definitions of bandwidth, e.g. 3 dB width or distance
to first zero in the transform. To make the statement reasonable, the pulses
2.0
lP(w)1 T= 'If
a= 1
-8 -6 6 8 w
• 1'=0<;
4
3.S
3
2.S
I 2
1.S
O.S
onvoga
;T_a=l
i 3
0_
8 -6 -4 4 6 8
""'¥
T=7f
a = 1/2
IP(w)1
0.4
0.2
3.S
2.S
.,..'"'
-Ii
loS
O.S
omega
need to be rectangular, and of width equal to the inter-pulse period. Here
IP(w)1 = a; sa a:w)
2(
where Sa x = (sin x)/x.

6.6 Use Eq. (6.37).
where ~ = radial velocity of ship. Therefore,

Vr = Vf3/4
The sign needs interpretation. For SAR, k in Eq. (6.33) is negative. Hence,
if the ship's motion is towards the platform (Wd is positive), the corresponding
time is positive, i.e. the ship is imaged at a later time, so is moved in the
along-track direction in the image (not the opposite way). So we know the
ship is moving towards the radar; the component in the direction is
4.5 km h- i (unless aliased). From the wake direction we can then calculate
the full velocity vector.
6.7 (a) -1,0,1,4,1,0, -1

(b) 1,0, -1. 4, -1,0,1
(c) 1, 0, 1, 0, 3, 0, - 1, 0, 0, 0, 0, 8, 0, 0, 0, 0, - 1, 0, 3, 0, 1, 0, 1
Chapter 7
7.1 The power received by a unity-gain antenna at 256 n. mile (~474 km)
would be '" -99 dB W. This is nearly 2 dB greater than the -101 dB
required to trigger the transponder, and the signal is therefore adequate
when the aircraft is in the centre of the beam. The excess power (1.7 dB)
would ensure that, as the interrogation beam swept over the aircraft,
0
the transponder would be triggered for about 3/4 either side of maximum
illumination. This must be near the limit of satisfactory performance and is
probably the reason why Cossor quote 256 n. mile as the maximum range
of the system.
7.2 Pr = - 95.6 dB W at the centre ofthe beam. Over the azimuth beamwidth
of 2.45° the signal would equal, or exceed, -98.6 dB W.
7.3 Rmax = 616.6 km =::: 333 n. mile. Losses amount to 12 dB.
7.4 The power x aperture product of the uplink (52.5 dB W + 0 dB) is

similar to the downlink (24 dB W + 27 dB). The main difference lies in the
greater sophistication of the ground-based receiver being able to operate at
signal levels ( - 110 dB W) that are smaller than the transponder receiver
on the aircraft ( -101 dB W).
7.5 From problem 7.2, Pr = 95.6 dB Wand SNR = 34.4 dB. We could there-
J
fore expect the intrinsic resolution of 2.45° to be improved by (2 x SNR)
to give an uncertainty of 0.03°. In practice, monopulse can improve upon
this figure when the aircraft is near the centre of the beam and the uncertainty
might be nearer half this value. A good guide to the dependence of the
angular error on the SNR, angle from bore-sight, etc., is to be found in
Chapter 10 of Stevens 1 •
Chapter 8
8.1 (a) 112.9km, (b) 130.4km, (c) 120.9km, (d) 233.2km.
8.2 35.1 m and 56.0 m respectively.
8.3 (a) The detection would be halved to 2250 m by the 6 dB diffraction loss.
(b) The detection range of a point target would be reduced by a factor
of 0.71.
8.4 500 km. This principle is used in altimetry (see Chapter 11) to remove
ionospheric effects and allow precise measurements of ocean surface height
from space.
Chapter 9
9.1 Z = 200(1 )1.6 = 200 = 23[dB mm 6/m- 3 ], sO
Constant = - 106.4 [dB m s - 1] 2.3 X 1011 m S-1
Pt = 58.1 [dB W] 650kW
GtGr = 74 [dB] 37 dB each way
Z = 23 [dB mm 6 m- 3 ]
,'!.l) A¢ = -29.1 [dB rad 2] 2° beam
T = - 57.0 [dB s] 2 JlS pulse
1/).,2 = 19.7 [dB m- 2] 2.9GHz
1/ R2 = -86.0 [dB m- 2] 20 km range
Pr = -103.7 [dB W]
The SNR would therefore be 33.3 dB. More detailed performance calculations
by Met. Office engineers arrive at a similar figure of 32.6 dB.
9.2 Z = 200(100)1.6 = 317 X 103 = 55 [dB mm 6 m -3]. This would increase

the SNR by 32 dB (the Met. Office performance calculations likewise show
a 32 dB enhancement).
9.3 Equation (9.7) can be reduced to

200(r) 1.6
Pr oc R2
The largest ratio is when r = 64 and R = 1 km, giving - 8.1 dB. The smallest
ratio is for r = 1/8 and R = 200 km, giving -97.5 dB. The dynamic range
requirement is therefore nearly 90 dB. This can be relaxed by using
range-dependent gain, and Siemens-Plessey do in fact use swept gain
following a 1/ R2 law to 200 km.
9.4 The R2 term increases the loss by only 0.1 dB and this is almost
compensated by an increase in the area of the layer illuminated by the
off-vertical beam. The cause of the 10 dB drop in power when viewing
off-vertical (an effect known as aspect sensitivity) may be an anisotropy of
the turbulence at the edges of the layer, although scattering from steps in
the refractive index may cause the same effect. Not all layers exhibit aspect
sensitivity.
9.5 We have
11 = 0.38C~)" -1/3
= 0.38 X 10- 17 x 0.55 = 2.1 x 10- 18

and so
C; = ARG'1
= 300 x 1259 x 2.1 x 10- 18 = 7.9 X 10- 13 = -12l.0dB
Therefore,
PI = 57.8 dB W
A2 = I5.0dB m 2
Lsys = -7.0 dB
C; = -121.0 dB
1/(16n2 ) = -22.0dB
I/R2 = -80.8dBm- 2
Pr = -158.0dB W
9.6 The MU transmitter power is 2.2 dB larger than for the SOUSY system.
The antenna gain, at 34 dB, is 3 dB larger, and so the SNR should be about
5.2 dB larger. This illustrates the efficiency of performing calculations in dB,
rather than multiplying linear terms.
9.7 The PRF must be 6250 Hz and the total number of pulses transmitted
in 10.5 s is 65536 (see also Eq. (10.1) in the next chapter). Averaging these
echo pulses in blocks of 512 leaves 128 samples for the FFT processing. Each
FFT cell will be 1/10.5 Hz wide and, as there are 128 points, the doppler
spectrum will be ± 6 Hz wide.
9.8 The component of the aircraft's horizontal velocity in the off-vertical

beam is 300 sin 12° = 62 m s -1 = 19 Hz doppler shift. This is outside the
doppler spectrum and the aircraft would not be observed. In the vertical
beam the aircraft would be observed as a very large low velocity target, which
would contaminate wind observations at that height.
9.9 Yes. The calculations in Section 9.4 hold for MST radar and show
meteors to be very bright targets, although somewhat short-lived compared
to the integration times usually used by MST radars. For a meteor to be
observed in a narrow vertical beam of an MST radar, it must be travelling
horizontally with a considerable path length through the atmosphere. Under
these cond!tions, many small meteors burn up before they reach the radar
beam and some larger meteors disintegrate through thermal shock. For these
reasons, although MST radars are used to observe meteors, few meteor
echoes are observed in the vertical beam.
Chapter to
to.t SRI quote the following figures: (a) 0.5°, (b) gain exceeds 30 dB between
8 and 28 MHz, (c) 93 dB W, (d) FOM of 104 dB J.
10.2 (a) AR = 3 km, R AO = 13 km, so area = 39 x 106 m 2 == 75.9 dB m 2 •

(b) 52.9 dB m 2 == 195 x 103 m 2 •
10.3 Next = 60 - (2 X 15) - 204 + 47 = -127 dB W
10.4
4n = 11.0 dB
PtGtGrt; = 104 dB J
(1 = 52.9 dB m 2
m= 6 dB
Is = -15 dB
l/Fa = -30 dB
l/kTo = +204 dB W- 1 Hz
1/UhA)2 = -275 dB m-2
SNR= 57.9 dB
10.5 The doppler shift corresponds to a velocity of 150 m s - 1. The SNR is

about 35 dB lower than the sea echo, giving an apparent RCS of 18 dB m 2 ;
because of the height factor h, the true RCS is 24 dB m 2 • This is therefore
a large aircraft travelling slowly.
Even when the ionospheric losses are undetermined, the sea clutter can
be used in the way above as an approximate method of calibrating the RCS
of an unknown target.
Chapter 11
11.1 The dominant Bragg line should appear at ±0.4 Hz in the doppler
power spectrum. The extra shift of 0.3 Hz implies a tidal current of 3 m s - 1
or nearly 6 knots. Ship speeds are generally restricted to 5 knots approaching
a harbour, and hence the knowledge of a 6 knot cross-current could aid the
pilot.
11.2 The travel time difference between two ranges separated by rs is
2
-[(R+rs)-R]
c
Hence
2rs ' '-- ,.'
"
c B it.
The range difference between near and far edges of the beam (in slant range) is
h h
RM-R = ---
m cos(O + P) cos (J
-h( cos 0 -
1P __1)
sin 0 cos 0
hP sin (J
cos 2 0
Therefore
2B
N - - hP tan 0 sec 0
c
Using values B = 5 X 108 , P= 0.06/0.6 = 0.1 rad, then
N-8secOtanO
The variation of N with 0 is shown in the plot below.
I:~
20t
~ 1;L-~---1-~~--L--....,.L---~
Incidence angle [deg]
11.3 We have (see diagram)

Area = 1t(d~ - di)

= 1t[S2 - (S - Ct)2]
= 1tcr(2s - Ct)
'" 2n:CTS
Minimum value of s is R. Maximum value of s is
(R2 + R2P2/4) 1/2 '" R(1 + P2/8)
0
For a 2 beamwidth, s varies by less than 0.02 per cent. So
Area'" 21tRcr
Time for leading edge to reach beam edge is
~~p2
c 8
Hence condition given.
For ERS-l: p = )./d = O.ot8 rad or 1.04 and R 0
= 7.85 x lOs m. Thus,
for pulse-limited operation,
r ~ 1.07 x 10- 7 s
The radar equation is
P'G 2C1A).2L
P, = I s
r (41t)3R 4
whereC1 = ReS/unit area, A = illuminated area, G = 41tA;/).2 and r = antenna
radius. So
P,1t 2r4 C1cr).2L
P, = I s
r 2R3
For a simple pulse and matched receiver, noise power = kToB - kTo/t, giving
212L
SNR = P,In: r C1cr.ll. s
2 4
3
2R kTo
= 0.0375PI C1
= 59.3 (17.7 dB)
for PI = 500 Wand C1 = 5 dB m 2 • In practice, the ERS-l altimeter uses a
20 ps chirp with 50 W Tx power.
11.4 Number of complex multiplications/second for real-time operation is

4R)'V
M = -3 -
d
Therefore, using subscripts a and s for air-borne and space-borne,
Ma = Ra v" (d s )3
Ms Rs V. da
1 1
=-x-xlO00
20 135
- 0.37
At C-band, Ma = t X 106
At L-band, Ma = i X 106
11.5 The return from the slant swath will occupy a time duration
2 2Rs
-(RM - Rm)=-
c c
Successive pulses must be at least this far apart, even with no margin, i.e.
PRF ~ c/2R s
But PRF ~ 2V/d, and so
Rs ~ cd/4V
For the aircraft:
8
R & 3 X 10 X 2 x 3600 = 180 km
s -..::: 4 x 300 x 1000
In practice, aircraft SARs normally use much narrower swaths. Air-borne
SARs therefore do not usually have to trade swath width for resolution.
By contrast, for the spacecraft:
3 5
d >- 4 x 7.5 X 10 x 10 = 10 m
r 3 X 108
Hence best possible resolution is 5 m in the along-track direction. The
trade-off between coverage and resolution is a serious design consideration.
11.6 Lay-over occurs when the top of a slope is nearer to the radar than
the bottom (see diagram). The limiting case occurs when the top and bottom
(and the whole of the slope) are imaged in the same place, i.e., are equidistant
from the radar. This is the line OX in the diagram. At the long ranges used
by SAR, this means that OS is nearly perpendicular to OX, or <p = (). Hence
any slope facing the satellite with a gradient exceeding 23° will suffer lay-over.
Slopes of this magnitude are common in hilly areas.
For shadowing, the radiation to X must be blocked by Y. This

corresponds to a slope of 67° on the slope away from the satellite (steep,
but not uncommon in mountainous terrain).
For the aircraft, incidence angles range from 78.7° to 84.3°. Hence
lay-over will occur only in very steep terrain, but shadowing will be common.
SAR operating at these angles can respond to very slight variations in
topography, such as you see near sunset as shadows lengthen. Note that for
the resolutions of the order of 1 m or so available to air-borne radar, lay-over
is always encountered in images of urban areas.
Chapter 12
12.1 200 V /10 mV implies a dynamic range of 86 dB, reduced to 69 dB by
the antenna losses. The attenuation rate and the depth requirement suggest
a 99 dB dynamic range. The signal must therefore be boosted by at least
30 dB if targets are to be detected to depths of 4.3 m. These figures are, in
fact, those given in reference 6, in which it is reported that objects were
detected to 3.5 m (the limiting factor being mainly television interference).
12.2 11.6 MHz (PRFs this high are not used in practice).
12.3 The signal appearing in the receiver may be only 17 dB below 200 V, or
as high as 28 V (reference 6). This would be sufficient to damage the receiver
if protection methods (attenuators and diode limiters) were not used.
12.4 For two-way propagation losses of 200 dB/l00 m (right-hand scale of

Fig. 12.1), the 99 dB budget in the answer to problem 12.1 would enable
the penetration to be extended to about 50 m.
12.5 At 10 MHz the losses are about 12 dB/l00 m, suggesting that depths
up to 825 m might be achieved. However, at this frequency, external noise
would dominate internal noise and the system performance could be worse.
On Mars, the noise level would be determined by galactic noise.
12.6 PRF = 60 kHz. A half-wave antenna would be 15 m long.
Chapter 13
13.1 (a) 1 kHz.
(b) 2 kHz.
(c) Monostatically the target motion decreases the path of the radio
signal on both the outward and the return leg. In this bistatic case, only the
outward path is shortening.
13.2 (a) 60 km (remember to allow for the time taken by the direct signal
travelling along the baseline).
(b) 39.7 km.
(c) 20.3 km.
13.3 (a) 1082 Hz (plus or minus, depending on whether the aircraft is

travelling towards, or away from, the receiver).
(b) 47S (from the cosine formula).
13.4 (a) 6.7 km (about 22000 ft). In practice, some pulse integration would
be needed for the doppler processing and the extra gain from this would
enable the aircraft to loiter at much higher altitudes.
(b) The direct signal would be about 41 dB higher than the echo. This
should cause no dynamic range problems for the receiver.
(c) A radar on board the aircraft would be able to use the 20 dB
antenna gain both ways. This would more than compensate for the additional
path losses, allowing a smaller transmitter to be used (assuming the same
RCS). However, it would give away the aircraft position.
Chapter 14
14.1 10.5 dB.
14.2 40 dB to -104 dB W.
14.370mW.
14.4 The noise power spectral density of 70 mW in a 1 MHz bandwidth

would have to be extended across a 20000 MHz bandwidth. The total noise
power would therefore have to be raised 43 dB to 1.4 kW.
14.5 10000 pulses.
14.6 A suitable PRF might be 100kHz (unambiguous range 1.5 km). The
answer to problem 14.5 implies an integration time of 0.1 s at this PRF. This
gives a speed resolution of 0.15 m s - 1, or 0.3 mph, sufficient to determine
whether a car is speeding.
14.7 (a) Radar warning receivers (RWRs) covering several bands used by
speed traps can be purchased, but they are expensive.
(b) A jamming device would be illegal and very expensive, unless the
exact radar frequency was known.
(c) Stealth materials are also expensive and only really effective against
a radar of known frequency. In any case, it would be difficult to conceal
some parts of the car such as the headlights, driver, radio antenna, etc.
Perhaps the cheapest solution is to keep to the speed limit. This also sa ves fuel.
Chapter 15
15.1 The antenna gain can be derived from either 475 (26.8 dB) plus 7.24 dB
for each element, or from G = 4nAI A2 , using the array area of 8330 m2 • Both
methods give a gain of 34 dB. The beamwidth AID = 6.45/103 = 3.6° is the
value quoted in the article.
15.2 Using Eq. (15.13) (appropriate for no weighting) we would expect a

spacing of a little over 0.6A. In fact, a triangular lattice of spacing 0.7A. is
used since triangular grids generally allow a wider range of antenna beam
steering than rectangular grids with the same element density 9. A further
sophistication of the MU system is that groups of 19 antenna elements are
collected into hexagonal sub-arrays.
15.3 Roughly 56° off-zenith for the vertical beam. The authors quote 'no
grating lobe is formed at beam positions within 40° of the zenith '.
15.4 Calculating 475 phase angles and setting up 475 phaseshifters 2500
times per second would require over one million such operations per second.
Instead, the MU system stores all phaseshift data required for beam steering
in a ROM (read-only memory) installed in each transmit module. The
appropriate values are then read out according to instructions from the
module controller.
BIBLIOGRAPHY
Throughout this book we have tried to restrict references to books which are relatively
easy to obtain. Occasionally we have referred to original publications in journals,
but these can be found in most university and industrial libraries. To help with finding
references, we list them for you (first author only) alphabetically below.
Abramowitz, M. Handbook of Mathematical Functions, Dover, New York, 1964.

Atlas, D. (ed.) Radar in Meteorology, Battan Memorial and 40th Anniversary Radar
Meteorology Conference, American Meteorology Society, 1990.
Barker, R. H. Group synchronizing of binary digital systems, in Communication
Theory, (ed.) W. Jackson, Academic Press, New York, pp. 273-287, 1953.
Barrick, D. E. Remote sensing of sea state by radar, chapter 12 in Remote Sensing
of the Troposphere, (ed.) V. E. Derr, NOAA Environmental Research Laboratories,
Boulder, Colorado, 1972.
Bar-shalom, Y. Tracking in a cluttered environment with probabilistic data
association, Automatica, 11,451-461, 1975.
Battan, L. J. Radar Meteorology, University of Chicago Press, Chicago, 1959.
- - - Radar Observation of the Atmosphere, University of Chicago Press, Chicago,
1973.
Bean, B. R. Radio Meteorology, Dover Publications, New York, 1968.
Berkowitz, R. S. Modern Radar: Analysis, Evaluation and System Design, Wiley, New
York, 1966.
Bracewell, R. N. The Fourier Transform and its Applications, McGraw-Hili, New
York, 1986.
Brigham, E. O. The Fast Fourier Transform, Prentice-Hall, Englewood Cliffs, NJ,
1974.
Brookner, E. (ed.) Practical Phased-Array Antenna Systems, Artech House, 1991.
Caldecott, R. Underground mapping of utility lines using impulse radar, lEE Proc.-F,
135(4),343-353, 1988.
CCIR, World Distribution and Characteristics of Atmospheric Radio Noise, CCIR
Report 322, 1963.
- - - Reference Atmosphere for Refraction, Recommendations and Reports of the
CClR, CCIR Rec. 369-3, V, ITU, Geneva, 1986.
Cole, H. W. Understanding Radar, BSP Professional Books, Oxford, 1985.
Colgrave, S. B. Track initiation and nearest neighbour incorporated into probabilistic
data association, IE Aust. IREE Aust., 6(1), 191-198, 1986.
364
BIBLIOGRAPHY 365
Connor, F. R. Antennas, Edward Arnold, London, 1972.

Costas, J. P. A study of a class of detection waveforms having nearly ideal
range-doppler ambiguity properties, Proc. IEEE, 71., 996-1009, 1984.
Doviak, R. J. Doppler Radar and Weather Observations, Academic Press, New York,
1984.
Eastwood, Sir E. Radar Ornithology, Methuen, London, 1967.
Elachi, C. Spaceborne Radar Remote Sensing: Applications and Techniques, IEEE
Press, New York, 1987.
Frank, R. L. Polyphase codes with good nonperiodic correlation properties, IEEE
Trans. Illformation Theory, IT-9, 43-45, 1963.
Fukao, S. The MU radar with an active phased array system: 1. Antenna and power
amplifiers, Radio Sci., 20(6),1155-1168, 1985.
Georges, T. M. Real-time sea-state surveillance with skywave radar, IEEE J. Ocean
Eng., OE-8(2), 97-103,1983.
Gjessing, D. T. Target Adaptive Matched II/umimation Radar, lEE Electromagnetic
Waves series 22, Peter Peregrinus, Hitchin, Herts, 1986.
Gonzales, C. A. Pulse compression techniques with application to HF probing of
the mesosphere, Radio Sci., 19, 871-877, 1984.
Gossard, E. E. Radar Observations of Clear Air and Clouds, Elsevier, Amsterdam,
1983.
Greenhow, J. S. Turbulence at altitudes of (80-100) km and its effects on long
duration meteor echoes, J. Atmos. Terr. Phys., 384-392, 1959.
Griffiths, J. Radio Wave Propagation and Antennas, Prentice-Hall, Englewood Cliffs,
NJ,1987.
Hall, M. P. M. (ed.) Radiowave Propagation, Peter Peregrinus for the lEE, Stevenage,
Herts, 1989.
Harmuth, H. F. Transmission of Information by Orthogonal Functions, Springer-
Verlag, Berlin, 1972.
- - - N onsinusoidal Waves for Radar and Radio Communication, Academic Press,
New York, 1981. .
Hinkley, E. D. (ed.) Laser Monitoring of the Atmosphere, Springer-Verlag, Berlin,
1976.
Hirsch, H. L. Practical Simulation of Radar Antennas and Radomes, Artech House,
Norwood, MA, 1988.
Jackson, M. C. The geometry ofbistatic radar systems, lEE Proc.-F, 133(7),604-612,
1986.
Jane's, Jane's Radar and Electronic Warfare Systems 1989-90, London, 1989.
Kolosov, A. A. Over-the-Horizon Radar, Artech House, MA, 1987.
Kovaly, 1. J. Synthetic Aperture Radar, Artech House, MA, 1976.
Leonov, A. I. Monopulse Radar, Artech House, Norwood, MA, 1986.
Levanon, N. Radar Principles, Wiley, New York, 1988.
Lipa, B. J. Extraction of sea state from HF radar: mathematical theory and modelling,
Radio Sci., 21(1), 81-100,1986.
Long, M. W. Radar Reflectivity of Land and Sea, Lexington Books, Lexington, MA.,
1975.
Lynn, P. A. Radar Systems, Macmillan, London, 1987.
Maffett, A. L. Topics for a Statistical Description of Radar Cross Section, Wiley, New
York, 1989.
Manasse, R. Range and velocity accuracy from radar measurements, MIT Lincoln
Lab. Report, 312-326, 1955.
Marcum, J. A statistical theory of target detection by pulsed radar, IRE Trans., 1T-6,
145-267, 1960.
McKinley,D. W. R. Meteor Science and Engineering, McGraw-Hill, New York, 1961.
Meeks, M. L. Radar Propagation at Low Altitudes, Artech House, MA, 1982.

Mischenko, Y. A. Over-the-Horizon Radar, Zagorizontnaya Radiolokatsiya, Military
Publishing House, Ministry of Defence, Moscow, 1972.
Nathanson, F. E. Radar Design Principles, McGraw-Hili, New York, 1969.
Neal, B. T. CH-the first operational radar, GEC J. Res., 3(2), 73-83, 1986.
Papoulis, A. Probability, Random Variables and Stochastic Processes, McGraw-HilI,
New York, 1985.
Pedrotti, F. L. Introduction to Optics, Prentice-Hall, Englewood Cliffs, NJ, 1987.
Picquenard, A. Radio Wave Propagation, Macmillan, London, 1974.
Radford, M. F. Electronically scanned antenna systems, Proc. lEE, 115(l1R),
1100-1112, 1978.
- - Whither radar, GEC J. Res., 3(2), 137-143, 1985.
Ramsay, D. A. The evolution of radar guidance, GEC J. Res., 3(3), 92-103, 1985.
Rhodes, D. R. Introduction to Monopu/se, McGraw-Hili, New York, 1959.
Rohan, P. Surveillance Radar Performance Prediction, Peter Peregrinus for the lEE,
Rice, S. O. Mathematical analysis of random noise, Bell Syst. Tech. J., 23 and 24,
1944; reprinted in Selected Papers on Noise and Stochastic Processes, (ed.) N. Wax,
Dover, New York, 1954.
Rotheram, S. Theory of SAR ocean wave imaging, in Satellite Microwave Remote
Sensing, Ellis Horwood, Chichester, 1983.
Rudge, A. W. (ed.) The Handbook of Antenna Design, vol. 2, lEE Electromagnetic
Wave series, Peter Peregrinus, Hitchin, Herts, 1983.
Rumsey, V. H. Frequency Independent Antennas, Academic Press, New York, 1966.
Scanlan, M. J. B. (ed.) Modern Radar Techniques, Collins, Glasgow, 1987.
Shafer, G. A Mathematical Theory of Evidence, Princeton University Press, Princeton,
NJ,1976.
Shearman, E. D. R. Radio science and oceanography, Radio Sci., 18(3),299-320,
1983.
Sherman, S. M. Monopulse Principles and Techniques, Artech House, Norwood, MA,
1984.
Schwartz, M. Information Transmission, Modulation and Noise, McGraw-Hill, New
York,1980.
Skolnik, M. I. Radar Handbook, McGraw-Hill, New York, 1970.
---Introduction to Radar Systems, McGraw-Hili, New York, 1985.
- - Radar Handbook, McGraw-Hill, New York, 1990.
Stevens, M. C. Secondary Surveillance Radar, Artech House, Norwood, MA, 1988.
Stremler, F. G. Introduction to Communication Systems, Addison-Wesley, Reading,
MA,1982.
Sugar, G. R. Radio propagation by reflection from meteor trails, Proc. IEEE, 52,
116-136, 1964.
Sweetman, B. Stealth Bomber, Airlife Publishing, Shrewsbury, 1989.
Swerling, P. Detection of fluctuating pulsed signals in the presence of noise, IRE
'Irans., 1T-3, 175-178, 1957.
- - Probability of detection for fluctuating targets, IRE 'Irans., IT-6, 269-308,
1960.
Swords, S. S. Technical History of the Beginnings of RADAR, Peter Peregrinus for
the lEE, Stevenage, Herts, 1986.
Toomay, J. C. Radar Principles for the Non-specialist, Van Nostrand Reinhold, New
York, 1989.
Ulaby, F. T. Microwave Remote Sensing: Active and Passive, vols 1-3,
Addison-Wesley, Reading, MA., 1981.
Wax, N. (ed.) Selected papers on Noise and Stochastic Processes, Dover, New York,
1954.
BIBLIOGRAPHY 367
Weeks, W. L. Antenna Engineering, McGraw-Hili, New York, 1968.

Wiley, R. G. Electronic Intelligence: The Analysis of Radar Signals, Artech House,
Dedham, MA, 1982.
Willis, N. J. Bistatic Radar, Artech House, Norwood, MA, 1991.
Woodward, P. M. A theory of radar information, Phil. Mag., 41, 1001, 1950.
- - - Probability and Information Theory, with Applications to Radar, Pergamon
INDEX
Abbreviations, 3, 330-2 Antennas (continued)

Accuracy, 15-18, 129 functions, 6-7
Acronyms, 330-2 LVA, 163, 168
Active arrays, 310-11 multistatic radar, 272
Active homing radar guidance patterns, 8
systems, 59 recent developments, 296
Active Microwave Instrument, 247 Skywave radar, 214
ADA software language, 312 SSR, 163
Adaptive doppler filtering, 323 surface-wave radar, 208
Adaptivity, 322-5 tracking radar, 49
Air-borne surveillance radar, 302 Anti-radiation missile (ARM), 288
Air-borne warning and control Array factor, 298
system, (AWACS) 30, 303 Artificial intelligence, 312, 325
Air defence radar, 206, 313 Atmospheric circulation models,
Air traffic controllers (ATC), 161-3, 237
313 Atmospheric effects, 173-8
rx-p filter 122-4 Atmospheric radars, 190-205
Ambiguity function, 138-43 applications, 190-1
Chirp, 146 Atmospheric research, 191
contour plot, 143, 146, 152 Atmospheric retardation, 178, 181
definition, 138 Attenuation:
examples, 142-3 atmosphere, 174
properties of, 139-41 ionosphere, 183
role of, 139 Autocorrelation, 88
Amplitude-comparison monopulse Autocorrelation functions (ACFs),
radars, 50-2 74-5, 90, 135, 137, 150
Analogue/digital (A/D) converter, AWACS (air-borne warning and
13, 16,31, 104,209,215,308 control system), 30, 303
Angels, 194-5 Azimuth angle, 7
Angle noise, 55 Azimuth bandwidth, 10
Angles, 265-7
Antenna gain, 6--11, 14 Band-limited signal, 129
Antennas, 3, 5 Band-limited white noise, 41, 75
aperture, 11 Bandwidth, 16--18,65, 129-31,260
beam width, 26--7 definition, 129-30
carrier-free radar, 264-7 developing concept of, 320-2
design, 26 Barker codes, 150, 152, 154
368
INDEX 369
Barrage jamming, 290 Clutter (continued)

Battlefield radar systems, 179-81 volume, 39
Beamforming, 214-15, 303, 306 Co-channel interference, 165
digital, 297, 308-10 Coherent integration, 209
Beamriding missiles, 58-9 Coherent processing, 91, 92
Beamsteering, 303 Coherent receiver, 32
Beamwidth factor, 299-300 COMINT,287
Biconical antennas, 265 Complementary motion, 283
Birds, 194-5 Compression ratio, 148
Bistatic doppler, 280-3 Conical scan radars, 49-50
Bistatic operation, 233 Constant false-alarm rate (CFAR),
Bistatic radar 271, 294 118
Bistatic radar equation, 275-7 Continuous-wave (CW)
Bistatic techniques, applications, modulation, 209
283 CONUS-B system, 214, 221-2
Blind speeds, 111 Conversion factors, 333--4
Bragg formula, 229 Convolution operation, 64
Bragg lines, 230-4 Correlation operation, 138
Bragg resonant scatter, 228, 231 Correlation receiver, 66, 84-90, 138
Bragg resonant sea waves, 229 Cosine-squared pulse, 57-8
Bragg ripple waves, 249
Bragg scattering, 228 Data fusion, 323--4
Brewster angle, 208 Data processing, 104-27,267-9
Burn through, 291 Deception jamming, 291
Decibels, 14, 335-6
Calibration target, 36 Decision making, 66-73
Carrier-free radar, 260-4 Delay line, 111
antennas, 264-7 Delta function, 75
Cauchy-Schwartz inequality, 84, 85, Detection probability, 116-18
141 Deterministic autocorrelation
C-band, 240, 241 function, 88
CCIR reference atmosphere, 177 Differential phaseshift keying
Centre of gravity, 55, 58 (DPSK),168
Chaff, 291 Diffraction:
Chain Home, 206 by the terrain, 178-9
Channel occupancy monitoring ionosphere, 185
receivers, 219 Digital approximation, 108
Chi-squared distribution, 236 Digital beamforming, 297, 308-10
Chi-squared variable, 36 Directivity, 7
Chirp radars, 18, 145-50 Discrete Fourier transform (DFT),
ambiguity function, 146 108, 114-15
Clear-air turbulence, 196-7 Dispersion, 146, 173
Clutter, 38--41, 63, 129, 267 Distributed clutter, 38
attenuation factor, 110 Doppler accuracy, 131-2
distributed, 38 Doppler formula, 229
fixing, 41 Doppler frequency, 20, 134
maps, 41 Doppler power spectrum, 230, 231,
point, 38 233
properties of, 106-7 Doppler shift, 19-21, 107,229,281,
spreading, 282 282
subsurface, 267 Doppler spectrum, 112, 209
surface, 38, 39, 267 analysis, 315
tuning, 282 Duplexer,3
Dynamic range, 211 Four-thirds earth, 176-7

Fourier series, 261
Echo frequency, 19 Fourier transform, 64, 74, 108, 129,
Echo pulses, 4, 13 131,300
Effective antenna noise factor, 220 Fourier transform pairs, 130, 134
Effective bandwidth, 130 Frank coding, 151-2, 154
Effective pulse duration, 130 Frequency agility, 55, 58
Efficiency factor, 9 Frequency diversity, 290
Electromagnetic (EM) spectrum, Frequency-domain ACF, 137
286 Frequency hopping, 290
Electromagnetic (EM) waves, 181 Frequency management, 219
Electronic counter-countermeasures Frequency-modulated continuous
(ECCM), 287, 290-2 wave, (FMCW) modulation 209,
Electronic countermeasures (ECM), 210, 212, 267
287, 290-2, 322 Frequency-modulated interrupted
Electronic support measures continuous wave (FMICW)
(ESM), 263, 287 modulation, 209, 210
Electronic surveillance, 287 Frequency modulation (FM), 131
Electronic warfare, 286-95 linear, 145-54
objectives and definitions, 286-8 Frequency resolution, 137-8
stealth applications, 292-4 Frequency span, 137
typical radar scenario, 287 Frequency synthesizer, 3
Elevation angle, 7 Frequency transfer function, 75
Elevation beam width, 10 Fruit (false replies unsynchronized
ELINT, 287, 292 in time), 165
Elliptical geometry, 278
Energy/noise ratio (E/N), 44 Gain, 7
Envelope detection, 92 Gainfactor, 11-12
Error function, 71 Gallium arsenide, 6, 311
Error variance, 123 Garbling, 165
Error voltage, 50 Gaussian inputs, 73
ERS-l satellite, 237-9, 241, 250, 251 Gaussian outputs, 73
Escort jammer (EJ), 288 Geoid, 240
European Incoherent Scatter Radar Grating lobes, 53, 300, 304
Facility (EISCAT), 181 Ground-probing radar, 257-70
European Space Agency, 237 applications, 257-8
Expendable jammer, 288 design, 258-60
Exponential PDF, 67, 97 Groundwave effect, 222
External noise, 13 Groundwave radar. (see
Externally noise-limited system, Surface-wave radar)
214 Guard cells, 116
False-alarm probability, 72, 81-4, Half-power beamwidth, 298

116-18 Hazard avoidance, 190
Faraday rotation, 185 Heisenberg uncertainty principle,
Fast Fourier transform (FFT), 107, 134
108, 112-15,209-10 High-frequency (HF) radar,S, 21,
Feeds, 296 40,210-12,228,231
Fibre optics, 307-8 problem areas, 223
First-order Doppler spectrum, 230 sea scattering, 226-30
Floodlight illumination, 208 High-frequency (HF) remote
Floodlighting, 272 sensing, 233-4
Fly-by-wire aircraft, 323 High-PRF radar, 30
INDEX 371
High resolution radar techniques, Lidar (light detection and ranging),

314-16 203
Home-on jammer (HOJ) missiles, Lightning strikes, 146
288 Linear arrays, 297, 306
Horns, 267 Linear frequency modulation,
Huygens sources, 9-10 145-54
Hydrometeors, 192-4 Linearity, 64
Hyperbolic geometry, 280 Lloyd's mirror, 166
Lobes, 166-7
Ice shelves, 240 Long-range surveillance radar, 26,
Identify, friend or foe (IFF) systems, 31
159 Look-ahead receivers, 219
Impulse response function, 64-6 Loss factor, 12-13
Incoherent processing, 91 Losses in radar systems, 43
Incoherent radar systems, 19 Low frequencies, 5
In-phase channel, 32 Low-PRF radars, 30
Input noise, 73, 85 Low probability of intercept (LPI),
Input power spectrum, 75 292
Input signal, 85 LV A (large vertical aperture)
Insects, 195 antennas, 163, 168
Integrated bias detector, 124
Integrated circuits, (Ies) 310, 322
Integration time, 19-21 Magellan spacecraft, 251
Interference fringes, 166 Marconi Radar System Martello 2723,
Interferometry, 315 322
Intermediate frequency (IF), 4, 66 Margin, 81
input, 94-5 Maritime radar, 171
Internal noise, 13,41 Matched filter, 65, 90-1, 136
Internally noise-limited system, 214 Maxwell's equations, 206, 258
Inverse Fourier transform, 74 Mean, 67, 70
Inverse synthetic aperture radar Mean noise power, 41-2
(ISAR),315 Medium-PRF systems, 30
Ionogram, 219 Mesosphere, 196,200
Ionosonde radar system, 185, 219 Mesosphere-stratosphere-
Ionosphere, 196, 212 troposphere (MST) radar,
OTH radar, 219 197-200
Ionospheric effects, 181-7 Meteor trail scattering theory,
Ionospheric propagation, 215 201-2
Ionospheric sounding, 219 Meteor wind radar, 200-3
Ionospheric wave, 206 Microwave optics, 308
Isodop, 280 Microwave radar, 21
Isotropic radiator, 7 Microwave scatterometry, 234-7
Middle atmosphere, 200
Mie region, 33
Jammer transmitter, 290
Mode interlacing, 163
Jump-jet, 19
Mode S, 168-9
Monopulse radar, 50-3
Kalman-Bucy filter, 124-5 Monopulse tracking, 50
Knife-edges, 178-9 . Mortar location, 180
Moving-target indication (MTI), 29,
Lag, 74 30, 107-12
L-band radar, 26, 27, 36,41,42, MU-radar, 200
193, 240, 249 Multifrequency illumination, 314
Multifunction electronically Pencil beam, 272

scanned adaptive radar Phase change, 297
(MESAR),310-11 Phase coding, 150-4
Multifunction radar, 311-12 Phase-comparison monopulse, 52
Multihead radar, 297,312-13 Phase delays, 306
Multi-looking, 247 Phase perturbations, 185
Multipath, 165-8 Phase scintillations, 186
Multiple observations, 91-3 Phase-sensitive detector, 50-2
Multiple spot jamming, 290 Phased arrays, 53, 297-308, 323
Multipolarization techniques, 314 (t),71-3
Multistatic radar, 271-85 Pioneer, 251
advantages, 273 Plan position indicator (PPI), 4,
antenna, 272 161,323
applications, 283-4 Planar arrays, 297, 306, 308
basic arrangement, 272 Plot extraction, 31, 56, 118-19
disadvantages, 273-5 Plot-track association, 119-20
technical aspects, 273 Point clutter, 38
Multistatic target location, 277-80 Power flux, 11-12
Power gain, 7-9, 298
NASA, 237 Power spectrum, 74, 75
NATO band names, 6 Precipitation, 194
Nodding scan, 56 Predetection integration, 209
Noise, 41-3, 63, 129, 131 Presummation, 209
Noise density, 41 Probabilistic data association
Noise distribution, 73-8 (PDA), 125
Noise figure, 42 Probability density function (PDF),
Noise input, 65 36-7, 66, 236
Noise jamming, 288-91 Probability theory, 63, 66-73
Noise power, 75, 77, 114 Propagation factor, 222
Noise sources, 13 Propagation of radio waves, 171-89
Non-linear filtering, 284 along surfaces, 207
Normalized array pattern, 298 Pulse chasing, 283
Nyquist rate, 129 Pulse compression, 145
Pulse doppler radar, 30
Ocean currents, 230-1 Pulse duration, 17, 129-31
Ocean waves, 226 definitions, 130
One-way transmission factor, 222 Pulse length, 31-3, 145,260
Optical fibres, 307-8 Pulse radar, 3
Optical region, 33 equation, 44
Output noise power, 77 Pulse repetition frequency (PRF),
Output power spectrum, 75 28-30,207,210,247
Over-the-horizon (OTH) radar, Pulse-shapes, 16-17, 57, 131
206-25 Pulse-to-pulse variation, 50
equation, 220-2
system losses, 221 Quadpolarized systems, 314
(see also Skywave propagation Quadrature channel, 32
and frequency management;
Skywave radar; Surface-wave Radar:
radar) applications, 1-2
basics of, 2
PACE radar, 181 block diagram of elementary
Parseval's theorem, 77, 130, 137 system, 3
P-band radar, 240 derivation, 1
INDEX 373
Radar (continued) Range error, 178

future of, 319-25 Range gate, 15
nature of, 1-2 Range gate stealer, 291
principle of, 1 Range measurement, 128
simple explanation, 3--4 Range resolution, 15-18, 145,260
Radar-absorbing material (RAM), Raster scan, 56
293 Rate of change of RCS, 97
Radar-absorbing paint, 167 Rayleigh distribution, 95, 106
Radar acoustic sounding system Rayleigh region, 33
(RASS),203 Rayleigh scattering, 192
Radar altimetry, 237--40 Receiver, 4
Radar bandwidth. (see Bandwidth) effects on noise distribution, 73-8
Radar clock, 3 Receiver channels, 32
Radar command to line of sight Reference cells, 116
(RCLOS),58-9 Refraction
Radar cross-section (RCS), 12, 29, atmosphere, 174--8
33-7,64, 135, 192, 195,203, ionosphere, 183
212, 222, 235, 237, 250 Refractive gradient, 175
fluctuation analysis, 315 Refractive index
rate of change of, 97 atmosphere, 175, 177
Radar detection probability, 81--4, ionosphere, 183, 185
87 neutral atmosphere, 196
Radar detection theory, 62-103 Relative bandwidth, 260
Radar echoes from the atmosphere, Remotely piloted vehicle (RPV), 288
191 Repeater jammer, 291
Radar equation, 11-15,43--4 Research radars, 194
ground-wave, 222 Resistively loaded dipoles, 265
and noise jamming, 288-90 Resolution, 15-18, 128, 134--8
OTH,222 definitions, 135-7
Radar frequency, 4--6 properties of, 135
band names, 6 Resonance region, 33
Radar guidance, 58-9 Retardation, atmospheric, 178, 181
Radar horizon, 172-3 Ripple waves, 249
Radar maps, 191 Root-mean-square (RMS) error, 18
Radar reflectivity factor, 192 Royal Signals and Radar
Radar remote sensing, 226--56 Establishment, 248
Radar software, 58 Rutherford Appleton Laboratory,
Radar systems testing, 36 193
Radar warning receiver (R WR)
display, 288 SABRE,181
Radar waveforms. (see Waveforms) Sampling, 31-3
Radio frequency (RF), 3 Satellite dishes, 296
Radio refractivity, 174--5 S-band radar, 26,42,191,193,310
Radio waves, 1, 4 Scattering mechanisms, 191-7
propagation of, 171-89 Scattering structures, 167
Radiosonde data, 181 Scintillations, 186
Raised cosine, 57 Sea clutter, 38
Range, 3 Sea ice, 233--4
Range accuracy, 15-18,57, 131-2, Sea scattering, high-frequency radar,
145 226--30
Range ambiguity, 28-30 Sea waves
Range bin, 15 spectrum, 233
Range/depth resolution, 259 trochoidal shape of, 227
Sea waves (continued) Staggered PRFs, 30

wind effects on, 227 Stand-off jammer, (SOJ) 288, 291
Seasat satellite, 238 Standard deviation, 67, 68, 70, 72-4,
Secondary surveillance radar (SSR), 134,236
159-70 STARE, 181
advantages over primary radar, Staring array, 272
164 Stealth aircraft, 167
antennas, 163 Stealth applications, 292-4
basic principles, 162-3 Step scanning, 212
origins of, 159 Steradians, 7
problems with, 164-5 Stratosphere, 196, 200
range performance, 161 Stratospheric-tropospheric (ST)
Self-screening jammers, 288 radars, 191
Semi-active homing, 59 Sub-clutter visibility (SCV), 40
Sequency,261 Sub-refraction, 178
Sequentiallobing, 49 Subsurface clutter, 267
Shannon-Whittaker sampling Subsurface radar. (see
theorem, 129, 247 Ground-probing radar)
Ship clutter, 38 Superposition property, 64
Ship surveiIIance, 171 Super-refracting ducts, 177-8
Ship tracking, 234 Super-resolution, 315
Short-range radar, 13-14, 33 Surface acoustic wave (SAW)
SI units, 4, 326-9 devices, 248
Sidelobes, 291, 300, 302 Surface clutter, 38, 39, 267
interrogation, 164 Surface-wave radar, 207-12
Siemens-Plessey type 46C weather antennas, 208
radar, 10 Surveillance radar, 91, 120, 171,
Sigma zero, 38-9 194,322
Signal detection, key elements of, 91 design considerations, 25-47
Signal input, 65 final design, 43-5
Signal plus noise distribution, 78-9 multifunction, 25
Signal processing, 104-27 (see also Secondary surveillance
Signal propagation delay, 4 radar
Signal-to-clutter ratio, 40 Swell,227
Signal-to-noise ratio (SNR), 13, 18, Swept gain, 31
21,40, 50, 65, 79-83, 85, 86, Sweriing cases, 36-7, 95-9
92,95, 114, 115, 118, 134 Symbols, 326-9
definitions, 79-81 Synthetic aperture radar (SAR), 69,
Simultaneous lobing, 50 186,235,240-52,314,315
Skywave mapping, 233 images, 148-50
Skywave propagation and inverse, 251-2
frequency management, 215-20 System transfer function, 64
Skywave radar, 207, 212-15
antennas, 214 Tangential fading, 112
OTH,294 Target, 3
relative performance, 216 Target decorrelation bandwidth, 55
Solid angles, 8 Target detection, 145
Space wave, 206 Target fluctuations, 95-9
Speckle phenomenon, 246 Test cell, 116
Specular reflections, 196 Thermal noise, 41, 70, 73
Speed measurement, 19 Theta-theta geometry, 279-80
Spot jamming, 290 Thresholding, 115-18
Squared ambiguity, 141 Time delay, 131
INDEX 375
Time error, 149 Velocity gate stealer, 291

Tomography, 315 Velocity measurement, 128
Total electron content (TEC), 183 Venera 15 and 16 spacecraft, 251
Track branching, 125 Venus, 251
Track initiation, 120-2 Vertical sounder, 219
Track-while-scan process, 48 Very large-scale integration (VLSI)
Tracking accuracy, 54-5 signal processing integrated
angular accuracy, 54 circuits, 322
Tracking noise, 54 Virtual height, 219
Tracking radar, 48-61 Viterbi algorithm, 125
antenna, 49 Volume clutter, 39
function, 48 Volume scattering, 196-8
range estimation, 56
scanning methods, 56 Walsh functions, 261
software, 58 Wave measurement, 231-2
tracking process, 56-8, 122-5 Wave motion, 226
velocity information, 58 Waveforms, 128-58
TRANSIT positioning system, 183 Waveheight spectrum, 227
Transmit-receive switch (T-R Weapon location radars, 180
switch), 3, 261 Weather forecasting, 191
Transmitter breakthrough, 267 Weighting, 302, 306
Transmitter power, 43 White noise, 41, 75
Transponder jammer, 291 Wideband radar systems, 314
Trocoid, 227 Wind effects on sea waves, 227
Troposphere, 191, 196,200 Wind in upper atmosphere, 200-3
Turbulence, 198, 200 Window function, 302
World Ocean Circulation
Experiment, 237
Unambiguous range and velocities,
28-30
X-band radar, 191, 193,240,248
US Naval Research Laboratory,
237
Young's slit experiment, 166
Variance, 67 Zero doppler contour, 280

Understanding Radar Systems

Uploaded by

Copyright:

Available Formats

Understanding Radar Systems

Uploaded by

Document Information

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

Understanding Radar Systems

Uploaded by

Copyright:

Available Formats

UNDERSTANDING RADAR SYSTEMS

SciTech Publishing, Inc

© 1999 by SciTech Publishing, Inc.

© 1992 by McGraw-Hill International (UK) Limited.

All rights reserved. No part of this book may be reproduced

Printed in the United States of America

SciTech books may be purchased at quantity discounts for educational,

2 Designing a surveillance radar 25

6 Designing radar waveforms 128

7 Secondary surveillance radar 159

8 Propagation aspects 171

9 Radar studies of the atmosphere 190

10 Over-the-horizon radar 206

11 Radar remote sensing 226

12 Ground-probing radar 257

13 Multistatic radar 271

13.9 Further reading 285

14 Electronic warfare 286

15 Recent developments 296

16 The future of radar 319

Simon Kingsley and Shaun Quegan

We are grateful to those companies who supplied us with details of

Simon Kingsley and Shaun Quegan

A rough-and-ready guide to radar and its capabilities.

1.1 WHAT IS RADAR?

Table t.t Applications of radar

1.2 A SIMPLE RADAR EXPLAINED

Stable local Coherent local

where R = range [m], C = velocity oflight = 3 x 108 m s - 1 and td = signal

1.3 OVERVIEW OF RADAR FREQUENCIES

where c is the velocity of light, which is about 3 x 108 m s - 1 in air or space

1. Antennas become very large at low frequencies; for example, at 1 MHz,

Despite all these problems, the rewards for long-range over-the-horizon

High Very high Ultra high Super high Extremely high

Wavelength (Aj 100 m 10m

improving with the development of gallium arsenide technology. The

1.4 ANTENNA GAIN

1. To 'beam' power in a given direction in order to increase the radar

Figure 1.3 Azimuth and elevation angles and beamwidths.

are expressed in degrees, then Eq. (1.3a) may be rewritten as

summation of the contributions from all the Huygens sources. We can

G~~= 4nd 1 d2 = 4nA []

Worked example The Siemens- Plessey type 46C weather radar is a

parabolic reflector antenna. If the antenna efficiency factor e is 0.5, what

SOLUTION The physical area of the antenna aperture A is given by

G = 4nAe = 4n x 2.34 == 10350 == 40.1 dB [J

All these arguments involve approximations. More rigorous calculations

1.5 THE RADAR EQUATION

The ability of a radar to detect the presence of a target is expressed in terms

Power flux at distance R = PI 2 (1.8 )

Gt • The power flux in the direction of the beam is now

Power flux at the target = -PtG

The amount of this returning power that is intercepted by the antenna is

The next move is to substitute for Ae by using Eq. "( 1.5) :

Although Eq. (1.13) is a complete description of the power received, it is still

Worked example A short-range surveillance radar operates at 3 GHz

Table 1.2 A worked example using linear values or decibels

Multiply these Add these

1'. 10000 [W] 40 [dB W] Beware mistaking k W for watts

R!ax 8.0 x 10 16 [m4] 169.0 [dB m 4]