THE BELL SYSTEM

TECHNICAL JOURNAL
VOLUME XXXIX JULY 1960 NUMBER 4

COPI/Tig1ll. 1960, A "",rican relepAD"" ano:! relegraph Company

The Theory and Design of Chirp Radars

By J. R. KLAUDER, A. C. PRICE,
S. DARLINGTON and W. J. ALBERSHEIM
(Manuscript received April 5, 1960)

A new radar technique has been developed that provides a solution for the
conflicting requirements of simultaneous long-range and high-resolution per-
formance in radar systems. This technique, called Chirp at Bell Telephone
Laboratories, recognizes that resolution depends on the transmitted pulse
bandwidth. A long high-duty-factor transmitted pulse, with suitable modula-
tion (linear frequency modulation in the case of Chirp), which covers a
frequency interval many times the inherent bandwidth of the envelope, is
employed. The receiver is designed to make optimum use of the additional
signal bandwidth. This paper contains many of the important analytical
methods required for the design of a Chirp radar system. The details of two
signal generation methods are considered and the resulting signal wave-
forms and power spectra are calculated. The required receiver character-
istics are derived and the receiver output waveforms are presented. The
time-bandwidth product is introduced and related to the effective increase
in the performance of Chirp systems. The concept of a matched filter is pre-
sented and used as a reference standard in receiver design. The effect of am-
plitude and phase distortion is analyzed by the method of paired echoes. One
consequence of the signal design is the presence of time side lobes on the
receiver output pulse analogous to the spatial side lobes in antenna theory.
A. method to reduce the time side lobesby weighting the pulse energy spectrum
is explained in terms of paired echoes. The weighting process is described,
and calculated pulse envelopes, weighting network characteristics and dele-
745

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 746 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

terious effects are presented. The effects of quadratic phase distortion are
analyzed and the resultant pulse envelopes are presented. The receiver re-
sponse characteristics in the presence of Doppler-shifted signals from moving
targets are examined. Schematic ambiguity diagrams are presented for cur-
rent signal designs.
FOREWORD

A recent declassification makes it possible to publish information con-

cerning an important advance in military electronics that has been pur-
sued in classified work at Bell Telephone Laboratories since 1951.
Traditionally, warfare has been a continuing competition between
armament and firepower. This competition has carried over to the elec-
tronics of warfare. In particular, radar has been hard put to keep pace
with the requirements determined by faster and higher-flying aircraft
and missiles and the various radar countermeasures that they require.
Therefore, anything that enhances the capability of radar, either in
range, range resolution or rate of acquisition has always been most wel-
come.
Simple pulsed radar is limited in range by the average power radiated,
in resolution by the pulse length and in acquisition time by the beam
width. The design of any radar involves a compromise among these
three factors-range, resolution and speed. Anything that eases this
compromise is of great interest.
During most of the war, the principal emphasis in development of
radar techniques and components was centered on increasing range and
range resolution performance by increasing transmitter peak power and
reducing pulse length. Thus, shortly after the end of the war, transmit-
ting tubes were in existence with average power capabilities in excess of
100 times that which could be used in straightforward short-pulse opera-
tion. At about this time a number of investigators realized that targets
separated in range by a distance M could be resolved by using an ap-
propriately modulated pulse provided that c/2N ;?; L1R, where c is the
velocity of propagation and L1f is the width of the frequency spectrum of
the transmitted pulse. The familiar case of short-pulse transmission
where the free-space wave train is of the order of twice the distance be-
tween targets to be resolved satisfies this condition. The important point
is that this criterion can also be satisfied with a variety of other signal
waveforms. Thus, a linearly frequency-modulated pulse with frequency
span set by the resolution required and with duration set by the energy
required in the pulse for range performance is at once suggested. By
transmitting a modulated pulse, it is possible to obtain range resolutions

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AKD DESIGN OF CHIRP RADARS 747

that are smaller than the transmitted pulse length by very large factors.
The range resolution is inherently limited to approximately the recipro-
cal of the signal bandwidth, and one can (in retrospect) imagine many
methods of pulse modulation that utilize a given handwidth as effi-
ciently as does a short pulse, but that do not involve the high peak pow-
ers of a very short pulse.
Such considerations led to early theoretical and experimental work on
various types of radar signals. Some of this work was concerned pri-
marily with frequency-modulated pulses. At Bell Telephone Laboratories
this followed ideas that were first proposed in 1947 by Darlington in
connection with waveguide transmission, and were later covered by a
patent.' This approach is also to be found in the work of Huttman,"
Sproule and Hughes," Caner,' and Dioke." * The early experimental
work at Bell Telephone Laboratories by A. F. Dietrich and O. E. De
Lange and by W. J. Albersheim in 1951 utilized a reflex klystron to
obtain the frequency-modulated or "Chirped" t transmitted signal. The
receiver was modified to obtain the resolution compatible with the trans-
mitted bandwidth.
Other work by C. C. Cutler envisaged more elaborate signals which
included frequency-modulated pulses as a special case. Analytical work
by A. W. Schelling clarified the ideas involved and clearly pointed up
the problems of "side lobes in range" which arise from pulse and spec-
trum shaping.
The fact that the pulse compression idea, so valuable to radar, grew
out of the work of a mathematician on a problem in waveguide commu-
nication is an illustration of how each branch of our complex technology
supports the other.
Fundamental work in this field has been supported since the Spring
of 1955 by the Weapons Guidance Laboratory of Wright Air Develop-
ment Division in Dayton, Ohio.
Today, the pulse compression technique is in use in a variety of very
important radar applications. Through its use, these radars transmit a
pulse with about 100 times the energy of a short pulse with equivalent
resolution and peak power. The paper that follows covers in detail the
design and analysis necessary in applying this technique, and a second
paper by Klauder? covers an aspect of this technique that is still evolv-
ing: the synthesis of a signal function to provide a specified ambiguity
diagram. It is interesting to note that here, too, the technique has been
* An investigation based on this patent is reported by Cook.·
t The appellation "Chirp" was first used by B. M. Oliver in an internal Bell
Laboratories Memorandum, "Not with a Bang, but a Chirp", in 1951.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 748 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

advanced because Klauder recognized the similarity between the proc-

esses of synthesis of a signal function to get a specified ambiguity dia-
gram and a method for the generation of quantum-mechanical wave
functions developed by E. Wigner.

I. INTRODUCTION

The specifications of modern, highly complex communication and de-

fense systems continually require more powerful and more sophisticated
radar or radar-like systems. Such radars must be capable of detecting
very small targets at greater ranges and with greater range and angular
resolution than ever before. To satisfy these demands, unusual innova-
tions in radar techniques have been required. This paper is devoted to
some of the theoretical and design considerations of one such innovation,
the so-called "Chirp" system, which has proved highly successful.
A familiar and intuitive figure of merit on the performance of a radar
is provided by the average transmitted power: the higher the transmitted
power, the greater the range of detectable targets, other things being
equal. Suppose one regards the pulse repetition frequency (PRF) as
being essentially fixed and seeks ways to profitably increase the trans-
mitted power in each pulse. Clearly, an amplification of each transmitted
signal constitutes the natural way to increase this power. However, the
usefulness of this approach is fixed by fundamental equipment limitations,
which frequently take the form of component breakdown. Furthermore,
this presents such a serious limitation that it is necessary to seek addi-
tional means of raising the transmitted power to satisfy the requirements
of modern systems.
To accomplish this it is necessary to use the only "dimension" re-
maining available: the time. One can surely increase the average power
if the duration of each pulse is increased. Again, this technique has its
limitations. First, there will be a maximum obtainable pulse width
limited typically by duty factor considerations, i.e., the fraction of time
the transmitting tubes will actually be in operation. However, there is a
second limitation dictated by the desired range resolution of the over-all
system, which determines the ability to separate multiple targets clus-
tered closely together. In numerous present and projected systems it is
an important fact that the pulse-width limitation imposed by the de-
sired range resolution may be several orders of magnitude less than the
limitation imposed by the present "state-of-the-art" in high duty-factor
tubes. Thus, high-power, high-resolution radars are, in principle, not
limited in a fundamental way by equipment breakdown. Rather, they
are affected by poor "signal design". What is required is a transmitted

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 749

signal that combines the large amplitude and long pulse width available
with existing apparatus, but retains the range resolution capabilities
inherent in a pulse of much shorter duration.
One solution to this problem was recognized by Darlington.' Although
his precise technique will not be pursued in detail, it does provide an
intuitive picture of the operation of an actual Chirp system. The follow-
ing section presents Darlington's model and a survey of the remaining
sections of the present paper.

II. DISCUSSION OF A SIMPLE CmRP MODEL; SURVEY AND SUMMARY OF

REMAINING SECTIONS

It is the large-frequency content, or bandwidth, of a short radar pulse

that accounts for its high resolution capabilities. It follows, as a conse-
quence of Fourier analysis, that a long pulse of constant carrier fre-
quency contains a narrow bandwidth and, therefore, possesses poor
resolution properties. However, the spectrum of this long signal can be
significantly broadened by introducing modulation. To utilize the trans-
mitting tubes efficiently, this modulation must take the form of fre-
quency modulation (FM). By this method one can introduce the fre-
quency-spread characteristic of a short pulse within the envelope of a
long-duration signal. The linearity of a radar system permits one to
realize the potential of this shorter pulse by a suitable phase equalization
in the radar receiver. Darlington's model for such a system is illustrated
in Fig. 1. The transmitted signal consists of a sequence of adjacent pulses
each possessing a unique carrier frequency, [« [see Fig. l(a)]. To realize
the short pulse potential, one imagines that the received signal is passed

t
w
(a)
t
c ?(
:> ..J I
t-
:i ~-'L
n.
x: T
~ a:
< 01;
.J ::ct- I
<
Z W
\!) Z
iiiL--.L_-'--;-.l..-.,....-J_....L_..L-----l
fi f2 f3 f4 r,
TIME_ FREQUENCY _

Fig. 1 - (a) Model Chirp radar signal composed of five adjacent pulses each
possessing a unique frequency. (b) Suitable delay equalizer for the signal in (a).
The output of this network is qualitatively a pulse of increased amplitude with a
pulse width of T seconds. The network is illustrated for the particular case where
f .. < 1..+ 1'

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 7.50 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

through a network possessing a delay versus frequency character, as

illustrated in Fig. l(b). Although the frequency h is the first to be re-
ceived, it is just this frequency that is delayed the longest amount by
the network. In this manner each pulse of distinct carrier frequency is
made to "wait" until the highest-frequency component arrives, where-
upon all the short pulses emerge simultaneously. Thus, following the
delay equalization, the original signal of Fig. l(a) will be compressed in
time, and, by energy conservation, the collapsed signal will be necessarily
increased in amplitude.
For several practical as well as theoretical reasons the signal chosen
for the Chirp system is characterized by a lineal' FM: illustrated, for
example, by Fig. 2. This important case consists of a rectangular en-
velope of T seconds duration [see Fig. 2(a)]. Within this envelope the

~l~f_'I j-E---------- T -----------1

1_
r---------- <b)
I I
! I
fo ~ I
I I
! I
~ I
- I-c---------- T ----------~

t 1I (el
I

I
w
~g
zt I-------f
!!t
1Il::::E
«

TIME _

Fig. 2 - (a) Ideal envelope of actual Chirp signal, of T seconds duration and
chosen to he of unit amplitude. (h) Instantaneous frequency vs, time characteris-
tic of Chirp signal i a hand of frequeneiea.A, centered atfo is linearly swept during
the pulse duration. (c) Schematic diagram of a signal having the properties indi-
cated in (a) and (h).

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 751

instantaneous frequency is modulated in a linear manner covering a

band of frequencies .:l, centered at a frequency 10 [see Fig. 2(b)]. Fig.
2(c) schematically illustrates a signal possessing a linear FM. It is
clear that this signal is a limiting form of the one illustrated in Fig.
lea). A suitable delay-equalizing network for the signal with linear FM
is given, therefore, by a limiting form of the equalizer in Fig. l(b). The
delay characteristic of this limiting network is shown in Fig. 3(a). The
response of the delay network to the rectangular pulse with linear FM
is discussed in Section 3.1 of this paper. This response envelope is given
analytically by the absolute value of .yD[(sin 'll'.:lt)/('ll'.:lt)] and is

TI
t
~
I
I
I
I
I
I
oJ
W
Cl
T
I
Cal
I
:<: I
cr
o
~ i I
--.L---+=
~
w
z
6,-=--.J
fa
FREQUENCY _

Fig. 3 - (a) Network delay vs. frequency characteristic suitable for phase
equalization of the Chirp signal in Fig. 2; ideally, this network is chosen to have
a flat loss characteristic. (b) Envelope of output response from the network in
(a); this pulse now has a pulse width about l/~ and an amplitude increase given
by VD, where D = T!:1 is called the dispersion factor.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 752 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

illustrated in Fig. 3(b). In this expression, D, called the dispersion factor,

is defined by the product Til and represents a parameter of fundamental
significance. The collapsed pulse width, T, is of the order of II Il. Conse-
quently, there has been a pulse width reduction given by the ratio
TIT~ Til = D,
and a signal amplification given by VD. Useful systems already pro-
posed will possess dispersion factors of 100, and even greater! Experi-
mental measurements already made exhibit output signals such as Fig.
3(b) for D = 120.
Along with the problem of constructing delay networks having the
characteristics shown in Fig. 3(a) there is also the problem of actually
generating a signal having the properties illustrated in Fig. 2. Two
techniques have been devised that are basically equivalent. First, an
"active" generation scheme may be used; for example, such a signal may
be obtained by suitable modulation of a voltage-tunable device. It was an
active generation technique that was used in early demonstrations of the
Chirp principle by A. F. Dietrich and O. E. DeLange and by W. J.
Albersheim in 1951. The second generation scheme, which will be called
"passive" generation, relies on using another network that will perform
the inverse operation of the network in Fig. 3(a). That is, the additional
network should generate a long FM output pulse from a very short
input pulse. Of course, the long pulse achieved in this manner may then
be amplified to achieve the desired transmission level. Both of these
important generation schemes will be considered from an analytical
point of view in this paper.
In Section 3.3 a brief review of "matched filter" theory is presented.
This theory provides a linear filtering scheme which yields the largest
obtainable signal-to-noise ratio (SIN), and serves principally as a
standard of comparison. Fortunately, there is a rather broad maximum
in the SIN, and filters based on the one illustrated in Fig. 3(a) provide
an excellent approximation to the "optimum" filter. Such approximate
filters or receivers are eminently more desirable from a practical point
of view.
In compressing the rectangular Chirp pulse, one unfortunately intro-
duces an undesirable feature, which is illustrated in Fig. 3(b): the spuri-
ous signals on both sides of the central pulse generally present unwanted
ambiguities. Curves of this shape occur as antenna patterns in the
theory of antenna design. By analogy, these spurious signals shall be
referred to as side lobes. Techniques for reducing the relative side-lobe
level have been extensively studied in antenna design. The analysis pre-
sented in Sections 3.3 and 3.4 to reduce the side lobes of the Chirp signal

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 753

will draw heavily on these antenna analogs. In particular, considerable

side-lobe reduction may be obtained by properly attenuating the fre-
quencies at the edges of the band 11. Such a reduction is accompanied
by a slight, but inevitable, increase in the pulse width.
The design of a practical radar system is not complete unless it is
accompanied by a set of acceptable equipment tolerances; Section 3.4
discusses this important problem. The basis of the analytical study in
this section is the paired echo concept introduced by MacColl. 8 Following
a review of MacColl's treatment of the problem, there is a discussion of
the effect on the collapsed pulse of phase and amplitude distortions.
Also in Section 3.4 there appears a detailed discussion of spectrum
shaping networks to reduce the side-lobe level based on Taylor's" ap-
proximation to the "ideal" Dolph-Tehebycheff signal. To present a clear
and intuitive picture, Taylor's model is analyzed with the help of paired-
echo theory. Effects on the pulse width and side-lobe level are presented,
as well as a discussion of the influence of Taylor weighting on the SIN.
The final section of this paper, Section 3.5, is devoted to the unusual
effects produced by targets moving in a radial direction with respect to
the radar. As is well known, this relative motion gives rise to a small
frequency shift of the returning signal j this phenomena is known as the
Doppler effect. The principle qualitative effect of this frequency shift
may be obtained by referring to Fig. 3(a), which shows the delay versus
frequency characteristic of the receiver network. One sees from this
figure that a uniform shift of the returning spectrum upwards (or down-
wards) in frequency will produce a collapsed pulse that will lead (or lag)
the unaffected signal's output time. These effects will be discussed in
terms of ambiguity diagrams introduced by Woodward. 10

III. DETAILED ANALYSIS

The practical realization of any radar, including the Chirp radar,

involves an enormous amount of complex equipment. It is gratifying
that analysis of the radar need not include the bulk of these details.
Much of the following analysis, therefore, deals with highly simplified
ideal transmission and reception processes. However, one must always
be sure that the ideal process can be essentially realized, at least in
principle. This is not the case for the frequently made replacement of
the true transmitted signal, which of course must be real, by a signal
that is a complex function of time. The validity of considering complex
signals is a simple consequence of the assumed linearity of the system.
However, complex signals are useful, not only because the system is
linear but also because they permit important simplifications regarding

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 754 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

the assumed receiving network characteristics to be made. In particular,

the analysis in Section 3.1 will be concerned only with the positive fre-
quency portion of signals and with networks having desired characteris-
tics in this frequency range. The true signal output is obtained as usual
by taking, say, the real part of the complex response.
The initial part of the analysis is devoted to studying the rectangular
signal with linear FM. Crudely speaking, this may be regarded as
assuming an "active" generation scheme; i.e., the problem of obtaining
the signal illustrated in Fig. 2 is left to the engineer. However, this is
not an accurate statement because, from an analytical point of view,
consideration of the long rectangular signal in no way prejudges its
means of generation. The discussion on the very important "passive"
technique to generate the long-duration signal can then draw on the
preceding analysis regarding the collapse process of a long FM pulse.
Furthermore, there are two related reasons why an intrinsic study of the
rectangular signal has special significance: (a) the rectangular envelope
represents the most efficient use of transmitting tubes (as remarked in
Section I), and (b) even the signals passively generated, which, for eco-
onomical tube operation are again subjected to amplitude-limiting, seem
to approach the characteristics of the ideal rectangular signal.

3.1 Analytical Discussion of the Reception and Collapse of the Chirp Radar
Signal
Let Cl(O denote the analytical waveform of a single pulse received
from an isolated, stationary point target. This real signal is chosen for
convenience to be the real part of a complex waveform EI(O, where

El
t(
(t) -- rec Tt) C2..i(Jol+kt 2/2)• (1)

As customary, fo denotes some suitable carrier frequency. The function,

rect z, introduced by Woodward, is defined by the following relations:
rect z = 1, iflzl <!
(2)
= 0, iflzl>~.
Therefore, the received signal (l) is conveniently, but arbitarily, selected
to be of unit amplitude. The particular phase characteristic in this
signal is chosen to duplicate the linear FM properties illustrated in Fig.
2. With the phase of (1) represented by tp = 21r(fot + kN2), the in-
stantaneous signal frequency is defined, as usual, by
1 dtp
fi = 21r dt = fo + kt. (3)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY ACIID DESIGX OF CHIRP RADARS 755

Thus, during the T-second interval of the pulse, the instantaneous

frequency changes in a linear fashion from i« - kT 12 to fo + kT12
[see Fig. 2(b)]. The net frequency sweep, ~, is then the difference of
these two values, or kT. This frequency interval ~ provides a con-
venient means to introduce natural units into the Chirp radar problem.
The natural variables in which to measure time and frequency are ex-
pressed as follows:
y = I~, a "time" variable, (4a)
x = f I~, a "frequency" variable. (4b)
In conjunction with these natural variables the dimensionless product
T ~ frequently appears. This product is called the dispersion factor and
is denoted hy D:
D == T~. (5)
The complex received waveform (1) takes the following form when ex-
pressed 'in terms of the natural time and frequency variables:

()- t(Y)
Ely-recDc 2ri(zOy+y'/2D) , (6)

where Xo == fol ~ and k = ~/T have been used. This signal is now D
units of "time" in duration.
The signal illustrated in (1) or (6) possesses, by construction, the
characteristics of Fig. 2. However, the qualitative argument regarding
the instantaneous frequency does not necessarily provide an accurate
description of the signal's frequency content. Such an accurate picture
is obtained by studying the signal's Fourier transform. Let A (1) denote
the Fourier transform of an arbitrary function of time, A(t).

l:
The transform of the signal in (1) becomes
2r i t
fl(J) = El(t)e- / dt

= 1.
T/2

-T[2
e2r i [(f o- f H+ k / 2 / 2) dt.
(7a)

Following some algebraic manipulations, (7a) becomes

fl(J) = . / T e- ir (f - f o) 2/ k [Z (7b)
11
( U 2) - Z(Ul)],
2~
where Z (u) is the complex Fresnel integral:

Z(u) = C(u) + is(u) = ioU eira2/2 da. (8)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 756 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

The arguments U2 and Ul are defined by

U2 = -2(j - /0) I~ + 1;l1, (9a)

Ul = -2(j - /0) v'~ - l1 ·

v'T2 (9b)

The shape of the frequency spectrum, as would be determined, for

example, by a spectrum analyser, contains only the amplitude informa-
tion. The amplitude of the signal spectrum in (7) is obtained simply by
taking the absolute value:

I fl(j) I = 12~ I Z(U2) - Z(Ul) I

(lOa)
~ {[C(U2) - C(Ul)]2 + [S(U2) - S(ttl)]21\

where C( u) and S ( u) are defined in (8) as the real and imaginary part
of Z ( u). In establishing the relative frequency content, a normalized
form of (lOa) is used. Figs. 4, 5 and 6 illustrate various spectra obtained
by plotting
1
V2 {[C(U2) - C(Ul)]2 + [S(U2) - S(ud]21\ (lOb)

where U2 and Ul appear in (9a) and (9b). Inspection of (9a) and (9b)
1.5.---------,-------,-----,-------,-------,

on
tIJ

~
II
o
II:
ou,
lU
o
=>
t: 0.51----------1f---------+-~...._.______+----_+----_1
...J
a,
:::;;
-c

°o!-------~':----___::L.,..----__=_'=,___---~----....J
0.2 0.4 0.6 0.8 1.0
(f-fo)!A

Fig. 4 - Spectral amplitude of a rectangular Chirp signal for D = 10.125. The

shape is symmetric about the point (f - !o)/t. = O.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 757
1.5 ~----"""'----~--------''''''-------r-----.....,

....
o
; 1.0 """"~!\-J~A-'I--+-l--\--I--+--\-------1f--------+--------j
o
II:
o
II.
au
Cl
::l
j 0.5f.-----+-----+---\------,1-----+--------j
Q.
J;
-c

OL-----.,.J------,-L.,....----......,..~----:l_:_----....,J
o 0.2 0.8 1.0

Fig. 5 - Spectral amplitude of a rectangular Chirp signal for D = 60.5.

shows that the dispersion factor, D, is the only auxiliary variable present
besides the "frequency" x - Xo , i.e. (f - 10)/ Ii. The values of D chosen
for the curves in Figs. 4, 5 and 6 were selected so that V2D would be
rational. In this manner full use (no interpolation) was made of the ex-
1.5r--------,-----,--------,---------,

....
N

0
~ 1.0
II
0
II:
o
IL
au
o
::l
!:: 0.51------+-----1---\----!-----------j
...J
Q.
J;
-c

oo~----:l."._----.,.L-----...,.L---....::.....---l
0.2 0.8

Fig. 6 - Spectral amplitude of a rectangular Chirp signal for D = 120.125.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 758 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

cellent Fresnel integral tables compiled by Van Wijngaarden and

Seheen." Since Z( -u) = -Z(u), it is easily established that the spec-
trum amplitude function in (lOb) is an even function of the frequency
variable x - Xo. Thus, in Figs. 4, 5 and 6 the spectrum shape is il-
lustrated only for frequencies greater than the carrier 10 ; a symmetric
reflection provides the spectrum for frequencies 1 below 10 . Qualitatively,
these curves show that, as the dispersion factor D is increased, the
spectrum shape becomes more nearly rectangular, with a total band-
width approaching Do. For smaller D values, there appears, from Fig. 4,
to be considerable frequency content outside a band Do wide centered at
10. However, a numerical integration has shown that almost 95 per cent
of the spectral energy is contained in the band Do for D's as low as 10;
when D becomes 100, approximately 98 to 99 per cent of the energy is
confined between fo - Do/2 and 10 +
Do/2. Only for extremely large D
values will the intuitive instantaneous frequency picture yield results
for the frequency spectrum with any quantitative accuracy. In practice,
the value of fo 1 whether at radio frequency (RF) or some intermediate
frequency (IF), will always be at least several times the approximate
bandwidth Do. So long as 10 continues to be several times Do, Fourier
analysis shows that the complex radar signal in (1) contains only posi-
tive frequencies to a very high degree of accuracy. As stated previously,
simplified receiver network characteristics may be assumed when the
ideal received waveform contains only positive frequencies.
The function of the receiving network may be conveniently divided
into two parts. First, there is a phase-equalizing network whose function
is to cancel the phase distortion intentionally introduced in the Chirp
signal illustrated in Fig. 2 and described analytically above. The char-
acteristics of the phase-equalizing network are shown in Fig. 3(a). For
an input given by (1), the response of this network will have a form,
illustrated in Fig. 3(b), that possesses a multiple-side-lobe (in time)
structure. The second function of the receiving network, then, will be to
reduce the relative amplitudes of these side lobes. This important second
network function will be discussed in detail in Section 3.4.2. It will be
found that the necessary corrective networks need affect only the ampli-
tude and not the phase characteristic of the signal spectrum. Thus, there
is also a natural splitting of the network characteristics from an ana-
lytical viewpoint.
The following expression is chosen for the analytical specification of a
network admittance function possessing a linear delay characteristic:
Y(f) = e . (11 )
ilrp(f-!O)2

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 759

This network is given a uniform amplitude oharacteristie with the par-

ticular amplitude value of unity, To see that the chosen phase charac-
teristic represents the linear delay property, it is necessary to investigate
the instantaneous delay of Y(n. The instantaneous delay is propor-
tional to the phase if; = 1rp(f - fO)2 differentiated with respect to fre-
quency. Let td denote the instantaneous delay; then

(12)

It should be noted that this linear delay characteristic has been chosen
so as to give zero delay to a frequency f equal to the carrier I«. This
choice is made for analytical convenience only, and a linear phase is
frequently ignored, since it gives rise only to a uniform, distortionless
signal delay. When p = i:', the differential delay over the band 6. is
given by p6. = 6./k = T. It is this particular case that is illustrated in
Fig. 3 (a). Throughout this paper only this optimum case is considered:

(13)

The detailed response of the lossless phase equalizer to the input. Chirp
signal, (1), will now be investigated. The complex response function will
be denoted by E(t) and its Fourier transform by E(f). The true, real
response e(t) is obtained, as usual, by taking the real part of E(t):
e(t) = Re E(t), From the definition of the network admittance function,
the output spectrum is given by

(14)

A well-known result of Fourier analysis permits the response E(t) to be

defined according to
(15)

Here Y(t), the network impulse response, is simply the Fourier trans-
form of the network function in (13). The star (*) denotes the convolu-
tion operation, which is defined for two arbitrary functions as follows:

(16a)

or equivalently

fez) *g(z) L: f{v )g(z - u) dv. (Ifib )

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 760 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

It is somewhat easier to calculate the response E( t) by using (15). First,

the impulse response, Yet), must be computed:

yet) = L: Y(f)e
27f i!t
df
(17)

Vi:
This is a well-known integral, t and (17) is evaluated as

yet) = e27fWot-ktS/2l. (18)

The complex output, E(t), is obtained by forming the convolution

product of yet) and h(t) given, respectively, in (18) and (1). Thus,

E(t) = Iii:. r
•
T 2
e2r i[! o l+ k, 2 / 2- tir,
11 T L
/ k ( I-r) S/ 2]
T/2

which simplifies considerably, due to cancellations in the exponent, to

E(t) = ~~ e
T I
-=- 27f i(fot-kt S/2l T/2
-T/2
e27f'ktT dT. L

Vi:
Carrying out the elementary integral that remains,

E(t) = 1r~t sin(1rktT)e27fiC!oHtS/2l,

which, when regrouped becomes

E(t) = v'Di sin 1r~t e2riC!ot-ktS/2l. (19)
1r~t

This response assumes the following form when expressed in terms of the
natural time variable y = ~t;
_ • InD sin 1ry 2..i(xOU-y2/2Dl+i.. /4
E( Y) - v u - - e . (20)
1ry

Following Woodward's nomenclature'" it is useful to introduce

• 8111 1ry
smcy ==-- (21)
1rY
as a convenient shorthand. The functions sine y and rect x are Fourier
transforms of one another. The envelope of the complex response i8
given by the absolute value of (19) and takes the form
t This transform is number 708.0 in Campbell and Foster.P

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 701

VD Isin1rtJ.t1rtJ.t I= VD I sine tJ.t I. (22)

It is this envelope that is shown in Fig. 3 (b) : this collapsed pulse has an
amplitude increase of VD over the input pulse and a new pulse width
T'" II tJ., measured at the half-power points (3-db points).
There are two important facts to be noted regarding the complex re-
sponse signal given in (19) and (20). First, there exists a residual FM
after phase equalization. By comparing (19) with the input signal h(t)
in (1) one easily sees that the residual FM occurs at the same rate, but
is reversed in sense. Otherwise stated, if k > 0, then the frequency of
the transmitted pulse rises during the pulse period, while the instan-
taneous frequency of the collapsed signal (19) falls, but at the same
absolute rate. This effect has been observed in a laboratory model for a
low dispersion factor (D ~ 10). One would expect to observe this effect
principally for low D's, as may be seen from the following qualitative
argument. Using (20), the instantaneous "frequency" is given by
xo - yiD. Now, about 90 per cent of the signal energy is confined in the
"time" interval I y I < 1 by the envelope sine y. If attention is confined
to this region of pronounced signal amplitude, the frequency will vary
only between Xo + liD and Xo - liD. From the discussion following
uoi», the value of Xo = fol tJ. will be seen to lie generally in the range
Xo > 2. For large D, the frequency in the significant region of the out-
put signal remains constant to an accuracy of about 1/Dxo. As D be-
comes larger, deviations from a constant frequency are therefore nearly
undetectable. A slight reservation to this point will be noted in Section
3.3 but, for the most part (especially in studying the side-lobe re-
duction in Section 3.4.2), the residual FM in the response will be ignored.
The second important point to be noted regarding the output signal
(19) pertains to the envelope, sine tJ.t. Although the initial envelope,
rect (tiT), and the final one, sine tJ.t, are both functions of time, there
does appear to be a Fourier transform relation between the functional
form of these envelopes. The two points just stressed above will now be
shown to be consequences of an important result of greater generality.
Assume for the moment that, instead of (1), the following complex
signal is transmitted:
(23)
This signal differs from (1) insofar as the envelope rect (tIT) has been
replaced by an arbitrary envelope denoted by E (t). The linear FM
characteristic remains unchanged. Attention is now turned to finding the

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 7fi2 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

response of the same lossless network given before in (13) to the modified
input" e'(O. The spectrum of the new output signal is given by
(j) Y(f)E'(j)
(24a)

The terms in the brackets constitute a perfect square, so that

E"(j) = L: E(T)e(rilk)((J-/oHrrJ! dr. (24b)

By applying another Fourier transform, the output signal e" (t) is

genera ted:

E"(t) = L: L: E( T )e2ri/l+ri/k(U-/oHT)
2
df dr. (25)

The integration over f is similar to one carried out previously, and (25)
becomes

(26)

The remaining integral in (26) essentially defines the Fourier transform

of E (t), i':10 that OUl' final expression for e" (t) becomes
e" (t) = Vki E( _kOehiC/ol-kl2/2). (27)

This result shows that the two points stressed above are special cases of
an exceedingly general result. That is, the output response in (27)
possesses the same lineal' FM property as the input signal but with the
reversed sense, and the Fourier-transform functional relation between
input and output envelopes is established as a general rule.
The preceding general analysis for the arbitrary envelope E(t) pro-
vides an excellent intuitive picture for the next topic to be discussed:
the generation of the long transmitted pulse by passive means.

s.z Passive Generation of the Long F ~f Transmitter Signal

In any scheme of generating the long-duration FM signal, the fore-
most objective is to secure a signal whose properties are similar to those
illustrated in Fig. 2. The passive generation means described in this
section can be no exception to this rule.
The previous section concluded with a general theorem relating the
input and output envelopes of a signal with linear FM when passed

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 763

through a lossless, linear delay equalizer. This theorem illustrates how

an input signal with a long-duration envelope becomes transformed by
the equalizer to a signal whose new envelope is much shorter in dura-
tion. In addition, implicitly contained in this theorem is the effect pro-
duced by an input signal that possesses a very shori-duration envelope.
Since the output and input envelopes are functionally connected through
the Fourier transform, the response envelope will be one of long-duration
when a narrow input signal is used. Inspection of (2:~) and (27) shows
that, if the initial envelope E(t) is chosen as sine Iii, then the envelope
ofthe equalizer output is given by Vk B( -1.'1) = Vk Ii-I red (kil Ii) =
D-! reet (tIT). The full signal, with its linear FM chnracteristic, may
then be amplified to achieve the proper transmission level.
There are two points that arise in connection with the passive gen-
eration scheme described above. The first deals with the particular
characteristics of the narrow-input signal whose envelope E(l) = sine
lit. One of the additional requirements for this signal, according to (23),
is a linear FM characteristic. In the discussion following (22) it was
shown, for all cases of practical instance, that the linear FM characteris-
tic would cause only a very slight deviation from a constant frequency.
Qualitatively, it appears that the properties of the output signal (27)
will remain essentially unchanged if the input signal contains no FM;
i.e., if (23) is replaced by e'Ct) = (sine lit)i T ifo l • Practically, it is of
course much easier to generate simply the required short envelope than
a signal with such delicate FM properties. It is the neglect of the linear
FM characteristic that is responsible for what analytical distinctions
there are in studying the passive generation scheme.
The second issue to be discussed relates to the particular sign of the
linear FM slope in the network output signal in (27). By comparing (27)
and (1) it is clear that the response signal's FM is directed oppositely to
the FM of the desired input in (1). This means that the same network
charocieristics cannot be used both to generate the long FM signal in the
transmitter and to collapse that signal in the receiver. What is clearly
needed, then, is two distinct network characteristics, one with k > 0
and the other with 1. < 0, so that the sum of the delays in these two
characteristics is a constant, independent of frequency. If one of these
characteristics is used in generating the long FM signal, then the re-
maining one represents the appropriate collapsing characteristic. The
reciprocal relation existing between these two network characteristics is
illustrated in Fig. 7.
In discussing the generation of the long FM signal by passive means
it was stated that, for practical reasons, a short, constant-frequency

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 764 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

-;r------
I
I
ca) ,--
: I
Cb)
I
I I
t t ,,
I
I I
I

I I
I
T ,
I

?( ?(
...J
I
I
I
I ...J
T
I
--.1__ IL___
W
I
I w
0 0
I

I
I
---..Y.-_ _
~---I:J.---~ fc-----I:J.----~
fo fo
FREQUENCY _ FREQUENCY -

Fig. 7 - (a) Equivalent network characteristics for the transmitter in the

passive generation scheme; a uniform amplitude characteristic is assumed along
with the phase characteristic illustrated. (b) Appropriate receiver network charac-
teristic to be used with the passive transmitter network in (a); note that the sum
of the delay introduced by the transmitter and receiver networks is a constant.

input signal is used whose envelope is given by sine t1t. The network
output envelope will no longer match that shown in Fig. 2, due to the
neglect of a small-input FM characteristic. It is important to determine
the quantitative effect that this approximation has on the network out-
put, since this is the characteristic signal transmitted by the radar. The
equivalent transmitter network characteristics are chosen in accord
with Fig. 7(a):
r'(f) = e- ir (f - / o)2/ k • (28)

The complex input to this network is denoted by q(t), where

E2(t) = (sine I:J.t)ehi/ol, (29)

and again the shorthand sine y has been used. The response of the trans-

1:
mitter network is given by

E;(t) = e2ri/IE2(J)Y'(f) df
(30)
= _1110+4/2 e2ril/l-CI-/o)2/2k) ti],
t1 /0-4/2
This result, like (7), may be expressed in terms of the complex Fresnel
integral Z ( u) defined in (8):

where the superscript star (*) denotes complex conjugate. The argu-
ments V2 and Vl are given by

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY A~D DESIG~ OF CHIRP RADARS 765

V2 = v'2 ( I~ t + ~ v'~T) , (32a)

Vl = v'2 ( I~ t - ~ ~) . (32b)
What is the relation between the desired rectangular envelope and the
true one given in (31)? Apart from a scale factor, this envelope is given
by

(33)
Notice that this envelope is of precisely the same functional form as
(lOb), which describes the amplitude of the frequency spectrum for the
rectangular Chirp signal in (1). Consequently, by merely reinterpreting
Figs. 4, 5 and 6, curves representing the transmitted signal envelope
may be obtained. To facilitate the reinterpretation the definitions of the
arguments used in calculating (lOb) are repeated here [see (9a) and
(9b)]:

U2 = I! [ -D(x - xo) + ~J,

Ul = vi! [ -D(x - xo) - ¥J.
A comparison with the new arguments in (32) for the transmitted en-
velope shows that the appropriate reinterpretation of the abscissa is
given by the transformation: (z - xo) -+ ylD = tf T: Thus, Fig. 4, for
example, shows one-half of the symmetric envelope for the case D =
10.125; the abscissa value of 0.5 is to be understood as the value of the
parameter tf T, This value of tiT marks the boundary in time of the
desired rectangular envelope. The value 1.0 on the ordinate represents
the corresponding amplitude of the ideal rectangular envelope. A similar
abscissa relabeling and reinterpretation applies to Figs. 5 and 6. These
latter two figures illustrate that, as D becomes larger, the transmitted
signal envelope approaches the rectangular envelope more closely. The
deviation from a rectangular shape is entirely due to the neglect of the
small linear FM in the input.
Mter the transmitted signal is reflected from, say, a point target, the
return signal undergoes phase equalization in the receiver. Although
Section 3.1 was devoted to the properties of a collapsed pulse, it is in-
structive to see what new shape is obtained when the transmitted en-
velope is no longer exactly rectangular. The calculation of the final

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 766 THE BELL SYSTEM TECH:-lICAL JOURNAL, JULY 1960

response signal is almost immediate. The short signal in the transmitter

is givcn in (29): f2(t) = (sine D..t)e2Trifol. The transmitter network that
spreads out this short pulse has a characteristic given in (28): Y'(f) =
e- i1rU-fo)Zlk. The receiver network characteristic to collapse the trans-
mitted signal appears in (1:3): Y(f) = eirU-fo)Zlk. The ideal receiver
output, then, is simply

E~(t) = L: Y(j)yl(f)E2(f)e211ifl £if

= L: E2(j)C211ifl df
(34)

= f2(t).
That is, the receiver equalizer output equals the short generating signal
f:
employed in the transmitter: (t) = (sine Ilt)ll1ifo I . It is to be noted
that the envelope of the response, sine D..t, achieved by passive generation
is equivalent to the envelope obtained from an "active" generation given
in (19) and (22). The only distinction in the two modes of generation
rests in the presence or absence of a small, residual linear FM. One pos-
sible consequence of this small distinction will be discussed in the next
section.

3.3 Some Signal-to-Noise Considerations in the Study of Chirp Radars

Up to this point the analysis has been confined only to uniformly
lossless delay-equalizing networks. In practice, of course, this idealization
cannot hold true, nor would one even want it to hold true. For a study
of the SIN properties of Chirp radars, additional networks must be used
which shape the spectrum by introducing some loss. Spectrum shaping
will be discussed again in Section 3.4.2 from a different point of view.

3.3.1 Maximum Signal-to-Noise and the Matched Filter

Since the presence of noise represents a degradation of signal informa-
tion, it is important that the ratio of signal to noise be made as large as
practical. Woodward" discusses several different criteria for just which
quantity should be maximized. For this paper, the peak signal power to
mean noise power ratio is selected as the definition of SIN. The noise
perturbing the ideal Chirp system is assumed to be additive and to have
the properties of so-called white gaussian noise. Such noise is (a) sta-
tistically independent at each frequency, (b) of uniform mean power
density independent of frequency and (c) gaussian in its amplitude dis-

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP R..-\.DARS 767

tribution at each frequency. With no loss of generality, the mean noise

power can be normalized to take the value one. Therefore, the total
noise output is given by the noise energy in a "typical" output noise
signal whose Fourier transform is denoted by Y(f)ii(f) , where Y(f)
denotes an arbitrary network characteristic. Using the criteria for mean
noise level, the noise power output becomes

N == J f(f)
I 1
2
I ii ( f) 1
2
df = J YU)
1
2
1 df, (35)

when expressed in terms of the arbitrary filter fCf). It is clear that

additional spectrum shaping of the linear delay network is necessary to
give meaning to the noise defined by (35).
In order to make signal-to-noise comparisons meaningful, both the
signals and the noise must be normalized. The common practice is to
normalize the signal power by dividing it by the total energy content in
2
the signal: f I f(t) 1 dt = f I E(f) [2 df, where f(t) now denotes an arhi-
trary signal chosen, for convenience, to have only positive frequencies.
The normalized peak signal power, S, is then defined by

S (maximum over t)
I JY(f)EU)e2~i!1 df 2
1

J E(f)
I
2
1 df

Finally, thc SIN is given by

(37)

where the maximum of the right-hand side occurs at t = t M j t M will in

general be a function of both E(f) and Y(f). Suppose now that Hf) is
fixed. For which filter Y(f) will the value of SIN achieve its maximum
value? This is a problem in the calculus of variations whose solution, for
present purposes, is summarized in the well-known Schwarz inequality,
which states, for two arbitrary functions f( z ) and g( z), that

If f(z)g(z) dz 1
2

(38)
J If(v) J g(w) 1 dw
~1.
2 2
1 dv I

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 768 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

The equality holds only whenf(z) is proportional to g*(z). Upon apply-

ing this theorem to (37), the maximum SIN is achieved when Y(f) is
proportional to 't*(f )e-2ri / IM • The exponential factor merely represents a
gross delay without distortion, which will be ignored. Consequently, the
r
matched filter, ",(f), for the signal E(t) is defined as
r m(f) == 't*(f). (39)
It should be noticed that the particular value of the maximum SIN is in
no way related to which specific signal and matched filter are used; any
signal coupled with its matched filter attains the same maximum value
of "one".
It is important to find the specification of the output response when a
matched filter is used. According to (39), the complex response Em(t) is
determined by
(40a)

which may be re-expressed with the aid of the convolution operator

defined in (16):
Em(t) = E*( -t)*E(t)

= L: E*(T - t)E(T) dT.

(40b)

It follows from (40b) that the complex response from a matched filter
is determined by a correlation process. As usual, the true, real response
is obtained by taking the real part of Em (t) in (40a) or (40b). The
matched filter for Chirp radars can now be studied with the aid of the
general theory just presented.

3.3.2 Matched Filter [or Chirp

3.3.2.1 Matched Filter [or the Active Generation Case. For the case of
active generation, the initial signal is E1(t) in (1), namely,

E1 ( t ) -- rec t(Tt) e 2.. iUo'+kl

l/Z)
>

where rect z is defined by (2). The matched filter for this signal, Y"'I (f),
may be readily obtained from the analysis of the spectrum of E1(t) in (7):
Yml(f) = 'tt(f)
(41)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 769

where U2 and Ul are given in (9). The matched filter impulse response is
also readily obtained by using the general relation: Y m (t) = E*( - t).
Therefore,

=rec
t)
t(1'e 21ri(fol-kl t/2)
.
(42)

The output response when the rectangular Chirp signal is passed through
its own matched filter may be computed with the aid of (40b) :

Em, (t)
= 1"" rect(r -Tree
ee
t) t(~)
T e 2.. il!ot+lk/2h
t-<k/2)(T-I)2)d
r,

When 0 ~ t ~ T,

f
T /2
Em, (t) = e2Jri (f ol- l
k l / 2) e2JrikTl dr.
I-T/2

Finally, after the remaining integration is performed,

Em, (t) = ...!-k e2.ifol sin s (ktT - kt2 ) . (43)

'll" t
It is readily determined that the envelope of Em, (t) is an even function
of time. Therefore,
2
T sin 1l"(.~ I t I - kt ) (44a)
1/"A I t I
represents the envelope of the matched filter response for times I t I ~ T.
For times I t I > T the envelope vanishes. When (44a) is expressed in
terms of the natural time variable, y = lit, the envelope becomes

_ sin e (I y I -~) (44b)

VD 'll" IY I '
where a normalization has been chosen so as to facilitate a direct com-
parison with the response envelope in (22) obtained by using the loss-
less phase equalizer [see (l3)}. The output from the simple linear delay
equalizer, VD(sin 'll"y)/'ll"Y, is a remarkable approximation to the re-
sponse from the more complicated matched filter, Ym.(f), especially as
the dispersion factor increases. The similarity between the two signals
does not mean that they possess similar values of SIN, for, as already
noted, the noise energy passed by the ideal linear delay equalizer is
infinite. The signal similarity does suggest, however, that a large SIN

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 770 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

value may be obtained by combining with the delay equalizer a shaping

network that limits the noise but does not seriously attenuate the strong
signal frequencies.
Ali an extreme example of such a shaping network, the SIN properties
will be investigated when an ideal filter. which suppresses all frequencies
outside the range If - fo I < 1l/2, is added to the delay equalizer. The
combined network is specified by

f'rect{f) = rect ~ ~ f o) eir(f-fol!/k. (45)

The total noise passed through Yrcct may be computed from the general
definition in (35):

N rect = JI - Yrcct{f) I2 df = Il. (46)

From symmetry arguments, the peak output signal value should occur
at t = O. The response at t = 0 is given by

(47a)

where U2 and Ul are linear functions of f. If natural variables are intro-

duced and a shift of the origin is made, (47a) is transformed into

Erect (0) = I~ L: {Z [ I~ (2x + I)]

(47b)
- Z [/~ (2x - I)]}dX,

where the definition of U2 and Ul is equivalent to that given in (9). The

integrals appearing in (47h) may he evaluated with the aid of the
following general formula:

I
b

..
Z(ax + (3) 1
dx = - (ab
a
+ (3)Z(ab + (3) - (aa + (3)Z(aa + (3)]

-i (ir(abt-Pl!/2
+ 'Ira eir(aa+Pl
l/2]
e - .

If this general formula is applied to the specific calculation required in

(47h), it follows that

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 771

It is simpler, and sufficient for present purposes, to confine attention to

D values that are even integers. With this restriction, (48) reduces to
froct(O) = y'WZh/2D),
and the peak signal power is given by
1 frect(O)
2
1 = 2D I Z( V2D)
2
1 • (49)
To complete the calculation for the normalized peak signal power, S,
it is necessary, according to (36), to divide (49) by f I flU) 12 dt =
f rect (tIT) dt = T. Finally, the SIN is
(50)

where Z ( u) is the complex Fresnel integral defined in (8). Any

decrease of SIN below the matched filter maximum value of "one"
represents a SIN degradation. In Fig. 8, the SIN degradation in decibels
[i.e. units of 10 loglo (SIN)} is illustrated for various values of D, the
dispersion factor. Only even integral values of D were used in calculating
this result; a smooth curve was then drawn between the calculated
points. Certainly no qualitative error is made in this process. As antici-
pated, Fig. 8 shows that, as D becomes larger, the simple filter charac-
teristics specified by Yroct approach those of the matched filter. The
extremely small SIN degradation, especially for large D, shows a clear
practical preference for characteristics like those of Yrect as compared to,
say, the complicated amplitude characteristic required in building a
matched filter. According to (39), the matched filter amplitude charac-
teristic must match the signal spectrum and, for example, would need
to resemble one of the spectra shown in Figs. 4, 5 or 6.
3.3.2.2 Matched Filter for the Passive Generation Case. Suppose, instead
of the rectangular envelope, that the transmitted waveform had an
envelope characteristic of the passive generation scheme discussed in
Section 3.2. Under ideal circumstances a short signal, (sine tJ.t)e2ri/o!, is
dispersed in the transmitter by a suitable linear delay equalizer. The
resulting transmitted signal has a spectrum proportional to

rec t(f - fo)

-6.- e
-irU-/oJ2/k
.

From the definition of a matched filter in (39), the filter r rect(f) in (45)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 772 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

----
o.1

I ~

.>

.:
I
0.2
V
~-~

0.3

/
f/I
...J
W
10

~
00.4
z
~I
-~~-

I
f/I

s
f/I

0.5 _. ._.-

t-
I

0.6

0.7 I i
o 40 so 120 160 200
DISPERSION FACTOR, D

Fig. 8 - Degradation in SIN from ideal maximum when the rectangular Chirp
signal passes through a delay equalizer and a sharp cutoff filter of width .1. The
loss decreases rapidly as the dispersion factor, D, increases.

actually represents the matched filter in the case of ideal passive genera-
tion. With this filter the response becomes simply (sine t:.t)e2fr i / o l , as
was noted in Section 3.2; now it appears that this particular network
also yields the maximum SIN value of "one".

3.3.3 Signal-to-Noise for a Shaping Network to Reduce Side Lobes

Frequently there are important advantages to be gained by trading a
small amount of SIN for certain desirable signal properties. An impor-
tant improvement in signal characteristics is obtained when the relative
side-lobe strengths are reduced, and it will be found that substantial
reduction in side lobes may be obtained at only a very small cost in SIN.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 773

Both the active and passive cases are treated simultaneously by assum-
ing the shaping-network input signal to be specified by
(sine ll.t) eh i (f ol- k l ' / 2}.

To obtain results pertaining to the passive generation, it suffices to set

k = 0, which is effectively equivalent to an infinite dispersion factor.
The specific shaping or weighting network to be studied is gaussian in
shape and given by
r,"'fllBU (I) _
- e-rO(f-/o}1 . (51)
Here, G is a parameter related to the amount of taper introduced by
Y/lBU(I). It is more convenient to use two other parameters, L and a,

which are linearly related to G:

L -- 20 1oglQe ro/1"/4
(52a)

where
(52b)
L represents the loss in decibels, imposed by YIIBU (I), at the band edges,
i.e, when f = fo ± ll./2. Equation (51) may be rewritten in terms of
x = II ll. and a = Gll. :
2

(53)
For convenience, the convolution theorem Egau(Y) YRau(Y)*E(Y), will
be expressed directly in natural units. For the input E(Y), a normalized
response of the linear delay equalizer is employed:
E(Y) = (sine y)e2ri(roU-U·/2D). (54)
The network impulse response, Ygau(Y), becomes

Y gau ()
Y =
L'"' r
'"" "'f
gau
( ) 2'1"iZll dx
X e
(55)
1 2rirou-ru"/a
= V;;;e .
If this result is combined with (54), the modified complex response is
given by

Egau()
Y = _
1j - e2rirOll L'"' (siSInez ) e -r[(u-z}2/ a+iz 2/D) dz. (56a)
va '"'

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 774 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

The integral in (560.) may be evaluated, and yields the following result
for Egau(Y):

(56b)
-{err[ 1/~ G- ~)J + err[ I~ G+ i:)]},
where c is a complex constant defined by
1 i
c = ~ + D' (57)

and "erf" denotes the usual error function, The peak signal, required

1:
for a SIN study, is attained at y = O. Since
2
sinc Y dy = 1,

the input energy in (54) has the value 1. Therefore, the peak normalized
signal power becomes

(58)

The noise power may be defined as

(59)

when expressed in terms of the natural frequency variable. If (53) and

(59) are combined, one obtains N~nu = I/V2a. From (57) and (58),
then,

(60)

In the case of ideal passive generation the effective value of D is infinity

in (54) and, therefore, also in (60). Under these circumstances, (60)
becomes

(61)

Actually this result is also quite accurate in the "active" case when D
remains finite. In practice, a typical value of L, the loss in decibels intro-
duced at the band edges, is of the order of 15-25 db; it follows from (520.)
that a ~ 3. For high-compression Chirp systems, whose D values are 100

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 775

and greater, the SIN in an active generation is well represented by (61).

Fig. 9 shows a plot of the BIN degradation for a range of different
gaussian weighting networks. It is noticed that there is a minimum loss
of about 0.5 db when L ~ 8.6 db. A detailed study of the full response
signal, (56b), has been made for the passive case; the results are approxi-
mately correct for actively generated pulses. Two gross signal features
have been studied: (a) the output pulse width measured at the 3-dh
level and (b) the relative level between the main signal peak and the
first adjacent side-lobe peak. The results of this study are summarized
in Figs. 10(a) and lOeb). Qualitatively, as the 10SH introduced at the
hand edges by the gaussian network is increased, there is (a) a cor-
responding increase in pulse width and (b) au increase in side-lobe dis-
crimination. The marked improvement in side-lobe discrimination is well
worth the slight pulse-width increase and the small cost in SIN. This
example illustrates that the maximum SIN attained only by a matched
filteI' is, in reality, a very broad maximum. This general principle is ex-
ploited in Section ;{.4.2, where the primary aim of weighting network'!
will be to improve signal properties using other networks more cfficient
than the simple gaussian one studied here.
The gaussian CUHe does provide a qualitative picture of the effect
produced by the residual FM in the active generation scheme. Suppose

0r----...,.-------,----------,

ell 1
...J
W
<0
U
w
o
~
ell
ell
3 2

3 o':-L-----:----~---__:'::__-----!.
4 8 12 18
LOSS AT BAND EDGES IN DECIBELS

Fig. 9 - Degradation in SIN from ideal maximum when the rectangular Chirp
signal passes through the delay equalizer and a smooth, gaussian-taper filter,
which introduces a loss of L db at the band edges.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 776 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

28
1.10
en (b)

f.OS
la)
./
V iJ
III
U
26 -

V
-
/~ ~ 24

1/
I:
I:; z

1/
~ 1.00 - 22 -
w /' w
c
en
./
V
:;)
.J t- 20
~ 0.95 -- f----- :::;
~-

w
> /
/ n,
~
/
f8
~ 0.90 w
V
9
.J III
W 16
a:
0.S5 1---- w //
c f4
iii V
0.80 12
o 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
LOSS AT BAND EDCOE IN DECIBELS

Fig. 10 - (a) Half-power pulse width increase vs. L, the gaussian filter band
edge loss. (b) The relative amplitude between the maximum of the first adjacent
side lobe and the central maximum of the output signal following shaping by a
gaussian filter.

one was so intent on reducing the side-lobe level that L, and conse-
quently a, was increased to a point where (61) no longer approximated
(60). In fact, consider the other extreme where a» D, so that (60) be-
comes essentially

(S/N).&u = ¥! I yQ)) r
erfG

¥!IZ(v1)12
(62)
= 2

The relative change, as compared to (61), is given by

which, for large D and consequently large a, is of the order of one.

Crudely speaking, therefore, the effect of the residual FM seems to have
no significant effect on the SIN. The modification of pulse width and
side-lobe level by the residual FM would require a detailed considera-
tion of the exact signal in (56b). This study is not warranted in view of
the many advantages inherent in passive generation.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 777

3.4 A St1ldy of Departures from I deal Chirp Systems

In this section, deviations from the ideal system behavior will be
analyzed. These deviations will include the planned departures, such
as the weighting network to reduce side lobes, and random deviations
introduced by imperfect system components. A unified approach has
been developed to predict the effects of both types of system irregulari-
ties.
The first topic, treated in Section 3.4.1, is the basic analytical tool,
"paired-echo theory", which is used in all of the analysis on system
distortions. The theory of spectrum weighting, Section 3.4.2, is then
presented with the aid of paired-echo theory. The next subject in the
sequence, Section 3.4.3, is a consideration of quadratic phase distortion
or improper equalization of the FM transmitted pulse in the receiver.
The last topic, Section 3.4.4, presents a treatment of the effects of mov-
ing targets on the collapsed-signal envelope characteristics. Although the
effect of moving targets is not strictly a system distortion, the analytical
investigation will also be presented with the aid of the paired-echo theory
developed in Section 3.4.1. For this reason, it is natural to include the
effects of moving targets in the present section.

3.4.1 Paired-Echo Concept

When an attempt is made to predict the performance of an actual
Chirp radar system that includes many elements, a very unwieldy in-
tegral is obtained, which can be solved only by numerical methods. This,
of course, gives no insight into how various distortional effects perturb
the system response. It was hoped that a solution could be obtained
that would provide a clear picture of the effects of distortion terms.
The characteristics of the Chirp signal permit the use of a previous solu-
tion to the problem of amplitude and phase distortion in linear trans-
mission systems. This solution was obtained in 1931 by MacCo1l8 of
Bell Telephone Laboratories and amplified in 1939 by Wheeler 13 and
Burrows." The solution of the distortion problem leads to a result
which is interpreted in terms of paired echos. This will be clarified in
the next section.
A linear transmission system is to be studied that has a steady-state
transfer admittance, [Yew)], defined as
Yew) = A(W)~jB("), (63)

where, as usual w = 27ft The steady-state amplitude response is A(w)

and the steady-state phase characteristic is B(w). In an ideal system,

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 778 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

A(w) would be a constant independent of frequency and B(w) would

increase linearly with frequency. The physical system will exhibit vari-
ations from the ideal behavior that can be described by a Fourier series
expansion about the frequency band of interest. This leads to a descrip-
tion of the system by the following equations:
A(w) = ao +L n
an cos new, (64)

B(w) = bow +L b; sin new. (65)

"
If all all and b; were zero except au and bu, the above equations would
describe an ideal transmission system. Any distorting influence arising
from a passive element can be described by these equations, as can
active elements operated in their linear regions. Thus, all of the elements
of a linear system can be represented in these terms.
MacColl's analysis" considers one term of the Fourier expansion given
by (64) and (65). The steady-state amplitude and phase characteristic
is given by:
A(w) = ao + al cos co, (66)
B(w) = bow + b, sin ew. (67)
Since a linear system is being considered, superposition will apply;
therefore, the resultant output for an input E(t) can be obtained as the
sum of the responses to the various terms of the Fourier series expansion.
The analysis obtains an output signal, let), for an input signal, E(t), to
a system having the characteristics given in (66) and (67):
let) = auJo(bl)E(t + bo) + J1(b1)
-[(au + ~) E(t + bo + c) - (ao - ~:) E(t + bo - c) ]

+J 2(b l) [(ao + 2b~l) su + bo + 2e) + (au - 2b~1) (68)

.s« + bu - 2c)] + Ja(b [(au + 3b~l) E(t + bu + 3c)

1)

- (au - ~~1) E (t + bu - 3e)] + ....

The functions .Iu( b. ) , .I 1 ( b) , J 2 (br) , ... are the usual Bessel func-
tions; the first four are shown graphically in Fig. 11.
The solution given in (68) provides some insight into the perturba-

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND OESIGX OF CHIRP RADARS 779

1.0
r--- r--
0.9

0.8
r--
r....... , I
Jo(b,)
..........
r-....
0.7

0.6 l.o

b,~ V
....-:J,.-
/ ~ J,(b,)
0.4
......
0.3
~~
~
./
0.2

./
/ ~ ..
0.1
V ........ ~ J 3(b1 )
o 1/
~

o 0.1 0.2 0.3 0.4 0.5 0.8 0.7 0.8 0.9 1.0 "I 1.2 1.3 1.4
b,

Fig. 11 - Bessel functions.

tions produced by distortion in a physical system. It can be seen that,

with only one sinusoidal distortion term, the output is composed of a
delayed replica of the input signal, E(t +
bo), modified in amplitude by
the coefficient, ao Jo(bd. The output also has an infinite series of terms
- echoes that occur in pairs, one preceding and one lagging the major
response - whose amplitude diminishes according to the Bessel func-
tion coefficients and whose separation from the major response is propor-
tional to the order of the Bessel function. In a well-designed system,
the coefficients u] and b, would be small. This leads to the following
approximation for the Bessel functions:
JO(b l ) = 1, (69)
J1(b l ) = !b1 , (70)
In(b 1 ) = 0, for n > 1. (71)
These approximations apply when
b1 < 0.4 radians. (72)
This leads to a considerably simpler expression for the output, l(t):

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 780 THE BELL SYSTEM TECHNICAL JOURNAL, JULy 1960

let) ,...., ao[E(t + bo) + Hadao + b1)E(t + b« + c)

(73)
+ Hadao - b1)E(t + bo - c)].
Thus it can be seen that a small sinusoidal distortion term produces a
single pair of echoes, one preceding and one lagging the maj or response of
the system. Notice also that an even function in the amplitude A (w).
i.e., a cosine ripple, gives rise to even symmetry in the echoes in the out-
put. This is to say that both amplitude echoes (ad2ao) will be of the
same polarity, either positive or negative. Similarly, an odd function of
phase gives rise to odd symmetry in the output echoes, i.e., echoes of
opposite polarity. Thus, a condition can be obtained where ad2ao is
equal to t bi and there will be r 0 echo preceding the main response. This
is a very important case and occurs in a minimum-phase network. The
phenomena, often called ringing, is familiar to anyone who has observed
the response of an unequalized low-pass filter to a very narrow pulse.
The coefficients of (73) have been calculated and are given in graphical
form in Figs. 12 and 13. It is apparent from these results that the design
of a system with low residual baseline clutter (or coherent noise) is a
difficult problem. If a 40-db baseline clutter level is a design objective,
each component, an or b; , of the Fourier series expansion of (64) and
(65) would have to be kept below the following low value for a mini-
mum phase condition:

20 loglo (1 + ::) < 0.085 db, (74)

i; < 0.57 degrees. (75)

Thus, it is seen that a 40-db clutter level is indeed extremely difficult to
attain.

3.4.2 Frequency Weighting to Reduce Side-Lobe Levels

3.4.2.1 Choice of Weighting Scheme. Before proceeding into a detailed
discussion of the weighting scheme employed in the present Chirp radar
system, a simplified qualitative picture of the mechanics of weighting
will be presented j this will permit a clearer understanding of the weight-
ing process. The paired-echo theory of the preceding section will be used
in the simplified description.
If a network with a raised-cosine amplitude response (as in Fig. 14)
and a linear phase characteristic is excited with a sine At input signal,
the output signal, shown in Fig. 15, can be derived by paired-echo theory.
The resultant output signal is the sum of a main delayed replica of the

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 781

10

~
-:
I,;'
V
V

-:
.,/

V V
-:
60
0.01
-:
0.02
/'

0.04 0.06 0.1 0.2

AMPLITUDE DEVIATION,
0.4 0.6 0.81.0
(1+ :~), IN DECIBELS
2 4

Fig. 12 - Amplitude of echo with amplitude distortion.

input signal (main response) and two echoes, one leading and one lagging
the main response. It is seen that the amplitude of the echoes is equal to
al/2ao and the time displacement of the echoes with respect to the main
response is equal to the period of the cosine ripple, 1/11. The sum of the
main response and the two echoes gives the resultant output signal, which
10

I
. /'"
! iI I

i i
/'"

~I
I
I
I
! !

V
I I

I
I
./ !
V .. I
I
:/ ! :

V I
I

i,
./
i
80 I I
0.1 0.2 0.4 0.6 0.8 1.0 2 4 6 8 10 20 40
PHASE DEVIATION, b, IN DEGREES

Fig. 13 - Amplitude of echo with phase distortion.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 782 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

0.4

0.2
OL.....,- -'- .......

fo-~ fa
FREQUENCY _

Fig. 14 - Network amplitude response.

has the desired lower side lobes at the expense of a slightly wider main
pulse and a small loss in signal-to-noise ratio over the matched filter
optimum.
The ideal weighting scheme should provide maximum attenuation of
the sidelobes with minimum broadening of the main pulse. An analogous
situation occurs in antenna theory with a line source. The antenna
designer attempts t minimize the spatial side lobes without broadening
the main lobe appreciably. As in the antenna case, Taylor weighting"
of the Chirp signal spectrum offers a good compromise between side-lobe
level and pulse-width increase.
3.4.2.2 Theory. The optimum weighting function to achieve low side
lobes with the least degradation of the pulse length is a so-called Dolph-
Tchebycheff'" function, which would give an output pulse of the form
Eo(t) = COS'lr V(M)2 - A.2. (76)
cosh 'irA
This is a signal of unit peak amplitude which has a uniform side-lobe
level, 'I = (cosh rA) -I (expressed in decibels by means of N = 20 loglo 'I).
This signal possesses infinite energy and is not physically realizable but
provides a suitable standard of comparison. The 3-db pulse width of
E o( t) is given by

{3o = ~ [CCOSh-1 '1)2 - (COSh-

1
~YJ (77)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 783

'{r\
1.0 r----,-.,..--,----,-.,..--,.---r-......- , - - - r -........-,.---r-........-,----,

1/ ~
'I \'
I

,I

II
:1
LEADING ~
:I/
I

ECHO I'V'~

-0.3 t-----+-+--t-----+-+--t---+--+--f---+--+--f---+--+----1I---1

-3 -2 -1 0 1 2 3 4
NORMALIZED TIME, llt

Fig. 15 - Weighting mechanics.

This relationship is plotted in Fig. 16 as normalized pulse length, f30 ,

versus N.
An approximation of the Dolph-Tchebycheff signal has been derived
by Taylor." When Taylor weighting functions are used, the output pulse

[It (1II-(12_::)
envelope takes the form

;-~~2t~ t)2)]
, 8
. A (7 )
E ()
flay t = SIllC ~t

where A is the same parameter that appears in (76), and

_2
2 n
o == A2+ (ii _ !)2'
(79)

The parameter, ii, will be described presently. The Taylor function,

EtlLy(t), of (78) is a compromise between the unweighted Chirp output

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 784 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960
45
.,...J
/
WEIGHTING FUNCTIONS
A -DOLPH -TCHEBYCHEFF
~ 40 f.----

l%:
B-TAYLOR (n=6)
U
l1J
C-TAYLOR (n = s)
D -GAUSSIAN ,/
""
<:>
~ 35 - '-- -

. /~ ~
IZ~ V~

~Vk?V
J"
30
V;
...J
l1J
>
UI
...J
2S

t.>~~V
UI
lD
o
...J

. / ~~ r>
UI 20
<:>
iii
~ 15
V" ~
~~
iii
UI
o
10 ~
!U. Q& 1& 1.1 1.2
t:.. t:.. t:.. b: Ii:"
NORMALIZED PULSE WIDTH

Fig. 16 - Pulse widening due to weighting.

signal envelope, sine I1t, and the Dolph-Tchebycheff signal envelope of

(76).
The spectrum-weighting function [Fourier transform of Etay(t)] is
most easily obtained by techniques similar to sampling theory. The
Fourier transform of Etay(t), Ytay(X), is zero for IXI > 0.5, where
X = (f - fo)/I1:

Etay(t) = it Y\ay
-t
(X, A, ii)e21fixtH dX. (80)

Sampling theory suggests the following substitution:

00

Y- tay (X , A ,ii ) -- "F

£..J ",e-2".i",X . (81)
-00

From this, one finds

= E F",[sinc (I1t - m)].

00

Etay(t) (82)
-00

It also follows that

F m = F(m, A, ii). (83)
The coefficients, F", , can be obtained as follows:
F« = 1, (84)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 785

O.5( _l)mH u-2m2

2) II
ii-I [ ]
r; r.; = II
(ii-I)' (
1 _ ~ ,,=1
1 - A2 + (n _ 1)2'
:.1
(85)
,,=1 n
for 0 < I m I < n. The prime in (85) means that n ;;= m; the symbol II
denotes a product of successive terms. Finally,
Fm = 0 for I m I ~ n: (86)
It follows that
ii-I

1 +2L m=1
F m cos 2'1rmX. (87)

The weighting network characteristic can be expressed in terms of the

parameters of a Chirp system:
2
+2 L
ii-1
Ytny(j, A, ii) = 1 P; cos ~ (j - fo), (88)
m=1 A
where fo is the center frequency of the IF signal if weighting is to be per-
formed at other than video frequencies.
3.4.2.3 Effect of Weighting on the Pulse Shape. In the previous section a
parameter, ii, was used but has not yet been defined; it is appropriate
to discuss it at this time. When a comparison is made of the Dolph-
Tchebycheff and Taylor pulse envelopes for the same design side-lobe
level (see Fig. 17), there are several important differences: The Taylor
pulse width, expressed at the 3-db pulse width, is
{3 = u{3o, (89)
where {3o is the Dolph-Tchebycheff 3-db pulse width given in (77) and the
quantity a is defined in (79). An important analytical difference is the
location of the zeros of the two functions. The Dolph-Tchebycheff en-
velope has all of its zeros occurring at nonintegral values of the argument
At, while the Taylor envelope has two distinct regions: a region of
nearly uniform side lobes where the zeros occur at non integral values of
the argument I At I ~ ii - 1, and a region of decaying side lobes where
the zeros always occur at integers 1 At I ~ ii. The central region of near-
uniform side lobes approximates the Dolph-Tchebycheff behavior. It is
essential to have the remote side lobes decay to keep their energy con-
tent down.
From the previous discussion it is easy to see the significance of the
parameter ii, since it gives a bound on the region of approximation to
Dolph-Tchebycheff behavior. Higher values of ii mean a closer approxi-

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 786 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960
I.

o~ - - DOLPH-TCHEBYCHEFF
- - - TAYLOR (n = 6)
o.9 \
- --- c---
I
I

0.6 -- I-

w 0.7
a
:::>
>-
:::;
~ 0.6 -
0(
.J
0(

~ 0.5
iii
a
w
N
:::; 0.4
~
«
o I
z 0.3
,
I

tfrrJtrf:
I
I

\ I: ~'- ,
0.2 0.01
\-~
I I \
I \
I \
I '
I \
,
1\ f'
~-.
I I \
f \ 'J
\: V
\

! \V V\ I V,
I I : \ I \ I \/ f
0.1 0.005
If \
~I ,
II; \
I I I I
o
I I VI o
o 2 3 4 5 6 7 6
NORMALIZED TIME, at
_ Fig. 17 - Comparison of Taylor and Dolph-Tchebycheff signal envelopes for
N = 40 db.

mation. The upper bound in the antenna case is determined by super-

gaining but, for the Chirp case, it is not desirable to use any larger value
of ii than the minimum permissible. When the minimum value of ii is
chosen, the required weighting characteristic is easily obtained with a
bandpass filter. This is evident from the results presented in Figs. 18
and 19, which are plots of the required loss functions for design side-lobe
levels of 40 and 47.5 db for various values of ii, ranging upward from the
minimum permissible value in both cases. In order to obtain the mini-
mum permissible value, Taylor employs the rule that, as ii is increased
by one, the value of 0' should not increase.
The required loss functions for side-lobe levels of 34, 37.5,40 and 47.5
db have been calculated for the minimum value of ii in each case and
are plotted in Fig. 20. The resultant pulse envelopes are shown in Figs.
21 through 24. The side-lobe levels have been amplified in some of the
figures to show the details.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AXn DESIGK OF CHIRP RADARS 787
o
r--
-2
r-..... -- -- -- ---

-4 ""'" !'..
11l
..J
w -6
"i\..
III
U
~
~
~ -8
~
Z -10
'\
« \
"1\
19
ur
-12
>
~ -14 - ---
..J
w
,j<n=
20
II:
-16 r - - t---- - -
\ ~ ...--'
¥"'4
2
-18 10 i-----
~ ~~
-20
o 0.1 0.2 0.3 0.4 0.5 0.6
(f - fo)
NORMALIZED FREQUENCY, --I!l.-

Fig. 18 - Taylor weighting characteristics for N = 40 db.

o
r-. <,
-2
I'-.
-4
-,
'\
-6

11l -8

\
..J
ur
III -10
U
w
a -12
\
~

~ -14 \
19

~ -16
\
~...J -18 f--- -- \
w
II:
-20
\
\ n=
20
-22

-24
\:J.r'4 12 I - - -
~~~
-26
o 0.1 0.2 0.3 0.4 0.5 0.6
(f-f. )
NORMALIZED FREQUENCY, ~

Fig. 19 - Taylor weighting characteristics for N = 47.5 db.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 788 THE BELL SYSTEM TECHNICAL JOURNAL, JUI.Y 1960

~~~
~
~~
!\l~\.
<'\'~
\I\~\~
1\ \'\
'\
\ ,
'~=34DB
\ n= 5
\. <, ~= 37.5DB
\ ... n=6

'" ~=40DB
\ ..... n=6

\
1\
\N=47.5DB
n=6
0.1 0.2 0.3 0.4 0.5 0.6
(f - f.)
NORMALIZED FREQUENCY'--T

Fig. 20 - Taylor weighting characteristics for various side-lobe levels.

3.4.2.4 Effect of Weighting on Signal-to-Noise Ratio. The concept of a

matched-filter radar has been explained in Section 3.3. Clearly, this is
the goal that one should strive for in the design of a radar system. If a
matched-filter receiver were designed using the theory described in
earlier parts of this paper, the output pulse would have the side lobes of
(sin z)lz, which are undesirable in many applications. The use of Taylor
weighting does not seriously effect the system detection capability when
compared to a matched-filter radar that utilizes the same transmitted
energy.
The loss in system performance of the Taylor-weighted Chirp radar
compared to a matched-filter radar is expressed most easily in terms of
the parameters F m • The SIN ratio of a Taylor-weighted signal relative
to the SIN ratio of the matched filter optimum is given by
?i-I
( S/ N ) matched = 1 +2 ""
£...JFm 2• (90)
(SIN) taylor .-1

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 789
1.0

~
\
0.9

0.8

w 0.7
o
~
:J
~ 0.6
«
..J
«
Z
o 0.5
iii
o
w
N
:J 0,4
-c
:::!
a:
o
Z 0.3

1\

\
0.2

o.,
o
o
\- 2
~ .-
345
-
NORMALIZED TIME, 61.
6
- 7 8

Fig. 21 - Taylor weighting pulse envelope for N = 34 db, ii = 5.

This expression has been evaluated (see Fig. 25) for a number of side-
lobe levels and values of the parameter N. A loss in SIN ratio of only
1.455 db is obtained when the design side-lobe level, ii, is 47.5 db below
the peak signal.

3.4.3 Tolerance of Chirp Pulses to Quadratic Phase Distortion

This section is concerned with the unwanted degradation of collapsed
Chirp pulses as a result of quadratic phase distortion, which could be the
result of an imperfect match of the transmitter and receiver delay-equal-
izer slopes. Some experimental evidence exists that heavily weighted
pulses possessing very low side lobes can tolerate more of this type of
distortion than their unweighted counterparts (sin z)jz. This section

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 790 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

1.

o.s
o~
, I
0.6 t------ r--- f

w 0.7
a
:J
I-
:;
~ 0.6
<{
..J
<{
Z
~ 0.5 1 - - - -
Ul
a
w
N
:; 0.4
<{
:::;
a:
o

-
z 0.3

_. ---- c'~~r~'-"-",,,,~,!_,~~\!:- ---. 0.02

\ rV\
0.2

'J VI \/ \1/\11\
0.1
1\ I\. 1"'\ " - 0.01

o 0
o 2 3 4 5 6 7 8
NORMALIZEO TIME, Ll.t

. Fig. 22 - Taylor weighting pulse envelope for il = 37.5 db, fi = 6.

shows analytically that unweighted pulses are degraded more than

heavily weighted pulses. The particular weighting function chosen for
this analysis is slightly less efficient than those previously described
(Section 3.4.2) but represents the Taylor-weighted optimum to a rea-
sonable approximation. It is believed that the general results of this
section are appropriate for any heavily weighted pulse.

3.4.3.1 Analysis of Quadratic Phase Distortion. The weighting scheme

described in Section 3.4.2 used a network having a transfer admittance
given by
- _ ~
Ytay(j, A, n) = 1 L: F m cos -~m
+ 2 .....1 ~
(j - fo). (91)

This network, when excited by the collapsed signal,

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESWN OF CHIRP RADARS 791

,
1.0

0.9
~ I
>-

I
I
---c-
!
I

i
o.e r-- ----
I --

w 0.7
o
....:::>
:::;
I
~ 0.6 --

<{
...J
<{
Z
<!l 0.5
iii
o
W
N
:::; 0.4
-c
:::!
a:
o
Z 0.3

\
-

\
0.2 -

\ V"~r-.
0.1 ~--~ 0.01

o o
o 2 3 4 5 6 7 8
NORMALIZED TIME, LI. t.
Fig. 23 - Taylor weighting pulse envelope for N = 40 db, n= 6.

e(t) = cos 27rfot sine ,j,t, (92)

gives rise to an output signal envelope, Etny(t), where
(n-I)

Etny(t) = 2:
m=-(n-I)
F m sine (,j,t - m). (93)

The desired weighting-network characteristic is described by (91),

but the coefficients F m decrease rapidly as m increases. In the particular
case considered in this section, N = 40 db and ii = () (the minimum
permissible value), the following ratios exist:

(94)

Thus, it can be seen that the echoes introduced in EtRy(t) hy the second
through fifth. terms in the expansion in (91) are at least an order of mag-

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 1.0

~ I

\
0.9

0.8 -- -

w 0.7
o
:::l
I-
:J
~ 0.6 !
«
...J
«
~ 0.5
Vi
o
W
N
:J 0.4
:
i
-~.

-c
~
0:
o \ I
\
Z 0.3 -
I

---

\ : I
0.2

\ ~~r<:
0.1 ---~ 0.01

o o
o 234 5 6 8
NORMALIZED TIME, Ll.t

Fig. 24 - Taylor weighting pulse envelope for IV = 47.5 db, n = 8.

'" -----
o~
I
N-200B

<,
0.2

--
UI
..J
w
!!! 0.4
U N:::;250B <,
w
o
:?; 0.6
-,
Z
<,
<, ~:::;150B
1Il 0.8
LL
o
~ 1.0
"" -,
~
o ,..N 400B
<,
~1.2
ew
o
1.4
-,
N:::; 47.5 OB
1.8 I
2 4 6 8 10 12 14 16 18 20

Fig. 25 - Signa.l-to-noise effects.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 793

nitude smaller than the echo introduced by the first term. For this
reason and also to simplify the analysis, only the first term of the ex-
pansion of (91) will be considered. Specifically, the following weighting-
network characteristic will be considered:

(95)

This network, when excited with an input pulse of the form of (92), has
the output envelope
E:ay(t) = sine (~t) + F 1 sine (~t - 1) + F 1 sino (~t + 1). (96)
A value of F 1 = 0.42 gives an output pulse in which all side lobes are 40
db or more below the peak signal. This output signal is shown in Fig. 26.
Let this pulse be subject to a quadratic phase distortion, Ifld , where

1.0

0.9
1\ -
\
f--

0.8

w 0.7
o
:::>
~
::::; 0.6
e,
~

: 0.5
4;
z
+--
I

CI 0.4
iii :
i
I
;

0.3

\
!
I

0.2

0.1
\ 'I

o \
w 0.06
..J
4;
~ 0.04
I
o
w
~ 0.02 I
~
i:S 0
o
I/"\. ~~ /"\.. """- ./
2345678 9 10
NORMALIZED TIME, bot

Fig. 26 - Quadratic phase distortion effects, weighted pulse, <p = o.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 794 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1900

(07)

This can he expressed as a linear delay distortion

(98)

It should be emphasized that the linear delay in (98) represents a

distortion over and above the planned linear delay introduced by the
Chirp receiving network in (13) .
.The distorted output-pulse envelope is found by use of the inverse
Fourier transform. Omitting constant values,

where
E dis (t) --
.:
-".to
Y-' ( ')
lay '" e
i.,'1
e
-iIT o2(.,')2JJ2 d
'" ,
r
(99)

r
1 + 2F 1 '" '
cos .1 (100)

and
(101 )
It is more convenient to express the distortion in terms of the phase
shift at the edge of the band, <1>, where
2
<I> = '11"2 To ; .1)2. (102)

This transforms (99) into

(103)

The general result of (103), apart from a constant, is expressed as

I Edi.(t) I = I C(P) - C(Q) + is(P) - is(Q)

+ F1[C(P + 71') + C(P - 71') - C(Q + 71') - C(Q - 71')] (104)

+ jF1[S(P + 71') + S(P - '11") - seQ + 71') - seQ - 71')] I,
where C and S are the Fresnel integrals defined in (8) and the arguments
are given by

1 (.1t )
P = V;' 2n + n , (105)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY AND DESIGN OF CHIRP RADARS 795

Q= J; (~: - n) , (106)

n = V2<11. (107)

Notice that the echo structure of the signal is unchanged hy the dis-
tortion process. The unweighted signal response is found very simply by
setting F 1 equal to zero.
3.4.3.2 Discussion of Results of Quadratic Phase Distortion Analysis. The
effect of quadratic phase distortion on weighted and unweighted pulses
has been calculated according to (104), and the results are shown in Figs.
26 through 36. With no weighting, the side-lobe structure is appreciably
distorted with a quadratic phase distortion as small as <II = 71'/8 = 450
(sec Fig. :H). Moreover, when <II = 271' (Fig. :36), the peak signal is
attentuated by more than 7 db, the pulse breaks up into two peaks, and
the 3-db pulse length increases tenfold. This result agrees with the rule
of thumb, long applied by antenna designers to uniformly illuminated
aperatures, that near field distortion starts at about 71'/8 < <II < 71'/4.
On the other hand, the results for quadratic distortion of a heavily
weighted pulse (Figs. 26 through 32) bear out the previous observation:
heavily weighted pulses can tolerate much more quadratic phase dis-
tortion than can unweightcd pulses. At <I> = '11", the peak attenuation is
only 1.5 db, the pulse broadening 40 per cent, and the side-lobe level
remains less than 38 db (Fig. 31). Even at <I> = 2'11", the pulse shape
exhibits no breakup. The pulse remains smooth with the peak signal
attenuated 4 db and the 3-db pulse width increased 2.3-fold; the peak
side-lobe level remains more than 36 db down.
In present-day radar systems such as Chirp, the design side-lobe level
is less than 40 db. In view of the results of this section, it is believed that
a quadratic phase distortion of <I> = 71' = 1800 can be tolerated at the
edges of the frequency band.

3.4.4 Problem of Fixed Weighting in the Reception of Doppler-Shifted

Signals
3.4.4.1 Introduction. The analysis thus far has assumed that there is
no relative motion between the radar and the target. If the target is
moving relative to the radar, the returned signal experiences a slight
shift in frequency (known as the Doppler shift) given by

v = c2v fRF, (l08)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 1.0

~ ~
0.9

0.8 \ 1.0

,~ I I I T

\ 0.9
I ~
t.".l

\
w 0.7
0
;:) 0.8~ I ttl

~
!::
...J 0.6
0-
0.7
.7 I
~ w
~ , o I1l
...J
-c
z
0.5 ~
::; 0.6
.6 L-
I
I
~
~
IL I
~ 0.4 ~
1/1 -c 0.5 .-
L ~
...J
\ ~

=
0.3 t':l

i\
Z
CI 0.4
.4
iii
0.2
I ~
&:
\ 0.3
I ...o
t"

1\
0.1
2
0.2

o I \
c::l
1.0)
~
\ .....
~~~
~t"
o ...c::l

ttmIHm
~ 0.04 .
ow O o.M
. ~

I~M:
oZ '
0.02 .....
f
x . ~
w 0 o
o I 2 3 4 5 6 7 8 9 to o I 2 3 4 5 6 7 8 9 10
NORMALIZED TIME, ~t NORMALIZED TIME, ~t

Fig. 27 - Quadratic phase distortion effects, weighted Fig. 28 - Quadratic phase distortion effects, weighted
pulse, If' = 'Ir/S. pulse, rp = ",/6.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictions apply.
 1.0

Q.9 1\

\
0.8I \
1.0
! I

0.9 I~ : l!l 0.7

\ ~
\
~
. -
t.".l
I
~ 0.6
~
'1
0.8 I- .- ::l!
< t.".l
o
I :i 0.5 :l:I

1
W
o 0.7 Z ><
"iii
:l
t: ,
,
0.4 >-
..J
-- -- Z

1
0.6
1\
Q.
:Ii
I :I
< 0.3I :I

\ :.:::
..J 0.5 .-
-e
sZ
I C/l
0.2,
Z

"iii
\
0.4
:
~
I
0.1
0.3

\ o
r-, ("'J
::z:
0.2 ;;
0.1 \ W
0.08 -e
::.l
>-
\ ....
I ..J
j e
o i ~ 0.06
>-
o
II)

la 0.04 1\ ::c
C/l

\
0.04
o
W
OW Z

-,
Z..J
~ 0.02

- -
~ ~ 0.02 )(
'\
XVI w
<; ~ .............. ~ .J'
o 10""
-8 o
W

o -2 34567 9 10 o 2 3 4 5 8 7 8 9 10
NORMALIZED TIME, .o.t NORMALIZED TIME, .o.t
-I
~
Fig. 29 - Quadratic phase distortion effects, weighted Fig. 30 - Quadratic phase distortion effects, weighted -t
pulse, '" = r/4. pulse, '" = .. /2.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictions apply.
 +
1.0

1.0 -J
0.9I to
00
0.9
0.8
\
\
0.8
w 0.7
~
,
Q

2 0.6 r--....
:::>
l!l t':I

\
0.7 I-
:::> t::I
l."J
it
I- ::l; I e-
0.6 -c
o.S \ e-
\,
.J

*
~ -c rn
O.S z -<
~
CI \ 0.4

\
~
t=J
iii 0.4
\ ..,==
\
0.3
t:l

\
0.3 t":l
0.2 :I:
Z
....
0.2
o. ~ t":l
>-
~ ....... e-
0.1 o I'- ...o
-, c::
o <; 0.08
l:l
III
III 0.06 .J ~
1\
.J ~ 0.06 _t"
<
U
III 0.04
filQ
...
1\
0.04
Q
III z \ ~
!i1
s
0.02
~ -- '-- ......
x
III 0
\ ~ ~ v--.... -' III
0.02

o
<, ~ ~
o
o 2 3 4 S 6 7 8 9- 10 a 2 3 4 5 II 7 8 9 10
NOR...... LIZED TI... E. 6t NOR...... LIZED TI ... E. 6t

Fig. 31 - Quadratic phase distortion effects, weighted Fig. 32 - Quadratic phase distortion effects, weighted
pulse, rp ~ 11". pulse, rp = 211".

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictions apply.
 1.0 1.0
I'
~
0.9I 0.9I

~
t.:l
0.8I 0.8I

,
~
::t
0.7 0.7 t.:l
w
w
o
o
o :.l
~ 0.6 E0.8 ><
:..
zo
.J
~ ~ 0.5
~ 0.5 I- -- -c
.J
«
.J o
~
Cl
0,4 ~ 0.4
~
o
iii
0.3
iii
0.3 z
o'oj
0.2
0.2 o
/\ rv \/ \ ::t

\/" \ ~
V\r. / \ r-. V"\-V-
0.1

V\ 1\V'\ V"\-I/"\.-1/
0.1
II \ I \
o o ...
:.l
o II-
o 2- 3 4 5 6
NORMALIZED TIME,
7
tot
8 9 10 2 3 4 5 S 7
NORMALIZED TIME, ~t
8 10
...
I::l

~
Fig. 33 - Quadratic phase distortion effects, unweighted Fig. 34 - Quadratic phase distortion effects, unweighted
pulse, <p = o. pulse, <p = 'Ir/8.

'l
;g

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictions apply.
 8
0.'
~
l.".l
0.11I
b:l
0.7 ~
in
~ 0.11
E ~
rt,M 0.5 ~

-
~

"
'":3

v
l.".l
~ 0,4 c
\
~ 0.4 ::>
f- ~
~
" ~
1\
iii 0.3 0.3

1\ :E
-c
:J.
~t"
...o
\/\
0.2 0.2
z
l!l ''""\. c::
\/\V\ V\ /'\l/
0.1 Li'i 0.1 ~
......... -r-, ~
o o ~~V S'
o 2 3 4 5 e 7 8 9 to o . 2- 3 4 5 6 7 e- 9- 10
NORMALIZED TIME, .o.t NORMALIZED TIME,.o.t
~
Fig. 35 - Quadratic phase distortion effects, unweighted
pulse, tp - 1r/2.
Fig. 36 - Quadratic phase distortion effects, unweighted
pulse, tp = 21r.
-~

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictions apply.
 THE THEORY AND DESIG'" OF CHIRP RADARS 801

where v is the relative radial velocity between the target and the radar, c
is the velocity of light, and fRF is the radio-frequency carrier of the
transmitted signal. Actually, each frequency in the transmitted spec-
trum will be shifted an amount proportional to that frequency; in prac-
tice, the spectrum width is usually such a very small percentage of the
RF carrier that it is assumed the entire spectrum is bodily translated an
amount v. Negligible error results from this assumption.
The radar return signal, shifted an amount v, is modulated down to
the IF range of the collapsing and weighting networks. The IF signal
resulting from a moving target has a different center frequency than
does the signal from a stationary target. Since the collapsing and
weighting networks were optimized for stationary targets, the effect of
these collapsing and weighting networks on the Doppler-shifted signals
must be calculated. The effect of the linear delay equalizer on Doppler-
shifted signals is discussed in Section 3.5. A principal result of that
analysis shows that collapsing networks may be constructed that will
not modify the collapsed pulse envelope. The present section considers
only the effect of the weighting process on Doppler-shifted signals when
the Doppler frequency shift, v, is a small fractional part of the system
bandwidth, ~.
3.4.4.2 A.nalysis. The paired-echo explanation of the weighting process
(Section 3.4.2) suggests another method of realizing the necessary
weighting charncteristic, in addition to a loss equalizer. The output of
the weighting network can be viewed as the superposition of a large
major response [delayed replica of the (sin z)jz input] pulse and a series
of ii - 1 leading echoes and ii - 1 lagging echoes. An equivalent realiza-
tion of the weighting network would be a delay line with [2(ii - 1) + 1]
taps, each with adjustable amplitude and 0 or 1800 phase shift, brought
together in a central summing network. If the tap spacing is equal to
the reciprocal of the signal bandwidth, ~, and the amplitude setting of
the mth tap is proportional to F m in the weighting network expansion
[see (91»), then the performance of the tapped delay line is identical to
the weighting network previously described. The tapped delay line
method of synthesizing a loss characteristic has been used in the past,16,17
and the resultant network is known as a transversal filter. The analysis
and results of this section are not essentially different from the work of
Bellows and Graham" on transversal filters.
As in Section 3.4.3, the weighting characteristic will be considered as

(109)

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 802 THE BELL SYSTEM TECHNICAL JOUR:'\'AL, JULY 1960

For a moving target the mean echo frequency is

fm = fo + I), (110)
where v has been defined in (108). The spectrum of the received signal
is translated and will be constant in the runge
t; - 11/2 < f < fill + 1l/2, (111)
and equal to zero elsewhere. With this fact, the output of the weighting
network is proportional to

E:a)(t) = /1.. +"'/2

1.. -6./2
[1 + 2FI cos 2: (j -
L.lo
fo)J c,"2r/1 df. (112)

An equivalent result is obtained by the passage of a signal with no

Doppler shift through a weighting network with a slightly shifted center
frequency. With this interpretation, (112) becomes
27r ]
E:'ay(t) = }
' / O+"" 2 [
1 + 2FI cos A (j - fo - v) (/2.. /1 df. (113)
10-"'/2 L.lo

The envelope of this function is given by

1 E:ay(t) 1 = I sino III + F ll sino (Ilt + 1) + F u sine (Ilt - 1)
(114)
+ iF12[ -sinc (Ilt + 1) + sino (Ilt - 1)]1,
where
F ll = F I cos (27rvl/l), (115)
and
(116)
For small values of viIl,
(117)
F 12 '" 21rvI fl.
Therefore, the real component of the side lobes remains unchanged. The
quadrature echoes hecome increasingly important as the percentage
Doppler shift, v/ Il, increases.
From the results of (114), it can be shown that Doppler shifts of less
than 5 per cent of the system bandwidth, Il, will cause very little degra-
dation of the weighted output pulse. The breakover point for serious
effects occurs for values of vi/l somewhat greater than 10 per cent. This
result has been verified by an independent calculation of the output

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY A:-.ID DESIGN OF CHIRP RADARS 803

pulse shape from a 47.5-db weighting network for a 10 per cent Doppler
shift. This calculation (Fig. 37) shows a O.2-db degrudntion of the peak
pulse amplitude and a 13-db increase in the side-lobe level to 34.5 db.

3.5 The Influence of JI[ oving Targets on Chirp Systems

This section continues the study of the effect of moving targets ini-
tiated in Section 3.4.4. Here attention is concentrated on the influence
of moving targets on the process of compression as opposed to the study
in 3.4.4 which was confined to the influence of moving targets on the
process of weighting. In the study in this section of the effect of moving
targets on a Chirp system it is initially assumed that only the linear-
delay phase equalizer is present. With this simplified receiver character-
istic, the quantitative effect of the Doppler shift may he readily found.
Inspection of the receiver network characteristics in Fig. 7(h) reveals
that the distortionless frequency translation caused by the Doppler
shift results in It corresponding time translation of the network output

1.0

0.9

0.8

0.7

l&J
o
:> 0.8
~

~
~ O.!!>
.J
~
!2 0.4
(/l

0.3

0.2 -- ~--- --- _!~~A~E~~O~£I!:!A~_~CA~~__ --- ---~ 0.04

0.1 0.02

-
'""""
~ .....
o o
o 2 4 6 8 10 12 14 18 18 20 22
NORMALIZED TIME, At

Fig. 37 - Weighting effects on Doppler-shifted signals.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 804 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

signal. According to the particular network slope illustrated in Fig. 7(b),

the response of an approaching target, with positive Doppler shift, will
emerge before the response of a hypothetical stationary target at the
same range. Likewise, the response of a receding target will lag the
hypothetical stationary target response. The amount of time lead or lag
may be determined from the following relation:
tz T
(118)
/I :;i'
where tl > 0 denotes a time lag and tz < 0 denotes a time lead. If (108)
for /I is combined with (118), it follows that

t l = _ 2vTfRF . (119)
c~

It is important to point out that t z is simply proportional to the relative

velocity of the moving target. The time shift given by (119) will give
rise to an error in the indicated range for evcry moving target. This range
error is determined hy

R = ~l = _ vTfnF (120)
• 2 ..:1'
and it has a different value depending on the particular velocity of each
individual target. Suppose, for a time interval of the order of tl , that the
velocity of each target remains essentially fixed. Because the targets are
moving, each one will traverse a range increment of liR in a time lit,
where liR = -vlit, with the minus sign signifying that approaching tar-
gets correspond to negative values of liR. If lit is chosen equal to TfRF/~,
then the range error in the signal at one time equals the range increment
incurred in a time lit = TfRF/~' Therefore, although the complete output
signal at anyone time does not represent a true distribution of the target
positions corresponding to that time, that same signal does represent a
true distribution of what the target positions will be TfRF/..:1 seconds
later. Thus, what might seem to be a complicated mixture of positive
and negative range errors can be viewed simply as an "error of inter-
pretation of the range data." (This time error is a constant of the radar
system and can easily be handled by a computer.) One can hardly over-
emphasize the importance of such an extremely simple interpretation to
unscramble the Doppler effect on the linear-FM Chirp signal!
The addition of shaping networks to reduce the side-lobe levels will not
modify the above interpretation in any qualitative manner. The return-
ing signal that has experienced a Doppler shift no longer passes through

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 Tim THEORY AND DESIG~ OF CHIRP RADARS 805

the central symmetric region of the weighting network. By forcing the

signal to pass through the filter away from the center, an over-all attenu-
ation is introduced accompanied by some modification of the signal de-
tails (see Section 3.4.4). The principal feature of concern here is the
inevitable signal attenuation and the time lead or lag caused by the
receiver delay equalizer.
A useful pictorial representation of this state of affairs is afforded by
the ambiguity diagrams of Woodward 1o (p. 118). Similar pictorial rep-
resentations have been discussed earlier by Gabor" and Ville,20 but the
interpretation that will be stressed here was first emphasized by Wood-
ward. Ambiguity diagrams are joint-response diagrams in both Doppler
frequency and response time. They are most commonly calculated for
the case of matched filters but need not be restricted to this ideal situa-
tion. Figs, 38(a), 38(b) and 38(c) illustrate, in a qualitative manner,
ambiguity diagrams for (a) a long constant-frequency signal, (b) a
short constant-frequency signal, and (c) a linear-FM Chirp signal. In
each, the matched-filter characteristics, or nearly so, are assumed. The
region of heavy shading schematically illustrates the region of strong
signal response. The long constant-frequency signal, Fig. 38(a), possesses
a good velocity resolution capability but a correspondingly poor range
resolution capability. The pattern generated by the short signal shown
in Fig. 38 (b) has the opposite characteristics; good range resolution and
poor velocity resolution. The interpretation to be placed on these am-
biguity diagrams may now be readily understood. The shaded region-
that is, the region of strong response - represents the "distribution of
ambiguity" in interpreting the response of a single point target 01' in
the ability to distinguish and identify several point targets separated
either in range or velocity or both, Fig. 38(c) shows a qualitative
ambiguity diagram for a linear-EM Chirp system. The Chirp scheme
possesses resolution ambiguities situated along an inclined axis. The in-
formation contained in a single returning Chirp signal does not provide
an unambiguous determination of both range and velocity, but repre-
sents only one quantity, which is a linear combination of these two
variables, Unless the target velocity is known a priori, it is necessary to
obtain several return signals so that a separate velocity determination
is possible. For a large class of applications, this method represents an
adequate solution to the ambiguity presented either by a Chirp signal
or by a short constant-frequency signal.
The question naturally arises as to why one should be confined to
ambiguity diagrams with the properties illustrated in Fig. 38. Why not
use a signal leading to a small, concentrated shaded region that possesses

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 806 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

fa)

t
):'
t-
o0
..J
W
2:-
>
u
z
w
::>
13
a: (b)
LL
I
1:1.

(c)

t-TIME (RANGE) -

Fig. 38 - (a) Schematic ambiguity diagram for a long constant-frequency

radar signal; the more shaded portions denote regions of greater signal response
with the time, t, and for variously moving targets with Doppler frequency II.
(b) Schematic ambiguity diagram for a short constant-frequency radar signal i
improved time resolution is secured at the expense of some discrimination against
moving targets. (c) Schematic ambiguity diagram for a typical chirp signal; an
approaching target with a positive Doppler shift has a response attenuated in
amplitude and shifted to an earlier time.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 THE THEORY Ai'lD DESIGN OF CHIRP RADARS 807

simultaneous range and velocity resolution? Although this question is

not. directly related to the study of Chirp systems, it is a fascinating
issue clearly encompassed by the spirit of the Chirp scheme. A companion
paper? considers this question in more detail and investigates an im-
portant class of ambiguity diagrams.

This paper has presented in detail the theory and design of Chirp
radars. Throughout the paper the FM characteristic intentionally im-
posed on the transmitted signal took but one form: a uniform linear
FM. Needless to say, such a simple specification as linear FM represents
just one of numerous, more complex FM schemes that might be used.
However, the simplicity of the Chirp radar would perhaps make it
easier to construct than any of the hierarchy of alternative methods.
This important feature of ease of construction increases in importance
when it is realized that it is theoretically possible to design a high-
performance linear-Fl\I Chirp radar whose efficiency is only slightly be-
low the ideal maximum; such is the result of the analysis presented in
this paper.
In summarizing the contents of this paper the following remarks can
be made:
i. Passive generation of the transmitted signal appears to be nearly
as efficient as an ideal active generation process, In addition, passive
generation is very desirable from a practical standpoint.
ii. The combination of receiver collapsing and weighting networks
provides a good approximation to a matched filter. Moreover, the de-
sirable signal property of very low side lobes is obtained by sacrificing
only a few decibels in the signal-to-noise ratio.
iii. Even in the presence of substantial Doppler shifts the output
signal maintains low side lohes. The most. significant effect on the
return signal is a time translation proportional to the velocity of the
moving target. But hy a simple data "reinterpretation" it is possible to
nullify the effect of this time translation. This significant property follows
only in the case of linear FM such as in Chirp,
iv. The design of Chirp rndars with dispersion factors of 100 is quite
reasonable; such radars are presently in the experimental stages of de-
velopment. With a dispersion factor of lOO, a threefold increase in range
is to he expected.
V.ACKNOWLEDGMENTS

The authors are indebted to C. N. Nebel for his advice and sugges-

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio
 808 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1960

tions during the course of this work. They also take pleasure in acknowl-
edging the assistance of the computing personnel and facilities of Bell
Telephone Laboratories at both Whippany and Murray Hill, N. J.

REFERENCES
1. Darlington, S., U. S. Patent No. 2,678,997, May 18, 1954.
2. Hiittman, E., German Patent No. 768,068, March 22, 1940.
3. Sproule, D. O. and Hughes, A. J., British Patent No. 604,429, July 5, 1948.
4. Cauer, W., German Patent No. 892,772, December 19, 1950.
5. Dicke, R. H., U. S. Patent No. 2,624,876, January 6, 1953.
6. Cook, C. E., Proc. LR.E., 48, 1960, p. 310.
7. Klauder, J., this issue, p. 809.
8. MacColl, L. A., unpublished manuscript.
9. Taylor, T. T., LR.E. Trans., AP·3, 1955, p. 316.
10. Woodward, P. M., Probability and Information Theoru, with Applications to
Radar, McGraw-Hili, New York, 1953.
11. Van Wijngaarden, A, and Scheen, W. L., Tables of Fresnel Integrals, N. V.
Noord-Hollandsehe Uitgevers Maatschappij, Amsterdam, 1949.
12. Campbell, G. A. and Foster, R. M., Fourier Integrals for Practical Applica-
tions, D. Van Nostrand Co., New York, 1942.
13. Wheeler, H. A., Proc. LR.E., 27, 1939, p. 359.
14. Burrow~ C. R., Proc. I.R.E., 27, 1939, p. 384.
15. Dolr.h, v. L., Proc. LR.E., 34, 1946, p. 335.
16. Kal man, H. E., Proc. I.R.E., 28, 1940, p. 302.
17. Graham, R. S. and Sperry, R. V., U. S. Patent No. 2,760, 164.
18. Bellows, B. C. and Graham, R. S., B.S.T.J., 36,1957, p. 1429.
19. Gabor, D.:J J.LE.E..:J 93, Pt. III, 1946, p. 429.
20. Ville, J., Cables & .rrans., 2,1948, p. 61.

d licensed use limited to: Universidad EAFIT. Downloaded on September 26,2020 at 14:53:50 UTC from IEEE Xplore. Restrictio