Geometric Comb Waveforms for

Reverberation Suppression
Henry Cox
Hung Lai
ORINCON Corporation
4350 North Fairfax Drive, Suite 470
Arlington, VA 22203
Abstract achieved by matched filtering and thereby limits the
The major problem for sonar systems operating in effectivenessof this approach.
shallow water is reverberation from the ocean bottom. A Doppler exploitation typically involves the use of a
new class of waveforms is described which combines high long shaded CW pulse of duration T in order to produce a
range resolution with excellent Doppler properties for reverberation spectrum with a large narrow peak at zero
detecting moving targets in stationary reverberation. The Doppler and a rapid falloff with increasing Dopplex shift.
waveforms consist of multiple simultaneously transmitted The idea is that target motion will shift the echo away
CW tone pulses. The tones are non-uniformly spaced in from the peak into a spectral region in which the
frequency to break up range ambiguities. A special reverberation power is low. The spectral or Doppler
subclass is rhe geometric comb in which the frequency resolution of the cw pulse is ~ - 1 making
, a long pulse
spacing between the adjacent components follows a desirable. However, the range resolution is proportional
geometric progression. Analyiic approximations can be to T so that good Doppler rejection of reverberation is
used to obtain the important properlies of this class, achieved at the expense of range resolution. The
resulting in a practical design methodology. The resulting reverberation spectrum of the CW pulse can be smeared by
waveforms have excellent Doppler properties and an source or receiver motion so that performance is best for
unambiguous high resolution range estimation capability. stationary. slowly moving or highly directional systems.
Techniquesfor reducing the peak-to-average power ratio The weaknesses of the long CW pulse are the large peak
are discwsed. of zero Doppler reverberation and the poor range
resolution. The combination of a wideband pulse and a
CW pulse in the same transmission is commonly used in
1 Introduction active sonar systems to take advantage of the
It is sometimes said that the three most important complementarynature of their characteristics.
problems for active sonars operating in shallow water are The goal of our new waveform is to achieve significant
reverberation, reverberation and reverberation. There are improvement over a long CW pulse for Doppler rejection
two standard signal design approaches for fighting of reverberation and for tracking. To motivate our
reverberation. The first exploits range resolution. The approach, consider a signal which is the sum of N
second exploits target Doppler to shift the echo into a simultaneously transmitted CW pulses with a uniform
valley of the reverberation spectrum. spacing s between adjacent tones. The spacing s is chosen
In the first approach, a spectrally flat wideband pulse is so that it is large compared to the Doppler shifts of
used. This spreads the reverberation power over the interest. We call this signal a uniform comb because of
bandwidth B of the pulse. When coherent matched filter its uniformly spaced spectral peaks. The reverberation
processing is used, the pulse is compressed into an spectrum would have N peaks, and near each peak there
equivalent duration of B-l. This provides a gain of 10 log would be a spectral component of the echo. The signal-
BT against reverberation over that which would be to-reverbemtion spectral ratio of each component would be
experienced for a CW pulse of the same duration T against nearly the same as for a single CW pulse. Coherent
a zero Doppler echo. The performance of the wideband addition of all these components provides a gain of 10 log
pulse is independent of the Doppler shift of the echo, N over the CW pulse. There are two severe problems
whereas the CW pulse performance increases with with the uniform comb which make it unattractive as a
increasing echo Doppler shift. Multipath time-spreading, sonar waveform, although it is useful for combating
which is usually severe in shallow water environments, fading in communication. These problems are illustrated
limits the amount of echo compression which can be in Fig. 1 which shows the signal, its spectrum and its

1185
1058-6393/95$4.00 0 1995 IEEE
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on September 10,2024 at 06:29:41 UTC from IEEE Xplore. Restrictions apply.
 autocorrelation function for a uniform Hanning shaded The autocorrelation is important because it is the range
comb with 34 components. First, the signal has large ambiguity function at zero Doppler. The expression (7)
narrow peaks separated by wide valleys resulting in a very does not depend on the geometric spacing and is
high peak-to-average power ratio. Since most sonar approximately valid for any multi-tone pulse in which
transmitters are peak power limited, with this type of T-' is much smaller than the minimum frequency
waveform one must greatly lower the average power or spacing. The peak at z = 0 has width (fN- fl)-' .
suffer peak distortion. Second. the autocorrelation The effect of non-uniform spacing is U) spread and
function has multiple peaks of nearly equal height causing reduce the ambiguous peaks of the autocorrelation
severe range ambiguity. The geometric comb waveform function of the uniform comb shown in Fig. 1. The
overcomes both of these problems while retaining good important result for geometric spacing is that the
SpeCtrd F O P X t h . ambiguous peaks are spread into plateaux with envelopes
of constant level. The properties of these plateaux were
2 The Geometric Comb derived by Chow [ 11 for the beampattems of geometrically
The geometric comb is a multi-tone pulse in which the spaced arrays. The derivation involves the Poisson sum
frequency spacings between successive components are a representation of a finite sum of cosines, used by Ishimaru
geometric progression. The basic signal is described by 121, combined with the stationary phase approximation for
N
x(t)=w(t)Ccos2nfn(t+a), O l t l T (1) evaluating the resulting integrals. The k - th ambiguity
n-1 plateau's envelope may be characterized as follows:
where w(t) is a shading function used to reduce spectral Start : k/Rs (84
sidelobes and a is a time shift parameter. All the stop : kls (8b)
components add in phase at t = - U . The frequencies f. 1

satisfy the following relationships: Relative Level : (NkIn R)- (W

f, = E, +s The basic properties of the autocorrelation function are
simply described in terms of N, R, s, and T, which greatly
fN=fN-l+r N-2 s simplifies waveform design. We have found that 1.6 is a
(2b)
good value for R. In the region where plateaux overlap,
so that r is the ratio of adjacent spacings and is assumed to
the components can sometimes add in phase giving a local
be slightly larger than one. The maximum spacing is
peak amplitude equal to the sum of the overlapping
f N - f , , =rN-'s=Rs (3) envelope amplitudes. Overlap of multiple plateaux at
where R is the ratio of the maximum to the minimum large values of 7/ s is countered by the shading function's
spacing. autocomelation, R,(r).
The series (2) may be summed to obtain the following Fig. 2 shows a Hanning shaded geometric comb
expression for f (R = 1.59,N = 34, s = 2.227, a =O)which may be
f, = f, + s(rN-'- 1) 1(r - 1) (4) directly compared with the uniform comb of Fig. 1 to see
An approximate expression for the average spacing may the effects of geometric spacing. The ambiguous peaks in
be obtained from (3) and (4) the autocorrelation function have all been suppressed by
s, = s(R - 1)I In R (5) more than 10 dB. The envelope is well described by (8).
Another useful property of the autocorrelation function is
2.1 Autocorrelation Function the deep valleys adjacent to the peak at T = 0. This
The attractiveness of the geometric comb as a signal permits the use of split window normalization in range
waveform is primarily due to its well behaved without being limited by the plateaux regions.
autocorrelation function.
1 2.2 The Peak-to-Average Power Ratio
R,(z)=-lw(t)w(t+ z) ~ c o s 2 7 r f n z +
oscillating
2 [nyl
It was noted earlier that the autocorrelation function was
essentially a shifted version of the basic waveform.
sum and differencefrequency termsldt (6) Hence, the suppression of ambiguous peaks in the
Hence for a smooth positive window function w(t) autocorrelation function also suppresses unwanted peaks
1 N
in the waveform, thereby improving the peak-to-average
R,(T)=-RR,(T)Ccos2d,T (7)
2 arl power ratio. The role of the parameter a in (1) is to shift
Comparing (7) and (l), the autocorrelation is Seen to be the peak at t = -a and the adjacent deep valleys so they
simply a shifted version of X ( I ) in which the window are not part of the transmitted waveform. For Hanning
function w(t) is replaced by its autocorrelation function weighting a shift may not be necessary since the values of
and multiplied by 0.5, the average power of a sine wave. the weighting function are very small near t = 0. These

1186

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on September 10,2024 at 06:29:41 UTC from IEEE Xplore. Restrictions apply.
 effects are easily seen in comparing the waveforms of 3 Alternative Processing Strategies
Figs. 1 and 2. The peak amplitude in the geometric comb An attractive feature of the geometric comb waveform is
is smdler by more than 10 dB when compared to the that it offers a number of alternative processing strategies.
uniform comb. These include matched filtering, incoherent sum across
In judging peak-to-average power ratio it is useful to frequency, M of N across frequency and various
define a factor F, the ratio of the peak-to-average power combinations. The reverberation spectrum will be far
ratio of the waveform of interest to that of a sine wave from white, hence there is significant potential gain by
with the same shading. For a geometric comb preceding any of these processors with an adaptive
F = (signal peak)' 1N (9) prewhitening filter. Rewhitening is also helpful for a
For the signal of Fig. 2, F is approximately 2.5. This long CW pulse.
compares with a value of 34 for the uniform comb for In conditions of s,.cre he-spread, the performance of
which F=N. the matched frlter cbo b; seriously degraded. Even in these
Examining the waveform of Fig. 2,it appears that the conditions, by inc Jherently combining frequencies the
peak-to-average power ratio could be further reduced by geometric comb can provide approximately 5 log N
passing the waveform through a limiter which clipped off improvement OVP the standard CW pulse. This compares
the peaks. This will cause a distortion of the signal to 10 log N improvement by matched filtering in non-
spectrum and partially fill in the valleys between the time-spread conditions. Thus, for a 30 component
spectral peaks. Adjusting the limiting threshold provides geometric comb we expect 7.5 to 15 dB improvement in
a tradeoff between peak-to-average power ratio and spectral signal-to-reverberation ratio over a CW pulse of the same
distortion. This effect is illustrated in Fig. 3 in which the duration.
limiting threshold is set at 7. Then F = 7' 134 = 1.44
which is remarkably close to that of a sine wave. 4 Conclusion
Limiting of the signal peaks has almost no effect on the C-...-.relric comb waveforms are very attractive for
autocorrelation function. The spectral fill-in is about 40 shal J water reverberation-limited situations. They have
dB down from the peaks. excellent Doppler projmties for reverberation suppression.
They can provide a unique high range resolution peak for
2.3 Doppler Ambiguity tracking, with range ambiguities more than 10 dB down.
The basic concept is to choose the frequency spacings They have a good peak-to-average power ratio. This can
so that they are larger than the Doppler shifts of interest. be further improved by peak limiting at the expense of
This avoids Doppler ambiguity. This may not always be some spectral distortion. Alternative processing schemes
possible so that it is useful to examine Doppler make it possible to achieve 7.5 to 15 dB gain over a CW
ambiguity. pulse depending on the severity of the multipath time-
Consider an echo from a target with radial velocity v so SIaead.
that each frequency fkis shifted by 'Lfkvlc. In the
special case in which 5 References
r = l+s/f, (10) [I] [Link], "On Grating Plateau of Nonuniformly
Spcued Arrays", IRE Trans Ant and Prop, (AP-13) pp 208-
the frequencies themselves, not just the spacings, form a
geometric progression, and 215, Mar 1965.
[2] A. Ishimaru, "Theory of Unequally-Spaced Arrays", IRE
f,+l = f f k (1 1)
Trans Ant and Prop, (AP-10)pp 691-702,Nov 1963.
In this case, the ratio of the Doppler shift to the spacing
is the same for each frequency component. That is

When 2(v / c)f, = s , all frequency components would be

simultaneously shifted to the frequency of the next higher
component. Usually we choose r > 1+ s 1f, , so that this
does not occur. The Doppler ambiguity is then reduced
and smeared over the region s I f

1187

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on September 10,2024 at 06:29:41 UTC from IEEE Xplore. Restrictions apply.
 Figure 1 Uniform Comb with Hanning:
Waveform, Spectrum, Autocorrelation
(N=34, Rz1.0, s=2.758, a = 0)

(&........... !............ !.............:............ :............ !..........

1s

.,* ........... ............(.............,..... ......................

-15
,........................................ ,............ .: ...........
0 1 2 3 4 s 6 7 I I 10

.................

T l m1-1

Figure 2 Geometric Comb with Hanning:

Waveform, Spectrum, Autocorrelation
(N=33, R=1.59, ~ = 2 . 2 2 7 , a = 0)

1188

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on September 10,2024 at 06:29:41 UTC from IEEE Xplore. Restrictions apply.
 IS
...............!................ !................!................!................!................ !................L..............

-12 ............... .................................................................. < ................ <................ <...............

-ISt0
1 2 3 4 5 6 7 8

Figure 3 Geometric Comb with Peak Limiting at 7:

Waveform, Spectrum, Autocorrelation
(N=34, R=1.59, ~=2.227, a = 0)

1189

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on September 10,2024 at 06:29:41 UTC from IEEE Xplore. Restrictions apply.