Sie sind auf Seite 1von 7

Chapter C-XI-1

Practical Radar Waveform Optimization Techniques for Colored Noise Mitigation

Matthew A. Ferrara, Jameson S. Bergin, Paul M. Techau

I. ABSTRACT to the interference. Previous noise mitigation techniques result


This paper presents practical radar waveform design algo- in waveforms with high signal-to-interference-plus-noise
rithms for colored noise mitigation. One algorithm uses an ratios (SINRs), but relatively poor waveform properties
iterative least squares approach to form a hybrid waveform (specifically, low compressed pulse resolutions and dynamic
that simultaneously maximizes SINR and minimizes peak amplitude modula). This paper concentrates on the problem of
sidelobe levels under the constraint of constant signal improving these waveform properties while maintaining
amplitude modulus. A second algorithm employs a simple favorable SINRs.
modification to the traditional Costas coded signal that III. BACKGROUND
increases SINR while maintaining a constant modulus and its
ideal ambiguity function. Another algorithm makes use of four We assume the additive colored noise (ACN) that distorts the
peak-sidelobe preserving properties inherent to biphase codes received signal is wide-sense stationary (w.s.s.). Given this
to choose the biphase modulated signal having the smallest assumption it is well-known that the SINR is maximized when
Euclidean distance from the SINR-optimal signal. These the received signal is first whitened to cancel the ACN and
algorithms are tested in a variety of interference scenarios, then detected using a filter, hw(t), matched to the 'whitened'
including wide band and narrow band interference sources as waveform [4]. That is, we have
well as realistic terrain scattered interference settings. Findings
from these tests are presented in terms of SINR gain over an
ordinary linear frequency modulated pulse and autocorrelation where * denotes linear convolution and ST (t) is the transmitted
sidelobe suppression. signal. In [3], [5], [6], it is shown that the SINR in this
situation is given by:
II. INTRODUCTION
1 iT2 Isw(t)12dt
SINR = 2"""
Waveform optimization is a technique that can be used to (Yw T1
adjust any or all of the radar transmitter degrees of freedom
where (Y~ is the power spectral density (PSD) of the whitened
(DoFs) based on observations of a dynamic environment
ACN and [T 1, T 2J is the observation interval.
caused by intentional and/or unintentional interference
sources. These DoFs include but are not limited to pulse shape, In a discrete-time application in which ST is a complex valued
pulse repetition frequency (PRF), bandwidth, and operating vector having N elements (i.e., the signal is sampled at N
frequency. While adaptivity on receive has been exploited for equally spaced points in time), we have that the whitening
quite some time [I], [2] waveform optimization provides the filter (to cancel the additive noise vector, n) is Hw = R~1/2,
promise of the development of a fully adaptive radar. Recent where Rn = E{ nn t} (t denotes the conjugate transpose). Thus
work has resulted in various techniques for applying this we have that Sw = HtST, and Equation (2) becomes:
optimization [3].
IIswl12 s~HwHtST S~R~lST
This paper addresses the use of radar waveform optimization SINR = -2- = 2 2 (3)
(Yw (Yw (Yw
in radio frequency interference (RFI) scenarios that include
either partial band interference or terrain-scattered interference
(hot clutter). Terrain-scattered interference can induce From Equation (3) we see that the discrete-time signal that
significant coverage losses. Without waveform optimization, maximizes the SINR is the least dominant eigenvector of the
the only solution may be to "burn through" which will impact interference (fast-time temporal) covariance matrix, Rn.
the radar time line. Waveform optimization can potentially
provide an alternative to this by matching to the target and Although the signal produced strictly through the
anti-matching eigenanalysis of Rn yields the optimal SINR, it does
3)
5)
4) -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15
0.2 0.25
time (ms)

onto the complex unit circle to attain constant modulus (CM).


Amplitude Modulus In Time At this point the SINR is tested, and if the SINR is
unsatisfactory, the new CM waveform is used as the input
waveform in the next iteration.

The basic algorithm is shown in Figure 2 and uses the


following notations:
1)s is a Nxl complex vector, where N is equal to the
number of time samples during the pulse.
2)B is the collection of M least dominant eigenvectors of
Fig. I. The autocorrelations and amplitude modulus of an the interference covariance matrix. Here, N>M and B forms
optimal SINR a M-dimensional basis.
waveform.
6)SCM is the waveform that has low correlation
not satisfy requirements for practical use such as low sidelobes.
correlation sidelobes and constant amplitude modulus. For This is of the same dimension as the vector s, but each
example, see the results from an SINR-optimal waveform in element lies on the complex unit cirle.
Figure 1. We see that the optimal waveform has very high
sidelobes and that its amplitude varies significantly in time. 4)LS(s,B) is the least squares approximation of the vector s
using the vectors in B.
[3], [6], [7], [8] consider an approach that addresses this issue
with the following observation (similar to the one originally 5)pes) is the result (another Nxl complex vector) of
contrived in [9]): given the eigenspectrum of Rn, {uo, Ul,"" element-wise multiplication of normalization factors
UN} (with the associated set of eigenvalues, [AD, AI, ... , AN]), against the vector s, i.e., the 2D projection of s onto the
eigenvectors associated with nearly equal eigenvalues yield complex unit circle.
nearly equal SINRs. If there are several least dominant
eigenvectors (and typically there are), then these least The stopping criterion used in the following simulations is the
dominant eigenvectors can be used to form a basis in which to difference in SINR gain over an LFM signal between
calculate a least squares approximation of a radar signal with consecutive iterates. This convergence threshold is chosen to
"good" waveform properties. be 10-4 dB. That is, when ISINR(i+l)'h iteration - SIN~'h
iteration I < 10-4 the loop terminates. Section VI
In [3], [6], [7], [8] this least squares approach yielded nearly demonstrates that the algorithm typically converges, but
optimal SINRs along with good pulse compression results. in some cases the convergence is slow. We note that the
However, the amplitude modula of the least squares fitted SINR was computed using an ideal interference
signals deviated significantly throughout the pulse. This paper covariance in all cases. Thus the results presented here
presents an iterative algorithm that normalizes each complex typically represent best case performance. In practice,
vector element [1 0] (thus guaranteeing constant amplitude where the interference covariance is not known we will
modulus) of a least squares fitted signal at the end of each be forced to use an estimate of the covariance. The
iteration. The iterative algorithm accepts any desired impact of using a sample covariance matrix within the
waveform as an input. Here, we consider common waveforms iterative procedure is an area for future work.
such as linear frequency modulated (LFM), Costas stepped-
frequency coded, and phase modulated (PM) signals. Two initial waveforms considered in this experiment are
IV. CONSTANT MODULUS WAVEFORMS: A TIME biphase and LFM signals. Their simplicity, low correlation
sidelobes, and constant amplitude modula make them good
DOMAIN ApPROACH
candidate starting waveforms. A third type of initial waveform
This algorithm extends the least squares approach to ensure is developed in the next section.
constant amplitude modulus. After the least squares
approximation, the new waveform is projected (element-wise)
7)
~ T4~
I
• n
c
• r
e
• a
s
• e
d
w
e time
l s

l

A biphase modulated (BPM) signal, SB, is transmitted at the
system's• center frequency. Given the desired pulse duration
and sample rate, the biphase signal having the lowest Only a slight modification to traditional Costas coding is

attainable peak-sidelobe (PSL) would be the one modulated by needed to increase the SINR of the Costas signal: the Costas
the minimum PSL (MPSL) code, provided that the Nyquist arrays are created by eliminating the indices that corresponded
Criterion is met. This MPSL code is chosen from a table of to frequency intervals deemed "noisy" by a predetermined
best known minimum-PSL codes (taken from [11] and [12]). noise threshold. The remaining interference-free intervals are
At this point the four PSL-preserving operations, conjugation, then re-indexed, dwell times increased, and a smaller Costas
reversal, constant multiplication, and ramping (see [11]) are code is implemented (see Figure 4 for an illustration). The
used to create a total of five candidate biphase signals. The adaptive Costas signal is denoted by 'Costas'.
biphase signal that is the best approximation to So is used and
is denoted by 'B'. [13].
If constant modulus is not necessary, an amplitude taper at
The LFM pulse used in this study has a constant chirp rate of k each Costas subpulse can be utilized to increase SINR. This is
= B;',where BW is the system bandwidth and T is the pulse because decreasing the power output near frequency jumps
duration. The LFM waveform is denoted by 'LFM'. would minimize the amount of smearing between subpulses.
V. CONSTANT MODULUS WAVEFORMS: A An example of this technique, which applies 32 dB Chebychev
amplitude attenuation at each frequency step, is included in
FREQUENCY this study. The Chebychev tapered signal is denoted by
DOMAIN ApPROACH 'CostasC'.
The algorithm from the previous section focused on the
eigenspectrum of the interference covariance matrix. Another Note that since Costas and CostasC have good autocorrelation
approach to colored noise mitigation is to analyze the properties (because they are Costas-coded signals), they
traditional power spectrum and define a noise threshold over constitute potentially good starting points for the iterative
the usable frequency range. The power spectrum can be algorithm presented in the previous section.
quickly calculated by applying a Fast Fourier Transform (FFT) VI. SIMULATION RESULTS
to the first row of the interference (temporal) covariance
A. Scenario 1: Fixed Bandwidth With Varying Offset
matrix. Once an average noise level is calculated for a set of
frequency intervals, a stepped-frequency approach can be used The case considered is a radar system with 1 MHz bandwidth
to effectively skip over the noisy/unusable intervals. This and an interference source having a 30 dB interference-tonoise
requires the need for nonlinear frequency steps, which ratio (INR) across a 150 kHz interval. The interference source
suggests that a predefined nonlinear frequency hopping was placed at center frequency then sequentially shifted until
sequence (such as a Costas sequence [13]) is applicable. This the source's center frequency was out of the effective system
type of signal has the advantage of being constant modulus. bandwidth. The interference covariance matrix is generated
from 301 samples equally spaced in time (the time between
A Costas array is a MxM matrix with the property that when samples is T = 1 j.Ls). At each stage the SINR gains (in dB)
shifted i rows and j columns and compared to its original over a standard LFM waveform (assuming both signals
position, there will be either only zero or one coincident 1 undergo a whitening filter) were recorded. The signals
between the two matrices (see Figure 3) [11]. These shifts obtained from the iterative algorithm are denoted with the
correspond to time and Doppler delays, and thus the definition subscript 'ITER,' the optimum-SINR signal is denoted by
of a Costas array produces the "thumbtack" shaped ambiguity SOPT, and the least squares approximation to the LFM
functions (AFs) for which Costas signals are well known [11], waveform for the interference scenario analyzed is denoted by
8) 50 ..
LFM ITER MHz bandwidth and the interference source has a 30 dB INR.
--- In Scenario 2 the interference source is located at the center
CostaslTER
45 - - frequency and its bandwidth is varied. The bandwidth of the
CostasC1TE interference source varies from 1 kHz to 999 kHz in 1 kHz
R --
BOPT1 increments.
TER
~
-J
:0. The interference covariance matrix is generated with the same
6 number of samples and time spacing as in Scenario 1. The
~ --OPT:
- resulting SINR gains are shown in Figure 6(a).
~ NotCM
0.5
o ---
c LFMLS: The bandwidth of the centered interference source had similar
.~ Not eM
~ - - Costas: , effects on all optimized signals. The relative gain of the
0.4 eM \- •.••.•. I. optimized signals versus the LFM signal increases as the
-- CostasC:
..m Not eM ..,. -' ••. ,/
•• " r ' "'t ,I 1" .••• interference bandwidth increases because the optimized
0:: -- LFM1TER:
0.3 eM waveforms concentrate more of the transmitted energy in
0.2 ·\~JW;..,·
BiphaselTER: unoccupied portions of the system bandwidth. The constant
eM ""-,i\~ modulus waveform CostasC1TER performed best, obtaining near
- CostaslTER: B. Scenario 2: optimal SINR gains in nearly all cases. The iterative algorithm
eM
- CostasC1TER: Fixed Offiet With improved SINR values for all input signals except the PM
eM Varying Bandwidth signals.
0.1 _
SOPT: eM As in Scenario 1, the
-- BOPT1TER: radar system has a 1
eM

the convergence criterion. Again, we see that in most cases the


number of iterations required is less than 10.
e. Scenario 3: Terrain Scattered Interference
Data from the White Sands scenario (also a 1 MHz system)
used in [8] was used to create a single interference covariance
matrix from 100 time samples (again, T = 1 j1.s).

The SINR results of each of the methods are recorded in Table


I, and the compressed pulses and amplitude modula are given
in Figures 7 and 8, respectively.
Fig. 5. SINR gain and iterations required as functions of
interference From the table, we see that CostasC1TER obtained the highest
frequency offset.
SINR of all the signals, both with constant and varying
amplitude modula. Figure 7 shows that the autocorrelation
sidelobes of the LFMLs signal were significantly increased by
the iterative procedure, and that LFM1TER has comparable
sidelobes to CostaslTER. Also, we see from Figure 7 that
In Figure 5(a) we see that of the two signals (LFMLs and although the iterative algorithm greatly improved the SINR
CostasC) with varying amplitude, LFMLs achieved the best gain for the biphase modulated signal, the good autocorrelation
SINR values. The three constant modulus signals with properties were diminished. Note that the sidelobes of the
consistently high SINR gains were LFM1TER, CostasITER, and CostasC signal were less affected by the iterative algorithm
CostasCITER. The CostasC1TER signal appears to yield the than those of the Costas signal.
most consistent results, although there are some cases where
either CostasITER and LFM1TER attain higher SINRs. It should
also be noted that while there were many cases in which VII. SUMMARY AND FUTURE WORK
BiphaselTER had the best SINR, the iterative procedure The results from the interference scenarios tested here show
demolished much of the sidelobe suppression obtained with that gains in SINR can be achieved while simultaneously
the original PM sequence. maintaining constant amplitude modulus and low range
sidelobes. The adaptive Costas array implementation appears
Figure 5(b) compares the number of iterations required to meet promising, but some questions remain unanswered: how
the convergence criterion. We see that in most cases the should the noise threshold be chosen? Also, there is a tradeoff
number of iterations required is less than 10. between SINR and peak sidelobe level (PSL) given a fixed
Figure 6(b) compares the number of iterations required to meet
9)
11)
10)
13)
12)
ba
nd
wi
dt
h.
OPT: 0
NotCM
- - - LFMLS: -2
Not CM
- - Costas: CM -4
-- CostasC:
Not CM -6
-- LFM1TER:
CM C -8
BiphaselTER D
~
: CM -- :;;
CostaslTER: CM ~
-- CostasCITER: CM 0
C -1
--BOPT:CM o -12
-- BOPTITER: eM
4 - - -OPT
-1 ,-, -, LfMLS
6
-1 --LfM'TER
--Costas
8
-2 --CostasC
-- Costas ITER
i __
CostasCITER
--sa
LF --BOPT
M1TE
R
400 - - - --
CostaslT
ER - -
CostasC
ITER
350 -- BOPT,TER
BOPT ITER \\
\
\
.\
\
C \,,\
~
250 D
g- ~
o:
.c :;
-40
~
o
1200 Co
Fig. 6. SINR gains and iterations required as functions of
interference

Waveform Gain(dB Constant Iteration Costas array is optimal in the SINR-PSL sense? Additionally,
SOPT 9.9641 No - subpulse amplitude tapering of the Costas signal improved
LFMLS 7.4191 No - both the SINR and PSL in the iterative algorithm case; is there
Se 0.5143 Yes -
See 1.6957 No - an ideal window to use, and if so, is it interference scenario
S8 0.4917 Yes - dependent?
BOPT 0.4917 Yes -
LFMITER 7.1028 Yes 57
Costas 7.0691 Yes 42 Across all experiments the iterative algorithm appeared to
CostasCIT 9.1143 Yes 36 perform best with the Chebychev weighted adaptive Costas
BiphaselTER 7.6277 Yes 5
Array as its initial signal. However, other starting waveforms
TABLE I may perform better than those considered here. Additionally,
SINR GAIN OVER LFM FOR the iterative algorithm seemed to have a detrimental effect on
SCENARIO 3 the sidelobe properties of the biphase coded signals. This
transmit time and sample rate, so how do we know which suggests that PM signals may not be suitable for this approach.
In all experiments the iterative algorithm converged. However,
convergence did not necessarily guarantee gains in SINR over
a LFM signal, and some signal choices lead to faster
convergence rates than others. In practice, sample covariance
matrices will be used to determine the convergence criterion
instead of the ideal covariance matrices used in this paper. The
impact of sample covariance errors on convergence is an area
for future work.

Future work will also consider variations on the iterative


scheme proposed here, as well as other model waveforms to be
used as input.

[I] L. E. Brennan and I. S. Reed, "Theory of adaptive radar,"


IEEE Transactions on Aerospace and Electronic Systems, vol.
9, March, 1973. [2] L. E. Brennan, J. D. Mallet, and I. S.
Reed, "Adaptive arrays in airborne MTI radar," IEEE
Transactions on Antennas and Propagation, vol. 24,
September, 1976.
[3] J. R. Guerci and S.U. Pillai, "Theory and application of
adaptive transmission (ATx) radar," Proceedings of the
Workshop on Adaptive Sensor Array Processing (ASAP), 14-
15 March 2000, Lexington, MA. [4] H. L. Van Trees,
"Detection, Estimation, and Modulation Theory: Part I," John
Wiley and Sons, Inc., New York, 1968.
[5] S. U. Pillai, H. S. Oh, D. C. Youla, and J. R. Guerci,
"Optimum transmit-receiver design in the presence of
signal-dependent interference and channel noise," IEEE
Transactions on Information Theory, vol. 46, no. 2,
March, 2000.
[6] J. S. Bergin and P.M. Techau, "An upper bound on the
performance gain of an adaptive transmitter," IEEE 2005
Radar Conference.
[7] J. S. Bergin, P. M. Techau, J. E. Don Carlos, and J. R.
Guerci, "Radar Waveform Optimization for Colored
Noise Mitigation," 2005 IEEE Radar Conference,
Alexandria, VA, 9-12 May, 2005.
[8] J. S. Bergin, P.M. Techau, and J.E. Don Carlos, "Adaptive
Transmitter Waveform: Analysis and Trade-oft's," ISL
Technical Report ISL-TR-02107, Vienna, VA, August
2002.
[9] J. R. Guerci, et aI, "Constrained optimum matched
illuminationreception radar," U.S. Patent 5,146,229,
September 8, 1992.
[10] w. C. Graustein, "Introduction to Higher Geometry," pp.
119-120, The Macmillan Company, 1933.
[II] N. Levanon and E. Mozeson, "Radar Signals," Wiley, NY
2004.
[12] M. A. Ferrara, "Near-Optimal Peak-Sidelobe Binary
Codes," submitted to Radar 2006.
[13] S. W. Golomb and H.Taylor, "Constructions and
Properties of Costas Arrays," Proceedings of the IEEE,
vol. 72, no. 9, 1143-1163, September 1984.

Das könnte Ihnen auch gefallen