Radar Signal Processing and Its Applications

273 Pages·2003·8.438 MB·English

by  Dale E. Nelson, Janusz A. Starzyk (auth.), Jian Li, Robert Hummel, Petre Stoica, Edmund G. Zelnio (eds.)

RADAR SIGNAL PROCESSING
AND ITS APPLICATIONS
edited by
Jian Li Petre Stoica
University of Florida Uppsala University
Robert Hummel Edmund G. Zelnio
Defense Advance Research AFRLISNA
Projects Agency
A Special Issue of
MULTIDIMENSIONAL SYSTEMS
AND SIGNAL PROCESSING
An International Journal
Volume 14: Nos. 1-3 (2003)
Springer Science+Business Media, LLC
MULTIDIMENSIONAL
SYSTEMS AND SIGNAL
PROCESSING
An International Journal
Volume 14, Nos. 112/3, January-July 2003
Special Issue: Radar Signal Processing and Its Applications
Guest Editors: Jian Li, Robert Hummel, Petre Stoica
and Edmund G. Zelnio
Editorial ...................................................... N. K. Bose 5
Guest Editorial .......... Jian Li, Robert Hummel, Petre Stoica and Edmund G. Zelnio 7
Wavelet Transformation and Signal Discrimination for HRR Radar Target
Recognition ................. Dale E. Nelson, Janusz A. Starzyk and D. David Ensley 9
2D HRR Radar Data Modeling and Processing ............................... .
· ..................................... Junshui Ma, Xun Du and Stanley C. Ahalt 25
Detection and Analysis of Anisotropic Scattering in SAR Data .................... .
· ........................... .A ndrew J. Kim, John W Fisher III and Alan S. Willsky 49
SAR Image Superresolution via 2-D Adaptive Extrapolation ...................... .
· .......................... Alejandro E. Brito, Shiu H. Chan and Sergio D. Cabrera 83
Multi-Channel Multi-Variate Equalizer Design ................................ .
· ........................................ Ravikiran Rajagopal and Lee Potter 105
Signal Processing for Large Bandwidth and Long Duration Waveform SAR .......... .
· ......................... . Zhiping Lin, Yonghong Zeng, Guoan Bi and Jocelyn Yeo 119
Target -Centered Models and Information-Theoretic Segmentation for Automatic
Target Recognition ................... Michael D. DeVore and Joseph A. O'Sullivan 139
Extraction of Three-Dimensional Motion and Geometric Invariants from Range
Dependent Signals ............. Mark A. Stuff, Pedro Sanchez and Martin Biancalana 161
A Wide-Band Approach to the Absolute Phase Retrieval in SAR Interferometry ....... .
· ...................................... N. Veneziani, F. Bovenga and A. Refice 183
Scattering-Based Tomography for HRR and SAR Prediction ...................... .
· ........................................ B. S. Denney and R. J. P. de Figueiredo 207
An Algorithm to Detect the Presence of 3D Target Motion from ISAR Data .......... .
· .......................................... Junfei Li, Hao Ling and Victor Chen 223
Experimental Evaluation of Adaptive Beamforming Methods and Interference
Models for High Frequency Over-the-Horizon Radar Systems ..................... .
· ..................................... G. A. Fabrizio, D. A. Gray and M. D. Turley 241
Contributing Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
ISBN 978-1-4419-5345-2 ISBN 978-1-4757-6342-3 (eBook)
DOI 10.1007/978-1-4757-6342-3
Library of Congress Cataloging-in-Publication Data
Radar Signal Processing and Its Applications / edited by Jian Li ... [et al.].
p.cm.
Reprinted from a special issue of MULTIDIMENSIONAL SYSTEMS AND
SIGNAL PROCESSING, an intemationaljoumal, Volume 14: Nos. 1-3;
January-July 2003.
Copyright © 2003 by Springer Science+Business Media New York
Originally published by Kluwer Academic Publishers in 2003
Softcover reprint ofthe hardcover Ist edition 2003
AII rights reserved. No part of this publication may be reproduced, stored in a
retrieval system or transmitted in any form or by any means, mechanical, photo
copying, recording, or otherwise, without the prior written permis sion of the
publisher, Springer Science+Business Media, LLC.
Permission for books published in Europe: [e-mail]
Permissions for books published in the United States of America: [e-mail]
Printed an acid-frec paper.
~ Multidimensional Systems and Signal Processing, 14,5,2003
.,.. © 2003 Kluwer Academic Publishers.
Editorial
On behalf of the readers, editors, associate editors and editorial board members of this
journal, I wish to express my appreciation and thanks to the four guest editors who
organized this comprehensive special issue on the challenging problems in radar that are
being tackled using technical devices originating in one and multidimensional signal
processing. The topics spanned are extensive and pertinent as summarized in the guest
editorial and detailed in the papers by the authors. We look forward to receiving comments
by the readers following perusal.
There was considerable emphasis on multidimensional systems and signal processing,
highlighted by a mini-symposium on the subject, at the Mathematical Theory of
Networks and Systems (MTNS 2002) biennial symposium held in August 2002 at the
University of Notre Dame in South Bend, Indiana. The symposium provided a meeting
ground of old as well as new members of the multidimensional systems and signal
processing community. The attention of the readers is also directed to the June 2002
special issue on multidimensional signals and systems in the IEEE Transactions on
Circuits and Systems I: Fundamental Theory and Applications. A future special issue on
super-resolution image reconstruction is being planned for appearance in IEEE Signal
Processing Magazine.
In the previous editorial I had mentioned that Marwan Simaan, one of our co-editors,
was moving to Auburn University in Alabama. I am now told that for family reasons, he
has decided to stay at the University of Pittsburgh. Finally, I wish to thank Dr. Chalie
Charoenlarpnopparut from the Sirindhorn International Institute of Technology, Thamma
sart University, Thailand, for accepting my invitation to become an associate editor of our
journal. He will be replacing Sanjit K. Mitra who served this journal creditably for many
years and expressed a desire to be relieved of his duties at an opportune time. My thanks to
Sanjit for his advice and help over the years.
N. K. Bose
Editor-in-Chief
Multidimensional Systems and Signal Processing, 14,7-8,2003
2003 Kluwer Academic Publishers.
Guest Editorial
Radar has been around since World War II. As the technology develops, radar has been
playing an increasingly more significant role in both military and civilian applications,
especially due to its all-weather and day-and-night capabilities. As the role and goal of
radar expand, many challenges emerge. The purpose of this special issue is to provide an
overview of how signal processing methods can help tackle these challenges. It is evident
from the papers that we have included into this special issue that radar is a fertile ground
for signal processing applications. We will briefly discuss below, in a topic-wise manner,
the papers included into this special issue.
As the radar bandwidth increases, the scattering centers of a target can be resolved to
various degrees depending on the radar bandwidth. One may wonder how to use high
range resolution (HRR) radar for automatic target recognition (ATR), especially for
moving target identification. Wavelet transformation has been considered for selecting
discriminant target features in (Nelson, Starzyk, and Ensley). For target scatterer extraction
from HRR radar, data modelling and parameter estimation are investigated in (Ma, Du,
and Ahalt).
Synthetic aperture radar (SAR) imaging is becoming increasingly more sophisticated. As
we extend the length of the synthetic aperture for better cross-range resolution and expand
the radar bandwidth for improved range resolution, aspect and frequency dependencies of
the scattering objects must be addressed. The detection and analysis of anisotropic
scattering are studied in (Kim, Fisher, and Willsky). Other interesting SAR related topics
include two-dimensional adaptive extrapolation for improved resolution (Brito, Chan, and
Cabrera), polarimetric calibration (Rajagopal and Potter), large bandwidth and long
duration waveform SAR (Lin, Zeng, Bi, and Yeo), and ATR using SAR imagery in
(DeVore and O'Sullivan).
Naturally we would like to use radar to collect as much information as possible,
including the three-dimensional (3-D) information. Topics related to 3-D information
retrieval and modelling include 3-D feature extraction of rigid moving body (Stuff,
Sanchez, and Biancalana), absolute phase retrieval for SAR interferometry (Veneziani,
Bovenga, and Refice), 3-D scattering model for HRR and SAR prediction (Denney and
Figueiredo), and detection of 3-D target motion from ISAR data (Li, Ling, and Chen).
Finally, adaptive beamforming is known to have better resolution and much better
interference rejection capability than the standard data-independent beamformers such as
the delay-and-sum beamformer. Beamforming is needed for high-frequency over-the
horizon (OTH) radar systems, which suffer from multipath and scattering and propagation
uncertainties. Various adaptive beamforming schemes are evaluated with respect to their
interference cancellation performances when applied to OTH radar systems in (Fabrizio,
Gray, and Turley).
8 GUEST EDITORIAL
Acknowledgments
We thank the authors who submitted papers to this special issue and the reviewers who
spent their time to evaluate these papers. We also thank Ms. Jennifer Evans for initiating
the development of this special issue and Dr. N. K. Bose for encouraging us along the way.
Finally, we gratefully acknowledge the assistance of Ms. Michelle Misner and Ms. Melissa
Sullivan of the Editorial Office.
Jian Li
Robert Hummel
Petre Stoica
Edmund G. Zelnio
.... Multidimensional Systems and Signal Processing, 14,9-24,2003
~ © 2003 Kluwer Academic Publishers.
Wavelet Transformation and Signal Discrimination
for HRR Radar Target Recognition
DALE E. NELSON [e-mail]
Air Force Research Laboratory, 2241 Avionics Circle, AFR1ISNAT, Building 620 Rm C2S69, Wright Patterson
AFB, OH 45433-7321
JANUSZ A. STARZYK [e-mail]
Department of Electrical Engineering and Computer Science, Ohio University, Stocker Center #347, Athens,
OH 45701
D. DAVID ENSLEY [e-mail]
United States Air Force, WR-ALCILUJE, 226 Cochran St, Robins AFB GA 31098-1622
Received November 7, 2000; Revised November 7, 2000; Accepted October 5, 2001
Abstract. This paper explores the use of wavelets to improve the selection of discriminant features in the
target recognition problem using High Range Resolution (HRR) radar signals in an air to air scenario. We show
that there is statistically no difference between four different wavelet families in extracting discriminatory
features. Since similar results can be obtained from any of the four wavelet families and wavelets within the
families, the simplest wavelet (Haar) should be used. We further show that a simple box classifier can be
constructed from the extracted features and that any feature that classifies four or less training signals can be
removed from the classifier without a statistically significant difference in the classifier performance. We use
the box classifier to select the 128 most salient pseudo range bins and then apply the wavelet transform to this
reduced set of bins. We show that by iteratively applying this approach, classifier performance is improved. The
number of times the feature reduction and transformation can be performed while producing improved classifier
performance is small and the transformed features are shown to quickly cause the performance to approach
an asymptote.
Key Words: rough sets, wavelets, automatic target recognition, high range resolution radar, feature selection
1. Introduction
Most of the work in HRR target recognition has been done by or sponsored by the military.
The approaches taken by various researchers as summarized by [8] appear to ignore the
benefits that can be gained by proper transformations of the input signal. The wavelet
transform is a new tool which has been used in image compression and more recently in
target recognition. When wavelet transforms are used for image compression the most
important goal is to minimize the loss of information. In ATR the most important objective
is to be able to separate the various target classes [7]. Some researchers have explored
the use of wavelets to provide a richer feature space [2], [3], [4], [7], [9], [12], [13].
However there is little evidence of widespread use of this technique. Mitchell himself
explored transformations but he limited the analysis to an autoregressive approach to clean
up (remove low information data) from the signature.
10 D. E. NELSON, J. A. STARZYK AND D. D. ENSLEY
Famili states that preprocessing the data "... conditions the input data to allow
easier subsequent feature extraction and increased resolution." [5]. In the past,
engineers have used transforms such as the Fourier transform to move the signal
from a time base to a frequency base [14]. Although this is useful for some
applications, target recognition of HRR signals improved only a little under this
transform. Wavelets bring a new tool to HRR signals classification. The benefits of
using wavelets, according to Strang, are related to a local character of those trans
formations " ...n ew transforms are much more local. An event stays connected to the
time when it occurs ....a time-frequency description" [11]. Researchers that have used
wavelets for target recognition (especially for HRR) have found that the original
feature space can be augmented by the wavelet coefficients and will yield a smaller set
of more robust features in the final classifier [9], [13]. In addition to computational
savings [4], investigators have also found that wavelet methods can improve radar
performance (Pcc) [12], [13]. However, as pointed out by Stirman [13] even with
improvement in Pcc there can be a bias of the wavelets toward one or two classes to
the detriment of others.
In considering wavelets for ATR, serious consideration must be given to the selection
of a wavelet family and a wavelet in the family. Lu explored this issue in the context of
image coders [15]. In his paper, Lu compared two wavelets, one from the Biorthogonal
family (B97) and the other from the Daubechies family (DS). Using two different
metrics, Lu observed a slight advantage of the B97 versus the DS. In this paper, using
the criterion of improving the probability of correct classification (Pcc), we show that
there is no statistical advantage of one family (out of four) over any other family. Any
difference in performance which can be observed in a particular application is due to the
statistical nature of normal variations in the data. Stirman, using wavelets for ATR,
explored the use of different wavelets from the Daubechies family, and found that results
were similar among the three wavelets [13]. In this paper we show that here is no
statistical advantage of one wavelet in a family over another in the same family, thus
generalizing Stirman's observation.
Once the input data is transformed, the process of feature selection for the given type
of classifier must begin. A very popular approach uses a quadratic classifier [S]. The
quadratic classifier uses statistics of the signal to be classified and compares them to
the statistics of a template for the various target classes. This method is fraught with
problems since it uses the entire signal and thus tries to match noise to noise. In an
effort to get around this problem, Mitchell [S] uses an autoregressive filter to remove
noise and then uses the filter to help select important range bins for classification. It is
not the purpose of this paper to explore the development of a classifier. However, in
order to have a means to test the usefulness of the data transforms, we must have a
classifier to test the performance. In order to reduce the problem of the quadratic
classifier, we have chosen to use the simple generalized box classifier [1], [10] from
which to evaluate results. Our main objective was to determine which, if any, family of
wavelets provided the best feature set for a classifier. A secondary objective was to
determine if further wavelet transformations would produce even better classification
results. This required the use of a method for down selecting features from which
WAVELET TRANSFORMATION AND SIGNAL DISCRIMINATION 11
to perform further wavelet analysis. In this paper, using wavelet transformations, we
will show:
1) wavelets are useful for HRR ATR,
2) wavelet coefficients as features improve classifier performance,
3) what family of wavelets are best,
4) what wavelet in a family is best, and
5) how to mitigate or eliminate wavelet bias towards some target classes.
2. Signal Characterization
When constructing a classifier, the designer is often able to rely on intuition to select
the best features to choose among the target classes. This works when the sensor used
is a "literal" sensor. That is, the sensor provides an output similar to what the human
senses are used to dealing with. When the sensor cannot do this, then automated means
must be used to select the best features. This paper uses High Range Resolution (HRR)
radar signals. A HRR signal is an n-dimensional vector x = (aI, a2, ... , an), where
ai E {O, 1, ... , 255}. The HRR radar provides a I-D picture of what the sensor is looking
at. HRR signals are particularly hard to use for target recognition, partly because the 3-D
world is projected into just one dimension. When this is done, there are many ambiguities
created which must be resolved. A further complication is that when HRR data is plotted
as signal strength vs. range bin, the resulting graph is almost impossible for a human to use
for target recognition, mostly because it is an image we are not used to interpreting. A
better representation would be to present the HRR signal as an audio signal (similar to
sonar) because humans recognize this kind of 1-D signal better. In fact, Szu points out that
the human auditory system uses wavelets [7].
A further complication to target identification using HRR is that the signals change
considerably with only a small change in azimuth and elevation. This is illustrated in
Figure 1.
The signals in Figure 1 are from two different targets. The signals shown for each target
were taken at 200 msec intervals. Their significant variations in a short time span illustrates
how difficult it would be to construct a target identification system based on these signals.
3. Wavelet Transforms
Wavelet transforms have been found useful in a variety of applications. This is because
they provide the analyst with an approximation of the signal and a detail of the signal as
well. This helps to identity small anomalies which might be useful. For a complete
description of wavelet analysis the reader should refer to [14] and [11]. A brief summary
of how the wavelets were used is presented here.
The I-D discrete wavelet transform (DWT) of a signal yields an approximation and a
detail of the original signal. Passing the original signal through a low-pass filter then down