Signal Processing For Multistatic Radar Systems: Adaptive Waveform Selection, Optimal Geometries and Pseudolinear Tracking Algorithms Do Ançay

Signal processing for multistatic radar
systems: adaptive waveform selection,

optimal geometries and pseudolinear
tracking algorithms Do■ançay
Visit to download the full and correct content document:
https://ebookmass.com/product/signal-processing-for-multistatic-radar-systems-adapti
ve-waveform-selection-optimal-geometries-and-pseudolinear-tracking-algorithms-dog
ancay/
SIGNAL
PROCESSING FOR
MULTISTATIC
RADAR SYSTEMS
SIGNAL
PROCESSING FOR
MULTISTATIC
RADAR SYSTEMS
Adaptive Waveform Selection,
Optimal Geometries and
Pseudolinear Tracking
Algorithms
NGOC HUNG NGUYEN

KUTLUYIL DOĞANÇAY
Academic Press is an imprint of Elsevier
125 London Wall, London EC2Y 5AS, United Kingdom
525 B Street, Suite 1650, San Diego, CA 92101, United States
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
Copyright © 2020 Elsevier Ltd. All rights reserved.
MATLAB® is a trademark of The MathWorks, Inc. and is used with permission.

The MathWorks does not warrant the accuracy of the text or exercises in this book.
This book’s use or discussion of MATLAB® software or related products does not constitute
endorsement or sponsorship by The MathWorks of a particular pedagogical approach or particular
use of the MATLAB® software.
No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording, or any information storage and
retrieval system, without permission in writing from the publisher. Details on how to seek
permission, further information about the Publisher’s permissions policies and our arrangements
with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency,
can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the
Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience
broaden our understanding, changes in research methods, professional practices, or medical
treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in
evaluating and using any information, methods, compounds, or experiments described herein. In
using such information or methods they should be mindful of their own safety and the safety of
others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors,
assume any liability for any injury and/or damage to persons or property as a matter of products
liability, negligence or otherwise, or from any use or operation of any methods, products,
instructions, or ideas contained in the material herein.
Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library
ISBN: 978-0-12-815314-7
For information on all Academic Press publications

visit our website at https://www.elsevier.com/books-and-journals
Publisher: Mara Conner

Acquisition Editor: Tim Pitts
Editorial Project Manager: John Leonard
Production Project Manager: Kamesh Ramajogi
Designer: Matthew Limbert
Typeset by VTeX
Contents
About the Authors ix

Preface xi
List of Abbreviations and Symbols xiii
1. Introduction 1
1.1. Historical background 1
1.2. Purpose and scope 4
1.3. Outline of book 6
Part 1. Adaptive waveform selection
2. Waveform selection for multistatic tracking of a maneuvering

target 13
2.1. Introduction and system overview 13
2.2. Bistatic radar measurements 14
2.3. Bistatic ambiguity function and Cramér–Rao lower bounds 15
2.4. Target tracking 17
2.5. Adaptive waveform selection 23
2.6. Simulation examples 24
2.7. Summary 31
2.8. Appendix 31
3. Waveform selection for multistatic target tracking in clutter 33

3.2. Tracking algorithm with probabilistic data association 35
3.5. Summary 43
4. Waveform selection for multistatic target tracking with Cartesian

estimates 45
4.2. Target position and velocity estimation in Cartesian coordinates 47
4.3. Cramér–Rao lower bounds 50
4.4. Target tracking with joint selection of radar waveform and Cartesian
estimate 51
4.6. Summary 61
v
vi Contents
5. Waveform selection for distributed multistatic target tracking 63

5.2. Algorithm description 65
5.3. Communication complexity 67
5.5. Summary 72
Part 2. Optimal geometry analysis
6. Optimal geometries for multistatic target localization with one

transmitter and multiple receivers 77
6.1. Introduction and problem formulation 77
6.2. Optimal geometry analysis 80
6.3. Examples 86
6.4. Simulations 89
6.5. Summary 95
6.6. Appendices 96
7. Optimal geometries for multistatic target localization by

independent bistatic channels 99
7.1. Introduction and problem formulation 99
7.2. Optimal geometry analysis 101
7.4. Summary 108
Part 3. Pseudolinear tracking algorithms
8. Batch track estimators for multistatic target motion analysis 113

8.1. Introduction 113
8.2. Problem formulation 115
8.3. Maximum likelihood estimator and Cramér–Rao lower bound 118
8.4. Pseudolinear estimator 120
8.5. Bias compensation for pseudolinear estimator 124
8.6. Asymptotically-unbiased weighted instrumental variable estimator 127
8.7. Asymptotic efficiency analysis 130
8.8. Computational complexity 133
8.9. Algorithm performance and comparison 133
8.10. Summary 138
8.11. Appendices 139
9. Closed-form solutions for multistatic target localization with

time-difference-of-arrival measurements 143
9.1. Introduction 143
Contents vii
9.2. Maximum likelihood estimator and Cramér–Rao lower bound 146

9.3. Three-stage least-squares solution 147
9.4. Bias analysis 152
9.5. Bias compensation techniques 154
9.6. Algorithm performance and comparison 158
9.7. Summary 162
Bibliography 163
Index 169
About the Authors
Ngoc Hung Nguyen received the B.E. degree (Hons.) in electrical and
electronic engineering from the University of Adelaide, Australia, in 2012,
and the Ph.D. degree in telecommunications from the University of South
Australia, Australia, in 2016. He is currently a Researcher with the Defence
Science and Technology Group (DST Group), Australia, and an Adjunct
Research Fellow with the University of South Australia. Before joining
DST Group in 2019, he was a Research Fellow and Lecturer with the
University of South Australia. His research interests include statistical and
adaptive signal processing, compressive sensing, and estimation theory with
emphasis on target localization and tracking, sensor management, and radar
imaging. Dr. Nguyen currently serves on the Editorial Board of Digital
Signal Processing.
Kutluyıl Doğançay received the B.S. degree with honors in electrical and
electronic engineering from Boğaziçi University, Istanbul, Turkey, in 1989,
the M.Sc. degree in communications and signal processing from Imperial
College, The University of London, UK, in 1992, and the Ph.D. degree in
telecommunications engineering from The Australian National University,
Canberra, ACT, Australia, in 1996. Since November 1999, he has been
with the School of Engineering, University of South Australia, where he is
a professor and discipline leader of electrical and mechatronic engineering.
His research interests span statistical and adaptive signal processing with ap-
plications in defence and communication systems. Dr. Doğançay received
the Best Researcher Award of School of Engineering, University of South
Australia, in 2015, and Tall Poppy Science Award of the Australian Institute
of Political Science in 2005. He was the Tutorials Chair of the IEEE Sta-
tistical Signal Processing Workshop (SSP 2014), and the Signal Processing
and Communications Program Chair of the 2007 Information, Decision
and Control Conference. He serves on the Editorial Board of Signal Pro-
cessing and the EURASIP Journal on Advances in Signal Processing. From
2009–2015 he was an elected member of the Signal Processing Theory
and Methods (SPTM) Technical Committee of the IEEE Signal Process-
ing Society. He is currently an associate member of the Sensor Array and
Multichannel (SAM) Technical Committee and a member of the IEEE
ComSoc Signal Processing for Communications and Electronics Technical
Committee. Dr. Doğançay is the EURASIP liaison for Australia.
ix
Preface
Radar is an electromagnetic sensor for detecting and ranging objects by

radiating electromagnetic waves to illuminate the scene of interest and
receiving reflected echoes from the objects. Since its invention at the be-
ginning of the 20th century, radar has played a fundamental and prominent
role in a wide range of applications in both the civilian and military do-
mains thanks to its capability to operate under all weather conditions, at
day and night, and to detect, track, and image targets with high accuracy
at long stand-off ranges. Multistatic radar is a particular type of radar that
incorporates multiple spatially distributed transmitters and receivers. The
spatial diversity provided by the separation of transmitters and receivers not
only offers several advantages over conventional monostatic radar, but it
also gives system designers extra degrees of freedom to optimize the radar
system design for specific applications. Moreover, great progress has been
made toward solving the complexity issues of multistatic radar, thus making
multistatic radar systems perfectly feasible in real-world situations.
The conventional definition of the ambiguity function with respect to
time delay and Doppler shift provides relatively little information about
multistatic radar performance because time delay and Doppler shift in the
case of multistatic radar have nonlinear relationships with target position
and velocity. Instead, the multistatic ambiguity function is often expressed
in terms of target position and velocity. In this form, the multistatic am-
biguity function is not only dependent on the radar waveform, but it is
also strongly influenced by the radar-target geometry. This emphasizes the
importance of geometry for the performance of multistatic radar.
Inspired by the unique peculiarities of multistatic radar in relation to
nonlinear signal processing, waveform and geometry, this book presents
modern signal processing techniques for multistatic tracking radar systems
with the pivotal theme of performance optimization via waveform adap-
tation, geometry optimization and pseudolinear tracking algorithms. The
core of this book is structured in three parts. Part 1 (Adaptive waveform se-
lection) presents several adaptive waveform selection algorithms for various
multistatic tracking problems. Part 2 (Optimal geometry analysis) addresses
the problem of radar-target geometry optimization for multistatic time-of-
arrival based target localization. Part 3 (Pseudolinear tracking algorithms)
focuses on the design and analysis of inherently-stable and low-complexity
xi
xii Preface
closed-form tracking and localization algorithms for passive multistatic

radar systems.
Simulation examples provided within the book are accompanied by
supplementary MATLAB codes which are available online at the webpage
of this book on the Elsevier website https://www.elsevier.com/books-
and-journals/book-companion/9780128153147. These MATLAB codes
are given on an as-is basis, and no warranties are claimed. The contrib-
utors of the codes will not be held liable for any damage caused.
The authors would like to extend their sincere gratitude to Professors
Fulvio Gini and Maria Sabrina Greco of the University of Pisa and Doctor
Hatem Hmam of the Australian Defence Science and Technology Group
for their expert peer-review and feedback on the book proposal. The au-
thors are grateful to the Editorial Project Manager of Academic Press, John
Leonard, and all of the publishing team for their support in the publication
of this book. In addition, the first author expresses his loving gratitude and
dedicates this book to his wife, Lien, for her love and encouragement. The
second author acknowledges the support and understanding of his family
during the writing of this book.
List of Abbreviations and Symbols
Abbreviations
2D Two-dimensional
3D Three-dimensional
AOA Angle-of-arrival
BCPLE Bias-compensated PLE
CRLB Cramér-Rao lower bound
DEKF Diffusion extended Kalman filter
EKF Extended Kalman filter
FDOA Frequency-difference-of-arrival
FIM Fisher information matrix
IMM Interacting multiple model
IV Instrumental variable
IVE IV estimator
LFM Linear frequency-modulated
LOS Line of sight
MIMO Multiple-input multiple-output
MLE Maximum likelihood estimator
MSE Mean squared error
PDA Probabilistic data association
PLE Pseudolinear estimator
RMSE Root mean squared error
Rx Receiver
TB Time-bandwidth
TDOA Time-difference-of-arrival
TLS Total least-squares
TMA Target motion analysis
TOA Time-of-arrival
Tx Transmitter
UAV Unmanned aerial vehicle
WIVE Weighted instrumental variable estimator
Symbols
Rn Set of real vectors of dimension n
x A scalar
x A vector
X A matrix
I N ×N or I An identity matrix of dimension N × N or of appropriate dimension
0N ×M or 0 Zero matrix of dimension of N × M or of appropriate dimension
1N ×M or 1 A matrix of ones of dimension N × M or of appropriate dimension
x(i) The ith element of vector x
xiii
xiv List of Abbreviations and Symbols
X (i ) The ith diagonal element of matrix X

X (i, j) or X [i,j] The (i, j)th element of matrix X
x(i : j) A sub-vector formed by the ith to jth elements of vector x
X (i, :) The ith row of matrix X
diag(x, y, . . . ) A diagonal matrix with diagonal entries x, y, . . .
diag(X , Y , . . . ) A block diagonal matrix with diagonal blocks X , Y , . . .
diag(x) A diagonal matrix formed by the elements of vector x
x̃ A noisy measurement of variable x
x̃ A noisy measurement of vector variable x
x̂ An estimate of variable x̂
x̂ An estimate of vector variable x̂
X ◦ or X ◦ The noise-free version of matrix X
· Euclidean norm
(·)T Matrix transpose
|·| Matrix determinant
trace(·) Matrix trace
log(·) Base-10 logarithm
ln(·) Natural logarithm
exp(·) Exponential function
sgn(·) Signum function
Element-by-element (Schur) product
E{·} Statistical expectation
CHAPTER 1
Introduction
Contents
1.1. Historical background 1
1.2. Purpose and scope 4
1.3. Outline of book 6
1.1 Historical background

Over its history of more than a century, radar has played a fundamental and
prominent role in a wide range of civilian and military applications thanks
to its capability to operate in all weather, day and night, and to detect, track,
and image targets with high accuracy at long stand-off ranges. Nowadays,
the use of radar can be found in weather forecasting, remote sensing and
mapping, astronomy, aerial and terrestrial traffic control, air defense and
weapon control, automotive sensor systems, high-resolution target imaging
and recognition, airborne collision avoidance systems, and eldercare and
assisted living, to name but a few [1–12].
Radar is essentially an electromagnetic sensor for detecting and ranging
objects by radiating electromagnetic wave to illuminate the scene of interest
and receiving reflected echoes from the objects. By processing the received
signal, radar can make a decision on whether one or more targets are present
in the scene of interest, and determine the position and velocity and possi-
bly the size, shape and features of the detected targets. In terms of system
geometry, radar can be categorized into three different types: monostatic
radar, bistatic radar and multistatic radar. Monostatic radar has a transmitter
and a receiver which are collocated at the same site. In contrast, bistatic
radar is a radar that operates with a transmitter and a receiver located sepa-
rately at different sites. Multistatic radar is a generalization of bistatic radar
by incorporating multiple transmitters and receivers at different locations
while having overlapping spatial coverage, as illustrated in Fig. 1.1.
By conventional definition, multistatic radar can be viewed as a sys-
tem of several individual transmitter–receiver pairs that operate indepen-
dently [13,14]. Detection, target estimation, and other high-level target
information from these transmitter–receiver pairs are communicated to
a central processor, where they are combined to improve detection and
Signal Processing for Multistatic Radar Systems Copyright © 2020 Elsevier Ltd.
https://doi.org/10.1016/B978-0-12-815314-7.00008-1 All rights reserved. 1
2 Signal Processing for Multistatic Radar Systems
Figure 1.1 An illustration of a multistatic radar system.
estimation performance. This type of data fusion is often considered as non-

coherent processing. For example, multistatic radar can use multilateration
to estimate the target kinematic state (i.e., position and velocity) based on
range, Doppler and/or angle measurements obtained independently at dif-
ferent transmitter–receiver pairs within the system. MIMO (multiple-input
multiple-output) radar with widely separated antennas is another type of
multistatic radar with a different design that distinguishes it from multistatic
radar by joint processing at the signal level for both transmission and re-
ception, as well as its close connection to MIMO communications [15,16].
However, this type of multistatic radar is not covered in this book.
The spatial diversity provided by the separation of transmitters and re-
ceivers not only brings a number of advantages to bistatic and multistatic
radars over monostatic radar, but also gives system designers extra degrees
of freedom to optimize the radar system design for specific applications [13,
14]. Since the receivers are not collocated with the transmitters, bistatic and
multistatic radars are able to counter retrodirective jammers which sense the
radar signals from the transmitters and direct jamming signals back towards
the transmitters. The bistatic/multistatic geometry also enhances the radar
cross section for stealthy targets, thus improving the detection performance.
Moreover, the target kinematic state can be estimated with a high accuracy
by multistatic radar thanks to the exploitation of multilateration from mea-
surements collected by multiple transmitter–receiver pairs, each having a
different bistatic geometry with respect to the target. In addition, multi-
static radar can avoid the unfavorable geometry of bistatic radar in which
the target is located near the line-of-sight between the transmitter and re-
ceiver.
Introduction 3
Bistatic and multistatic radars can incorporate their own dedicated

transmitters or exploit illuminators of opportunity from other transmis-
sion sources [13,14,17]. The illuminators of opportunity may come from
existing radar systems (either cooperative or non-cooperative) or from com-
mercial non-cooperative broadcast and communications signals. The radar
systems that utilize illuminators of opportunity are commonly known as
passive. In a passive configuration, bistatic and multistatic radar can oper-
ate in civilian areas, particularly in large cities with multiple airports, where
radar transmission may not be allowed. In addition, exploiting the illumina-
tors of opportunity from other existing transmission sources eliminates the
construction and operation costs for transmitters and associated equipment.
These advantages of bistatic and multistatic radar come at the expense of
increased complexity for system hardware and signal processing, particularly
in terms of direct-path signal cancellation and transmitter–receiver syn-
chronization. However, great progress has been made towards solving these
complexity issues thanks to the tremendous amount of research dedicated
to the topics of beamforming, integrated circuit design, and the Global Po-
sitioning System, thus making bistatic and multistatic radar systems perfectly
feasible in real-world situations [13].
Bistatic and multistatic radars have experienced periodic resurgences
throughout the history of radar. The very first radars were of the bistatic
type. Prior to and during World War II, several continuous-wave bistatic
radars were developed and deployed in many countries including the
United States, United Kingdom, Soviet Union, France, Germany, Japan
and Italy. Since the invention of radar duplexer, which allows the use of
pulsed waveforms with a single common antenna for both transmission
and reception, monostatic radar has dominated the field of radar research
mainly due to the advantages of single-site operation. Consequently, all
bistatic radar work was discontinued by the end of World War II. The first
resurgence of bistatic and multistatic radar started in the 1950s with applica-
tions in air defense and ballistic missile launch warning, tactical semiactive
homing missiles, and test range instrumentation and satellite tracking. The
second resurgence happened in the 1970s and 1980s. A number of experi-
mental bistatic radar systems were tested (but not deployed) in response to
the retro-jamming and antiradiation missile threats. The Multistatic Mea-
surement System for ballistic missile tracking, which is a multistatic radar
hitchhiker, was deployed in the 1980s. The first experiment of using broad-
cast transmitters as illuminators of opportunity was conducted during this
resurgence period. Since the beginning of the third resurgence in the
mid-1990s, extensive research studies have been devoted to bistatic and

multistatic radars including image focusing and motion compensation tech-
niques for bistatic Synthetic Aperture Radar, adaptive clutter cancellation
methods for bistatic moving target indication, and exploitation of commer-
cial broadcast transmitters for passive radar. A comprehensive review of the
history of radar, including bistatic and multistatic radars, is available in [13,
14,17].
1.2 Purpose and scope

The ambiguity function is an important and indispensable tool for radar
performance analysis [8,18,19]. The ambiguity function essentially charac-
terizes the auto-correlation of the complex envelope of the radar waveform
with its shifted copy in time and frequency. The point target response of
the waveform is expressed by the ambiguity function as a two-dimensional
function of time delay and Doppler shift (or equivalently target range and
radial velocity). Based on the shape of the ambiguity function, radar perfor-
mance can be evaluated in terms of estimation accuracy, target resolution,
and clutter cancellation. In monostatic radar, where time delay and Doppler
shift are proportional to target range and radial velocity, respectively, the
ambiguity function is only dependent on the radar waveform without hav-
ing any influences from the radar-target geometry. In terms of time delay
and Doppler shift, the ambiguity function for bistatic and multistatic radars
remains identical to that of monostatic radar. However, this definition of
ambiguity function provides relatively little information about radar perfor-
mance because time delay and Doppler shift have nonlinear relationships
with target position and velocity parameters [20,21]. Depending on the
application, different coordinate systems can be used to characterize target
position and velocity, for example, the North-referenced system, the Carte-
sian system, and spherical coordinates. Therefore, for bistatic and multistatic
radars, the ambiguity functions are often expressed in terms of target po-
sition and velocity in these coordinate systems. In such forms, the bistatic
and multistatic ambiguity functions are not only dependent on the radar
waveform, but also strongly influenced by the radar-target geometry [20,
21]. This highlights the importance of geometry to the performance of
multistatic radar.
The focus of this book is on modern signal processing techniques for
multistatic tracking radar systems. Inspired by the unique peculiarities of
multistatic radar as highlighted above, the pivotal theme of the topics pre-
Introduction 5
sented in this book is performance optimization via waveform adaptation,

geometry optimization and pseudolinear tracking algorithms. The core of
this book is structured into three parts:
• Part 1: Adaptive waveform selection (Chapters 2–5) presents a
number of adaptive waveform selection algorithms for various multi-
static tracking scenarios including maneuvering target tracking in Chap-
ter 2, target tracking in clutter in Chapter 3, target tracking with Carte-
sian estimates in Chapter 4, and distributed target tracking in Chapter 5.
The objective is to optimize tracking performance via minimizing the
mean squared error (MSE) of the target kinematic state estimate in
the Cartesian coordinates. This is achieved by minimizing the trace of
the target state error covariance matrix which can be computed from
the update equations of the tracking algorithms using the Cramér–Rao
lower bound (CRLB) of radar measurement errors.
• Part 2: Optimal geometry analysis (Chapters 6–7) addresses the
problem of radar-target geometry optimization for multistatic target
localization. The optimal geometries for time-of-arrival (TOA)-based
localization are analytically derived for two multistatic configurations:
(i) a single transmitter and multiple receivers in Chapter 6, and (ii) mul-
tiple independent bistatic channels in Chapter 7. The optimal geometry
analyses are conducted based on minimizing the area of estimation con-
fidence region, which is equivalent to maximizing the determinant of
the Fisher information matrix (FIM).
• Part 3: Pseudolinear tracking algorithms (Chapters 8–9) focuses
on the design and analysis of inherently-stable and low-complexity
closed-form tracking and localization algorithms for passive multistatic
radar systems. Closed-form batch estimators are presented in Chap-
ter 8 for multistatic target motion analysis of a constant-velocity tar-
get utilizing angle-of-arrival (AOA), time-difference-of-arrival (TDOA)
and frequency-difference-of-arrival (FDOA) measurements under the
framework of pseudolinear estimation and instrumental-variable esti-
mation. The method of closed-form least-squares estimation and its
bias-reduced variants are covered in Chapter 9 for multistatic target lo-
calization with TDOA measurements.
We assume that the reader has a background in digital signal processing,
linear and matrix algebra, and probability and random processes, as well as
a basic understanding of radar system and signal processing (e.g., [1,2]), and
estimation and tracking (e.g., [22–24]).
1.3 Outline of book

A brief overview and summary of the remaining chapters of the book is
given below.
Part 1: Adaptive waveform selection

Chapter 2: Waveform selection for multistatic tracking of a ma-
neuvering target
This chapter considers the problem of adaptive waveform selection for mul-
tistatic tracking of a maneuvering target. An interacting multiple model
algorithm incorporating extended Kalman filters is employed to deal with
the nonlinearity between the radar observation vector and the target state
vector and to handle multiple models of target dynamics. Three wave-
form selection schemes are presented based on minimizing the traces of
the most-likely, minimax and combined covariance matrices of the inter-
acting multiple model algorithm. Monte Carlo simulation examples are
presented to demonstrate the performance advantages of the waveform se-
lection schemes over conventional fixed waveforms.
Chapter 3: Waveform selection for multistatic target tracking in
clutter
This chapter presents an optimal waveform selection algorithm for mul-
tistatic target tracking in a cluttered environment. Due to the false-alarm
measurements caused by clutter, the exact track error covariance matrix is
not available during the waveform selection process (i.e., prior to the ac-
tual waveform transmission). Therefore, an expected track error covariance
matrix computed via the modified Riccati equation is exploited for wave-
form selection. Simulation examples corroborate the effectiveness of the
presented waveform selection algorithm.
Chapter 4: Waveform selection for multistatic target tracking with
Cartesian estimates
This chapter is concerned with multistatic tracking using Cartesian esti-
mates, where only desired measurements from two receivers in the system
are selected and utilized to compute a Cartesian state estimate for target
dynamics which is then processed by a linear Kalman tracker. Joint adaptive
selection of radar waveform and Cartesian estimate is presented to optimize
tracking performance. The CRLBs for Cartesian estimates are derived and
exploited in the development of the joint adaptive selection scheme. The
tracking system under consideration provides three important advantages;
viz., (i) minimal requirement for bandwidth and power consumption in
Introduction 7
transmitter–receiver communication links by utilizing only one Cartesian

estimate for target tracking, (ii) performance optimization brought about
by the joint adaptive selection of radar waveform and Cartesian estimate,
and (iii) inherent benefits of linear estimation, e.g., stability, from the use of
linear Kalman filter. The performance advantages of the presented tracking
system are demonstrated via simulation examples.
Chapter 5: Waveform selection for distributed multistatic target
tracking
This chapter addresses the waveform selection problem in the context of
distributed multistatic target tracking, where the receivers form a connected
network in which they can communicate their radar measurements and
track estimates to their immediate neighbors. A method of performing
joint target tracking and waveform selection over the receivers in a fully de-
centralized manner is presented. Specifically, target tracking and waveform
selection are performed at each receiver exploiting the information data
available in its closed neighborhood. Such a distributed processing strategy
offers many important advantages in terms of cost savings for the opera-
tion and maintenance of the system, as well as increasing the robustness
of the system to link or node failures. Simulation examples are given to
corroborate the performance of the presented method.
Part 2: Optimal geometry analysis

Chapter 6: Optimal geometries for multistatic target localization
with one transmitter and multiple receivers
This chapter analyzes the optimal geometries for the problem of target
localization using TOA measurements collected from a multistatic radar
system with one transmitter and multiple receivers. The optimal geometry
analysis is conducted based on maximizing the determinant of the FIM
which effectively minimizes the area of the estimation confidence region.
Rotating the coordinate system so that the bearing angle of the transmitter
relative to the target is zero, an optimal geometry is formed by placing
some receivers collinearly with the target at the bearing angle of 60 degrees
and the remaining receivers at −60 degrees. The accuracy of the analytical
findings is confirmed via simulation studies including numerical solutions
and sensor trajectory optimization.
Chapter 7: Optimal geometries for multistatic target localization
by independent bistatic channels
This chapter derives the optimal geometries for the target localization prob-
lem using TOA measurements collected from a multistatic radar system
consisting of multiple independent bistatic channels. The optimal angular

geometries for this multistatic configuration are obtained when the target
is collinear with the transmitter and receiver of each bistatic channel and
the target is located at either end. The problem reduces to optimizing an-
gular separation between different bistatic channels which is readily solved
by referring to the results of optimal angular separation for AOA localiza-
tion. Simulation examples on sensor trajectory optimization are presented
to verify the analytical findings.
Part 3: Pseudolinear tracking algorithms

Chapter 8: Batch track estimators for multistatic target motion
analysis
This chapter considers the problem of target motion analysis for a constant-
velocity target by a passive multistatic radar system using AOA, TDOA and
FDOA measurements. Four batch estimators are presented including the
pseudolinear estimator (PLE), bias-compensated PLE (BCPLE), weighted
instrumental-variable estimator (WIVE) and maximum likelihood estima-
tor (MLE). Compared to the iterative MLE, the PLE, BCPLE and WIVE
are closed-form with inherent stability and computation advantages. The
PLE suffers from severe bias problems due to the correlation between the
measurement matrix and pseudolinear noise vector. The BCPLE alleviates
the PLE bias by estimating and subtracting the PLE instantaneous bias. The
BCPLE estimate is incorporated into the WIVE to produce asymptotically
unbiased estimates of target motion parameters. The WIVE is analytically
shown to be asymptotically efficient under small measurement noise. The
performance of the estimators is evaluated via numerical simulation exam-
ples. The WIVE produces the best performance among the closed-form
estimators, while exhibiting performance on a par with the computation-
ally more demanding MLE.
Chapter 9: Closed-form solutions for multistatic target localiza-
tion with time-difference-of-arrival measurements
This chapter focuses on closed-form least-squares estimation techniques
for multistatic target localization using TDOA measurements. To over-
come the computational burden and instability of the MLE, the nonlinear
TDOA measurement equations are algebraically rearranged into a set of
linear equations by introducing a nuisance parameter, thus enabling linear
least-squares to be performed in closed-form. Such a transformation of the
TDOA measurement equations creates two technical challenges. First, the
nuisance parameter is dependent on the true target position and this de-
Introduction 9
pendence must be taken into account in order to obtain efficient estimates.

Second, the measurement matrix in the linearized equations is correlated
with the pseudolinear noise vector, leading to bias problems in the linear
least-squares solution. In this chapter, we cover a range of algorithms that
can effectively tackle these two challenges. Simulation examples are pre-
sented to corroborate the performance benefits of the algorithms.
PART 1
Adaptive waveform
selection
Background and purpose
Waveform diversity is the concept of dynamically adapting the radar wave-
form to optimize the radar performance and confront the nonstationarity
and uncertainty of the environment [8,18]. Waveform diversity has been
exploited across different aspects of radar signal processing from detection,
tracking to classification (see e.g., [8,18] and the references therein). Various
waveform parameters can be used for waveform diversity, including pulse
repetition frequency, carrier frequency, bandwidth, pulse width, amplitude,
coding, spatial characteristics, polarization, jitter, and frequency shaping.
While waveform adaptation and agility have been only recently applied to
synthetic sensing systems like radar and sonar (which have less than 100
years of history), waveform diversity in fact has been being exploited in
nature by echolocating mammals like whales, dolphins and bats for millions
of years.
In target tracking, the tracker incorporates information about the tar-
get (e.g., range, Doppler and/or angle information) in a sequential manner,
while the radar-target geometry and sensing environment are constantly
changing due to radar and/or target motion. Therefore, dynamic adaptation
of waveforms is especially advantageous in target tracking applications as it
11
allows the transmitted waveform to be adjusted on a pulse-to-pulse basis to

accommodate changes in tracker requirements [25–32]. Waveform adap-
tation in target tracking is motivated by the fact that different waveforms
have different resolution properties and thus induce different errors in radar
measurements. The idea is to design or select the waveform to ensure small
measurement errors in the dimensions of the target kinematic state where
the uncertainty of the tracker is large while higher levels of errors may be
tolerated for other dimensions with less uncertainty. Dynamic waveform
adaptation is commonly considered under the framework of control theory,
where the choice of the waveform to be transmitted in the next time step is
made by optimizing a waveform-dependent cost function such as the mean
squared error of the target state estimation error [28–34]. This formulation
of the problem results in a feedback loop where the waveform selected at
the end of the current time step affects the radar and tracker performance in
the next time step which in turn influences the next waveform selection.
For the simple case of one-dimensional target tracking with a linear ob-
servation model, closed-form solutions are available [28]. However, brute
force optimization techniques are often required in most modern tracking
scenarios because of the nonlinearity of observation models, the presence
of clutter or the imperfection of detection [29–32]. For this reason, the
problem is commonly referred to as adaptive waveform selection. The ob-
jective is to select the best waveform candidate within a library of available
waveforms which can optimize the tracking performance at the next time
step. A waveform library may consist of different waveform classes, or sim-
ply a collection of waveforms of the same type with different waveform
parameters.
In this part of the book, we present several adaptive waveform selection
algorithms for different multistatic tracking scenarios based on the control
theoretical approach:
• Chapter 2: maneuvering target tracking,
• Chapter 3: target tracking in clutter,
• Chapter 4: target tracking with Cartesian target state estimates,
• Chapter 5: target tracking in a distributed multistatic system.
The objective is to minimize the MSE of the target kinematic state estimate
in the Cartesian coordinates by minimizing the trace of the target state error
covariance matrix which can be computed from the update equations of
the tracking algorithms using the CRLB of radar measurement errors.
CHAPTER 2
Waveform selection for

multistatic tracking of
a maneuvering target
Contents
2.2. Bistatic radar measurements 14
2.3. Bistatic ambiguity function and Cramér–Rao lower bounds 15
2.4. Target tracking 17
2.4.1 Target dynamic model 17
2.4.2 Observation model 19
2.4.3 Interacting multiple model – extended Kalman filter 20
2.7. Summary 31
2.8. Appendix 31
2.1 Introduction and system overview

This chapter considers the problem of adaptive waveform selection for ma-
neuvering target tracking with a multistatic radar system consisting of a
single transmitter (Tx) and multiple receivers (Rx). The overall system con-
figuration is depicted in Fig. 2.1. At each receiver, time delay, Doppler shift
and angle-of-arrival of the received signal (which is reflected back from the
target) are estimated by an antenna array. The receivers then communicate
their measurements to a central processor located at the transmitter site to
perform target tracking and waveform selection. To track a maneuvering
target, the interacting multiple model (IMM) algorithm [23,24,35] is em-
ployed. After that, waveform selection is performed to determine the next
waveform to be transmitted in order to minimize the MSE of the target
state estimate.
For the sake of simplicity, we consider a two-dimensional (2D) tracking
scenario with the radar-target geometry depicted in Fig. 2.2. It is straight-
forward to extend the problem to three-dimensional (3D) scenarios. In
Fig. 2.2, pk = [px,k , py,k ]T and vk = [vx,k , vy,k ]T denote the unknown target
position and velocity at discrete-time instant k ∈ {0, 1, 2, . . . }, respectively.
https://doi.org/10.1016/B978-0-12-815314-7.00010-X All rights reserved. 13
Figure 2.1 System configuration of multistatic tracking radar under consideration.
The transmitter position is denoted by t = [tx , ty ]T , and the receiver posi-

tions by ri = [rx,i , ry,i ]T (i = 1, . . . , N ) where i represents the receiver index.
2.2 Bistatic radar measurements

Each receiver cooperates with the transmitter to form a bistatic channel.
The time delay, Doppler shift and angle-of-arrival measurements obtained
by receiver i at time instant k are given by
τ̃i,k = τi,k + eτ,i,k , (2.1a)
˜ i,k = i,k + e,i,k , (2.1b)
θ̃i,k = θi,k + eθ,i,k , (2.1c)
where
1
τi,k = pk − t + pk − ri , (2.2a)
c

f◦ (pk − t)T vk (pk − ri )T vk
i ,k = + , (2.2b)
c pk − t pk − ri

py,k − ry,i
θi,k = tan−1 . (2.2c)
px,k − rx,i
Here, f◦ is the transmission frequency, c is the speed of signal propagation,
tan−1 (·) denotes the 4-quadrant arctangent, and · is the Euclidean norm.
Waveform selection for multistatic tracking of a maneuvering target 15
Figure 2.2 Target tracking geometry in 2D.
In (2.1), eτ,i,k , e,i,k and eθ,i,k are measurement errors modeled as zero-mean
Gaussian random variables.
Since the radar measurements can achieve their CRLBs approximately
under mild noise conditions, we follow [28–31,36] to assume that the vari-
ances of radar measurement errors eτ,i,k , e,i,k and eθ,i,k can be approximated
by their corresponding CRLBs. It is important to emphasize that the radar
measurement errors and CRLBs are known to be dependent on the radar
transmitted waveform [19]. This dependence is discussed further in the next
section, where the CRLBs for time delay and Doppler shift are explicitly
expressed as functions of transmitted waveform parameters.
2.3 Bistatic ambiguity function and Cramér–Rao lower

bounds
The ambiguity function is defined as the absolute value of the correlation
between the complex envelope u(t) of the radar transmitted waveform and
its time-frequency shifted copy [19], i.e.,

∞
(τH , H ) = u(t − τA )u∗ (t − τH )e−j2π(H −A )t dt . (2.3)
−∞
Here, τA and A denote the actual time delay and Doppler shift, respectively,
while τH and H denote the hypothesized time delay and Doppler shift,
respectively. The ambiguity function (τH , H ) is maximized when τH =
τA and H = A . Here, we assume that the narrowband condition is satisfied,
i.e., the time-bandwidth product of the transmitted signal is much smaller

than the signal propagation speed divided by the time derivative of the
total bistatic range to the target. The concept of ambiguity function was
originally developed for monostatic radar. However, in terms of measuring
time delay and Doppler shift, this ambiguity function is also applied to
bistatic radar [8,37–39]. Since time delay and Doppler shift are used directly
as measurement inputs to the tracker, we can characterize the accuracy of
these measurements based on the ambiguity function (τH , H ) in (2.3).
Given a sufficiently high SNR, the Fisher information matrix {τ,} ,
i.e., the inverse of the CRLB matrix C{τ,} , for time delay and Doppler shift
measurements is a function of both the SNR and the second derivatives of
the ambiguity function (τH , H ) [19]:
⎡ ⎤
∂ 2 (τH , H ) ∂ 2 (τH , H )

⎢ ∂τH2 ∂τH ∂H ⎥
{τ,} = C− ⎢ ⎥
{τ,} = −2η ⎣ ∂ 2 (τ , )
1
(2.4)
H H ∂ (τH , H )
2 ⎦
∂H ∂τH ∂H2
τH =τA , H =A
where η denotes the SNR at the receiver. Note that, consistent with the
fact that the monostatic and bistatic ambiguity functions have the same
form with respect to time delay and Doppler shift, the Fisher information
matrix {τ,} and the CRLB matrix C{τ,} for time delay and Doppler shift
given in (2.4) hold for both monostatic and bistatic cases.
Since the sharpness of the ambiguity function is determined by the
transmitted waveform as seen in (2.3), the Fisher information matrix {τ,}
and thus the CRLB matrix C{τ,} are also dependent on the transmitted
waveform. Therefore, the CRLB matrix C{τ,} can be written explicitly as
a function of the SNR at the receiver as well as the transmitted waveform
parameters ψ :
C{τ,} (η, ψ) = −{τ,}

1
(η, ψ). (2.5)
On the other hand, the CRLB for angle-of-arrival measurement ob-

tained by an antenna array is independent of the transmitted waveform [40,
41], and it is inversely proportional to the SNR at the receiver, i.e.,
σθ2
Cθ (η) = . (2.6)
η
Although these results were originally derived for monostatic radar, they
also hold for bistatic radar because the operation of bistatic radar is basically
the same as that of monostatic radar in measuring time delay, Doppler shift
and angle-of-arrival.
Since the CRLB for angle-of-arrival measurement is independent of the
CRLBs for time delay and Doppler shift measurements, the overall CRLB
matrix for time delay, Doppler shift and angle-of-arrival measurements ob-
tained at receiver i at time instant k is given by
Ci,k (ηi,k , ψ k ) = diag(C{τ,} (ηi,k , ψ k ), Cθ (ηi,k )) (2.7)
where ηi,k is the SNR at receiver i at time instant k, and ψ k is a vector

consisting of the transmitted waveform parameters at time instant k.
2.4 Target tracking

The target tracking process is performed in a centralized manner at the
transmitter site after receiving radar measurements from all receivers via
communication links.
2.4.1 Target dynamic model

The target motion is assumed to be governed by the kinematic model
xk+1 = Fxk + wk (2.8)
where xk is the target kinematic state vector at time instant k, F is the

state transition matrix and wk ∼ N (0, Q) is the process noise modeled as
independent zero-mean Gaussian random variable. The target kinematic
state vector may include the target position, velocity, acceleration and/or
other target motion parameters, while the expressions for F and Q are
dependent on the target dynamics. In what follows, we summarize three
widely used kinematic models in the literature including the nearly constant
velocity model, nearly constant acceleration model and nearly coordinated
turn model [23,24,42].
Nearly constant velocity model

The state vector is defined by xk = [pTk , vTk ]T . The transition matrix F is

I TI 2×2
F = 2×2 (2.9)
02×2 I 2×2
and the process noise covariance matrix Q is

⎡ ⎤
T3 T2
⎢ Qo Q
2 o⎥
Q = ⎣ 32 ⎦ (2.10)
T
Q TQo
2 o
where I and 0 are identity and zero matrices, respectively, with their di-
mensions specified in the superscript, and T is the sampling interval. Here,
Qo = diag(qx , qy ), with qx and qy being the power spectral densities of the
process noise in x- and y-coordinates, corresponds to the level of the un-
known target maneuver.
Nearly constant acceleration model

The state vector is given by xk = [pTk , vTk , aTk ]T where ak = [ax,k , ay,k ]T de-
notes the acceleration of the target at time instant k. The expressions of F
and Q for this model are
⎡ ⎤
T2
⎢ I 2×2 TI 2×2 I 2×2 ⎥
F=⎢
⎣02×2 I 2×2
2
TI 2×2 ⎦
⎥ (2.11)
02×2 02×2 I 2×2
and
⎡ ⎤
T5 T4 T3
⎢ 20 Qo Q
83 o
Q
62 o ⎥
⎢ 4 ⎥
⎢T T T ⎥
Q = ⎢ Qo Q Qo ⎥ . (2.12)
⎢ 8 32 o 2 ⎥
⎣T3 T ⎦
Q Q TQo
6 o 2 o
Nearly coordinated turn model
The state vector is xk = [pTk , vTk ]T . The transition matrix F is given by
⎡ sin(ωT ) 1 − cos(ωT ) ⎤
1 0 −
⎢ ω ω ⎥
⎢ 1 − cos(ωT ) sin(ωT ) ⎥
⎢ ⎥
F = ⎢0 1 ⎥ (2.13)
⎢ ω ω ⎥
⎣0 0 cos(ωT ) − sin(ωT ) ⎦
0 0 sin(ωT ) cos(ωT )
and the process noise covariance matrix Q given by

⎡ ⎤
Q11 Q12 Q13 Q14
⎢Q Q24 ⎥
⎢ 21 Q22 Q23 ⎥
Q = q⎢ ⎥ (2.14)
⎣Q31 Q32 Q33 Q34 ⎦
Q41 Q42 Q43 Q44
with
2 (ωT − sin(ωT ))
Q11 = Q22 = (2.15a)
ω3
Q12 = Q21 = Q34 = Q43 = 0 (2.15b)
1 − cos(ωT )
Q13 = Q24 = Q31 = Q42 = 2
(2.15c)
ω
ωT − sin(ωT )
Q14 = Q41 = (2.15d)
ω2
ωT − sin(ωT )
Q23 = Q32 = − (2.15e)
ω2
Q33 = Q44 = T . (2.15f)
In the above model, ω is the angular turn rate and is presumed to be a

known constant. For the case that the turn rate ω is not a known constant,
the coordinated turn model becomes nonlinear where the turn rate ω is
incorporated as an element of an augmented target state vector.
Multiple models
In practice, a single dynamic model cannot always capture the real complex
motion of the target because the target may move straight with a constant
velocity, make turns or accelerate. In this sense, it is desirable to use the
IMM algorithm [23,24,35,43] to handle multiple models of target dynamics
simultaneously. In this chapter, we let M denote the number of models
used to characterize the target dynamics, and F m and Qm denote the state
transition matrix and process noise covariance matrix of the mth dynamic
model.
2.4.2 Observation model

Multiplying (2.1a) with c and (2.1b) with c /f◦ gives the expressions for the
total bistatic range and range-rate measurements, respectively, for the target
with respect to the transmitter and receiver i:
d̃i,k = di,k + ed,i,k , (2.16a)

ρ̃i,k = ρi,k + eρ,i,k , (2.16b)
where
d̃i,k = c τ̃i,k , di,k = c τi,k , ed,i,k = ceτ,i,k , (2.17a)

ρ̃i,k = c ˜ i,k /f◦ , ρi,k = c i,k /f◦ , eρ,i,k = ce,i,k /f◦ . (2.17b)
Stacking d̃i,k , ρ̃i,k and θ̃i,k together forms a measurement vector at receiver i
at time instant k as
ỹi,k = [d̃i,k , ρ̃i,k , θ̃i,k ]T = yi,k + ei,k = [di,k , ρi,k , θi,k ]T + [ed,i,k , eρ,i,k , eθ,i,k ]T .
(2.18)
By stacking ỹ1,k , . . . , ỹN ,k from all receivers, the overall measurement vector
at time instant k is given by
ỹk = [ỹT1,k , . . . , ỹTN ,k ]T = yk + ek = [yT1,k , . . . , yTN ,k ]T + [eT1,k , . . . , eTN ,k ]T . (2.19)
Since radar measurement errors are independent across different receivers,

the covariance matrix of ek is given by
Rk = E{ek eTk } = diag{R1,k , . . . , RN ,k } (2.20)
where the covariance matrix Ri,k at each receiver is computed from the
corresponding CRLB matrix Ci,k as
Ri,k = Ci,k . (2.21)
Here, is defined by = diag(c , c /f◦ , 1) which reflects the transformations

from {τ̃i,k , ˜ i,k } to {d̃i,k , ρ̃i,k }.
Note that Ci,k is a function of ηi,k and ψ k as discussed in Section 2.3;
thus Rk is a function of η1,k , . . . , ηN ,k and ψ k . Since the main interest here is
to optimize tracking performance by waveform adaptation, we only write
Rk (ψ k ) explicitly as a function of ψ k (i.e., the transmitted waveform pa-
rameter vector).
2.4.3 Interacting multiple model – extended Kalman filter

Since the radar measurement vector yk is a nonlinear function of the target
state xk (see (2.1), (2.2) and (2.16)–(2.19)), a nonlinear tracking algorithm
must be used. Among various options available, the extended Kalman
filter (EKF) is computationally most efficient compared to other more so-
phisticated nonlinear Kalman filtering algorithms such as the sigma-point
Kalman filters including the unscented and cubature Kalman filters, and
the particle filters. For the target tracking problem under consideration,
the EKF can produce a good tracking performance thanks to the advantage
of spatial diversity given by multistatic radar. The procedure for calculating
the EKF estimates for the mth target dynamic model is given by
x̂m,k+1|k = F m x̂m,k|k , (2.22a)

Pm,k+1|k = F m Pm,k|k F Tm + Qm , (2.22b)
Sm,k+1 (ψ k+1 ) = H m,k+1 Pm,k+1|k H Tm,k+1 + Rk+1 (ψ k+1 ), (2.22c)
K m,k+1 (ψ k+1 ) = Pm,k+1|k H Tm,k+1 Sm−,1k+1 (ψ k+1 ), (2.22d)
z̃m,k+1 = ỹk+1 − yk+1 (x̂m,k+1|k ), (2.22e)
x̂m,k+1|k+1 (ψ k+1 ) = x̂m,k+1|k + K m,k+1 (ψ k+1 )z̃m,k+1 , (2.22f)
Pm,k+1|k+1 (ψ k+1 ) = (I − K m,k+1 (ψ k+1 )H m,k+1 )Pm,k+1|k, (2.22g)
where x̂m,k|j and Pm,k|j are the state estimate and error covariance,
respectively, at time instant k given measurements through time in-
stants, 0, 1, . . . , j, and H m,k+1 = H k+1 (x̂m,k+1|k ) is the Jacobian matrix H k+1
of yk+1 with respect to xk+1 evaluated at x̂m,k+1|k . The expression for the
Jacobian matrix H k+1 is provided in Section 2.8 (Appendix).
To track a maneuvering target, we employ the IMM algorithm, which
can incorporate multiple models of target dynamics. The IMM algorithm
is a technique of combining multiple state hypotheses by running multiple
filters in parallel to obtain a better state estimate for targets with changing
dynamics [23,24,35,43]. Specifically, the IMM algorithm treats the dynamic
motion of the target as multiple switching models:
xk+1 = F (mk+1 )xk + wk (mk+1 ) (2.23)
where mk+1 is a finite-state Markov chain (mk+1 ∈ {1, . . . , M }) which fol-

lows the transition probabilities plm for switching from model l to model m
and the covariance matrix of the process noise wk (mk+1 ) is governed
by E{wk (mk+1 = l)wTk (mk+1 = l)} = Q(mk+1 = l) [35].
Given the model-conditioned state estimate x̂m,k|k and error covari-
ance Pm,k|k and the model probability estimate (denoted as μm,k|k ) for all
models m = 1, 2 . . . , M from previous time instant k, the state estimation of
the IMM algorithm at time instant k + 1 proceeds as follows.
1. Mixing of state estimates:
– Compute the predicted model probabilities

M
μm,k+1|k = plm μl,k|k . (2.24)
l=1
– Compute the conditional model probabilities

1
μl|m,k|k = plm μl,k|k . (2.25)
μm,k+1|k
– Compute the mixed estimates and covariances

M
x̂m† ,k|k = μl|m,k|k x̂l,k|k , (2.26a)
l=1
M

Pm† ,k|k = μl|m,k|k Pl,k|k + (x̂l,k|k − x̂m† ,k|k )(x̂l,k|k − x̂m† ,k|k )T .
l=1
(2.26b)
2. Model-conditioned updates:
- The mixed estimate x̂m† ,k|k and covariance Pm† ,k|k are used as inputs
for the mth EKF to compute the state estimate x̂m,k+1|k+1 and error
covariance Pm,k+1|k+1 at time instant k + 1.
3. Model likelihood computations:

−1/2 1 T
m,k+1 = 2π S m,k+1
exp − z̃ (S )−1
z̃ (2.27)
m,k+1 m,k+1 m,k+1
2
where | · | denotes the determinant.
4. Model probability updates
1
μm,k+1|k+1 = μm,k+1|k m,k+1 (2.28)
κ
where κ is the normalization factor given by

M
κ= μl,k+1|k l,k+1 . (2.29)
l=1
5. Combination of state estimates:

M
x̂IMM,k+1|k+1 = μm,k+1|k+1 x̂m,k+1|k+1 , (2.30a)
m=1
M
PIMM,k+1|k+1 = μm,k+1|k+1 Pm,k+1|k+1 + x̂m,k+1|k+1 x̂Tm,k+1|k+1 ,
m=1
(2.30b)
where x̂m,k+1|k+1 = x̂m,k+1|k+1 − x̂IMM,k+1|k+1 .
2.5 Adaptive waveform selection

The main motivation behind adaptive waveform selection is that the track-
ing performance (i.e., the state estimation error covariance matrix) is de-
pendent on the transmitted waveform parameters. To see this relationship
more clearly, we recall from (2.20) that, at time instant k + 1, the mea-
surement covariance matrix Rk+1 (ψ k+1 ) is a function of the transmitted
waveform parameter vector ψ k+1 . As a result, the error covariance of fil-
tered state estimate Pm,k+1|k+1 (ψ k+1 ) for the mth EKF component of the
IMM-EKF algorithm is explicitly dependent on ψ k+1 as seen in (2.22).
Consequently, the combined state error covariance matrix PIMM,k+1|k+1 is
also a function of ψ k+1 . Based on this dependence, the tracking perfor-
mance can be optimized by adaptively adjusting the parameters of the
waveform transmitted at the next time instant k + 1.
The criterion for our adaptive waveform selection scheme is to mini-
mize the MSE of the target state estimate which is equivalent to minimizing
the trace of the state estimate’s error covariance matrix. Since there exist
more than one error covariance matrix for the maneuvering target track-
ing problem under consideration, i.e., P1,k+1|k+1 , P2,k+1|k+1 , . . . , PM ,k+1|k+1
for the hypothesized models and PIMM,k+1|k+1 for the combined state esti-
mate, one matrix must be chosen to characterize the tracking performance.
Three options are available [44]:
(i) The most-likely covariance which corresponds to the dynamic model
with the highest predicted probability of being the correct model:
PMost-likely,k+1|k+1 = Pm ,k+1|k+1 (2.31)
where
m = arg max μm,k+1|k . (2.32)

m
(ii) The minimax covariance which corresponds to the covariance with

the largest trace value:
PMinimax,k+1|k+1 = Pm ,k+1|k+1 (2.33)
where

m = arg max trace Pm,k+1|k+1 . (2.34)
m
(iii) The predicted combined covariance matrix PIMM,k+1|k defined by

M
x̂IMM,k+1|k = μm,k+1|k x̂m,k+1|k , (2.35a)
m=1
PIMM,k+1|k

M

= μm,k+1|k Pm,k+1|k+1 +(x̂m,k+1|k −x̂IMM,k+1|k )(x̂m,k+1|k −x̂IMM,k+1|k )T .
m=1
(2.35b)
Note that the error covariance of filtered IMM estimate, PIMM,k+1|k+1 , can-
not be used because the measurement vector ỹk+1 is not available during
the waveform selection step at time instant k.
By denoting Pk+1|k+1 as the covariance matrix chosen from the three
above covariance matrices for using in adaptive waveform selection, the
waveform optimization problem is defined as
opt
ψ k+1 = arg min trace Pk+1|k+1 (ψ) , (2.36)
ψ∈
where is a library of possible waveforms. The waveform library may

include a number of different radar waveform classes or a single radar wave-
form class with various waveform parameters. Note that this waveform
selection step is performed at the end of the processing at time instant k
and prior to the waveform transmission at the next time instant k + 1.
2.6 Simulation examples

Fig. 2.3 depicts the simulated tracking scenario with one transmitter lo-
cated at t = [0, 0]T m and four receivers located at r1 = [20 000, 0]T , r2 =
[10 000, 15 000]T , r3 = [20 000, −5000]T and r4 = [0, 10 000]T m. The tar-
get follows a nearly constant velocity trajectory with qx = qy = q = 10 m2 /s3
from initial position p0 = [27 000, 7000]T m with initial velocity v0 =
[−400, −200]T m/s. From t = 11 s, the target performs a turn by uniformly
changing its velocity to [200, −200]T m/s at t = 12 s. After the turn, the
target travels with a nearly constant velocity model again with q = 10 m2 /s3 .
In this simulation, the transmitter emits Gaussian linear frequency-
modulated (LFM) pulses. The complex envelope of LFM pulse is given
by [19]
Figure 2.3 Simulated target tracking geometry.
1/4
1 1
s̃(t) = exp − − j2π b t2 (2.37)
πλ2 2λ2
where λ is the Gaussian pulse length parameter and b = F /(2Ts ) is

the frequency modulation rate. Here, F is the frequency sweep and Ts
is the effective pulse duration which is approximated by 7.4338λ [28].
We adapt the waveform based on λ and F , i.e., the transmitted wave-
form parameter vector of ψ = [λ, F ]T , with λ ∈ {20, 30, 40, 50, 60} µs
and F ∈ {0.1, 0.325, 0.55, 0.775, 1} MHz. The waveform selection is per-
formed via a grid search. The transmission frequency is set to fc = 12 GHz,
the pulse repetition interval and sampling interval to T = 200 m s, and the
speed of signal propagation to c = 3 × 108 m/s.
For the LFM pulses, the CRLB matrix C{τ,} for time delay and Doppler
shift is given by [28]
⎡ ⎤
1 2λ
2 −4bλ2
C{τ,} = ⎣ 1 ⎦. (2.38)
η −4bλ2 + 8b2 λ2
2π 2 λ2
Therefore, the measurement error covariance matrix Ri,k becomes
⎡ ⎤
4bc 2 λ2
2 2
⎢ 2c λ − 0⎥
1 ⎢ ⎢ f◦ ⎥
R i ,k = 4bc 2 λ2 c2 1 ⎥. (2.39)
η i ,k ⎢
⎣ − + 8b2 λ2 0⎥
⎦
f◦ f◦2 2 2
2π λ
0 0 σθ2
The SNR at receiver i at time instant k is modeled as

d04
η i ,k = (2.40)
pk − t2 pk − ri 2
with d0 = 80 000 m. In the simulations, we set σθ = 0.04 rad.

In this section, the IMM algorithm utilizes two nearly constant veloc-
ity models with different process noise levels to characterize hypothesized
target dynamics:
Model 1: a small process noise for modeling uniform motion,
Model 2: a large process noise for modeling maneuvering motion (e.g.,
turning or acceleration).
Specifically, we set qx = qy = q = 10 m2 /s3 for model 1 and q = 100 000 m2 /s3
for model 2. In a more general situation, other target dynamic models
can be added to the list of hypothesized models. This extension will only
require additional tracking filters to be run in parallel. The initial model
probabilities are set to μ1,0|−1 = 0.9 for model 1 and to μ2,0|−1 = 0.1 for
model 2. The transition probability matrix between models 1 and 2 is set to

0.9 0.1
[plm ] = . (2.41)
0.2 0.8
The track initialization is set to x̂1,0|−1 = x̂2,0|−1 = x̂0|−1 which is drawn from
a normal distribution with mean x0 and covariance P0|−1 = diag(10002 ,
10002 , 502 , 502 ).
For comparison purposes, the root mean squared error (RMSE) of the
target state estimate x̂IMM,k|k is computed from NMC = 2000 Monte Carlo
simulation runs using

NMC
1/2
1 2
n n
RMSEk = x̂IMM,k|k − xk (2.42)
NMC n=1
where xkn and x̂IMM,

n
k|k denote the true and estimated target state vector at
time instant k in the nth Monte Carlo run.
Adaptive waveform versus fixed waveform

Fig. 2.4 compares the RMSE performance of the adaptive waveform se-
lection schemes presented in Section 2.5 versus two fixed waveforms with
minimum and maximum time-bandwidth (TB) products. The compari-
son also includes the adaptive waveform selection scheme whose decision
Figure 2.4 Performance comparison of various waveforms for IMM-EKF.
is made based on minimizing the trace of the observation error covari-

ance matrix Rk+1 . Note that the performance deterioration occurring at
t = 11–12 s is caused by the turning motion of the target during that period.
We observe from Fig. 2.4 that the adaptive waveform selection schemes,
which minimize the traces of the most-likely and combined covariance
matrices, exhibit the best RMSE performance among the waveforms un-
der consideration. Specifically, after the turning motion of the target, the
RMSE performance of these two selection schemes recovers much faster
than the other waveforms. It is interesting to observe from Fig. 2.5 that
these selection schemes select alternating sequences of λ and F during
the track initialization period and the track recovery period for the target
turn before the selected waveforms settle down to the maximum TB wave-
form. In contrast, the minimum TB waveform produces the worst tracking
performance. Although outperforming the minimum TB waveform, the
maximum TB waveform cannot provide a tracking performance as good
as the adaptive waveform selection schemes based on the most-likely and
combined covariance matrices.
The adaptive waveform selection scheme that minimizes the trace of the
minimax covariance matrix does not perform very well compared to the
other two selection schemes mentioned above. The main reason for this is
that the objective of this selection scheme is to minimize the trace of the
‘largest’ covariance matrix while the ‘largest’ covariance matrix may not
Figure 2.5 Patterns of selected waveform parameters for different adaptive waveform
selection schemes. (A) Minimizing combined state covariance; (B) Minimizing most-
likely state covariance; (C) Minimizing minimax state covariance; (D) Minimizing obser-
vation covariance.
be the best representative covariance matrix for the IMM algorithm. It is

shown in Fig. 2.6 that the dynamic model which has the largest covariance
trace often has a lower probability to be the correct model.
We also observe that the adaptive waveform selection scheme based
on the observation error covariance matrix exhibits a poor RMSE per-
formance. A similar observation was obtained in the monostatic radar
setting [28]. In particular, the RMSE of this selection scheme is almost
identical to that of the minimum TB waveform. This can be explained by
noting that this scheme always selects the minimum TB waveform as shown
in Fig. 2.5.
IMM-EKF versus EKF

We now examine the performance of the adaptive waveform selection
scheme if only one EKF is utilized for target tracking. Here, the EKF makes
Figure 2.6 Model probabilities and traces of state covariances for the uniform-motion
and maneuvering-motion models if the waveform selection is performed based on the
minimax state covariance matrix.
use of a nearly constant velocity model with q = 10 m2 /s3 . In this case, the
most-likely, minimax and combined covariance matrices become the same
as only one target dynamic model is used. Fig. 2.7 plots the RMSE per-
formance of the adaptive waveform selection method in comparison with
the minimum and maximum TB waveforms. Although the adaptive wave-
form selection outperforms the fixed waveforms in the first ten seconds,
it performs poorly after the target turn. In addition, we also observe that
the alternating sequences of λ and F no longer occur during the track
recovery period after the target turn as in the IMM-EKF case. We can
explain this by noting the inconsistency between the RMSE of the target
state estimate and the trace of the target state error covariance matrix as
shown in Fig. 2.7. We can observe that the target state error covariance
matrix Pk|k computed by the EKF does not reflect the degradation in the
real RMSE performance arising from the event of target turn. Unfortu-
nately, the waveform selection scheme decides the waveform purely based
on this covariance matrix, and thus is not able to adapt the waveform to
cope with the variation in target dynamics. This consequently results in a
poor RMSE performance of the waveform selection scheme when a varia-
tion in target dynamics occurs. Fig. 2.8 clearly demonstrates the advantage
of the use of IMM in conjunction with adaptive waveform selection.
Figure 2.7 Simulation results with EKF. (A) RMSE performance; (B) Target state error co-
variance; (C) Patterns of selected waveform parameters.
Figure 2.8 Performance comparison between IMM-EKF and EKF.
Multistatic radar versus bistatic radar

This simulation compares the RMSE performance between the multistatic
radar system and four individual bistatic radars. All cases utilize the IMM-
EKF algorithm jointly with the adaptive waveform selection scheme that
Figure 2.9 Performance comparison between multistatic and bistatic radars.
minimizes the trace of the combined covariance matrix. We observe from

Fig. 2.9 that, by exploiting radar measurements from all receivers, the mul-
tistatic radar system significantly outperforms the bistatic radars, particularly
during the variation in target dynamics between t = 11 s and 12 s.
2.7 Summary
This chapter has considered the problem of adaptive waveform selection
for multistatic tracking of a maneuvering target. Three waveform selection
schemes were presented based on minimizing the traces of the most-likely,
minimax and combined state estimation error covariance matrices of the
IMM algorithm. The performance advantages of the presented waveform
selection schemes over conventional fixed waveforms were demonstrated
by Monte Carlo simulation examples. The selection schemes based on the
most-likely and combined covariance matrices were observed to exhibit
the best RMSE performance and thus are the algorithms of choice among
all the waveform schemes considered.
2.8 Appendix
The Jacobian matrix H k of yk with respect to xk is
H k = [H T1,k , . . . , H TN ,k ]T , (2.43)
where
⎡ ⎤
∂ d i ,k ∂ d i ,k
⎢ ∂ px,k 0 0 ⎥
⎢ ∂ py,k ⎥
⎢ ∂ρi,k ∂ρi,k ∂ρi,k ∂ρi,k ⎥
H i ,k = ⎢
⎢ ∂ px,k
⎥ (2.44)
⎢ ∂ py,k ∂ vx,k ∂ vy,k ⎥
⎥
⎣ ∂θi,k ∂θi,k ⎦
0 0
∂ px,k ∂ py,k
for the state vector xk = [pTk , vTk ]T , or

⎡ ⎤
∂ d i ,k ∂ d i ,k
⎢ ∂ px,k 0 0 0 0⎥
⎢ ∂ py,k ⎥
⎢ ∂ρi,k ∂ρi,k ∂ρi,k ∂ρi,k ⎥
H i ,k = ⎢
⎢ ∂ px,k 0 0⎥
⎥ (2.45)
⎢ ∂ py,k ∂ vx,k ∂ vy,k ⎥
⎣ ∂θi,k ∂θi,k ⎦
0 0 0 0
∂ px,k ∂ py,k
for the state vector xk = [pTk , vTk , aTk ]T . The derivative terms in (2.44)
and (2.45) are given by
∂ d i ,k px,k − tx px,k − rx,i
= + , (2.46a)
∂ px,k pk − t pk − ri
∂ d i ,k py,k − ty py,k − ry,i
= + , (2.46b)
∂ py,k pk − t pk − ri
∂ρi,k (py,k − ty )(vx,k (py,k − ty ) − vy,k (px,k − tx ))
= (2.46c)
∂ px,k pk − t3
(py,k − ry,i )(vx,k (py,k − ry,i ) − vy,k (px,k − rx,i ))
+ , (2.46d)
pk − ri 3
∂ρi,k (px,k − tx )(vy,k (px,k − tx ) − vx,k (py,k − ty ))
=
∂ py,k pk − t3
(px,k − rx,i )(vy,k (px,k − rx,i ) − vx,k (py,k − ry,i ))
+ , (2.46e)
pk − ri 3
∂ρi,k px,k − tx px,k − rx,i
= + , (2.46f)
∂ vx,k pk − t pk − ri
∂ρi,k py,k − ty py,k − ry,i
= + , (2.46g)
∂ vy,k pk − t pk − ri
∂θi,k py,k − ry,i
=− , (2.46h)
∂ px,k pk − ri 2
∂θi,k px,k − rx,i
= . (2.46i)
∂ py,k pk − ri 2
CHAPTER 3
Waveform selection for

multistatic target tracking in
clutter
Contents
3.2. Tracking algorithm with probabilistic data association 35
3.5. Summary 43
3.1 Introduction and system overview

This chapter considers adaptive waveform selection for multistatic target
tracking in a cluttered environment where the target detection probabil-
ity is less than one and unwanted false-alarm measurements are introduced
by clutter. In this chapter, we consider a multistatic radar system with one
transmitter (located at position t) and multiple receivers (located at posi-
tions ri , i = 1, 2, . . . , N) as depicted in Fig. 3.1. Each receiver incorporates
a probabilistic data association (PDA) algorithm [35] to obtain a local track.
The local tracks from all receivers are communicated to the transmitter,
which then performs track combining and waveform selection. The ad-
vantages of this distributed tracking scheme are twofold. First, local target
tracking processes are performed in parallel by all receivers to reduce com-
putation time. Second, the communication load between the transmitter
and receivers is significantly reduced since only the local track informa-
tion is transferred from the receivers to the transmitter instead of the raw
measurement data, which may contain a large amount of false alarms.
For the sake of simplicity, we consider a single model for the target
dynamics:
xk+1 = Fxk + wk , wk ∼ N (0, Q) (3.1)
even though it is straightforward to extend the problem to maneuvering
target tracking using the IMM algorithm as in Chapter 2.
Assuming that the time delay and Doppler shift measurements (i.e., τ̃i,k
and ˜ i,k ) are available at each receiver i at time instant k, the total bistatic
https://doi.org/10.1016/B978-0-12-815314-7.00011-1 All rights reserved. 33
Another random document with
no related content on Scribd:
ground in a pouring rain, with the result that he caught a violent chill,
pneumonia supervened and carried him off within the week.
As soon as he was buried the daughter shook the dust of
Brinkstone off her feet; she had been profoundly unhappy ever since
the Larchesters had left.
“She was civil enough to all the folk around, but she had made no
friend in the place except Miss Lettice,” the old man explained. “So
we didn’t hear anything of her after she left any more than we did of
the other two. But I did hear from a theatrical gentleman who stayed
here for a few days a couple of years later, and with whom I used to
gossip a bit, as I’ve done with you, sir, that there was a Miss Alma
Buckley on the stage, and from the description he gave me of her I
should say it was the same. I heard from Mr. Larchester that she was
very fond of play-acting, but that the old man was a bit strict in his
notions, a regular attendant at church and all that sort of thing, and
he kept a pretty tight hand on her.”
Sellars pigeon-holed the name in his memory. This Alma Buckley
might be useful to him if he could get hold of her. On the stage,
according to Dobbs—well, she could not be a particularly well-known
actress, or he would have heard of her, as he was a great theatre-
goer. And besides, all this happened a great many years ago; it was
very unlikely she was still pursuing her theatrical career.
“Now tell me, Dobbs, after the Larchesters left, do you know if the
friendship was kept up at all—I mean, of course, in the way of
correspondence?”
Mr. Dobbs answered this question in the affirmative. Mrs.
Simpson, the then landlady, used to chat a little with Alma Buckley
when they met in the village, and he distinctly remembered being
told some three months after the Larchesters had left, that at one of
these meetings the girl had mentioned she had heard from her
friend, and that Mr. Larchester was going from bad to worse, and
that things were growing very hard for them, in a pecuniary sense.
Sellars went back to London and, of course, paid an early visit to
the detective. It had been arranged that he should not write during
his absence, but deliver his news in one budget on his return.
“Well, you’ve got some very important information,” said Lane
when the young man had finished. “Strange that I should have been
suspicious of that nephew story almost from the beginning. Now, it is
evident there is some mystery in which both Mrs. Morrice and Sir
George are concerned, which has led them to concoct this lying tale.
And this young Brookes, if he is not the relation it is pretended he is,
who and what the deuce is he?”
“That’s what we’ve got to find out,” said Sellars. “This Mrs. Morrice
is evidently a queer fish, and, of course, her pretended brother-in-law
is another. I suppose Morrice really knows very little about her; they
say he married her abroad, and, of course, she could fudge up any
tale, mixing up truth and fiction as she liked.”
“And yet Morrice must be a shrewd old bird, or he wouldn’t be
where he is,” observed the detective. “You would think before he
married a woman he would have made exhaustive inquiries about
her. Unless, of course, he does know certain facts, and winks at the
nephew business, thinking it doesn’t concern anybody but
themselves. But, of course, I incline to the belief he doesn’t know. He
has the highest reputation for integrity, and it is more than
improbable he would lend his countenance to such an imposture,
even if it were an innocent one, which I very much doubt.”
“So do I,” agreed Sellars. “Now as soon as I can I will get hold of
this Alma Buckley and see if she knows anything, and if so, if she will
impart it to me. But I am not very hopeful in that quarter. It’s a deuce
of a long time ago, and she may be out of business or dead.”
“If she’s dead, of course we are done. But as long as people are
alive we can generally get at them sooner or later,” said Lane with a
knowing smile.
“I quite agree, they can’t hide themselves for long if one is
sufficiently persevering. Well, now about this Clayton-Brookes. We
have established that he is as queer a fish as Mrs. Morrice; we want
to know a good bit about him, don’t we?”
“What is the general report about him—I mean, of course,
amongst the circles in which he moves?”
Sellars paused a moment or two before he answered. He had
heard a good deal about the man, of course, but up to the present he
had not taken any particular interest in him.
“The general impression is that he is very well-off, not from his
property, because it is well-known that was so heavily encumbered
by his father that it would take more than a lifetime to redeem it. He
is supposed to have come into a fortune from a man named Clayton,
whose name he assumed, either out of compliment or because it
was a condition imposed.”
“Have you ever heard any details of this man Clayton, who was so
obliging as to leave him a fortune?”
Sellars shook his head. “None. You know how easily people
swallow a story when it is properly prepared and ladled out in a
circumstantial way. Clayton may be as much a myth as the nephew
for aught we know. You see how readily that has been accepted. You
would say, at first hand, that a man would be afraid to invent a
marriage in his own family; that there would be dozens of people
who knew Archibald Brookes had no wife, and would come forward
and say so.”
“He was helped in that case by the man having cut himself adrift
for so many years, that nobody was likely to know anything about
him. But now concerning this Clayton—if we could get to know who
the man was; there is such a place as Somerset House, there are
such things as wills there. We could soon get what we wanted.”
“I’ll try my old friend at White’s,” suggested Sellars, which in a few
days he did, but not with any brilliant results. All he could learn was
that the man Clayton was a very distant connection of the Brookes
family, that he had made his money in sheep-farming in Australia.
“Obviously he knows just what Clayton-Brookes thinks it is good
for him to know,” observed the young man when he reported to
Lane; “and he has been told in order that he may communicate his
knowledge to anybody who is a bit curious. We are done, I am
afraid, in that direction.”
“I agree,” was Lane’s rejoinder. “Well, the resources of civilization
are not yet exhausted, as was once remarked by a very famous
man. I must employ other methods. Now, of course, you don’t
happen to know the name of his bank?”
“No, I don’t, but I can get it like a shot. He deals with the same
bookmaker as I do, but in a much larger way. We are great pals, my
‘bookie’ and I; I’ve done him several good turns in the way of
information about people who want to open accounts with him. He’ll
tell me for the asking.”
That was the great utility of Sellars in such a complicated business
as that of Lane’s. If he could not give you the precise information you
required, his acquaintance was so varied, his ramifications were so
wide, that he could get it for you from somebody else in a very short
time.
Within a couple of hours the detective was informed that Sir
George Clayton-Brookes banked at the Pall Mall Branch of the
International Bank.
Mr. Lane reached for his hat. “I’ll just step down to my man and put
a little inquiry through as to the gentleman’s financial status.
Fortunately for me, the sub-manager is in charge just now, and like
you and your ‘bookie,’ we are great pals. He’ll do more for me than
the manager, who is a very orthodox person and a bit of an old
stick.”
The report came in double-quick time. The wealthy Sir George,
who betted high and gambled for big stakes according to general
rumour, was considered by the custodians of his money not to be
good enough for five hundred pounds.
“Another illusion shattered,” said the detective with a grim smile
when he next saw young Sellars. “This promises to be a very
interesting case. We are unearthing a few queer things, aren’t we?
The Clayton business is a myth, of course; there never was such a
person, or if there was, he never left a fortune to our friend. It is
admitted that his income from his estates is practically nil, and the
evidence of your very useful waiter confirms that. We also know that
he passes off a spurious nephew, for some sinister purpose
obviously. The man is a ‘wrong-un’ and lives by his wits, that is pretty
evident.”
Sellars could not help laughing. It was a bit comical to find that the
magnificent Sir George was not good enough for five hundred
pounds. Sellars’ bankers would have given him a reference for that
amount, and he lived by his wits too. But then it was in a respectable
way, and he did not invent spurious relations.
“I think we had better set about old Morrice himself next,” he said.
“What’s the odds on finding something fishy about him, in spite of his
high reputation?”
Lane smiled. “By gad, when you’ve been in this line as long as I
have, I’m hanged if you’ve got much belief left in anybody. It is
marvellous the queer things we do unearth, many of them of little
actual importance to the case, when we once start a long
investigation.”
“Well, what’s the next move on the board?” queried his colleague.
He began to feel great interest in the Morrice mystery; if it went on as
it had begun, there promised to be some surprising developments.
He was not so astonished about Sir George. He was not popular,
partly, perhaps, on account of the wildness of his youth, and Sellars
himself had been repelled by the man; he had always thought there
was something a little sinister about him.
But the discovery that Mrs. Morrice, that pleasant, gracious
woman who made such an admirable hostess in the big house in
Deanery Street, was a party to such an extraordinary fraud, had
fairly taken his breath away. He recalled the old waiter’s admiration
for her as a girl, of his pity for her lonely life, his disgust with her
soddened father. She must have changed very much from the girl
who lived in the little village of Brinkstone and ate her heart out in
these sordid surroundings.
“I hardly quite know,” was Lane’s answer to the question put by his
young lieutenant. “I want thoroughly to digest all that very important
information you got for me, and make up my mind as to how we are
going to utilize it. But certainly one of the first steps is to discover
who this so-called Archie Brookes really is. It only wants two days to
Christmas; I’m not going to work on it any more till after the holidays.
Then you’ll try to get into touch, if possible, with this woman Alma
Buckley, who is a very strong link with Mrs. Morrice’s past.”
Lane was spending his Christmas in the bosom of his family.
Sellars, as became a young man of his position, was due at a smart
country house. They would meet in the New Year.
Richard Croxton passed the festive season with his kind old nurse.
Rosabelle kept up a smiling face that hid a very aching heart in
Switzerland. It was not a cheerful Christmas for any member of the
Morrice family; they missed greatly the familiar figure that had been
with them for years.
CHAPTER X
THE SAFE IS ROBBED AGAIN
I N the New Year, Sellars, having spent a most enjoyable

Christmas, and fortified both in spirits and body by the season’s
jollity and good cheer, set to work to discover whether Miss Alma
Buckley was still in the land of the living, and if so, where she was to
be found.
It has been said that nobody knew better how to set about getting
information that he was in need of than this agreeable young man-
about-town, who had never been credited by his ordinary
acquaintances with ability beyond the average.
Amongst his various clubs was a very unpretentious and
Bohemian one called “The Strollers.” As its name implied, its
members were mainly recruited from the theatrical profession, but it
also admitted within its portals musicians, artists, journalists, authors,
and a few people who were great admirers of the arts but did not
practise any of them as a means of livelihood. If Miss Alma Buckley
was still in the profession, he would find somebody here who knew
her, or at any rate knew of her.
We know that, strictly speaking, his proper milieu was the
fashionable world, his proper place for relaxation a club like White’s
or Boodles’; but he was a young man of catholic tastes, and he was
also entitled to call himself a journalist, if his activities in that
profession were not very great. He was also very fond of people who
“did something,” whether in music or art or literature. Therefore “The
Strollers” suited him very well when he got a bit bored with exclusive
society, and the rather banal talk of fashionable and semi-
fashionable people.
The subscription was very moderate, the entrance fee equally
reasonable; he met there men who could talk well, a few quite
brilliantly. Once a week during certain seasons they held an
entertainment at which there was quite a respectable array of talent.
To this very delightful little place he repaired one evening in search
of information about Miss Alma Buckley.
He inquired of two theatrical members, not of the very highest rank
in their profession, but neither of these gentlemen had ever heard of
the lady in question. They suggested that she might probably be on
the provincial stage.
The third time, however, he was more lucky. He came across a
rather well-known music-hall artist, one Tom Codlin, who reddened
his nose and leaned decidedly to the vulgar side on the boards, but
who was a very quiet, decorous fellow off. He knew the name at
once.
“Alma Buckley, of course, known her since I first took up the
business; must be a good ten years older than I am, I should say,
makes up wonderfully, too. Saw her a few months ago in one of the
Stein shows, and I was surprised to see how well she wore. No
particular talent, no particular line, but generally gets an
engagement, even when cleverer people are out.”
“Has she ever been on the stage, the real stage, I mean?” asked
young Sellars. Mr. Codlin shook his head.
“Never. She started in the halls when quite a young girl and has
stuck there ever since.”
“Do you know if she’s playing now, and if so, where?”
Codlin had no idea; he had not seen her name for some time in
any of the bills. She might be in the provinces, she might have gone
for a tour abroad.
He thought a moment, and then added: “The best thing you can do
if you want to get hold of her is to go to her agent, ‘Mossy’
Samuelson, as we call him. I’ll scribble his address on my card; he
knows me well. That will get you the entrée at once, for he’s an
awfully busy chap, and if he doesn’t know you, will keep you waiting
for hours.”
The next day found Sellars presenting his club friend’s card to a
small, sharp-looking boy in the rather dingy front room of a house in
a street off the Strand. A communicating door led to Mr. “Mossy”
Samuelson’s private sanctum, where he received his clients. A lot of
women, mostly young, but a few middle-aged, were waiting to see
the great man. Sellars thought that if all these people had to go in
before him he would have to wait for hours. He did not of course
know the ways of busy theatrical agents, that they do not see half
the people who are waiting for an audience, only come out and
dismiss most of them with a brief “Nothing for you to-day.”
“Tell him I won’t keep him a minute,” he whispered to the sharp-
looking boy, slipping half a crown into his rather grubby but
appreciative palm.
Mr. Codlin’s card had a magical effect. In less than a minute the
boy appeared in the opening of the doorway and beckoned him in. A
gentleman with a pronounced Hebraic aspect sat in solemn state at
a big table, wearing the shiniest tall hat that Sellars had ever seen on
a human head. He doffed this resplendent article when he observed
the young man remove his own.
“Good-morning, sir; good-morning. Sorry I can’t give you more
than a minute or two. I’ve got three contracts to draw before one
o’clock, and there’s half the music-hall profession waiting in the other
room to see me. If it hadn’t been for Mr. Codlin’s card, couldn’t have
given you a second.”
In view of Mr. Samuelson’s evident importance, Sellars adopted a
most respectful tone. “Very kind of you, sir, very kind indeed. I will
come to the point at once. You are the agent of Miss Alma Buckley, I
am told.”
“I am, sir; been her agent for the last twenty years. What can I do
for you?”
“I want to see the lady on some private and important business.
You could not, of course, give me her address?”
“We never give addresses of our clients; clean against the rules,
sir.” The little, keen, beady eyes looked at him inquiringly. The young
man belonged to White’s Club and looked what Mr. Samuelson
would describe in his own words, “a toff.” What could such a person
want with a middle-aged, undistinguished music-hall artist?
“I quite appreciate that, Mr. Samuelson. Would you be good
enough to forward a letter for me?”
“Very ’appy, sir,” replied the Hebraic gentleman affably. “But it’s no
good your sending it yet. Miss Buckley is returning from South Africa;
at this moment she’s on the ocean, and she’s not due in London for
another ten days. Send it then, and I will take care it reaches her.
Good-morning, sir; ’appy to have met you.”
He held out a podgy hand, and the interview terminated. It was a
bit of a check, this waiting for ten days, for Sellars was getting very
keen on the Morrice case. But there was no help for it, and it was
always on the cards that Miss Buckley might refuse to receive him,
or if she did, might decline to give information about her old friend.
Rosabelle returned to London with her uncle and aunt, very glad to
get home again. Under ordinary conditions she would have enjoyed
herself hugely at Mürren, for she was a thorough open-air girl, and
delighted in every form of sport. But the sight of other people’s gaiety
made her sad when she was so miserable herself. Mrs. Morrice, too,
seemed very unhappy and restless during what should have been
such a festive season. Rosabelle thought that Mrs. Morrice must
have been fonder of Richard than she had believed.
The first visit she paid, even before she went to see her lover, was
to the offices of Gideon Lane. This man, with his strong resolute
face, was her only hope; she had longed to be back in London so
that she might be near him; his propinquity to her gave her a sense
of comfort.
“I don’t want to make myself a nuisance, Mr. Lane, but I simply
could not keep away,” she explained by way of greeting. “You have
not been idle during our absence, I am sure. Are you any nearer to
discovering the true criminal? Have you found out anything at all?”
It was an awkward question for the detective to reply to. A very
great deal had been discovered during the time that had elapsed
between her departure for Mürren and her return to London; startling
facts at present known only to himself and Sellars.
If she had been a hard-headed practical man instead of an
emotional girl wrought up to a pitch of almost unendurable tension by
the serious plight of her lover, he might have been disposed to make
a clean breast of it. But for the moment he dared not trust her.
Guided by her feelings, she might act impulsively and spoil all his
plans.
“I will be frank with you, Miss Sheldon, as far as I can be, as I dare
be, at this juncture. Certain things have been discovered of
considerable importance. What they are, the precise nature of them,
even a hint, I dare not indulge in for the present, not until I know
much more.”
She knew the man well by now; it was useless to attempt to shake
his determination. When he had once made up his mind, it was like
beating against an open door.
“Will you at least tell me this, to ease my suspense,” she said at
length. “What you have discovered so far, does it tell against or in
favour of Mr. Croxton?”
There was a perceptible pause before he answered. Caution was
so ingrained in the man, his habit of carefully weighing every word,
his dread of expressing an opinion before he was fully justified, had
become so deeply rooted that he could hardly ever exhibit complete
frankness. Optimism was, of course, a mood unknown to him.
“If certain nebulous suspicions which are slowly forming in my
mind turn out to be correct, the result of my discoveries, so far, is
rather in favour of Mr. Croxton.”
That was all she could get out of him for the present, all she
would, in all probability, ever get out of him till he had fully and finally
solved the Morrice mystery. It was not great comfort, but it was better
than nothing.
The next day she went to Petersham, and received a warm
welcome from her lover, whose heart had been aching for her during
those weary days of separation. Small as were the crumbs of
comfort which Lane had given her, she made the most of them, and
heartened Dick considerably by her assurance that all would come
right in the end.
“There are things about the man that irritate me—just an impulsive
woman with more heart than head—his slowness, his caution, his
dislike to speak positively; but I do believe in him, in his capacity, his
ability. If the mystery is to be solved by human agency, I am
convinced he will solve it.”
The lovers had a fairly happy day, considering the depressing
circumstances. They took a long walk through busy Kingston, over
the Thames glistening in the winter sunshine, to Hampton Court,
where they had lunch at that best of old-fashioned hotels, the Mitre.
They got back to Petersham late in the afternoon, and Richard’s
kindly old nurse had a dainty tea ready for them. And too soon the
hour of parting came and her lover put her in a taxi and kissed her
fervently as they said good-bye.
“God bless you, my darling; need I tell you how I appreciate your
faith in me? But for your visits, and I count the hours till you come, I
think I should go mad. And yet I must not reckon on them. One day
your uncle will forbid them.”
The steadfast girl smiled bravely. “We will talk of that when the
time comes, my poor old boy.”
She returned his fond caress. “Remember, Dick, whether Lane
succeeds or fails, it will make no difference to me. We are
sweethearts now, and we are going to be sweethearts till the end.”
During her long absence startling events had happened in the big,
old-fashioned house in Deanery Street.
Gideon Lane had spent a busy morning away from his office. It
was three o’clock before he found time to snatch a hasty lunch.
When he got back it was close upon four. His clerk had an urgent
message for him.
“Mr. Morrice of Deanery Street has rung up three times during your
absence, sir. The last time he left word for you to go round as soon
as you came in. He said it was of the utmost importance.”
A taxi soon conveyed the detective to the financier’s house. He
found Morrice in his room in a great state of anger and excitement.
“Another robbery, Mr. Lane, this time a small one. A bundle of
Treasury notes and a quantity of Swiss bank-notes have been
abstracted, to the value of two hundred and eighty pounds. This time
I am determined to get to the bottom of it. If you are agreeable, you
shall act for me as well as for the other parties. You have no
objection to that, I suppose?”
Lane bowed. “None at all, sir. Whoever employs me does so with
the same object—to bring home the guilt to the right person.”
There were finger-marks on the safe as before. These were duly
photographed. They were identical with the previous ones, those of
the expert safe-breaker known as “Tubby” Thomas.
And “Tubby” Thomas, as they knew beyond the possibility of
doubt, was safely locked up in Dartmoor.
CHAPTER XI
A RIFT IN THE CLOUDS
C OULD there be two men the whorls of whose fingers were the
same? Scotland Yard says impossible.
“I do not question that this is a very wonderful safe, Mr. Morrice,”
the detective remarked quietly to the financier in a subsequent
conversation, “but it is evidently not as invulnerable as you and the
makers thought it.”
“Once know the secret of its mechanism, and the rest is easy,”
retorted Morrice, a little nettled at this depreciatory reference to the
wonderful invention, to the perfecting of which he had contributed not
a little himself with his own ingenious suggestions.
He explained to Lane a few of its marvels. To begin with, it was the
only safe of its kind that had ever been manufactured. The
combinations of the times when it would open would run into
millions. Supposing you worked on tens, for instance, that is ten,
twenty, thirty, the days of the month, it would only open if both keys
were applied to the same keyhole!
The detective listened politely, but he was not very interested in
learning how the thing worked; his object was to find out if there was
anybody besides young Croxton and his employer who could have
become acquainted with the secret of its working.
“It seems to me that it is really a matter of exercising ordinary
common-sense,” observed the angry banker. “Two men alone know
the secret, myself and Richard Croxton, therefore either of us could
open it after having obtained surreptitiously a duplicate of the key
held by the other. Let us assume, for a moment that you, acting on
Croxton’s behalf, say that I was the thief, that from some sinister
motive I stole my own property. Well, you are perfectly entitled to that
opinion, as an opinion.”
“I have not expressed it,” said the detective quietly.
“I know you haven’t,” snapped Morrice. He was in a very angry
mood to-day, and inclined to let his temper run away with him. “But I
also know that gentlemen of your profession cast your net wide
when you start, in the hope of catching some very unlikely fish. Of
course, I could have opened it and cast it upon Croxton, if I were so
disposed. But where is my motive for robbing myself? I can
understand certain circumstances which might induce a man to
commit such an act, and cleverly provide a scapegoat. Men set fire
to their own warehouses to get the insurance money. Why? Because
they are on the verge of bankruptcy and that money will save them.
A desperate man might steal his own money for similar reasons, to
place it beyond the reach of his creditors, so that he should not go
forth to the world a beggar. But these motives are absent in my own
case. I am more than solvent; I don’t wish to speak in any boasting
spirit, but I have more money than I know what to do with.”
Lane thought for such a practical man of the world, and
possessing, as he did, such a clear logical brain, he was indulging in
rather superfluous observations. Besides, he had referred to one
aspect of the situation as it affected himself—the absence of
financial embarrassment. If one chose to argue with him, one could
cite from the annals of crime instances of more than one other
motive that had impelled men to commit crimes which they artfully
fixed upon innocent persons, whom for some subtle reason they
wished to remove from their path, or on whom they desired revenge.
The next words, however, showed why the banker had
volunteered such an elaborate defence of himself.
“I am therefore eliminated, at any rate to my own satisfaction.” His
tone was still angry, as if he inwardly resented Lane’s rather
lukewarm attitude. “There is only left this young man whom I have
treated as a son, whom I have loaded with benefits, whom I have
preserved from the consequences of his criminal acts, his dastardly
ingratitude to me. He alone, beyond myself, knew the secret of this
safe’s mechanism, therefore he alone could open it, unless, which of
course is possible, he employed a confederate whom he took into
his confidence. Is that common-sense or not, Mr. Lane?” he
concluded in a slightly calmer tone.
“Perfectly common-sense so far as it carries us, Mr. Morrice,” was
the detective’s judicial reply.
“So far as it carries us,” cried Morrice with a slight return of his
previous explosive manner. “I do not understand you.”
Lane smiled. It was a slightly superior smile, prompted by the
thought that these clever business men, excellent and keen as they
were in their own pursuits, did not exhibit that logical mind which is
the great equipment of a trained investigator.
“By that expression I mean simply this, that when you definitely
assume Mr. Croxton’s guilt you are acting on the presumption that
nobody but you and he knows the secret of this mechanism. Can
you prove that?”
“Of course I can’t prove it in a way that might satisfy you, but I do
know that I have never told anybody else. There is of course the
maker,” he added sarcastically. “Perhaps you are including him in
your calculations.”
“I think I will consent readily to his elimination,” replied Lane in his
quiet, not unhumorous way. “Now, Mr. Morrice, we will discuss this
matter without heat. I am now employed by you as well as Mr.
Croxton, and I have only one object in view. But I must conduct my
investigation in my own way, and I want to go a little deeper than we
have yet gone.”
Morrice was impressed by the grave authoritative manner of the
man; he showed a strong touch of that quality which we notice in
eminent judges, successful barristers and all properly qualified
members of the legal profession, a patient pursuit of facts, a strongly
developed power of deduction from whatever facts are presented, a
keen faculty of analysis.
“Now, Mr. Morrice, I shall ask you a question or two for my own
enlightenment. From the little you have explained to me, the
mechanism of this safe is extremely complicated. Did you and Mr.
Croxton carry all the details of it in your heads, or had you some
written memorandum of its working to refer to in case of a temporary
lapse of memory?”
Morrice was quick enough immediately to see the drift of that
question, and his manner changed at once. “Thank you for that
suggestion, Mr. Lane; I fear I have shown a little impatience. For all
practical purposes, we did carry it in our heads, but I have in my
possession, as you surmise, a written key to which reference could
be made in the event of our requiring an elaborate combination.”
“And that key is still in your possession?”
“Yes. I keep it in the safe in my dressing-room. I looked this
morning, and it was there.”
The detective ruminated over this latest piece of information. While
he was doing so, Morrice spoke again, with just a little hesitation, as
if he knew that what he was going to say would cut the ground a little
from under his strongly expressed theory of Croxton’s guilt.
“You ought to know the whole truth, Mr. Lane, and you shall have it
for what it is worth. This present memorandum—I will speak of it by
that name—is one that I wrote out from memory. I had an original
one, perhaps just a trifle fuller, but I lost it, that is to say I could not
find it amongst my papers, some two years ago.”
Still clinging obstinately to his theory, he added a few comments
which, needless to say, did not make much impression on his
listener, who went into possibilities and probabilities perhaps just a
trifle too elaborately for the ordinary man.
“You know how easily papers get lost or mislaid. It is as likely as
not that the original memorandum will turn up in the last place I
should expect to find it. And if it got loose and was swept up by some
careless servant, it would get into the hands of the dustman. To the
ordinary person it would, of course, be quite unintelligible.”
To this Lane simply remarked that when a paper of importance
had disappeared it was quite impossible to prophesy into whose
hands it would fall. The dustman was a comforting theory, but it was
no part of his business to adopt comforting theories that did away
with the necessity to think. If “Tubby” Thomas had not been safely
locked up at Dartmoor for the last two years, he would have been
pretty certain that by some felonious means it had come into the
possession of that accomplished safe-breaker.
His position had changed since Mr. Morrice had summoned him to
Deanery Street on the occasion of the second burglary. He was now
representing the financier as well as Richard Croxton. In a way he
was glad, for Richard Croxton was poor according to Rosabelle, and
this promised to be an expensive investigation. To Morrice money
did not matter; he would not be stopped from ascertaining the truth
by lack of funds.
But in another way it was awkward. They had already found out
about Mrs. Morrice that, in conjunction with Sir George Clayton-
Brookes, the supposed wealthy baronet, she was passing off as her
nephew a young man who had no claim to the title. In course of time
Morrice would have to be acquainted with that suspicious fact, and
whatever degree of affection the banker felt for his wife, whether he
loved her very much or hardly at all, it would be a terrible blow to
him, either to his love or pride, or both.
Lane had a long talk with Sellars over the latest development of
the Morrice mystery. The young man strongly maintained that it
greatly strengthened the presumption of Croxton’s innocence, and
although the detective, with his usual habit of caution, did not take
quite such a decided view as his more impetuous lieutenant, he
readily admitted that it told in his favour, that any man possessing
the legal mind must concede as much.
“The more I can find out about him and his habits,” Sellars
remarked, “the more it seems unlikely, although not, of course,
impossible, that he should have done this thing. As far as I can learn,
he has been in love with Miss Sheldon for years, and his life has
simply been bound up with the Morrice family. They entertained very
largely, and he always showed up at their entertainments, was at
every dinner-party they gave, just like a son of the house. He seems
to have very few young men friends, but they are all of a most
respectable type. He doesn’t gamble, he doesn’t drink. Other women
don’t come into the case, for he is hardly ever a yard away from his
lady-love. Does he seem the kind of man to get himself into a hole
out of which he can only extricate himself by robbery?”
And Lane was forced to agree that if the good report of one’s
fellows could establish innocence, young Richard Croxton was
already satisfactorily whitewashed. But of course, in the opinion of
this eminent practitioner, all this was negative evidence, not positive.
Rosabelle, who was duly informed of the loss of the original
memorandum—for Lane was at bottom a very kind-hearted man and
thought he could give the harassed girl this crumb of comfort without
jeopardizing his future action—was very jubilant. She was also
pleased that her uncle had appointed the detective to prosecute his
investigations on his behalf. It would mean that Lane would not be
hampered for money.
She went over to Petersham the next day to tell her lover what had
happened, and succeeded in infecting him with her own hopeful
spirit.
“And is Mr. Morrice still as bitter against me as ever, does he still
believe as firmly in my guilt?”
Rosabelle was not very sure of the financier’s real thoughts, but
she gave the best answer she could.
“You know, my dear old Dick, how obstinately he clings to an idea
when he has once taken it into his head, but I fancy he is a bit
shaken.”
CHAPTER XII
SIR GEORGE’S VALET
I N a back street behind the busy thoroughfare of Piccadilly there is

a small, quiet-looking public-house which is a great rendezvous for
male servants of a superior class. Thither in their leisure moments
repair chauffeurs, butlers, valets in good service, to take a moderate
amount of refreshment—such men seldom drink to excess—to chat
over the news of the day, and very frequently to comment to each
other on the characters and doings of their employers.
On the evening following the day on which Lane had held that long
conversation with Mr. Morrice recorded in the last chapter, two men
sat in a corner of the snug little bar, drinking whisky and soda and
talking together in confidential tones.
The one party to the conversation had “gentleman’s gentleman”
written all over his manner and appearance. The deferential voice,
the trick of lowering his eyes when he spoke, proclaimed the valet
who had served in good families. He had not the smug pomposity of
the butler nor the breezy open-air demeanour of the driver.
His companion was no other than our old friend the detective, who
did not, like his companion, exhibit any signs of his calling, as far as
appearance went; he might have been anything—respectable
bookmaker, prosperous commercial traveller, well-to-do shopkeeper,
whatever you pleased.
In the pursuit of his professional duties Lane penetrated into
several circles, outside the higher ones. These he left to Sellars
principally, although he had two other occasional assistants of the
same class, but less zealous and capable. His object in coming to
this quiet little place to-night was to extend his acquaintance, one
formed a few nights previously, with the man who was drinking now
at his expense, who rejoiced in the popular name of Simmons and
was valet to Sir George Clayton-Brookes.

Signal Processing For Multistatic Radar Systems: Adaptive Waveform Selection, Optimal Geometries and Pseudolinear Tracking Algorithms Do Ançay

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Signal Processing For Multistatic Radar Systems: Adaptive Waveform Selection, Optimal Geometries and Pseudolinear Tracking Algorithms Do Ançay

Uploaded by

Copyright:

Available Formats

Signal processing for multistatic radar

systems: adaptive waveform selection,

NGOC HUNG NGUYEN

MATLAB® is a trademark of The MathWorks, Inc. and is used with permission.

Library of Congress Cataloging-in-Publication Data

British Library Cataloguing-in-Publication Data

For information on all Academic Press publications

Publisher: Mara Conner

About the Authors ix

Part 1. Adaptive waveform selection

2. Waveform selection for multistatic tracking of a maneuvering

3. Waveform selection for multistatic target tracking in clutter 33

4. Waveform selection for multistatic target tracking with Cartesian

5. Waveform selection for distributed multistatic target tracking 63

Part 2. Optimal geometry analysis

6. Optimal geometries for multistatic target localization with one

7. Optimal geometries for multistatic target localization by

Part 3. Pseudolinear tracking algorithms

8. Batch track estimators for multistatic target motion analysis 113

9. Closed-form solutions for multistatic target localization with

9.2. Maximum likelihood estimator and Cramér–Rao lower bound 146

Radar is an electromagnetic sensor for detecting and ranging objects by

closed-form tracking and localization algorithms for passive multistatic

X (i ) The ith diagonal element of matrix X

1.1 Historical background

Figure 1.1 An illustration of a multistatic radar system.

estimation performance. This type of data fusion is often considered as non-

Bistatic and multistatic radars can incorporate their own dedicated

mid-1990s, extensive research studies have been devoted to bistatic and

1.2 Purpose and scope

sented in this book is performance optimization via waveform adaptation,

1.3 Outline of book

Part 1: Adaptive waveform selection

transmitter–receiver communication links by utilizing only one Cartesian

Part 2: Optimal geometry analysis

consisting of multiple independent bistatic channels. The optimal angular

Part 3: Pseudolinear tracking algorithms

pendence must be taken into account in order to obtain efficient estimates.

allows the transmitted waveform to be adjusted on a pulse-to-pulse basis to

Waveform selection for

2.1 Introduction and system overview

Figure 2.1 System conﬁguration of multistatic tracking radar under consideration.

The transmitter position is denoted by t = [tx , ty ]T , and the receiver posi-

2.2 Bistatic radar measurements

Figure 2.2 Target tracking geometry in 2D.

2.3 Bistatic ambiguity function and Cramér–Rao lower

i.e., the time-bandwidth product of the transmitted signal is much smaller

C{τ,} (η, ψ) = −{τ,}

On the other hand, the CRLB for angle-of-arrival measurement ob-

Ci,k (ηi,k , ψ k ) = diag(C{τ,} (ηi,k , ψ k ), Cθ (ηi,k )) (2.7)

where ηi,k is the SNR at receiver i at time instant k, and ψ k is a vector

2.4 Target tracking

2.4.1 Target dynamic model

xk+1 = Fxk + wk (2.8)

where xk is the target kinematic state vector at time instant k, F is the

Nearly constant velocity model

and the process noise covariance matrix Q is

Nearly constant acceleration model

and the process noise covariance matrix Q given by

In the above model, ω is the angular turn rate and is presumed to be a

2.4.2 Observation model

d̃i,k = di,k + ed,i,k , (2.16a)

d̃i,k = c τ̃i,k , di,k = c τi,k , ed,i,k = ceτ,i,k , (2.17a)

PMost-likely,k+1|k+1 = Pm ,k+1|k+1 (2.31)

m = arg max μm,k+1|k . (2.32)

where is a library of possible waveforms. The waveform library may

where λ is the Gaussian pulse length parameter and b = F /(2Ts ) is