(D) Antenna Array Signal Processing For Multistatic Radar Systems - Belfiori - 2013

i
i
MainFrancescoThesis 2013/6/12 14:07 page i #1
Antenna Array Signal Processing for

Multistatic Radar Systems
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page ii #2
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page iii #3
Antenna Array Signal Processing for

Multistatic Radar Systems
Proefschrift
ter verkrijging van de graad van doctor

aan de Technische Universiteit Delft,
op gezag van de Rector Magnificus prof.dr.ir. K.C.A.M. Luyben,
voorzitter van het College voor Promoties,
in het openbaar te verdedigen
op vrijdag 12 juli 2013 om 10:00 uur
door
Francesco BELFIORI
Laurea Specialistica in Ingegneria delle Telecomunicazioni
Universit`
a degli studi di Roma La Sapienza
geboren te Roma, Italie
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page iv #4
Dit proefschrift is goedgekeurd door de promotor:

Prof.ir. P. Hoogeboom
Samenstelling promotiecomissie:
Rector Magnificus,
Prof. ir. P. Hoogeboom,
Prof. Dr. L. Ferro-Famil,
Prof. Dr. A. Yarovoy,
Prof. Dr.-Ing. J. Ender,
Prof. Dr. F. Le Chevalier,
Dr. W. van Rossum,
voorzitter
Technische Universiteit Delft, promotor
Universite de Rennes 1
Technische Universiteit Delft
Universitat Siegen - Fraunhofer FHR
Technische Universiteit Delft - Thales
TNO
This research was supported by TNO under contract DenV-017.

ISBN 978-94-6191-782-9
Antenna Array Signal Processing for Multistatic Radar Systems.
Dissertation at Delft University of Technology.
c 2013 by Francesco Belfiori.
Copyright
All rights reserved. No parts of this publication may be reproduced or transmitted in
any form or by any means, electronic or mechanical, including photocopy, recording,
or any information storage and retrieval system, without permission in writing from
the author.
Author e-mail: francesco.belfiori@gmail.com
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page v #5
A mia madre
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page vi #6
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page vii #7
Contents
1 Introduction
1.1 Passive Coherent Locator (PCL) systems . . . . . . . . . . . . . . . .
1.2 The Multiple-Input Multiple-Output radar concept . . . . . . . . . . .
1.3 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 Antenna theory and array pattern synthesis
2.1 Main antenna parameters . . . . . . . . . . .
2.1.1 Directivity . . . . . . . . . . . . . . .
2.1.2 Efficiency . . . . . . . . . . . . . . . .
2.1.3 Gain . . . . . . . . . . . . . . . . . . .
2.2 Antenna array pattern synthesis . . . . . . .
2.2.1 Linear array pattern synthesis . . . . .
2.2.2 Circular array pattern synthesis . . . .
2.3 Pattern synthesis in non ideal arrays . . . . .
2.3.1 Mutual coupling . . . . . . . . . . . .
2.3.2 Illumination errors . . . . . . . . . . .
2.4 Summary . . . . . . . . . . . . . . . . . . . .
1
4
5
7
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
9
10
10
11
11
14
17
20
20
21
22
3 Digital beamforming for PCL

3.1 Digital processing scheme . . . . . . . . . . . . . . . . . .
3.2 Mutual coupling compensation . . . . . . . . . . . . . . .
3.2.1 Analytical description . . . . . . . . . . . . . . . .
3.2.2 Optimisation approach for the C matrix evaluation
3.3 DBF for circular arrays . . . . . . . . . . . . . . . . . . .
3.3.1 Phase modes theory . . . . . . . . . . . . . . . . .
3.3.2 Proposed algorithm . . . . . . . . . . . . . . . . .
3.3.3 Array pattern comparisons . . . . . . . . . . . . .
3.4 Direct path interference suppression . . . . . . . . . . . .
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
23
23
24
24
26
27
27
29
31
34
35
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page viii #8
viii
4 PCL radar description and experimental results
4.1 Overview of the PCL system . . . . . . . . . . . .
4.2 Antenna array analysis . . . . . . . . . . . . . . . .
4.2.1 Array element characterisation . . . . . . .
4.2.2 Circular array characterisation . . . . . . .
4.3 PCL receiver . . . . . . . . . . . . . . . . . . . . .
4.3.1 Dynamic range analysis . . . . . . . . . . .
4.3.2 Digital down conversion . . . . . . . . . . .
4.4 Experimental results . . . . . . . . . . . . . . . . .
4.4.1 MC compensation . . . . . . . . . . . . . .
4.4.2 Direct path interference suppression . . . .
4.4.3 Range/Doppler processing . . . . . . . . . .
4.4.4 CFAR detector and plots extraction . . . .
4.5 System future improvements . . . . . . . . . . . . .
4.6 Summary . . . . . . . . . . . . . . . . . . . . . . .
Contents
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5 Coherent MIMO array theory

5.1 Coherent MIMO array pattern synthesis . . . . . . . . . . . .
5.1.1 Fourier-like transform representation of a MIMO array
5.2 Waveform diversity/orthogonality concept . . . . . . . . . . .
5.3 Effect of the illumination errors on the pattern synthesis . . .
5.3.1 Simulated results . . . . . . . . . . . . . . . . . . . . .
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
37
37
39
39
41
43
44
46
47
47
49
51
52
54
54
. . . . .
pattern
. . . . .
. . . . .
. . . . .
. . . . .
57
57
60
62
64
70
72
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6 MIMO signal processing: RADOCA test board and experimental

results
75
6.1 RADOCA MIMO radar description . . . . . . . . . . . . . . . . . . . . 76
6.1.1 Antenna and PCB design . . . . . . . . . . . . . . . . . . . . . 78
6.2 Board calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.2.1 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . 82
6.3 Moving target detection in TDM MIMO radars . . . . . . . . . . . . . 84
6.3.1 Doppler speed impact . . . . . . . . . . . . . . . . . . . . . . . 86
6.3.2 Multi domain signal analysis . . . . . . . . . . . . . . . . . . . 87
6.3.3 Effect of the random selection of the active transmitter . . . . 91
6.4 High resolution techniques applied to coherent MIMO arrays . . . . . 94
6.4.1 The MUltiple SIgnal Classification (MUSIC) method . . . . . 94
6.4.2 2D-MUSIC algorithm description . . . . . . . . . . . . . . . . . 95
6.4.3 Simulated and Experimental Results . . . . . . . . . . . . . . . 98
6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page ix #9
Contents
ix
7 Conclusion and outlook

103
7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.2 Recommendations and future work . . . . . . . . . . . . . . . . . . . . 106
A PCL system measurements
109
A.1 Receiver channel gains . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A.2 Channel noise figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
A.3 Element patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
B Illumination error effects on the synthesis of MIMO array pattern 115
C Basic theory of FMCW
121
List of Acronyms and Symbols
123
Bibliography
136
Summary
137
Samenvatting
139
Authors publications
141
About the author
143
Acknowledgements
145
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page x #10
Contents
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page xi #11
List of Figures
1.1
Sketch of an analog beamformer based on phase shifter components. .
1.2
Basic digital beamforming scheme. . . . . . . . . . . . . . . . . . . . .
2.1
Generic antenna array geometry. . . . . . . . . . . . . . . . . . . . . .
12
2.2
Linear antenna array geometry scanning on the xOz plane. . . . . . .
14
2.3
Normalised array pattern behaviors along the azimuthal plane ( = 90 ). 17
2.4
Circular antenna array geometry. . . . . . . . . . . . . . . . . . . . . .
18
2.5
3D view of the array factor for an 8-elements UCA (linear scale). . . .
19
2.6
Array factor for an 8-elements UCA along the azimuthal ( = 90 , left)

and the elevation ( = 0o , right) planes. . . . . . . . . . . . . . . . . .
19
3.1
Digital signal processing scheme of the PCL radar. . . . . . . . . . . .
24
3.2
First kind Bessel functions of different orders. . . . . . . . . . . . . . .
29
3.3
Normalised UCA pattern and mask function fd () of the desired pattern behavior. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
Results of the proposed side lobe reduction method for a (a) UCA with
radius r = 0.48, (b) UCA with radius r = 0.36. . . . . . . . . . . . .
31
Results of the phase modes side lobe reduction method for a (a) UCA
with radius r = 0.48, (b) UCA with radius r = 0.36. . . . . . . . . .
32
Comparison between the phase modes and the proposed DBF technique syntheses of a circular array pattern with radius r = 0.36 and
SLL=19dB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
33
4.1
Block diagram of the PCL system. . . . . . . . . . . . . . . . . . . . .
38
4.2
TNO circular array for passive radar applications. . . . . . . . . . . .
39
4.3
Simulated element pattern gain of a stand alone dipole considered as a

single radiating element using CST: (a) 3D plot and (b) cut along the
elevation plane ( = 90 ). . . . . . . . . . . . . . . . . . . . . . . . . .
40
3.4
3.5
3.6
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page xii #12
xii
LIST OF FIGURES
4.4
4.5
4.6
4.7
4.8
4.9
4.10
4.11
4.12
4.13
4.14
4.15
4.16
4.17
4.18
5.1
5.2
5.3
5.4
6.1
6.2
6.3
6.4
6.5
Simulated element pattern gain of a single dipole in the circular array

configuration using CST: (a) 3D plot and (b) cut along the azimuthal
plane ( = 90 ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Comparison between the measured element patterns for three different
array channels and the CST simulated data. . . . . . . . . . . . . . . .
Array construction model used in the CST simulator. . . . . . . . . .
Simulated circular array pattern gain using CST: (a) 3D plot and (b)
blue line: cut along the azimuthal plane ( = 90 ) red line: cut
along the elevation plane ( = 90 ). . . . . . . . . . . . . . . . . . . .
Rack of the PCL system analog receiver: (a) front view (b) internal view.
Block diagram of a receiving channel. . . . . . . . . . . . . . . . . . .
FM bandwidth input signals measured with the 6-th dipole of the array.
Representation of: a uniformly (a) and a sparsely (b) filled FM band.
(a) Sxx scattering parameters for the 8 channels of the array , (b) S1x
scattering parameters with respect to the first element of the array. . .
(a) Relative (with respect to first array channel) phase shifts after digital conversion, (b) Measured signal amplitudes after digital conversion.
Cartesian reference system comparison between the un-/calibrated and
the theoretical patterns in dB scale for (a) transmission point 1 and
(b) transmission point 2. . . . . . . . . . . . . . . . . . . . . . . . . . .
Effect of the DBF nulling procedure on the array pattern behavior. . .
Matched filter output Range-Doppler map. . . . . . . . . . . . . . . .
GO-CFAR Time vs Range output map (a) and Overlapping of the
CFAR detections with the ADSB available tracks data (b). . . . . . .
GO-CFAR Doppler Velocity vs Range output map (a) and Overlapping
of the CFAR detections with the ADSB available tracks data (b). . . .
Periodical array configuration of transmitting and receiving elements.
Different MIMO pattern contributions and realised pattern synthesis. .
Error affected array pattern comparisons for (a) 2 = 2 = 0.01, (b)
2 = 0.01 and 2 = 0.1, (c) 2 = 2 = 0.1. . . . . . . . . . . . . . . .
Analysis of the upper bound conditions for the average error patterns
for (a) 2 = 2 = 0.001, (b) 2 = 0.01 and 2 = 0.1. . . . . . . . . . .
X-Band radar test board with MIMO functionality developed at TNO.
Details of the USB 2.0 connector (a), boxed test board (b) and radar
system layout while connected to the data acquisition notebook (c). .
Virtual element relative positions of the MIMO array board. . . . . . .
Layout of /4 (8mm) spaced microstrip fed quasi-Yagi antenna. . . . .
Scattering parameter measurements of the /4 spaced elements. . . .
40
41
42
42
43
44
46
47
48
48
49
50
52
53
53
58
61
70
72
77
77
78
79
79
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page xiii #13
LIST OF FIGURES
6.6
6.7
6.8
6.9
6.10
6.11
6.12
6.13
6.14
6.15
6.16
6.17
6.18
6.19
6.20
6.21
6.22
6.23
6.24
xiii
Required spacings of the receiver elements for increasing number of the

transmitters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Measured and expected phase behavior of the reference scatterers. . . 82
Measured and average (a) phase offset and (b) fractional amplitude
values for the reference scatterers. . . . . . . . . . . . . . . . . . . . . 82
Effect of the calibration on the pattern synthesis (a) first scatterer (b)
second scatterer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Effect of the tapering on the calibrated and un-calibrated patterns (a)
first scatterer (b) second scatterer. . . . . . . . . . . . . . . . . . . . . 83
Example of a FMCW transmission scheme for a TDM MIMO array. . 85
FMCW Range/Doppler processing (a) FMCW Range/Doppler processing with a 3 stages MTI canceler (b). . . . . . . . . . . . . . . . . 86
TDM MIMO 3D matrix data structure. . . . . . . . . . . . . . . . . . 88
Ambiguity functions of a target with status vector v0 = [50m, 20o , 3m/s]
in a sequential TDM transmission mode: Range/Doppler (a) and Angle/Doppler (b) maps. . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Angle/Doppler map for a target with status vector v0 = [50m, 20o , 3m/s]
in a conventional FMCW radar with transmitted pulse length of NT T . 90
Ambiguity functions of a target with status vector v0 = [50m, 20o , 3m/s]
in a random TDM transmission mode: range/Doppler (a) and azimuth/Doppler (b) maps. . . . . . . . . . . . . . . . . . . . . . . . . . 91
Ambiguity functions of two targets with status vectors v0 = [50m, 20o , 3m/s]
and v1 = [50m, 35o , 10m/s] in a sequential (a) and random (b) TDM
transmission mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Effect of the number of integrated sweeps on the sidelobes level: (a) 16
MIMO sweeps (b) 128 MIMO sweeps. . . . . . . . . . . . . . . . . . . 92
Data matrix samples and scanning window procedure. . . . . . . . . . 96
Simulated scenario with 6 targets spaced by 0.6m in range and 5o in
angle: DBF processing (a) 2D-MUSIC processing (b). . . . . . . . . . 98
Cut at the 20.6m range bin for the 3 scatterers case (a); cut at the 20m
range bin for the 2 scatterers case (b). . . . . . . . . . . . . . . . . . . 99
Cut at the 5o angular bin for the 3 scatterers case (a); cut at the 0o
angular bin for the 2 scatterers case (b). . . . . . . . . . . . . . . . . . 99
Scenario for the measured data set collection with highlighted targets. 100
Real data scenario: DBF processing (a) 2D-MUSIC processing (b). . . 101
A.1 Set-up of the gain measurements. . . . . . . . . . . . . . . . . . . . . . 109

A.2 Set-up of the noise figure measurements. . . . . . . . . . . . . . . . . . 110
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page xiv #14
xiv
LIST OF FIGURES
A.3 Element pattern measurement setup: reference transmitter (a), PCL
array ARx and reference transmitter AT x (b), measurements geometry
(c) and rotating platform with scaled plane (d) . . . . . . . . . . . . . 113
C.1 Transmitted and received sawtooth modulated signals . . . . . . . . . 121
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page xv #15
List of Tables
3.1
3.2
3.3
Azimuthal resolution performance . . . . . . . . . . . . . . . . . . . .

Angular resolution comparison for different SLL taperings . . . . . . .
Illumination efficiency comparison for different SLL taperings . . . . .
32
32
33
4.1
4.2
ICS-554B digitiser board specifications . . . . . . . . . . . . . . . . . .

Average power levels at the input of the analog receiver . . . . . . . .
38
45
6.1
6.2
6.3
Main to Maximum Sidelobe Level . . . . . . . . . . . . . . . . . . . . .

Radar parameters selection for the simulated scenario . . . . . . . . .
Target positions for the simulated scenario . . . . . . . . . . . . . . . .
93
98
99
A.1 Gain values of the receiver channels . . . . . . . . . . . . . . . . . . . 110

A.2 Noise figure values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
A.3 Noise figure dependency with the attenuators configuration . . . . . . 111
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page xvi #16
xvi
LIST OF TABLES
i
i
i
MainFrancescoThesis 2013/6/12 14:07 page 1 #17
Chapter 1
Introduction
The basic tasks of a radar system conceive the detection of a target and the estimation of its range, i.e. the distance from the radar. This information is retrieved
by transmitting a probing signal inside the scene of interest, and by analysing the
signal that is reflected by the object. In the case of classical monostatic radars, a
single antenna is used at both the transmitting and the receiving sections. When the
angular information is needed, the antenna is mechanically rotated and the direction
of arrival of the target is obtained.
Modern radars employ transceiver sections that are composed by multiple radiators,
which are referred to as array of antennas, in order to enhance the overall system
capabilities. With respect to single antenna-based radars, one of the main advantages
offered by such systems is the possibility of combining the signals received by the
different channels and retrieving the angular information belonging to various directions [1]. This allows for non mechanical scanning of the beams. A higher flexibility
for surveillance tasks exploitation can then be achieved. Aside from the targets of interest, the radar scenario is characterised by the presence of unwanted signals that can
degrade the detection performance of the desired echoes. Those signals can be generated by either environmental reflections (clutter, multipath) or intentional interferers
(jammer). The latter class of signals are normally referred to as Electronic Counter
Measure (ECM) and they are present in military scenarios with the aim of deceiving
the sensor, increasing the noise floor of the radar or saturating the related receiver.
Since these disturbances are usually located in well defined sectors and the degrees
of freedom provided by the multiple antennas allow the synthesis of directional nulls
in the array radiation pattern, several Electronic Counter Counter Measure (ECCM)
techniques can be adopted to suppress or mitigate the impact of the interfering signals
on the radar system [2].
i
i
i
1. Introduction
Figure 1.1: Sketch of an analog beamformer based on phase shifter components.
Early array based radar systems used to perform pattern synthesis by means of analog components, such as phase shifters1 (PS) or variable time delay lines (TDL) [2,3],
electronically controlled by a computer unit. As shown in Fig.1.1, the signals are then
summed at the receiver level, which yields a single beam, and digitally converted before being available to the radar signal processor. The choice about using either PS or
TDL, aside from the inherent cost of the systems based on the latter ones, is driven by
the transmitted pulse characteristics with respect to the array fill time [4]. In order to
really benefit from all the array elements, the signal has indeed to be present on each
of them before summation. If TDL are considered, this condition can be satisfied by
a proper adjustment of each acquisition channel delay. In the case of PS, the signal
duration has to be greater than the time needed by the electromagnetic wave to cover
the array extension. By considering a linear array geometry, about which more details
are provided in Ch.2, it can be written that:
T =
L sin
,
c
(1.1)
being L the array length, the angle of arrival of the signal and c the speed of
light. The quantities and T represent the pulse duration and the array fill time
respectively. Since the pulse duration is inversely proportional to its bandwidth B,
by means of a proportionality term kB , (1.1) can also be written as:
B
c
.
kB L sin
(1.2)
1 The wording phased arrays was indeed introduced to refer to these type of analog systems.
Nowadays it is kept and it is more generally referred to any class of antenna arrays which perform
signal processing at an element level.
i
i
i
Figure 1.2: Basic digital beamforming scheme.
If we divide both sides of (1.2) by the carrier frequency fc , we obtain:

B

,
fc
kB L sin
(1.3)
where is the wavelength. Expression (1.3) highlights the dependency between the
signal fractional bandwidth fBc and the array system characteristics. If (1.3) is satisfied, the system is considered operating under narrow band assumption and PS can
be used in place of TDL, which are then exploited in the case of wideband arrays.
Thanks to the evolution of the radar electronics, state of the art systems can apply
digital conversion at an element level (see Fig.1.2). In this case, each receiving channel
has an independent front-end and an Analog to Digital Converter (ADC). Maximum
freedom is then provided to the signal processing section for the application of the
digital processing schemes, that are conventionally referred to as Digital Beam Forming (DBF) algorithms. In the digital domain, the antenna array can electronically
be steered to any direction. Multiple and closely spaced antenna beams can be synthesised and independently pointed and shaped. A more accurate control of both
the sidelobes behavior and the directional nulling is obtained thanks to calibration
techniques that can directly be implemented in the digital domain [2]. With respect
to surveillance operations, the DBF processing allows following the trajectory of the
detected targets by means of narrow beams, while the sensed area being illuminated
with a wide beam. Since the two operations are carried on at the same time, on
the one hand, the performance of the radar tracker is improved, on the other hand,
there is no reduction of the update rate of the sensed area. The electronic steering
i
i
i
1. Introduction
capability also represents an essential requirement in case of wideband and/or near

field applications2 [5, 6].
Latest developments of sensing techniques have led to the realisation of novel radar
concepts for security, safety and surveillance applications based on phased array technology [79]. A growing interest in that sense is represented by the definition of
proper array processing approaches for passive and Multiple-Input Multiple-Output
(MIMO) radars.
1.1
Passive Coherent Locator (PCL) systems
Active monostatic radars work by transmitting a known signal and receiving the
reflection of illuminated objects. As a consequence, while they detect a target, they
can also be detected by hostile sensing systems because of their own transmission.
To improve the covertness of the operation, bistatic systems were deployed already
in the early days of radar development, with the transmitter located far from the
receiver. While such a configuration allowed protecting the receiver, the transmitter
could still be detected and damaged by the enemy. Thanks to the use of transmitters
of opportunity, such as radio, mobile phone and TV broadcasters, a Passive Coherent
Locator (PCL) does not produce any emissions and is inherently covert.
The target location is retrieved in terms of a bistatic range, which is the sum of the
distance from the transmitter to the illuminated object and from this object to the
receiver. Given the bistatic range, the object is located on an ellipse having as focal
points the transmitter and the receiver. To resolve the object position, the direction
of arrival of the reflected signal is needed. For this purpose, passive radars typically
contain an antenna array and the target bearing angle is retrieved by means of array
signal processing [10]. A crucial issue which influences the system performance relates
to the capability of separating the relatively small signal reflected by the object and
the direct signal produced by the emitter of opportunity. The large difference in
strength between these two signals makes the design of passive radar system rather
cumbersome [11]. Furthermore, the accuracy also depends on the emitted waveform
and the related ambiguity function characteristics [12].
The covertness characteristic and the bistatic observation geometry, the latter one
particularly suitable against stealth targets [13], have stimulated the research about
PCL for military purposes. More recently air and vessel traffic control applications
have also been considered [1416].
Experimental PCL programs have been developed in several research institutes and
industries in the world. Among them, the Lockheed Martin Silent Sentry in USA,
the french HA-100 developed by Thales, the CORA PCL system designed at the
2 The
distinction between the near and the far field systems will be presented in Ch.2.
i
i
i
1.2 The Multiple-Input Multiple-Output radar concept
Fraunhofer institute in Germany and the italian AULOS realised by Selex-ES. TNO Defence, Safety and Security has been actively working on passive radar systems since
2002 [17]. Different experimental systems have been designed and manufactured for
research purposes. The latest demonstrator is based on a circular array configuration
that, with respect to linear array systems, offers the benefit of an omnidirectional
coverage [18]. As a consequence, the direct signal is always present since there is
not a physical separation between the channels which are aimed at acquiring the
surveillance and the reference signals. The two channels are indeed synthesised by
means of DBF techniques, that are also used to perform a preliminary suppression of
the direct signal. With respect to this system, the following research objectives have
been addressed:
The identification of a proper calibration technique for the PCL system, accounting for both the analog and the digital sections of the receiver. For the
application of any DBF scheme, it is indeed essential to have a complete understanding of the antenna configuration and the RF chain of the radar.
The derivation of a novel DBF method for the synthesis of the reference and
the surveillance beams. Since the target direction of arrival estimation depends
on the angular resolution provided by the array, this parameter has represented
a driving point for the selection of the preferred approach. A comparison with
existing technique, aimed at showing the benefits of the proposed method, has
been carried out. The effective direct signal nulling capability that is achievable
by means of DBF has also been evaluated.
The assessment of the PCL system by means of real data acquisition and performance validation. To that aim, the entire signal processing chain has been
implemented and the detection performance compared with ground truth data
sets.
1.2
The Multiple-Input Multiple-Output radar concept
Unlike phased array radars, where the radiating elements transmit a scaled version
of the same waveform, the MIMO paradigm is based on the possibility of employing
multiple emitters in order to transmit probing signals that are orthogonal to each
other [19]. At the receiver side, these uncorrelated waveforms can be separated to
enable processing of each independent transmitter/receiver pair.
The orthogonality among the probing signals can be achieved in different domains [20].
The most obvious way is by allocating each emitter to a separate time slot, in the so
called Time Division Multiplexing (TDM); drawbacks of this method are the scarce
capability of dealing with fast moving targets and the increased requirement in data
i
i
i
1. Introduction
handling (the transmitted waveforms must separately be digitised). Another approach

is based on the Frequency Division Multiplexing (FDM), in which case each waveform
belongs to a different sub-band. Here the main issue is represented by the difficulty to
ensure the desired coherency between pulses which are located at different frequencies.
The last category is represented by the Code Division Multiplexing (CDM), that is
an inheritance of the wireless communications where the MIMO paradigm was firstly
applied. In this case digital codes with low cross-correlation profiles are exploited to
modulate the basic waveform.
MIMO radars can also be classified on the basis of the transmitters and receivers
spatial distribution. The case of widely separated radiators is normally referred to
as statistical MIMO [21]. Such systems exploit the intrinsic angular diversity of the
system layout in order to obtain uncorrelated reflections of the target. The processing of the multiple responses helps in mitigating the rapid signal fluctuations of the
target Radar Cross Section (RCS). An improved target detection and estimation performance is then achieved [22, 23]. The other class of MIMO radars, namely coherent
MIMO radars, is denoted by the presence of antenna elements placed in proximity
to each other. By assuming each transmitter/receiver couple excited by the same
target scattering response, i.e. the target response keeps its coherency among the
multiple transmitted waveforms [2426], the different contributions can be coherently
combined leading to the synthesis of an extended number of virtual array channels.
Array signal processing techniques based on the exploitation of these additional channels ensure a better angular resolution of the radar system and an increased number
of detectable targets [27].
The RADOCA radar demonstrator, where RADOCA stands for RAdar DOme CAmera, is the result of one of the research projects led at TNO in the framework of
coherent MIMO systems. The idea consists of combining a camera with a radar in
order to perform the detection and classification of slow moving targets in private
environments. To this aim, the radar has to provide the information (azimuth and
range) about the target to the camera, that can then be steered towards the object
and finalize its identification. Design requirements in terms of: system size, angular
resolution and spatial adaptivity for clutter mitigation have driven the choice to the
realisation of a TDM based MIMO radar. The research activity within this framework
has focused on:
The theoretical characterisation of the coherent MIMO radar concept by the assessment of advantages and disadvantages with respect to conventional phased
array based radars. Specifically, a proper dimensioning of the MIMO system, in
both ideal and non ideal conditions, has been studied and the expected performance for the angular resolution and the array pattern sidelobes behavior have
been retrieved.
i
i
i
1.3 Outline of the Thesis
The estimation of a calibration procedure aimed at compensating for the array

element mismatching. As in the case of phased arrays, the application of DBF
techniques and the correct behavior of the radar depend on a proper calibration
of the array antenna section. An approach that exploits the MIMO configuration
has been studied to this purpose.
The analysis of the limiting factors which are introduced by the selected TDM
scheme. Their impact on the target identification have been analysed and a
transmitter selection procedure aimed at overcoming this limitation has been
proposed.
The evaluation of high resolution array processing techniques for radar performance enhancement. Thanks to the exploitation of the extended number of
degrees of freedom provided by the coherent MIMO array, a novel technique
has been implemented and the better range/angle discrimination capability has
been assessed.
1.3
Outline of the Thesis
The remainder of the thesis is organised as follows:

Chapter 2 introduces the principles of general antenna and array antenna theory.
The main concepts to assess and to describe the array pattern behavior are
presented. Causes of pattern degradation, i.e. mutual coupling and element illumination errors, are explained and referred throughout the text to characterise
the radar systems under investigation.
Chapter 3 focuses on the algorithm for the mutual coupling compensation of the
circular array which is used as a PCL sensor. A set of internal and external
measurements are used for the compensation procedure and the related signal
processing operations are illustrated. An array pattern shaping technique is also
illustrated. The performance of the proposed method is then compared with
the phase modes technique and the specific benefits are highlighted.
Chapter 4 presents the PCL radar that has been the object of the digital beamforming research on circular arrays. The antenna section of the system is analysed
and the design choices of the receiver are depicted. The techniques illustrated
in Ch.3 are then applied and assessed by means of experimental validations.
Also, the entire signal processing chain, leading to target detection and plot extraction, is presented and the radar performance is compared with Automatic
Dependent Surveillance Broadcast (ADSB) ground truth data.
i
i
i
1. Introduction
Chapter 5 discusses the topic of digital beamforming in connection with the coherent MIMO theory. The procedure leading to the design of this kind of radar is
presented and the concept of waveform diversity is introduced. Then, the characteristics of the MIMO array pattern and the degradation due to the presence
of element illumination errors are addressed. The higher sensitivity to the illumination errors of the MIMO arrays, with respect to conventional linear arrays,
is described and proven by the retrieved analytical formalism.
Chapter 6 describes the RADOCA radar demonstrator that has been realised and
the results of the signal processing techniques which have been developed: the
calibration procedure, the transmission approach aimed at extending the unambiguous Doppler interval for the TDM based MIMO radars and the novel high
resolution technique. Both simulated and experimental results are presented
and discussed.
Chapter 7 summarises the main achievements of the research activity described
in this thesis and provides an overview of the challenges which can still be
the objective of further analyses, both from hardware and signal processing
perspectives.
i
i
i
Chapter 2
Antenna theory and array
pattern synthesis
The chapter is concerned with the description of the main antenna and array antenna parameters that will be used throughout the text. In the first section, the basic definitions of
the antenna radiating properties are considered. Then, specific attention is given to the introduction of the array pattern concepts with respect to the linear and the circular shape
configurations. With the aim of characterising the effect of the non ideal behavior of the
array elements, the consequences due to the presence of mutual element interactions and
aperture illumination errors are considered in the last section of the chapter.
2.1
Main antenna parameters
The classical definition of antenna is provided by the IEEE Standard Definitions of

Terms for Antennas (IEEE Std 145-1983) [28] where it is defined as the part of a
transmitting or receiving system which is designed to radiate or to receive electromagnetic waves. It then results to be the transitional structure between the free-space
and the guided propagations of an electromagnetic wave. From the circuital point of
view, by exploiting the Thevenin representation of a guided propagation channel, the
antenna can be described by means of the impedance [29]:
ZA = (RL + Rr ) + jXA ,
(2.1)
being RL the resistance associated to the dielectric and conduction losses of the antenna structure, Rr the radiation resistance and XA is the reactance which represents
the imaginary part of the impedance. Any design is performed in order to obtain,
i
i
i
10
2. Antenna theory and array pattern synthesis
under ideal conditions, the conjugate matching between the antenna characteristic
impedance and the internal impedance of the source. In this way, the maximum
power is delivered to the antenna.
Besides its function as a receiving or transmitting energy device, most of the applications require the optimization of the radiated energy into some specific directions,
while it has to be minimized into others. Here the main parameters which are used
to describe the directional behavior of an antenna are introduced, since they will be
referred to later in the thesis. A more detailed and extensive description can be found
in [2932].
2.1.1
Directivity
The directivity function D(, ) of an antenna is defined as the ratio between the
radiation intensity U (, )1 in a given direction and the radiation intensity averaged
all over the directions. Being the average radiation intensity equal to the total power
radiated by the antenna divided by 4, it follows that
D(, ) =
U (, )
4U (, )
=
,
U0
Prad
(2.2)
R
1
U (, )d and is the solid angle. From (2.2) it is clear that, in
where U0 = 4
the case of an isotropic radiator, the directivity is unity as the quantities U (, ) and
U0 are equal to each other.
Under the far field assumption2 , the directivity can also be expressed in terms of the
electric field component as:
D(, ) = R
2.1.2
4E(, )2
.
E(, )2 d
(2.3)
Efficiency
A number of efficiencies can be related to an antenna; the overall efficiency is

normally taken as the combination of all of them and it is written as:
= r c d ,
(2.4)
where r is the reflection efficiency, which depends on the mismatch between the
transmission line and the antenna, c and d are the conduction and the dielectric
1 The
radiation intensity represents the power radiated by an antenna per unit solid angle [30].
2
far field region is assumed at a distance greater than 2D

being D the diameter of the
sphere which includes the antenna and the wavelength. This approximation keeps its validity at
the frequency bands of VHF and above. At lower frequencies, the starting distance of the far field
region is dictated by the most stringent value between 10D and 10.
2 The
i
i
i
2.2 Antenna array pattern synthesis
11
efficiencies respectively and they are related to the dissipation losses.

The efficiency represents a quality factor of the antenna performance as it indicates
how much of the input power Pin is effectively radiated by the antenna:
Prad = Pin .
2.1.3
(2.5)
Gain
A parameter which takes into account both the directional and the efficiency properties of an antenna is the gain, which is defined as the ratio between the radiation
intensity in a given direction and the radiation intensity that would be obtained if
the power accepted by the antenna were radiated isotropically. Being the radiation
intensity corresponding to an isotropic radiator equal to the input accepted power
divided by 4 and by considering (2.5) we have:
G(, ) =
4U (, )
4U (, )
=
.
Pin
Prad
(2.6)
The relation between the antenna gain and the antenna directivity can then be written
as:
G(, ) = D(, ).
(2.7)
The gain is then equal to the directivity in the case of an ideal radiator which presents
no losses, i.e. efficiency equal to unit. It must also be said that the IEEE standard
definition does not usually include the reflection efficiency in (2.7), limiting therefore
the efficiency to the dielectric and the conduction contributions.
2.2
Antenna array pattern synthesis
Multiple independent radiating elements can be arranged into a certain geometrical

configuration in order to improve the performance of the transceiver section. These
antenna configurations are usually referred to as arrays. The main advantage provided by the antenna arrays is represented by their electronic scanning capability.
Furthermore, an opportune tapering of the excitation coefficients of the different elements can be used to control the shape and the sidelobes of the array pattern as
it will be shown in Sec.2.2.1 and Sec.2.2.2. The array pattern control also conceives
the possibility of simultaneously steering the main beam in a certain direction while
a null is placed towards the direction of an interfering/jamming signal.
The behavior of the array pattern mainly depends on five parameters [29]:
geometry, i.e. the spatial configuration of the array
i
i
i
12
Figure 2.1: Generic antenna array geometry.
spacing between the multiple elements

amplitude excitation of each element
phase excitation of each element
element pattern of the radiators
The resulting electric field excited by a radiating element in a given point of the space
depends on the distance from the element itself and the related angular coordinates.
If the point is taken at a very far distance from the element, then the electric field
can be represented as the product between an angular function, fn (, ), which is
referred to as the element pattern and a radiation term which has spherical wave
dependance with the range [4]. By considering the geometry illustrated in Fig.2.1,
which represents a generic distribution and composition of array elements, the electric
far field produced the by the n-th element can be written as:
En (, , r) = fn (, )
ejkRn
,
Rn
(2.8)
being k the free space wave number and:

Rn =
(x xn )2 + (y yn )2 + (z zn )2
(2.9)
i
i
i
13
the distance between the n-th array element and the point where the electric field
is measured. If this distance is considered in the far field of the array3 the related
electric field wavefront can be assumed to be planar and (2.9) can be approximated
by:
Rn = R r ~rn ,
(2.10)
where r is the versor to the point where the electric field is measured and ~rn is position
vector of the n-th array element. The origin of the reference system can be arbitrarily
chosen. According to (2.10), the expression in (2.8) can be rewritten as:
En (, , r) = fn (, )
ejkR jk(r~rn )
e
.
R
(2.11)
The total electric field can then be retrieved by superposition of all the element
contributions:
N
ejkR X
fn (, )ejk(r~rn ) .
(2.12)
E(, , r) =
R n=1
Since the angular properties of the electric field in (2.12) are measured on a sphere
jkR
of constant radius, the term e R can be neglected and (2.12) can be written as
a function of the angular variables only. Thus, if all the elements share the same
pattern characteristic and different complex excitations an are considered for the
multiple elements of the array, (2.12) becomes:
P (, ) = f (, )
N
X
an ejk(r~rn )
(2.13)
n=1
which represents the array pattern. The aim of the an tapering coefficients is to
perform both the array pattern shaping and steering. More details about the weights
selection are provided in Sec.2.2.1 and Sec.2.2.2 with respect to linear and circular
array configurations. These geometries are of particular interest since they are the
chosen structures for the design of the radar systems that are illustrated in Ch.3 and
Ch.6.
According to (2.13), the array pattern can be seen as the product between two main
components: the element pattern and the array factor :
F (, ) =
N
X
an ejk(r~rn )
(2.14)
n=1
which describes the geometrical characteristics of the array.

2
, being L the largest

array far field region is considered to start at a distance R = 2L
dimension of the array. However, it is pointed out in [4, 33, 34] that, in order to measure very low
3 The
sidelobes and array patterns with deep nulled regions, the more stringent limit of R =
be taken as reference.
10L2
should
i
i
i
14
2.2.1
Linear array pattern synthesis
The linear array structure, or uniform linear array (ULA) when the an coefficients
have the same amplitudes, foresees a deployment of the antenna elements on a straight
line, as it is shown in Fig.2.2 for an array of N elements. According to this geometry,
the steering capability of the array is limited to the xOz plane ( = 0), whereas the
angular performance on the xOy plane is dictated by the behavior of the element
pattern only.
The analysis of the array factor for this configuration can be done by taking for
simplicity a phase reference at the first element, then (2.14) takes the form:
F () =
N
1
X
an ejknd sin () ,
(2.15)
n=0
where d is the inter-element spacing between the radiators. The series in (2.15) has
a known expression4 and it can then be rewritten as (the tapering coefficients are
assumed equal to one):
F ()
=
=
1 ej
2
N d sin()
ej N d sin() ej N d sin() ej N d sin()
2
ej d sin() ej d sin() ej d sin()
1 ej d sin()

sin N d sin()

(2.16)
ejref
sin d sin()
=
Figure 2.2: Linear antenna array geometry scanning on the xOz plane.
PA1
a=0
ejax =
1ejAx
.
1ejx
i
i
i
15
and it is also valid:

sin N d sin()
sin N d sin()
j
ref
=e

F () = e
N
sin d sin()
N sin d sin()
h
i
ejref N sinc N d sin() ,
jref
(2.17)
where the contribution of the array gain N , which depends on the total elements
number, has been highlighted.
When the complex excitations are chosen to synthesise the main beam towards a
specific direction 0 , which is obtained by selecting an = ejknd sin(0 ) , it can easily be
retrieved that (2.16) becomes:

sin N d(sin() sin(0 ))

.
F () = ejref
(2.18)
sin d(sin() sin(0 ))
Grating lobes
By analysing (2.18) it is clear that the element contributions can coherently add up
together at different angles. Those angles correspond to the zeros of the denominator:
d| sin() sin(0 )| = p
with p = 1, 2, ...
(2.19)
and, as a result, multiple beams of the same amplitude, usually referred to as grating
lobes, can arise in the field of view of the array. This occurrence can be avoided by
properly choosing the inter element spacing d. For a given maximum pointing angle
0 , if only a single beam, i.e. the main beam, has to be synthesised, the following
condition must hold:
N 1
1
d
<
,
(2.20)
N 1 + | sin(0 )|
which is a conservative constraint as it imposes that the entire grating lobe must be
outside of the scanning region of interest.
Beamwidth and sidelobes
The pattern beamwidth, which corresponds to the physical angular resolution of the
array, can be referred to either the Half Power Beamwidth (HPB) or to the First Null
Beamwidth (FNB). The former is measured at the -3dB intersection points from the
array pattern main peak, which for large arrays5 and by considering the boresight
5 For
a generic array length the expression which provides the 3dB beamwidth is:
= 0.5. However, it is shown in [4] that for N 3 the beamwidth variation
d sin(
sin( N
3dB ))
N sin( d
sin(3dB ))
with respect to (2.21) is less than 5% and it becomes lower than 1% for N 7.
i
i
i
16
pointing direction are given by:

1
3dB = sin
0.443
Nd

,
(2.21)
the latter, by considering (2.18), is given by the solution of:
d(sin() sin(0 )) =
(2.22)
which after few steps6 results to be:

F N B = sin1
N d cos(0 )

.
(2.23)
Equation (2.23) also highlights the resolution degradation which is introduced by the
pattern steering. By combining (2.21) and (2.23), the relation between the HPB and
the FNB is retrieved:
HP B ' 0.886F N B .
(2.24)
A useful descriptor of the array pattern behavior is the Side Lobe Level (SLL), which
represents the ratio between the highest side lobe and the main beam amplitudes. In
the case of large arrays, the SLL is independent of the main beam angle [35] and, if
a uniform excitation is applied throughout the elements, a -13.2dB level is obtained.
The SLL can be decreased by applying non uniform tapering masks at the cost of
both a reduced illumination efficiency:
PN
1 ( n=1 |an |)2
i =
(2.25)
PN
2
N
n=1 |an |
and a degraded angular resolution. Any novel technique that is implemented in order
to improve the SLL of an antenna array should also be analysed by taking into account
these effects.
Fig.2.3 provides an example of all the concepts that have just been introduced. Three
normalised pattern behaviors, belonging to a 16 elements linear array, are illustrated.
The first plot shows the uniform illumination case with an inter-element spacing
d = /2. The FNB and the HPB are depicted. In the second curve, while keeping
the uniform illumination, the array spacing has been increased to the value of d =
leading to the synthesis of a grating lobe at = 90o . In the last plot, a Chebishev
tapering mask [36] providing an SLL=-30dB has been applied instead of a uniform
one. The broadening of the main beam can be noticed and a resulting illumination
efficiency i = 0.86 is obtained.
6 d
[sin[( 0 ) + 0 ]]
sin(F N B ) cos(0 ).
[sin( 0 ) cos(0 ) + sin(0 ) sin(0 )]
i
i
i
17
Figure 2.3: Normalised array pattern behaviors along the azimuthal plane ( = 90o ).
2.2.2
Circular array pattern synthesis
The main advantage of using a circular array configuration is represented by its symmetry on the plane where the array is deployed. As a result, the pattern characteristics
do not present an high variation while the beam is electronically steered. For this
reason, the circular arrays are particularly suited for applications that require a 360o
coverage as it is for direction finding systems, auxiliary antennas, communication
bridges and so on.
In Fig.2.4 the typical geometry of an N elements circular array is shown. By following similar steps to the ones which have led to the representation of the array factor
for the linear array case, the Rn distance between the n-th array element and the
observation point in the far field region is equal to:
Rn =
p
R2 + a2 2aR cos n ,
(2.26)
being a the array radius and R the distance between the array center and the point
where the electric field is measured. Considering that R a, Rn can be approximated
in the following way:
Rn
' R a cos n = R a(
ar a
n )
= R a(
ax cos n + a
y sin n )
(
ax sin cos + a
y sin sin + a
z cos )
= R a sin cos( n ),
(2.27)
i
i
i
18
z
R
q
Rn
ar
a
xn
fn
an
n
2
f
x
Figure 2.4: Circular antenna array geometry.
where n = 2
N (n 1).
The complex excitation coefficients that are used to perform both the shaping and
the steering of the array pattern can explicitly be written as:
wn = In ejn .
(2.28)
By considering (2.15), (2.27) and (2.28), the array factor in the circular array case is
then given by:
N
X
F (, ) =
In ej[ka sin cos(n )+n ] .
(2.29)
n=1
The synthesis of the array pattern towards a specific direction (0 , 0 ) can be obtained
by selecting n = ka sin 0 cos(0 n ) and it consequently follows that:
F (, ) =
N
X
In ejka[sin cos(n )sin 0 cos(0 n )] .
(2.30)
n=1
The pattern of the array factor for an 8-elements Uniform Circular Array (UCA),
having therefore In = 1, is presented in Fig.2.5, whereas the cuts along the azimuthal
and the elevation planes are shown in Fig.2.6 The first sidelobe in the UCA case shows
a relative amplitude of 7.9dB, which is the typical value for this class of arrays. A
traditional way to perform the analysis and the pattern shaping of UCA systems is
i
i
i
19
Figure 2.5: 3D view of the array factor for an 8-elements UCA (linear scale).
Figure 2.6: Array factor for an 8-elements UCA along the azimuthal ( = 90o , left) and the elevation
( = 0o , right) planes.
based on the phase modes theory, which will be presented in Sec.3.3.1. By exploiting
that representation, in [37,38] an extensive analysis of the parameters selection, which
allows avoiding distortions of the synthesised pattern, is conducted.
Specifically in [38], the condition to prevent the grating lobes is proven to be:
d<
(2.31)
being d the element spacing on the circle.
i
i
i
20
2.3
Pattern synthesis in non ideal arrays
The array pattern analyses, which have been presented in Sec.2.2.1 and Sec.2.2.2, are
based on the ideal behavior of the array manifolds. However, real arrays are affected
by different sources of errors that introduce degradations in terms of sidelobes level,
pointing accuracy and angular resolution of the synthesised pattern. Although it is
possible to compensate for some of these errors, as it is described in Sec.2.3.1 and
with more details in Sec.3.2, proper design choices are required in order to overcome
the physical limitations produced by the random error distributions of the system,
which cannot be corrected. This case is analysed in Sec.2.3.2 and it is extended in
Ch.5 for a particular type of arrays.
2.3.1
Mutual coupling
The expression retrieved in (2.13) is based on the assumption that the element pattern
is the same for all the radiators. However, the gain of an isolated element may
substantially vary when the same element is placed inside an array. This gain variation
also depends on the specific location of the radiator: the behavior at the edges of the
structure is quite different from the one in the center. For this reason, a reliable
evaluation of the array pattern behavior cannot be exempted from the assessment
of the mutual electromagnetic influences, which are usually referred to as Mutual
Coupling (MC) [39, 40], among the array radiating elements.
The element pattern affected by MC can be written as the product between the
expected ideal pattern fi (, ) and a spatial factor that accounts for the coupled
element contributions [4]:
"
fn (, ) = fi (, ) 1 +
#
X
Snm ejk(~rm ~rn )r ,
(2.32)
being Snm the scattering coefficient [41]. The pattern in (2.32) is measured in a
controlled environment and by transmitting with the n-th element while all the other
elements are closed on matched loads. The total coupling effect is quantified by
collecting all the scattering contributions in the so called MC matrix [42, 43], which
is then used to retrieve the ideal behavior of the array pattern [4447]. Due to its
dependance on several parameters, it is preferable to perform the direct measurement
of the MC matrix instead of predicting its coefficient values. In Ch.3 a novel technique
aimed at estimating the MC matrix is presented with respect to the case of a circular
array in an interfering environment.
i
i
i
2.3 Pattern synthesis in non ideal arrays
2.3.2
21
Illumination errors
An array illumination distribution, which is aimed at providing a very low sidelobes

array pattern, may result in a poor SLR because of the presence of amplitude and
phase errors introduced by the array imperfections [2, 4]. This type of errors, which
are caused by the tolerance and quantization limits of the array devices (feeding
network, phase shifters, element orientations and positions, ...), can be treated as a
random and uncorrelated process7 . The sidelobes level degradation is produced by
the random sidelobes that are generated and that add to the expected ones.
If the errors for the N elements array are assumed to be independent but identically
distributed, and by referring to the linear array case, the AF affected by illumination
errors can be written from (2.15) as:
Fe () =
N
1
X
an (1 + n )ej(knd sin +n ) ,
(2.33)
n=0
where n and n are the fractional amplitude and phase errors associated to the
n-th element respectively. The AF is now a random variable and it depends on the
statistical distributions of n and n . The mean value of the AF is:
hFe ()i =
N
1
X
NX

an ejknd sin ejn +
hn i ejknd sin ejn ,
n=0
n=0
(2.34)
where the second addend in the sum is equal to zero if the presence of systematic errors
is avoided, i.e. the random processes are not biased. The first contribution represents
the product between the ideal AF and the characteristic function8 evaluated in = 1.
For a uniform phase error distribution:
p(n ) =
1
rect (),
(2.36)
sin()
.
(2.37)
(2.33) can then be written as:

hFe ()i = F ()
With respect to the field intensity, which is proportional to the array factor as shown
in (2.13), it is clear from (2.37) that the effect of the illumination errors only consists
7 For an extended analysis, the correlated errors case could also be considered and it would result
in higher sidelobes at fixed locations [2].
8 The characteristic function is defined as:
Z

C() = ejx =
ejx px (x)dx.
(2.35)
i
i
i
22
of a little attenuation of the expected ideal pattern.

In the case of the power pattern, which is related to the radiation intensity through
(2.2)-(2.3), its representation can be retrieved by taking the average value of the
product between (2.33) and its complex conjugate value:
D
E
2
|Fe ()| = hFe () Fe ()i .
(2.38)
The solution of (2.38) has been firstly retrieved in [48], then further investigated
in [49, 50]. Here the final expression is provided:
D
E

2
2
|Fe ()| = |F ()| + n2 + 2n i .
(2.39)
An extension to the case of virtually synthesised arrays is retrieved in Ch.5, where
also more details about the analytical solution are given. The average power pattern
that is shown in (2.39) is the sum of two contributions: the ideal power pattern and
a term depending on both the amplitude and the phase error distributions. It can
be observed that it is independent of the angular coordinates whereas it is directly
proportional to the illumination efficiency i . According to the mentioned characteristics, the second term in (2.39) introduces a uniform rise of the sidelobes level all
over the angular domain of the pattern; the effect is however of minor relevance in
the main beam and in the near sidelobe regions [49].
It is also of interest to observe that, by referring to (2.25) which shows a 1/N dependency with the array elements number, the larger the number of elements, the lower
will be the degradation of the sidelobes produced by the illumination errors.
2.4
Summary
The aim of this chapter was to provide the basic theoretical notions that allow describing the behavior of an antenna in both the isolated and the array configuration
cases. Attention has been given to the characterisation of the array pattern synthesis
for the linear and the circular array structures. At first, the ideal pattern behaviors
have been retrieved and the expected performance in terms of beamwidth, characteristic sidelobes level and shape have been discussed. Secondly, the main causes of
pattern degradation at both the antenna level (MC) and the system front end (illumination errors) have been presented. These analyses pave the way for the system
investigations in Ch.3, Ch.5 and Ch.6 where real radar demonstrators based on the
above mentioned configurations are considered.
i
i
i
Chapter 3
Digital beamforming for PCL
The signal processing steps required by a radar system are multiple and they have to deal
with the receiver chain at different levels. A first distinction can be done on the basis of the
analog and the digital sections that are involved. The first one, which is mainly based on
the front-end system architecture, will be analysed in the next chapter. Here the attention is
focused on the description of the digital processing blocks that represent the back-end part of
the system. A further distinction is required. The digital section can indeed be considered as
composed by two different macro blocks: the one belonging to the DBF processing, which is
tailored on the specific array characteristics, and the one related to the conventional signal
processing steps of the radar system. In this chapter, the entire attention is given to the
analysis of the implemented DBF procedures for the PCL system, which represents the core
of the research activity. However, in order to fully characterise the system performance, also
the remaining signal processing steps have been implemented and they will be shown in Ch.41 .
3.1
Digital processing scheme
A sketch of the digital processing steps that are involved in the PCL system operations is presented Fig.3.1. The sub-division into the two macro-blocks referred to as
the DBF and the specific radar signal processing is also highlighted. With respect to
the DBF part, the first operation is aimed at compensating for the MC interactions
among the antennas of the circular array. The non ideal behavior of the receiver
elements is corrected by means of the estimation of the MC matrix. This operation is
considered as a part of the DBF section since it allows the application of the beamforming algorithms, which otherwise would not provide the expected performance.
1 This chapter is based on articles [J1], [C8], [C9] and [R1] (a list of the authors publications
is included at the end of this dissertation, p. 141).
i
i
i
24
3. Digital beamforming for PCL
Figure 3.1: Digital signal processing scheme of the PCL radar.
Specifically, a method for the sidelobes shaping is proposed and the related results
are compared with a well established technique, which is especially tailored for circular array patterns. The pattern shaping technique, in conjunction with the digital
steering, is used to retrieve the reference and the surveillance channels of the system.
While the surveillance beam is synthesised, a digital null is placed in the direction of
the reference transmitter. The spatial nulling is part of the Direct Path Interference
(DPI) suppression, which is aimed at reducing the jamming effect of the direct signal
in the surveillance channel, and it represents the last step of the processing referred
to as the DBF.
The overall signal processing chain is then completed by: a digital filtering operation,
based on the Least Mean Squared (LMS) principle, which integrates the spatial DPI
nulling; the Matched Filter (MF), based on the cross correlation between the surveillance and the reference channels; the Constant False Alarm Rate (CFAR) thresholding
that provides the output target detections.
3.2
Mutual coupling compensation
Most of the MC compensation techniques require either a detailed knowledge of the

electrical characteristics of the antenna [51], or the possibility to measure them in a
controlled environment, e.g. an anechoic chamber, [52, 53]. The former case, as it is
highlighted in Sec.4.2.1, cannot be considered due to the lack of information about
the commercial components of the reference system; the latter one, due to the size
of the entire structure, is not applicable either. The compensation technique that
is proposed in this section, in order to overcome the mentioned issues, is based on
measurements aimed at obtaining a preliminary estimation of the MC matrix. Then,
a further refinement is obtained by means of an optimisation approach.
3.2.1
Analytical description
In real arrays, the element pattern is affected by the neighboring elements and, if we
refer to a circular array, it is an angular function of the element position. By recalling
i
i
i
3.2 Mutual coupling compensation
25
the characterisation presented in Ch.2, the far-field array pattern for a N-element
circular array can be written as:
P () =
N
X
an fn ( n )ejkr cos(n ) .
(3.1)
n=1
The dependency with the elevation angle is omitted for simplicity. Similarly to [42],
the total received voltage at the m-th element can be written as a weighted sum of
the contributions of all the array elements:
vm ()
N
X
= cmm Em fi ( m ) +
cmn En fi ( n )
n,n6=m
N
X
cmn En fi ( n ),
(3.2)
n=1
being Em the electric field which excites the m-th element port and fi the ideal element
pattern. The mutual coupling coefficient cmn therefore represents the proportionality
term that relates the induced voltage on channel m to the total voltage on channel
n. The desired voltage on channel m is then given by:
d
vm
() = Em fi ( m ).
(3.3)
If we compare (3.2) and (3.3) and by taking into account that Em = Em0 ejkr cos(m )
we obtain the same expression as [42] for the linear case which relates the desired
voltage signal and the real (affected by mutual coupling) one. By means of matrix
notation it can be written as:

d
v1
c11 c12 . . . c1N
v1
v2 c21 c22 . . . c2N v2d
(3.4)
. = .
..
..
.. .. ,
.. ..
.
.
. .
vN
cN 1
cN 2
...
cN N
d
vN
or in a more compact way as:

vd = C1 v.
(3.5)
The MC matrix C can be written as a function of the antenna scattering parameters

S [4, 41]:
C = S + I,
(3.6)
where I is the identity matrix.
The evaluation of the coupling coefficients in (3.5) can be performed in an analytical
or numerical way, depending on the antenna type and the array configuration [29].
i
i
i
26
If we refer to real systems, the effect of the feeding network and the analog/digital
front-end must also be considered, since it introduces discrepancies with respect to
the conventional representation. By considering the cables section as a reciprocal and
isolated structure, its contribution on the MC matrix can be modeled by a diagonal
matrix T of the transmission coefficients2 . From (3.6) we have:
C = T(S + I).
(3.7)
Equation (3.7) fully describes the matrix term which must be applied in order to
compensate for the non ideal behavior of the analog section of the system. Since the
presented correction is applied in the digital domain, e.g. after digital conversion of
the received signals, the effect of the phase and amplitude errors introduced by the
digital receiver should also be taken into account. The final correction matrix can
then be expressed as the following product:
C = P[T(S + I)],
(3.8)
where P is the digital section compensation matrix.
3.2.2
Optimisation approach for the C matrix evaluation
The expression retrieved in (3.8) is valid when no other signal sources, which can
generate local interferences to the radiation pattern, are present at the array location.
However, the environment in which the TNO passive radar is located, the top of the
tower at The Hague laboratory, is heavily affected by multipath due to the presence
of several large metallic structures. Moreover, the FM signals coming from the local
radio stations produce additive interferences and, as a consequence, an inaccurate
measurement of the main components of the MC matrix. For this reason, the C
matrix coefficients have been refined by means of an optimisation approach.
As proposed in [54], a reference transmitter has been moved around the PCL, at
distances ensuring the far field condition, and the signals from the known positions
have been acquired. For a far field monochromatic source, the received data vector
can be written as:
x(t) = C[sa(t) + n(t)],
(3.9)
where a(t) is the amplitude of the received signal and n(t) the noise realisation vector.
The quantity s represents the steering vector of the array that, on the azimuthal plane
and for an angle of arrival m , takes the form:
s = [ejkr cos(m ) , ejkr cos(m 1 ) , ..., ejkr cos(m N 1 ) ]T ,
(3.10)
2 The transmission coefficients represent the attenuation and the phase offset values to which the
signals propagating in the cables are affected.
i
i
i
3.3 DBF for circular arrays
27
where ()T is the transposed operator. As mentioned before, the m value is known.
By considering M transmission points, all of them characterised by the same C matrix,
the following objective function can be written:
O(C, a1 , a2 , ..., aM ) =
M
X
kxm Csm am k2
(3.11)
m=1
which represents a non linear system of equations whose unknowns are the signal amplitudes [a1 , a2 , ..., aM ] and the coupling matrix elements in C. The xm vector is the
collected data vector having the reference transmitter at the m-th location around the
array. A Broyden-Fletcher-Goldfarb-Shanno (BFGS) Quasi Newton method [55, 56]
has been applied to minimise the objective function in (3.11). In order to facilitate the optimisation process, the values of C have been initialised according to the
preliminary estimation provided by (3.8).
3.3
DBF for circular arrays
In passive radar applications the requirement on the SLL can be a very stringent
constraint, when considering that the ratio between the reference signal and a target
echo can be in the order of 90dB or more. Thus the patterns obtained in the previous
section are not able to satisfy it. A further reduction can be achieved by means of a
non uniform tapering on the array channels. Several DBF techniques for circularly
shaped phased arrays are based on phase mode excitations theory [5760]. Such theory
aims at synthesising a virtual uniform linear array that can then be weighted with
conventional tapering functions. However, the synthesis of the linear array results in
a degradation of the pattern beamwidth and often in an illumination efficiency loss.
If the second consequence can still be tolerated in PCL radars, thanks to the power
strength of the signal which are exploited, a further deterioration of the angular
resolution of the system should be avoided. To this aim a novel sidelobes shaping
technique has been developed. The regions of interest of the circular array pattern
are identified by an optimisation mask, as it is proposed in [54]. Then, the closed
form analytical expression for the optimal beamforming vector is retrieved. The
phase modes theory is briefly presented in Sec.3.3.1 whereas in Sec.3.3.2 the proposed
technique is discussed and the comparisons of the two methods are illustrated.
3.3.1
Phase modes theory
A well known technique for the array pattern shaping of circular and cylindrical arrays
is based on the phase modes theory. The array pattern of a circular array is a periodic
function in the interval [0, 2] and this characteristic allows representing it in terms
i
i
i
28
of a complex Fourier series. Referring to (3.1) we can write:
P () =
= Pp () =
p=
being:
1
Cp =
2
Cp ejp ,
(3.12)
p=
P ()ejp d.
(3.13)
Each term of the series in (3.12) is normally referred to as phase mode of the radiation
pattern [1] and it has a 2p phase variation as varies from 0 to 2. From (3.12) it
follows that, in order to synthesise a directional far field pattern Pp (), a p-th order
phase mode ejp must be produced in the far field. The Cp coefficients can generally
be evaluated as a sum of the an excitations, but that sum cannot be solved in a closed
form for all the an . However, a specific set of symmetrical excitation currents can be
identified [4] and computed as:
an =
apn =
p=
Ap ejpn .
(3.14)
p=
The terms of the sum in (3.14) are referred to as the phase mode excitation currents.
By combining (3.1) and (3.14), the p-th phase mode pattern takes the form:
Pp () =
N
X
Ap ejpn ejkr cos(n ) .
(3.15)
n=1
By introducing the Bessel function approximation for the array factor exponential
term and neglecting a constant term of no interest, (3.15) can be rewritten as:
Pp () Ap j p Jp (kr)ejp ,
(3.16)
where Jp () is the first kind Bessel function of order p. The behavior of the Bessel
functions, which is shown in Fig.3.2, also introduces a limitation on the maximum
number of phase modes that can be excited on the array. Decreasing rapidly to zero
when the order of the function exceeds its argument, the maximum phase mode order
is estimated as the maximum value of the argument which is: P = bkrc.
The Ap coefficients can now be selected in order to synthesise the far field pattern of
the array. By choosing the coefficients as:
Ap =
ejp0
,
j p Jp (kr)
(3.17)
the array pattern resembles the pattern of a linear array [57]. The only difference is
that, whereas in the linear case the azimuth scanning angle is related to a sin()
i
i
i
29
First kind Bessel functions
1
J0(x)
J1(x)
J2(x)
J3(x)
0.5
J4(x)
J5(x)
-0.5
0
10
Figure 3.2: First kind Bessel functions of different orders.
function, providing therefore a possible steering sector of 180 , in this case the scanning capability has no angular restrictions over the 360 . This result is in accordance
with the circular shape of the array.
Referring to the N excitations that must be applied to the circular array, they can
be evaluated by combining (3.14) and (3.17) and by considering the upper bound of
the effective number of phase modes that can be excited, it results:
an =
P
X
ejp(n 0 )
.
j p Jp (kr)
(3.18)
p=P
The tapering introduced by (3.18) produces a virtual transformation of the circularly

shaped array into a linear array structure, resulting in a SLL of 13dB. A secondary
effect resides in the possibility of over imposing a conventional tapering window in
order to further reduce the SLL.
3.3.2
Proposed algorithm
The array factor of the UCA that has been introduced in Ch.2, can be written according to (3.10) in the compact form:
F () = sH a(),
(3.19)
a() = [ejkr cos() , ejkr cos(1 ) , ..., ejkr cos(N 1 ) ]T
(3.20)
where
represents the array manifold and the superscript ()H is the Hermitian operator. The
angular behavior of the array factor is characterised by the presence of three different
i
i
i
30
Figure 3.3: Normalised UCA pattern and mask function fd () of the desired pattern behavior.
regions: the sidelobe, the transition and the main beam regions. In accordance with
the angular location of these sectors, it is possible to define a mask function fd ()
that describes a desired array factor behavior. An example is given in Fig.3.3 where
the generic UCA pattern for an 8 elements array with r = 0.48 is also plotted. The
aim of a digital beamforming procedure is to identify the steering vector that, at the
same time, minimises the contribution from the side lobes and focuses the energy into
the main beam region without a considerable degradation of the angular resolution.
If the mask function is taken as the objective of the pattern synthesis, the previous
conditions can be expressed by the following minimisation:
O(s) = min{ksH a() fd ()k2 }.
s
(3.21)
The effect of the minimisation in the sidelobes region can be emphasised by a diagonal
matrix of weights ; thus (3.21) can be rewritten as:
O(s) = min{[sH a() fd ()][sH a() fd ()]H }
s
(3.22)
and by omitting the dependency with the angle :

O(s) = min{sH aaH s sH afdH fd aH s + fd fdH }.
s
(3.23)
This approach, for the sidelobe shaping of a UCA pattern, was proposed in [54].
In that case, the steering vector was obtained by applying a numerical optimisation
approach similar to the one presented in Sec.3.2.2. Here, a closed form expression for
the optimal beamformer is retrieved.
As a function of the steering vector s, which is the only unknown in the second term
of (3.22), the desired minimisation argument is obtained by imposing the equality to
i
i
i
31
zero of the gradient [61, 62]:

s O(s) = aaH s afdH .
(3.24)
From (3.24), and by writing:

aaH = R,
(3.25)
the final expression of the UCA steering vector takes the form:
s = R1 afdH .
3.3.3
(3.26)
Array pattern comparisons
-10
-10
-20
-20
Normalised pattern [dB]
Different functions have been applied, according to (3.26), providing different sidelobe reduction (SLR) levels. Results of the DBF algorithm, for two circular arrays
with different radius lengths, are presented in Fig.3.4, whereas the 3dB beamwidth
apertures of the synthesised patterns, with respect to the unweighted UCA excitations, are summarised in Tab.3.1. The radius of r = 0.36 has been chosen as a
comparison term as it shows, among the analysed cases, the best angular resolution of
the array pattern. It is interesting to observe that for r = 0.36 the implemented DBF
algorithm produces a slightly better angular resolution than the uniformly weighted
case. On the other hand, an array radius of 0.48 is still preferable w.r.t. the mutual
coupling for the reasons that will be discussed in Sec.4.2.
The phase mode weights, with additional tapering windows, have also been applied
to the circular arrays with different radius presented in the previous paragraph. In
Fig.3.5 the obtained array patterns are shown whereas a comparison between the
-30
-40
-50
-60
UCA Pattern
19dB SLR
23dB SLR
27dB SLR
-70
-80
50
100
150
200
Azimuth angle []
(a)
-30
-40
-50
-60
UCA Pattern
19dB SLR
23dB SLR
27dB SLR
-70
250
300
350
-80
50
100
150
200
250
300
350
Azimuth angle []
(b)
Figure 3.4: Results of the proposed side lobe reduction method for a (a) UCA with radius r = 0.48,
(b) UCA with radius r = 0.36.
i
i
i
32
Table 3.1: Azimuthal resolution performance
SLR
Uniform Tapering
8dB
r = 0.36
r = 0.48
57.6o
43o
Proposed Tapering
19dB 23dB 27dB
41.6o
43o
45o
46.7o
50.2o
51.2o
-10
-10
-20
-20
angular resolutions realised with the phase mode excitations and the tapering of the
proposed DBF algorithm is listed in Tab.3.2.
In both cases, for the different values of the radius length, and for all the obtained
SLL reductions the proposed DBF method provides better angular resolutions than
the phase modes technique. A visual example of the resolution improvement is illustrated in Fig.3.6. The array patterns synthesised with the phase modes algorithm
and the proposed DBF technique, in the case of r = 0.36 and for an SLL=19dB,
are shown. In order to further asses the validity of the proposed pattern synthesis
-30
-40
-50
-60
UCA Pattern
19dB SLR
23dB SLR
27dB SLR
-70
-80
50
100
150
200
-30
-40
-50
-60
UCA Pattern
19dB SLR
23dB SLR
27dB SLR
-70
250
300
350
-80
50
Azimuth angle []
100
150
200
250
300
350
Azimuth angle []
(a)
(b)
Figure 3.5: Results of the phase modes side lobe reduction method for a (a) UCA with radius
r = 0.48, (b) UCA with radius r = 0.36.
Table 3.2: Angular resolution comparison for different SLL taperings
Array radius
r = 0.36
r = 0.48
Algorithm
Phase Modes
Proposed DBF
Phase Modes
Proposed DBF
SLL reduction
19dB 23dB 27dB
53.2o
41.6o
55.7o
43o
56.2o
45o
58.4o
46.7o
59o
50.2o
60.9o
51.2o
i
i
i
33
Figure 3.6: Comparison between the phase modes and the proposed DBF technique syntheses of a
circular array pattern with radius r = 0.36 and SLL=19dB.
method, the taper efficiency of the array weights is also evaluated in the two cases.
Being the illumination efficiency i defined as:
i =
PN
n=1
PN
|an |)2
n=1
|an |2
(3.27)
the results for all the considered configurations are shown in Tab.3.3. With reference
to the depicted values, the proposed algorithm shows an improvement in terms of
angular resolution gain which is, on average, equal to 14.4%; at the same time,
the illumination efficiency is, also on average, 5.3% lower than the phase modes one.
For the application purpose, due to the poor directivity of the circular array pattern,
an improvement of the angular resolution is preferable. The quality of the proposed
method is then confirmed. For the following processing steps, the tapering mask that
ensures an SLL reduction of -19dB has been used.
Table 3.3: Illumination efficiency comparison for different SLL taperings
Array radius
r = 0.36
r = 0.48
Algorithm
Phase Modes
Proposed DBF
Phase Modes
Proposed DBF
SLL reduction
19dB 23dB 27dB
0.97
0.82
0.81
0.82
0.97
0.89
0.83
0.81
0.98
0.93
0.84
0.76
i
i
i
34
3.4
Direct path interference suppression
The Direct Path Interference (DPI) reduction is based on two different steps: the
synthesis of an adapted pattern with a null in the direction of the reference illuminator
and a digital filter which provides the further reduction of the DPI signal.
If the absence of targets is initially considered and according to [2]:
x(t) = [v1 , v2 , ..., vI ]a(t) + n(t) = Va(t) + n(t),
(3.28)
where x(t) = [x1 (t), x2 (t), . . . , xN (t)]T is the N channels received signal data vector,
vi represents the steering vector related to the i-th interference, a(t) is the vector of
the interference amplitudes and n(t) the vector of noise realisations. The interference
source directions are then collected into the V matrix which is not necessarily a full
rank matrix. The modeling of the environment depends on the representation of the
data covariance matrix:
M
= < x(t)xH (t) >=< [Va(t) + n(t)][aH (t)VH + nH (t)] >

= V < a(t)aH (t) > VH + n2 I = VPVH + n2 I,
(3.29)
being P the covariance matrix of the interference sources complex amplitudes and n2
the noise variance, which is assumed to have a gaussian distribution. The interference
rejection in the array subspace is achieved by including the inverse of (3.29) into the
surveillance channel signal acquisition:
xs (t) = sH M1 x(t).
(3.30)
The expression of M1 can be evaluated in a closed form. By applying the ShermanMorrison-Woodbury3 formula, from (3.29) we have:
"
#
1

VP
1 H
1
1
H
M
=
I 2 I + 2 V VP
V
n2
n
n
1 H i
1 h
=
I VP n2 I + VH VP
V
2
n
1 H i
1 h
2 1
H
=
I
P
+
V
V
V .
(3.31)
n
n2
Thanks to the broadcasting channel selection procedure, which will be shown in
Sec.4.3.2, the presence of a single source of interference can be assumed. As a conse-
3 (A
+ UVT )1 = A1 A1 U(I + VT A1 U)1 VT A1 .
i
i
i
3.5 Summary
35
quence, (3.31) can be simplified:

1
"
#
2
1
1
n
H
H
Iv
+v v
v
n2
I2
H
1
vv
I 2
n
Hv
n2
+
v
2
I

IN R
1
H
I
vv ,
n2
1 + IN RvH v
(3.32)
where I2 is the interference (broadcaster) power and IN R is the Interference to Noise

Ratio. Experimental results of the DPI suppression by means of pattern nulling are
presented in Ch.4.
3.5
Summary
The chapter has introduced the digital processing chain of a PCL system with DBF
capabilities. The core of the analysis has regarded:
an advanced calibration procedure for the circular array in an interfering environment. The novelty of the technique is represented by the exploitation of two
sequential steps: a measurement phase and an optimisation algorithm. The former comprises both internal and external data acquisitions, aimed at retrieving
a preliminary evaluation of the system non ideal behavior. The latter exploits
that estimation in order to obtain a refined version of the MC matrix.
an original DBF method for the array pattern shaping. The related results
have been compared with the phase modes approach, an algorithm specifically
tailored for circular arrays. The proposed tapering technique ensured better
results in terms of angular resolution of the synthesised patterns, whereas a
negligible reduction of illumination efficiency has been calculated.
The steps regarding the signal processing, which is required by the passive radar
system, will be taken into account in Ch.4, together with the experimental results that
have been achieved. However, a relevant consideration about the proposed DBF steps
can already be drawn: the achieved angular resolution, when only DBF is applied, is
obviously below the standard expected from a surveillance radar system. An obvious
improvement from this point of view can be represented by the exploitation of high
resolution techniques based on either monopulse or spectral estimation approaches.
i
i
i
36
i
i
i
Chapter 4
PCL radar description and
experimental results
The chapter provides a functional description of the PCL system that has been designed
and developed at TNO. A detailed analysis of the antenna array is conducted from both the
element pattern and the array configuration perspectives. The FM analog receiver is then presented and the multiple RF components are depicted. Considerations about the degradation
of the receiver noise figure, which can affect the passive radar system in presence of strong
input signals, are highlighted. Experimental results, obtained by applying the MC compensation and pattern shaping techniques that have been presented in Ch.3, are illustrated. At
last, the remaining steps of the signal processing chain, which lead to the target detection
and the plot extraction, are considered. Suggestions for future system improvement and final
remarks about the PCL design are outlined in the concluding section of the chapter1 .
4.1
Overview of the PCL system
The PCL system, which is the object of the research study, can be described by taking
into account the three main sections into which it is divided and that are illustrated
in Fig.4.1. The antenna section, which performs the acquisition of the FM signals
and is deeply described in Sec.4.2, consists of a circular array of 8 half-wavelength
dipoles. Each dipole has an independent receiving channel; all the amplifying and
filtering stages are integrated into Printed Circuit Board (PCB) inside the analog
receiver rack (Sec.4.3). The rack also contains the power supply component and the
1 This chapter is based on articles [J1], [C8], [C9] and [R1] (a list of the authors publications
is included at the end of this dissertation, p. 141).
i
i
i
38
4. PCL radar description and experimental results
Receiver unit Ch1
Ch1
Ch2
Ch3
Antenna Array
Ch4
Receiver unit Ch4
ICS-554B
Data Acquisition Board 1
Clock
Receiver unit Ch5
Ch5
Ch6
Ch7
Ch8
Receiver unit Ch8
ICS-554B
Data Acquisition Board 2
Clock
8
A/D Conversion on
PC board
Sampling clock unit
Power Supply
Analog Receiver
Figure 4.1: Block diagram of the PCL system.
clock unit. The sampling clock is buffered and split into two (amplitude and phase
equal) signals for both the four-channel ICS-554B digitiser PC boards [63]. The board
specifications are listed in Tab.4.1. The boards are directly mounted on a separate
workstation where all the digitised data are collected. The signal processing is then
performed on a separate machine.
Table 4.1: ICS-554B digitiser board specifications
Type
ICS-554B
Input impedance
Full scale input level
Number of bits
Analog bandwidth
Minimum sampling rate
Maximum sampling rate
Clock input level
50 SMA
1.2Vpp
14
2-200MHz
30MHz
100MHz
-3...+6dBm
5.56dBm in 50
-3dB bandwidth
i
i
i
4.2 Antenna array analysis
4.2
39
Antenna array analysis
The analysis of the array characteristics regards both the single radiating element, the
element behavior in the array setup and the array pattern synthesis taking into account the overall structure. The antenna array in Fig.4.2 consists of 8 half-wavelength
dipoles, produced by Aldena [64], that are placed in a circular configuration. The
dipole bandwidth defined with respect to a return loss of 10dB is about 19% centered at the operating frequency within the FM band (88-108MHz). Each dipole is
equipped with a Gamma match [29] that allows a fine tuning of the input impedance
to optimise the matching at the reference frequency. This feature is useful when the
antenna is inserted into an array environment because the MC affects the antenna
input impedance. The dipoles have been tuned to operate at 96.8MHz, the broadcast frequency of Radio 3 in the Netherlands, which has been selected as emitter of
opportunity due to its high transmitted power (100kW).
4.2.1
Array element characterisation
Since the exact characteristics of the selected antenna were not disclosed by the manufacturer, simulations were carried out using a Finite Difference Time Domain (FDTD)
based commercial software package: CST Microwave Studio [65]. A dipole of radius
r = 0.003 and length l = 0.45, where = 309.93cm at the frequency of 96.8MHz,
was considered for the simulations. The dipole was fed with an ideal discrete edge
port with 50 input impedance. Fig.4.3.(a) shows the calculated radiation pattern
of the dipole oriented along the z-axis. The cut along the elevation plane, considered
for = 90 , is presented in Fig.4.3.(b). The antenna is matched with a return loss of
10dB in the frequency band 91.9-105MHz centered at 97.85MHz (13% band), and it
has a maximum gain of 2.02dB.
Figure 4.2: TNO circular array for passive radar applications.
i
i
i
40
0
30
30
60
90
60
10
90
120
120
150
150
180
(a)
(b)
Figure 4.3: Simulated element pattern gain of a stand alone dipole considered as a single radiating
element using CST: (a) 3D plot and (b) cut along the elevation plane ( = 90 ).
The stand alone behavior depicted in Fig.4.3 changes when the radiating element
is part of an array. A discriminating factor is represented by the array radius: the
shorter is the radius with respect to the wavelength, the more directive the element
pattern, but the higher is the MC. The last effect introduces a degradation in the
element pattern behavior that has to be evaluated before applying any digital beamforming scheme. Different array configurations have been investigated by taking into
account both the EM behavior of the structure and the following signal processing
steps, as discussed in Ch.3. The multiple aspects have been analysed and as a result
of this trade-off a radius a = 0.48 was chosen. The simulated element radiation
0
30
30
60
90
60
10
90
120
120
150
150
180
(a)
(b)
Figure 4.4: Simulated element pattern gain of a single dipole in the circular array configuration using
CST: (a) 3D plot and (b) cut along the azimuthal plane ( = 90 ).
i
i
i
4.2 Antenna array analysis
41
Normalised element pattern [dB]
-2
-4
-6
-8
Channel 8
Channel 4
Channel 6
Simulated Pattern
-10
-12
-14
-16
-18
-20
0
50
100
150
200
250
300
350
Azimuth angle []
Figure 4.5: Comparison between the measured element patterns for three different array channels
and the CST simulated data.
pattern, in the array configuration, is shown in Fig.4.4.(a). Due to the mutual interaction of the elements, the maximum gain of the single dipole changes from 2.02dB
to 4.5dB. The simulation has been performed with a simplified geometry: the feeding
network has not been taken into account and the dipoles have been fed with ideal
lumped ports. In order to asses the validity of the simulation, measurements aimed
at evaluating the element pattern of the real array system have been conducted at
the reference frequency of 96.8MHz. The element pattern of the array elements has
been characterised in an open air far-field measurement setup, by placing the array
on a rotating platform and using a biconical dipole antenna as reference transmitter.
A detailed description of the procedure is described in Appendix A whereas the results are presented in Fig.4.5. Three different array elements have been considered.
The related element patterns have been aligned along the same pointing angle for an
easy comparison. A good matching with the element pattern obtained with the CST
simulation can be observed.
4.2.2
Circular array characterisation
A preliminary analysis of the array pattern synthesis has also been performed at an
early stage of the PCL system design. As the absolute gain of the element embedded in the array antenna could not be measured at the time of the element pattern
characterisation, because a standard gain antenna was not available, an investigation
aimed at qualifying the eight elements array gain was carried out.
The final array model used in the simulation is shown in Fig.4.6. It includes: a central mast with a metallic tripod, a circular ring that holds the dipoles and part of the
concrete surface which supports the entire structure. At the PCL current location,
the base of the system is surrounded by a square steel frame and this component has
i
i
i
42
Figure 4.6: Array construction model used in the CST simulator.
also been introduced in the simulation.

The inclusion of the mast was driven by the EM analysis of the array. It was noticed
that the effect of the tripod on the element patterns resulted in a de-pointing of the
array pattern towards the ground. Cause of the misalignment was the resonating
effect of the tripod itself at the working frequency of the PCL. The extension of the
mast above the plane of the array solved the de-pointing issue by reestablishing the
correct radial pointing of the array pattern. The increase of the single dipole gain,
which has been presented in Sec.4.2.1, is also influenced by the presence of the mast.
Fig.4.7 presents the result of the simulated array pattern synthesis. The realised antenna pattern gain has a maximum equal to 9.6dB and if we consider the gain of
the single element (4.5dB), it can be concluded that no more than 3 elements effec-
0
30
30
60
90
60
10
90
120
120
150
150
180
(a)
(b)
Figure 4.7: Simulated circular array pattern gain using CST: (a) 3D plot and (b) blue line: cut
along the azimuthal plane ( = 90 ) red line: cut along the elevation plane ( = 90 ).
i
i
i
4.3 PCL receiver
43
tively produce a contribution to the array pattern synthesis. The simulated circular
array pattern in Fig.4.7 is taken as reference in Sec.4.4.1 for the validation of the MC
compensation algorithm presented in Ch.3.
4.3
PCL receiver
The signals received by the antennas are independently amplified and filtered by the
8-channel analog receiver in Fig.4.8. The analog chain was designed to be flexible
and reliable in different working environments, also in presence of relatively strong
unwanted signals. Main characteristic of the receiver is the frequency agility in the
whole FM radio band (88-108MHz). Since the system does not perform analog down
conversion to a lower frequency and the entire FM radio band is digitally converted,
any transmitting channel can be acquired without any performance degradation. The
sampling clock has been chosen in order to obtain a high sampling frequency, which
is optimal for the digital processing, and to preserve the spectral purity during the
baseband conversion. A sampling frequency of fs = 80MHz proved to be a good
choice for two reasons: from the theoretical point of view, the signal spectrum is
located between fs and the Nyquist frequency (fN = 120MHz) such that the under
sampling also provides the down conversion to the baseband, and from the practical
point of view, 80MHz components can be found off-the-shelf. The building blocks of
each analog receiving channel are shown in Fig.4.9. They consist of: (1) a variable
attenuator providing a maximum attenuation of 10dB, which can be introduced if
very strong signals are expected at the input of each channel (this attenuator is not
used in the current setup), (2) a 20MHz bandpass filter centered around 98MHz, that
prevents intermodulation of strong incoming out-of-band signals with signals inside
the FM-band; (3) a 10dB pre-attenuator that can be used to reduce the total gain
of the receiver if the input signals were causing saturation of the ADC (this preattenuator is normally switched off); (4) a low noise amplifier, which provides 22dB
(a)
(b)
Figure 4.8: Rack of the PCL system analog receiver: (a) front view (b) internal view.
i
i
i
44
-10dB
ON/OFF
-10dB
ATT
ATT
22dB
-6dB
ON/OFF
-10dB
ATT
ATT
22dB
-6dB
ATT
9
10
Figure 4.9: Block diagram of a receiving channel.
gain; (5) a high-pass filter with a cut-off frequency of 88MHz, which suppresses the
signals on the low side of the FM-band; (6) a fixed 6dB attenuator, that isolates the
high and low pass filter; (7) a low pass filter with a cut-off frequency of 108MHz, which
suppresses the signals on the high side of the FM-band; (8) a 10dB post-attenuator
to control the gain of the receiver. It does not severely influence the noise figure of
the system; (9) a 22dB gain amplifier and (10) a fixed 6dB attenuator that has three
purposes: preventing the digitiser from being damaged by too strong signals from
the amplifier; improving the S11 at the output of the receiver and providing a short
circuit protection to the receiver output.
4.3.1
Dynamic range analysis
Passive radar systems are known for the needed relatively large Dynamic Range (DR).
Transmitters of opportunity are used for the detection and the bistatic ranging of the
targets and these transmitters are typically broadcast stations for television and radio.
The direct signal from the transmitter of opportunity is much larger than the scattered
signal from the object to be detected. Moreover, whereas in active radar systems the
transmitted signal is known, in passive radar systems this signal is unknown and needs
to be sampled just as the target signal.
The receiver DR must be higher than the maximum Signal to Noise Ratio (SNR),
which is the ratio of the maximum received power Smax to the level of equivalent
input thermal noise power N . For a wideband receiver, such that of TNO passive
radar, which is in principle sensitive to all FM broadcast stations, Smax is the sum of
the power received from these stations:
DR =
Smax
,
N
(4.1)
with N = kT0 BF where k is the Boltzmann constant, T0 is the standard temperature

of 270K, B is the receiver bandwidth, F is the system noise figure. The Smax in (4.1)
is computed as:
!
X
Smax = max
Si ,
(4.2)
time
i
i
i
4.3 PCL receiver
45
being:
Si =
Pi Gi (, )Gr (, )2
,
(4)2 LRi2
(4.3)
and it is calculated with respect to the transmitter parameters: gain Gi as a function

of elevation and azimuth angles, transmitted power Pi , distance to the receiver
Ri and atmospheric losses L. Considering the large amount of broadcast stations
operating in urban environment (Fig.4.10), it appears from (4.3) that a high DR is
required for any passive radar system.
The DR of the signal is defined as the ratio (linear scale) or difference (dB scale)
between the highest signal level and lowest signal level. Throughout the radar system
these levels can change. This means that the DR can be different at the antenna from
the DR at the ADC. In the following it is assumed that the DR of the hardware is
sufficient in order not to influence the DR at the ADC input: 1dB compression levels
are not reached and no third order intermodulation products are larger than the noise
level. Then for the passive radar system at TNO, being the total FM band sampled
at 80Ms/s and 14 bits and representing each bit a factor of two in amplitude or a
factor of four (6dB) in power, a theoretical maximum DR of 84dB is available.
Power consideration about the broadcasting stations in the FM bandwidth
The maximum signal level at the ADC input corresponds to the total power received
in the FM-band at the passive radar location, a 40m high tower at TNO premises
in The Hague. Multiple transmitters and multiple stations operate in the working
band. In order to evaluate the impact of all these transmitters on the signal level,
an oscilloscope was used to estimate the amplitude of the signals received from each
dipole. The measurement of the 6-th channel is presented in Fig.4.10 where also the
reference transmitter power has been highlighted. The average power levels have been
measured for all the array channels and their values are listed in Tab.4.2. With reference to the analog channel analysis illustrated in Sec.4.3, the minimum gain provided
by the receiving channel is Gmin 6.5dB. According to the values in Tab.4.2 and by
considering that the maximum input power to the ADC, in order to avoid clipping
effects, is 8dBm (by direct estimation) we concluded that at the current array location both available attenuators in the receiver had to be switched on during the data
acquisition.
From the radar point of view, this constraint has a high impact on the system per-
Table 4.2: Average power levels at the input of the analog receiver
Power (dBm)
Ch. 1
2.54
Ch. 2
1.34
Ch. 3
2.46
Ch. 4
2.73
Ch. 5
2.78
Ch. 6
-2.82
Ch. 7
-6.49
Ch. 8
0.38
i
i
i
46

-45dBm
0
-10
-43dBm
Power [dBm]
-20
-30
-40
-50
-60
-70
-80
-90
-100
85
90
95
100
105
110
Frequency [Mhz]
Figure 4.10: FM bandwidth input signals measured with the 6-th dipole of the array.
formance as the noise figure of the whole system is degraded to the value of 17dB. As
a consequence, the detection sensitivity is also decreased due to the additional noise
that is introduced. However, since the receiver hardware could not be modified, the
reduced sensitivity was tolerated.
4.3.2
Digital down conversion
Due to physical limitations in data handling and storage capacity of the digital section of the current system, the digitiser boards work in Digital Down Conversion
(DDC) mode. The DDC performs a down sampling of the digitised signal by selecting a specific channel (96.8MHz carrier of FM3 radio from Lopik) within the filtered
spectrum. The output sampling rate of the DDC is 250ks/s that, by considering the
80Ms/s input data rate of the ADC, yields a decimation factor of 320. The introduced
decimation is equivalent to a processing gain of 25dB that approximately corresponds
to a gain in SNR and in DR of 25dB or 4 bits2 . Therefore, the theoretically achievable
DR after the DDC is 109dB (18 bits). A setback is that the suppression level of the
out-of-band signals, FDDC , although large is still finite.
The out-of-band signals are suppressed at the DDC output by about 70dB (measured
signal level relative to an injected out-of-band signal). By assuming a completely and
2 A rigorous analysis should take into account that the noise component, after the filtering stages
described in Sec.4.3, has a uniform distribution only within the selected 20MHz of the FM spectrum.
Indeed, the behavior outside the mentioned bandwidth depends on the combined response of both
the low-pass and the high-pass filters. However, the residual out-of-band noise signal is weaker than
the one inside, therefore the 25dB SNR gain can be assumed a conservative value.
i
i
i
4.4 Experimental results
47
Power
88MHz
108MHz
Nc channels
(a)
(b)
Figure 4.11: Representation of: a uniformly (a) and a sparsely (b) filled FM band.
uniformly filled FM band, Fig.4.11.(a), the maximum power reduction factor Fpow is
given by:
Fpow =
FDDC (BF M Bchannel ) + Bchannel

Bchannel
,
BF M
BF M
(4.4)
where Bchannel is the bandwidth occupied by a single channel, selected by the DDC,
and BF M is the whole FM bandwidth. However, the power distribution within the
band of interest is not uniform, Fig.4.11.(b), and Nc separate equal-power channels
can be distinguished. According to that, (4.4) becomes:
Fpow =
FDDC (Nc 1) + 1
1
.
Nc
Nch
(4.5)
The Nc value was estimated by considering all the stations for which the received
power, measured at the input of the analog receiver, was greater or equal to the
Lopik value minus 3dB. In our case this yields Nc = 20 and a reduction factor of
about 13dB. Thus, after the DDC, the thermal noise is reduced by a factor of 320
while the maximum expected power is reduced by a factor 20. This corresponds to
an increase in DR of a factor 16 (12dB) or 2 bits and not the expected increase of
4 bits from noise consideration alone. As a consequence, the actually available DR
after the DDC is 96dB (16 bits).
4.4
4.4.1
Experimental results
MC compensation
The experimental validation of the proposed technique for the MC compensation

is based on a set of internal and external measurements. By referring to the first
ones, a network analyzer has been used to get the preliminary estimation of C, as
i
i
i
48
-5
-10
Scattering parameter [dB]
-15
-20
-25
S
-30
S
S
-35
-40
S
S
-45
12
13
14
15
16
17
18
-50
-55
90
92
94
96
98
100
102
104
106
108
110
Frequency [Mhz]
(a)
(b)
Figure 4.12: (a) Sxx scattering parameters for the 8 channels of the array , (b) S1x scattering
parameters with respect to the first element of the array.
it is presented in (3.7), with respect to the S matrix component. Since it was not
possible to directly measure the scattering parameters at the input of the antenna
elements, the contribution of the cables is included in the measurements that are
shown in Fig.4.12. The plots highlight an overall behavior which is below -10dB in
the bandwidth 88MHz-102MHz and below -8dB in the bandwidth 102MHz-108MHz.
A frequency agility of the array elements could then be appreciated in the band of
interest for the passive radar application.
The estimation of the P matrix has been retrieved by connecting a stable transmitter
to an 8-channels splitter. The output signals of the splitter have directly fed the
(a)
(b)
Figure 4.13: (a) Relative (with respect to first array channel) phase shifts after digital conversion,
(b) Measured signal amplitudes after digital conversion.
i
i
i
49
0
Uncalibrated Pattern
Calibrated Pattern
Theoretical Pattern
-10
-15
-20
-10
-15
-20
-25
-25
-30
-30
-35
50
100
150
Calibrated Pattern
Theoretical Pattern
-5
-5
200
250
300
350
-35
50
Azimuth angle []
(a)
100
150
200
250
300
350
Azimuth angle []
(b)
Figure 4.14: Cartesian reference system comparison between the un-/calibrated and the theoretical
patterns in dB scale for (a) transmission point 1 and (b) transmission point 2.
digital receiver. The amplitudes and relative phases of the measured digital signals
are plotted in Fig.4.13. It can be observed that the amplitude variations among the
different channels represents a negligible effect, which is in the order of 0.8dB at the
reference frequency, whereas the measured relative phase shifts, which are produced
by the sampling with two parallel boards, show an higher variation which is both
frequency and channel dependent. The measurements in Fig.4.12 and Fig.4.13 have
been used to initialise the algorithm for the minimisation of the objective function
in (3.11). A total number of 6 locations around the array have been chosen and
some results of the overall compensation are illustrated in Fig.4.14 for two different
transmission points. Effects of the calibration include the removal of the pattern
asymmetries and the reduction of the SLL to the expected theoretical value of -8dB.
If we compare the obtained figures, differences between the compensated diagrams
mostly affect the region far from the main beam pointing. Such discrepancies depend
on both the variability of the multipath interferences with respect to the angular
position of the transmitter, and the convergency of the optimisation algorithm that is
mostly influenced by the main beam region, where the signal amplitudes are stronger.
4.4.2
Direct path interference suppression
As already pointed out in Ch.3, the whole DPI suppression procedure consists of two
steps: the synthesis of a spatial null into the surveillance beam pattern in the direction
of the reference transmitter; the application of a digital filter for the removal of the
residual direct signal component inside the surveillance channel. The result of the
achieved suppression level, thanks to the DBF nulling, is presented in Fig.4.15. The
i
i
i
50

0
-10
25dB
-20
-30
-40
-50
Calibrated Pattern
Calibrated Pattern + SLR
Calibrated Pattern + SLR +Nulling
-60
-70
50
100
150
200
250
300
350
Azimuth angle []
Figure 4.15: Effect of the DBF nulling procedure on the array pattern behavior.
figure also shows the calibrated array pattern, obtained by electronic steering, in the
uniform excitations case and with the pattern shaping method proposed in Sec.3.3.2.
The effective suppression provided by the DBF nulling, due to the width of the array
pattern, is in the order of 25dB. For conventional processing scenarios, which at
best allow using 1s of integration time for an effective bandwidth of 50kHz, the
autocorrelation function of an FM signal is characterised by range/Doppler sidelobes
at a level of 40-50dB. Given that the direct signal itself can be 80-90dB higher than
the expected reflection from a real target, it is clear that a suppression of 25dB is not
enough for the PCL system.
The further suppression of the residual reference signal xr (t), that according to the
used notation can be expressed as:
xr (t) = vH x(t),
(4.6)
in the surveillance channel xs (t), can be achieved by a digital filtering subtraction.

The implemented filtering procedure is based on an adaptive Gradient Least Mean
Squared (GLMS) filter [6668] and, since the interference is not constant, the filter
has to continuously adapt its coefficients to the interference variation. In the following
description of the filter the discrete-time notation is used, with k as the time index.
At time instant k, the N tap delayed inputs xk = [xk , . . . , xkN +1 ]T of the reference
signal are fed to the filter with weights wk = [wk (0), . . . , wk (N 1)]T . The output
i
i
i
51
signal at time k is then:

yk =
N
1
X
wk (n)xkn = wkH xk .
(4.7)
n=0
If we define the error signal ek as the difference between the filtered signal yk and the
desired signal dk xr (t):
ek = dk wkH xk = dk yk ,
(4.8)
the interference cancelation is realised when (4.8) is minimal. In order to minimise

the error power, the weights vector wk is adapted according to the LMS optimisation
algorithm:
ek
wk+1 = wk
ek = wk + xH
(4.9)
k ek .
wk
ek
ek represents the gradient of the function 21 e2k and the scalar
Where the product w
k
term is called the learning rate of the filter. This learning rate depends on the
variability of the signal which must be removed. Experimental tests have shown that
a learning rate = 0.005 and a number of filter taps N = 48 are suitable values for
the data set under test. Typical additional suppression provided by the filter with
the mentioned characteristics is in the order of 30-40dB.
4.4.3
Range/Doppler processing
The next step in the processing chain of the PCL is represented by the exploitation
of the reference signal in order to perform the target detection inside the surveillance
channel. By applying the conventional Matched Filter (MF) theory, the range/Doppler maps of the target positions can be retrieved. In this section a brief summary
of the theory is presented, whereas a more detailed description can be found in the
referred radar literature [69, 70].
The optimum detector, in presence of white Gaussian noise, is represented by a filter
with impulse response h(t) of the form:
h(t) = Ga s (T t),
(4.10)
being s(t) the signal we want to detect, T the delay of the filter, which is equivalent
to the signal duration, and Ga the arbitrary gain of the filter. In both the active and
the passive radar cases, the signal to be detected is a delayed and Doppler shifted
copy of the reference signal xr (t):
s(t) = xr (t )ej2fD t .
(4.11)
The filter output can then be evaluated as a function of the introduced delay and
i
i
i
52
Bistatic range [km]
150
100
50
0
-200
-150
-100
-50
50
100
150
200
Doppler frequency [Hz]
Figure 4.16: Matched filter output Range-Doppler map.
the Doppler shift; after normalisation it can be written as:

Z

Z(, fD ) = |z(, fD )|2 =
2

xs (t)xr (t )ej2fD t dt .
(4.12)
The formulation in (4.12) shows that the MF receiver is equivalent to the Fourier
transform of the product between the surveillance channel signal and a delayed conjugate version of the reference signal. The practical software implementation makes
use of this equivalence. In Fig.4.16 the output of the MF is presented for an acquired
data set of 1s duration. Multiple targets can be distinguished, with Doppler components not equal to zero, and they have been included highlighted in the dash-line
circles. On the other hand, zero-Doppler detections can also be noticed in the figure
and they are associated to multipath reflections of the direct signal coming from static
objects.
4.4.4
CFAR detector and plots extraction
The amplitude values of (4.12) have been passed through a CFAR detector based
on the Greatest-Of (GO) principle [71]. The reference window size, in both range
and Doppler dimensions, and the detection threshold were empirically estimated over
different range/Doppler maps at the output of the MF. The estimation led to a
[17x17] window, having therefore M=8 cells on each side of the cell under test, with a
threshold 0 = 6dB. The threshold value has been chosen to maintain the detection
probability in view of the reduced sensitivity of the system caused by the degraded
noise figure. The higher false alarm rate was tolerated.
The outputs of the CFAR algorithm are shown in Fig.4.17.(a) and Fig.4.18.(a) for
the range/time and range/Doppler cases respectively. The detection performance
i
i
i
53
160
140
140
120
120
Bistatic range [km]
Bistatic range [km]
160
100
80
60
100
80
60
40
40
20
20
50
100
150
200
250
300
350
400
50
100
150
200
250
300
350
400
Time [s]
Time [s]
(a)
(b)
160
160
140
140
120
120
100
100
Bistatic range [km]
Bistatic range [km]
Figure 4.17: GO-CFAR Time vs Range output map (a) and Overlapping of the CFAR detections
with the ADSB available tracks data (b).
80
60
80
60
40
40
20
20
0
-400
-300
-200
-100
100
Doppler velocity [m/s]
(a)
200
300
400
0
-400
-300
-200
-100
100
200
300
400
Doppler velocity [m/s]
(b)
Figure 4.18: GO-CFAR Doppler Velocity vs Range output map (a) and Overlapping of the CFAR
detections with the ADSB available tracks data (b).
was compared with the track data provided by an Automatic Dependent Surveillance
Broadcast (ADSB) transponder and recorded at the same time of the experiment.
This technology is not yet available on all flights that explains why only some of
the detected tracks match with the information included in the ADSB files. The
comparisons between the detections and the ground truth data are illustrated in
Fig.4.17.(b) and Fig.4.18.(b). A good match of the radar plots with the track overlay
can be distinguished for the entire extension of the available data set.
i
i
i
54
4.5
System future improvements
The PCL demonstrator is the third generation of passive radar systems that have
internally been studied at TNO during the last years. The main difference with the
previous systems is represented by the use of the same antenna array to perform the
acquisition of both the reference and the surveillance channels. This characteristic
has driven the design choices which have been illustrated in this chapter and which
have led to the current system configuration.
Some limiting factors have also been mentioned and those can be considered as objectives for future improvements of the current system. Specifically:
A complete exploitation of the available agile receiver depends on the data
throughput. The upgrade of the data acquisition Local Area Network (LAN)
would allow the simultaneous collection of multiple baseband signals. A multilateral system could then be realised, with an expected gain in terms of radar
coverage and resolution in both the range and the angular domains.
With respect to the previous issue, the system agility does also depend on the
array elements. The current dipoles are indeed tuned at the reference frequency
of 96.8MHz. Still showing a good behavior over the full FM-band, wideband
radiating elements would be recommended in order to avoid variability of the
overall system performance due to the selection of different broadcasters.
The noise figure of the PCL system is currently degraded by the introduction
of the double attenuation step, which is needed in order to avoid the clipping of
the ADC. This degradation has a direct impact on the radar sensitivity. Stateof-the-art PC digitiser boards could be installed into the PCL working station
to overcome this limitation.
4.6
Summary
This chapter has provided a top-down description and a performance assessment of

the PCL radar system that has been realised at TNO.
A detailed study of the antenna section has been presented. The analysis resulted, by
means of simulations and direct measurements, in the characterisation of the radiating
properties of the antenna elements in both the stand alone and the array configurations. Then, the 8-channel analog receiver has been considered. The receiver RF
chain, aimed at correctly selecting the FM bandwidth and properly amplifying the
input signals for the ADC, has been illustrated. The impact of the digital signal
decimation on the DR has been discussed by highlighting the difference among the
expected and the realised improvements.
i
i
i
4.6 Summary
55
The experimental validation of the system has firstly been based on the capability
of retrieving the proper pattern characteristics of the circular array under analysis.
The proposed MC compensation technique has been capable of restoring the pattern
symmetry and the theoretical SLL of a circular array. The consistency of the result
has been confirmed by the application of the proposed pattern shaping method and
the DPI spatial nulling, which have shown the expected performance in both cases.
The final validation of the PCL operation followed after the implementation of a complete back-end processing section to the system. Assessed processing techniques have
been taken into account, since the main interest of the research was not driven by
the investigation of novel algorithms. However, the implementation of the mentioned
processing blocks has been a crucial part for a complete evaluation of the PCL system
capabilities.
i
i
i
56
i
i
i
Chapter 5
Coherent MIMO array theory
The distinction between the statistical and the coherent MIMO arrays has been introduced
in Ch.1. Here the attention is focused on the analysis of the latter case with respect to the
synthesised virtual array pattern. The procedure to select the mutual positions between the
transmitting and the receiving elements, which is needed in order to avoid the presence of
grating lobes, and the design choices aimed at obtaining the required angular resolution are
discussed. The concepts of signal orthogonality and waveform diversity, as a preliminary
condition for the exploitation of the MIMO approach, are also presented. In the last section
of the chapter, the MIMO array pattern affected by illumination errors is analysed and the
analytical representation, in a closed form expression, is retrieved. Simulated results are
shown and a comparison with the case of a conventional ULA system is illustrated1 .
5.1
Coherent MIMO array pattern synthesis
A coherent MIMO array consists of a set of transmitting and receiving elements

that are placed in proximity of each other. According to this configuration and by
assuming a far field target to have the same RCS response to the probing signals
coming from the multiple transmitters, at the receiver side, the different transmitterreceiver channels can be coherently combined. An array pattern associated to an
extended virtual array can then be synthesised [24, 27, 72].
To better clarify this assertion, let us refer to a single target scenario and to the
MIMO array configuration which is shown in Fig.5.1. A receiving linear array of NR
elements is considered and it is placed in one of the interleaving spaces of a periodic
sequence of transmitting linear arrays. Each of the transmitting arrays consists of NT
1 This chapter is based on the article [J2] and on research material currently in preparation for
submission (a list of the authors publications is included at the end of this dissertation, p. 141).
i
i
i
58
5. Coherent MIMO array theory
elements. DT is the distance between two sequential transmitting arrays and MT is

the total number of arrays acting as transmitters. The receiver elements spacing is
dR , whereas the one between two transmitters is dT .
The target echo at the j-th receiver can be represented as:
!
NX
T MT
aT,i ()si (t) + n(t)
(5.1)
xj (t) = aR,j ()
i=1
and then collected into the NR 1 vector:

xR (t) = aR ()aT H ()s(t) + n(t),
(5.2)
where is the complex response of the target RCS, aR () and aT () are the receiving
and the transmitting steering vectors2 respectively, n(t) represents the additive noise
contribution and:
s(t) = [s1 (t), s2 (t) . . . sNT MT (t)]
(5.3)
is the vector of the transmitted waveforms. The virtual array synthesis depends on
the capability of identifying the signals belonging to the different emitters. Probing
signals, which are orthogonal with each other in a certain domain, are then employed
at transmission. This characteristic is usually referred to as waveform diversity or
waveform orthogonality [7376] and further details about this concept are provided
in Sec.5.2.
Once all the emitter contributions have been separated and by neglecting the noise
term, the angular dependency of (5.2) is represented by the combined steering vector:
a() = aR () aT H (),
(5.4)
Figure 5.1: Periodical array configuration of transmitting and receiving elements.

2 Here
the transmitting steering vector has to be meant as a result of the signal processing which
is performed at the receiver side. There is indeed no focusing, i.e. no steering, of the beam during
the transmission.
i
i
i
5.1 Coherent MIMO array pattern synthesis
59
where represents the Kronecker product. As a consequence, the virtual array

pattern behavior depends on the mutual positioning of the antenna elements.
By exploiting the exponential notation, as it has been done in Ch.2 for the ULA case,
and with respect to the structure in Fig.5.1, we have:
FT ()
NX
T 1
ejref [
ej
2
ndT
sin()
n=0
+ ej
2
DT
sin()
NX
T 1
ej
2
ndT
sin()
+ ...
n=0
ej
2
DT (MT 1) sin()
NX
T 1
ej
2
ndT
sin()
n=0
NX
T 1
ejref [
ej
2
ndT
sin()
n=0
MX
T 1
ej
2
mDT
sin()
],
(5.5)
m=0
where ejref is the arbitrary phase reference of the array and a pointing direction
0 = 0o is assumed.
By looking at the retrieved expression of the array factor in transmission, it is interesting to observe that it can be described as the combination of two contributions:
a first term that is associated to the spacing of transmitter elements dT , as for any
linear array, and a second term which is related to the periodicity distance DT among
the multiple transmitter arrays. According to this analysis, (5.5) can be rewritten as:
FT () =
sin [ NT dT sin()] sin [ MT DT sin()]
.
sin [ dT sin()]
sin [ DT sin()]
(5.6)
Similarly, the array factor associated to the receiver array is given by:
FR () =
sin [ NR dR sin()]
,
sin [ dR sin()]
(5.7)
leading to the final expression of the MIMO array factor which is:
F ()
= FT () FR ()
sin [ NT dT sin()] sin [ MT DT sin()] sin [ NR dR sin()]
. (5.8)
=
sin [ dT sin()]
sin [ DT sin()]
sin [ dR sin()]
The expression presented in (5.8) refers to a MIMO array based on linear radiating
i
i
i
60
elements. The related behavior depends on the periodicity parameters (dR ,dT , DT )
and, if the aim is to resemble to characteristic of a conventional linear array, they
have to be properly selected. Since the achievable angular resolution is dictated by
the combined aperture of the different arrays and the appearance of grating lobes must
be avoided, two conditions have to be imposed on (5.8). At first and by referring to
the configuration in Fig.5.1, if we consider a dR spacing larger dT , then grating lobes
are generated in the receiver array pattern and they must be compensated by the
nulls of the transmitting array pattern:
(
NT dT sin() = k
dR = NT dT .
(5.9)
dR sin() = k
Secondly, the side-lobes of the receiving array must be attenuated by the periodicity
of the transmitting array series:
(
DT sin() = k
D T = NR d R .
(5.10)
NR dR sin() = k
By considering (5.8), (5.9) and (5.10), the expression of the MIMO array factor becomes:
sin [ MT NR dR sin()]
F () =
sin [ dT sin()]
sin [ MT NR NT dT sin()]
.
(5.11)
=
sin [ dT sin()]
It can be noticed that (5.11) represents the array pattern of a linear array of MT NR NT
elements whereas the effective number of elements is equal to NT MT +NR . This result
represents the additional gain, in terms of degrees of freedom, provided by the MIMO
processing with respect to the conventional array techniques. By considering the
element spacing dT , that is present in (5.11), the achieved 3dB aperture associated
to the transmitting and receiving array is given by:
3dB =
.
NR NT MT dT
(5.12)
A synthesis result example is shown in Fig.5.2 where the contributions belonging to

the different patterns have been highlighted. The following physical array parameters
have been chosen: dT = 0.5, MT = 2, NR = 8 and NT = 4.
5.1.1
Fourier-like transform representation of a MIMO array pattern
The transmitter and receiver arrays can also be seen as two linear filters aimed at
performing spatial sampling of the signal of interest. In the spatial domain the wavelength can be considered as the independent variable and according to that and to
i
i
i
5.1 Coherent MIMO array pattern synthesis
61
Figure 5.2: Different MIMO pattern contributions and realised pattern synthesis.
the geometry in Fig.5.1, the sampling function which represents the array behavior
of the receiver antennas can be expressed via a sequence of Dirac pulses of the form:
sR (, ) =
NX
R 1
( nr dR sin() Rref ),
(5.13)
nr =0
where Rref is a spatial delay associated with the position of the array. In the transmitters section the periodicity related to the term DT can be represented as a convolution product between two pulse sequences:
sT (, )
NX
T 1
( nt dT sin() Tref )
nt =0
MX
T 1
( mDT sin()),
(5.14)
m=0
being Tref a common spatial delay reference for the transmitter sections.
According to (5.13) and (5.14), and considering the linear characteristic of the domain,
the impulse response of the overall system is:
sM IM O (, ) = sR (, ) sT (, ),
(5.15)
i
i
i
62
that can also be represented in a transformed domain by means of the Fourier operator:
Z
F{} = ej d.
(5.16)
Thanks to the analytical properties of both the Dirac sequences and the Fourier
transform, (5.13) and (5.14), after a few passages, can be rewritten as:
SR (, ) = ejRref
sin(
2 NR dR sin())
sin(
2 dR sin())
(5.17)
and:
ST (, ) = ejTref
sin(
2 NT dT sin())
sin(
2 dT sin())
sin(
2 MT DT sin())
,
sin(
2 DT sin())
that are equivalent to (5.6) and (5.7) for =
5.2
(5.18)
2
.
Waveform diversity/orthogonality concept
Although the topic of MIMO radars has started to gain a large interest in the last
years, it is useful to clarify the concept of waveform diversity, which is the basic
property of the transmitted waveforms in a MIMO system. The term waveform
diversity refers to the property of the radar probing signals to be separable, i.e.
orthogonal, in a certain domain. By associating a specific waveform to each of the
transmitters and by exploiting this orthogonality feature, the contributions coming
from the different emitters can be retrieved at the receiver side. A further clarification
is needed.
The signal model under the coherent case assumption, as it has been presented in (5.2),
refers to a far field point scatterer. By considering that the condition of separability
between the waveforms is applied at the receiver, from the theoretical point of view
and in the most general case, it can be written:
Z
si ((t )) sj ( 0 (t 0 ))dt = ij i (, 0 , , 0 ),
(5.19)
T
where i is the complex ambiguity function [70, 77] related to the i-th waveform and
=
1 vD /c
1 + vD /c
(5.20)
is the speed stretching factor [78]. The radial speed of the target is represented by
vD . It is known that the result of the integral in (5.19) is independent of the actual
i
i
i
5.2 Waveform diversity/orthogonality concept
63
values (, ) and it can be expressed by means of the relative values = 0 and

= 0 [79]:
Z
si (t) sj ((t ))dt = ij i (, )
(5.21)
T
which is usually just simply written as:

Z
si (t) sj ((t ))dt = ij i (, ).
(5.22)
The requirement expressed by (5.22) is often replaced by the less stringent orthogonality condition among the transmitted waveforms:
Z
si (t)sj (t)dt = ij i (0, 0),
(5.23)
T
that does not take into account for the delays and the speed components introduced
by the target reflections [80]. To include these contributions and by means of the
following narrow band assumption:
s((t )) s(t )ej2fD t ,
the expression in (5.23) can be rewritten as:
Z
si (t)sj (t + )ej2fD t dt = ij i (, fD ).
(5.24)
(5.25)
As correctly pointed out in [81], by considering for simplicity in the explanation the
case of zero Doppler speed, the following equivalence holds:
Z

si (t)sj (t + )dt = F 1 Si (f ) Sj (f ) ,
(5.26)
T
where F 1 represents the inverse Fourier transform. It is clear from (5.26) that the
orthogonality condition between the two signals is achieved by having non overlapping
spectra of the waveforms in the frequency domain. A softer constraint can be
obtained by limiting the validity of (5.26) within the time interval of the target
response. To this aim, in [81] a method based on the application of a weighting
function3 is proposed.
3 In
this case (5.25) becomes:

Z
h( )si (t)sj (t + )ej2fD t dt = ij i (, fD ),
(5.27)
where h( ) may represent either a delay dependant weighting function or, as in the case of the
referred paper, a pattern nulling provided by the different observing angles of the target of interest.
i
i
i
64
If (5.25) is valid, the contribution of the m-th transmitter can be retrieved from (5.2):
Z
xm (t) = xR (t)sm ( 0 )d 0 ,
(5.28)
and by iterating the procedure for all the waveforms, the NR NT MT 1 received
MIMO vector can be synthesised. Since the behavior of the virtual array depends
on the capability to separate the channels at the receiver side, a combined pattern
degradation due to the non ideal separation among the channels is expected.
It is worth of mention that the analysis presented in this paragraph has considered
MIMO transmissions without a focusing capability of the transmitting array, i.e. omnidirectional transmissions. If agility on transmission is available, hybrid techniques
which combine spatially orthogonal beams and signal orthogonality can be implemented [20].
5.3
Effect of the illumination errors on the pattern synthesis
It has been shown in Ch.2 that the array pattern synthesis, and the behavior of
the related SLL, is subjected to a series of errors which depends on the subsystem
components. The presence of these errors degrades the expected performance of the
array in the way that has been presented in Sec.2.3.2. In addition to the consequences
which follow from the non ideal separation of the multiple channels, MIMO arrays
are also influenced by the illumination errors. Their impact on the pattern synthesis
is here after illustrated. The omnidirectional antenna element case is considered in
the following text; the terms array pattern and array factor are then used in an
equivalent way.
By referring to the linear array case and by means of the same notation which has
been used in Ch.2, the MIMO array factor can be written as:
F ()
=
=
FT () FR ()
N
1
X
an (1 + n )ej(kndT sin +n )
n=0
M
1
X
bm (1 + m )ej(kmdR sin +m ) .
(5.29)
m=0
Here N represents the number of transmitters and M the number of receivers4 . Accordingly, the fractional amplitude errors are n and m whereas the phase error
4 The
equivalence with Fig.5.1 is obtained by considering NT = 1 and MT = NR .
i
i
i
5.3 Effect of the illumination errors on the pattern synthesis
65
components are written as n and m .

The effect of the illumination errors on the pattern sidelobes is related to the behavior
of the power pattern, which is obtained by multiplying (5.29) with its complex conjugate value. By rearranging the different series which are present in the expression,
we obtain:
2
|F ()|
=
=
F () F ()
N
1 M
1 N
1 M
1
X
X
X
X
an an0 bm bm0 (1 + n )(1 + n0 )(1 + m )(1 + m0 )
n=0 m=0 n0 =0 m0 =0
ejk[(nn )dT +(mm )dR ] sin() ej(n n0 ) ej(m m0 ) .
(5.30)
Due to the random characteristic of the expression in (5.30), as in the conventional

array case, the pattern behavior depends on the statistical distributions of the amplitude and phase illumination errors. In order to retrieve the expected performance of
the virtual array sidelobes, the mean value estimation of (5.30) must be computed.
Four different contributions can be distinguished in accordance with the multiple combinations of the series indexes (n, n0 , m and m0 ). The overall average array pattern
can indeed be written as:
D
E
D
E
D
E
2
2
2
|F ()|
=
|F ()| (n6=n0 ) + |F ()| (n=n0 )
(m6=m0 )
(m=m0 )
D
E
D
E
2
2
|F ()| (n6=n0 ) + |F ()| (n=n0 ) .
(m=m0 )
(5.31)
(m6=m0 )
Let us now consider the first term in (5.31). By referring to the inequalities of the
indexes and to (5.30), being
h(1 + n )(1 + n0 )(1 + m )(1 + m0 )i = 1,
it results:
D
E
2
|F ()| (n6=n0 )
(m6=m0 )
N
1
X
M
1
X
1
N
1 M
X
X
(5.32)
D
ED
E
an an0 bm bm0 ej(n n0 ) ej(m m0 )
n=0
m=0 n0 =0 m0 =0
(n6=n0 ) (m6=m0 )
ejk[(nn )dT +(mm )dR ] sin() ,
(5.33)
since the absence of systematic errors is assumed, i.e. the random distributions of the
illumination errors have mean values equal to zero5 . The mean values in (5.33) can
be evaluated by applying the change of variable y = n n0 (in a similar way for the
other contribution), that allows writing:
E Z
D
ej(n n0 ) =
ejy p(y)dy.
(5.34)
5 According
to this assumption we have h(1 + i )i = 1 for i = n, n0 , m, m0 .
i
i
i
66
The probability density function p(y) must be defined in accordance with the one of
i , that here is assumed to be gaussian6 for i. In this case, being n and n0 two
statistically independent random variables and with the same normal distribution
N (0, 2 ), p(y) results [82]:
2
p(y) =
y2
1
e 4 .
2
(5.35)
By combining (5.34) and (5.35):

D
ej(n n0 )
=
=
and by recalling that [83]:

Z
2
1
ejy e
y2
4 2
cos(y)e
dy
y2
4 2
dy,
(5.36)
cos(2px)ea
x2
dx =
p22
e a ,
2a
(5.37)
after few steps we obtain the following final expression:

D
E
2
ej(n n0 ) = eT
(5.38)
and similarly:
D
E
2
ej(m m0 ) = eR .
(5.39)
The subscripts ()T and ()R have been introduced to distinguish among the transmitter and the receiver components respectively.
The results in (5.38) and (5.39) can now be used to simplify the expression in (5.33):
D
E
2
|F ()| (n6=n0 )
e(R +T )
(m6=m )
N
1
X
M
1
X
n=0
m=0
(n6=n0 ) (m6=m0 )
N
1 M
1
X
X
n0 =0
ejk[(nn )dT +(mm )dR ] sin() .
an an0 bm bm0
m0 =0
(5.40)
By considering the argument of the series in (5.40), it can be noticed that it represents
the difference between the ideal MIMO array pattern and the mixed terms which are
6 p(
i)
2
i
2
i
2 2
i
i
i
i
67
denoted by the other combinations of the indexes in (5.30). We can then write:
D
E
2
|F ()| (n6=n0 )
(m6=m0 )
N
1 M
1
X
X
2
2
|an |2 |bm |2
e(T +R ) |F ()|2i
n=0 m=0
|
{z
}
(n=n0 )
(m=m0 )
M
1
X
"
2
|FT ()|2i
|bm |
m=0
N
1
X
#
2
|an |
n=0
{z
(n6=n0 )
(m=m0 )
N
1
X
|FR ()|2i
|bm |2
m=0
{z
}
|an |2
n=0
= e
M
1
X
"
(n=n0 )
(m6=m0 )
(2 +2 )
T
"
|F ()|2i +
N
1 M
1
X
X
|an |2 |bm |2
n=0 m=0
|FT ()|2i
M
1
X
|bm |
|FR ()|2i
m=0
N
1
X
#
2
|an |
(5.41)
n=0
where the subscript ()i refers to the ideal pattern behavior. For the transmitter and
the receiver array components in (5.41), the same representation, as the difference
between the ideal pattern and the residual terms, has been used. By following analogue steps, the other addends in (5.31) can be evaluated; for (n = n0 ) and (m = m0 )
we obtain
N
1 M
1
D
E
X
X
2
2
2
|an |2 |bm |2 ,
|F ()| (n=n0 ) = (1 + T )(1 + R )
(m=m0 )
(5.42)
n=0 m=0
being 2T and 2R the mean square values of the fractional amplitude errors for the
transmitter and the receiver arrays respectively.
i
i
i
68
The remaining terms are represented by:

D
E
2
|F ()| (n6=n0 )
eT (1 + 2R )
(m=m0 )
M
1
X
N
1 N
1
X
X
|bm |2
n=0 n0 =0
(n6=n0 )
m=0
(1 +
2R )
M
1
X
an an0 ejk[(nn )dT ] sin()
"
2
|FT ()|2i
|bm |
m=0
N
1
X
#
|an |
(5.43)
n=0
and:
D
E
2
|F ()| (n=n0 )
= eR (1 + 2T )
(m6=m )
N
1
X
|an |2
(1 +
2T )
N
1
X
M
1
X
bm bm0 ejk[(mm )dR ] sin()
m=0 m0 =0
(m6=m0 )
n=0
M
1
X
"
|an |
|FR ()|2i
M
1
X
#
|bm |
(5.44)
m=0
n=0
The expressions in (5.41), (5.42), (5.43) and (5.44) can now be used to evaluate the
analytical form of the average array pattern in (5.31). All the steps involved in the
pattern evaluation are depicted and explained in Appendix B, here the final expression
of the normalised pattern is presented:
D
E
2
' |F ()|2iN ORM +
|F ()|
N ORM
M
1
P
h
i
|bm |2 2R + 2R
+ |FT ()|2i m=0

N 1 M 1
P P
2
|an ||bm |
n=0 m=0
NP
1
h
i
|an |2 2T + 2T
n=0
+ |FR ()|2i
NP
1 M
1
P
2 .
|an ||bm |
(5.45)
n=0 m=0
Three main terms can be distinguished. Whereas the first one is the ideal behavior
of the normalised MIMO array pattern, the other two contributions can be described
as weighted sums of the ideal patterns belonging to the transmitter and the receiver
arrays. The weighting coefficients result to be dependant of the receiver array
errors distribution, for what it concerns the ideal transmitter array pattern, and in
a reciprocal way for the receiver array side. This result differs from what has been
observed in Ch.2 for a conventional linear array.
In the ULA case it was indeed possible to retrieve an expression of the average pattern
that was completely angle independent. With respect to the MIMO system, this can
i
i
i
69
only be done in terms of an inequality condition that therefore yields to an upper

bound for (5.45).
M 1
2
N 1
2
P
P
2
2
|bm | , from (5.45)
|an | and |FR ()|i
If we consider that |FT ()|i
m=0
n=0
it can straightforwardly be written:

D
E
2
|F ()|
|F ()|2iN ORM +
N ORM
h
i
M 1
!2 P |b |2 2 + 2
m
R
R
m=0
|an |
N 1 M 1
2
P P
n=0
|an ||bm |
N
1
X
n=0 m=0
M
1
X
m=0
!2
|bm |
NP
1
n=0
h
i
|an |2 2T + 2T
N 1 M 1
P P
2
(5.46)
|an ||bm |
n=0 m=0
which, in the particular case of uniform illumination masks and after few simplification
steps, leads to:
i
h
i
h
D
E
N 2R + 2R + M 2T + 2T
2
.
(5.47)
|F ()|
|F ()|2iN ORM +
NM
N ORM
A further assumption can be taken with respect to the statistical characteristics of the
illumination errors. If the moments of the distributions are assumed to be identical
at both the receiver and the transmitter sides, then the final representation of the
upper bound for the average power pattern is obtained:
h
i
D
E
(N + M ) 2 + 2
2
|F ()|
|F ()|2iN ORM +
.
(5.48)
NM
N ORM
By referring to the expressions in (5.46), (5.47) and (5.48), some considerations must
be drawn. The approximation is based on the maximum value that can be assumed
by the transmitter and receiver array patterns. Since this value refers to the pointing
direction of the two patterns, the offset between (5.45) and (5.46) increases in the
sidelobes region far from the main lobe. As a consequence, yet being a conservative
approximation for the average power pattern behavior, the imposed limit may result
into an over dimensioned constraint for the system. This risk can be avoided by
a deeper analysis of the individual array patterns which can help in identifying a
better approximation for the upper bound in (5.48). An example is provided in the
following section where simulated array patterns based on the presented analysis are
also shown.
i
i
i
70
5.3.1
Simulated results
The reference structure for the simulation is a MIMO array with N = M = 6. The
receiver elements spacing is dR = 0.5 whereas the spacing of the transmitters, according to (5.10), is dT = 3. The term of comparison is represented by a ULA
with a total number of elements equal to N M = 36. Similar error distributions
are assumed at both the receiver and the transmitter sides; the same applies to the
conventional linear array. A Monte Carlo approach [84] has been used to generate the
array patterns affected by the illumination errors.
Some results with different values for the moments of the error distributions are plotted in Fig.5.3. Each plot illustrates the normalised behaviors of: the Ideal MIMO
array pattern; the Average Error MIMO pattern, obtained by evaluating the mean
(a)
(b)
(c)
Figure 5.3: Error affected array pattern comparisons for (a) 2 = 2 = 0.01, (b) 2 = 0.01 and
2 = 0.1, (c) 2 = 2 = 0.1.
i
i
i
71
of all the patterns produced in the Monte Carlo runs; the Theoretical Error MIMO,
which represents the analytical expression in (5.45) and the Theoretical Error ULA,
which depicts the effect of (2.39) on the reference ULA. All the different sub-figures
confirm the validity of (5.45). The plots obtained by averaging the Monte Carlo runs
indeed fit with the closed form expression that has been retrieved.
It was described in Sec.2.3.2 that the presence of illumination errors introduces a uniform raise of the array pattern sidelobes in the ULA case. This effect can be observed
in all the plots of Fig.5.3 and it is due to the fact that the error contributions in (2.39)
are completely independent of the array pattern. If we refer to the MIMO case, and
specifically to (5.45), this condition does not hold. As a consequence, the appearance
of coherent sidelobes can be noticed. The locations of these sidelobes coincide with
the ones of the grating lobes of the transmitting array , which can be computed by
means of (2.19)
p
| sin() 1| =
with p = 1, 2, 3.
(5.49)
3
and they are equal to: 1 = 19.5o , 2 = 41.8o and 3 = 90o .
The raise of the grating lobes is explained by the fact that the illumination errors
introduce a de-pointing of the transmitter and receiver array patterns. The impact of
this de-pointing is negligible for what it concerns the main beam of the MIMO pattern.
On the other hand, the nulls of the receiver array pattern, which are supposed to
compensate for the grating lobes of the transmitting antennas, are also shifted. This
pointing offset produces a non ideal cancellation of the mentioned grating lobes and,
consequently, a considerable degradation of the synthesised MIMO array pattern.
This consideration allows further analysing the expressions retrieved in Sec.5.3. It
resulted that the level of the sidelobes, in the region which is far from the pointing
direction of the array, is mainly dictated by the weighted7 grating lobes produced by
the array with a larger element spacing. According to that it is reasonable to assume
2
|FR ()|i = 0 in (5.45) and by following the same steps we have:
i
h
D
E
2 + 2
2
.
(5.50)
|F ()|
|F ()|2iN ORM +
M
N ORM
The upper bound in (5.50) has been written with respect to the transmitter array,
since it is the one with a larger element spacing in the considered example; in the
opposite case it is obviously valid the substitution M N for (5.50). The inequalities
in (5.48) and (5.50) are compared in Fig.5.4. By referring to (5.45) and to Fig.5.4, a
number of considerations can be drawn:
the inequality in (5.50) represents a less stringent upper bound than the one
proposed in (5.48); it can indeed be observed that the requirement imposed
7 The
weighting depends on the tapering which is applied to the array.
i
i
i
72
(a)
(b)
Figure 5.4: Analysis of the upper bound conditions for the average error patterns for (a) 2 = 2 =
0.001, (b) 2 = 0.01 and 2 = 0.1.
by (5.48) overestimates the level of the sidelobes which are produced by the
illumination errors;
(5.50) does not require the a-priori knowledge of the original array pattern
behavior, that was the key condition for the approximation of (5.45). However
the proper fitting of the curves at the grating lobe locations suggests a correct
evaluation of the performance of the synthesised MIMO array pattern.
whereas the angular resolution of a MIMO array depends on the product between the transmitter and receiver number of elements, the sidelobes behavior
of the same array, when illumination errors are considered, is only inversely
proportional to the number of elements of the array with a lower spacing.
as a consequence of the previous statement and by referring to the same number
of both physical and virtual array elements, the sidelobes performance of a
conventional linear array have been proved to be always better than the ones
that can be achieved by the equivalent MIMO array structure.
5.4
Summary
The chapter has dealt with the analysis of the coherent MIMO arrays with particular
focus towards: the array pattern synthesis, the waveform diversity concept and the
theoretical characterisation of the illumination errors effect on the pattern synthesis.
Specifically, the virtual array concept has been introduced. The design choices that
are required for the dimensioning of such type of systems, in order to obtain well
i
i
i
5.4 Summary
73
defined performance in terms of angular resolution, have been highlighted. With respect to the synthetic array retrieval, the duality between the element space and the
Fourier/wavelength domain has also been shown.
By referring to the waveform diversity, and the related orthogonality concept, a clarification has been pointed out. It has been remarked that, even though the requirement about the probing signals separation is applied on the transmission side, from
a theoretical point of view, the consistency of the MIMO processing is based on the
capability of separating the signals at the receiver side. Therefore, a rigorous formalism should imply obtaining the orthogonality among the waveforms at reception.
Finally, a complete investigation about the impact of the illumination errors on the
MIMO arrays has been presented. Analytical expressions aimed at estimating the
expected pattern behaviors have been retrieved in closed forms. The study has allowed defining some relevant characteristics of the MIMO arrays and it has revealed
differences with respect to the conventional ULA systems. Among them, it has been
demonstrated that, whereas coherent MIMO arrays provide better angular resolution
with a fewer number of physical elements, when illumination errors are considered, a
degraded performance in terms of the achieved sidelobes level is expected.
i
i
i
74
i
i
i
Chapter 6
MIMO signal processing:
RADOCA test board and
experimental results
The chapter presents the radar test board that has been realised within the RADOCA project
at TNO and the studies which have characterised its experimental validation. The studies
were aimed at evaluating the highest performance which could be achieved by the MIMO
radar demonstrator in both the Doppler, the range and the azimuth domains. At first, the
design choices are illustrated; the antenna section is described with specific interest towards
the positioning of the antenna elements. This analysis is undertaken from both the RF and
the mutual location perspectives. As described in Ch.3, an issue that must be accounted for
an array based radar regards the definition of a reliable calibration technique. With respect
to the realised design and by considering the MIMO characteristic, a method and the related
performance are presented. The description then leads to the detection of the slow moving
targets, which is the main objective of the radar. Due to the transmission scheme that is used
to implement the waveform diversity, a limitation is introduced in terms of the maximum
unambiguous Doppler speed that can be measured. A transmitters selection technique to
overcome this issue is suggested and the simulated outcome is illustrated. The last section
of the chapter is dedicated to the study of high resolution techniques. By referring to the
MUSIC method, the validation of such techniques has been confirmed also in the case of
MIMO arrays. Furthermore, a novel two-dimensional algorithm has been developed and both
simulated and experimental results are shown. Final remarks and a summary of the main
i
i
i
76
6. MIMO signal processing: RADOCA test board and experimental results
achievements are depicted in the last section of the chapter1 .
6.1
RADOCA MIMO radar description
The design of the MIMO radar demonstrator, from an antenna section point of view,
was aimed at both achieving the required performance in terms of angular resolution
and obtaining a test board that could work as a conventional phased array. This
second characteristic was considered of particular interest for two reasons: it could
effectively demonstrate the added value of the MIMO processing by directly comparing the different configurations; it increases the system robustness in case of failure of
one of the transmitters, i.e. the absence of grating lobes within the scanning region
is ensured (yet with a reduced angular resolution).
According to these premises, the MIMO radar demonstrator in Fig.6.1 has been realised. The board consists of 2 transmitter pairs, one per each side of the array, and
of 8 receiving channels which are placed in the middle. The radar operates at X-band
in FMCW mode. The waveform can be adjusted in terms of bandwidth and pulse
duration: the maximum sweep rate and the maximum transmitted bandwidth are
20MHz/10s and 1.6GHz respectively. The sampling frequency of the ADC is set
to 1.4MHz. The MIMO orthogonality condition is realised in the time domain by
sequentially switching on the transmitters. Short range applications were addressed
during the design phase of the radar and the related requirement resulted in a maximum transmitted power of 9dBm. A direct benefit is represented by the possibility
to use a standard USB 2.0 to power the system, which is shown in Fig.6.2.(a). The
low data rate that characterises the FMCW system allows using the same connection
for the data acquisition. Thanks to all the mentioned features, the dimensions of the
system could be reduced, Fig.6.2.(b), and a conventional notebook, Fig.6.2.(c), can
be used to control the radar.
The transmitter and the receiver spacings are equal to /2. In order to have a correct
spacing of the MIMO element channels, /4 is the distance between the inner transmitters and the outer receivers. More details about the antenna section design are
provided in Sec.6.1.1. The selected receiver elements spacing is in contrast with the
analysis in Ch.5 about the dimensioning of the real arrays for the realisation of the
MIMO system. A distance of should indeed be used. However, the actual design
allows the system to be used in a Single-Input Multiple-Output (SIMO) mode with
only one active transmitter, which is one of the requirements for the demonstrator.
As a drawback, an overlap of the virtual channels is introduced and it is illustrated in
Fig.6.3. Out of 32 virtual available MIMO elements, only 18 channels are effectively
1 This chapter is based on articles [C1], [C2], [C3] and [C4] (the author publications list is
included at the end of this dissertation, p. 141).
i
i
i
6.1 RADOCA MIMO radar description
77
Figure 6.1: X-Band radar test board with MIMO functionality developed at TNO.
(a)
(b)
(c)
Figure 6.2: Details of the USB 2.0 connector (a), boxed test board (b) and radar system layout while
connected to the data acquisition notebook (c).
i
i
i
78
Figure 6.3: Virtual element relative positions of the MIMO array board.
independent and contribute to the angular resolution of the overall system. The redundant elements can obviously still be used in order to improve the SNR level.
6.1.1
Antenna and PCB design
The coherent MIMO approach, aimed at the synthesis of digital beams, requires a
uniform /4 spacing between the elements of the virtual array, as it acts like an active
phased array with colocated antennas. From the physical elements point of view, this
condition is achieved by placing the transmitter array at a /4 distance from the first
element of the receiver array. In order to facilitate the integration of the antennas
with the front-end electronics, a microstrip folded dipole (quasi-Yagi) antenna has
been designed.
The isolation between the /2 spaced elements in digitally steered array systems
is a central issue and the requirement becomes more stringent when the antenna
elements are closer. Recent studies [8588] referring to similar layouts have shown
average isolation levels in the order of 20dB for closely packed and strongly coupled
antennas. Here, the 9dBm transmitted power should not cause compression and thus
de-sensitization of the receiver. This condition requires an effective 15dB isolation.
Since a straightforward printed dipole antenna on a Rogers RO4350 [89] substrate
already measures 10mm in width, the effective required spacing of 8mm (/4 at the
reference frequency of 9.4GHz) together with the 15dB isolation cannot be achieved
easily.
The basic antenna design uses an open dipole radiator in combination with a single
director. The reflector is formed by the truncated ground plane of the microstrip. This
i
i
i
6.1 RADOCA MIMO radar description
79
Figure 6.4: Layout of /4 (8mm) spaced microstrip fed quasi-Yagi antenna.
basic design was already described in [90]. In order to realise the /4 spacing, the
open ends of the dipole have been bent over 90o to reduce the width of the dipole from
10mm to 7.2mm. The reduction in effective width supports an increase of the azimuth
beamwidth as required. The actual structure resulted in a 0.8mm spacing between
the TX and RX antennas which would normally result in a very low isolation of much
less than 10dB, leading to an un-usable receiving channel. However, the problem
is solved by the balanced nature of the antenna. The overall coupling between two
neighboring antennas is from center to center and can be seen as an inphase signal.
The ending sections of the dipoles of the two antennas that are /4 apart introduces
an out-of-phase signal to be coupled. As a result, an effective suppression of the /4
spaced transmitter-receiver couples is obtained. The basic straightforward design is
shown in Fig.6.4.
The balun for the antenna, effectively connecting the 50 Ohms front-end electronics
to the radiating antenna is formed by a simple /2 microstrip line, meandered to fit
Figure 6.5: Scattering parameter measurements of the /4 spaced elements.
i
i
i
80
the narrow spacing requirement. The measured result of the /4 spaced antenna pair
shows an isolation of better than 15dB and a better than -10dB input reflection over
an 8GHz to 12GHz frequency range as it is shown in Fig.6.5. The /2 spaced antennas
exhibit more than 25dB of isolation over the same frequency range. The estimated
antenna gain is around 4dBi. The front-end electronics and the antenna require only
a single board of 0.254mm thickness for all interconnections. For mechanical stability
reasons, the final design consists of a 4-layer board with a total thickness of 1.4mm.
6.2
Board calibration
The importance of the calibration procedure, and the related implications on the
performance of an array based radar system, has already been analysed in Ch.2 and
Ch.3. Calibration techniques aimed at compensating for the errors that derive from
the MC effect are based on the evaluation of the MC matrix. Those techniques
require antenna pattern measurements for each element of the array [91]- [92]. As it
has been shown in Ch.4, a reference transmitter from different and known angular
positions can also be used to perform the matrix estimation. Here a technique, which
exploits the design characteristics of the coherent MIMO system described in Sec.6.1,
is illustrated.
By referring to (5.2) and (5.4), the angular response of a far field point scatterer can
be represented by the vector:
x = aR () aT () + n,
(6.1)
where is the complex amplitude of the target located at the angular position and
n is the vector of the white gaussian noise components. In order to highlight the
effects of the calibration errors, (6.1) can be rewritten as:

x = AT ejT 1 ej1 2 ej2 ejkd sin 3 ej3 ejk2d sin
i
N M ejN M ejk(N M 1)d sin + n,

(6.2)
where = AT ejT , 1 . . . N M and ej1 . . . ejN M represent the amplitude and phase
channel errors respectively, k = 2
and d is the spacing of virtual MIMO elements.
By considering that the evaluation of the calibration error components is meaningful
in terms of the mutual dependance between the different channels, it is possible to
normalise (6.2) for a given complex value without loosing any information. If the
normalisation is performed with respect to the first channel value, we have:

2 ej2 jkd sin 3 ej3 jk2d sin
=
e
e
x
1
1 ej1
1 ej1

N M ejN M jk(N M 1)d sin
,
e
+n
(6.3)
1 ej1
i
i
i
6.2 Board calibration
81
that can be written in an equivalent way as:

1 a2 ej2 ejkd sin a3 ej3 ejk2d sin
i
,
aN M ejN M ejk(N M 1)d sin + n
(6.4)
being ai = 1i and eji = ej(i 1 ) .

According to (6.4), the phase errors can be seen as offsets from the linear behavior of
the steering vector phase; the amplitude errors are the fractional values which deviate
from the unit. As already seen in Sec.3.2.1, those errors are the effect of a complex
weighted sum of the interferences coming from all the channels of the real array. Thus,
the m-th element of (6.4) is represented by:
x
m =
cmm xim
N
X
cmn xin = am ejm ejk(m1)d sin + n

m.
(6.5)
n,n6=m
With respect to a MIMO system, we know that the spacing between similar elements,
for instance the receivers, is proportional to the number and the spacing of the other
elements [24]; an example of this positioning is depicted in Fig.6.6. The choice is
made in order to avoid the overlapping of the virtual elements. When the distance
between the elements becomes larger with respect to the wavelength, it is reasonable
to consider the mutual coupling interaction negligible. Thus, we can write from (6.5):
cmm xim am ejm ejk(m1)d sin
(6.6)
cmm am ejm ,
(6.7)
and then:
that can be directly measured from the observations. Considerations drawn to retrieve
(6.7) are equivalent to say that for MIMO array composed by many transmitter/receiver elements, the mutual coupling matrix which characterises the real structure
reduces to a diagonal matrix.
Figure 6.6: Required spacings of the receiver elements for increasing number of the transmitters.
i
i
i
82
Figure 6.7: Measured and expected phase behavior of the reference scatterers.
6.2.1
Experimental results
Design choices of the MIMO radar, related to the integration of all the RF components
on a single layer board, do not allow to perform data acquisitions in receiving mode
only, i.e. without switching on the transmitters. The array transmitters have then
been used for the experiment and the reflection from three different scatterer points
have been considered as references. The choice of multiple reference points was meant
to compensate for the signal fluctuations caused by the presence of noise. Average
values have then been exploited for the compensation. As described in Sec.6.1, the
spacing between the receiver elements of the board is equal to /2, which produces
(a)
(b)
Figure 6.8: Measured and average (a) phase offset and (b) fractional amplitude values for the reference scatterers.
i
i
i
6.2 Board calibration
(a)
83
(b)
Figure 6.9: Effect of the calibration on the pattern synthesis (a) first scatterer (b) second scatterer.
an overlapping of the MIMO virtual elements. A secondary effect, introduced by this

spacing value, is that the linear correspondence in (6.7) is not completely satisfied and
a partial degradation of the compensation algorithm is introduced. For the purpose
of the experiment this degradation is tolerated.
The radar was set to transmit the maximum bandwidth of 1.6GHz with a sweep
duration of 2.6ms and the responses of the 18 independent MIMO elements were considered. In Fig.6.7 the phase behaviors of the measured scatterer points are depicted
and compared with the expected linear behavior; the expected values are computed
(a)
(b)
Figure 6.10: Effect of the tapering on the calibrated and un-calibrated patterns (a) first scatterer
(b) second scatterer.
i
i
i
84
as linear fitting out of the measurements. By following this procedure, the knowledge
of the target locations in the angular domain is not required. According to the estimated linear fitting, the cmm coefficients, which are used for the compensation, are
evaluated and presented in Fig.6.8.(a) and Fig.6.8.(b) for the phase and amplitude
quantities respectively. The values have then been used to restore the ideal shape of
the antenna pattern for the considered MIMO array structure; results of the calibration are illustrated in Fig.6.9 for two out of the three scatterers under analysis; similar
results also characterise the third reflector. The quality of the calibration method can
be observed and an overall improvement greater than 2dB can be distinguished in all
the plots. It is also interesting to observe the effect of the compensation method when
a side lobe reduction window is applied to the array. Fig.6.10 confirms the improvement in the pattern shaping that is provided by the proposed calibration. The effect
is particularly clear in the Fig.6.10.(b) where the achieved sidelobes level is equal to
-20dB and it is 6dB lower than the un-calibrated case.
6.3
Moving target detection in TDM MIMO radars
This section investigates the effect that a TDM based waveform orthogonality produces on the moving target detection capabilities. With respect to the system presented in Sec.6.1, the reference waveform which is considered for the analysis is the
FMCW. Exhaustive descriptions of the FMCW theory and some applications to modern radar systems can be found in the literature [69, 77, 93, 94]; a summary is also
presented in Appendix C. Here the beat signal:
1
sb (t) = ej2[fc + t 2 ] ,
(6.8)
as a result of the mixing operation between the transmitted and the received signals,
is considered. In (6.8), fc represents the carrier frequency, = B
T being B the
transmitted bandwidth and T the pulse duration and is the round trip time delay
associated to the target position. The beat signal is a sinusoidal signal with a range
(R) - beat frequency (fb ) dependence given by:
fb = =
2R
,
c
(6.9)
the maximum unambiguous range is then related to the sampling scheme which is
applied at reception. If a real sampling fs is considered, the maximum unambiguous
range results to be:
c
Rmax =
fs .
(6.10)
4
i
i
i
6.3 Moving target detection in TDM MIMO radars
85
Figure 6.11: Example of a FMCW transmission scheme for a TDM MIMO array.
When a moving target with radial velocity vD is taken into account, the time delay
associated to its position becomes:
=
2
2
(R + vD t) = 0 + vD t,
c
c
(6.11)
and by substituting (6.11) into (6.8), the beat frequency becomes:

fb 0 +
2vD
fc = 0 + fD .
c
(6.12)
Now, the measured beat frequency corresponds to the sum of two different components that depend on both the range and the speed information of the target. This
dependency introduces ambiguities in the range Doppler response of the target; the
main peak response is indeed shifted with respect to the true position. Different
waveform techniques, based on non-linear frequency modulations [95], can resolve the
issue. However, in the case of moderate velocities (or equivalently for short pulse durations) the shift introduced by the additional Doppler component can be tolerated.
Under the aforementioned assumption, the range information can be associated to the
frequency shift in the fast time (fs ) dimension, whereas the Doppler velocity can be
retrieved by means of a phase comparison on the slow time (T ) dimension. The phase
variation of the scatterer response, which is experienced in two consecutive sweeps, is
equal to:
= 2fD T,
(6.13)
and by considering that the unambiguous phase interval is [ ], the maximum
unambiguous Doppler frequency and speed result to be:
fD =
1
2T
vD =
.
4T
(6.14)
By referring to the MIMO case with a TDM approach and to Fig.6.11, the synthesis
of an entire MIMO baseline requires NT effective sweeps that are then equivalent to a
single transmission for the virtual array. As a consequence, the maximum unambiguous Doppler estimations in (6.14) are degraded by a factor of NT which depends on
i
i
i
86
(a)
(b)
Figure 6.12: FMCW Range/Doppler processing (a) FMCW Range/Doppler processing with a 3
stages MTI canceler (b).
the number of the active transmitters. This condition represents a severe limitation
on the performance of MIMO radar systems based on the mentioned transmission
scheme. A visual example is provided in Fig.6.12, where the radar demonstrator has
been set to transmit at fc = 9.25GHz a waveform with pulse length T = 363s and
bandwidth B = 500MHz. In Fig.6.12.(b) a 3 points Moving Target Indicator (MTI)
[69,77] is also applied to highlight the Doppler components belonging to target while
the suppression of the static clutter is performed.
Since the 4 transmitters are used and sequentially switched on according to the
approach in Fig.6.11, it can be seen that the unambiguous Doppler interval is equal
to vD = 4.9m/s. For the proposed case, the moving target of interest, consisting
of a slow manoeuvring car, could still be detected. Yet the degradation with respect
to a conventional FMCW radar, that would ensure a Doppler speed interval in the
range [19.6 19.6]m/s, is considerable. In Sec.6.3.1 and Sec.6.3.2 more details about
the effect of the Doppler component on the mentioned system are provided and in
Sec.6.3.3 a transmission scheme to overcome the mentioned limitation is proposed.
6.3.1
Doppler speed impact
The sketch in Fig.6.11 has helped in defining the difference between a conventional
and a TDM MIMO basic pulse. The latter has been noticed to be constituted by NT
effective pulses, each of them allows synthesising a part of the overall MIMO array.
By referring to the FMCW modulation principle, the beat signal belonging to the
i
i
i
87
m-th sub-array can be written as:

1
xm (t) = ej2[f0 m +m t 2 m ] ,
(6.15)
where m is now the round trip delay associated to the target for the MIMO sub-array
under analysis. If we consider the status vector v = [R0 , , vD ], which characterises
the distance, the azimuth position and the radial velocity of the target respectively,
the time delay takes the form (under far-field point scatterer assumption):

2
mdT + ndR
m =
R0 +
sin + vD t0 ,
(6.16)
c
2
being mdT the relative position of the m-th transmitter and ndR the relative position
of the n-th receiver; t0 represents the time of the performed measurement and it can
be expressed as:
t0 = T (m + kNT 1),
(6.17)
with k = [0, . . . , NM IM O 1], where NM IM O is the number of MIMO sweeps that are
considered for the processing. According to (6.17) and by neglecting the contribution
due to R0 , equation (6.16) can be rewritten in the following way:

dT
2
sin + vD T m
m =
c
2

ndR
+
sin + vD T (kNT 1) .
(6.18)
2
It must be noticed from (6.18) that the target speed vD is multiplied by both T
and NT T and this dependance affects the processing steps that are performed after
the matched filtering. Furthermore, the variable m is related to both the angular
information and the radial velocity; that effectively results in a de-pointing of the
array pattern associated to transmitting array2 .
6.3.2
Multi domain signal analysis
A conventional mean to analyse the performance of a radar system, in terms of achievable resolutions, ambiguities location and sidelobes shape, is based on the evaluation
2 If
we consider the m dependency in (6.18), we can relate it to the generic array factor:
F () =
N
1
X
an ej
2 n d
sin +vD T ]
(6.19)
n=0
By performing the array steering towards the direction 0 , without accounting for the speed component, the effective pointing direction results to be:

2vD T
00 = sin1 sin 0
.
(6.20)
d
i
i
i
88
Figure 6.13: TDM MIMO 3D matrix data structure.
of the waveform ambiguity function. With respect to the time domain division, the
virtual baseline synthesis requires the information from all the transmitters and it
implies that multiple basic pulses must be received. This leads to the definition of
the reference waveform:
s(t) = [s1 (t), s2 (t), . . . , sNT (t)]T .
(6.21)
By referring to the expressions (5.23)-(5.28) retrieved in Sec.5.2, we can write in a

matrix form that:
Z +
2

H
j2fD t

(6.22)
(v)
=
xR (t)s (t + )e
dt .
The analysis which is performed hereinafter is aimed at describing the behavior of

(6.22) in both the range/Doppler and azimuth/Doppler domain; the second analysis
depends on the fact that we are considering a TDM MIMO array, thus we can have
different information according to different steering angles. With respect to these
premises and by referring to (6.16)-(6.22) we can then define:
Z +
2

2v
j2 D t

dt
(v)| (R=R0 ) =
(6.23)
s(t)s (t + )e
i
i
i
89
and:
(v)| (=0 )
Z

=
s(t)s (t + )e
j2
2vD
2

dt ,
(6.24)
that are the ambiguity functions when the variation in the time delay is considered
in both the range and the angular domains.
The data structure of a TDM MIMO array can be represented by means of a 3D
matrix as illustrated Fig.6.13. The evaluation of the ambiguity functions, thanks to
the FMCW transmission, can be performed as it follows:
(6.23) is computed by firstly applying a FFT along the fast time dimension and
by selecting the 2D matrix corresponding to the range bin of the target. For a
given scanning angle , the matrix is multiplied by the steering vectors of the
transmitter and receiver elements; the sum along the receiver direction is done.
Finally, in order to estimate the Doppler contents of the signal, a second FFT
is performed along the transmitter channels direction.
15
100
15
10
90
10
80
70
60
10
50
15
40
50
100
150
Range [m]
(a)
200
250
Doppler speed [m/s]
Doppler speed [m/s]
(6.24) evaluation almost follows a reverse approach. At first, the steering vectors
associated to the target angle of arrival are multiplied by the 3D data matrix;
then the summation along the receivers direction is done and the corresponding
2D matrix is retrieved. Finally, the 2D FFT is computed in order to obtain the
desired function.
105
100
95
90
85
5
80
10
75
15
70
30
80 60 40 20
0
20 40
Azimuth angle [o]
60
80
(b)
Figure 6.14: Ambiguity functions of a target with status vector v0 = [50m, 20o , 3m/s] in a sequential
TDM transmission mode: Range/Doppler (a) and Angle/Doppler (b) maps.
i
i
i
Doppler speed [m/s]
90
15
105
10
100
95
90
0
85
5
80
10
75
15
70
80 60 40 20
0
20 40
Azimuth angle [o]
60
80
Figure 6.15: Angle/Doppler map for a target with status vector v0 = [50m, 20o , 3m/s] in a conventional FMCW radar with transmitted pulse length of NT T .
Let us now consider the case of a MIMO array with NT =NR =4. The transmitted
waveform has T =405s, B=150MHz and, for the beat signal, a sampling frequency
fs =1.4MHz has been chosen. By considering these values, the maximum unambiguous Doppler speeds are vD = 4.9m/s and the unambiguous range Ru '285m.
For a target status vector v0 = [50m, 20o , 3m/s], the range/Doppler and the azimuth/Doppler responses are presented in Fig.6.14. With respect to Fig.6.14.(a), it can be
noticed that side peaks are present in the Doppler domain and they are related to the
quantity vD (kNT T ) in equation (6.18). The spacing between the effective speed and
the side peaks is indeed equal to:
vof f =
.
2NT T
(6.25)
However, the amplitude of these spikes is negligible in the considered case3 . On the
other hand, if we refer to the azimuth/Doppler domain and to the results obtained in
Fig.6.14.(b), the effect introduced by the offset (6.25) is highlighted and an effective
reduction of the unambiguous Doppler interval can be clearly noticed. This result is
not in contrast with the theory as, in order to synthesise the MIMO virtual baseline,
the acquisition time is increased by a factor of NT . On the other hand, the presented
example differs from the case of a conventional FMCW radar with a pulse length of
NT T as it is shown in Fig.6.15. In this case, the ambiguities of the real target are
effectively located at Doppler speeds that are proportional to /(2NT T ).
3 We refer to the behavior of the target response. A similar consideration cannot be drawn in
presence of strong clutter reflections.
i
i
i
6.3.3
91
Effect of the random selection of the active transmitter
In the previous section it has been discussed the effect of the periodic term:
NT =
2
vD (kNT T ),
c
(6.26)
introduced by the use of multiple transmitters in a TDM MIMO system. It was noticed that the sequential selection of the active emitter effectively produces a reduction
of the unambiguous Doppler interval. This limitation, when multiple MIMO sweeps
are considered in the processing (which is always the case when accurate estimation
of the target speed has to be performed), can partially be overcome by introducing a
random selection of the active transmitter.
By referring to (6.34) we can write the time delay associated to the i-th transmission
as:

2
dT xi + dRxn
i =
R0 +
sin + vD t ,
(6.27)
c
2
where T xi is randomly chosen and:
t = T (i 1) +
k1
,
fs
(6.28)
15
100
15
105
10
90
10
100
80
70
60
10
50
10
15
40
15
50
100
150
Range [m]
(a)
200
250
30
Doppler speed [m/s]
Doppler speed [m/s]
being i = 1 . . . NT NM IM O and k = 1 . . . T fs . The random transmitter sequence is

realised in order to have the same number of transmissions by each element. In this
way, as in the sequential transmission mode, the contributions for the different virtual
sub-arrays are identically weighted.
Results of the randomization on the range/Doppler and the azimuth/Doppler do-
95
90
85
80
75
70
80 60 40 20
0
20 40
Azimuth angle [o]
60
80
(b)
Figure 6.16: Ambiguity functions of a target with status vector v0 = [50m, 20o , 3m/s] in a random
TDM transmission mode: range/Doppler (a) and azimuth/Doppler (b) maps.
i
i
i
105
15
Doppler speed [m/s]
10
95
90
85
80
10
100
10
95
90
85
5
80
10
75
75
15
105
15
100
Doppler speed [m/s]
92
15
70
70
80 60 40 20
0
20 40
Azimuth angle [o]
60
80
80 60 40 20
0
20 40
Azimuth angle [o]
(a)
60
80
(b)
Figure 6.17: Ambiguity functions of two targets with status vectors v0 = [50m, 20o , 3m/s] and
v1 = [50m, 35o , 10m/s] in a sequential (a) and random (b) TDM transmission mode.
mains are presented in Fig.6.16. From the comparison with the sequential approach,
two considerations can be drawn: in the range/Doppler case, the low amplitude side
peaks are now completely removed; in the azimuth/Doppler one, the introduced randomization produces a spreading of the ambiguous peaks and allow retrieving the
position of the target in the larger Doppler interval related to the single FMCW pulse
duration.
It is of interest to observe the results of the proposed scheme for the multiple targets
102
120
15
15
100
X: 20 Y: 3
Value: 102.5dB
10
98
96
Doppler speed [m/s]
Doppler speed [m/s]
10
5
10
92
10
90
15
15
80 60 40 20
0
20 40
o
Azimuth angle [ ]
(a)
60
80
110
94
X: 43 Y: 6.5
Value: 91.0dB
115
X: 41 Y: 13
Value: 102dB
5
105
X: 20 Y:3
Value: 121dB
100
80 60 40 20
0
20 40
Azimuth angle [o]
60
80
(b)
Figure 6.18: Effect of the number of integrated sweeps on the sidelobes level: (a) 16 MIMO sweeps
(b) 128 MIMO sweeps.
i
i
i
93
Table 6.1: Main to Maximum Sidelobe Level
MIMO Sweeps
MMSL [dB]
8
16
64
128
9.5
11.5
16
19
case that is presented in Fig.6.17. Here the scenario presents an additional target with
status vector v1 = [50m, 35o , 10m/s], which locates itself in the ambiguous Doppler
region of the sequential transmission. It is clear from the processing results that,
whereas with a conventional approach (Fig.6.17.(a)) a correct positioning of the target is not achievable, with the randomised (Fig.6.17.(b)) method the location in the
azimuth/Doppler map can be resolved.
As observed in the azimuth/Doppler results, which are presented in Fig.6.16 and
Fig.6.17, the introduced randomization produces a spreading of the ambiguous peaks.
The amplitude of the random peaks has been checked for different values of the integrated sweeps and it has been noticed that the average level of the unwanted peaks
decreases as the number of integrated sweeps increases. Such behavior is illustrated
by the comparison in Fig.6.18, where the cases of 16 and 128 integrated sweeps are
considered.
The MMSL (Main to Maximum Sidelobe Level) varies according to the same behavior ( N1 ) and Table 6.1 depicts the values for different numbers of integrated
sweeps for the simulated cases. It can be seen that the value of the highest sidelobe
remains quite high for a number of integrated MIMO sweeps lower than 64, which
implies that the proposed method for extending the unambiguous Doppler range can
only be used when long integration time is available. On the other hand, the randomization of the transmitters sequence can be used in conjunction with a conventional
transmission scheme in a subsequent bursts mode. In this way, whereas the conventional sequential transmission is used to perform the targets detection in the extended
Doppler interval, the random burst allows resolving the ambiguities on the identified
objects.
It must also be highlighted that no degradation of the system performance in terms
of both angular, Doppler and range resolutions has been measured when the different
transmitters are equally used during the data acquisition.
i
i
i
94
6.4
High resolution techniques applied to coherent MIMO

arrays
The previous section has analysed a possible mean to improve the Doppler speed retrieval capability of a coherent MIMO radar based on a TDM transmission scheme.
The attention is now given to the other domains that characterise the target status
vector: the range and the angular ones.
As it has been discussed in Ch.2, the angular resolution of an antenna system is
dictated by the working frequency (wavelength) and the antenna aperture [29]. The
range resolution depends on the characteristics of the transmitted signal, specifically
on the waveform bandwidth [70]. The aforementioned quantities represent the physical resolutions of a radar system. The application of spectral analysis techniques to
the angular domain in the one-dimensional (1D) case [96,97] and the two-dimensional
(2D) extension (azimuth and elevation) estimation [98, 99] have demonstrated the
improvement related to the high resolution processing. Similar approaches have also
been used in the range dimension [100].
In Sec.6.4.2, a novel technique that combines both the angular and the range high
resolution processing is presented. Since the algorithm is based on the MUltiple SIgnal Classification (MUSIC) estimator, the related basic theory is briefly illustrated in
Sec.6.4.1.
6.4.1
The MUltiple SIgnal Classification (MUSIC) method
The MUSIC algorithm [101104] is a parametric estimator for the localization of

signals in white Gaussian noise. Since the algorithm performs a decomposition of the
received data covariance matrix, in order to evaluate the eigenvalue amplitudes and
distribution, it is referred to as a subspaces technique.
The data matrix X is collected by an array of sensors. The size of X is [M, N ]
being M the number of sensors and N the number of acquired snapshots at a given
range bin (the application of the algorithm to the angular domain is considered). The
covariance matrix results to be:
Rx =
1
2
XXH = BRs BH + w
IM ,
N
(6.29)
2
where Rs is the signal sources covariance matrix, w
is the noise power and IM is the
identity matrix. The steering vectors related to the target locations are collected into
steering matrix:
B = [b(1 ) b(2 ) . . . b(s )] ,
(6.30)
and the structure of the vectors depends on the geometric characteristics of the array
as it has been explained in Ch.2.
i
i
i
6.4 High resolution techniques applied to coherent MIMO arrays
95
The MUSIC algorithm principle is based on the capability of separating the subspaces
related to the noise and the signal components. In order to retrieve this information,
an Eigenvalue Decomposition (EVD) of the Rx matrix is performed. The number
of eigenvalues belonging to the signal components is then evaluated. Most common
techniques to estimate the number of signals in white Gaussian noise environment
are: the Minimum Description Length (MDL) and the Akaike Information Criterion
(AIC) [105107]. The performance of both the estimators considerably decreases
when the scatterers are correlated, i.e. the signals are coherent with each other.
In this case the matrix in (6.29) is rank deficient and the signal eigenvalues can be
wrongly associated to the noise subspace. The issue can be solved by applying spatial
smoothing techniques [108, 109] to the received data matrix in order to decorrelate
the signals while restoring the high resolution capability of the MUSIC processing.
The identification of the signal subspace allows isolating the matrix Wn of the noise
eigenvectors. These eigenvectors are orthogonal to the signal subspace that is spanned
by eigenvectors of the form of b(). As a result, the squared Euclidean distance:
d2 = bH ()Wn Wn H b()
(6.31)
is minimal for all the values of which correspond to the target DOAs. It is then
possible to identify the function:
PM () =
1
,
bH ()Wn Wn H b()
(6.32)
that presents peak values in the same directions and is normally referred to as the
MUSIC pseudo spectrum.
6.4.2
2D-MUSIC algorithm description
The 2D (range and azimuth) information retrieval process, by means of high resolution algorithms, can be approached in two different ways. The first solution consists
of a 2 times 1D implementation: the MUSIC processing is applied on the angle
and the range dimensions separately and then the two sub-products are combined in
an incoherent way. The second solution allows retrieving the 2D information about
the object at the same processing step; it can then be defined as a fully coherent
2D processing and the novel algorithm, which has been implemented, belongs to this
category.
The implementation of the MUSIC algorithm model in the 2D case implies the definition of a new steering vector that includes both the range and the angular information
about the target. In order to determine this steering vector, the FMCW model, which
has been used in Sec.6.3, has been taken as reference and extended to the case of multiple targets. The beat signal received at the m-th channel, in presence of K targets,
i
i
i
96
takes then the form:

xm (t) =
K
X
k ej2[fc k (m)+k (m)t 2 k (m) ] + w(t),
(6.33)
k=1
being w(t) the noise component and where:

k (m) =
2
[Rk + md sin(k )]
c
(6.34)
allows highlighting both the range (Rk ) and the bearing (k ) positions of the target.
The model refers to a receiver sensor based on a liner array configuration.
The signal in (6.33) is then sampled with a sample frequency fs ; for a single sweep
we have the digital signal:
xm (n) =
K
X
k ej2[fc k (m)+k (m) fs 2 k (m) ] + w(n),
(6.35)
k=1
with n = 0...N 1 and N = T fs . It must be highlighted that, contrary to Sec.6.4, here

N represents the number of samples in the fast time (range) dimension. By collecting
the contributions from all the M channels, the X spatial/time sampled data matrix,
of size M N and that is associated to a single FMCW sweep is then retrieved. The
mentioned matrix is considered for the description of the 2D-MUSIC algorithm. The
processing steps can be summarised as it follows:
X is scanned by a window of dimensions [l1 l2 ] as shown in Fig.6.19. In total
we have p1 = M l1 + 1 positions that belong to the spatial domain of the
signal and p2 = N l2 + 1 positions associated to the time domain of the signal.
Each sub-matrix is reshaped into a [l1 l2 1] vector and collected into the new
According to the used notation, the dimensions of X
are [l1 l2
data matrix X.
p1 p2 ].
Figure 6.19: Data matrix samples and scanning window procedure.
i
i
i
97
The data smoothed covariance matrix can be computed as:

C=
i
1 h H
X
H ) J ,
XX + J(X
2p1 p2
(6.36)
being:
0
..
.
J=
0
1
0
.
0 ..
... 0
...
..
0
(6.37)
the [l1 l2 l1 l2 ] transition matrix [109]. After EVD operation, the (l1 l2 K)
can be identified.
eigenvectors associated to the noise subspace W
Finally, the 2D-MUSIC spectrum is evaluated as:
S(i ) =
1
W
H a(i )
aH (i )W
(6.38)
where the steering function vector a(i ) is equal to:

a(i ) =
(6.39)
h
l2 1
ej2[fc i (0)] , ..., ej2[fc i (0)+i (0) fs ],
ej2[fc i (1)] , ..., ej2[fc i (1)+i (1)
l2 1
fs ,]
ej2[fc i (l1 1)] , ..., ej2[fc i (l1 1)+i (l1 1)
l2 1
fs ]
iT
and the Residual Video Phase (RVP) [110] contribution has been neglected.
The maximum number of signals that can be detected is associated to the number
of degrees of freedom, i.e. number of elements, provided by the array. The use of
coherent MIMO radars then represents an interesting application, since this number is
larger than the physical elements one. The presented processing can straightforwardly
be applied to the MIMO systems; in that case the M value represents the number of
virtual channels that have been retrieved.
i
i
i
98
6.4.3
Simulated and Experimental Results
The validity of the algorithm has been tested on both simulated and real data scenarios. With respect to the first case and in order to resemble the system presented in
Sec.6.1, the radar parameters selection in Tab.6.2 has been used. The single channel
SNR has been set to 20dB and 10 MIMO sweeps have been integrated to compute
the covariance matrix in (6.36). The simulated scenario consists of 6 equal RCS point
targets that present an angular separation of 5o and a range separation of 0.6m; both
values do not allow to discriminate and separate the targets neither in range nor in
angle. The specific target positions (range/angle) are listed in Tab.6.3 and the angular reference is taken perpendicular to the x-axis at the 0m point. In Fig.6.20 the
results of the conventional DBF processing and the proposed 2D-MUSIC algorithm
are presented respectively. By referring to the latter one, the values l1 = 10 and
l2 = 280 have been used to compute (6.36). It can easily be noticed that, whereas
the standard processing does not provide a clear identification of the number and the
location of the targets, the high resolution method allows discriminating the presence
of the 6 different objects.
Table 6.2: Radar parameters selection for the simulated scenario
Number of transmitters
Carrier frequency
Transmitted bandwidth
Angular resolution
NT =4
fc =9.4GHz
B=250MHz
=7.2o
22
20.5
15
20
20
25
19.5
0
5
21.5
Y[m]
21
10
NR =4
fs =1.4MHz
T =363s
R=60cm
22
21.5
Y[m]
Number of receivers
Sampling frequency
Sweep duration
Range resolution
21
10
20.5
15
20
20
25
19.5
30
19
30
19
35
18.5
35
18.5
0
X[m]
(a)
40
0
X[m]
40
(b)
Figure 6.20: Simulated scenario with 6 targets spaced by 0.6m in range and 5o in angle: DBF
processing (a) 2D-MUSIC processing (b).
i
i
i
99
Table 6.3: Target positions for the simulated scenario
[R1 , 1 ] = [20.6m, 5o ]
[R4 , 4 ] = [20m, 5o ]
(a)
[R2 , 2 ] = [20.6m, 0o ]
[R5 , 5 ] = [20m, 0o ]
[R3 , 3 ] = [20.6m, 5o ]
[R6 , 6 ] = [19.4m, 5o ]
(b)
Figure 6.21: Cut at the 20.6m range bin for the 3 scatterers case (a); cut at the 20m range bin for
the 2 scatterers case (b).
More details about the processing results are shown in Fig.6.21 and Fig.6.22, where
the angular and range cuts at the peak locations of the MUSIC spectrum are presented. The DBF output curves at the same positions are also plotted. It can be
(a)
(b)
Figure 6.22: Cut at the 5o angular bin for the 3 scatterers case (a); cut at the 0o angular bin for the
2 scatterers case (b).
i
i
i
100
seen that the exact angular (5o , 0o and 5o ) and range (19.4m, 20.0m and 20.6m)
values have been identified for all the targets and the reduced sidelobes level introduced by the 2D-MUSIC method can also be appreciated. For the presented scenario,
the number of targets has been given as an input to the processing; thus it has not
directly been estimated from the simulated data set. However, in the real application,
it has been retrieved from the acquired data. Being not the purpose of this work to
analyse the behavior of the different signal estimation algorithms, the MDL method
proposed in [105] has been chosen. The scenario for the real data test is shown in
Fig.6.23 where the main features have been highlighted. The scene is characterised
by the presence of two handrails, multiple air pipe unions and a corner reflector that
was used as reference scatterer. The MIMO radar test board has been set to transmit
the following FMCW waveform: up-sweep duration 363s, down-sweep duration 37s
and transmitted bandwidth B = 600MHz. The down-sweep of the pulse is not considered during the processing but it is required by the board controller to ensure the
correct switching time from one transmitter to the other. Its value for the designed
board is normally taken equal to 10% of the up-sweep duration.
Results of the DBF and the 2D-MUSIC processing are shown in Fig.6.24. The data
smoothing factors for the second case have been chosen equal to l1 = 16 and l2 = 200;
the main reasoning resides on limiting the overall computation time required by the
processing. The optimal selection of the two parameters is an issue which has not
Figure 6.23: Scenario for the measured data set collection with highlighted targets.
i
i
i
6.5 Summary
101
(a)
(b)
Figure 6.24: Real data scenario: DBF processing (a) 2D-MUSIC processing (b).
been analysed in this work and it will be taken into account in the follow-on of the
research. By referring to the targets that have been considered in Fig.6.23, the improvement provided by the proposed two dimensional high resolution technique, with
respect to the DBF one, can be appreciated. Not only the positioning and the separation between the different objects is better assessed, but it can be noticed how the
handrail on the left side of the image is better highlighted by the 2D-MUSIC processing. It is then possible to state that the validity of the method is also confirmed by
its application to the experimental data set.
6.5
Summary
The chapter has introduced the compact and coherent MIMO test board that has
been realised at TNO. The design choices, which have led to the definition of the antenna section characteristics, have shown the capability of operating in two different
configurations: as a conventional phased array and as a MIMO array. The achieved
antenna element isolation, as direct consequence of the proposed layout, has been
capable of ensuring the expected working behavior of the radar. The application of
the signal processing techniques, which have been described throughout the chapter,
has confirmed both the validity of the implemented processing models and the quality
of the realised system.
Specifically, by exploiting the real array configurations in MIMO radar, which consists of receiving array elements spaced by a distance greater than the wavelength, a
simple calibration model has been retrieved. The model is based on the evaluation
of the calibration coefficients as offset values from the estimated theoretical phase
i
i
i
102
and amplitude behavior of the array. An a-priori knowledge of the target position in
the azimuth plane is not necessary. Yet accounting for the calibration performance
degradation, due to the redundant spacing of then antenna elements, the results have
shown a good compensation of the array non linearities and a good retrieval of the
ideal array pattern shape.
With respect to a FMCW modulation, the analytical description of the received signal
for the virtual MIMO array has been retrieved. By considering the phase behavior
of the signal, a double relation between the sweep duration and the target Doppler
speed has been observed. As a direct result, an effective reduction of the unambiguous Doppler interval is introduced by the sequential transmissions that are needed to
obtain the waveform diversity. A possible way to restore the full extent of the Doppler
region has been investigated and the introduction of a random selection of the active
transmitter has shown interesting results for this purpose. On the one hand, under
the assumption that all the transmitters were uniformly used, no degradation of the
system resolution was encountered. On the other hand, for a limited number of integrated sweeps, the level of the produced side lobes in the azimuth/Doppler domain
is not sufficiently low to ensure good detection performance. Surveillance systems
where relatively slow moving targets need to be detected can be considered suitable
applications for the presented technique. The RADOCA system falls into this category since one of its main tasks is the detection of human beings.
The last section of the chapter focuses on the topic of high resolution signal processing. An extension of the conventional MUSIC processing to the 2D case for both
azimuth and range information retrieval has been presented. By exploiting the characteristics of the array steering vectors in the time and the spatial domains, a fully
coherent algorithm has been defined. The model has been presented with respect
to the FMCW waveform but it can in principle be generalised to any transmitted
waveform. The validity of the proposed method has been assessed by means of both
simulations and experiments. In the former case, all the targets have been spaced by
angular and range distances lower than the theoretical resolutions; thus not allowing
a correct separation between them with conventional processing techniques. In the
latter case, a data acquisition was performed in an outdoor environment characterised
by the presence of multiple objects without the a-priori knowledge of the exact locations of the different targets. A comparison with the DBF technique has been carried
out and the improvements of the proposed method have clearly arisen in both cases.
Further developments of the algorithm can be identified. First of all it is known that
the spectrum obtained with the high resolution techniques does not refer to the real
amplitudes of the received signals, it is indeed commonly referred to pseudo-spectrum.
An estimation technique of the effective target reflectivity is then required, above all
when high resolution imaging applications are foreseen.
i
i
i
Chapter 7
Conclusion and outlook
The research that has been described in this thesis was conducted over a period of
four years in a cooperation activity between the Delft University of Technology and
the radar group of TNO - Defence, Security and Safety in the Hague. The project
focussed on the study of novel digital beamforming and signal processing algorithms
for array based radar systems.
In the first part of the thesis, the case of a circular array, meant to be used as a Passive Coherent Locator (PCL), has been investigated. The referred structure allows
simultaneously retrieving, by means of digital beamforming synthesis, the reference
signal coming from the chosen transmitter of opportunity and the signal reflected
by a possible target which is present in the region of interest. The synthesis of the
digital beams, which is aimed at retrieving the mentioned signals, must meet specific
requirements. Sidelobes level, interference suppression and angular resolution are parameters of interest while evaluating the beamforming performance. The research
activity has considered all these aspects, developed a novel technique for the array
pattern shaping and confirmed the radar performance with a data acquisition campaign. Furthermore, since the application of any digital beamforming scheme requires
a proper calibration of the array section, an ad-hoc procedure has been implemented.
The second part of the thesis has analysed the application of digital beamforming and
signal processing algorithms to the coherent Multiple-Input Multiple-Output (MIMO)
radars. This novel class of radar sensors has been the object of several research studies
in the last years. The applicability of conventional processing techniques, the limitations arising from the non ideal orthogonality among the transmitted probing signals
and the effective performance enhancement provided by the synthesised virtual array
i
i
i
104
7. Conclusion and outlook
are still under investigation. These concepts have been considered with a specific focus, on the one hand, on the analytical characterisation of the array pattern synthesis
and its degradation due to array non linearities introduced by the illumination errors;
on the other hand, novel techniques aimed at calibrating the real array system and
improving the related radar performance have been developed.
The technical introduction of the research study is included in Ch.2. Here all the
basic notions about antenna and antenna array theories are provided. These concepts have then been used, throughout the thesis, both to analyse the characteristics
of the antenna sections of the radar demonstrators and to qualify the performance of
the beamforming algorithms. The causes of pattern synthesis degradation, such as
antenna coupling and element illumination errors, have also been introduced. These
descriptions have then been referred to in the specific cases of interest for both the
circular and the MIMO arrays.
7.1
Conclusions
The research activity concerning the design and the application of array processing
and digital beamforming schemes to passive and coherent MIMO radar systems led
to the following results and conclusive remarks:
Internal and external measurements can effectively be used for PCL array digital
calibration. A technique for the compensation of the mutual coupling effect in
the digital domain has been presented. The proposed method accounts for
both the RF and the digital sections of the radar receiver. By combining a
set of internal and external measurements with an optimisation approach, the
expected shape for the PCL array pattern can be retrieved and made available
to the digital beamformer.
A novel digital beamforming technique for circular arrays has been invented and
better performance, with respect to classical UCA beamformer, has been confirmed. The analytical expression of the weights for circular array pattern shaping has been retrieved in a closed form. The comparison with the phase modes
theory approach has shown the better performance of the proposed method in
terms of achieved angular resolution of the synthesised array pattern.
The potentiality of the UCA PCL system has been confirmed. The entire digital
signal processing chain of a PCL radar has been developed. The system validation has been carried out thanks to the comparison of the radar detections
with the ground truth data provided by an ADSB receiver. The quality of the
i
i
i
7.1 Conclusions
105
resulting overlay between the two data streams has confirmed the capabilities
of the PCL radar.
The sidelobes performance of a conventional array is always better than the one
achieved by the equivalent MIMO array structure. The impact of the illumination errors on the synthesis of coherent MIMO array patterns, for the linear
array case, has been characterised in an analytical and closed form expression.
According to the related study, it has been possible to state that: although
the angular resolution of a MIMO array depends on the product between the
transmitter and receiver number of elements, the sidelobes behavior of the same
array, when illumination errors are considered, is inversely proportional to the
number of elements of the array with a lower spacing.
A TDM MIMO array has been designed and the related calibration procedure has
been implemented. A coherent TDM MIMO radar has been realised on a PCB.
Due to the high level of integration, an extensive characterisation of the antenna
section could not be obtained. A technique for the estimation of the calibration
coefficients of the virtual elements, which exploits the MIMO configuration and
can directly be applied on field measurements, has been proposed. The achieved
level of sidelobes reduction has validated the suggested solution.
The available Doppler interval in TDM MIMO systems can be re-established.
The use of a TDM transmission scheme produces a reduction of the unambiguous Doppler interval that is proportional to the number of transmitters. This
issue has been solved by introducing a random selection procedure for the active radiator within the transmission sequence. Simulations have confirmed the
validity of the approach and the possibility to re-establish the original Doppler
interval extension in the case of TDM MIMO radars and for slow moving targets.
The MUSIC algorithm can be extended to a multidimensional domain in a fully
coherent way. An innovative 2D high resolution algorithm, based on the MUSIC
processing, has been developed. By combining at the same time the spatial and
the fast time sampling capabilities of the array, the angular and range information about the target are retrieved as a result of a fully coherent processing.
The validity and increased performance provided by the algorithm have been assessed by means of simulations and real data acquisitions. The experiments and
the comparison with conventional signal processing approaches have highlighted
the benefits of the proposed method.
i
i
i
106
7.2
Recommendations and future work
Although being a topic that has largely been analysed since several decades, the
subjects of array processing, digital beamforming and the issues related to their application to modern systems continue to be of high interest in the radar community.
Array based radar systems, which provide a digitisation capability at a channel level,
indeed offer extended signal processing flexibility. However, in order to fully exploit
these type of systems, proper design choices are required from both the hardware
point of view and the signal processing approach.
The data throughput which is produced by fully digital systems can be a driving
factor for the following processing steps. By referring to the PCL system presented
in Ch.4, it was indeed necessary to perform a digital decimation, and consequently a
single transmitter selection, in order to handle the channel data streams provided by
the ADC. The mentioned compromise represents a stringent limitation for the PCL
performance, since it prevents the possibility of realising a multilateral/multistatic
system by exploiting the different radio stations which are available. A considerable
gain in terms of coverage and both range and angular resolutions can be achieved by
upgrading the data links of the current system.
The frequency agility of the system does depend on the radiating characteristic of the
array elements. With the aim at preventing any variability of the PCL performance
related to the selection of transmitters on the all FM band, wideband radiators would
be recommended to improve the EM behavior of the system antenna section.
Concerning the digital processing, the array pattern shaping procedure in Ch.3 represents the first step for improving the angular resolution of the PCL. As a further
development, the application of high resolution techniques for improved angular and
range resolution can be considered. The results obtained by the two dimensional approach with the MIMO demonstrator obviously represent a reliable perspective.
The opportunities offered by the MIMO processing applied to the radar field are
continuously evolving. The additional degrees of freedom that are offered by the coherent MIMO systems are particularly suitable for the application of digital beamforming
schemes. In that sense, the RADOCA demonstrator has confirmed the validity of this
combination.
One of the main research interests in this field regards the behavior of the sidelobes
with respect to the selected orthogonality approach and target characteristics. The
analysis conducted in Ch.5 has indeed tackled this problem from an antenna section
point of view and a preliminary answer from the mentioned perspective has been
provided. By referring to the proposed high resolution technique based on the MUSIC method, research activities towards different directions can be foreseen. At first,
i
i
i
7.2 Recommendations and future work
107
it must once again be noticed that the output of the algorithm represents a pseudo
amplitude value. It would be of large interest to retrieve, for the detected targets,
an estimation of the true reflectivity signature. In that way, high resolution imaging
applications by means of a stationary system can be performed. It has then been
discussed the importance of applying a data smoothing technique, in presence of coherent targets, to the received slow and fast time data samples in order to avoid a
rank deficiency of the related covariance matrix. The optimal dimensioning of the
averaging window has to be investigated. On the one hand, it should prevent an excessive reduction of the available degrees of freedom in both domains (angle, range);
on the other hand, the target decorrelation should still be ensured.
Research activities have started to investigate the potentiality of the MIMO/DBF
processing in combination with SAR acquisition geometries. In that sense, the possibility of synthesising the virtual channels would allow reducing the data throughput
of the radar system, that is a desirable advantage in airborne as well as in space
based configurations. Equivalently, the multiple degrees of freedom offered by the
MIMO processing would represent an added value for interferometric, tomographic
and GMTI signal processing. Under this perspective, further studies must be conducted in order to ensure the coherency among the multiple acquisitions.
Enhanced performance can be achieved by combining the orthogonal transmissions
with the spatial selectivity, i.e. by transmitting different waveforms in different directions. This approach, usually referred to as space-time coding or colored transmission,
allows overcoming some limitations related to the wide beam illumination (spreading
of the clutter spectrum, uniqueness of the transmitted waveform in multi-task systems, degraded multipath rejection capability). At the same time, since a wide area
is scanned, there is no reduction of the system coverage at the receiver side. A significant effort is required, at both the system design and the signal processing levels, in
order to ensure the electronic steering capability at transmission and the agile synthesis/selection of the colored waveforms. The related research activity, which can be
seen as an extension of the MIMO work produced in this thesis, obviously represents
a challenging opportunity for future investigations.
i
i
i
108
i
i
i
Appendix A
PCL system measurements
The PCL system measurements have concerned with the characterisation of the analog receiver components and the antenna element patterns. By referring to the first
ones, that are presented in Sec.A.1 and Sec.A.2, two specific evaluations have been
performed: the RF gains and the noise figures for each of the channels of the passive
radar subsystem. Specifically, the variations of the channel behaviors, due to the introduction of the attenuator blocks, have been checked. In Sec.A.3, the measurement
procedure leading to the estimation of the element patterns in the array set-up are
presented.
A.1
Receiver channel gains
A reference signal has directly been injected into the passive radar receiver in order
to perform the measurements. To improve the isolation between the signal generator
and the receiver itself, an additional attenuator has been included in the chain as
Figure A.1: Set-up of the gain measurements.
i
i
i
110
A. PCL system measurements
Table A.1: Gain values of the receiver channels
Channel
number
Both attenuators
OFF
Pre-Attenuator
ON
Both attenuators
ON
1
2
3
4
5
6
7
8
26.8dB
26.8dB
27.4dB
27.3dB
26.6dB
27.0dB
26.8dB
26.9dB
16.6dB
16.6dB
17.3dB
17.1dB
16.4dB
16.8dB
16.6dB
16.7dB
6.7dB
6.7dB
7.3dB
7.1dB
6.5dB
6.9dB
6.7dB
6.7dB
depicted in Fig.A.1. The attenuation provided by the attenuator was equal to 10dB.
The behavior of the 8 channels has been checked by simultaneously switching on/off
the pre and post attenuators of the passive receiver. Table A.1 summarises the measurement results in terms of gain provided at the output of the receiver. The external
attenuation is not included in the results.
A.2
Channel noise figures
The noise figures have been measured in order to check the effect of the attenuator
series which are present in the receiver rack. The values have been estimated according
to the setup in Fig.A.2. An HP346B noise source [111] has been taken as reference
and the noise floor has been retrieved by means of an HP8970A noise meter [112],
that has been connected to the output of the receiver. The frequency of 100MHz
has been considered and, at that frequency, an average gain of 26.8dB has been
measured for the channels with the attenuators switched off. This value is in line
Figure A.2: Set-up of the noise figure measurements.
i
i
i
A.3 Element patterns
111
Table A.2: Noise figure values
Channel
number
Both attenuators
OFF
Post-Attenuator
ON
1
2
3
4
5
6
7
8
4.5dB
4.6dB
4.6dB
4.6dB
4.6dB
4.5dB
4.5dB
4.5dB
7.0dB
7.1dB
7.0dB
6.9dB
7.4dB
7.1dB
7.1dB
7.1dB
with the previous measurement set-up in Sec.A.1. Tab.A.2 shows the measured noise
figures for all the receiver channels. The pre-attenuator was switched off during these
measurements as it is the non-preferred one, due to its placement just before the
low noise amplifier; anyway, its effect has been evaluated on one of the channels
and taken as representative of all the others. The values are presented in Tab.A.3 As
expected, the degradation introduced by the pre-attenuator is equal to the attenuation
it provides (10dB) and it leads to a final value for the noise figure, when both the
attenuators are considered, of roughly 17dB.
A.3
Element patterns
Due to the size of the array structure, it was not possible to perform the measurements
in the anechoic chamber available at TNO. The element pattern of the array elements
have been characterised in an open air far field measurement setup, by placing the
array on a rotating platform and using a bow-tie antenna as reference transmitter.
The following procedure has been used:
A reference transmitter, consisting of a biconical dipole antenna Fig.A.3.(a),
was connected to the first port of a spectrum analyzer whereas a channel of the
PCL array Fig.A.3.(b) was connected to the second one. All the other channels
Table A.3: Noise figure dependency with the attenuators configuration
Channel
number
Both attenuators
OFF
Pre-Attenuator
ON
Both attenuators
ON
4.5dB
14.5dB
17.0dB
i
i
i
112

were characteristically loaded.
The transmitter and the array were located according to the geometry in Fig.A.3.(c),
at a distance of ca. 25m which ensured operation in the far field. The array
was placed on a rotating platform Fig.A.3.(d).
While keeping the same position of the transmitter, the array was turned to
cover the 360 degrees on the azimuthal plane and the S21 parameter was measured.
The procedure was repeated for the other channels
i
i
i
A.3 Element patterns
113
(b)
(a)
(d)
(c)
Figure A.3: Element pattern measurement setup: reference transmitter (a), PCL array ARx and
reference transmitter AT x (b), measurements geometry (c) and rotating platform with scaled plane
(d)
i
i
i
114
i
i
i
Appendix B
Illumination error effects on the
synthesis of MIMO array pattern
Proof that:
2
h|F ()| iN ORM
'
|F ()|2iN ORM
(B.1)
h
i
h
i
M
1
N
1
P
P
|FT ()|2i
|bm |2 2R + 2T + |FR ()|2i
|an |2 2T + 2R
m=0
n=0
N 1 M 1
P P
2
|an ||bm |
n=0 m=0
The four main contributions, related to the different combinations of the (m, n) indexes, have been retrieved in Ch.5. For a better understanding of the steps, they are
here recalled:
"
N
1 M
1

X
X
2
(2 +2 )
T
R
h|F ()| i (n6=n0 ) = e
|F ()|2i +
|an |2 |bm |2
(m6=m0 )
n=0 m=0
|FT ()|2i
M
1
X
N
1
X
#
2
|an |
(B.2)
N
1 M
1

X
X
2
|an |2 |bm |2
h|F ()| i (n=n0 ) = (1 + 2T )(1 + 2R )
(B.3)
|bm |
|FR ()|2i
m=0
(m=m0 )
n=0
n=0 m=0
"
#
M
1
N
1

X
X

2
2
2
2
2
h|F ()| i (n6=n0 ) = e T (1 + R )
|bm | |FT ()|i
|an |
2
(m=m0 )
m=0
(B.4)
n=0
i
i
i
116
B. Illumination error effects on the synthesis of MIMO array pattern
"
#
N
1
M
1

X
X

2
2
2
2
2
h|F ()| i (n=n0 ) = e R (1 + T )
|an | |FR ()|i
|bm |
2
(m6=m0 )
n=0
(B.5)
m=0
By combining (B.2) and (B.3):

2
h|F ()| i1
=
+
e(T +R ) |F ()|2i
N
1 M
1
X
X
i
h
2
2
|an |2 |bm |2 (1 + 2R )(1 + 2T ) + e(T +R )
n=0 m=0
(2 +2 )
T
"
|FT ()|2i
M
1
X
|FR ()|2i
|bm | +
N
1
X
m=0
#
2
|an |
(B.6)
n=0
and by combining (B.4) and (B.5):

2
h|F ()| i2
= e(T ) (1 + 2T )
M
1
X
|bm |2 |FT ()|2i
m=0
(2 )
T
(1 +
2T )
M
1
X
|bm |
m=0
2
N
1
X
M
1
X
+ e(R ) (1 + 2R )
N
1
X
|an |2
n=0
|an |2 |FR ()|2i
n=0
e(R ) (1 + 2R )
m=0
2
|bm |2
N
1
X
|an |2 .
(B.7)
n=0
The terms h|F ()| i1 and h|F ()| i2 can now be collected and normalised by the
quantity:
!2
N
1 M
1
X
X
(2 +2 )
T
R
=e
|an ||bm | .
(B.8)
n=0 m=0
This normalisation leads to:
i
i
i
=
|F ()|2iN ORM

M
1
NP
1
h
i
P
2
2
2
2
2
2
(
+
)
|FT ()|i
|bm | + |FR ()|i
|an |
1
1 M
(1 + 2R )(1 + 2T ) + e T R NX
X
m=0
n=0
2
2
|a
|
|b
|
n
m
N 1 M 1
2
N 1 M 1
2
P P
P P
( 2 + 2 )
n=0 m=0
|an ||bm |
e T R
|an ||bm |
n=0 m=0
n=0 m=0
2
e(T ) (1 + 2T )
+
e
(2 +2 )
T
M
1
P
m=0
|bm |2 |FT ()|2i
N 1 M 1
P P
e(T ) (1 + 2T )
(2 )
R
(1 + 2R )
+
e
(2 +2 )
T
NP
1
n=0
(2 )
R
(1 + 2R )
M
1
P
|bm |2
m=0
NP
1
|an |2
n=0
(B.9)
n=0 m=0
|F ()|2iN ORM

M
1
NP
1
i
h
P
2
2
2
2
2
2
)
(
+
|FT ()|i
|bm | + |FR ()|i
|an |
1 M
1
(1 + 2R )(1 + 2T ) + e T R NX
X
m=0
n=0
2
2
|a
|
|b
|
n
m
2
2
N 1 M 1
N 1 M 1
P P
P P
( 2 + 2 )
n=0 m=0
e T R
|an ||bm |
|an ||bm |
n=0 m=0
e
+
|an |2
n=0
N 1 M 1
2
2
P P
(2 +2 )
T
R
e
|an ||bm |
|an ||bm |
n=0 m=0
2
|bm |2
m=0
NP
1
n=0 m=0
|an |2 |FR ()|2i
N 1 M 1
P P
M
1
P
N 1 M 1
2
2
P P
(2 +2 )
T
R
e
|an ||bm |
|an ||bm |
n=0 m=0
h|F ()| iN ORM
h|F ()| iN ORM
(2
T
(1 + 2T )
M
1
P
m=0
|bm |2 |FT ()|2i e

(2
T
+2 )
R
N 1 M 1
P P
n=0 m=0
(1 + 2T )
2
|an ||bm |
M
1
P
m=0
|bm |2
NP
1
|an |2
n=0
117
n=0 m=0
(2
T
i
i
i
n=0
|an |2 |FR ()|2i e(R ) (1 + 2R )

(2 +2 )
T
N 1 M 1
P P
M
1
P
|bm |2
m=0
NP
1
|an |2
n=0
(B.10)
2
|an ||bm |
n=0 m=0
|F ()|2iN ORM
NP
1 M
1
P
|an | |bm |
i
h
2
2
(1 + 2R )(1 + 2T )e(T +R ) + 1 n=0 m=0
2
NP
1 M
1
P
|an ||bm |
|FT ()|2i
M
1
P
e(R ) (1 + 2T )
+
m=0
N 1 M 1
P P
|FR ()|2i
NP
1
|an |
n=0
2
|an ||bm |
n=0 m=0
|bm |2 |FT ()|2i e(R ) (1 + 2T )

N 1 M 1
P P
|bm | +
m=0
n=0 m=0
M
1
P
M
1
P
NP
1
|bm |2
m=0
|an |2
n=0
2
|an ||bm |
n=0 m=0
2
e(T ) (1 + 2R )
+
NP
1
n=0
|an |2 |FR ()|2i e(T ) (1 + 2R )

N 1 M 1
P P
M
1
P
m=0
|bm |2
NP
1
|an |2
n=0
2
|an ||bm |
n=0 m=0
h|F ()| iN ORM
|F ()|2iN ORM
NP
1 M
1
P
|an |2 |bm |2
i
h
n=0 m=0
(2 +2 )
2
2
R + 1
+ (1 + R )(1 + T )e T
2
N 1 M 1
P P
|an ||bm |
n=0 m=0
|FT ()|2i
+
M
1
P
m=0
|bm |2 e
(2 )
R
(1 + 2T ) 1 + |FR ()|2i
N 1 M 1
P P
n=0 m=0
NP
1
2
|an ||bm |
n=0
h 2
i
|an |2 e(T ) (1 + 2R ) 1
(B.11)
h|F ()| iN ORM
NP
1
118
e(R ) (1 + 2R )
i
i
i
(B.12)
NP
1 M
1
P
|an |2 |bm |2
i
h 2
n=0 m=0
(2 )
( )
2
2
e R (1 + T ) + e T (1 + R )
2
NP
1 M
1
P
|an ||bm |
n=0 m=0
h|F ()| iN ORM
= |F ()|2iN ORM
NP
1 M
1
P
|an |2 |bm |2 h
i
2
2
2
2
n=0 m=0
+
(1 + 2R )(1 + 2T )e(T +R ) + 1 eR (1 + 2T ) eT (1 + 2R )

2
NP
1 M
1
P
|an ||bm |
n=0 m=0
|FT ()|2i
+
M
1
P
m=0
i
i
h 2
h 2
NP
1
|an |2 e(T ) (1 + 2R ) 1
|bm |2 e(R ) (1 + 2T ) 1 + |FR ()|2i
n=0
N 1 M 1
P P
2
|an ||bm |
(B.13)
n=0 m=0
If we consider the presence of small errors, i.e. 2 ' 0, then the following approximations hold:
2
e(R ) ' 1 + 2R
e(T ) ' 1 + 2T
(B.14)
and by neglecting the second order terms which derive from (B.13), we obtain:
2
h|F ()| iN ORM
n=0 m=0
119
' |F ()|2iN ORM

1
NP
1 M
P
|an |2 |bm |2 h
i
2
2
2 + 2 + 1 (1 + 2 + 2 ) (1 + 2 + 2 )
1
+
+
+ n=0 m=0
R
T
T
R
2
T
R
R
T
NP
1 M
1
P
|an ||bm |
i
i
i
m=0
h
i
h
i
NP
1
2
2
|bm |2 2R + m + |FR ()|2i
|an |2 2T + n
n=0
N 1 M 1
P P
2
|an ||bm |
n=0 m=0
h|F ()| iN ORM ' |F ()|2iN ORM +
|FT ()|2i
M
1
P
m=0
h
i
i
h
NP
1
|bm |2 2R + 2T + |FR ()|2i
|an |2 2T + 2R
n=0
N 1 M 1
P P
n=0 m=0
Q.E.D.
2
|an ||bm |
(B.15)
That finally leads to:
M
1
P
120
|FT ()|2i
i
i
i
Appendix C
Basic theory of FMCW
A frequency modulated continuous wave (FMCW) signal like the one presented in
Fig.C.1, also referred to as sawtooth modulated signal, is characterised by the instantaneous transmitted frequency:
ft (t) = fc
B
B
+ t = f0 + t
2
T
(C.1)
being fc the carrier frequency, B the transmitted bandwidth, T the pulse duration
and with 0 < t < T . For simplicity in the description, the expression with the initial
frequency f0 and the sweep rate is considered. According to (C.1), the transmitted
signal is represented by:
2
st (t) = ej2[fc t+ 2 t ] .
(C.2)
Figure C.1: Transmitted and received sawtooth modulated signals
i
i
i
122
C. Basic theory of FMCW
Under narrow band assumption, the received signal is a time delayed version of the
transmitted one:
2
sr (t) = ej2[fc (t )+ 2 (t ) ]
(C.3)
where, in the case of a stationary target, = 2Rc 0 being c the speed of light and R0
the range location of the target.
In the case of homodyne FMCW radar, the received signal is then mixed with the
transmitted one producing the signal:
2
sb (t) = ej2[fc + t 2 ]
(C.4)
characterised by the so called beat frequency:

fb =
(C.5)
which is directly related to the range information. Therefore, unlikely pulse radars
where the maximum unambiguous range is dictated by the pulse repetition interval
(time), in FMCW systems it depends on the frequency fs which is used to sample
the received signal. According to the Nyquist criterion, the maximum unambiguous
range is:
c fs
Rmax =
(C.6)
2 2
The range resolution is directly related to the frequency resolution, i.e. to the observation time, and it results:
R =
c 1
c
c
f =
=
.
2
2 T
2B
(C.7)
If the case of a moving target characterised by constant velocity vD , the time delay
takes the form:
2
2vD
= (R0 + vD t) = 0 +
t
(C.8)
c
c
and by replacing (C.8) in (C.4), we obtain:
sr (t) = e

v2
2v
2v
vD cD t2
j2 fc 0 12 02 +(0 +fc cD 0 cD )t+ 2
c
Under the assumption that
vD
c
(C.9)
1, the resulting beat frequency component is:
2vD
= 0 + fD
(C.10)
c
being fD the Doppler frequency shift induced by the target movement. As a result,
an ambiguity is introduced in the range estimation of the target since the peak of the
signal response is shifted with respect to the true target location. Several modulation
techniques have been proposed in the literature to overcome this limitation of the
conventional FMCW system [95].
fb ' 0 + fc
i
i
i

Acronyms
ADC
Analog to Digital Converter
ADSB
Automatic Dependent Surveillance Broadcast
AIC
Akaike Information Criterion
CDM
Code Division Multiplexing
CFAR
Constant False Alarm Rate
DBF
Digital Beam Forming
DDC
Digital Down Conversion
DPI
Direct Path Interference
DR
Dynamic Range
ECCM
Electronic Counter Counter Measure
ECM
Electronic Counter Measure
EVD
Eigenvalue Decomposition
FDM
Frequency Division Multiplexing
FDTD
Finite Difference Time Domain
FMCW
Frequency Modulated Continuous Wave
FNB
First Null Beamwidth
GMTI
Ground Moving Target Indicator
HPB
Half Power Beamwidth
i
i
i
124
INR
Interference to Noise Ratio
LAN
Local Area Network
LMS
Least Mean Squared
MC
Mutual Coupling
MDL
Minimum Description Length
MF
Matched Filter
MTI
Moving Target Indicator
PCB
Printed Circuit Board
PCL
Passive Coherent Locator
RADOCA RAdar DOme CAmera

RCS
Radar Cross Section
RVP
Residual Video Phase
SAR
Synthetic Aperture Radar
SIMO
Single Input Multiple Output
SLL
Side Lobe Level
SLR
Side Lobe Reduction
SNR
Signal to Noise Ratio
TDM
Time Division Multiplexing
UCA
Uniform Circular Array
ULA
Uniform Linear Array
Symbols
(, fD )
Ambiguity function
Antenna efficiency
Conduction efficiency
Dielectric efficiency
i
i
i

i
Illumination efficiency
Reflection efficiency
Wavelength
Mutual coupling matrix
Digital section compensation matrix
Scattering parameters matrix
Transmission coefficients matrix
Interference source directions matrix
Azimuth angle
I2
Interference power
n2
Noise power
Elevation angle
Speed stretching factor
D(, )
Antenna directivity
125
E(, , r) Electric field

F (, )
Array factor
fc
Carrier frequency
fi (, )
Ideal element pattern
fN
Nyquist frequency
fn (, )
Element pattern
fs
Sampling frequency
G(, )
Antenna gain
Free space wave number
P (, )
Array pattern
U (, )
Radiation intensity
i
i
i
126
vD
Radial (Doppler) speed
xs (t)
Surveillance channel signal
xr (t)
Reference channel signal
i
i
i
Bibliography
[1] A. Rudge, K. Milne, A. Olver, and P. Knight, The Handbook of Antenna Design,
IEE Electromagnetic Waves Series 16, Ed., Exeter, 1983.
[2] A. Farina, Antenna-Based Signal Processing Techniques for Radar Systems,
Artech House, Ed., Norwood, MA, 1991.
[3] E. Giaccari and C. Penazzi, A family of radars for advanced systems, Alta
Frequenza, vol. 58, pp. 97114, Apr. 1989.
[4] R. Mailloux, Phased Array Antenna Handbook, 2nd ed., Artech-House, Ed.,
Boston, MA, 2005.
[5] A. Yarovoy, T. Savelyev, P. Aubry, P. Lys, and L. Ligthart, UWB arraybased sensor for near-field imaging, Microwave Theory and Techniques, IEEE
Transactions on, vol. 55, no. 6, pp. 12881295, Jun. 2007.
[6] L. Giubbolini, A microwave imaging radar in the near field for anti-collision
(MIRANDA), Microwave Theory and Techniques, IEEE Transactions on,
vol. 47, no. 9, pp. 18911900, Sept. 1999.
[7] F. Nijenbauer and M. P. G. Otten, High resolution processing with an active phased array SAR, in Geoscience and Remote Sensing Symposium, 1999.
IGARSS 99 Proceedings. IEEE 1999 International, vol. 1, Jun. 2009, pp. 5052.
[8] J. J. M. De Wit, W. Van Rossum, and F. M. A. Smits, Sapphire: A novel building mapping radar, in Microwave Conference, 2009. EuMC 2009. European,
Oct. 2009, pp. 18961899.
[9] A. Huizing, M. P. G. Otten, W. Van Rossum, R. Van Dijk, A. P. M. Maas, E. H.
Van der Houwen, and R. Bolt, Compact scalable multifunction RF payload
for UAVs with FMCW radar and ESM functionality, in Radar Conference Surveillance for a Safer World, 2009. RADAR. International, Oct. 2009, pp.
16.
i
i
i
128
Bibliography
[10] N. Willis, Bistatic Radar, SciTech Publishing Inc., Ed., Raleigh, NC, 2005.
[11] H. Griffiths and C. Baker, Passive coherent location radar systems. part 1:
performance prediction, Radar, Sonar and Navigation, IEE Proceedings -, vol.
152, no. 3, pp. 153159, Jun. 2005.
[12] C. Baker, H. Griffiths, and I. Papoutsis, Passive coherent location radar systems. part 2: waveform properties, Radar, Sonar and Navigation, IEE Proceedings -, vol. 152, no. 3, pp. 160168, Jun. 2005.
[13] H. Kuschel, J. Heckenbach, S. Muller, and R. Appel, Countering stealth with
passive, multi-static, low frequency radars, Aerospace and Electronic Systems
Magazine, IEEE, vol. 25, no. 9, pp. 1117, Sept. 2010.
[14] D. OHagan, H. Kuschel, M. Ummenhofer, J. Heckenbach, and J. Schell, A
multi-frequency hybrid passive radar concept for medium range air surveillance, Aerospace and Electronic Systems Magazine, IEEE, vol. 27, no. 10, pp.
615, Oct. 2012.
[15] G. Fabrizio, F. Colone, P. Lombardo, and A. Farina, Adaptive beamforming
for high-frequency over-the-horizon passive radar, Radar, Sonar Navigation,
IET, vol. 3, no. 4, pp. 384405, Aug. 2009.
[16] R. Zemmari, M. Daun, and U. Nickel, Maritime surveillance using GSM passive
radar, in Radar Symposium (IRS), 2012 13th International, May 2012, pp. 76
82.
[17] S. Gelsema, Development and application of passive radar systems at TNO,
in Passive Covert Radar, 5th Multi-National Conference on, 2007.
[18] D. Caratelli, I. Liberal, and A. Yarovoy, Design and full-wave analysis of conformal ultra-wideband radio direction finders, Microwaves, Antennas Propagation, IET, vol. 5, no. 10, pp. 11641174, Jan. 2011.
[19] E. Fishler, A. Haimovich, R. Blum, D. Chizhik, L. Cimini, and R. Valenzuela,
MIMO radar: an idea whose time has come, in Radar Conference, 2004.
Proceedings of the IEEE, Apr. 2004, pp. 7178.
[20] W. Melvin and J. Scheer, Principles of Modern Radar - Volume 2 Advanced
Techniques, SciTech Publishing Inc., Ed., Raleigh, NC, 2013.
[21] A. Haimovich, R. Blum, and L. Cimini, MIMO radar with widely separated
antennas, Signal Processing Magazine, IEEE, vol. 25, no. 1, pp. 116129, Dec.
2008.
i
i
i
Bibliography
129
[22] E. Fishler, A. Haimovich, R. Blum, R. Cimini, D. Chizhik, and R. Valenzuela,

Performance of MIMO radar systems: advantages of angular diversity, in
Signals, Systems and Computers, 2004. Conference Record of the Thirty-Eighth
Asilomar Conference on, vol. 1, Nov. 2004, pp. 305309.
[23] E. Fishler, A. Haimovich, R. Blum, J. Cimini, L.J., D. Chizhik, and R. Valenzuela, Spatial diversity in radars-models and detection performance, Signal
Processing, IEEE Transactions on, vol. 54, no. 3, pp. 823838, Mar. 2006.
[24] J. Li and P. Stoica, MIMO radar with colocated antennas, Signal Processing
Magazine, IEEE, vol. 24, no. 5, pp. 106114, Sept. 2007.
[25] X. Zhuge and A. Yarovoy, A sparse aperture MIMO-SAR-based UWB imaging
system for concealed weapon detection, Geoscience and Remote Sensing, IEEE
Transactions on, vol. 49, no. 1, pp. 509518, Jan. 2011.
[26] , Sparse multiple-input multiple-output arrays for high-resolution nearfield ultra-wideband imaging, Microwaves, Antennas Propagation, IET, vol. 5,
no. 13, pp. 15521562, Mar. 2011.
[27] A. Hassanien and S. Vorobyov, Transmit/receive beamforming for MIMO radar
with colocated antennas, in Acoustics, Speech and Signal Processing, 2009.
ICASSP 2009. IEEE International Conference on, Apr. 2009, pp. 20892092.
[28] IEEE standard definitions of terms for antennas, IEEE Std 145-1983, 1983.
[29] C. Balanis, Antenna Theory: Analysis and Design, 3rd ed., Wiley-Interscience,
Ed., Hoboken, NJ, 2005.
[30] , Modern Antenna Handbook, Wiley-Interscience, Ed., Hoboken, NJ, 2008.
[31] W. Stutzman and G. A. Thiele, Antenna Theory and Design, 2nd ed., WileyInterscience, Ed., Hoboken, NJ, 1998.
[32] R. Elliot, Antenna Theory and Design, Wiley-Interscience, Ed., Hoboken, NJ,
2003.
[33] P. Hacker and H. Schrank, Range distance requirements for measuring low and
ultralow sidelobe antenna patterns, Antennas and Propagation, IEEE Transactions on, vol. 30, no. 5, pp. 956966, Sept. 1982.
[34] R. Hansen, Measurement distance effects on low sidelobe patterns, Antennas
and Propagation, IEEE Transactions on, vol. 32, no. 6, pp. 591594, Jun. 1984.
[35] , Phased Array Antennas, Wiley-Interscience, Ed., Hoboken, NJ, 2001.
i
i
i
130
Bibliography
[36] P. Lynch, The Dolph Chebyshev Window: A Simple Optimal Filter, Monthly
Weather Review, vol. 125, pp. 655660, 1997.
[37] P. James, Polar patterns of phase-corrected circular arrays, Electrical Engineers, Proceedings of the Institution of, vol. 112, no. 10, pp. 18391848, Oct.
1965.
[38] R. Fenby, Limitations on directional patterns of phase-compensated circular
arrays, Radio and Electronic Engineer, vol. 30, no. 4, pp. 206222, Oct. 1965.
[39] J. Allen and B. Diamond, Mutual coupling in array antennas, Lincoln
Lab., Massachusetts Inst. Technol., Tech. Rep., Mar. 1966. [Online]. Available:
http://www.dtic.mil/dtic/tr/fulltext/u2/648153.pdf
[40] H. Aumann, A. Fenn, and F. Willwerth, Phased array antenna calibration
and pattern prediction using mutual coupling measurements, Antennas and
Propagation, IEEE Transactions on, vol. 37, no. 7, pp. 844850, Jul. 1989.
[41] R. Collin, Array Theory for Waveguiding Systems, McGraw-Hill, Ed., New York,
1966.
[42] H. Steyskal and J. Herd, Mutual coupling compensation in small array antennas, Antennas and Propagation, IEEE Transactions on, vol. 38, no. 12, pp.
19711975, Dec. 1990.
[43] H. Aumann and F. Willwerth, Phased array calibrations using measured element patterns, in Antennas and Propagation Society International Symposium,
1995. AP-S. Digest, vol. 2, Jun. 1995, pp. 918921.
[44] H. Zhiyong, C. Balanis, and C. Birtcher, Mutual coupling compensation in
UCAs: Simulations and experiment, Antennas and Propagation, IEEE Transactions on, vol. 54, no. 11, pp. 30823086, Nov. 2006.
[45] T. Svantesson, Modeling and estimation of mutual coupling in a uniform linear
array of dipoles, in Acoustics, Speech, and Signal Processing, 1999. Proceedings., 1999 IEEE International Conference on, vol. 5, Mar. 1999, pp. 29612964.
[46] K. Dandekar, H. Ling, and G. Xu, Experimental study of mutual coupling
compensation in smart antenna applications, Wireless Communications, IEEE
Transactions on, vol. 1, no. 3, pp. 480487, Jul. 2002.
[47] G. Borgiotti and Q. Balzano, Mutual coupling analysis of a conformal array
of elements on a cylindrical surface, Antennas and Propagation, IEEE Transactions on, vol. 18, no. 1, pp. 5563, Jan. 1970.
i
i
i
Bibliography
131
[48] J. Ruze, The effect of aperture errors on the antenna radiation pattern, Il
Nuovo Cimento, vol. 9, no. 3, pp. 364380, Mar. 1952.
[49] R. Collin and F. Zucker, Antenna Theory, McGraw-Hill, Ed. Inter University
Electronics Series.
[50] J. Allen, L. Cartledge, W. Delaney, J. Di Bartolo, A. Fallo, W. Ince, M. Siegel,
J. Sklar, S. Spoerri, J. Teele, and D. Temme, Phased array radar studies,
Lincoln Lab., Massachusetts Inst. Technol., Tech. Rep., Nov. 1961.
[51] H. T. Hui, An effective compensation method for the mutual coupling effect
in phased arrays for magnetic resonance imaging, Antennas and Propagation,
IEEE Transactions on, vol. 53, no. 11, pp. 35763583, Nov. 2005.
[52] L. Sodin, Frequency-independent approximate compensation of mutual coupling in a linear array antenna, Antennas and Propagation, IEEE Transactions
on, vol. 57, no. 8, pp. 22932296, Aug. 2009.
[53] S. Yang and Z. Nie, Mutual coupling compensation in time modulated linear
antenna arrays, Antennas and Propagation, IEEE Transactions on, vol. 53,
no. 12, pp. 41824185, Dec. 2005.
[54] M. Malanowski and K. Kulpa, Digital beamforming for passive coherent location radar, in Radar Conference, 2008. RADAR 08. IEEE, May 2008, pp.
16.
[55] D. Shanno, Conditioning of quasi-Newton methods for function minimization,
Mathematics of Computation, vol. 24, no. 111, pp. 647656, Jul. 1970.
[56] C. Broyden, The convergence of a class of double-rank minimization algorithms, IMA Journal of Applied Mathematics, vol. 6, no. 1, pp. 7690, May
1970.
[57] T. Rahim and D. Davies, Effect of directional elements on the directional
response of circular antenna arrays, Microwaves, Optics and Antennas, IEE
Proceedings H, vol. 129, no. 1, pp. 1822, Feb. 1982.
[58] H. Aumann, Phase mode analysis of directional elements in a circular array, in
Antennas and Propagation Society International Symposium, 2008. AP-S 2008.
IEEE, Jul. 2008, pp. 14.
[59] J. Guy and D. Davies, Studies of the Adcock direction finder in terms of
phase-mode excitations around circular arrays, Radio and Electronic Engineer,
vol. 53, no. 1, pp. 3338, Jan. 1983.
i
i
i
132
Bibliography
[60] T. Rahim, Analysis of the element pattern shape for circular arrays, Electronics Letters, vol. 19, no. 20, pp. 838840, Aug. 1983.
[61] D. H. Brandwood, A complex gradient operator and its application in adaptive
array theory, Microwaves, Optics and Antennas, IEE Proceedings H, vol. 130,
no. 1, pp. 1116, Feb. 1983.
[62] J. Hudson, Adaptive Array Principles, IEE Electromagnetic Wave Series 11,
Ed., London, UK, 1989.
[63] ECRIN, GE fanuc embedded systems ics-554, http://www.ecrin.com/
datasheets/GEFIP/ics-554.pdf, 2009, [Online; accessed October-2012].
[64] ALDENA, AST.01.02.235 VHF band FM radio broadcasting, http://www.
aldena.it/images/catalogo/pdf/29 ast010223x aldena v2 2011.pdf, 2009, [Online; accessed September-2012].
[65] CST Microwave Studio, User manual version 5.0, CST gmbh, 2005, Darmstadt, Germany.
[66] S. Haykin, Adaptive Filter Theory, 4th ed., Prentice Hall - Information and
Systems Sciences Series, Ed., New Jersey, 2002.
[67] J. G. Proakis and D. K. Manolakis, Digital Signal Processing, 4th ed., Prentice
Hall, Ed., New Jersey, 2007.
[68] S. Gelsema, TNO-DV A187 - Modelling and analysis of passive radar systems:
A threat to modern air operations? TNO - Innovation for Life, Tech. Rep.,
Jun. 2007.
[69] M. I. Skolnik, Radar Handbook, 3rd ed., McGraw-Hill, Ed., New York, 2008.
[70] N. Levanon and E. Mozeson, Radar Signals, 4th ed., Wiley-Interscience, Ed.,
Hoboken, NJ, 2004.
[71] H. Rohling, Radar CFAR thresholding in clutter and multiple target situations, Aerospace and Electronic Systems, IEEE Transactions on, vol. AES-19,
no. 4, pp. 608621, Jul. 1983.
[72] A. Hassanien and S. Vorobyov, Phased-MIMO radar: A tradeoff between
phased-array and MIMO radars, Signal Processing, IEEE Transactions on,
vol. 58, no. 6, pp. 31373151, Jun. 2010.
[73] M. Wicks, E. Mokole, S. Blunt, R. Schneible, and V. Amuso, Principles of
Waveform Diversity and Design, SciTech Publishing Inc., Ed., Raleigh, NC,
2010.
i
i
i
Bibliography
133
[74] B. Friedlander, On the role of waveform diversity in MIMO radar, in Signals,

Systems and Computers (ASILOMAR), 2011 Conference Record of the Forty
Fifth Asilomar Conference on, Nov. 2011, pp. 15011505.
[75] , Waveform design for MIMO radars, Aerospace and Electronic Systems,
IEEE Transactions on, vol. 43, no. 3, pp. 12271238, Jul. 2007.
[76] K. Forsythe and D. Bliss, Waveform correlation and optimization issues for
MIMO radar, in Signals, Systems and Computers, 2005. Conference Record of
the Thirty-Ninth Asilomar Conference on, Nov. 2005, pp. 13061310.
[77] M. Richards, J. Scheer, and W. Holm, Principles of Modern Radar - Volume I
Basic Principles, SciTech Publishing Inc., Ed., Raleigh, NC, 2010.
[78] L. Weiss, Wavelets and wideband correlation processing, Signal Processing
Magazine, IEEE, vol. 11, no. 1, pp. 1332, Jan. 1994.
[79] T. Derham, S. Doughty, C. Baker, and K. Woodbridge, Ambiguity functions for
spatially coherent and incoherent multistatic radar, Aerospace and Electronic
Systems, IEEE Transactions on, vol. 46, no. 1, pp. 230245, Jan. 2010.
[80] G. Krieger, N. Gebert, and A. Moreira, Multidimensional waveform encoding:
A new digital beamforming technique for synthetic aperture radar remote sensing, Geoscience and Remote Sensing, IEEE Transactions on, vol. 46, no. 1,
pp. 3146, Jan. 2008.
[81] G. Krieger, M. Younis, S. Huber, F. Bordoni, A. Patyuchenko, J. Kim,
P. Laskowski, M. Villano, T. Rommel, P. Lopez-Dekker, and A. Moreira, Digital beamforming and MIMO SAR: Review and new concepts, Synthetic Aperture Radar, 2012. EUSAR. 9th European Conference on, pp. 1114, Apr. 2012.
[82] W. Bryc, Normal Distribution Characterizations with Applications - Lecture
Notes in Statistics, Cincinnati, 2005.
[83] A. Prudnikov, Y. Brychkov, and O. Marichev, Integrals and Series, Gordon and
Breach Science Publisher, Ed., Amsterdam, 1986.
[84] G. Fishman, Monte Carlo - Concepts, Algorithms and Applications, Springer,
Ed. Stanford, CA: Springer Series on Operations Research, 2000.
[85] J. Rahola and J. Ollikainen, Analysis of isolation of two-port antenna systems
using simultaneous matching, in Antennas and Propagation, 2007. EuCAP
2007. The Second European Conference on, Nov. 2007, pp. 15.
i
i
i
134
Bibliography
[86] A. Mak, C. Rowell, and R. Murch, Isolation enhancement between two closely
packed antennas, Antennas and Propagation, IEEE Transactions on, vol. 56,
no. 11, pp. 34113419, Nov. 2008.
[87] J.-N. Hwang and S.-J. Chung, Isolation enhancement between two packed antennas with coupling element, Antennas and Wireless Propagation Letters,
IEEE, vol. 10, pp. 12631266, 2011.
[88] S.-C. Chen, Y.-S. Wang, and S.-J. Chung, A decoupling technique for increasing the port isolation between two strongly coupled antennas, Antennas and
Propagation, IEEE Transactions on, vol. 56, no. 12, pp. 36503658, Dec. 2008.
[89] ROGERS, RO4350 high frequency laminates, http://www.cadxservices.com/
guides/pdfs/roger4003.pdf, 1999, [Online; accessed January-2013].
[90] W. Deal, N. Kaneda, J. Sor, Y. Qian, and T. Itoh, A new quasi-Yagi antenna
for planar active antenna arrays, Microwave Theory and Techniques, IEEE
Transactions on, vol. 48, no. 6, pp. 910918, Jun. 2000.
[91] L. Kuehnke, Phased array calibration procedures based on measured element
patterns, in Antennas and Propagation, 2001. Eleventh International Conference on (IEE Conf. Publ. No. 480), vol. 2, Apr. 2001, pp. 660663.
[92] W. Keizer, Fast and accurate array calibration using a synthetic array approach, Antennas and Propagation, IEEE Transactions on, vol. 59, no. 11, pp.
41154122, Nov. 2011.
[93] A.
Meta,
Signal
processing
of
FMCW
synthetic
aperture
radar data,
Ph.D. dissertation,
Delft University of Technology, 2006. [Online]. Available: http://repository.tudelft.nl/view/ir/uuid%
3A24352ff9-c11a-46c9-87d4-4d9d8968ed81/
[94] J. Figueras i Ventura, Design of a high resolution X-band doppler
polarimetric radar, Ph.D. dissertation, Delft University of Technology, 2009. [Online]. Available: http://repository.tudelft.nl/view/ir/uuid%
3Ad90b9ad6-237b-435d-9dc5-5660d9e7fbdd/
[95] H. Rohling, Automotive radar systems for civilian applications, Radar Symposium (IRS), 2012 13th International, May 2012, dedicated workshop.
[96] A. Swindlehurst, P. Stoica, and M. Jansson, Exploiting arrays with multiple
invariances using MUSIC and MODE, Signal Processing, IEEE Transactions
on, vol. 49, no. 11, pp. 25112521, Nov. 2001.
i
i
i
Bibliography
135
[97] R. Roy, A. Paulraj, and T. Kailath, Estimation of signal parameters via rotational invariance techniques - ESPRIT, in Military Communications Conference - Communications-Computers: Teamed for the 90s, 1986. MILCOM
1986. IEEE, vol. 3, Oct. 1986, pp. 41.6.141.6.5.
[98] C. Mathews and M. Zoltowski, Eigenstructure techniques for 2-d angle estimation with uniform circular arrays, Signal Processing, IEEE Transactions on,
vol. 42, no. 9, pp. 23952407, Sept. 1994.
[99] M. Zoltowski and C. Mathews, Beamspace root-MUSIC for rectangular arrays,
circular arrays, and nonredundant linear arrays, in Signals, Systems and Computers, 1991. 1991 Conference Record of the Twenty-Fifth Asilomar Conference
on, Nov. 1991, pp. 556560 vol.1.
[100] M. Abou-Khousa, D. Simms, S. Kharkovsky, and R. Zoughi, High-resolution
short-range wideband FMCW radar measurements based on MUSIC algorithm, in Instrumentation and Measurement Technology Conference, 2009.
I2MTC 09. IEEE, May 2009, pp. 498501.
[101] R. Schmidt, Multiple emitter location and signal parameter estimation, Antennas and Propagation, IEEE Transactions on, vol. 34, no. 3, pp. 276280,
Mar. 1986.
[102] P. Stoica, P. Handel, and A. Nehoral, Improved sequential MUSIC, Aerospace
and Electronic Systems, IEEE Transactions on, vol. 31, no. 4, pp. 12301239,
Oct. 1995.
[103] Q. Cheng and Y. Hua, Performance analysis of the MUSIC and pencil-MUSIC
algorithms for diversely polarized array, Signal Processing, IEEE Transactions
on, vol. 42, no. 11, pp. 31503165, Nov. 1994.
[104] Y. Huang, Tomographic processing of polarimetric and interferometric SAR
data for urban and forestry remote sensing, Ph.D. dissertation, University of
Rennes 1, 2011.
[105] M. Wax and T. Kailath, Detection of signals by information theoretic criteria,
Acoustics, Speech and Signal Processing, IEEE Transactions on, vol. 33, no. 2,
pp. 387392, Apr. 1985.
[106] E. Fishler, M. Grosmann, and H. Messer, Detection of signals by information
theoretic criteria: general asymptotic performance analysis, Signal Processing,
IEEE Transactions on, vol. 50, no. 5, pp. 10271036, May 2002.
i
i
i
136
Bibliography
[107] B. Nadler, Nonparametric detection of signals by information theoretic criteria:

Performance analysis and an improved estimator, Signal Processing, IEEE
Transactions on, vol. 58, no. 5, pp. 2746 2756, May 2010.
[108] B. Rao and K. Hari, Effect of spatial smoothing on the performance of MUSIC
and the minimum-norm method, Radar and Signal Processing, IEE Proceedings F, vol. 137, no. 6, pp. 449458, Dec. 1990.
[109] S. Pillai and B. Kwon, Forward/backward spatial smoothing techniques for
coherent signal identification, Acoustics, Speech and Signal Processing, IEEE
Transactions on, vol. 37, no. 1, pp. 815, Jan. 1989.
[110] W. Carrara, R. Goodman, and R. Majewski, Spotlight Synthetic Aperture
Radar: Signal Processing Algorithms, Artech-House, Ed., Norwood, MA, 1995.
[111] Hewlett-Packard,
HP346B
noise
source,
http://www.
home.agilent.com/en/pd-1000001299%3Aepsg%3Apro-pn-346B/
noise-source-10-mhz-to-18-ghz-nominal-enr-15-db?cc=NL&lc=dut,
2010,
[Online; accessed February-2013].
[112] , HP8970 noise meter, http://www.airlink.dk/Dokumenter/HP8970A.
pdf, 2010, [Online; accessed February-2013].
i
i
i
Summary
Antenna Array Signal Processing for Multistatic Radar
Systems
The introductions of Digital Beam Forming (DBF), original signal exploitation and
waveform multiplexing techniques have led to the design of novel radar concepts. Passive Coherent Locator (PCL) and Multiple-Input Multiple-Output (MIMO) sensors
are two examples of innovative approaches. Beside the inherent benefits of this class
of sensors, related to the use of emitters of opportunity and the waveform diversity
for the PCL and the MIMO cases respectively, an additional gain is achieved by using
ad-hoc array signal processing techniques.
The well-known advantages provided by the antenna arrays, or phased arrays, are,
among others, the possibility of synthesising arbitrarily steered beams, especially in
the case of DBF, the capability of shaping the array pattern sidelobes region and
the opportunity of placing nulls towards specific directions. The last feature allows
suppressing, or mitigating, the echoes of unwanted signals (clutter, jammer, interferences) that are present inside the radar scenario and that can degrade the detection
performance of the system. With DBF, the interferences suppression can adaptively
be done in real time.
This thesis has focused on the development of novel array processing algorithms,
with emphasis on antenna pattern optimisation. Also, attention was paid to resolution enhancement and clutter suppression techniques exploiting the available radar
architectures. A PCL and a MIMO array system have been investigated in order to
validate the proposed techniques.
Specifically, the first part of the thesis deals with the development of the array signal
processing chain of a PCL system based on a circular array. The performance of digital beamforming algorithms highly depends on the behavior of the antenna section
of the array; indeed, an inaccurate characterisation of the element channels and their
mutual influence degrades the shape of the synthesised array pattern. A complete
analysis of the circular array is presented in the text and a technique for the mutual
i
i
i
138
Summary
coupling compensation, based on an optimisation approach, was retrieved. The effectiveness of the method was validated. Furthermore, a new pattern shaping technique
for the sidelobes reduction was developed. Due to the poor angular resolution of the
PCL array, the criteria for estimating the quality of the retrieved algorithm was represented by the achievable sidelobes suppression level for a given angular resolution.
A comparison with the phase modes approach has proven the better performance of
the proposed method. Concerning the validation of the PCL system, the entire signal
processing chain of the sensor was developed and the related behavior was successfully assessed by the comparison with the ground truth data provided by an ADSB
transponder.
The second part of the thesis refers to coherent MIMO radars and to the benefits
they provide with respect to conventional array-based radars. The linear array configuration was taken as reference and a MIMO test board, the RADOCA system, has
been realised and studied. By first recalling the peculiarity of such systems, i.e. the
possibility of synthesising virtual array structures, the research activity has investigated the effect that illumination errors produce on the related pattern behavior. The
analytical expression of the errors affected pattern was retrieved and the comparison
with conventional linear array systems showed the higher degradation they produce
on MIMO based arrays. The focus was then moved to the identification of calibration
and array processing techniques, capable of exploiting both the MIMO radar principle
and the additional degrees of freedom provided by the virtual channels. By referring
to the RADOCA demonstrator, a calibration procedure was retrieved and validated
thanks to real data acquisitions. Then, with the aim of enhancing the system performance, two objectives were reached: the extension of the maximum unambiguous
Doppler interval, by means of a random selection of the radar transmitter sequence,
and the identification of a novel 2D high resolution technique based on the MUSIC
algorithm. The comparisons with traditional approaches highlighted the benefits of
the proposed techniques.
i
i
i
Samenvatting
Antenna Array Signal Processing for Multistatic Radar
Systems
De introductie van Digitale Bundelvorming (Digital Beamforming, DBF),
signaalexploitatie en het gebruik van meerdere golfvormen hebben geleid tot nieuwe
radar concepten. Passive Coherent Locator (PCL) en Multiple-Input MultipleOutput (MIMO) sensoren zijn twee voorbeelden van zulke innovatieve methoden.
Inherente voordelen van deze sensoren zijn gerelateerd aan toevallig aanwezige zenders
voor PCL en golfvormdiversiteit in het geval van MIMO. Daarnaast is met behulp
van ad-hoc signaalverwerkingstechnieken nog extra voordeel te behalen.
Bekende voordelen van antenne arrays zijn onder andere de mogelijkheid tot
bundelsturing in elke richting (DBF), controle over het zijluspatroon en de
mogelijkheid tot onderdrukking van signalen in specifieke richtingen. Dit laatste punt
maakt het mogelijk om ongewenste signalen die de prestaties van de radar kunnen
verlagen, zoals clutter, jamming, en interferentie, te onderdrukken. DBF maakt het
mogelijk om interferentie adaptief te onderdrukken in real-time.
De focus van deze scriptie is op de ontwikkeling van nieuwe algoritmes voor arrays
met nadruk op optimalisering van het antennepatroon. Daarnaast komen technieken
voor het verbeteren van de resolutie en het onderdrukken van clutter aan bod, die
gebruik maken van de beschikbare, innovatieve radararchitectuur. Om de voorgestelde
technieken te valideren zijn een PCL en een MIMO arraysysteem onderzocht.
In het eerste deel van dit proefschrift wordt de ontwikkeling van de
signaalverwerkingsketen van een PCL systeem behandeld dat gebaseerd is op een
circulair array. De prestaties van DBF algoritmes zijn sterk afhankelijk van het
gedrag van de antennes in het array. Een onnauwkeurige karakterisering van de
signaalkanalen en hun wederzijdse benvloeding degraderen de vorm van het
gecombineerde antennepatroon. De tekst bevat een complete analyse van het
circulaire array. Met behulp van optimalisatietechnieken was het mogelijk een nieuwe
methode voor de compensatie van de wederzijdse benvloeding te ontwikkelen.
i
i
i
140
Samenvatting
Vervolgens is de effectiviteit van deze methode gevalideerd. Daarnaast is een nieuwe

techniek ontwikkeld om zijlussen te onderdrukken met behulp van patroonvorming
bij een minimale verbreding van de hoofdlus. De reden hiervoor is de beperkte
hoekresolutie van het PCL array. In een vergelijking met de zogenaamde phase
modes approach is de betere prestatie van het voorgestelde algoritme aangetoond.
De volledige PCL signaalverwerkingsketen is ontwikkeld en vervolgens gevalideerd
met behulp van data van een ADSB transponder.
Het tweede deel van dit proefschrift behandelt coherente MIMO radars en de
voordelen hiervan ten opzichte van conventionele radars met een antenne-array.
Hiervoor is een lineair array als referentie genomen. Een MIMO testboard, het
zogenaamde RADOCA systeem, is ontwikkeld en bestudeerd. Vervolgens worden
de bijzonderheden van een dergelijk systeem beschreven, namelijk de mogelijkheid
om een virtueel array te maken. Het effect dat belichtingsfouten (bijvoorbeeld faseof positioneringsfouten) hebben op het bundelpatroon is onderzocht en hiervoor is een
analytische uitdrukking afgeleid. Het blijkt dat een MIMO array gevoeliger is voor
belichtingsfouten dan een conventioneel lineair array. Vervolgens zijn kalibratie- en
array-processingtechnieken gedentificeerd, die gebruik maken van de extra
vrijheden die het MIMO principe biedt. Dankzij het RADOCA demonstratiemodel
was het mogelijk om een kalibratieprocedure te ontwikkelen en te valideren gebaseerd
op opgenomen data. Met het oog op het vergroten van de prestaties van het
systeem zijn de volgende twee verbeteringen gerealiseerd: het verlengen van het
maximale niet-ambigue Dopplerinterval, door middel van random selectie van de
volgorde van radarzenders; het verhogen van de resolutie door een nieuw 2D-MUSIC
algoritme. In een directe vergelijking met de traditionele technieken is de meerwaarde
van bovenstaande algoritmes aangetoond.
i
i
i
Journal Papers
J1 F. Belfiori, S. Monni, W. van Rossum and P. Hoogeboom, Antenna Array Characterisation and Signal Processing for an FM Radio-Based Passive Coherent
Location Radar System, Radar, Sonar and Navigation, IET, Vol. 6, Is. 8, pp.
687-696, Oct. 2012.
J2 F. Belfiori, W. van Rossum and P. Hoogeboom, Coherent MIMO Array Design
with Periodical Physical Element Structures, Antennas and Wireless Propagation Letters, IEEE, Vol. 10, pp. 1341-1344, 2011.
J3 F. Belfiori, W. van Rossum and P. Hoogeboom, A Coherent MUSIC Technique for Range/Angle Information Retrieval: Application to an FMCW MIMO
Radar, Invited paper - to appear in Radar, Sonar and Navigation, IET, Special
Issue on Bistatic and MIMO Radars and their Applications in Surveillance and
Remote Sensing.
Conference Papers
C1 F. Belfiori, W. van Rossum and P. Hoogeboom, Application of 2D MUSIC Algorithm to Range-Azimuth FMCW Radar Data, in Radar Conference (EuRAD),
2012 European, Oct. 2012, pp. 242-245, Amsterdam, Netherlands.
C2 F. Belfiori, W. van Rossum and P. Hoogeboom, 2D-MUSIC Technique Applied
to a Coherent FMCW MIMO Radar, in Radar Systems (Radar 2012), IET International Conference on, Oct. 2012, pp. 1-6, Glasgow, Scotland. (Awarded
with the best student paper prize).
C3 F. Belfiori, W. van Rossum and P. Hoogeboom, Array Calibration Technique for
a Coherent MIMO Radar, in Radar Symposium (IRS), 2012 13th International,
May 2012, pp. 122-125, Warsaw, Poland.
i
i
i
142
C4 F. Belfiori, W. van Rossum and P. Hoogeboom, Random Transmission Scheme

Approach for a FMCW TDMA Coherent MIMO Radar, in Radar Conference
(RADAR), 2012 IEEE, May 2012, pp. 178-183, Atlanta, USA.
C5 F. Belfiori, L. Anitori, W. van Rossum, M. Otten and P. Hoogeboom, Digital
Beam Forming and Compressive Sensing Based DOA Estimation in MIMO Arrays, in Radar Conference (EuRAD), 2010 European, Oct. 2011, pp.285-288,
Manchester, UK.
C6 F. Belfiori, N. Maas, P. Hoogeboom and W. van Rossum, DMA X-Band FMCW
Radar for Short Range Surveillance Applications, in Antennas and Propagation
(EUCAP), Proceedings of the 5th European Conference on, Apr. 2011, pp. 483487, Rome, Italy.
C7 F. Belfiori, P. Hoogeboom, Analysis of a Novel MIMO System for Security Applications, in Phased Array Systems and Technology (ARRAY), 2010 IEEE
International Symposium on, Oct. 2010, pp. 338-343, Waltham, USA.
C8 F. Belfiori, S. Monni, W. van Rossum and P. Hoogeboom, Side-Lobe Suppression
Techniques for a Uniform Circular Array, in Radar Conference (EuRAD), 2010
European, Sep. 2010, pp. 113-116, Paris, France.
C9 F. Belfiori, S. Monni, W. van Rossum and P. Hoogeboom, Mutual Coupling
Compensation Applied to a Uniform Circular Array, in Radar Symposium
(IRS), 2010 11th International, Jun. 2010, pp. 1-4, Vilnius, Lithuania.
Patent
P1 P. Hoogeboom and F. Belfiori, Antenna System, Radar Device and Radar Method
with 360 Degree Coverage, European Patent Application No. 10155992.0, Sep.
2011.
Technical Report
R1 F. Belfiori, S. Monni, W. van Rossum and D. Deurloo, Passive Radar Advantages
and Threats, Technical Report, TNO, The Hague, Netherlands, Dec. 2010.
i
i
i
About the author

Francesco Belfiori was born in Roma, Italy, on September 6, 1982. In September
2001 he started his Telecommunication Engineering studies at the Università degli
studi di Roma - La Sapienza. He received the Bachelor degree and the Master degree
from the same university, both summa cum laude, in 2004 and 2007 respectively. He
graduated with a thesis entitled SHARAD radar sounder simulation: efficient scattering approximation from a coherent surface. The activity was carried on at the
SHARAD Operation Centre (SHOC) located at the Thales Alenia Space facilities in
Roma.
From June 2007 to January 2008 he worked as a software engineer at ElsagDatamat
S.p.a.; he cooperated at the development of the mission planning tool for the NH-90
helicopter in force of the Italian Army. In 2008 he joined the Electronics Division of
DAppolonia S.p.A. as an R&D engineer. During this period he conducted applied
research activities in the fields of electronics, radar and remote sensing, signal and
image processing.
In February 2009 he started working on his PhD degree in a joint cooperation between the Delft University of Technology and TNO Defence, Security and Safety, in
the Netherlands. The subject of the research was about the application of digital
beamforming and signal processing techniques to passive and MIMO radar systems.
During his PhD work, in 2012, he spent three months at the Institut d Electronique
et de Telecommunications de Rennes (IETR) at the Universite de Rennes 1, France,
where he worked in cooperation with Prof. L. Ferro-Famil.
He is a student member of IEEE and serves as a reviewer for IEEE, IET, Elsevier
and Springer journals.
Award
Best student paper prize at the 2012 IET - International Conference on Radar Systems (RADAR2012), Glasgow, UK. Awarded for the article entitled: 2D-MUSIC
Technique Applied to a Coherent FMCW MIMO Radar.
i
i
i
144
About the author
i
i
i
Acknowledgements
Although a very high weight is on your own shoulders, on the PhD way you meet
many people that help you in carrying it on. The achievements you reach at the end
are then always the result of combined efforts and, even if they bring your name, you
know you would not have done much without that support.
First of all, I would like to thank my supervisor Prof. Peter Hoogeboom, who gave me
the possibility to challenge myself with the research activity on array signal processing. He secured founding for this research and he introduced me to the radar groups
of TNO and of the Delft University of Technology.
I am especially thankful to Dr. Wim van Rossum, my daily supervisor at TNO. He
has been a mentor in these four years and he has always been capable of providing
critical feedbacks and keen remarks during our discussions. It has been a pleasure to
work with you and I hope to keep the same passion and dedication for research you
have shown to me.
At the Delft University of Technology, many thanks go to Prof. Francois Le Chevalier
for the several advices and suggestions about my research. One of them also brought
me to the lovely city of Rennes, where I was glad to work at the IETR laboratory of
the University of Rennes 1 in cooperation with Prof. Laurent Ferro-Famil. He has
been truly inspiring to me and I really look forward to finalising the activity we have
started together.
Felt gratitude goes to the TNO radar group and especially to its Italian creek: Daniela,
Stefania, Bernadetta, Lorenzo, Giampiero and Daniele. You have been as a new family for me and, after knowing you, I finally realised what the brain drain means.
Same feeling I share with all the other great scientists and very good friends I had the
fortune to meet at TU Delft. It is a long list and definitely worth of being written.
Dana, Tobias, Diego, Alexey, Dima, Teun, Karolina, Ricardo, Guendalina and Igor:
the time spent together simply had no price.
Among others, I want to express my deep heart gratitude to my paranymphs: my
sorellina italiana Laura and frère francais Yann. Thanks for being so close to me in
these years, for sharing the many doubts, concerns and pains about our PhD tracks.
i
i
i
146
Acknowledgements
It has been a long way for the all of us but at the end we all managed.
The city of Delft is truly an agglomerate of multiple cultures and the time I spent
here, learning about the different habits and traditions, was just great. Though, it
would have not been so without the presence of the many dutch and international
friends I met. Special thanks then go to Jitske, Leen, Renske, Illy, the thai couple
Anne and Martijn and the bundes one Christian and Kati. A list like that cannot
obviously be complete without an italian component: Elena, Riccardo and the metacrew: Alex, Simone, Adriano and Christian; thanks for the enjoyable time we have
spent together in front of a well arranged table, drinking beer and all complaining
about our politically devastated country.
Living abroad could result in breaking down the connections with the people you
left behind. I feel really lucky by not having faced anything of that, thanks to the
presence of all my friends in Rome. Every time I come back, you just behave as I had
never left, regardless the last occasion we have met or talked to each other. I could
never thank you enough for that.
Finally, I would like to thank those who share the most important part of my heart:
my father Pietro, my brother Stefano and my sister in love Cristiana with their little
Matteo. The choice to leave you was one of the most though I had ever taken and it
often led to rethinking. If I am here today is only because of your constant support
and encouragement: I will forever be in debt with you.
The very last thought is for my mother, Anna, to whom this thesis is dedicated: I
know you have followed me along the way and I hope that today you are proud of
me.
Infine vorrei ringraziare coloro che occupano il posto pi`
u importante nel mio cuore:
mio padre Pietro, mio fratello Stefano e mia cognata Cristiana con il loro piccolo
Matteo. La scelta di lasciarvi è stata una delle pi`
u difficili che io abbia mai preso ed
è spesso stata motivo di ripensamento. Se sono qui oggi è solamente grazie al vostro
supporto e incoraggiamento costanti: sar`
o sempre in debito con voi.
L ultimo pensiero è per mia madre, Anna, a cui questa tesi è dedicata: so che mi hai
seguito lungo la strada e spero che oggi tu sia orgogliosa di me.
i
i

(D) Antenna Array Signal Processing For Multistatic Radar Systems - Belfiori - 2013

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

(D) Antenna Array Signal Processing For Multistatic Radar Systems - Belfiori - 2013

Uploaded by

Copyright:

Available Formats

i

Antenna Array Signal Processing for

Antenna Array Signal Processing for

ter verkrijging van de graad van doctor

Dit proefschrift is goedgekeurd door de promotor:

This research was supported by TNO under contract DenV-017.

3 Digital beamforming for PCL

5 Coherent MIMO array theory

6 MIMO signal processing: RADOCA test board and experimental

7 Conclusion and outlook

List of Acronyms and Symbols

About the author

Sketch of an analog beamformer based on phase shifter components. .

Basic digital beamforming scheme. . . . . . . . . . . . . . . . . . . . .

Generic antenna array geometry. . . . . . . . . . . . . . . . . . . . . .

Linear antenna array geometry scanning on the xOz plane. . . . . . .

Normalised array pattern behaviors along the azimuthal plane ( = 90 ). 17

Circular antenna array geometry. . . . . . . . . . . . . . . . . . . . . .

3D view of the array factor for an 8-elements UCA (linear scale). . . .

Array factor for an 8-elements UCA along the azimuthal ( = 90 , left)

Digital signal processing scheme of the PCL radar. . . . . . . . . . . .

First kind Bessel functions of different orders. . . . . . . . . . . . . . .

Block diagram of the PCL system. . . . . . . . . . . . . . . . . . . . .

TNO circular array for passive radar applications. . . . . . . . . . . .

Simulated element pattern gain of a stand alone dipole considered as a

Simulated element pattern gain of a single dipole in the circular array

Required spacings of the receiver elements for increasing number of the

A.1 Set-up of the gain measurements. . . . . . . . . . . . . . . . . . . . . . 109

Azimuthal resolution performance . . . . . . . . . . . . . . . . . . . .

ICS-554B digitiser board specifications . . . . . . . . . . . . . . . . . .

Main to Maximum Sidelobe Level . . . . . . . . . . . . . . . . . . . . .

A.1 Gain values of the receiver channels . . . . . . . . . . . . . . . . . . . 110

Figure 1.1: Sketch of an analog beamformer based on phase shifter components.

Figure 1.2: Basic digital beamforming scheme.

If we divide both sides of (1.2) by the carrier frequency fc , we obtain:

capability also represents an essential requirement in case of wideband and/or near

Passive Coherent Locator (PCL) systems

1.2 The Multiple-Input Multiple-Output radar concept

The Multiple-Input Multiple-Output radar concept

handling (the transmitted waveforms must separately be digitised). Another approach

1.3 Outline of the Thesis

The estimation of a calibration procedure aimed at compensating for the array

Outline of the Thesis

The remainder of the thesis is organised as follows:

Main antenna parameters

The classical definition of antenna is provided by the IEEE Standard Definitions of

2. Antenna theory and array pattern synthesis

A number of efficiencies can be related to an antenna; the overall efficiency is

far field region is assumed at a distance greater than 2D

2.2 Antenna array pattern synthesis

efficiencies respectively and they are related to the dissipation losses.

Antenna array pattern synthesis

Multiple independent radiating elements can be arranged into a certain geometrical

2. Antenna theory and array pattern synthesis

Figure 2.1: Generic antenna array geometry.

spacing between the multiple elements

being k the free space wave number and:

2.2 Antenna array pattern synthesis

which describes the geometrical characteristics of the array.

, being L the largest

2. Antenna theory and array pattern synthesis

Linear array pattern synthesis

ej N d sin() ej N d sin() ej N d sin()