Digital Beamforming Focal Plane Arrays For Radio Astronomy and Space-Borne Passive Remote Sensing

Thesis for the Degree of Doctor of Philosophy
Digital Beamforming Focal Plane Arrays

for Radio Astronomy and Space-Borne
Passive Remote Sensing
Oleg Iupikov
Department of Electrical Engineering

Chalmers University of Technology
Göteborg, Sweden 2017

Digital Beamforming Focal Plane Arrays for Radio Astronomy and Space-
Borne Passive Remote Sensing
Oleg Iupikov
ISBN 978-91-7597-579-5
© Oleg Iupikov, 2017.

Doktorsavhandlingar vid Chalmers tekniska högskola
Ny serie nr 4260
ISSN 0346-718X
Department of Electrical Engineering

Antenna group
Chalmers University of Technology
SE412 96 Göteborg
Sweden
Telephone: +46 (0)31 772 1000
Email: oleg.iupikov@chalmers.se
AT X.
Typeset by the author using L E
Chalmers Reproservice
Göteborg, Sweden 2017
To my family
Look deep into nature, and then you will understand everything better
-Albert Einstein
Abstract
Dense Phased Array Feeds (PAFs) for reector antennas have numerous advantages
over traditional cluster feeds of horns in a one-horn-per-beam conguration, espe-
cially in RF-imaging applications which require multiple simultaneously formed and
closely overlapping beams. However, the accurate analysis and design of such PAF
systems represents a challenging problem, both from an EM-modeling and beamform-
ing optimization point of view. The current work addresses some of these challenges
and consists of two main parts.
In the rst part the mutual interaction eects that exist between a PAF consisting
of many densely packed antenna elements and an electrically large reector antenna
are investigated. For that purpose the iterative CBFM-PO method has been devel-
oped. This method not only allows one to tackle this problem in a time-ecient and
accurate manner, but also provides physical insight into the feed-reector coupling
mechanism and allows to quantify its eect on the antenna impedance and radiation
characteristics. Numerous numerical examples of large reector antennas with var-
ious representative feeds (e.g. a single dipole feed and complex PAFs of hundreds
of elements) are also presented and some of them are validated experimentally. In
order to analyze electrically large feeds eciently, a domain-decomposition approach
to Krylov subspace iteration, where macro basis functions (or characteristic basis
functions) on each subdomain are naturally constructed from the dierent segments
of the generating vectors, is also proposed.
The second part of the thesis is devoted to the optimization of PAF beamformers
and covers two application examples: (i) microwave satellite radiometers for accurate
ocean surveillance; and (ii) radio telescopes for wide eld-of-view sky surveys. Based
on the initial requirements for future antenna systems, which are currently being
formulated for these applications, we propose various gures-of-merits and describe
the corresponding optimal beamforming algorithms that have been developed. Stud-
ies into these numerical examples demonstrate how optimal beamforming strategies
can help to greatly improve the antenna system characteristics (e.g. beam eciency,
side-lobe level and sensitivity in the presence of the noise) as well as to reduce the
complexity of the beam calibration models and overall phased array feed design.
Keywords: phased array feeds, reector antenna feeds, beamforming, feed-reector

interaction, radio telescopes, spaceborne radiometers.
i
ii
Preface
This thesis is in partial fulllment for the degree of PhD of Engineering at Chalmers
University of Technology.
The work that has resulted in this thesis was carried out between December 2011
and March 2017 and has been performed within the Antenna group at the Department
of Electrical Engineering, Chalmers. Professor Marianna Ivashina has been both the
examiner and main supervisor, and Associate Professor Rob Maaskant has been the
co-supervisor.
The work has been supported by a project grant System Modelering och Optimer-
ing av Gruppantenner för Digital Lobformning from the Swedish Research Council
(VR), Swedish National Space Board (SNSB) project grant 202-15 Antenna-Array
Digital-Beamforming and Calibration Methods for the Next Generation Multi-Beam
Space-borne Radiometers for Ocean Observation within the framework of the SNSB
2015-R open call for Space and Earth Observation Research, and a grant Study on
Advanced Multiple-Beam Radiometers (contract 4000107369-12-NL-MH) from the
European Space Agency (ESA).
iii
iv
Acknowledgments
First and foremost, I wish to thank my supervisor Prof. Marianna Ivashina for the
opportunity to work on challenging and relevant research topics, and for her contin-
uous guidance and encouragement during these years. I thank her for her patience
and signicant amount of time she spent in discussing the challenging phased-array
feed problems, reviewing my research papers and technical reports. I am very thank-
ful Prof. Ivashina for teaching me to be an independent researcher by giving me
a chance to be creative in my work. I would also like to thank my co-supervisor
Assoc. Prof. Rob Maaskant for numerous fruitful discussions related to my work,
and for the innite support in such complicated topics as numerical methods for
electromagnetic modeling, electromagnetic theory, and other technical topics I could
have a question on. All in all, it had been my honor to have worked with both of You.
You are as much as mentors to me as are friends. I would like to thank Prof. Per-
Simon Kildal for welcoming me to the Antenna group. Also, thanks to all of you I
have met my beloved Esperanza :)
Thanks to my colleagues at the Onsala Space Observatory, especially to Prof. John

Conway and Dr. Miroslav Panteleev, for providing me with interesting depart-
ment service tasks related to the Square Kilometer Array (SKA) project, which was
also benecial for my PhD project as I could improve large parts of my Matlab code.
I would also like to thank Drs. Kees van 't Klooster and Benedetta Fiorelli
from ESA, Drs. Knud Pontoppidan, Per Heighwood Nielsen and Cecilia Cap-
pellin from TICRA, Prof. Niels Skou from Technical University of Denmark for the
interesting and fruitful collaboration on new satellite radiometers, and Dr. Andre
Young from Stellenbosch University for common work on calibration techniques for
radio telescopes. I thank Prof. Christophe Craeye for inviting me to Université
Catholique de Louvain and for interesting collaboration which had led to our joint
journal publication on numerical methods.
I would like to acknowledge Dr. Wim van Cappellen from ASTRON, The
Netherlands, for providing us with measurements of the Vivaldi antenna PAF (APER-
TIF), that were made at the Westerbork Synthesis Radio Telescope. These measure-
ment results have been very benecial for validating my numerical models.
My special thanks go to all the former and current colleagues of the Electrical
Engineering Department for creating a nice and enjoyable working environment.We've
had a lot of fun and enjoyable moments both at work and afterwork time.
v
Acknowledgments
Bart Smolders, and the

Finally, I would like to thank the faculty opponent, Prof.
committee members, Prof. Joakim Johansson, Prof. Dirk de Villiers, Dr. Mauro
Ettorre, and Dr. Miroslav Pantaleev, for the time invested in reviewing my thesis.
This has been a great help in improving the quality and readability of this thesis.
And of course, my most sincere gratitude to my parents, sister and my own family
Esperanza, Marc and Tania for granting me a new sense of life and accompa-
nying me in this scientic journey.
Oleg
vi
List of Publications
This thesis is based on the work contained in the following appended papers:
Paper A
O. Iupikov, R. Maaskant, and M. Ivashina, Towards the Understanding of the
Interaction Eects Between Reector Antennas and Phased Array Feeds, in Proceed-
ings of the International Conference on Electromagnetics in Advanced Applications,
ICEAA 2012, Cape Town, South Africa, September 2012, pp. 792795.
Paper B
O. Iupikov, R. Maaskant, and M. Ivashina, A plane wave approximation in the
Proceedings of the
computation of multiscattering eects in reector systems, in
7th European Conference on Antennas and Propagation, EUCAP 2013, Gothenburg,
Sweden, April 2013, pp. 38283832.
Paper C
O. Iupikov, R. Maaskant, M. Ivashina, A. Young, and P.S. Kildal, Fast and Ac-
curate Analysis of Reector Antennas with Phased Array Feeds including Multiple
Reections between Feed and Reector, IEEE Transactions on Antennas and Prop-
agation, vol.62, no.7, 2014, pp. 34503462.
Paper D
C. Cappellin, K. Pontoppidan, P. H. Nielsen, N. Skou, S. S. Søbjærg, M. Ivashina,
O. Iupikov, A. Ihle, D. Hartmann, and K. v. 't Klooster, Novel Multi-Beam Radiome-
Proceedings of the 8th European Conference
ters for Accurate Ocean Surveillance, in
on Antennas and Propagation, EUCAP 2014, The Hague, The Netherlands, April
2014, pp. 15
Paper E
O. Iupikov, M. Ivashina, K. Pontoppidan, P. H. Nielsen, C. Cappellin, N. Skou,
S. S. Søbjærg, , A. Ihle, D. Hartmann, and K. v. 't Klooster, Dense Focal Plane
Proceedings of the 8th European
Arrays for Pushbroom Satellite Radiometers, in
Conference on Antennas and Propagation, EUCAP 2014, The Hague, The Nether-
lands, April 2014, pp. 15
vii
List of Publications
Paper F
A. Young, M.V. Ivashina, R. Maaskant, O.A. Iupikov, D.B. Davidson, Improving
the Calibration Eciency of an Array Fed Reector Antenna Through Constrained
Beamforming, IEEE Transactions on Antennas and Propagation, vol.61, no.7, July
2013, pp. 35383545.
Paper G
O. A. Iupikov, C. Craeye, R. Maaskant, M. V. Ivashina, Domain-Decomposition
Approach to Krylov Subspace Iteration, EEE Antennas and Wireless Propagation
Letters, vol.15, 2016, pp. 14141417.
Paper H
C. Cappellin, K. Pontoppidan, P. H. Nielsen, N. Skou, S. S. Søbjærg, A. Ihle,
M. V. Ivashina, O. A. Iupikov, K. v. 't Klooster, Design of a push-broom multi-
Proceedings of the 9th European
beam radiometer for future ocean observations, in
Conference on Antennas and Propagation, EUCAP 2015, Lisbon, Portugal, April
2015, pp. 15.
Paper I
O. A. Iupikov, M. V. Ivashina, K. Pontoppidan, P. H. Nielsen, C. Cappellin,
N. Skou, S. S. Søbjærg, A. Ihle, D. Hartmann, K. v. 't Klooster, An Optimal Beam-
forming Algorithm for Phased-Array Antennas Used in Multi-Beam Spaceborne Ra-
Proceedings of the 9th European Conference on Antennas and Propa-
diometers, in
gation, EUCAP 2015, Lisbon, Portugal, April 2015, pp. 15.
Paper J
M. V. Ivashina, O. A. Iupikov, C. Cappellin, K. Pontoppidan, P. H. Nielsen,
N. Skou, S. S. Søbjærg, B. Fiorelli, Enabling High-sensitivity Near-land Radiometric
Measurements With Multi-beam Conical Scanners Employing Phased Arrays, in
Proceedings of the 36th ESA Antenna Workshop on Antennas and RF Systems for
Space Science, ESA/ESTEC, The Netherlands, 6-9 October 2015, pp. 16.
Paper K
O. A. Iupikov, M. V. Ivashina, N. Skou, C. Cappellin, K. Pontoppidan, K. v. 't
Klooster, Multi-Beam Focal Plane Arrays with Digital Beamforming for High Preci-
sion Space-Borne Remote Sensing, Under review for IEEE Transactions on Antennas
and Propagation, 2017.
Paper L
O. A. Iupikov, A. A. Roev, M. V. Ivashina, Prediction of Far-Field Pattern
Characteristics of Phased Array Fed Reector Antennas by Modeling Only a Small
Part of the Array Case Study of Spaceborne Radiometer Antennas, in Proceedings
of the 11th European Conference on Antennas and Propagation, EUCAP 2017, Paris,
France, April 2017, pp. 14.
viii
Contents
Abstract i
Preface iii
Acknowledgments v
List of Publications vii
Contents ix
I Introductory Chapters
1 Introduction 1
1.1 Next generation radio telescopes . . . . . . . . . . . . . . . . . . . . . 1
1.2 Satellite radiometers for Earth observations . . . . . . . . . . . . . . . 3
1.3 Antenna arrays in satellite communication and telecommunication sys-
tems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Modeling, design and calibration challenges of novel Phased Array Feeds 6
1.5 Goal and outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . 7
2 Electromagnetic Analysis of Reector Antennas with Phased Ar-

ray Feeds Including Feed-Reector Multiple Reection Eects 9
2.1 Analysis method: formulation and validation of the iterative CBFM-
PO approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Acceleration techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Single plane wave approximation of the reector eld . . . . . 12
2.2.2 Plane wave spectrum (PWS) approach . . . . . . . . . . . . . 15
2.2.3 Near-eld interpolation (NFI) technique . . . . . . . . . . . . . 18
2.2.4 Analysis of PWE and NFI errors and simulation times . . . . . 20
2.3 Experimental verication of the CBFM-PO approach with acceleration
techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Numerical studies for dierent types of reector antenna feeds . . . . . 21
ix
Contents
2.5 Analysis of antenna arrays using Krylov subspace iteration with domain
decomposition approach . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3 Optimum Beamforming Strategies for Earth Observations 31

3.1 Performance requirements . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 Reector antenna design . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Optimum PAF beamformers . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.1 Standard maximum signal-to-noise ratio beamformer MaxSNR 36
3.3.2 Standard Conjugate Field Matching beamformer CFM . . . 37
3.3.3 Maximum Sensitivity, Minimum Distance to Land beamformer
MSMDL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.4 Advanced Maximum Beam Eciency beamformeR AMBER 41
3.3.5 Comparison of the beamformers . . . . . . . . . . . . . . . . . 43
3.4 Optimum PAF architectures . . . . . . . . . . . . . . . . . . . . . . . 47
3.4.1 Initial PAF layout . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4.2 Array size and inter-element spacing . . . . . . . . . . . . . . . 48
3.4.3 Dynamic range of beamformer weights . . . . . . . . . . . . . 50
3.5 Radiating element trade-o study . . . . . . . . . . . . . . . . . . . . 50
3.5.1 Requirements for the array radiator . . . . . . . . . . . . . . . 51
3.5.2 Array layouts and analysis method . . . . . . . . . . . . . . . 54
3.5.3 Comparison of the dipole, patch and Vivaldi elements . . . . . 56
3.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4 Beamforming Strategy for Beam Shape Calibration of PAF-equipped

Radio Telescope 63
5 Conclusions and recommendations for future work 65
A Radiometer characteristics for the PAFs of the patch-excited cups
and Vivaldi elements 69
References 77
II Included Papers
Paper A Towards the Understanding of the Interaction Eects Be-
tween Reector Antennas and Phased Array Feeds 91
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2 Analysis methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
x
Contents
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Paper B A Plane Wave Approximation In The Computation Of

Multiscattering Eects In Reector Systems 101
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
2 Modeling procedure and numerical results . . . . . . . . . . . . . . . . 102
3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Paper C Fast and Accurate Analysis of Reector Antennas with

Phased Array Feeds including Multiple Reections between Feed
and Reector 113
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
2 Iterative CBFM-PO Formulation . . . . . . . . . . . . . . . . . . . . . 115
3 Acceleration of the Field Computations . . . . . . . . . . . . . . . . . 120
3.1 Plane Wave Spectrum Expansion FFT . . . . . . . . . . . . 121
3.2 Near-Field Interpolation . . . . . . . . . . . . . . . . . . . . . 122
4.1 Validation of the Iterative Approach . . . . . . . . . . . . . . . 125
4.2 Field Approximation Errors . . . . . . . . . . . . . . . . . . . 128
4.3 Feed-Reector Antenna System Performance Study . . . . . . 131
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Paper D Novel Multi-Beam Radiometers for Accurate Ocean Surveil-

lance 145
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
2 Optical Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
2.1 Conical scanning radiometer antenna . . . . . . . . . . . . . . 147
2.2 Torus push-brom radiometer antenna . . . . . . . . . . . . . . 147
3 Antenna Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 149
3.1 Acceptable cross-polarization . . . . . . . . . . . . . . . . . . . 149
3.2 Acceptable side lobes and distance to coast . . . . . . . . . . . 149
4 Feed Array Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.1 Conical scanning radiometer antenna . . . . . . . . . . . . . . 149
4.2 Torus push-brom radiometer antenna . . . . . . . . . . . . . . 151
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Paper E Dense Focal Plane Arrays for Pushbroom Satellite Ra-

diometers 159
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
xi
Contents
2 Antenna Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 161

3 FPA-system design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3.1 Antenna array model . . . . . . . . . . . . . . . . . . . . . . . 162
3.2 Beamforming algorithms . . . . . . . . . . . . . . . . . . . . . 163
3.3 Parametric study . . . . . . . . . . . . . . . . . . . . . . . . . 163
4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5 Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Paper F Improving the Calibration Eciency of an Array Fed Re-

ector Antenna Through Constrained Beamforming 173
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
2 Antenna Pattern Model . . . . . . . . . . . . . . . . . . . . . . . . . . 175
3 Beamforming Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
3.1 Number of Constraints and Pattern Calibration Measurements 177
3.2 Constraint Positions . . . . . . . . . . . . . . . . . . . . . . . . 177
3.3 Constraints Vector . . . . . . . . . . . . . . . . . . . . . . . . 178
4.1 Beam Directivity and Side Lobe Levels . . . . . . . . . . . . . 180
4.2 Calibration Performance . . . . . . . . . . . . . . . . . . . . . 181
4.3 Comparison of MaxDir and LCMV beamformers . . . . . . . . 182
5 Conclusions and Recommendations . . . . . . . . . . . . . . . . . . . . 183
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Paper G Domain-Decomposition Approach to Krylov Subspace It-

eration 191
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
2 Segmented Krylov subspace as MBFs . . . . . . . . . . . . . . . . . . 192
3 CBFM with restarts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Paper H Design of a push-broom multi-beam radiometer for future

ocean observations 205
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
2 Antenna requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3 Push-broom antenna . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
3.1 Antenna geometry . . . . . . . . . . . . . . . . . . . . . . . . . 207
3.2 Feed array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4 Feed array design principles . . . . . . . . . . . . . . . . . . . . . . . . 208
xii
Contents
5 RF performance results . . . . . . . . . . . . . . . . . . . . . . . . . . 209

5.1 Central beam at Ku band . . . . . . . . . . . . . . . . . . . . . 209
5.2 Reduction of feed array rows along φ . . . . . . . . . . . . . . 210
5.3 Total feed arrays for C-, X- and Ku-band . . . . . . . . . . . . 211
5.4 Additional performance checks . . . . . . . . . . . . . . . . . . 212
6 Mechanical design of the push-broom torus antenna . . . . . . . . . . 212
7 Feeding network and receiver issues . . . . . . . . . . . . . . . . . . . 213
7.1 Existing state-of-the-art components . . . . . . . . . . . . . . . 213
7.2 Realistic components within a 5 years time frame . . . . . . . 214
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Paper I An Optimal Beamforming Algorithm for Phased-Array An-

tennas Used in Multi-Beam Spaceborne Radiometers 217
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
2 Array design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
3 Optimal beamforming algorithm . . . . . . . . . . . . . . . . . . . . . 219
3.1 Generic formulation . . . . . . . . . . . . . . . . . . . . . . . . 219
3.2 Computation acceleration . . . . . . . . . . . . . . . . . . . . . 219
3.3 Iterative procedure for constraints on the dynamic range of the
weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
4 Parametric studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
4.1 Beamformer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
4.2 PAF size and nal radiometer characteristics . . . . . . . . . . 222
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Paper J Enabling High-sensitivity Near-land Radiometric Measure-

ments With Multi-beam Conical Scanners Employing Phased Ar-
rays 229
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
2 Limitations of horn feeds . . . . . . . . . . . . . . . . . . . . . . . . . 230
3 Arrays feeds: CFM beamforming . . . . . . . . . . . . . . . . . . . . . 232
4 Arrays feeds: Max. Sensitivity - Min. Distance-to-Land beamforming 235
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Paper K Multi-Beam Focal Plane Arrays with Digital Beamforming

for High Precision Space-Borne Remote Sensing 241
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
2 From oceonagraphic requirements to antenna system specications . . 243
2.1 Spatial resolution (FP) ⇒ reector diameter . . . . . . . . . . 244
2.2 Bias (∆T) ⇒ acceptable cross-polarization power . . . . . . . 244
xiii
Contents
2.3 Distance to coast (Dc ) ⇒ acceptable side lobes and cross-polarization

power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
2.4 Radiometric resolution (∆Tmin ) ⇒ number of beams . . . . . . 246
3 Reector antenna design . . . . . . . . . . . . . . . . . . . . . . . . . 247
3.1 Reector geometries . . . . . . . . . . . . . . . . . . . . . . . . 247
3.2 Reector surface technology . . . . . . . . . . . . . . . . . . . 248
4 Limitations of cluster feeds of horns . . . . . . . . . . . . . . . . . . . 249
5 Dense Focal Plane Arrays . . . . . . . . . . . . . . . . . . . . . . . . . 251
5.1 Array models and congurations . . . . . . . . . . . . . . . . . 251
5.2 Optimization procedure for element excitation . . . . . . . . . 252
5.3 Antenna patterns and radiometric characteristics . . . . . . . . 254
6 Receiver considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 256
7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
Paper L Prediction of Far-Field Pattern Characteristics of Phased

Array Fed Reector Antennas by Modeling Only a Small Part of
the Array Case Study of Spaceborne Radiometer Antennas 265
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
2 Antenna geometry and specications . . . . . . . . . . . . . . . . . . . 266
3 Array antenna design . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
4 Analysis methodology and numerical results . . . . . . . . . . . . . . . 268
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
xiv
Part I
Introductory Chapters
Chapter
1 Introduction
Since recently, several types of so-called dense Phased-Array Feed (PAF) systems
for reector antennas have been designed for applications in future instruments for
radio astronomy, Earth surface and space observations [111]. The main advantage of
these PAFs over conventional single-horn feeds and cluster feeds of horns is that the
inter-element separation distance of such dense PAFs can be much smaller than one
wavelength to allows the formation of multiple closely overlapping beams with high
eciency [12]. Another advantage is that these PAFs can be equipped with digital
beamformers providing an individual complex excitation per array antenna element
and hence can realize an optimal illumination of the reector aperture [1318]. These
advantageous properties are of great importance both for radio astronomy and Earth
observation applications requiring fast and wide eld-of-view (FOV) surveys.
1.1 Next generation radio telescopes

The eectiveness of performing wide-eld surveys is characterized by the telescope's
survey speed, i.e., the speed at which a certain volume of space can be observed with
a given sensitivity. The survey speed is proportional to the size of the instantaneous
FOV and the frequency bandwidth, weighted by the sensitivity squared [19]. Present-
day aperture synthesis radio telescopes have a limited observation capability due to
the fact that only a small part of the sky can be observed simultaneously, which
therefore results in a low survey speed. In contrast, using PAFs as a reector antenna
feed allows: (i) to increase the receiving sensitivity of the reector antenna due to
better illumination of the dish, and; (ii) to form multiple simultaneous beams, which
can be closely overlapped, as a result of overlapping sub-arrays forming these beams,
to provide a continuous FOV [20]. An example of such system which recently became
operational is APERTIF [7], which is developed by The Netherlands Institute for
Radio Astronomy (ASTRON) and illustrated in Fig. 1.1.
The FOV of conventional telescopes with single-beam feeds is limited to one half-
power beamwidth, where the sensitivity takes the maximum value along the beam
axis and gradually decreases from its center. To image a larger region of the sky,
1
Chapter 1. Introduction
Figure 1.1: APERTIF project [7], which aims to increase the eld-of-view of the Westerbork Syn-
thesis Radio Telescope (WSRT) with a factor 25, as is illustrated in the bottom-left inset. This
performance gain is achieved by placing a receiver array in the focus of each parabolic dish of the
WSRT, instead of the single receiver element that the current system employs [26].
astronomers use the mosaicing technique [21]. With this technique, a telescope
performs many observations by mechanically steering (scanning) the dish such that
the main lobes of the beams generated in subsequent observations closely overlap and
form an almost continuous beam envelope when superimposed. The large-eld image
is therefore formed by composing a mosaic of smaller sized overlapping images taken
during these observations. According to Nyquist's eld-sampling theorem, a uniform
sensitivity of the combined image is achieved when the beam separation is equal to
or smaller than one half of the half-power beamwidth [22]. A larger spacing between
the observations results in a sensitivity ripple over the FOV [17, 23]. The maximum
allowable ripple will depend on the particular science case.
PAFs can provide many closely overlapping beams in one snapshot, thereby
greatly improving the size of the FOV. However, to meet the required eld-sampling
limit with a cost-eective number of PAF beams, their shapes should be optimized
and the maximum achievable receiving sensitivity, as well as minimum receiver and
antenna noise [24, 25], should be traded against the maximum tolerable sensitivity
ripple over the FOV.
In addition to a continuous FOV and high sensitivity, high polarization discrimi-

nation is required for large-eld surveys [2729]. For this purpose, the incident eld
is sampled by two orthogonally polarized receptors or beams. In radio astronomy,
the polarization purity of the resulting images is established after extensive oine
2
1.2. Satellite radiometers for Earth observations
calibration of the data. In this respect, two antenna design aspects are of particular
importance: the stability (i.e. variation over time) of the co- and cross-polarized
beams; and the orthogonality of the two beams in the direction of incidence. This
requires that the beams are formed simultaneously and span a 2-D basis along which
the incident eld is decomposed. Future PAF-equipped telescopes are potentially
accurate polarimeters thanks to the exibility that digital beamforming oers. How-
ever, although the orthogonality of the beam pair in the direction of observation may
be improved electronically, it is important that the intrinsic polarization character-
istics of the beams are suciently good to minimize such corrections as they may
compromise the receiving sensitivity.
Another important concern about radio telescopes is their calibration procedure.
This requires accurate models of the instrumental parameters and propagation con-
ditions, which vary over time, so that the model parameters have to be determined
during the observation time through a number of calibration measurements [30]. To
perform calibration of radio telescopes eciently, the number of model parameters
should be minimal. One of the instrumental parameters that needs accurate char-
acterization is the radiation pattern of the antenna, which is especially challenging
for future array based multiple beam radio telescopes due to complexity of such
instruments and increased size of the FOV.
To be able to characterize all beams inside the FOV by means of a simple beam
model, beamforming techniques can be used to create similarly shaped beams [31,32].
However, this leads to a loss in the receiving sensitivity requiring us to employ more
advanced but still simple beam models. An attempt to develop such beam model in
conjunction with constrained beamforming technique is made in this work.
1.2 Satellite radiometers for Earth observations

Besides radio astronomy applications, PAFs are used in other applications, such as
remote sensing of the atmosphere and the Earth's surface [33, 34]. However, there are
some important dierences in requirements for the instruments in these applications.
For example, receivers for Earth remote sensing are typically designed to measure high
brightness temperatures (75−300 K) along with short integration times, while in radio
astronomy very low brightness temperatures are of interest and the integration time
can reach many hours. Therefore, the receiving sensitivity is one of the key instrument
characteristics, in particular for radio astronomy applications. On the other hand,
for Earth remote sensing applications, such as the assessment of ocean parameters
(salinity, sea surface temperature, ocean vector wind), additional specications for
high beam eciency and measurement accuracy near a coast line are required [10,35].
Recent advances in phased array antenna technologies and low-cost active elec-
tronic components open up new possibilities for designing Earth observation instru-
ments, in particular those used for radiometric measurements. Nowadays, two de-
3
Figure 1.2: Operational principle of a push-broom microwave radiometer, which includes an o-set
toroidal reector antenna fed by a multi-beam focal plane array of horns arranged perpendicular
to the ight direction of the spacecraft. Dierent areas of the ocean-surface are scanned as the
spacecraft ies forward.
sign concepts of microwave radiometers are in use: push-broom and whisk-broom

scanners [36]. Push-broom scanners have an important advantage over whisk-broom
scanners in providing larger FOV with higher sensitivity, owing to the fact that these
systems can observe a particular area of the ocean for a longer period of time with
multiple simultaneous beams. However, the drawback of pushbroom designs based
on conventional focal plane arrays of horns in one-horn-per-beam conguration [37]
or clusters with simplistic beamforming schemes [38] is the FOV varying sensitiv-
ity. This variation occurs due to the dierences between scanned beams, as these are
formed by dierent horns or clusters, and their large beam separation distance on the
oceanic surface, which is caused by a large separation distance between the horns.
This drawback may be signicantly reduced by employing dense PAFs consist-

ing of many electrically small antenna elements utilizing advanced beamforming
schemes [1517]. This technology has been extensively studied during the last decade
in the radio astronomy community, and several telescopes are currently being equipped
with dense PAFs [7, 39, 40]. While those systems aim at providing scan ranges of
about 5 − 10 beamwidths, for applications as herein considered, the desired scan
range (swath range of the radiometer) is one order of magnitude larger [10]. To
achieve this large scan-range performance, more complex reector optics and PAF
4
1.3. Antenna arrays in satellite communication and telecommunication...
designs are required. For push-broom radiometers, various optics concepts have been
investigated [37], and the optimum solution has been found to be an oset toroidal
single reector antenna, such as illustrated in Fig. 1.2. This reector structure is ro-
tationally symmetric around its vertical axis, and thus is able to cover a wide swath
range. However, its aperture eld exhibits signicant phase errors due to the non-
ideal (parabolic) surface of the reector, which requires the use of a more complex
feed system.
1.3 Antenna arrays in satellite communication and

telecommunication systems
Recent developments in antenna technologies and low-cost active electronic compo-
nents open a possibility to produce aperture antenna arrays and PAFs for communi-
cation applications as well, in order to improve dierent aspects of radio links. For
example, authors in [41] suggest to use PAF for a high-gain reector antenna to com-
pensate pointing errors due to the unwanted movement of the antenna mast and to
ease the antenna installation. A beam squint compensation in circularly polarized
oset reector antennas is proposed in [42] by employing a sequentially rotated PAF
(Fig. 1.3).
Advanced PAF systems are also being considered as potential candidates for large-
scale array antenna systems for 5G wireless infrastructure, as they can oer unique
electronic beamforming approaches. The PAF beamformers hybridize analog and
digital beamforming in an optimal manner through the combination of the best of
two worlds, i.e. robustness, low-cost and design simplicity of reector antennas with
Figure 1.3: Prototype of the array-fed oset reector antenna [42], and the array of sequentially
rotated aperture-coupled microstrip antennas in the focal plane of the prototype.
5
the exibility of phased arrays. The recently funded Horizon 2020 Innovative Train-
ing Network SILIKA [43] is dealing with quasi-optically beamformed array antennas
such as Focal Line Arrays and Focal Plane Arrays, where each antenna element sees
multiple users within a sector of the total coverage area, and where each active array
element consists of a silicon integrated circuit with integrated antenna radiator.
1.4 Modeling, design and calibration challenges of

novel Phased Array Feeds
The design of the above-mentioned highly complex PAF systems requires the de-
velopment of accurate and ecient modeling techniques. This is a challenging task
considering the size of the reector used in radio astronomy and Earth observation
applications, which can be hundreds of wavelengths in diameter, as well as the size of
the PAF, which is too small to be analyzed with an innite array simulation approach
(which also has limitations on the excitation schemes), but too large for the direct
usage of full-wave methods implementing plain MoM or FDTD techniques that run
on standard computing platforms.
During the last decades, a number of analytical and numerical techniques have
been developed to model feed-reector interaction eects. For example, in [44], the
multiscattered eld between the feed and reector is approximated by a geometric
series of on-axis plane wave (PW) elds, each of which is scattered by the antenna
feed due to its incident PW at each iteration, and where the amplitudes of these PWs
are known in closed-form for a given reector geometry. This method is very fast and
insightful, while MoM-level accuracy can be achieved for single-horn feeds, but not
for array feeds as demonstrated in Paper B. An alternative approach is to use more
versatile, though more time-consuming, hybrid numerical methods combining Phys-
ical Optics or Gaussian beams for the analysis of reectors with MoM and/or Mode
Matching techniques for horn feeds [45, 46]. The recent article [47] has introduced a
PO/Generalized-Scattering-Matrix approach for solving multiple domain problems,
and has shown its application to a cluster of disjoint horns. This approach is generic
and accurate, but may require the lling of a large scattering matrix for electrically
large PAFs and/or multifrequency front-ends (MFFEs) that often includes a large
extended metal structure [48]. Other hybrid methods, which are not specic for solv-
ing the present type of problems, make use of eld transformations, eld operators,
multilevel fast multipole approaches (MLFMA), and matrix modications [4952].
Recently, a Krylov subspace iterative method has been combined with an MBF-
PO approach for solving feed-reector problems [53], and complementary to this, an
iteration-free CBFM-PO approach has been presented by S. Hay, where a modied
reduced MoM matrix for the array feed is constructed by directly accounting for the
presence of the reector [54]. However, most of these methods are either complicated
6
1.5. Goal and outline of the thesis
or slow, or do not allow for the extraction of the feed-reector interaction eect in a
systematic manner.
Besides the eciency and simplicity of modeling techniques, their accuracy is of
great importance too. For example, present-day radio telescopes with single-beam
6
feeds can achieve a dynamic range upward to 10 : 1 along the on-axis beam direction.
However, the o-axis dynamic range is severely limited by uncertainties and temporal
instabilities in the beam patterns caused by gain drift in PAF channels, mispointing
and mechanical deformations of the dishes, as well as by station-to-station beam pat-
terns dierences [55, 56]. A number of calibration techniques for dealing with these
eects have been proposed and used in practical systems [21, 30, 57, 58]. For novel
PAF-based telescopes, the beam calibration is a new challenging eld and there is
not yet a clear consensus on what constitutes a good beam pattern. Furthermore,
the mutual coupling between the PAF and the dish(es) of a reector antenna gives
rise to a frequency dependent ripple in the antenna radiation and impedance char-
acteristics [59], which exacerbates the calibration. Accurate system models can help
alleviating the beam calibration problem.
In conclusion, the challenges in modeling, designing and calibrating novel PAFs,
are:
complexity to accurately model a large antenna array of complex antennas,

including mutual coupling between array elements; accurate modeling of an
antenna radiation eciency can be challenging as well [24];
cumbersomeness of analyzing a combined PAF-reector structure due to the

large size of the reector and mutual coupling between them (multi-scale prob-
lem);
development of optimal beamforming algorithms that provide performance re-

quirements on multiple antenna characteristics (e.g. beam eciency, side-lobe
level, sensitivity, etc), while realizing easy-to-calibrate beam shapes and main-
taining minimum complexity of the array design (minimum number of elements,
similarity of sub-arrays, etc);
Calibration of the Phased array feed (PAF) based radio telescope, which largely
depends on the accuracy of the antenna beam model.
1.5 Goal and outline of the thesis

The herein presented work is devoted to address the following challenges: (i) the de-
velopment of a reector antenna model, which accounts for the feed-reector coupling
and provides physical insight in the coupling processes, and the analysis of several
reector antennas for dierent types of feeds and determining which of these feeds
7
are preferred in terms of low feed-reector coupling and overall antenna performance;
(ii) the design of PAFs for an oset toroidal reector antenna and the development
of optimal beamforming algorithms for accurate radiometric measurements; (iii) im-
proving the calibratibility of the beam shape of a radio telescope.
This thesis is organized as follows. In Chapter 2 a general CBFM-PO model of a
reector antenna system is developed. This model is based upon the Jacobi method
for solving a system of linear equations iteratively. The Characteristic Basis Function
Method (CBFM) is used to model the feed, while the Physical Optics (PO) approach
is used to model the current on the reector at each iteration.
To speed-up the method, several acceleration techniques are developed: the eld
scattered from the reector is expanded in a Plane Wave Spectrum (PWS), while
the eld radiated/scattered by the feed is computed at few near-eld points only
and then interpolated in order to nd the PO current distribution on the reector
surface. This allows us to simulate a reector antenna 5 − 100 times faster than a
pure CBFM-PO approach.
Afterwards, the developed method is used to model large reector antennas (38λ
and 118λ) fed by dierent types of feeds: (i) a single dipole above a ground plane;
(ii) a 20-elements dipole array; (iii) a 121-element dipole array; (iv) a 121-element
Vivaldi array; (v) a classical pyramidal horn with aperture size of ∼1λ, and; (vi) the
same horn with extended ground plane, which could represent a feed cabin of the
reector antenna.
Chapter 3 describes a PAF design procedure and several beamforming strate-
gies for the application of satellite radiometers observing the sea surface, where the
requirements for such radiometers are specied and translated into performance g-
ures in terms of antenna characteristics. Two beamformer algorithms are developed
to meet the tight radiometer requirements, and compared to the commonly used
Conjugate-Field-Matching beamformer. An algorithm to limit the weights dynamic
range keeping them optimal is also developed and numerically tested.
Next, a design procedure of an optimal PAF is presented, including analysis of
the required array size and inter-element spacing, as well as trade-o study between
several candidates for the array radiating element. In the later study several analysis
approaches are described and compared, which allow for a time-ecient analysis of
radiometers with PAFs.
Numerical results for the designed radiometer equipped with the optimal PAF are
presented for each type of beamformer.
In a short Chapter 4 it is shown how a constrained beamforming strategy can be
used to improve the calibration eciency of the PAF beam shape of a radio telescope.
The conclusions and recommendations are described in Chapter 5.
8
Chapter
2 Electromagnetic Analysis
of Reector Antennas with
Phased Array Feeds Includ-
ing Feed-Reector Multiple
Reection Eects
The characterization of feeds in unblocked reectors and on-axis beams can be han-
dled by the traditional spillover, illumination, polarization and phase subeciency
factors dened for rotationally symmetric reectors in [60], and be extended to in-
clude excitation-dependent decoupling eciencies of PAFs [20, 61]. The current work
investigates the eects of aperture blockage and multiple reections on the system
performance in a more generic fashion than it was done in [44] and [62] for rotationally
symmetric antennas and single-pixel feeds.
2.1 Analysis method: formulation and validation

of the iterative CBFM-PO approach
The herein proposed analysis method is based on the Jacobi method intended to solve
a system of linear equations in an iterative manner. Suppose that the MoM matrix
equation for the entire reector antenna (including both the dish and the feed) is
given by
ZI = V, (2.1)
where Z is the MoM matrix of size K ×K and V is a K ×1 excitation vector.

This matrix can be decomposed into matrix blocks as
Z Zrf Ir
rr r
V
f = , (2.2)
fr
Z Z
I Vf
9
Chapter 2. Electromagnetic Analysis of Reflector Antennas with Phased...
where Zrr Z are the MoM matrix self-blocks of the reector and feed, re-
and
1 r f
spectively , and V and V are the corresponding excitation vectors. The matrix
Zrf = (Zfr )T contains the mutual reactions involving the basis functions on the feed
r
and reector. The unknown current expansion coecient vectors are denoted by I
f
and I .
It can be shown that the solution to Eq. (2.2) can be written as an innite
geometric series (see Paper C for the derivation), which, in turn, can be represented
by the recursive scheme:
Reector Feed
∞ ∞
f
r r Ifn
X X
I = In (2.3a) I = (2.4a)
n=0 n=0
In+1 = −(Zrr )−1 Zrf Ifn

r
Ifn+1 = −(Z ) Zfr Irn
−1
(2.3b) (2.4b)
r rr −1 r
If0 = (Z ) Vf
−1
I0 = (Z ) V (2.3c) (2.4c)
The cross-coupled recursive scheme as formulated by Eqs. (2.3) and (2.4) is ex-
emplied in Fig. 2.1 as a ve-step procedure, in which the problem is rst solved in
isolation to obtain Ir0 and If0 . Afterwards, the feed current If0 is used to induce the
r
reector current I1 , which is then added up to the initial reector current. Likewise,
the initial reector current Ir0 is used to induce the feed current If1 , which is then
added to the initial feed current, and so forth.
Rather than computing the reector and feed currents through the large-size MoM
matrix blocks Zrr , Zrf , Zfr , and Z , additional computational and memory ecient
techniques can be employed for the rapid computation of these currents at each
iteration. Here, the Physical Optics (PO) current is used on the reector surface
and the Characteristic Basis Function Method (CBFM, [64]) is invoked as a MoM
enhancement technique for computing the current on the feed. Please see the Paper C
for details on how this is done.
The above described approach has been validated using the MoM solver as part of
the CAESAR software [64, 65] and the commercial software FEKO [66] (c.f. Paper C
for details).
2.2 Acceleration techniques

The above-described approach allows us to simulate reector antennas employing
electrically large reectors fed by complex feeds like PAFs of hundreds of Vivaldi an-
tennas. However, the approach requires the eld to be computed at numerous points
on both the feed and the reector surfaces, thereby rendering the eld computations
1 Here Z includes the eect of the antenna port terminations [63].
10
2.2. Acceleration techniques
Step (i) V Step (ii)
Zload
If0
Transmit case:
Ir0 = 0 Ir1
Step (iii) Step (iv)
If1
Ir2
V
Zload
Step (v)
If = If0 + If1 + If2 + . . .
Ir = Ir0 + Ir1 + Ir2 + . . .
Figure 2.1: Illustration of the cross-coupled iterative scheme for the multiscattering analysis of the
feed-reector interaction eects, as formulated by Eqs. (2.3) and (2.4): (i) The antenna feed radiates
in the absence of reector; (ii) the radiated eld from the feed scatters from the reector; (iii) the
scattered reector eld is incident on the terminated feed and re-scatters; (iv) the re-scattered eld
from the feed is incident on the reector; etc. (v) the nal solution for the current is the sum of the
subsequently induced currents.
inecient, in particular for complex-shaped electrically large feed antennas employ-

ing hundreds of thousands of low-level basis functions. Similarly, one has to cope
11
with a computational burden when calculating the PO equivalent current on elec-

trically large reectors. In this section a few enhancement techniques are presented
that accelerate the eld computations while maintaining high accuracy.
2.2.1 Single plane wave approximation of the reector eld

The method described here relies on the fact that the eld scattered from the reector
resembles a plane wave (PW), and therefore can be dened by a single PW mode
amplitude. In [44] this amplitude is expressed analytically at each iteration for a
given reector geometry, and the scattered eld of the feed is approximated by a
geometric series of elds scattered by the antenna feed due to an incident plane wave
with known amplitude. With reference to Fig. 2.2, the total radiation pattern of the
feed Etot (including feed-reector coupling) can be expressed as
− r10 A(0) exp (−jk2r0 )

Etot (θ, φ, r) = Er (θ, φ, r) + 1 Es (θ, φ, r), (2.5)
1+ A (0) exp (−jk2r0 )
r0 s
where Er and Es are the radiation and scattering far-eld patterns of the feed in
isolation correspondingly, and A(0) and As (0) are values of the co-polarization com-
ponent of these elds in the on-axis direction [see Fig. 2.2(a) and 2.2(b)]; r0 is the
distance between the reector apex and the phase reference point with respect to
which Er and Es are dened.
Plane Plane
wave wave
r0 r0
A(0) A(0) As(0) As(0)
Er Er Es Es Etot Etot
Phase ref. Phase ref. Phase ref. Phase ref.
point point point point
(a) The radiation (b) The scattering (c) The total pattern of the
pattern of the feed pattern of the feed feed including coupling with
on transmit due to an incident the reector
unit PW from the
direction of the
reector
Figure 2.2: Semi-analytical PW approximation as described in [44].
12
However, as shown in Paper B, this semi-analytical approach works well only when
the feed is small w.r.t. the reector and when it has low-scattering properties. If the
feed becomes electrically large and high-scattering (such as for conventional multi-
frequency front-ends in radio telescopes), the accuracy of this method deteriorates. In
order to improve the accuracy, the plane wave coecient can be computed numerically
at each iteration. To do so, the eld scattered from the reector is sampled in the
focal plane, and the PW coecient is computed as an average of the sampled eld
values on a regular grid (see Paper B for the derivation):
K
1 X ref
α≈ E (rk ), (2.6)
K k=1 p
where Epref is the dominant p-component of the focal eld, and the set {rk }K
k=1 are K
sample points, which are assumed to be located on a uniform grid in the focal plane.
In summary, the plane-wave-enhanced MoM/PO method consists of the following
steps: (i) the antenna feed currents are computed through a method-of-moments
(MoM) approach by exciting the antenna port(s) in the absence of the reector;
(ii) these currents generate an EM eld which induces PO-currents on the reector
surface; (iii) the PO currents create a scattered eld that is tested at only a few
points in the focal plane; (iv) the eld intensity at the sample points is averaged in
accordance with (2.6), and the obtained value is used as the expansion coecient for
the plane wave traveling from the reector towards the feed; (v) this incident plane
wave induces a new current distribution on the feed structure. The steps (ii)(v) are
repeated until a convergence condition is met.
The following three types of feeds are used to illuminate a reector antenna: (i) a
pyramidal horn with aperture diameter in the order of one wavelength; (ii) a pyrami-
dal horn with extended ground plane, and; (iii) an 121-element dual-polarized dipole
array (see Fig. 2.3). All antennas are impedance power-matched, so that the an-
tenna component [67] of their corresponding radar cross-section (RCS) is minimized.
However, the residual component of the RCS of the horn with ground plane is still
high due to the extended metal structure surrounding it, so that this feed is a high
scattering antenna and strong feed-reector coupling can be expected.
The above feeds are used to illuminate two parabolic reectors with aperture
diameters 38λ and 118λ, and the errors introduced by the PW approximation in the
focal eld and scalars antenna characteristics are computed as
rP
ref − E mod |2
|Ep;k p;k
k
1 = rP × 100% (2.7)
ref |2
|Ep;k
k
|f ref − f mod |
2 = × 100%, (2.8)
|f ref |
13
where
ref and E mod are the k -th sample of the discretized p-components of the actual
Ep;k p;k
ref
focal E-eld E (x, y) and the focal eld modeled by a plane wave E
mod (x, y) respec-
tively; f
ref and f
mod is the gain or antenna input impedance, reference and modeled
values, respectively. The MoM/PO results without the plane wave approximation
are used as the reference solution.
Table 2.1: Errors due to the plane wave approximation
Focal eld Gain Gain Impedance

(on-axis) (@−3 dB)
Reector
38λ 118λ 38λ 118λ 38λ 118λ 38λ 118λ
diameter D
Feed: Pyramidal horn

Parameter
3.91 1.23 1.98 0.62 3.99 2.16 15.05 4.66
variation, %
Method: Error, %
Method 1 0.3 0.05 0.28 0.05 0.36 0.14 1.37 0.18
Method 2 0.1 0.04 0.16 0.04 0.3 0.13 0.09 0.03
Feed: Pyramidal horn with extended ground plane

Parameter
139.3 39.1 19.2 3.4 29.4 3.56 43.4 6.1
variation, %
Method: Error, %
Method 1 37.7 1.29 12.7 0.1 10.1 0.17 18.5 0.2
Method 2 11.9 0.48 2.23 0.07 4.71 0.15 12.46 0.11
Feed: 121-element dual-polarized dipole array

Parameter
8.45 3.28 1.84 0.28 3.68 0.73 5.8 1.7
variation, %
Method: Error, %
Method 1 0.61 0.11 0.21 0.03 0.15 0.02 0.34 0.08
Method 2 0.44 0.1 0.12 0.03 0.08 0.03 0.58 0.05
(a) Horn (b) Horn with gnd plane (c) Dipole array
Figure 2.3: Considered feed geometries (in addition to the dipole feed with PEC ground plane):
(a) a classical pyramidal horn with aperture length ∼ 1λ; (b) the same horn but with extended
ground plane (∼3.7λ), where the ground plane may model the presence of a large feed cabin; (c) an
antenna array consisting of 121 0.45λ-dipoles above a ground plane of the same size; (d) the same
array, but with the dipoles replaced by wideband tapered slot Vivaldi antennas.
14
The above errors that have been computed for both the semi-analytical and the
numerical PW-approximation approaches are summarized in Table 2.1. We will refer
to the semi-analytical method as Method 1, while the proposed above approach is
denoted as Method 2.
The total simulation time (10 frequency points) for the 38λ reector fed by the
considered feeds is shown in Table 2.2.
Table 2.2: Total simulation time
Horn Horn with Dipole array Vivaldi

gnd plane array
MoM-PO, no 9 min 05 sec 59 min 21 sec 71 min 09 sec 197 min 04 sec
approximations (100%) (100%) (100%) (100%)
0 min 39 sec 1 min 12 sec 4 min 49 sec 33 min 58 sec
Method 1
(7%) (2%) (7%) (17%)
Method 2
(28%) (23%) (27%) (34%)
By analyzing Table 2.1 and Table 2.2 the following observations can be made:
Method 1 is numerically ecient and accurate for small feeds (whose sizes are
in the order of one wavelength) and for low-scattering feeds, but fails in case
of large high-scattering feeds, such as MFFEs, because the focal eld produced
by the feed scattering pattern has a high level and a highly tapered shape;
Method 2 provides a better prediction of all the system parameters, since it

accounts for the actual shape of the scattering pattern when tting the plane
wave to it; however, it is slower than Method 1;
Both methods are accurate in case of large reectors, because the multiscat-
tering eects are less pronounced (see Parameter variation in Table 2.1), and
the eld scattered from the reector is close to a plane wave at all iterations.
For the focal eld distribution plots and more detailed discussions, see Paper B.
2.2.2 Plane wave spectrum (PWS) approach

Further improvement of the accuracy can be achieved by expanding the sampled focal
eld into a plane wave spectrum (PWS) [6870].
With reference to Fig. 2.4, a grid of sampling points in the xy -plane P in front of
the feed at z=0 is chosen for the expansion of the PO radiated eld in terms of a
PWS. Each PW propagates to a specic observation point r on the feed where the
eld E i,f is tested. This process of eld expansion and PW propagation is realized
15
through the application of the truncated Fourier Transform pair [68]
Zymax Zxmax
1
A(kx , ky ) = E i,f (x, y, z = 0)ej(kx x+ky y) dx dy (2.9a)
2π
−ymax −xmax
max kymax
Zkx
1
Z
E i,f (r) = A(kx , ky )e−jkz z e−j(kx x+ky y) dkx dky (2.9b)
2π
−kxmax −kymax
where
k 2 > kx2 − ky2
p 2
kp− kx2 − ky2 if
kz = , (2.10)
−j kx2 − ky2 − k 2 otherwise.
and where the spectrum of PWs is limited to only those that are incident on the feed
from directions within an angle subtended by the reector and seen from the center
of the plane P (see Fig. 2.4).
The magnitude of the co-polarized spatial frequency spectrum |Aco (kx , ky )| com-
puted for small and large sampling plane sizes are shown in Fig. 2.5. It exhibits
several interesting features: (i) as expected, the dominant spectral component corre-
sponds to the on-axis PW, for which kx = ky = 0, while the second strongest set of
xmax
r xfmax
Ei,f ẑ
P ŷ
d
x̂
z=0 kxmax x̂ ∆y
∆x
n̂
Figure 2.4: The FFT-enhanced PWS expansion method for the fast computation of the feed current
due to the E -eld from the reector. Firstly, the incident eld E i,f is sampled in the xy plane P
in front of the feed in order to obtain the sampled PWS A(kx , ky ); Secondly, each spectral PW
propagates to an observation point r on the feed where E i,f is tested to compute the induced feed
current.
16
−70 −60 −50 −40 −30 −20 −10 0
dB
0 0
30 30
−10 −10
20 20
−20 −20
10 10
−30 −30
ky
ky
0
−40 −40
−10 −10
−50 −50
−20 −20
−60 −60
−30 −30
−70 −70
−20 0 20 −20 0 20
kx kx kxmax
(a) (b)
Figure 2.5: (a) The magnitude of the spatial frequency spectrum |Aco (kx , ky )| (i.e. plane wave
spectrum) for the 38λ reector fed by the dipole array in case the FFT grid size is equal to size of
the feed, and (b) when it is eight times the feed size.
PWs originate from the rim of the reector, as observed by the spectral ring structure
2 2 max )2 = (k max )2 ; (ii) the magnitude of the PWs originating from
for which kx +ky = (kx y
the rim is polarization dependent, in fact, it is seen that, since the feed is X -polarized,
the feed eld interacts more at the top and bottom segments of the rim.
The approximation of the reector eld by a PWS introduces an error, 1 , in the
surface current of the feed. The relative error between the current expansion coef-
cient vectors Iapprox and Iref for the iterative CBFM-PO solution with and without
eld approximations, respectively is computed as
 
s s
|Iiref − Iiapprox |2 |Iiref |2  × 100%.
X X
1 =  (2.11)
i i
Fig. 2.6 illustrates the relative error computed as a function of the FFT sampling
plane size P when the PWS is employed for expanding the reector radiated eld
(for PWS parameters see Paper C), and when only the dominant on-axis PW term
is used. As expected, the error decreases for an increasing sampling plane size, since
more spectral PW terms are taken into account while the eect of the FFT-related
periodic continuation of the spatial aperture eld decreases. Henceforth, we choose
the sampling plane size equal to that of the feed, for which the feed current error is
about −35 dB for all the considered feeds, while it represents a good compromise from
both a minimum number of sampling points and accuracy point of view. Conversely,
if only the dominant on-axis PW term is used to approximate the reector eld, the
error increases when the plane P becomes larger. This is due to the tapering of
17
−10
−20
Error in currents, dB
−30 Only
dominant
PW is used
−40
−50 Dipole array Reflector

Horn with ext. ground plane edge
−60 Horn
PWS
Horns Max. array or is used
−70
aperture ext. ground plane
size dimension
−80
1 2 3 5 7 10 20 30 40
Sample plane size, λ
Figure 2.6: The relative error in induced feed currents [cf. (2.11)] as a function of the FFT sampling
plane size P.
the reector scattered eld which becomes more pronounced when the plane size P
increases, so that the PW amplitude A(kx , ky ) is underestimated when using the eld
averaging in (2.9a) for kx = ky = 0, as opposed to the direct on-axis point sampling
method that has been presented in [44] and overviewed in Sec. 2.2.1.
2.2.3 Near-eld interpolation (NFI) technique

While the previous section describes how the PWS-expanded E -eld from the re-
ector accelerates the computation of the induced feed current, this section explains
how the reector incident H -eld can be computed for the rapid determination of
the induced PO current. For this purpose, the radiated H -eld from the feed is rst
computed at a coarse grid on the reector surface (white circles in Fig. 2.7), after
which the eld at each triangle is determined on the reector (yellow square mark-
ers) through an interpolation technique. This interpolation technique de-embeds the
initially sampled eld to a reference sphere with radius R whose origin coincides with
the phase center of the feed to assure that the phase of the de-embedded eld will be
slowly varying. Consequently, relatively few sampling points are required for the eld
interpolation, after which the interpolated elds are propagated back to the reector.
In summary, and with reference to Fig. 2.7, the H -eld interpolation algorithm
for determining the reector PO current
1. Denes a grid on the reector surface (white circles) for computing the H -eld.
2. De-embeds the H -eld to a reference sphere around the feed phase center (green
18
initial eld sampling points

feed phase center
de-embedded eld points
interpolation points
R nal eld testing points
∆θ
Hqsph
dm sph
Hm dq
Hm H i,r (rqr )
Figure 2.7: The near-eld interpolation technique for the rapid determination of the induced PO
current on the reector.
points):
sph = H d ejkdm ,
Hm (2.12)
m m
where dm is the distance between the reector surface and the sphere of radius
R along the line connecting the mth sample point on the reector and the feed
phase center.
3. Computes the elds on the sphere in the same directions as the reector triangle
centroids are observed (blue square markers) through interpolating the elds
at the adjacent (green) points.
4. Propagates the eld to the reector surface; that is, at the q th triangle, the
H -eld
H i,r (rqr ) = Hqsph d−1

q e
−jkdq
. (2.13)
5. Computes the reector PO current (see e.g. [71, p. 343])
The error in the reector current as a function of the sampling grid density is
depicted in Fig. 2.8. It shows that the error in the resulting induced reector current
depends on the angular step size ∆θ and ∆φ of the initial eld sampling grid (before
interpolation). As expected, the error increases when the sampling grid coarsens.
Furthermore, the error is larger for larger feeds, especially for high-scattering ones,
for which the scattered elds (i.e. 2nd iteration and further) vary more rapidly than
for smaller low-scattering antennas for which a coarser grid can be applied.
19
Horn with ext. ground plane Dipole array Horn
0 0
∆φ = 2.5 deg ∆θ = 2.5 deg
−20 −20
Error, [dB]
Error, [dB]
−40 −40
−60 −60
−80 −80
0 2 4 6 8 10 0 2 4 6 8 10
∆θ, [deg] ∆φ, [deg]
(a) (b)
Figure 2.8: The interpolation error in the 38λ reector current as a function of (a) the sampling
step ∆θ, and (b) the sampling step ∆φ of the near elds of the feed.
2.2.4 Analysis of PWE and NFI errors and simulation times

Table 2.3 shows how the simulation time of a plain iterative CBFM-PO (or MoM-
PO) approach reduces, and Table 2.4 summarizes the relative errors in both the cur-
rents and relevant antenna characteristics when the eld approximations of Sec. 2.2.1
are used. The errors have been computed according to (2.8) and (2.11). Note that
the PWS approximation leads to the small relative error of 0.28% in the surface cur-
rent of the high-scattering feed for the 38λ reector, while if only a single on-axis
PW is used, the relative error is found to be two orders larger (see Sec. 2.2.1). It is
also observed that, when applying the eld approximations for both the reector and
feed, the relative error in the considered antenna characteristics remains less than
1%, while the computational speed advantage is signicant (see Table 2.3), i.e., a
factor 5 to 100, depending on the reector size and feed complexity.
Table 2.3: Total simulation time (for D = 118λ reector)
Horn Horn with

ground plane Dipole array Vivaldi array
MoM-PO, no 3906 min
70 min (100%) 192 min (100%) 801 min (100%)
approx. (100%)
PWS approx. 27 min (39%) 63 min (33%) 190 min (24%) 1312 min (34%)
NFI approx. 57 min (81%) 152 min (79%) 548 min (68%) 2108 min (54%)
Both approx. 13 min (19%) 17 min (9%) 16 min (2%) 33 min (1%)
20
2.3. Experimental verification of the CBFM-PO approach with...
Table 2.4: Errors due to applying the eld approximations, %

Feed sur- Reector Gain Gain
face cur- surface (on-axis) (@−3 dB) Impedance
rent current
Reector 38λ 118λ 38λ 118λ 38λ 118λ 38λ 118λ 38λ 118λ

PWS approx. 0.09 0.02 0.11 0.03 0.09 0.03 0.07 0.02 0.16 0.04
NFI approx. 0.01 <0.01 0.06 0.06 0.05 0.04 0.01 0.02 0.01 <0.01
Both approx. 0.09 0.02 0.13 0.07 0.13 0.07 0.07 0.04 0.15 0.04

PWS approx. 0.28 0.02 0.41 0.02 0.06 0.01 0.09 0.01 0.44 0.04
NFI approx. 0.3 0.01 1.01 0.16 0.16 0.07 0.37 0.07 0.52 0.02
Both approx. 0.53 0.03 1.02 0.16 0.15 0.08 0.34 0.07 0.88 0.05

PWS approx. 0.05 0.02 0.1 0.02 0.03 0.01 0.01 0.01 0.03 0.01
NFI approx. 0.02 0.01 0.21 0.20 0.09 0.07 0.12 0.13 0.02 0.01
Both approx. 0.06 0.02 0.23 0.21 0.10 0.07 0.13 0.14 0.05 0.02
2.3 Experimental verication of the CBFM-PO ap-

proach with acceleration techniques
In addition to several cross-validations of the CBFM-PO approach using commercial
software (see Paper C), a practical antenna system has been modeled and the com-
puted illumination eciency ηill is compared to measurements. Fig. 2.9 shows ηill of
a 118λ reector antenna (D = 25 m, F/D = 0.35), either fed by the Vivaldi array
feed (APERTIF, [7]), or a single horn antenna. The numerically computed results are
compared to measurements carried out at the Westerbork Synthesis Radio Telescope
(WSRT) [7]. As one can see, the agreement is very good. The size of the simulated
ground plane has been chosen equal to the size of the feed cabin (≈ 1×1 m). The
fact that ηill is higher for the array feed than for the horn antenna nicely demon-
strates the superior focal eld sampling capabilities of dense PAFs. Furthermore, one
can also observe a rather strong ripple in ηill for the case of the horn feed with the
extended ground plane. This ripple is caused by the relatively high feed scattering
of the reector eld.
2.4 Numerical studies for dierent types of reec-

tor antenna feeds
In this section several feeds are considered as part of a reector antenna with aperture
diameter 38λ. Several antenna characteristics, such as the radiation pattern, the
21
Measured (Vivaldi array) Measured (Horn feed)

CBFM−PO (Vivaldi array) CBFM−PO (Horn feed)
80
75
Illumination efficiency, [%]
70
65
60
55
50
45
1.295 1.3 1.305 1.31 1.315 1.32
Frequency, [GHz]
Figure 2.9: Illumination eciencies of the 118λ reector antenna, either fed by the 121 Vivaldi PAF,
or the single-horn feed. The CBFM-PO simulated results are compared to the measured ones for
a 25 m reector antenna of the Westerbork Synthesis Radio Telescope [7]. Bottom of the gure: a
photo of the experimental PAF system placed at the focal region of the reector, and an image of
a smaller-scale PAF-reector model.
receiving sensitivity, and the aperture- and focal eld distributions are analyzed using
the CBFM-PO approach. First, we will show how the feed-reector coupling aects
the eld distribution in the aperture of the reector when fed by the horn with
extended ground plane or the dipole antenna array of the same size [see Fig. 2.3(b)
and (c)]. Afterwards, the model of the antenna system will be extended to include
the spillover and antenna-LNA noise mismatch characteristics, so that the receiving
sensitivity can be analyzed.
The aperture eld distributions at two frequency points corresponding to the

minimum and maximum aperture eciencies are shown in Fig. 2.10 and 2.11 for the
horn and the dipole array feeds, respectively. It is pointed out that the antenna
elements are loaded by a complex impedance, which is accounted for directly when
solving for the antenna feed currents through the CBFM. This is done through the
modication of the diagonal elements of the MoM matrix corresponding to the port
basis functions as described in [63, p. 223]. The impedance of the loads has been
chosen to maximize the array decoupling eciency [61], which yields the optimum
load impedance of 60.88 − 8.12j and 147.4 + 45.6j Ω for the horn and array case,
respectively. For the horn case this implies the ideal power-matched case.
As one can observe from the gures, the aperture eld at the 2nd iteration, i.e., due
to the scattered eld of the feed [Fig. 2.10(c) and Fig. 2.11(c)], is about 20 dB lower
for the array feed, thereby rendering the eld variation negligible. On the contrary,
for the horn feed, the peak and dip of the eld is clearly seen in the aperture center.
22
2.4. Numerical studies for different types of reflector antenna feeds
(a) Total aperture eld, (b) Total aperture eld, (c) Aperture eld at 2nd iter-
f @ min ηap f @ max ηap ation, f @ max ηap
Figure 2.10: The eld distribution in the aperture of a 38λ reector fed by the horn with extended
ground plane.
(a) Total aperture eld, (b) Total aperture eld, (c) Aperture eld at 2nd iter-
f @ min ηap f @ max ηap ation, f @ max ηap
Figure 2.11: The eld distribution in the aperture of a 38λ reector fed by the array of 121 half-
wavelength dipoles.
This leads to a signicant variation of the aperture eciency ηap over frequency, viz.
19.6% versus 0.6% for the array.
For some applications, such as for radio astronomy, a reector antenna works
purely in receiving mode, and other system characteristics, such as the system noise
temperature Tsys and the receiving sensitivity Ae /Tsys , become important. The main
contributors to Tsys that are dependent on multiscattering eects, are the spillover
noise temperature Tspill and the noise temperature due to the noise mismatch between
2
the antenna(s) and LNA(s) , Tcoup . In order to compute Tcoup , the equivalent one-
port system representation is used as described in [72]. By using this extension to the
CBFM-PO approach, the next step is to consider the two relatively small feeds shown
in Fig. 2.12. The antenna array ports are connected to Low Noise Ampliers (LNAs)
which are also part of the antenna-receiver model. Two beamforming scenarios for
the array are considered: (i) a singly-excited embedded element, and; (ii) a fully-
excited antenna array employing the Conjugate Field Matching (CFM) beamformer
for maximizing the gain of the secondary far-eld beam.
The computed aperture eciency ηap , system noise temperature Tsys , and the
2 in case of phased array feeds Tcoup also takes the excitation scheme and coupling between the
array elements into account.
23
(a) A single dipole above a PEC (b) A dual-polarized array of

ground plane 20 dipole antenna elements
Figure 2.12: The considered dipole antenna feeds. The dipole length is 0.47λ and the ground plane
size is 1.66λ × 1.33λ
2 10
1
Tsys variation, %
ηap variation, %
5
0
−1 0
One dipole
−2
Dip array, CFM
Dip array, one excited −5
−3 One dipole
Dip array, CFM
−4 −10 Dip array, one excited
1.39 1.4 1.41 1.42 1.43 1.44 1.45 1.39 1.4 1.41 1.42 1.43 1.44 1.45
Frequency, GHz Frequency, GHz
(a) (b)
5
Dip array, CFM
Sensitivity variation, %
Dip array, one excited

One dipole
−5
1.39 1.4 1.41 1.42 1.43 1.44 1.45
Frequency, GHz
(c)
Figure 2.13: The aperture eciency, the system noise temperature, and the resulting receiving
sensitivity of the reector antenna system as a function of frequency.
resulting receiving sensitivity Ae /Tsys are shown in Fig. 2.13. By analyzing these
gures, one can conclude that the aperture eciency varies with frequency much
more for the case of a single element due to a fact that a lot of energy scatters from
the ground plane behind the dipole.
The feed-reector interaction phenomenon leads not only to the variation in ηap ,
but also leads to a variation in Tsys . These variations are comparable for the the single
dipole and array feeds, and have a major impact on the sensitivity ripple. Although
Tsys is similar for both feeds, the mechanism of forming the ripple is dierent; when
the reector is fed by the feed shown in Fig. 2.12(a), the radiation pattern of the
feed is breathing over frequency, resulting in the variation of the spillover noise
24
2.5. Analysis of antenna arrays using Krylov subspace iteration with...
temperature Tspill , while for the feed in Fig. 2.12(b) the main contribution to the Tsys
variation is caused by the variation Tcoup . See Paper A for more details.
2.5 Analysis of antenna arrays using Krylov sub-

space iteration with domain decomposition ap-
proach
To calculate the feedreector interaction the described above method involves solv-
ing for the electric current on the feed, for which CBFM method was used. There
are other methods that are capable of handling electrically large antennas, such as
the multilevel fast multipole method (MLFMM), the Full Orthogonalization Method
(FOM), or the Generalized Minimal Residual Method (GMRES) [73]. The latter two
methods are iterative in nature and make use of Krylov subspace iteration to generate
so called generating vectors, which (after orthogonalization) can be considered as
macro-basis functions, the coecients of which are determined through the GMRES
approach.
In this section we propose a domain-decomposition approach to Krylov subspace
iteration, where MBFs (or CBFs) on each subdomain are naturally constructed from
the dierent segments of the generating vectors.
Formulation of the method

Consider the Method of Moments (MoM) matrix equation:
ZI = e, (2.14)
where Z is the N × N MoM matrix; e is the N × 1 excitation vector and I is a vector

containing the unknown expansion coecients for the elementary basis functions.
Accordingly, the reduced CBFM system of equations can be written as
Z̃Ĩ = ẽ, (2.15a)
Z̃i,j = KH
i Zi,j Kj , (2.15b)
ẽi = KH
i ei , (2.15c)
where i, j = 1 . . . M are sub-domain indices; M is the number of subdomains; H

is the Hermitian operator and Ki is the set of MBFs. The method proposed here
consists of selecting as MBFs on a given subdomain the corresponding segments of
the generating vectors. Those segments correspond to entries associated to basis
functions dened on the subdomain of interest. Hence, the newly proposed MBF
selection reads: h i
(1) (2) (P )
Ki = ki = ei | ki | . . . | ki , (2.16)
25
in which the generating vector k is formed iteratively as
k(p+1) = Zk(p) for p = 1 . . . P − 1, (2.17)
where index i refers to the MBF vector entries related to subdomain i.

It is important to point out that the most computationally expensive part of the
MoM matrix reduction (2.15b), namely the matrix-matrix product Zi,j Kj , can be
carried out during the subspace construction (2.17). For this purpose, (2.17) is built
2 2 (p)
from M smaller matrix-vector products resulting in the M vectors vi,j , expressed
as
(p) (p)
vi,j = Zi,j kj . (2.18)
Segment i of the vector k(p+1) (at the next iteration) is obtained by a simple summa-
(p)
tion of vectors vi,j as
(p+1) (p)
X
ki = vi,j . (2.19)
j
(p)
If the vectors vi,j are concatenated in a matrix Q as
h i
(1) (2) (p)
Qi,j = vi,j | vi,j | ... | vi,j , (2.20)
then the MoM matrix reduction (2.15b) can be rewritten as
Z̃i,j = KH H
i Zi,j Kj = Ki Qi,j , (2.21)
which allows one to reduce the time involved in (2.15)-(2.17) by almost a factor two ,
as compared to a straight-forward implementation. The appendix in Paper G explains
how (2.21) can be modied when the set of MBFs needs to be orthogonalized.
Fig. 2.14(a) visualizes the matrix Q, which consists of vectors v, and Fig. 2.14(b)
shows the matrix segment Qi,j .
CBFM with restarts

The accuracy of the CBFM can be signicantly improved down to machine preci-
sion by introducing a restart procedure similar to that used in a restarted GMRES
method [73]. The main dierence with GMRES is that the subspace is restarted on
every subdomain. The restart algorithm is presented in Sec.III of Paper G.
Numerical results
The proposed approach is compared to the GMRES algorithm in terms of the relative
error in the surface current (reference is exact MoM solution) versus the solving
complexity. The complexity is herein dened as the number of elementary operations
26
2.5. Analysis of antenna arrays using Krylov subspace iteration with...
VM(1) VM(2) (P) VM

(P)
(P) V2
V1
Q1,2
VM,1(P)
1
V1,1(P)
V1,1(1)
V1,1(2)
1
VM,2(P)
isub-domain
V1,2(1)
V1,2(2)
V1,2(P)
RWG
VM,M(P)
V1,M(P)
V1,M(1)
V1,M(2)
N
1 iteration P
(a) (b)
Figure 2.14: (a) Matrix Q and its vectors, and (b) its sub-matrices Qi,j .
ab+ (oating point product of complex scalar numbers and summation with another
complex number), required to solve the problem.
Fig. 2.15(a) and (b) show two similar antenna arrays consisting of Vivaldi ra-
diators, where in the rst case the elements are electrically interconnected, and in
second array there is a small air-gap between the elements. Dierent colors denote
dierent subdomains in which the arrays are divided. First, a reference surface cur-
rent distribution on each array was found using pure method-of-moments. Then, the
current was computed using GMRES and the proposed CBF approach. Finally, an
error between each of these and the reference solution was obtained, the result of
which is depicted in Fig. 2.15(c). The round markers denote restart positions. From
the gure one can see a clear advantage of the proposed CBFM approach as it takes
about 1.5 to 2.6 times less of elementary operations to reach same level of the error,
which was chosen to be −50 dB.
Of course, the benet of using this approach varies depending on the type of
structure under investigation. We have considered two more structures in Paper G:
i) a sphere at a resonant frequency, and; ii) a rectangular plate, both of which were
divided in subdomains of dierent sizes. It turned out that in the worst case (like
for the rectangular plate) the iterative CBF approach essentially provides the same
accuracy as GMRES, but for the best case (the sphere at resonant frequency), the
time to reach convergence for the CBFM can be more than 3 times shorter.
27
(a) connected array. (b) disconnected array.
GMRES: connected array (P=18)

0 CBFM: connected array (P=20)
Error in current, [dB]
GMRES: disconnected array (P=16)

CBFM: disconnected array (P=12)
-20
-40
-50
-60
-80
61.69 161.52
-100
0 20 40 60 80 100 120 140 160
Complexity, 10 9
(c) GMRES and CBFM convergence.
Figure 2.15: Numerical example: (a) a connected, and; (b) disconnected 121-element dual-polarized
Vivaldi array, divided into 121 subdomains and excited by a delta-gap voltage sources at each
antenna element. Subgure (c) compares the convergence rates of restarted GMRES and CBF
method. The restart positions are indicated by circles.
Discussion
When well-preconditioned, GMRES converges very rapidly (i.e. within a few tens
of iterations), almost irrespective of the number of unknowns. As explained in [74],
GMRES amounts to solving a reduced system of equations, whose size (i.e. number
of degrees of freedom, DoFs), corresponds to the number of iterations. For large
problems, this solution takes a negligible time as compared to that involved in the
mat-vec operations. This means that, without signicant increase in the computation
time, one can aord more DoFs, as is the case with the approach proposed here, since
the number of DoFs now corresponds to the number of mat-vecs P multiplied by the
number of subdomains. Without any specic matrix-vector multiplication, solving
3
the reduced system of equations has a complexity (P M ) (here it is worthwhile
mentioning that there exist methods to reduce this exponent, see e.g. [75]), while the
2
complexity of the mat-vecs is P N . The increase in computational time is therefore
2 2 2
small as long as P M Nsd , where Nsd is the average number of elementary basis
functions per subdomain.
28
2.6. Conclusions
2.6 Conclusions
To conclude the research that has been presented in this chapter, we highlight the
following observations:
The feed-reector interaction (standing wave) eects give rise to oscillations in

the system characteristics with frequency ∆f = c/(2F ), where c is speed of
light and F is the reector focal distance. This results in the heart beating
eect the change of the beamwidth and gain, as well as Tsys variation over
frequency.
An FFT-enhanced Plane Wave Spectrum (PWS) approach has been formu-

lated in conjunction with the Characteristic Basis Function Method, a Jacobi
iterative multiscattering approach, and a near-eld interpolation technique for
the fast and accurate analysis of electrically large array feed reector systems.
Numerical validation (presented in Paper C) has been carried out using the
multilevel fast multipole algorithm method available in the commercial FEKO
software.
The scattering from the feed is minimal for power-matched antenna loads (more
critical for PAFs) and when size of the surrounding metal structure is minimized
(more critical for single-port feeds, especially in MFFEs).
The electromagnetic coupling between the reector antenna and the dipole
PAFs under study have a minor impact on the antenna beam shape and aperture
eciency, as opposed to that of a single dipole feed. The nite ground plane
behind the single dipole, which is part of the feed supporting structure and is
often much larger than one antenna element, but comparable to the size of a
PAF, is a reason for this dierence.
The (active) impedance matching of the strongly-coupled PAF elements appears

to be more sensitive to the feed-reector interaction eects, as a result of which
the receiver noise temperature increases.
The sensitivity variation is mainly driven by the variation in the system noise
temperature, of which the main contribution is due to the noise mismatch of
the considered PAF array elements with LNAs. Therefore, in order to reduce
the sensitivity ripple of reector antennas with PAFs, major attention should
be paid to the noise matching and its stability over time in the presence of a
reector when designing a PAF system.
The conclusion in [48] states that the Radar Cross-Section (RCS) of the feed is
the determining factor in magnitude of the standing wave eect. This is true
only for the aperture eciency variation, but it does not apply to the noise
29
characteristics (Tspill , Tcoup ). Other factors showing why the RCS is not a good
gure of merit to quantify the standing wave eect in receiving systems are
that the the RCS does not account for the relative size of the feed w.r.t. the
reector, and that it assumes a uniform PW eld radiated by the reector.
A domain-decomposition technique into Krylov subspace iteration has been

developed. This method is similar to the CBFM, here the MBFs are generated
by simple segmentation of the pre-computed vectors of the Krylov subspace.
The achieved convergence is faster than with GMRES by a factor ranging from
1.05 (the rectangular plate with large subdomains) to 2.6 (the connected Vivaldi
array) while keeping the same accuracy.
30
Chapter
3 Optimum Beamforming
Strategies for Earth Obser-
vations
It has been argued in Sec. 1.2 that push-broom congurations for satellite radiometers
are advantageous for Earth observation systems when equipped by PAFs. Therefore,
the goals of the work presented in this chapter are to determine: (i) to what extent
the performance-limiting factors of push-broom radiometers can be reduced by using
dense PAFs employing advanced beamforming schemes; (ii) the minimum required
complexity of the PAF design (size, number of elements and their arrangement in the
feed as well as the number of active receiver channels), and; (iii) what beamforming
strategy to use for meeting the instrument specications for future radiometers [10].
Finally, a trade-o analysis will be performed in order to choose the radiating antenna
element for the PAF.
3.1 Performance requirements

Before describing the push-broom array design and beamforming scenarios, the re-
quirements for such radiometers are described rst and how they are related to the
antenna system requirements.
In February 2013 the ESA contract 4000107369-12-NL-MH was awarded to the
team consisting of TICRA, DTU-Space, HPS, and Chalmers University. The rst
workpackage of the contract involved the review of ocean sensing performance pa-
rameters, which in turn resulted in the requirements for future satellite radiometers as
shown in Table 3.1 [10, 76]. Derivation of these requirements is presented in Paper K.
The table indicates that the radiometer should operate at three narrow frequency
bands: C-band (6.9 GHz), X-band (10.65 GHz) and Ku-band (18.7 GHz). The
instrument must be dual-polarized and have a receiving sensitivity in the 0.22 −
0.3 K range. The overall error of the sea temperature measurement should not
exceed 0.25 K. The maximum allowed footprint size is 20 km for C- and X-band,
and 10 km for the Ku-band. Under footprint we understand the region of the sea
31
Chapter 3. Optimum Beamforming Strategies for Earth Observations
Table 3.1: Radiometer requirements

Freq., Bandwidth, Polari- Sensiti- Accuracy Resolution Dist. to land
[GHz] [MHz] zation vity, [K] ∆T , [K] FP, [km] DL , [km]
6.9 300 V, H 0.30 0.25 20 5-15
10.65 100 V, H 0.22 0.25 20 5-15
18.7 200 V, H 0.25 0.25 10 5-15
that is illuminated by the antenna beam from −3 to 0 dB level with respect to the
beam maximum. Additionally, the instrument should satisfy the above-described
requirements even when the observation is as close as 15 km from the coast line. The
latter requirement is called distance to coast and explained with the aid of Fig. 3.1.
The brightness temperature of the land surface is assumed to be TL = 250 K.
Assume next that we wish to measure the sea at horizontal polarization for which
the brightness temperature is around TH = 75 K (the brightness temperature of
the vertical polarization is higher, i.e. 150 K, and therefore it is less aected by
the erroneous power signal from land). It can be shown that the requirement for
the maximum error ∆T = 0.25 K can be satised only if the power of the beam
in the cone with half-angle θc is 99.72 % of the total power incident on the Earth's
surface [35]. This determines the distance to coast Dc , which is dened as the angular
dierence θc − θ3dB projected on the Earth surface, i.e.,
Dc = Y sin θc − Y sin θ3dB ≈ (θc − θ3dB )Y, (3.1)
where Y is the distance from the satellite to the observation point on the Earth (see
Fig. 3.2 for the satellite orbit parameters). Therefore, to nd the distance-to-coast
Footprint
Rotation
Sea
%iofitheib -3 d B
99.72 ittingi eami
axis
Thi=i75iK theiEa
ow e rih rthi
FPS
p Velocityv
vector
- 3i d B
dB
i FPL =
Beami θ 3 αv θE
45. =v
m
vk
θci 07°
Rv=v6378vkm Hv=v817vkm
peak
43
vv
12
=v
direction
νv=v
Yv
Nadirv
53°
nce
Dista D c
st
to a
c o
X = 880 km
face
h sur
Land Ea r t
il ine
Coast Tlandi=i250iK
Figure 3.2: Geometrical parameters of the ra-
Figure 3.1: Denition of the Distance to coast diometer antenna system located on the Earth
radiometer requirement. A typical radiation orbit. The notation of the major and minor axes
pattern of the torus reector antenna is shown of the footprint ellipse is shown in the top-left
as background. insertion.
32
3.2. Reflector antenna design
characteristic, the angles θc and θ3dB are found rst from the antenna compound
beam and Eq. (3.1) is used afterwards.
Since the radiometer must be able to measure the brightness temperature of both
polarizations separately, an error is introduced due to the received power of the cross-
polarized component of the incident eld. It is shown in [35] that this power must
not exceed 0.34 % of the co-polarized power, in order to satisfy the maximum error
requirement ∆T = 0.25 K. Since the brightness temperature of the sky is very low
and the amount of power radiated towards the sky is small, it suces to compute the
antenna total radiation pattern only at the angular range subtended by the Earth
(θ = 0 . . . θE from the Nadir direction).
Another requirement for the radiometer is the sampling resolution, which sets
requirements on the maximum size of the footprint (FP). The footprint will have an
elliptical shape due to oblique incidence of the radiated eld on the Earth's surface
as shown in the top-left insertion of Fig. 3.2. The longitudinal and transverse to the
movement direction axes of the ellipse are denoted as FPL and FPS, correspondingly.
The footprint size, FP, is determined as the average of FPS and FPL:
FPS + FPL
FP = , (3.2)
2
where FPS is related to the half-power beamwidth (HPBW) as
FPS = Y × HPBWtransv , (3.3)
and FPL is
Y × HPBWlong
FPL = , (3.4)
cos ν
where HPBWtransv and HPBWlong are the longitudinal and transverse beamwidths
to the movement vector directions; and ν is the incidence angle.
Another characteristic of the radiometer radiation pattern is the beam eciency,
which is usually dened as the relative power within the main beam down to the
−20 dB contour level. A high beam eciency is generally synonymous with a good
quality antenna. However, a low beam eciency antenna may not necessarily rep-
resent a bad antenna. For example, for the radiometer, the feed spillover past the
reector edge reduces the beam eciency, but it illuminates the cold sky and does
no harm; it is the radiation towards the Earth that makes a signicant impact, and
must therefore be taken in account.
3.2 Reector antenna design

The design procedure of the push-broom reector is described in [37] and has been
developed by TICRA. In short, and with the reference to Fig. 3.3, the surface of the
reector (blue dots) is created by rotation of the parabolic prole (black dots), dened
33
Axis of rotation
F
Fo Pa
ca l rab
line ( ola
ar c) Z CS
X ic
ol Y
r ab ile
Pa rof
p
Figure 3.3: Design procedure of a parabolic torus reector: the parabolic prole (black circles at the
bottom), dened in the coordinate system Parabola CS and with focal point F, is rotated around
the green axis of rotation which itself is tilted with respect to the parabola axis. This transforms
the prole focal point F to the focal line (arc) along which a PAF will be positioned.
in the coordinate system Parabola CS and with focal point F, around the green
axis of rotation which is tilted with respect to the parabola axis. The reector rim
(edge of the red area) is chosen based on the requirements on the projected aperture
area and maximum scan angle. The latter parameter also denes the size of the PAF
along the focal arc, which is created by rotating the focal point F around the axis of
rotation.
Due to the rotational symmetry of the reector, it is natural to locate the array
antenna elements in a polar grid with the origin located at the point where the axis
of rotation intersects the plane of the focal arc. The layout of such an array is shown
in Fig. 3.4. The reector focal arc is denoted by the black curve to show the position
of the array relative to the reector.
3.3 Optimum PAF beamformers

To outline the optimization procedure for the PAF beamformers considered in this
work, we utilize the generalized system representation as shown in Fig. 3.5 for N
actively beamformed PAF antennas [17]. The PAF system is subdivided into two
blocks: (i) the frontend including the reector, array feed and Low Noise Ampliers
∗ N
(LNAs), and; (ii) the beamformer with complex conjugated weights {wn }n=1 and an
ideal (noiseless/reectionless) power combiner realized in software.
Here, w
H
= [w1∗ , . . . , wN
∗
] is the beamformer weight vector, H is the Hermitian
conjugate-transpose, and the asterisk denotes the complex conjugate. Furthermore,
34
3.3. Optimum PAF beamformers
Figure 3.4: Preliminary layout of the PAF for the push-broom reector. The black arc shows the
position of the focal arc of the reector.
a = [a1 , . . . , aN ]T is the vector holding the transmission-line voltage-wave amplitudes

at the beamformer input (the N LNA outputs). Hence, the ctitious beamformer
output voltage v (across Z0 ) can be written as v = wH a, and the receiver output
2
power as |v| = vv ∗ = (wH a)(wH a)∗ = (wH a)(aT w∗ )∗ = wH aaH w, where the propor-
Figure 3.5: Generalized representation of the PAF reector antenna system.
35
tionality constant has been dropped as this is customary in array signal processing
and because we will consider only ratio of powers.
Although each subsystem can be rather complex and contain multiple internal sig-
nal/noise sources, it is characterized externally (at its accessible ports) by a scattering
matrix in conjunction with a noise- and signal-wave correlation matrix. Accordingly,
the signal-to-noise ratio (SNR) can be expressed as
wH Pw
SNR = , (3.5)
wH Cw
that is, the SNR function is dened as a ratio of quadratic forms, where P = eeH is
the signal-wave correlation matrix, which is a one-rank positive semi-denite matrix
T
for a single point source; the vector e = [e1 , . . . , eN ] holds the signal-wave amplitudes
at the receiver outputs and arises due to an externally applied electromagnetic plane
wave Ei ; and C is a Hermitian spectral noise-wave correlation matrix holding the cor-
∗ ∗
relation coecients between the array receiver channels, i.e., Cmk = E{cm ck } = cm ck .
Here, cm is the complex-valued voltage amplitude of the noise wave emanating from
channel m, which includes the external and internal noise contributions inside the
frontend block in Fig. 3.5, and the overbar denotes time average. We consider only a
narrow frequency band, and assume that the statistical noise sources are (wide-sense)
stationary random processes which exhibit ergodicity, so that the statistical expec-
tation can be replaced by a time average (as also exploited in hardware correlators).
Below, we will rst discuss two standard signal processing algorithms, which are
then used as the starting point to develop the two customized push-broom radiometer
beamformers.
3.3.1 Standard maximum signal-to-noise ratio beamformer

MaxSNR
The well-known closed-form solution that maximizes (3.5) for the point source case,
where P is of rank 1, is given by [77]
wMaxSNR = C−1 e, with SNR = eH wMaxSNR , (3.6)
where the principal eigenvector e corresponds to the largest eigenvalue of P. If we

assume a noiseless antenna system, the matrix C will contain the noise correlation
coecients only due to external noise sources (received noise), and its elements can
be calculated through the pattern-overlap integrals between fn (Ω) and fm (Ω), which
are the nth and mth embedded element pattern of the array, respectively [17], i.e.,
Z
Cmn = Text (Ω)[fm (Ω) · fn∗ (Ω)] dΩ, (3.7)
36
where Text (Ω) is the brightness temperature distribution of the environment. The
proportionality constants between the right-hand side and the noise waves on the
left-hand side are omitted.
3.3.2 Standard Conjugate Field Matching beamformer CFM

The CFM beamformer maximizes the received signal power at its output, i.e.,
wH Pw

max , (3.8)
w wH w
which is also equivalent to maximizing the directivity. The trivial solution to that is
H
(provided that P = ee )
wCFM = e. (3.9)
However, since this beamformer assumes a noiseless system and a uniform incident
plane wave from the direction of observation, it will provide a sub-optimal solution
for practical systems. In particular, it allows no control on the radiation outside of
the main beam.
To overcome this issue, a slightly dierent CFM approach is used, in which we let
a tapered plane wave with a Gaussian amplitude distribution be incident from the
observation direction, calculate the focal eld at location of the PAF elements and
use the conjugate values of this eld as excitation coecients. Thus, the excitation
coecient are calculated as

wCo = ECo (xCo , yCo )
∗
CFM2 foc i∗ (3.10)
wXp = EXp
h
(x Xp , yXp )
CFM2 foc
where wCo
CFM2 and wXp
CFM2 are vectors holding the excitation coecients of co- and
cross-polarized components, correspondingly; x{Co,Xp} and y{Co,Xp} are vectors with
x- and y -coordinates of the co- and cross-polarized PAF elements in the focal plane;
iT
Xp Xp Xp
ECo
Co h
Co
T
and foc = E foc,1 , . . . Efoc,N and E foc = E foc,1 , . . . Efoc,N are vectors holding the
focal eld components at position of each PAF element and corresponding to their
polarization.
In contrast to (3.9), the CFM approach in (3.10) assumes that the PAF elements
are isotropic within the reector subtended angle (there are no array EEPs incor-
porated in this beamformer, and the array radiators are just point linear/circular
polarized sources), but this assumption is still good because a typical small radiator
used in dense PAFs has a wide beamwidth and nearly spherical phase front of the
◦
radiated eld within the reector subtended angle (±24 in our case).
This CFM formulation allows us to choose the taper of the incident plane wave in
order to control such beam parameters as side-lobes (related to the distance-to-coast
radiometer characteristic) and beamwidth (related to the footprint size).
37
This beamforming approach has been used by our co-authors in Paper D, where
the optimal taper value has been determined through a study how it aects the
radiometer performance. Although the radiometer characteristics will then satisfy
the system performance specications, the PAF requires us to employ too many
antenna elements (almost a factor 2 more as compared to the customized beamformers
presented in the following subsections 3.3.3 and 3.3.4), which is not feasible for a
realistic satellite system due to an excessive power consumption.
3.3.3 Maximum Sensitivity, Minimum Distance to Land beam-

former MSMDL
A major drawback of the MaxSNR and CFM beamformers is that these maximize
the sensitivity/directivity without constraints imposed on the side-lobes and cross-
polarization levels, or have a very limited control over them. This means that the
required values of the distance-to-coast and the maximum allowable cross-polarization
power cannot be guaranteed, especially not for a non-parabolic surface of the reector.
To overcome this limitation one could consider the MaxSNR beamformer [Eq. (3.6)]
which is used to maximize the beam eciency (dened at the −20 dB level), while
minimizing the power received from other directions.
For this purpose, the function Text (Ω) in Eq. (3.7) is chosen such that it has low
Figure 3.6: The Text (Ω) mask-constraint functions dened for the calculation of the antenna noise
correlation matrices C1 due to the noise sources within the Earth angular region (see the inset in
the left upper corner) and C2 due to the noise sources in the sky region (see the inset in the right
upper corner). The toroidal reector fed with a PAF is shown in the middle of the illustration,
where the multiple secondary beams point to the Earth.
38
temperature values in the region of the expected main lobe (down to −20 dB level)
and high values outside of this region. In this way, we maximize the beam eciency
dened at the −20 dB level while minimizing the side-lobe and cross-polarization
powers outside of this region, as required for the radiometers.
The top-left insertion in Fig. 3.6 shows the function Text (Ω), where the cold
region around the main lobe has an elliptical shape with major a and minor b semi-
axes. Later in the results section we will refer to this ellipse as mask ellipse, because
it is a stepped function.
The temperature of the hot region is chosen to be as high as 1000 K to strongly
suppress the side lobes in order to satisfy the distance-to-coast requirement. This
value can also be a beamformer control parameter for optimization radiometer char-
acteristics.
In order to use the beamformer to realize scanned beams (pink rays in Fig. 3.6),
the noise temperature distribution function Text (Ω) can be assumed the same for each
of them, but the matrix C needs to be recomputed.
Acceleration of the computations

When constructing the matrix C, one should realize that its lling can be an extremely
time-consuming procedure as it requires computation of all secondary EEPs over the
entire sphere and evaluation of the pattern overlap integral (3.7) for all combinations
of EEPs (see Tabl. 3.2). In order to speed-up the computational process, we have
therefore represented the matrix C as a sum of two contributions, matrices C1 and
C2 that can be calculated relatively quickly. The rst matrix is obtained by using the
secondary EEPs computed in a limited angular range around the main lobe region,
while the second matrix is used for correcting for the spillover eects and is evaluated
through the primary feed patterns. A similar approach has been recently published
in [78] and used to calculate the antenna noise temperature of a Gregorian antenna
system. The brightness temperature distribution functions Text (Ω) corresponding to
C1 and C2 are illustrated in the top-left and top-right insets of Fig. 3.6, respectively.
Table 3.2 cross-compares the computation time at Ku-band (18.7 GHz) that is
needed for the simulations (using GRASP10, [79]) of the secondary patterns over
the entire sphere (when computing the matrix C through the brute-force approach)
and over the reduced region with the post-correction for the spillover eect (when
computing the matrices C1 and C2 through the proposed approach). There is an
obvious advantage in using the latter approach, especially for systems that employ a
large number of beams and operating at high frequencies.
Constraints on the dynamic range of the amplitude weights

Realistic (digital) receivers have a limit on the dynamic range of the weight coecients
due to limited resolution of analog-to-digital converters (ADC) and temporal stability
39
Table 3.2: Computational time of the matrix C at Ku-band (18.7 GHz)

Brute-force approach Proposed approach
Computing 156 Computing 156
Computing C1 and
secondary EEPs Computing C secondary EEPs
C2
(full sphere) (small angular range)
∼9 months no data 3 hours 5 min/beam
of receiver channels between calibrations.

In order to account for such realistic receivers, the MSMDL beamformer, as de-
scribed above, has been further extended so as to include constraints on the dynamic
range of the weights. This beamforming algorithm is implemented through an iter-
ative procedure that modies the reference weighting coecients (as determined by
the MSMDL beamformer), while trying to maintain the shape of the PAF pattern as
close as possible to the reference one. This will ensure that the radiometer parame-
ters are as close as possible to those obtained with the reference set of weights. The
corresponding algorithm is listed as follows:
[w(1) ]H Pw(1)
At the rst iteration (q = 1) the sensitivity function
[w(1) ]H C(1) w(1)
is maximized
(1) (1)
to determine the reference weight vector w . The matrix C is computed as
described above for the standard MaxSNR beamformer (with no constraints on
the dynamic range of the weight amplitudes).
[w(q) ]H Pw(q)
At iteration q = 2, 3 . . . the sensitivity function
[w(q) ]H C(q) w(q)
is maximized to
determine the new weight vector w(q) , where P is the signal covariance matrix
(q)
(computed only once, for the 1st iteration), C is the noise covariance matrix,
whose diagonal elements are a function of the weight vector w(q−1) obtained
after the previous iteration, i.e.,
 (q−1) (q−1) (q−1) (q−1) 

C11 f (|w1 |) C12 ··· C1N
(q−1) (q−1) (q−1) (q−1)
 C21 C22 f (|w2 |) · · · C2N 
C(q) (w(q−1) ) =  ,
 
. . .. .
. . . .
 . . . 
(q−1) (q−1) (q−1) (q−1)
CN 1 CN 2 · · · CN N f (|wN |)
(3.11)
where f is a receiver function that needs to be provided as an input to the
algorithm; it should have such a behavior that the lower the weight of the array
antenna element, the higher the function value is (which physically corresponds
to an increase in the noise temperature of the corresponding receiver channel).
In the numerical examples presented hereafter, a lter function is used whose
values are close to zero when the weights magnitude |wi | are higher than wconstr ,
and which has a sharp linear increase near wconstr . In this way f is similar
to inverse step function near wconstr (Fig. 3.7). Here wconstr is the value of
40
the amplitude weight constraint, which is typically in the order of −30 dB to

−40 dB.
1.5
f ( |w| )
1
wconstr
0.5
−60 −50 −40 −30 −20 −10 0
|w|, dB
Figure 3.7: The function f used in the numerical examples presented hereafter.
Check whether the magnitude of all weights are either higher than wconstr or
negligibly low (i.e. −80 dB in this work). If this condition is satised, the
iterative procedure is terminated. The channels with negligible weights are
switched-o, while the resulting set of weight coecients is considered to be
the nal one.
3.3.4 Advanced Maximum Beam Eciency beamformeR

AMBER
In order to reduce the distance to coast Dc , we maximize the power contained in the
main beam Pmb over the solid angle Ωmb divided by the total radiated power Ptot .
That is, our objective is to nd the weights w such that max (Pmb /Ptot ).
w
The power in the main beam of the total radiated eld Etot (θ, φ) is calculated as
ZZ Z Z XN
2

Pmb = |Etot (θ, φ)|2 dΩmb = wi Ei (θ, φ) dΩmb , (3.12)

Ωmb Ωmb i=1
where Ei (θ, φ) is i-th secondary (after reection from the dish) embedded element
pattern (EEP). The proportionality constant 1/(2η) in front of the integral is omitted.
It can be shown that the integrand is a quadratic form which we write in a matrix
form:  
ZZ ZZ
Pmb = wH Xw dΩmb = wH  X dΩmb  w = wH Aw, (3.13)
Ωmb Ωmb
where X is a Hermitian matrix, which is function of the direction, i.e.,

 
E1 (θ, φ) · E1∗ (θ, φ) . . . E1 (θ, φ) · EN
∗
(θ, φ)
. .. .
X(θ, φ) =  . . , (3.14)
 
. . .
∗ ∗
EN (θ, φ) · E1 (θ, φ) . . . EN (θ, φ) · EN (θ, φ)
41
and, therefore, the elements of the matrix A are calculated as so-called pattern overlap
integrals, i.e.,
ZZ ZZ
Ej∗ dΩmb EiCO Ej∗CO + EiXP Ej∗XP

Aij = Ei · = dΩmb , (3.15)
Ωmb Ωmb
where each EEP Ei is decomposed in co- and cross-polarized components EiCO and
EiXP , respectively.
We aim at maximizing the beam eciency, as well as to minimize the power in
cross-polarized eld component. Therefore, only the power in the co-polarized eld
∗
component should be maximized. This leads to removal of the EiXP Ej term from
XP
(3.15) in the optimization process, i.e. the elements of the matrix A are computed
now as ZZ
Aij = EiCO Ej∗CO dΩmb , (3.16)
Ωmb
Following the same procedure for calculating Pmb , we can calculate the total
radiated power Ptot as
Ptot = wH Bw, (3.17)
where the elements of the matrix B are calculated similar to (3.15), but with inte-
gration over full sphere, i.e.,
ZZ
EiCO Ej∗CO + EiXP Ej∗XP

Bij = dΩ. (3.18)
4π
For a given solid angle Ωmb it is thus desired to nd the weight coecients w that
maximizes the following ratio of quadratic forms:
Pmb wH Aw
= H , (3.19)
Ptot w Bw
It can be shown that the maximum value of this ratio is the maximum eigenvalue
λ of the expression
Aw = λBw, (3.20)
and that the vector holding the optimal complex-valued excitation coecients is given
by the corresponding eigenvector. In MATLAB [80] this can be coded as
[W , D] = eig (A ,B); % generalized eigenvalue decomposition ; matrix W holds
eigenvectors
Lam = diag (D); % extract the vector holding the eigenvalues
w = conj (W (: , Lam == max ( Lam )) ); % take eigenvector corresponding to the maximum
eigenvalue
Note that the AMBER beamformer is parametric, i.e. the solid angle Ωmb must be
dened before computing the weight coecients. Similar to the MSMDL beamformer,
Ωmb is dened by the major semi-axis and axis ratio of an ellipse centered around
the expected main lobe. We will use them as the beamformer parameters when
calculating the radiometer characteristics presented in the following sections.
42
250
200
0
150
|Gco |, |Gxp |, [dB]

100
−10
50
Y, [mm]
-50
−20
-100
-150
-200
−30
-250 −50 0 50
-100 0 100
X, [mm] θ, [deg]
(a) (b)
Figure 3.8: (a) The layout of the optimal (for a one beam) 6 × 13 array feed of the torus reector
antenna when the radiating element is a crossed half-wavelength dipole antenna, and (b) cuts of the
array embedded element patterns. The cuts are shown of the co-polar patterns for E- and H-planes,
as well as cross-polar pattern for diagonal plane (φ = 45◦ ), and the green area denotes the reector
subtended angle.
3.3.5 Comparison of the beamformers

In order to compare the beamformers, we analyze the push-broom radiometer equipped
with the optimal PAF of crossed dipole elements (for a single beam). Details on the
PAF optimization procedure will be discussed in the following sections, and for now
the array is assumed to be given. The layout of such array and embedded element
patterns of its elements are shown in Fig. 3.8.
The array was analyzed using the HFSS software [81] and the obtained primary
EEPs were imported into GRASP software [79] to calculate the secondary EEPs (after
reection from the reector), which in turn were used to perform the beamforming.
After determining the excitation coecients, the compound beam was computed and
the radiometer characteristics as discussed in Sec. 3.1 were calculated.
As described in Sec. 3.3.2, the CFM beamformer has one parameter the taper of
the incident plane wave (PW). Fig. 3.9 shows the calculated focal eld distribution,
the corresponding weight coecients and the radiation patterns of the primary and
secondary compound beams when the incident PW taper is equal to −50 dB, −30 dB
and −10 dB. As one can see from the gure, the focal eld becomes less blurry and
its side lobes increase as the PW taper decreases (the PW becomes more uniform).
This results in better reector illumination eciency (and consequently increased
directivity), narrower beamwidth (≡ footprint size), but at the same time the side lobe
level of the secondary beam increases, which degrades distance-to-coast characteristic.
This can be seen more clearly in Fig. 3.10 (top row), where the main radiometer
43
PW taper = −50 dB PW taper = −30 dB PW taper = −10 dB
Optimum Beamforming Strategies for Earth Observations
Co-polarized Cross-polarized Co-polarized Cross-polarized Co-polarized Cross-polarized

0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40
0.1
0.1
0.1
0.1
0.1
0.1
weight coecients
Focal eld and
X, [m]
X, [m]
X, [m]
X, [m]
X, [m]
X, [m]
0
0
−0.1
−0.1
−0.1
−0.1
−0.1
−0.1
0.2 0.1 0 −0.1 −0.2 0.2 0.1 0 −0.1 −0.2 0.2 0.1 0 −0.1 −0.2 0.2 0.1 0 −0.1 −0.2 0.2 0.1 0 −0.1 −0.2 0.2 0.1 0 −0.1 −0.2
Y, [m] Y, [m] Y, [m] Y, [m] Y, [m] Y, [m]
0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40
-29.1 -25.1 -21.3 -22.1 -17.3 -9.3 -7.0 -9.3 -17.3 -22.1 -21.3 -25.1 -29.1 -34.8 -35.3 -35.3 -34.8 -23.8 -21.5 -18.6 -14.7 -18.4 -12.9 -8.4 -12.9 -18.4 -14.7 -18.6 -21.5 -23.8 -39.6 -34.8 -32.0 -32.0 -34.8 -39.6 -20.7 -15.6 -17.9 -12.5 -9.5 -24.1 -8.7 -24.1 -9.5 -12.5 -17.9 -15.6 -20.7 -35.1 -31.1 -36.6 -26.7 -26.7 -36.6 -31.1 -35.1
-28.1 -23.4 -21.5 -18.9 -9.5 -4.1 -2.4 -4.1 -9.5 -18.9 -21.5 -23.4 -28.1 -35.8 -31.3 -33.0 -33.0 -31.4 -35.8 -25.0 -21.2 -16.3 -16.0 -15.1 -5.6 -2.9 -5.6 -15.1 -16.0 -16.3 -21.2 -25.0 -30.2 -30.0 -30.0 -30.2 -20.1 -20.2 -15.2 -9.1 -15.4 -8.9 -3.5 -8.9 -15.4 -9.1 -15.2 -20.3 -20.1 -33.3 -37.8 -28.0 -26.0 -26.0 -28.0 -37.8 -33.3
-27.9 -24.3 -20.9 -12.2 -5.2 -1.3 0.0 -1.3 -5.2 -12.2 -20.9 -24.3 -27.9 -39.8 -32.9 -30.2 -32.7 -32.7 -30.2 -32.9 -39.8 -26.0 -20.9 -18.7 -17.0 -7.3 -1.7 0.0 -1.7 -7.3 -17.0 -18.8 -20.9 -26.0 -33.7 -28.7 -30.3 -30.3 -28.7 -33.7 -24.7 -19.2 -14.1 -17.8 -11.2 -2.2 0.0 -2.2 -11.2 -17.8 -14.2 -19.2 -24.7 -33.2 -26.0 -27.7 -27.7 -26.0 -33.3
-29.7 -24.7 -16.9 -9.5 -4.2 -1.1 -0.1 -1.1 -4.2 -9.5 -16.9 -24.8 -29.7 -37.8 -33.0 -31.5 -34.6 -34.6 -31.5 -33.0 -37.8 -27.9 -24.6 -20.5 -11.9 -5.3 -1.6 -0.4 -1.6 -5.3 -11.9 -20.5 -24.6 -28.0 -33.2 -30.7 -33.4 -33.4 -30.7 -33.2 -25.7 -22.1 -25.1 -14.6 -5.9 -2.1 -1.2 -2.1 -5.9 -14.6 -25.1 -22.1 -25.7 -39.6 -31.8 -28.8 -33.1 -33.1 -28.8 -31.8 -39.6
-31.1 -23.7 -16.3 -10.3 -6.1 -3.6 -2.7 -3.6 -6.1 -10.3 -16.3 -23.7 -31.1 -39.4 -35.9 -35.2 -38.8 -38.8 -35.2 -35.9 -39.4 -32.5 -26.8 -19.2 -12.6 -8.1 -5.5 -4.7 -5.5 -8.1 -12.6 -19.2 -26.8 -32.5 -37.3 -36.4 -36.4 -37.3 -30.8 -29.2 -20.6 -14.1 -10.8 -9.6 -9.7 -9.6 -10.8 -14.1 -20.6 -29.2 -30.8 -37.2 -36.3 -36.3 -37.2
-32.7 -25.5 -19.2 -14.3 -10.9 -8.8 -8.1 -8.8 -10.9 -14.3 -19.2 -25.5 -32.7 -36.4 -29.6 -23.6 -19.1 -16.0 -14.2 -13.6 -14.2 -16.0 -19.1 -23.6 -29.6 -36.4 -35.3 -30.1 -28.8 -29.4 -35.0 -30.3 -29.3 -30.4 -35.1 -29.4 -28.8 -30.2 -35.3
Spillover efficiency = 99.03 % Relative cross-pol. level = -30.92 dB Spillover efficiency = 99.12 % Relative cross-pol. level = -29.1 dB Spillover efficiency = 96.56 % Relative cross-pol. level = -21.16 dB
0 0 0 0 0 0
105 90 75 105 90 75 105 90 75 105 90 75 105 90 75 105 90 75
120 40 60 120 60 120 40 60 120 40 60 120 40 60 120 60
40 40
135 45 135 45 135 45 135 45 135 45 135 45
patterns
Primary
30 −10 30 −10 30 −10 30 −10 30 −10 30 −10

150 30 150 30 150 30 150 30 150 30 150 30
20 20 20 20 20 20
165 10 15 165 10 15 165 10 15 165 10 15 165 10 15 165 10 15
180 0 0 −20 180 0 0 −20 180 0 0 −20 180 0 0 −20 180 0 0 −20 180 0 0 −20
195 345 195 345 195 345 195 345 195 345 195 345
210 330 210 330 −30 210 330 210 330 −30 210 330 210 330 −30
−30 −30 −30
225 315 225 315 225 315 225 315 225 315 225 315
240 300 240 300 240 300 240 300 240 300 240 300
255 270 285 255 270 285 255 270 285 255 270 285 255 270 285 255 270 285
−40 −40 −40 −40 −40 −40
Directivity = 46.19 dBi; ∆θ3dB = [0.982 × 0.904] deg Relative cross-pol. level = -37.53 dB Directivity = 47.65 dBi; ∆θ3dB = [0.832 × 0.747] deg Relative cross-pol. level = -36.91 dB Directivity = 49.4 dBi; ∆θ3dB = [0.643 × 0.607] deg Relative cross-pol. level = -37.36 dB
4 4 4 4 4 4
40 3 40 3 3
3 3 40 40 3
40 40
Secondary
2 2 2 2 2 2
patterns
30 30
30 30
30 30
Elevation, [deg]
Elevation, [deg]
Elevation, [deg]
Elevation, [deg]
Elevation, [deg]
Elevation, [deg]
1 1 1 1 1 1
20 20
20 20
0 0 0 0 0 20 0 20
−1 10 −1 10 −1 −1 −1 −1
10 10
10 10
−2 0 −2 0 −2 −2 −2 −2
0 0
0 0
−3 −3 −3 −3 −3 −3
−10 −10 −10 −10
−4 −4 −4 −4 −4 −10 −4 −10
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
Azimuth, [deg] Azimuth, [deg] Azimuth, [deg] Azimuth, [deg] Azimuth, [deg] Azimuth, [deg]
Chapter 3.
Figure 3.9: CFM beamformer results for 3 plane wave tapers: (from top to bottom) co- and cross-polarized components of the focal eld (white
crosses denote dipole elements); excitation coecients of corresponding array elements; primary compound beam illuminating the reector
(white line denotes the reector rim); secondary compound beam.
44
3.3.
Distance to coast, [km] Relative cross-polar power, [%] Average footprint size, [km] Beam eciency, [%]
100 FPL 100

40
Requirement 0.4 FPS
dB)
80 Requirement 35 Average FP 98
0.3 Requirement 97.19
60 30
For uniform aperture
−30
47.84 96
25 22.49
40 0.2
20
94
20
Footprint, [km]
0.1 15
Beam efficiency, [%]
Distance to coast, [km]

0
-30 0.04041 -30 92
-30
−60 −50 −40 −30 −20 −10 -30 10 −60 −50 −40 −30 −20 −10
0 −60 −50 −40 −30 −20 −10
Plane wave taper, [dB] Plane wave taper, [dB]
CFM beamformer
−60 −50 −40 −30 −20 −10
Plane wave taper, [dB]
(PW taper
Relative power in cross-polar, [%]
Plane wave taper, [dB]
km % km %
1.7 0.4 1.7 100
24 1.7 1.7
24 98
1.6 1.6 1.6 1.6
1.5 20 22 1.5 1.5 1.5 98
0.3
1.4 1.4 0.17 1.4 22 1.4
Optimum PAF beamformers
20
1.3 20 1.3 96
1.3 1.3
1.2 18 1.2 0.2 1.2 20 1.2 98
17 15.6 km 0.22% 21.8 km 98%
1.1 16 1.1 0.1 1.1 1.1 94
7
1 1 0.1 1 18 1
0.9 14 −2
0.1 0.9 0.9 97
MSMDL
0.9 5 · 10 22 92
0.8 12 0.8 0.8 0.8 96
beamformer
19 18
16 95 93 90
0.7 0.7 0.7 21 20 0.7
10 0 90
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
Major semi-axis of mask ellipse, [deg]

0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
Axis ratio of the mask ellipse, [-] Axis ratio of the mask ellipse, [-] Axis ratio of the mask ellipse, [-] Axis ratio of the mask ellipse, [-]
km % km %
1.7 0.4 1.7 100
24 1.7 1.7
1.6 1.6 24 98
1.6 1.6
1.5 22 1.5 1.5 1.5 98
0.3
1.4 20 1.4 1.4 22 1.4
20 20
1.3 1.3 1.3 1.3 96
18 30
29
1.2 20 1.2 0.012% 0.2 1.2 28 20 1.2 98.6%
17 km 27 23.6 km
1.1 16 1.1 1.1 26 1.1 94
25
1 1 1 24 18 1
0.9 14 0.9 0.1 0.9 0.9
AMBER
23 92
0.8 12 0.8 0.8 0.8 98
beamformer
22 21 16 96 95
0.7 0.7 0.7 20 19 0.7 97
10 0 90
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7

0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
Figure 3.10: Comparison of three beamformers in terms of the main radiometer characteristics, i.e. distance to coast, relative power in
cross-polarized eld component; average footprint size, and beam eciency.
45
CFM (PW taper = −30 dB) MSMDL AMBER
−5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40
coecients
0 0 0
Weight
-23.8 -21.5 -18.6 -14.7 -18.4 -12.9 -8.4 -12.9 -18.4 -14.7 -18.6 -21.5 -23.8 -39.6 -34.8 -32.0 -32.0 -34.8 -39.6 -29.2 -26.7 -31.9 -18.2 -18.3 -22.3 -12.7 -22.4 -18.3 -18.3 -32.0 -26.7 -29.1 -37.3 -35.7 -22.8 -19.7 -22.6 -18.2 -15.3 -17.3 -21.5 -19.7 -22.5 -31.2
-25.0 -21.2 -16.3 -16.0 -15.1 -5.6 -2.9 -5.6 -15.1 -16.0 -16.3 -21.2 -25.0 -30.2 -30.0 -30.0 -30.2 -30.9 -26.3 -20.2 -19.0 -20.0 -9.0 -6.2 -8.9 -20.0 -19.0 -20.1 -26.3 -30.8 -30.9 -28.6 -19.4 -18.0 -13.9 -7.7 -5.6 -7.2 -12.9 -17.2 -18.9 -27.3 -36.1 -36.7 -36.4 -36.2 -36.1
-26.0 -20.9 -18.7 -17.0 -7.3 -1.7 0.0 -1.7 -7.3 -17.0 -18.8 -20.9 -26.0 -33.7 -28.7 -30.3 -30.3 -28.7 -33.7 -37.4 -27.2 -22.2 -29.0 -9.8 -2.7 -0.7 -2.7 -9.8 -29.1 -22.2 -27.1 -37.4 -33.8 -27.7 -22.1 -16.1 -7.7 -2.7 -1.0 -2.4 -7.3 -15.3 -20.4 -25.2 -33.2 -36.1 -30.8 -31.7 -32.0 -30.4 -35.2
-27.9 -24.6 -20.5 -11.9 -5.3 -1.6 -0.4 -1.6 -5.3 -11.9 -20.5 -24.6 -28.0 -33.2 -30.7 -33.4 -33.4 -30.7 -33.2 -32.2 -32.7 -39.8 -14.7 -6.0 -1.4 0.0 -1.4 -6.0 -14.6 -39.1 -32.9 -32.8 -35.1 -22.3 -11.9 -5.3 -1.4 0.0 -1.2 -5.0 -11.6 -21.5 -35.0 -38.9 -32.7 -29.8 -31.2 -31.6 -29.4 -32.5 -38.5
-32.5 -26.8 -19.2 -12.6 -8.1 -5.5 -4.7 -5.5 -8.1 -12.6 -19.2 -26.8 -32.5 -37.3 -36.4 -36.4 -37.3 -35.0 -23.2 -13.1 -7.9 -4.9 -3.9 -4.9 -7.8 -13.1 -23.3 -34.1 -30.0 -28.4 -19.1 -12.1 -7.2 -4.0 -2.8 -3.7 -6.7 -11.6 -18.4 -27.4 -30.7 -38.6 -34.5 -32.4 -34.4 -35.0 -32.5 -34.1 -38.5
-36.4 -29.6 -23.6 -19.1 -16.0 -14.2 -13.6 -14.2 -16.0 -19.1 -23.6 -29.6 -36.4 -34.7 -28.6 -22.5 -18.4 -15.1 -13.6 -13.0 -13.6 -15.1 -18.4 -22.3 -28.4 -33.9 -35.5 -31.3 -23.9 -18.5 -14.6 -11.9 -10.7 -11.5 -13.9 -17.5 -22.3 -28.2 -31.5
Spillover efficiency = 99.12 % Relative cross-pol. level = -29.1 dB Spillover efficiency = 98.73 % Relative cross-pol. level = -37.27 dB Spillover efficiency = 99.01 % Relative cross-pol. level = -30.61 dB
0 0 0 0 0 0
105 90 75 105 90 75 105 90 75 105 90 75 105 90 75 105 90 75
120 40 60 120 40 60 120 40 60 120 60 120 40 60 120 60
40 40
135 45 135 45 135 45 135 45 135 45 135 45
patterns
Primary
30 −10 30 −10 30 −10 30 −10 30 −10 30 −10

150 30 150 30 150 30 150 30 150 30 150 30
20 20 20 20 20 20
165 10 15 165 10 15 165 10 15 165 10 15 165 10 15 165 10 15
180 0 0 −20 180 0 0 −20 180 0 0 −20 180 0 0 −20 180 0 0 −20 180 0 0 −20
195 345 195 345 195 345 195 345 195 345 195 345
210 330 210 330 −30 210 330 210 330 −30 210 330 210 330 −30
−30 −30 −30
225 315 225 315 225 315 225 315 225 315 225 315
240 300 240 300 240 300 240 300 240 300 240 300
255 270 285 255 270 285 255 270 285 255 270 285 255 270 285 255 270 285
−40 −40 −40 −40 −40 −40
Directivity = 47.65 dBi; ∆θ3dB = [0.832 × 0.747] deg Relative cross-pol. level = -36.91 dB Directivity = 48.07 dBi; ∆θ3dB = [0.78 × 0.738] deg Relative cross-pol. level = -28.22 dB Directivity = 46.76 dBi; ∆θ3dB = [0.958 × 0.825] deg Relative cross-pol. level = -44.39 dB
4 4 4 4 4 4
3 40 3 40 3 3 40 3 40 3 40
40
2 2 2 2 2 2
30 30 30 30 30
30
Elevation, [deg]
Elevation, [deg]
Elevation, [deg]
Elevation, [deg]
Elevation, [deg]
Elevation, [deg]
1 1 1 1 1 1
20 20 20 20 20
20
0 0 0 0 0 0
Secondary patterns
−1 10 −1 10 −1 −1 10 −1 10 −1 10
10
−2 −2 −2 −2 −2 −2
0 0 0 0 0
0
−3 −3 −3 −3 −3 −3
and cuts
−10 −10 −10 −10

−10 −10
−4 −4 −4 −4 −4 −4
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
Azimuth, [deg] Azimuth, [deg] Azimuth, [deg] Azimuth, [deg] Azimuth, [deg] Azimuth, [deg]
Co-polar pattern cuts Co-polar pattern cuts Co-polar pattern cuts
50 50 50
40 Phi=0 40 Phi=0 40 Phi=0
Phi=45 Phi=45 Phi=45
|Gco |, [dB]30 30 30
|Gco |, [dB]
|Gco |, [dB]
Phi=90 Phi=90 Phi=90
20 20 20
10 10 10
0 0 0
−10 −10 −10
−20 −20 −20
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
Chapter 3.
θ, [deg] θ, [deg] θ, [deg]

Figure 3.11: Comparison of three beamformers: (top) weight coecients; (middle) primary radiation patterns; (bottom) secondary radiation
patterns.
46
3.4. Optimum PAF architectures
characteristics are shown as a function of the PW taper.
Fig. 3.10 show the main radiometer characteristics as a function of considered

CFM, MSMDL and AMBER beamformer parameters, which are the PW taper for
CFM beamformer and the mask ellipse parameters (major semi-axis and axis ratio)
for the MSMDL and AMBER beamformers. The chosen beamformer parameters are
denoted by markers in the gure, and the corresponding weight coecients, primary
and secondary radiation patterns are shown in Fig. 3.11.
Having analyzed the gures one can conclude that there is a trade-o between the
distance-to-coast and the footprint characteristics, which is valid for all the beam-
formers. This is easy to explain from a physics point-of-view: the smaller footprint
we want, the larger part of reector should be illuminated, which results in higher
side lobes and, correspondingly, a larger distance-to-coast value. From Fig. 3.10 we
can see that the CFM beamformer does not allow us to satisfy both requirements at
the same time, while for the MSMDL and the AMBER beamformers we can choose
a point in the parametric space such that both radiometer characteristics get very
close to the requirements.
Table 3.3 summarizes the radiometer performance for all three beamformers with
their optimal parameters. It is pointed out that these calculations have been per-
formed at C-band (6.9 GHz).
Table 3.3: Radiometric characteristics of the push-broom system for three types of beamformers
at C-band, when the PAF consists of dipole elements, the EEPs of which are calculated using the
FEM method in HFSS.
Radiometer characte- Requirement CFM (PW
ristic taper −30dB) MSMDL AMBER
Distance to coast, [km] <15 47.8 15.6 17.0
Rel. cross-pol. power, [%] <0.34 0.04 0.22 0.01
Beam eciency, [%] 97.2 98.0 98.6
Footprint, [km] <20 22.5 21.8 23.6
For the X- and Ku-bands (10.65 GHz and 18.7 GHz) all requirements for the ra-
diometer characteristics are fully satised (see Paper K) using MSMDL beamformer.
Concluding the beamformer comparison we can say that the MSMDL is more
benecial when a measurement is performed close to a coast line as it allows smaller
distance-to-coast and footprint values, while measuring far from the coast line, the
CFM and AMBER may be a better choice thanks to their high polarization purities.
3.4 Optimum PAF architectures

In this section we will show how the initial array layout has been chosen and optimized
to minimize the required number of elements.
47
3.4.1 Initial PAF layout

In order to choose the initial layout of the array, we let a tapered plane wave be
incident from the direction of observation on the reector antenna, after which the
vector EM eld in the plane of the array is computed. The magnitude of the E-eld
is shown as a background color in Fig. 3.12. The initial size of the array has been
chosen such that it covers an area where the eld intensity exceeds (−15 . . . − 20) dB,
while the initial inter-element spacing has been chosen to be 0.5λ. This element
spacing is expected to lead to a high beam eciency, while minimizing the spillover
loss [82]. The taper of the incident plane wave has been chosen −30 dB at the reector
rim. This value is shown to be optimal from the radiometer characteristics point-of-
view [Paper D], when Conjugate Field Matching (CFM) beamforming is used. Since
we will use more advanced beamformers, the focal eld distribution will dier from
the one shown in Fig. 3.12, so that the optimal array size can be dierent as well.
Therefore we may need dierent number of antenna array elements to sample this
eld suciently well in order to satisfy the radiometer requirements. Under optimal
array we understand an array employing a minimum number of antenna elements,
while all performance requirements of the radiometer equipped with a such array
remain satised.
3.4.2 Array size and inter-element spacing

To optimize the initial array layout in conjunction with the MSMDL beamformer
described in Sec. 3.3.3, the main characteristics of the radiometer are studied as a
function of the inter-element spacing between the array elements and the array size
in the radial direction. The array size in the azimuthal direction is not a parameter
Figure 3.12: Initial layout of the PAF for the push-broom reector: red and green lines denote the
ρ- and φ-polarized array elements correspondingly, while the black arc shows the position of the
focal arc of the reector. The E-eld distribution in the array plane (when a tapered plane wave is
incident on the reector from the direction of observation) is shown as the background color, [dB].
48
3.4. Optimum PAF architectures
of interest in the optimization, since the array will need to form multiple beams in
this direction and sub-arrays for the neighbouring beams will partially overlap. This
work is presented in Paper E, while the performance of the radiometer at the X-band
is summarized in Table 3.4. In this case the radiometer is subsequently equipped
with: (i) a horn feed (its radiation pattern is modeled as a Gaussian beam), (ii) the
initial array with the CFM beamformer, and (iii) the optimized array feed employing
the MSMDL beamformer (unconstrained dynamic range of the beamformer weight
amplitudes).
PAF with CFM PAF with

Gaussian feed BF (uniform MSMDL BF
model PW) 15 × 29 × 2 6 × 13 × 2 elem.
elem. del = 0.5λ del = 0.75λ
PAF element excita-

tion coecients
Reector illumina-
tion patterns
Beam eciency [%] 84.2 85.1 98.4

XP-power, [%]
0.39 1.01 0.12
(<0.34% is req.)
Dist. to land, [km]
87.8 116.6 13.4
(<15 km is req.)
Beam width, [deg] 0.600 0.351 0.538
Footprint (FP), [km]
16.9 10.5 15.9
(<20 km is req.)
FP ellipticity 1.38 2.14 1.21
Table 3.4: Radiometer characteristics for dierent PAFs and beamformers at X-band (10.65 GHz).
As expected, dense PAFs have obvious benets in achieving the required minimum
distance-to-coast and footprint roundness, while meeting all the other radiometer
requirements at the same time. The minimum size of the PAF sub-array has been
found to be 6 × 13 elements (for each polarization) with the inter-element separation
distance in the order of del = 0.75λ.
It generally known that for maximizing the illumination eciency of the reector
the focal eld should be sampled by an array with the element spacing ≤ 0.5λ [82].
For the considered applications, however, a reduced eciency is acceptable as long
as the primary requirements (such as distance-to-coast, cross-polar power, footprint
size) are satised and the total number of the array elements are minimized. From
the other hand, del > 0.75λ leads to the grating lobes, and hence signicant drop in
the antenna beam eciency. Furthermore, the grating lobes may be directed towards
49
a strong noise source in the sky (e.g. sun, moon) and hence increase the measurement
error.
3.4.3 Dynamic range of beamformer weights

The eect of the limited dynamic range (DR) on the radiometric performance has also
been investigated. It has been found that the minimum DR required to satisfy the
radiometric requirements is 30 dB. If we reduce it further, rst the cross-polarization
power will be aected since beamforming cannot compensate it anymore. Further-
more, with the weights DR less than 20 dB the distance-to-coast characteristic will
be degraded as well.
To ensure a minor eect of the DR limitation on the radiometer performance we

will use a DR value equal to 40 dB, which is still realistic for receiver systems.
3.5 Radiating element trade-o study

The purpose of this section is to perform a trade-o analysis of several PAF radiator
candidates and select the best one in terms of the radiometer performance and feed-
ing/fabrication simplicity of the array. The input for this study is the requirements
for the radiator as dened in Sec. 3.5.1.
Three dierent antenna technologies have been considered for the analysis (see
Fig. 3.13): (i) a crossed-dipole antenna, (ii) a patch-excited cup antenna developed
by RUAG [83], and (iii) a tapered-slot antenna (Vivaldi antenna) [Paper L].
(a) (b) (c)
Figure 3.13: Considered radiating elements for the PAF: (a) a crossed-dipole antenna (HFSS model);
(b) RUAG's patch-excited cup antenna [83]; (c) Vivaldi antenna [Paper L].
50
3.5. Radiating element trade-off study
3.5.1 Requirements for the array radiator

Overview of the radiating element requirements
While selecting the radiating element for the PAF, both the electrical performances
and mechanical issues of the elements should be considered. These requirements are
summarized in Table 3.5.
Back radiation power

The back radiation power is dened here as the power of the total radiated eld of
the reector antenna system contained in the angular range subtended by the Earth,
◦ ◦
i.e., in the range of θ between 0 and 62 from the nadir, see Fig. 3.2.
Due to the high tapering of the feed pattern, the diraction eects at the reector
edge are negligibly small, and the power of the diracted eld propagating towards
the Earth is assumed to be zero. Therefore, the main contribution to the radiometer
back radiation is the back radiation of the array feed. The power of the back-radiated
eld can be decomposed into powers containing co- and cross-polarized components
of the eld. Let us consider them separately.
Co-pol power in the back radiation
As it was shown in Paper K, the distance-to-coast characteristic is dened by the

cone angle, inside which the radiometer beam contains at least 99.72% power of the
co-polarized eld component. This follows from the Eq. (4) in Paper K, which is
Pland
∆T ≥ (Tland − Th ) , (3.21)
Pco
where ∆T = 0.25 K is the accuracy requirement; Th = 75 K is the brightness tem-
perature of the sea surface (horizontal polarization); Tland = 250 K is the brightness
temperature of the land surface; and Pland is power radiated towards the hot land,
i.e.,
Pco − Pc
Pland = , (3.22)
2
where Pco is the co-polarization received power within the angular region subtended
by the Earth; and Pc is the power contained in the beam cone with semi-angle θc
(= angle between the beam center and the closest point at a coast line, see also
Fig. 3.1).
If we make the same assumption as in Paper K that only half of transmitted
power in reciprocal transmitting situation is outside of the beam cone (including the
back radiation) and thus incident on the hot land, then we can write Pland as
Pcob + Pcov − Pc
Pland = , (3.23)
2
51
Table 3.5: Radiating element requirements

Electrical performance
C-band: 6.8. . .7.0 GHz

Frequency band Dual-band operation is an advantage (+ L-band)
The amplitude of the active reection coecient should not ex-

ceed -10dB (the reference impedance of 50 Ohm) when the ele-
ment is in the nal array environment and optimum beamforming
coecients are applied. For radiating elements with negligible
Matching condition
mutual coupling eects (when they are in an array) the active
reection coecient can be replaced by the standard passive re-
ection coecient.
Beam width
Should be wider than the reector subtended angle (±24 )
◦
Cross-polarization
Suciently low in the angular range subtended by the reector
◦
(±24 ) *
level
Suciently low in the angular range subtended by Earth (180 ±

Back-radiated power 62◦ ) *
Mechanical considerations
Simplicity of the feeding network (feeding lines and interconnec-

tions)
Single-ended output ports of the antenna elements are preferred
Feeding network (no baluns)
Sucient distance between ports of the elements in an array to
connect coaxial cables
Minimum amount of dielectric material, metal-only antenna

structure is preferred in order to reduce losses and noise re-
Use of dielectric ceived while avoiding possible problems with the accommodation
to space
Element size must be smaller than 0.75λ in the horizontal plane

Size of the element (plane of the array)
* Requirements on both the cross-polarization power and back-radiated power are important
for the entire array (not for a single element as in conventional non-dense arrays), since they
directly aect the radiometric characteristics. It is impossible to dene these requirements
for a radiating element in isolation, since the resulting values will depend not only on the
element type, but also on the array topology, excitation scheme and supporting structure
around the array. Therefore, at this stage, we can dene these requirements for the entire
array with a particular array topology, excitation scheme, and no supporting structure. The
calculation methodology and two case studies are in this section.
where we split the total co-polar power over the Earth Pco into the power in the back
radiation, Pcob , and the power in the main beam vicinity, Pcov , which should have
a size sucient to capture most of the power around the main beam; we used θmax
such that about 10 side-lobes are accounted for. Substituting (3.23) in (3.21) leads
52
to
Pcov Pc 2∆T Pcob

− ≤ − , (3.24)
Pco Pco Tland − Th Pco
Poutv
that is, the power outside the cone in vicinity of the main lobe,
Pco
= PPcocov − PPcoc ,
− Pcocob = 0.0028 − Pcocob . We can rewrite this power in
2∆T P P
should be less than
Tland −Th
percent, relative to the total power in the co-polarized component, i.e.,
rel = 0.28% − P rel .

Pout (3.25)
v cob
Knowing
rel ,
Pout the corresponding cone angle θc can be found from the radiation
v
pattern, after which the distance-to-coast is calculated using Eq. (3.1).
Fig. 3.14 shows the dependence of the distance-to-coast characteristic Dc of the
relative back radiated power
rel .
Pco The gure shows that if we allow for the maximum
b
distance to coast of 20 km, the maximum acceptable co-polar back radiation power
is about 0.11%. The curves on the gure were obtained using MSMDL beamformer.
70
Distance to coast, [km]
60
50
Vivaldi EEP
40 Dipole EEP
30
20
20
17.20
16.13 0.111 0.119
10 0.28
0 0.05 0.1 0.15 0.2 0.25 0.3
Ralative back radiation power, Pcorelb , [%]
Figure 3.14: Distance-to-coast as a function of back-radiated power of the PAF feed, consisting of
either Vivaldi or dipole radiating elements, excited with the weight coecients obtained using the
MSMDL beamformer.
Note that the maximum of the acceptable back-radiated power is 0.28%, other-
wise the accuracy requirement of 0.25 K cannot be satised, even if measurements
are performed far from the coast line.
Cross-pol power in the back radiation
The radiometer requirement on the temperature measurement accuracy of 0.25 K

implies that the power in the cross-polar component over the entire angle subtended
by the Earth must be less than 0.34% of total power within this angle. For instance,
if the cross-polar power generated by the reector antenna that is fed by a PAF of
dipole or Vivaldi antennas is 0.1%, the maximum allowed cross-polar power in the
53
array back radiation is about 0.34 − 0.1 = 0.24%. It is pointed out that this value
depends on the beamforming algorithm used.
Therefore, the back radiation of the radiating element should be such that the
total back radiation of the array after beamforming (both co- and cross-polar powers)
satisfy the requirements described in this section.
Cross-polarization requirements
Previous studies by Chalmers and TICRA [84] show that the cross-polar power is a
minor issue for the push-broom conguration. This is due to:
The relatively low XP generated by the torus reector itself (e.g. XP power is
0.23% when the reector is fed by a Gaussian feed with taper -30 dB and zero
cross-polar level, versus 0.87% for the conical scanner with a similar feed);
The relatively low feed XP power inside the feed-to-reector subtended angle.
In Sec. 3.5.3 it will be shown that the cross-polar power of the reector fed by PAF
of Vivaldi antenna elements is even better than when a PAF of dipole elements is used,
despite the fact that the cross-polar level of the Vivaldi EEP is higher. This is due to
the beamformer, which compensates the cross-polar component by means of exciting
the orthogonal elements as well (see weight coecients in Figs. 3.19 and A.6). Owing
to this property of the beamformer, the cross-polar level of the radiating element
is not an issue for the most commonly used radiators. A more detailed analysis is
presented in Sec. 3.5.3.
3.5.2 Array layouts and analysis method

The torus reector discussed in Sec. 3.2 is fed by an array feed consisting of 6 × 13 ra-
diators arranged in radial type grids, as illustrated in Fig. 3.15. As shown in Sec. 3.4.2
and Paper I, 6 rows of elements are the minimum number necessary to generate one
beam of the push-broom radiometer, satisfying the radiometric requirements of the
instrument.
For PAF excitation scenarios, the antenna elements located at the edges of the
array have signicantly (−12 dB and less) lower weighting coecients relative to the
elements in the center. This implies that dierences in embedded element pattern
shapes, introduced by edge eects, will likely have relatively weak contribution to
the total compound beam of the array when all elements are active. Hence, to speed
up the simulations, one could assume that EEPs of the elements near the edges are
similar to an element in the center of the array, see the Modeling approach II below.
This approach was used in the previous radiometer project, however it was not vali-
dated (except for some initial results in Paper L for the conical radiometer). Hence,
54
250 250 200
200 200 150

150 150
100
100 100
50
50 50
Y, [mm]
Y, [mm]
Y, [mm]
0 0 0
-50 -50
-50
-100 -100
-100
-150 -150
-200 -200 -150
-250 -250 -200

-100 0 100 -100 0 100 -50 0 50 100
X, [mm] X, [mm] X, [mm]
(a) (b) (c)
Figure 3.15: The layouts of the 6 × 13 array feed of the torus reector when the radiating antenna
element is: (a) a crossed half-wavelength dipole antenna; (b) a dual-polarized patch-excited cup
antenna, and; (c) a Vivaldi antenna (the Y-polarized elements in the 7th column are passive).
here we perform validation tests for the push-broom radiometer, and in particular
consider three modeling approaches for the computation of the EEPs of the array
elements (see also Fig. 3.16):
1. Modeling approach I: All EEPs are identical, obtained from an innite array
simulation and shifted to the positions of array elements in the layout;
2. Modeling approach II: The same as Modeling approach I, but where the
EEPs are obtained from the central element of a nite 5×5 rectangular array,
which allows for better estimation of the EEPs (especially its cross-polarization
component);
3. Modeling approach III (Reference): All individual EEPs are obtained

from a full-wave simulation of the full 6 × 13 array of dual-polarized elements.
The rst approach is the fastest one, however, as can be seen from the EEPs (see
e.g. Fig. 3.17), its accuracy is limited. On the other hand, and as expected, the
full-wave full-array approach is very time-consuming.
In next section we show the results for all radiators using the above described
approaches.
55
Identical EEPs equal to the

Individual EEPs from a
Identical EEPs from an central EEP of a 5x5 nite
full-wave array (Approach
innite array (Approach I) rectangular array (Approach
III)
II)
Figure 3.16: HFSS models to calculate EEPs for the 6x13 array: (a) Innite array simulations
(Approach I); (b) a small-scale nite array simulation and phase-shifted versions of the central
element EEP (Approach II), and; (c) a complete large-scale array simulation (Approach III).
3.5.3 Comparison of the dipole, patch and Vivaldi elements

In this section we present the numerical results for the dipole-element arrays. The
results for patch- and Vivaldi antenna arrays are given in the Appendix A. The set
of results is composed of the following gures:
1. (Fig. 3.17) EEPs of the array including: (i) the co- and cross-polarization com-
ponent contour plots of the central element, and; (ii) the E- and H-plane cuts
of co-polarized E-eld component, as well as D-plane cuts of cross-polarized
E-eld component. In case of the small-scale array (Modeling approach II) the
cuts are shown for the central element only, while in case of the full-scale array
model (Modeling approach III) the cuts are shown for every array element.
2. (Fig. 3.18) Contour plots of the main radiometer characteristics (distance to

coast, relative cross-polar power, beam eciency and average footprint) as a
function of two beamformer parameters, i.e., the major semi-axis and axis ratio
of the reector antenna beamwidth at the −20 dB level. The optimum set
of the beamformer parameters are chosen such that all the above mentioned
characteristics are within the required specications, and is indicated by black
marker, along with its corresponding value.
3. (Fig. 3.19) The amplitude of the optimal weighting coecients corresponding

to the above chosen beamformer parameters. They are shown for both the co-
and cross-polarized antenna elements.
Additionally, in Fig. 3.20, the resultant radiation patterns of the PAF are shown
for the dipole array, when the antenna array elements are excited using optimal
beamforming weights, as well as the corresponding radiation pattern of the whole
56
antenna system that is used to evaluate the radiometer characteristics. Radiation

patterns for the patch and Vivaldi arrays are visually very similar and therefore not
shown.
Table 3.6 summarizes achieved radiometer performance for the three arrays with
the EEPs computed by three methods described above. The results for the full-wave
Vivaldi array are not available due to limited computational resources.
Table 3.6: Radiometric characteristics of the push-broom system for three types of PAF with EEPs
each calculated using three methods.
Radiometer characte- Requirement Approach
Inf.array
I: Approach II: Approach III:
Small array
ristic Full-wave
Dipole array
Distance-to-coast, [km] <15 15.3 15.5 16.5
Beam eciency, [%] 98.5 98.3 98.1
Footprint, [km] <20 22.6 22.2 22.2
Footprint ellipticity 1.54 1.57 1.54
Patch-excited cup array
Distance-to-coast, [km] <15 15.3 15.2 16.1
Beam eciency, [%] 98.3 98.3 98.1
Footprint, [km] <20 22.5 22.7 23.1
Vivaldi array
Distance-to-coast, [km] <15 16.6 16.6 N/A
Rel. cross-pol. power, [%] <0.34 0.17 0.10 N/A
Beam eciency, [%] 98.0 98.3 N/A
Footprint, [km] <20 22.0 22.2 N/A
Footprint ellipticity 1.51 1.59 N/A
As one can see, the power of the cross-polarized component is not an issue re-
gardless of the radiating element type and analysis approach. The distance-to-coast
and footprint size exceed a little the requirement, but is still much better than if a
horn feed is used (see Papers K and E for the push-broom radiometer results with
a horn feed). From Fig. 3.18 it follows that we could reduce the footprint size, but
this would result in a unacceptably large distance-to-coast characteristic and reduced
beam eciency. This trade-o eect is expected because in order to achieve a smaller
footprint we need to over-illuminate the reector (reduce the illumination taper at
the reector edge), which leads also to increased side-lobe and spillover levels, which
in turn aect the distance-to-coast and the beam eciency, respectively.
Another interesting observation can be made about the cross-polarization power
for each radiating element. Despite the cross-polarization level within the reector
subtended angle is the lowest for the PAF of dipole elements and the largest for the
Vivaldi PAF (see 2nd column in Figs. 3.17, A.1 and A.4), the power contained in
the cross-polarized eld component after beamforming behaves in opposite way, i.e.,
57
Identical EEPs from an innite array Identical EEPs from a 5x5 nite Individual EEPs from a full-wave array
(Approach I) rectangular array (Approach II) (Approach III)
0 0 0

−10 −10 −10
−20 −20 −20
−30 −30 −30
−50 0 50 −50 0 50 −50 0 50
Central element: Nearly central element:
0 0 0
105 909075 105 909075 105 909075
120 60 120 60 120 60
135 75 45 135 75 45 135 75 45
60 60 60
150 45 30 150 45 30 150 45 30
30 −10 30 −10 30 −10
165 15 165 15 165 15
15 15 15
180 0 0 180 0 0 180 0 0
195 345 195 345 195 345
−20 −20 −20
210 330 210 330 210 330
225 315 225 315 225 315
240 300 240 300 240 300
255 270 285 255 270 285 255 270 285
−30 −30 −30
0 0 0
105 909075 105 909075 105 909075
120 60 120 60 120 60
135 75 45 135 75 45 135 75 45
60 60 60
150 45 30 150 45 30 150 45 30
30 −10 30 −10 30 −10
165 15 165 15 165 15
15 15 15
180 0 0 180 0 0 180 0 0
195 345 195 345 195 345
−20 −20 −20
210 330 210 330 210 330
225 315 225 315 225 315
Chapter 3.
240 300 240 300 240 300

255 270 285 255 270 285 255 270 285
−30 −30 −30
Figure 3.17: Embedded element patterns of the dipole array. Cuts of the co-polar patterns for E- and H-planes, as well as cross-polar pattern
for diagonal plane (φ = 45◦ ) are shown on top, and the green area denotes the reector subtended angle.
58
Relative cross-polar power,
Distance to coast, [km] Average footprint size, [km] Beam eciency, [%]
[%]
3.5.
km % km %
0.4 100
1.6 24 1.6 1.6 1.6
24
1.5 22 1.5 1.5 1.5 98
1.4 1.4 0.17 0.3 1.4 1.4
20 22
20
1.3 17 1.3 1.3 1.3 96
15.3 km 18 0.2% 22.6 km 98%
1.2 1.2 0.2 1.2 20 1.2
1.1 16 1.1 1.1 23 1.1 94
1 14 1 1 18 1
0.17 0.1 98
0.9 0.9 0.9 0.9 92
12 0.1 −2
10 22 20 16 93
(Approach I)
0.8 0.8 5· 0.8 21 19 18 0.8 97 96 95 90
10 0 90
an innite array
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7

0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
Identical EEPs from

km % km %
0.4 100
1.6 24 1.6 1.6 1.6
24 98
1.5 20 1.5 1.5 1.5
22 98
1.4 1.4 0.3 1.4 1.4
20 22
20
1.3 1.3 1.3 1.3 96
15.5 km 18 0.18% 22.2 km 98%
1.2 1.2 0.2 1.2 20 1.2
17
1.1 16 1.1 1.1 1.1 94
0.17
1 14 1 1 18 1
0.1 0.1 98
92
Radiating element trade-off study
0.9 12 0.9 0.9 22 0.9

5 · 10−2
20 18 16 0.8 97 6 95 90
0.8 0.8 0.8 21 19 9 93
(Approach II)
10 0 90
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7

0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
Axis ratio of the mask ellipse, [-] Axis ratio of the mask ellipse, [-]
nite rectangular array

Identical EEPs from a 5x5

km % km %
0.4 100
1.6 24 1.6 1.6 1.6
24 98
1.5 22 1.5 1.5 1.5 98
20
1.4 20 1.4 0.17 0.3 1.4 1.4
20 22
1.3 1.3 1.3 1.3 96
16.5 km 18 0.2% 22.2 km 98%
1.2 1.2 0.2 1.2 20 1.2
17
1.1 16 1.1 1.1 1.1 98 94
1 14 1 1 18 1
0.17 0.1
0.9 0.9 0.1 0.9 0.9 92
12 22
−2
20 18 16 0.8 97 96 95 93 90
0.8 0.8 5 · 10 0.8 21 19
(Approach III)
10 0 90
a full-wave array
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7

0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
Individual EEPs from

Figure 3.18: Radiometer characteristics as function of two beamformer parameters, i.e. major semi-axis and axis ratio of the antenna main
59
lobe. The black marker shows the chosen optimum parameters and corresponding characteristic's value.
Identical EEPs from an innite array Identical EEPs from a 5x5 nite rectangular Individual EEPs from a full-wave array
(Approach I) array (Approach II) (Approach III)
Co-polarized elements, [dB] Co-polarized elements, [dB] Co-polarized elements, [dB]
0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40
-29.5 -26.0 -30.7 -17.4 -16.4 -24.1 -9.4 -24.1 -16.4 -17.4 -30.7 -26.0 -29.5 -29.2 -27.1 -30.8 -17.2 -16.9 -23.1 -9.1 -23.1 -17.0 -17.2 -30.8 -27.1 -29.2 -29.3 -26.1 -29.8 -17.4 -16.7 -22.3 -9.2 -22.4 -16.7 -17.4 -29.8 -26.1 -29.3
-27.2 -22.5 -20.9 -23.9 -13.7 -9.4 -10.9 -9.4 -13.7 -23.9 -20.9 -22.5 -27.2 -27.4 -22.1 -22.4 -27.8 -12.9 -8.9 -10.5 -8.9 -12.9 -27.8 -22.4 -22.1 -27.4 -27.7 -21.9 -21.0 -23.7 -14.2 -9.9 -11.0 -9.9 -14.2 -23.7 -21.0 -22.0 -27.7
-34.5 -36.6 -21.0 -21.1 -15.9 -3.2 0.0 -3.2 -15.9 -21.1 -21.0 -36.6 -34.5 -33.0 -21.8 -22.0 -15.6 -3.1 0.0 -3.1 -15.6 -22.0 -21.8 -33.0 -34.6 -39.8 -21.0 -21.0 -14.2 -3.0 0.0 -3.0 -14.2 -21.0 -21.0 -39.6 -34.6
-31.4 -33.2 -27.2 -11.6 -5.4 -2.6 -2.0 -2.6 -5.4 -11.6 -27.2 -33.2 -31.4 -30.7 -36.9 -27.0 -10.8 -4.9 -2.2 -1.5 -2.2 -4.9 -10.8 -27.0 -36.9 -30.7 -33.0 -32.7 -25.2 -11.6 -5.6 -2.8 -2.2 -2.8 -5.6 -11.5 -25.2 -32.7 -33.0
-33.1 -26.6 -18.1 -9.8 -5.4 -4.1 -5.4 -9.8 -18.1 -26.6 -33.1 -34.2 -24.9 -18.2 -9.9 -5.7 -4.4 -5.7 -9.9 -18.2 -24.9 -34.3 -32.5 -39.0 -26.4 -17.6 -10.1 -5.5 -4.3 -5.5 -10.1 -17.6 -26.4 -39.0 -32.4
-36.9 -30.1 -21.3 -17.2 -14.0 -13.9 -12.7 -13.9 -14.0 -17.2 -21.3 -30.1 -36.9 -37.0 -29.7 -21.4 -17.2 -13.9 -13.5 -12.4 -13.5 -13.9 -17.2 -21.4 -29.7 -37.0 -34.1 -29.0 -20.6 -16.5 -13.3 -13.4 -12.2 -13.4 -13.3 -16.5 -20.5 -28.9 -34.1
Cross-polarized elements, [dB] Cross-polarized elements, [dB] Cross-polarized elements, [dB]
0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40
Figure 3.19: Amplitude of the weighting coecients, [dB], of the 6 × 13 elements for the chosen beamformer parameters (black color indicates
Chapter 3.
passive elements): (top) co-polarized and (bottom) cross-polarized elements
60
3.6. Conclusions
Feed pattern Secondary pattern
Spillover efficiency = 99.3 %;

Directivity = 47.87 dBi; ∆θ3dB = [0.806 × 0.747] deg
0 4
105 90 75
120 60 3 40
40
135 45 −5
30 2
150 30 30
20 −10 1
Elevation, [deg]
165 15
10 20
0
180 0 0 −15
−1 10
195 345
−20 −2
0
210 330
−3
225 315 −25 −10
−4
240 300 −4 −3 −2 −1 0 1 2 3 4
255 270 285 Azimuth, [deg]
−30
Figure 3.20: (left) PAF radiation pattern, illuminating the reector aperture, where the white line
denotes the reector rim, and; (right) radiation pattern of the reector antenna illuminated by the
PAF (central beam). The patterns are for the full-wave dipole array. Radiation patterns for other
arrays and calculation approaches are visually very similar.
it is the smallest for the Vivaldi PAF (see the "Approach II" column in Table 3.6).
This can be explained by the capability of the beamformer to use orthogonal array
elements to compensate for the cross-polarized component of the secondary eld.
This can be seen from Figs. 3.19, A.3 and A.6, where the cross-polarized elements
are most strongly excited for the Vivaldi array.
In summary, we can say that the numerical results demonstrate that all con-
sidered radiating elements perform well in the array environment and meet the ra-
diometer specications when excited according to the optimum beamforming strategy
(MSMDL, [Paper I]). They weakly depend on the modeling approach for the array
antenna element, in the sense that the nal beams of the PAF-fed reector antenna
are almost identical, though the primary embedded element patterns and their ex-
citations derived by the three approaches dier. The three approaches require very
dierent computation times, as well as the time for setting up the array geometry in
software.
Since all elements were found to meet the radiometer requirements, the nal choice
of the nal array element should be mainly based on the mass and cost gures, as
well as possible multi-band considerations for future potential ocean missions.
3.6 Conclusions
Existing space-borne microwave radiometers that are used for the assessment of ocean
parameters like salinity, temperature, and wind can provide valid observations only
61
up to ∼ 100 km from the coastline, and hence do not allow for monitoring of the
coastal areas and ice-edge polar seas, or for measuring under extreme wind and
weather conditions. To achieve the desired precision, as required for future missions,
we propose digitally-beamforming phased array feeds (PAFs) previously not used
in space-borne applications employed either in a traditional conical-scan oset
parabolic reector antenna or in a wide-scan torus reector system.
When synthesized and excited according to the proposed optimum beamforming
procedure aiming at minimizing the signal contamination given by the side-lobe and
cross-polarization levels of antenna beams over the land the number of PAF antenna
elements and associated receivers can be kept to a minimum. In this procedure,
the input parameters include the number of array elements, their positions and the
secondary embedded element patterns (EEPs), which are computed after illuminating
the reector antenna. The output parameters are the optimal complex-valued element
excitations. Although the primary EEPs are generally not identical due to the array
antenna mutual coupling and edge truncation eects, for the considered PAFs with
more than 100 dipole antenna elements and inter-element spacing of 0.75λ, it has
been found sucient to use a single primary EEP. That is, the one for a central
element of the array as the source for each secondary EEP to accurately predict the
achievable radiometric characteristics.
1
For both types of radiometers , the realized resolutions are at least twice higher
than those realized by present-day systems; the distance-to-coast is as short as 6-16
km, depending on the frequency band. This excellent performance was shown to
be impossible with traditional multi-frequency PAFs of horns in one-horn-per-beam
congurations, as these cannot compensate for the high cross-polarization levels of
o-axis beams in conical-scanners and lead to unacceptably high side-lobes due to
severe focal-eld under-sampling eects in torus reector systems.
1 results for the conical scanner see in Paper K
62
Chapter
4 Beamforming Strategy for
Beam Shape Calibration of
PAF-equipped Radio Tele-
scope
In this chapter we will come back to the radio astronomy application of phased array
feeds and show how a constrained beamformer may simplify the calibration of a beam
shape.
Calibration of radio telescopes requires accurate models of the instrumental pa-

rameters and propagation conditions that aect the reception of radio waves [30].
These eects vary over time and the model parameters have to be determined at the
time of observation through a number of calibration measurements. Furthermore,
the calibration measurements should complete in a relatively short time and may be
repeated often over the course of an observation during which the instrumental and
atmospheric conditions can change signicantly. One of the instrumental param-
eters that needs accurate characterization is the radiation pattern of the antenna,
which is especially challenging in the arena of future array based multiple beam ra-
dio telescopes [8587], both due to the complexity of these instruments, as well as
the increased size of the Field-of-View (FoV). Calibrating for the radiation pattern of
a multi-beam PAF-based radio telescope largely depends on the accuracy of the pat-
tern model, and the availability of suitable reference sources to solve for the unknown
parameters in the pattern model.
The proposed idea on improving the calibration eciency of a radio telescope ra-
diation pattern is to conform the beamformed far-eld patterns to a two-parameter
physics-based analytic reference model through the use of a linearly constrained min-
imum variance (LCMV) beamformer. Through this approach, which requires only a
few calibration measurements, an accurate and simple pattern model is obtained.
The rst term of the Jacobi-Bessel (JB) series solution of reector antenna far
63
Chapter 4. Beamforming Strategy for Beam Shape Calibration of PAF-...
eld patterns [88, 89] is used as a reference pattern:
J1 (ka sin θ)
FA (θ, φ) ∝ ≡ jinc(ka sin θ), (4.1)
ka sin θ
where a is the reector aperture radius; k is the free space wavenumber. This model
has been extended to account for a beam width (parameter s in the equation below)
and the phase gradient of a scanned beam (parameter Ψ):
F (s, Ψ; θ, φ) = jinc(ksa sin θ)ejΨ sin θ cos(φ−φ0 ) , (4.2)
in which s and Ψ control the the amplitude and phase distributions of the reference
pattern, respectively.
The reference pattern (4.2) is used to dene directional constraints in a LCMV
beamformed PAF, for which the weights applied to the elements of the PAF are
calculated according to [90] [77, p. 526]
−1 H −1
H
= gH GH C−1 G

wLCMV G C (4.3)
in which xH means the complex conjugate transpose of x, C is the noise covari-

ance matrix, g is the constraints vector, and G is the directional constraint matrix.
For L elements in the array and constraints enforced in the K dierent directions
{Ω1 , Ω2 , . . . , ΩK }, G is an L×K matrix in which the ith column contains the signal
response vector of the array due to a plane wave incident from direction Ωi , and
the corresponding element gi in the vector g is the constraint value enforced on the
pattern in that direction. The choice of these constraint parameters comes from the
reference pattern (4.2).
An eect of the model parameters s and Ψ on the resulting beam characteristics
(directivity, side-lobes level) and the error between the actual LCMV beamformed
pattern and its model (4.2) are presented in Paper F, where the APERTIF PAF [7]
has been used to feed an oset Gregorian reector based on the MeerKAT radio
telescope reector antenna [91].
It is shown in Paper F that this beamforming approach has several performance
benets including circularly symmetric scanned beams over a wide FoV, even for
non-symmetric reector antennas. For the example of the MeerKAT oset Gregorian
antenna, this strategy resulted in multiple beams with aperture eciency above 70 %
that could be approximated down to the 10 dB level as a single analytic function with
an error of less than 5 %. In comparison with a conventional MaxDir beamformer,
this would reduce the average pattern calibration model error by more than 50 %.
64
Chapter
5 Conclusions and recommenda-
tions for future work
During last decades phased array feeds (PAFs) for reector antennas have been proven
to have numerous advantages over single-pixel feeds or clusters of them. However,
many unsolved questions remain, among them: "What is the mechanism governing
the PAF-reector interaction and how does it aect the reector antenna charac-
teristics, such as its radiation pattern, directivity, receiving sensitivity, etc?" In the
rst part of the current work an attempt to answer this question is made. For this
purpose a CBFM-PO Jacobi-iterative approach has been developed to model a large
reector antenna (with a diameter exceeding 100 wavelengths) that is fed by a com-
plex PAF. This approach, in combination with the proposed acceleration techniques,
not only allows one to solve electrically large antenna systems accurately and time-
eciently, but it also provides a physical insight in the feed-reector mutual coupling
mechanism. Several numerical computations have been performed including for a
real-world PAF system (i.e. a prototype of APERTIF system of the Westerbork
Synthesis Radio Telescope located in The Netherlands) and demonstrated excel-
lent agreement with the measurements. As a part of this study on PAFs for radio
astronomy, it has been shown how advanced beamforming algorithms can be used to
reduce the calibration complexity of the beam shape, while maintaining high receiving
sensitivity of radio telescopes equipped with PAFs.
The second part of the thesis is devoted to a feasibility study of PAFs in satellite
radiometers for remote sensing of the sea surface. In the current work the push-broom
radiometer with a toroidal reector has been considered and the following questions
have been addressed:
to what extent can the performance of push-broom radiometers be enhanced

by using dense PAFs and what are their performance-limiting factors?
what beamforming algorithms should be used to approach a certain optimality

criterion on the receiving characteristics of the radiometer?
what is the minimum complexity of the PAF design (size, number of elements
and their arrangement in the feed as well as the number of active receiver
65
Chapter 5. Conclusions and recommendations for future work
channels) that is required for meeting the instrument specications at which

future radiometers aim?
what radiating element types are most suitable for such radiometer applica-
tions?
To answer these questions several optimum beamforming methods have been con-
sidered, including a conventional Conjugate Field Matching (CFM) method and
two new methods which have been developed in this work: Maximum-radiometric
Sensitivity-to Minimum-Distance-to-Land (MSMDL) beamformer; and Advanced Max-
imum Beam Eciency beamformer (AMBER). The latter are specialized optimum
beamforming algorithms aiming at minimizing the signal contamination caused by
the side-lobes and cross-polarization of antenna beams covering the land, when mea-
suring the brightness temperature of the sea with a certain footprint.
The proposed beamforming solutions have been evaluated for the torus reector
antenna, which has the projected aperture of 5 × 7.5 m and the focal length of
5 m. It has been found that the MSMDL beamformer has the best performance in
terms of minimum distance-to-coast for the required footprint size, which is given by
the reector antenna aperture. The CFM and AMBER beamformers are preferred
when high polarization purity is required (e.g. the relative cross-polar power was
found to be in order of 0.01% for the AMBER and 0.2% for MSMDL beamformers,
respectively); with the dierence that AMBER leads to a more compact array with
almost twice fewer active antenna elements as compared to CFM in order to achieve
similar distance-to-coast.
Furthermore, it has been shown that when the PAF antenna elements are located
along the focal line of the torus reector (i.e. synthesize a moon-shaped array
layout), the optimum beamforming coecients are virtually identical for all sub-
◦
arrays generating multiple beams over a wide scanning range (±20 for the present
◦
study case, and potentially up to ±180 ). This is an advantage of the torus geometry
over conventional parabolic reectors, though this wide scanning range can be realized
only in a single dimension. A drawback of the torus conguration is that it requires
a very large number of the PAF antenna elements and associated receivers in the
present case 1332, 1836 and 3060 elements for 58, 89, and 156 beams at C-, X-
and Ku-bands, respectively. (In comparison, the L-band PAF illuminating a prime-
focus parabolic reector of the APERTIF radio telescope has ∼ 100 Vivaldi antenna
elements producing 37 beams.) Therefore, a modied torus reector geometry can
be further considered that has shorter focal length, and hence more compact focal
eld distribution.
Several types of PAF elements have been studied, including a crossed-dipole an-
tenna, patch-excited cup antenna, and tapered-slot antenna elements. To cross-
compare these elements, we used the above array layout synthesis procedure and
beamforming algorithms optimizing the radiometer characteristics while minimizing
66
the numbers of array elements. It was found that for dense PAFs (where the inter-
element-separation distance is ∼ 0.75λ and the total number of elements is large)
the type of the antenna element has a minor impact on the radiometer characteris-
tics, as opposed to the array beamforming method. Thanks to the large number of
degrees of freedom in beamforming (i.e. the fact that all array elements are excited
with their individual complex-valued coecients), the relative dierence between the
array embedded element patterns are compensated for in the beam forming process.
At present, a research project funded by the European Space Agency, is manufac-
turing a test 7 × 5 × 2 element PAF breadboard, that has been designed by using the
proposed array synthesis methodology and optimum beamforming algorithms. This
project is carried out in collaboration with TICRA and DTU-Space (Denmark).
67
68
Appendix
A Radiometer characteristics for
the PAFs of the patch-excited
cups and Vivaldi elements
69
0 0 0
Appendix A. Radiometer characteristics for the PAFs of the patch-...

−10 −10 −10
−20 −20 −20
−30 −30 −30
−50 0 50 −50 0 50 −50 0 50
Central element: Nearly central element:
0 0 0
105 909075 105 909075 105 909075
120 60 120 60 120 60
135 75 45 135 75 45 135 75 45
60 60 60
150 45 30 150 45 30 150 45 30
30 −10 30 −10 30 −10
165 15 165 15 165 15
15 15 15
180 0 0 180 0 0 180 0 0
195 345 195 345 195 345
−20 −20 −20
210 330 210 330 210 330
225 315 225 315 225 315
240 300 240 300 240 300
255 270 285 255 270 285 255 270 285
−30 −30 −30
0 0 0
105 909075 105 909075 105 909075
120 60 120 60 120 60
135 75 45 135 75 45 135 75 45
60 60 60
150 45 30 150 45 30 150 45 30
30 −10 30 −10 30 −10
165 15 165 15 165 15
15 15 15
180 0 0 180 0 0 180 0 0
195 345 195 345 195 345
−20 −20 −20
210 330 210 330 210 330
225 315 225 315 225 315
240 300 240 300 240 300
255 270 285 255 270 285 255 270 285
−30 −30 −30
Figure A.1: Patch array: Embedded element patterns. Cuts of the co-polar patterns for E- and H-planes, as well as cross-polar pattern for
diagonal plane (φ = 45◦ ) are shown on top, and the green area denotes the reector subtended angle.
70
Relative cross-polar power,
Distance to coast, [km] Average footprint size, [km] Beam eciency, [%]
[%]
km % km %
0.4 100
1.6 24 1.6 1.6 1.6
24
1.5 22 1.5 1.5 1.5 98
1.4 1.4 0.17 0.3 1.4 1.4
20 22
20
1.3 1.3 1.3 1.3 96
15.3 km 18 0.2% 22.5 km 98%
1.2 17 1.2 0.2 1.2 20 1.2
16 23 94
1.1 1.1 1.1 1.1
0.17
1 14 1 1 18 1 98
0.1
0.9 0.9 0.1 0.9 0.9 92
12 16
22 20 95 90
(Approach I)
0.8 0.8 5 · 10−2 0.8 21 19 18 0.8 97 96 93
10 0 90
an innite array
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7

0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
Identical EEPs from

km % km %
0.4 100
1.6 24 1.6 1.6 1.6
24
1.5 22 1.5 1.5 1.5 98
1.4 1.4 0.3 1.4 1.4
20 22
20
1.3 17 1.3 1.3 1.3 96
15.2 km 18 0.16% 22.7 km 98%
1.2 1.2 0.2 1.2 20 1.2
1.1 16 1.1 1.1 1.1 94
23 18 98
1 14 1 0.1 1 1
0.9 0.9 0.9 0.9 92
7
12 97
0.
21 16 93
0.1
0.8 0.8 0.8 22 20 19 18 0.8 96 95 90
(Approach II)
5·
10 0 90
0.5
0.8
1.1
1.4
1.7
0.5
0.8 1
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7

0.65
0.95
1.25
1.55
0.65
0.95 10 −2
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55


km % km %
0.4 100
1.6 24 1.6 1.6 1.6
17 24
1.5 22 1.5 1.5 1.5 98
1.4 1.4 0.3 1.4 1.4
20 22
17 20
1.3 1.3 1.3 24 1.3 96
16.1 km 18 0.14% 23.1 km 98%
1.2 1.2 0.2 1.2 20 1.2
1.1 16 1.1 0.17 1.1 1.1 98 94
1 14 1 0.17 1 18 1
0.1 23 97
0.9 0.9 0.9 0.9 92
12 0.1 −2 96
0.8 0 22 21 20 19 16 0.8 95 93 90
0.8 0.8 18
(Approach III)
5·1
10 0 90
a full-wave array
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7
0.5
0.8
1.1
1.4
1.7

0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55
0.65
0.95
1.25
1.55

Figure A.2: Patch array: Radiometer characteristics as function of two beamformer parameters, i.e. major semi-axis and axis ratio of the
71
antenna main lobe. The black marker shows the chosen optimum parameters and corresponding characteristic's value.
Co-polarized elements, [dB] Co-polarized elements, [dB] Co-polarized elements, [dB]
0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40
-29.7 -29.6 -30.4 -18.9 -20.6 -20.8 -13.4 -20.9 -20.7 -18.9 -30.5 -29.6 -29.6 -28.9 -29.0 -29.2 -18.9 -22.4 -18.4 -12.3 -18.4 -22.5 -18.8 -28.8 -29.2 -28.8 -28.8 -31.2 -26.2 -18.1 -23.1 -20.2 -14.2 -20.0 -23.2 -18.1 -26.2 -30.9 -28.8
-34.0 -27.5 -20.5 -19.9 -19.0 -8.4 -5.8 -8.5 -19.0 -19.9 -20.4 -27.5 -34.1 -33.6 -29.0 -19.9 -19.1 -19.5 -7.6 -4.9 -7.6 -19.0 -19.4 -19.8 -28.3 -34.2 -25.8 -20.5 -20.0 -17.1 -8.1 -5.7 -8.1 -17.0 -19.9 -20.4 -25.7
-36.3 -26.6 -22.7 -26.1 -9.1 -2.5 -0.6 -2.5 -9.1 -26.1 -22.7 -26.6 -36.5 -39.0 -25.0 -20.9 -33.2 -9.2 -2.4 -0.4 -2.3 -9.0 -29.6 -21.5 -25.0 -38.6 -31.2 -25.1 -24.8 -20.4 -8.0 -2.3 -0.6 -2.3 -8.0 -20.4 -24.8 -25.0 -31.2
-32.4 -33.9 -33.5 -13.9 -5.6 -1.3 0.0 -1.3 -5.6 -13.9 -33.3 -34.0 -32.8 -31.5 -27.9 -15.3 -6.0 -1.4 0.0 -1.4 -5.9 -14.8 -29.2 -32.1 -30.7 -36.7 -26.1 -12.1 -5.1 -1.3 0.0 -1.2 -5.0 -12.1 -26.1 -37.2 -31.0
-36.2 -22.1 -13.0 -7.6 -4.6 -3.6 -4.6 -7.6 -13.0 -22.2 -35.3 -34.4 -25.2 -14.0 -8.2 -4.9 -3.8 -4.9 -8.1 -13.9 -24.2 -32.8 -37.4 -34.5 -20.2 -12.2 -7.4 -4.6 -3.7 -4.6 -7.4 -12.2 -20.2 -34.8 -36.1
-36.7 -28.6 -22.3 -17.5 -14.3 -12.6 -11.9 -12.6 -14.3 -17.5 -22.2 -28.5 -35.9 -31.1 -23.5 -18.4 -15.1 -13.4 -12.7 -13.4 -15.1 -18.4 -23.3 -31.2 -39.7 -35.3 -28.6 -22.0 -17.7 -14.6 -13.0 -12.4 -13.0 -14.7 -17.7 -21.9 -28.3 -34.4
Cross-polarized elements, [dB] Cross-polarized elements, [dB] Cross-polarized elements, [dB]
0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40 0 −5 −10 −15 −20 −25 −30 −35 −40
-33.5 -34.8 -38.0 -36.7
-34.1 -32.5 -33.8 -35.9 -34.4 -36.6
-35.3 -34.6 -37.9 -37.1 -35.7
-37.4 -34.8 -36.9
-39.8
-36.8 -37.1
Figure A.3: Patch array: Amplitude of the weighting coecients, [dB], of the 6 × 13 elements for the chosen beamformer parameters (black
color indicates passive elements): (top) co-polarized and (bottom) cross-polarized elements
72
0 0
−10 −10
N/A
−20 −20

−30 −30
−50 0 50 −50 0 50
θ, [deg] θ, [deg]
Central element:
0 0
105 909075 105 909075
120 60 120 60
135 75 45 135 75 45
60 60
150 45 30 150 45 30
165 30 15
−10
165 30 15
−10
15 15
180 0 0 180 0 0 N/A
195 345 195 345
−20 −20
210 330 210 330
225 315 225 315
240 300 240 300
255 270 285 255 270 285
−30 −30
0 0
105 909075 105 909075
120 60 120 60
135 75 45 135 75 45
60 60
150 45 30 150 45 30
165 30 15
−10
165 30 15
−10
15 15
180 0 0 180 0 0 N/A
195 345 195 345
−20 −20
210 330 210 330
225 315 225 315
240 300 240 300
255 270 285 255 270 285
−30 −30
Figure A.4: Vivaldi array: Embedded element patterns. Cuts of the co-polar patterns for E- and H-planes, as well as cross-polar pattern for
diagonal plane (φ = 45◦ ) are shown on top, and the green area denotes the reector subtended angle.
73
Distance to coast, [km] Relative cross-polar power, [%] Average footprint size, [km] Beam eciency, [%]


km % km %
0.4 100
Identical EEPs from
1.6 24 1.6 1.6 1.6

24
an innite array
1.5 22 1.5 1.5 1.5

20 98
(Approach I)
1.4 1.4 0.3 1.4 1.4

20 22
20
1.3 1.3 0.17 1.3 1.3 96
16.6 km 18 0.17% 22 km 98%
1.2 1.2 0.17 0.2 1.2 20 1.2
17
1.1 16 1.1 1.1 1.1 94
98
1 14 1 0.1 1 18 1
0.1
22 0.9 92
0.9 12 0.9 0.9
16 97
0.8 5 · 10−2 21 20 19 0.8 96 95 93 90
0.8 0.8 18
10 0 90
0.5
0.65
0.8
0.95
1.1
1.25
1.4
1.55
1.7
0.5
0.65
0.8
0.95
1.1
1.25
1.4
1.55
1.7
0.5
0.65
0.8
0.95
1.1
1.25
1.4
1.55
1.7
0.5
0.65
0.8
0.95
1.1
1.25
1.4
1.55
1.7


km % km %
0.4 100
1.6 24 1.6 1.6 1.6
24 98
0.1
20
1.5 22 1.5 1.5 1.5

20 98
(Approach II)
1.4 1.4 0.3 1.4 1.4

20 0.17 23 22
1.3 1.3 1.3 1.3 96
17 16.6 km 18 0.1% 22.2 km 98%
1.2 1.2 0.2 1.2 20 1.2
1.1 16 1.1 1.1 1.1 94
1 14 1 1 18 1 98
0.1
0.9 0.9 0.9 0.9 92
12
0.5 0.34
22 20 18 16 0.8 97 96 95 93 90
0.8 0.8 0.8 21 19
10 0 90
0.5
0.65
0.8
0.95
1.1
1.25
1.4
1.55
1.7
0.65
0.8
0.95
1.1
1.25
1.4
1.55
1.7
0.5
0.65
0.8
0.95
1.1
1.25
1.4
1.55
1.7
0.5
0.65
0.8
0.95
1.1
1.25
1.4
1.55
1.7
a full-wave array
(Approach III)
N/A N/A N/A N/A
Figure A.5: Vivaldi array: Radiometer characteristics as function of two beamformer parameters, i.e. major semi-axis and axis ratio of the
74
antenna main lobe. The black marker shows the chosen optimum parameters and corresponding characteristic's value.
Co-polarized elements, [dB] Co-polarized elements, [dB]

0
0
−5 −10 −15 −20 −25 −30 −35 −40
−5 −10 −15 −20 −25 −30 −35 −40
-28.1 -32.5 -24.7 -17.8 -35.3 -14.7 -14.8 -34.7 -17.6 -24.4 -32.6 -28.1
-29.4 -27.7 -31.5 -20.8 -20.6 -24.9 -15.2 -25.1 -20.5 -20.8 -31.2 -27.5 -29.2
-35.3 -22.8 -18.3 -22.7 -12.1 -6.1 -6.1 -12.1 -22.3 -18.1 -22.6 -34.4
-30.8 -29.3 -20.8 -18.3 -25.2 -11.2 -7.9 -11.3 -25.3 -18.3 -20.9 -29.2 -30.7
N/A
-35.4 -26.4 -21.0 -29.3 -11.5 -3.8 -1.7 -3.8 -11.6 -29.3 -20.9 -26.3 -35.3 -29.4 -22.8 -24.2 -14.2 -4.6 -0.8 -0.8 -4.6 -14.2 -23.8 -22.8 -29.5
-30.7 -28.2 -15.7 -6.1 -1.4 0.0 -1.5 -6.2 -15.8 -29.4 -31.4 -8.5 -2.6 -0.0 0.0 -2.6 -8.5
-28.5 -35.2 -20.2 -20.1 -35.8 -28.8
-32.4 -23.9 -12.6 -6.8 -3.7 -2.8 -3.7 -6.9 -12.7 -24.2 -39.1 -32.2 -5.4 -3.6 -3.6 -5.3
-37.5 -29.2 -15.8 -9.1 -9.0 -15.9 -29.1 -37.0
-35.6 -27.9 -20.3 -15.5 -12.0 -10.3 -9.7 -10.3 -12.0 -15.4 -20.2 -27.3 -34.6
-31.2 -24.6 -18.8 -15.1 -12.7 -11.7 -11.7 -12.7 -15.0 -18.7 -24.5 -30.6
Cross-polarized elements, [dB] Cross-polarized elements, [dB]

0
0
−5 −10 −15 −20 −25 −30 −35 −40
−5 −10 −15 −20 −25 −30 −35 −40
-39.4 -37.5 -34.1 -35.1 -37.1 -33.8 -36.5 -39.2
-39.1 -38.5 -36.5 -35.2 -36.1 -36.0 -35.4 -36.5 -34.7

N/A
-38.9 -39.4 -29.3 -28.9 -29.2 -28.6 -37.7
-30.0 -24.7 -25.8 -25.8 -24.3 -29.7
-29.1 -25.1 -27.3 -27.6 -25.0 -28.7
-36.3 -28.3 -26.3 -29.5 -29.6 -26.5 -29.0 -36.8
Figure A.6: Vivaldi array: Amplitude of the weighting coecients, [dB], of the 6 × 13 elements for the chosen beamformer parameters (black
color indicates passive elements): (top) co-polarized and (bottom) cross-polarized elements
75
76
References
[1] W. V. Cappellen, J. G. B. de Vaate, K. Warnick, B. Veidt, R. Gough, C. Jackson,
and N. Roddis, Phased array feeds for the square kilometre array, in General
Assembly and Scientic Symposium, 2011 XXXth URSI, Istanbul, Turkey, Aug.
2011, pp. 14.
[2] J. Fisher and R. Bradley, Full-sampling array feeds for radio telescopes, in Proc.
SPIE, Radio Telescopes, vol. 4015, Munich, Germany, Jul. 2000, pp. 308318.
[3] M. Ivashina, J. bij de Vaate, R. Braun, and J. Bregman, Focal plane arrays for
large reector antennas: First results of a demonstrator project, in Proc. of the
SPIE, Astronomical Telescopes and Instrumentation, vol. 5489, Glasgow, UK,
Jun. 2004, pp. 11291138.
[4] D. Cavallo, A. Neto, G. Gerini, and G. Toso, On the potentials of connected slots
Proc. of 30th ESA Antenna
and dipoles in the presence of a backing reector, in
Workshop on Antennas for Earth Observation, Science, Telecommunication and
Navigation Space Missions, Noordwijk, The Netherlands, May 2008, pp. 407
410.
[5] D. R. DeBoer, R. G. Gough, J. D. Bunton, T. J. Cornwell, R. J. Beresford,

S. Johnston, I. J. Feain, A. E. Schinckel, C. A. Jackson, M. J. Kesteven, A. Chip-
pendale, G. A. Hampson, J. D. O'Sullivan, S. G. Hay, C. E. Jacka, T. W. Sweet-
nam, M. C. Storey, L. Ball, and B. J. Boyle, Australian SKA pathnder: A
high-dynamic range wide-eld of view survey telescope, Proc. IEEE, vol. 97,
no. 8, pp. 15071521, Aug. 2009.
[6] B. Veidt, T. Burgess, R. Messing, G. Hovey, and R. Smegal, The DRAO phased
array feed demonstrator: Recent results, in 13th Int. Symp. on Antenna Tech-
nology and Applied Electromagnetics and the Canadian Radio Science Meeting,
ANTEM/URSI 2009, Ban, Canada, Feb. 2009, pp. 14.
[7] W. A. van Cappellen and L. Bakker, APERTIF: Phased array feeds for the
IEEE International Symposium on
Westerbork synthesis radio telescope, in
Proc. Phased Array Systems and Technology (ARRAY), Boston, Oct. 2010, pp.
640647.
77
References
[8] M. Arts, M. Ivashina, O. Iupikov, L. Bakker, and R. van den Brink, Design
of a low-loss low-noise tapered slot phased array feed for reector antennas,
in Proc. European Conference on Antennas and Propag. (EuCAP), Barcelona,
Spain, Apr. 2010, pp. 15.
[9] X. Bosch-Lluis, I. Ramos-Perez, A. Camps, J. F. Marchan-Hernandez,

N. Rodriguez-Alvarez, E. Valencia, M. A. Guerrero, and J. M. Nieto, Initial
results of a digital radiometer with digital beamforming, in IEEE International
Geoscience and Remote Sensing Symposium, IGARSS 2008, vol. 2, Boston, Mas-
sachusetts, U.S.A., Jul. 2008, pp. II485II488.
[10] C. Cappellin, K. Pontoppidan, P. Nielsen, N. Skou, S. S. Søbjærg, A. Ihle,

D. Hartmann, M. Ivashina, O.Iupikov, and K. v. t Klooster, Novel multi-beam
radiometers for accurate ocean surveillance, in Proc. European Conference on
Antennas and Propag. (EuCAP), The Hague, The Netherlands, Apr. 2014, pp.
14.
[11] O. A. Iupikov, M. V. Ivashina, K. Pontoppidan, P. H. Nielsen, C. Cappellin,

N. Skou, S. S. Søbjærg, A. Ihle, D. Hartmann, and K. v. t Klooster, Dense focal
plane arrays for pushbroom satellite radiometers, in Proc. European Conference
on Antennas and Propag. (EuCAP), Hague, The Netherlands, Apr. 2014, pp.
15.
[12] M. Ivashina and C. G. M. van 't Klooster, Focal elds in reector antennas and
associated array feed synthesis for high eciency multi-beam performances,
TIJDSCHRIFT-NERG, vol. 68, no. 1, pp. 1119, 2003.
[13] W. Brisken and C. Craeye. (2004, Jan.) Focal-plane array beam-forming

and spill-over cancellation using vivaldi antennas. EVLA memo 69.
[Online]. Available: http://http://www.aoc.nrao.edu/evla/geninfo/memoseries/
evlamemo69.pdf
[14] C. Craeye, W. Brisken, and X. Dardenne, Simulated radiation characteristics of

wideband focal-plane arrays and their impact on the system gain-to-temperature
ratio, in Proc. of the JINA 2004 Conference, Nice, France, Nov. 2004.
[15] B. D. Jes, K. F. Warnick, J. Landon, J. Waldron, D. Jones, J. R. Fisher,

and R. Norrod, Signal processing for phased array feeds in radio astronomical
telescopes, IEEE J. Selected Topics in Signal Processing, vol. 2, no. 5, pp. 635
646, Oct. 2008.
[16] D. Hayman, T. Bird, K. Esselle, and P. Hall, Experimental demonstration of

focal plane array beamforming in a prototype radiotelescope, IEEE Trans. An-
tennas Propag., vol. 58, no. 6, pp. 19221934, Jun. 2010.
78
References
[17] M. V. Ivashina, O. Iupikov, R. Maaskant, W. A. van Cappellen, and T. Oost-

erloo, An optimal beamforming strategy for wide-eld surveys with phased-
array-fed reector antennas, IEEE Trans. Antennas Propag., vol. 59, no. 6, pp.
18641875, Jun. 2011.
[18] M. Elmer, B. Jes, K. F. Warnick, J. Fisher, and R. Norrod, Beamformer de-

sign methods for radio astronomical phased array feeds, IEEE Trans. Antennas
Propag., vol. 60, no. 2, pp. 930914, Feb. 2012.
[19] (2003, Sep.) Ska memo 40: Figure of merit for ska survey speed. [Online].
Available: https://www.skatelescope.org/uploaded/51368_40_memo_Bunton.
pdf
[20] M. V. Ivashina, M. Kehn, P.-S. Kildal, and R. Maaskant, Decoupling eciency

of a wideband Vivaldi focal plane array feeding a reector antenna, IEEE Trans.
Antennas Propag., vol. 57, no. 2, pp. 373382, Feb. 2009.
[21] A. Thompson, J. Moran, and G. Swenson, Interferometry and Synthesis in Radio

Astronomy. New York: Wiley, 2001.
[22] J. F. Johansson, Theoretical limits for aperture eciency in multi-beam antenna

systems, Chalmers University of Technology (Sweden), Dept. of Radio Space
Sciences, Technical Report 161, May 1988.
[23] M. V. Ivashina and C. G. M. van't Klooster, Focal elds in reector antennas

and associated array feed synthesis for high eciency multi-beam performances,
in Proc. 25th ESA Workshop on Satellite Antenna Technologies, Noordwijk, The
Netherlands, Sep. 2002, pp. 339346.
[24] R. Maaskant, D. J. Bekers, M. J. Arts, W. A. van Cappellen, and M. V. Ivashina,

Evaluation of the radiation eciency and the noise temperature of low-loss
antennas, IEEE Antennas Wireless Propag. Lett., vol. 8, no. 8, pp. 15361225,
Jan. 2009.
[25] E. E. M. Woestenburg, L. Bakker, and M. V. Ivashina, Experimental results

for the sensitivity of a low noise aperture array tile for the SKA, IEEE Trans.
[26] (2017) APERTIF project (ASTRON) Ocial website. [Online]. Available:

https://www.astron.nl/general/apertif/apertif
[27] K. F. Warnick, M. V. Ivashina, S. J. Wijnholds, and R. Maaskant, Polarime-

try with phased array antennas: theoretical framework and denitions, IEEE
Trans. Antennas Propag., vol. 60, no. 1, pp. 184196, Jan. 2012.
79
References
[28] S. Wijnholds, M. Ivashina, R. Maaskant, and K. F. Warnick, Polarimetry with

phased array antennas: Sensitivity and polarimetric performance using unpolar-
ized sources for calibration, IEEE Trans. Antennas Propag., vol. 60, no. 10, pp.
46884698, Oct. 2012.
[29] T. Carozzi and G. Woan, A fundamental gure of merit for radio polarimeters,
IEEE Trans. Antennas Propag., vol. 59, no. 6, pp. 20582065, Jun. 2011.
[30] O. Smirnov, Revisiting the radio interferometer measurement equation. ii.

calibration and direction-dependent eects, Astronomy and Astrophysics, vol.
527, Feb. 2011. [Online]. Available: http://arxiv.org/pdf/1101.1765v3.pdf
[31] O. A. Iupikov, M. V. Ivashina, and O. M. Smirnov, Reducing the complexity of

the beam calibration models of phased-array radio telescopes, in Proc. European
Conference on Antennas and Propag. (EuCAP), Rome, Italy, Apr. 2011, pp.
930933.
[32] A. Young, R. Maaskant, M. V. Ivashina, and D. B. Davidson, Performance

evaluation of far eld paterns for radio astronomy application through the use of
the Jacobi-Bessel series, in Proc. Int. Conf. on Electromagn. in Adv. Applicat.
(ICEAA), Cape Town, Sep. 2012, pp. 884887.
[33] P. Valle, G. Orlando, R. Mizzoni, F. Heliere, and K. van 't Klooster, P-band
feedarray for biomass, in Proc. European Conference on Antennas and Propag.
(EuCAP), Prague, Czech Republic, Mar. 2012, pp. 34263430.
[34] P. Cecchini, R. Mizzoni, G. Orlando, F. Heliere, and K. van 't Klooster, A

dual band array-fed reector scansar antenna, in Proc. European Conference
on Antennas and Propag. (EuCAP), Prague, Czech Republic, Mar. 2012, pp.
34313435.
[35] Antenna requirements document, TICRA (Denmark), Technical Report S-

1580-03, Jul. 2013.
[36] (2013, Sep.). [Online]. Available: http://earthobservatory.nasa.gov/Features/

EO1/eo1_2.php
[37] P. Nielsen, K. Pontoppidan, J. Heeboell, and B. L. Stradic, Design, manufacture

and test of a pushbroom radiometer, inAntennas and Propagation, 1989. ICAP
89., Sixth International Conference on (Conf. Publ. No.301), Coventry, United
Kingdom, Apr. 1989, pp. 126130.
[38] R. Hoferer and Y. Rahmat-Samii, RF characterization of an inatable parabolic

torus reector antenna for space-borne applications, IEEE Trans. Antennas
Propag., vol. 46, no. 10, pp. 14491457, Oct. 1998.
80
References
[39] S. G. Hay, J. D. O'Sullivan, J. S. Kot, C. Granet, A. Grancea, A. R. Forsyth,

and D. H. Hayman, Focal plane array development for ASKAP (australian SKA
pathnder), in Proc. European Conference on Antennas and Propag. (EuCAP),
Edinburgh, UK, Nov. 2007, pp. 15.
[40] K. F. Warnick, High eciency phased array feed antennas for large radio tele-
scopes and small satellite communication terminals, in Proc. European Confer-
ence on Antennas and Propag. (EuCAP), Gothenburg, Sweden, Apr. 2013, pp.
448449.
[41] A. N. H. Al-Rawi, A. Dubok, M. H. A. J. Herben, and A. B. Smolders, Point-

to-point radio link variation at e-band and its eect on antenna design, in in
Proc. of Progress In Electromagnetics Research Symposium (PIERS), Prague,
Czech Republic, Jul. 2015, pp. 913917.
[42] A. Zamanifekri and A. B. Smolders, Beam squint compensation in circularly

polarized oset reector antennas using a sequentially rotated focal-plane array,
IEEE Antennas and Wireless Propagation Letters (AWPL), vol. 14, pp. 815818,
Dec. 2015.
[43] (2017, May) Silicon-based Ka-band massive MIMO antenna systems for
new telecommunication services. [Online]. Available: http://cordis.europa.eu/
project/rcn/205533_en.html
[44] A. Moldsvor and P.-S. Kildal, Systematic approach to control feed scattering
and multiple reections in symmetrical primary-fed reector antennas, IEEE
Trans. Antennas Propag., vol. 139, no. 1, pp. 6571, Sep. 1992.
[45] P. Bolli, G. Gentili, L. Lucci, R. Nesti, G. Pelosi, and G. Toso, A hybrid pertur-
bative technique to characterize the coupling between a corrugated horn and a
reector dish, IEEE Trans. Antennas Propag., vol. 54, no. 2, pp. 595603, Sep.
2006.
[46] M. Bandinelli, F. Milani, G. Guida, M. Bercigli, P. Frandsen, S. Sorensen,

B. Bencivenga, and M. Sabbadini, Feed-array design in presence of strong scat-
tering from reectors, in Proc. European Conference on Antennas and Propag.
(EuCAP), Rome, Italy, Apr. 2011, pp. 38443848.
[47] C. D. Giovampaola, E. Martini, A. Toccafondi, and S. Maci, A hybrid

PO/generalized-scattering-matrix approach for estimating the reector induced
mismatch, IEEE Trans. Antennas Propag., vol. 60, no. 9, pp. 43164325, Sep.
2012.
81
References
[48] W. A. van Cappellen and L. Bakker. (2010) Eliminating sensitivity ripples in

prime focus reectors with low-scattering phased array feeds. [Online]. Available:
http://csas.ee.byu.edu/docs/Workshop/BYU_StandingWaves_Cappellen.pdf
[49] R. E. Hodges and Y. Rahmat-Samii, An iterative current-based hybrid method

for complex structures, IEEE Trans. Antennas Propag., vol. 45, no. 2, pp. 265
276, Feb. 1997.
[50] C. S. Kim and Y. Rahmat-Samii, Low prole antenna study using the physical
optics hybrid method (POHM), in Proc. IEEE AP-S International Symposium,
Ontario, Canada, Jun. 1991, pp. 13501353.
[51] J. M. Taboada and F. Obelleiro, Including multibounce eects in the moment-

method physical-optics (MMPO) method, Micr. Opt. Technol., vol. 32, no. 6,
pp. 435439, 2002.
[52] U. Jakobus and F. M. Landstorfer, Improved PO-MM hybrid formulation

for scattering from three-dimensional perfectly conducting bodies of arbitrary
shape, IEEE Trans. Antennas Propag., vol. 43, no. 2, pp. 162169, Feb. 1995.
[53] B. Andrés-Garciá, D. Gonzalez-Ovejero, C. Craeye, L. Garciá-Muñoz, and

D. Segovia-Vargas, An iterative MoM-PO method based on a MBF/Krylov
approach, in Proc. European Conference on Antennas and Propag. (EuCAP),
Barcelona, Spain, Apr. 2010, pp. 14.
[54] S. Hay, R. Mittra, and N. Huang. (2010) Analysis of reector and feed
scattering and coupling eects on the sensitivity of phased array feeds. [Online].
Available: http://csas.ee.byu.edu/docs/Workshop/BYUSGH.pdf
[55] G. Harp, R. F. Ackermann, Z. J. Nadler, S. Blair, M. Davis, M. C. H. Wright,

J. R. Forster, D. DeBoer, W. J. Welch, S. Atkinson, D. Backer, P. R. Backus,
W. Barott, A. Bauermeister, L. Blitz, D. C. J. Bock, G. Bower, T. Brad-
ford, C. Cheng, S. Croft, M. Dexter, J. Dreher, G. Engargiola, E. D. Fields,
C. Heiles, T. Helfer, J. Jordan, S. Jorgensen, T. Kilsdonk, C. Gutierrez-Kraybill,
G. Keating, C. Law, J. Lugten, D. H. E. MacMahon, P. McMahon, O. Milgrome,
A. Siemion, K. Smolek, D. Thornton, T. Pierson, K. Randall, J. Ross, S. Shostak,
J. Tarter, L. Urry, D. Werthimer, P. Williams, and D. Whysong, Primary beam
and dish surface characterization at the Allen telescope array by radio hologra-
phy, IEEE Trans. Antennas Propag., vol. 59, no. 6, pp. 20042021, Jun. 2011.
[56] W. Van Cappellen and M. Ivashina, Temporal beam pattern stability of a radio
astronomy phased array feed, in Proc. European Conference on Antennas and
Propag. (EuCAP), Rome, Italy, Apr. 2011, pp. 926929.
82
References
[57] R. Maaskant, M. V. Ivashina, S. J. Wijnholds, and K. F. Warnick, Ecient

prediction of array element patterns using physics-based expansions and a single
far-eld measurement, IEEE Trans. Antennas Propag., accepted for publica-
tion.
[58] A. Young, R. Maaskant, M. V. Ivashina, D. I. L. de Villiers, and D. B. Davidson,

Accurate beam prediction through characteristic basis function patterns for
the MeerKAT/SKA radio telescope antenna, IEEE Trans. Antennas Propag.,
vol. 61, no. 5, pp. 24662473, May 2013.
[59] O. A. Iupikov, R. Maaskant, M. Ivashina, A. Young, and P. Kildal, Fast and

accurate analysis of reector antennas with phased array feeds including multiple
reections between feed and reector, IEEE Trans. Antennas Propag., vol. 62,
no. 7, Jul. 2014.
[60] P.-S. Kildal, Factorization of the feed eciency of paraboloids and cassegrain
antennas, IEEE Trans. Antennas Propag., vol. 33, no. 8, pp. 903908, Aug.
1985.
[61] N. M. Kehn, M. V. Ivashina, P.-S. Kildal, and R. Maaskant, Denition of uni-

fying decoupling eciency of dierent array antennas case study of dense focal
plane array feed for parabolic reector, Int. Journal of Electronics and Com-
munications (AEU), vol. 64, no. 5, pp. 403412, May 2010.
[62] P.-S. Kildal, S. Skyttemyr, and A. Kishk, G/T maximization of a paraboloidal

reector fed by a dipole-disk antenna with ring by using the multiple-reection
approach and the moment method, IEEE Trans. Antennas Propag., vol. 45,
no. 7, pp. 11301139, Jul. 1997.
[63] S. N. Makarov, Antenna and EM Modeling With MATLAB. New York: John
Wiley and Sons, Inc., 2002.
[64] R. Maaskant, Analysis of large antenna systems, Ph.D. dissertation, Electrical

engineering, Eindhoven University of Technology, Eindhoven, 2010. [Online].
Available: http://alexandria.tue.nl/extra2/201010409.pdf
[65] R. Maaskant, M. V. Ivashina, O. Iupikov, E. A. Redkina, S. Kasturi, and D. H.

Schaubert, Analysis of large microstrip-fed tapered slot antenna arrays by com-
bining electrodynamic and quasi-static eld models, IEEE Trans. Antennas
Propag., vol. 56, no. 6, pp. 17981807, Jun. 2011.
[66] (2007) EM Software & Systems S.A. (Pty) Ltd, Stellenbosch, South Africa,
FEKO, Suite 6.0. [Online]. Available: http://www.feko.info
83
References
[67] B. A. Munk, Finite Antenna Arrays and FSS. Danvers, Massachusetts: John
Wiley & Sons, Inc., 2003.
[68] J. J. H. Wang, An examination of the theory and practices of planar near-eld
measurement, IEEE Trans. Antennas Propag., vol. 36, no. 6, pp. 746753, Jun.
1988.
[69] J. P. McKay and Y. Rahmat-Samii, Compact range reector analysis using the
plane wave spectrum approach with an adjustable sampling rate, IEEE Trans.
Antennas Propag., vol. 39, no. 6, pp. 746753, Jun. 1991.
[70] R. C. Rudduck, D. C. Wu, and M. Intihar, Near-eld analysis by the plane-wave

spectrum approach, IEEE Trans. Antennas Propag., vol. 21, no. 2, pp. 231234,
Mar. 1973.
[71] C. A. Balanis, Advanced Engineering Electromagnetics. New York: John Wiley

and Sons, Inc., 1989.
[72] M. V. Ivashina, R. Maaskant, and B. Woestenburg, Equivalent system rep-

resentation to model the beam sensitivity of receiving antenna arrays, IEEE
Antennas Wireless Propag. Lett., vol. 7, no. 1, pp. 733737, Jan. 2008.
[73] Y. Saad and M. H. Schultz, GMRES: a generalized minimal residual algorithm

for solving nonsymmetric linear systems, SIAM Journal on Scientic and Sta-
tistical Computing, vol. 7, no. 3, pp. 856869, Jul. 1986.
[74] C. Craeye, J. Laviada, R. Maaskant, and R. Mittra, Macro Basis Function

framework for solving Maxwell's equations in surface integral equation form,
The FERMAT Journal, vol. 3, pp. 116, 2014.
[75] A. Boja«czyk, Complexity of solving linear systems in dierent models of com-

putation, SIAM Journal on Numerical Analysis, vol. 21, no. 3, pp. 591603,
Jun. 1984.
[76] Technical note 1, TICRA (Denmark), DTU-Space (Denmark), Technical Re-
port S-1580-02, Mar. 2013.
Optimum Array Processing Part IV of Detection, Estimation,

[77] H. L. van Trees,
and Modulation Theory. New York: Wiley, 2002.
[78] D. I. L. de Villiers and R. Lehmensiek, Rapid calculation of antenna noise tem-

perature in oset gregorian reector systems, IEEE Trans. Antennas Propag.,
vol. 63, no. 4, pp. 15641571, Apr. 2015.
[79] (2017) General Reector Antenna Software Package (GRASP) Ocial

website. [Online]. Available: http://www.ticra.com/products/software/grasp
84
References
[80] (2017) Mathworks MATLAB Ocial website. [Online]. Available: https:

//mathworks.com/products/matlab.html
[81] (2017) High Frequency Structural Simulator (HFSS) ocial website. [Online].
Available: http://www.ansys.com/products/electronics/ansys-hfss
[82] M. Ivashina, M. Kehn, and P.-S. Kildal, Optimal number of elements and el-
ement spacing of wide-band focal plane arrays for a new generation radio tele-
scope, in IEEE Trans. Antennas Propag., Edinburgh, UK, Nov. 2007, pp. 17.
[83] J. Johansson and P. Ingvarson, Array antenna activities at RUAG space: An

overview, in Proc. European Conference on Antennas and Propag. (EuCAP),
Gothenburg, Sweden, Apr. 2013, pp. 666669.
[84] Study on advanced multiple-beam radiometers, ESA contract No

4000107369/12/NL/MH, Final report, TICRA (Denmark), DTU-Space
(Denmark), Chalmers (Sweden), HPS (Germany), Technical Report S-1580-D2,
Sep. 2014.
[85] S. Ellingson, T. Clarke, A. Cohen, J. Craig, N. Kassim, Y. Pihlstrom, L. Rickard,

and G. Taylor, The long wavelength array, IEEE Trans. Antennas Propag.,
vol. 97, no. 8, pp. 14211430, Aug. 2009.
[86] M. D. Vos, A. Gunst, and R. Nijboer, The lofar telescope: System architecture
and signal processing, IEEE Trans. Antennas Propag., vol. 97, no. 8, pp. 1431
1437, Aug. 2009.
[87] The SKA website. [Online]. Available: http://www.skatelescope.org/
[88] V. Galindo-Israel and R. Mittra, A new series representation for the radiation
integral with application to reector antennas, IEEE Trans. Antennas Propag.,
vol. 55, no. 5, pp. 631641, Sep. 1977.
[89] Y. Rahmat-Samii and V. Galindo-Israel, Shaped reector antenna analysis using

the Jacobi-Bessel series, IEEE Trans. Antennas Propag., vol. 28, no. 4, pp. 425
435, Jul. 1980.
[90] O. I. Frost, An algorithm for linearly constrained adaptive array processing,
Proceedings of the IEEE, vol. 60, no. 8, pp. 926935, Aug. 1972.
[91] I. Theron, R. Lehmensiek, and D. de Villiers, The design of the meerkat dish
optics, in Proc. Int. Conf. on Electromagn. in Adv. Applicat. (ICEAA), Cape
Town, Sep. 2012, pp. 539542.
85
86
Part II
Included Papers
Paper A
Towards the Understanding of the Interaction

Eects Between Reector Antennas and Phased
Array Feeds
O. A. Iupikov, R. Maaskant, and M. Ivashina
inProceedings of the International Conference on Electromagnetics in

Advanced Applications, Cape Town, WP, South Africa, September 2012.
The layout of this paper has been revised in order to comply with the rest of
the thesis.
Towards the Understanding of the Interaction
Eects Between Reector Antennas and Phased
Array Feeds
Abstract
A computationally ecient numerical procedure has been developed and

used to analyze the mutual interaction eects between an electrically large
reector antenna and a phased array feed (PAF). The complex electro-
magnetic behavior for such PAF systems is studied through a few simple
and didactical examples, among which a single dipole antenna feed, a
singly-excited antenna in an array of 20 dipoles, and a fully-excited array.
These examples account for the eects of the ground plane, active load-
ing (low noise ampliers), and beamforming scenario, and are used to
illustrate the dierences between single-port feeds and PAFs.
1 Introduction
For many practical applications it is required to accurately model the beam patterns
of reector antennas. Several factors can cause the actual beam to dier from the ide-
ally designed one due to inaccuracies of the antenna system model. For instance, one
often neglects or only partly takes into account the eects of the feed supporting
structure and reector-feed interactions. A rigorous analysis of such electrically large
antenna structures represents a challenging electromagnetic problem, especially when
the reector is fed with a phased array feed (PAF) consisting of many strongly cou-
pled antenna elements. During the last few years a number of pioneering studies have
been carried out towards the development of more complete numerical models [14]
while, at the same time, knowledge has been acquired through experimental stud-
ies [5, 6]. For example, in [6] it has been observed that the magnitude of the receiving
sensitivity ripple as a function of frequency caused by the feed-reector interactions
is signicantly smaller for a PAF of wideband Vivaldi antennas than it is for a horn
feed. It has been suggested that the smaller radar cross section (RCS) of Vivaldi
PAFs is a reason for this improvement. However, the fact that there exist dierences
in the EM coupling mechanisms for dierent phased-array and single-element feeds,
and how this aects the system design procedure, is not yet fully understood. The
objective of the present work is therefore to investigate this phenomenon in more
detail.
91
Paper A. Towards the Understanding of the Interaction Effects Between...
2 Analysis methodology
First, we examine a single dipole antenna feed above a nite ground plane, after which
an array of dipole elements is considered, as shown in Fig. 1(a) and (b), respectively.
The antenna array ports are connected to Low Noise Ampliers (LNAs) which are also
part of the antenna-receiver model. Two beamforming scenarios are considered: (i)
a singly-excited embedded element, and; (ii) a fully-excited antenna array employing
the Conjugate Field Matching (CFM) beamformer for maximizing the gain of the
secondary far-eld beam. This beamforming array system is analyzed in combination
with a parabolic reector of 8 m in diameter (∼ 38λ @ f = 1.42 GHz), F/D = 0.35.
(a) (b)
Figure 1: The considered dipole antenna feeds: (a) a single dipole; and (b) a dual-polarized array of
20 dipole antenna elements. The dipole length is (0.47λ) and the ground plane size is (3.3λ × 2.65λ)
To account for the mutual coupling between the feed and reector antenna in the
described system, a rapidly converging iterative procedure has been developed. It
consists of the following steps: (i) the antenna feed currents are computed through a
method-of-moments (MoM) approach by exciting the antenna port(s) in the absence
of the reector; (ii) these currents generate an EM eld which induces PO-currents
on the reector surface; (iii) the PO currents create a scattered eld that, in turn,
induces currents on the feed structure. The steps (ii) and (iii) are repeated until
the multiply induced currents which form the total current when summed has
converged. Afterwards, the antenna radiation pattern, the input impedance (matrix)
and derived antenna parameters aecting the receiving sensitivity can be computed.
It is worthwhile to mention that the antenna elements in our study are loaded
by LNAs, so that we will account for this loading when solving for the antenna feed
currents through the MoM. This is done through the modication of the diagonal
elements of the MoM matrix corresponding to the port basis functions as described
in [7, p. 223]. The impedance of the loads, and thus the input impedance of the
LNAs, has been chosen real-valued. Next, the (passive) reection coecient of the
antenna was minimized, which yielded the optimum load resistance of 80 and 140 Ω
for the single dipole and array case, respectively.
92
3. Numerical Results
To quantify the performance degradation of the antenna system due to the

interaction eects we analyze the antenna eciencies as well as the system noise
temperature contributions, both of which aect the receiving sensitivity Ae /Tsys [8],
i.e.,
Ae Aph ηap ηrad

= LNA (1)
Tsys ηrad Tspil + (1 − ηrad ) Tamb + TEq
where Aph and Ae are the physical and eective areas of the reector antenna,
respectively; Tsys the system noise temperature; ηap the aperture eciency; Tspil
the spillover noise temperature contribution; ηrad the antenna radiation eciency
(herein assumed 100%);
LNA the receiver
Tamb = 290 K the ambient temperature; TEq
noise temperature due to LNAs with minimum noise temperature Tmin , a component
which is independent from the antenna, and the noise coupling component Tcoup , due
to the impedance noise mismatch between the LNAs and the antenna elements [8].
In the next section it will be shown which of the above contributions are most
aected by the feed-reector interaction eects.
3 Numerical Results
The frequency-varying receiving sensitivity, which is caused by the interaction eects,
gives rise to a standing wave component between feed and reector with oscillation
period ∆f = 2F/c, where c is the speed of light [3]. Fig. 2 presents the computed
current distributions on the ground plane of the three feeds at two frequency points
leading to the minimum and maximum antenna aperture eciency within one period
of the oscillation. For the case of the single dipole [see Fig. 2(a)], one can clearly
see a signicant dierence between the areas supporting large currents on the ground
plane at these frequencies, as a result of which the corresponding far-eld patterns
of the feed dier in shape and beamwidth [see Fig. 3(a)].
Upon comparing the left- and right-hand-side subgures in Fig. 2, one observes
that the groundplane for the single-dipole case has a predominant eect on the scat-
tering mechanism. On the contrary, when the eld from the reector illuminates the
antenna array (the physical area of which is comparable to the size of the ground
plane), part of this eld is blocked by the dipoles. Therefore, the dierences be-
tween the feed patterns for the dipole arrays in Fig. 3(b) and (c) are less pronounced,
regardless of the beamforming scenario.
Next, we present the results for the system sensitivity and its subeciencies for
the three considered antenna feeds.
Fig. 4(a)(c) shows the aperture eciency and its dominant contributions, i.e.,
the spillover eciency ηspil and the taper illumination eciency ηtap ; and Fig. 4(d)
compares the respective frequency variations of ηap due to the standing wave phe-
nomenon. It is readily seen that the aperture eciency variation is less than 1% for
93
−30 −30
−35 −35
−40 −40
−45 −45
−50 −50
−55 −55
−60 −60
(a) Isolated dipole
−30 −30
−35 −35
−40 −40
−45 −45
−50 −50
−55 −55
−60 −60
(b) Embedded dipole
−30 −30
−35 −35
−40 −40
−45 −45
−50 −50
−55 −55
−60 −60
(c) Fully-excited array (CFM)
Figure 2: Current distributions on the ground plane of the feeds for two frequency points corre-
sponding to the minimum (left column) and maximum (right column) of the aperture eciency.
the two PAF cases, since the illumination pattern remains almost constant, whereas
this variation is approximately three times larger for the single dipole case, due to
the scattering mechanism dierences as described above.
A similar analysis has been performed for the system noise temperature Tsys (see
Fig. 5). Note that, for the embedded element case, Tsys is not aected much by
the standing wave phenomenon, since the input impedance of a centralized dipole
array element varies only little with frequency and is therefore well-matched (after
optimally loading the array elements), as opposed to the single dipole antenna. Also,
94
0 freq = 1.405 GHz 0

freq = 1.405 GHz
freq = 1.435 GHz
freq = 1.435 GHz
−5 −5
Level, dB
Level, dB
−10 −10
−15 −15
−20 −20
−200 −100 0 100 200 −200 −100 0 100 200
θ, deg θ, deg
(a) Isolated dipole (b) Embedded dipole

0
freq = 1.393 GHz
freq = 1.417 GHz
−5
Level, dB
−10
−15
−20
−200 −100 0 100 200
θ, deg
(c) Fully-excited array (CFM)
Figure 3: Primary patterns in φ = 45◦ cross-section.
ηspil ηtap ηap
100 100
90 90
Efficiencies, %
Efficiencies, %
80 80
70 70
60 60
1.39 1.4 1.41 1.42 1.43 1.44 1.45 1.39 1.4 1.41 1.42 1.43 1.44 1.45

100
2
90 1
Efficiencies, %
ηap variation, %
0
80
−1
70 One dipole
−2
−3
60 Dip array, CFM
−4
1.39 1.4 1.41 1.42 1.43 1.44 1.45 1.39 1.4 1.41 1.42 1.43 1.44 1.45
(c) Fully-excited array (CFM) (d) Comparison of the ηap variation
Figure 4: The aperture eciency and its dominant contributions. The solid and dotted lines are for
with and without accounting for feed-reector interactions, respectively.
95
Tsys Tmin Tsp Tcoup
50 50
40 40
30 30
T, K
T, K
20 20
10 10
0 0
1.39 1.4 1.41 1.42 1.43 1.44 1.45 1.39 1.4 1.41 1.42 1.43 1.44 1.45

50
10
40 Tsys variation, %
5
30
T, K
0
20
Dip array, CFM
10 −5
One dipole
0 −10 Dip array, one excited
1.39 1.4 1.41 1.42 1.43 1.44 1.45 1.39 1.4 1.41 1.42 1.43 1.44 1.45
(c) Fully-excited array (CFM) (d) Comparison of the Tsys variation
Figure 5: System noise temperature and its dominant contributions.
when beamforming is performed, the input impedance of each antenna array element
(scan impedance) will dier from its optimal noise-match impedance, and therefore
becomes more sensitive to the feed-reector coupling. This results to higher Tcoup
and a stronger frequency variation. Hence, and in contrast to the systems employing
single antenna feeds, the noise temperature due to mismatch eects, Tcoup , is the
dominant contribution to Tsys in case of PAF systems.
5
Dip array, CFM
Sensitivity variation, %

One dipole
−5
1.39 1.4 1.41 1.42 1.43 1.44 1.45
Frequency, GHz
Figure 6: System sensitivity variation.
The sensitivity variation for all three cases is shown in Fig. 6. Although both
ηap and Tsys vary signicantly for the system with a single dipole (i.e. −4% to
96
4. Conclusions
1.5%; and −5.5% to 3%, respectively), they partly compensate each other, leading
to approximately the same sensitivity variation for all three feeding schemes.
4 Conclusions
The electromagnetic coupling between the reector antenna and a single dipole feed
was found to have a signicant eect on the antenna beam shape and aperture ef-
ciency, as opposed to the dipole PAFs. Our study indicates that the nite ground
plane behind the single dipole, which is part of the feed supporting structure and of-
ten much larger than one antenna element, but comparable to the size of a PAF, is a
reason for this dierence. However, the (active) impedance matching of the strongly-
coupled PAF elements appears to be more sensitive to the feed-reector interaction,
which has an impact on the receiver noise temperature. Similar conclusions were
drawn from the numerical analysis of the checkerboard PAF of patch antennas [4],
whereas these eects were found to be much smaller for the larger experimentally
characterized array of 121 tapered-slot antenna elements [6]. The latter dierence
will be examined in more detail in future studies.
References
[1] N.-T. Huang, R. Mittra, M. Ivashina, and R. Maaskant, Numerical study of a
dual-polarized focal plane array (FPA) with vivaldi elements in the vicinity of a
large feed box using the parallelized FDTD code GEMS, in Proc. IEEE AP-S
International Symposium, Charleston, South Carolina, Jun. 2009, pp. 14.
[2] C. Craeye, Analysis of complex phased array feeds and their interaction with a
cylindrical reector, in Proc. IEEE AP-S International Symposium, San Diego,
California, Jul. 2008, pp. 14.
[3] M. A. Apeldoorn, Characterizing reector-feed interaction for parabolic

reector antennas, Master's thesis, International Research Centre for
Telecommunications and Radar (IRCTR), Delft University of Technology, Delft,
The Netherlands, 2011. [Online]. Available: http://repository.tudelft.nl/assets/
uuid:d6605869-37e0-49a6-8bc4-817855ab3710/MSc-Thesis-MAApeldoorn.pdf
[4] S. Hay, R. Mittra, and N. Huang. (2010) Analysis of reector and feed scattering
and coupling eects on the sensitivity of phased array feeds. [Online]. Available:
http://csas.ee.byu.edu/docs/Workshop/BYUSGH.pdf
[5] M. Ivashina. (2008) Dutch FPA progress characterization of eciency, system

noise temperature and sensitivity of focal plane arrays. [Online]. Available:
http://ska2008.ivec.org/www/Marianna_Ivashina.pdf
97

[8] M. V. Ivashina, R. Maaskant, and B. Woestenburg, Equivalent system represen-

tation to model the beam sensitivity of receiving antenna arrays, IEEE Antennas
Wireless Propag. Lett., vol. 7, no. 1, pp. 733737, Jan. 2008.
98
Paper B
A Plane Wave Approximation In The Computation

Of Multiscattering Eects In Reector Systems
inProceedings of the 7th European Conference on Antennas and

Propagation, EUCAP 2013, Gothenburg, Sweden, April 2013.
the thesis.
A Plane Wave Approximation In The
Computation Of Multiscattering Eects In
Reector Systems
Abstract
A hybrid MoM/PO method for the analysis of multiple scattering eects

between reector and large feeds, such as dense multibeam phased array
feeds or multifrequency front-ends (MFFEs) in which higher frequency
feeds operate in the vicinity of an extended metal structure, has been
presented and studied. This paper evaluates the accuracy and compu-
tational eciency for the MoM/PO method with and without using a
uniform plane wave approximation of the reector scattered eld.
1 Introduction
Prime-focus reector antennas are widely used for radio astronomy, satellite and radio
link communication thanks to their relatively low cost as compared to that of more
complex oset- and multi-reector systems. When designing these antennas, one
focuses on the optimization of the antenna feed to realize high gain, low sidelobes,
and low spillover loss for the selected reector, often under stringent dimensional
constraints to minimize the aperture blockage and frequency variation of the antenna
characteristics due to multiple scattering eects of electromagnetic waves traveling
between the feed and reector antenna.
During the last decades, a number of analytic and numerical techniques have been
developed to model feed-reector interaction eects. For example, in [1] the scattered
eld of the feed is approximated by a geometric series of elds scattered by the an-
tenna feed due to an incident plane wave at each iteration, where the amplitudes
of these plane waves are expressed analytically for a given reector geometry. This
method is very fast and, for the case of a horn feed with an aperture diameter in the
order of one wavelength, has been demonstrated to have an accuracy comparable to
that of a MoM approach. An alternative to this method is the use of more rigorous
(though more time-consuming) hybrid numerical methods combining Physical Optics
or Gaussian beams for the analysis of reectors with the Method of Moments and/or
Mode Matching techniques for radiating horns feeds [2, 3]. The recent article [4]
has introduced the PO/Generalized-Scattering-Matrix approach for solving multiple
domain problems, and has shown its application to a cluster of a few horns. This
101
Paper B. A Plane Wave Approximation In The Computation Of...
approach is generic and accurate, but requires the lling of a large scattering matrix
that can be time consuming especially for more complex feed systems, such as (i)
multifrequency front-ends (MFFEs) in which higher frequency feeds operate in the
vicinity of an extended metal structure, or (ii) dense multibeam phased array feeds
(PAFs [5, 6]). On the other hand, the above-mentioned analytic method may be in-
accurate for these systems, due to a much larger physical area and higher complexity
of radiation/scattering mechanisms (the plane wave approximation may not hold).
To examine this multiple domain problem with MFFEs and PAFs, we propose to use
a hybrid MoM/PO approach as described in [7]. While in [7] the eld is computed
at each mesh cell of the feed and reector structure, herein we investigate the ap-
proximation of the eld scattered by the reector with a (single) uniform plane wave
dened over the area of the feed. As will be shown in this paper, the scattered eld
computed through integration of the reector PO currents needs to be known only
at a few points in the focal plane region in order to determine the plane wave expan-
sion coecient in an accurate manner. This signicantly reduces the simulation time
relative to a direct MoM/PO solution.
2 Modeling procedure and numerical results

The MoM/PO method [7] consists of the following steps: (i) the antenna feed cur-
rents are computed through a method-of-moments (MoM) approach by exciting the
antenna port(s) in the absence of the reector; (ii) these currents generate an EM
eld which induces PO-currents on the reector surface; (iii) the PO currents create
a scattered eld that, in turn, induces currents on the feed structure. The steps (ii)
and (iii) are repeated until the sum of the multiply induced currents which forms
the total current has converged (typically, for low-scattering feeds, 2-3 iterations
are enough to achieve an error less than 1% relative to a MoM solution). Afterwards,
we can determine the antenna radiation pattern, the input impedance (matrix), and
derived antenna parameters aecting the receiving sensitivity.
The third step of this procedure is the most time-consuming since it requires the
eld computation (integrating of PO currents) at each mesh cell of the feed. To
alleviate this computational burden, the eld scattered from the reector can be
expanded into a plane wave spectrum, each spectral component of which induces
a current on the feed. This approach is much faster since it does not require the
integration of the reector currents at each basis function of the feed; the smoothly-
varying eld has to be tested at a few points only to nd the expansion coecients
of the corresponding plane wave modes. The incident eld on the feed is then tested
through these plane wave modes.
The model Emod of the actual focal eld Eref of the reector antenna, due to a
radiating PO current on the reector, can be expanded into a set of plane wave modes
102
2. Modeling procedure and numerical results
{En } as
N
Emod =
X
αn En . (1)
n=1
The least squares error between the actual eld Eref and the modeled eld (1) can
be expressed as
(α) = hEref − Emod (α), Eref − Emod (α)i (2)
where
ZZ
ha, bi = aH b dS; (3)
Sa
(. . .)H is the Hermitian operator; and Sa is the area constituting the support of the
vector function a.
It can be shown that, the solution that minimizes is obtained through solving
the matrix expression
Aα = b (4)
where α = [α1 , α2 , . . . , αN ]T ;
Amn = hEm , En i and bm = hEm , Eref i (5)
for m, n = 1, 2, ..., N .
Since the scattered eld from large parabolic reectors resembles a plane wave in
the vicinity of the antenna feed, it is sucient to employ only a single plane wave
expansion function [1]. Hence, we can solve Eq. (4) analytically for the coecient
α1 :
hE1 , Eref i
α1 = . (6)
hE1 , E1 i
If we choose the plane wave expansion function to have unit amplitude, the coecient
α1 will be equal to
1
ZZ
α1 = Epref dS (7)
Af
Af
where the subscript p denotes the dominant component of the eld Eref ; Af is the
area in the focal plane occupied by the feed. Eq. (7) can be evaluated numerically
using the midpoint integration rule, i.e.,
K
1 X ref
α1 ≈ E (rk ), (8)
K k=1 p
103
where the set {rk }K

k=1 are K sample points, which are assumed to be located on a
uniform grid.
In summary, the plane-wave-enhanced MoM/PO method consists of the following

steps: (i) the antenna feed currents are computed through a method-of-moments
(MoM) approach by exciting the antenna port(s) in the absence of the reector;
(ii) these currents generate an EM eld which induces PO-currents on the reector
surface; (iii) the PO currents create a scattered eld that is tested at only a few
points in the focal plane; (iv) the eld intensity at the sample points is averaged in
accordence with (8), and the obtained value is used as the expansion coecient for
the plane wave traveling from the reector towards the feed; (v) this incident plane
wave induces a new current distribution on the feed structure. The steps (ii)(v) are
repeated until a convergence condition is met.
The following three types of feeds were used to illuminate a reector antenna:
(i) a pyramidal horn with aperture diameter in the order of one wavelength, (ii) a
pyramidal horn with extended ground plane, and (iii) an 121-element dual-polarized
dipole array [see Fig. 1(a)]. All antennas are impedance power-matched, so that
antenna component [8] of their corresponding radar cross-section (RCS) is minimized.
However, the residual component of the RCS of the horn with ground plane is still
high due to the extended metal structure surrounding it, so that this feed is a high
scattering antenna and strong feed-reector coupling can be expected.
The corresponding E- and H-plane focal eld distribution cuts at the 1st and 2nd
iterations are shown in Figs. 2(b) and (c), respectively, each for the reector antenna
with the semi-subtended angle of 70 deg and a respective diameter of 38λ and 118λ.
This result clearly demonstrates that the eld scattered by the reector diers slightly
from a uniform plane wave, where the largest variation in amplitude is about 0.81.5
dB (and 46 degrees in phase) over the area occupied by the feed (the vertical dashed
line). The ripples in the focal plane eld at the 1st iteration are due to diraction
eects from the reector edges when it is illuminated by the primary feed pattern and,
as expected, are more pronounced for the electrically smaller reector, regardless of
the type of the feed. It is also observed that, at the 2nd iteration, when the scattered
eld component of the feed is incident on the reector, the focal eld distribution due
to the horn feed remains rather uniform, but becomes more tapered for the case of the
electrically larger feeds (the PAF and horn with the extended ground plane) because
of the much narrower scattered patterns of these feeds [see Fig. 1(b)]. Thus, larger
errors due to the plane wave approximation can be expected for these feed structures.
Another important observation is that the shapes of the scattered patterns and the
corresponding focal elds at the 2nd iteration are rather similar in case of the PAF
and horn with the extended ground plane, as the result of the equal aperture areas.
This similarity, however, does not imply that modeling errors due to the plane wave
approximation will be close as well. This can be readily seen from Table 1, where
the errors in the total focal eld and several antenna characteristics such as the gain,
104
2. Modeling procedure and numerical results
the gain at -3 dB level, and the antenna input impedance (in case of an array the
input impedance of the most excited antenna element) are summarized.
The errors in focal eld and scalars antenna characteristics are computed as
rP
ref − E mod |2
|Ep;k p;k
k
1 = rP × 100% (9)
|E ref |2
p;k
k
|f ref − f mod |
2 = × 100%, (10)
|f ref |
where
ref
Ep;k and
mod
Ep;k are the k -th sample of the discretized p-components of the
ref mod
focal E-eld E and E respectively; f ref and f mod is the gain or antenna input
impedance, reference and modeled values respectively. The MoM/PO results without
the plane wave approximation are used as the reference solution.
Table 1: Errors due to a plane wave approximation
Focal eld Gain Gain Impedance

(on-axis) (@−3 dB)
Reector
38λ 118λ 38λ 118λ 38λ 118λ 38λ 118λ
diameter D

Parameter
3.91 1.23 1.98 0.62 3.99 2.16 15.05 4.66
variation, %
Method: Error, %
Method 1 0.3 0.05 0.28 0.05 0.36 0.14 1.37 0.18
Method 2 0.1 0.04 0.16 0.04 0.3 0.13 0.09 0.03

Parameter
139.3 39.1 19.2 3.4 29.4 3.56 43.4 6.1
variation, %
Method: Error, %
Method 1 37.7 1.29 12.7 0.1 10.1 0.17 18.5 0.2
Method 2 11.9 0.48 2.23 0.07 4.71 0.15 12.46 0.11

Parameter
8.45 3.28 1.84 0.28 3.68 0.73 5.8 1.7
variation, %
Method: Error, %
Method 1 0.61 0.11 0.21 0.03 0.15 0.02 0.34 0.08
Method 2 0.44 0.1 0.12 0.03 0.08 0.03 0.58 0.05
105
Table 2: Total simulation time

gnd plane array
MoM-PO, no 9 min 05 sec 59 min 21 sec 71 min 09 sec 197 min 04 sec
approximations (100%) (100%) (100%) (100%)
Method 1
(7%) (2%) (7%) (17%)
Method 2
(28%) (23%) (27%) (34%)
The above values were also computed using the method described in [1], where
the plane wave coecient α1 is computed analytically from the eld intensity in the
on-axis direction of both the original and the scattered feed pattern due to an incident
plane wave. We will refer to this method as Method 1 while the herein proposed
approach is denoted as Method 2.
The total simulation time (10 frequency points) for the 38λ reector fed by the
considered feeds is shown in table 2. Virtually all simulation time is consumed by
the eld computation on the reector surface for obtaining its PO currents, while the
computation of the currents on the feed due to the currents on the reector is more
than 1000 times faster when a plane wave approximation is used.
By analyzing Table 1 and Table 2 the following observations can be made:
Method 1 is numerically ecient and accurate for small feeds (whose size is in
the order of one wavelength) and for low-scattering feeds, but fails in case of
large high-scattering feeds, such as MFFEs, because the focal eld produced by
the feed scattering pattern has a high level and a highly tapered shape;
Method 2 provides a better prediction of all the system parameters, since it

accounts for the actual shape of the scattered pattern when tting the plane
wave to it; however, it is slower than Method 1;
Both methods are accurate in case of large reectors because (i) the multiscat-
tering eects are less pronounced (see Parameter variation in Table 1), and
(ii) the eld scattered from the reector is close to a plane wave at all iterations.
3 Conclusions
A hybrid MoM/PO method for the analysis of multiple scattering eects between the
reector and large feeds, such as PAFs and MFFEs, has been presented and studied.
It has been shown that, although the eld scattered by the parabolic reector diers
slightly from that of a uniform plane wave, the plane wave approximation can be used
to predict the main antenna parameters with an error less than a few percent relative
106
3. Conclusions
(a)
0 0 0
primary primary
−10 −10 scattered −10
primary scattered
|E|, dB
|E|, dB
|E|, dB
−20 scattered −20 −20

−30 −30 −30
−40 −40 −40
−50 −50 −50
−80 −60 −40 −20 0 20 40 60 80 −80 −60 −40 −20 0 20 40 60 80 −80 −60 −40 −20 0 20 40 60 80
θ, deg θ, deg θ, deg
(b)
Figure 1: (a) EM models of the reector antenna feeds, including (when viewing from left to right)
the pyramidal horn feed (with the aperture diameter of one wavelength) without and with the
extended ground plane and the phased array feed of 121 half wavelength dipole antenna elements;
and (b) the corresponding primary eld patterns of the feeds and their scattered eld patterns due
to the eld incident from the reector at the 1st iteration.
Feed type Focal eld at 1st iteration Focal eld at 2nd iteration
0 0 38λ dish, E−plane
38λ dish, H−plane
−10
−1 118λ dish, E−plane
|E|, dB
|E|, dB
118λ dish, H−plane 118λ dish, H−plane
−20
118λ dish, E−plane
−2
38λ dish, E−plane −30
−3 −40
−0.5 0 0.5 −0.5 0 0.5
X, m X, m
0 0
−10
−1
|E|, dB
|E|, dB

−2 118λ dish, H−plane
38λ dish, E−plane −30
−3 −40
−0.5 0 0.5 −0.5 0 0.5
X, m X, m
0 0
−10
|E|, dB
|E|, dB
−20 118λ dish, H−plane

−30
−3 38λ dish, H−plane −40
−0.5 0 0.5 −0.5 0 0.5
X, m X, m
(a) (b)
Figure 2: (a)-(b) The focal plane elds of the reector antenna on transmit for the feeds shown in
Fig. 1(a). The plots in Fig. 2(a) are for the elds computed at the 1st iteration, when the reector
is illuminated by the primary eld of each of the considered feeds, and the results in Fig. 2(b)
are for the elds obtained at the 2nd iteration, when the illumination source is the scattering eld
component of the feed due to the scattered eld from the reector at the 1st iteration.
107
to a direct MoM/PO approach, while reducing the computational time signicantly.

It has also been shown that, for electrically large high-scattering feeds (exceeding
23 λ in diameter), the plane wave approximation gives rise to an increased error,
since the scattered eld from a reector at the 2nd iteration is tapered and has a
large amplitude. In the latter a spectrum of plane waves can be considered, which is
planned as future work. Among the antenna characteristics, the input impedance is
found to be the most sensitive to errors.
References
[1] A. Moldsvor and P.-S. Kildal, Systematic approach to control feed scattering and
multiple reections in symmetrical primary-fed reector antennas, IEEE Trans.
Antennas Propag., vol. 139, no. 1, pp. 6571, Sep. 1992.
2006.
[3] M. Bandinelli, F. Milani, G. Guida, M. Bercigli, P. Frandsen, S. Sorensen, B. Ben-

civenga, and M. Sabbadini, Feed-array design in presence of strong scattering
from reectors, in Proc. European Conference on Antennas and Propag. (Eu-
CAP), Rome, Italy, Apr. 2011, pp. 38443848.

2012.
[5] S. Hay, R. Mittra, and N. Huang. (2010) Analysis of reector and feed scattering
and coupling eects on the sensitivity of phased array feeds. [Online]. Available:
http://csas.ee.byu.edu/docs/Workshop/BYUSGH.pdf
[6] M. V. Ivashina, O. Iupikov, R. Maaskant, W. A. van Cappellen, and T. Oosterloo,

An optimal beamforming strategy for wide-eld surveys with phased-array-fed
reector antennas, IEEE Trans. Antennas Propag., vol. 59, no. 6, pp. 18641875,
Jun. 2011.
[7] O. Iupikov, R. Maaskant, and M. Ivashina, Towards the understanding of the

interaction eects between reector antennas and phased array feeds, in Proc.
Int. Conf. on Electromagn. in Adv. Applicat. (ICEAA), Cape Town, Sep. 2012,
pp. 792795.
108
References
109
110
Paper C
Fast and Accurate Analysis of Reector Antennas

with Phased Array Feeds including Multiple
Reections between Feed and Reector
O. A. Iupikov, R. Maaskant, M. Ivashina, A. Young, and P.S. Kildal
IEEE Transactions on Antennas and Propagation, vol.62, no.7, 2014
the thesis.
Fast and Accurate Analysis of Reector Antennas
with Phased Array Feeds including Multiple
Reections between Feed and Reector
O. A. Iupikov, R. Maaskant, M. Ivashina, A. Young, and P.S. Kildal
Abstract
Several electrically large Phased Array Feed (PAF) reector systems
are modeled to examine the mechanism of multiple reections between
parabolic reectors and low- and high-scattering feeds giving rise to frequency-
dependent patterns and impedance ripples. The PAF current is expanded
in physics-based macro domain basis functions (CBFs), while the reector
employs the Physical Optics (PO) equivalent current. The reector-feed
coupling is systematically accounted for through a multiscattering Jacobi
approach. An FFT expands the reector radiated eld in only a few
plane waves, and the reector PO current is computed rapidly through a
near-eld interpolation technique. The FEKO software is used for several
cross validations, and the convergence properties of the hybrid method
are studied for several representative examples showing excellent numer-
ical performance. The measured and simulated results for a 121-element
Vivaldi PAF, which is installed on the Westerbork Synthesis Radio Tele-
scope, are in very good agreement.
1 Introduction
Focal plane arrays can be used to form multiple reector beams covering a wide eld-
of-view (FoV) and large bandwidth. Among these feeds, one can distinguish between
a cluster of horns yielding one beam per feed [1, 2], and the more densely packed
beamforming array antennas commonly referred to as Phased Array Feeds (PAFs)
capable of providing a continuous FoV of simultaneous beams. Examples that ben-
et from these technologies are radars and terrestrial communications; while since
recently, PAFs have also been developed for astronomical and geoscientic instru-
ments, as well as for commercial satellite communication terminals [36]. Thanks to
their electronic beamforming capabilities, these new systems potentially enable much
faster studies of the Earth and Space than currently possible and are an attractive
alternative to bulky mechanically beam steered antennas.
The characterization of feeds in unblocked reectors and on-axis beams can be
handled by the traditional spillover, illumination, polarization and phase sube-
ciency factors dened for rotationally symmetric reectors in [7], and be extended to
113
Paper C. Fast and Accurate Analysis of Reflector Antennas with Phased...
Figure 1: Reector antenna and a Phased Array Feed (PAF) system.
include excitation-dependent decoupling eciencies of PAFs [8, 9]. The present paper
investigates the eects of aperture blockage and multiple reections on the system
performance in a more generic fashion than in [10] and [11] for rotationally symmetric
antennas.
An accurate analysis of these PAF systems, which include an array of many

closely-spaced antenna elements and an electrically large reector (see e.g. Fig. 1),
requires a modeling approach for the entire feed-reector structure accounting for
the array mutual coupling and the multiple scattering eects between the reector
and the feed, whose aperture diameter can be in the order of several wavelengths for
multi-beam applications [12, 13]. These eects give rise to a ripple in the antenna
impedance and radiation characteristics over frequency leading to impedance mis-
match eects and a periodically perturbed beam shape [1418]. The level of these
variations depends on several factors related to the reector geometry and feed de-
sign, among which the blockage area of the reector aperture caused by the feed, the
antenna array scattering characteristics [19, Sec. 2.2], the weighting coecients of the
beamforming network, and the presence of the (metal) structure in the vicinity of
the feed [20]. In order to solve these challenging problems, a method is needed that
is fast and physically-insightful for understanding how the EM coupling mechanism
between the PAF and reector antenna impacts the overall system performance.
During the last decades, a number of analytical and numerical techniques have
been developed to model feed-reector interaction eects. For example, in [10] the
multiscattered eld is approximated by a geometric series of on-axis plane wave (PW)
eld scattered by the antenna feed due to an incident PW at each iteration, where
the amplitudes of these PWs are expressed analytically for a given reector geometry.
This method is very fast and insightful, while MoM-level accuracy can be achieved
114
2. Iterative CBFM-PO Formulation
for single-horn feeds, but not for array feeds as demonstrated in this paper. An
alternative approach is to use more versatile, though more time-consuming, hybrid
numerical methods combining Physical Optics or Gaussian beams for the analysis of
reectors with MoM and/or Mode Matching techniques for horn feeds [21, 22]. The
recent article [23] has introduced the PO/Generalized-Scattering-Matrix approach
for solving multiple domain problems, and has shown its application to a cluster of
a few horns. This approach is generic and accurate, but may require the lling of a
large scattering matrix for electrically large PAFs and/or multifrequency front-ends
(MFFEs) that often have an extended metal structure [17]. Other hybrid methods,
which are not specic for solving the present type of problems, make use of eld
transformations, eld operators, multilevel fast multipole approaches (MLFMA), and
matrix modications [2427].
Recently, a Krylov subspace iterative method has been combined with an MBF-
PO approach for solving feed-reector problems [28], and complementary to this,
an iteration-free CBFM-PO approach has been presented by Hay, where a modied
reduced MoM matrix for the array feed is constructed by directly accounting for the
reector [16].
Among the above methods, the iterative methods have shown to be most useful
for gaining insight in the feed-reector multiscattering eects. In the present paper,
we therefore employ the Jacobi iterative approach as a simplied version of the full
orthogonalization method (FOM [28]), and combine it with an CBFM-PO approach
enhanced by eld expansion (see also [18]) and interpolation techniques. The method
is shown to converge within a few iterations.
The paper is arranged as follows: rst, the numerical approach is formulated and
then validated through a few representative examples, after which the eld expan-
sion and interpolation techniques are described along with a numerical accuracy and
eciency assessment; second, the performance and the multiscattering mechanism be-
tween electrically large reector antennas and several fundamentally dierent types
of feeds, including single-pixel horn feeds as in practical MFFEs, and 121-element
PAFs of dipoles and tapered slot Vivaldi antennas are studied for dierent port ter-
mination schemes. The predicted system sensitivity is in very good agreement with
the measurements of a single horn and Vivaldi PAF system feeding one of the 25-m
Westerbork Synthesis Radio Telescope reector antennas [29].
2 Iterative CBFM-PO Formulation

The below proposed iterative CBFM-PO approach is based upon the Jacobi method
for solving a system of linear equations in an iterative manner [30, 31].
Suppose the Method of Moments (MoM) matrix equation of the entire antenna
115
system comprised of both the parabolic reector and the antenna feed is given by
ZI = V, (1)
where the elements of the K ×K MoM matrix Z and K ×1 excitation vector V are
computed as
Zpq = hfp , E s (fq )i, Vp = −hfp , E i i (2)
for p, q = 1, 2, . . . K .
Furthermore, fp,q are the K basis/test functions for the
current/eld (Galerkin method); E
i,s is the incident/scattered electric eld, and
RR
ha, bi = Sa ∩Sb
[a · b] dS is the symmetric product, where Sa and Sb are the sup-
ports of the vector functions a and b, respectively. The expansion coecient vector
T
is given by I = [I1 , . . . IK ] , where T denotes the transposition operator.
To allow for a multiscattering analysis between the feed and reector, the MoM
matrix equation in (1) is rst partitioned into matrix blocks as
Z Zrf Ir
rr r
V
f = (3)
fr
Z Z
I Vf
where Zrr Z are the MoM matrix self-blocks of the reector and feed, re-
and
1 r f
spectively , and V and V are the corresponding excitation vectors. The matrix
rf fr T
Z = (Z ) contains the mutual reactions involving the basis functions on the feed
r
and reector. The unknown current expansion coecient vectors are denoted by I
f
and I . Next, Eq. (3) is written as
rr r r
0 Zrf

Z 0 I V
+ f = . (4)
0 Z Zfr
0 I Vf
[Zrr , 0; 0, Z ]
−1
Upon multiplying both sides by , the nal solution for the combined
problem can be obtained as
r rr !−1 r
0 Zrf
−1
I 1 0 Z 0 I0
= + . (5)
If 0 1 0 Z
Z fr
0 If0
where 1
is the identity matrix, and where the initial expansion coecient vector for
r rr −1 r f −1 f
the reector current I0 = (Z ) V , while for the feed current I0 = (Z ) V . These
initial currents are obtained by solving the reector and antenna feed problems in
isolation. It is observed that Eq. (5) is of the form
I = 1 + (Zd )−1 Zo
−1
I0 (6)
1 Here Z includes the eect of the antenna port terminations [32].
116
where
rr
0 Zrf

d Z 0 o
Z = and Z = fr . (7)
0 Z Z 0
P∞ n
Upon using the matrix equivalent of the scalar innite geometric series n=0 r =
−1
(1 − r) , where |r| < 1 for the series to converge, Eq. (6) can be rewritten in terms
of the innite series
∞
−(Zd )−1 Zo I0
X n
I= (8)
n=0
def
where the spectral radius ρ((Zd )−1 Zo ) = max(|λi |) d −1 o
of the matrix (Z ) Z with
i
eigenvalues {λi } must be smaller than unity for the series to converge. The phys-
ical multiscattering interpretation of the geometric series in (8) is apparent when
expanding it as:
∞
− (Zd )−1 Zo I (Zd )−1 Zo I
2 X
I = I0 0 + 0 + ... = In (9)
n=0
where the last summation is supposed to add up successively smaller contributions for
the currents on the reector and antenna feed in order to converge. It is conjectured
d −1 o
that ρ((Z ) Z ) 1 for the practical reector antenna systems that we consider,
since most of the energy is radiated out after each iteration and where the feeds
have relatively small aperture areas (weak reector-feed coupling), so that the sum
converges within a few iterations (cf. Sec. 4.1 and 4.3). Finally, using (7), the innite
series summation in Eq. (9) can be written in the cross-coupled recursive scheme
Reector Feed
∞ ∞
r r If = Ifn
X X
I = In (10a) (11a)
n=0 n=0
r rr −1 rf f
In+1 = −(Z ) Zfr Irn
f −1
In+1 = −(Z ) Z In (10b) (11b)
Ir0 = (Zrr )−1 Vr0 (10c) If0 = (Z ) Vf0

−1
(11c)
where Vr0 = Vr and Vf0 = Vf are the initial excitation voltage vectors of the reector
and the feed, respectively (in transmit situation Vr0 = 0).
The cross-coupled recursive scheme as formulated by Eqs. (10) and (11) is ex-
emplied in Fig. 2 as a ve-step procedure, in which the problem is rst solved in
isolation to obtain Ir0 and If0 . Afterwards, the feed current If0 is used to induce the
r
reector current I1 , which is then added up to the initial reector current. Likewise,
the initial reector current Ir0 is used to induce the feed current If1 , which is then
117
Step (i) V Step (ii)
Zload
If0
Transmit case:
Ir0 = 0 Ir1
Step (iii) Step (iv)
If1
Ir2
Zload
Step (v)
If = If0 + If1 + If2 + . . .
Ir = Ir0 + Ir1 + Ir2 + . . .
Figure 2: Illustration of the cross-coupled iterative scheme for multiscattering analysis of the feed-
reector interaction eects, as formulated by Eqs. (10) and (11): (i) The antenna feed radiates in
the absence of reector; (ii) the radiated eld from feed scatters from the reector; (iii) the scattered
reector eld is incident on the terminated feed and re-scatters; (iv) the re-scattered eld from the
feed is incident on the reector; etc. (v) the nal solution for the current is the sum of the induced
currents.
118
added to the initial feed current, and so forth. It is pointed out that this recursive
scheme can be used for any pair of radiating and/or scattering objects, provided that
the system is weakly coupled due to radiation and/or dissipation losses in order
to obtain a convergent solution.
Rather than computing the reector and feed currents through the large-size MoM
matrix blocks Zrr , Zrf , Zfr , and Z , additional computational and memory ecient
techniques can be used for the rapid computation of these currents at each iteration;
we propose to employ the Physical Optics (PO) current on the reector and invoke
the Characteristic Basis Function Method (CBFM, [33]) as a MoM enhancement
technique for computing the current on the feed.
Ifn+1 = (Z ) Vfn , where
−1
Note that (11b) represents the MoM matrix solution
Vfn = −Zfr Irn is the voltage excitation vector of the feed at iteration n. Hence, one
fr f
can obviate the construction of the large matrix Z by directly computing Vn . This
i,f
is done through testing the incident electric eld En (r) by the P basis functions
{fpf }Pp=1 supported by the feed, i.e.,
Ifn+1 = − Z hEni,f , f1f i, hEni,f , f2f i, . . . , hEni,f , fPf i

−1 T
(12)
where Eni,f is taken equal to the E -eld radiated by the PO current Jrn on the reector,
i,r
which is directly known through the reector incident H -eld Hn , so that there is
r
no need to compute the basis function coecients In explicitly.
For electrically small triangular cells on the reector surface (with edge length
< 0.2λ), the smoothly-varying PO current can be considered constant over each cell,
so that the electric eld produced by the q th reector triangle at the pth observation
point, E i,f
n,pq , can be computed through the near-eld formula for an incremental
electric current source, i.e. [34, p. 102],
−jηk e−jkrpq
E i,f
n,pq = [C1;pq `n,q − C2;pq (`n,q · r̂pq )r̂pq ] (13)
4π rpq
where
1 1
C1;pq = 1 + − , C2;pq = 3C1;pq − 2, (14)
jkrpq (krpq )2
and where the dipole moment is computed as
r A ,
`n,q = Jn,q with Aq the area of q th
q
reector triangle (q = 1, 2, . . . , Q). Hence, by using the expression for the PO current
for
r
Jn,q [35, p. 343], we nd that
`n,q = 2Aq n̂q × Hni,r (rqr ), (15)
where rqr ∈ S q th triangle on the reector surface S (cf. Fig. 3);

is the centroid of the
n̂q is the normal to the reector surface of the q th triangle, and Hni,r is the incident
H -eld generated by the feed current at iteration n. Using (15) and (13), the incident
119
E -eld in (12) is readily computed as

i,f =
En,p
PQ i,f .
En,pq The computation of (11b)
q=1
can be further accelerated as explained in Sec. 3.1.
Once Vfn is known, the current on the feed Ifn+1 at the next iteration can be
computed through solving the linear system of equations Z Ifn+1 = Vfn . For complex-
shaped and electrically large antennas, such as the wideband tapered slot antenna
array feeds [13], it becomes necessary to use both memory- and time-ecient methods,
such as the CBFM. The CBFM solves the current Ifn+1 through the following set of
equations:
 f CBF CBF
 In+1 = J In+1
CBF CBF CBF
I =Z Vn , (16)
 n+1
CBF CBF T f
Vn = (J ) Vn
where ZCBF = (JCBF )T Z JCBF is the CBFM-reduced MoM matrix of the feed; JCBF =
[J1 |JCBF
CBF
2 | . . . |JCBF
L ] is the column-augmented matrix of Characteristic Basis Func-
CBF
tions (CBFs), i.e., Jl is the set of CBFs (pre-dened expansion coecient vectors)
on the l th macro domain of the feed, and l = 1 . . . L, where L is number of macro
domains on the feed. Specic details on the generation of CBFs can be found in [33],
where the feed is analyzed as a phased array antenna in the absence of the reector.
Also, it is worth pointing out that the computation of ZCBF (i.e. the CBF coupling
terms) is performed in a time-ecient manner through utilizing the Adaptive Cross
Approximation (ACA) algorithm [36].
3 Acceleration of the Field Computations

The above-described iterative CBFM-PO approach requires the eld to be computed
at numerous points on both the feed and the reector surfaces, thereby rendering the
eld computations inecient, in particular for complex-shaped electrically large feed
antennas employing hundreds of thousands of low-level basis functions. Similarly,
one has to cope with a computational burden when calculating the PO equivalent
current on electrically large reectors.
However, it has been shown that the PO radiated eld for on-axis beams can be
approximated rather accurately through a single plane wave (PW) eld [10, 37]. This
observation opts for employing a Plane Wave Spectrum (PWS) to speed up the eld
computations [3840]. In fact, the on-axis PW corresponds to the Geometrical Optics
(GO) contribution of the PO-radiated eld (originating from the stationary phase
point), as will be demonstrated in Sec. 3.1, while the higher-order PWs are needed
to model the edge-diracted elds from the rim of the reector, which are associated
with the end-point contributions of the PO current in the radiation integral.
Furthermore, one can accelerate the computation of the PO current itself by using
an interpolation technique of the near-eld antenna feed pattern as detailed below.
120
3. Acceleration of the Field Computations
xmax
r xfmax
Ei,f ẑ
P ŷ
d
x̂
z=0 kxmax x̂ ∆y
∆x
n̂
Figure 3: The FFT-enhanced PWS expansion method for the fast computation of the feed current
due to the E -eld from the reector. Firstly, the incident eld E i,f is sampled in the xy plane P
in front of the feed in order to obtain the sampled PWS A(kx , ky ); Secondly, each spectral PW
propagates to an observation point r on the feed where E i,f is tested to compute the induced feed
current.
3.1 Plane Wave Spectrum Expansion FFT

With reference to Fig. 3, a grid of sampling points in the xy -plane P in front of the
feed at z = 0 is chosen for the expansion of the PO radiated eld in terms of a PWS.
Each PW propagates to a specic observation point r on the feed where the eld E i,f
is tested. This process of eld expansion and PW propagation is realized through
the application of the truncated Fourier Transform pair [38]
Zymax Zxmax
1
A(kx , ky ) = E i,f (x, y, z = 0)ej(kx x+ky y) dx dy (17a)
2π
−ymax −xmax
max max
Zkx Zky
1
E i,f (r) = A(kx , ky )e−jkz z e−j(kx x+ky y) dkx dky (17b)
2π
−kxmax −kymax
where
k 2 > kx2 − ky2
p 2
kp− kx2 − ky2 if
kz = , (18)
−j kx2 − ky2 − k 2 otherwise.
and where the spectrum of PWs is limited to only those that are incident on the feed
from directions within an angle subtended by the reector and seen from the center
121
of the plane P (see Fig. 3); hence, the maximum wavenumbers kxmax and kymax in (17b)
are chosen to be equal to
  
F

8
kxmax = kymax = k sin tan−1  h D i  (19)
F 2 F −1

16 D
1− d
−1
where k = 2π/λ is the free-space wavenumber; F and D are the focal distance and
diameter of the parabolic reector, respectively; and d is the distance between the
plane P and the geometrical focal plane of the reector. Since the maximum spectral
components kxmax and kymax are known, the minimum step size ∆x and ∆y for the
spatial sampling of the eld is found from Nyquist's sampling theorem:
∆x = π/kxmax , ∆y = π/kymax . (20)
Furthermore, if (17) is evaluated through a Fast Fourier Transform (FFT), the dis-
cretely sampled eld functions are periodic in both the spatial and frequency domains.
To minimize the eld artifacts that are associated with this periodicity,xmax and ymax
f
must be chosen suciently large, that is, at least equal to the maximum size xmax and
f
ymax of the feed coordinates. The examination of how the error of the feed current
depends on xmax and ymax is presented in Sec. 4.2.
As a result, the total number of sampling points in the x and y directions are Nx =
2xmax /∆x and Ny = 2ymax /∆y , respectively, and the spectral spacings and the spatial
extents are related through ∆kx = 2kx
max /N = π/x max
x max and ∆ky = 2ky /Ny =
π/ymax .
3.2 Near-Field Interpolation

While the previous section describes how the PWS-expanded E -eld from the re-
ector accelerates the computation of the induced feed current, this section explains
how the reector incident H -eld can be computed for the rapid determination of
the induced PO current. For this purpose, the radiated H -eld from the feed is rst
computed at a coarse grid on the reector surface (white circles in Fig. 4), after
which the eld at each triangle is determined on the reector (yellow square mark-
ers) through an interpolation technique. This interpolation technique de-embeds the
initially sampled eld to a reference sphere with radius R whose origin coincides with
the phase center of the feed to assure that the phase of the de-embedded eld will be
slowly varying. Consequently, relatively few sampling points are required for the eld
interpolation, after which the interpolated elds are propagated back to the reector.
In summary, and with reference to Fig. 4, the H -eld interpolation algorithm for
determining the reector PO current
1. Denes a grid on the reector surface (white circles) for computing the H -eld.
122
initial eld sampling points

feed phase center
de-embedded eld points
interpolation points
R nal eld testing points
∆θ
Hqsph
dm sph
Hm dq
Hm H i,r (rqr )
Figure 4: The near-eld interpolation technique for the rapid determination of the induced PO
current on the reector.
2. De-embeds the H -eld to a reference sphere around the feed phase center (green
points):
sph = H d ejkdm ,
Hm (21)
m m
where dm is the distance between the reector surface and the sphere of radius
R along the line connecting the mth sample point on the reector and the feed
phase center.
3. Computes the elds on the sphere in the same directions as the reector triangle
centroids are observed (blue square markers) through interpolating the elds
at the adjacent (green) points.
4. Propagates the eld to the reector surface; that is, at the q th triangle, the
H -eld
H i,r (rqr ) = Hqsph d−1

q e
−jkdq
. (22)
5. Computes the reector PO current by using (15).
Sec. 4.2 examines the error in the reector current as a function of the sample grid
density, in addition to the improvement in computation time that this method oers.
4 Numerical Results
In this section, we start with the validation of the proposed iterative MoM-PO ap-
proach for a relatively strongly coupled feed-reector system, comprised of a small
123
(a) (b)
(c) (d)
Figure 5: Considered feed geometries (in addition to the dipole feed with PEC ground plane): (a) a
classical pyramidal horn with aperture length ∼1λ; (b) the same horn but with extended ground
plane (∼3.7λ), where the ground plane may model the presence of a large feed cabin; (c) an antenna
array consisting of 121 0.45λ-dipoles above a ground plane of the same size; (d) the same array, but
with the dipoles replaced by wideband tapered slot Vivaldi antennas.
reector (D = 14λ) fed by a dipole antenna over a ground plane for which we examine
the convergence rate of the solution for the antenna input impedance. Furthermore,
we validate the frequency-dependent radiation characteristics of a dipole array feed
through the commercially available software FEKO [41]. Afterwards, a relative error
analysis of the antenna transmit characteristics is performed when the acceleration
techniques in Sec. 3 are utilized. Finally, a more practical study is carried out, where
the impact of the feed-reector coupling on the performance of the antenna reector
system for dierent types of low- and high-scattering feeds is analyzed and discussed.
For the latter study, two parabolic reectors with diameters D = 38λ and 118λ are
considered, in conjunction with the four types of feeds that are shown in Fig. 5. It
is shown that the measured and simulated results for a 121-element Vivaldi PAF,
which is installed on the Westerbork Synthesis Radio Telescope, are in very good
agreement.
The MoM computations have been carried out on a 64-bit openSUSE Linux server
(kernel version: 2.6.37.6-0.20-desktop), equipped with 144 GB of RAM and two quad-
core Intel(R) Xeon(R) E5640 CPUs, each operating at 2.67 GHz. The FEKO Suite
6.0 EM solver runs on an Ubuntu Linux server (kernel-release: 2.6.32-21-server),
equipped with a Dual Core AMD Opteron Processor 275 at 2.2 GHz with 16 GB of
124
−14
−16
−18
Γact, [dB]
−20
MoM (2λ ground plane)
−22
MoM−PO (2λ ground plane)
−24 MoM (1λ ground plane)
−26 MoM−PO (1λ ground plane)
1 2 3 4 5 6 7 8 9
Number of iterations
(a)
10
0
Gain, [dBi]
−10
MoM (w/o interaction)

−20
MoM (with interaction)
MoM−PO (2 iterations)
−30
−40
−150 −100 −50 0 50 100 150
θ, [deg]
(b)
Figure 6: The convergence of the feed radiation characteristics in the presence of the reector as a
function of the number of Jacobi iterations, in terms of: (a) the dipole input reection coecient,
and; (b) the dipole illumination pattern at 1 GHz (ground plane size is 2λ × 2λ). The convergence
as a function of the dipole load impedance is analyzed for a dipole antenna array feeding a 38λ
reector.
RAM.
4.1 Validation of the Iterative Approach

For validating the implemented iterative MoM-PO approach, a relatively small reec-
tor (D = 14λ, F/D = 0.35) fed by a 0.5λ-dipole spaced 0.25λ above an 1λ × 1λ and
a 2λ × 2λ PEC ground plane has been simulated, both by the proposed iterative and
plain MoM approach. The dipole reection coecient as a function of the iteration
125
−10 0
MLFMM (FEKO), array only
feed−reflector
−11 MLFMM (FEKO), with reflector interaction effect
−10
CBFM, array only
−12 CBFM−PO, with reflector
Gain, [dB]
Γact, [dB]
−20
−13
−14 −30 MLFMM (FEKO), 1.405 GHz

CBFM−PO, 1.405 GHz
−15 MLFMM (FEKO), 1.435 GHz
−40
CBFM−PO, 1.435 GHz
−16
1.395 1.4 1.405 1.41 1.415 1.42 1.425 1.43 1.435 1.44 1.445 −8 −6 −4 −2 0 2 4 6 8
Frequency, [GHz] θ, [deg]
(a) (b)
500 18
200
100 Opt 16
Imaginary part of Z load, [Ohm]
50
20 14
10
5
3 12
0
−3
−5
10
−10
−20
−50 8
−100
−200 6
−500
3 5 10 20 50 100 200 500 1k
Real part of Z load, [Ohm]
(c)
Figure 7: (a) The magnitude of the active reection coecient of the most excited antenna array
dipole element feeding a 38λ reector as a function of frequency, and; (b) reector antenna radiation
pattern, simulated in FEKO (MLFMM) and using the described iterative CBFM-PO approach; (c)
the number of required iterations for reaching convergence (error in feed current less than 0.5%).
Interesting fact: the round marker indicates the impedance that maximizes the decoupling eciency
(=power-matched case) when the array feed is used as a broadside-scanned aperture array, which
also happens to coincide with the minimum number of iterations (=low multiscattering eect).
count is shown in Fig. 6(a). Even though the feed-reector coupling is relatively large
due to a relatively large blockage area of the high-scattering feed, convergence of the
impedance down to 0.1% relative error level, measured as a change between the last
two iterations, is seen to occur within 5 and 9 iterations for the 1λ × 1λ and 2λ × 2λ
PEC ground planes, respectively. This error n at iteration n is computed as
 
s .sX
X
n =  |Iin − Iin−1 |2 |Iin |2  × 100%. (23)
i i
The small residual error of order 1% is a result of the PO-approximated reector

current. Fig. 6(b) shows how the forward gain of the dipole illumination pattern
changes due to the feed-reector coupling as the number of iterations increases.
126
−10
−20
Error in currents, dB
−30 Only
dominant
PW is used
−40
−50 Dipole array Reflector

Horn with ext. ground plane edge
−60 Horn
PWS
Horns Max. array or is used
−70
aperture ext. ground plane
size dimension
−80
1 2 3 5 7 10 20 30 40
Sample plane size, λ
(a)
−70 −60 −50 −40 −30 −20 −10 0
dB
0 0
30 30
−10 −10
20 20
−20 −20
10 10
−30 −30
ky
0
ky
0
−40 −40
−10 −10
−50 −50
−20 −20
−60 −60
−30 −30
−70 −70
−20 0 20 −20 0 20
kx kx
(b) (c)
Figure 8: (a) The relative error in induced feed currents [cf. (24)] as a function of the FFT sampling
plane size P; (b) the magnitude of the spatial frequency spectrum |Aco (kx , ky )| (i.e. plane wave
spectrum) for the 38λ reector fed by the dipole array in case the FFT grid size is equal to size of
the feed, and (c) when it is eight times the feed size.
For cross-code validation purposes, a larger and more complex 38λ reector
(F/D = 0.35) fed by an 121-dipole array feed has been analyzed [cf. Fig. 5(c)], both
by the proposed iterative approach and the commercial FEKO software. Fig. 7(a,b)
demonstrates a good agreement between the reector antenna radiation patterns (in-
cludes the feed blockage eect) and the magnitudes of the computed active reection
coecients as a function of frequency, where the frequency interval ∆f of the os-
cillation period is consistent with the electrical distance between the feed and the
127
reector vertex, i.e., ∆f = c/2F . Here, the optimal port termination that maximizes
the array decoupling eciency [8] was found through Matlab's fminsearch uncon-
strained nonlinear optimization routine (Nelder-Mead simplex direct search method)
and was found to be 147 + 45.6j Ω. Thus far, practical PAF antenna elements have
been optimized in phased array mode, broadside scan, using periodic boundary con-
ditions in EM simulation software [42], hence, here too, the co-polarized elements
of the array feed are excited in-phase to determine the optimal port loading. This
optimal impedance is marked on Fig. 7(c) (and Fig. 11), where its plot shows the
number of iterations required to obtain an error in the dipole array feed current
between the two last iterations less than 0.5% as a function of the array loading.
Note the interesting fact that the minimum number of iterations (=lowest multiscat-
tering eect) occurs when the array is optimally loaded (=power matched), which is
in accordance with our expectations, and this applies even though the antenna load
impedance has been found for the aperture-array-excited case.
4.2 Field Approximation Errors

Sec. 3.1 and 3.2 describe a eld expansion and interpolation technique for acceler-
ating the feed-reector interaction computations, respectively. In this section, we
analyze the reector induced feed current when the eld from the reector is ex-
panded in terms of a truncated spectrum of plane waves, and compute the error in
the feed current relative to a direct full-wave solution where the number of eld
modes radiated by the reector equals the number of reector triangles (=number
of incremental dipole sources on the reector). The distance d between the feed and
the sampling plane P (cf. Fig. 3) has been chosen equal to 0.5λ in all PWS com-
putations; in fact, our study shows that the selection of d in the range of 0.1 . . . 5λ
has a negligible (< 0.7%) eect on the antenna characteristics, such as the aperture
eciency, even when the size of the plane P is kept the same. Both the relative
error of the feed induced PO-reector current and how the near-eld interpolation
grid density aects this error will be analyzed afterwards. Furthermore, the errors in
the feed and reector currents, as well as those in the gain of the antenna reector
system and the input impedance of the feed, will be summarized in a table.
The relative error between vector (or matrix) quantities such as between the cur-
rent expansion coecient vectors Iapprox and Iref for the iterative CBFM-PO solution
with and without eld approximations, respectively is computed as
 
s s
|Iiref − Iiapprox |2 |Iiref |2  × 100%,
X X
1 =  (24)
i i
while the relative error for scalar functions (antenna gain, impedance characteristics,
128
Table 1: Errors due to applying the eld approximations, %

Feed Reector Gain Gain
surface surface (on-axis) (@−3 dB) Impedance
current current

PWS
0.09 0.02 0.11 0.03 0.09 0.03 0.07 0.02 0.16 0.04
approx.
NFI
0.01 <0.01 0.06 0.06 0.05 0.04 0.01 0.02 0.01 <0.01
approx.
Both
0.09 0.02 0.13 0.07 0.13 0.07 0.07 0.04 0.15 0.04
approx.

PWS
0.28 0.02 0.41 0.02 0.06 0.01 0.09 0.01 0.44 0.04
approx.
NFI
0.3 0.01 1.01 0.16 0.16 0.07 0.37 0.07 0.52 0.02
approx.
Both
0.53 0.03 1.02 0.16 0.15 0.08 0.34 0.07 0.88 0.05
approx.

PWS
0.05 0.02 0.1 0.02 0.03 0.01 0.01 0.01 0.03 0.01
approx.
NFI
0.02 0.01 0.21 0.20 0.09 0.07 0.12 0.13 0.02 0.01
approx.
Both
0.06 0.02 0.23 0.21 0.10 0.07 0.13 0.14 0.05 0.02
approx.
Table 2: Total simulation time (for D = 118λ reector)

gnd plane array
MoM-PO, no 192 min 801 min 3906 min
70 min (100%)
approx. (100%) (100%) (100%)
27 min 63 min 190 min 1312 min
PWS approx.
(39.0%) (32.9%) (23.8%) (33.6%)
57 min 152 min 548 min 2108 min
NFI approx.
(81.3%) (79.4%) (68.5%) (54.0%)
13 min
Both approx. 17 min (9.0%) 16 min (2.0%) 33 min (0.9%)
(19.2%)
etc.), is computed as
2 = |Aref − Aapprox | |Aref | × 100%.

(25)
Fig. C.8(a) illustrates the relative error in the feed surface current as a function of
the FFT sampling plane size when the PWS is employed for expanding the reector
129
Horn with ext. ground plane Dipole array Horn
0 0
∆φ = 2.5 deg ∆θ = 2.5 deg
−20 −20
Error, [dB]
Error, [dB]
−40 −40
−60 −60
−80 −80
0 2 4 6 8 10 0 2 4 6 8 10
∆θ, [deg] ∆φ, [deg]
(a) (b)
Figure 9: The interpolation error in the 38λ reector current as a function of (a) the sampling step
∆θ, and (b) the sampling step ∆φ of the near elds of the feed.
radiated eld (for PWS parameters see Sec. 3.1), and when only the dominant on-
axis PW term is used. As expected, the error decreases for an increasing sampling
plane size, since more spectral PW terms are taken into account while the eect of the
FFT-related periodic continuation of the spatial aperture eld decreases. Henceforth,
we choose the sampling plane size equal to that of the feed, for which the feed
current error is about −35 dB for all the considered feeds, while it represents a good
compromise from both a minimum number of sampling points and accuracy point of
view. Conversely, if only the dominant on-axis PW term is used to approximate the
reector eld, the error increases when the plane P becomes larger. This is due to
the tapering of the reector scattered eld which becomes more pronounced when the
plane size P increases, so that the PW amplitude A(kx , ky ) is underestimated when
using the eld averaging in (17a) for kx = ky = 0, as opposed to the direct on-axis
point sampling method that has been presented in [10].
Note that the magnitude of the co-polarized spatial frequency spectrum |Aco (kx , ky )|
in Figs. 8(b) and (c) exhibit several interesting features; as expected, the dominant
spectral component corresponds to the on-axis PW, for which kx = ky = 0, while the
second strongest set of PWs originate from the rim of the reector, as observed by
2 2 max )2 = (k max )2 .
the spectral ring structure for which kx + ky = (kx y
Regarding the interpolation method for the radiated near-elds of the feed (Sec. 3.2),
Figs. C.9(a) and (b) show that the error in the resulting induced reector current
depends on the angular step size ∆θ and ∆φ of the initial eld sampling grid (before
interpolation). As expected, the error increases when the sampling grid coarsens.
Furthermore, the error is larger for larger feeds, especially for high-scattering ones,
for which the scattered elds (i.e. 2nd iteration and further) vary more rapidly than
130
for smaller low-scattering antennas for which a coarser grid can be applied.
Table 1 summarizes the relative errors in both the currents and relevant antenna
characteristics, while Table 2 shows how the simulation time of a plain iterative
CBFM-PO (or MoM-PO) approach reduces when the eld approximations of Sec. 2
are used. Note that the PWS approximation leads to a small relative error in the
surface current of the high-scattering feed for the 38λ reector, i.e. 0.28%, while
if only a single on-axis PW is used, the relative error is found to be two orders
larger [37]. It is also observed that, when applying the eld approximations for both
the reector and feed, the error in the considered antenna characteristics remains less
than 1%, while the computational speed advantage is signicant, i.e., a factor 5 to
100, depending on the reector size and feed complexity.
4.3 Feed-Reector Antenna System Performance Study

The performance of several reectors fed by low- and high-scattering feeds is studied
in detail in this section. It is shown how the frequency ripple in the antenna radi-
ation characteristics is formed and how the feed termination aects the magnitude
of this ripple. The system performance and pros and cons of the dierent feeds are
summarized in a table and discussed from a multiscattering point of view.
Fig. 10 illustrates the level of the total (including feed-reector interaction) and
the scattered eld distributions in the aperture of a 38λ reector fed by the horn
with an extended ground plane, the dipole array, and the Vivaldi array, for both the
short-circuited (left column) and the power-matched (right column) loading schemes.
Although the short-circuited case is not very practical, it does showcase how two very
dierent loading scenarios aect the aperture eld variation, and how it depends on
the type of the feed. The two solid lines in each sub-gure show the extrema that the
aperture eld distribution attains within one period of the ripple's frequency interval
∆f = c/2F . The dashed lines show the aperture eld due to the scattered eld of the
feeds. Clearly, for array feeds, the aperture eld distribution is strongly dependent on
the antenna port termination; the re-scattered elds from the array feeds aect the
aperture eld distribution signicantly when the antenna ports are short-circuited,
as opposed to the power-matched array feeds, whose scattered elds are signicantly
weaker. Note the dierences in results for the horn with extended ground plane, for
which the dominant part of the scattered eld is primarily attributed to the metallic
ground plane, while the impedance mismatch of the horn itself has only a minor eect
(i.e. the residual component of the Radar Cross Section is large, but the antenna
component is small [19]).
From the above analysis, one concludes that more Jacobi iterations are required
to reach convergence for feeds that are poorly impedance matched as they tend to
scatter a larger portion of the incident eld (stronger multiscattering eects). It is
therefore likely that the number of Jacobi iterations is closely related to the magnitude
131
f @ min ηap f @ max ηap 2nd iteration (f @ max ηap)
0 0
|E|, [dB]
|E|, [dB]
−20 −20
−40 −40 PM
SC
−4 −2 0 2 4 −4 −2 0 2 4
X, [m] X, [m]
(a) Horn with ground plane
0 0
|E|, [dB]
|E|, [dB]
−20 −20
−40 SC −40 PM
−4 −2 0 2 4 −4 −2 0 2 4
X, [m] X, [m]
(b) Dipole array
0 0
|E|, [dB]
|E|, [dB]
−20 −20
−40 SC −40 PM
−4 −2 0 2 4 −4 −2 0 2 4
X, [m] X, [m]
(c) Vivaldi array
Figure 10: Distribution of the eld in the aperture of a 38λ reector fed by: (a) horn with extended
ground plane; (b) dipole array, and; (c) Vivaldi array. Left and right columns correspond to the
short-circuited (SC) and average power-matched (PM) feeds, respectively.
132
of the ripple on the antenna radiation characteristics; this fact is demonstrated in

Fig. 11, which shows the aperture eciency, mismatch factor [8, 43] and their ripples
as a function of the port termination impedance. The ripple Rν for a frequency-
dependent parameter ν(f ) is herein dened as
maxf [∆ν(f )] − minf [∆ν(f )]

Rν = × 100%, (26)
meanf νwith_coup (f )
where∆ν(f ) = νwith_coup (f ) − νno_coup (f ) is the dierence between the considered

parameter ν , with and without accounting for the feed-reector coupling. The con-
sidered frequency band is herein taken relatively narrow as it corresponds to one
period of the ripple only, i.e., ∆f = c/2F . We further point out that these results
apply to a feed that is excited at all its ports such as to realize a maximum gain
pattern of the combined feed-reector system, hereafter referred to as the Conjugate
Field Match (CFM) beamformer. Furthermore, to be able to compare the results
with the commonly employed uniformly excited array case analyzed above, the CFM
excitations are xed and determined only once for the optimal antenna port loading,
i.e., pertaining to the uniformly excited array.
Table 3: Maximum parameter dierence due to feed-reector coupling eect w.r.t. the cases when
no coupling is taken in account, %
Feed Reector Gain Gain
surface surface (on-axis) (@−3 dB) Impedance
current current
Pyramidal
7.9 2.5 4.2 1.3 2.0 0.6 4.0 2.2 15.1 4.7
horn
Horn with
ext. ground 23.2 3.5 65.1 11.9 19.2 3.4 29.4 3.6 43.4 6.1
plane
Dipole
13.8 4.2 3.2 0.8 1.8 0.3 3.7 0.7 5.8 1.7
array
Vivaldi
14.1 4.1 3.4 1.0 1.9 0.3 3.4 0.4 4.6 1.4
array
One concludes from Fig. C.11(a) that the aperture eciency is a function of the
antenna port loading, and that the impedance for which ηap attains a maximum
is close to the optimal power-match impedance found in Sec. 4.1 for the uniformly
excited array case. This apparently even holds in the absence of the feed-reector
interactions, in which case the array illumination pattern has changed slightly due
to perturbed array embedded element patterns while the CFM excitation coecients
remain unaltered. In Sec. 4.1 we maximized the decoupling eciency to nd the
optimal port loading. For the present CFM all-excited array case the decoupling
eciency reduces to the mismatch factor ηmis . The maximum of ηmis does, however,
not coincide with the earlier optimal load impedance primarily due to the dierence
133
Mean aperture efficiency, no interaction, [−] Aperture efficiency ripple, [%]

500 500 5
200 0.8 200

100 Max 100 Opt

50 50 Min ripple 4
20 Opt 0.78 20
10 10
5 5 3
3 0.76 3
0 0
−3 −3
−5 −5 2
−10 0.74 −10
−20 −20
−50 −50 1
−100 0.72 −100
−200 −200
−500 0.7 −500 0
20 50 100 200 500 1k 20 50 100 200 500 1k
Real part of Z load, [Ohm] Real part of Z load, [Ohm]
(a) Mean ηap (no interaction) (b) ηap ripple, %

Mean missmatch efficiency, no inter, [−] Missmatch efficiency ripple, [%]
500 500 10
200 200
Min ripple
0.8
100 Opt 100

50 50 8
20 0.7 20 Opt
10 Max 10
5 5 6
3 3
0 0.6 0
−3 −3
−5 −5 4
−10 0.5 −10
−20 −20
−50 −50 2
−100 0.4 −100
−200 −200
−500 0.3 −500 0
20 50 100 200 500 1k 20 50 100 200 500 1k
Real part of Z load, [Ohm] Real part of Z load, [Ohm]
(c) Mean ηmis (no interaction) (d) ηmis ripple, %
Figure 11: Eect of the antenna port loading on (a) the aperture eciency without feed-reector
coupling, and (b) aperture eciency ripple when the feed-reector coupling is present; (c),(d) the
same for the mismatch eciency. A 38λ reector is fed by the 121-element dipole array. The round
marker denotes the optimal load impedance that maximizes the decoupling eciency (cf. Sec. 4.1).
in array excitation schemes. Nonetheless, the observed quantities are only weakly
dependent on impedance variations around their maximums. As for the feed-reector-
induced ripple of ηap and ηmis [Fig. C.11(b) and Fig. C.11(d)], we can conclude that
the ηmis ripple is more sensitive to variations in the array loading relative to the
ripple in ηap . In practice, however, when the amplier/LNA impedance changes up
to 10-20%, this only weakly aect ηap and ηmis and their ripple.
Table 3 and 4 summarize the maximum dierence in mean values and ripple,
respectively, of several other relevant antenna radiation characteristics when the
feed-reector coupling is taken into account. For the computation of this dier-
ence Eq. (24) is used, where the superscripts ref and approx denote in this case
the considered antenna parameter after the 1st (no coupling) and nal iteration,
134
Table 4: System characteristics (and their ripple) over frequency band
38λ reector
Horn Horn + gnd Dipole array Vivaldi array
ηill 0.71 (7.2%) 0.67 (34.1%) 0.86 (1.0%) 0.92 (0.6%)
ηmis 0.992 (1.0%) 0.987 (5.1%) 0.830 (1.2%) 0.910 (0.9%)
Tsp 7.7 K (18%) 6.8 K (39%) 4.2 K (16.8%) 8.8 K (9.6%)
118λ reector
Horn Horn + gnd Dipole array Vivaldi array
ηill 0.71 (2.2%) 0.72 (4.1%) 0.85 (0.4%) 0.92 (0.2%)
ηmis 0.999 (0.2%) 0.999 (0.2%) 0.853 (0.5%) 0.926 (0.4%)
Tsp 7.7 K (6.0%) 7.2 K (6.8%) 3.8 K (5.7%) 8.7 K (3.4%)
Measured (Vivaldi array) Measured (Horn feed)

CBFM−PO (Vivaldi array) CBFM−PO (Horn feed)
80
75
Illumination efficiency, [%]
70
65
60
55
50
45
1.295 1.3 1.305 1.31 1.315 1.32
Frequency, [GHz]
Figure 12: Illumination eciencies of the 118λ reector antenna, either fed by the 121 Vivaldi PAF,
or the single-horn feed. The CBFM-PO simulated results are compared to the measured ones for
a 25 m reector antenna of the Westerbork Synthesis Radio Telescope [4]. Bottom of the gure: a
photo of the experimental PAF system placed at the focal region of the reector, and an image of
a smaller-scale PAF-reector model.
respectively, and where the summations are taken over frequency samples. Hence,
this table allows us to estimate how strong the feed-reector coupling is and how it
aects the antenna characteristics. As expected, the high-scattering horn feeds cause
stronger multiscattering eects, which is further excercebated for smaller dishes due
to the larger relative blockage area. The dierence in the antenna characteristics
and their ripples are largest for the case of the 38λ reector fed by the horn with
135
extended ground plane, while these values are comparable and weakly dependent on
the antenna element type in case of the array feeds.
Table 4 shows the mean values of various antenna radiation characteristics as
well as their ripple caused by the multiscattering phenomenon, where the reector
antenna is assumed to be pointed at zenith for the computation of the spillover noise
temperature Tsp . Upon comparing the values in the table, one conludes that the
spillover noise temperature Tsp is most sensitive to the feed-reector coupling, which
may be of importance in radio astronomy applications where high receiving sensitivity
is required.
Fig. 12 shows the illumination eciencies ηill of a 118λ reector antenna (D = 25
m, F/D = 0.35), either fed by the Vivaldi array feed, or a single horn antenna.
The numerically computed results are compared to measurements at the Westerbork
Synthesis Radio Telescope (WSRT) [4]. As one can see, the agreement is very good.
In the simulations, the size of the ground plane has been chosen equal to the size of
the feed cabin (≈ 1×1 m). The fact that ηill is higher for the array feed than for
the horn antenna nicely demonstrates the superior focal eld sampling capabilities of
dense phased array feeds. Furthermore, one can also observe a rather strong ripple
in ηill for the case of the horn feed with extended ground plane. This ripple is caused
by the relatively high feed scattering of the reector eld.
5 Conclusions
An FFT-enhanced Plane Wave Spectrum (PWS) approach has been formulated in
conjunction with the Characteristic Basis Function Method, a Jacobi iterative multi-
scattering approach, and a near-eld interpolation technique for the fast and accurate
analysis of electrically large array feed reector systems. Numerical validation has
been carried out using the multilevel fast multipole algorithm method available in
the commercially available FEKO software.
This physics-based numerical modeling oers the possibility to pull the feed-
reector interaction eects apart in a systematic manner and has demonstrated that:
(i) a relation exists between the number of Jacobi iterations and the magnitude of
the ripple on the frequency-dependent antenna radiation characteristics introduced
by the feed-reector coupling; (ii) the on-axis plane wave of the reector eld and
the ones originating from the reector rim are the strongest PWS components; (iii)
the reector-feed-induced ripple reduces when the array port termination is near a
power-matched situation; (iv) the array feeds demonstrate a higher illumination ef-
ciency than a single-horn feed with extended ground plane as a result of a better
synthesized illumination pattern, and; (v) the level of the ripple as a function of fre-
quency is smaller due to a smaller fraction of the scattered eld from the array feed.
The latter two ndings have also been observed in measurements [4] for a horn feed
and a 121-element Vivaldi PAF system installed at the Westerbork Synthesis Radio
136
References
Telescope (118λ-diameter), where we have shown that the relative dierence between
the simulated and measured antenna eciencies is only in the order of a few percent.
Acknowledgment
The authors thank Wim van Cappellen from ASTRON for providing the measurement
data of the APERTIF PAF system.
References
[1] B. Veidt. (2006) SKA memo 71: Focal-plane array architectures: Horn clusters
vs. phased array techniques. [Online]. Available: http://www.skatelescope.org/
uploaded/29162_71_Veidt.pdf


no. 8, pp. 15071521, Aug. 2009.
Westerbork synthesis radio telescope, in IEEE International Symposium on
Proc. Phased Array Systems and Technology (ARRAY), Boston, Oct. 2010, pp.
640647.
448449.

15.
137
[7] P.-S. Kildal, Factorization of the feed eciency of paraboloids and cassegrain
antennas, IEEE Trans. Antennas Propag., vol. 33, no. 8, pp. 903908, Aug.
1985.
[8] N. M. Kehn, M. V. Ivashina, P.-S. Kildal, and R. Maaskant, Denition of uni-

fying decoupling eciency of dierent array antennas case study of dense focal
plane array feed for parabolic reector, Int. Journal of Electronics and Com-
munications (AEU), vol. 64, no. 5, pp. 403412, May 2010.
[9] M. V. Ivashina, M. Kehn, P.-S. Kildal, and R. Maaskant, Decoupling eciency

of a wideband Vivaldi focal plane array feeding a reector antenna, IEEE Trans.
[10] A. Moldsvor and P.-S. Kildal, Systematic approach to control feed scattering
and multiple reections in symmetrical primary-fed reector antennas, IEEE
Trans. Antennas Propag., vol. 139, no. 1, pp. 6571, Sep. 1992.
[11] P.-S. Kildal, S. Skyttemyr, and A. Kishk, G/T maximization of a paraboloidal

reector fed by a dipole-disk antenna with ring by using the multiple-reection
approach and the moment method, IEEE Trans. Antennas Propag., vol. 45,
no. 7, pp. 11301139, Jul. 1997.
[12] S. G. Hay, J. D. O'Sullivan, and R. Mittra, Connected patch array analysis

using the characteristic basis function method, IEEE Trans. Antennas Propag.,
vol. 59, no. 6, pp. 18281837, Jun. 2011.
[13] R. Maaskant, M. V. Ivashina, O. Iupikov, E. A. Redkina, S. Kasturi, and D. H.

Schaubert, Analysis of large microstrip-fed tapered slot antenna arrays by com-
bining electrodynamic and quasi-static eld models, IEEE Trans. Antennas
Propag., vol. 56, no. 6, pp. 17981807, Jun. 2011.
[14] A. Popping and R. Braun, The standing wave phenomenon in radio telescopes,
frequency modulation of the wsrt primary beam, Astronomy and Astrophysics,
vol. 479, no. 3, pp. 903913, Mar. 2008.
[15] M. A. Apeldoorn, Characterizing reector-feed interaction for parabolic

reector antennas, Master's thesis, International Research Centre for
Telecommunications and Radar (IRCTR), Delft University of Technology, Delft,
The Netherlands, 2011. [Online]. Available: http://repository.tudelft.nl/assets/
uuid:d6605869-37e0-49a6-8bc4-817855ab3710/MSc-Thesis-MAApeldoorn.pdf
[16] S. Hay, R. Mittra, and N. Huang. (2010) Analysis of reector and feed
scattering and coupling eects on the sensitivity of phased array feeds. [Online].
Available: http://csas.ee.byu.edu/docs/Workshop/BYUSGH.pdf
138
References

[18] O. Iupikov, R. Maaskant, and M. Ivashina, Towards the understanding of the

interaction eects between reector antennas and phased array feeds, in Proc.
pp. 792795.
[20] N.-T. Huang, R. Mittra, M. V. Ivashina, and R. Maaskant, Numerical study of

a dual-polarized focal plane array (FPA) with vivaldi elements in the vicinity of
a large feed box using the parallelized FDTD code GEMS. in Proc. IEEE AP-S
International Symposium, Charleston, USA, Jun. 2009, pp. 15223965.
2006.
[22] M. Bandinelli, F. Milani, G. Guida, M. Bercigli, P. Frandsen, S. Sorensen,

B. Bencivenga, and M. Sabbadini, Feed-array design in presence of strong scat-
tering from reectors, in Proc. European Conference on Antennas and Propag.
(EuCAP), Rome, Italy, Apr. 2011, pp. 38443848.

2012.
[24] R. E. Hodges and Y. Rahmat-Samii, An iterative current-based hybrid method

for complex structures, IEEE Trans. Antennas Propag., vol. 45, no. 2, pp. 265
276, Feb. 1997.
[25] C. S. Kim and Y. Rahmat-Samii, Low prole antenna study using the physical
optics hybrid method (POHM), in Proc. IEEE AP-S International Symposium,
Ontario, Canada, Jun. 1991, pp. 13501353.
[26] J. M. Taboada and F. Obelleiro, Including multibounce eects in the moment-

method physical-optics (MMPO) method, Micr. Opt. Technol., vol. 32, no. 6,
pp. 435439, 2002.
139
[27] U. Jakobus and F. M. Landstorfer, Improved PO-MM hybrid formulation

for scattering from three-dimensional perfectly conducting bodies of arbitrary
shape, IEEE Trans. Antennas Propag., vol. 43, no. 2, pp. 162169, Feb. 1995.
[28] B. Andrés-Garciá, D. Gonzalez-Ovejero, C. Craeye, L. Garciá-Muñoz, and

D. Segovia-Vargas, An iterative MoM-PO method based on a MBF/Krylov
approach, in Proc. European Conference on Antennas and Propag. (EuCAP),
Barcelona, Spain, Apr. 2010, pp. 14.
[29] W. A. van Cappellen, L. Bakker, and T. A. Oosterloo, Experimental results of

the APERTIF phased array feed, in General Assembly and Scientic Sympo-
sium, 2011 XXXth URSI, Istanbul, Turkey, Aug. 2011, pp. 14.
[30] Y. Brand, A. K. Skrivervik, J. R. Mosig, and F. E. Gardiol, New iterative

integral equation technique for multilayered printed array antennas, in Mathe-
matical Methods in Electromagnetic Theory, Kharkov, Ukraine, Jun. 1998, pp.
615617.
[31] A. C. Polycarpou, Evaluation of stationary block iterative techniques for the

solution of nite arrays using the FE-BI method and domain decomposition, in
Proc. European Conference on Antennas and Propag. (EuCAP), Nice, France,
Nov. 2006, pp. 16.

[34] P.-S. Kildal, Foundations of Antennas: a Unied Approach. Lund: Studentlit-

teratur, 2000.
[35] C. A. Balanis, Advanced Engineering Electromagnetics. New York: John Wiley

and Sons, Inc., 1989.
[36] R. Maaskant, R. Mittra, and A. G. Tijhuis, Fast analysis of large antenna

arrays using the characteristic basis function method and the adaptive cross
approximation algorithm, IEEE Trans. Antennas Propag., vol. 56, no. 11, pp.
34403451, Nov. 2008.
[37] O. A. Iupikov, R. Maaskant, and M. V. Ivashina, A plane wave approximation

in the computation of multiscattering eects in reector systems, in Proc. Eu-
ropean Conference on Antennas and Propag. (EuCAP), Gothenburg, Sweden,
Apr. 2013, pp. 38283832.
140
References
[38] J. J. H. Wang, An examination of the theory and practices of planar near-eld
measurement, IEEE Trans. Antennas Propag., vol. 36, no. 6, pp. 746753, Jun.
1988.
[39] J. P. McKay and Y. Rahmat-Samii, Compact range reector analysis using the
plane wave spectrum approach with an adjustable sampling rate, IEEE Trans.
Antennas Propag., vol. 39, no. 6, pp. 746753, Jun. 1991.
[40] R. C. Rudduck, D. C. Wu, and M. Intihar, Near-eld analysis by the plane-wave

spectrum approach, IEEE Trans. Antennas Propag., vol. 21, no. 2, pp. 231234,
Mar. 1973.
[41] (2007) EM Software & Systems S.A. (Pty) Ltd, Stellenbosch, South Africa,
FEKO, Suite 6.0. [Online]. Available: http://www.feko.info

141
142
Paper D
Novel Multi-Beam Radiometers for Accurate Ocean

Surveillance
C. Cappellin, K. Pontoppidan, P. H. Nielsen, N. Skou, S. S. Søbjærg,
M. Ivashina, O. Iupikov, A. Ihle, D. Hartmann, and K. v. 't Klooster
Proceedings of the 8th European Conference on Antennas and

in
Propagation, EUCAP 2014, The Hague, The Netherlands, April 2014.
the thesis.
Novel Multi-Beam Radiometers for Accurate
Ocean Surveillance
C. Cappellin, K. Pontoppidan, P. H. Nielsen, N. Skou, S. S. Søbjærg, M. Ivashina,
O. Iupikov, A. Ihle, D. Hartmann, and K. v. 't Klooster
Abstract
Novel antenna architectures for real aperture multi-beam radiometers
providing high resolution and high sensitivity for accurate sea surface
temperature (SST) and ocean vector wind (OVW) measurements are in-
vestigated. On the basis of the radiometer requirements set for future
SST/OVW missions, conical scanners and push-broom antennas are com-
pared. The comparison will cover reector optics and focal plane array
conguration.
1 Introduction
The assessment of ocean parameters like salinity, sea surface temperature and ocean
vector wind based on spaceborne microwave radiometer measurements is an impor-
tant and challenging task, not only concerning geophysical algorithms but also con-
cerning technical aspects. A thorough and very recent review of ocean sensing was
carried out by ESTEC and leading oceanography expert groups worldwide, produc-
ing the instrument requirements that future radiometers shall aim at, according to
Table 1. The satellite height above the Earth and the incidence angle are assumed
equal to 817 km and 53 deg, respectively.
Table 1: Radiometer characteristics for the conical scan antenna at C-, X- and Ku-band.
Freq., Bandwidth, Polari- Sensiti- Bias, Resolution, Dist. to

[GHz] [MHz] zation vity, [K] [K] [km] coast, [km]
6.9 300 V, H 0.30 0.25 20 5-15
V, H 0.22 0.25 20 5-15
10.65 100
S3 , S4 0.22 0.25 20 5-15
V, H 0.25 0.25 10 5-15
18.7 200
S3 , S4 0.25 0.25 10 5-15
It is seen that SST and OVW are measured from C to Ku band, with a desired
ground resolution of around 20 km at C- and X-band, and 10 km at Ku-band. The
desired sensitivity is around 0.22 K. It is easily derived, see the procedure described
145
Paper D. Novel Multi-Beam Radiometers for Accurate Ocean Surveillance
Figure 1: Conical scan (left) and push-broom (right) scenario.
in [1], that a radiometer with antenna aperture of around 5 m provides the required
ground resolution but cannot achieve the desired sensitivity in a traditional single
radiometer channel/beam concept, even with the state-of-the-art noise performance
of receivers available in the market. The required sensitivity can only be met by
considering several simultaneous beams in the along- and across-track, in either a
push-broom system, or in a multi-beam scanning system, as depicted in Fig. 1.
The push-broom system achieves very high sensitivity since all across track foot-
prints are measured simultaneously by their own receivers [2]. The antenna has the
clear advantage of being stationary, but the number of beams and receivers is very
high. An advanced feed design and reector are necessary, and its light-weight me-
chanical realization is challenging. The multi-beam scanning system achieves high
sensitivity by measuring each footprint several times followed by integration. The
antenna is mechanically smaller than an equivalent torus, but presents numerous
challenges in order to achieve a well-balanced rotation at satellite level [3]. Again,
an advanced feed design is necessary.
In February 2013 the ESA contract 4000107369-12-NL-MH was awarded the team
consisting of TICRA, DTU-Space, HPS and Chalmers University. The purpose of
the activity is to identify the antenna requirements for a conical scanning and a push-
broom radiometer for accurate SST and OVW measurements, and to make a trade-o
of such two antennas, with respect to reector optics, focal plane array conguration,
ultra-light mesh reector technology, mechanical stability, and calibration and RFI
mitigation techniques. The purpose of the present paper is to describe the reector
optics and feed array design used for the trade-o, for a conical scanning and a
push-broom radiometer antenna satisfying the requirements of Table 1. The paper is
organized as follows: In Section 2 the optical design is described, while in Section 3
the antenna requirements derived from the radiometer requirements are highlighted.
The feed array design is nally given in Section 4.
146
2. Optical Design
2 Optical Design
Following the procedure described in [1] it was found that a reector antenna with
projected aperture of around 5 m provides the required ground resolution. A coni-
cal scanning and a torus-push-broom antenna implementation were then considered.
They are described in more detail in the following subsections.
2.1 Conical scanning radiometer antenna

The conical scanning antenna is an oset paraboloid with projected aperture D of
5 m. The clearance is set to 1 meter in order to provide space for the feed cluster
and the focal length f is set to 3 m in order to make the design more compact. For a
swath of 1500 km, the sensitivity of Table 1 can be achieved by:
2 beams along track at 6.9 GHz;
3 beams along track and 7 beams across track at 10.65 GHz;
5 beams along track and 6 beams across track at 18.7 GHz.
The number of beams in the along track direction is selected such that they cover the
same strip width on the Earth. The antenna rotates at 11.5 RPM and the radiometer
has a for-and-aft look.
2.2 Torus push-brom radiometer antenna

The push-broom antenna is a torus reector with projected aperture D of 5 m. The
torus is obtained by rotating a section of a parabolic arc around a rotation axis.
The focal length of the parabolic generator is also 5 m. A possible way of obtaining
the torus is shown in Fig. 2: the feed axis is selected parallel to the rotation axis,
implying that all feed element axes are parallel and orthogonal to the focal plane.
The feed array becomes therefore planar, simplifying the mechanical and electrical
design. The reector rim is found by the illuminated rotated aperture up to the
outmost scan positions, see Fig.3.
The antenna shall be able to provide a scan of ±20◦ corresponding to a swath
width of 600 km. The nal design is shown in Fig. 4, where the projected reector
aperture is 5 m by 7.5 m. It is recalled that the swath of the torus push-broom
was reduced from 1500 km to 600 km in order to decrease the horizontal size of the
reector from 11 m to 7.5 m. This also reduces the feed array size and simplies the
electrical and mechanical realization.
It is noted that the sensitivity provided by the torus push-broom is always one
degree of magnitude higher than the one provided by the conical scanner. This is at
the expenses of a very large number of beams, and correspondingly large number of
receivers. For a swath of 600 km we need:
147
Figure 2: Torus design. Figure 3: Rim trace for toroidal push-broom

antenna design.
Figure 4: Torus push-broom antenna with projected aperture D of 5 m, three feeds located at 0◦
◦
and ±20 , f /D = 1, and swath of 600 km.
58 beams across track at 6.9 GHz;
156 beams across track at 18.7 GHz.
The antenna is stationary in contrast to the conical scan antenna.
148
3. Antenna Requirements
3 Antenna Requirements
3.1 Acceptable cross-polarization
The requirement for the cross polarisation is not given directly in Table 1. We know,
however, that the radiometer shall operate with two linear polarisations, vertical and
horizontal, and that the accuracy indicated in the column Bias in Table 1 shall be
achieved. It can be shown that the required ∆T ≤ 0.25 K implies that the cross
polar power must not exceed 0.33 % of the total power on the Earth.
3.2 Acceptable side lobes and distance to coast

Table 1 states that the radiometer shall operate satisfactorily within 5 − 15 km from
the coast. It is assumed that this distance Dc is measured from the 3 dB footprint of
the beam. The reason behind the requirement is that the brightness temperature of
land areas is much higher than the brightness temperature of the sea, which is what
we want to measure. Assume that the coast is located at the angle θc from boresight.
It turns out, with∆T ≤ 0.25 K, that the power from boresight up to θc shall contain
99.71 % of the total power on the Earth. The value of θc is determined by integration
of the power pattern.
4 Feed Array Design

4.1 Conical scanning radiometer antenna
To design a feed array for the conical scanning radiometer antenna of Section 2.1, and
at the same time compensate for the cross-polar component generated by the small
f /D, a single feed per beam approach is not possible. A feed array with many closely
spaced elements is a good candidate. A feed is here understood as the collection
of the elements used to generate a particular beam. In the following, we will assume
that:
The feed array element is a half wave dipole above an innite ground plane;
The feed array elements are arranged in a square grid with a spacing of 0.75
wavelengths;
Each feed can be represented by a sub array of 5 by 5 elements.
We wish to design the feed arrays for the three frequencies and to calculate the
properties of the least scanned and the most scanned beams. This will require the
following steps:
1. Determine the necessary feed array size for each of the bands C, X and Ku;
149
Figure 5: Feed arrays located in the focal plane.
2. Position the feed arrays in the focal plane.
The conical scan antenna is a focusing system and the half power beam width
is inversely proportional to the frequency. With this in mind, and the previously
mentioned required number of beams, it is a simple task to determine the size of the
feed arrays for the three frequencies. The result is shown in Fig. 5. The layout is
selected such that the scan, measured in beam widths, for the most scanned beam
has been minimized. It is noted that the required number of beams is obtained by
assuming that the beams overlap at the −3 dB cross-over points.
To calculate the performance of the conical scan antenna we select the least
scanned and the most scanned beam for each frequency. The feed positions cor-
responding to these beams are indicated by small black crosses in Fig. 5. In order
to nd the feed array excitations necessary to generate these beams, the following
procedure is used:
1. Illuminate the reector with a Gaussian beam with correct direction and ori-
entation;
2. Calculate the focal plane eld;
3. Determine the top 30 dB co- and cross-polar element excitations.
The direction of the Gaussian beam in step 1 is given directly by the selected
beam. The orientation of the beam is especially important for the scanned beams:
it must be such that the beam on the Earth is vertically and horizontally polarized.
The Gaussian beam incident on the reector has a taper of 20 dB. The focal plane
eld is calculated in step 2. This eld is used to calculate the excitation of the array
elements in step 3 applying the Conjugate Field Matching (CFM) method. Only the
150
4. Feed Array Design
elements with excitations from the maximum value and down to 30 dB below the
maximum value are included, in order to account for realistic receivers.
The radiometer characteristics for the six beams of the conical scan antenna are
summarized in Table 2. It is seen that the X and Ku band beams satisfy the require-
ments of Table 1, relative to distance to coast, footprint and cross-polar power. The
performances for the C-band beams are not acceptable with respect to cross polar-
ization, while the distance to coast is around 20 km, slightly more than the required
15 km. The design and performances of the above feed array was obtained both by
TICRA and Chalmers, following the same procedure.
Number of cross- Peak Dist. to

Beam active polar directivity Footprint coast
elements power
x-dir y-dir % dBi km km
C_1 52 23 0.72 48.13 21.34 20.7
C_2 52 23 0.74 48.15 21.29 19.14
X_1 26 10 0.18 52.08 13.79 10.03
X_2 40 16 0.30 51.98 13.76 15.45
Ku_1 21 12 0.11 56.96 7.87 5.73
Ku_2 31 16 0.24 56.57 7.93 13.28
4.2 Torus push-brom radiometer antenna

To design the feed array for the torus push-broom antenna, a slightly dierent pro-
cedure than the one described in Section 4.1 is necessary. This is due to the fact
that the push broom reector is not a paraboloid and the antenna is not a focusing
system for which results obtained at one frequency can easily be scaled to another
frequency.
As a starting point, the inuence of the taper of the Gaussian beam incident on
the reector is investigated at 10 GHz. The center beam is considered. The taper is
varied from 20 dB to 60 dB in steps of 10 dB. The associated focal plane elds are
shown to the left in Fig. 6 for the 20 dB and the 60 dB cases. It is seen that the
extent of the eld decreases as the taper of the incident beam increases. This means
that if we can use a higher taper of the incident Gaussian beam we can apparently
reduce the size of the feed array.
A large feed array covering the same part of the focal plane as Fig. 6 is now
generated. The element spacing is 0.75 wavelengths = 22.5 mm and the number of
elements is 73 in both directions. The total number of elements is actually 2 × 73 ×
73 = 10658 because there are two orthogonal dipoles at each element location.
The focal plane elds in Fig. 6 are used to determine, again with the Conjugate
Field Matching (CFM) method, the excitations of all the 10658 dipole elements,
151
Figure 6: Focal plane eld (left) and far eld (right) for incident beam tapers of −20 dB (top) and
−60 dB (bottom). The center beam is considered.
which then are used to generate the radiated center beam. The co polar component
of the calculated far elds is shown to the right in Fig. 6 and it is seen, as expected,
that the beam becomes broader as the taper increases, and, at the same time, the
side lobes become smaller.
The radiometer characteristics for the center beams of Fig. 6 are shown in Table 3.
We see that the cross polarisation requirement is always perfectly met. The footprint
and the distance to coast increase as the taper increases. The results here are for
10 GHz which is close to the X-band frequency, where the requirement to footprint
and distance to coast is 20 km and 15 km, respectively.
The results in Table 3 include all the elements in the feed array. It is of course of
interest to reduce the number of active elements. Fig. 6 shows that the extent of the
eld in the focal plane decreases as the incident beam taper increases so from a feed
array size point of view it is better to use a high input taper. Table 3 shows that
with an incident taper of 50 dB a very acceptable beam is obtained. It is therefore
attempted to use this focal plane eld but only use those elements in the feed array
with an excitation larger than a certain value below the maximum. Table 4 shows
the results obtained when this limit is set to 40, 30 and 20 dB below the maximum.
It is seen that with a 30 dB limit both the cross polarisation, the footprint and
the distance to coast meet the requirements and the number of active elements are
152
4. Feed Array Design
Table 3: Radiometer characteristics for the toroidal push broom antenna at 10 GHz for varying
taper of the incident beam.
Taper of Number of co- cross- Peak Foot- Dist. to

incident active polar polar direc- print coast
eld elements power power tivity
dB ρ-dir φ-dir % % dBi km km
20 5329 5329 98.83 0.07 53.27 12.08 7.93
30 5329 5329 99.74 0.03 51.84 14.16 9.89
40 5329 5329 99.93 0.02 50.68 16.12 11.51
50 5329 5329 99.98 0.01 49.74 17.93 12.89
60 5329 5329 99.99 0.01 48.96 19.60 14.11
Table 4: Radiometer characteristics for an incident eld taper of 50 dB and excitation limits of 40,
30 and 20 dB.
Taper of Number of co- cross- Peak Foot- Dist. to

incident active polar polar direc- print coast
eld elements power power tivity
dB ρ-dir φ-dir % % dBi km km
40 292 56 98.96 0.02 49.63 18.31 12.77
30 155 2 99.79 0.12 49.46 18.66 13.60
20 69 0 99.35 0.14 49.09 19.49 40.18
reduced from 10658 to 157, i.e. 155 in the radial direction and 2 in the azimuthal
direction.
The experience gained at 10 GHz is used to design the feed array in the three
bands, following pretty much the same procedure. It is recalled that it is necessary
to tilt the direction of the incident beams such that the focal plane elds for the
dierent frequencies are located side by side, leading to the feed array parameters
shown in Table 5. Again, it is assumed that beams overlap at the −3 dB cross-over
points. The feed arrays are shown in Fig. 7 and the radiometer characteristics for
the center beam are presented in Table 6. It is seen that the performance meet the
requirements except for the distance to coast at C-band.
Table 5: Table used to determine the necessary size of the feed arrays for the push broom torus
reector antenna.
Wave- Distance Element Element Total

Freq. length between ρmax ρmin
Nρ Nφ element
elements number
GHz mm mm mm mm
6.90 43.48 32.61 4078 3732 11 93 1023
10.65 28.17 21.13 4566 4314 12 163 1956
18.70 16.04 12.03 4292 4108 16 271 4336
153
Figure 7: The three feed arrays for the push broom torus reector antenna.
Table 6: Radiometer characteristics for the three frequency bands.
Exci- Number co- cross- Peak Dist.

Freq. Input
taper tation of active polar polar direc- Foot-
print to
limit elements power power tivity coast
GHz dB dB x-dir y-dir % % dBi km km
6.90 30 30 133 4 99.56 0.20 48.13 21.82 27.59
10.65 50 30 161 2 99.68 0.10 49.96 17.61 13.29
18.70 40 30 351 0 99.73 0.17 55.56 9.28 13.00
The feed array designed by Chalmers for the torus push-broom antenna is de-
scribed in detail in [4]. The feed array element is a Vivaldi antenna and the element
spacing is 0.7 wavelength. The number of active elements and their weight coecients
is found with a customized beam former that aims to realize the best trade-o be-
tween the maximum beam eciency and the minimum sidelobe and cross-polarization
power. To include constraints on the dynamic range of the beamformer in the course
of optimization, the customized beamforming algorithm proposed in [4] has been fur-
ther extended through the use of an iterative procedure. This procedure modies
the reference weights, as determined for the beamformer without constraints, while
aiming to maintain the radiometer characteristics as close as possible to the refer-
ences ones for a specied value of the dynamic range. The performances obtained by
Chalmers coincided with Table 6, except for the distance to land at C-band which
was 16.9 Km, and thus met the requirements. The total number of elements of the
complete feed array for one polarization was 888, 1224 and 2184, for the C- X- and
Ku-band respectively, thus smaller than the number of elements obtained by TICRA
and reported in the last column of Table 5.
154
5. Conclusions
5 Conclusions
The reector optics and feed array designs of a conical scanning and push-broom
radiometer antenna for future SST/OVW missions were described. The conical scan-
ning is a traditional oset paraboloid with reduced f /D rotating at 11.5 RPM, while
the push-broom is a stationary torus reector, with projected aperture of 5 m by
7.5 m. The feed array of the conical scan antenna was obtained by considering half
wave dipoles above an innite ground plane, with a spacing of 0.75 wavelengths. The
array excitations were obtained by CFM, considering a Gaussian beam with taper of
20 dB impinging on the reector. The performances of the least and most scanned
beams met all the requirements at X- and Ku-band. The performances for the C-
band beams were not acceptable with respect to cross polarization, and the distance
to coast was slightly more than the required 15 km. The feed array of the torus
push-broom antenna was derived by TICRA in a way similar to the one used for the
conical scan, while Chalmers developed a customized beam former to optimize the
maximum beam eciency and the minimum sidelobe and cross-polarization power,
including constraints on the dynamic range of the beamformer. The performances of
the center beam obtained by Chalmers met all the requirements at all three frequency
bands, while TICRA obtained a slightly larger distance to coast at C-band and used
more antenna elements. The present results must be considered preliminary.
References
[1] N. Skou and D. L. Vine, Microwave Radiometer Systems: Design & Analysis.
Artech House, 2006.

and test of a pushbroom radiometer, in Antennas and Propagation, 1989. ICAP
Kingdom, Apr. 1989, pp. 126130.
[3] M. Mobrem, E. Keay, G. Marks, and E. Slimko, Development of the large aper-
ture reector/boom assembly for the SMAP spacecraft, in ESA/ESTEC Work-
shop on Large Deployable Antennas, Noordwijk, The Netherlands, Oct. 2012.

on Antennas and Propag. (EuCAP), Hague, The Netherlands, Apr. 2014, pp. 15.
155
156
Paper E
Dense Focal Plane Arrays for Pushbroom Satellite

Radiometers
O. Iupikov, M. Ivashina, K. Pontoppidan, P. H. Nielsen, C. Cappellin,
N. Skou, S. S. Søbjærg, , A. Ihle, D. Hartmann, and K. v. 't Klooster
Proceedings of the 8th European Conference on Antennas and

in
Propagation, EUCAP 2014, The Hague, The Netherlands, April 2014.
the thesis.
Dense Focal Plane Arrays for Pushbroom Satellite
Radiometers
O. Iupikov, M. Ivashina, K. Pontoppidan, P. H. Nielsen, C. Cappellin, N. Skou,
S. S. Søbjærg, , A. Ihle, D. Hartmann, and K. v. 't Klooster
Abstract
Performance of a dense focal plane array feeding an oset toroidal reec-

tor antenna system is studied and discussed in the context of a potential
application in multi-beam radiometers for ocean surveillance. We present
a preliminary design of the array feed for the 5-m diameter antenna at
X-band. This array is optimized to realize high antenna beam eciency
(∼ 95%) over a wide scan range (±20◦ ) with very low side-lobe and cross-
polarization levels.
1 Introduction
Recent advances in phased-array antenna technologies and low-cost active electronic
components open up new possibilities for designing Earth observation instruments, in
particular those used for radiometric measurements. Nowadays, two design concepts
of microwave radiometers are in use: push-broom and whisk-broom scanners [1].
Push-broom scanners have an important advantage over whisk-broom scanner in
providing larger eld-of-view with higher sensitivity, owing to the fact that these
systems can look at a particular area of the ocean for a longer time with multiple
simultaneous beams. This concept is illustrated on Fig. 1, where one can see several
beams, arranged perpendicular to the ight direction of the spacecraft. However, the
drawback of pushbroom designs based on conventional focal plane arrays (FPAs) of
horns in one-horn-per-beam conguration [2] or clusters with simplistic beamforming
[3] is the varying sensitivity. This variation occurs due to the dierence between
the scanned beams (as these are formed by dierent horns/clusters) and their large
separation on the ocean surface, as the result of the large separation between the
horns.
This drawback may be signicantly reduced by employing dense FPAs, i.e. phased-
array feeds consisting of many electrically small antenna elements, with advanced
beamforming [4]. This technology has been extensively studied during the last
decade in the radio astronomy community, and several telescopes are currently being
equipted with dense FPAs [57]. While those systems aim to provide the scan range
159
Paper E. Dense Focal Plane Arrays for Pushbroom Satellite Radiometers
of about 5 − 10 beamwidths, for applications as herein considered, the desired scan

range (swath range of the radiometer) is one order of magnitude larger [8]. Therefore,
to achieve this performance, more complex designs of the reector optics and FPA are
required. For push-broom radiometers, various optics concepts have been investigated
[2], and the optimum solution has been found to be an oset toroidal single reector
antenna, such as illustrated on Fig. 1. This reector structure is rotationally sym-
metric around a vertical axis, and thus is able to cover a wide swath range. However,
its aperture eld exhibits signicant phase errors due to the non-ideal (paraboloid)
surface of the reector as compared to that of classical paraboloids. The phase
errors cause degradation of the antenna beam eciency and increase the side-lobe
and cross-polarization levels. These degradations, in turn, limit the radiometer char-
acteristics (such as the minimum distance to coast at which the measurement data
remains usable) as well as worsen the situation with Radio Frequency Interference
(RFI) that is problematic at many radiometer bands [9].
The purpose of this work is, therefore, (i) to determine to what extent the
performance-limiting factors of push-broom radiometers can be reduced by using
dense FPAs with advanced beamforming; and (ii) what is the minimum complexity
of the FPA design (size, number of elements) that is required for meeting the instru-
Figure 1: Operational principle of a push-broom microwave radiometer, which includes an o-set

toroidal reector antenna fed with a multi-beam focal plane array of horns arranged perpendicular
to the ight direction of the spacecraft. Dierent areas of the ocean-surface are scanned as the
spacecraft ies forward.
160
2. Antenna Requirements
ment specications at which future radiometers aim [8]. To address these questions,
we have created an initial numerical model of the array that is based on the MoM-
CBFM-model in [4]; the elements of this array represent tapered-slot antennas, as
designed for the FPA system in [5]. To perform the parametric study, we have im-
plemented this model for dierent array sizes and inter-element separation distances
varying from 0.5 to 1 wavelength. For the evaluation of the radiometer characteris-
tics, two beamforming methods have been considered that aim to optimize the beam
eciency with the minimum distance to land and cross-polarization loss.
2 Antenna Requirements
In February 2013 the ESA contract 4000107369-12-NLMH was awarded the team
consisting of TICRA, DTU-Space, HPS and Chalmers University. The group com-
prises experts in reector antennas design and analysis, passive microwave radiometry,
mechanical and thermal analysis of ultra-light mesh reector technology, and radio
astronomy with the knowledge of dense focal plane arrays designs. As a part of this
activity, we perform a preliminary design study of a pushbroom antenna, as shown
on Fig. 1 with conventional FPAs of horns, as well as novel dense FPAs with active
beamforming.
To identify the best design for the targeted application, we use the list of antenna
system requirements that has been derived in [8], based on the the instrument spec-
ications for accurate sea surface temperature and ocean vector wind measurements.
This list includes the values for the required half-power beamwidth (and correspond-
ing footprint on the sea-surface), acceptable cross-polarization power, as well as the
minimum distance to coast at which the radiometer stops working correctly. It can
be shown that in order to meet the requirements for radiometer characteristics (max-
imum allowed error of the measured sea brightness temperature ∆T < 0.25K and the
distance to coast < 15km) the power incident on the land must be less than 0.14%
of the total power hitting the Earth. This requirement leads to stringent constraints
on both the side-lobe and cross-polarization levels of the antenna beams.
At present, the pushbroom antenna which can satisfy these requirements is a torus
◦
antenna with projected aperture of 5 m, (±20 ) scan, the focal length to diameter
ratio f/D=1 and swath of 600 km. This antenna has been designed by TICRA; it
achieves the swath on the Earth equal to600 km, assuming the satellite altitude
above the Earth of 817 km and the incidence angle of 53◦ . This antenna should
operate at C-band (6.9 GHz), X-band (10.65 GHz) and Ku-band (18.7 GHz) with
bandwidth 300 MHz, 100 MHz and 200 MHz, respectively. The analysis in this paper
is performed at X-band only.
161
−30 −25 −20 −15 −10 −5 0

,[dB]
del = 0.5λ del = 0.7λ del = 1.0λ
−5
−10
−15
−20
−25
−30
Figure 2: Eect of the inter-element separation distance del on (top) the optimized amplitude weights
of the FPA sub-array elements for the centre beam, as determined for the customized beamformer
maximizing the beam eciency (@−20 dB) with constraints on the side-lobe and cross-polarization
levels towards the Earth, and (bottom) the resultant illumination patterns of the reector antenna.
The array size is xed to Lx × Ly = 7λ × 14λ.
3 FPA-system design
3.1 Antenna array model
As a starting point of the design procedure, we have considered a sub-array for the
centre beam. The selected initial model of this sub-array represents a dual-polarized
antenna array consisting of 15 × 29 × 2 interconnected tapered-slot antenna elements
with the inter-element distance varying from 0.5λ to 1.0λ. This model is based on the
MoM-CBFM model of the 8×9×2 element array in [4]. To reduce the computational
time for our parametric studies, we have simplied this original model by assuming
that all embedded element patterns are identical to that of the central element of the
nite array. The sub-array embedded element patterns have been imported into the
162
3. FPA-system design
reector antenna software GRASP10 to compute the secondary embedded element

patterns (after reection from the dish), which, in turn, have been used to simulate
the overall receiving system (according to the procedure in [4]), and to optimize its
beamforming weights. It is worth mentioning that this analysis and optimization
procedure accounts for the eects of the array mutual coupling, elements loading, as
well as the signal and noise properties of the terminating ampliers. The later eects
have not been considered yet and are left for the future work.
3.2 Beamforming algorithms

For this study, we have implemented two types of signal-processing beamforming
algorithms: (i) standard maximum directivity beamformer (see Eq.3 in [4]), which
is equivalent (under certain conditions) to the Conjugate Field Matching (CFM)
beamformer, as commonly applied to conventional FPAs of horns; and (ii) customized
beamformer that has been formulated so as to maximize the beam eciency (within
the -20dB area), subject to constraints on the total radiated power towards the
coastal region. The latter approach is expected to lead to the optimal radiometer
performance in terms of the minimum distance to land, minimum cross-polarization
and side-lobe levels, and thus improved resistance to RFI.
Figure 3 illustrates the examples of the optimized weight coecients for the sub-
array elements for the CFM and constrained beamformers, where the corresponding
aperture-eld distributions and footprint patterns of the antenna are presented below.
As expected, the CFM beamformer leads to the highest directivity (the larger area
of the reector aperture that is illuminated eciently), as compared to that of the
constrained beamformer. On the other hand, the latter has a signicantly improved
shape of the footprint and much lower side-lobe level.
3.3 Parametric study

The analysis of the weighting coecients on Fig. 3 shows that many elements are
weakly excited, so the size of the initially selected sub-array can be reduced along
x-direction. Note that in y -direction, the sub-array cannot be smaller, since these
elements will be used to form the scanning beams. We have, therefore, performed a
parametric study by looking into the smaller arrays sizes, as well as dierent values
of the element separation distances.
Inter-element separation distance

Fig. 2 and Fig. 4 present the rst set of the results obtained for the array 7λ × 14λ
and constrained beamformer that illustrate the eect of the element separation
distance del on the optimized weights and corresponding aperture-eld distribution
of the reector. As observed, the most dense FPA provides a very ne sampling of
163
−5
−10
−15
−20
−25
−30
(a) CFM-BF weights (b) Customized-BF weights
(c) CFM-BF illumination pattern (d) Customized-BF illumination pattern
(e) CFM-BF footprint (f ) Customized-BF footprint
Figure 3: Comparison of two beamforming algorithms for the FPA sub-array for the centre beam:
(a, b) the array element amplitude weight coecients for the CFM beamformer (CFM-BF) and
customized beamformer (Customized-BF), where each block represents an element and the black
line shows the focal line of the torus reector, and the the corresponding (c, d) reector aperture
illumination patterns and (e, f ) footprint patterns on the sea-surface.
the array aperture eld, resulting in the well-behaved illumination of the reector,
whereas the eld produced by the most sparse array with 1λ-spaced elements exhibits
164
100

95
90
85
80
75 Beam efficiency
70
65 Power hitting reflector
60
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Distance to land, [km] Element spacing, λ
20
Requirement
18 Distance to land
16
14
12
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Element spacing, λ
20
Footprint, [km]
18 Requirement FPL (Phi=90) Average FPS (Phi=0)

16
14
12
10
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Element spacing, [λ]
0.4
in cross−polar, [%]
Relative power
0.3
Requirement
0.2
XP relative power
0.1
0
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Element spacing, λ
Figure 4: Radiometer characteristics as function of the FPA element spacing (del ) for the case of
Lx = 7λ, including (from top to bottom) the antenna beam eciency (dened within the −20 dB
region), distance to land at which the radiometer should stop working correctly, averaged footprint
and relative cross-polarization power loss in the entire region.
the grating lobes. The importance of the array density can be also seen from the
computed radiometer parameters that are shown on Fig. 4 as function of del . It is
interesting to see that the beam eciency and cross-polarization power are aected
most when the array becomes sparse (del > 0.7λ) because the array aperture eld
gets under-sampled and the grating lobes start to appear , while the minimum
distance from land remains small almost over the entire region of del (and within the
required value of 15 km) thanks to the low side-lobes in the coastal region that are
forced by the beamformer.
165
−30 −25 −20 −15 −10 −5 0

,[dB]
Lx = 3.5λ Lx = 4.5λ Lx = 7.0λ
(8 X -elements) (10 X -elements) (15 X -elements)
−5
−10
−15
−20
−25
−30
Figure 5: Eect of the array size Lx on (top) the optimized amplitude weights of the FPA sub-
array elements for the centre beam, as determined for the customized beamformer maximizing the
beam eciency (@−20 dB) with constraints on the side-lobe and cross-polarization levels towards
the coastal region, and (bottom) the resultant illumination patterns of the reector antenna. The
distance between the array elements is xed to del = 0.5λ.
Array size
The second set of the parametric study results is illustrated on Fig. 5 and Fig. 6,
that show the eect of the sub-array size along x-direction, for the case of del = 0.5λ.
These results demonstrate that all the radiometer parameters are sensitive to change
of the array size, and their values degrade when it becomes smaller. This observation
is expected, since the larger arrays have more degrees of freedom that the smaller
ones. In general, the minimum size of the array along x-direction should be ∼ 4.9λ
to realize the beam eciency higher than ∼ 91% with the distance to coast according
to the requirements. For del = 0.7λ, this would corresponds to 8 × 21 × 2 elements
in total for the center sub-array. Interestingly enough, the beam eciency of twice
166
Number of elements along X−axis

5 7 9 11 13 15
100

95
90
85
Beam efficiency
80
Power hitting reflector
75
70
65
60
2.0 3.0 4.0 5.0 6.0 7.0
Array size along X−axis, [λ]

5 7 9 11 13 15
Distance to land, [km]
20
Distance to land
18
Requirement
16
14
12
2.0 3.0 4.0 5.0 6.0 7.0

5 7 9 11 13 15
20
Footprint, [km]
18
16
14
12
Requirement FPL (Phi=90) Average FPS (Phi=0)
10
2.0 3.0 4.0 5.0 6.0 7.0

5 7 9 11 13 15
0.4
Relative power
0.3
Requirement
0.2
XP relative power
0.1
0
2.0 3.0 4.0 5.0 6.0 7.0
Figure 6: Radiometer characteristics as function of the number of elements (array size) alone X-axis
for the case of del = 0.5λ, including (from top to bottom) the antenna beam eciency (dened
within the −20 dB region), distance to land at which the radiometer should stop working correctly,
averaged footprint and relative cross-polarization power loss in the entire Earth region.
167
larger sub-array (Lx = 9λ) would be only a few percent higher (∼ 96%) with the
similar values of other considered radiometer parameters.
4 Conclusions
Table 1: Radiometer characteristics for dierent FPAs

Gauss. FPA with FPA with FPA with
feed CFM-BF Cust-BF Cust-BF
model 15 × 29 × 2 15 × 29 × 2 8 × 21 × 2
elem. elem. elem.
del = 0.5λ del = 0.5λ del = 0.7λ
Beam eciency [%] 84.2 85.1 94.9 92.0
XP-power, [%]
0.39 1.01 0.03 0.02
(<0.33% is req.)
Dist. to land, [km]
87.8 116.6 14.0 15.9
(<15 km is req.)
Beam width, [deg] 0.600 0.351 0.512 0.538
Footprint (FP), [km]
16.9 10.5 14.4 14.9
(<20 km is req.)
FP ellipticity 1.38 2.14 1.33 1.22
Table 1 summarizes the X -band performance parameters of the pushbroom ra-

diometer that employs a torus reector antenna with the 5-m diameter projected
aperture for dierent types of FPAs (i.e conventional FPAs of horns in one-horn-per-
beam-conguration that are herein represented by Gaussian beams, and dense FPAs
of tapered-slot antenna elements with dierent beamforming scenarios). As expected,
dense FPAs have obvious benets in achieving the required minimum distance to
coast and footprint roundness, while meeting all other radiometer requirements. The
minimum size of the FPA sub-array has been found to be 8 × 21 elements (for each
polarization) with the inter-element separation distance in the order of del = 0.7λ,
for the considered initial model of the array.
5 Acknowledgment
The present work has been carried out in the framework of the Advanced Multi-Beam
Radiometers project that is a collaborative eort between TICRA, DTU-Space (Den-
mark), HPS (Germany) and Chalmers, funded by European Space Agency (ESA).
The toroidal push-broom reector antenna used for our study has been designed by
TICRA. The authors would like to acknowledge the Swedish Research Council for
providing partial support to this work through the VR project grant.
168
References
References
[1] (2013, Sep.). [Online]. Available: http://earthobservatory.nasa.gov/Features/
EO1/eo1_2.php

Kingdom, Apr. 1989, pp. 126130.


Jun. 2011.
IEEE International Symposium on Proc.
Westerbork synthesis radio telescope, in
Phased Array Systems and Technology (ARRAY), Boston, Oct. 2010, pp. 640647.
448449.
[7] S. G. Hay, J. D. O'Sullivan, J. S. Kot, C. Granet, A. Grancea, A. R. Forsyth,

and D. H. Hayman, Focal plane array development for ASKAP (australian SKA
pathnder), in Proc. European Conference on Antennas and Propag. (EuCAP),
Edinburgh, UK, Nov. 2007, pp. 15.

14.
[9] C. Kidd, Radio frequency interference at passive microwave earth observation

frequencies, Int. Journal of Remote Sensing, vol. 27, no. 18, pp. 38533865, Sep.
2006.
169
170
Paper F
Improving the Calibration Eciency of an Array

Fed Reector Antenna Through Constrained
Beamforming
A. Young; M. V. Ivashina; R. Maaskant; O. A. Iupikov and
D. B. Davidson
IEEE Transactions on Antennas and Propagation, vol.61, no.7, 2013
the thesis.
Improving the Calibration Eciency of an Array
Fed Reector Antenna Through Constrained
Beamforming
A. Young; M. V. Ivashina; R. Maaskant; O. A. Iupikov and D. B. Davidson
Abstract
Calibrating for the radiation pattern of a multi-beam Phased Array

Feed (PAF) based radio telescope largely depends on the accuracy of the
pattern model, and the availability of suitable reference sources to solve
for the unknown parameters in the pattern model. It is shown how the
eciency of this pattern calibration for PAF antennas can be improved by
conforming the beamformed far eld patterns to a two-parameter physics-
based analytic reference model through the use of a Linearly Constrained
Minimum Variance (LCMV) beamformer. Through this approach, which
requires only a few calibration measurements, an accurate and simple
pattern model is obtained. The eects of the model parameters on the
directivity and sidelobe levels of multiple scanned beams are investigated,
and these results are used in an example PAF beamformer design for the
proposed MeerKAT antenna. Compared to a typically used Maximum
Directivity (MaxDir) beamformer, the proposed constrained beamform-
ing method is able to produce beam patterns over a wide Field-of-View
(FoV) that are modeled with a higher degree of accuracy and result in a
signicant reduction in pattern calibration complexity.
1 Introduction
Calibration of radio telescopes requires accurate models of the instrumental param-
eters and propagation conditions that aect the reception of radio waves [1]. These
eects vary over time and the model parameters have to be determined at the time
of observation through a number of calibration measurements. Furthermore, the
calibration measurements should complete in a relatively short time and may be re-
peated often over the course of an observation during which the instrumental and
atmospheric conditions can change signicantly. One of the instrumental param-
eters that needs accurate characterization is the radiation pattern of the antenna,
which is especially challenging in the arena of future array based multiple beam ra-
dio telescopes [24], both due to the complexity of these instruments, as well as the
173
Paper F. Improving the Calibration Efficiency of an Array Fed Reflector...
increased size of the Field-of-View (FoV). Above the requirement that the radiation
pattern should be accurately known, currently developed techniques for the pattern
calibration of these devices also emphasize the need for beams
1 over the FoV that are
similar in shape, and that each beam varies smoothly with time, frequency, and over
the main beam angular region [5]. Such beams can be described by simpler models,
which reduce the number of pattern model parameters that need to be solved for,
and also simplify the complexity of direction dependent calibration which is vitally
important for future radio telescopes [610]. However, achieving patterns exhibit-
ing these qualities, while also meeting the already stringent sensitivity requirements,
presents a dicult task.
Previously, beamforming techniques have been used to create similarly shaped

beams over the FoV by conforming them to an elliptical reference pattern, but at the
cost of a signicant loss in sensitivity [6] (up to 25%). An initial study has shown
that this loss can be reduced by applying the same beamforming technique, but using
a reference pattern that more closely matches the natural radiation characteristics
of large aperture antennas [11]. Therein, the rst term of the Jacobi-Bessel (JB)
series solution of reector antenna far eld patterns [12, 13] was used as a reference
pattern to dene directional constraints in a Linearly Constrained Minimum Variance
(LCMV) beamforming Phased Array Feed (PAF). It was found that this rst JB-
term is sucient to model the patterns of a prime focus single reector antenna over
a wide FoV of up to 5 beamwidths, over which the sensitivity reduction was less
than 10%. However, when considering a larger scan range, phase aberration eects
cause deformation of the radiation patterns to such an extent that this beam model
is no longer accurate. Furthermore, when applying this model to an oset reector
antenna for which the asymmetric geometry exacerbates the deformation of scanned
patterns [14, cf. Figs. 1 and 3], the inclusion of more physics-based information is
necessary.
Here, the reference pattern of [11] is extended to model the widening of the
scanned beam as well as the change in the phase distribution for an oset dual-
reector antenna by introducing two additional model parameters. It will be shown
that this model allows for the accurate characterization of multiple beams over a
wide FoV without the need to perform additional calibration measurements. The
eects of the model parameters on the directivity and sidelobe levels are investigated
for a proposed design of the MeerKAT radio telescope reector antenna [15]. An
LCMV beamformer is designed based on the results of this study, and its performance
evaluated through comparison with a Maximum Directivity (MaxDir) beamformer.
1 Often referred to as the direction-dependent gain or primary beam in the radio interferometer
community.
174
2. Antenna Pattern Model
2 Antenna Pattern Model

The reference pattern employed in [11] to constrain the main beam shape of a scanned
reector is based on the JB-series solution for modeling reector antenna far eld
patterns. The rst term in this series is the near-boresight approximation of the co-
polarized far eld pattern radiated by a circular aperture with a uniform amplitude
and phase distribution [16], i.e.,
J1 (ka sin θ)
FA (θ, φ) ∝ ≡ jinc(ka sin θ) (1)
ka sin θ
where a is the aperture radius, k is the free space wavenumber, and J1 is the Bessel
function of the rst kind of order one. Patterns radiated by more general aperture
eld distributions, including o-axis patterns of a scanned reector are represented
as a sum of (possibly) many more JB-terms. However, the rst term in the series
is still dominant over an angular region around the beam maximum. To obtain a
pattern function that applies to more general aperture eld distributions, certain
modications to the reference pattern (1) are required as detailed below.
In order to control the beamwidth of the pattern model, an angular scaling pa-
rameter s is introduced by letting a → sa, which enables accounting for widening
of the beam due to under-illumination of the reector aperture or coma aberration
2
when scanning [17, 18] . In this sense a distinction can be made between the physical
aperture radius a, and an eective aperture radius sa, where s . 1.
Another limitation of (1) is that it assumes a constant phase distribution of the
beam pattern. This implies that the phase reference of the pattern coincides with
the phase center of the antenna, dened here for a small angular region of the far
eld around the main beam center. Whereas this condition is easily satised for an
on-axis beam of a prime focus reector, the proper choice for the phase reference is
not straightforward for scanned beams. In the latter case it is more convenient to
keep the phase reference xed at the center of the projected aperture and to account
for a phase variation over the main beam through multiplying the pattern model by
Fψ (θ, φ) = exp (jΨ sin θ cos(φ − φ0 )) (2)
in which Ψ is a constant that determines the phase gradient, and φ0 denes the
direction of the phase center shift. The value of φ0 can be determined by noting that
for a scanned beam the phase center shift is in the scan plane. It can be shown that
the value of Ψ is proportional to the phase center shift projected orthogonally to the
direction of observation [19].
2 The pattern deformations for o-axis scanning are known to be asymmetrical, and since the
analytic model is used here to constrain the pattern shape so that it is easily modeled, we elect to
use a circularly symmetric pattern model.
175
Combining (1) and (2) gives the extended reference pattern model
3
F (s, Ψ; θ, φ) = jinc(ksa sin θ)ejΨ sin θ cos(φ−φ0 ) (3)
in which the the amplitude and phase distributions of the reference pattern are con-
trolled independently by the parameters s and Ψ, respectively. Note that (3) will
serve as a reference pattern for deriving the directional constraints in an LCMV beam-
former, as well as a pattern calibration model to describe the realized beamformed
pattern.
3 Beamforming Strategy
An LCMV beamformer is implemented which minimizes the power received by the
antenna due to noise subject to linear constraints that conform the co-polarized
pattern shape to the reference pattern in (3). The beamformer weights applied to
the elements of the PAF are calculated according to [20] [21, p. 526]
−1 H −1
H
= gH GH C−1 G

wLCMV G C (4)
in which xH means the complex conjugate transpose of x, C is the noise covari-

ance matrix, g is the constraints vector, and G is the directional constraint matrix.
For L elements in the array and constraints enforced in the K dierent directions
{Ω1 , Ω2 , . . . , ΩK }, G is an L×K matrix in which the ith column contains the signal
response vector of the array due to a plane wave incident from direction Ωi , and
the corresponding element gi in the vector g is the constraint value enforced on the
pattern in that direction. The choice of these constraint parameters is discussed in
the following subsections.
In this study the performance of the LCMV beamformer is compared to that
for the standard MaxSNR beamformer (no directional constraints). In this case the
beamformer weights are calculated according to [22] [21, p. 450]
wMaxSNR = C−1 v (5)
where v is the signal response vector of the array due to a plane wave incident from
the direction of interest. In this study a noiseless system is assumed which means
that the noise correlation matrix C can be taken equal to the identity matrix, and
therefore the weights in (5) maximize the received signal power. It can be shown
that this is approximately equivalent to maximizing the directivity, if the antenna
exhibits low loss and low scattering, as is the case for the PAF used herein. Therefore
the beamformer using the weights in (5) shall hereafter be referred to as a MaxDir
(Maximum Directivity) beamformer.
3 Henceforth we assume that (θ, φ) are dened in a local coordinate system for each beam in
which the maximum is at θ = 0.
176
3. Beamforming Strategy
3.1 Number of Constraints and Pattern Calibration Mea-

surements
Each of the weights applied to the PAF elements presents a complex Degree of Free-
dom (DoF) available for optimizing the beamformed pattern, and for each constraint
enforced on the pattern shape the number of DoFs available to maximize the directiv-
ity is reduced. The implication of this is that constraints should be selected carefully
to obtain the desired pattern shape while retaining enough freedom in the system to
achieve a suciently high directivity.
1.5
1
θ sin φ [degrees]
0.5
0
θc
-0.5
-1
-1.5
-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

θ cos φ [degrees]
Figure 1: Beams arranged over the FoV to enable reuse of constraint directions between adjacent
beams. Nominal half-power contours (HPBW = 1◦ ) and constraint positions of each beam shown
as solid lines and crosses, respectively.
Furthermore, the number of constraints has an impact on the calibration eciency

because a pattern calibration measurement is required for each constraint direction to
determine the signal response vector of the PAF [22]. Since these measurements can
become time consuming, we need to minimize the number of constraint directions to
ensure that the system parameters do not drift signicantly during this procedure. It
is worth pointing out that since both the amplitude and phase of the signal response
vectors are needed, this may require the use of an auxiliary antenna to recover the
phase information in addition to a natural celestial calibration source [23].
3.2 Constraint Positions

We aim to conform the beam to the reference pattern down to a certain level below the
beam maximum, so we choose to position the constraints within the corresponding
177
angular region. Also, the total required number of pattern calibration measurements
may be reduced by positioning the constraint directions at the centers of adjacent
beams, as shown in Fig. 1. This allows the reuse of measurement data between
multiple beams which is readily available in this type of measurement. In this example
six constraints are enforced in a circularly symmetric fashion around, and an angular
distance θc from the beam center for each beam. This arrangement results in a
ne enough sampling of the FoV since the half-power beams overlap [22], and the
constraints are enforced around the -8 to -5 dB level. In this case only 37 pattern
calibration measurements are needed to realize a total of 19 constrained beams over
the FoV, which is a minor increase over that for unconstrained beamforming as in (5).
The 18 additional measurements are necessary for the constraints enforced around
the edge of the FoV.
3.3 Constraints Vector

The constraints vector g in (4) is formed by evaluating the reference pattern in (3)
at the beam center and the directions of constraints Ωi = {θc , φi }, i.e.,
(
F (s, Ψ; 0, 0) for i=1
gi = (6)
F (s, Ψ; θc , φi ) for i = 2, 3, . . . , 7,
where the selection of the model parameters s and Ψ has to be made for each scan
direction to account for the beam widening and the increasing phase gradient over
the main lobe region. In order not to compromise the beam sensitivity too much,
it is natural to derive the initial physics-based values s = s0 and Ψ = Ψ0 from the
reference patterns realized by the MaxDir beamformer, i.e.,
√
ae,MaxDir λ DMaxDir
s0 = = (7a)
a 2πa
∂ψMaxDir
Ψ0 = (7b)
∂θ θ=0,φ=φ0
where ae,MaxDir is the eective aperture radius, and DMaxDir and ψMaxDir are the
directivity and phase pattern over the main lobe region, respectively, of the MaxDir
beam. Using thus obtained values for the parameters s and Ψ result in rotationally
symmetric beams that have sensitivities close to the MaxDir beams. However, this
choice leads to a sidelobe level (SLL) that can be relatively high for certain (o-axis)
beams. Hence, the optimum values for s and Ψ may be slightly dierent from s0 and
Ψ0 depending upon the required antenna beam performance, such as minimum beam
sensitivity and maximum allowable SLL, as explained below for a numerical example.
178
0
-10 s0
-0.2
Normalized directivity [dB]
-0.4
1st Sidelobe level [dB]

-0.6 -15
-0.8
s0
-1 -20
◦
-1.2 0 0◦
0.75◦ 0.75◦
-1.4 -25
1.5◦ 1.5◦
-1.6 2.25◦ 2.25◦
3◦ 3◦
-1.8 -30
0.65 0.7 0.75 0.8 0.85 0.9 0.95 0.65 0.7 0.75 0.8 0.85 0.9 0.95
Beamwidth scaling parameter s Beamwidth scaling parameter s
(a) (b)
0
-10 Ψ0
-0.2
Normalized directivity [dB]
-0.4
1st Sidelobe level [dB]

-0.6 -15
-0.8
Ψ0
-1 -20
-1.2 0◦ 0◦
0.75◦ 0.75◦
-1.4 -25
1.5◦ 1.5◦
-1.6 2.25◦ 2.25◦
3◦ 3◦
-1.8 -30
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30
Phase gradient Ψ − Ψ0 [radians] Phase gradient Ψ − Ψ0 [radians]
(c) (d)
Figure 2: Eect of model parameters on beam pattern performance. (a) and (c) show the directivity
of scanned LCMV patterns relative to that of the on-axis MaxDir pattern for various values of s and
Ψ, respectively; (b) and (d) show the highest SLL of scanned LCMV patterns for various values of
s and Ψ, respectively. Markers indicate the results for s = s0 and Ψ = Ψ0 for each scan direction.
4 Numerical Results
In this section we investigate the trade-o eects of the beam model parameters s and
Ψ on the directivity and SLL. After choosing s and Ψ, the beam model accuracy is
examined as the dierence between the resulting LCMV-beamformed pattern and the
reference beam. As a numerical example, we present results for an oset Gregorian
geometry based on the MeerKAT radio telescope reector antenna [15] by employing
simulated primary far-eld patterns of the APERTIF PAF [6]. The reector has a
projected diameter of 13.5 m (64λ at 1.42 GHz) and an equivalent focal length to di-
ameter ratio (F/D ) of 0.55. The APERTIF PAF is a dual-polarized array composed
of 121 tapered slot antenna elements. Here all elements in the array (both polar-
179
izations) are employed to produce patterns on the sky for each nominal polarization
(as opposed to a bi-scalar beamfomer wherein only elements of one polarization are
used, cf. [24]). Results presented here are for only one nominal polarization, as the
results for either polarization are very similar. The numerical results are shown for
the operating frequency of 1.42 GHz at which the half-power beamwidth (HPBW)
◦
is approximately 1 , and results were obtained using a toolbox interface [22] to the
GRASP software.
4.1 Beam Directivity and Side Lobe Levels

Fig. 2(a) shows the directivity of the LCMV-scanned patterns relative to the corre-
sponding MaxDir patterns as a function of s (with Ψ = Ψ0 ) over a scan range of
4
3 beamwidths in the symmetry plane . Markers indicate the results for the initial
values s = s0 that were derived from the MaxDir beams. As expected, the highest
directivity is achieved if s is close to s0 , except for far o-axis patterns where it occurs
for slightly smaller values of s. This is attributed to the fact that the computation
of s0 is based solely on the directivity of the MaxDir elliptically-shaped pattern,
while s0 is applied to rotationally symmetric patterns pertaining to the same eec-
tive aperture size. It is also observed that when choosing s = s0 , the loss incurred by
constrained beamforming is relatively small (< 0.4 dB) over the entire FoV. Letting
s→1 results in the edge illumination taper approaching 0 dB as the primary (feed)
pattern widens, and a subsequent decrease in directivity due to increasing spillover
loss.
The eect of s on the 1st SLL performance of the LCMV patterns is shown
in Fig. 2(b). As one can see, the choice of s = s0 leads to a signicant variation
of the SLLs of the scanned beams where the minimum (for the on-axis direction)
and maximum (for a scan angle of 3 beamwidths) are around -17 dB and -12 dB,
respectively. Decreasing the value of s improves the SLL, albeit at a moderate cost of
a reduction in directivity. This is in accordance with the familiar trade-o between
directivity and SLL for reector antennas. The 2nd SLL is aected similarly to the
1st SLL when s is varied, and these results are therefore not shown.
The eects of Ψ on the relative directivity and SLL of the LCMV beamformed
patterns were also investigated and the results are shown in Figs. 2(c) and (d). In
these gures, the abscissae represent the dierence Ψ − Ψ0 calculated for each scan
direction. Over the FoV, the value of Ψ0 decreases monotonically from 0 rad for the
on-axis pattern to -33 rad for the farthest o-axis scanned pattern, indicating a steady
shift of the antenna phase center from the phase reference point. Choosing Ψ close to
Ψ0 as opposed to setting it to zero, thereby eectively reducing the pattern model
to (1) resulted in a signicant improvement in directivity of far o-axis scanned
4 Although results are only shown for scanning in a single plane, the conclusions are valid for
scanning in all φ-directions.
180
patterns. This underlines the importance of using a proper reference pattern function
such as (3) which represents a more accurate description of the (o-axis) radiation
characteristics of the antenna. The eect of small variations of Ψ around the value
Ψ0 on the relative directivity and SLLs was found to be less pronounced than the
eect of parameter s, so that, generally, the choice Ψ = Ψ0 yielded the best results.
4.2 Calibration Performance

A deviation of the actual beam shape from the one predicted through calibration
measurements sets constraints on the dynamic range of the mosaicked images. Al-
though the relationship between the desired dynamic range and pattern calibration
error is very complex (and typically requires the analysis of the error propagation
eects in the image plane [25, 26]), the required accuracy of the pattern model can be
approximately derived from the required image delity [27] which is limited by the
maximum error present in the beam model. In this section we will therefore use the
maximum normalized error in the complex voltage pattern within the 10 dB region
when approximating realized beam patterns with (3) as a measure of beamshape
calibratibility.
The eect of s on the calibratibility of LCMV beamformed patterns is shown in

Fig. 3(a). Using the MaxDir equivalent value s = s0 the maximum error ranges from
0.7 % for on-axis up to 4 % for the widest scan angle. Decreasing the parameter over
the range s . s0 is seen to slowly increase the model error, whereas increasing s > s0
is seen to have a more dramatic eect on the model accuracy for wider scan directions
due to the increase of the 1st SLL above the 10 dB level. In Fig. 3(b) the eect of Ψ
on the calibratibility is shown and for this gure of merit the optimal choice for the
phase gradient is as before Ψ = Ψ0 .
Since the constrained beamformer ensures that the realized beam conforms exactly
to (3) at the constraint positions the model error is smallest in the vicinity of these
points and the placement of constraints may be optimized to minimize this error
within a certain angular region. The eect of θc on the pattern calibratibility is
shown in Fig. 3(c). Markers indicate the power level relative to pattern maximum
that correspond to the values of θc . The optimal placement of constraints is seen to
be around the -5 dB to -7 dB level.
Furthermore, to examine how the technique performs over a range of frequencies

we repeated the above described analysis at several frequencies within the antenna
array operation band from 1 to 1.75 GHz. In this study, frequency-dependent param-
eters were scaled for each of these frequencies (scan directions, positions of constraints
for LCMV beamforming, etc.). The results obtained have shown that at the lower
frequencies the FoV is limited by the size of the array (that is the case for any type of
the PAF beamformers), although the results for the on-axis and closer scanned direc-
tions are similar to those at 1.42 GHz. Hence, the advantages of using the proposed
181
LCMV-based beamforming are also applicable at lower frequencies over a relatively

smaller FoV. For higher frequencies, the results for all scan directions (within the
FoV of ±3 beamwidths) are very similar to those at 1.42 GHz. Based on these ob-
servations, we can conclude that the proposed beamforming technique ensures the
smooth characteristics of the resulting FoV calibration over a wide frequency band,
and does not require additional constraints due to frequency variation.
4.3 Comparison of MaxDir and LCMV beamformers

Using the results from Section 4.1, an LCMV beamformer was implemented to pro-
◦
duce a number of beams over a dense grid within an angular region θ < 3 . For
each LCMV beam the value of s was chosen such that the 1st SLL is below -17 dB,
and Ψ = Ψ0 as calculated from a MaxDir beam towards the same scan direction.
The performance of the MaxDir and LCMV produced beams were then compared for
every scan direction within the angular region of interest.
In Fig. 4(a) the aperture eciencies achieved with the respective beamformers are
shown as a function of scan direction. The asymmetry of the results over the FoV is a
◦
consequence of the oset geometry and wide scanning towards φ = 0 is seen to result
in the largest reduction in eciency. A FoV was dened for each beamformer as the
region within which the aperture eciency is greater than 70%, the size of which was
23.6 and 19.3 square degrees for the MaxDir and LCMV beamformers, respectively.
The boundary of each FoV is indicated on the plots in Fig. 4 as a solid black line,
and the results presented below were calculated within the respective regions for the
two beamformers.
The beamformers were compared by considering the maximum pattern calibration

model error for each of the dened beams over the FoV, which is shown in Fig. 4(b).
For the MaxDir beamformer this error ranges from 1.6% up to 10.4%, whereas for
the LCMV beamformer the same error ranges from 0.5% up to 4.3% and presents
a considerable improvement in accuracy. One prominent factor contributing to the
relatively large model error for MaxDir beams, especially at wider scan angles, is
the asymmetry of these patterns. As a comparison, the aspect ratio of the half-
power contours of the MaxDir beams may be as high as 1.15:1, whereas for the
LCMV beams this ratio is less than 1.01:1 for all scan directions. The symmetry of
constrained beams therefore also present a signicant advantage in terms of reducing
the complexity of direction-dependent calibration [6].
Finally, the maximum 1st and 2nd SLLs are shown as a function of scan direction
in Figs. 4(c) and (d), respectively. Compared to the MaxDir beams, the LCMV beams
have 1st SLLs that are 0.8 dB lower and 2nd SLLs that are 1.0 dB lower, on average
over the FoV. The 2nd SLL is of particular interest in the case of MeerKAT, for which
the maximum is specied as -23 dB (L-band). The LCMV beamformer meets this
◦ ◦
specication over most of the FoV (except for wide scanning in the φ ≈ 135 , 225
182
5. Conclusions and Recommendations
directions), whereas the MaxDir beams exceed this limit over a much larger region.
In order to quantify the trade-o in sensitivity for this reduction in sidelobes through
constrained beamforming, LCMV beams were also realized to yield 1st SLLs within
0.2 dB of that for the MaxDir beams. Following this approach the size of the FoV
could be increased by 4.6% to 20.2 square degrees.
5 Conclusions and Recommendations

A constrained beamforming technique that conforms multiple patterns on the sky
to a physics-based analytic far eld function was presented as a method to improve
the calibration eciency of an array fed reector antenna. The eects of the two
parameters in the analytic model on the pattern performance were investigated, and
a procedure by which these parameters could be optimized was proposed. This beam-
forming approach was shown to have several performance benets including circularly
symmetric scanned beams over a wide FoV, even for non-symmetric reector anten-
nas. For the example of the MeerKAT oset Gregorian antenna, this strategy resulted
in multiple beams with aperture eciency above 70% that could be approximated
down to the 10 dB level as a single analytic function with an error of less than 5%. In
comparison with a conventional MaxDir beamformer, this would reduce the average
pattern calibration model error by more than 50%. Finally, the proposed beamform-
ing strategy was found to be eective across a wide frequency band by simply scaling
all frequency dependent parameters.
Future work will include the assessment of the proposed beamforming in the
presence of external and internal noise sources, as well as experimental demonstration
for a practical system.
References
[1] O. Smirnov, Revisiting the radio interferometer measurement equation. ii.
calibration and direction-dependent eects, Astronomy and Astrophysics, vol.
527, Feb. 2011. [Online]. Available: http://arxiv.org/pdf/1101.1765v3.pdf
[2] S. Ellingson, T. Clarke, A. Cohen, J. Craig, N. Kassim, Y. Pihlstrom, L. Rickard,

and G. Taylor, The long wavelength array, IEEE Trans. Antennas Propag.,
vol. 97, no. 8, pp. 14211430, Aug. 2009.
[3] M. D. Vos, A. Gunst, and R. Nijboer, The lofar telescope: System architecture
and signal processing, IEEE Trans. Antennas Propag., vol. 97, no. 8, pp. 1431
1437, Aug. 2009.
[4] The SKA website. [Online]. Available: http://www.skatelescope.org/
183
[5] O. M. Smirnov, B. S. Frank, I. P. Theron, and I. Heywood, Understanding the

impact of beamshapes on radio interferometer imaging performance, in Proc.
pp. 586590.
[6] O. A. Iupikov, M. V. Ivashina, and O. M. Smirnov, Reducing the complexity of

the beam calibration models of phased-array radio telescopes, in Proc. European
Conference on Antennas and Propag. (EuCAP), Rome, Italy, Apr. 2011, pp.
930933.
[7] B. D. Jes, K. F. Warnick, J. Landon, J. Waldron, D. Jones, J. R. Fisher,

and R. Norrod, Signal processing for phased array feeds in radio astronomical
telescopes, IEEE J. Selected Topics in Signal Processing, vol. 2, no. 5, pp. 635
646, Oct. 2008.
[8] M. Elmer, B. Jes, K. F. Warnick, J. Fisher, and R. Norrod, Beamformer de-

sign methods for radio astronomical phased array feeds, IEEE Trans. Antennas
Propag., vol. 60, no. 2, pp. 930914, Feb. 2012.
[9] S. G. Hay, SKA eld of view de-rotation using connected array beam scanning
with constant beam shape, in Proc. 25th Asia-Pacic Microwave Conference,
Melbourne, Dec. 2011, pp. 11741177.
[10] S. J. Wijnholds, K. J. B. Grainge, and R. Nijboer, Report on calibration and

imaging working group activities, Calibration and Imaging Working Group,
Technical Report, 2012.
[11] A. Young, R. Maaskant, M. V. Ivashina, and D. B. Davidson, Performance

evaluation of far eld paterns for radio astronomy application through the use of
the Jacobi-Bessel series, in Proc. Int. Conf. on Electromagn. in Adv. Applicat.
(ICEAA), Cape Town, Sep. 2012, pp. 884887.
[12] V. Galindo-Israel and R. Mittra, A new series representation for the radiation
integral with application to reector antennas, IEEE Trans. Antennas Propag.,
vol. 55, no. 5, pp. 631641, Sep. 1977.
[13] Y. Rahmat-Samii and V. Galindo-Israel, Shaped reector antenna analysis using

the Jacobi-Bessel series, IEEE Trans. Antennas Propag., vol. 28, no. 4, pp. 425
435, Jul. 1980.

184
References
[15] I. Theron, R. Lehmensiek, and D. de Villiers, The design of the meerkat dish
optics, in Proc. Int. Conf. on Electromagn. in Adv. Applicat. (ICEAA), Cape
Town, Sep. 2012, pp. 539542.
[16] R. C. Hansen, A one-parameter circular aperture distribution with narrow

beamwidth and low sidelobes, IEEE Trans. Antennas Propag., vol. 24, no. 4,
pp. 477480, Jul. 1976.
[17] J. Ruze, Lateral-feed displacement in a paraboloid, IEEE Trans. Antennas

Propag., vol. 13, no. 5, pp. 660665, Sep. 1965.
[18] W. A. Imbriale, P. G. Ingerson, , and W. C. Wong, Large lateral feed displace-

ments in a parabolic reector, IEEE Trans. Antennas Propag., vol. 22, no. 6,
pp. 742745, Nov. 1974.
[19] Y. Y. Hu, A method of determining phase centers and its application to elec-
tromagnetic horns, J. Franklin Inst., Technical Report, Jan. 1966.
[20] O. I. Frost, An algorithm for linearly constrained adaptive array processing,
Proceedings of the IEEE, vol. 60, no. 8, pp. 926935, Aug. 1972.
Optimum Array Processing Part IV of Detection, Estimation,

[21] H. L. van Trees,

18641875, Jun. 2011.
[23] P. G. Smith, Measurement of the complete far-eld pattern of large antennas

by radio-star sources, IEEE Trans. Antennas Propag., vol. 14, no. 1, pp. 616,
Jan. 1966.
[24] W. van Cappellen, S. Wijnholds, and L. Bakker, Experimental evaluation of

polarimetric beamformers for an l-band phased array feed, in Proc. European
Conference on Antennas and Propag. (EuCAP), Prague, Czech Republic, Mar.
2012, pp. 634637.
[25] S. J. Wijnholds, Calculating beam pattern inaccuracies and their implications,

in in 3GC-II Workshop, Albufeira, Portugal, Sep. 2011.
[26] O. M. Smirnov and M. V. Ivashina, Element gain drifts as an imaging dynamic

range limitation in PAF-based interferometers, in General Assembly and Sci-
entic Symposium, 2011 XXXth URSI, Istanbul, Turkey, Aug. 2011, pp. 14.
185
[27] G. Harp, R. F. Ackermann, Z. J. Nadler, S. Blair, M. Davis, M. C. H. Wright,

J. R. Forster, D. DeBoer, W. J. Welch, S. Atkinson, D. Backer, P. R. Backus,
W. Barott, A. Bauermeister, L. Blitz, D. C. J. Bock, G. Bower, T. Brad-
ford, C. Cheng, S. Croft, M. Dexter, J. Dreher, G. Engargiola, E. D. Fields,
C. Heiles, T. Helfer, J. Jordan, S. Jorgensen, T. Kilsdonk, C. Gutierrez-Kraybill,
G. Keating, C. Law, J. Lugten, D. H. E. MacMahon, P. McMahon, O. Milgrome,
A. Siemion, K. Smolek, D. Thornton, T. Pierson, K. Randall, J. Ross, S. Shostak,
J. Tarter, L. Urry, D. Werthimer, P. Williams, and D. Whysong, Primary beam
and dish surface characterization at the Allen telescope array by radio hologra-
phy, IEEE Trans. Antennas Propag., vol. 59, no. 6, pp. 20042021, Jun. 2011.
186
References
101 s0
Error [%]
0◦
0.75◦
1.5◦
100 2.25◦
3◦
0.65 0.7 0.75 0.8 0.85 0.9 0.95

Beamwidth scaling parameter s
(a)
101
Error [%]
0◦
0.75◦
1.5◦
100 2.25◦
Ψ0 3◦
-30 -20 -10 0 10 20 30

Phase gradient Ψ − Ψ0
(b)
0◦
0.75◦ -10 dB
1.5◦
2.25◦
101 3◦
-8 dB
Error [%]
100 -3 dB
-5 dB
0.3 0.4 0.5 0.6 0.7 0.8 0.9
Constraint position θc [degrees]
(c)
Figure 3: Maximum normalized error over the 10 dB beamwidth of the LCMV beamformed patterns
when approximated by the analytical function (3), and using the same parameter values as was used
to dene directional constraints. The error is shown as a function of the parameter (a) s, (b) Ψ−Ψ0 ,
and (c) θc . Default values for these parameters are s = s0 , Ψ = Ψ0 , and θc = 0.75◦ .
187
MaxDir LCMV
80
3 3
Aperture efficiency [%]

2 2
θ sin φ [degrees]
θ sin φ [degrees]
1 1
0 0 70
-1 -1
-2 -2
-3 -3
60
-4 -2 0 2 4 -4 -2 0 2 4
θ cos φ [degrees] θ cos φ [degrees]
(a)
15
3 3
Model error [%]

2 2
θ sin φ [degrees]
θ sin φ [degrees]
1 1 10
0 0
-1 -1 5
-2 -2
-3 -3
0
-4 -2 0 2 4 -4 -2 0 2 4
(b)
-15
3 3
2 2 1st SLL [dB] -16
θ sin φ [degrees]
θ sin φ [degrees]
1 1
-17
0 0
-18
-1 -1
-2 -2 -19
-3 -3
-20
-4 -2 0 2 4 -4 -2 0 2 4
(c)
-21
3 3
2 2 -22
2nd SLL [dB]
θ sin φ [degrees]
θ sin φ [degrees]
1 1
-23
0 0
-24
-1 -1
-2 -2 -25
-3 -3
-26
-4 -2 0 2 4 -4 -2 0 2 4
(d)
Figure 4: Comparison of LCMV and MaxDir beamformers over a θ ≤ 3◦ angular region based on
(a) aperture eciency, (b) maximum beam model error, (c) 1st SLL, and (d) 2nd SLL. Figures of
merit are shown as functions of beam steering direction over the FoV. Solid lines on all plots indicate
the FoV within which aperture eciency is above 70% for each beamformer. The asymmetry in the
results is due to the oset geometry of the antenna.
188
Paper G
Domain-Decomposition Approach to Krylov

Subspace Iteration
O. A. Iupikov; C. Craeye; R. Maaskant and M. V. Ivashina
EEE Antennas and Wireless Propagation Letters, vol. 15, 2016
the thesis.
Domain-Decomposition Approach to Krylov
Subspace Iteration
O. A. Iupikov; C. Craeye; R. Maaskant and M. V. Ivashina
Abstract
Krylov subspace iterative techniques consist of nding the solution of

a scattering problem as a linear combination of generating vectors ob-
tained through successive matrix-vector multiplications. This paper ex-
tends this approach to domain-decomposition. Here, on each subdomain
a subspace is obtained by constructing the segments of each generating
vector associated with the subdomain, and by weighting these segments
independently, which provides more degrees of freedom. The method is
tested for scattering by a sphere and a rectangular plate, as well as radi-
ation from connected arrays with strongly coupled antenna elements. It
is shown that substantial computational savings can be obtained for the
sphere and the array. This opens up new perspectives for faster solutions
of multi-scaled problems.
1 Introduction
Conventional iterative techniques, such as the Full Orthogonalization Method (FOM)
or the Generalized Minimal Residual Method (GMRES) [1], have proved their ca-
pability and eciency to solve large-scale electromagnetic problems. They dene a
set of current distributions on the whole domain through successive matrix-vector
multiplications (mat-vecs) and then solve for their expansion coecients in an it-
erative manner. However, these methods become computationally expensive for a
large number of generating vectors. To avoid this, a restart procedure is often used,
which in addition helps to improve the condition number of the generated system of
equations, thereby improving the accuracy of the method.
Many improvements on the GMRES method can be found in the literature. For
example, in [2] an adaptive deation strategy is proposed, which retains useful in-
formation at the time of a restart to avoid stagnation and improve the convergence
rate.
The generating vectors in GMRES can also be seen as Macro Basis Functions
(MBFs) [3, 4]. A similar approach is used in domain-decomposition methods like
the Characteristic Basis Functions Method (CBFM) [5] and the Synthetic Functions
method (SFX) [6]. A major dierence between them is that MBFs in GMRES (or
191
Paper G. Domain-Decomposition Approach to Krylov Subspace Iteration
FOM) are dened on the whole computational domain and belong to a Krylov sub-
space, while CBFM-like techniques split the structure into subdomains and analyze
them in isolation through the denition of set of independent MBFs on each subdo-
1
main, obtained by exciting the subdomain in various ways . Assuming that MBFs
are obtained using a multiple-scattering (between subdomains) methodology, a rule
of thumb is proposed in [7] stating that both FOM and CBFM provide a similar
accuracy when the number of iterations in FOM is equal to the average number of
MBFs per subdomain in CBFM. However, in some cases, the CBFM yields better
accuracy, owing to the fact that it provides more degrees of freedom (DoFs).
In this paper we propose a domain-decomposition approach to Krylov subspace

iteration, where MBFs (or CBFs) on each subdomain are naturally constructed from
the dierent segments of the generating vectors.
The paper is organized as follows. In Sec. 2 we formulate a reduced system of

equations built from segments of the generating vectors as MBFs, while avoiding
extra mat-vecs. Next, an algorithm for the restart procedure is described in Sec. 3.
The proposed approach is validated in Sec. 4 for the case of a perfectly conducting
sphere, a rectangular plate, and an array of electrically connected and disconnected
tapered-slot antennas. A discussion about complexity and accuracy of the approach
follows in Sec. 5 and conclusions are drawn in Sec. 6.
2 Segmented Krylov subspace as MBFs

Consider the Method of Moments (MoM) matrix equation:
ZI = e, (1)
where Z is the N × N MoM matrix; e is the N × 1 excitation vector and I is a vector

containing the expansion coecients for the elementary basis functions. Accordingly,
the reduced CBFM system of equations can be written as
Z̃Ĩ = ẽ, (2a)
Z̃i,j = KH
i Zi,j Kj , (2b)
ẽi = KH
i ei , (2c)
where i, j = 1 . . . M are sub-domain indices; H is the Hermitian operator and Ki

is the set of MBFs. The method proposed here consists of selecting as MBFs on
a given subdomain the corresponding segments of the generating vectors. Those
1 More about the relationship between CBFM-like approaches and Krylov subspace iterative
methods can be found in [7].
192
3. CBFM with restarts
segments correspond to entries associated to basis functions dened on the subdomain

of interest. Hence, the newly proposed MBF selection reads:
h i
(1) (2) (P )
Ki = ki = ei | ki | . . . | ki , (3)
in which generating vector k is formed iteratively as
k(p+1) = Zk(p) for p = 1 . . . P − 1, (4)
where index i refers to the MBF vector entries related to subdomain i.

It is important to point out that the most computationally expensive part of
the MoM matrix reduction (2b), namely the matrix-matrix product Zi,j Kj , can be
carried out during the subspace construction (4). For this purpose, (4) is built from
(p)
M 2 smaller matrix-vector products resulting in the M 2 vectors vi,j , expressed as
(p) (p)
vi,j = Zi,j kj . (5)
Segment i of the vector k(p+1) (at the next iteration) is obtained by a simple summa-
(p)
tion of vectors vi,j as
(p+1) (p)
X
ki = vi,j . (6)
j
(p)
If the vectors vi,j are concatenated in a matrix Q as
h i
(1) (2) (p)
Qi,j = vi,j | vi,j | . . . | vi,j , (7)
then the MoM matrix reduction (2b) can be rewritten as
Z̃i,j = KH H
i Zi,j Kj = Ki Qi,j , (8)
which allows one to reduce by a factor close to two the time involved in (2)-(4), as
compared to a straight-forward implementation. The appendix explains how (8) can
be modied when the set of MBFs needs to be orthogonalized.
3 CBFM with restarts

The accuracy of the CBFM method can be signicantly improved down to machine
precision by introducing a restart procedure similar to that used in a restarted GM-
RES method [1]:
Step 1. Initialize the nal solution In = 0.
Step 2. Set the excitation vector e in (2c) to the initial excitation vector e0 .
193
Step 3. Build and solve the reduced system of equations (2a), compute the
solution Ij = Kj Ĩj for j = 1, . . . , M . Note that the reduced system of equations
can be built progressively, similar to the internal iterations in GMRES.
Step 4. Add the result to the nal solution, In = In + I.
Step 5. Compute the residue r = e0 − ZIn .
Step 6. Set the excitation vector e to the residue r and go to Step 3 until the
required residue is reached.
The main dierence with GMRES is that the subspace is restarted on every
subdomain.
4 Numerical results
In this section the proposed approach is compared to the GMRES algorithm in terms
of an error in surface current versus the solving complexity. The complexity is de-
ned herein as the number of elementary operations ab+ (oating point product of
complex scalar numbers and summation with another complex number), required to
solve the problem, while the relative error in the surface current is computed as
 
s s
X approx
− Inref |2 |Inref |2  ,
X
= 20 log10  |In (9)
n n
where Iapprox is the current expansion coecient vector, obtained using the proposed
approach or restarted GMRES; and Iref is the reference solution, obtained by direct
solution of the MoM matrix equation (1).
The structures considered hereafter are subdivided into subdomains to have nearly
equal and compact surfaces; for antenna arrays each subdomain is chosen to be a
single antenna element. One possible way to improve the division into subdomains
is through the so-called graph-partitioning technique (see e.g. [8]).
The system of equations (1) is assumed to be preconditioned for both CBFM
and GMRES using the preconditioner described in [9], during which auxiliary subdo-
mains are considered [7] in order to deal with the nearest interactions. Furthermore,
a simplied version of GMRES is used [10], which implements the Gram-Schmidt
orthogonalization of the vectors in the Krylov sub-space instead of using the Arnoldi
iteration. This approach has a similar complexity as the original GMRES algorithm,
while it is structurally closer to the CBFM.
For each geometry considered below a series of simulations have been performed
for dierent numbers of CBFM-generating vectors and dierent numbers of internal
iterations (between restarts) for GMRES, and the best convergence curves of both
194
4. Numerical results
(a) 96 subdomains. (b) 384 subdomains.
0
-50
-100
GMRES: 384 subdomains (P=24)
CBFM: 384 subdomains (P=12)
-150
0 10 20 30 40 50 60 70 80 90 100
Complexity, 10 9
Figure 1: Numerical example 1: A sphere with radius 1.58λ, divided into (a) 96 subdomains and (b)
384 subdomains, and excited by an incident plane wave. Subgure (c) compares the convergence
rates of restarted GMRES and CBFM. The restart positions are indicated with circles.
methods are compared. Under an iteration for CBFM approach we understand
hereafter a procedure consisting of (i) building the reduced system of equations of
size P M × P M, which involves P mat-vecs, and (ii) solving this system.
Three geometries are considered:
(Fig. 1) A sphere with radius 1.58λ, divided into M = 96 or M = 384 subdo-

mains. The number of elementary basis functions (RWG) is N = 30720. The
sphere is excited by a plane wave.
(Fig. 2) A rectangular plate with size 12λ, M = 144 or M =

divided into
256 subdomains. The number of RWG basis functions is N = 42960. The plate
is excited by a plane wave under 45 degrees incidence.
195
(a) 144 subdomains. (b) 256 subdomains.
0
-50
-100
-150
0 10 20 30 40 50 60 70
9
Complexity, 10
Figure 2: Numerical example 2: A rectangular plate with size 12λ, divided into (a) 144 subdomains
and (b) 256 subdomains, and excited by a plane wave under 45 deg incidence. Subgure (c) compares
the convergence rates of restarted GMRES and CBFM. The restart positions are indicated with
circles.
(Fig. 3) A 121-element dual-polarized array of both connected and disconnected

Vivaldi tapered slot antennas. The numbers of RWG basis functions are N =
41975 and N = 39325 respectively. The array is uniformly excited by delta-gap
voltage sources at each antenna element. The connected Vivaldi array has been
designed by ASTRON [11].
Fig. 1 demonstrates the convergence rate of the newly dened iterative CBFM and
GMRES for the sphere. If one aims at an accuracy in the surface current of e.g. 50 dB,
the domain-decomposition approach is more than twice faster than GMRES, i.e. with
twice smaller operations count. The convergence in case of 96 subdomains is faster
for both methods, and this can be explained by the inuence of the preconditioner,
which accounts for all adjacent neighbours of each subdomain. This is true as long
as the solution time of the reduced system of equations is small compared to the
matrix-vector product needed to produce that system of equations. As explained in
Section 5, this supposes that the number of subdomains M remains small compared
2 2
to F , where F > 1 is the DoF reduction factor , which is satised in all numerical
2 F = N /P is average ratio between numbers of elementary basis functions and MBFs on each
sd
196
4. Numerical results
examples considered here.

Similar observations are made for the square plate shown in Fig. 2. However, the
advantage of using the iterative CBFM is not signicant in this case as compared to
the more strongly coupled subdomains in the other examples.
(a) connected array. (b) disconnected array.
GMRES: connected array (P=18)

0 CBFM: connected array (P=20)
GMRES: disconnected array (P=16)

CBFM: disconnected array (P=12)
-20
-40
-50
-60
-80
61.69 161.52
-100
0 20 40 60 80 100 120 140 160
Complexity, 10 9
Figure 3: Numerical example 3: (a) A connected and (b) disconnected 121-element dual-polarized
Vivaldi array, divided into 121 subdomains, and excited by a delta-gap voltage sources at each
antenna element. Subgure (c) compares the convergence rates of restarted GMRES and CBFM.
The restart positions are indicated with circles.
The more complicated Vivaldi array case is demonstrated in Fig. 3. As expected,

the disconnected array is a much easier case for numerical analysis, since no current
can physically ow from one subdomain to another, which reduces mutual coupling,
such that the convergence is much faster. The gure also shows that for the connected
subdomain
197
array the proposed domain-decomposition approach is more than a factor two faster,
as compared to a conventional GMRES approach.
In all numerical examples the CBFM reaches an accuracy better than −50 dB in
only 1 to2 iterations (for 0 to 1 restarts), with the number of mat-vecs per iteration
equal to P as indicated in the legends of Figs. 13. GMRES requires 1 to 5 restarts
to achieve similar accuracy levels.
It worth noting that we used an integral error in the surface current as a main
gure of merit in this study. However, antenna characteristics, such as the antenna
impedance and radiation pattern, are most commonly used by antenna designers. The
relation between respective errors is not straightforward, however it can be assumed
that the error in surface current and the error in antenna characteristics are of same
order (see e.g. the approximation error analysis in [12], where dierent reector
antenna feeds are considered).
5 Discussion
When well preconditioned, GMRES converges very rapidly (i.e. within a few tens
of iterations), almost irrespective of the number of unknowns. As explained in [10],
GMRES amounts to solving a reduced system of equations, whose size (i.e. number
of DoFs), corresponds to the number of iterations. For large problems, this solution
takes a negligible time as compared to that involved in the mat-vec operations. This
means that, without signicant increase in the computation time, one can aord
more DoFs, as is the case with the approach proposed here, since the number of
DoFs now corresponds to the number of mat-vecs P multiplied by the number of
subdomains. Without any specic matrix-vector multiplication, solving the reduced
3
system of equations has a complexity (P M ) (here it is worth to mention that there
are methods to reduce this exponent, see e.g. [13]), while the complexity of mat-vecs
2 2 2
is P N . The increase of computational time is hence small as long as P M Nsd ,
2
where Nsd is the average number of elementary basis functions per subdomain.
It is pointed out that the gained accuracy does not seem to be commensurate
with the increase of the degrees of freedom. More precisely, the achieved accuracy
is not as good as that we may expect from GMRES when the number of iterations
equals the total number of CBFs in the problem (in that case GMRES exploits the
same number of DoF, at the expense of an excessive number of mat-vecs). This is
probably due to the possible slight discontinuity between current distributions on
contiguous subdomains; part of the newly generated DoFs may actually be needed
to correct this deciency.
In the very worst case, i.e., when the iterative CBF approach essentially provides
the same accuracy as GMRES, one has two methods, one based on GMRES and
one based on CBFs, with comparable accuracies when the number of iterations in
the former is equal to the number of CBFs per subdomains in the latter. That
198
6. Conclusions
equality is obtained by construction of the proposed method, since one new CBF per
segment from the new generating vector is created at every iteration. It is interesting
to notice that this equality precisely corresponds to the rule of thumb delineated
from numerical experiments in [7] where MBFs (or CBFs) were created in a multiple-
scattering fashion, and it is shown here that this rule of thumb constitutes a lower
bound for the capabilities of the iterative CBF (or MBF) approach.
It appears that a clear advantage beyond this rule is obtained with CBFs when
as proposed here the CBFs on a given subdomain are simply taken as the segments
of the generating vectors (which correspond to the subdomain of interest). Other
(either purely algebraic or more physical) ways of creating the CBFs may allow us to
further benet from the larger number of DoFs created through the subdomain-based
approach.
6 Conclusions
This work has introduced a domain-decomposition technique into Krylov subspace
iteration, such as in GMRES for instance. This method is similar to the CBFM, here
the MBFs are generated by simple segmentation of the pre-computed vectors of the
Krylov subspace. The achieved convergence is faster than with GMRES by a factor
ranging from 1.05 (the rectangular plate with large subdomains) to 2.6 (the connected
Vivaldi array) while keeping the same accuracy. This opens new perspectives for the
solution of multi-scaled radiation and scattering problems.
Appendix
To keep a well-conditioned reduced system of equations, the set K of MBFs should
be orthogonalized by means of, e.g., a QR-decomposition. This slightly complicates
the acceleration technique described in the Sec. 2. The updated acceleration can be
carried out in the following way.
Let us denote the orthogonalized matrix K as Ko , then (2b) becomes
Z̃i,j = Koi H Zi,j Koj . (10)
After performing the QR-decomposition for each sub-domain j , Kj = Koj Rj , Eq. (10)
can be rewritten as
Z̃i,j = Koi H Zi,j Kj R−1 oH −1

j = Ki Qi,j Rj , (11)
which only involves small matrices, and based on (7), is the nal expression of the
(i, j) block of the reduced MoM matrix.
199
References
[1] Y. Saad and M. H. Schultz, GMRES: a generalized minimal residual algorithm
for solving nonsymmetric linear systems, SIAM Journal on Scientic and Sta-
tistical Computing, vol. 7, no. 3, pp. 856869, Jul. 1986.
[2] D. N. Wakam, J. Erhel, and W. D. Gropp, Parallel adaptive deated GMRES,

in The 20th International Conference on Domain Decomposition Methods, UC
San Diego, in La Jolla, California, Feb. 2011, pp. 631638.
[3] E. Suter and J. R. Mosig, A subdomain multilevel approach for the ecient
MoM analysis of large planar antennas, Micr. Opt. Technol., vol. 26, no. 4, pp.
270277, Aug. 2000.
[4] I. Stevanovic and J. R. Mosig, Subdomain multilevel approach with fast MBF
interactions, in Proc. IEEE AP-S International Symposium, Monterey, Califor-
nia, Jun. 2004, pp. 367370.
[5] V. Prakash and R. Mittra, Characteristic basis function method: A new tech-
nique for ecient solution of method of moments matrix equations, Micr. Opt.
Technol., vol. 36, no. 2, pp. 95100, Jan. 2003.
[6] L. Matekovits, G. Vecchi, G. Dassano, and M. Orece, Synthetic function anal-

ysis of large printed structures: the solution space sampling approach, in Proc.
IEEE AP-S International Symposium, vol. 2, Boston, Massachusetts, Jul. 2001,
pp. 568571.
[7] N. Ozdemir, D. Gonzalez-Ovejero, and C. Craeye, On the relationship between

multiple-scattering Macro Basis Functions and Krylov subspace iterative meth-
ods, IEEE Trans. Antennas Propag., vol. 61, no. 4, pp. 20882098, Apr. 2013.
[8] R. Mitharwal and F. P. Andriulli, On the multiplicative regularization of graph

laplacians on closed and open structures with applications to spectral partition-
ing, IEEE Access, vol. 2, pp. 788796, 2014.
[9] N. Ozdemir, D. Gonzalez-Ovejero, and C. Craeye, A near-eld preconditioner

preserving the low-rank representation of method of moments interaction ma-
trices, in Proc. Int. Conf. on Electromagn. in Adv. Applicat. (ICEAA), Torino,
Italy, Sep. 2013, pp. 133136.
[10] C. Craeye, J. Laviada, R. Maaskant, and R. Mittra, Macro Basis Function

framework for solving Maxwell's equations in surface integral equation form,
The FERMAT Journal, vol. 3, pp. 116, 2014.
200
References
[12] O. A. Iupikov, R. Maaskant, M. Ivashina, A. Young, and P. Kildal, Fast and

accurate analysis of reector antennas with phased array feeds including multiple
reections between feed and reector, IEEE Trans. Antennas Propag., vol. 62,
no. 7, Jul. 2014.
[13] A. Boja«czyk, Complexity of solving linear systems in dierent models of com-

putation, SIAM Journal on Numerical Analysis, vol. 21, no. 3, pp. 591603,
Jun. 1984.
201
202
Paper H
Design of a push-broom multi-beam radiometer for

future ocean observations
C. Cappellin, K. Pontoppidan, P. H. Nielsen, N. Skou, S. S. Søbjærg,
A. Ihle, M. V. Ivashina, O. A. Iupikov, and K. v. 't Klooster
in Proceedings of the 9th European Conference on Antennas and

Propagation, EUCAP 2015, Lisbon, Portugal, April 2015.
the thesis.
Design of a push-broom multi-beam radiometer
for future ocean observations
C. Cappellin, K. Pontoppidan, P. H. Nielsen, N. Skou, S. S. Søbjærg, A. Ihle,
M. V. Ivashina, O. A. Iupikov, and K. v. 't Klooster
Abstract
The design of a push-broom multi-beam radiometer for future ocean ob-

servations is described. Such a radiometer has the big advantage of being
fully stationary on the platform and provides a sensitivity one order of
magnitude higher than a traditional conical scanning radiometer. Thanks
to a dense focal plane array and a dedicated optimization procedure, the
radiometric performance can be optimized and the instrument can accu-
rately measure in C, X and Ku band and as close as 15 km from the coast
line.
1 Introduction
The oceanographic community has strong interest in high spatial resolution. Current
microwave radiometers in space operating at C-band (6.9 GHz) or at higher frequency
provide a spatial resolution of around 50 km, whereas less than 20 km is desirable.
Current capabilities provide measurements not closer than around 100 km from the
shore-line, because of the signal contamination by the antenna side-lobes illuminating
the land. There is a strong desire to reduce this distance to 5-15 km.
The instrument requirements for future radiometers measuring sea surface tem-
perature (SST) and ocean vector wind (OVW) are summarized in Table 1. The
instrument shall operate in three well separated bands, C-band (6.9 GHz), X-band
(10.65 GHz) and Ku-band (18.7 GHz). The required 20 km resolution, i.e. 3 dB
footprint, at C-band leads to a large antenna aperture of around 5 m in diameter.
This is considerably larger than any radiometer system antenna own hitherto.
The conical scanning antenna rotates around a vertical axis, and the coverage of
the Earth is obtained partly by the movement of the satellite and partly by the rota-
tion. For the push-broom system there are no moving parts but the antenna radiates
as many beams as required to cover the swath. The push-broom system achieves very
high sensitivity since all across track footprints are measured simultaneously. The
antenna has the clear advantage of being stationary, but the number of beams and
receivers is very high.
205
Paper H. Design of a push-broom multi-beam radiometer for future ocean...
Freq., Bandwidth, Polari- Sensiti- Accuracy, Resolution, Dist. to

[GHz] [MHz] zation vity, [K] [K] [km] coast, [km]
6.9 300 V, H 0.30 0.25 20 5-15
V, H 0.22 0.25 20 5-15
10.65 100
S3 , S4 0.22 0.25 20 5-15
V, H 0.25 0.25 10 5-15
18.7 200
S3 , S4 0.25 0.25 10 5-15
The trade-o between a conical scanning and a pushbroom scanning relative to

reector optics and feed array design was described in [1], and indicated the push-
broom antenna as the most promising candidate.
The purpose of the present paper is to focus on the detailed design of the push-
broom antenna radiometer.
The paper is organized as follows: in Sec. 2 the antenna requirements are sum-
marized, in Sec. 3 the geometry of the push-broom antenna and its focal plane array
are described. The principles behind the optimization of the focal plane array are
given in Sec. 4, while the detailed RF performances of the antenna are given in Sec. 5.
Finally, Sec. 6 describes the mechanical realization of the push-broom reector and
Sec. 7 summarizes important results on the feeding network and the necessary power.
2 Antenna requirements
The requirements for the radiometer antenna can be derived from the radiometric
requirements of Table 1.
One requirement concerns the cross-polarization of the antenna. The radiometer

shall measure brightness temperatures in two linear polarizations, vertical and hori-
zontal, and with an accuracy of 0.25K. It can be demonstrated [1] that this will be
fullled when the cross-polar power received from the Earth does not exceed 0.33%
of the total power coming from the Earth for that polarization state.
The instrument must be able to measure as close as 5-15 km from the coast. The
brightness temperature of the sea is between 75 and 150K, whereas the land is 250K.
The power in the pattern over the land shall be suciently small. It can be found [1]
that the required accuracy is obtained when the coast line is located outside a cone
around the main beam containing 99.72% of the total power on the Earth. In order
to obtain a small distance to coast it is therefore of interest to reduce this cone.
The satellite height above the Earth and the incidence angle are assumed equal to
817 km and 53◦ , respectively. The required swath width was initially set to 1500 km.
It was however realized very early in the study that, as far as push-broom systems are
concerned, this will lead to a very large antenna. It was therefore decided to reduce
206
3. Push-broom antenna
Figure 1: Push-broom torus design.
Figure 2: Torus push-broom antenna with

projected aperture D and focal length f of
5 m. The three feed positions represent scan
directions of 0◦ and ±20◦ . The feed array is
shown in blue.
the swath width to 600 km. Even with this reduction the radiometer will represent
a major advancement in the study of the oceans.
3 Push-broom antenna
3.1 Antenna geometry
For the push-broom system a torus reector has been considered with projected
aperture D of 5 m. The torus surface is obtained by rotating a section of a parabolic
arc around a rotation axis, as shown in Fig. 1. The focal length f of the parabolic
generator has been selected as 5 m. The angle α between the rotation axis and the
parabola axis is connected to the orbit geometry, including the satellite height and
the required incidence angle on ground. The distance p from the parabola vertex to
the rotation axis is a function of f and α. The feed axis is selected parallel to the
rotation axis, implying that all feed element axes are parallel and orthogonal to the
focal plane. The feed array becomes therefore planar, simplifying the mechanical and
electrical design. The reector rim is found by intersecting the torus surface by the
feed cone up to the outmost scan positions.
The antenna shall be able to provide a scan of ±20◦ corresponding to a swath
207
width of 600 km. The nal design is shown in Fig. 2, where the projected reector
aperture is 5 m by 7.5 m.
3.2 Feed array

The sensitivity provided by the torus push-broom is always one degree of magnitude
better than the one provided by the conical scanner. This is at the expenses of a very
large number of beams, and correspondingly large number of receivers. For a swath
of 600 km we need:
156 beams across track at 18.7 GHz.
The array elements are arranged in a ρφ-grid around the rotation axis. The
distance between the elements is approximately the same in the ρ-direction and in
the φ-direction, and set equal to 0.75 wavelength. This distance was proven to be
the optimal distance. For analysis purposes the array elements are assumed to be
half-wave dipoles located a quarter of a wavelength above an innite ground plane.
Each element consists of both an x- and a y -directed dipole with separate ports.
They only radiate in the upper half space above the ground plane.
4 Feed array design principles

Two dierent methods have been applied in order to determine the excitation coef-
cients of the feed array: the Conjugate Field Matching (CFM) and a direct opti-
mization of the distance to coast.
It was realized early in the project that a traditional one beam-per-feed arrange-
ment was not possible and a dense focal plane array was needed. This means that
many array elements take part in the formation of one beam and the same array
element takes part in the formation of many beams. The composite feed array must
be excited by a multi-mode beam-forming network.
The feed array design was investigated almost independently by both TICRA and
Chalmers. Initially, TICRA used the Conjugate Field Matching method to determine
the feed element excitations. It turned out that this approach was too restrictive and
gave rise to a number of feed elements larger than the one obtained by Chalmers [1]
with a dedicated optimization procedure.
Since behind each feed element sit two receivers for the respective orthogonal
polarization states, it is of high importance to reduce the number of receivers to the
minimum necessary, in order to minimize the power consumption and the complexity
of the feed array. An alternative technique was thus developed by TICRA, by which
208
5. RF performance results
the array excitations are obtained by directly minimizing the distance to coast. It
turns out that the optimization can be formulated as an eigenvalue problem, where
the eigenvalue represents the maximum radiated power inside a given cone and the
eigenvector holds the excitations to generate this eld. The number of elements along
the ρ and φ direction must be given as input to the algorithm.
The optimization method used by TICRA is similar to the one developed by

Chalmers [2]. The dierence lies in the way the cost function is dened: it is the
ratio of the power inside and outside the angular cone for TICRA; while it is the ratio
of the power inside a specied small region to the noise power outside this region for
Chalmers. The radiometric performances obtained by the two algorithms are very
similar.
5 RF performance results
5.1 Central beam at Ku band
The reector surface is not a paraboloid and the performances are therefore expected
to be most critical at the highest frequency. In this section the central beam at
Ku-band, 18.7 GHz, is thus presented.
The feed array has 8 elements in the ρ-direction and 21 elements in the φ-direction,
as indicated by Chalmers. The total number of array elements to generate the central
beam is therefore 168. The element excitations are determined by TICRA's optimiza-
tion approach described earlier and the result shows that 99.72% of the power from
◦
the antenna is contained inside a cone with half angle 0.5 . The synthesized excita-
tion coecients in amplitude are shown in Fig. 3 and the far eld from the feed array
at 18.7 GHz is shown in Fig. 4. It is seen that the radiation outside the reector rim
is very low leading to a spill-over of only 0.05%. The far-eld pattern of the antenna
is depicted in Fig. 5. It is evident that this pattern is not rotationally symmetric and
one could therefore get the impression that the actual orientation of the coast line
would be very important for the instrument performance quality. When the -30 dB
◦
contour is plotted, one can nd that it is nearly a circle with radius 0.5 . This circle
contains 99.72% of the power and the coast line can therefore be located anywhere
outside this circle and its orientation is not important.
The radiometer characteristics are shown in Table 2 and include not only the
centre frequency but also the two band ends in order to demonstrate that the per-
formance is almost constant across the entire band. It is noticed that the distance to
coast at Ku-band is only 7 km.
209
Figure 3: Excitation coecients for the centre beam for minimum distance to coast.
Figure 4: Far-eld radiation pattern of the Figure 5: Image plot of the co-polar far eld
feed array for the centre beam. of the centre beam for the push-broom an-
tenna optimised for low distance to coast.
Table 2: Radiometer characteristics for the central ku-band beam for the push-broom antenna
optimized for low distance to coast.
Frequency, [GHz] Cx-power], [%] Footprint, [km] Dist. to coast, [km]

18.6 0.15 10.34 7.16
18.7 0.14 10.32 7.08
18.8 0.14 10.31 7.02
5.2 Reduction of feed array rows along φ

It was demonstrated in the previous section that a feed array with 8 rows along φ
gives a distance to coast of 7 km which is actually better than the required 15 km.
It was investigated if it is possible to reduce the number of feed array rows to 7 or 6
and still maintain an acceptable performance.
The excitations are determined such that 99.72% of the power is contained in
the smallest possible cone around the beam peak. Using a smaller number of rows
generates a more elliptical illumination on the reector and a more elliptical far-eld
beam. The radiometer characteristics are summarized in Table 3 where it is seen
210
5. RF performance results
that with 6 rows the footprint is slightly larger than 10 km and the distance to coast
is smaller than 10 km, thus acceptable.
Table 3: Radiometer characteristics for the central ku-band beam for dierent number of feed array
rows along φ.
Feed array rows Number of Cx-power, Footprint, Dist. to

along phi active [%] [km] coast, [km]
elements
x-dir y -dir
8 rows 168 0 0.14 10.34 7.16
7 rows 147 0 0.11 10.32 7.08
6 rows 126 0 0.08 10.31 7.02
5.3 Total feed arrays for C-, X- and Ku-band

It was demonstrated in the previous section that acceptable performance for the
centre beam at Ku-band can be achieved with a feed array containing 6 elements in
the ρ-direction and 21 elements in the φ-direction.
We will use the same principles as for Ku-band to design the feed arrays for the
centre beams for C- and X-band, still using 6 × 21 elements. The element excitations
are determined by optimizing the distance to coast and the calculated results are
summarized in Table 4. It is seen that the footprint size is slightly too large at C-
and Ku-band and the distance to coast exceeds slightly the required 15 km at C-band.
Table 4: Radiometer characteristics for the centre beam at C-, X- and Ku-band.
Frequency, [GHz] Cx-power], [%] Footprint, [km] Dist. to coast, [km]

C-band 0.20 23.26 16.41
X-band 0.14 16.53 12.28
Ku-band 0.08 10.86 9.19
Having determined the feed array for the centre beam the complete feed array can
be readily designed. The Ku-band feeds are located close to the focal circle of the
push-broom torus and the feed arrays for C- and X-band are located on either side of
the Ku-band array. The three feed arrays are shown in Fig. 6. The total number of
array elements is 1284, 1956 and 3156 for C-, X- and Ku-band, respectively. If 8 rows
along φ instead of 6 were used the number of elements becomes 1616, 2480 and 4320.
(The latter numbers are used for the power estimates in Sec. 7.) These numbers
clearly show that the number of rows along φ is an important design parameter for
the push-broom system.
211
Figure 7: Mechanical realization of the torus

Figure 6: Feed arrays for C-, X- and Ku- push-broom reector dish.
band.
5.4 Additional performance checks

A number of dierent detailed investigations were carried out. The results can be
summarized as follows:
Scan performance. The antenna is able to scan up to 20◦ to both sides with-
out any severe scan degradations. It was found that the excitations are practically
identical for all the beams but shifted in the φ-direction according to the actual scan
direction.
Bandwidth investigations. The required bandwidth is 300, 100 and 200 MHz
at C-, X- and Ku-band, respectively. The radiometer performance is almost constant
over the bands.
Sensitivity to excitation inaccuracies. The feed element excitations can only
be realized to certain accuracy. Two types of excitation errors were investigated.
It was found that excitation errors up to 10% for each separate element and up to
3% of the largest excitation are acceptable. This is very well in line with another
observation, namely that it is acceptable to discard all elements with an amplitude
lower than 30 dB below the strongest excited element.
Redundancy aspects. It was nally demonstrated that it is possible to re-
optimize the excitation coecients in case of a receiver failure and in this way remedy
the consequences of the failure.
6 Mechanical design of the push-broom torus an-

tenna
The mechanical realization of the torus reector proposed by HPS consists of a double
layer pantograph and two triangular wire-band nets, one in the front and one in the
212
7. Feeding network and receiver issues
back. The corners of the triangles of the two nets are connected by adjustment wires,
as shown in Fig. 7. The front net forms the support of the reector.
Initially it was assumed that the front net would be covered with a knitted metal
mesh in order to provide the necessary RF reection. It was realized, however, that
the triangular facets would generate high and unacceptable grating lobes unless the
triangles were made very small, i.e. 100 mm size. Consequently, it was proposed to
construct the reector as a doubly curved CFRS (Carbon Fibre Reinforced Silicon)
surface. The triangular net is maintained to support the CFRS but the size of the
triangles can be much larger, around 400 mm.
7 Feeding network and receiver issues

In this section the receiver resource demands, especially concerning power consump-
tion, will be evaluated, rst using existing state-of-the-art components, and second
using values that can be realistically expected within a 5 years' time frame.
7.1 Existing state-of-the-art components

As illustrated in Fig. 8, each feed array element is assumed to be connected to a direct
detection circuit, one for each polarization channel and assigned its own receiver and
A/D converter. Hence, the total number of components in a single receiver must
be multiplied onto the number of feed array elements and polarization, and a major
concern is the total number of components in the system, with respect to mass, size
and especially power consumption.
Figure 8: Dense feed array receiver system.
The A to D converter is the most critical component, as it is traditionally the

largest and by far the most power consuming. However, over the past decade tech-
nology has developed rapidly. Most components are wideband, or similar between
the three frequency bands of interest, and thus a common overview has been made
for the dierent component types included in each receiver. The result of this shows
213
that a realistic power budget based on state-of-the-art components will result in a

power consumption of approximately 850 mW per receiver at X-band and 1100 mW
at Ku-band and C-band. The total power budget is dominated by X- and Ku-band
due to the many receivers at these frequencies, and for global power budget estimates
we can assume 1 W per receiver. Adding also power for the beam forming network
and RFI processor we end up with 1.38 W per receiver. With 8416 elements this
gives a total power of 11.6 kW, which is not realistic now or in the near future.
7.2 Realistic components within a 5 years time frame

Already now A/D converters able to sub-sample signals up to X-band are available
in research labs and within very few years Ku-band is also served. The development
concerning ampliers is also impressive, especially when it comes to noise gure at
high frequencies and power consumption. For global power budget estimates we can
within a few years assume 49 mW per receiver, including beam forming network and
RFI processor. This amounts to a total power consumption of 8416 × 49 mW = 412
W, which is certainly realistic.
References
14.

N. Skou, S. S. Søbjærg, A. Ihle, D. Hartmann, and K. v. t Klooster, An optimal
beamforming algorithm for phased-array antennas used in multi-beam spaceborne
radiometers, in Proc. European Conference on Antennas and Propag. (EuCAP),
Lisbon, Portugal, Apr. 2015, pp. 15.
214
Paper I
An Optimal Beamforming Algorithm for

Phased-Array Antennas Used in Multi-Beam
Spaceborne Radiometers
O. A. Iupikov, M. V. Ivashina, K. Pontoppidan, P. H. Nielsen,
C. Cappellin, N. Skou, S. S. Søbjærg, A. Ihle, D. Hartmann,
K. v. 't Klooster

Propagation, EUCAP 2015, Lisbon, Portugal, April 2015.
the thesis.
An Optimal Beamforming Algorithm for
Phased-Array Antennas Used in Multi-Beam
Spaceborne Radiometers
O. A. Iupikov, M. V. Ivashina, K. Pontoppidan, P. H. Nielsen, C. Cappellin,
N. Skou, S. S. Søbjærg, A. Ihle, D. Hartmann, K. v. 't Klooster
Abstract
Strict requirements for future spaceborne ocean missions using multi-

beam radiometers call for new antenna technologies, such as digital beam-
forming phased arrays. In this paper, we present an optimal beamforming
algorithm for phased-array antenna systems designed to operate as focal
plane arrays (FPA) in push-broom radiometers. This algorithm is formu-
lated as an optimization procedure that maximizes the beam eciency,
while minimizing the side-lobe and cross-polarization power in the area
of Earth, subject to a constraint on the beamformer dynamic range. The
proposed algorithm is applied to a FPA feeding a torus reector antenna
(designed under the contract with the European Space Agency) and tested
for multiple beams. The results demonstrate an improved performance in
terms of the optimized beam characteristics, yielding much higher spatial
and radiometric resolution as well as much closer distance to coast, as
compared to the present-day systems.
1 Introduction
Recent advances in phased-array antenna technologies and low-cost active electronic
components open up new possibilities for designing Earth observation instruments.
One example of such technologies is digital beamforming phased-array feeds (PAFs)
(often referred to as dense focal plane arrays [1]). Using PAFs is especially attrac-
tive in spaceborne radiometers in so-called push-broom conguration [2], where a
large number of beams cover a wide region (swath) of the Earth simultaneously to
achieve high sensitivity. For such radiometers, various optics concepts have been
investigated [3], and the optimum solution has been found to be an oset toroidal
single reector antenna. This reector structure is rotationally symmetric around its
vertical axis, and thus is able to cover a wide swath range. However, its aperture
eld exhibits signicant phase errors due to the non-ideal (parabolic) surface of the
217
Paper I. An Optimal Beamforming Algorithm for Phased-Array Antennas...
reector that lead to the beam deformation. Accurate compensation for these ef-
fects requires the use of a moon-shaped PAF (as shown on g.1) as well as dedicated
beamforming algorithms. Development of such an optimal algorithm is the objective
of this paper.
feed arrays
B
B
B
B
B
B
HH
H torus reector rim
Figure 1: A schematic layout of the moon-shaped phased-array feeds for X-, Ku- and C-bands that
are located in the focal eld region of the torus reector. The arrays comprise dual-polarized dipole
antenna elements, denoted by the red and green lines. The black curve denotes the focal arc of the
torus reector.
2 Array design
An initial design of the PAF has been reported in [4]; where the array elements
are arranged on the rectangular grid. For the current study, we have re-arranged
the element positions along the focal-eld arc of the torus reector (see Fig. 1).
This re-arrangement has led to the moon-shaped layout of the present PAF enabling
similar focal-eld distributions that are resulted from dierent incident directions
upon the apertures of the corresponding sub-arrays. Thanks to this advantageous
property, optimization of the beamformer weights for multiple beams reduces to the
optimization of a single set of weights for one beam only, and most importantly, to the
virtually identical beam shapes over the wide observation range. Furthermore, the
new design consists of dual-polarized 0.5λ-dipole antennas, having higher polarization
purity, as compared to the tapered-slot antenna elements used in the array in [4].
To simplify the modeling of the array for this study, we have made the following
assumptions: (i) all array elements have the same radiation patterns; (ii) no mutual
coupling and edge truncation eects are accounted for the array, and; (iii) the dipoles
218
3. Optimal beamforming algorithm
are located above an innite ground plane. In the future studies, these simplications
will be eliminated.
3 Optimal beamforming algorithm

3.1 Generic formulation
The proposed optimization algorithm is formulated as the maximum Signal-To-Noise
beamformer (MaxSNR) [5], where the antenna eciency - dened for a given direction
of observation corresponding to the beam center - is maximized, while minimizing
the power received from the other directions. The weight vector for this beamformer
can be written as
wMaxSNR = C−1 e, with SNR = eH wMaxSNR , (1)
where the vector e = [e1 , . . . , eN ]T holds the signal-wave amplitudes at the receiver
outputs and arises due to an externally applied electromagnetic plane wave Ei ; and
C is a Hermitian spectral noise-wave correlation matrix holding the correlation coef-
cients between the outputs of the array receiving system.
If we assume a noiseless receiver system, the matrix C represents the antenna noise
correlation matrix, which contains the noise correlation coecients due to external
noise sources (that are present in the region of observation on the Earth as well as
outside). The elements of C can be calculated through the pattern-overlap integrals
between fn (Ω) and fm (Ω), which are the nth and mth embedded element pattern
(EEP) of the array (dened after the reection from the dish), respectively [6], i.e.,
Z
Cmn = Text (Ω)[fm (Ω) · fn∗ (Ω)] dΩ, (2)
where Text (Ω) is the brightness temperature distribution of the environment. To meet
the radiometer requirements [2], the function Text (Ω) is chosen such that it has low
temperature values in the region of the expected main lobe (down to −20 dB level)
and high values outside of this region. In this way, we realize the maximization of
the beam eciency dened at the −20 dB level while minimizing the side-lobe
and cross-polarization power outside of this region, as required for the radiometers.
3.2 Computation acceleration

When constructing matrix C, one should realize that its lling can be an extremely
time-consuming procedure as it requires computation of all secondary EEPs over the
entire sphere and evaluation of the pattern overlap integral (2) for all combinations of
EEPs (see Tabl. 1). In order to speed-up the computational process, we have therefore
represented the matrix C as a sum of two contributions, matrices C1 and C2 that
219
Figure 2: The Text mask-constrained functions dened for the calculation of the antenna noise
correlation matrices C1 due to the noise sources in the Earth region (see the inset in the left upper
corner) and C2 due to the noise sources in the sky region (see the inset in the right upper corner).
The toroidal reector fed with a PAF is in the middle of the illustration, where the multiple beams
point to the Earth.
can be calculated relatively fast. The rst matrix is obtained by using the secondary
EEPs computed in a limited angular range around the main lobe region, while the
second matrix is used for correcting for the spillover eects and evaluated through
the primary feed patterns. The brightness temperature distribution functions Text (Ω)
corresponding to C1 and C2 are illustratively shown in the insets of Fig. 2.
The table below cross-compares the computational time at Ku band that is needed
for the simulations (using GRASP) of the secondary patterns over the entire sphere
(when computing the matrix C through the brute-force approach) and over the re-
duced region with the post-correction for the spillover eect (when computing the
matrices C1 and C2 through the proposed approach). There is obvious advantage in
using the latter approach, especially for the systems with a large number of beams
and high operational frequencies.
Table 1: Computational time of the matrix C at Ku-band (18.7 GHz)

Brute-force approach Proposed approach
Computing sec. EEPs Computing C Computing sec. EEPs Computing C1 and C2
∼9 months no data 3 hours 5 min/beam
220
3. Optimal beamforming algorithm
3.3 Iterative procedure for constraints on the dynamic range

of the weights
The proposed beamformer, as described in sub-section III.A, has been extended so
as to include constraints on the dynamic range of the weights that should not exceed
a certain value. For this purpose, we have implemented an iterative procedure that
modies the reference weighting coecients (as determined by the MaxSNR beam-
former), while aiming to maintain the shape of the PAF pattern as close as possible
to the reference one. This ensures that the radiometer parameters are as similar
as possible to those obtained with the reference set of weights. The corresponding
algorithm is listed as follows:
At the rst iteration (q = 1)

the reference weight vector w(1) is calculated
(1)
using (1) with the initial noise-wave correlation matrix C .
At iteration q = 2, 3 . . . a new weight vector w(q) is computed using the noise

(q)
covariance matrix C , diagonal elements of which are a function of the weight
(q−1)
vector w obtained after the previous iteration, i.e.,
(q) (q−1)
Cnn = Cnn f (|wn(q−1) |) (3)
where f is a receiver function that needs to be provided as an input to the

algorithm; it should have such a behaviour that the lower the weight of the array
antenna element, the higher the function value is (which physically corresponds
to an increase in the noise temperature of the corresponding receiver channel).
In the numerical results presented hereafter, a lter function is used whose
values are close to zero when the weights magnitude|wi | are higher than wconstr ,
and which has a sharp linear increase near wconstr .
f is similar to In this way,
the inverse step function near wconstr (Fig. 3). Here, wconstr is the value of
the amplitude weight constraint, which is typically in the order of −30 dB or
−40 dB.
2
1.5
f ( |w| )
1
wconstr
0.5
−60 −50 −40 −30 −20 −10 0
|w|, dB
Figure 3: The function f used in the iterative procedure described in section III.C.
Check whether all the weights are higher than wconstr , or negligibly low (i.e.
−80 dB in this work). If this condition is satised, the iterative procedure is
terminated. The channels with negligible weights are switched-o, while the
resulting set of weight coecients is considered to be the nal one.
221
In order to use the beamformer for scanned beams, the noise temperature distri-
bution function Text (Ω) must be provided for each of them and the matrix C needs
to be recomputed.
More detailed on the formulation and implementation of the beamformer can be
found in [7].
4 Parametric studies
4.1 Beamformer
The proposed beamformer has two parameters for dening the cold ellipse of the
mask-constrained function Text (Ω) that are used for the computation of C1 : the ellipse
major semi-axis a and the axis ratio a/b (see Fig. 2, top-left inset). Since the area
of the ellipse is related to the area of the main lobe over which the received power
is maximized, and the size of the foot print is known from specications, the range
of practical values for a and the axis ratio a/b is relatively small, and hence the
parametric study to nd the optimal values is not time-consuming.
The considered radiometer characteristics [2] as functions of these parameters have
been computed and the most critical ones, i.e. the distance-to-land and footprint size,
are shown in Fig. 4. As a trade-o between the required values of the distance-to-
land (< 15 km) and the footprint size (< 10 km), the following best values have been
chosen: a = 0.535 and a/b = 1.3.
25 11
0.575 0.575
Major semi−axis, [deg]
Major semi−axis, [deg]
10
0.565 0.565
10
15
10.5
20
0.555 20 0.555
0.545 0.545 10
0.535 0.535
0.525 15 0.525 9.5
0.515 0.515
0.505 0.505 9
0.495 10 0.495
0.485 0.485 8.5
0.475 0.475
5 8
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1
Axis ratio of the mask ellipse, [−] Axis ratio of the mask ellipse, [−]
(a) (b)
Figure 4: (a) Distance-to-land, [km]; and, (b) footprint size, [km], as functions of a and a/b used
for the denition of the mask-constrained function Text (Ω) as shown on Fig. 2.
4.2 PAF size and nal radiometer characteristics

For the beamformer with the parameters as obtained above, we have studied a range
of PAF designs which have a number of rows varying from 5 to 8. The computed
222
4. Parametric studies
radiometer characteristics for the range of rows are shown in Fig. 5. As one can see,
to satisfy all radiometer requirements, the minimum number of rows in the PAF must
be equal to 6.
99 30

25 Distance to land
98 Requirement
20
15
97
10
96 5
5 6 7 8 5 6 7 8
Number of rows Number of rows
(a) (b)
11
Rel. power in cross−pol., [%]

0.2
Footprint, [km]
10 0.15 XP relative power

Requirement
FPL (Phi=90)
9 Average 0.1
FPS (Phi=0)
Requirement 0.05
8 5 6 7 8
5 6 7 8 Number of rows
Number of rows
(c) (d)
Figure 5: Radiometer characteristics, including the beam eciency, distance to land, footprint and
relative cross-polarization power, vs. the number of rows in the PAF.
Table 2: Final radiometer characteristics at Ku-band (18.7 GHz)

Radiometer charac- Require- Gaussian PAF 6 × 19 × 2 elem.
teristic ment feed del = 0.75λ
Beam eciency [%] 73.9 97.9
XP-power, [%] 0.34 0.38 0.06
Dist. to land, [km] 15 90.8 14.6
Beam width, [deg] 0.38 0.357
Footprint (FP), [km] 10 10.5 9.8
FP ellipticity 1.13 1.12
The optimized set of weight coecients are shown in Fig. 6. The corresponding
pattern of the phased-array feed and the pattern of the entire reector antenna system
for the on-axis beam are shown in Fig. 7. We can observe the very ne shape of the
illumination pattern across the reector aperture, and well-behaved nal beam with
the minimized side-lobe levels. The levels of the side lobes are dierent, though, over
the angular region; that results in the angular dependence of the distance-to-land
parameter, which becomes a function of the coast line position. Since the footprint
on the Earth resulted from this beam, is not symmetric either, we have investigated
whether the distance-to-land requirement is satised for all possible locations on
223
−40 −30 −20 −10 0
Figure 6: The array element amplitude weight coecients, [dB], as obtained with the proposed
beamforming algorithm. Each block represents an element of the array.
(a) (b)
Figure 7: (a) The optimized pattern of the PAF when illuminating the aperture of the torus reector,
[dB], and (b) the corresponding nal beam of the entire reector antenna system for the case of the
center beam, [dBi].
30
5 rows
25 6 rows
7 rows
20 8 rows
Requirement
15
10
5
0 50 100 150 200 250 300 350
Coast angle, [deg]
Figure 8: Distance-to-land as a function of angle at which the coast line is approached by the beam
for dierent array sizes.
the land line with respect to the beam footprint. As the data on Fig. 8 show, the
PAF with 6 rows satises this criterion for all possible positions.
The corresponding radiometer characteristics for the on-axis beam are summa-
rized in Table 2. Thanks to the rotational symmetry of the reector and the moon-
shaped array layout, the scanned beams will have similar characteristics.
224
5. Conclusion
5 Conclusion
An optimal beamforming algorithm for phased-array antennas, such as considered
for the next generation multi-beam radiometers, has been presented and evaluated
for a currently designed prototype system. It yields well behaved multiple beams
which satisfy strick requirements to the footprint on the Earth, minimized power in
the side-lobes and cross-polarization as well as the distance-to-coast. The proposed
algorithm is formulated in a closed form and enables dierent performance trade-os.
References
[1] M. Ivashina and J. Bregman, A way to improve the eld of view of the radio
telescope with a dense focal plane array, in Proc. of the Int. conf. on Microwave
and Telecommunication Technology, Sevastopol, Ukraine, Sep. 2002.

14.

Kingdom, Apr. 1989, pp. 126130.

on Antennas and Propag. (EuCAP), Hague, The Netherlands, Apr. 2014, pp. 15.
[5] H. L. van Trees,Optimum Array Processing Part IV of Detection, Estimation,


Jun. 2011.
[7] O. Iupikov, Phased-array-fed reector antenna systems for radio astronomy and
Earth observations, Licentiate Thesis, Institutionen för signaler och system,
Antenner, Chalmers tekniska högskola, Göteborg, Oct. 2014. [Online]. Available:
http://publications.lib.chalmers.se/records/fulltext/206295/206295.pdf
225
226
Paper J
Enabling High-sensitivity Near-land Radiometric

Measurements With Multi-beam Conical Scanners
Employing Phased Arrays
M. V. Ivashina, O. A. Iupikov, C. Cappellin, K. Pontoppidan,
P. H. Nielsen, N. Skou, S. S. Søbjærg, B. Fiorelli
in Proceedings of 36th ESA Antenna Workshop on Antennas and RF

Systems for Space Science, ESA/ESTEC, The Netherlands, 6-9 October
2015
the thesis.
Enabling High-sensitivity Near-land Radiometric
Measurements With Multi-beam Conical Scanners
Employing Phased Arrays
M. V. Ivashina, O. A. Iupikov, C. Cappellin, K. Pontoppidan, P. H. Nielsen,
N. Skou, S. S. Søbjærg, B. Fiorelli
Abstract
In the recent study, carried out under the ESA contract 4000107369/12/
NL/MH Study on Advanced Multiple-Beam Radiometers, we investi-
gated the use of dense phased arrays feeds (dPAFs) for conical scan
and push-broom radiometer congurations. It has been found that such
systems can satisfy all the challenging requirements for the future Earth
observation missions, but need large arrays with many antenna elements.
To determine the minimum number of elements and their excitations, we
have developed a dedicated optimization procedure and applied it to the
dPAFs for the push-broom case. This procedure is based on the beam-
forming approach that jointly optimizes for the maximum beam eciency
(or the maximum beam sensitivity) and minimum distance-to-land. The
goal of this work is to repeat the same procedure for the conical-scan ra-
diometer, for which we initially used a simplied Conjugate Field Match-
ing based beamforming approach.
1 Introduction
Existing spaceborne microwave radiometers typically use conical-scan (CS) reector
antennas with horn feeds. Such systems, operating at C-band (6.9 GHz) or at higher
frequency, provide a spatial resolution of around 50 km, whereas less than 20 km
is desirable [1, 2]. Furthermore, accurate radiometric measurements are currently
possible at not closer than around 100 km from the shore-line, because of the signal
contamination by the antenna side-lobes illuminating the land (see Table 1). There
is a strong desire to reduce it to 5-15 km.
The required 20 km resolution, i.e. 3-dB footprint, at C-band leads to a large
antenna aperture of around 5 m in diameter that is considerably larger than any
radiometer system antenna own hitherto. Moreover, the required short distance-to-
land (i.e. 5-15 km) can only be achieved by replacing the conventional feed technology
with novel dense phased arrays feeds (dPAFs), which are capable of producing many
simultaneously-formed and closely-spaced beams. A feasibility study of dPFAs was
229
Paper J. Enabling High-sensitivity Near-land Radiometric Measurements...
Table 1: Characteristics of the existing radiometers (ANSR-E and WindSat). The desired values
for future observation missions are shown in parentheses.
Band Footprint, [km] Sensitivity, [K] Dist.to coast, [km]

C 55 (20) 0.3 (0.30) 100 (5-15)
X 35 (20) 0.5 (0.22) 100 (5-15)
Ku 20 (10) 0.5 (0.22) 100 (5-15)
conducted by the authors (in collaboration with HPS, Germany), where dPAFs were
investigated in both CS and more advanced push-broom congurations. This study
has demonstrated that with dPFAs we could satisfy all the challenging radiometric
requirements, but at the expenses of a large number of array elements [3, 4]. To deter-
mine the optimum number of elements and their excitations, we developed a dedicated
procedure maximizing the beam eciency or the beam sensitivity (as formulated by
TICRA and CHALMERS, respectively), while minimizing the distance-to-land [5, 6].
First, this procedure was applied to reduce the number of elements in the dPAFs
for the push-broom antenna, while for the conical scanner which needs relatively
smaller array feeds we used a simplied Conjugate Field Matching (CFM) based
approach. In this paper, we apply this dedicated procedure to the conical-scan case
and show its advantages.
2 Limitations of horn feeds

Horn antenna feeds used in radiometer systems are typically designed to produce an
illumination pattern with strong taper toward the edge of the reector to trade high
beam eciency, and low cross-polarization power against low side lobes, to achieve
a short distance to land. Figures 1(b-h) show typical antenna patterns as well as
radiometer parameters at C-band, which were obtained for the conical scanner (see
the description of antenna geometry in [3]). The antenna is a paraboloid with circular
projected aperture, and it has the projected aperture diameter of 5 m, clearance of
1 m and focal length of 3 m. The antenna was analyzed with the Gaussian beam
model of the feed having a varying illumination taper and aperture diameter of the
corresponding horn antenna [7, 8]. The positions and aperture diameters of the horns
in the focal plane are depicted on Figure 1(a) by circles of dierent sizes corresponding
to three taper values at C, X and Ku bands. As one can see, the horn feeds have
large aperture diameter (> one wavelength), and thereby large oset positions with
respect to the focal point, hence degradation in performance of such o-axis beams
can be expected, especially at the C-band.
These simulations have been used as the starting point of our research on dPAFs
in order to quantitatively illustrate fundamental limitations of conventional feed tech-
nologies. Indeed, as the results in Figures 1(e-h) show, the cross-polarization power
230
2. Limitations of horn feeds
Focal point
To form taper -50 dB @ θ sub=35°
(a) (b)
(c) (d)
50 4
Requirement Requirement
Relative power
40
30 2
20
10 0
−60 −50 −40 −30 −20 −10 −60 −50 −40 −30 −20 −10
Gaussian beam taper @ 35 deg, [dB] Gaussian beam taper @ 35 deg, [dB]
(e) (f )
40 FPL (along track) 50

Footprint, [km]
Directivity, [dB]
FPS (across track)

30 Average
48
Requirement
20 46
−60 −50 −40 −30 −20 −10 −60 −50 −40 −30 −20 −10
Gaussian beam taper @ 35 deg, [dB] Gaussian beam taper @ 35 deg, [dB]
(g) (h)
Figure 1: (b) Simulated locations and aperture diameters of the horn feeds at C, X and Ku frequency
bands for three dierent values of the illumination taper; (b) Illumination pattern of the Gaussian
feed with the optimal relative taper toward the edge of reector of -25dB and (c-d) Co- and cross-
polarization patterns of the reector antenna. (e-h) Radiometer characteristics, i.e. the distance-to-
land, relative cross-polarization power, footprint size and directivity as function of the illumination
taper of the Gaussian feed.
231
for the conical-scan radiometer can only be minimized by strongly tapering the feed
pattern, but this leads to the increase of the footprint size and distance-to-land, and
hence the diculties to satisfy the sensitivity requirements (more horn/receivers may
be needed). The shortest distance-to-land that can be achieved with this tapering
approach is ∼ 20 km for the taper value of −25dB, for which the realized cross-
polarization power is at least 3 times higher than the desired 0.34%. The radiometer
characteristics for the optimal taper value are also summarized in Table 3.
3 Arrays feeds: CFM beamforming

During the project `Study on Advanced Multiple-Beam Radiometers we found that
in order to meet the sensitivity requirement for a swath of 1500 km, the conical-
scan antenna needs 2 beams at C-band, 21 beams at X-band and 30 beams at Ku-
band [3, 9]. The layouts of the considered dPAFs along with their relative positions in
the focal plane of the reector are illustrated on Figure J.1(a) and their dimensions
is summarized in Table 2.
Table 2: Number of antenna elements in the arrays
Number of elements
Band Array size
X-orient. Y-orient. Total
C 6.0 × 5.3λ 64 63 127
X 6.0 × 11.3λ 128 135 263
Ku 8.3 × 10.5λ 165 168 333
In the course of this study, some assumptions and simplications were made in
order to limit the complexity of the phased array feed. One of these assumption
was that the array has identical embedded element patterns, which were modeled
for the case for a half wavelength dipole antenna array with 0.75 wavelength inter-
element separation distance, located above an innite ground plane. Furthermore,
for the conical scanner, a simple beamforming method was applied to determine the
optimal excitation coecients of the feed array elements. This method is based on
the Conjugate Field Matching (CFM) approach that has been conventionally used
as a feed synthesis technique for horn feeds [10]. The core of this approach is to
analyze the focal region eld of a reector antenna (in the absence of the feed) for an
incident plane wave (PW), and then calculate the desired size of the feed aperture,
which should conjugately match the truncated reference focal-eld distribution. One
of the limitations of this CFM method is that the optimal taper of the incident plane
wave (that is typically used to control side-lobe levels) is a priory not known, and
if not determined correctly can yield an over-estimated feed size or unsatisfactory
beam performance. This limitation can be critical, especially when applying CFM to
232
3. Arrays feeds: CFM beamforming
50
0.4 beam 1
beam 1
Relative power
40 beam 2
beam 2
30 0.2 Requirement
Requirement
20
10 0
−50 −45 −40 −35 −30 −25 −20 −50 −45 −40 −35 −30 −25 −20
Incident PW taper @ reflector edge, [dB] Incident PW taper @ reflector edge, [dB]
(a) (b)
40 FPL (along track) 50

Footprint, [km]
beam 1
Directivity, [dB]
FPS, (across track) beam 2
30 Average 48
20 46
−50 −45 −40 −35 −30 −25 −20 −50 −45 −40 −35 −30 −25 −20
PW taper @ reflector edge, [dB] Incident PW taper @ reflector edge, [dB]
(c) (d)
0
−44.0 −34.9 −26.7 −26.5 −29.8 −54.6 −44.9 −47.4
−5
−35.9 −20.3 −12.9 −11.5 −16.0 −27.1 −60.7 −50.9
−10
−23.7 −11.6 −4.5 −3.1 −7.1 −17.0 −30.4 −51.5
−15
−16.6 −7.1 −1.3 0.0 −3.5 −11.3 −23.4 −38.9
−20
−13.6 −6.2 −2.4 −1.8 −4.2 −10.0 −19.2 −31.6
−25
−13.3 −8.6 −6.8 −7.2 −8.6 −12.2 −18.9 −28.7
−30
−16.0 −14.1 −14.4 −14.7 −16.5 −17.8 −21.3 −28.4
−35
−21.7 −23.4 −23.4 −25.4 −25.4 −27.8 −27.4 −29.6
−40
(e) (f )
(g) (h)
Figure 2: CFM beamforming approach: (a-d) Radiometer characteristics, i.e. the distance-to-land,
relative cross-polarization power, footprint size and directivity as function of the incident plane wave
taper; (e) Excitation coecients of the PAF at C-band, dB, as obtained with the CFM approach
for the plane wave taper of -30 dB as an example, (f ) corresponding illumination pattern, (g-h) co-
and cross-polarization patterns of the reector antenna.
233
Table 3: Radiometer characteristics for dierent feeds

Array feed I, Array feed II, Array feed I,
Radiometer Require- Horn
CFM beamformer CFM beamformer MSMDL beamf.
characteristic ment feed
Beam 1 Beam 2 Beam 1 Beam 2 Beam 1 Beam 2
Number of antenna ele-
64 + 63 = 127 80 + 77 =157 64 + 63 = 127
ments (2 polarizations)
Distance to land, [km] < 15 19 20 18 16 16 14 15
Rel. cross-pol. power, [%] < 0.34 1.04 0.21 0.19 0.19 0.19 0.29 0.23
Beam eciency, [%] 97 95 95 95 95 96 95
Footprint (average), [km] < 20 21 21 21 21 21 20 20
Footprint ellipticity 1.6 1.7 1.7 1.7 1.7 1.4 1.4
design PAFs, which can have a large number of active antenna elements and exhibit
strong mutual coupling eects.
Number of elements X+Y: 80 + 77 = 157

Number of elements X+Y: 64 + 63 = 127
-0.2
-0.25
Y, [m]
-0.3
-0.35
-0.4
-0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2
X, [m]
Figure 3: Original (black) and enlarged (green) dual-polarized phased array feeds for C-band.
Figure 2(a-d) presents the results of a parametric study aimed to determine the
optimal value of the plane wave taper that could meet the performance specications
at C-band (see Array I in Table 3). As seen, with the given array size and utilized
CFM optimization approach, the performance upper bound in terms of the distance-
to-land is 18 and 20 km that is similar to that with the Gaussian feed, while the
relative cross-polarization power is signicantly better (0.20% vs. 1.04%). Figure 2(e-
h) shows the element excitation coecients and antenna patterns for the optimum
PW taper value of -30dB.
As one can notice from the Figure 2(e), the excitation coecients at the left
edge of the array have relatively high amplitude values (> −14 dB). Therefore, an
improvement in the radiometer performance can be expected by adding additional
column of antenna elements. The enlarged in such way array is shown on Figure 3,
where newly added elements are denoted by green color. The radiometer character-
istics for this array (Array II) are also summarized in Table 3. A better performance
is observed now, where most of the requirements are satised, though with the larger
array of 30 more antenna elements.
234
4. Arrays feeds: Max. Sensitivity - Min. Distance-to-Land beamforming
0
−44.7 −32.4 −25.7 −22.9 −24.3 −28.8 −39.1 −44.2
−5
−36.6 −21.8 −12.5 −9.5 −12.0 −20.2 −39.7 −41.5
−10
−36.7 −16.9 −5.8 −2.1 −5.4 −16.5 −33.4 −37.4
−15
−27.9 −13.9 −3.9 −0.0 −3.2 −12.5 −26.8 −46.2
−20
−21.0 −11.4 −4.8 −2.5 −4.0 −10.0 −21.3 −42.4
−25
−19.2 −11.5 −8.4 −7.6 −7.4 −9.9 −17.1 −29.6
−30
−20.1 −14.7 −14.0 −13.6 −13.9 −13.6 −17.3 −25.5
−35
−23.1 −19.7 −19.2 −20.1 −21.0 −20.5 −20.6 −26.0
−40
(a) (b)
(c) (d)
Figure 4: MSMDL beamforming approach: (a) Excitation coecients of the PAF at C-band, (b)
corresponding illumination pattern, (c-d) co- and cross-polarization patterns of the reector antenna.
All values are in dB.
4 Arrays feeds: Max. Sensitivity - Min. Distance-

to-Land beamforming
In contrast to the traditional CFM beamforming, new approaches based on optimum
array signal processing methods are more accurate, and capable of handling a large
number of array elements, large number of beams and complex performance speci-
cations. An example of such methods is a customized beamforming method that
jointly optimizes for maximum sensitivity and minimum distance-to-land, which was
proposed and applied to the push-broom radiometer concept [5, 6]. In this paper
we apply this customized beamforming to the conical scanner and present obtained
results.
The proposed optimization method is formulated as the maximum Signal-To-

Noise beamformer (MaxSNR) [11], where the antenna sensitivity (or eciency in the
noise-less case) - dened for a given direction of observation corresponding to the
beam center - is maximized, while minimizing the power received from the directions
of side-lobes. The weight vector for this beamformer can be written as wMaxSNR =
235
C−1 e with with SNR = eH wMaxSNR , where where the vector e = [e1 , . . . , eN ]T holds
the signal-wave amplitudes at the receiver outputs and arises due to an externally
applied electromagnetic plane wave Ei ; and C is a Hermitian spectral noise-wave
correlation matrix holding the correlation coecients between the outputs of the
array receiving system.
If we assume a noiseless receiver system, the matrix C represents the antenna noise
correlation matrix. The elements of C can be calculated through the pattern-overlap
integrals between fn (Ω) and fm (Ω), which are the nth and mth embedded element
pattern of the array (dened after the reection from the dish), respectively [12], i.e.,
Cmn = Text (Ω)[fm (Ω) · fn∗ (Ω)] dΩ, where Text (Ω) is the brightness temperature dis-
R
tribution of the environment. To meet the radiometer requirements [3], the function
Text (Ω) is chosen such that it has low temperature values in the region of the expected
main lobe and high values outside of this region. In this way, we realize the maxi-
mization of the beam eciency, while minimizing the side-lobe and cross-polarization
power outside of this region.
The resulting excitation coecients and antenna patterns for the smaller array
(Array I) are shown on Figure 4 and the radiometer characteristics are summarized in
Table 3. These results clearly demonstrate the advantage of the MSMDL beamform-
ing approach to determine the minimum number of antenna elements, as compared
to the CFM approach. Further minor reduction in the array size could be possible
for the MSMDL beamformer, but this has to be still studied.
5 Conclusions
An advantage of novel phased array feeds with respect to traditional single-horns is

their capability of satisfying complex performance specications by optimally beam-
forming the signals received by a large number of array elements. As shown in
this paper, an optimal beamforming approach, such as the Maximum Sensitivity -
Minimum Distance-to-Land approach can lead to a major improvement in the per-
formance of the present and ying technologies. For the conical-scan antenna, the
distance to land of 15 km can be achieved with an array of 127 half-wavelength dipole
antenna elements. For comparison, a beamforming method based on the conventional
Conjugate Field Matching approach would provide similar performance with a larger
array, in our case with 30 more elements. In future work we plan to re-evaluate the
resultant radiometer characteristics with a more accurate array model accounting for
the mutual coupling and edge truncation eects, as well as try to further minimize
the number of antenna elements.
236
References
Acknowledgments
The authors acknowledge the Swedish Research Council for providing partial support
to this work as well as Kees van 't Klooster for many fruitful discussions.
References
[1] P. W. Gaiser, The WindSat spaceborne polarimetric microwave radiometer:
Sensor description and early orbit performance, IEEE Trans. Geo. Rem. Sens-
ing, vol. 42, no. 11, pp. 591603, Nov. 2004.
[2] Amsr-e instrument description. [Online]. Available: http://nsidc.org/data/

docs/daac/amsre\_instrument.gd.html

14.

15.
[5] C. Cappellin, K. Pontoppidan, P. H. Nielsen, N. Skou, S. S. Søbjærg, A. Ihle,

M. V. Ivashina, O. A. Iupikov, and K. v. 't Klooster, Design of a push-broom
multi-beam radiometer for future ocean observations, in Proc. European Con-
ference on Antennas and Propag. (EuCAP), Lisbon, Portugal, Apr. 2015, pp.
15.

beamforming algorithm for phased-array antennas used in multi-beam space-
borne radiometers, in Proc. European Conference on Antennas and Propag.
(EuCAP), Lisbon, Portugal, Apr. 2015, pp. 15.
Foundations of Antenna Engineering A Unied Approach for

[7] P.-S. Kildal,
Line-Of-Sight and Multipath. Kildal Antenn AB, 2015.
[8] T. Teshirogi and T. Yoneyama, Modern Millimeter-wave Technologies. IOS

Press, 2001.
237
[9] N. Skou, S. S. Søbjærg, S. S. Kristensen, C. Cappellin, K. Pontoppidan,

M. Ivashina, A. Ihle, and K. v.'t Klooster, Future spaceborne ocean mis-
sions using high sensitivity multiple-beam radiometers, in IEEE International
Geoscience and Remote Sensing Symposium (IGARSS), Québec City, Québec,
Canada, Jul. 2014.

[11] H. L. van Trees,Optimum Array Processing Part IV of Detection, Estimation,


18641875, Jun. 2011.
238
Paper K
Multi-Beam Focal Plane Arrays with Digital

Beamforming for High Precision Space-Borne
Remote Sensing
O. A. Iupikov, M. V. Ivashina, N. Skou, C. Cappellin, K. Pontoppidan,
K. v. 't Klooster
Under review for IEEE Transactions on Antennas and Propagation,

2017
the thesis.
Multi-Beam Focal Plane Arrays with Digital
Beamforming for High Precision Space-Borne
Remote Sensing
O. A. Iupikov, M. V. Ivashina, N. Skou, C. Cappellin, K. Pontoppidan,
K. v. 't Klooster
Abstract
The present-day ocean remote sensing instruments that operate at low

microwave frequencies are limited in spatial resolution and do not al-
low for monitoring of the coastal waters. This is due the diculties of
employing a large reector antenna on a satellite platform, and generat-
ing high-quality pencil beams at multiple frequencies. Recent advances
in digital beamforming focal-plane-arrays (FPAs) have been exploited in
the current work to overcome the above problems. A holistic design pro-
cedure for such novel multi-beam radiometers has been developed, where
(i) the antenna system specications are derived directly from the re-
quirements to oceanographic surveys for future satellite missions; and (ii)
the numbers of FPA elements/receivers are determined through a dedi-
cated optimum beamforming procedure minimizing the distance to coast.
This approach has been applied to synthesize FPAs for two alternative
radiometer systems: a conical scanner with an o-set parabolic reector,
and stationary wide-scan torus reector system; each operating at C, X
and Ku bands. Numerical results predict excellent beam performance for
both systems with as low as 0.14 % total received power over the land.
1 Introduction
Microwave radiometry is a highly versatile method of remote sensing, capable of de-
livering measurements of a variety of geophysical properties of the ocean and atmo-
sphere, even through clouds. The retrieval methods distinguish the individual eects
of dierent geophysical properties by using the frequency and polarization state of
the microwave radiation detected by the antenna. Despite such versatility, the ex-
ploitation of microwave radiometry in Earth observation has been constrained by the
diculties of generating antenna beams with low side-lobes and cross-polarization,
and accomodating several feeds operating at dierent frequencies, when deploying the
antenna on a satellite platform [1]. In particular, for high resolutions demanded by
oceanographers, the current antenna designs would need to be scaled up to a physical
241
Paper K. Multi-Beam Focal Plane Arrays with Digital Beamforming for...
size that is too large to be achievable or aordable within typical Earth observation
infrastructure budgets. For this reason, space agencies have been seeking solutions
to overcome what seems at present to be an unpassable barrier to further signicant
improvement of a whole class of remote sensing methods.
The European Space Agency (ESA) is currently considering the ocean missions
where extreme weather, climate variability, coastal and marginal-ice-zone studies
are strong drivers [2, 3]. These studies require a very high radiometric resolution, i.e.
around 0.25 Kelvin, and at the same time a high spatial resolution approaching 20 km
at C and X bands and 10 km at Ku band (see Table 1) [4]. This desired performance
represents a signicant improvement compared with existing space-borne radiometer
systems, such as AMSR-E and WindSat [5,6]. They feature spatial resolutions around
55 km, 35 km, and 20 km at the C, X, and Ku bands, and the radiometric resolution
provided by AMSR-E is 0.3 K at C band and 0.6 K at X and Ku band, while for
WindSat it is around 0.7 K. Moreover, future systems are required to provide valid
observations up to very short distances from the coastline, i.e. 5-15 km, while the
existing systems can observe only up to ∼ 100 km.
It can be shown that the desired spatial resolution calls for a reector antenna with
∼ 5 m aperture diameter [7]; that is considerably larger than any radiometer antenna
own hitherto. On the other hand, for all three frequency bands the bandwidths
are limited to a few hundreds of MHz, that makes it possible (at least in theory)
to achieve very low noise temperatures of the receivers. However, even the most
optimistic receiver noise properties cannot ensure the required radiometric resolution
when considering a single beam scanning system (see Fig. 1(a)). For a scanner, the
only solution is to employ several independent beams per frequency, and improve
radiometric resolution by integrating several footprints. This calls for a large number
of overlapping beams in the present case up to 30 beams at Ku-band. An alternative
is a push-broom system [8, 9], where many beams cover the swath simultaneously,
as illustrated in Fig. 1(b). Using traditional feeds, each antenna beam is associated
with its own receiver, and high radiometric resolution is achieved thanks to the fact
that the signals associated with multiple across-track footprints do not have to be
multiplexed through a single receiver. Radiometric resolution is no longer a problem,
but a more complicated antenna design (a tilted parabolic torus reector) is needed
as well as many beams for the present case up to 156 at Ku-band. Realizing these,
while correcting for the antenna eld distortions causing the well-known triangular
footprints and their large separation on the Earth [8, 9], represents a great challenge.
Also the implementation of this concept should be feasible regarding the resource
requirements, i.e. the size, mass and power consumption.
As demonstrated in this study (see Sec. 4), the above radiometric requirements
cannot be fullled by using traditional cluster feeds of horns (in one-horn-per-beam
conguration), employed at such multi-frequency radiometer antennas. Recent stud-
ies supported by ESA [1013] have identied a promising solution that originates
242
2. From oceonagraphic requirements to antenna system specifications
from the eld of radio astronomy [1420], where instrument designs have evolved to
meet the high-sensitivity and large-coverage requirements of ground based observa-
tories exploring the universe without the above challenges. This solution is based on
`dense' focal plane arrays (FPAs), where many small antenna elements take part in
the formation of each beam (so that each beam can be optimized for high performace,
even far o-axis beams) and the same element takes part in the formation of multiple
beams (so that the footprints overlap), thanks to digital beamforming. Although
the basic principles of these systems are rather similar to those in radio astronomy,
there are many dierences, which are related to application specic requirements.
These requirements will be discussed in Sec. 2, and transtated into antenna system
specications and beam characteristics to optimize for. The reector antenna ge-
ometries used in this study are briey described in Sec. 3. Section 5 will cover the
synthesis of FPAs for such systems, and include the following original contributions:
(i) a dedicated optimum-beamforming algorithm minimizing the distance to coast;
(ii) optimized antenna patterns and radiometric parameters as obtained for the
half-wavelength dipole element FPAs that fulll all above requirements with al-
most twice less elements in comparison to the conventional conjugate-eld-matching
optimization approach [10]; and (iii) validation of the simplied array model with the
assumed identical embeded element patterns [10, 12] across the full MoM model for
the purpose of the FPA synthesis. Finally, digital receiver resource requirements will
be considered in Sec. 6.
2 From oceonagraphic requirements to antenna sys-

tem specications
The requirements for future missions in Table 1 are dened in terms of performance
metrics for oceanographic surveys, i.e. spatial resolution, radiometric resolution,
bias and distance to coast. Since these terms are not commonly known by antenna
designers, next we will summarize their denitions and use these to derive the antenna
system specications.
Table 1: Radiometric requirements for future ocean missions
Band Radiometric Spatial

Freq., Polari- Bias, Dist.to coast,
width, resolution, resolution,
[GHz] zation [K] [km]
[MHz] [K] [km]
6.9 300 V, H 0.30 0.25 20 5-15
V, H,
10.65 100 0.22 0.25 20 5-15
S3 , S4
V, H,
18.7 200 0.25 0.25 10 5-15
S3 , S4
243
Figure 1: Operational principle of (left) the conical scan and (right) push-broom microwave ra-
diometers for ocean remote sensing.
2.1 Spatial resolution (FP) ⇒ reector diameter

The required spatial resolution in Table 1 is dened in terms of the footprint (FP)
on the Earth's surface:
FP = (Y × θ3dBT + Y × θ3dBL / cos ν)/2, (1)
where θ3dBL and θ3dBT are the half-power beamwidths of the antenna main beam
along the elevation and azimuth directions, respectively, ν is the incidence angle as
measured from the normal to the Earth's surface and Y is the distance from the
satellite to the observation point on the Earth.
The FP is directly related to the antenna beamwidth, and hence determines its
◦
aperture diameter. This diameter should be at least 5 m for the present case (ν = 53
and Y = 1243 km) in order to realize the FP of 20 km at C-band. Since for the
considered system, the same antenna is used at dierent bands, the same FP cannot
be obtained at both C- and X-bands. The required FP shall therefore be considered
a guideline and values both slightly above and below are acceptable. The important
factor is that the beam crossover points should be at the -3 dB level. This means
that if the FP is reduced, more beams are needed to cover a particular region on the
Earth.
2.2 Bias (∆T) ⇒ acceptable cross-polarization power

Bias is a systematic error of the measured brightness temperature of the sea. For
full polarization radiometers, ∆T
is typically driven by polarization leackage. The
◦
approximate values of the see temperature for the incidence angle 53 are Tv = 150 K
and Th = 75 K in vertical and horizontal polarizations, respectively. To measure
244
2. From oceonagraphic requirements to antenna system specifications
Th , one can select the co-polar component as the horizontal polarization. The cross-
polarization component of the pattern, however, will pick up the vertical component
of the radiation from the sea, which has a temperature of 150 K. Using the assumption
that the amount of radiation received from the sky is negligible, it is sucient to
consider the antenna pattern in the angular region covering the Earth only, and
hence compute the total temperature as Tb = Tv Pcross + Th Pco , where Pco and Pcross
are the co- and cross-polarization received powers in the angular region of the Earth.
Then, ∆T can be found as
∆T = Tb − Th = (Tv − Th )Pcross , (2)
where Pcross is the acceptable relative cross-polarization power of the antenna pattern
that coverers the Earth. Using (2), one can show that the requirement for the ∆T =
0.25 K can be satised only if Pcross does not exceed 0.34 %.
2.3 Distance to coast (Dc ) ⇒ acceptable side lobes and cross-

polarization power
Table 1 states that Dc should be 5-15 km, when measured from the FP. The reason
behind this requirement is that the brightness temperature of the land is much higher
than that of the sea. This means that the power in the antenna pattern over land
must be suciently small. In order to assess the inuence from the land the cross
polarization can be neglected. The brightness temperature of the land surface is
about Tland = 250 K. Assuming the measurements at horizontal polarization, the
sea temperature is around Th = 75 K. If there is no land below the satellite, the
radiometer will receive an amount of power proportional to Th Pco . If the satellite
covers both the land and sea regions, the power from the sea is Th (Pco − Pland ), where
Pland is the relative co-polarization power in the land region. The signal from the
land is Tland Pland . The measured temperature and ∆T are therefore
Pco − Pland Pland

Tb = Th + Tland , (3)
Pco Pco
Pland
∆T = Tb − Th = (Tland − Th ) . (4)
Pco
We will now determine Dc by the help of Fig. 2, where we have assumed a straight
coastline and a circular symmetric beam with the beamwidth of θ3dB . The beam is
located over the sea and the distance from the peak to the coast is indicated by the
angle θc , while the power in the cone with semi-angle θc is denoted by Pc . The power
outside this cone is Pco − Pc and approximately half of this power will fall on the
land, so we have Pland = (Pco − Pc )/2. Substituting this into (4) gives
Pc 2∆T
=1− . (5)
Pco Tland − Th
245
Inserting the required ∆T ≤ 0.25K in (5) gives
Pc 2 × 0.25
≥1− = 0.9972. (6)
Pco Tland − Th
This equation shows that the required accuracy is obtained when the coastline is
located outside a cone around the main beam containing 99.72% of the total power
on the Earth. Hence, in order to reduceDc , one should minimize this cone. Then Dc
can be dened as the angular dierence θc − θ3dB projected on the Earth surface, i.e.,
Dc = Y sin θc − Y sin θ3dB ≈ (θc − θ3dB )Y. (7)
It should be noted that for non-symmetric patterns, the same procedure can be used,
where the resulting distance to coast should be an average value for all antenna
pattern cuts.
2.4 Radiometric resolution (∆Tmin ) ⇒ number of beams

Radiometric resolution is the smallest change in input brightness temperature that
can be detected. For a full-polarization radiometer it can be found as
Tsys Trec + Tb
∆Tmin = √ = √ , (8)
Nb Bτ Nb Bτ
where τ is the integration interval, B is the radiometer eective bandwidth, Trec is
the receiver noise temperature, and Nb is the number of beams. Since Th Tv , it is
more aected by the erroneous power signal from land.
The required ∆Tmin can be achieved by making a trade-o between Nb for a given
reector diameter, and complexity of the feed. For a conically scanning antenna,
rotating at 11.5 RPM, Nb in the along-track direction is selected such to cover the
same strip width on the Earth at each frequency band. To reach the required ∆Tmin
we need:
2 beams along track at 6.9 GHz
3 beams along track and 7 beams across track at 10.65 GHz
5 beams along track and 6 beams across track at 18.7 GHz
For a push-broom case, the antenna is stationary, and its ∆Tmin is about one
order of magnitude better than the one for the scanner. This is at the expense of a
very large Nb , and correspondingly large number of receivers. For a swath of 600 km
we need:
58 beams across track at 6.9 GHz
246
3. Reflector antenna design
Sea
%iofitheib
Thi=i75iK 99.72 ittingi eami
e r ih theiEa
pow rthi
- 3i d B
i
dB
Beami θ 3
peak θci
nce
Dista D c
st
to c a
o
Land
iline
Coast Tlandi=i250iK
Figure 2: Footprint falling on the sea near a coast: illustration for the denition of the distance to
coast Dc .
For both cases listed above, we have considered a FP overlap of ∼ 30% both
along track and across track to assure accurate sampling of the temperature scene
on-ground, and the values of B and Trec as shown in the Table 1 and Table 2 [7].
Table 2: Assumed noise characteristics of the receiver
Conical scanner Push-broom

NF Trec NF Trec
C band 2.5 dB 226 K 3.5 dB 359 K
X band 2.5 dB 226 K 3.5 dB 359 K
Ku band 3.0 dB 290 K 4.0 dB 438 K
3 Reector antenna design

3.1 Reector geometries
To nd a suitable type of the antenna that is capable of fullling all performance re-
quirements, while having as compact as possible design, we have investigated dierent
reector systems, including conventional o-set parabolic reectors with circular and
247
Figure 3: Obtaining a torus from a parabolic arc.
elliptical apertures as the conical scanner, and toroidal single- and dual-reector an-
tennas for the push-broom concept. The sections below describe the selected conical
scanner and push-broom antenna implementations.
The conical scan antenna is a conventional oset paraboloid with projected aper-
ture D of 5 m and circular rim. The clearance is set to 1 meter in order to provide
space for the feed cluster and the focal length f is set to 3 m in order to make the
design more compact.
The push-broom antenna is a torus reector with projected aperture D of 5 m.
The torus is obtained by rotating a section of a parabolic arc around a rotation axis.
The focal length of the parabolic generator is also 5 m. A possible way of obtaining
the torus is shown in Fig. 3: the feed axis is selected parallel to the rotation axis,
implying that all feed element axes are parallel and orthogonal to the focal plane.
The array feed becomes therefore planar, simplifying the mechanical and electrical
◦
design. The antenna shall be able to provide a scan of ±20 corresponding to a swath
width of 600 km. The reector rim is found by intersecting the torus surface by the
◦ ◦
feed cone up to the out-most scan positions of 20 and −20 (see Fig.3 in [10]). The
antenna projected aperture is 5 × 7.5 m.
3.2 Reector surface technology

The conical scan and push-broom antennas can be realized by a deployable double
layer pantograph technology with two triangular wire-band curved nets, like in an
AstroMesh reector [21,22]. The corners of the triangles of the two nets are connected
by adjustment wires. The front net forms the support of the reector. Initially it
was assumed that the front net would be covered with a knitted metal mesh in order
to provide the necessary RF reection. It was realized, however, that the triangular
facets would generate high and unacceptable grating lobes which would deteriorate
the beams and the short distance to coast unless the triangles were made very small,
248
4. Limitations of cluster feeds of horns
Figure 4: Mechanical realization of the torus push broom reector dish.
i.e. 100 mm size. Consequently, it was proposed to construct the reector as a doubly
curved CFRS (Carbon Fibre Reinforced Silicon) surface (see Fig. 4.). The triangular
net is maintained to support the CFRS but the size of the triangles can be much
larger, around 400 mm.
4 Limitations of cluster feeds of horns

Cluster feeds for space-borne multi-frequency radiometers are typically designed to
provide a strong illumination taper toward the edge of the reector (when seen in
transmit situation) in order to maximize the antenna beam eciency and minimize
the side-lobe and cross-polarization power [25]. This approach, however, leads to (i)
the lower spatial resolution due to the widening of the footprint; and (ii) the diculty
to accommodate several feeds due to their large apertures, and hence several bands.
Figs 5(a-c) and 5(d-f ) illustrate these limitations for the considered scanner and push-
broom systems, respectively. As seen, Pcross of the scanner can only be minimized by
employing a feed with the aperture diameter larger than 5λ and illumination taper
◦
that is < 60 dB at 35 . This gives FP > 30 km and Dc > 23 km at C-band, while
FP = 20 km and Dc = 5 − 15 km are desired. The shortest Dc that can be achieved
is ∼ 20 km, for which the realized Pcross is at least 3 times higher than the desired
0.34%. At higher frequency bands, realizing the required Dc is not a problem, as
the side-lobe levels can be signicantly reduced (see Figs 6(c)) by under-illuminating
the reector aperture, while providing FP=10 km. However, the cross-polarization
power is not acceptable.
For the push-broom system, the dependence of the radiometer characteristics
from the illumination taper is similar to that of the scanner, and even larger feed
apertures are needed due to the more shalow surface of the reector. The main
challenges for this system are attributed to the complex shape of the torus reector,
and, as the result, more complex focal eld (compare Figs K.8(a) and K.8(b)). The
249

Horn diameter to realize corresponding taper, λ
[] Horn diameter to realize corresponding taper, λ
[]
5.3 4.8 4.3 3.7 3.0 2.2 7.8 7.2 6.4 5.6 4.5 3.2
50 50
C-band C-band
40 X-band 40 X-band
Ku-band Ku-band
Dc, [km]
Dc, [km]
30 30
20 20
10 10
0 0
-60 -50 -40 -30 -20 -10 -60 -50 -40 -30 -20 -10
Gauss beam taper @ 35 deg, [dB] Gauss beam taper @ 23.9 deg, [dB]
(a) (d)

[]
5.3 4.8 4.3 3.7 3.0 2.2 7.8 7.2 6.4 5.6 4.5 3.2
2 2
C-band C-band
X-band X-band
1.5 Ku-band 1.5 Ku-band
Pcross, [%]
Pcross, [%]
1 1
0.5 0.5
0 0
-60 -50 -40 -30 -20 -10 -60 -50 -40 -30 -20 -10
(b) (e)

[]
5.3 4.8 4.3 3.7 3.0 2.2 7.8 7.2 6.4 5.6 4.5 3.2
35 35 C-band
C-band
X-band X-band
30 30
Ku-band Ku-band
25 25 Requirement
FP, [km]
Requirement
FP, [km]
C,X-bands C,X-bands
20 20
15 15
Ku-band
10 10 Ku-band
5 5
-60 -50 -40 -30 -20 -10 -60 -50 -40 -30 -20 -10
(c) (f )
Figure 5: Radiometer characteristics, i.e. the distance-to-land, relative cross-polarization power and
footprint size, as function of the illumination taper of the Gaussian feed for (a-c) the conical scanner
and (d-f ) push-broom antenna conguration. The corresponding aperture diameter of the optimal
circular horn [23, 24] is shown on the top axis.
high coma-side lobes and non-circular main lobe of the focal eld distribution of the
torus reector (see Fig. K.8(b)) cannot be accurately sampled by a single (horn)
antenna feed; and this is the reason of the high side-lobe of the antenna far-eld
pattern (see Figs. 7(a-c)), and hence too large distance-to-coast. In contrast, dense
FPAs are capable of handeling these complexities, as will be demonstrated in the
following section.
250
5. Dense Focal Plane Arrays
C-band X-band Ku-band

57 57 57
55.93
50 xp, φ=0◦ 50 51.39 xp, φ=0◦
50
47.7 xp, φ=45◦ φ=45◦
◦
xp,
xp, φ=45◦
40 xp, φ=90 40 ◦ 40
xp, φ=90
φ=90◦
[dBi]
[dBi]
[dBi]
◦ xp,
co, φ=0 ◦
co, φ=0
◦ co, φ=0◦
co, φ=45 ◦
30 30 co, φ=45 30 φ=45◦
|Gco |, |Gxp |,
|Gco |, |Gxp |,
|Gco |, |Gxp |,
◦ co,
co, φ=90 ◦
co, φ=90
co, φ=90◦
Conical scanner
20 20 20
10 10 10
0 0 0
−5 −5 −5
−4 −3 −2 −1 0 1 2 3 4 −3 −2 −1 0 1 2 3 −2 −1 0 1 2
(a) (b) (c)

57 57 57
56.66
50 50 51.76 50
48.86 xp, φ=0◦
xp, φ=45
◦
xp, φ=45 ◦ xp, φ=45◦
40 40 40
φ=90◦
[dBi]
[dBi]
[dBi]
◦
xp, φ=90 xp, φ=90◦ xp,
co, φ=0
◦
co, φ=0◦ co, φ=0◦
30 30 30 φ=45◦
|Gco |, |Gxp |,
|Gco |, |Gxp |,
|Gco |, |Gxp |,
◦
co, φ=45 co, φ=45◦ co,
co, φ=90
◦
co, φ=90◦ co, φ=90◦
20 20 20
10 10 10
0 0 0
−5 −5 −5
−4 −3 −2 −1 0 1 2 3 4 −3 −2 −1 0 1 2 3 −2 −1 0 1 2
(d) (e) (f )
Figure 6: Far-eld pattern cuts for the conical scanner antenna at (a,d) C-band, (b,e) X-band, and
(c,f ) Ku-band, when the feed is (a-c) the Gaussian horn feed illuminating the reector edge with
the taper -30 dB, and (d-e) FPA with the optimum beamforming.
5 Dense Focal Plane Arrays

5.1 Array models and congurations
Based on the requirements derived in Sec. 2, three FPAs of half-wavelength dipole an-
tenna elements covering C-, X- and Ku-bands have been designed for each radiometer.
First, we computed the focal elds of several plane waves corresponding to the desired
beam directions, and then used these to derive the minimum aperture sizes of FPAs
and their positions in the focal regions, as shown in Fig. 8. After that, a parametric
study was carried out to determine the minimum needed Nel and the corresponding
inter-element separation distance del . Note that to reduce the computational time,
we have simplied the original MoM array model by assuming that all embedded
element patterns (EEPs) are identical to that of the central element (the validity of
this assumption will be conrmed in Sec. 5.3). The EEPs for each unique set of Nel
and element positions were imported into the reector antenna software GRASP10
to compute the secondary EEPs, which, in turn, were used to determine the optimum
element excitation coecients that will be discussed further. Table 3 summarizes the
results of this parametric study. As one can see, for the conical scanner we need 127,
263 and 333 antenna elements for the C- X- and Ku-band, respectively, to provide 2,
21 and 30 beams. Since the radiometric resolution of the push-broom system is much
higher (due to many more beams), as one can expect, this comes at the expenses
251
C-band X-band Ku-band

57 57 57
xp, φ=0◦
50 50 xp, φ=0◦ 50 53.17 xp, φ=45◦
xp, φ=0◦ 49.65 xp, φ=45◦ co, φ=0◦
46.65 ◦
xp, φ=45 co, φ=0◦ co, φ=45◦
40 40 40
Push-broom radiometer
◦
φ=45◦
[dBi]
[dBi]
[dBi]
co, φ=0 co, co, φ=90◦
◦
co, φ=45 co, φ=90◦
30 30 30
|Gco |, |Gxp |,
|Gco |, |Gxp |,
|Gco |, |Gxp |,
◦
co, φ=90
20 20 20
10 10 10
0 0 0
−5 −5 −5
−4 −3 −2 −1 0 1 2 3 4 −3 −2 −1 0 1 2 3 −2 −1 0 1 2
(a) (b) (c)

57 57 57
54.63
50 50 50
50.59
47.44
φ=0◦ xp, φ=0◦
xp, φ=0◦ xp,
xp, φ=45◦
40 ◦ 40 xp, φ=45◦ 40
[dBi]
[dBi]
[dBi]
xp, φ=45
◦ co, φ=0◦ co, φ=0◦
co, φ=0
φ=45◦ co, φ=45◦
30 ◦ 30 co, 30
|Gco |, |Gxp |,
|Gco |, |Gxp |,
|Gco |, |Gxp |,
co, φ=45
◦ co, φ=90◦ co, φ=90◦
co, φ=90
20 20 20
10 10 10
0 0 0
−5 −5 −5
−4 −3 −2 −1 0 1 2 3 4 −3 −2 −1 0 1 2 3 −2 −1 0 1 2
(d) (e) (f )
Figure 7: Far-eld pattern cuts for the push-broom radiometer antenna at (a,d) C-band, (b,e) X-
band, and (c,f ) Ku-band, when the feed is (a-c) the Gaussian horn feed illuminating the reector
edge with the taper -30 dB, and (d-e) FPA with the optimum beamforming.
Table 3: Number of elements

Array grid rectangular polar
C band 64 + 63 = 127 6 × 111 × 2 = 1332

X band 128 + 135 = 263 6 × 153 × 2 = 1836
Ku band 165 + 168 = 333 6 × 255 × 2 = 3060
of more elements. It is important to note that the required numbers of elements,

determined through this optimization procedure, are almost twice smaller than when
applying a conventional conjugate-eld-matching optimization approach (see Table
3 in [10]).
For both systems, the optimal del is near 0.75λ; this value satises the grating-
lobe free condition [11] and also minimizes the active impedance variation of antenna
elements due to their non-identical excitation [26, 27].
5.2 Optimization procedure for element excitation

In Sec. 2, it has been shown that the antenna far-eld beam should contains 99.72%
of the total power within a circular cone with half angle θc to realize the desired Dc .
The goal is, therefore, to determine the excitation coecients such that the angle θc
252
(a) (b)
Figure 8: Focal eld distributions of multiple plane waves incident on (a) the conical scan reector
antenna and (b) torus reector antenna at C-, X- and Ku-bands, as calculated using the Physical
Optics software GRASP10. For each band, the eld distributions are shown for two beam directions
and overlaid with the array grids.
becomes as small as possible, i.e. Dc is minimized.

The far eld from the reector antenna can be written as
Nel
X
Efar (θ, φ) = αi Efar,i (θ, φ), (9)
i=1
where Efar,i is the eld due to element i, Nel is the total number of elements; and αi
is the corresponding complex excitation coecient. The radiated power within the
cone of half-angle θc can be written as
Z 2π Z θc
Pc (θc ) = |Efar (θ, φ)|2 sin θ dθ dφ, (10)
0 0
If the expression (9) is inserted in (10) it is seen that it becomes a quadratic

polynomial in the αi variables and can be written in the form
Pc (θc ) = αH Aα, (11)
where α = [α1 , α2 , . . . , αN ]T and H is Hermitian operator. The matrix A is Hermitian

of size Nel × Nel such that the expression in (11) becomes a real number. Note that
the matrix A is a function of θc .
The power Pc (θc ) in (10) must be related to the total radiated power from the feed
array. This power, Ptot , can be computed from the expression (10) if θc is replaced
253
by π/2 and the reector patterns Efar,i are replaced by the array element patterns
Efar,array,i . Again the power Ptot becomes a quadratic polynomial in the variables α
such that
Ptot = αH Cα, (12)
For a given value of θc it is thus desired to nd the excitations α that maximize the
ratio
Pc (θc ) αH Aα
= H , (13)
Ptot α Cα
It can be shown that the maximum value of this ratio is the maximum eigenvalue λ
of the expression
Aα = λCα, (14)
and that the vector holding the complex excitation coecients are given by the
corresponding eigenvector.
Spillover loss
10 0 towards the 0
sky is 3.2%
Subtended angle
φ=0◦ −10 φ=0◦

5 −5
φ=45◦ φ=45◦
Subtended angle
φ=90◦ −20 φ=90◦
[dBi]
0 Co-polar, E-plane −10

[dB]
[dB]
Co-polar, H-plane
−30
|Gco |, |Gxp |,
Xp-polar, D-plane
−5 −15
|Gco |,
|Gco |,
−40
−10 −20
−50
−15 −25
−60
−20 −30 −70

−90 −75 −60 −45 −30 −15 0 15 30 45 60 75 90 −90 −75 −60 −45 −30 −15 0 15 30 45 60 75 90 −4 −3 −2 −1 0 1 2 3 4
θ, [deg]
θ, [deg]
θ, [deg]
(a) (b) (c)
Figure 9: (a) All embedded element patterns of the C-band FPA for the conical scanner at E-, H-
and D-planes, as obtained through the Method of Moments in CAESAR software [28], where the
bold lines correspond to the central antenna element of the array; (b) beamformed far-eld pattern
cuts of the FPA within the reector subtended angle region for the conical scan antenna, and (c)
far-eld pattern cuts of the reector antenna for beam 1. The solid lines correspond to the MoM
array model, dashed lines represent the model with the assumed identical embedded element patters
of the array, and the thin solid lines show the relative normalized dierence between the antenna
patterns obtained with the above models.
The present optimisation method is similar to the one reported in [12] which is
based on a more general Signal-To-Noise-Ratio algorithm but simpler to implement.
Since for the considered application scenario, the optimization is strongly driven by
the acceptable side-lobe and cross-polarization power of the antenna, the radiometric
performances obtained by the two algorithms are very similar.
5.3 Antenna patterns and radiometric characteristics

Dense FPAs oer more degrees of freedom in beam-forming, as compared to conven-
tional feeds, and thereby can provide highly optimized beams with more circular-
symmetric main lobes and much lower cross polarization and side-lobe levels, as
254
demonstrated in Figs 6(d-f ) and 7(d-f ). This results in signicantly better radiomet-
ric characteristics for both systems. As one can see in Table 4, the realized Dc of the
conical scanner is 6.6-14 km and Pcross is only 0.10-0.15% (i.e. about one order of
magnitude better than the horn feed); for the push-broom radiometer, the respective
quantities are less than 16 km (while the horn feed cannot fulll this requirement)
and 0.08-0.12% (i.e. 3 times better than the horn feed). Furthermore, the latter sys-
tem has wide scan-range performance, where the characteristics of all multiple beams
within the angular range of ±20 deg are virtually identical, thanks to the symmetry
of the torus reector in the azimuthal plane and the moon-like shape of the FPA that
matches the focal line of the reector (see Fig. 8(b)).
Table 4: Radiometric characteristics of the conical scanner and push-broom systems for the Gaussian
horn and FPA. The values in brackets are for the full MoM array model, and the other values are
when assuming identical embedded element patterns
Radiometer Require- HornConical scanner Push-broom

Horn FPA
characteristic ment feed FPA
Beam 1 Beam 2 feed
C-band
Distance to land, [km] <15 19.2 14.2 (14.2) 14.2 (14.2) 41.4 16.1
Rel. cross-pol. power, [%] <0.34 1.04 0.15 (0.06) 0.10 (0.07) 0.23 0.08
Beam eciency, [%] 97.2 95.6 (96.0) 95.6 (96.0) 96.1 97.8
Footprint, [km] <20 21 19.6 (19.6) 19.6 (19.6) 25.3 23.1
Footprint ellipticity 1.64 1.43 (1.44) 1.44 (1.44) 1.57 1.48
X-band
Distance to land, [km] <15 13.0 9.7 9.8 55.3 13.4
Rel. cross-pol. power, [%] <0.34 0.89 0.10 0.10 0.22 0.12
Beam eciency, [%] 97.7 98.2 97.4 95.0 98.4
Footprint, [km] <20 14.5 14.2 14.2 17.3 15.9
Footprint ellipticity 1.64 1.32 1.38 1.48 1.21
Ku-band
Distance to land, [km] <15 7.6 6.6 6.6 53.2 13.4
Rel. cross-pol. power, [%] <0.34 0.95 0.03 0.07 0.22 0.08
Beam eciency, [%] 97.7 97.4 97.2 84.5 98.0
Footprint, [km] <10 8.6 8.0 8.2 11.1 10.0
Footprint ellipticity 1.67 1.24 1.35 1.27 1.05
The accuracy of the above analysis (that is based on the assumption of identical
array element patterns) has been evaluated by cross-comparing the antenna patterns
and corresponding radiometric characteristics with those obtained through the full
MoM model. Fig. 9 shows the results for C-band, as the worse case scenario among
the considered ones. As seen, the relative dierence between the far-eld patterns
obtained with the simplied and more rigorous FPA models is negligible, so as the
dierence between the corresponding sets of radiometric characteristics (see Table 4).
This observation might appear count-intuitive, given a signicant variation between
the embedded element patterns (EEPs) of the array, as shown in Fig. K.9(a). How-
255
ever, one should realize that the optimal pattern of the feed leading to the minimum
distance to land represents a combined eect of the EEPs and element excitation coef-
cients. Hence, when the optimization algorithm is applied to the set of non-identical
EEPs, the excitation coecients are modied with respect to that determined for the
identical EEP case. For the considered arrays with more than 100 dipole antenna
elements, the resultant optimal feed patterns have been found very similar for both
array models (see the example for C-band in Fig. 9(b,c)). This observation, however,
may not be valid for arrays with fewer and denser-spaced elements.
6 Receiver considerations
In this section, we briey consider receiver resource requirements in order to see if
implementation of the present antenna concept is feasible and realistic. We con-
sider the receiver where the signals from dierent antenna elements contribute to
more than one beam, and each antenna element is connected to its own receiver,
followed by an A/D converter. The beam-forming process takes place in an Field
Programmable Gate Array (FPGA), using complex digital multipliers and adders.
Both the scanner and the push-broom system require a large number of elements to
fulll the radiometric requirements. Hence resource requirements concerning the size,
mass and especially power consumption, is an important issue.
A study of state-of-the-art microwave components, assuming a super-heterodyne
receiver (see Fig.7 in [29]), has been carried out. It has been found that at the
considered frequency bands, most components are small and light weight, and thus
volume and mass are not deemed to be a problematic issue. Power consumption has
dropped dramatically over the past decade, and 1 W per receiver is now a realistic
estimate. Furthermore, the output signals from FPA elements have to be optimally
combined in a dedicated beamforming network to form the desired antenna beams.
This involves a number of FPGAs and the average power consumption is estimated to
be 0.24 W per receiver. Future radiometers must include intelligent RFI detection and
mitigation processors. Based on a representative case study of such a processor [30],
the power consumption can be estimated to be 0.14 W per receiver.
In summary, the power estimate is: 1 + 0.24 + 0.14 = 1.38 W per receiver, using
present state-of-the-art components. The total number of receivers is 6228 in the
push-broom case. This results in a total power consumption of 8.6 kW, which is not
realistic today. For the scanner with 723 receivers, the estimate is 1000 W a large
number, but feasible.
The present study is a preparation for the future, and it is of interest to base
a power budget on realistic developments over a 5 years time frame. Already now,
A/D converters able to sub-sample signals up to X-band are available in research
labs, and within very few years Ku band is also possible. Thus we do not need the
super-heterodyne layout, and the local oscillator and its power consumption, can be
256
7. Conclusions
avoided. The new, fast A/D converters use very small signal levels typically around
−35 dBm, and hence not much gain is needed in the receiver (also saving on power).
The development concerning amplier power consumption is also impressive. For
global power budget estimates we can within a few years assume ∼ 35 mW per
receiver. If we assume a similar reduction for processing circuitry, the result is 9 mW
for the beam forming network, and 5 mW for the RFI processor, i.e. 49 mW per
receiver. For the push-broom system this amounts to a total power consumption of
305 W, which is certainly realistic. For the scanner the estimate is about 35 W.
7 Conclusions
Existing space-borne microwave radiometers that are used for the assessment of ocean
parameters like salinity, temperature, and wind can provide valid observations only
up to ∼ 100 km from the coastline, and hence do not allow for monitoring of the
coastal areas and ice-edge polar seas, and measuring under extreme wind and weather
conditions. To achieve the desired precision, as required for future missions, we
propose digitally-beamforming dense focal plane arrays (FPAs) previously not used
in space-borne applications, employed either in a traditional conical-scan o-set
parabolic reector antenna or in a wide-scan torus reector system.
When synthesized and excited according to the proposed optimum beamforming
procedure aiming to minimize the signal contamination given by the side-lobes
and cross-polarization of antenna beams covering the land, the number of the
FPA antenna elements and associated receivers can be kept to minimum. In this
procedure, the input parameters include the number of array elements, their positions
and the secondary embedded element patterns (EEPs), which are computed after
the illumination of the reector antenna, and the output parameters are the optimal
complex-valued element excitations. Although, the primary EEPs are generally not
identical, due to the array antenna mutual coupling and edge truncation eects, for
the considered FPAs with more than 100 dipole antenna elements and inter-element
spacing of 0.75λ, it has been found sucient to use a single primary EEP, i.e. the
one for a central element of the array, as the source of the secondary EEPs for all
elements in order to accurately predict the achievable radiometric characteristics.
For both types of radiometers, the realized resolutions are at least twice higher
than the values provided by the current systems, and the distance to coastline is
as short as 6-15 km. This excellent performance was shown to be impossible with
traditional multi-frequency FPAs of horns in one-horn-per-beam conguration, as
these cannot compensate for the high cross-polarization of o-axis beams in conical-
scanners, and produce unacceptably high side-lobes due to severe focal-eld under-
sampling eects in torus reector systems.
Our analysis of realistic developments of digital processors predicts acceptable
receiver resources budget for such multi-beam radiometers within a 5 years time
257
frame.
References
[1] C. Prigent, F. Aires, F. Bernardo, J.-C. Orlhac, J.-M. Goutoule, H. Roquet, and
C. Donlon, Analysis of the potential and limitations of microwave radiometry
for the retrieval of sea surface temperature: Denition of MICROWAT, a new
mission concept, Geophys. Res. Oceans, vol. 118, pp. 30743086, Jun. 2013.
[2] E. V. Zabolotskikh, L. M. Mitnik, and B. Chapron, New approach for severe

marine weather study using satellite passive microwave sensing, Geophys. Res.
Lett., vol. 40, pp. 33473350, Jul. 2013.
[3] N. Reul, B. Chapron, T. Lee, C. Donlon, J. Boutin, and G. Alory, Sea surface
salinity structure of the meandering Gulf stream revealed by SMOS sensor,
Geophys. Res. Lett., vol. 41, pp. 31413148, May 2014.
[4] STSE-high resolution microwave wind and temperature (MICROWAT).

[Online]. Available: https://www.yumpu.com/en/document/view/10779401/
download-document-microwat
[5] P. W. Gaiser, The WindSat spaceborne polarimetric microwave radiometer:

Sensor description and early orbit performance, IEEE Trans. Geo. Rem. Sens-
ing, vol. 42, no. 11, pp. 591603, Nov. 2004.
[6] Amsr-e instrument description. [Online]. Available: http://nsidc.org/data/

docs/daac/amsre\_instrument.gd.html
Artech House, 2006.


and test of a pushbroom radiometer, inAntennas and Propagation, 1989. ICAP
Kingdom, Apr. 1989, pp. 126130.

14.
258
References

15.
[12] , An optimal beamforming algorithm for phased-array antennas used in
multi-beam spaceborne radiometers, in Proc. European Conference on Antennas
and Propag. (EuCAP), Lisbon, Portugal, Apr. 2015, pp. 15.

15.
[14] J. Fisher and R. Bradley, Full-sampling array feeds for radio telescopes, in Proc.
SPIE, Radio Telescopes, vol. 4015, Munich, Germany, Jul. 2000, pp. 308318.
[15] K. F. Warnick, D. Carter, T. Webb, B. D. Jes, J. Landon, V. Asthana,

M. Elmer, R. Norrod, D. A. Roshi, and J. R. Fisher, Towards a high sensi-
tivity cryogenic phased array feed antenna for the Green Bank Telescope, in
XXXth URSI General Assembly and Scientic Symposium, Istanbul, Turkey,
Aug. 2011, pp. 14.

no. 8, pp. 15071521, Aug. 2009.
[17] B. Veidt, T. Burgess, R. Messing, G. Hovey, and R. Smegal, The DRAO phased
array feed demonstrator: Recent results, in 13th Int. Symp. on Antenna Tech-
nology and Applied Electromagnetics and the Canadian Radio Science Meeting,
ANTEM/URSI 2009, Ban, Canada, Feb. 2009, pp. 14.

18641875, Jun. 2011.
[19] Y. Wu, X. Zhang, B. Du, C. Jin, L. Zhang, and K. Zhu, Design of antenna array
for the L-band phased array feed for FAST, in Antennas Propagation (ISAP),
259
2013 Proceedings of the International Symposium on, Nanjing, China, Oct. 2013,
pp. 1113.
[20] K. F. Warnick, R. Maaskant, M. V. Ivashina, D. B. Davidson, and B. D. Jes,

High-sensitivity phased array receivers for radio astronomy, Proceedings of the
IEEE, vol. 104, no. 3, pp. 607622, Mar. 2016.
[21] M. W. Thomson, The AstroMesh deployable reector, in Antennas and Prop-

agation Society International Symposium, 1999. IEEE, vol. 3, Orlando, Florida,
Jul. 1999, pp. 15161519.
[22] K. van't Klooster, E. Medzmariashvili, S. Tserodze, L. Datashvili, and N. Tsig-

nadze, Large deployable reector conguration for spacebased applications in
telecommunications, science, and remote sensing, in International Scientic
Conference on Advanced Lightweight Structures and Reector Antennas, Tbilisi,
Georgien, Oct. 2009, pp. 7582.
[23] T. Teshirogi and T. Yoneyama, Modern Millimeter-wave Technologies. IOS

Press, 2001.
[24] P.-S. Kildal, Foundations of Antenna Engineering A Unied Approach for

Line-Of-Sight and Multipath. Kildal Antenn AB, 2015.
[25] S. Contu and F. M. Marinelli, The antenna system for the multi-frequency
imaging microwave radiometer: M.I.M.R, inAntennas and Propagation Society
International Symposium, 1994. AP-S. Digest, vol. 3, Seattle, WA, USA, Jun.
1994, pp. 20542057.
[27] K. F. Warnick, D. Carter, T. Webb, J. Landon, M. Elmer, and B. D. Jes,

Design and characterization of an active impedance matched low-noise phased
array feed, IEEE Trans. Antennas Propag., vol. 59, no. 6, pp. 18761885, Jun.
2011.

[29] N. Skou, S. S. Søbjærg, S. S. Kristensen, C. Cappellin, K. Pontoppidan,

M. Ivashina, A. Ihle, and K. v.'t Klooster, Future spaceborne ocean mis-
sions using high sensitivity multiple-beam radiometers, in IEEE International
260
References
Geoscience and Remote Sensing Symposium (IGARSS), Québec City, Québec,

Canada, Jul. 2014.
[30] N. Skou, S. Kristensen, A. Kovanen, and J. Lahtinen, Processor breadboard

for on-board RFI detection and mitigation in MetOp-SG radiometers, in Geo-
science and Remote Sensing Symposium (IGARSS), 2015 IEEE International,
Milan, Italy, Jul. 2015, pp. 14451448.
261
262
Paper L
Prediction of Far-Field Pattern Characteristics of

Phased Array Fed Reector Antennas by Modeling
Only a Small Part of the Array Case Study of
Spaceborne Radiometer Antennas
O. A. Iupikov, A. A. Roev, M. V. Ivashina

Propagation, EUCAP 2017, Paris, France, April 2017.
the thesis.
Prediction of Far-Field Pattern Characteristics of
Phased Array Fed Reector Antennas by
Modeling Only a Small Part of the Array Case
Study of Spaceborne Radiometer Antennas
O. A. Iupikov, A. A. Roev, M. V. Ivashina
Abstract
In this work we present an approach for the prediction of far-eld

pattern characteristics of phased array fed reector antennas by modeling
only a small part of the array. In this approach, the simulated EEPs of the
FPA are modeled as the phase-shifted versions of the simulated embedded
element pattern (EEP) of the central element, and thereafter combined
with the optimum weighting coecients in order to nd the total pattern
of the feed. Although, the EEPs of dense array antennas are generally not
identical (due to the array antenna mutual coupling and edge truncation
eects), for typical FPA excitation scenarios, where the array edge ele-
ments have relatively low weights to produce the desired illumination of
the reector, this simplied approach has been found suciently accurate.
1 Introduction
Recent advances in radio-frequency and digital electronics have allowed for the design
of novel antenna systems, which have superior beamforming capabilities. Examples
of such systems are spaceborne antennas for ocean surveillance and satellite commu-
nication; these systems are capable to provide multiple high-eciency beams (with
extremely low side-lobes or cross-polarization) and operate at several frequency bands
(typically L-, C-, X and Ku-bands), while having a compact single-antenna design.
These challenging requirements can be met by using dense focal plane arrays (FPAs)
feeding a reector (or a lens), or directly-radiating sparse irregular arrays [1, 2]. How-
ever, there are common problems with such large and multi-scale antenna designs,
including fast and accurate electromagnetic analysis as well as cost-ecient prototype
development. Dierent approaches have been proposed to overcome these problems
for the sparse arrays, where performance of the whole antenna system is evaluated
through the analysis of a small part of it (e.g. [1, 2]).
265
Paper L. Prediction of Far-Field Pattern Characteristics of Phased...
In this work we address this problem for the case of FPA systems, and in particular
present a validated simplied approach where a reduced-size FPA simulations are used
to predict the performance of the whole array feeding the reector antenna.
2 Antenna geometry and specications

To demonstrate the proposed approach we have considered a conical-scan oset
parabolic reector antenna (projected aperture diameter is 5 m, focal length is 3 m
and clearance is 1 m) with the 67-element array feed. This antenna system is currently
being considered for potential future ocean missions by ESA [3]. The requirements
for this mission are given in Table 1, [4], in terms of standard performance metrics
for oceanographic surveys. For the given satellite altitude and incidence angle, the
radiometric requirements can be transfered [5] to the antenna system specications
as shown in Table 2.
Table 1: Radiometric requirements for future ocean missions

Band Radiometric Spatial Dist.to
Freq., Polari- Bias,
width, resolution, resolution, coast,
[GHz] zation [K]
[MHz] [K] [km] [km]
L-band:
1.404 − 19 V, H 0.15 0.25 100 50-100
1.423
C-band:
6.8 − 7.0 200 V, H 0.30 0.25 20 15-20
7.2 − 7.4
Table 2: Antenna requirements
Antenna characteristic L-band C-band

Number of beams 4 2
Cross-polar. power over the Earth < 0.34 %
Power over the land < 0.28 %
Projected aperture diameter of the re-

5 m
ector
In previous system-level studies, we have applied this simplied approach to cross-

compare dierent radiometer system concepts, i.e. a traditional conical-scan o-set
parabolic reector antenna vs. a wide-scan torus reector system [6, 7], as well as to
perform parametric studies for the FPAs to dene the optimal number of antenna
elements, inter element spacing, and the arrangement of FPAs operating at dierent
bands [6, 8, 9]. In the current work, we validate this approach for the case of a wide-
band Vivaldi antenna element FPA feeding the conical-scan reector antenna, and
266
3. Array antenna design
use for this purpose the requirements in Table 1. To simplify the prototyping phase,
our focus will be on the high frequency performance only (C-band), for which the
small-size array demonstrator has only 24 elements, while the operational bandwidth
of the designed full-scale array covers both L- and C-bands.
3 Array antenna design

The Vivaldi antenna element in [10], which most closely meets the wide-band require-
ments of the project, was chosen as a reference: it has the relative bandwidth greater
than 6:1 over wide scan range (±45 deg). Since the geometry of the referred TSA
in [10] is for the frequency band of 0.41.6 GHz, we have scaled up this design with
some modications related to the following practical implementation aspects:
to improve the mechanical stability;
to improve the matching for the reference impedance of 50 Ohm (in opposite
to the original design, where 70 Ohm LNAs are used).
Thus a new element geometry of a dual-polarized phased array has been optimized
and analyzed with the aid of periodic boundary conditions. The slotline width, rate
of exponential slotline, cavity length, stub radius and stripline width were chosen as
variable parameters. The main goal was to achieve the impedance matching condition
with magnitude of the active reection coecient less than −10 dB within ±45 deg
scan range. The optimization have been performed with the commercially available
EM software HFSS and CST.
The nal antenna and feed geometries with dimensions are shown in Fig. L.1(a)
and L.1(b), respectively. Tapered slot prole is determined by curve:
y = C1 eRx + C2 , (1)
where R is the rate of exponential slotline, and coecients C1 and C2 are dened as
− y1
y2
C1 = (2)
− eRx1
Rx
e 2
y1 eRx2 − y2 eRx1
C2 = , (3)
eRx2 − eRx1
where points (x1 , y1 ) and (x2 , y2 ) determine a slot width in the excitation region and
the aperture, respectively.
Based on the simulations, a prototype of the small-scale dual-polarized array,
comprising 24 elements, was designed and manufactured (Fig. 2). The array antenna
structure consists of 4 orthogonally placed brass sheets with 3 TSA elements per
polarization. All elements are mounted on the 250x250 mm aluminum ground plane.
Each element is excited directly by a PCB feed with the SMA connector located
under the ground plane.
267
(a) (b)
x1 y1 x2 y2 R
13.21 0.25 82.57 13.92 0.04
Figure 1: Geometrical dimensions of (a) the proposed TSA element and (b) feeding plate. All
dimensions are given in [mm].
4 Analysis methodology and numerical results

For typical FPA excitation scenarios, the antenna elements at the edge of the array
have signicantly (-5...-15 dB) lower weighting coecients relatively to the elements
in the center. This implies that the dierences in the embedded element pattern
shapes, introduced by the edge eects, will have relatively weak contribution to the
total compound beam of the array when all elements are excited. This motivates our
assumption on the identical EEPs that can be taken to be the same as the pattern of
an element in the center. Such approach can greatly speed up the numerical analysis
of a reector antenna system, which is very important for optimization.
The antenna specications (see Table 2) dene the required array layout and
aperture area, which are shown in Fig. L.3(a). In order to validate the proposed
analysis approach, we have used the full-wave simulation results for this array as the
reference for the following simplied models:
1. Simplied model I, where FPA EEPs are phase-shifted versions of the EEP
of the central element (element No.18), which was obtained for the full-scale
array;
2. Simplied model II, where FPA EEPs are phase-shifted versions of the EEP
of the central element (element No.5), which was obtained for the small-sized
array, shown in Fig. L.3(b).
Figure 4 shows the EEPs for all these cases.

Figure L.3(c) shows the weighting coecients for Simplied models I and II have
been found through the dedicated optimum beamforming procedure detailed in [11]
that aims to satisfy the radiometric requirements. The coecients for the small-sized
268
4. Analysis methodology and numerical results
Figure 2: Photo of the manufactured reduced prototype.
array have been chosen to be a sub-set of the calculated coecients that correspond
to the most strongly excited elements; they are shown in Fig. L.3(d).
To cross-compare the array performances, we have used the active reection co-
ecient [12] of the central element, when all antenna elements are excited with a
certain complex-valued weight, as well as the radiometric characteristics specied in
Table 3.
The full-sized and small-scaled arrays have been modeled using a full-wave ap-
proach and the active reection coecient of the most excited elements are shown
in Fig. 5. The red curve (a) corresponds to the fully-excited full-sized array; dashed
curve (b) is for the same array when only 24 elements (highlighted in Fig. L.3(a)) are
active; and the blue curve (c) corresponds to the most excited element of the small
array, when the same weight coecients are used as for the previous case.
As one can see, the curves (a) and (b) are nearly identical. This is expected,
since they are for the same EM model of the full-sized array, and the array elements
outside the highlighted area are weakly excited, so they have negligible eect on
the central element active reection coecient. The result (c) diers from (b) since
the edge truncation eects are stronger in the smaller array. Nevertheless, the overall
prediction of the reference reection coecient (a) is good enough for such a strongly-
coupled antenna array.
The total primary- and secondary patterns of the array, i.e. the pattern before
and after reection from the dish) are cross-compared for the above cases in Fig. 6
and Fig. 7, respectively.
One can see the overall shape of the co-polar pattern of the reference full-wave
array model has been predicted rather well with both simplied models, however the
cross-polar components obtained with the latter appear to be higher. Similar obser-
269
29 30 31 32 33 34 35
39 43 47 51 55 59 63 67
22 23 24 25 26 27 28 10 11 12
38 42 46 50 54 58 62 66 15 18 21 24
15 16 17 18 19 20 21 7 8 9
37 41 45 49 53 57 61 65 14 17 20 23
8 9 10 11 12 13 14 4 5 6
36 40 44 48 52 56 60 64 13 16 19 22
1 2 3 4 5 6 7 1 2 3
(a) (b)
−40
0 −30 −20 −10 0
-31.5 -27.3 -30.2 -33.5 -36.9 -37.4
0
−10
-30.4 -10.1 -4.7 -9.4 -22.0 -33.8 -27.3 -30.2 -33.5
−10
-28.8 -5.2 0.0 -4.3 -18.4 -27.5 -31.5 −20
-10.1 -4.7 -9.4
−20
-30.8 -10.0 -4.6 -9.3 -21.9 -34.5 -5.2 0.0 -4.3
−30
−30
-31.1 -27.3 -30.2 -33.6 -36.5 -36.6 -10.0 -4.6 -9.3

−40 −40
(c) (d)
Figure 3: (a,c) Full-size array and (b,d) small-sized array layouts, and the corresponding weight-
ing coecients of the horizontally-polarized elements at 6.9 GHz (weighting coecients of the
orthogonally-polarized elements are not shown due to their low values), in [dB]
vations can be made for the radiometric characteristics, cross-compared in Table 3,

where the distance to coast, beam width, footprint size and beam eciencies have
very similar values for all models, while the cross-polarization powers are a bit pes-
simistic for Simplied models I and II. More close investigation of the latter eects
indicates the sensitivity of the presently used optimum beamforming approach to the
variations and asymmetries of the cross-polarization patterns. This will be studied
in our future work.
270
5. Conclusions
10
|Gco |, |Gxp |, [dBi]

0
−5
−10 Subtended angle

Co-polar, E-plane
−15 Co-polar, H-plane
Xp-polar, D-plane
−20
−90 −75 −60 −45 −30 −15 0 15 30 45 60 75 90
θ, deg
Figure 4: (solid lines) The E-, H- and D-plane embedded element pattern (EEP) cuts of the 67-
element array at C-band, simulated with the nite element method in HFSS software (reference
case), where the bold lines denote the EEP of the central element (no. 18) of the full-size array,
used for Simplied model I; and the dashed lines denote the EEP of the central element (no. 5) in
the small-sized array, used for Simplied model II.
Table 3: Final radiometer characteristics at C-band (6.9 GHz)

Radiometer characte- Require- Reference Simplied Simplied
ristic ment model model I model II
Beam eciency [%] 96.6 97.8 96.5
Cross-polar. power, [%] < 0.34 0.19 0.34 0.71
Distance to coast, [km] ≤ 15 11.4 14.5 13.6
Beam width, [deg] 0.648 0.664 0.647
Average footprint, [km] 20 18.8 19.5 18.6
5 Conclusions
The simplied modeling approach assuming identical embedded element patterns of
the phased array feed illuminating a large reector has been validated for the case
of a conical scan radiometer antenna fed with a strongly coupled Vivaldi antenna
element array. It has been shown that rather signicant dierences between the
embedded element patterns, introduced by the edge truncation eects, have relatively
weak contribution to the total compound beam of the array, when all elements are
excited to provide optimum illumination. As the result, radiometer characteristics
derived from the antenna far-eld pattern, such as the beam eciency, footprint, and
distance to coast can be predicted almost as equally well as with the full-wave array
model that is important for the antenna system optimization and array prototype
development phase. When applying this approach to applications with stringent
requirements on the cross-polarization, one could expect pessimistic estimation of its
271
0
(a) 67-elem array, all active
(b) 67-elem array, 24 active
(c) 24-elem array, all active
−10
|Γact |, [dB]
−20
−30
6.7 6.8 6.9 7 7.1 7.2 7.3 7.4 7.5
Frequency, [GHz]
Figure 5: Central active reection coecient for (a) full-size array, when all elements are excited to
form the optimum beam; (b) full-size array, when only 24 most strongly excited elements are used
in the calcultion; and (c) 24-element array with the same weight coecients as for the previous case.
The operating frequency bands are shown as green strips.
0
Subtended angle
−5 Reference model
−10 Simplified model I
Simplified model II
−15
−20
−25
−30
−35
−40
−60 −40 −20 0 20 40 60
θ, [deg]
Figure 6: Comparison of the total primary patterns obtained for the reference full-wave array
model and Simplied models I and II. Solid and dashed lines show the co-polarized (at φ = 0◦ ) and
◦
cross-polarized (at φ = 45 ) eld components, respectively.
0
Reference model
−10 Simplified model I
Simplified model II
−20
−30
−40
−50
−60
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3
θ, [deg]
Figure 7: Comparison of the total secondary patterns obtained for the reference full-wave array
model and Simplied models I and II. Solid and dashed lines show the co-polarized (at φ = 0◦ ) and
◦
cross-polarized (at φ = 45 ) eld components, respectively.
levels and the sensitivity to the optimum element excitation choice.
272
References
Acknowledgment
The present work has been funded by the Swedish National Space Board. The ra-
diometer requirements have been derived by the team consisting of TICRA and DTU-
Space (Denmark).
References
[1] C. Bencivenni, M. Ivashina, R. Maaskant, and J. Wettergren, Synthesis of max-
imally sparse arrays using compressive-sensing and full-wave analysis for global
earth coverage, IEEE Trans. Antennas Propag., vol. PP, no. 99, pp. 16, 2016.
[2] L. Poli, P. Rocca, G. Gottardi, and A. Massa, Design of simplied large array
structures for preliminary experimental validation, in 10th European Conference
on Antennas and Propagation (EuCAP), Davos, Switzerland, Apr. 2016, pp. 14.
[3] (2015) Earth Explorer Mission EE-9. [Online]. Available: http://explorercall.

esa.int/index.php/15-mission-ee9
[4] F. Collard et al., Sea surface temperature, wind and salinity (TWIST), pro-
posal for the Earth Explorer Mission EE-9, Unpublished.
Artech House, 2006.

14.
[7] O. A. Iupikov, M. V. Ivashina, C. Cappellin, and N. Skou, Digital-beamforming

array antenna technologies for future ocean-observing satellite missions, in Proc.
IEEE AP-S International Symposium, Fajardo PR, Jul. 2016, pp. 12.

15.

273
References

15.
[10] G. W. Kant, P. D. Patel, S. J. Wijnholds, M. Ruiter, and E. van der Wal, EM-
BRACE: A multi-beam 20,000-element radio astronomical phased array antenna
demonstrator, IEEE Trans. Antennas Propag., vol. 59, no. 6, pp. 19902003,
Jun. 2011.

beamforming algorithm for phased-array antennas used in multi-beam space-
borne radiometers, in Proc. European Conference on Antennas and Propag.
(EuCAP), Lisbon, Portugal, Apr. 2015, pp. 15.

274

Digital Beamforming Focal Plane Arrays For Radio Astronomy and Space-Borne Passive Remote Sensing

Uploaded by

Copyright:

Available Formats

Digital Beamforming Focal Plane Arrays For Radio Astronomy and Space-Borne Passive Remote Sensing

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

Digital Beamforming Focal Plane Arrays For Radio Astronomy and Space-Borne Passive Remote Sensing

Uploaded by

Copyright:

Available Formats

Thesis for the Degree of Doctor of Philosophy

Digital Beamforming Focal Plane Arrays

Department of Electrical Engineering

Göteborg, Sweden 2017

© Oleg Iupikov, 2017.

Department of Electrical Engineering

Keywords: phased array feeds, reector antenna feeds, beamforming, feed-reector

Thanks to my colleagues at the Onsala Space Observatory, especially to Prof. John

Bart Smolders, and the

List of Publications vii

2 Electromagnetic Analysis of Reector Antennas with Phased Ar-

3 Optimum Beamforming Strategies for Earth Observations 31

4 Beamforming Strategy for Beam Shape Calibration of PAF-equipped

Paper B A Plane Wave Approximation In The Computation Of

Paper C Fast and Accurate Analysis of Reector Antennas with

Paper D Novel Multi-Beam Radiometers for Accurate Ocean Surveil-

Paper E Dense Focal Plane Arrays for Pushbroom Satellite Ra-

2 Antenna Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 161

Paper F Improving the Calibration Eciency of an Array Fed Re-

Paper G Domain-Decomposition Approach to Krylov Subspace It-

Paper H Design of a push-broom multi-beam radiometer for future

5 RF performance results . . . . . . . . . . . . . . . . . . . . . . . . . . 209

Paper I An Optimal Beamforming Algorithm for Phased-Array An-

Paper J Enabling High-sensitivity Near-land Radiometric Measure-

Paper K Multi-Beam Focal Plane Arrays with Digital Beamforming

2.3 Distance to coast (Dc ) ⇒ acceptable side lobes and cross-polarization

Paper L Prediction of Far-Field Pattern Characteristics of Phased

1.1 Next generation radio telescopes

In addition to a continuous FOV and high sensitivity, high polarization discrimi-

1.2 Satellite radiometers for Earth observations

sign concepts of microwave radiometers are in use: push-broom and whisk-broom

This drawback may be signicantly reduced by employing dense PAFs consist-

1.3 Antenna arrays in satellite communication and

1.4 Modeling, design and calibration challenges of

 complexity to accurately model a large antenna array of complex antennas,

 cumbersomeness of analyzing a combined PAF-reector structure due to the

 development of optimal beamforming algorithms that provide performance re-

1.5 Goal and outline of the thesis

2.1 Analysis method: formulation and validation

where Z is the MoM matrix of size K ×K and V is a K ×1 excitation vector.

In+1 = −(Zrr )−1 Zrf Ifn

2.2 Acceleration techniques

1 Here Z includes the eect of the antenna port terminations [63].

Step (i) V Step (ii)

Step (iii) Step (iv)

Ir = Ir0 + Ir1 + Ir2 + . . .

inecient, in particular for complex-shaped electrically large feed antennas employ-

with a computational burden when calculating the PO equivalent current on elec-

2.2.1 Single plane wave approximation of the reector eld

− r10 A(0) exp (−jk2r0 )

Figure 2.2: Semi-analytical PW approximation as described in [44].

Table 2.1: Errors due to the plane wave approximation

Focal eld Gain Gain Impedance

Feed: Pyramidal horn

Feed: Pyramidal horn with extended ground plane

Feed: 121-element dual-polarized dipole array

Table 2.2: Total simulation time

Horn Horn with Dipole array Vivaldi

 Method 2 provides a better prediction of all the system parameters, since it

Keywords: phased array feeds, reector antenna feeds, beamforming, feed-reector

2 Electromagnetic Analysis of Reector Antennas with Phased Ar-

Paper C Fast and Accurate Analysis of Reector Antennas with

Paper F Improving the Calibration Eciency of an Array Fed Re-

sign concepts of microwave radiometers are in use: push-broom and whisk-broom

This drawback may be signicantly reduced by employing dense PAFs consist-

complexity to accurately model a large antenna array of complex antennas,

cumbersomeness of analyzing a combined PAF-reector structure due to the

development of optimal beamforming algorithms that provide performance re-

1 Here Z includes the eect of the antenna port terminations [63].

inecient, in particular for complex-shaped electrically large feed antennas employ-

2.2.1 Single plane wave approximation of the reector eld

Focal eld Gain Gain Impedance

Method 2 provides a better prediction of all the system parameters, since it

2.2.3 Near-eld interpolation (NFI) technique

initial eld sampling points

5. Computes the reector PO current (see e.g. [71, p. 343])

Table 2.3: Total simulation time (for D = 118λ reector)

Table 2.4: Errors due to applying the eld approximations, %

2.3 Experimental verication of the CBFM-PO ap-

2.4 Numerical studies for dierent types of reec-

The aperture eld distributions at two frequency points corresponding to the

The feed-reector interaction (standing wave) eects give rise to oscillations in

An FFT-enhanced Plane Wave Spectrum (PWS) approach has been formu-

The (active) impedance matching of the strongly-coupled PAF elements appears

A domain-decomposition technique into Krylov subspace iteration has been

3.2 Reector antenna design

Figure 3.5: Generalized representation of the PAF reector antenna system.

3.3.2 Standard Conjugate Field Matching beamformer CFM

3.3.4 Advanced Maximum Beam Eciency beamformeR