International Do toral S hool in Information
and Communi ation Te hnology
DISI- University of Trento
Advan ed Analysis and Synthesis Methods
for the Design of Next Generation
Refle tarrays
Angelo Gelmini
Tutor:
PaoloRo a, Asso iate Professor
University of Trento
Advisor:
Gia omoOliveri, Asso iate Professor
University of Trento
s ientist
I want to thank also all my familiy, my un les, ousins and Daniel for the
interest and the moral support.
An important thank-you goes to Giulia that has tolerate and support me
beyond every di ulties and also to her parents that have a ept me in the
family.
Spe ial thankstomy friendsfor thebeautiful momentsandfor havinguseful
and amusing onversations.
To my olleguesand friends of the ELEDIAgroupand toallthe people that
support me.
The design of ree tarray surfa e urrents that satisfy both radiation and
user-dened antenna feasibility onstraints is addressed through a novel paradigm
whi htakesadvantageofthenon-uniquenessofinverse sour e(IS)problems. To
thisend, thesynthesis isformulatedinthe IS frameworkanditsnon-measurable
solutions are employed as a design DoF. Thanks to the adopted framework,
a losed-form expression for the design of ree tarray surfa e urrents is
de-rived whi h does not require any iterative lo al/global optimization pro edure
and whi h inherently satises both the radiation and the feasibility design
on-straints. The features and potentialities of the proposed strategy are assessed
through sele ted numeri alexperimentsdealing with dierent ree tarray
aper-ture types/sizesand forbiddenregion denitions.
Keywords
Ree tarray Synthesis, Non-RadiatingCurrents, Inverse Sour e Problems,
[C1℄ G.Oliveri,F.Apolloni,A.Gelmini,E.Bekele,S.Ma i,andA.Massa,
Numeri alhomogenization andsynthesis ofwave polarizersthrough
the material-by-design paradigm, 9th European Conferen e on
An-tennas and Propagation (EUCAP 2015) (ISBN 978-8-8907-0185-6),
Lisbon, Portugal, pp. 1-4, April12-17, 2015.
[C2℄ N.Anselmi,M.Donelli,A.Gelmini,G.Gottardi,G.Oliveri,L.Poli,
P. Ro a, L.Tenuti, and A. Massa, Design and optimization of
ad-van edradarand ommuni ationssystemsandar hite tures
ELE-DIAResear h Center, Atti XXI Riunione Nazionale di
Elettromag-netismo(XXI RiNEm), Parma, pp. 164-167,12-14 Settembre 2016.
[C3℄ G.Oliveri, M. Salu i,A. Gelmini,L. Poli, P. Ro a, and A. Massa,
SARarraysynthesisfornextgenerationEarthobservationsystems,
11th European Conferen e on Antennas and Propagation (EUCAP
2017)(no. 978-88-907018-7-0/17/$31.00
2017IEEE),Paris,Fran e, pp. 2312-2314, Mar h 19-24, 2017.
[C4℄ G.Oliveri, A.Gelmini,M. Salu i,D. Bres iani,and A.Massa,
Ex-ploitingnon-radiating urrents in ree tarray antenna design, 11th
European Conferen e on Antennas and Propagation (EUCAP 2017)
(no. 978-88-907018-7-0/17/$31.00
2017 IEEE), Paris, Fran e, pp. 88-91,Mar h 19-24, 2017.
[C5℄ A.Gelmini,G. Gottardi,and T. Moriyama, A ompressive
sensing-based omputational method for the inversion of wide-band ground
penetrating radar data, 7th International Workshop on New
Com-putational Methods for Inverse Problems (NCMIP 2017), Ca han,
e tarray antennasimpli ationthrough non-radiating urrents
syn-thesis, 2017IEEE AP-S InternationalSymposium andUSNC-URSI
RadioS ien eMeeting (no. 978-1-5386-3284-0/17/$31.00
2017IEEE), San Diego,California,USA, pp. 1185-1186, July 9-15, 2017.
[C7℄ A. Gelmini, M. Salu i, G. Oliveri, and A. Massa, Robust
diagno-sis of planar antenna arrays through a Bayesian ompressive
sens-ing approa h, 6th Asia-Pa i Conferen e on Antennas and
Prop-agation (APCAP 2017), Xi'an, 2017, pp. 1-3. doi:
10.1109/AP-CAP.2017.8420936.
[C8℄ M.Salu i,A.Gelmini,G.Oliveri,andA.Massa,Exploitationof
fre-quen ydiversity inGPRimagingthrough aninnovative
onstrained-BCSmethod, 12thEuropean Conferen e onAntennasand
Propaga-tion (EUCAP 2018) , London, UnitedKingdom April9-13, 2018.
[C9℄ A. Gelmini, M. Salu i, G. Oliveri, and A. Massa, Innovative
syn-thesisof ree tarrays within the non-radiatinginverse sour e
frame-work,12thEuropeanConferen eonAntennasandPropagation
(EU-CAP 2018), London, United Kingdom, pp. 1-4, April 9-13, 2018
(DOI:10.1049/ p.2018.0864).
[C10℄ M. Bertolli,M. Donelli, A. Massa, G. Oliveri, A. Polo, F. Robol, L.
Poli,A.Gelmini,G.Gottardi,M.A.Hannan,L.T.P.Bui,P.Ro a,
C. Sa hi, F. Viani, T. Moriyama, T. Takenaka, and M. Salu i,
Computationalmethodsforwirelessstru tural healthmonitoringof
ultural heritages, 8th International Conferen e on New
Computa-tionalMethodsforInverseProblems(NCMIP2018),Ca han,Fran e,
May 25,2018.
[C11℄ M.Salu i,G.Oliveri,A.Gelmini,andA.Massa,Over oming
feasi-bility onstraints in ree rattary design by exploiting non-radiating
urrents, 2018 IEEE AP-S International Symposium and
USNC-URSIRadioS ien eMeeting Boston,Massa hussets,USA,July8-13,
ationsELEDIAResear hCenter, Atti XXIIRiunioneNazionale
diElettromagnetismo(XXII RiNEm),Cagliari,pp. 325-328,3-6
Set-tembre 2018.
[C13℄ N. Anselmi, R. Azaro, P. Bui, A. Gelmini, G. Gottardi,A. Hannan,
G.Oliveri, L. Poli, A. Polo, F. Robol, P. Ro a, M. Salu i,and A.
Massa, Antenna Synthesis and Optimization ELEDIA Resear h
Center, Atti XXIIRiunione Nazionale di Elettromagnetismo (XXII
RiNEm),Cagliari, pp. 333-336,3-6Settembre 2018.
[C14℄ M. Salu i, A. Gelmini, G. Oliveri, and A. Massa, From
inverse-sour e problems to ree tarray design - An innovative approa h for
dealingwithmanufa turingandgeometri al onstraints, 13th
Euro-pean Conferen e on Antennas and Propagation (EUCAP 2019).
[C15℄ G.Oliveri,M.Salu i,A.Gelmini,and A.Massa,
Computationally-e ient synthesis of advan ed ree tarrays through a
system-by-designtool,13thEuropeanConferen eonAntennasandPropagation
(EUCAP 2019).
[C16℄ A. Polo, M. Salu i, A. Gelmini, G. Gottardi, G. Oliveri, P. Ro a,
and A. Massa, Advan ed tea hing in EM - Towards an integration
of theoreti al skills and appli ative/industrial skills, 13th European
Conferen e on Antennas and Propagation (EUCAP 2019).
[C17℄ G. Oliveri, M. Salu i, A. Gelmini, and A. Massa, Frontiers in
re-e tarray design, 2019 IEEE AP-S International Symposium and
USNC-URSI Radio S ien e Meeting, Atlanta,Georgia,USA.
[C18℄ G.Oliveri, A.Gelmini, G.Gottardi,and M. Salu i,
Metamaterial-by-design-A Paradigmfortheindustrialsynthesisof EM
manipula-tion devi es, 2019 IEEE International Conferen e on Mi rowaves,
[R1℄ A.Gelmini,G. Gottardi,and T. Moriyama, A ompressive
sensing-based omputational method for the inversion of wide-band ground
penetratingradardata, JournalofPhysi s: Conferen eSeries,,vol.
904, pp. 1-7, 2017 (extended version submitted torevision by an
in-ternational omiteeof[C5℄)(DOI:10.1088/1742-6596/904/1/012002).
[R2℄ M. Salu i,A. Gelmini, L. Poli,G. Oliveri, and A. Massa,
Progres-sive ompressive sensing for exploiting frequen y-diversity in GPR
imaging, Journal of Ele tromagneti Waves and Appli ations, vol.
32,no. 9,pp. 1164-1193,2018(DOI:10.1080/09205071.2018.1425160).
[R3℄ M. Salu i, A. Gelmini, G. Oliveri, and A. Massa, Planar arrays
diagnosisbymeansofanadvan edBayesian ompressivepro essing,
IEEE Transa tions on Antennas and Propagation, vol. 66, no. 11,
pp. 5892-5906, November2018 (DOI: 10.1109/TAP.2018.2866534).
[R4℄ M.Salu i,A. Gelmini,G.Oliveri, N. Anselmi,and A. Massa,
Syn-thesisof shaped beamree tarrays with onstrainedgeometryby
ex-ploiting non-radiatingsurfa e urrents, IEEE Transa tions on
An-tennas and Propagation, vol. 66, no. 11, pp. 5805-5817, November
2018(DOI: 10.1109/TAP.2018.2869036).
[R5℄ M. Bertolli,M. Donelli, A. Massa, G. Oliveri, A. Polo, F. Robol,L.
Poli,A.Gelmini,G.Gottardi,M.A.Hannan,L.T.P.Bui,P.Ro a,
C. Sa hi, F. Viani, T. Moriyama, T. Takenaka, and M. Salu i,
Computationalmethodsforwirelessstru tural healthmonitoringof
ulturalheritages, Journal of Physi s: Conferen e Series,vol. 1131,
pp. 1-7,2018 (extended version submittedtorevision by an
Instantaneous brain stroke lassi ation and lo alization from real
s attering data, Mi rowaveand Opti alTe hnologyLetters, vol. 61,
1 Introdu tion 1
2 Problem Formulation 11
2.1 Radiation fromsurfa e urrent . . . 11
2.2 Dis retization of the ree tarray surfa e . . . 15
2.3 Inverse Sour e problemdenition . . . 18
3 Non-Measurable Currents-based Solution Method 19 3.1 Field dis retization . . . 20
3.2 Trun ated SingularValue De omposition . . . 22
3.3 Synthesis approa h . . . 26
3.4 Non-measurable/Non-radiatingdenition . . . 27
3.5 Forbidden region onstraint denition . . . 28
3.6 Final losed-formformulation . . . 30
4 Method Assessment 33 4.1 Error metri s . . . 33
4.2 Square ree tarray:
55 × 55
elements . . . 344.2.1 Step-by-step pro edure with lower dimensionality ase . . 35
4.2.2 Analysisvs. variousforbiddenregionshapeskeepingsame order dimension . . . 44
4.2.3 Changingthedimensionofthesametypeofforbiddenregion 48 4.2.4 Large dimension and omplex topology forbiddenregion . 52 4.3 Re tangular ree tarray:
81 × 69
elements . . . 564.3.1 Large dimension and omplex topology of the forbidden region . . . 59
test ase . . . 65
4.1 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Performan e Assessment - Varying the geometry. . . 45
4.2 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Performan e Assessment - Fixed geometryvarying the dimension. 51
4.3 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Performan e Assessment -Complex and large geometries. . . 55
4.4 Re tangularAperture (
M ×N = 81×69
,∆x = ∆y = 3.07×10
−1
λ
)
1.1 Geometry omparisonof(a)PhasedArrayAntenna,(b)Ree tor
Antenna, ( )Ree tarray Antenna. . . 2
1.2 Ree tarray Antenna implemented by (a)[56℄ and by (b)Pozar
(http://www.e s.umass.edu/e e/pozar/ree t.jpg). . . 4
1.3 Geometry of the ree tarray antenna elements: (a)[15℄, (b)[64℄,
( )[65℄,(d)[66℄. . . 7
2.1 Geometry of the ree tarray antenna. . . 12
2.2 Co-polar and ross-polar unit ve tor following the Ludwig third
denition. . . 17
3.1 Example of singular values distribution
ψ
w
,w = 1, ..., W
, taking intoa ount a trun ation orderH
and a trun ation thresholdτ
. 25 3.2 Geometry of the ree tarray antenna. . . 284.1 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Ree tarray geometry. . . 34
4.2 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
,
K = 11
) - Example of forbidden regionΦ
, E-Shape forbidden region withK = 11
numberof elements. . . 35 4.3 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
- Plotof the referen e urrent (a) magnitude
J
ref
x
(x, y)
and (b) phase∠J
ref
x
(x, y)
andradiatedeld( )magnitudeF
CO
ref
(u, v)
and (d) phase∠F
ref
CO
(u, v)
. . . 36 4.4 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
4.5 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Normalizederror
ξ
varying the SVD thresholdτ
. . . 38 4.6 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
,
K = 11
) - Plots of (a)( )(e)(g) the magnitude and (b)(d)(f)(h) thephaseof(a)(b)J
ref
x
(r)
andsynthesized( )(d)J
MN
x
(r)
,(e)(f)F
CO
ref
,and (g)(h)F
MN
CO
. . . 40 4.7 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
,
K = 11
) - Plots of (a)( ) the magnitude and (b)(d) the phase of (a)(b)J
N R
x
(r)
and synthesized ( )(d)F
N R
CO
. . . 41 4.8 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
,
K = 11
) - Plots of (a)( ) the magnitude and (b)(d) the phase of the synthesized (a)(b)J
x
(r)
and ( )(d)F
CO
(r)
. . . 43 4.9 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Denitionof forbidden regions
Φ
. . . 44 4.10 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Plotsof
|J
x
(r)|
assuming(a)Cross-shaped(K = 28
),(b) Ring-shaped (K = 32
), ( ) Cir ular Ring-shaped (K = 36
) and (d) Cir le-shaped (K = 37
) forbiddenregions. . . 46 4.11 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Plotsof
∆F
CO
(u, v)
whenassuming(a)Cross-shaped(K = 28
), (b)Ring-shaped(K = 32
),( )Cir ularRing-shaped(K = 36
) and (d) Cir le-shaped (K = 37
)forbidden regions. . . 47 4.12 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
- Denition of forbidden regions
Φ
keeping the same shape but varying the dimension: (a)K = 4
, (b)K = 25
, ( )K = 49
and (d)K = 100
. . . 48 4.13 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Plots of
|J
x
(r)|
assuming dierent dimension of a Square-shape (a)K = 4
, (b)K = 25
, ( )K = 49
and (d)K = 100
forbidden regions. . . 494.14 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
- Plots of
∆F
CO
(u, v)
when assuming dierent dimension of a Square-shape (a)K = 4
, (b)K = 25
, ( )K = 49
and (d)4.15 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
,
Square-shaped forbiddenregion)-Behaviourof
ξ
and∆t
versusK
. . . 51 4.16 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
- Denition of forbidden regions
Φ
with omplex shape and large dimension: (a) Triangle-shapedK = 55
nearer to the orner, (b)Triangle-shapedK = 55
,( )ELEDIA-shapedK = 54
and (d) Diamond-shapedK = 115
. . . 52 4.17 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Plots of
|J
x
(r)|
assuming dierent dimension of a Square-shape (a)K = 4
, (b)K = 25
, ( )K = 49
and (d)K = 100
forbidden regions. . . 534.18 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
- Plots of
∆F
CO
(u, v)
when assuming dierent dimension of a Square-shape (a)K = 4
, (b)K = 25
, ( )K = 49
and (d)K = 100
forbiddenregions.. . . 54 4.19 Re tangularAperture (M ×N = 81×69
,∆x = ∆y = 3.07×10
−1
λ
)
- Ree tarray geometry. . . 56
4.20 Re tangularAperture (
M ×N = 81×69
,∆x = ∆y = 3.07×10
−1
λ
)
- Plotof the referen e urrent (a) magnitude
J
ref
x
(x, y)
and (b) phase∠J
ref
x
(x, y)
and radiated eld ( ) magnitudeF
CO
ref
(u, v)
and (d) phase∠F
ref
CO
(u, v)
and the minimum-norm solution (e) magnitudeJ
MN
x
(x, y)
and (f) phase∠J
MN
x
(x, y)
and radiated eld (g) magnitudeF
MN
CO
(u, v)
and (h) phase∠F
MN
CO
(u, v)
. . . . 58 4.21 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
- Denition of forbidden regions
Φ
with omplex shape and large dimension: (a) ELEDIA-shapedK = 54
and (b) Diamond-shapedK = 115
. . . 59 4.22 Re tangularAperture (M ×N = 81×69
,∆x = ∆y = 3.07×10
−1
λ
)
-Plots of(a)( )the magnitudeandand (b)(d)thephase of
J
x
(r)
when assuming (a)(b) ELEDIA-shaped (K = 54
) and ( )(d) Diamond-shaped (K = 115
) forbiddenregions. . . 60 4.23 Re tangularAperture (M ×N = 81×69
,∆x = ∆y = 3.07×10
−1
λ
)
-Plotsof
∆F
CO
(u, v)
whenassuming(a)ELEDIA-shaped(K =
4.24 Square Aperture (
M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Denition of forbiddenregions
Φ
with dierent shapes: (a) E-shapedK = 11
, (b) Cross-shapedK = 28
, ( ) Ring-shapedK = 32
, (d) Cir ular Ring-shapedK = 36
and (e) Cir le-shapedK = 37
. . . 62 4.25 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Plots of
|J
x
(r)|
assuming an(a) E-shapedK = 11
, (b) Cross-shapedK = 28
, ( ) Ring-shapedK = 32
, (d) Cir ular Ring-shapedK = 36
and (e) Cir le-shapedK = 37
forbidden. . . 63 4.26 Square Aperture (M × N = 55 × 55
,∆x = ∆y = 3.73 × 10
−1
λ
)
-Plotsof
∆F
CO
(u, v)
whenassumingan(a)E-shapedK = 11
,(b) Cross-shapedK = 28
, ( ) Ring-shapedK = 32
, (d) Cir ular Ring-shapedK = 36
and (e) Cir le-shapedK = 37
forbidden region. . . 644.27 Re tangularAperture(
M ×N = 81×69
,∆x = ∆y = 3.07×10
−1
λ
)
Introdu tion
Antennas abletoexhibithigh gainsand arefullyshaped patternsare of
fun-damental importan e in radar, satellite remote sensing and
long-distan e/high- apa ity ommuni ationsystems [1℄-[6℄.
In order to meet su h ambitiousperforman e requirements,the te hnologies
traditionallyemployed areree tor antennas[6℄andphasedarrays[7℄,[8 ℄. Infa t
both te hnologies an a hieve a very high gain and are able togenerate shaped
patterns. In ree tor antenna the high gain apability is a hieved thanks to the
abilityto on entratetheeldthatis omingfromthefeedertoasingledire tion
inspa e(duetothe paraboli ree tor),whilethe shaped beam an beobtained
by shaped prole ree tors (adding some bumps into the paraboloid ree tor
surfa e) [9℄.
In [9℄itisproposed touse anoptimizationpro edurethattakesininput the
typeof the ree tor surfa e ( ir ular, ellipsoid, square, re tangular, oni et ..),
the feed onguration (horn antenna, array antennas, et ..) and position (at
the enter or shifted),the ree tor onguration(single ree tor, dual ree tor,
Cassegrain,et ..) andthe radiation hara teristi s. The optimizer omputes the
radiatedpattern applying the theory of generalized dira tion, i.e. the physi al
theory of dira tionis used toanalyse the antennaand produ ea ost fun tion
to quantify the mat hing of the radiation hara teristi s.If the ost fun tion is
not minimized, the optimizer reates anew trialsolution.
In phased arrays, high gain and a properly shaped beam are given by the
position of the elements (regular latti e, sparse latti e, random latti e, et ..),
Figure 1.1: Geometry omparison of (a) Phased Array Antenna, (b) Ree tor
Antenna, ( ) Ree tarray Antenna.
tapering,thinning,phasesynthesis,timemodulatedarray, lusteringte hniques,
et ..)[7 ℄[8℄.
Though able toa hievethe desiredrequirements,bothsolutionshave
signi- ant drawba ks. Ree tor antennas exhibit high manufa turing omplexity, are
di ulttobeimplemented as re ongurable antennas(unless me hani al
steer-ing is onsidered, whi h is typi allyavoided in spa e appli ations) and are also
hara terizedby non- onformalshapes[6℄. Moreover, spa eappli ation ree tor
antennas suer manufa ture toleran e and deformation problems [10℄ that an
severely ae t the antenna operational. Phased antennaarrays are expensive in
terms of fabri ation and power onsumption (and, onsequently, need
tempera-ture ontrol, not suitablefor spa eappli ations). In addition,su hantennas are
also heavy due to support and feeding network, and their design is not trivial
[7℄[8℄.
In order to deal with the aforementioned issues, ree tarray antennas have
emerged as a possible solution to yield high-gain shaped beam antennas with
lowrealization osts, at/ onformalshapes, andlow- onsumptionfeednetworks
[6℄,[11℄. Thanks to their potentials and exibility [6℄,[12℄-[36℄, the design of
shaped-beam ree tarray antennas has be ome a very a tive resear h eld and
several methodologieshave been proposed tothis end [37℄-[52℄.
apa-bilityof ombiningthe positivefeatures of both lassi alree tor antennas(i.e.,
highgain, low ost andeasyfabri ation)[6℄and phasedarrays (i.e.,
re ongura-bility and lowprole)[7℄. Typi ally, they onsist of a planar array of mi rostrip
pat hes printed on a ground-ba ked diele tri substrate and illuminated by a
feeder (e.g., a horn antenna, or also a phased array). Size, shape and
arrange-mentofthemetalli pat hesareproperlydesignedsu hthattheeldree tedby
thepassive/a tivesurfa emeetsthedesiredpatternfeatures(e.g.,steeringangle,
sidelobe level, bandwidth,et ..)[6 ℄. As a main onsequen e, ree tarrays do not
requirethe use ofabulkyparaboli dis , whilethe tuningofthe radiatedeldis
obtained without the need for expensive beam-formingnetworks or me hani al
steering[14℄.
The rst exampleof aree tarray antennawasproposed atthe beginningof
60's by Berry [11℄, who proposes to use trun ated waveguides as ree ting
ele-ments. Thesewaveguideshavedierentlengthsthat areable toimposeaproper
phase shift to obtain a desired ree ted pattern. The ree tarrays produ ed
with this te hnology an a hieve good performan es and an handle very high
power (no diele tri substrate) at the ost of using a heavy stru ture. For this
reason, only with the development of the mi rostrip te hnology in the late 80's
the ree tarray arose asa leadingte hnology.
Before mi rostrip te hnology, another kind of ree ting stru ture was
ana-lyzed: the Spiralphase ree tarray [53℄. In this work four arms of spirals are
onne ted with swit hing diodes that a tivate a dierent pair of arms and thus
permit to ontrol the s an angle of the ir ular polarized ree tarray. However,
due to the diodes ir uitand the spiral avity (
λ/4
), the stru ture be omes too bulky topermit ane ient implementation.The simplestdesign of a ree tarray is proposed in [54℄ and [55℄ and
imple-mented by [56℄,[57℄ [Fig. 1.2(a)℄, and onsists in mi rostrip pat hes with xed
shapeanddierentadaptingstubs. Sin ethesestubshavedierentlengths,they
anprovideadierentimpedan e, and onsequently adierentphase shift. The
major problem with the stub te hnology is that this method is inherently
nar-rowband,sin ethe stubstru turemustbedimensionedforaspe i wavelength,
and parasiti oupling withadja ent elements ouldbea possible issue.
Pozar et al. in [58℄,[59℄ and Chaharmir et al. [60℄ propose to introdu e, in
the same planar stru ture, pat hes with dierent dimension, rotation, or even
(a) (b)
Figure 1.2: Ree tarray Antenna implemented by (a)[56℄ and by (b)Pozar
(http://www.e s.umass.edu/e e/pozar/ree t.jpg).
Thiskindofdesignsolvestheproblemsofthestubs,improvesthebandwidthand
allowsthe designertohaveabetter ontrolonpolarization. However, atrade-o
must be taken into a ount when designing ree tarray antennas that radiated
shaped beams. Ifaparti ularshapedbeamisdesiredinorderto overonlysome
regionsof theEarth (e.g. abeam that an over northernEuropewithout
send-ing power on sea areas) the phase distribution on the ree tarray aperture has
a non-smooth behavior. This meansthat adja ent elements ouldhave a
signif-i antly dierent phase shift and this implies very dierent and omplex shapes.
As a onsequen e, manufa turing osts are high (also due to manufa ture
toler-an es),and there may alsobe problems involvingin orre t shape denition and
oupling. This kind of design is improved by En inar et al. in [15℄,[16℄,[37℄,[61℄
that propose to design shaped beams by using more layers (2 or 3) of dierent
shaped pat hes and exploiting an optimizationte hnique in order to dene the
bestphasedistributionondierentlayers. Thiskindofdesign,basedonmultiple
layers, animprovetheperforman eandde reasethe omplexityofea hsingular
layer, although the overall stru ture is still omplex, expensive to manufa ture,
and it ould behard to inserta ontrolnetwork for beam-steeringappli ations.
In general, ree tarray layouts are usually synthesized by a two-step
pro e-dure inwhi h:
omputed;
(b) the feed and ree ting elements (e.g., mi rostrip pat hes) able to
approxi-mately generate su h urrentsare dedu ed/designed.
Several methods have been developed in the literature to solve (b) for various
unit ell geometries and ar hite tures [2℄,[6 ℄,[15℄,[59℄,[62℄-[66℄. On the ontrary,
veryfew approa hes have been proposed toaddress (a) [37℄-[39℄.
Oneexampleofsolutioninliteraturethatdealswithstep (b)is[59℄,inwhi h
theauthorsdes ribeamethodof omputingthephaseresponse hara teristi sfor
asquarepat hmi rostripand thensynthesizingthe pat h distributiontoobtain
apen ilbeamin dierent ree tarray ongurations: squared ree tarray oset
beam having the feeder in broadside, ir ular diameter ree tarray with both
feeder and far-eld maximum inbroadside, square ree tarray withprime fo us
re tangularhorn and square ree tarray in Cassegrian onguration.
Instead [15℄ [Fig. 1.3(a)℄, in order to enlarge the operational bandwidth of
the system, a multi-layer stru ture is employed. In parti ular the number of
layer isset to 2 and a simple square pat h is sele ted to have the desired phase
shift. More in detail, the size of the side of the square pat h an vary the phase
response of the spe i ell and by xing the ratio between the ell in the two
layers(theupperlayerpat hesare
0.7
timesthelowerlayerones)theree tarray issynthesizedusingthesimplephasedelay ompensation(thephaseofthepat hhas to ompensatethe same travelling time that shouldbe o urred in ase ofa
ree tor) and good performan e are obtained within
16.67%
of the operational bandwidth.When ree tarray pat hes are designed, a problem that an o ur is that
their phase response doesnot over the full
360
◦
phase range. Toover ome this
problem[64℄ [Fig. 1.3(b)℄propose to use a kind of stru ture that is y li . This
kind of element omes ba k to the original geometry shape when a full phase
rangeis overed. Infa tthe proposed elementisaphoenixelement (i.e. alled
phoenix for its rebirthing apabilities) that is omposed by a entered square
pat h of xed dimension, an external ring of xed dimension that delimits the
element with the adja ent ones and a variable ring that an move from the
inner tothe outer. Furthermore,this elementis designed to be metal-only,thus
thee ien yoftheree tarraysin eit an overthefullphaserangeanditdoes
not require any diele tri substrate. Nonetheless, the onne ting strips exhibits
some drawba ks in ontrolling the ross-polarization.
Tobetter ontrolthe ross polarization,butmaintainingthe fullrange over,
it is proposed by [65℄ [Fig. 1.3( )℄ to use two dierent y les to dene the
element. Firstly the element is made by dipole rossed with same arms (to
ensuredual-polarization),whosewidth ishalfof their length. To implementthe
y le, the length is in reased until the element tou hes the adja ent ones, then
the elementgeometry hangesand be omes agrid. The se ondstep of the y le
is done keeping xed the length and vary only the width of the arms unless the
metalizationdisappear, then the y le restart asa rossed dipole. The designed
ree tarray using this elements an handle both polarizations and demonstrate
tohave anoperationalbandwidth of
11.1%
.Ree tarrayare used alsofornon-mi rowave appli ationmovingtothe
tera-hertzdomain[66℄[Fig. 1.3(d)℄. Inthisdomainstubsormany ells withdierent
shapes an not be manufa tured (or are too expensive). Thus, it is proposed
touse metalblo ks with dierent height inorder to ompensate the phase with
respe t toareferen e plane. Inthis way itisthesameasifthe physi albehavior
of a ree tor is obtained by sampling and then applying a modulus operation
with respe t to the wavelength atthe heights of the blo ks. It is demonstrated
thatwith thisapproa hagoodgain an bea hieved and thepatternbehavioris
quitestabletothefrequen yband(
30%
)obtainingalsoaverygoodperforman e inantenna e ien y due tothe absen e of diele tri s.Consideringstep(a),theexploitationoflo aloptimizationstrategies(su has
theInterse tionApproa h [37℄,[39℄)has been proposed asarst stepofashaped
beam ree tarray synthesis [37℄,[39℄. However, su h methodologies an be
om-putationallyexpensive(espe iallyif wideapertures are athand)and their
ee -tiveness and onvergen e rate stronglydepend onthe hoi e of theinitialization
point [37℄. Alternatively, ray-tra ing te hniques have been proposed to dedu e
theree tarray surfa e urrentsstartingfromtheknowledgeofapreviously
syn-thesizedshaped ree toraordingthedesiredbeam pattern[38℄. Unfortunately,
su ha strategy doesnot allowthe designer to spe ify any feasibility onstraints
onthe solution (e.g., presen e of forbiddenregions inthe array aperture) and
"Square" Patch Dual Layer
g
(1)
g
(2)
"Square" Patch Dual Layer
g
(1)g
(2)(a) (b)
( ) (d)
Figure 1.3: Geometry of the ree tarray antenna elements: (a)[15℄, (b)[64℄,
( )[65℄, (d)[66℄.
In the Interse tion Approa h [67℄ two sets are onsidered: the rst set is
omposedbyalltheradiationpatternthatrespe tstherequiredspe i ationand
the se ond ontains all the radiation pattern that the ree tarray an radiate.
Roughly,thesynthesispro eduremakes ontinueproje tionofthepatternsinthe
twosetsfromtherstsettothese ond,untilthemismat hbetweentheproje ted
patterns is almost null. Thus, as the dimension of the ree tarray in reases the
dimension of the sets in reases as well, and this is one of the drawba ks of
te hniques explained in [37℄,[39℄. While [37℄ has a ree tarray made with three
layerof squared pat hes and an a hieve verygoodperforman e in overing the
SouthAmeri aregionwithabandwidthof
10%
,and[39℄(thathas alsoused the FFT toin reasee ien y oftheapproa h) ana hievegoodperforman ewitharee tarray made ofbla k boxes(itdoes not takeintoa ountthe real element,
onlyitsree tion oe ient)synthesizinganisoux patternand ashaped-beam
for the Europe overing with aDire t Broad astSatellite (DBS).
Froma dierentperspe tive,itisknown thatthe relationbetween the
ree -tarray urrentsand theirradiatedpatterns an beee tivelymodeledexploiting
and the unknowns are the surfa e urrents.
In [69℄ and [70℄ the problem addressed is toretrieve the urrent distribution
that radiates a measured eld. In the rst one it is minimized the distan e of
re onstru ted Equivalent Magneti Current (EMC) by the near-eld measured
ina ylindri alway usingtheMarquardtalgorithm,and inthe se onditsalmost
the same but taking into a ount a near-eld measured on a spheri al surfa e
(hen ethree omponents of the eld, instead of onlytwo).
In [71℄ the problem is to re onstru t equivalent urrents distribution using
integral equationalgorithm. Usingthe integralequation the authorsare ableto
re onstru tthe urrentoveruser-dened surfa es, notonly ylindri alor
spheri- alsurfa e (thatare easierto omputeusingthe tangentialeldsandthe
Equiv-alen e Prin iple)but also, for example,on the surfa e of a horn antenna.
In[72℄ metalli bodies are re onstru tedasequivalent urrents. Inparti ular
the Sour e Re onstru tion method is applied to the retrievalof metal obje t in
an investigation domain and use a minimization (using a Conjugate Gradient
method)ofa ostfun tionthat,takingintoa ounttheTikhonov regularization
and the normalization of the equations terms, of the
L
2
-norm of the measured and re onstru ted eld (by the radiationof the equivalent urrent).Intheframeworkofinverses atteringandantennadiagnosis/ hara terization
[69℄-[72℄,su haproblemisknowntobeill-posedbe auseofthenon-uniquenessof
theradiationoperator[73℄,whi hisrelatedtotheexisten eofnon-me
asurable/non-radiating urrents [74℄-[76℄. While this feature an be an issue in traditional
inverse problems requiring suitable ountermeasures [74℄-[76℄, it a tually
repre-sents a degree-of-freedom (DoF) in the framework of onstrained ree tarray
design. In fa t,bysuperimposingasuitably designednon-measurable urrentto
anavailable(e.g., minimum-norm[74℄-[77℄)solutionoftheIS problem,a urrent
ould be synthesized whi h radiates the desired far-eld pattern, and omplies
with the user-dened onstraints.
A ordingtosu h onsiderations,aninnovativeparadigmtosynthesize
ree -tarray surfa e urrents[i.e.,toaddressstep (a)℄isproposedwhi h,byleveraging
on the non-uniqueness of the IS problem as a design DoF, enables to dedu e
solutionsalsosatisfyinguser-dened antenna feasibility onstraints (e.g.,on the
presen e and shape of forbiddenregions in the aperture). To this end, the
the IS problemisrstly derived, and then(ii)asuitablenon-measurable sour e
is omputedsothattheresultingsurfa e urrent[i.e.,thesuperpositionofthe
so-lutions(i) and (ii)℄ omplieswith the user-dened requirements. Thanks tothe
features of the proposed formulation, a losed-formexpression is nally derived
for both the minimum-norm and the non-measurable urrents whi h does not
require any iterative lo al/global optimization pro edure and whi h inherently
satisesboth the radiationand the feasibilitydesign onstraints.
Inparti ular,itisproposedtoapplytheSingularValueDe omposition(SVD)
to a dened Green's operator. The out ome of this pro ess are two set of
or-thonormal bases and a matrix of singular values. This output has to be
ana-lyzed inorder to nd agoodtrade-o between, onone side, the pre isionof the
minimum-norm urrent able to radiate the desired eld; on the other side, the
possibility to have the greatest number of non-measurable bases. This analysis
it is done by dening a variable threshold on the value of the singular values
and olle tingdierent ombinationof orthonormalbases that are linked to the
singularvaluesabove or belowthe threshold.
The innovative methodologi al ontributions of the paper therefore in lude
theintrodu tion,forthersttimetothebestoftheauthorknowledge,ofa
ree -tarray surfa e urrentsynthesisparadigmwhi hleveragesonthenon-uniqueness
of the IS problemand the existen e of non-measurable urrents to improve the
features of the obtained solution (e.g., in terms of feasibility), and the
intro-du tion of expli it losed-form expressions for the omputation of ree tarray
surfa e urrentsaordingadesiredfar-eldpatternand omplyingwith
The thesis isorganized asfollows. After the formulationof the shaped-beam
onstrained ree tarray urrents synthesis problem (Chapter 2), the proposed
designmethodisillustratedanditsnal losed-formsolutionisderived (Chapter
3). A set of numeri al examplesbased on realisti ree tarray ar hite tures are
then illustrated to assess the ee tiveness and potentialities of the onsidered
design paradigm (Chapter 4). At the end are presented the on lusion and
Problem Formulation
In this hapter the problem formulation is explained throgh the understanding
of the radiation problem in a ree tarray antenna and how to formulate it as
an Inverse Sour e problem. In parti ular, rstly it is des ribed the radiation
problemformulationanditsdis retizationontheree tarraysurfa e(thatisnot
ontinuous) and then the problem to nd the urrent that generate a spe ied
radiatedeld is formulated asanInverse Sour e problem.
2.1 Radiation from surfa e urrent
We onsider a ree tarray antenna, oriented like in Fig. 2.1, with both ground
plane and pat hes made by a Perfe t Ele tri Condu tor separated by a layer
of substrate withstandard omplex permittivityvalue
ε = ε
0
ε
r
(1 − j tan δ)
and illuminatedby a feeder positionedinr
f
= (x
f
, y
f
, z
f
)
that infar-eld generates a plane-wave that has a relative angular position(θ
inc
, φ
inc
)
(see Fig. 2.1). The in ident ve tor for ea h ell of the ree tarray isν
inc
(r) =− (sin θ
inc
cos φ
inc
,
sin θ
inc
sin φ
inc
, cos θ
inc
)
.The in ident plane wave ona ree tarray element an bemodelas:
"
E
θ
inc
E
inc
φ
#
=
"
E
θ
0
E
0
φ
#
e
−jk
(
ν
inc
(r)·r
)
(2.1) whereE
0
isthe ve torthat des ribes amplitudeand polarizationofthe in ident plane-wave,r
is the position of the ree tarray element,k = 2πf √µε
,µ
,ε
arex
z
y
U
T
I
y
'
x
'
:
5HIOHFWDUUD\
6XUIDFH
)HHG
ܚ
ଵଵ
ܚ
ܚ
ெே
Figure2.1: Geometry of the ree tarray antenna.
the free-spa e wave number, permeability, and permittivity, respe tively, and
f
isthe frequen y.The presen e of the groundeddiele tri slab and ofthe layerprinted pat hes
generates dierent kind of eld that are ba k-radiated. The total eld that is
present in the region of the spa e in front of the ree tarray antennas an be
des ribed asthe sum of these ontributes:
E
tot
= E
inc
+ E
RGDS
+ E
RP P
(2.2)The term
E
RGDS
indi atesthe ree tedeld by the innitegrounded diele -tri slab without any kindof pat hprinted on, and an be deniteas:E
RGDS
= RE
0
e
jk(x sin θ
inc
cos φ
inc
+y sin θ
inc
sin φ
inc
−z cos θ
inc
)
(2.3)where matrix
R
is the diagonal ree tion matrix, and its non-null entriesR
θθ
andR
φφ
are dened asin [79℄.The other term
E
RP P
represents the ree ted eld when the mi rostrip pat hes are present. On this pat hes, made of PEC, theE
inc
indu es asur-fa e urrentthat radiates aeld dened as:
E
RP P
= SE
0
e
jk(x sin θ
inc
cos φ
inc
+y sin θ
inc
sin φ
inc
−z cos θ
inc
)
(2.4)
where
S
isthe s attering matrix and its oe ients hara terize the ree tion:S
=
"
S
θθ
S
θφ
S
φθ
S
θθ
#
(2.5)Ea h s attering oe ient is dened as the ratio between the s attered and
in ident eld of the mi rostrip surfa e for ea hpolarization:
S
ji
=
E
RP P
j
(z = 0)
E
i
inc
(z = 0)
j, i = {θ, φ}
(2.6)These oe ients an be omputed for ea h mi rostrip pat h and then used
to obtain the surfa e urrent on the ree tarray aperture. Sin e we want to
dened the urrenton the mi rostrip surfa e
J
s
:J
s
= ˆ
n × H
(2.7)(where
n
ˆ
isthe normaltothe surfa e)and wehavedened allthe termsin(2.2) we an express alsothe total magneti eld as:H =
1
η
ν
inc
× E
inc
+
1
η
ν
ref l
× E
RGDS
+
1
η
ν
ref l
× E
RP P
(2.8) whereν
ref l
is the spe ular ree tion dire tion (Snell's law on a PEC) of the
in iden e dire tion
ν
inc
, and
η
is the free-spa e impedan e.In ase of far-eld, following [80℄, the radiatedeld by an ele tri urrent
J
s
an be approximates as:
E
RAD
(r) ≈ −jη
exp (−jkr)
2λr
(N
θ
(θ, φ) + N
φ
(θ, φ))
(2.9) where the radiationve torin arthesian oordinatesN
an be expressed asN (θ, φ) =
R R
Ω
J
x
(r) ˆ
x + J
y
(r) ˆ
y
×
exp j
2π
λ
(x sin θ cos φ + y sin θ sin φ)
dx dy
and
λ
is the wavelength.In order to be easier omputed the radiation ve tors an be expressed in
spheri al oordinates form:
(
N
θ
(θ, φ) = N
x
(θ, φ) cos θ cos φ + N
y
(θ, φ) cos θ sin φ
N
φ
(θ, φ) = −N
x
(θ, φ) sin φ + N
y
(θ, φ) cos φ
(2.11)
In this way, bu substituting(2.11) in (2.9) we an obtain:
(
E
RAD,θ
(r) =
−jη
exp(−jkr)
2λr
(N
x
(θ, φ) cos θ cos φ + N
y
(θ, φ) cos θ sin φ)
E
RAD,φ
(r) =
−jη
exp(−jkr)
2λr
(−N
x
(θ, φ) sin φ + N
y
(θ, φ) cos φ)
(2.12)
The equations(2.12)and (2.10) ompletelydes ribethe far-eldradiationof
an indu ed urrent from a feeder on the ree tarray surfa e. In order to make
possible the utilizationof these equation and the problemdenition we need to
2.2 Dis retization of the ree tarray surfa e
Lets assume to dis retize the ree tarray surfa e in a regular latti e, and
ea h ell of the latti e in ludes only one ree tarray element. Thus, the
ree -tarray onsists of agrid of
M × N
elements withunit ells of size∆x × ∆y
and in order to dis retize (2.10) we will apply a pixel-basis fun tionP
entered atr
mn
,
m −
M
2
∆x, n −
N
2
∆y, 0
,m = 1, ..., M
,n = 1, ..., N
tothe urrent distributionand we obtain:J
q
(r)
,
M
X
m=1
N
X
n=1
J
mn
q
P
mn
(r)
q ∈ {x, y}
(2.13)In parti ular, the x- omponent of the urrent an beexpressed as:
J
x
|
mn
= −
ν
inc
x
(r
mn
) sin(θ)
η
(1 + S
θθ
(r
mn
)) E
inc
θ
(r
mn
) +
+S
θφ
(r
mn
) E
inc
φ
(r
mn
)
i
+
−
ν
inc
z
(r
mn
) cos(θ) cos(φ)
η
(1 − S
θθ
(r
mn
)) E
inc
θ
(r
mn
) +
−S
θφ
(r
mn
) E
inc
φ
(r
mn
)
i
+
+
ν
inc
z
(r
mn
) sin(φ)
η
(1 − S
φθ
(r
mn
)) E
inc
θ
(r
mn
) +
−S
φφ
(r
mn
) E
inc
φ
(r
mn
)
i
(2.14)and the y- omponent as:
J
y
|
mn
= −
ν
inc
y
(r
mn
) sin(θ)
η
(1 + S
θθ
(r
mn
)) E
inc
θ
(r
mn
) +
+S
θφ
(r
mn
) E
inc
φ
(r
mn
)
i
+
+
ν
z
inc
(r
mn
) cos(θ) sin(φ)
η
(1 − S
θθ
(r
mn
)) E
inc
θ
(r
mn
) +
−S
θφ
(r
mn
) E
φ
inc
(r
mn
)
i
+
+
ν
z
inc
(r
mn
) cos(φ)
η
(1 − S
φθ
(r
mn
)) E
inc
θ
(r
mn
) +
−S
φφ
(r
mn
) E
inc
φ
(r
mn
)
i
(2.15)where the ree tion matrix of the ground plane is substituted by
1
sin e it is perfe tly ree ting, while the omponent z is obviously null (J
z
|
mn
= 0
,m =
Substituting (2.13)in (2.10):
N
q
(θ, ϕ) =
P
M
m=1
P
N
n=1
exp [jkˆ
rr
mn
] J
q
mn
×
R
Ω
mn
P
mn
(r) exp
jkˆ
rr
Ω
mn
dΩ
mn
q ∈ {x, y}
(2.16)where
Ω
mn
is the area of themn
-th re tangular pixel. Due to the presen e of the pixel-basisfun tion the integralin the radiationve tor formulabe omes:Z
x
n
+
∆x
2
x
n
−
∆x
2
Z
y
m
+
∆y
2
y
m
−
∆y
2
exp
j
2π
λ
x sin θ cos φ
exp
j
2π
λ
y sin θ sin φ
dx dy
(2.17)where
(x
n
, y
m
)
isthe enter of themn
-three tarray element/re tangularpixel. Moreover, with some simple step an be proven that this integral an be solvedas:
exp
j
2π
λ
x sin θ cos φ
exp
j
2π
λ
y sin θ sin φ
×
4
∆x∆y
sinc
k∆x
2
sin θ cos φ
sinc
k∆y
2
sin θ sin φ
(2.18)
Nowsubstituting the integral solution(2.18) in (2.16)we nallyhave:
N
q
(θ, ϕ) =
∆x∆y
4
sinc
k∆x
2
sin θ cos φ
sinc
k∆y
2
sin θ sin φ
×
P
M
m=1
P
N
n=1
J
q
mn
exp [jk (x
n
sin θ cos φ + y
m
sin θ sin φ)]
(2.19)
Thisnalequationisimportantbe auseitdes ribes, inadis retizedway,the
radiationve torsand thuswe an ompute thefar-eldhaving anon ontinuous
urrent denition.
Theusualway todes ribethe far-eldpatterninree tarrayantennasystem
isusing the thirdLudwig denition [78℄[6℄[81℄ of the oordinatesystem.
The far eld radiated by a ree tarray displa ed on a surfa e
Ω
(Fig. 2.1) an be modeled as[68℄E
(r) ≈
jµf
2
exp (−jkr)
r
[F
CO
(θ, ϕ)
p
b
CO
+ F
CX
(θ, ϕ)
b
p
CX
]
(2.20) wherer = |r|
,r
= (r sin θ cos ϕ, r sin θ sin ϕ, r cos θ)
is the position ve tor, and the o-polarb
p
CO
and ross-polarp
b
CX
unitve torsagreewiththethirddenition of Ludwig[78℄[6℄[81℄(Fig. 2.2)(
b
p
CO
= cos (ϕ) b
θ
− sin (ϕ) b
ϕ
b
p
CX
= sin (ϕ) b
θ
+ cos (ϕ)
ϕ
b
(2.21)Figure 2.2: Co-polar and ross-polar unit ve tor following the Ludwig third
denition.
with some manipulation the o-polar and ross-polar pattern an be retrieved
from(2.20) and (2.12):
F
CO
(θ, ϕ) =
{1 + cos
2
(ϕ) [1 − cos (θ)]} N
x
(θ, ϕ) +
+ [cos (θ) − 1] sin (ϕ) cos (ϕ) N
y
(θ, ϕ) +
− sin (θ) cos (ϕ) N
z
(θ, ϕ)
(2.22)
F
CX
(θ, ϕ) =
[cos (θ) − 1] sin (ϕ) cos (ϕ) N
x
(θ, ϕ) +
+
1 + sin
2
(ϕ) [1 − cos (θ)]
N
y
(θ, ϕ) +
− sin (θ) sin (ϕ) N
z
(θ, ϕ)
(2.23)
At this point we have dened how to ompute the referen e pattern in the
standard omponents we an move to the synthesis problem denition. Sin e
therearemanyworksintheliteraturethatdealwiththeproblemtondaproper
te hnology (e.g. printed pat hes type, number of layers, all-metal stru tures,
et ...)[2℄,[6℄,[15℄,[59℄,[62℄-[66℄ toobtainthe wanted value ofthe s attering matrix
S
for a given urrent distribution, now we do not take this step into a ount (step (b))and wewill ontinue handlingonlythe problemrelated tothe surfa e2.3 Inverse Sour e problem denition
A ording to the previous formulation, the synthesis of the onstrained
sur-fa e urrents of a ree tarray withdesired far-eld shaped-beam pattern
E
ref
(r)
,
jµf
2
exp (−jkr)
r
h
F
CO
ref
(θ, ϕ)
p
b
CO
+ F
CX
ref
(θ, ϕ)
p
b
CX
i
(2.24)an be formulated asan inverse sour e problem.
The inverse sour e problemis dened as:
Constraint-Geometry Ree tarray-Currents Synthesis problem
(CG-RCS)and its denition is:
Find the surfa e urrent
J (r)
(or its numeri al ounterpartJ
q
,
J
mn
q
; m = 1, ..., M
,n = 1, ..., N}
,q ∈ {x, y}
), whose radiatea far-eld,E
,whoseasso iated o-polarand ross-polar omponent,tthe following referen epattern mat hing ondition:(
F
CO
ref
(θ, ϕ) = F
CO
(θ, ϕ)
F
CX
ref
(θ, ϕ) = F
CX
(θ, ϕ)
(2.25)
having that
J
q
∈ S
q
,q ∈ {x, y}
.WhereS
q
,q ∈ {x, y}
, are the fea-sibility sets, whi h a ount for the onstraints provided by theend-user/designer(i.e.,thepresen eofforbidden regions intheaperture).
For example, if
Φ
identies the arbitrary-shaped user-dened 2-D forbidden re-gion (withintheree tarrayΦ ∈ Ω
),thefeasibility onditionstatethatJ (r) = 0
ifr ∈ Φ
(i.e.: in numeri al form:J
mn
q
= 0
,q = {x, y}
, ifr
mn
∈ Φ
). It is worth
remarking that many te hniques an be adopted for the synthesis of feed and
asso iated ree tarray elements (step (b)) (depending on the sele ted unit- ell
geometry[2℄,[6℄,[15℄,[59℄,[62℄-[66℄)on e
J
q
,q ∈ {x, y}
, has been found by solving the above problem.Non-Measurable Currents-based
Solution Method
In this hapter the solution method, that solve the Inverse Sour e problem, is
explained. Inparti ular,aftersomemathemati al omputationneeded toobtain
a matrix formulation of the problem, it is applied a Trun ated Singular Value
De omposition (T-SVD) in order to obtain the minimum-norm solution. This
solution an radiate the desired eld but an not deal withfeasibility onstraint
(e.g. forbidden region). In order to over ome this problem, it is proposed to
superimpose the non-radiating/non-measurable urrents, that are derived from
theT-SVD,totheminimum-norm solution. Moreover,giventhedenitionofthe
handledfeasibility onstraint (i.e. forbidden region),the losed-formformulation
3.1 Field dis retization
In order to address the CG-RCS problem, taking in onsideration that the
z- omponentofthe urrentisnotpresent,itsdis retizedversionisrstly omputed
bysubstituting(2.22)and(2.23),in(2.25),samplingitinasetof
L
angles(θ
l
, ϕ
l
)
,l = 1, ..., L
as follows
F
CO
ref
(θ
l
, ϕ
l
) = {1 + cos
2
(ϕ
l
) [1 − cos (θ
l
)]} N
x
(θ
l
, ϕ
l
) +
+ [cos (θ
l
) − 1] sin (ϕ
l
) cos (ϕ
l
) N
y
(θ
l
, ϕ
l
)
F
CX
ref
(θ
l
, ϕ
l
) = [cos (θ
l
) − 1] sin (ϕ
l
) cos (ϕ
l
) N
x
(θ
l
, ϕ
l
) +
+
1 + sin
2
(ϕ
l
) [1 − cos (θ
l
)]
N
y
(θ
l
, ϕ
l
)
l = 1, ..., L
(3.1)
whi h, by exploiting (2.19), an berewritten as
F
CO
ref
(θ
l
, ϕ
l
) = Γ (θ
l
, ϕ
l
) ({1 + cos
2
(ϕ
l
) [1 − cos (θ
l
)]}
×
P
M
m=1
P
N
n=1
J
mn
x
e
mn
(θ
l
, ϕ
l
) +
+ [cos (θ
l
) − 1] sin (ϕ
l
) cos (ϕ
l
)
×
P
M
m=1
P
N
n=1
J
y
mn
e
mn
(θ
l
, ϕ
l
)
F
CO
ref
(θ
l
, ϕ
l
) = Γ (θ
l
, ϕ
l
) ([cos (θ
l
) − 1] sin (ϕ
l
) cos (ϕ
l
)
×
P
M
m=1
P
N
n=1
J
mn
x
e
mn
(θ
l
, ϕ
l
) +
+
1 + sin
2
(ϕ
l
) [1 − cos (θ
l
)]
×
P
M
m=1
P
N
n=1
J
y
mn
e
mn
(θ
l
, ϕ
l
)
l = 1, ..., L
(3.2)where,for easy of ompa tness:
e
mn
(θ
l
, ϕ
l
)
, exp [jk
0
(m∆x sin θ
l
cos ϕ
l
+ n∆y sin θ
l
sin ϕ
l
)]
(3.3)and
Γ (θ
l
, ϕ
l
) =
4
∆x∆y
sinc
k∆x
2
sin θ cos φ
sinc
k∆y
2
sin θ sin φ
(3.4)In order to further simplify the notation and to better handle the problem
we need toexpress the equation in matrix form.
following matrix equation
F
ref
= GJ
(3.5) whereF
ref
,
h
F
ref
CO
, F
ref
CX
i
T
,F
ref
t
=
n
F
t
ref
(θ
l
, ϕ
l
) , l = 1, ..., L
o
,t ∈ {CO, CX}
,J
, [J
x
, J
y
]
T
, and:G ,
"
G
CO,x
G
CO,y
G
CX,x
G
CX,y
#
(3.6)is the
(2 × L) × (2 × P )
overall Green matrix (·
T
being the transpose operator)
featuringthe sub-matri es:
G
CO,x
, {Γ (θ
l
, ϕ
l
) e
mn
(θ
l
, ϕ
l
) {1 + cos
2
(ϕ
l
) [1 − cos (θ
l
)]}
G
CO,y
= G
CX,x
, {Γ (θ
l
, ϕ
l
) e
mn
(θ
l
, ϕ
l
) [cos (θ
l
) − 1] sin (ϕ
l
) cos (ϕ
l
)
G
CX,y
, {Γ (θ
l
, ϕ
l
) e
mn
(θ
l
, ϕ
l
)
1 + sin
2
(ϕ
l
) [1 − cos (θ
l
)]
m = 1, ..., M, n = 1, ..., N, l = 1, ..., L
(3.7)
where
Γ
was dened in (3.4) ande
mn
in(3.3).
Now we have all the formulation ready for start ta king the Inverse Sour e
problem.
The problem to retrieve a urrent distribution from a eld is well-known to
beill-posed. Thismeansthat multiple urrent distribution anradiatethe same
eld. Inthe literature, oneof the mostused tooltoobtaina minimum-norm(or
generalized)solution of the system is the regularizationand inversion te hnique
Trun ated Singular Value De omposition. Using this algorithm the obtained
solution is the ones that best represents the radiated eld with the smallest
3.2 Trun ated Singular Value De omposition
Inorderto omputetheminimumnormsolutioniswell-knownthe pro edure
based onthe trun ated version of the SingularValue De omposition
(SVD)[73℄-[77℄.
Weassumethatthenumberofthe ree tarrayelements(
2 ×P = 2×M ×N
) is less than the number of eld samples (2 × L
). Given thatψ
2
1
, ψ
2
2
, ..., ψ
2
W
(
W
, min {2 × L, 2 × P }
)arethepositiveeigenvalues ofsymmetri matrixG
∗
G
(where
∗
indi atethe onjugatetranspose)and
c
1
, c
2
, ..., c
2×P
the orresponding orthonormaleigenve tors:G
∗
Gc
j
= ψ
j
2
c
j
c
∗
j
c
k
= δ
jk
j, k = 1, ..., 2 × L
(3.8) being:ψ
j
b
j
= Gc
j
j = 1, ..., 2 × L
(3.9)substituting (3.8)in (3.9) it an be obtained:
G
∗
b
j
= ψ
j
c
j
j = 1, ..., 2 × L
(3.10)multiplying left and right side of (3.10) for
G
and thanks to (3.9) immediately follow that:GG
∗
b
j
= ψ
j
2
b
∗
j
b
j
= δ
jk
j, k = 1, ..., 2 × L
(3.11)
Equations(3.10)and(3.11)showntheorthonormalpropertiesofthetwomatri es
B
andC
.In matrix notation(3.9) an bewritten as:
an be noti edthat
Ψ
isa diagonalmatrix in the form:Ψ =
ψ
1
0
. . . . . .0
ψ
W
W
×W
(3.13)another note is that the singular values are ordered in des ending order (i.e.,
ψ
w
≥ ψ
w+1
,w = 1, ..., W − 1
).Writing nowthe minimum-norm urrent as aweighted sum as:
J
MN
= Cγ
with
γ
=
γ
1
. . . . . .γ
W
(3.14)from(3.9) we an obtain that:
γ
j
ψ
j
b
j
= γ
j
Gc
j
j = 1, . . . , W
(3.15)and using the ration expressed in(3.5):
F
ref
=
W
X
j=1
γ
j
ψ
j
b
j
(3.16)Sin e
ψ
i
b
j
= 0
forj > W
, itis simply to derive that only the rstW c
w
bases that are used to des ribeJ
MN
are measurable.
For these bases the oe ients
γ
j
are given using:γ
j
= ψ
−1
j
b
∗
j
F
ref
j = 1, ..., W
(3.17) and substituting (3.17)in (3.14):J
MN
=
W
X
j=1
ψ
−1
j
b
∗
j
F
ref
c
j
(3.18)Dually inmatrix form (3.18) be ame:
J
MN
= CΨ
−1
B
∗
F
ref
(3.19)This resulting problem is well-known [74℄-[77℄ to be not well-posed due to the
ill- onditioningof the
G
matrix.The solution instability o urs due to the fa t that some singular values
ψ
are mu h lower in magnitude with respe t to the rst one (ψ
1
). This problem an be measured by using the ondition number that is dened as:d =
ψ
1
ψ
W
(3.20)
This value measures the instability of the problem. In fa t as higher is the
value, as higher is the instability, and this means that the a small variation in
the
F
ref
generate a greatvariationin
J
MN
.
In literature this problem iswell-known [74℄-[77℄ and the solution isto use a
trun ated version of the SVD.
It is dened
H
as the trun ation order, and is omputed as:H
, arg
min
w
ψ
w
ψ
1
− τ
s.t.
ψ
w
ψ
1
≥ τ
(3.21)where
τ
beingtheasso iateduser-denedSVD trun ationthreshold. The thresh-oldτ
, the trun ation orderH
and an example of singular value behaviorψ
are shown in Fig. 3.1. By sele ting the value of the SVD trun ation thresholdτ
theuser impli itlydenes the pre isiononthe reprodu tionthe far-eldand theinstability of the urrent, thus lower value of the threshold means better
repro-du tion of the far-eld but also higher variation in the solution (e.g. urrent
distribution withhigh spa evariations).
Then,are omputedthetrun atedversionofmatri es
C
,B
andΨ
bysele ting the rstH
bases of the orresponding sets:B
τ
= {b
h
, h = 1, ..., H}
C
τ
= {c
h
, h = 1, ..., H}
Ψ
τ
= diag (ψ
h
, h = 1, ..., H)
(3.22)
^ŝ
ŶŐ
Ƶů
Ăƌ
s
Ăů
ƵĞ
ƐŶ
Žƌ
ŵ
Ăů
ŝnjĞ
Ě
;ɗ
௪
Ȁɗ
ଵ
Ϳ
Ě
^ŝŶŐƵůĂƌ ǀĂůƵĞ ŝŶĚĞdž͕ǁ
Ϭ
ͲŝŶĨ
ʏ
,
Figure3.1: Exampleofsingularvalues distribution
ψ
w
,w = 1, ..., W
,takinginto a ount atrun ation orderH
and a trun ationthresholdτ
.as:
J
MN
, C
τ
Ψ
−1
τ
B
∗
τ
F
ref
(3.23) This kind of solution (minimum-norm) an be a hieved also in other ways(not only T-SVD), however the trun ation operation on the SVD give us a set
of bases that willradiate anulleld outsideof the supportand thus, ea h basis
an beinterpreted asadierent urrentwith dierentshapethat donotradiate