International Do toral S hool in Information
and Communi ation Te hnology
DISI- University of Trento
Novel Design Con epts for Un onventional
Antenna Array Ar hite tures in Next
Generation Communi ations Systems
Giorgio Gottardi
Advisor:
PaoloRo a, Asso iate Professor
University of Trento
In this work, the formulationand the implementationof innovative
method-ologi alparadigmsforthedesignofun onventional arrayar hite turesforfuture
generation ommuni ation systems has been addressed. By exploiting the
po-tentialities of the odesign strategy for elementary radiators in an irregularly
lustered array ar hite tures and by introdu ing an innovative apa ity-driven
design paradigm, the proposed methodologiesallow toee tively design
un on-ventional arrayar hite tureswithoptimaltrade-osintermsofperforman eand
omplexity/ osts. The odesignsynthesisstrategyisproposedtosolvethearising
massive multi-obje tivedesign problem aimed at tting the multiple obje tives
and requirementson the free-spa e performan e of the array ar hite ture.
Af-terward, the apa ity-drivendesignparadigmisformulatedandimplementedfor
thedesignofMIMO arrayar hite turestomaximizethequalityofthe
ommuni- ationsystem inrstpla einsteadof onsideringfree-spa e gures-of-merit. A
setofnumeri alresultshasbeenprovided(i)tovalidatetheproposedparadigms
inreal-appli ations enarios and (ii)toprovide insightsonthe ee tiveness, the
limitationsand the potentialities ofproposed design methodologies.
Keywords
5G ommuni ations,5Gradiatingsystems,integer- odedgeneti algorithm(GA),
[R1℄ G. Gottardi, L. Poli, P. Ro a, A. Montanari, A. Aprile, and A.
Massa, Optimal monopulse beamforming for side-looking airborne
radars, IEEE Antennas Wireless Propag. Lett., vol. 16, pp.
1221-1224,2017.
[R2℄ G. Oliveri, G. Gottardi, F. Robol, A. Polo, L. Poli, M. Salu i, M.
Chuan, C. Massagrande, P. Vinetti, M. Mattivi, R. Lombardi, and
A. Massa, Co-design of un onventional array ar hite tures and
an-tennaelementsfor5Gbasestations, IEEETrans. AntennasPropag.
-Spe ialIssueon'AntennasandPropagationAspe tsof5G
Commu-ni ations,' vol. 65,no. 12,pp. 6752-6767, De ember 2017.
[R3℄ M.Salu i,G.Oliveri,N.Anselmi, G.Gottardi,andA. Massa,
Per-forman e enhan ement of linear a tive ele troni ally-s annedarrays
by means of MbD-synthesized metalenses, Journal of
Ele tromag-neti Waves and Appli ations, vol. 32, no. 8, pp. 927-955,2018.
[R4℄ 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,
"Computational methods for wireless stru tural health monitoring
of ultural heritages," Journal of Physi s: Conferen e Series, vol.
1131,pp. 1-7, 2018.
[R5℄ A. Massa, G. Gottardi, and E. Rajo-Iglesias, "CS-Based
omputa-tionalmethodsfor inverse problems arising inarrays pro essing and
design," Journal of Physi s: Conferen e Series, vol. 1131, pp. 1-7,
2018.
[R6℄ M. Salu i, L. Tenuti, G. Gottardi, A. Hannan, and A. Massa, A
System-by-Design method for e ient linear array miniaturization
throughlow- omplexityisotropi lenses, Ele troni sLetters,vol. 55,
no. 8,pp. 433-434, April2019.
[R7℄ G. Oliveri, G. Gottardi, and A. Massa, A new meta-paradigm for
the synthesis of antenna arrays for future wireless ommuni ations,
IEEE Trans. Antennas Propag., vol. 67, no. 6, pp. 3774-3788, June
2019.
[R8℄ N.Anselmi,G.Gottardi,G.Oliveri,andA.Massa,Atotal-variation
sparseness-promoting method for the synthesis of ontiguously
lus-tered linear ar hite tures, IEEE Trans. Antennas Propag., vol. 67,
row., Antennas Propag.,vol. 11,no. 13,pp. 1841-1845, O t. 2017.
[R10℄ A.Gelmini,G.Gottardi,andT. Moriyama,"A ompressive
sensing-based omputational method for the inversion of wide-band ground
penetratingradar data," Journal of Physi s: Conferen e Series, vol.
904, pp. 1-7, 2017.
[R11℄ G. Gottardi and T. Moriyama, "Indoor dete tion of passive targets
re ast as an inverse s attering problem," Journal of Physi s:
Con-feren e Series, vol. 904, pp. 1-7, 2017.
[R12℄ M.Salu iandG.Gottardi,"Advan esonmulti-s aleMbDsynthesis
ofWAIMs foradvan ed phasedarrays,"Journal of Physi s:
Confer-en e Series, vol. 963, pp. 1-4, 2018.
[R13℄ N. Anselmi and G. Gottardi, Re ent advan es and urrent trends
inmetamaterials-by-design, Journal of Physi s: Conferen e Series,
vol. 963, pp. 1-3, 2018.
[R14℄ P. Ro a, G. Gottardi, M. Bertolli,F. Robol, T. Moriyama, and T.
Takenaka,"Pro essingGPRdatawithinverses atteringapproa hes,"
Journal of Physi s: Conferen e Series, vol. 1131, pp. 1-7, 2018.
[R15℄ G.GottardiandL.Poli,"Human hestimagingbyreal-time
pro ess-ing of ele tri al impedan e data tomography," Journal of Physi s:
[C1℄ F. Viani,N.Anselmi, M. Donelli,P.Garofalo, G.Gottardi,G.
Oliv-eri, L.Poli,A. Polo, P. Ro a, M. Salu i,L. Tenuti, and A. Massa,
On the role of information in inversion and synthesis & hallenges,
tools,andtrends,"2015IEEEMediterraneanMi rowaveSymposium
(MMS'2015),Le e,Italy,pp. 1-4, November30-De ember2,2015.
[C2℄ L. Poli, P. Ro a, G. Gottardi, and A. Massa, Design of simplied
largearray stru tures for preliminaryexperimentalvalidation, 10th
European Conferen eonAntennasand Propagation(EUCAP2016),
Davos,Switzerland, pp. 1-4, April 11-15,2016.
[C3℄ L. Poli, N. Anselmi, G. Gottardi, P. Ro a, and A. Massa,
"Proba-bilisti interval method for phased array sensitivity analysis," 2016
IEEE AP-S International Symposium and USNC-URSI Radio
S i-en e Meeting, Fajardo, Puerto Ri o, pp. 919-920, June 26 - July 1,
2016.
[C4℄ G.Oliveri,E.Bekele,M.Salu i,L.Tenuti,G.Gottardi,T.Moriyama,
T. Takenaka, F. Bilotti, A. Tos ano,A. Massa, A system-by-design
approa h to the synthesis of mantle loaks for large diele tri
ylin-ders, PIERS 2016, Shangai, China, August 8-11, pp. 3144-3145,
2016.
[C5℄ A.Massa, N.Anselmi,G.Gottardi,G.Oliveri,L.Poli,P.Ro a,M.
Salu i,and L. Tenuti, Un onventional te hniques for the synthesis
of modern antenna arrays, 11th European Conferen e on
Anten-nas and Propagation (EUCAP 2017), Paris, Fran e, pp. 2843-2845,
Mar h 19-24, 2017.
[C6℄ G.Gottardi,N.Ebrahimi,P.Ro a,andA.Massa,Optimal
synthe-sisof monopulse beamforming weights for airborne radards through
onvexoptimization,2017InternationalAppliedComputational
Ele -tromagneti sSo iety Symposium, ACES 2017, Firenze, Italy, pp.
1-2,Mar h 26-30,2017.
[C7℄ N. Anselmi, P. Ro a, M. Salu i, G. Gottardi, and A. Massa, A
mask mat hing tiling optimization method for lustered phased
ar-rays, 2017 IEEE AP-S International Symposium and USNC-URSI
Radio S ien e Meeting, San Diego, California, USA, pp. 1045-1046,
July9-15, 2017.
[C8℄ G. Oliveri, P. Ro a, L. Poli, G. Gottardi, N. Anselmi, M. Salu i,
R. Lombardi, M. Chuan, M. Mattivi, P. Vinetti, F. Morgia, and
9-15, 2017.
[C9℄ N.Anselmi,P.Ro a,G.Gottardi,M.Salu i,andA.Massa,Tiling
optimizationoforthogonal-polygonshaped apertureforphasedarray
antennas, 2017 IEEE AP-S International Symposium and
USNC-URSIRadioS ien e Meeting,San Diego,California,USA,pp.
2025-2026, July 9-15, 2017.
[C10℄ L. Tenuti, P. Ro a, M. Salu i, G. Gottardi, and A. Massa,
In-novative optimization-based design of UWB planar arrays for
grat-inglobesredu tion, 2017IEEE AP-SInternational Symposiumand
USNC-URSIRadioS ien eMeeting,SanDiego,California,USA,pp.
2017-2018,July 9-15, 2017.
[C11℄ P.Ro a,N.Anselmi,M.Salu i,G.Gottardi,L.Poli,andA.Massa,
A novel analyti beam steering approa h for lustered phased
ar-ray ar hite tures, 2017 IEEE AP-S International Symposium and
USNC-URSI Radio S ien e Meeting, San Diego, California, USA,
pp. 2013-2014,July 9-15, 2017.
[C12℄ F. Robol, G. Gottardi, M. Salu i, G. Oliveri, and A. Massa,
De-signofnot h-enhan ed ompa tprintedantennasfor5G
ommuni a-tions, 2017 IEEE AP-SInternational Symposium and USNC-URSI
Radio S ien e Meeting, San Diego, California, USA, pp. 2321-2322,
July9-15, 2017.
[C13℄ G.Gottardi,G.Oliveri, andA. Massa,New antenna design on ept
for future generation wireless ommuni ation systems, 12th
Euro-peanConferen eonAntennasandPropagation(EUCAP2018),
Lon-don, United Kingdom, pp. 1-4, April9-13, 2018.
[C14℄ A.Massa, M. Bertolli,G.Gottardi,A. Hannan, D. Mar antonio,G.
Oliveri, A. Polo, F. Robol, P. Ro a, and F. Viani, Compressive
sensing as applied to antenna arrays: synthesis, diagnosis, and
pro- essing,2018IEEEInternationalSymposiumonCir uits&Systems
(ISCAS 2018), Firenze, Italy, pp. 1-5, May 27-30, 2018.
[C15℄ 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,
"Computational methods for wireless stru tural health monitoring
of ultural heritages," 8th International Conferen e on New
Com-putational Methods for Inverse Problems (NCMIP 2018), Ca han,
tionalmethodsfor inverse problems arising inarrays pro essing and
design,"8th International Conferen e onNew Computational
Meth-ods for Inverse Problems (NCMIP 2018), Ca han, Fran e, May 25,
2018.
[C17℄ G. Gottardi, G. Oliveri, D. Cunial, and A. Massa, Designing new
generationantennasfor5GMiMosystems-A newperspe tivein
ar-raysynthesis,2018IEEEAP-SInternationalSymposiumand
USNC-URSIRadioS ien eMeeting,Boston,Massa hussets,USA,pp.
2175-2176,July 8-13, 2018.
[C18℄ 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, AttiXXIRiunione Nazionaledi
Elettromag-netismo(XXI RiNEm), Parma, pp. 164-167,12-14 Settembre 2016.
[C19℄ G. Gottardi, L. Poli, M. A. Hannan, and P. Ro a, "Wideband
phased arrays optimal design through onvex programming," 2016
IEEE AP-S International Symposium and USNC-URSI Radio
S i-en e Meeting, Fajardo, Puerto Ri o, pp. 767-768, June 26 - July 1,
2016.
[C20℄ A.Gelmini,G.Gottardi,andT. 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,
Fran e,May 12,2017.
[C21℄ G. Gottardi and T. Moriyama, "Indoor dete tion of passive targets
re ast as an inverse s attering problem," 7th International W
ork-shoponNew ComputationalMethodsfor InverseProblems(NCMIP
2017),Ca han, Fran e, May 12,2017.
[C22℄ G.Gottardi,L.Turrina,N.Anselmi,G.Oliveri,andP.Ro a,Sparse
onformalarraydesignformultiplepatternsgenerationthrough
multi-task bayesian ompressive sensing, 2017 IEEE AP-S International
SymposiumandUSNC-URSIRadioS ien eMeeting,SanDiego,
Cal-ifornia,USA,pp. 429-430,July 9-15,2017.
[C23℄ L.T. P.Bui, N.Anselmi,G. Gottardi,L.Poli,andP.Ro a,
Wide-band phasedarrays synthesis with maximum bandwidth through
it-erative onvexoptimization, 2017IEEEAP-SInternational
Sympo-siumandUSNC-URSIRadioS ien e Meeting,SanDiego,California,
national Symposium and USNC-URSI Radio S ien e Meeting, San
Diego,California,USA, pp. 2625-2626, July 9-15, 2017.
[C25℄ M. A. Hannan, M. Salu i, G. Gottardi, L. Poli, and P. Ro a,
Advan ed time-modulated array synthesis for dire tional
modula-tion optimization, 2017 IEEE AP-S International Symposium and
USNC-URSIRadioS ien eMeeting,SanDiego,California,USA,pp.
2027-2028,July 9-15, 2017.
[C26℄ M.Salu i, G.Gottardi,D. Pin hera, and M.D. Migliore, Antenna
measurements by design: a novel paradigm for antenna testing,
6th Asia-Pa i Conferen e onAntennas and Propagation(APCAP
2017),Xi'an, China, pp. 1-3, O tober16-19, 2017.
[C27℄ P. Ro a, G. Gottardi, M. Bertolli,F. Robol, T. Moriyama, and T.
Takenaka,"Pro essingGPRdatawith2DBayesian ompressive
sens-ing inverse s attering approa hes," 8th International Conferen e on
New Computational Methods for Inverse Problems (NCMIP 2018),
Ca han, Fran e, May 25,2018.
[C28℄ G. Gottardi and L. Poli, " Human hest imaging by real-time
pro- essing of ele tri al impedan e data tomography," 8th International
Conferen e on New Computational Methods for Inverse Problems
(NCMIP 2018), Ca han, Fran e,May 25,2018.
[C29℄ G.Gottardi,M.A.Hannan,and A.Polo,"Amulti-fo using ontrast
sour e Bayesian ompressive method for solving inverse s attering
problems,"9thInternationalWorkshoponNewComputational
Meth-ods for Inverse Problems (NCMIP 2019), Ca han, Fran e, May 24,
2019.
[C30℄ G. Gottardi, M. A. Hannan, and A. Polo, "NDT/NDE by means
of a probabilisti dierential ompressive sensing method," 9th
In-ternational Workshop on New Computational Methods for Inverse
Problems (NCMIP 2019), Ca han, Fran e, May 24, 2019.
[C31℄ G.Gottardi,M.A.Hannan,B.Li,A.Polo,M.Salu i,and F.Viani,
"PCA-Based inversion of WiFi signal for robust devi e-free indoor
targetdete tion,"9thInternationalWorkshoponNewComputational
Methods for Inverse Problems (NCMIP 2019),Ca han, Fran e,May
24,2019.
[C32℄ G.Gottardi,M.A.Hannan,andA.Polo,"Compressivesensing
Interna-lems (NCMIP 2019), Ca han, Fran e, May 24,2019.
[C33℄ F.Boulos,L.Dall'Asta,G.Gottardi,M.A.Hannan,A.Polo,andA.
Salas-San hez, "A onvex optimization-based inversion method for
the Synthesis of monopulse linear arrays," 9th International
Work-shop on NewComputational Methods for InverseProblems (NCMIP
2019),Ca han, Fran e, May 24,2019.
[C34℄ F. Boulos, L. Dall'Asta, G. Gottardi, M. A. Hannan, A. Polo, A.
Salas-San hez, and M. Salu i, "A omputational inversion method
for interferen e suppression in re ongurable thinned ring arrays,"
9th International Workshop on New Computational Methods for
In-verse Problems (NCMIP 2019),Ca han, Fran e, May 24, 2019.
[C35℄ G. Gottardi, M. A. Hannan, A. Polo, M. Salu i, and F. Viani,
"Frequen y-basedinversionofasinglewirelesslinkforindoorpassive
targetdete tion,"9thInternationalWorkshoponNewComputational
Methods forInverse Problems (NCMIP 2019),Ca han, Fran e,May
1 Introdu tion 1
2 Free-Spa e-Oriented Antenna Design Problem 7
2.1 Mathemati al Formulation . . . 8
3 Codesign Synthesis Strategy 13 3.1 Codesign Methodology -Formulation . . . 14
3.1.1 Array Clustering Step (Integer-Coded SO-GA). . . 15
3.1.2 AntennaElement Synthesis Step(
ε
-MOEA) . . . 183.1.3 Codesign Algorithmi Implementation . . . 20
3.2 Numeri al Assessment . . . 22
3.2.1 Calibrationof the Extended Finite Array Model . . . 22
3.2.2 Element Synthesis and Array Clustering . . . 24
4 Capa ity-Driven Antenna Design Problem 43 4.1 Capa ity-Oriented MIMO System Design Meta-Paradigm - F or-mulation . . . 44
5 Capa ity-Driven Synthesis Methodology 51 5.1 Methodologi alCustomizationandImplementationProofs-of-Con ept 52 5.1.1 FP Array Ar hite tures . . . 52
5.1.2 CC Array Ar hite tures . . . 53
5.2 Numeri al Results and ComparativeAnalysis . . . 58
5.2.1 Synthesis of Fully-Populated Layouts . . . 59
5.2.2 Design of ContiguouslyClustered Arrangements . . . 73
3.1 Array lusteringphase(Re tangularlatti e,
0.492λ
0
×0.651λ
0
unit ell,N
= 18 × 14
) - E ien y and gain performan e assuming isotropi ,stand-alone,andembeddedradiatorsinFig. 3.5(b)and omparisonswith CST full-wave modeling. . . 35
3.2 Array lusteringphase(Re tangularlatti e,
0.492λ
0
×0.651λ
0
unit ell,N
= 28 × 16
) - E ien y and gain performan e assuming isotropi ,stand-alone,andembeddedradiatorsinFig. 3.5(b)and omparisonswith CST full-wave modeling. . . 42
5.1 CC Ar hite ture (
N
= 55
,Q
= 22
,V
= 6
) - Aperture tiling des riptors. . . 555.2 FP Ar hite ture Design (NLOS IMT-A UMi,
N
= 32
,L
= 16
,SN R
= 20
[dB℄)- Average performan e indexes. . . 65 5.3 CC Ar hite ture Design (NLOS IMT-A UMi - S enario 1,L
=
1.1 Layout of (a) onventional array ar hite ture and (b) thinned,
( ) lustered, and (d) sparse un onventional arrayar hite tures. 2
2.1 Elementaryradiator top-view(b) and side-view ( ). . . 8
3.1 Flow hart of the proposed odesign strategy (yellow boxes
high-lightad-ho ustomized operators). . . 15
3.2 Sket hoftheinteger- odedSO-GA de oding te hnique for luster
pla ement. . . 17
3.3 Extended nite model. . . 19
3.4 Extended nite model alibration (Re tangularlatti e,
0.492λ
0
×
0.651λ
0
unit ell) - Behaviour of|S
k1
(f )|
versus frequen y when varying thenumberof ringsH
surroundingthe referen eantenna when (a)k
= 2
and (b)k
= 5
. . . 23 3.5 Antenna element synthesis phase (Re tangular latti e,0.492λ
0
×
0.651λ
0
unit ell) - Plots of the representative points of a set ofε
-MOEA solutions in thee
Φ
1
, e
Φ
2
plane and asso iated Pareto
fronts (a), and geometries of sele ted Pareto-optimal tradeo
so-lutions (b)( ) asmodeled inCST Mi rowave Studio. . . 24
3.6 Antenna element synthesis phase (Re tangular latti e,
0.492λ
0
×
0.651λ
0
unit ell) - Behaviour ofE
(θ, ϕ; f )
for (a)( ) tradeo solution in Fig. 3.5(b) and (a)( ) tradeo solution in Fig. 3.5( )when (a)(b)
f
= f
min
and ( )(d)
f
= f
max
. . . 25
3.7 Antenna element synthesis phase (Re tangular latti e,
0.492λ
0
×
0.651λ
0
unit ell)-Behaviourofthefull-wavenumeri alsimulated (a)|S
11
(f )|
,(b)ε
(f )
,( )realizedgain,(d)W
AZ
(f )
andW
EL
(f )
, and (e)|S
k1
(f )|
(k
= 2, ..., 9
)forthe tradeo ongurationinFig. 3.5(b). . . 263.8 Antenna element synthesis phase (Re tangular latti e,
0.492λ
0
×
0.651λ
0
unit ell)-Behaviourofthefull-wavenumeri alsimulated (a)|S
11
(f )|
,(b)ε
(f )
,( )realizedgain,(d)W
AZ
(f )
andW
EL
(f )
, and (e)|S
k1
(f )|
(k
= 2, ..., 9
)forthe tradeo ongurationinFig. 3.5( ). . . 273.9 Array lusteringphase(Re tangularlatti e,
0.492λ
0
×0.651λ
0
unit ell,N
= 18 × 14
, Fig. 3.5(b) embedded radiator) - Plots of the representative pointsofaset ofSO-GA solutionsinthe(Φ
SO
, Q)
-plane and asso iated Pareto fronts (a), and layouts of sele tedPareto-optimal tradeo lustered arrangements (ea h olor
iden-tifyingadierent luster)with(b)
Q
= 196
,( )Q
= 144
, and(d)Q
= 76
. . . 28 3.10 Array lusteringphase(Re tangularlatti e,0.492λ
0
×0.651λ
0
unitell,
N
= 18 × 14
,Q
= 196
),(θ
0
, ϕ
0
) = (90, 60)
[deg℄ - Plots ofP (θ, ϕ; θ
0
, ϕ
0
; f )
assumingthe embedded element inFig. 3.5(b) at(a)f
= f
min
and (d)
f
= f
max
and(g)asso iated uts versus
ϕ
and omparisons with isotropi and stand-alone radiators. 303.11 Array lusteringphase(Re tangularlatti e,
0.492λ
0
×0.651λ
0
unit ell,N
= 18 × 14
,Q
= 196
),(θ
0
, ϕ
0
) = (112.5, 0)
[deg℄ - Plots ofP (θ, ϕ; θ
0
, ϕ
0
; f )
assumingthe embedded element inFig. 3.5(b) at(a)f
= f
min
and (d)
f
= f
max
and(g)asso iated uts versus
ϕ
and omparisons with isotropi and stand-alone radiators. 31 3.12 Array lusteringphase(Re tangularlatti e,0.492λ
0
×0.651λ
0
unitell,
N
= 18 × 14
,Q
= 196
),(θ
0
, ϕ
0
) = (112.5, 60)
[deg℄- Plots ofP (θ, ϕ; θ
0
, ϕ
0
; f )
assumingthe embedded element inFig. 3.5(b) at(a)f
= f
min
and (d)
f
= f
max
and(g)asso iated uts versus
ϕ
and omparisons with isotropi and stand-alone radiators. 323.13 Array lusteringphase(Re tangularlatti e,
0.492λ
0
×0.651λ
0
unit ell,N
= 18 × 14
) - Pattern uts ofP (θ, ϕ; θ
0
, ϕ
0
; f )
assuming the embedded element in Fig. 3.5(b) and omparisons withisotropi and stand-alone radiators when (a)(b)
(θ
0
, ϕ
0
) =
(90, 60)
[deg℄,( )(d)(θ
0
, ϕ
0
) = (112.5, 0)
[deg℄,and(e)(f)(θ
0
, ϕ
0
) =
(112.5, 60)
[deg℄for (a)( )(e)Q
= 144
,and (b)(d)(f)Q
= 76
. . . 333.14 Array lusteringphase(Re tangularlatti e,
0.492λ
0
×0.651λ
0
unit ell,N
= 18 × 14
,Q
= 196
,(θ
0
, ϕ
0
) = (112.5, 0)
[deg℄) - Plots simulatedP (θ, ϕ; θ
0
, ϕ
0
; f )
and omparisons with CST full-wave modeling. . . 343.15 Array lusteringphase(Re tangularlatti e,
0.492λ
0
×0.651λ
0
unit ell,N
= 28 × 16
, Fig. 3.5(b) embedded radiator) - Plots of the representative pointsofaset ofSO-GA solutionsinthe(Φ
SO
, Q)
-plane and asso iated Pareto fronts (a), and layouts of sele tedPareto-optimal tradeo lustered arrangements (ea h olor
iden-tifyingadierent luster)with(b)
Q
= 348
,( )Q
= 252
, and(d)3.16 Array lustering phase (Re tangular latti e,
0.492λ
0
× 0.651λ
0
unit ell,N
= 28 × 16
),Q
= 348
- PlotsP (θ, ϕ; θ
0
, ϕ
0
; f )
as-suming the embedded element in Fig. 3.5(b) and omparisonswith isotropi and stand-alone radiators when (a)
(θ
0
, ϕ
0
) =
(90, 60)
[deg℄, (b)(θ
0
, ϕ
0
) = (112.5, 0)
[deg℄, and ( )(θ
0
, ϕ
0
) =
(112.5, 60)
[deg℄. . . 38 3.17 Array lustering phase (Re tangular latti e,0.492λ
0
× 0.651λ
0
unit ell,
N
= 28 × 16
),Q
= 252
- PlotsP (θ, ϕ; θ
0
, ϕ
0
; f )
as-suming the embedded element in Fig. 3.5(b) and omparisonswith isotropi and stand-alone radiators when (a)
(θ
0
, ϕ
0
) =
(90, 60)
[deg℄, (b)(θ
0
, ϕ
0
) = (112.5, 0)
[deg℄, and ( )(θ
0
, ϕ
0
) =
(112.5, 60)
[deg℄. . . 39 3.18 Array lustering phase (Re tangular latti e,0.492λ
0
× 0.651λ
0
unit ell,
N
= 28 × 16
),Q
= 152
- PlotsP (θ, ϕ; θ
0
, ϕ
0
; f )
as-suming the embedded element in Fig. 3.5(b) and omparisonswith isotropi and stand-alone radiators when (a)
(θ
0
, ϕ
0
) =
(90, 60)
[deg℄, (b)(θ
0
, ϕ
0
) = (112.5, 0)
[deg℄, and ( )(θ
0
, ϕ
0
) =
(112.5, 60)
[deg℄. . . 40 3.19 Array lusteringphase(Re tangularlatti e,0.492λ
0
×0.651λ
0
unitell,
N
= 28 × 16
,Q
= 348
,(θ
0
, ϕ
0
) = (112.5, 0)
[deg℄) - Plots simulatedP (θ, ϕ; θ
0
, ϕ
0
; f )
and omparisons with CST full-wave modeling. . . 414.1 Problem Formulation - Multi-user multi-antenna downlink
s e-nario (a). Detailsof(a)the transmittingandthere eiving
anten-nas and of (b)the logi als hemeof the linearmulti-beamfeeding
ar hite ture ( ). . . 45
5.1 CC Ar hite ture (
N
= 55
,Q
= 22
,V
= 6
) - Sket h of a CC aperture(Tab. II5.1). . . 545.2 Ben hmark S enarios - Spatiallo ations of the BS (i.e., a
trans-mittingarrayof
S
radiatingelements)andoftheL
user-terminals in the Madrid grid urban environment (NLOS IMT-A UMi ):(a)S enario 1, (b)S enario2, and ( )S enario 3. . . 59
5.3 FP Ar hite ture Design (NLOS IMT-A UMi - S enario 1,
N
=
32
,L
= 16
,SN R
= 20
[dB℄) - Plots of (a)C
χ,l
, (b)D
χ,l
, and ( )SLL
χ,l
versus thel
-th(l
= 1, ..., L
) re eiverindex (χ
∈ {V, H}
). 60 5.4 FP Ar hite ture Design (NLOS IMT-A UMi - S enario 1,N
=
32
,L
= 16
) - Dire tivity patterns synthesized by (a)( )(e) themax C
and (b)(d)(f) themax D
methods when (a)(b)b
= 1
, ( )(d)b
= 3
,and (e)(f)b
= 16
. . . 615.5 FP Ar hite ture Design (NLOS IMT-A UMi - S enario 1,
N
=
32
,L
= 16
) -φ
0
ut of the dire tivity pattern when (a)b
= 1
(φ
0
= 16
[deg℄), (b)b
= 3
(φ
0
= −16
[deg℄), and ( )b
= 16
(φ
0
= 4
[deg℄). . . 62 5.6 FP Ar hite ture Design (NLOS IMT-A UMi,N
= 32
,L
= 16
,b
= 1
) - Dire tivity patterns radiated by (a)(b) themax C
and ( )(d)themax D
arrays for (a)( )the S enario2 and (b)(d) the S enario 3. . . 635.7 FP Ar hite ture Design (NLOS IMT-A UMi,
N
= 32
,L
= 16
,b
= 1
)-Plots of(a)(b)magnitude,W
ψ,n,b
(i)
(f )
,and ( )(d)phase,∠W
ψ,n,b
(i)
(f )
,ex itations (n
= 1, ..., N
;ψ
∈ {V, H}
) for(a)( ) theS enario 2 and (b)(d) the S enario 3. . . 64
5.8 FP Ar hite ture Design (NLOS IMT-A UMi - S enario 1,
N
=
32
,L
= 16
)-Plotsof(a)( )(e)W
ψ,n,b
(i)
(f )
and(b)(d)(f)∠W
(i)
ψ,n,b
(f )
versus the
n
-th (n
= 1, ..., N
) antenna array index (ψ
∈ {V, H}
) (a)(b)b
= 1
, ( )(d)b
= 3
,and (e)(f)b
= 16
. . . 66 5.9 FP Ar hite ture Design (NLOS IMT-A UMi,N
= 32
,L
= 16
,SN R
= 20
[dB℄) - Plots of (a)(b)C
χ,l
, ( )(d)D
χ,l
, and (e)(f)SLL
χ,l
versus thel
-th (l
= 1, ..., L
) re eiver index (χ
∈ {V, H}
) for(a)( )(e) the S enario2 and (b)(d)(f) the S enario 3. . . 675.10 FP Ar hite ture Design (NLOS IMT-A UMi - S enario 1,
L
=
16
,SN R
= 20
[dB℄) - Behaviour of (a)C
ave
, (b)D
ave
, and ( )SLL
ave
versusN
. . . 68 5.11 FP Ar hite ture Design (NLOS IMT-A UMi - S enario 1,N
=
72
,SN R
= 20
[dB℄) - Behaviour of (a)C
ave
, (b)D
ave
, and ( )SLL
ave
versusL
. . . 70 5.12 FP Ar hite ture Design (N
= 32
,L
= 16
) - Plot ofC
ave
versusSN R
fordierentpropagations enarios (urbans enario-NLOS IMT-A UMi; rural s enario - LOS 3GPP RMa and NLOS3GPP RMa). . . 71
5.13 FP Ar hite ture Design (
N
= 32
,L
= 16
,b
= 1
) - Dire tivity patternsgenerated by (a)( )themax C
and(b)(d)themax D
arrays in orresponden e with (b)( ) the LOS 3GPP RMa and(d)(e)theNLOS3GPPRMa propagationenvironmentsandthe
orresponding
φ
0
- uts (φ
0
= 16
[deg℄). . . 72 5.14 CC Ar hite ture Design (NLOS IMT-A UMi -S enario1,N
=
64
,L
= 16
,Q
= 32
,V
= 2
,SN R
= 20
[dB℄) - CC layouts synthesized with (a) themax C
and (b) themax D
methods. . 73 5.15 CC Ar hite ture Design (NLOS IMT-A UMi -S enario1,N
=
64
,L
= 16
,Q
= 32
,SN R
= 20
[dB℄) - Plots of (a)C
χ,l
, (b)D
χ,l
, and ( )SLL
χ,l
versus thel
-th (l
= 1, ..., L
) re eiver index (χ
∈ {V, H}
). . . 745.16 CC Ar hite ture Design (NLOS IMT-A UMi - S enario1,
N
=
64
,L
= 16
,Q
= 32
,SN R
= 20
[dB℄) - Dire tivity patterns radiated by (a)(b) themax C
and ( )(d) themax D
arrays when (a)( )b
= 1
and (b)(d)b
= 16
. . . 75 5.17 CC Ar hite ture Design (NLOS IMT-A UMi - S enario1,N
=
64
,L
= 16
,Q
= 32
,SN R
= 20
[dB℄) - Plots of (a)(b) mag-nitude,W
ψ,n,b
(i)
(f )
, and ( )(d) phase,∠W
(i)
ψ,n,b
(f )
, ex itations(
n
= 1, ..., N
;ψ
∈ {V, H}
) when (a)( )b
= 1
and (b)(d)b
= 16
. . 76 5.18 CC Ar hite ture Design (NLOS IMT-A UMi - S enario1,N
=
64
,L
= 16
,Q
= 32
,SN R
= 20
[dB℄) -φ
0
ut of the dire tivity pattern when (a)b
= 1
(φ
0
= 16
[deg℄) and (b)b
= 16
(φ
0
= 4
[deg℄). . . 775.19 CC Ar hite ture Design (NLOS IMT-A UMi - S enario 1,
L
=
16
,Q
N
= 0.5
,SN R
= 20
[dB℄) - Behaviour of (a)C
ave
, (b)D
ave
, and ( )SLL
ave
versusN
. . . 78 5.20 CC Ar hite ture Design (NLOS IMT-A UMi - S enario 1,L
=
16
,Q
N
= 0.5
,SN R
= 20
[dB℄,b
= 1
) - CC layouts (a)(b)(e)(f) and asso iated dire tivity patterns ( )(d)(g)(h) synthesized with(a)( )(e)(g) the
max C
and (b)(d)(f)(h) themax D
methods for BS arrays of (a)-(d)N
= 128
and (e)-(h)N
= 144
elements. 79 5.21 CC Ar hite ture Design (NLOS IMT-A UMi GF - S enario 1,L
= 16
,Q
N
= 0.5
,b
= 1
)-θ
0
- ut (θ
0
= 92
[deg℄) of the dire tivity pattern when (a)N
= 64
,(b)N
= 128
, and (b)N
= 144
. . . 80 5.22 CC Ar hite ture Design (N
= 64
,L
= 16
,Q
= 32
) - Plot ofC
ave
versus
SN R
for dierent propagation s enarios (urban s enario - NLOS IMT-A UMi; rural s enario - LOS 3GPP RMa andIntrodu tion
The next generation of mobile wireless te hnologies, ommonly labeled as
5G ,isexpe tedtodelivermulti-gigabit-per-se onddatarateswhileminimizing
osts, onne tion laten y, and power onsumption[1℄[2℄[3℄. To t these
require-ments,signi antadvan eswithrespe ttoprevious-generationsystems[2℄[4℄are
expe tedandrequiredfromthe te hnologi al, themethodologi al,and the
ar hi-te tural viewpoints. This onsideration holds true even more for the radiating
segment of 5G base stations (BSs) [2℄[3℄[4℄[5℄ as these systems are expe ted to
rely more and more on omplex multi-input multi-output (MIMO) antenna
ar-rayar hite tures toguarantee unpre edented levelofexibilityand performan e
[6℄[7℄[8℄ (i.e., supporting real-time adaptive beam re onguration/user tra king
features and MIMO pro essing) to address the needs of next-generationmobile
wirelesssystemsintermsofdataratesandnetwork apa ity[2℄. Moreover, these
latterobje tiveswillhavetobe obtained withinexpensivear hite tures and
an-tennaelementstominimizethefabri ationandthemaintenan e osts[2℄[3℄[4℄[5℄.
Towards this aim, a tive ele troni ally s anned arrays (AESAs) [9℄[10℄ are a
promisingte hnologytoaddress5G BS antennadesign[3℄[11℄[12℄[13℄[14℄[15℄that
are required to have a wide steering angle, a wide bandwidth, MIMO
apabili-ties [9℄[10℄,and real-time beam ontrol. Indeed, a broad lass ofstate-of-the-art
synthesismethodshavebeen formulated,andsu essfullyapplied,forthedesign
of AESA onventional ar hite tures to optimize the free-spa e performan e
of the radiated pattern [9℄[10℄[16℄[18℄[19℄[20℄[21℄[22℄[23℄[32℄. The exploitation of
free-spa e gures-of-merit,su hasthedire tivity(D),thehalf-powerbeamwidth
(HPBW) or the sidelobe level (SLL) [9℄[10℄[16℄[17℄[19℄[20℄[21℄[22℄[23℄[32℄ omes
fromtheideathat, a ordingtotheFriis'equation,theBS radiatedpowermust
be fo used along the terminal dire tion to maximize the link quality and the
information transfer. Thus, the signal-to-noise ratio (SNR) at the re eiver an
be improved, forinstan e, by in reasing the antennagain and redu ing the SLL
(b)
( )
(d)
Figure1.1: Layout of (a) onventional array ar hite ture and (b) thinned, ( )
Generally speaking, onventional,or fully-populated,AESA ar hite tures
[Fig. 1.1(a)℄are hara terizedfromaregularpla ementofradiatingelementson
a regular latti e, or grid, and an independent ontrol of amplitude and phase,
of eitherthe transmitted orthe re eived signals, onea hradiatingelement (i.e.,
a transmit/re eive module, TRM, for ea h radiating element is provided) [9℄.
Despite the fa t that fully-populated ar hite tures provide optimal/full pattern
ontrol,these solutionsgenerally turnout to beexpensive ar hite tures [10℄[24℄,
due to the high number of TRMs in the feeding network, and therefore not
suitable for the onsidered appli ation. To redu e the ost of AESA
ar hite -tures still yielding satisfa tory free-spa e performan es, un onventional
ar hi-te tures, su h as thinned [Fig. 1.1(b)℄, lustered [Fig. 1.1( )℄, or sparse [Fig.
1.1(d)℄ arrays ar hite tures have been introdu ed [9℄ [25℄. Clustered
ar hite -tures [Fig. 1.1( )℄ inparti ular, have been widely used in the literature asthey
allows to a hieve AESA ar hite tures with good trade-os between omplexity
andperforman e [9℄-[39℄. Indeed, lusteredphasedarraysare hara terizedfrom
anar hite turewheremultipleelementsaregroupedinto lusters(orsub-arrays)
and ea h luster is fed from a single TRM [Fig. 1.1( )℄, leading to an
ar hite -ture whose number of ontrol points (i.e., the number of TRMs) is lower than
the number of radiating elements. Along with the redu ed omplexity of the
feeding network, the osts of lustered ar hite tures an be further redu ed if
adopted sub-arrays have similar and simple shapes (e.g., the same number of
grouped elements or identi al shapes) that are suitable for modular assembling
[26℄. Unfortunately, lusteredar hite turesmaybeae tedbysevere limitations
in the operationalbandwidth and in the re onguration exibility. Indeed, the
redu ed number of ontrol points, and the unavoidable quantization of
ampli-tude and phase at the element-level, leads to a limited ontrol on the radiated
patternanderrorsinthe apertureillumination[27℄. Su hadrawba k is
parti u-larlyevidentwhentheapertureissimplypartitionedintore tangularsub-arrays
of equal shapes and orientations (i.e., a regularly lustered layout is adopted)
as the periodi errors in the aperture illumination generate undesired grating
lobes in the radiation pattern when steered [9℄[27℄. To avoid these drawba ks,
a-periodi arrangements of sub-arrays (i.e., irregular lustered ar hite tures )
havebeen shown toa hieveasigni antlymoderationofthe quantizationphase
lobesinthe radiationpatternasthey allowtobreakthe quantizationperiodi ity
[9℄[28℄.
The advantages of irregular lustered ar hite tures have been rstly shown
in [29℄[30℄[31℄, where su h ar hite tures have been introdu ed to in rease the
operational bandwidth of the array ar hite tures, showing that the a-periodi
arrangement of sub-arrays allows to mitigate the level of the undesired
side-lobes. Su essively, several e ient lo al-sear h te hniques have been proposed
to ompute the optimal pla ement of sub-arrays subje t to onstraints on the
radiation pattern, for instan e a Geneti Algorithm (GA)-based approa h [32℄,
to be sele ted from an a-priori dened alphabet) have been also introdu ed to
enhan e the modularity of the designed ar hite ture. [37℄[38℄[39℄. Although
ef- ient methods for the design of jointly lustered ex itations magnitudes and
phasesare available,the design of the system nowat handis stillvery
halleng-ingtasksin e, tothebest of theauthor's knowledge,state-of-the-art-approa hes
are employed only for very limited s an angles/bands [10℄[38℄, while more
om-plexarrangements omprisingmanymorephaseshiftersthanmagnitude ontrols
are adopted if wider elds-of-view (su h as those of interest in5G) are required
[9℄[39℄.
In this thesis the design of un onventional lustered array ar hite tures
has been addressed by introdu ing new methodologiesand new methodologi al
paradigms inthe design pro ess. Firstly,the simultaneoussynthesis of irregular
lustered ar hite tures and antenna elements, for 5G BSs, is dis ussed and an
innovative odesign pro ess in whi h the antenna element, embedded in the
-nite layout, andthearrayar hite ture (i.e.,theradiatorsdispla ementandtheir
feedings)aresimultaneouslysynthesized by takingintoa ountreal-systems
im-pa tisformulated. Indeed,the proposed odesign strategyisaimedatee tively
handlingtheinterrelationshipsarisinginthe5G s enariobetweenthesingle
radi-atorandthearrayar hite turesin e(i)thearraygeometrydire tlymodiesthe
mutual ouplingee ts among the antenna elements that ae t the 5G
gures-of-merit of the single array element (e.g., the impedan e, the isolation, and the
radiationpattern) and, in turn, (ii) the antenna response (e.g., the beamwidth
and the patternslope) impa ts the sidelobe ontroland the steering features of
the resulting array layout. In the se ond part, an innovative meta-paradigm is
introdu ed to design un onventional apa ity-driven antenna ar hite tures, for
MIMO appli ations, to maximize the ommuni ation quality depending on the
a tual Green's fun tion of the whole propagation s enario, in luding the
an-tennasystems, insteadof onsideringstandard free-spa e onditions andrelated
performan e indexes. With respe t to the free-spa e-oriented odesign
strat-egy, the expression of the link quality measure is expli itly derived from the
time-domainmodelofthe ele tromagneti environmentand the problemof
syn-thesizing MIMO arrays is formulated as the maximization of the quality of the
ommuni ation system. A ordingly,the proposed apa ity-driven design
meta-paradigmis then ustomized to two relevant and representative ex itation-only
synthesis problems, on erned with fully-populated and ontiguously- lustered
Thesis outline
The thesis is organized as follows, Firstly, the 5G base station design problem
is mathemati ally formulated as the optimization of free-spa e parameters in
Chapter 2. Then the odesign synthesis pro ess is formulated and detailed in
Chapter 3toee tively addressthe arisingMMO design problem. InChapter 4
the apa ity-drivendesignparadigmisformulatedandtheimplemented
apa ity-driven synthesis methodology is applied to the design of fully-populated and
ontiguously- lusteredarrangementsinChapter 5. Finallysome on lusions are
Free-Spa e-Oriented Antenna
Design Problem
In this Chapter, the mathemati al formulation for the design of un onventional
arrayar hite tures,featuringirregularly lusteredex itations,isaddressed. More
in details, the 5G base station design problem,aimed at tting multiple
obje -tives/requirementsonthe free-spa e performan e ofthearrayar hite ture and
of the the array element radiation, is formulated as a massive-multi-obje tive
[MMO℄problem. Consequently, the MMO ost fun tionand thetermsen oding
the synthesisobje tivesof theMMO designproblemare dened and
2.1 Mathemati al Formulation
Letus onsideraplanararrangementof
N
elementsdispla edintheyz
-planeand lusteredinQ
sub-arraysof ontiguouselements[Fig. 1.1( )℄asindi atedbythe membershipve torc
, {c
n
∈ [0, Q − 1] ; n = 1, ..., N}
, whosen
-thentryc
n
= q
indi atesthemembershipof then
-tharrayelementtotheq
-th luster. Sin e all the elements of theq
-th luster are fed with the same ex itationmagnitude,a
q
, and phase,ψ
q
(θ
0
, ϕ
0
; f )
, whi h depends on the steering angle(θ
0
, ϕ
0
)
and the working frequen yf
, the far-eld power pattern radiated by the array is given by [9℄[10℄P (θ, ϕ; θ
0
, ϕ
0
; f ) =
P
Q−1
q=0
{a
q
exp [jψ
q
(θ
0
, ϕ
0
; f )]
P
N−1
n=0
δ
qc
n
E
n
(θ, ϕ; f ) exp j
2π
λ
r
n
· ˆr
o
2
(2.1)where
λ
is the wavelength atf
,δ
qc
n
is the Krone ker delta fun tion (δ
qc
n
= 1
ifc
n
= q
andδ
qc
n
= 0
otherwise), andˆr = sin (θ) sin (ϕ) ˆ
y
+ cos (θ) ˆz.
(2.2) Moreover,r
n
= y
n
y
ˆ
+ z
n
ˆz
is the position of the then
-th (n
= 1, ..., N
) array element, whileE
n
(θ, ϕ; f )
is the orresponding radiation pattern, whi h is a fun tion of theM
-size ve tor of its geometri al des riptorsg
, {g
m
, m
= 1, ..., M} .
(2.3) Alongwith the far-eld powerpatternP (θ, ϕ; θ
0
, ϕ
0
; f )
, the dire tivity patternD
(θ, ϕ; θ
0
, ϕ
0
; f ) =
4πP (θ, ϕ; θ
0
, ϕ
0
; f )
R
π
0
R
2π
0
[P (θ, ϕ; θ
0
, ϕ
0
; f ) sin (θ)] dθdϕ
(2.4)and inparti ular,the dire tivityvalue alongthe steeringangle,
D
(θ
0
, ϕ
0
; f )
, has been onsidered in the synthesis pro ess as a fundamental metri of thefree-spa e performan e of the ar hite ture.
(a) (b)
PROBLEM
To design a 5G base station antenna, manifold free-spa e-based
obje -tives/requirements pertainingthe array pattern(i.e., sidelobe mask omplian y
for allthe steeringangles and withinthe whole frequen y band of interest) and
the array element radiation features (i.e., impedan e mat hing, isolation from
surrounding elements, polarization ellipti ity, gain, and beamwidth) must be
tted [3℄[40℄[14℄.
A ording toabove onsiderations, the synthesis athand turns out to
inher-entlybeaMMO problemwhenformulatedasthat ofofthe optimalsetupof the
unknown ve tors
n
c, a
, {a
q
, q
= 1, ..., Q} , Ψ, g
o
su h that
{c, a, Ψ, g}
opt
= arg
min
{c, a, Ψ, g}
[Φ (c, a, Ψ, g)]
(2.5) whereΨ
,
ψ
q
(θ
0
, ϕ
0
; f ) ; q = 1, ..., Q; θ
0
∈
θ
0
min
, θ
max
0
,
ϕ
0
∈
ϕ
min
0
, ϕ
max
0
, f
∈
f
min
, f
max
(2.6) is the set of ex itation phases of the
Q
lusters in the rangesf
∈
f
min
, f
max
,θ
0
∈
θ
min
0
, θ
0
max
, andϕ
0
∈
ϕ
min
0
, ϕ
max
0
of the 5G requirements. Moreover,Φ
(c, a, Ψ, g)
, {Φ
l
(c, a, Ψ, g) , l = 1, ..., L}
(2.7) is the MMO ost fun tion set omprisingL
terms en oding the synthesis obje -tives as follows1
:
•
Array Sidelobe Mask Complian y Term (l
= 1
):Φ
1
(c, a, Ψ, g)
,
(f
max
−f
min
)(ϕ
max
1
−ϕ
min
)(θ
max
−θ
min
)
×
×
R
f
f
min
max
R
θ
max
0
θ
min
0
R
ϕ
max
0
ϕ
min
0
max
θ,ϕ
{R [P (θ, ϕ; θ
0
, ϕ
0
; f ) −
P
T
(θ, ϕ; θ
0
, ϕ
0
; f )
dθ
0
dϕ
0
df
(2.8) whereP
T
(θ, ϕ; θ
0
, ϕ
0
)
is the target sidelobe when the array is steered to-wards(θ
0
, ϕ
0
)
andR [·]
istheramp fun tion(R [·]
, [·]×H [·]
,H [·]
being the Heaviside fun tion);•
Element Impedan eMat hing Term (l
= 2
):Φ
2
(c, a, Ψ, g) =
= Φ
2
(c, g)
,
R
f max
f min
R
[
|S
11
(f )|−
|
S
T
11
|]
df
(f
max
−f
min
)
(2.9) whereS
11
(f )
andS
T
11
arethesynthesizedand thetargetantennaree tion oe ient atthe input port atthe frequen yf
,respe tively;1
Forthesakeofnotation ompa tness,thedependen yof alltheradiationquantities(i.e.,
pattern,gain,sidelobe,ree tion oe ient,et .) ontheDoFsve tor
(c, a, Ψ, g)
isomitted hereinafter.•
Element Polarization Term (l
= 3
):Φ
3
(c, a, Ψ, g) = Φ
3
(c, g)
,
R
f
max
f
min
ε
(f ) − ε
T
2
df
(f
max
− f
min
)
whereε
(f )
andε
T
arethe synthesized andthe targetantennapolarization
ellipti ity angle atthe frequen y
f
, respe tively;•
Element Beamwidth Terms (l
= 4, 5
):
Φ
4
(c, a, Ψ, g) =
= Φ
4
(c, g)
,
R
f max
f min
R
[
W
AZ
(f )−W
T
AZ
]
df
(f
max
−f
min
)
Φ
5
(c, a, Ψ, g) =
= Φ
5
(c, g)
,
R
f max
f min
R
[
W
EL
(f )−W
T
EL
]
df
(f
max
−f
min
)
(2.10) whereW
AZ
(f ) =
arg min
ϕ∈[0,ϕ
M
]
E(θ
M
,ϕ;f )
E(θ
M
,ϕ
M
;f )
− 0.5
−
arg min
ϕ∈
[
−
π
2
,ϕ
M
]
E(θ
M
,ϕ;f )
E(θ
M
,ϕ
M
;f )
− 0.5
(2.11) andW
EL
(f ) =
arg min
θ∈[θ
M
,π]
E(θ,ϕ
M
;f )
E(θ
M
,ϕ
M
;f )
− 0.5
−
arg min
θ∈[0,θ
M
]
E(θ,ϕ
M
;f )
E(θ
M
,ϕ
M
;f )
− 0.5
(2.12)are the antenna half-power beamwidths in azimuth and elevation, while
W
T
AZ
and
W
T
EL
the orrespondingtargetvalues, beingθ
M
=
π
2
, ϕ
M
= 0
the element
broadsidedire tion,and
E
(θ, ϕ; f ) =
1
N
P
N
−1
n=0
E
n
(θ, ϕ; f )
istheaverageelement fa tor;•
Element Realized Gain Term (l
= 6
):Φ
6
(c, a, Ψ, g)
,
=
R
f max
f min
R
G
T
−
4πκ1(f)κ2(f)E
(
θM,ϕM;f
)
R 2π
0
R π
0 [E(θ,ϕ;f ) sin(θ)]dθdϕ
df
(f
max
−f
min
)
(2.13) whereG
T
is the target realized gain value, while
κ
1
(f )
andκ
2
(f )
are the average elemente ien y and the average mismat hfa tor, respe tively;•
Element Isolation terms (l
= 7, ..., L
):Φ
l
(c, a, Ψ, g)
,
=
R
f max
f min
R
[
|S
k1
(f )|−
|
S
T
k1
|]
df
(f
max
−f
min
)
k
= l − 6; l = 7, ..., L
(2.14) whereS
k1
(f )
andS
T
k1
are the synthesized and the target isolation oe- ients between the referen e antenna and itsk
-thneighborelement at the frequen yf
,respe tively.PROBLEM
Although5G standards have notyetbeen o iallypublished,aset offrequen y
bands have already been released [41℄, and guidelines regarding the envisaged
performan eandindi atorstobeoptimizedfor3Dbeamforming,a tiveantenna,
and massive MIMO systems have already been dis ussed [3℄[40℄[14℄[42℄. Thus,
owing to its full generality, the formulated design problem and the proposed
synthesis strategy willbeseamlessly exploitable/ ustomized a ording tofuture
Codesign Synthesis Strategy
In this Chapter, the odesign synthesis strategy of un onventional array
ar hi-te tures, featuring irregularly lustered ex itations and an elementary antenna
onsisting ofa avity-ba ked planar pat h stru ture witha spline-based ontour
[43℄[44℄[Fig. 2.1(a)-(b)℄isaddressed. The ombinationofspline- ontouredpat h
antennas[43℄[44℄with avity ba kinghas been hosen asitisexpe tedtoenable
awideexibility(when omparedtostandardgeometries[43℄[44℄)toaddressthe
5G MMO problem as well as to allow a low inter-element oupling with
te h-nologi ally simple and quite inexpensive stru tures. A ordingly, the proposed
odesign methodology is introdu ed and detailed. Following the No-free-lun h
Theorem[45℄,ad-ho sear hstrategiesaresele tedtoe ientlyaddresstheabove
5G design sub-problems. Namely,
•
asingle-obje tiveGeneti Algorithm(SO-GA)[50℄[32℄[46℄ featuringan in-novative integer oding of the variables and ad-ho operators for handlingthe irregular lustering [Fig. 1.1( )℄.
and
•
aε
-MOEAapproa h [47℄[48℄ aimedatoptimizingthe shapeof the elemen-taryradiator [Fig. 2.1(a)-(b)℄to meet the 5G ele tri alrequirements (i.e,inputimpedan e,beamwidth,isolation,polarization,et ....) when
embed-ded in a nite array and not alone or in an innite periodi arrangement
asusually done in the state-of-the-art literature.
Sele ted numeri alexamples, drawn fromanexhaustive designpro ess,are then
presented toassessthe advantagesandthe ee tivenessoftheproposed odesign
3.1 Codesign Methodology - Formulation
Solving the MMO problem in (2.5) with a single step pro edure is not feasible
be ause of the unavailability of ee tive and e ient MO optimization
strate-gies when
L
≥ 4
, the omplex and highly non-linear nature of the obje tive fun tions/termsin (2.7), and the large number of DoFs,{c, a, Ψ, g}
, resulting in a too huge and omputationally untra table sear h spa e. To address su hhallenges, the MMO problem in (2.5) has been split into two (still onne ted)
simplerones on erned with (i) the array design and (ii) the element synthesis.
However, solving the two problems is still a very hallenging task as it yield to
the derivation of ompletely new designs, both for the array and for the
ele-mentary radiator, with respe t to traditionalsolutions [2℄. Indeed, to the best
of the author's knowledge, (i) no methodologies exist that allow the ontrol of
the sidelobe prole of sub-arrayed layouts over wide s anning angles and
band-widthswhenthe ex itationmagnitudes andphases are jointly lustered and (ii)
spline- ontoured pat hes have never been designed in a 5G MMO s enario by
taking into a ount their intera tions with the surrounding elements in view of
the reliable base stationprototyping [43℄[44℄ 1
To address su h hallenges, the proposed odesign pro edure exploits the
om-binationof
•
asingle-obje tive (SO)array lusteringproblemwhere theDoFs omprise the lustering s hemes
and the ex itation ve torsa
andΨ
, the SO ost fun tionbeing dened, for a xed radiatorgeometry (g
= g
), asfollowsΦ
SO
(c, a, Ψ)
, Φ
1
(c, a, Ψ, g) ;
(3.1)•
aMMO radiatingelementdesignproblemwheretheDoFsarethegeometry parametersg
,whiletheMMO ostfun tion,foraxedsetoftheremaining parameters (c, a, Ψ
), is given by
Φ
M M O
(g)
,
Φ
l
c, a, Ψ, g
; l = 2, ..., L
.
(3.2)Itisworthremarkingthat(3.1)and (3.2)are generallytightly oupledproblems,
buteven moreinthe5Gs enario, anda ordinglyanalternatetwo-step odesign
pro ess (Se t. 3.1.3) is onsidered by rst optimizing the ost fun tion
Φ
SO
(3.1) wheng
= g
(Array Clustering Phase - Se t. 3.1.1) and then (Antenna Element Synthesis Phase - Se t. 3.1.2)Φ
M M O
(3.2) is minimized for a xed DoFs ongurationc, a, Ψ
(Fig. 3.1). 1
Conventionalspline- ontouredantennas havebeenproposedonlyforstand-alone
3.1.1 Array Clustering Step (Integer-Coded SO-GA)
In order to properly address the SO problem (3.1) of the Array Clustering
step, let us rst noti e that, for a given membership ve tor
c
and a xed setup ofg
(g
= g
), the optimal values of the array ex itations in both amplitude,a
, and phase,Ψ
, for auser-spe ied frequen yf
(f
∈
f
min
, f
max
)and asteering dire tion(θ
0
,ϕ
0
)(θ
0
∈
θ
min
0
, θ
0
max
,ϕ
0
∈
ϕ
min
0
, ϕ
max
0
),are unique and they an be found by solving the following onvex problem [49℄[a (c) , ψ (θ
0
, ϕ
0
; f ; c)]
opt
,
= arg min
a(c), ψ(θ
0
,ϕ
0
; f ; c)
[D (θ
0
, ϕ
0
; f )]
−1
s.t
P (θ, ϕ; θ
0
, ϕ
0
; f ) ≤ P
T
(θ, ϕ; θ
0
, ϕ
0
; f ) .
(3.3)
Indeed,
Φ
SO
[a (c) , ψ (θ
0
, ϕ
0
; f ; c)]
opt
= 0
. A ordingly, sin e the solution of the onvexproblem(3.3)isuniqueand it an befoundwithstate-of-the-arton-vex programmingtools [49℄, itturns out that the DoFs of the Array Clustering
phaseare onlythe membershipsofthe arrayelementslisted inthe integer
mem-bershipve tor
c
and theoptimizationproblemathand anbe lassiedasaSO dis rete one.Figure3.1: Flow hart of the proposed odesign strategy (yellowboxes highlight
As for the determinationof the unknown integer ve tor
c
, it annot be ignored that the hoi e ofc
(i.e., the type, the size, and the nature of the array lus-ters/tiles) signi antly impa ts on the osts of the resulting arrangements [9℄,thus it annot be totallyfree and the designer must be allowed to spe ify what
lass/es of lusters(e.g., the sizes and/orthe shapes aswellas
ontiguous/non- ontiguous types) has/have to be used as building blo k/s for the arising 5G
ar hite ture. Therefore, the 5GArray Clustering problemis formulatedas that
of nding the best ombinationof
Q
lusters taken from an alphabet ofD
user-dened sub-arrays [39℄ su h that (2.8) is minimized.Due to dis rete nature of the problem unknowns,
c
n
,n
= 1, ..., N
, it would be quitenaturalto onsider asoptimization solver[50℄[51℄[52℄abinary GA [32℄[46℄inwhi hthe membershipof the generi
n
-th array element,q
,is en oded intoa stringof⌈log
2
D⌉
bits [39℄. However, this hoi ewouldbenumeri allyinee tive sin e it would result in GA individuals with hromosomes of⌈log
2
D⌉ × Q
bits (i.e.,proportionaltothe sizeof the alphabetandthe aperture)potentiallyaus-ing GA onvergen e issues [50℄ be ause of their long length when implemented
for medium/large arrays as those in 5G appli ations. In order to prevent these
latter, an Integer-Coded SO-GA strategy is adopted hereinafter, the shape/size
hoi e of the
q
-th luster in thep
-th (p
= 1, ..., P
;P
being the GA population size)GA individualatthei
-thiteration(i
= 1, ..., I
;I
beingthe maximum num-ber of GA iterations),d
p,i
,
d
p,i
q
; q = 1, ..., Q
, being en oded in an integer
variable,
d
p,i
q
∈ [1, D]
thusredu ing theoverall lengthofthe hromosometoQ
as wellas the dimension ofthe sear hing spa e from2
(⌈log
2
D⌉×Q)
down to
D
Q
. The
pri e to pay for this hoi e isthe need to ustomize the standard GA operators
tosu h anInteger-Coded SO-GA.
More spe i ally (Fig. 3.1),
•
Mutation - If thep
-th individual is sele ted for mutation (η
being the mutationprobability),theq
-thgeneofthe orresponding hromosome(i.e., theq
-th tile of thep
-th lustering solution) is randomly mutated with probabilityµ
by settingd
p,i
q
toa randomintegervalue inthe range[1, D]
;•
De oding (Array Tiling)-Atthei
-thiteration,ea hgened
p,i
q
(q
= 1, ..., Q
) isrstly de oded intothe orresponding lustershape/size/orientationa - ordingtoastrategyinspiredby[39℄and onsideringthe alphabetlook-up
table in Fig. 3.2. In short, the
p
-th lustered arrangement orresponding to thep
-th GA individual at thei
-th iteration,c
p,i
, is generated by
pla -ingthe
Q
lusters within the aperturesu h that the rst tile islo ated in the top-rightedgeof the aperture,while the followingones are distributedover the array surfa e through a ir ular ( lo kwise) pla ement by taking
into a ount the lo ation of already lustered elements and exploiting a
Figure 3.2: Sket h of the integer- oded SO-GA de oding te hnique for luster
pla ement.
•
CostFun tion Evaluation -The optimalex itations[a (c
p,i
) , Ψ (c
p,i
)]
opt
of
the lustering s heme
c
p,i
are omputed by solving (3.3) with a standard
onvex programming tool [49℄ and the orresponding ost fun tion value
Φ
p,i
SO
is determined by substitutingc
p,i
and
[a (c
p,i
) , Ψ (c
p,i
)]
opt
in (3.1),
Φ
p,i
SO
, Φ
SO
(c
p,i
, a
(c
p,i
) , Ψ (c
p,i
))
;•
Convergen e Che k - The GA pro edure isiterateduntil eitheri
= I
orΦ
(I)
SO
−
P
I
i=I−b
I
Φ
(i)
SO
Φ
(I)
SO
≤ τ
(3.4) whereΦ
(I)
SO
, min
p=1,...,P ;i=1,...,I
Φ
p,i
SO
is the ost fun tion value of the
globalbestindividualattheiteration
I
,whileI
b
andτ
arethe onvergen e window and the onvergen e threshold, respe tively.As for the Sele tion and Crossover operators, standard roulette-wheel
imple-mentationsand (integer- oded)single-point rossoverwith rossoverprobability
χ
are adopted [50℄.It isworth pointing out that hoosing su han optimizationstrategy (i)enables
theusertoa-priori spe ify, bydeningthetilealphabet,the lustershapes
om-pliantwith the te hnologi al onstraints of the referen e 5G implementationas
itminimizesthe dimension of the sear h spa e,thus mitigatingthe onvergen e
issues of the optimization whenlarge apertures are of interest.
3.1.2 Antenna Element Synthesis Step (
ε
-MOEA)As for the se ond step of the odesign pro ess, it onsists of a MMO
ontinu-ous optimization problem sin e it involves the joint minimization of
L
− 1
ost fun tion terms [i.e.,Φ
M M O
(g)
- 3.2℄ to determine the optimal values of the ( ontinuous) geometri al parameters of the array element.Akeyme hanismofthis pro eduralstep istheso- alledantennageometry
gen-erator (AGG),whi his responsible toen ode/de ode the shapeof the antenna
elementstartingfromthe orrespondingve torof geometri alparameters
g
[43℄. The requirements for su h an operator are that, on the one hand, a wide setof stru tures should be modeled toallowthe explorationof dierent geometries
duringthe design phase[Fig. 2.1(a)℄and, onthe otherhand, a smallnumberof
des riptors,
M
, shouldbeenoughfor faithfullyidentifyingthe radiatorshape in order to keep low the dimensionality of the optimization problemthus avoidingomplex and time-expensive sear hes in the sampling the sear h spa e looking
for the optimalsolution. A ording tothese inputs and following the guidelines
adoptedforthe designofstand-alonewidebandantennasystems[43℄,the hosen
AGG operator is the result of a ombination of parametri and spline-based
des riptors used, the former, to des ribe the avity radius,
g
1
, and the avity height,g
2
,aswellastheprobefeedpositionintheyz
-plane,(g
3
, g
4
)
[Fig. 2.1(b)℄ and, the latter, toen ode omplex antenna ontours through spline urves withM−4
2
ontrol points in theyz
-plane,g
m
, g
m+
M −4
2
,m
= 5, ..., M −
M−4
2
[Fig. 2.1(a)℄. Of ourse,su hamodellingwouldbeenough(atleastinprin iple)whendealingwith stand-alone radiators[43℄, but itis unsatisfa tory when addressing
synthesis problems su h as that for 5G systems and, even more, when the
de-sign result is not limited to a lab- ontrolled experiment, but it must be lose
to the nal manufa turing, thus numeri ally predi ting real world impa ts and
guaranteeing a reliable working. More in detail, unlike the stand-alone design,
the solution of (3.2) must manage shape/size onstraints that vary depending
on the solution to the SO dis rete problem (3.1) iteratively arried out in the
odesign array lustering phase (Se t. 3.1.1). Moreover, the evaluation of the
mutual intera tionsamong nearby radiatorswhen arranged inanitearray
lay-out is mandatory (3.2), as well. To properly fa e with those di ulties, it is
needed to ouplethe AGG operator withanotherone [here alledSystem
Char-a terization (SC)operator℄abletomodeltheelementaryradiatorbytakinginto
a ount spa ing onstraints oming from the array lustering blo k and to
re-liably estimate the inter-element isolation/ oupling parameters by onsidering
the nite size of the nal arrangements, that is, without re urring to simplied
periodi models for the array stru ture. Towards this end, an extended nite