Si e-Speci ic RIS Deploymen in Cellula Ne wo ks
ia Calib a ed Ray T acing
Sina Bey aghi†⋆, Ja ad Shabanpou ♭, Gio anni Ge aci⋆, Paul Almasan†, and Angel Lozano⋆
⋆Uni . Pompeu Fab a, Ba celona, Spain †Tele ónica Resea ch, Ba celona, Spain
♭Uni . Aal o, Espoo, Finland
Abs ac —This wo k in oduces a ully-au oma ed RIS de-
ploymen s a egy alida ed h ough a digi al win, powe ed by
Sionna ay acing, o a UK ci y. On a scene calib a ed wi h
measu ed da a, he me hod join ly op imizes RIS placemen ,
o ien a ion, con igu a ion, and BS beam o ming ac oss 4G,
5G, and hypo he ical 6G equencies. Candida e RIS si es a e
iden i ied ia sca e ing-based ays, while use clus e ing educes
deploymen o e head. Resul s show ha meaning ul co e age
enhancemen equi es dense, la ge-ape u e RIS deploymen s,
aising ques ions abou he p ac icali y and cos o la ge-scale
RIS adop ion.
I. INTRODUCTION
Consis en connec i i y is a key challenge in mode n
wi eless sys ems, especially in dense u ban en i onmen s
whe e obs uc ions and mul ipa h p opaga ion deg ade pe -
o mance [1]. These issues wo sen a highe equencies due
o inc eased pa hloss and limi ed di ac ion. Recon igu able
in elligen su aces (RIS) ha e eme ged as a p omising solu-
ion. By e lec ing signals owa d unde se ed a eas wi hou
addi ional ansmi powe , RIS o e a po en ially ene gy-
e icien means o enhance co e age [2], [3]. Howe e , la ge-
scale adop ion by mobile ne wo k ope a o s (MNOs) emains
unce ain due o unclea cos -pe o mance adeo s.
Indeed, RIS placemen in ol es a high-dimensional combi-
na o ial p oblem, wi h many candida e si es, o ien a ion con-
s ain s, and si e-speci ic p opaga ion. T ial-and-e o me hods
a e unscalable. Tackling his challenge equi es an au oma ed
and scalable amewo k, and an en icing enable o such a
amewo k is a digi al win p o iding a p ecise digi al eplica
o he adio en i onmen [4].
P e ious s udies on RIS placemen and design o e alu-
able insigh s bu su e om key limi a ions. Many ocus
on single-use equipmen (UE) o single-cell scena ios using
idealized o s ochas ic models ha a e unable o cap u e si e-
speci ic de ails [5]–[9]. O he s use simpli ied e lec ion models
This wo k was in pa suppo ed by H2020-MSCA-ITN-2020 META
WIRELESS (G an Ag eemen : 956256), by he SNS JU Ho izon Eu ope
P ojec unde G an Ag eemen No. 101139161 (INSTINCT), by he 6G-
Machine In elligence based Radio Access In as uc u e (6G-MIRAI) p ojec
unde G an Ag eemen No. 101192369, by he Spanish Resea ch Agency
h ough g an s PID2021-123999OB-I00 and CNS2023-145384, by ICREA, by
he Ma ia de Maez u Uni s o Excellence P og amme (CEX2021-001195-M),
and by he Spanish Minis y o Economic A ai s and Digi al T ans o ma ion
and he Eu opean Union Nex Gene a ionEU h ough UNICO-5G I+D p ojec s
TSI-063000-2021-138 (SORUS-RIS) and TSI-063000-2021-59 (RISC-6G).
ha igno e pola iza ion and angula losses [2], [10]–[12], o
neglec p ac ical cons ain s like building geome y, ma e ial
p ope ies, and o ien a ion easibili y [13], [14]. Few combine
ay acing wi h empi ical da a, and none o e a scalable,
au oma ed solu ion ac oss mul iple equencies and ne wo k
layou s.
This pape p esen s a ully au oma ed, da a-d i en ame-
wo k o si e-speci ic RIS deploymen in cellula ne wo ks.
A adio digi al win is applied o join ly op imize RIS
placemen , o ien a ion, phase con igu a ion, and base s a ion
(BS) beam o ming. Unlike analy ical o heu is ic me hods, he
amewo k suppo s mul i-cell, mul i-band deploymen s and
inco po a es a calib a ed ay- acing engine o model ma e ial-
and geome y-speci ic elec omagne ic in e ac ions.
The amewo k is alida ed using da a om a UK comme -
cial deploymen . Simula ions span 4G, 5G, and a hypo he ical
6G sys em a 2GHz, 3.5GHz, and 10 GHz, espec i ely,
using he open-sou ce Sionna engine [15]. Ma e ial p ope ies
a e calib a ed wi h measu ed UE ecei ed powe . The main
con ibu ions a e as ollows:
•Using measu ed UE da a, ma e ial p ope ies a e cali-
b a ed o imp o e ay- acing accu acy.
•A ay-based me hod join ly op imizes RIS placemen ,
phase con igu a ion, and BS beam o ming, inco po a ing
o ien a ion and geome y cons ain s.
•Co e age imp o emen s a e e alua ed ac oss equencies
as a unc ion o RIS ape u e, elemen coun , and densi y.
•A cos -pe o mance analysis e eals ha signi ican co -
e age gains equi e ex ensi e deploymen o la ge RIS
uni s, ques ioning economic iabili y.
To os e ep oducibili y, he ull simula ion amewo k and
RIS deploymen algo i hms a e a ailable in open sou ce.1
II. NETWORK, RIS, AND PROPAGATION MODELS
A. Cellula Ne wo k Model
The digi al win eplica es he layou o a comme cial
cellula ne wo k deployed by a leading MNO. The ocus a ea
con ains 12 BSs anging om 18 o 56 m in heigh , each
hos ing h ee sec o s o a o al o 36 cells. The a ea spans
1340 m ×1390 m in he UK. Figu e 1 shows a 3D ende ing
o he en i onmen .
1h ps://gi hub.com/Tele onica-Scien i ic-Resea ch/DDRD
Fig. 1: 3D isualiza ion o he a ea p oduced wi h OpenS ee Map,
wi h black pins indica ing BS si es.
1) An ennas: Each BS ea u es a plana a ay o e ically
pola ized elemen s. The con igu a ion o each BS is speci ied
by i s e ical il angle and i s ho izon al bea ing angle, bo h
de e mined by he MNO. A single adio- equency chain eeds
each an enna elemen and i abides by he 3GPP adia ion
pa e n, wi h hal -powe beamwid hs o 65◦in azimu h and
10◦in ele a ion [16].
2) Radio Deploymen s: Th ee sys ems a e conside ed:
•4G a 2 GHz (FR1)
•5G a 3.5 GHz (FR1)
•6G a 10 GHz (FR3)
Fo each, he pa ame e s (adop ed om comme cial p oduc s,
ei he exis ing o unde de elopmen ) a e de ailed in Table I.
No e ha only single pola iza ion is assumed, esul ing in
elemen coun s ha a e hal hose o dual pola iza ion.
Reusing exis ing si e g ids is essen ial o he inco po a ion
o new spec um o be economically easible, as u he si es
would lead o conside able expenses and p olong deploymen
imelines [17]. Hence, The same BS coo dina es a e applied
o all h ee sys ems.2
3) Beam o ming Codebook: Each BS adop s a wo-
dimensional disc e e Fou ie ans o m (DFT) beam o ming
codebook pe sys em. This codebook speci ies a se o o hogo-
nal beams on gi en angula di ec ions; each such beam is cha -
ac e ized by he K onecke p oduc o wo one-dimensional
DFT ec o s, one o azimu h and one o ele a ion, namely
wmh,m =wmh⊗wm ,(1)
whe e mh, m a e he ho izon al and e ical beam indices.
2The 6G sys em ope a es a a highe equency and wi h mo e bandwid h
han i s 4G/5G coun e pa s, bu i is con igu ed wi h a lowe ansmi powe .
This e lec s a ole o 6G as a non-s andalone capaci y-boos ing laye wi h
possibly discon inuous co e age.
TABLE I: Sys em ea u es [18]
Fea u e 4G 5G 6G
Ca ie equency [GHz] 2 3.5 10
Bandwid h [MHz] 20 100 200
Sec o s pe si e 3 3 3
Pola iza ion Ve ical Ve ical Ve ical
Plana a ay opology 2×2 4×8 4×16
Beam o ming codebook size 4 32 64
TX powe pe cell [dBm] 43 49 44
TX powe pe subca ie [dBm] 12.2 13.85 8.85
Numbe o subca ie s 1200 3276 3276
Noise powe pe subca ie [dBm] -132 -129 -126
4) UEs: UEs a e deployed on a g id o 2 m ×2 m iles a
a heigh o 1.5 m, wi h he ecei ed powe a e aged pe ile.
The ocus is on ou doo UEs as he main use case o cellula
ne wo ks, hough he me hod ex ends o indoo scena ios gi en
loo plans and ou doo - o-indoo ay acing suppo . Each UE
employs a single e ically pola ized iso opic an enna and is
se ed indi idually pe ime- equency esou ce, isola ing he
e ec s o scheduling.
B. Physically consis en RIS Model
The physically consis en RIS model in [19] is adop ed,
which ea s he RIS as a con inuous e adia ing su ace wi h a
spa ial modula ion unc ion based on ay op ics. This cap u es
ini e su ace size, angula e adia ion, and ene gy conse -
a ion. P ecisely, he spa ial modula ion unc ion Γ(xR, yR)
desc ibes he complex ans o ma ion applied by he RIS a
each su ace poin (xR, yR), combining ampli ude and phase
modula ion in o [20]
Γ(xR, yR) = R√ηA(xR, yR)ejφ(xR,yR),(2)
whe e Raccoun s o oughness and loss, ηis he su ace
e iciency, and A(xR, yR)con ols e adia ed powe .
C. Si e-speci ic Ray T acing and Co e age Calcula ion
The p opaga ion among BSs, UEs, and RIS uni s is digi-
ally eplica ed using Sionna [21], an open-sou ce ay- acing
simula o . To balance ealism and e iciency, he ay acing
employs a Fibonacci shoo -and-bounce me hod wi h 107 ays
pe cell, up o ou bounces pe ay, and i includes specula
e lec ion, di ac ion, and sca e ing.
To compu e he la ge-scale channel gain be ween each BS
and UE, he se o co esponding ays is conside ed. As
ad anced, he egion o in e es is pa i ioned in o squa e iles
Cp,q o size 2 m ×2 m, whe e each ile co esponds o a UE.
Fo each BS sec o and DFT beam index m, he di ec ional
channel gain a ile (p, q)is compu ed as
G ,m,p,q =1
4ZZCp,q h∗
,p,q(x, y)w ,m
2dxdy, (3)
whe e h ,p,q(x, y)is he channel ec o a posi ion (x, y),
ob ained om all ay ypes (LoS, e lec ed, di ac ed, and
sca e ed), w ,m is he DFT p ecoding ec o , and 4is he ile
a ea.
The e e ence signal ecei ed powe (RSRP) o ile (p, q),
se ed by BS sec o and beam index m, is hen
RSRPp,q =G ,m,p,qP ,(4)
whe e P is he ansmi powe pe subca ie , as speci ied in
Table I. The UE a he cen e o each ile selec s i s se ing
BS and beam by maximizing he RSRP ac oss all sec o s and
beam indices.
III. SCENE MATERIAL CALIBRATION
Accu a e ay acing equi es no only p ecise geome y bu
also ealis ic elec omagne ic ma e ial p ope ies—some hing
pa icula ly di icul a ci y scale. While p io wo k ocused
on small, con olled en i onmen s [21], we a ge dense u -
ban a eas whe e de aul ma e ial se ings lead o signi ican
misma ches. Using a la ge ou doo measu emen da ase om
a UK ci y—co e ing BS con igu a ions (e.g., o ien a ion, il )
and spa ial ecei ed powe —ma e ial pa ame e s a e calib a ed
o align simula ed and expe imen al adio co e age.
Speci ically, building ma e ial pa ame e s a e calkib a ed by
minimizing he di e ence be ween simula ed and measu ed
RSRP alues. Using he Sionna ay ace and he ac ual
BS con igu a ions, co e age is i s simula ed assuming all
buildings a e made o conc e e. Th ee ma e ial p ope ies—
ela i e pe mi i i y, conduc i i y, and su ace sca e ing—
a e ea ed as lea nable a iables. UE measu emen da a is
agg ega ed o e 10 ×10 m2 egions wi h a leas 20 samples,
enabling s able RSRP compa isons. Fo each egion, su ound-
ing buildings wi hin 100 me e s a e join ly calib a ed by
compa ing a e age simula ed and measu ed RSRP. G adien s
a e compu ed using an Adam op imize , and ma e ials a e
i e a i ely upda ed. The op imiza ion uns o 600 s eps pe
cell, g adually co e ing he ci y while p ese ing al eady
calib a ed a eas. To ensu e meaning ul upda es, ou lie s and
egions wi h pe sis en misma ches a e il e ed ou . Region
me ging and e o h esholds a e applied o manage complexi y
and adap o u ban densi y a ia ions.
The design o his calib a ion app oach is shaped by bo h
modeling simpli ica ions and p ac ical da a limi a ions. Since
he ocus is on minimizing RSRP e o a he han e ie ing
physically exac ma e ial alues, he op imize adjus s pa ame-
e s o bes ma ch he model’s p edic ions wi h measu emen s.
Calib a ion is pe o med o e spa ial egions ins ead o in-
di idual samples o mi iga e de ice- ela ed RSRP a iabili y.
Regions wi h ex eme misma ches a e excluded, as hese a e
ypically due o missing scene geome y. The op imize eac s
adap i ely: i simula ed RSRP o e shoo s, ma e ial p ope ies
a e d i en owa d ee-space condi ions; i i unde shoo s,
e lec i i y is inc eased; in bo h cases, wi hin cons ained
bounds. Pe sis en e o s a he ex emes igge egion exclu-
sion. And, o manage u ban complexi y, adjacen buildings a e
me ged, and egion-speci ic h esholds a e applied based on
densi y, acknowledging ha calib a ion ole ance a ies ac oss
he ci yscape.
A. Valida ion
Ma e ial calib a ion was conduc ed o e a cen al u ban
a ea measu ing 1122 m ×710 m. To assess i s e ec i eness,
70 ep esen a i e a ge egions we e selec ed ac oss he a ea.
In each egion, he a e age measu ed RSRP was compa ed
agains simula ed alues, bo h be o e and a e calib a ion.
A de ailed analysis o p edic ion e o s is p o ided in Fig.2,
wi h ou lie egions excluded. P io o calib a ion, RSRP
p edic ions ended o unde es ima e co e age, wi h a mean
e o −5.69 dB, median o −5.37 dB, and s anda d de ia ion
o 5.71 dB. A e calib a ion, he e o dis ibu ion na owed
subs an ially, wi h a nea -ze o mean o −0.32 dB, median o
−0.13 dB, and s anda d de ia ion o 2.57 dB, con i ming he
e ec i eness o he calib a ion app oach.
20 15 10 5 0 5 10
E o (Simula ed - Empi ical) [dB]
0
2
4
6
8
10
12
Numbe o A eas
E o Be o e Calib a ion
E o A e Calib a ion
Fig. 2: Dis ibu ion o RSRP p edic ion e o s.
IV. DATA-DRIVEN RIS DEPLOYMENT
This sec ion p esen s he da a-d i en s a egy o la ge-scale
RIS deploymen on he digi al win.
A. Co e age E alua ion and UE Clus e ing
The baseline ne wo k co e age is e alua ed by deploying
BSs and dis ibu ing UEs on a g id, as desc ibed in Sec. II.
Fig. 3 shows he RSRP hea maps. UEs wi h an RSRP below
−100 dBm a e decla ed in ou age [22], esul ing in ou age
a es o 2.85%, 1.65%, and 6.07% a 2 GHz, 3.5 GHz, and
10 GHz, espec i ely. The goal o s a egic RIS deploymen
is o imp o e RSRP a hese ou age loca ions, bu deploying
a dedica ed RIS o each UE is in easible due o he la ge
numbe and dis ibu ion o ou age UEs.
To add ess his scalabili y challenge, nea by ou age UEs a e
g ouped in o clus e s so ha each RIS se es a clus e ins ead
o an indi idual UE. A hie a chical BIRCH algo i hm [23]
is employed, building compac clus e ing ea u e ees. Each
clus e is ep esen ed by
"U,
U
X
i=1
si,
U
X
i=1∥si∥2#,(5)
whe e Uis he numbe o UEs and si∈R2deno es he
2D posi ion o UE i(cen e o a ile). The h eshold Tis
a key design pa ame e ha de ines he maximum allowable
Euclidean dis ance (in me e s) be ween da a poin s wi hin a
(a) 4G deploymen a 2 GHz (b) 5G deploymen a 3.5 GHz (c) 6G deploymen a 10 GHz
Fig. 3: RSRP hea maps ac oss he conside ed u ban a ea o 4G, 5G, and 6G.
Fig. 4: Numbe o clus e s o med by he BIRCH algo i hm s.
pa ame e Tac oss he conside ed equency bands.
clus e in he BIRCH algo i hm. A smalle Tyields mo e
clus e s and RISs, while a la ge T educes he coun bu
enla ges co e age pe RIS. Fig. 4 illus a es how he clus e
coun changes wi h Tac oss equencies.
B. RIS Candida e Loca ions
This sec ion p esen s he algo i hm o he iden i ica ion
o candida e RIS loca ions. Fo each clus e o ou age UEs
p oduced by BIRCH, one RIS is deployed o enhance co e age
by aligning wi h he e lec ion poin o ele an sca e ed
ays. This s udy assumes RIS uni s can be deployed on any
building su ace wi h a alid sca e ing poin , p o iding an
uppe bound on pe o mance by igno ing p ac ical cons ain s
such as accessibili y o egula ions.
The app oach iden i ies candida e RIS si es by ac i a ing
only single-bounce sca e ing ays and inc easing ay shoo -
ings o 3×107 o dense angula sampling.
Among he ays ha each a gi en clus e cen oid, he
ones e ained a e hose ha (i) each i a e one sca e ing
e en and (ii) c ea e a alid LoS pa h om he BS ia he
sca e poin . Candida e RIS loca ions a e hen de e mined
om he co esponding sca e poin s. I mul iple candida es
a e ound, he one wi h he sho es 3D dis ance o he clus e
cen oid is selec ed. I he cen oid’s RSRP inc eases, UE-le el
gains wi hin he clus e a e assessed. A clus e is ma ked as
RIS-e ec i e i a leas 60% o i s UEs expe ience an RSRP
imp o emen . This balances RIS loca ion e iciency wi h he
compu a ional complexi y o he op imiza ion.
C. RIS Con igu a ion and BS Beam o ming
Once a candida e loca ion is selec ed, he RIS mus be
o ien ed, con igu ed, and inco po a ed in o he ne wo k.
•Each RIS is ins alled along he building wall a he
designa ed loca ion, ace ou wa ds.
•Fo each combina ion o BS, UE, and RIS loca ions, he
spa ial modula ion unc ion in Sec. II-B is compu ed.
•The se ing BS selec s he beam ha maximizes he
RSRP a he RIS loca ion by i e a ing o e all a ailable
DFT beams.
D. Re-Clus e ing and Re-Associa ion
Fo he emaining ou age UEs o which he candida e
RIS loca ions p o ed ine ec i e, wo allback s a egies a e
employed: e-clus e ing and RIS e-associa ion.
1) Re-Clus e ing: The emaining ou age UEs a e eg ouped
wi h BIRCH using a smalle h eshold T o o m mo e
compac clus e s. The new h eshold is de e mined i e a i ely
o op imize he ade-o be ween co e age and deploymen
cos . Fo each new clus e , he cen oid poin is ex ac ed, and
he me hod is eapplied o ind he RIS loca ion. Al hough
lowe ing Timp o es pe o mance, i also inc eases he numbe
o RIS uni s, aising capi al cos s o he MNO.
2) RIS Re-Associa ion: Some ou age UEs may be in a eas
whe e exis ing RIS uni s a e e ec i e bu we e no ini ially
assigned o hei clus e . To a oid edundan deploymen s,
hese UEs can be eassigned o exis ing RISs i a LoS link
is easible. The p ocess consis s o ou s eps:
1) De e mine which deployed RIS uni s main ain a LoS link
o each ou age UE.
2) Fo each po en ial RIS, iden i y he BSs ha ha e a LoS
connec ion o i .
3) F om he easible RIS op ions, selec he one closes o
he UE.
(a) 4G a 2 GHz (b) 5G a 3.5 GHz (c) 6G a 10 GHz
Fig. 5: RSRP enhancemen om deploying he maximum numbe o RIS.
(a) 4G a 2 GHz (b) 5G a 3.5 GHz (c) 6G a 10 GHz
Fig. 6: RSRP enhancemen s. numbe o deployed RIS.
4) F om he BSs linked o he chosen RIS, pick he one
nea es o i .
This p ocess e ines he associa ion o UEs, RISs, and BSs,
boos ing RSRP while a oiding u he RIS deploymen s.
V. SIMULATION RESULTS
Fo p elimina y es ing o he deploymen me hod, a e y
la ge RIS ape u e (11.24 ×11.24 m) is conside ed, wi h hal -
wa eleng h elemen spacing. Ou age UEs a e clus e ed using
BIRCH wi h a h eshold o T= 15 m.
1) 4G a 2 GHz: Ou age UEs ep esen 2.85% o he o al
popula ion. Wi h T= 15 m, 246 clus e s a e o med, each
associa ed wi h one RIS. This ini ial RIS placemen b ings
68.85% o ou age UEs back in o co e age. Re-clus e ing he
esidual UEs wi h T= 10 m eco e s an ex a 2.73%, while
RIS e-associa ion u he imp o es co e age by 6.61%, esul -
ing in a o al eco e y o 78.97% o ou age UEs. Figu e 5a
illus a es he CDF o he enhanced RSRP alues.
2) 5G a 3.5 GHz: Ou age UEs cons i u e 1.65% o he
en i e ne wo k. Using he same me hod, RIS deploymen
ini ially es o es co e age o 66.10% o hese UEs. Applying
e-clus e ing wi h T= 10 m adds ano he 3.80%, while
RIS e-associa ion con ibu es an addi ional 5.67%, esul ing
in an o e all eco e y o 75.25%. Figu e 5b shows hese
imp o emen s, accomplished wi h 197 deployed RIS uni s.
3) 6G a 10 GHz: A his equency, 393 RIS uni s a e
deployed o add ess he 6.07% o UEs ini ially in ou age. The
RIS deploymen alone es o es co e age o 46.02% o hese
UEs, wi h e-clus e ing adding ano he 6.02%. In pa icula ,
RIS e-associa ion is especially e ec i e in his scena io,
eco e ing an addi ional 12.55%, o a o al o 64.59% o
ou age UEs b ough back in o co e age, as shown in Fig. 5c.
While hese ini ial indings show no able co e age gains,
hese en ail a e y la ge RIS ape u e and many uni s, as
each clus e is assigned a sepa a e RIS. The emainde o his
sec ion e alua es how es ic ing ape u e size and deploying
RISs o only a subse o clus e s in luences pe o mance.
A. Impac o RIS Densi y
In he p e ious analysis, each clus e was associa ed wi h
a sepa a e RIS, esul ing in 246, 197, and 393 RIS uni s a
2, 3.5, and 10 GHz, espec i ely. Howe e , many clus e s a e
qui e small—o e hal con ain ewe han six UEs— e lec ing
he spa ial spa si y o ou age a eas. To add ess his, clus e s a e
now anked by hei UE coun , and RIS uni s a e inc emen ally
alloca ed o he op-Nclus e s as N a ies. Figu es 6a, 6b,
and 6c show he esul ing CDFs o RSRP o di e en alues
o N, showing ha p io i izing la ge clus e s helps balance
co e age gains agains deploymen cos s. Ne e heless, e en
wi h p io i iza ion, achie ing signi ican imp o emen s s ill
needs deploying a subs an ial numbe o RIS uni s.
Fig. 7: Pe cen age o ou age UEs eco e ed wi h di e en RIS
ape u e sizes in a ious adio deploymen s.
B. Impac o RIS Ape u es
Nex , we s udy he impac o a ying he RIS ape u e,
which di ec ly de e mines he numbe o RIS elemen s. The
elemen spacing emains ixed a hal a wa eleng h. This
e alua ion is conduc ed o he maximum numbe o RIS uni s
p e iously iden i ied, namely, 246, 197, and 393. Figu e 7
p esen s he pe cen age o ou age UEs eco e ed as a unc-
ion o he ape u e, wi h he co esponding numbe o RIS
elemen s also indica ed. The esul s show ha pe o mance is
mo e obus o a ia ions in ape u e han o changes in RIS
deploymen densi y.
VI. CONCLUSION
This wo k has in oduced an au oma ed, da a-d i en ame-
wo k o e alua ing la ge-scale RIS deploymen s in cellula
ne wo ks. By in eg a ing si e-speci ic ay acing, ou age use
clus e ing, and ay-based heu is ics, he amewo k join ly
de e mines RIS placemen , o ien a ion, con igu a ion, and BS
beam o ming. Pe o mance assessmen s in 4G, 5G, and 6G
scena ios, g ounded in a calib a ed digi al win o an u ban en-
i onmen , highligh he adeo be ween co e age gains and
deploymen cos s. Meaning ul co e age enhancemen neces-
si a es deploying many la ge RIS uni s pe squa e kilome e ,
aising ques ions abou hei cos -e ec i eness in wide-a ea
ou doo sys ems.
REFERENCES
[1] S. Islam, A. Zainab Abdulsalam, B. Anil Kuma , M. Kam ul Hasan,
R. Kolandaisamy, and N. Sa ie, “Mobile ne wo ks owa d 5G/6G:
Ne wo k a chi ec u e, oppo uni ies and challenges in sma ci y,” IEEE
open j. Commun. Soc., ol. 6, pp. 3082–3093, 2025.
[2] M. Di Renzo, A. Zappone, M. Debbah, M.-S. Alouini, C. Yuen,
J. de Rosny, and S. T e yako , “Sma adio en i onmen s empowe ed
by econ igu able in elligen su aces: How i wo ks, s a e o esea ch,
and he oad ahead,” IEEE J. Sel. A eas Commun., ol. 38, no. 11, pp.
2450–2525, 2020.
[3] C. Huang, A. Zappone, G. C. Alexand opoulos, M. Debbah, and
C. Yuen, “Recon igu able in elligen su aces o ene gy e iciency in
wi eless communica ion,” IEEE T ans. Wi eless Commun., ol. 18, no. 8,
pp. 4157–4170, 2019.
[4] W. Liu, Y. Fu, Z. Shi, and H. Wang, “When digi al win mee s 6G:
Concep s, obs acles, and esea ch p ospec s,” IEEE Commun. Mag.,
ol. 63, no. 3, pp. 16–22, 2025.
[5] C. You and R. Zhang, “Wi eless communica ion aided by in elligen
e lec ing su ace: Ac i e o passi e?” IEEE Wi eless Commun. Le e s,
ol. 10, no. 12, pp. 2659–2663, 2021.
[6] P.-Q. Huang, Y. Zhou, K. Wang, and B.-C. Wang, “Placemen op imiza-
ion o mul i-IRS-aided wi eless communica ions: An adap i e di e -
en ial e olu ion algo i hm,” IEEE Wi eless Commun. Le e s, ol. 11,
no. 5, pp. 942–946, 2022.
[7] H. Lu, Y. Zeng, S. Jin, and R. Zhang, “Ae ial in elligen e lec ing
su ace: Join placemen and passi e beam o ming design wi h 3D beam
la ening,” IEEE T ans. Wi eless Commun., ol. 20, no. 7, pp. 4128–
4143, 2021.
[8] W. Huang, W. Ding, C. Kai, Y. Yi, and Y. Huang, “Join placemen and
beam o ming design o IRS-enhanced mul iuse MISO sys ems,” IEEE
T ans. Commun., ol. 70, no. 10, pp. 6678–6692, 2022.
[9] G. B anca i, E. F. Pupo, O. Chukhno, N. Chukhno, M. Mu oni,
and G. A ani i, “Recon igu able in elligen su ace deploymen and
o ien a ion in beyond 5G mul icas ne wo ks,” in 2023 IEEE In e na-
ional Symposium on B oadband Mul imedia Sys ems and B oadcas ing
(BMSB), 2023, pp. 1–6.
[10] J. Shabanpou , V. Lene s, G. Le osey, S. T e yako , and C. Simo ski,
“Enginee ing o in elligen e lec ing su aces: Re lec ion locali y and
angula s abili y,” IEEE T ans. An ennas P opag., ol. 72, no. 5, pp.
4198–4211, 2024.
[11] S. Sandh, H. Radpou , B. Raine , M. Ho e , and T. Zemen, “Ray
acing algo i hm o econ igu able in elligen su aces,” a Xi p ep in
a Xi :2402.13034, 2024.
[12] H. Na, M. M. Taygu , and T. F. Eibe , “A mul iple huygens su ace-
based ay acing amewo k wi h GPU accele a ion,” IEEE T ans.
An ennas P opag., ol. 72, no. 1, pp. 183–196, 2024.
[13] C. You, B. Zheng, W. Mei, and R. Zhang, “How o deploy in elligen
e lec ing su aces in wi eless ne wo k: BS-side, use -side, o bo h
sides?” J.Commn.Ne , ol. 7, no. 1, pp. 1–10, 2022.
[14] Z. Kang, C. You, and R. Zhang, “IRS-aided wi eless elaying: Deploy-
men s a egy and capaci y scaling,” IEEE Wi eless Commun. Le e s,
ol. 11, no. 2, pp. 215–219, 2022.
[15] J. Hoydis, F. A. Aoudia, S. Camme e , M. Nimie -Da id, N. Binde ,
G. Ma cus, and A. Kelle , “Sionna RT: Di e en iable ay acing o
adio p opaga ion modeling,” a Xi :2303.11103, 2023.
[16] 3GPP Technical Repo 38.901, “S udy on channel model o equencies
om 0.5 o 100 GHz (Release 16),” Dec. 2019.
[17] H. Holma, H. Viswana han, and P. Mogensen, “Ex eme massi e MIMO
o mac o cell capaci y boos in 5G-ad anced and 6G,” Nokia Bell Labs,
Whi e pape , 2021.
[18] D. López-Pé ez, N. Pio esan, and G. Ge aci, “Capaci y and powe
consump ion o mul i-laye 6G ne wo ks using he uppe mid-band,”
a Xi p ep in a Xi :2411.09660, 2024.
[19] V. Degli-Espos i, E. M. Vi ucci, M. D. Renzo, and S. A. T e yako ,
“Re adia ion and sca e ing om a econ igu able in elligen su ace: A
gene al mac oscopic model,” IEEE T ans. An ennas P opag., ol. 70,
no. 10, pp. 8691–8706, 2022.
[20] E. M. Vi ucci, M. Albani, S. Kod a, M. Ba bi oli, and V. Degli-
Espos i, “An e icien ay-based modeling app oach o sca e ing om
econ igu able in elligen su aces,” IEEE T ans. An ennas P opag.,
ol. 72, no. 3, pp. 2673–2685, 2024.
[21] J. Hoydis, F. A. Aoudia, S. Camme e , F. Euchne , M. Nimie -Da id,
S. T. B ink, and A. Kelle , “Lea ning adio en i onmen s by di e en-
iable ay acing,” IEEE T ans. Mach. Lea n. Commun. Ne w, ol. 2,
pp. 1527–1539, 2024.
[22] Tel onika Ne wo ks Wiki, “RSRP and RSRQ.” [Online]. A ailable:
h ps://wiki. el onika-ne wo ks.com/ iew/RSRP_and_RSRQ
[23] T. Zhang, R. Ramak ishnan, and M. Li ny, “BIRCH: An e icien da a
clus e ing me hod o e y la ge da abases,” in ACM SIGMOD In ’l Con .
Managemen o Da a, 1996, p. 103–114.
