Beam alignment for mmWave and THz: systematic review

Madhekwana, Sundi e; Usman, Muhammad A slan; Ayyub, Ah isham; Poli is,
Ch is os
A icle — Published Ve sion
Beam alignmen o mmWa e and THz: sys ema ic e iew
Telecommunica ion Sys ems
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: Madhekwana, Sundi e; Usman, Muhammad A slan; Ayyub, Ah isham; Poli is,
Ch is os (2025) : Beam alignmen o mmWa e and THz: sys ema ic e iew, Telecommunica ion
Sys ems, ISSN 1572-9451, Sp inge US, New Yo k, NY, Vol. 88, Iss. 3,
h ps://doi.o g/10.1007/s11235-025-01318-7
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/323473
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h p://c ea i ecommons.o g/licenses/by/4.0/
Telecommunica ion Sys ems (2025) 88:87
h ps://doi.o g/10.1007/s11235-025-01318-7
Beam alignmen o mmWa e and THz: sys ema ic e iew
Sundi e Madhekwana1,2 ·Muhammad A slan Usman1·Ah isham Ayyub1·Ch is os Poli is1
Accep ed: 28 May 2025
© The Au ho (s) 2025
Abs ac
This compac s udy in es iga es published esea ch on beam managemen amewo ks o la ge an enna a ays in mmWa e and
e ahe z (THz) sys ems. The sys ema ic li e a u e e iew (SLR) ocused on beam alignmen and ini ial access amewo ks o
mmWa e and THz. The in es iga ion unco e ed 596 ele an a icles, including jou nals and con e ence pape s om di e en
sou ces using de ined c i e ia, leading o 73 inal s udies pos - il e ing, ollowing ou SLR-de ined 3-s age sc eening p ocess.
We explained he axonomy o mmWa e and THz beam alignmen amewo ks by classi ying hem in o beam sweeping-
based, con ex in o ma ion-based, comp essi e sensing-based, and Machine Lea ning/A i icial In elligence (ML/AI)-based
amewo ks, as well as In eg a ed Sensing and Communica ion (ISAC)-based amewo ks, which combine ISAC wi h o he
amewo ks. In addi ion, we p esen ed esea ch gaps by iden i ying and analyzing he limi a ions o he cu en amewo ks.
Las ly, challenges and possible oppo uni ies o u u e esea ch we e highligh ed.
Keywo ds Beam alignmen ·MmWa e ·THz ·Comp essi e sensing ·Beam sweeping ·ML/AI ·Con ex in o ma ion ·
Beam managemen
1 In oduc ion
1.1 Backg ound
The nex -gene a ion wi eless ne wo k echnology, 6G, has
he po en ial o e olu ionize he da a use expe ience [1]
as we know i om cu en echnologies like 4G and 5G.
The as bandwid h a ailabili y in he mmWa e and THz
egions [2] pa es he way o new applica ions ha we e
p e iously un hinkable, such as holog aphic communica-
ion and applica ions in he me a e se domain. Howe e , he
main d awbacks and challenges a e due o i s small co e -
age ange. The mmWa e and THz signals su e high pa h
BSundi e Madhekwana
[email p o ec ed]
BMuhammad A slan Usman
[email p o ec ed]
Ah isham Ayyub
[email p o ec ed]
Ch is os Poli is
[email p o ec ed]
1Facul y o Enginee ing, Compu ing and he En i onmen ,
Kings on Uni e si y London, London KT1 2EE, UK
2Baseband Pla o m So wa e Depa men , Nokia, Ulm 89081,
Ge many
losses. To add ess hese challenges, beam o ming echnolo-
gies wi h la ge an enna a ays o mul i-an enna sys ems a e
u ilized o ex end he co e age and enhance da a h oughpu
[3]. Mul i-an enna sys ems, also known as Mul iple-Inpu
Mul iple-Ou pu (MIMO) sys ems [4], u ilize mul iple an en-
nas a bo h he ansmi e and ecei e o signi ican ly
enhance communica ion pe o mance. This echnology is
pa icula ly c ucial o beam alignmen in Te ahe z (THz)
communica ions. The sho wa eleng hs in he THz band
enable a high numbe o an ennas in a small space, esul -
ing in na ow beams ha pose challenges in main aining a
s able connec ion be ween he ansmi e and ecei e . By
employing mul iple an ennas, MIMO sys ems can gene -
a e highly di ec ional beams ha can be p ecisely s ee ed
owa ds he in ended ecei e . This beam o ming capabil-
i y acili a es e icien signal ansmission and ecep ion,
e ec i ely o e coming he limi a ions o na ow beams and
ensu ing eliable communica ion wi hin he THz spec um.
Howe e , he use o mul i-an enna sys ems a hese equen-
cies makes he beams highly di ec ional, so he beams need
o be aligned be o e any link can be es ablished. The beam
alignmen p ocess es ablishes and main ains a communica-
ion link be ween a ansmi e and a ecei e by aligning
he ansmi e and ecei e beams. This p ocess becomes
pa icula ly c ucial in he mmWa e (30 GHz -300) GHz)
0123456789().: V,- ol 123
87 Page 2 o 53 S. Madhekwana e al.
[5] and THz (0.1-10THz) [6] spec um ange because o he
cha ac e is ics o hese equencies, including high pa h loss,
ulne abili y o blockages, and suscep ibili y o a mosphe ic
condi ions [7]. In lowe - equency wi eless sys ems, omni-
di ec ional an ennas o simple sec o ized an ennas ha e been
used, as he wide beam pa e ns can accommoda e mobili y
and p o ide co e age o e a la ge a ea. Howe e , using na -
ow beam o ming an ennas in mmWa e and THz sys ems
enables highly di ec ional ansmission, which o e s se -
e al ad an ages, including inc eased spa ial euse, imp o ed
capaci y, and educed in e e ence. Beam alignmen plays
a undamen al ole in achie ing he bene i s o di ec ional
ansmission in wi eless communica ion sys ems. I in ol es
wo main p ocesses: beam aining and beam acking.
•Beam aining - This is he i s s ep in beam align-
men , whe e he ansmi e and ecei e explo e di e en
beam di ec ions o ind he bes beam pai wi h he high-
es ecei ed signal s eng h [8]. The ansmi e ypically
sends a se ies o aining signals wi h di e en beam
di ec ions, using a codebook o de ined p o ocol, and he
ecei e measu es he signal s eng h o each beam di ec-
ion [9,10] and selec s he bes one. Th ough eedback o
explici signalling, he ansmi e adjus s i s beam di ec-
ion o align wi h he ecei e ’s beam di ec ion, which
yields he s onges ecei ed signal. Codebook and p o-
ocol design in THz sys ems is essen ial o achie e obus
beam aining and o e come challenges such as beam
squin , whe e beams op imized o he cen e equency
LoS gain a o he equencies, causing misalignmen
and pe o mance deg ada ion. [11] in oduced a spe-
cialized wideband codebook ha ensu es each beam
main ains su icien gain ac oss he en i e bandwid h,
enabling low-complexi y beam aining ha a oids ull
channel es ima ion and complex eal- ime op imiza ion.
By maximizing he wo s -case wideband beam gain and
ailo ing spa ial zones, his design deli e s consis en sig-
nal quali y e en a co e age edges while simpli ying
beam alignmen and educing aining o e head. Fu -
he mo e, THz massi e MIMO sys ems equi e p ecise
beam alignmen o comba high- equency challenges
and se e e pa h losses, ye adi ional na ow beams a e
imp ac ical unde dominan LoS condi ions and weak
pilo signals. [12] in oduces algo i hms ha gene a e
wide-beam codewo ds wi h b oad angula co e age, la
main lobes, and supp essed side lobes, enabling e icien
beam aining wi hou ull CSI and educing eal- ime.
•Beam acking -[2,13]: This is pe o med o con-
inuously upda e he beam di ec ions as en i onmen al
condi ions change beyond he ini ial alignmen . This is
pa icula ly impo an in mobile scena ios o en i on-
men s wi h mo ing obs acles. Many di e en me hods
can be used o beam acking. [14] and [15] gi e such
examples o me hods ha can be employed o Beam
aining and beam acking. [14] in oduces a uni ied
3D beam aining and acking p ocedu e o THz com-
munica ions ha bypasses he need o eal- ime CSI.
The au ho s used a hie a chical 3D codebook wi h wide
beams o b oad co e age and na ow beams o high
gain, enabling a g id-based aining p o ocol ha e i-
cien ly iden i ies he op imal beam pai wi h a ewe
es s han an exhaus i e sea ch. Once he beam pai is
se , a g id-based hyb id acking p o ocol employing wo
dynamic modes o e ine and p edic he beam. Whe eas
[15] in oduced an exhaus i e me hod o es ima e angles
o bo h di ec LoS and In elligen Re lec ing Su ace
(IRS)- e lec ed links, due o he high complexi y in THz
massi e MIMO sys ems, i p oposes a low-complexi y
coope a i e beam aining scheme. This app oach com-
bines a pa ial sea ch a he IRS–scanning key angle
di e ences in he sine space–wi h a e na y- ee hie a -
chical sea ch a he BS and use using wo specialized
codebooks o e icien ly na ow down he bes na ow-
beam pai . Once aining es ablishes an op imal beam
pai , beam acking con inuously upda es he pai based
on ecei ed signal quali y, ensu ing sus ained high beam-
o ming gain in dynamic. Beam acking helps main ain
he alignmen o he communica ion link, compensa ing
o beam misalignmen due o ac o s such as use mobil-
i y, blockages, o an enna misalignmen .
Beam alignmen echniques can a y depending on sys em
equi emen s, channel condi ions, and a ailable esou ces.
Va ious amewo ks and me hods ha e been p oposed o
achie e e icien and accu a e beam alignmen [16]. Some o
hese amewo ks include codebook-based echniques, com-
p essed sensing-based app oaches, machine lea ning-based
me hods, hyb id beam o ming, channel es ima ion-based
echniques, pilo -based beam alignmen , and adap i e beam
alignmen . Howe e , none o hese amewo ks ha e been
s anda dized, implemen ed in eal-wo ld sys ems, o p o en
adequa e o la ge-scale an enna a ays. The pe o mance
e alua ion o beam alignmen echniques in ol es se e al
me ics, such as signal- o-noise a io (SNR), bi e o a e
(BER), h oughpu , beam misalignmen angle, beam aining
o e head, ene gy e iciency, and la ency. Compa a i e analy-
sis o di e en beam alignmen echniques helps assess hei
s eng hs, limi a ions, and ade-o s ac oss a ious scena ios
and deploymen con ex s [10].
Challenges in beam alignmen a ise om he cha ac e is-
ics o mmWa e and THz equencies, including mul i-pa h
p opaga ion, in e e ence, ha dwa e cons ain s, mobili y,
and en i onmen al ac o s [2,10]. While mmWa e and THz
channels sha e se e al simila cha ac e is ics, he e ec s
o pa h loss, a mosphe ic abso p ion, and sca e ing a e
signi ican ly mo e p onounced in THz equencies [17].
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 3 o 53 87
Consequen ly, THz communica ion sys ems equi e mo e
sensi i e, apid, and complex alignmen echniques han
mmWa e due o he na ow beams and ex eme signal a en-
ua ion.
These challenges necessi a e esea ch and de elopmen
e o s o o e come obs acles and imp o e he obus ness
and e iciency o beam alignmen echniques. As wi eless
communica ion sys ems con inue o e ol e, in eg a ing beam
alignmen amewo ks wi h ad anced ne wo king echnolo-
gies, such as 5G/6G and he In e ne o Things (IoT), becomes
essen ial. C oss-laye op imiza ion, secu i y conside a ions,
machine lea ning ad ancemen s, and s anda diza ion e o s
u he con ibu e o he con inuous imp o emen o beam
alignmen in wi eless communica ion sys ems. O e all,
beam alignmen plays a c i ical ole in enabling eliable
and e icien communica ion in high- equency wi eless sys-
ems. Ongoing esea ch and ad ancemen s in his ield aim
o add ess challenges, op imize pe o mance, and acili-
a e deploymen . Ou mo i a ion o his compac e iew
o exis ing beam alignmen amewo ks is o explo e unex-
ploi ed oppo uni ies and po en ial challenges. The sys em-
a ic e iew aims o con ibu e o:
•The knowledge o he design o alignmen amewo ks.
•Imp o emen o beam alignmen amewo ks.
The s udy in ends o answe he ollowing esea ch ques-
ions (RQs), as shown in Table 1.
1.2 P incipal con ibu ions and o ganiza ion
We analyzed a ious exis ing beam managemen esea ch
wo ks ha co e beam alignmen amewo ks wi h a s ong
ocus on la ge an enna a ays, mmWa e, and THz sys ems.
We in es iga ed he s udies in e ms o hei con ibu ion o
he a eas depic ed in he igu e 1: The pu pose o his s udy
is o:
•De elop a axonomy o beam alignmen amewo ks and
ca ego ize exis ing beam alignmen amewo ks.
•Compa e hem and iden i y hei limi a ions.
•Examine hei p ospec s o s anda diza ion
•E alua e hei sui abili y in mee ing he challenges aced
in beam alignmen and he a eas depic ed in he ig-
u e 1. The conduc ed esea ch examined he exis ing
beam alignmen amewo ks in ecognized jou nals and
con e ences.
The de eloped axonomy ca ego izes he beam alignmen
solu ions and desc ibes each amewo k o gi e an o e iew
o exis ing di icul ies and he la es exis ing echniques.
All he selec ed amewo ks ha e been e alua ed in e ms
o beam o ming a chi ec u e, beam alignmen amewo ks,
Fig. 1 Focus o esea ch a ea Con ibu ions
En i onmen , mobili y, numbe o beams, pe o mance me -
ics, key indings, e c. Addi ionally, he limi a ions o he
amewo ks we e iden i ied as well as he exis ing esea ch
gaps.
Sec ion 2 gi es an o e iew o he p e ious wo ks. Sec ion
3 ca ego izes he esul s o exis ing solu ions and echniques
ha help sol e beam alignmen challenges in o i e di e -
en classes. Sec ion 4 discusses he la es esea ch on beam
alignmen amewo ks in de ail. In his sec ion, we e iew
he esea ch wo ks wi hin he de ined amewo ks, p o iding
exis ing solu ions and e alua ing he iden i ied esea ch p ob-
lems. Sec ion 5 o e s a quali a i e compa ison and ou lines
he limi a ions o he amewo k ypes. Sec ion 6 p esen s
u u e di ec ions and oppo uni ies. Finally, we conclude in
Sec ion 7.
1.3 Findings o e iew
In his sec ion, we p o ide an o e iew o he indings
o he in es iga ion. Acco ding o [19–21], beam align-
men amewo ks can be implemen ed in a ious ways:
exhaus i e sea ching, ou -o -band in o ma ion exchange ia
senso s moun ed on he ansmi e and ecei e nodes, com-
p essi e sensing, machine lea ning, and In eg a ed Sensing
and Communica ion (ISAC)-based beam alignmen . These
app oaches can be used in nume ous ways o de elop beam
alignmen (BA) amewo ks ha add ess speci ic p ob-
lema ic scena ios. Table 2ca ego izes he BA amewo ks
iden i ied based on hei implemen a ion ea u es. To acil-
i a e his classi ica ion, esea che s decided o c ea e he
ollowing ca ego ies:
•Beam sweeping
•Con ex In o ma ion
•Comp essi e Sensing
•ML/AI
•ISAC beam alignmen
Fo his s udy, beam sweeping, con ex in o ma ion, com-
p essi e sensing, ML/AI, and ISAC beam alignmen we e
conside ed. We conside ed he beam sweeping echnique
as any me hod ha uses ei he exhaus i e beam sea ching,
adap i e sweeping me hod, o hie a chical beam sweeping
123
87 Page 4 o 53 S. Madhekwana e al.
Table 1 Resea ch Ques ions
Ques ion Mo i a ion
RQ1: Which beam selec ion o alignmen amewo ks exis o
mmWa e and THz?
The nex -gene a ion wi eless ne wo k sys ems will use he as
bandwid h a ailable in mmWa e and THz egions wi h highly
di ec ional beams o compensa e o he high pa h losses su e ed by
such signals. Howe e , he T ansmi e and ecei e ’s di ec ional
beams mus be aligned [18]. The beam sweeping amewo k [16]
was s anda dized o beam alignmen in 5G NR wi h he assump ion
o a maximum o 64 beams. Howe e , nex -gene a ion wi eless
sys em beyond 5G is an icipa ed o use e y la ge an enna a ays
wi h up o 1024 an enna elemen s and e y di ec ional beams. The
beam-sweeping amewo k will no be adequa e due o he high
beam alignmen la ency, high ene gy consump ion and high
o e head ha g ows wi h an inc easing numbe o beams. Al hough
he e is a subs an ial body o li e a u e on his opic, u he esea ch
is needed. This ques ion is c ucial o es ablishing exis ing
amewo ks and he challenges hey ha e add essed, de e mining he
s a us quo, iden i ying knowledge gaps, and explo ing po en ial
s anda diza ion solu ions.
RQ2: How do hese F amewo ks compa e o each o he ? This ques ion aims o compa e he di e en amewo ks o iden i y
common issues ha ha e been add essed and o highligh
di e ences. Addi ionally, gaps a e iden i ied by showing he issues
ha ha e no been add essed by any o he es ablished amewo ks.
RQ3: Wha a e he challenges and limi a ions o hese amewo ks? Challenges in beam alignmen a ise om he cha ac e is ics o
mmWa e and THz equencies, including mul i-pa h p opaga ion,
in e e ence, ha dwa e cons ain s, mobili y, and en i onmen al
ac o s. These challenges necessi a e esea ch and de elopmen
e o s o o e come obs acles and imp o e he obus ness and
e iciency o beam alignmen echniques. Al hough many o he
challenges ha e been add essed by di e en app oaches, some o
hese will spill o e o 6G [10]. Speci ically, he challenges and
limi a ions ha a e no add essed by exis ing amewo ks need o be
iden i ied h ough hese ques ions.
RQ4: Wha a e he u u e di ec ions and esea ch oppo uni ies? A e es ablishing he exis ing solu ions and hei limi a ions, he
u u e di ec ions and esea ch oppo uni ies will be appa en , he eby
answe ing his ques ion.
echnique. Any echnique ha uses in-band o ou -o -band
signalling o posi ion, di ec ion, o o he in o ma ion o he
ansmi e o ecei e was conside ed a con ex in o ma-
ion echnique. Comp essi e sensing echniques a e hose
ha exploi spa si y cha ac e is ics o he mmWa e o THz
channel. App oaches ha use any o he machine lea ning
echniques we e classi ied as ML/AI. App oaches ha com-
bine any o he p ima y amewo ks wi h In eg a ed Sensing
and Communica ion (ISAC) we e classi ied as ISAC-based
beam alignmen o hyb id.
The summa y o esea ch dis ibu ion pe ca ego y is
depic ed and summa ized in igu e 2. Table 2shows he s ud-
ies ha all in o each ca ego y and he dis ibu ion o he
selec ed a icles acco ding o he amewo k ca ego iza ion.
Machine lea ning has he highes numbe o samples ound.
Figu es 2and 3illus a e he dis ibu ion o publica-
ions by amewo k and yea o publica ion, espec i ely.
An ini ial classi ica ion was conduc ed o highligh key clus-
e s wi hin eme ging beam alignmen echniques. These
Table 2 Beam Alignmen Classi ica ion
Classi ica ion Re e ences No.o Re s
Comp essi e Sensing [22–31]10
Beam Sweeping [10,32–40]10
Con ex In o ma ion [41–55]16
Machine Lea ning [56–82]26
ISAC Beam Alignmen [83–93]11
ca ego ies add ess a ious challenges by le e aging and
cap u ing he unique cha ac e is ics o mmWa e and THz sig-
nals. No ably, he da a shows a signi ican inc ease in esea ch
ac i i y om 2022 onwa d. I is also impo an o empha-
size ha amewo ks in ol ing Machine Lea ning (ML) and
ISAC ha e become dominan in ecen esea ch, indica ing
hei g owing impo ance in his ield.
123

Beam alignmen o mmWa e and THz: sys ema ic... Page 5 o 53 87
Fig. 2 Dis ibu ion o A icles by F amewo k
Fig. 3 Dis ibu ion o A icles by yea
2 Rela ed wo ks
The e is a as body o ecen li e a u e ela ed o ou wo k
ha has in es iga ed beam managemen and alignmen o
mmWa e and THz bands. These s udies include [17,94–
100], and a e desc ibed as ollows:
[17] and [94] highligh ed majo challenges in beam man-
agemen o mmWa e, discussing he ends and issues
behind hese challenges. Some o he iden i ied challenges
include ine icien beam sweeping, mul i-panel op imiza ion,
high mobili y, and DL/UL beam co espondence ailu es.
The au ho s p oposed po en ial solu ions such as comp es-
si e sensing, machine lea ning (ML), ein o cemen lea ning
(RL), mul i-base s a ion coope a ion, and obus codebook
design. Howe e , hei wo k p ima ily ocused on chal-
lenges ela ed o s anda dized 5G NR beam managemen
based on beam sweeping, and echnical implemen a ion
was no ho oughly add essed. [95] Wang e al. p o ided a
comp ehensi e o e iew o mmWa e echnologies in sce-
na ios such as 5G ne wo ks, ixed wi eless access (FWA),
ehicle- o-e e y hing (V2X), and indoo ne wo ks. The
wo k highligh ed challenges encoun e ed in hese scena ios,
including high pa h loss, blockage, a mosphe ic abso p-
ion, and ha dwa e limi a ions. Al hough he pape o e ed
a b oad o e iew o mmWa e, i did no co e he THz
bands o aspec s like scalabili y, ene gy e iciency, and in e -
ope abili y, and i lacked de ailed echnical dep h in beam
managemen . [96] Yi e al. p esen ed an in-dep h o e iew
o beam aining and acking algo i hms o mmWa e in
scena ios such as 5G ne wo ks, FWA, V2X, and indoo and
u ban en i onmen s. The s udy highligh ed challenges such
as high pa h loss, blockage, mobili y, and eal- ime acking
and discussed how blind beam aining, p io in o ma ion-
aided beam aining, and ML-based mechanisms can educe
aining o e head. The au ho s ocused mainly on mmWa e,
wi h less emphasis on b oade 5G/6G con ex s, such as THz
bands. Howe e , aspec s like low la ency, beam managemen ,
scalabili y, and ene gy e iciency we e no ully add essed.
[97] Khan e al. p o ided ex ensi e co e age o ML ech-
niques speci ic o beam managemen o bo h mmWa e and
THz bands.
The s udy explo ed how o add ess challenges like da a
a ailabili y, compu a ional complexi y and model gene al-
iza ion using supe ised lea ning, ein o cemen lea ning,
and ede a ed lea ning. Scena ios such as au onomous ehi-
cles, sma ci ies, UAVs, and indoo ne wo ks we e also
conside ed. Despi e he ho ough explo a ion o ML appli-
ca ions in beam managemen , non-ML app oaches we e no
co e ed. [98] Xue e al. ocused on ML-based beam manage-
men o bo h mmWa e and THz bands, examining cu en
achie emen s wi h an emphasis on a i icial in elligence (AI),
in eg a ed sensing and communica ion (ISAC), and econ ig-
u able in elligen su aces (RIS). They also highligh ed he
use o mul i-agen collabo a ion echniques like ede a ed
lea ning and ans e lea ning. Despi e he ex ensi e analysis
o ML-based beam managemen , he pape did no p o ide
a b oade analysis o beam managemen , as i neglec ed
ad ancemen s in non-ML-based app oaches.
[99] Häge e al. ocused on design aspec s ha op imize
beam aining o mobile use s in 6G mmWa e ne wo ks.
The au ho s examined he impac o mobili y bo h du ing
and a e beam aining, p oposing a ade-o app oach in
he design o beam aining sys ems. This app oach balances
he quali y o he communica ion link, he delay in oduced
by he beam aining p ocess, and he co e age a ea. The
design should inco po a e a quali y-o -se ice (QoS)-awa e
beam sea ch algo i hm ha dynamically adjus s he beam
aining p ocess acco ding o he equi ed link quali y. Such
algo i hms enable a balance be ween speed and obus ness,
which is pa icula ly c ucial o mobile use s who need quick
ye eliable link es ablishmen . Howe e , he s udy p ima -
ily concen a ed on design aspec s, wi h less emphasis on
123
87 Page 6 o 53 S. Madhekwana e al.
b oad heo e ical amewo ks o scena ios in ol ing non-
mobile use s. [100] p esen ed a new app oach o educing
mul i-use in e e ence in IRS-assis ed massi e MIMO sys-
ems. I uses a space-o hogonal scheme ha decouples he
p ecode and decode design. One pa cancels in e e ence
using ze o- o cing echniques. The o he pa maximizes he
a e using singula alue decomposi ion (SVD) and powe
alloca ion. To lowe he complexi y o join beam o ming,
he pape in oduces wo IRS phase-shi design me hods:
wa e - illing segmen ma ching (WSM) and phase i e a i e
e olu ion (PIE). These me hods balance pe o mance wi h
compu a ional e iciency. The scheme achie es nea -op imal
sum- a e pe o mance wi h much lowe complexi y han
an exhaus i e sea ch. The pape ’s s eng hs include b eak-
ing down a complex p oblem in o simple subp oblems and
using IRS-pa ial ze o- o cing o imp o e spa ial mul iplex-
ing gains. Howe e , i elies on accu a e channel models and
dominan LoS assump ions, which may limi i s use in sca -
e ing en i onmen s. Scalabili y emains a conce n o e y
la ge sys ems.
While exis ing esea ch has made signi ican con ibu-
ions in speci ic a eas o ins ance, ML, we belie e he e
is a gap in p o iding comp ehensi e co e age o beam align-
men and managemen s a egies. Ou wo k aims o ill
his gap by o e ing a b oade analysis o ad ancemen s in
beam managemen , explo ing bo h ML/AI-d i en echniques
and adi ional app oaches. This wide scope allows us o
add ess a ious me hodologies and inno a ions, highligh -
ing p og ess in all possible di ec ions.
3 Beam alignmen amewo ks and
axonomy
The selec ed a icles we e classi ied by cons uc ing me a-
da a o ms o o ganize he de ails and include ou de ined
ema ks o hose a icles. We collec ed only he me ada a
ha add essed ou esea ch ques ions, which included he
i le, esea ch objec i es, me hodology, beam o ming a chi-
ec u e, beam alignmen amewo ks en i onmen , mobili y,
numbe o beams, pe o mance me ics, e c. Figu e 2 shows
he dis ibu ion o he publica ions acco ding o amewo ks.
An ini ial classi ica ion was made conce ning ou esea ch
opic o demons a e ou s anding clus e s o he eme ging
beam alignmen echniques ca ego ies sol ing di e en chal-
lenges by u ilizing and cap u ing cha ac e is ics and aspec s
o he mmWa e and THz signals. AI/ML domina es he
anks.
3.1 Beam sweeping
We classi ied he published li e a u e on beam sweeping in o
exhaus i e beam sweeping and hie a chical beam sweeping.
3.1.1 Exhaus i e beam alignmen
In an exhaus i e beam-sweeping amewo k, bo h he ans-
mi e and ecei e scan h ough all he beams o iden i y
he beam wi h he maximum powe and mos sui able o
communica ion [10]. The exhaus i e beam sweeping was
s anda dized by he 3GPP o 5G NR in Release 15 [10,94].
The beam alignmen o e head and delay inc ease wi h he
numbe o beams when mmWa e and THz a e in ope a ion.
These challenges igge he need o imp o emen in his
amewo k.
3.1.2 Hie a chical and adap i e beam sweeping
The esea ch in his a ea in es iga es how he s anda d-
ized beam-sweeping app oach can be imp o ed in e ms
o o e head, delay, and ene gy e iciency. The hie a chical
amewo k’s beam sweeping is done adap i ely. A i s ,
wide beams a e swep h ough, and a beam wi h highe
powe is hen subsequen ly swep wi h na owe beams o
de e mine he op imal beam accu a ely [34]. Howe e , hie -
a chical beam sweeping is a common o m o adap i e beam
sweeping, allowing he sys em o p og essi ely ocus om
wide o na ow beams. The implemen a ion o beam sweep-
ing can be codebook-based o ha dwa e-based, e.g., whe e
some an ennas a e swi ched o o p oduce la ge beams, and
hen all an ennas a e used o na ow beams. The main d aw-
backs o he amewo k a e he inaccu acy caused by he
wide beams ha may no de ec mobile s a ions a he cell
edges [101] and he dependence on DL/UL beam co espon-
dence, which in oduces in lexibili y and blockage [10].
3.2 Con ex in o ma ion
As highligh ed in beam sweeping-based BA, he o e head
and delay g ow wi h he numbe o beams used by he
ansmi e and ecei e . Al hough he hie a chical app oach
educes o e head and delay me ics, i is suscep ible o e o s
and may no be sui able o applica ions like ul a- eliable
low-la ency communica ion (URLLC). The con ex in o -
ma ion app oach in ends o educe hese me ics u he by
using ou -o -band in o ma ion ha is communica ed o bo h
he ecei e and he ansmi e [41] o in o m each o he o
hei bea ings and whe e o sea ch o he beams om each
o he . The con ex in o ma ion amewo k esea ch can be
u he classi ied by he con ex in o ma ion used o imple-
men he amewo k, as shown in igu e 4and he subsec ions
below:
3.2.1 Loca ion-awa e beam alignmen
The loca ion-awa e amewo ks exploi he posi ion in o -
ma ion o he base s a ion and mobile s a ion [41]. The base
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 7 o 53 87
Fig. 4 Beam Alignmen Taxonomy
s a ion o mobile s a ion de i es he loca ion in o ma ion
om GPS de ices ha a e in eg a ed in o hose en i ies.
Once hese en i ies ob ain hei posi ions, hey beacon hem
owa ds he ansmi e o ecei e so ha hey know which
a ea o sea ch o sui able beams.
3.2.2 Came a-based alignmen
The came a-based amewo k uses he came a as a senso
o images o objec s in he en i onmen o he ansmi e
o ecei e o de e mine he objec sending he signal and i s
di ec ion [19].
3.2.3 Ligh de ec ion and anging
Ligh de ec ion and anging (LiDAR) echnology is used
o acqui e he con ex ual in o ma ion o he ansmi e and
ecei e [19,102]. The in o ma ion is used o de e mine he
di ec ion o he beam.
3.2.4 Ou -o -band measu emen s
Ou -o -band measu emen s is any ype o signal ha can be
sen o he ansmi e o ecei e o p edic he di ec ion o
he sou ce o he signal [20,101].
Theo e ically, a con ex in o ma ion-based app oach o
beam alignmen has a high po en ial o educe la ency,
o e head, and ene gy consump ion. Howe e , he main sho -
coming o such a amewo k is ha he ex a equipmen
needed o he in-band o ou -o -band communica ion makes
he equipmen (base s a ion and mobile s a ion) mo e expen-
si e. Hence, he indus y may be eluc an o implemen such
an app oach.
3.3 Comp essi e sensing (CS)
CS is a echnique o signal p ocessing ha makes i possible
o econs uc images o signals om a compa a i ely small
numbe o samples o measu emen s [103,104]. CS elies on
he spa si y p inciple and op imiza ion algo i hms o signal
eco e y. Spa se signals ha e a small numbe o non-ze o
componen s, while a signi ican numbe o he componen s
a e ze os o e y close o ze o. I is o en quan i ied using he
L1-no m o he signal, which is he sum o he absolu e al-
ues o i s componen s. Comp essi e sensing has applica ions
in a ious ields, including image p ocessing, medical imag-
ing, and da a comp ession. The equa ion below o malizes
comp essi e sensing ma hema ically:
y=x=a=a (see [105] o de ails)
Whe e y is he low dimensional measu emen s, = andom mea-
su able ma ix, = ans o m basis, a=spa se coe icien s
123
87 Page 8 o 53 S. Madhekwana e al.
The econs uc ion o he signal p oblem will be sol ed as
an op imiza ion p oblem o he o m:
minal1,(1)
such ha :
y=a(see [105] o de ails)
Comp essi e sensing educes he aining o e head o
beam alignmen by le e aging he spa ial clus e ing o mul-
ipa hs in he channel [20]. The bes beam is selec ed by
p obing and measu ing a small se o andomly selec ed
beams om he whole codebook.
Al hough he comp essi e sensing echnique may educe
o e head and ene gy consump ion by p obing ewe beams,
i is complex due o he need o p ecise modelling and
econs uc ion algo i hms and may ace s anda diza ion and
in e ope abili y challenges.
3.4 Machine lea ning (ML) and a i icial in elligence
(AI) (ML/AI))
ML and AI a e used o enhance ini ial beam alignmen
in wi eless communica ion sys ems [106]. ML models a e
ained o es ima e and p edic wi eless channel cha ac-
e is ics, such as signal s eng h, mul ipa h p opaga ion,
and in e e ence. These channel cha ac e is ics a e c i ical
o achie ing op imal beam alignmen . ML models lea n
om his o ical da a, adap ing o en i onmen al ac o s like
changes in su oundings, in e e ence, o obs acles du ing
wi eless communica ion. ML/AI sys ems dynamically adjus
beam pa ame e s based on changing condi ions, op imiz-
ing he ini ial alignmen . Fu he mo e, weigh s and phase
pa ame e s can be adjus ed o as e beam alignmen . [107]
p esen ed he s a us quo, oppo uni ies, and challenges o
deep lea ning, including Con olu ional Neu al Ne wo ks
(CNNs). Rein o cemen lea ning ains sys ems o make
decisions abou ini ial beam alignmen , lea ning op imal
s a egies h ough ial and e o . ML and AI enable wi e-
less communica ion sys ems o achie e as e , mo e accu a e
ini ial beam alignmen , ensu ing adap abili y o di e se en i-
onmen al condi ions and enhancing o e all eliabili y and
pe o mance. The machine lea ning and a i icial in elligence
(ML/AI) esea ch wo k o beam alignmen is u he classi-
ied acco ding o he machine lea ning model ypes, namely
supe ised, ein o cemen , and deep lea ning, as depic ed in
igu e 4.
3.4.1 Rein o cemen Lea ning
Acco ding o [108], Rein o cemen lea ning is a me hod
wi hin AI sys ems whe e he sys em lea ns om i s expe i-
ences o imp o e i s decision-making. This app oach empha-
sizes lea ning h ough ial and e o , ma king a signi ican
shi in he capabili ies o a i icial in elligence sys ems.
3.4.2 Supe ised lea ning
Supe ised lea ning in ol es using p e iously collec ed da a,
along wi h labeled in o ma ion, o p edic ou comes and
h ough a aining p ocess, machine lea ning algo i hms
de elop a unc ion o an icipa e ou pu alues. This sys-
em can hen gene a e p edic ions o new inpu da a a e
su icien aining, as de ined by he au ho s o [109]. By
compa ing p edic ed ou comes wi h ac ual esul s, e o s
a e iden i ied and used o e ine he model. In con as ,
unsupe ised lea ning is used when da a lacks labels o clas-
si ica ions, aiming o unco e unde lying pa e ns wi hou
p ede ined ou pu s by analyzing and de i ing unc ions om
unlabeled da a, hus disco e ing hidden pa e ns h ough
obse a ion and analysis [109].
3.4.3 Deep lea ning
Deep lea ning uses compu a ional models consis ing o
nume ous laye s o unde s and da a by o ming ep esen a-
ions wi h a ious le els o complexi y using neu al ne wo ks
[110].
The main d awback o he ML/AI app oach is i s high
complexi y due o he loads o da a needed o aining he
models and he need o ongoing model imp o emen s as
he e y expensi e ha dwa e necessa y o unning in e ence
engines.
3.5 ISAC beam alignmen amewo ks
In eg a ed Sensing and Communica ion (ISAC) in eg a es
wi eless communica ion and sensing in one sys em. I s goal
is o enhance e iciency by sha ing spec um, ha dwa e, and
signals be ween he wo asks. ISAC is pa icula ly aluable
in ad anced ne wo ks like 5G and 6G beam alignmen ame-
wo ks, whe e high da a a es and eal- ime en i onmen al
awa eness a e c ucial. ISAC can be combined wi h di e -
en o he beam alignmen amewo ks o o m subca ego ies.
The h ee key subca ego ies o ISAC a e ISAC and Machine
Lea ning (ML), ISAC and Beam Sweeping, and ISAC and
Con ex In o ma ion.
3.5.1 ISAC and machine lea ning
Machine Lea ning (ML) enhances ISAC by using da a mod-
els o imp o e sensing and communica ion. Fo example,
[31] in eg a es ISAC wi h compu e ision and ex ended
Kalman il e s (EKF). Resul ing in imp o ed ini ial access
(IA) and beam acking in UAV ne wo ks The dual iden i y
associa ion (DIA) app oach u he enhances beam- acking
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 15 o 53 87
sho e swi ch delays esul in highe h oughpu . The s udy
also sugges s ha he cu en RSRP-based beam selec ion
migh no always choose he bes beam, indica ing a need
o eassess beam managemen c i e ia in 5G mmWa e ne -
wo ks. Finally, [39] p esen s a beam alignmen s a egy o
THz communica ion sys ems, using an i e a i e p ocess o
op imize beam di ec ions and maximize da a a es. I ack-
les challenges like se e e pa h loss and he need o p ecise
alignmen in high- equency en i onmen s. Despi e i s ben-
e i s in da a a e and accu acy, he app oach aces challenges
wi h compu a ional complexi y and eal- ime easibili y. The
s udy ad ances beam managemen in nex -gene a ion wi e-
less ne wo ks, especially in he con ex o eme ging THz
echnologies.
Beam sweeping exhibi s a ious challenges and consid-
e a ions. Fi s ly, i in oduces la ency, which, while ole able
in some applica ions, becomes c i ical in low-la ency sce-
na ios. This la ency issue is compounded by he o e head
incu ed in e ms o ime and equency esou ces, impac ing
he ne wo k’s o e all e iciency. Mo eo e , beam sweeping
is sensi i e o en i onmen al changes, wi h ac o s like obs a-
cles and in e e ence a ec ing alignmen accu acy and speed,
po en ially leading o subop imal pe o mance in dynamic
se ings. Scalabili y is a conce n, pa icula ly in scena ios
wi h many use s o de ices, as managing mul iple simul a-
neous beams and ensu ing e icien alignmen o each use
poses challenges. Fu he mo e, he cons an adjus men o
beams h ough sweeping can con ibu e o inc eased ene gy
consump ion, aising conce ns abou he ne wo k’s sus ain-
abili y and ene gy e iciency.
Finally, he implemen a ion and managemen o beam
sweeping p ocedu es, especially in millime e-wa e and
THz equencies, in ol e complexi y, equi ing sophis ica ed
algo i hms, signalling mechanisms, and ha dwa e, which
could escala e deploymen and main enance cos s.
4.2 Con ex in o ma ion amewo ks
In wi eless communica ion, con ex in o ma ion encom-
passes de ails abou he en i onmen , including de ice
loca ions, obs acle p esence, de ice mo emen , and o e -
all ne wo k condi ions, which is aluable o op imizing
communica ion pa ame e s. Beam alignmen is essen ial in
millime e-wa e and THz communica ion, u ilizing di ec-
ional beams h ough beam o ming o enhance signal ocus
and eliabili y. A con ex -awa e beam alignmen app oach
in eg a es addi ional con ex ual in o ma ion, such as p ecise
de ice loca ions, obs acle awa eness, and de ice dynam-
ics (especially o mobile de ices), acili a ing p oac i e
adjus men s o op imal beam di ec ion and in e e ence
educ ion. The con ex in o ma ion app oach o en employs
adap i e algo i hms o moni o and analyze con ex ual ac-
o s, dynamically adjus ing beam o ming pa ame e s o
be e communica ion pe o mance. Fu he mo e, con ex-
ual da a plays a c ucial ole in Machine Lea ning (ML).
I p o ides addi ional in o ma ion ha helps models make
mo e accu a e p edic ions. This enables in o med decisions
abou beam alignmen and adap i e esponses o changing
condi ions. Tables 5and 6show he collec ed me ada a o
he selec ed Con ex In o ma ion amewo ks esea ch s ud-
ies ha add essed ou esea ch ques ions, which include
yea , i le [Re e ence], Resea ch objec i es, Me hodology,
Beam o ming A chi ec u e, Beam Alignmen amewo ks
En i onmen , Mobili y, Numbe o beams, Pe o mance me -
ics, e c.
[42] explains he ansmission model o beam o ming
using a Bu le ma ix and pa ame e s cha ac e izing adio
spa ial p opaga ion. I in oduces he beam alignmen p ob-
lem as a mul i-hypo hesis es ing issue, discussing a ious
de ec ion amewo ks, including ixed-leng h de ec ion and
a iable-leng h de ec ion (SCET), wi h SCET p esen ed as
an adap i e and obus imp o emen o e ixed-leng h me h-
ods. The pape de ails an expe imen al se up a TU D esden
o mmWa e expe imen a ion wi h hyb id beam o ming,
e alua ing beam selec ion me hods in di e en scena ios.
The esul s demons a ed he adap abili y and e iciency o
SCET and SCT in single-pa h and mul i-pa h scena ios com-
pa ed o ixed-leng h me hods. The pape also highligh ed
he adap abili y and e iciency o he SCET amewo k in
ealis ic mmWa e scena ios. The s udy acknowledged i s
obus ness and supe io i y o e ixed-leng h me hods. [41]
add esses he beam alignmen challenges in millime e-wa e
(mmWa e) communica ion, whe e high da a ansmission
and educed in e e ence a e c ucial. The p oposed algo i hm
le e ages he loca ion in o ma ion o mobile use s and po en-
ial e lec o s o e icien ly align beams in mmWa e-band
communica ions. The key idea is o enable he base s a ion
and mobile use o wo k oge he o scan a limi ed numbe o
beams wi hin he e o bounda y o he noisy loca ion in o -
ma ion. This ini ial se o beams guides he sea ch o u u e
beams, educing alignmen o e head, powe consump ion,
and ime esou ces. The algo i hm’s e ec i eness is demon-
s a ed h ough simula ion esul s, demons a ing imp o ed
sys em pe o mance e en when dealing wi h noisy loca ion
da a. The pape compa es he p oposed app oach wi h exis -
ing me hods, highligh ing i s adap abili y, e iciency, and i s
educed complexi y. O e all, he esea ch aims o enhance
he beam alignmen p ocess in mmWa e and THz commu-
nica ion, pa icula ly in mobile en i onmen s, by exploi ing
loca ion-awa e coo dina ion.
[43] explo es "F equency-Dependen Beam o ming”, le e -
aging ixed and a iable leng h es s wi h a ocus on mmWa e
communica ion sys ems. I in oduces an exhaus i e sea ch
(ES) based es ing app oach, including a Va iable Leng h
Tes ing Scheme (VLT) u ilizing ES and a Gene alized Likeli-
hood Ra io Tes (GLRT) conside ing c oss-co ela ion alues
123

87 Page 16 o 53 S. Madhekwana e al.
and noise a iance es ima es, along wi h a e mina ion h esh-
old de e mined by he likelihood o a alse ala m. The
F equency Dependen Beam o ming (FDB) sec ion employs
a linea sys em model wi h Rician ading, adap ing GLRT
o hypo hesis es ing, and in oducing a leng h-dependen
h eshold based on an F-dis ibu ion. Simula ion esul s com-
pa e Fixed Leng h Tes ing (FLT) wi h ue ime delay o
di e en scena ios, emphasizing he ad an ages o a i-
able leng h es ing, pa icula ly in highe SNR egimes.
The conclusion highligh s he ime e iciency o he p o-
posed equency-dependen beam o ming, achie ing powe
loss measu es in medium o high SNR, and sugges s u he
esea ch o imp o ing ime e iciency and educing powe
losses in challenging condi ions.
[44] ocuses on beam alignmen s a egies o mmWa e
massi e MIMO sys ems o enhance communica ion pe o -
mance, explo ing a ious codebooks, including o hogonal,
hie a chical, and equency-dependen ypes. I discusses
he op imal beam o me choice, ocusing on he o hogo-
nal codebook a e a speci ic numbe o obse a ions (N)
and emphasizing he minimum equi ed samples o he
Leas Squa es (LS) solu ion. Simula ions conside ed a chan-
nel wi h mul iple pa hs and a ecei e wi h a 64-elemen
uni o m linea a ay. The pape compa es ixed and a iable-
leng h beam selec ion echniques using he Bu le codebook
and in oduces he Sequen ial Compe i ion and Elimina ion
Tes (SCET) o adap i e es leng h and obus ness. The
hie a chical codebook (MLVL) and F equency-Dependen
Beam o ming (FDB) codebook a e e alua ed, showing ha
a iable-leng h de ec ion, especially wi h SCET, enhances
adap i i y, esul ing in a e gains and delay educ ion. Di e -
en codebook ypes impac de ec ion pe o mance di e en ly,
wi h o hogonal codebooks bene i ing om elimina ion
mechanisms and equency-dependen codebooks elimina -
ing he need o beam swi ching. [45] in es iga ed he
ini ial beam acquisi ion/alignmen in mmWa e communica-
ion sys ems. The au ho s p oposed a aining p o ocol ha
uses a equency-selec i e beam p obing ne wo k o scan
all beam o me s om a pa icula codebook. This app oach
exploi s he spa se na u e o mmWa e channels, mapping
each beam o me o di e en equencies. The pape com-
pa es wo ha dwa e designs o pa allel beam aining and
e alua es hei easibili y. The esul s demons a ed ha
equency-selec i e beam p obing ou pe o ms exhaus i e
sea ch in e ms o e ec i e ansmission a es in apidly
changing en i onmen s, c ucial o mmWa e sys ems wi h
la ge beam o ming codebooks and sho cohe ence imes.
The s udy highligh s he impo ance o educing iming o e -
head in communica ion sys ems, especially in scena ios wi h
high mobili y and spa ial esolu ion. The p oposed app oach
demons a es ad an ages in ime e iciency and ansmission
a es. I o e s po en ial bene i s in spa se sca e ing en i-
onmen s common in mmWa e communica ions. The pape
sugges s explo ing con inuous pilo subca ie s du ing da a
ansmission o imp o ed channel acking o u u e wo k.
[46] in oduces an inno a i e apid access me hod o
mmWa e sys ems in cellula ne wo ks, aiming o educe
he complexi y o beam aining. This is achie ed h ough
he u iliza ion o e o-di ec i e a ays (RDA) and a Zado -
Chu (ZC) sequence-based p eamble. The p oposed me hod
in ol es wo s ages: in he i s pa , he base s a ion (BS)
pe o ms ansmi (Tx) beam sweeping wi h ac i a ed RDA,
and i a mobile s a ion (MS) is de ec ed in any o he beam
di ec ions, he RDA e ansmi s he signal o he BS. In
he second pa , he BS ansmi s he p eamble h ough he
chosen Tx beam, and he MS pe o ms Rx beam sweep-
ing o measu e he SNR. The pape demons a ed a no able
dec ease in complexi y when compa ed o he adi ional
exhaus i e sea ch app oach. Simula ion esul s con i med
he echnique’s abili y o accu a ely de ec beam alignmen
pa ame e s, highligh ing i s po en ial in add essing he com-
plex challenges o beam aining. The p oposed app oach,
le e aging e o-di ec i e a ays and Zado -Chu sequences,
o e s a p omising solu ion o e icien ini ial access in
mmWa e ne wo ks.
[47] ackles he beam alignmen challenge in mmWa e
communica ion ne wo ks by p oposing a g oundb eaking
amewo k aided by Ul a-wideband (UWB). The inno a-
ion lies in in eg a ing UWB echnology o es ima e op imal
angles o beam alignmen , elimina ing he necessi y o
exhaus i e space sea ching and add essing la ency and com-
plexi y issues. The Mul i-F equency MUSIC (MF-MUSIC)
algo i hm expands he adi ional MUSIC algo i hm in o he
equency domain, enabling accu a e angle es ima ion wi h
a small numbe o UWB an ennas. The pape ecognizes
p ac ical challenges wi h limi ed an ennas on Comme cial-
O -The-Shel (COTS) de ices and p o ides a solu ion,
enhancing i s eal-wo ld applicabili y. While comp ehensi e
nume ical e alua ions and compa isons a e s ong, hey lack
eal-wo ld alida ion and ely on assump ions.
The main con ibu ions o [48] include a no el app oach o
coo dina e compu a ion using accele ome e o a ion angles,
add essing azimu h e o s. The pape p oposes an e icien
associa ion o mobile s a ions (MS) wi h base s a ions (BS)
h ough a ansmi ed uplink con ol signal, conside ing
powe , esponse ec o s, and he geome ic channel. Two
b oadcas ing app oaches, single and mul i-b oadcas ing,
acili a e da a dissemina ion o mul iple MS use s. Pe -
o mance e alua ion indica es ad an ages such as signi i-
can ly educed access imes, high success a es, and ene gy
e iciency in compa ison o exis ing schemes. Limi a ions
include po en ial challenges in eal-wo ld implemen a ion
and he de elopmen al s age o mmWa e echnology. In hei
u u e wo k, hey sugges explo ing ou ages caused by block-
age and mobili y.
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 17 o 53 87
Table 5 Con ex In o ma ion based F amewo k esea ch s udies
Re Objec i es Me hod A chi e c u e F ame wo k En i on men Mobili y Beam N .
[42] E alua es he pe o mance
o a sequen ial
compe i ion and
Elimina ion Tes (SCET) -
a a iable leng h beam
selec ion amewo k
Expe im en ally Analog BF Con ex
in o ma ion
based
LOS and NLOS N/A 64
[41] De e mines he op imal
beams by u ilising he
use ’s posi ion, he base
s a ion’s loca ion, o
ce ain obs uc ions on he
channel
Simul a ion Analog BF Con ex
in o ma ion
based
NLOS N/A 64
[43] Achie es beam alignmen
by equipping he
ansmi e wi h a
equency-dependen
beam o me ha can
simul aneously e i y all
spa ial angles and,
consequen ly, any
beam o me ha migh
come om a compa able
codebook.
Simula ion Hyb id BF F equency-
dependen
beam o m-
ing
NLOS N/A 16,32, 64
[44] Compa es he beam
alignmen accu acy and
acquisi ion ime o h ee
ypes o o hogonal bu le
ma ix s a egy,
hie a chical/mul ile el
sea ch and
equency-dependen
beam o ming amewo k
using bo h ixed and
a iable leng h
Simul a ion Hyb id BF Con ex
in o ma ion
based
NLOS N/A 8,16, 32,64
123
87 Page 18 o 53 S. Madhekwana e al.
Table 5 con inued
Re me ics Scena io F equen cy Me i s Deme i s Beam Pa e n
[45] Used a s a egy ha simul aneously
p obes all beams om a
codebook o de e mine AoA o
AoD and exploi s he spa si y o
mmWa e channels. The me hod
maps e e y beam o me o he
codebook o mul iple equencies
using a equency-selec i e beam
p obing ne wo k and spec al
analysis a he ecei e o
de e mine a ou able
beam o me s AoA o AoD
Simul a ion Analog BF F equency-selec i e
beam p obing
NLOS N/A 8
[46] P oposed a e odi ec i e a ay
(RDA)-based ini ial access
me hod o lowe o e head
associa ed wi h he beam aining
o e head. The RDA ansmi s he
incoming signal back o he
sou ce wi h a di ec i i y ha
ma ches he size o he an enna
a ay.
Simul a ion Analog BF Re odi e c i e a ay
(RDA)
NLOS N/A BS:256, MS:8
[47] P oposed a mmWa e
communica ion a chi ec u e wi h
join ly si ua ed UWB an ennas o
ind he op imal angles o
mmwa e beam alignmen .
Simul a ion Analog BF co-loca ed UWB
an ennas
NLOS N/A -
[48] When a new MS en e s a cell, i
scans he channel o a beacon
indica ing he bea ing o he BS,
b oadcas by a nea by MS ha
has accomplished beam
associa ion. Then, using a digi al
compass, he loca ion o he MS
is e lec ed in he adjus ed
coo dina es o he BS.
Simul a ion digi al BF P obing beacon NLOS N/A -
[42] SNR and La ency Cellula , UAV, FWA mmWa e Reduces aining
o e head, adap able o
a ying channel
condi ions
Compu a ional
complexi y and
ha dwa e impai men s
impac accu acy
O hogonal beams using a Bu le Ma ix se up
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 19 o 53 87
Table 5 con inued
Re me ics Scena io F equen cy Me i s Deme i s Beam Pa e n
[41] Th ough pu Cellula , V2X, UAV mmWa e Reduced alignmen
o e head, enhanced
alignmen accu acy
Compu a ional o e head
equi es p ecise
loca ion da a
Adap i e beams based
on loca ion da a
[43] SNR Cellula , UAV mmWa e, sub-THz Reduces beam
alignmen ime, highe
SNR, adap able o
channel condi ions
Compu a ional
complexi y in
mul i-pa h scena ios,
high ha dwa e cos
Adap i e beam pa e ns
basedon imedelay
ac oss di e en
equencies
[44] SNR Cellula , V2X mmWa e Reduces beam- aining
o e head, adap able o
SNR condi ions
Compu a ional
complexi y, eliance
on p io knowledge o
SNR
O hogonal beams ia
Bu le Ma ix,
F equency-dependen
beams
[45] SNR and h oughpu Cellula , V2V, FWA mmWa e Reduces beam
alignmen ime, be e
pe o mance in apidly
changing
en i onmen s
Inc eased ha dwa e
complexi y due o
delay elemen s
F equency-dependen
beam pa e ns u ilizing
OFDM modula ion
[46] SNR,DP Cellula mmWa e Fas e beam alignmen
wi h educed
complexi y, accu a e
de ec ion
Highe ha dwa e cos ,
complexi y in
managing
e odi ec i e a ays
Re odi ec i e beam
pa e ns
[47] SNR, Th oughpu , AC Cellula mmWa e Signi ican imp o emen
in AoA es ima ion
accu acy wi h ewe
an ennas, educ ion o
communica ion delays
Pe o mance d ops in
NLoS scena ios,
equi es a calib a ion
s ep o
Highly di ec ional
[48] SNR, EE, ST Cellula mmWa e Fas access imes, high
success a es, educed
ene gy consump ion
Highe cos , complex
beam o ming
s uc u e, azimu h il
e o s
Omni-di ec ional and
di ec ional
BF= Beam Fo ming, RI=Rank indica o , PMI=p ecoding ma ix Index, CQI =channel quali y indica o , RSRP = Re e ence Signal Recei ed Powe , RSS = Recei ed Signal S eng h, BDA = beam
de ec ion accu acy, MDP = miss de ec ion p obabili y, DP = De ec ion P obabili y, C = complexi y, LOS = line o sigh , NLOS = Non line o Sigh , SNR = Signal o noise a io, UAV = Unmanned
ae ial ehicle, FWA = ixed wi eless access, V2X = Vehicle o any hing, MS = mobile s a ion, BS = bases a ion
123
87 Page 20 o 53 S. Madhekwana e al.
Table 6 Con inued Con ex In o ma ion based F amewo k esea ch s udies
Re Objec i es Me hod A chi e c u e F ame wo k En i o nmen Mobil i y Beam N .
[49] The s udy used s ochas ic geome y and S uden ’s
-dis ibu ion s a is ics analy ical model o examine
he e ec o uplink powe con ol on he downlink
beam alignmen e o s in an mmWa e ne wo k as
hey depend on se e al link ac o s
Anal y ical Analog BF Uplink powe
con ol
NLOS N/A -
[50] The Au ho s p oposed an ene gy-con olled THz
pulse-le el beam swi ching called TRPLE. The
impulse adio, which emi s pulses las ing
em oseconds, allows beam di ec ion con ol a he
pulse le el a he han he packe le el. TRPL
amewo k sol es he pulse- o-pulse and symbol- o
symbol sepa a ion pa ame e s o maximize he da a
a e while mee ing he in e e ence equi emen s
Simul a ion Analog BF pulse-le el
beam
swi ching
NLOS N/A -
[61] The s udy e alua ed he pe o mance o
beam o ming le e aging space di ision mul iple
access (SDMA) depending on he design, i.e. a ay
pa e n, numbe o a ay elemen s and angle o
sepa a ion. Signal o in e es and signal o no
in e es scena ios we e de ined o di e en base
s a ions’ angles o sepa a ion and numbe o
an enna elemen s.
Simula ion Analog BF loca ion-
awa e
LOS N/A 128
[51] E icien ini ial access and wi eless powe | ans e o
ene gy-neu al de ices
Simula ion Analog BF Loca ion LOS, NLOS N/A a ia ble
[52] Reduce beam alignmen ime o mmWa e
communica ions using hi d-pa y came a da a
Expe i men al Analog BF came a LOS, NLOS Yes 31
[53] Reduce aining o e head and enhance beam o ming
accu acy
Simula ion,
Expe i
men al
Hyb id BF Loca ion LOS, NLOS Yes Va ia ble
[54] Use CV and ML o beam managemen , educing
CSI eliance
Simula ion Hyb id BF Came a LOS Yes a ia ble
[55] C ea e a Digi al Twin (DT) o imp o e beam
managemen
Expe i
men al,
Simula ion
Hyb id BF Came a LOS, NLOS Yes No speci ied
[49] SNR, CP Cellula mmWa e Reduces UE
powe con-
sump ion,
main ains
downlink
SNR in
NLOS
Deg ada ion
in LOS
SNR wi h
FPC, beam
alignmen
e o s in
LOS
Di ec ional,
sec o -
based
123

Beam alignmen o mmWa e and THz: sys ema ic... Page 21 o 53 87
Table 6 con inued
Re me ics Scena io F equen cy Me i s Deme i s Beam Pa e n
[50] DR, SNR Small cells,
WLAN
THz High da a a e (167
Gbps), e ec i e o
long dis ances (up
o 20m), e icien
mul iplexing
High complexi y in
beam con ol,
limi ed by pulse
sepa a ion,
molecula
abso p ion
Highly di ec ional
wi h na ow beams
[61] SNR Cellul a ,
D2D, 5G
Ul a-Dense
Ne wo ks
mmWa e Signi ican
in e e ence
supp ession,
enhanced spa ial
mul iplexing o
ul a-dense
ne wo ks
Requi es accu a e
posi ioning,
compu a ional
complexi y
inc eases wi h
mo e dense
en i onmen s
Di ec ional wi h
spa ial sepa a ion
[51] PG, Powe
Densi y
Indoo WPT mmWa e E icien powe
deli e y using
beam di e si y;
scalable o dense
en i onmen s
Dependence on
en i onmen
awa eness; beam
sweeping inc eases
se up ime and
ene gy
consump ion
Di ec ional
[52]AC,Time
Consump-
ion
Indoo , Sma
Ci y
mmWa e Reduces beam
alignmen ime by
up o 1/50
compa ed o
adi ional me hods
Reliance on came a
a ailabili y and
placemen ,
p ocessing
o e head
Di ec ional
[53] Th oug hpu ,
T aining
o e head
Cellul a mmWa e Reduces eal- ime
channel aining,
imp o es
communica ion
a es e en wi h
mode a e loca ion
e o s
Relies on loca ion
accu acy,
en i onmen al
changes a ec
pe o mance
Di ec ional
[54]AC,
O e head
Cellul a ,
UAV
mmWa e Reduces o e head
and imp o es
eal- ime beam
alignmen
Depends on came a
da a and
en i onmen al
condi ions
Di ec ional
[55] AC Cellul a mmWa e Highe accu acy in
beam alignmen
educes o e head
Came a dependency,
high compu a ional
o e head
Digi al win wi h
adap i e mapping
BF= Beam Fo ming, RI=Rank indica o , PMI=p ecoding ma ix Index, CQI =channel quali y indica o , RSRP = Re e ence Signal Recei ed Powe , RSS = Recei ed Signal S eng h, BDA = beam
de ec ion accu acy, MDP = miss de ec ion p obabili y, DP = De ec ion P obabili y, C = complexi y, LOS = line o sigh , NLOS = Non line o Sigh , SNR = Signal o noise a io, UAV = Unmanned
ae ial ehicle, FWA = ixed wi eless access, V2X = Vehicle o any hing, MS = mobile s a ion, BS = base s a ion, AC = accu acy, PG = pa h gain, WPT = wi eless powe ans e
123
87 Page 22 o 53 S. Madhekwana e al.
[49] in es iga es he impac o uplink powe con ol
on downlink beam alignmen e o s in mmWa e cellu-
la ne wo ks, employing s ochas ic geome y and S uden ’s
-dis ibu ion s a is ics. The au ho s c ea ed an analy ical
amewo k ha conside s link pa ame e s, uplink powe
con ol, and SNR co e age. The esul s demons a e a
subs an ial educ ion in use equipmen (UE) powe con-
sump ion h ough e ec i e uplink powe con ol wi hou
comp omising downlink SNR co e age. The s udy highligh s
ac ional powe con ol o uplink pilo signal ansmission,
an unde explo ed a ea in mmWa e sys ems. Key con i-
bu ions include explo ing he in luence o uplink powe
con ol on downlink beam alignmen e o s, in oducing an
analy ical model, and showcasing signi ican UE powe con-
sump ion educ ion wi h minimal e ec s on downlink SNR
co e age. Acknowledging limi a ions ela ed o assump-
ions, he pape sugges s u u e esea ch di ec ions in ol ing
eal-wo ld alida ion, dynamic powe con ol s a egies, and
ex ensions o addi ional scena ios o a comp ehensi e unde -
s anding o powe con ol mechanisms in eme ging wi eless
communica ion echnologies. In summa y, he esea ch p o-
ides aluable insigh s in o op imizing powe consump ion
in mmWa e cellula ne wo ks while ensu ing sa is ac o y
downlink SNR co e age.
[50] ocuses on ad ancing Te ahe z (THz) communi-
ca ion by de eloping and op imizing a Medium Access
Con ol (MAC) p o ocol amewo k. The au ho s aim o
ailo he MAC amewo k o he downlink in THz ne -
wo ks, conside ing THz channel cha ac e is ics and empha-
sizing pulse-le el beam-swi ching using la ge an enna a ays
based on g aphene o mac o-scale communica ion. The
p oposed amewo k aims o es ablish a "pseudo wi ed"
wi eless link h ough ocused ansmission ia line-o -sigh
o e lec ed pa hs. The pape explo es key componen s,
including he op imiza ion o In e -Pulse Sepa a ion (IPS)
and In e -Symbol Sepa a ion (ISS), e ising ansmission
scheduling, discussing beam acquisi ion s a egies, add ess-
ing beam swi ching delay, and p esen ing nume ical esul s
o bo h LOS and NLOS.
[61] in es iga es he applica ion o beam o ming and
spa ial mul iplexing in he con ex o 5G Ul a-Dense Ne -
wo ks (UDN), pa icula ly emphasizing he ansi ion o
mmWa e. The s udy highligh s adap i e beam o ming (AB)
and spa ial signal p ocessing as c ucial elemen s in o e -
coming mmWa e limi a ions. The pape explo es adap i e
beam o ming scena ios, add essing challenges like beam
alignmen and p esen ing analog, digi al, and hyb id me h-
ods. The esea ch e alua es spa ial mul iplexing pe o mance
h ough simula ions employing he Leas Mean Squa es
(LMS) algo i hm. I ocuses on he in e e ence supp es-
sion a e (ISR) unde a ious condi ions. Key con ibu ions
include assessing loca ion-awa e beam o ming and de elop-
ing a simula ion model o unde s and he impac o angula
sepa a ion and an enna elemen s on in e e ence supp es-
sion. In conclusion, he esea ch signi ican ly ad ances he
unde s anding o beam o ming and spa ial mul iplexing in
5G UDN, p o iding aluable insigh s o u he explo a ion
and p ac ical implemen a ion.
Con ex -based beam alignmen is an op imiza ion ech-
nique in wi eless communica ion sys ems ha u ilizes con-
ex ual in o ma ion such as use loca ions, en i onmen al
condi ions, in e e ence le els, and ne wo k dynamics o
dynamically adjus beam pa ame e s. While his app oach
aims o enhance sys em pe o mance, eliabili y, and e i-
ciency, i aces se e al common limi a ions. These include
he complexi y in oduced o sys em design, challenges in
achie ing eal- ime adap a ion, eliance on accu a e con ex-
ual in o ma ion, conce ns ega ding in e ope abili y ac oss
di e se sys ems, and po en ial secu i y and p i acy issues
associa ed wi h sensi i e con ex ual da a.
To add ess hese challenges, common ecommenda ions
include he de elopmen o obus algo i hms, in eg a ion
wi h a i icial in elligence and machine lea ning o imp o ed
adap abili y, and collabo a i e s anda diza ion e o s o
ensu e consis ency, eal-wo ld alida ion o assess e ec i e-
ness in a ious en i onmen s, and he implemen a ion o
s ong secu i y measu es, such as enc yp ion and au hen i-
ca ion, o sa egua d sensi i e in o ma ion.
4.3 Comp essi e sensing (CS) amewo ks
CS is employed o ake ad an age o he mmWa e chan-
nels’ in insic spa si y. The spa si y cha ac e is ics allow o
he accu a e eco e y o dominan pa hs wi h ewe aining
esou ces. I also educes he aining o e head by p o id-
ing a means o eco e essen ial channel in o ma ion wi h a
spa se se o measu emen s. Addi ionally, comp essi e sens-
ing can be combined wi h o he echniques (e.g., Kalman
Fil e ing [25]) o adap o he dynamic na u e o ime- a ying
channels, ensu ing accu a e beam alignmen o e changing
condi ions. Tables 7and 8show a lis o s udies included
in his sec ion. The selec ed s udies in Tables 7and 8col-
lec i ely p o ide a con empo a y unde s anding o he s a us
quo in Comp essi e Sensing (CS) beam alignmen . These
s udies showcase he po en ial o CS-based amewo ks in
add essing challenges associa ed wi h mmWa e communi-
ca ion. No ably, he in eg a ion o CS wi h Kalman Fil e ing,
as demons a ed by Kun-Hsien Lin, p esen s a p omising
app oach o accu a e channel eco e y wi h educed o e -
head. E an Kho dad’s explo a ion o beam alignmen in
dynamic en i onmen s emphasizes he impo ance o adap -
abili y and a swi ching mechanism be ween beam aining
and acking o balance accu acy and o e head.
[25]’s wo k u he unde sco es he po en ial o CS-
Kalman Fil e ing in eg a ion in achie ing accu a e beam
alignmen wi h signi ican ly lowe Mean Squa ed E o
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 23 o 53 87
(MSE) compa ed o adi ional me hods. Howe e , he s a-
us quo also e eals c i ical aspec s ha wa an a en ion.
The impac o CS-based amewo ks in highly mobile sce-
na ios and hei obus ness unde ex eme channel dynamics
equi e u he in es iga ion. Real-wo ld deploymen chal-
lenges, such as ha dwa e cons ain s and synch oniza ion
issues, a e ecognized bu no ex ensi ely discussed. Adap -
abili y o di e se en i onmen s and comp ehensi e analysis
o CS pe o mance unde a ying signal- o-noise a ios
(SNR) emain a eas o mo e in-dep h in es iga ion. The
ade-o be ween accu acy and o e head educ ion, as well
as compa a i e assessmen s wi h al e na i e beam alignmen
echniques, need mo e de ailed conside a ion. In conclusion,
he s a us quo o CS-based beam alignmen is p omising, wi h
no able ad ancemen s in in eg a ion s a egies and pe o -
mance imp o emen s. Ye , he e exis s a need o con inued
esea ch o add ess challenges in mobili y, eal-wo ld imple-
men a ion, en i onmen al adap abili y, SNR sensi i i y, and
a comp ehensi e unde s anding o ade-o s in p ac ical
mmWa e communica ion scena ios.
4.4 ML/AI amewo ks
The applica ion o machine lea ning (ML) and a i icial
in elligence (AI) p inciples in beam alignmen in ol es a
da a-d i en app oach, u ilizing his o ical da a o ain models
on ele an ea u es, such as signal s eng h and en i on-
men al ac o s. Supe ised lea ning is employed o asks
wi h labelled da ase s, while unsupe ised lea ning aids in
unde s anding da a pa e ns. Rein o cemen lea ning ains
sys ems o make beam alignmen decisions h ough ial and
e o . ML models adap in eal ime o dynamic en i onmen-
al changes, and op imiza ion algo i hms adjus beam o ming
pa ame e s o op imal pe o mance. Deep lea ning, pa icu-
la ly Con olu ional Neu al Ne wo ks (CNNs), may be used
o image p ocessing and pa e n ecogni ion. The p inciples
also include he c ea ion o adap i e sys ems ha dynami-
cally adjus beam pa ame e s based on changing condi ions
and p edic i e analy ics o o ecas ing po en ial misalign-
men s. Toge he , hese p inciples enhance he au oma ion,
op imiza ion, and adap abili y o beam alignmen p ocesses,
imp o ing e iciency and eliabili y in a ious echnologies.
Tables 9,10,11 and 12 show he collec ed me ada a o
he selec ed ML/AI amewo k-based esea ch s udies ha
add essed ou esea ch ques ions, which include yea , [Re -
e ence], Resea ch objec i es, Me hodology, Beam o ming
A chi ec u e, Beam Alignmen amewo ks En i onmen ,
Mobili y, Numbe o beams, Pe o mance me ics, e c. The
wo k o [56] explo es beam alignmen in mmWa e massi e
MIMO sys ems using ML. The p oposed amewo k using
ML, in oduces a neu al ne wo k (NN) ained o line in
simula ed en i onmen s ha uses pa ial beams o o ecas
he beam dis ibu ion ec o . Based on he NN p edic-
ions, all use s’ beams a e hen simul aneously aligned. The
amewo k demons a ed supe io o e all pe o mance, con-
side ing o al aining ime slo s and
spec al e iciency. The app oach elimina es he need o
his o ical use loca ion da a du ing NN aining, con ibu -
ing o educed sys em o e head. The esea ch indica es ha
ML echniques can signi ican ly enhance beam alignmen by
imp o ing e iciency and eliabili y.
[57] p oposes a new app oach o o e come he challenges
o beam alignmen in mmWa e sys ems, which a e c ucial o
add essing p opaga ion issues a high equencies. The pape
in oduces a dis ibu ed beam-alignmen s a egy u ilizing an
ad e sa ial mul i-a med bandi s (MAB) app oach, le e ag-
ing a single bi o eedback indica ing whe he he SINR
su passes a p ede e mined cu o . The main b eak h ough
is a edesigned ewa d unc ion ha d aws inspi a ion om
he mmWa e channels’ spa se s uc u e, e icien ly ein o c-
ing good beam di ec ions and penalizing poo ones. The
p oposed algo i hm, MEXP3, builds upon he exponen ial
weigh s (EXP3) algo i hm, demons a ing heo e ical gua -
an ees o eg e . The esul s o he mmWa e nume ical
simula ions wi h use mobili y showcase he supe io pe -
o mance o MEXP3 when compa ed o EXP3 and o he
policies, emphasizing i s adap abili y o dynamic wi eless
en i onmen s.
[58] sugges s BsNe , a deep lea ning-based beam selec-
ion echnique. The mo i a ion s ems om he challenges
posed by he na ow and highly di ec ional beam cha ac e -
is ics o mmWa e bands, making beam alignmen cos ly and
challenging. BsNe add esses his by ea ing beam align-
men as an image econs uc ion p oblem, le e aging deep
neu al ne wo ks (DNNs) o beam domain image econ-
s uc ion. The app oach in ol es o line aining and online
p edic ion s ages, signi ican ly dec easing he online beam
selec ion o e head. The pape in oduces he concep o
eigen-beam ex ac ion and uses a lea ning-based me hod
o BsNe aining, achie ing ema kable esul s in e ms o
scalabili y, obus ness, and pe o mance. Simula ion esul s
demons a e BsNe ’s supe io i y o e exis ing me hods, such
as Modi ied Rosenb ock’s Di ec Sea ch (MRDS) and local
lea ning-based clus e ing algo i hm wi h ea u e selec ion
(LLC s), o e ing high spec al e iciency while minimiz-
ing he sea ch o e head. The p oposed app oach exhibi s
p omising po en ial o e icien beam selec ion in mmWa e
communica ion scena ios, demons a ing adap abili y o a -
ious en i onmen s and scena ios. [59] p oposes a machine
lea ning solu ion o beam managemen in 5G NR using
geoloca ion side in o ma ion. The me hod models he map-
ping be ween he geoloca ions o use equipmen (UE) and
he beams o cells ha se e hem in a mul iuse , mul i-
cell en i onmen by using suppo ec o machines (SVMs).
Fu he mo e, a mul iuse scheduling app oach is shown ha
educes eal- ime channel s a e in o ma ion (CSI) eedback
123
87 Page 24 o 53 S. Madhekwana e al.
Table 7 Comp essi e Sensing F amewo k esea ch s udies
Re Objec i es Me hod A chi e c u e F ame wo k En i on men Mobili y Beam N .
[22] Demons a ed ha limi ing
andomness in comp essed
sampling o local se s achie es
obus ness o s uc u ed e o s
due o ca ie equency o se
Simula ion Analog BF Comp ess i e sensing LOS N/A 64
[23] Aims o op imize beam alignmen
in mul iuse mmWa e MIMO
sys ems unde p ac ical low SNR
condi ions by in es iga ing
comp essi e sensing echniques,
p oposing a ial-based p o ocol
and a no el de e minis ic
cons uc ion o he CS sensing
ma ix, and demons a ing
supe io pe o mance compa ed
o exis ing me hods.
Simula ion Analog BF Comp ess i e sensing LOS, NLOS Yes Va iable
[24] De elop a de e minis ic sensing
ma ix o millime e -wa e beam
alignmen using a
K onecke -based s uc u e,
exploi ing he spa se na u e o
mmWa e channels.
Simula ion Analog BF Comp ess i e sensing NLOS, LOS N/A a iable
[25] Aimed o add ess he challenges o
acking and aining beams in a
ime- a ying millime e-wa e
channel while minimizing
o e head by combining
Comp essed Sensing (CS) and
Kalman Fil e ing.
Simula ion Analog BF Comp ess i e sensing NLOS, LOS N/A Va iable
[26] To educe o e head in beam
alignmen o mmWa e/THz
sys ems using comp essi e
sensing.
Simula ion Hyb id BF Comp ess i e Sensing LOS N/A 32
[27] To educe compu a ional
complexi y in mmWa e beam
alignmen using a de e minis ic,
spa se sensing ma ix
Simula ion Hyb id BF Comp es si e Sensing LOS N/A a iable
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 31 o 53 87
mo e da a. Howe e , he app oach is hea ily elian on high-
quali y aining da a, and i s compu a ional complexi y may
limi eal- ime applica ions in esou ce-cons ained en i on-
men s. Addi ionally, scaling o la ge ne wo ks wi h mo e
APs and use s p esen s challenges in managing mul iple
beams simul aneously. [81] p esen s a deep lea ning me hod
called deep egula ised wa e o m lea ning (DRWL) o
imp o e beam p edic ion in non-coope a i e mmWa e sys-
ems, add essing he challenge o limi ed aining da a. Using
da a augmen a ion echniques like cyclic ime shi (CTS)
and signal splicing, DRWL p edic s op imal beams wi hou
coo dina ion be ween ansmi e and ecei e , educing he
need o exhaus i e beam sweeping. The me hod imp o es
beam p edic ion accu acy e en wi h ewe aining sam-
ples, main aining high pe o mance by le e aging wa e o m
cha ac e is ics. Howe e , he model in oduces compu a-
ional complexi y, making eal- ime deploymen di icul
in esou ce-cons ained en i onmen s, and i s pe o mance
depends on inpu signal quali y. Fu he es ing is needed o
con i m i s e ec i eness in dynamic, eal-wo ld scena ios.
[82] explo es using compu e ision (CV) and machine
lea ning (ML) o imp o e mmWa e beam managemen . T a-
di ional sys ems ely on channel s a e in o ma ion (CSI),
leading o high o e head and delays. This new app oach
uses came a da a o selec beams wi hou needing CSI. The
amewo k ocuses on imp o ing e iciency, obus ness, and
scalabili y wi h ML models. Simula ions show be e beam
alignmen and educed compu a ional cos s. Howe e , he
sys em depends on eliable came a da a, which may be
impac ed by blocked iews o poo condi ions. Labeling da a
o ML aining is also challenging. The sys em aces u he
issues wi h mul iuse en i onmen s, in eg a ion complexi y,
and en i onmen al sensi i i y.
The esea ch pape s spanning [56–60,62–69], [70–82]
collec i ely p o ide a ho ough examina ion o ML/AI-based
beam alignmen echniques. These me hodologies, exempli-
ied by inno a ions like HBA (Hie a chical Beam Align-
men ), a e designed o op imize beam o ming o imp o ed
communica ion pe o mance, demons a ing p omising ou -
comes in e ms o eg e pe o mance, scalabili y, and beam
alignmen la ency educ ion. Howe e , he pape s iden i y
se e al
limi a ions, including eliance on speci ic channel models
ha may no ully ep esen eal-wo ld mmWa e com-
plexi ies, posing challenges o gene aliza ion o di e se
scena ios. The impac o mobili y, dynamic obs acles, and
en i onmen al changes on beam alignmen is also insu i-
cien ly add essed. P ac ical implemen a ion aspec s, such as
compu a ional complexi y and ene gy e iciency, a e o en
o e looked, wi h deploymen challenges needing mo e a en-
ion. The ade-o be ween explo a ion and exploi a ion in
lea ning algo i hms, especially in dynamic mmWa e chan-
nels, equi es u he explo a ion. In essence, while ML/AI-
based beam alignmen echniques exhibi signi ican p og ess
in op imizing mmWa e communica ion, challenges pe sis
in gene aliza ion, p ac ical implemen a ion, and adap abili y
o dynamic scena ios, necessi a ing u u e wo k o add ess
hese limi a ions and explo e eal-wo ld deploymen chal-
lenges, mobili y, and ene gy e iciency o a mo e holis ic
solu ion in mmWa e communica ion sys ems.
4.5 ISAC beam alignmen amewo ks
In eg a ed Sensing and Communica ion (ISAC) is an eme g-
ing echnology. I combines wi eless communica ion and
sensing wi hin one sys em. ISAC o e s new ways o
op imize beam alignmen . Which is essen ial o e icien
da a ansmission and en i onmen al awa eness. Machine
Lea ning (ML), beam sweeping, and con ex in o ma ion
amewo ks can enhance ISAC. These me hods boos pe -
o mance in dynamic en i onmen s. The ollowing esea ch
highligh s ad ancemen s and inno a ions in ISAC, p o id-
ing an o e iew o key e o s d i ing his ield o wa d, see
Tables 13 and 14.
In [83], Zhiqiang Xiao e al. p esen ed a beam o ming
echnique o mmWa e in eg a ed sensing and communica-
ion (ISAC) sys ems, aiming o c ea e a cos -e ec i e analog
beam o ming app oach ha enables a MIMO ISAC sys-
em o simul aneously sense a ge s and communica e wi h
use equipmen (UE). Thei esea ch in oduces a lexible
double-beam codebook o c ea e sepa a e beams o sensing
and communica ion. They also p opose simul aneous beam-
sweeping schemes. These include single and double-beam
me hods o op imize he p ocess and educe o e head.
[84] p esen s an ISAC sys em using Recon igu able In el-
ligen Su aces (RIS). I imp o es communica ion and a ge
sensing e iciency. The au ho s p opose a Simul aneous
Beam T aining and Ta ge Sensing (SBTTS) scheme. This
scheme dis inguishes RIS om a ge s by accumula ing
ene gy in di e en domains. I educes aining o e head.
A Posi ioning and A ay O ien a ion Es ima ion (PAOE)
scheme u he imp o es e iciency wi h a as sea ch algo-
i hm. The sys em pe o ms well in bo h line-o -sigh and
non-line-o -sigh scena ios. Howe e , i aces complexi y,
noise sensi i i y, and pilo o e head challenges. Deploy-
men is also di icul due o la ge ne wo ks’ ha dwa e needs,
eal- ime con ol, and scalabili y. [85] in oduces he In e-
g a ed Sensing and Communica ion (ISAC) beam alignmen
me hod o Te ahe z (THz) ne wo ks. The me hod, called
Join Synch oniza ion Signal Block (SSB) and Re e ence
Signal (RS)-based Sensing (JSRS), educes line-o -sigh
(LoS) blockages and co ec s beam misalignmen caused
by use mobili y. Wi h high- esolu ion THz sensing, JSRS
de ec s and p e en s misalignmen ea ly, imp o ing ne wo k
co e age. The pape shows JSRS enhances beam align-
men and co e age, especially in u ban ehicle- o-e e y hing
123

87 Page 32 o 53 S. Madhekwana e al.
Table 9 ML/AI based F amewo k esea ch s udies
Re Objec i es Me hod A chi e c u e F ame wo k En i o nmen Mobil i y Beam N .
[56] The a icle sugges ed a me hod o beam alignmen
wi h pa ial beams ha uses machine lea ning and
doesn’ equi e any p io knowledge, like use
loca ion da a.
Simula ion Hyb id BF ML-NN NLOS N/A 5
[57] To be conside ed aligned, a signal mus each a
ce ain SINR h eshold be o e i can be conside ed
aligned by a dis ibu ed beam alignmen scheme.
Simula ion Hyb id BF ML-RL NLOS Yes 32
[58] P oposed a deep neu al ne wo k (DNN) solu ion ha
equi ed no knowledge o he channel bu ea ed
he beam selec ion as an image econs uc ion
p oblem and he DNN is used o cons uc a beam
domain image. The solu ion consis s o o line
aining and online p edic ion. Eigen-beam
heo em en y ea ed as s a ing alue o beam
domain image econs uc ion (BDIR). The o -line
Eigen-beam ex ac ion educes he o e head o
online-beam alignmen .
Simula ion Analog and Hyb id
BF
ML-DNN NLOS N/A 32,64, 128,256
[59] P oposed a mmWa e beam alignmen solu ion ha
uses geoloca ion side in o ma ion and le e aging
Suppo Vec o Machine (SVM) model o
mapping UE loca ion and co e ing beam in a
mul i-use mul i-cell scena io. Subsequen ly, he
beam assignmen in o ma ion om use -adjacen
cells helps o educe he la ency o channel s a e
in o ma ion (CSI) eedback.
Simula ion Analog and Hyb id
BF
ML- Supe ised LOS N/A 8
[60] P oposed ML-based beam alignmen amewo k ha
is ained o o ecas he bes access poin (AP) and
he mos sui able beam choice gi en an UE’s GPS
coo dina es. Whe eby he GPS coo dina es a e
epo ed by he UE o by some o he echnique.
Simula ion Hyb id and Analog
BF
ML-DCNN NLOS N/A 64
[62] The au ho s de i ed an AP beam- aining amewo k
by using he spa ial co ela ion be ween a ious
beams in addi ion o he abili y o he Deep
Con olu ional Neu al Ne wo k (DCNN) o e ie e
ea u es The AP amewo k only p obes a ixed se
o he en i e beam space and de e mines he bes
Simula ion Analog BF DCNN amewo k NLOS N/A 128
[63] The au ho s p oposed a CNN s a egy o pe o ming
beam selec ion be ween he ansmi e and
ecei e . The employed CNN skip connec ions and
hype pa ame e op imiza ion o balance be ween
accu acy and compu a ional complexi y.
Simula ion Analog BF CNN amewo k NLOS Yes 61
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 33 o 53 87
Table 9 con inued
Re me ics Scena io F equen cy Me i s Deme i s Beam Pa e n
[56] SNR, SE Cellula mmWa e Reduces aining ime,
does no equi e use
loca ion da a, e icien
beam alignmen
High compu a ional
complexi y, equi es
simula ed aining
en i onmen s
Di ec ional
[57] SNR Cellula mmWa e Low-la ency beam
alignmen , op imal
pe o mance in spa se
mmWa e channels
High compu a ional
complexi y, equi es
uning o ewa d
pa ame e s o a ying
condi ions
Di ec ional Beam Pa e n
[58] SNR Cellula mmWa e Reduces beam sea ch
o e head by up o
90%, main ains 99%
o spec al e iciency
compa ed o
exhaus i e sea ch
Requi es ex ensi e
aining, po en ial
compu a ional
complexi y in
eal- ime applica ions
Di ec ional Beam Pa e n
[59] SNR Cellula , V2X mmWa e Reduces la ency by up
o 50%, dec eases
signaling o e head by
34%, and imp o es
beam pai ing accu acy.
High eliance on
geoloca ion da a,
complexi y in
mul i-cell scena ios,
compu a ional
o e head in aining
Di ec ional Beam Pa e n
[60] BPA Cellula mmWa e Reduces sea ch space
o op imal AP and
beam alignmen ,
obus agains
dynamic sca e e s and
loca ion unce ain y
High compu a ional
complexi y, only
conside s MISO
scena io, equi es
mo e de elopmen o
UE-side beam
aining.
Di ec ional Beam Pa e n
[62] SNR, BPA, La en cy Cellula , V2X mmWa e Reduces compu a ional
complexi y by 15%,
high accu acy (70.4%)
in beam selec ion
Requi es ex ensi e
aining da ase ,
compu a ional
complexi y in
eal- ime deploymen
Di ec ional, Adap i e
[63] SNR Cellula mmWa e Reduces signaling
o e head by 90%,
high p edic ion
accu acy wi h 10%
beam soundings
Compu a ional
complexi y, equi es
ca e ul model uning
o p e en o e i ing
Di ec ion
BF= Beam Fo ming, RSRP = Re e ence Signal Recei ed Powe , RSS = Recei ed Signal S eng h, BPA = beam P edic ion accu acy, SE = Spec al E iciency, ML = Machine Lea ning, RL =
Rein o ced Lea ning , SNR = Signal o Noise Ra ion, NN = neu al Ne wo ks, DNN= Deep NN, DCNN Deep Con olu ional NN, EE =Ene gy E iciency, UDN =Ul a-Dense Ne wo ks (UDN),
D2D = De ice- o-De ice (D2D), SBP =Spa ial Beam P edic ion,TBP=Tempo al Beam P edic ion, DL = Deep Lea ning, LSTM = Long and Sho Memo y, SL=supe ised Lea ning
123
87 Page 34 o 53 S. Madhekwana e al.
Table 10 Con inued ML/AI based F amewo k esea ch s udies - con inued
Re Objec i es Me hod A chi e c u e F ame wo k En i o nmen Mobil i y Beam N .
[64] The au ho s p oposed o ain Ma ix
ac o iza ion ML and Nonnega i e
Ma ix ac o iza ion models by p obing
a iny subse o a massi e beam’s
codebook o TX and RX o p edic he
SNRs o he beams o he codebook.
Subsequen ly, hey de i ed equa ions o
op imizing he ma ix ac o iza ion
me hod
Simula ion Analog BF ML-NN NLOS N/A 1024
[65] The au ho s s udied a
si e-speci ic-sounding codebook and
designed a NN ame ha uses
measu emen s om he sounding
codebook o p edic he bes beam. The
NN amewo k model cap u ed he
si e-speci ic beams ha a e associa ed
wi h he pa icula cha ac e is ics o he
p opaga ion en i onmen .
Simula ion Analog BF ML - NN NLOS N/A 128
[66] The au ho s p oposed a g idless beam
alignmen amewo k ha p edic s
nea -op imal beam weigh s om a
con inuous beam se using unsupe ised
aining ha uses ew beams o channel
sounding.
Expe im en al Analog BF ML-NN NLOS N/A a iable
[67] P oposed a deep ein o cemen
lea ning-based beam alignmen
amewo k ha can swi ch be ween
di e en beam alignmen me hods
depending on he adio channel he eby
achie ing good powe and spec al
e iciency by using channel
ci cums ances o egula e ac i e RF
chains
Simula ion Hyb id BF ML-DRL NLOS Yes 64
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 35 o 53 87
Table 10 con inued
Re Objec i es Me hod A chi e c u e F ame wo k En i o nmen Mobil i y Beam N .
[68] Imp o ed he Kolmogo o model (KM)
by in oducing disc e e mono onic
op imiza ion o educe complexi y and
imp o e scalabili y. Fu he mo e, es ing
ad anced hypo heses can also be
conduc ed using he
Kolmogo o -Smi no (KS) c i e ion,
which equi es no subjec i e h eshold
se ing compa ed o equency
es ima ion used in KM
Simula ion Analog BF ML-KM NLOS N/A 16
[69] The au ho s p oposed a hie a chical beam
alignmen algo i hm by using
knowledge abou he in e -beam
co ela ion s uc u e and channel
a ia ion. The au ho s concep ualised
he beam alignmen p oblem as a
"mul i-a med bandi " p oblem and
sol ed by cha ac e izing he beam
co ela ion s uc u e in he
mul i-channel as a mul imodal unc ion.
Addi ionally, use he p e iously gained
knowledge o he channel luc ua ions o
p ope ly conside ewa d unce ain y
Simula ion Analog BF ML- MAB NLOS N/A 512
[70] Op imize beam managemen and CSI
acquisi ion using ML in X-MIMO
sys ems o 6G ne wo ks
Simula ion Hyb id BF ML-NN LOS, NLOS Yes 8, 16, 32
123
87 Page 36 o 53 S. Madhekwana e al.
Table 10 con inued
Re me ics Scena io F equen cy Me i s Deme i s Beam Pa e n
[64] SNR Cellula mmW a e, THz Reduces beam sweeping
o e head by 3x,
achie es o e 90%
alignmen accu acy
High aining
complexi y, dependen
on si e-speci ic
aining da a
Di ec ional Beam Pa e n
[65] SNR Cellula mmWa e Reduces sweeping
o e head by 21x,
achie es 0.32 dB o
he heo e ical uppe
bound
High compu a ional
complexi y, equi es
si e-speci ic aining
Di ec ional
[66] SNR Cellula mmWa e Reduces aining
o e head, adap s o
en i onmen al
changes, maximizes
bo h EE and SE
High compu a ional
complexi y, sensi i e
o changes in
eal-wo ld
en i onmen s, elies
on deep lea ning
Di ec ional
Adap i e
[67] EE, SE Cellula mmWa e Reduced compu a ional
complexi y, imp o ed
beam alignmen
pe o mance
High complexi y in
p e ious KM models,
compu a ional
limi a ions
hie a chical beam pa e n
[68] SNR Cellul a , WLAN, FWA, UAV mmWa e Signi ican ly educed
compu a ional
complexi y, imp o ed
beam alignmen
pe o mance
High compu a ional
complexi y wi h la ge
an enna a ays
O e lapping, hie a chical
[69] SNR UDN, D2D mmWa e Imp o ed in e e ence
supp ession,
loca ion-awa e
communica ion,
op imized
beam o ming o
spa ial mul iplexing in
dense en i onmen s
Requi es accu a e
posi ioning and
synch oniza ion,
compu a ional
complexi y in
high-densi y
en i onmen s
Di ec ional
[70] RSRP, SNR, SE Cellul a mmWa e Signi ican gains in
spec al e iciency,
adap abili y, educed
eedback equi emen s
Inc eased compu a ional
complexi y, challenges
in eal- ime ope a ions
and gene aliza ion.
Dynamic and Adap i e
BF= Beam Fo ming, RSRP = Re e ence Signal Recei ed Powe , RSS = Recei ed Signal S eng h, BPA = beam P edic ion accu acy, SE = Spec al E iciency, ML = Machine Lea ning, RL =
Rein o ced Lea ning , SNR = Signal o Noise Ra ion, NN = neu al Ne wo ks, DNN= Deep NN, DCNN Deep Con olu ional NN, EE =Ene gy E iciency, UDN =Ul a-Dense Ne wo ks (UDN),
D2D = De ice- o-De ice (D2D), SBP =Spa ial Beam P edic ion,TBP=Tempo al Beam P edic ion, DL = Deep Lea ning, LSTM = Long and Sho Memo y, SL=supe ised Lea ning
123

Beam alignmen o mmWa e and THz: sys ema ic... Page 37 o 53 87
Table 11 Con inued ML/AI based F amewo k esea ch s udies
Re Objec i es Me hod A chi e c u e F ame wo k En i on men Mobil i y Beam N .
[71] Op imize beam acking and a e
adap a ion in mmWa e sys ems using
ein o cemen lea ning
Simula ion Hyb id BF ML-RL LOS, NLOS Yes Va ia ble
[72] Op imize ini ial access and beam
alignmen o secu i y and e iciency
using deep lea ning
Simula ion Real-wo ld da a Hyb id BF ML-DL LOS, NLOS Yes Limi e d se
[73] Op imize beam aining a THz
equencies using MAB algo i hms o
as and adap i e beam selec ion
Simula ion Hyb id BF ML-RL LOS, NLOS Yes Va iab le
[74] Enhance beam managemen wi h AI/ML
o imp o e p edic ion accu acy, educe
la ency, and op imize pe o mance in
5G-Ad anced
Simula ion Su ey Hyb id BF ML- SBP ML-TBP) LOS, NLOS Yes Va iab le
[75] Op imize beam alignmen in mmWa e
sys ems using a deep lea ning-based
hie a chical amewo k
Simula ion Hyb id BF ML-DL LOS, NLOS Yes Va iab le
[76] Imp o e ene gy e iciency in mmWa e
ne wo ks using he A2C lea ning
amewo k o beam selec ion and
powe op imiza ion
Simula ion Hyb id BF ML-RL LOS, NLOS Yes Va iab le
123
87 Page 38 o 53 S. Madhekwana e al.
Table 11 con inued
Re me ics Scena io F equen cy Me i s Deme i s Beam Pa e n
[71] Th oug hpu , Ou age du a ion Cellul a mmWa e 182% h oughpu gain,
educed ou age
du a ion, low
o e head, adap able o
mobili y
Compu a ional
complexi y in
eal- ime ope a ions,
uning o ATS
pa ame e s o
pe o mance.
Dynamic and Adap i e
[72] BPA, RSRP Cellul a mmW a e 75% educ ion in beam
sea ch space, 99.66%
accu acy, imp o ed
secu i y, ene gy
e iciency
Reduced RSRP by 6.5
dB o secu i y,
ade-o be ween
signal s eng h and
secu i y
Dynamic and Adap i e
[73] Adap a ion Speed, SE Cellul a THz Imp o ed spec al
e iciency, as e
adap a ion, and
supe io pe o mance
wi h con ex ual da a.
Highe complexi y o
con ex ual algo i hms,
la ency ade-o s wi h
P obing-LinUCB
Dynamic and Adap i e
[74] BPA RSRP La en cy Cellul a mmWa e 63.5% inc ease in
p edic ion accu acy,
educed o e head, and
be e high-mobili y
pe o mance
High compu a ional
complexi y, di icul ies
wi h model
gene aliza ion, p i acy
conce ns
Dynamic and Adap i e
[75] BPA, SE, O e head Cellul a mmWa e Highe accu acy wi h
ewe measu emen s,
educed o e head,
enhanced spec al
e iciency
High compu a ional
complexi y, need o
ex ensi e aining
da a, challenges in
eal- ime deploymen
Dynamic and Adap i e
[76] EE, O e head, SNR Cellul a mmWa e Mo e han doubles
ene gy e iciency,
adap s o a ious
condi ions, and
in eg a es in o Open
RAN
High compu a ional
complexi y, ex ensi e
aining da a,
scalabili y issues in
mul i-use
en i onmen s
Dynamic and Adap i e
BF= Beam Fo ming, RSRP = Re e ence Signal Recei ed Powe , RSS = Recei ed Signal S eng h, SE = Spec al E iciency, ML = Machine Lea ning, RL = Rein o ced Lea ning , SNR = Signal
o Noise Ra ion, NN = neu al Ne wo ks, DNN= Deep NN, DCNN Deep Con olu ional NN, EE =Ene gy E iciency, UDN =Ul a-Dense Ne wo ks (UDN), D2D = De ice- o-De ice (D2D), SBP
=Spa ial Beam P edic ion,TBP=Tempo al Beam P edic ion, DL = Deep Lea ning, LSTM = Long and Sho Memo y, SL=supe ised Lea ning, BPA=Beam P edic ion Accu acy
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 39 o 53 87
Table 12 Con inued ML/AI based F amewo k esea ch s udies
Re Objec i es Me hod A chi e c u e F ame wo k En i o nmen Mobili y Beam N .
[77] Op imize beam selec ion in mmWa e
MIMO sys ems using a deep con ex ual
bandi lea ning amewo k
Simula ion Hyb id BF ML-DL LOS, NLOS Yes Va iab le
[78] Imp o e beam acking e iciency and
educe o e head using LSTM in
mul i-cell mmWa e en i onmen s
Simula ion Analog and Digi al BF ML-DL and -LSTM LOS, NLOS Yes Va iab le
[79] Imp o e beam selec ion e iciency and
educe o e head in dynamic,
blockage-p one en i onmen s
Simula ion Hyb id BF ML-NN LOS, NLOS Yes Va iab le
[80] Use ML algo i hms o op imize beam
managemen in D-MIMO sys ems and
educe measu emen complexi y
Simula ion Hyb id BF ML-SL LOS, NLOS Yes Va iab le
[81] Imp o e beam p edic ion wi h limi ed
da a using deep lea ning in
non-coope a i e mmWa e en i onmen s
Simula ion Hyb id BF ML-DL LOS, NLOS Yes a iab le
[82] Use CV and ML o beam managemen ,
educing CSI eliance
Simula ion Hyb id BF ML-SL LOS Yes a iab le
123
87 Page 40 o 53 S. Madhekwana e al.
Table 12 con inued
Re me ics Scena io F equen cy Me i s Deme i s Beam Pa e n
[77] Th oug hpu , O e head Cellul a mmWa e Imp o ed beam
selec ion e iciency,
educes exhaus i e
sea ch, adap able o
dynamic en i onmen s
High compu a ional
complexi y, equi es
signi ican aining
da a, limi ed
eal-wo ld alida ion
Dynamic and Adap i e
[78] Powe Consump ion, O e head, BPA Cellul a mmWa e Reduced powe
consump ion and
o e head, high
accu acy in beam
acking, adap able o
en i onmen s
Compu a ional
complexi y, eliance
on aining da a, ixed
s adap i e cons ain
complexi ies
Dynamic and Adap i e
[79] O e head, SNR, BPA Cellul a mmW a e 90% accu acy, educes
beam sea ch ime by
86.4%, pe o ms well
in dynamic
en i onmen s like
u ban V2X
Compu a ional
complexi y, da a
dependency, pa ame e
op imiza ion o
a ious en i onmen s.
Dynamic and Adap i e
[80] O e head, BPA Cellul a mmW a e Reduces measu emen
o e head, imp o es
beam selec ion
accu acy, e icien in
dynamic en i onmen s
Da a dependency,
compu a ional
complexi y, challenges
in scaling o la ge
ne wo ks
Dynamic and Adap i e
[81] BPA, O e head Cellul a mmWa e High p edic ion
accu acy wi h limi ed
aining samples,
educed beam
sweeping o e head
High compu a ional
complexi y,
pe o mance
dependen on signal
quali y, limi ed
eal-wo ld es ing
Dynamic and Adap i e
[82] BPA, O e head Cellul a UAV mmWa e Reduces o e head and
imp o es eal- ime
beam alignmen
Depends on came a da a
and en i onmen al
condi ions
P ede ined beam codebook (YOLO, e c.)
BF= Beam Fo ming, RSRP = Re e ence Signal Recei ed Powe , RSS = Recei ed Signal S eng h, SE = Spec al E iciency, ML = Machine Lea ning, RL = Rein o ced Lea ning , SNR = Signal
o Noise Ra ion, NN = neu al Ne wo ks, DNN= Deep NN, DCNN Deep Con olu ional NN, EE =Ene gy E iciency, UDN =Ul a-Dense Ne wo ks (UDN), D2D = De ice- o-De ice (D2D), SBP
=Spa ial Beam P edic ion,TBP=Tempo al Beam P edic ion, DL = Deep Lea ning, LSTM = Long and Sho Memo y, SL=supe ised Lea ning, BPA=Beam P edic ion Accu acy
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 47 o 53 87
Table 15 con inued
Fea u e/ F amewo k Beam Sweeping Con ex In o ma ion Comp essi e Sensing ML/AI ISAC F amewo ks
Robus ness Limi ed obus ness in dynamic
en i onmen s
Can enhance obus ness by
conside ing
en i onmen al ac o s
Robus in spa se channel
condi ions bu may
s uggle in highly
dynamic scena ios
Can adap and imp o e
obus ness o e ime wi h
con inuous lea ning
High, sensi i e o
en i onmen bu imp o ed
beam o ming accu acy
and highly adap able
S anda diza ion Po en ial Gene ally simple and widely used,
making i easie o s anda dize ac oss
di e en sys ems
S anda diza ion may be
easible o ce ain
aspec s, bu con ex
in o ma ion can be di e se
and applica ion-speci ic
May ace challenges in
s anda diza ion due o
a ia ions in algo i hms
and econs uc ion
me hods.
S anda diza ion is
challenging due o he
di e si y o machine
lea ning models and he
need o ailo ed solu ions
Mode a e o High, elian
on 5G like s anda ds
andnlikely o be adop ed
in RIS aided 6G S anda ds
In e ope abili y Gene ally, mo e in e ope able as i elies
on basic beam adjus men mechanisms
In e ope abili y can be a
challenge i di e en
sys ems use di e se
con ex in o ma ion
sou ces and p ocessing
me hods
In e ope abili y challenges
may a ise due o
a ia ions in sensing
ha dwa e and algo i hms.
In e ope abili y is a conce n
due o a ia ions in
machine lea ning models,
a chi ec u es, and aining
da a
High, scals o la ge
ne wo ks wi h mul iple
e lec ing su aces and
UEs
Scalabili y Highly scalable as i in ol es basic beam
adjus men s, sui able o a ious
deploymen scales
Scalabili y challenges may
a ise i con ex
in o ma ion sou ces
become oo complex o
di e se
Scalabili y depends on he
e iciency o sensing
ha dwa e and algo i hms,
po en ially acing
limi a ions in la ge-scale
deploymen s
Scalabili y can be an issue
due o he compu a ional
demands, especially
du ing he aining phase
High asi in eg a es well
wi h exis ing mmWa e
se ups in la ge scale
sys ems
Ne wo k Scena ios All All All All All
123

87 Page 48 o 53 S. Madhekwana e al.
5 Compa ison and limi a ions o exis ing
amewo ks
In his sec ion, RQ2 is add essed. Table 15 shows a quali-
a i e analysis o pe o mance me ics used o compa e all
he amewo ks iden i ied in his e iew. In compa ing he
amewo ks o beam alignmen , each amewo k o e s dis-
inc ad an ages and challenges.
Beam Sweeping, cha ac e ized by i s sys ema ic adjus -
men o beam di ec ion, is simple o implemen and widely
used, making i easily scalable and in e ope able. Howe e ,
i s limi ed obus ness in dynamic en i onmen s and highe
la ency in inding op imal alignmen may be d awbacks.
Con ex In o ma ion, which inco po a es addi ional en i-
onmen al da a, enhances obus ness bu may su e om
complexi y and in e ope abili y challenges, pa icula ly wi h
di e se sou ces o con ex in o ma ion.
Comp essi e sensing, elying on spa se channel exploi a-
ion, o e s obus ness in ce ain condi ions bu equi es
complex algo i hms and may ace challenges in s anda diza-
ion and in e ope abili y due o a ia ions in ha dwa e and
me hods.
ML/AI-based app oaches le e age machine lea ning o
adap i e s a egies, p omising imp o ed obus ness o e
ime. S ill, hey demand high compu a ional esou ces and
may s uggle wi h s anda diza ion and in e ope abili y due
o di e se models and aining da a.
ISAC sys ems p esen a p omising solu ion o u u e ne -
wo ks, combining communica ion and sensing. They o e
bene i s like be e e iciency and pe o mance. Howe e ,
challenges such as compu a ional complexi y, in as uc u e
equi emen s, and eal- ime p ocessing mus be add essed.
As esea ch p og esses, ISAC will likely play a majo ole in
shaping 6G and o he ad anced ne wo ks. Machine lea ning-
based and ISAC-based amewo ks a e bo h scalable and
in e ope able wi h o he amewo ks, howe e , hey a e
hea y in e ms o compu a ion. Despi e hea y compu a ional
esou ce equi emen s and complexi y, hey ha e a po en ial
o s anda diza ion.
In essence, while each amewo k p esen s unique s eng hs,
hey also pose dis inc complexi ies and limi a ions ha a e
shown in Table 16.
6 Fu u e di ec ions and esea ch
oppo uni ies
This sec ion will add ess RQ4. The 3GPP Release 15 in o-
duced he beam sweeping-based p o ocol o 5G o add ess
he beam alignmen [10,94]. Howe e , as a highe equency
spec um and la ge an enna a ays a e deployed in 5G and
6G, his app oach won’ be adequa e. The beam-sweeping
app oach won’ add ess all he challenges ha a e associa ed
wi h beam alignmen . Ins ead, he o e head, complexi y and
la ency associa ed wi h beam sweeping scale up wi h he size
o he beam space [10]. Some o he challenges and gaps ha
need o be sol ed in add essing beam alignmen we e iden-
i ied in [10,111–113] and can be summa ized as ollows:
6.1 Dynamic and mobile en i onmen s
Many exis ing beam alignmen amewo ks p ima ily ocus
on s a ic scena ios and may no adequa ely add ess he chal-
lenges posed by dynamic en i onmen s and mobile use s.
Resea ch is needed o de elop beam alignmen echniques
ha can e icien ly ack and adap o Fas -mo ing use s o
changing en i onmen al condi ions, ensu ing con inuous and
eliable connec i i y [10,112,113].
6.2 Robus ness agains en i onmen al ac o s
En i onmen al ac o s, such as blockages, e lec ions, di ac-
ion, and a mosphe ic condi ions, can signi ican ly impac
he pe o mance o beam alignmen amewo ks. De eloping
obus beam alignmen echniques ha can handle a ying
en i onmen al condi ions, including ad e se wea he condi-
ions, dense u ban en i onmen s, and indoo scena ios, is an
impo an esea ch a ea [113].
6.3 In e e ence mi iga ion
In e e ence om neighbou ing sys ems o coexis ing ne -
wo ks can hinde beam alignmen pe o mance. Resea ch is
needed o de elop in e e ence-awa e beam alignmen ame-
wo ks ha can e ec i ely mi iga e in e e ence e ec s and
main ain high-quali y communica ion links in c owded spec-
um en i onmen s [94], [111].
6.4 Ene gy e iciency conside a ions
Beam alignmen echniques o en in ol e ene gy-in ensi e
ope a ions, such as beam aining and acking. Conside ing
he ene gy cons ain s o ba e y-powe ed de ices and he
g owing demand o ene gy-e icien communica ion sys-
ems, esea ch is needed o explo e ene gy-e icien beam
alignmen me hods and p o ocols wi hou comp omising
he pe o mance and eliabili y o he communica ion links
[112].
6.5 S anda diza ion and in e ope abili y
While se e al beam alignmen amewo ks ha e been p o-
posed, he e is a lack o s anda diza ion and in e ope abili y
ac oss di e en sys ems and endo s. E o s a e equi ed
o de elop common s anda ds and p o ocols o beam align-
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 49 o 53 87
men , acili a ing seamless in eg a ion and in e ope abili y
among di e en de ices, ne wo ks, and echnologies [113].
6.6 Real-wo ld deploymen and alida ion
Many exis ing beam alignmen amewo ks ha e been e al-
ua ed h ough simula ions o limi ed expe imen al se ups.
Mo e esea ch is needed o alida e hese amewo ks in
eal-wo ld deploymen s, conside ing p ac ical cons ain s,
eal-li e channel condi ions, and di e se deploymen sce-
na ios. This would p o ide aluable insigh s in o he ac ual
pe o mance, limi a ions, and scalabili y o beam alignmen
echniques [10].
6.7 Secu i y and p i acy conside a ions
Beam alignmen amewo ks may be ulne able o secu i y
h ea s, including ea esd opping, spoo ing, o unau ho ized
beam manipula ion. Fu he esea ch is needed o in es i-
ga e secu i y and p i acy aspec s ela ed o beam alignmen
and de elop obus mechanisms o ensu e secu e and p i a e
communica ion in beam o ming-based sys ems [111], [113].
Add essing hese esea ch gaps will con ibu e o he
de elopmen o mo e e icien , eliable, and adap i e beam
alignmen amewo ks, pa ing he way o he success ul
deploymen o mmWa e and THz communica ion sys ems
in a ious scena ios and applica ions.
7 Conclusion
The SLR was conduc ed on published esea ch ocusing on
beam alignmen and ini ial access amewo ks o mmWa e
and THz. The SLR iden i ied he mos ele an 73 esea ch
pape s acco ding o he au ho ’s knowledge du ing he e iew
pe iod. 596 pape s we e ini ially selec ed using he de ined
sea ch s ings om digi al lib a ies such as IEEE, Sp inge ,
Else ie , and F ancis & Taylo . Subsequen ly, a h ee-s age
il e ing mechanism was applied o he i le, abs ac , and ull
ex , including he assessmen o pape quali y. The chosen
s udies we e hen explo ed, so ed and e alua ed acco ding
o he de ined RQs.
In his s udy, we p esen ed s a e-o - he-a amewo ks o
beam alignmen in mmWa e and THZ, and we explained he
axonomy o he beam alignmen amewo ks by classi ying
hem in o beam sweeping-based, con ex in o ma ion-based,
comp essi e Sensing-based machine Lea ning/A i icial In el-
ligence (ML/AI) -based and ISAC-based amewo ks. To
imp o e beam alignmen in e ms o la ency educ ion,
ene gy consump ion educ ion and o e head, he exis ing
li e a u e body was explo ed o iden i y he mos com-
mon beam alignmen amewo ks. While each amewo k
p esen s unique s eng hs, hey also pose dis inc complex-
i ies and limi a ions. Beam Sweeping and Con ex In o -
ma ion o e simplici y and obus ness enhancemen s bu
may s uggle wi h dynamic en i onmen s and in e ope -
abili y issues. Comp essi e sensing p o ides obus ness in
spa se channels bu demands complex algo i hms and aces
s anda diza ion challenges. ML/AI app oaches p omise con-
inuous imp o emen in obus ness bu come wi h high
compu a ional demands and po en ial in e ope abili y hu -
dles. ISAC sys ems p esen a p omising solu ion o u u e
ne wo k beam managemen , combining communica ion and
sensing. They o e bene i s like be e e iciency and pe -
o mance. Howe e , challenges such as compu a ional com-
plexi y, in as uc u e equi emen s, and eal- ime p ocessing
mus be add essed. Thus, he choice among hese ame-
wo ks hinges on ac o s such as deploymen en i onmen ,
compu a ional esou ces, and he need o adap abili y and
s anda diza ion.
In addi ion, we p esen ed esea ch gaps by iden i ying and
analyzing he limi a ions o he cu en amewo ks. Las ly,
challenges and possible oppo uni ies o u u e esea ch
we e highligh ed.
Acknowledgemen s I hank Nokia o suppo ing his wo k.
Au ho con ibu ions Sundi e Madhekwana was esponsible o esea ch,
p epa a ion, p esen a ion and c i ical e alua ion o he wo ks p esen ed
in his SLR. Muhammad A slan Usman con ibu ed owa ds he esea ch
planning, supe ision and echnical edi ing. Ah isham Ayyub con-
ibu ed owa ds da a isualisa ion and p esen a ion o he a icles his
SLR co e s. Ch is os Poli is p o ided his o e sigh , supe ision and
men o ship in c i ically e iewing his a icle o i s scien i ic igou .
Da a A ailabili y No da ase s we e gene a ed o analysed du ing he
cu en s udy.
Decla a ions
Compe ing in e es s The au ho s decla e no compe ing in e es s.
Open Access This a icle is licensed unde a C ea i e Commons
A ibu ion 4.0 In e na ional License, which pe mi s use, sha ing, adap-
a ion, dis ibu ion and ep oduc ion in any medium o o ma , as
long as you gi e app op ia e c edi o he o iginal au ho (s) and he
sou ce, p o ide a link o he C ea i e Commons licence, and indi-
ca e i changes we e made. The images o o he hi d pa y ma e ial
in his a icle a e included in he a icle’s C ea i e Commons licence,
unless indica ed o he wise in a c edi line o he ma e ial. I ma e ial
is no included in he a icle’s C ea i e Commons licence and you
in ended use is no pe mi ed by s a u o y egula ion o exceeds he
pe mi ed use, you will need o ob ain pe mission di ec ly om he copy-
igh holde . To iew a copy o his licence, isi h p://c ea i ecomm
ons.o g/licenses/by/4.0/.
Re e ences
1. Uwaechia, A. N., & Mahyuddin, N. M. (2020). A comp ehensi e
su ey on millime e wa e communica ions o i h-gene a ion
123
87 Page 50 o 53 S. Madhekwana e al.
wi eless ne wo ks: Feasibili y and challenges. IEEE Access, 8,
62367–62414.
2. Xue, Q., Ji, C., Ma, S., Guo, J., Xu, Y., Chen, Q., & Zhang,
W. (2024). A su ey o beam managemen o mmwa e and hz
communica ions owa ds 6g. IEEE Communica ions Su eys &
Tu o ials., 26(3), 1520–1559.
3. Zhang, J., Bjö nson, E., Ma haiou, M., Ng, D. W. K., Yang, H., &
Lo e, D. J. (2020). P ospec i e mul iple an enna echnologies o
beyond 5g. IEEE Jou nal on Selec ed A eas in Communica ions,
38(8), 1637–1660.
4. Ning, B., Tian, Z., Mei, W., Chen, Z., Han, C., Li, S., Yuan, J., &
Zhang, R. (2023). Beam o ming echnologies o ul a-massi e
mimo in e ahe z communica ions. IEEE Open Jou nal o he
Communica ions Socie y, 4, 614–658.
5. Al-Falahy, N., & Alani, O. Y. (2019). Millime e wa e equency
band as a candida e spec um o 5g ne wo k a chi ec u e: A su -
ey. Physical Communica ion, 32, 120–144.
6. Pe o , V., Pya ae , A., Mol chano , D., & Kouche ya y, Y.
(2016). Te ahe z band communica ions: Applica ions, esea ch
challenges, and s anda diza ion ac i i ies. In: 2016 8 h In e -
na ional Cong ess on Ul a Mode n Telecommunica ions and
Con ol Sys ems and Wo kshops (ICUMT), pp. 183–190. IEEE.
7. Wang, X., Kong, L., Kong, F., Qiu, F., Xia, M., A non, S., &
Chen, G. (2018). Millime e wa e communica ion: A comp ehen-
si e su ey. IEEE Communica ions Su eys I& Tu o ials., 20(3),
1616–1653.
8. Na enge ile, N. (2022). Channel es ima ion and beam aining
wi h machine lea ning applica ions o millime e-wa e commu-
nica ion sys ems.
9. A aoui, W., Bou aqia, K., & Sabi , E. (2022). Ini ial access
i& beam alignmen o mmwa e and e ahe z communica ions.
IEEE Access, 10, 35363–35397.
10. Heng, Y., And ews, J.G., J., M., V., V., A., A., L., N.B., & Zhang,
J.C. (2021). Six key challenges o beam managemen in 5.5 g
and 6g sys ems. IEEE Communica ions Magazine 59(7).
11. Ning, B., Mei, W., Zhu, L., Chen, Z., & Zhang, R. (2024). Code-
book design and pe o mance analysis o wideband beam o ming
in e ahe z communica ions. IEEE T ansac ions on Wi eless
Communica ions.
12. Ning, B., Wang, T., Huang, C., Zhang, Y., & Chen, Z. (2023).
Wide-beam designs o e ahe z massi e mimo: Sca-a p and s-
sa . IEEE In e ne o Things Jou nal, 10(12), 10857–10869.
13. Duan, Q., Kim, T., & Ghauch, H. (2021). Km lea ning o
millime e -wa e beam alignmen and acking: P edic abili y and
in e p e abili y. IEEE Access, 9, 117204–117216.
14. Ning, B., Chen, Z., Tian, Z., Han, C., & Li, S. (2021). A uni ied
3d beam aining and acking p ocedu e o e ahe z communi-
ca ion. IEEE T ansac ions on Wi eless Communica ions, 21(4),
2445–2461.
15. Ning, B., Chen, Z., Chen, W., Du, Y., & Fang, J. (2021). Te a-
he z mul i-use massi e mimo wi h in elligen e lec ing su ace:
Beam aining and hyb id beam o ming. IEEE T ansac ions on
Vehicula Technology, 70(2), 1376–1393.
16. Gio dani, M., Mezza illa, M., & Zo zi, M. (2016). Ini ial access in
5g mmwa e cellula ne wo ks. IEEE communica ions Magazine,
54(11), 40–47.
17. Ta a ia, H., Sha i, M., Molisch, A. F., Dohle , M., Sjöland, H., &
Tu esson, F. (2021). 6g wi eless sys ems: Vision, equi emen s,
challenges, insigh s, and oppo uni ies. P oceedings o he IEEE,
109(7), 1166–1199.
18. Yi, W., Zhiqing, W., & Zhiyong, F. (2024). Beam aining and
acking in mmwa e communica ion: A su ey. China Commu-
nica ions.
19. Zecchin, M., Mashhadi, M. B., Jankowski, M., Gündüz, D.,
Koun ou is, M., & Gesbe , D. (2022). Lida and posi ion-aided
mmwa e beam selec ion wi h non-local cnns and cu iculum
aining. IEEE T ansac ions on Vehicula Technology, 71(3),
2979–2990.
20. Ali, A., González-P elcic, N., & Hea h, R. W. (2017). Millime e
wa e beam-selec ion using ou -o -band spa ial in o ma ion. EEE
T ansac ions on Wi eless Communica ions, 17(2), 1038–1052.
21. Li, X., Fang, J., Duan, H., & Li, H. (2019). A spa se encoding and
phaseless decoding app oach o as mmwa e beam alignmen .
ICASSP 2019-2019 IEEE In e na ional Con e ence on Acous ics,
Speech and Signal P ocessing (ICASSP). IEEE.
22. Mye s, N.J., & Hea h, R.W. (2019). Localized andom sam-
pling o obus comp essi e beam alignmen . ICASSP 2019-2019
IEEE In e na ional Con e ence on Acous ics, Speech and Signal
P ocessing (ICASSP). IEEE.
23. Kho dad, E., Collings, I. B., Hanly, S. V., & Cai e, G. (2022). Com-
p essi e sensing-based beam alignmen schemes o ime- a ying
millime e -wa e channels. IEEE T ansac ions on Wi eless Com-
munica ions, 22(3), 1604–1617.
24. Kho dad, E., Collings, I.B., & Hanly, S.V. (2019). A k onecke -
based spa se comp essi e sensing ma ix o millime e wa e
beam alignmen . 2019 13 h In e na ional Con e ence on Signal
P ocessing and Communica ion Sys ems (ICSPCS). IEEE.
25. Lin, K. H., & Liu, K. H. (2022). A no el beam alignmen
scheme o mobile millime e -wa e communica ions based on
comp essed sensing aided-kalman il e . IEEE Open Jou nal o
he Communica ions Socie y, 3, 1515–1527.
26. Masoumi, H., Ve haegen, M., & Mye s, N.J. (2023). In-sec o
comp essi e beam alignmen o mmwa e and hz adios. a Xi
p ep in a Xi :2308.13268
27. Kho dad, E., Collings, I.B., & Hanly, S.V. (2019). A k onecke -
based spa se comp essi e sensing ma ix o millime e wa e
beam alignmen . In: 2019 13 h In e na ional Con e ence on Sig-
nal P ocessing and Communica ion Sys ems (ICSPCS), pp. 1–5.
IEEE.
28. Mye s, N. J., Mezghani, A., & Hea h, R. W. (2018). Swi -link:
A comp essi e beam alignmen algo i hm o p ac ical mmwa e
adios. IEEE T ansac ions on Signal P ocessing, 67(4), 1104–
1119.
29. Kho dad, E., Collings, I. B., Hanly, S. V., & Cai e, G. (2022). Com-
p essi e sensing-based beam alignmen schemes o ime- a ying
millime e -wa e channels. IEEE T ansac ions on Wi eless Com-
munica ions, 22(3), 1604–1617.
30. Yan, H., & Cab ic, D. (2019). Comp essi e ini ial access and
beam o ming aining o millime e -wa e cellula sys ems. IEEE
jou nal o selec ed opics in signal p ocessing, 13(5), 1151–1166.
31. Song, Y., Ge, C., Qiu, L., & Zhang, Y. (2023). 2ace: Spec-
al p o ile-d i en mul i- esolu ional comp essi e sensing o
mmwa e channel es ima ion. In: P oceedings o he Twen y-
ou h In e na ional Symposium on Theo y, Algo i hmic Foun-
da ions, and P o ocol Design o Mobile Ne wo ks and Mobile
Compu ing, pp. 41–50.
32. Fe nandes, F., Rom, C., Ha ebek, J., & Manchón, C.N. (2022).
Imp o ing beam managemen signalling o 5g n sys ems using
hyb idbeam o ming. IEEE Wi eless Communica ions and Ne -
wo king Con e ence (WCNC). IEEE.
33. Li, Y. N. R., Gao, B., Zhang, X., & Huang, K. (2020). Beam man-
agemen in millime e -wa e communica ions o 5g and beyond.
IEEE Access, 8, 13282–13293.
34. Qian, J., Masou os, C., Tanabe, K., Sasaki, E., Zein, N., & Ma u-
mo o, T. (2022). Beam-pa e n assis ed low-complexi y beam
alignmen o ixed wi eless mmwa e xhaul. ICC 2022-IEEE
In e na ional Con e ence on Communica ions. IEEE.
35. Gio dani, M., Polese, M., Roy, A., Cas o , D., & Zo zi, M. (2018).
Ini ial access amewo ks o 3gpp n a mm-wa e equencies.
17 h Annual Medi e anean Ad Hoc Ne wo king Wo kshop (Med-
Hoc-Ne ). IEEE.
123
Beam alignmen o mmWa e and THz: sys ema ic... Page 51 o 53 87
36. Khi wadka , P., Shahsa a i, S., Hücümeno˘glu, M.C., & Pal, P.
(2023). Max-min hie a chical codebooks o millime e -wa e
channel sensing wi h bandi algo i hms. In: 2023 57 h Asiloma
Con e ence on Signals, Sys ems, and Compu e s, pp. 387–392.
IEEE.
37. Ram, A., López, M., Rod , I., Sø ensen, T. B., Ba io, S. C.,
Padilla, P., Valenzuela-Valdés, J. F., Mogensen, P., e al. (2024).
F 2 5g ne wo ks o indus ial scena ios: Expe imen al cha -
ac e iza ion and beam managemen p ocedu es in ope a ional
condi ions. IEEE T ansac ions on Vehicula Technology., 73(9),
13513–13525.
38. Ba on-Hyppoli e, A., Ab eu, J.V., San os, J.F., DaSil a, L.A.,
& Kibilda, J. (2023). Adap i e beam managemen o secu e
mmwa e communica ions using so wa e-de ined adios. In:
MILCOM 2023-2023 IEEE Mili a y Communica ions Con e -
ence (MILCOM), pp. 243–244. IEEE.
39. Huang, W., Zhang, W., & Wang, C. (2023). I e a i e achie able
a e adap i e beam alignmen o hz communica ion sys ems. In:
2023 8 h In e na ional Con e ence on Communica ion, Image and
Signal P ocessing (CCISP), pp. 450–456. IEEE.
40. Feng, Y., Wei, J., Dinh, P., Ghoshal, M., & Kou sonikolas, D.
(2023). Beam managemen in ope a ional 5g mmwa e ne wo ks.
In: P oceedings o he 7 h ACM Wo kshop on Millime e -Wa e
and Te ahe z Ne wo ks and Sensing Sys ems, pp. 13–18.
41. O ikumhi, I., Kang, J., Pa k, C., Yang, J., & Kim, S. (2018).
Loca ion-awa e coo dina ed beam alignmen in mmwa e com-
munica ion. 2018 56 h Annual Alle on Con e ence on Commu-
nica ion, Con ol, and Compu ing (Alle on). IEEE.
42. Ma andi, M.K., Wei, S., Khodapanah, B., Ra e, W., & Fe weis, G.
(2022). Expe imen al e alua ion o a a iable leng h beam selec-
ion amewo k in a us p based es bed wi h mmwa e on ends
and bu le ma ices. IEEE In e na ional Symposium on Phased
A ay Sys ems I& Technology (PAST). IEEE.
43. Jans, C., Ra e, W., & Fe weis, G. (2022). Fas beam alignmen ia
ue ime delay equency dependen beam o ming using ixed and
a iable leng h es s. IEEE In e na ional Symposium on Phased
A ay Sys ems I& Technology (PAST). IEEE.
44. Ma andi, M.K., Jans, C., Ra e, W., & Fe weis, G. (2021). E al-
ua ion o de ec ion accu acy and e iciency o conside ed beam
alignmen s a egies o mm-wa e massi e mimo sys ems. 55 h
Asiloma Con e ence on Signals, Sys ems, and Compu e s. IEEE.
45. Jans, C., Song, X., Ra e, W., & Fe weis, G. (2020). F equency-
selec i e analog beam p obing o millime e wa e communica-
ion sys ems. IEEE Wi eless Communica ions and Ne wo king
Con e ence (WCNC). IEEE.
46. Kim, Y., Lee, H., & Cho, Y.S. (2020). Fas ini ial access ech-
nique o millime e -wa e sys ems using e odi ec i e a ays.
In e na ional Con e ence on In o ma ion and Communica ion
Technology Con e gence (ICTC).
47. Wang, J., Chen, X., Liu, X., & Dudek, G. (2021). Uwb-assis ed
as mmwa e beam alignmen . ICC 2021-IEEE In e na ional Con-
e ence on Communica ions. IEEE.
48. Aldalbahi, A., Siasi, N., Mazin, M.A. A. & Jasim. (2022). Digi al
compass o mul i-use beam access in mmwa e cellula ne -
wo ks. Digi al Communica ions and Ne wo ks.
49. Zia, M.S., Blough, D.M., & Wei naue , M.A. (2022). Uplink
powe con ol and sn -dependen beam alignmen e o s in
mmwa e cellula ne wo ks. IEEE 33 d Annual In e na ional Sym-
posium on Pe sonal, Indoo and Mobile Radio Communica ions
(PIMRC). IEEE.
50. Lin, J., & Wei naue , M. A. (2019). Pulse-le el beam-swi ching
o e ahe z ne wo ks. Wi eless Ne wo ks, 25(6), 3047–3062.
51. Deu schmann, B.J., Wilding, T., La sson, E.G., & Wi isal, K.
(2022). Loca ion-based ini ial access o wi eless powe ans e
wi h physically la ge a ays. In: 2022 IEEE In e na ional Con-
e ence on Communica ions Wo kshops (ICC Wo kshops), pp.
127–132. IEEE.
52. Zhang, T., Wang, Y., Ouyang, M., Xing, L., & Gao, F. (2024).
Thi d-pa y came a aided beam alignmen eal-wo ld p o o ype
o mmwa e communica ions. In: 2024 IEEE Wi eless Commu-
nica ions and Ne wo king Con e ence (WCNC), pp. 1–6. IEEE.
53. Wu, D., Zeng, Y., Jin, S., & Zhang, R. (2023). En i onmen -awa e
hyb id beam o ming by le e aging channel knowledge map. IEEE
T ansac ions on Wi eless Communica ions., 23(5), 4990–5005.
54. Zheng, K., Yang, H., Ying, Z., Wang, P., & Hanzo, L. (2023).
Vision-assis ed millime e -wa e beam managemen o nex -
gene a ion wi eless sys ems: Concep s, solu ions, and open chal-
lenges. IEEE Vehicula Technology Magazine, 18(3), 58–68.
55. A nold, M., Majo , B., Massoli, F.V., So iaga, J.B., & Behboodi,
A. (2024). Vision-assis ed digi al win c ea ion o mmwa e beam
managemen . a Xi p ep in a Xi :2401.17781.
56. Ma, W., Qi, C., & Li, G. Y. (2020). Machine lea ning o beam
alignmen in millime e wa e massi e mimo. IEEE Wi eless Com-
munica ions Le e s, 9(6), 875–878.
57. Cha aa, I., Belmega, E.V., & Debbah, M. (2020). Exploi ing
channel spa si y o beam alignmen in mmwa e sys ems ia
exponen ial lea ning. IEEE In e na ional Con e ence on Commu-
nica ions Wo kshops (ICC Wo kshops). IEEE.
58. Lin, C.H., Kao, W.C., Zhan, S.Q., & Lee, T.S. (2019). Bsne :
A deep lea ning-based beam selec ion me hod o mmwa e
communica ions. IEEE 90 h Vehicula Technology Con e ence
(VTC2019-Fall). IEEE.
59. A in e, M., Ta a es, M., & Sama dzija, D. (2019). Beam man-
agemen in 5g n using geoloca ion side in o ma ion. 53 d Annual
Con e ence on In o ma ion Sciences and Sys ems (CISS). IEEE.
60. Heng, Y., & And ews, J. G. (2021). Machine lea ning-assis ed
beam alignmen o mmwa e sys ems. IEEE T ansac ions on Cog-
ni i e Communica ions and Ne wo king, 7(4), 1142–1155.
61. Laza e , V., & Fokin, G. (2020). Beam o ming and spa ial mul-
iplexing pe o mance e alua ion in 5g ul a-dense ne wo ks. In:
In e na ional You h Con e ence on Elec onics, Telecommunica-
ions and In o ma ion Technologies: P oceedings o he YETI
2020, S . Pe e sbu g, Russia, pp. 709–717. Sp inge .
62. Wang, Z., Shan, W., Zhang, L., Ma, S., Xiao, M., Wei, N., & Li, S.
(2022). In elligen beam aining wi h deep con olu ional neu al
ne wo k in mmwa e communica ions. GLOBECOM 2022-2022
IEEE Global Communica ions Con e ence. IEEE.
63. Vankayala, S.K., Kuma , S., Thi umulana han, D., Ma hu , A.,
Yoon, I. S., & Kommineni. (2022). Deep-lea ning based beam
selec ion echnique o 6g millime e wa e communica ion. IEEE
33 d Annual In e na ional Symposium on Pe sonal, Indoo and
Mobile Radio Communica ions (PIMRC). IEEE.
64. K a i, A., Ghauch, H., & Rekaya, G. (2022). Ma ix ac o iza-
ion o blind beam alignmen in massi e mmwa e mimo. IEEE
Wi eless Communica ions and Ne wo king Con e ence (WCNC).
IEEE.
65. Heng, Y., Mo, J., & And ews, J. G. (2022). Lea ning si e-speci ic
p obing beams o as mmwa e beam alignmen . IEEE T ansac-
ions on Wi eless Communica ions, 21(8), 5785–5800.
66. Heng, Y., & And ews, J.G. (2022). G id-less mmwa e beam
alignmen h ough deep lea ning. GLOBECOM 2022-2022 IEEE
Global Communica ions Con e ence. IEEE.
67. TNa enge ile, T.J., Pa as, P., & Ra na ajah, T. (2022). Deep ein-
o cemen lea ning-based beam aining wi h ene gy and spec al
e iciency maximisa ion o millime e-wa e channels. EURASIP
Jou nal on Wi eless Communica ions and Ne wo king 2022(1),
110
68. Duan, Q., Kim, T., Ghauch, H., & Wong, E.W. (2020). Enhanced
beam alignmen o millime e wa e mimo sys ems: A kol-
mogo o model. GLOBECOM 2020-2020 IEEE Global Com-
munica ions Con e ence. IEEE.
123
87 Page 52 o 53 S. Madhekwana e al.
69. Yang, P., Wu, W., Zhang, N., Shen, X., Yang, P., Wu, W...., & Shen,
X. (2021). Machine lea ning-based beam alignmen in mmwa e
ne wo ks. Millime e -Wa e Ne wo ks: Beam o ming Design and
Pe o mance Analysis, 37–71.
70. D ei ue s , R.M., & Hea h, R.W. (2024). Hie a chical ml code-
book design o ex eme mimo beam managemen . IEEE T ansac-
ions on Machine Lea ning in Communica ions and Ne wo king.
71. K unz, M., Aykin, I., Sa ka , S., & Akgun, B. (2023). Online
ein o cemen lea ning o beam acking and a e adap a ion in
millime e -wa e sys ems. IEEE T ansac ions on Mobile Compu -
ing, 23(2), 1830–1845.
72. Ozma , U., Yazici, M. A., & Demi kol, M. F. (2023). Secu e ini ial
access and beam alignmen using deep lea ning in 5g and beyond
sys ems. IEEE Access., 12, 46–59.
73. Dinh-Van, S., Mo, Y.K., Zidan, J., Fowle , D.S., E ans, A.,
Higgins, M.D., e al. (2024). Rapid beam aining a e ahe z
equency wi h con ex ual mul i-a med bandi lea ning. In: 2024
IEEE In e na ional Con e ence on Indus ial Technology (ICIT),
pp. 1–7. IEEE.
74. Xue, Q., Guo, J., Zhou, B., Xu, Y., Li, Z., & Ma, S. (2023).
Ai/ml o beam managemen in 5g-ad anced. a Xi p ep in
a Xi :2309.10575
75. Yang, J., Zhu, W., Tao, M., & Sun, S. (2023). Hie a chical beam
alignmen o millime e -wa e communica ion sys ems: A deep
lea ning app oach. IEEE T ansac ions on Wi eless Communica-
ions., 23(4), 3541–3556.
76. Dan as, Y., I u ia-Ri e a, P.E., Zhou, H., Ba and, M., Elsayed,
M., Gaigalas, R., & E ol-Kan a ci, M. (2023). Beam selec ion o
ene gy-e icien mmwa e ne wo k using ad an age ac o c i ic
lea ning. In: ICC 2023-IEEE In e na ional Con e ence on Com-
munica ions, pp. 5285–5290. IEEE.
77. Mohseni a ani, M., Ali, S., Raja he a, N., & La a-Aho, M.
(2023). Deep con ex ual bandi s lea ning-based beam selec-
ion o mmwa e mimo sys ems. In: 2023 IEEE In e na ional
Medi e anean Con e ence on Communica ions and Ne wo king
(Medi Com), pp. 211–216. IEEE.
78. Shah, S. H. A., & Rangan, S. (2023). Ls m-aided selec i e beam
acking in mul i-cell scena io o mmwa e wi eless sys ems.
IEEE T ansac ions on Wi eless Communica ions, 23(2), 890–907.
79. Sheng, Y., Xu, J., & Tao, X. (2023). A wo-s age p edic ion-based
beam selec ion algo i hm in mmwa e massi e mimo sys ems. In:
2023 IEEE 34 h Annual In e na ional Symposium on Pe sonal,
Indoo and Mobile Radio Communica ions (PIMRC), pp. 1–6.
IEEE.
80. Hegde, D.N., Sa ajlic, M., Sa ka , A., e al. (2023). Machine
lea ning (ml)-assis ed beam managemen in millime e (mm)
wa e dis ibu ed mul iple inpu mul iple ou pu (d-mimo) sys-
ems. a Xi p ep in a Xi :2401.05422
81. Huang, H., Gui, G., Gacanin, H., Yuen, C., Sa i, H., & Adachi, F.
(2023). Deep egula ized wa e o m lea ning o beam p edic ion
wi h limi ed samples in non-coope a i e mmwa e sys ems. IEEE
T ansac ions on Vehicula Technology, 72(7), 9614–9619.
82. Zheng, K., Yang, H., Ying, Z., Wang, P., & Hanzo, L. (2023).
Vision-assis ed millime e -wa e beam managemen o nex -
gene a ion wi eless sys ems: Concep s, solu ions, and open chal-
lenges. IEEE Vehicula Technology Magazine, 18(3), 58–68.
83. Xiao, Z., Chen, S., & Zeng, Y. (2024). Simul aneous mul i-beam
sweeping o mmwa e massi e mimo in eg a ed sensing and com-
munica ion. IEEE T ansac ions on Vehicula Technology., 73(6),
8141–8152.
84. Chen, K., Qi, C., Dob e, O. A., & Li, G. Y. (2023). Simul aneous
beam aining and a ge sensing in isac sys ems wi h is. IEEE
T ansac ions on Wi eless Communica ions., 23(4), 2696–2710.
85. Chen, W., Li, L., Chen, Z., Liu, Y., Ning, B., & Quek, T. (2024).
Isac-enabled beam alignmen o e ahe z ne wo ks: Scheme
design and co e age analysis. IEEE T ansac ions on Vehicula
Technology.
86. Cui, Y., Nie, J., Cao, X., Yu, T., Zou, J., Mu, J., & Jing, X. (2024).
Sensing-assis ed high eliable communica ion: A ans o me -
based beam o ming app oach. IEEE Jou nal o Selec ed Topics
in Signal P ocessing.
87. Jiang, H., Wang, Z., & Liu, Y. (2024). Sense- hen- ain: A no el
beam aining design o nea - ield mimo sys ems. a Xi p ep in
a Xi :2402.15634
88. Li, L., Chen, W., Chen, Z., Hu, T., Mei, W., & Ning, B. (2024).
Enhancing e ahe z communica ions co e age wi h isac-assis ed
beam managemen . IEEE Wi eless Communica ions, 31(1), 34–
40.
89. Ta iq, S., A e o, B.E., Khalid, U., Kim, S., Duong, T.Q., & Shin,
H. (2024). Deep quan um- ans o me ne wo ks o mul i-modal
beam p edic ion in isac sys ems. IEEE In e ne o Things Jou nal.
90. Yanpeng, C., Qixun, Z., Zhiyong, F., Wen, Q., Ying, Z., Zhiqing,
W., & Ping, Z. (2024). Sensing-assis ed accu a e and as beam
managemen o cellula -connec ed mmwa e ua ne wo k. China
Communica ions.
91. Chen, H., Liu, L., Xue, S., Sun, Y., & Pang, J. (2023). Ac i e
sensing o beam managemen : A deep-lea ning app oach. In:
2023 IEEE Wi eless Communica ions and Ne wo king Con e -
ence (WCNC), pp. 1–6. IEEE.
92. Jiang, T., Soh abi, F., & Yu, W. (2023). Ac i e sensing o wo-
sided beam alignmen and e lec ion design using ping-pong
pilo s. IEEE Jou nal on Selec ed A eas in In o ma ion Theo y,
4, 24–39.
93. Yang, J., Wen, C.-K., Yang, X., Xu, J., Du, T., & Jin, S. (2023).
Mul i-domain coope a i e slam: The enable o in eg a ed sens-
ing and communica ions. IEEE Wi eless Communica ions, 30(1),
40–49.
94. Li, Y. N. R., Gao, B., Zhang, X., & Huang, K. (2020). Beam man-
agemen in millime e -wa e communica ions o 5g and beyond.
IEEE Access., 8, 13282–13293.
95. Wang, X., Kong, L., Kong, F., Qiu, F., Xia, M., A non, S., & Chen,
G. (2018). Millime e wa e communica ion: A comp ehensi e
su ey. IEEE Communica ions Su eys & Tu o ials, 20(3), 1616–
1653.
96. Yi, W., Zhiqing, W., & Zhiyong, F. (2024). Beam aining and
acking in mmwa e communica ion: A su ey. China Commu-
nica ions.
97. Khan, M. Q., Gabe , A., Schulz, P., & Fe weis, G. (2023).
Machine lea ning o millime e wa e and e ahe z beam man-
agemen : A su ey and open challenges. IEEE Access, 11,
11880–11902.
98. Khan, M. Q., Gabe , A., Schulz, P., & Fe weis, G. (2023).
Machine lea ning o millime e wa e and e ahe z beam man-
agemen : A su ey and open challenges. IEEE Access, 11,
11880–11902.
99. Häge , S., Sabbah, B., & Wie eld, C. (2024). Design aspec s
o e icien beam aining in 6g millime e -wa e ne wo ks wi h
mobile use s. cni.e i . u-do mund.de.
100. Ning, B., Wang, P., Li, L., Chen, Z., & Fang, J. (2022). Mul i-
i s-aided mul i-use mimo in mmwa e/ hz communica ions: A
space-o hogonal scheme. IEEE T ansac ions on Communica-
ions, 70(12), 8138–8152.
101. Heng, Y., & And ews, J.G. (2022). G id-less mmwa e beam
alignmen h ough deep lea ning. GLOBECOM 2022-2022 IEEE
Global Communica ions Con e ence. IEEE.
102. Klau au, A., González-P elcic, N., & Hea h, R. W. (2019). Lida
da a o deep lea ning-based mmwa e beam-selec ion. IEEE
Wi eless Communica ions Le e s, 8(3), 909–912.
103. Jain, J.K., & Chauhan, D. (2023). Inno a i e model o in e ne
o hings o indus ial applica ions. In: Oppo uni ies and Chal-
123

Beam alignmen o mmWa e and THz: sys ema ic... Page 53 o 53 87
lenges o Indus ial IoT in 5G and 6G Ne wo ks, pp. 95–118. IGI
Global, ???.
104. Mye s, N. J., Mezghani, A., & Hea h, R. W. (2018). Swi -link:
A comp essi e beam alignmen algo i hm o p ac ical mmwa e
adios. IEEE T ansac ions on Signal P ocessing, 67(4), 1104–
1119.
105. Candès, E. J. (2006). Comp essi e sampling. P oceedings o he
In e na ional Cong ess o Ma hema icians., 3, 1433–1452.
106. Cousik, T.S., Shah, V.K., E pek, T., Sagduyu, Y.E., & Reed, J.H.
(2022). Deep lea ning o as and eliable ini ial access in ai-
d i en 6g mm wa e ne wo ks. IEEE T ansac ions on Ne wo k
Science and Enginee ing.
107. Ma, K., Wang, Z., Tian, W., Chen, S., & Hanzo, L. (2021). Deep
lea ning o beam-managemen : S a e-o - he-a , oppo uni ies
and challenges. a Xi p ep in a Xi :
108. Su on, R.S., & Ba o, A.G. (2018). Rein o cemen Lea ning: An
In oduc ion. MIT p ess, ???.
109. Sa a anan, R., & Suja ha, P. (2018). A s a e o a echniques on
machine lea ning algo i hms: a pe spec i e o supe ised lea ning
app oaches in da a classi ica ion. In: 2018 Second In e na ional
Con e ence on In elligen Compu ing and Con ol Sys ems (ICI-
CCS), pp. 945–949. IEEE.
110. LeCun, Y., Bengio, Y., & Hin on, G. (2015). Deep lea ning.
na u e, 521(7553), 436–444.
111. Chowdhu y, M.Z., Shahjalal, S., Ahmed, M., & Jang, Y.M. (2020).
6g wi eless communica ion sys ems: Applica ions, equi emen s,
echnologies, challenges, and esea ch di ec ions. IEEE Open
Jou nal o he Communica ions Socie y. 1, 957–975
112. Huang, C., Hu, S., Alexand opoulos, G. C., Zappone, A., Yuen,
C., Zhang, R., & Debbah, M. (2020). Holog aphic mimo su aces
o 6g wi eless ne wo ks: Oppo uni ies, challenges, and ends.
IEEE Wi eless Communica ions, 27(5), 118–125.
113. Khan, M. Q., Gabe , A., Schulz, P., & Fe weis, G. (2023).
Machine lea ning o millime e wa e and e ahe z beam man-
agemen : A su ey and open challenges. IEEE Access., 11,
11880–11902.
Publishe ’s No e Sp inge Na u e emains neu al wi h ega d o ju is-
dic ional claims in published maps and ins i u ional a ilia ions.
123