Foreword: This dissertation stems from an ANRT CIFRE (Convention Industrielle de
FormationparlaReherhe/IndustrialAgreementforTrainingthroughResearh)agreement
betweenTeleomParisTeh/EURECOM,Sophia-Antipolis, andtheRadioAessNetworks
(RESA)groupatFraneTeleomResearhandDevelopment,Paris. Theondutedresearh
workwas fullyfundedbyFraneTeleomResearh andDevelopment (OrangeLabs).
Themain fous ofthe thesisis user seletionand linear preoding in multiuser
multi-antennasystemswithlimitedfeedbak. Weprovidebelowanoutlineofthedissertationand
desribetheontributionsmadeineahhapter.
Chapter 2-Multi-antennaBroadast Channels
Inthis hapter,wereviewreentfundamental ndingsin MIMO broadasthannels. The
general multi-antenna system model is introdued and apaity results for the broadast
hannelarepresentedunderdierentassumptionsonthequality/amountofCSIT.We
em-phasizeontheardinalimportaneofCSITandtheroleofmultiuserdiversityforahieving
losetooptimumapaity. Capaitysalinglawsforopportunistishedulingunder
dier-ent hannel statistial distributions are provided. The apaity growth for networks with
path loss and fading is a ontribution of this hapter. Finally, we present in detail
lin-earpreodingstrategiesombinedwithshedulingusing limitedfeedbak,whihforms the
building blokofthedissertation. Theadvantages anddrawbaksof thissetting are
iden-tied, motivatingourworkand thesolutionsproposed inthesubsequenthapters. Partof
•
D.Gesbert,M.Kountouris,R.W.Heath,Jr.,C.-B.Chae,andT.Sälzer,"FromSingle Userto MultiuserCommuniations: Shiftingthe MIMO Paradigm,"in IEEE SignalProessingMagazine,Speial IssueonSignalProessingforMultiterminalCommun.
Systems,vol.24,no.5,pp. 36-46,Sept. 2007.
Chapter 3- Enhaned MultiuserRandomBeamforming
Theontributionsofthishapteraretwo-fold: Intherstpart,weprovideanunpublished
exatsum-rateanalysis of onventionalrandom beamforming (RBF)[9℄. Capaitysaling
lawsfortheinterferene-limitedregion(highSNR)arederivedusingextremevaluetheory,
showingtheardinalimportaneofmultiuserdiversityinthisregime. Intheseondpart,a
limitedfeedbak-basedshedulingandbeamformingsenariothatbuildsonRBFis
onsid-ered. Weintrodueatwo-stageframeworkthatdeouplestheshedulingandbeamforming
designproblems in twophases. Several renementstrategies, inluding beam power
on-trolandbeamseletion,areproposed,oeringvariousfeedbakredutionandperformane
tradeos. TheommonfeatureoftheseshemesistorestorerobustnessofRBFwithrespet
tosparsenetworksettings(lowtomoderatenumberofativeusers),attheostofmoderate
omplexityinrease.
Theworkin thishapterhasbeenpublished in:
•
M. Kountouris and D. Gesbert, "Robust multi-user opportunisti beamforming for sparse networks," in Pro. 6th IEEE Workshop on Signal Proessing Advanes inWirelessCommuniations(SPAWC2005),pp. 975-979,NewYork,USA,June5-8,
2005(invitedpaper).
andwillappearin:
•
M.Kountouris,D.Gesbert,andT.Sälzer,"EnhanedMultiuserRandom Beamform-ing: Dealingwiththenotsolargenumberofusersase,"IEEEJournalonSel. AreasinCommuniations(JSAC),SpeialIssueonLimitedFeedbakWirelessComm.
Net-works,Ot. 2008.
Chapter 4- ExploitingChannel Struture in MIMO Broadast Channels
Inthis hapter,we onsidermultiuser MIMO hannels orrelatedin either time orspatial
domain,and provideseveraltehniquesthat inreasethesystemthroughputbyexploiting
thehannel struture. Intime orrelatedhannels, anopportunisti beamforming sheme
exploitinghannelmemoryisproposed. Thisshemeisshowntolltheapaitygap with
optimum unitary preoding with full CSIT for slow time-varying hannels. In spatially
orrelated hannels, a maximum likelihood (ML) oarse hannel estimation framework is
established,whih eetivelyombines slowlyvaryingstatistial CSIT -assumedavailable
atthetransmitter -with instantaneouslow-ratefeedbak. A greedyuserseletion sheme
andalow-omplexitySDMA eigenbeamformingtehniquebasedonmultiuser interferene
bounds are also proposed and evaluated. It is demonstrated that, in wide-area ellular
networks,salarCSITfeedbakissuienttoahievenear-optimalthroughputperformane
ifitisproperlyombinedwithlong-termstatistialknowledge.
Theworkin thishapter hasbeenpublished in:
•
M.KountourisandD.Gesert,"Memory-basedopportunistimulti-userbeamforming,"in Pro. of IEEEInternationalSymposium onInformation Theory (ISIT2005), pp.
•
M. Kountouris, D. Gesbert, and L. Pittman, "Transmit Correlation-aided Oppor-tunistiBeamformingandSheduling,"in Pro. of14thEuropeanSignalProessingConferene(EUSIPCO),Florene,Italy,September4- 8,2006(invitedpaper).
•
D.Gesbert,L.Pittman,andM.Kountouris,"TransmitCorrelation-aidedSheduling inMultiuserMIMONetworks,"inPro. IEEEInternationalConfereneonAoustis,Speeh, andSignalProessing(ICASSP2006),Vol.4,pp. 249-252,Toulouse,Frane,
May14-19,2006.
•
M. Kountouris, R. de Franiso, D. Gesbert, D.T.M. Slok, and T. Sälzer, "Low omplexityshedulingandbeamformingformultiuserMIMO systems,"in Pro. 7thIEEEWorkshoponSignalProessingAdvanesinWirelessCommuniations(SPAWC
2006),Cannes,Frane,July2-5,2006.
Chapter 5-LimitedFeedbak Broadast Channelsbased on Codebooks
Thishapterdealswithlimitedfeedbakstrategiesutilizingvetorquantizationodebooks.
In partiular, the problem of eient, sum-rate maximizing hannel quality information
(CQI) feedbak design is addressed. We proposed several salar feedbak metris that
inorporate information on the hannel gain,the hannel diretion, and the quantization
error. Thesemetrisarebuiltuponboundsontheinstantaneousinter-userinterferene,and
anbeinterpreted asreliableestimatesofthe reeived SINR.It is shown thatsalar CQI
feedbakombined withhanneldiretional information (CDI) andeientuser seletion
algorithmanahieveasigniantfrationoftheapaityofthefullCSITasebyexploiting
multiuser diversity. An adaptive sheme transiting from SDMA to TDMA transmission
modeisproposedandisshowntoahievelinearsum-rategrowthatanySNRrange.
Theworkin thishapterhasbeenpublished in:
•
M.Kountouris,R. deFraniso, D. Gesbert, D.T.M.Slok, andT.Sälzer, "Eient metrisforshedulingin MIMObroadasthannels withlimitedfeedbak,"in Pro.IEEEInternationalConfereneonAoustis,Speeh,andSignalProessing(ICASSP
2007),Honolulu,USA,April15-20, 2007.
•
M.Kountouris,R.deFraniso,D.Gesbert,D.T.M.Slok,andT.Sälzer,"Multiuser diversity-multiplexingtradeoinMIMObroadasthannelswithlimitedfeedbak,"inPro. of40thAsilomarConfereneonSignals,Systems&Computers,PaiGrove,
CA,USA,Ot. 29-Nov. 1,2006(invitedpaper).
andaeptedto:
•
M.Kountouris,R.deFraniso,D.Gesbert,D.T.M.Slok,andT.Sälzer,"Exploiting MultiuserDiversityin MIMOBroadastChannelswithLimitedFeedbak,"aeptedtoIEEETrans. onSignalProessing,August2007(underrevision).
Chapter 6-Feedbak Redution using Ranking-basedFeedbak
In this hapter, a low-rate representation of CSIT feedbak parameters, referred to as
ranking-based feedbak, is identiedas ameansto further ompress thereported hannel
feedbak. This representation enables the sheduler to identify users that are
instanta-neouslyonthehighestpeakwithrespettotheirownhanneldistributions,independently
restoredin heterogeneousnetworks withi.ni.d. hannel statistisamong users. Thework
inthishapterhasbeenpublishedin:
•
M.Kountouris,T.Sälzer,andD.Gesbert,"ShedulingforMultiuserMIMODownlink Channels with Ranking-based Feedbak," EURASIP Journalon Advanes in SignalProessing,SpeialIssueonMIMOTransmissionwithLimitedFeedbak,Marh2008.
Chapter 7- SystemAspets in MultiuserMIMO Systems
Thishapterfousesonseveralsystemissuesanddesignhallengesthatariseinreal-world
wireless systems. We disuss the main pratial and implementation hallenges that one
mayfaewhendeployingtehniquesasthoseproposedinChapters3-6. Emphasisisputon
fairnessissuesand theproportionalfairsheduling(PFS)rule isgeneralizedformultiuser
systemsettings,inludingOFDM,SDMA,multiellnetworks,et. Partoftheseresultshas
beenpublishedin:
•
M.Kountourisand D. Gesbert, "Memory-based opportunisti multi-user beamform-ing,"in Pro. ofIEEEInternationalSymposiumonInformationTheory(ISIT2005),pp. 1426-1430,Adelaide,Australia,September4-9,2005.
Patents
Inadditionto theabovepubliations,ourresearhwork resultedin thefollowingpatents:
•
PCTWO2007057568,"Informationenodingforabakwardhannel,"(assigned)•
FR 2893474, "Method of information enoding for abakwardhannel of a SDMA system,userterminalandbasestationofsuhasystem," (assigned).•
"Feedbakommuniationfrom aterminalto atransmitter toredue inter-beam in-terferene,"(led,Jan. 2008).Multi-antenna Broadast
Channels
In this hapter, we review multiuser MIMO ommuniations fousing on the more
hal-lengingdownlink,theso-alledbroadasthannel(BC).Thegeneralmulti-antennasystem
modelisintroduedandknownapaityresultsforthebroadasthannelarepresented
un-derdierentassumptionsregardingtheamountofCSIT.Informationtheoretiresultsshed
lightontheardinalimportaneofCSITandsheduling,aswellasontheroleofmultiuser
diversityforahievingtheoptimumsystemapaity. Capaitysalinglawsforopportunisti
sheduling under dierent hannel models are investigated. Several approahes inluding
non-linearand linearhannel-awarepreodingare reviewed, disussingdesign hoiesand
performane tradeos. Emphasis is given on low-omplexity, linear preoding strategies
ombinedwithshedulingusinglimitedfeedbak,whihformthebuildingblokofthe
dis-sertation.Thelimitedfeedbakmodelthatweadoptandinvestigateinsubsequenthapters
ispresentedindetailanditslimitationsareidentied.
2.1 The Wireless Channel
Thewireless radiohannel isapartiularlyhallengingmediumforreliablehigh-rate
om-muniations. Apartfrombeingsubjettonoise,interfereneandseveralotherimpairments,
thewireless medium is aboveall amultipath time-varying hannel. A signaltransmitted
overaradio hannel issubjetto thephysial lawsof eletromagnetiwavetheory, whih
ditatethat multiplepaths ourasaresultofreetion onlargesurfaes(e.g. buildings,
walls, and ground), diration on edges, and sattering on various objets. Therefore, a
reeived signal is a superposition of multiple signalsarriving from dierent diretions at
dierent time instanes and with dierent phases and power. These paths may ombine
onstrutively or destrutively, reating a multi-tap hannel impulse response, with eah
tap havingrandom phase and time-varying amplitude. We rst review the physial
phe-nomena that attenuate the signalpower. Foramoredetailed presentation,the interested
readerisreferredto [13℄.
2.1.1 Path loss
Pathlossisarange-dependenteetandisduetothedistane
d
betweenthereeiverandthetransmitter. Inidealfreespae,thereeivedsignalpowerisdesribedbytheFriisequation
and follows aninverse square lawpowerloss. Several deterministiand empirial models
have been developed for various ellular environments (miroells, maroells, pioells,
et.), suh as Okumura-Hata, Walsh-Ikegami, and their COST-231 extensions,
plane-earthandlutter fatormodel[13℄. Ageneripathlossmodelisgivenby
L = βd − ǫ
(2.1)where
ǫ
isthepathlossexponentandβ
isasalingfatorthataountsforantennahar-ateristis and average hannel attenuation. The pathlossexponentvaries normallyfrom
2to6,dependingonthepropagationenvironment. Fortheaseof fullspeularreetions
from groundis4,whileforbuildingsandindoorenvironmentsitantakevaluesfrom4to
6.
2.1.2 Shadowing
Shadowing, also known as marosopi or long-term fading, results from large obstales
blokingthemain signalpathbetweenthe transmitterandreeiver,and isdetermined by
theloalmeanofafastfadingsignal. Therandomshadowingeets,whihareinuened
by antennaheights, operatingfrequenyand thefeaturesof thepropagationenvironment,
maybemodeled aslog-normaldistributedwithprobabilitydensityfuntion (PDF):
p(x) = 1 xσ √
2π e (log 2σ x−µ)2 2 x > 0
(2.2)where
µ
andσ
arethemeanandstandarddeviationoftheshadowing'slogarithm.2.1.3 Fading
Fading,oftenreferredtoasmirosopiorsmall-salefading, resultsfrom theonstrutive
ordestrutivesuperpositionofmultipathsanddesribestherapidsignalutuationsofthe
amplitudes,phases,ormultipathdelays. Thestatistialtimevaryingnatureofthereeived
envelopeisommonlydesribedbythefollowingthree fadingdistributions:
Rayleighfading
Rayleighfading is areasonable model when there is nodominantpropagation path (non
line-of-sight, NLOS) betweenthe transmitter andthe reeiverandis used to desribe the
amplitude ofasignalwhen there is alargenumberof independentsattered omponents.
Applying the entral limit theorem, the hannel impulse response an be onsidered asa
omplex-valued Gaussian proessirrespetiveof thedistribution of the individual
ompo-phaseevenlydistributedbetween0and2
π
radians. TheenvelopeofthereeivedsignalwillthereforebeRayleighdistributed withPDFgivenby
p(x) = 2x Ω e − x
2
Ω x > 0
(2.3)where
Ω = E { x 2 }
istheaveragereeivedpower.Rieanfading
Ifadiret,possiblyaline-of-sight(LOS),pathexists,theassumptionofazero-meanfading
proess does no longer hold and the distribution of the signal amplitude is modeled as
Riean. The Riean distribution is often dened in terms of the Riean fator
K
whihdenotestheratio ofthepowerin the meanomponentofthehannel(diret path)to the
powerin thesatteredpaths. TheRieanPDFisgivenby
p(x) = 2x(K + 1)
Ω e − K − (K+1)x
2
Ω I 0 2x
r K(K + 1) Ω
!
x > 0
(2.4)where
Ω = E { x 2 }
andI 0 (x)
is the zero-order modied Bessel funtion of the rst kinddenedas
I 0 (x) = 1 2π
Z 2π 0
e − x cos θ dθ
(2.5)Nakagami fading
Ageneralfading distributionthat ts wellwith empirialmeasureddata is theNakagami
distributiongivenby
p(x; m) = 2m m x 2m − 1
Γ(m)Ω e − mx Ω 2 x > 0
(2.6)where
Ω
is the average reeived powerandm = E{ x 2 Ω − 2 Ω 2 }
. Them
fator determines theseverityof fading, i.e. for
m = ∞
there isno fading. Form = 1
the distribution in (2.6) reduestoRayleighfading,whileform = (K+1) 2 /(2K+1)
thedistributionisapproximately RieanfadingwithfatorK
.2.1.4 Channel Seletivity
Multipathpropagationresultsinthespreadingofthesignalindierentdimensionsaeting
signiantlythereeivedsignal. These dimensionsaretime(Dopplerspread),spae(angle
spread)andfrequeny(delayspread).
Dopplerspread and timeseletive fading
Themotionof thetransmitter,thereeiverorthesatterersresultsin timeseletivity, i.e.
a single tone spreads in frequeny over a nite spetral bandwidth. The variations due
to Doppler shiftsare spei to eah path anddepend on their angle with respet to the
movingdiretion ofthetransmitter/reeiver. DierentDopplershiftsleadto theso-alled
Dopplerspread, whih is themaximumfrequenyspread amongall Dopplershifts, and is
givenby
f m = v λ c
(2.7)
where
v
isthemobilespeedandλ c
isthearrierwavelength.Howfastthehanneldeorrelateswithtimeisspeiedbythetemporalautoorrelation
funtion. The Doppler power spetrum
ρ d (f d )
is dened as the Fouriertransform of thetemporalautoorrelationfuntion ofthehannelresponseto aontinuouswave
ρ d (f d ) =
( 1
πf m √
1 − (f d /f m ) 2 ∀ f d ∈ [ − f m , f m ]
0
elsewhere(2.8)
ThemostommonlyusedmodelfortheautoorrelationfuntionistheClarke-Jakes'model,
whihassumesuniformlydistributedsatterersonairlearoundtheantenna
ρ d (τ) = J 0 (2πf m τ)
(2.9)where
J k
isthek-thorderBesselfuntionoftherstkindandτ
isthesampling interval.Ameasureofthetimeseletivityisthehanneloherenetime
T c
,denedastheintervalover whih the hannel remains stronglyorrelated. The shorter the oherene time, the
fasterthehannelhangesovertime. Theoherenetimeisastatistialmeasureandsatises
T c ∼ 1 f m
(2.10)
AsweshowinChapter4,theshedulerantakeadvantageofthetimeseletivityandbenet
fromtheresultinghannelredundany(timediversity),asameanstofurtherompressthe
hannelfeedbakorsuessivelyrenetheshedulingdeisions.
Delay spread and frequeny seletivefading
Delayspreadis aused whenseveral delayedand saled versions of thetransmitted signal
arriveatdierenttimeinstantsatthereeiver. Thetimedierenebetweenthemaximum
multipath delay
τ max
(typiallythearrivaltimeof theLOS omponent)andtheminimumpathdelay
τ min
isalleddelayspread. Delayspreadausesfrequenyseletivefadingasthehannelatslikeatapped-linelter. Therangeoffrequeniesoverwhihthehannelanbe
onsidered`at'denestheoherenebandwidth
B c
anddependsontheform ofthepowerdelay spetrum (rms delay spread). A hannel is haraterized as at orfrequeny
non-seletiveifthesignalbandwidth
B
issigniantlysmallomparedtothehanneloherenetime, i.e.
B << B c = 1/τ max
. Inthe subsequent hapters,only at fading hannelsareonsidered.
Angle spread and spae-seletivefading
Angle spreadat thereeiver/transmitterrefersto thespreadin angles ofarrival(AoAs) /
angles of departure (AoDs) of the multipath omponent at the reeive/transmit antenna
array,respetively. Thedierentdiretions ofarrivalleadtospatialseletivitythat implies
that signalamplitudedependsonthespatialloationoftheantennaarray. Spaeseletive
fadingisharaterizedbytheoherenedistane
d c
,whihisthemaximumdistanebetweentwo antenna elementsfor whih the fading remains strongly orrelated. An upperbound
fortheoherenedistaneisgivenby
d c ≤ λ c
2 sin(∆θ max /2)
(2.11)where