• Aucun résultat trouvé

On achievable rates in a multi-antenna gaussian broadcast channel

N/A
N/A
Protected

Academic year: 2022

Partager "On achievable rates in a multi-antenna gaussian broadcast channel"

Copied!
1
0
0

Texte intégral

(1)

On Achivable Rates in a Multi-Antenna Gaussian Broadcast Channel

GiuseppeCaire

Eurecom,Sophia-Antipolis,France

ShlomoShamai

Technion-32000Haifa,Israel

Abstract | A Gaussian broadcast channel with r

single-antennareceivers and t antennas at the trans-

mitter isconsidered. Both transmitter and receivers

haveperfectknowledgeofthechannel. Despiteappar-

entsimplicity,thismodelisingeneralanon-degraded

broadcast channel, for which the capacity region is

not fully known. We propose a novel transmission

schemebasedon\rankedknowninterference"(RKI).

Inbrief,thetransmitterdecomposesthechannelinto

an ordered (or ranked) set of interference channels

forwhichtheinterferencesignalofthei-thchannelis

generatedasalinearcombinationofthesignalstrans-

mitted in channels j < i. In this way, known tech-

niques of coding for non-causally known interference

canbeappliedtomaketheinterferenceineachchan-

nelharmlesswithoutfurtherpowerpenalty. Weshow

that the proposed scheme is throughputwise asymp-

toticallyoptimalforbothlowandhighSNR.Wepro-

vide a modication of the basic RKI scheme which

achievesoptimalthroughputfor all SNRs inthe spe-

cial case of 2 antenna and 2 users. For independent

Rayleigh fading closed-form throughput expressions

areobtained in variouscases ofinterest and numeri-

calexamplesareprovidedfortheinnite-dimensional

Rayleigh channel.

I. Problem statement

The t1:::r Gaussian boradcast channel (GBC) is de-

scribedbyy = Hx+zwhere the notationt1:::r desig-

nates that the r receivers cannot cooperate. Here, x 2 C t

denotesthevectorofsymbolstransmittedinparallelfromthe

tantennasinany channeluse, H 2C rt

denotes thechan-

nelmatrix,whosei;j-thelementisthecomplexchannelgain

from antenna j to antenna i, y 2 C r

denotes the vector of

receivedsignalsattherreceiversandz2C r

isthenoisevec-

tor, N

C (0;N

0

I). The channelmatrixH is knownto the

transmitterandtoallreceiversandtheinputisconstrainedto

satisfy trace (E[xx H

]) E where E isthe maximumallowed

totaltransmitenergyperchanneluse.

WeareinterestedinthechannelthroughputR gbc

(orrate-

sum),denedasthesumofallachievableindividualuserrates.

We shallconsider the following scenarios: i)H is determin-

istic andxed;ii)His xedduring thetransmissionofeach

code word,but it is randomly selectedaccording to agiven

probabilitydistribution(compositechannel).

II. The \rankedknowninterference" scheme

Let H = GQ be a QR-type decomposition of H where

G2C rm

islowertriangularandQ2C mt

hasorthonormal

rows (we dene m

= minfr;tg). The transmitted signal is

obtainedasx=Q H

u. Letg

i;j

denotethe(i;j)-thelementof

Gandletdi=jgi;ij 2

.Theoriginalchannelisturnedintothe

setofinterferencechannelsy

i

=g

i;i u

i +

P

j<i g

i;j u

j +z

i ,for

i=1;:::;m,whilenoinformationissenttousersm+1;:::;r.

H H H

uiare allgeneratedbythetransmitter: foreachi-thchannel

u

i

isthewantedsignaland P

j<i g

i;j u

j

isaninterferencesignal

non-causallyknowntothetransmitter. Withsomecare[5 ],we

canapplytheresults of[2 ]or(equivalently)themorerecent

results of[3 ], and show thatbyusinganappropriatecoding

strategy,the capacityofeachi-thchannelisequalto thatof

thevirtualinterference-freechannely

i

=u

i +z

i :

Proposition 1. Themaximumachievable throughputof

theRKIschemeisgivenbyR rki max

= P

m

i=1 [log(d

i )]

+ where

isthesolutionof P

i=1 [ 1=d

i ]

+

=

=E=N

0

.

III.Results

We denote by R zf max

and by R

coop max

the maximum

achievable throughputs of a conventional system based on

zero-forcing beamforming [4] and of a system where the re-

ceivers are allowed to cooperate (equivalent to the single-

user multiple-antenna case [1 ]). For H given and constant

wehave thefollowing results: 1) ForanychannelmatrixH,

R rki max

R zf max

;2) Foranychannel matrixH withfull

row-rankr,limA!1 R

coop max

R rki max

=0;3)Forany

channelmatrixH,lim

A!0 R

rki max

R zf max

=1.

Notice that for H of rank r the RKI is asymptoti-

cally throughputwise optimalfor large SNR,since obviously

R

coop max

R gbc

.

A generalized RKIstrategy where x = Q H

Ru, where R

is anoptimizeduppertriangularmatrixis shownto achieve

themaximumthroughputofthe21:::2GBC,foranySNR

andanymatrixH[5 ].

InthecaseofrandomHwithi.i.d. Rayleighfading,dier-

entRKIstrategieswithequal-powerandconstant-rateactive

usersarecomparativelyinvestigatedforalargearray(i.e.,for

r;t!1 and r=t xed). It is demonstrated that theequal-

powerschemeperformsveryclosetothewaterllingRKIand

approachestheupperboundofcooperativerateformoderate

and high SNRs[5 ]. The results essentially hold when H is

availabletothetransmitteronly[5 ].

References

[1] E.Telatar,\Capacityofmulti-antennaGaussianchannels,"Eu-

ropeanTrans.onTelecomm.ETT,Vol.10,No.6,pp.585-596,

Nov.1999.

[2] M.Costa,\Writing ondirtypaper,"IEEETrans.onInform.

Theory,Vol.29,No.3,pp.439-441,May1983.

[3] U. Erez, S. Shamai and R. Zamir, \Capacity and lattice-

strategiesforcancellingknowninterference,"ISITA2000,Hon-

olulu,Hawaii(USA),Nov.2000.

[4] P.Baier,M. Meurer,T.WeberandH. Troeger, \Jointtrans-

mission(JT),analternativerationaleforthedownlinkoftime

divisionCDMAusingmulti-elementtransmitantennas,"IEEE

6th Int.Symp. onSpread-Spectrum Tech.&Appl. (ISSSTA),

NJIT,N.J.(USA),Sept.2000.

[5] G.CaireandS.Shamai,\Onachievableratesinamulti-antenna

Gaussian broadcast channel," submitted to IEEE Trans. on

Inform.Theory,April2001.

Références

Documents relatifs

This analysis shows that a practical and sensible strategy to the downlink of a wireless system consists of hybrid TDMA and space-time multiplexing where only a subset of active

interference channels for which the interference signal in the i-th channel is a linear.. combination of the signals transmitted in channels j

We assume that each user estimates the channel gains and accordingly the achievable rate on each sub-channel, and for each antenna in the case of DAS and CAS with spatial

5 shows the spectral efficiency (SE) as a function of the number of users accommodated, using the proposed MMA algorithm on a frequency selective channel with correlated

Massive MIMO makes the pathwise approach viable: the (cross-link) beamformers (BF) can be updated at a reduced (slow fading) rate, parsimonious channel representation facilitates

Massive MIMO makes the pathwise approach viable: the (cross-link) BF can be updated at a reduced (slow fading) rate, parsimonious channel representation facilitates not only uplink

In this section we apply the asymptotic results obtained in the previous sections to a MIMO-BC with K users and an M - antenna UCA at the transmitter; the asymptotic eigenvalues of

The covariance information of the target users and in- terfering users can be acquired exploiting resource blocks where the desired user and interference users are known to be