http://www.paper.edu.cnOn the Channel Capacity of MultipleInput Multiple Output Systems with1Channel Estimation ErrorJun Wang*Shihua Zhu Lei WangSchool of Electronics and Information EngineeringXi’an Jiaotong UniversityXi’an,P.R.China 710049AbstractIn order to study the impact of channel estimation error on the channel capacity of aMultiple Input Multiple Output(MIMO)system,a novel method is proposed toexplore the channel capacity under correlated Rayleigh fading.The system model isconstructed based on the channel estimation error at the receiver side.And using theproperties of Wishart distribution,the lower bound on the channel capacity is derivedwhen the MIMO channel is full rank.And then a method is proposed to select theoptimum set of transmit antennas based on the lower bound on the mean channelcapacity.The proposed method based only on the fading correlation and not theinstantaneous channel realization.And also its computational complexity is low andtends to be realized.The simulation results show that the channel capacity of MIMOsystem is sensitive to the channel estimation error.The channel capacity is maximizedwhen the signal-to-noise increases to a certain point.Antenna selection of transmittercan increase the channel capacity of MIMO system by about 1bit/s in a flat fadingenvironment with rank deficient channel.
Keywords:Multiple Input Multiple Output,Channel Capacity,Spatial Correlation,Antenna Selection1 IntroductionMultiple Input Multiple Output(MIMO)systems provide significant capacity gains over single antennasystems for fading channels through an increased spatial dimension.In[1-2],the author has shown thatthe channel capacity of MIMO system scales linearly with the number of antennas.Recently,mostworks in the literature have investigated the channel capacity with the assumption about the perfectchannel state information at the receiver.However,practical systems seldom have perfect channelknowledge.The imperfect channel state information has an important impact on the MIMO channelcapacity[3-4].In[4],the variation in channel capacity due to small perturbations that do not change the1 Support by the National Doctoral Foundation of China under Grant 20020698024.
1http://www.paper.edu.cnrank of the channel is studied.And the authors have adopted the first order perturbation methods toderive the upper and lower bounds on the MIMO channel capacity.The method is too complex and doesnot consider the model parameter.
In this paper,the system model is constructed based on the channel estimation error at the receiver side.
We study the impact of channel estimation error on the channel capacity of MIMO systems undercorrelated fading.And then using the properties of Wishart distribution,the lower bound on the channelcapacity is derived when the MIMO channel is full rank.When the channel is rank deficient,weproposed a new method to select the optimum set of transmit antennas based on the lower bound of themean channel capacity.The proposed method based only on the fading correlation and not theinstantaneous channel realization.And also its computational complexity is low and tends to berealized.
In this paper,the symbol I N denotes the identity matrix of size N;E(i)the expectation operator;Hi the complex-conjugate;and i denotes the element at the intersection of the ith row and jth()[]ijcolumn of a matrix.
2 System ModelConsider the M-transmit antenna by N-receive antenna MIMO communication system.The receivedsignal is given byy=+Hx n(1)Where y is the N×1channel output;x is the M×1channel input;H is a M×N channel transfer matrix,with element hij denoting the complex fading between the ith receive antenna and the jth transmitantenna;n is a N×1 vector of additive white Gaussian noise(AWGN),with i.i.d.distribution2Nc(0,σn).
In this paper,only the correlation at the receiver end is considered.Recall that H is a complexHGaussian random matrix with E[]HH=M R.R is the spatial fading correlation matrix.If theprobability density function of angular energy is1p()θ=ϕ−∆≤≤+∆θϕ2∆Where 2∆is the range of angles about a central angel-of-departureϕ.The spatial correlation is12ϕ+∆⎛⎞πdR(djd)=exp⎜⎟sin(θ)θ2∆∫ϕ−∆⎝⎠λ12∆⎛⎞πd=Θ+Θexp⎜⎟jt()sin()cos()cos(tdt)sin()2∆∫−∆⎝⎠λFor the uplink,∆is usually small.We can approximate cosz≈1and sin zz≈which gives2http://www.paper.edu.cn⎛⎞⎛⎞22ππdd∆Rd()≈exp⎜⎟⎜⎟j sin(ϕ)sinc cos(ϕ)(2)⎝⎠⎝⎠λλWhere d is the distance between two antennas;λis the carrier wavelength.
The spatial fading correlation matrix R is constructed based on(2).And then the channel matrix H forthe case in which we have correlated transmit antenna is modeled asHUR=12(3)Where the entries of U are i.i.d complex circular symmetric Gaussian with mean 0 and variance 1.
We assume that the receiver can measure R accurately.And after obtaining R,the receiver performsMMSE estimation ofU,i.e.,^UU+=ε(4)^^WhereU is the estimate for U,U andεare uncorrelated,and their entries are i.i.d complex Gaussian,^2with⎛⎞2 andε∼ 0,σ.
⎜⎟U∼ c()0,1+σε()ij c(ε)⎝⎠ij^*The correlation coefficient of Uij and U ij is^⎛⎞*21(5)ρσ=+E⎜⎟UUij ij 1ε=⎝⎠2()1+σεTherefore,we can get that⎡⎤^^11^E UU==µρU=U(6)⎣⎦⎢⎥2 1+σ21+σεε^σ2⎛⎞ε(7)D⎜⎟UUij=2⎝⎠1+σε^The actual matrixU can be written in terms of U as1^(8)U=2 U+η1+σεWhere the entries ofηare i.id complex Gaussian variables with zero-mean and2 22varianceση=+σσεε(1)。
^From the above analysis,the actual MIMO channel matrix H can be written in terms of H as⎛⎞1^^12(9)HURH=⎜⎟2+=+ηα⎝⎠1+σε3http://www.paper.edu.cn^^1 12 12where H=2 UR,αη=R。
1+σε3 Capacity Bound with Channel Estimation ErrorThe received signals can be written as⎛⎞^y=⎜⎟Hxn++α⎝⎠Let Lx+n=α,thenH2 11HH H H11 H H22σηL=+Tr E()ααxx nn=E tr xx RηR+nnMM{}{(())}2 2=σηtr(QR)+σn(10)The lower bound on the channel capacity of MIMO system with channel estimation error is[5]^^H⎡⎤⎛⎞-1CE≥log2⎢⎥det⎜⎟I+RL HQH(11)⎣⎦⎝⎠H H。
where R L=E(LL);Qxx=E[]The lower bound on the channel capacity is obtained byreplacing L by independent zero-mean additive Gaussian noise with the same power2constraint tr()R LL=Mσ.
Using(11),we may,therefore,write the lower bound on the channel capacity as⎡⎛⎞^^H⎤2 2(12)CE12=+^log⎢det⎜⎟I+HQH()σσnηtr()QR⎥H⎣⎝⎠⎦^2Thus using the expressing of H andση,(12)can be written as⎡⎤⎛⎞⎢⎥⎜⎟^^HRUQURH 212CE=log⎢⎥det⎜⎟I+12^2H⎢⎥⎜⎟22⎛⎞2σε⎢⎥⎜⎟()1tr++σσε⎜⎟n 2()QR⎜⎟1+σ⎣⎦⎢⎥⎝⎠⎝⎠ε⎡⎤⎛⎞^^H⎢⎥⎜⎟UQURE log det I+(13)=^2 2U⎢⎥⎜⎟2222⎢⎥⎜⎟()()11tr+++σσεεnσσε()QR⎣⎦⎝⎠()These instructions apply to everyone,regardless of the formatter being used.
4http://www.paper.edu.cn3.1 Channel Capacity of MIMO System when Channel is full RankIf the channel state information at transmitter is not available,the optimal transmission strategy is toevenly allocate power to each antenna.We assume that the total transmitted power is P,Hthen Q=E()xx=()PNI N.So(13)can be simplified as⎡⎤⎛⎞^^H⎢⎥⎜⎟P UURCE=log det I+12^2U⎢⎥⎜⎟2222⎢⎥⎜⎟11tr+++σσεεNPnσσε()R⎝⎠(()())⎣⎦⎡⎤⎛⎞^^H⎢⎥⎜⎟PUUR≈E log det^2⎢⎥⎜⎟2U 11tr+++σσ222NP2σσR⎢⎥⎜⎟(()εεn()ε())⎣⎦⎝⎠⎛⎞⎛⎞^^H⎜⎟P⎛⎞=++NElog2222 log()det()R^log⎜⎟det⎜⎟UU⎜⎟2222 U⎜⎟()11t+++σσεεNPn()σσεr()R⎝⎠⎝⎠⎝⎠(14)^U 2Where the entries of are i.i.d complex Gaussian variables with zero-mean and variance1+σε.So,as^^H~2。,well known to all that WUU=+∼WMN(),1()σεIN Using the properties of Wishart distribution[6]we can get thatNN2222Ndet()W=+det()()1σχεεI N∏∏21()Mi−+−=+()1σχ21()Mi+ii==11(15)Using(15),we can derive that⎛⎞⎛⎞^^H NΓ−+’(Mi1)ENlog detUU=+1 log 1+σ2+(16)^22⎜⎟⎜⎟()()ε∑U⎝⎠⎝⎠i=1Γ−+()Mi1ln2Finally,using(15)and(16),we can get the lower bound on the channel capacity as⎡⎤⎛⎞P NΓ−+’()Mi1CN≈+⎢⎥1log⎜⎟+logdetR+(17)12222 2()()∑⎢⎥⎜⎟1tr++σσNPσR i=1Γ−+Mi1ln2⎣⎦⎝⎠()εεn()()3.2 Method of Transmit Antenna Selection When MIMO Channel is RankDeficientIn a low scattering environment,the MIMO channel tends to be rank deficient.Therefore,it is judicioususe of fewer transmit antennas to reduce the system cost and coding and decoding complexity ofspace-time codes.In previous works[7],the authors have shown that in a low rank environment,systemcapacity can be increased by using only as many transmit antennas as the rank of the channel.Thereforeconsidering rank deficient MIMO channel with rank r,we select the optimal antennas that increasing^^channel capacity from all the antennas.From H=UR12,we know that the rank of MIMO channel5http://www.paper.edu.cn^depends on the rank of transmit spatial correlation matrix,i.e.rank()H=rank()Rt.The channel^12between the r selected transmit antennas and M receive antennas can be expressed as H Ft=UR.
Where Rt(r×r)is the submatrix of R corresponding to the selected transmit antennas.The entries of^2U are i.i.d complex Gaussian variables with zero-mean and variance1+σε.We suppose that afterantenna selection,the total transmitted power is also P.
Next,the method of transmit antenna selection will be given based on transmit spatial correlationmatrix.At the receiver end,using the selection algorithm,computes the optimal set of transmitantennas and relays to the transmitter using a low rate feedback path.
From the spatial correlation matrix,we can know that tr(Rt)=r.Therefore,using this to(17)we canget the channel capacity of MIMO system after antenna selection is⎡⎤⎛⎞P rΓ’()Mi−+1CN≈+⎢⎥1 log⎜⎟+log det()R+ln 222⎜⎟222()t∑⎢⎥()1++σσεεrPrnσi=1Γ()Mi−+1⎣⎦⎝⎠⎡⎤⎛⎞P rrΓ’()Mi−+1=+Nr⎢⎥1log⎜⎟+log+ln222,⎜⎟222∑∑()ii⎢⎥()1++σσεεrPrnσii==11Γ()Mi−+1⎣⎦⎝⎠(18)In the process of derivation of(18),we use the QR decomposition of spatial correlation matrix,i.e.RQrtt=.And rii,is the ith diagonal entry of r。
Note that the channel capacity has three distinct terms.The first and third terms is function of thechannel rank,number of transmit antennas and SNR at the receiver.The second term is entirelydependent on the spatial correlation at the transmitter end.Therefore,based on equation(18),we givethe method of transmit antenna selection in low scattering environment.This method based only on the。