Sie sind auf Seite 1von 15

1.

Capacity of MIMO for Ethiotelecoms LTE


Since the capacity, C, of a MIMO system is a function of channel matrix H, if H
is random, so is C. Therefore, when the channel varies randomly C is a
random variable that can be characterized in terms of all the various statistics
that are applicable to any random variable. The simplest statistic is the mean
of the capacity. In MIMO and information theory literature, the mean value of
the capacity is called the ergodic capacity. Ergodic capacity is an
appropriate capacity metric for channels that vary quickly, where the channel
is ergodic over the duration of one codeword.we show that when there is no
direct path between the transmitter and receiver as in some highly built dense
area in Addis Ababa , the magnitude of the elements of H is Rayleigh
distributed. This means that Re{hij} and Im{hij} are N(0, ). Let the
magnitude of the elements of H be Rayleigh distributed with = 1/2, which
implies that E{|hij|}= 1. If frequency-selective Rayleigh fading channel
impulse response has no corre-lations, then its ergodic capacity is the same
as that of the frequency flat Rayleigh fading channel. In LTE frequency
selective fading typically takes the approach of dividing the channel
bandwidth into parallel flat fading channels.When the channel is time-varying
channel capacity has multiple definitions, depending on what is known about
the channel state or its distribution at the transmitter and/or receiver.

1.1A Set of Parallel Independent AWGN Channels


When multiple antennas are present simultaneously at both the transmitter
and receiver, the capacity achieving scheme consists of sending multiple
symbols per transmission period. The transmission and reception of each
symbol relies on pre- and post-processing that is matched to the underlying
structure of the channel based on its singular value decomposition (SVD).
This pre- and post-processing allows for the extraction of a spatial route for
communication (simply a SISO channel) for each transmitted symbol. Multiple
pairs of pre- and post-processing create multiple spatial routes. Those
multiple spatial routes are independent and the MIMO system becomes
equivalent to a set of independent SISO(traditional) channels. Hence the
capacity becomes the sum of the capacity of each SISO channel.
1.1.1Singular Value Decomposition of H
The SVD decomposition of the channel matrix is fundamental in
understanding MIMO systems:
(a) it extracts the equivalent independent AWGN channel structure,
(b) it gives the maximum number of streams that can be multiplexed
simultaneously, or rank of the matrix
(c) it provides a very simple way to compute the capacity which becomes the
sum of AWGN channel capacity.

The SVD of the channel matrix H is:

. Eq1.1
Both MRMR matrix U and MTMT matrix V are unitary matrices. _ is a
MRMT diagonal matrix with nonnegative singular values k, k = 1, . . . , Mmin,
where Mmin =min(MT,MR). For convenience, the singular values are ordered
decreasingly: 1 2 Mmin . The ks are called the eigenmodes of the
channel.
Singular Values and Channel Energy: The following relationship between
channel energy and singular values will be useful.
. Using the relationships, the following result is obtained:

Eq1.2

..Eq1.3
Where Q=U
Each independent channel is also called an eigenchannel as the associated
channel coefficient is an eigenvalue of the channel matrix. Alternatively it is
sometimes referred to as a subchannel. The entire MIMO channel is equivalent
to the set of all the eigenchannels, each of which has a different SNR.
each inputoutput relationship whichdescribes an AWGN channel is given by
yk = k xk + nk Eq.1.4
Furthermore, as the additive noises nk are all independent from each other,
those AWGN subchannels are all independent from each other, forming a set
of parallel AWGN channels. This means that an optimal coding can be done
independently for each AWGN subchannel. Thus, the capacity of the MIMO
system is the sum of the individual capacities.
In this thesis, capacity is given for the following two cases.
1. The receiver knows the instantaneous value of the CSIR. The transmitter
knows the instantaneous value of the CSIT.This is called closed loop MIMO
used for low and medium speed users.
2. The receiver knows the instantaneous value of the CSIR. The transmitter
does not know the instantaneous value of the CSIT but knows its distribution.
This is called open loop MIMO and used for high speed users
,
The MIMO channel capacity can be given by its time average. In practice, we
assume that the random channel is an ergodic process. Then, we should
consider the following statistical notion of the MIMO channel capacity:

.Eq.1.5.

1.2 Open loop MIMO


Under the CSIR only assumption, the transmitter does not have knowledge
about the communications channel, so there is no basis for transmitting
signals in any sort of preferential way on different antennas. This fact has two
implications:
1. There is no reason to transmit more energy on one antenna than
another; thus, the average signal power should be the same on each
transmit antenna.
2. There is no reason to introduce correlation or dependence between
antennas.
The capacity of a MIMO channel with only CSIR can be interpreted as the
sum of r SISO channels, each having power gain, i, i = 1, . . . , r, where
the effective transmit power of a SISO channel is 1/Nt times the total
actual transmit power.

Substituting (since the channel is Rayleign) into eq 1.5 we


get and Ex is the power of the transmitted signal
...Eq1.6

where

.Eq1.7

.Eq1.8

Eq.1.9

1.3Closed loop MIMO


In closed loop MIMO, the channel is assumed time invariant (TI) and known
at the transmitter and the receiver. Under those assumptions, the transmitter
has knowledge of the communications channel matrix, so it will be found that
equi-power and uncorrelated transmitted signals are not optimal and that by
using the transmitters knowledge of the channel .In this principle, more
power is allocated to the channel that is in a good condition and less power or
none to the bad channels.

1.3.1Waterfilling( Waterpouring) Algorithm


Let us introduce the following quantities defined for each eigenchannel as:

Let Pk be the transmit power of eigenchannel k (the power in xk). Pkk can be
seen as the SNR of the kth eigenchannel. The capacity of each eigenchannel
with transmit power Pk is the capacity of an AWGNchannel with SNR = Pkk: it
is equal to log2(1 + Pkk). Pk is adjusted to maximise the capacity of the MIMO
system while complying with the overall transmit power constraint:
Pk P . In general, Pk depends on all nonzero singular values, through the
power constraint.The capacity of the MIMO system is the sum of the individual
capacities with optimized transmit power per eigenchannel.
The optimisation problem can be solved using the method of Lagrangian
multipliers. The results on capacity can be summarised as follows
Since OFDM changes the whole spectrum into narrow band.
1. Set the iteration count p to 1 and compute by solving the following
equation:

2. Using the value of obtained above, solve for the power, Popt i , for the ith
eigen-channel using the following equation:

3. If the power is allocated to the channel with the lowest gain (i.e., if Popt

, discard that channel by setting Popt rp+1 = 0 and rerun the algorithm with
the iteration count p incremented V by 1.
4. Repeat steps 13 until all channels have been allocated power.
The received signal y can be rewritten in a matrix form as follows:

.Eq.2.1
..Eq.2.2

the output signal which is post coded in the receiver can be written as

.Eq.2.3

Where .Eq.2.4
Using the singular value decomposition

.Eq.2.5
which is equivalent to the following r virtual SISO channels, that is,

.Eq.2.6

.Eq.2.7
Using water filling algorithm we can get

.Eq.2.8
MATLAB_ Program 1: Ergodic Channel Capacity for Various
Antenna Configurations when CSI unknown at the transmitter
% This program plots Ergodic_Capacity_vs_SNR.m
%compute the ergodic capacity
%of the MIMO channel as SNR is varied,
%when CSI is not known at the transmitter side(for High speed Users)
%the ergodic channel capacity as varying the number of antennas,
%under the same conditions

clear all, close all


SNR_dB=[0:5:20]; SNR_linear=10.^(SNR_dB/10.);
N_iter=1000;
for Icase=1:5
if Icase==1, nT=4; nR=4; % 4x4
elseif Icase==2, nT=2; nR=2; % 2x2
elseif Icase==3, nT=1; nR=1; % 1x1
elseif Icase==4, nT=1; nR=2; % 1x2
else nT=2; nR=1; % 2x1
end
n=min(nT,nR); I = eye(n);
C(Icase,:) = zeros(1,length(SNR_dB));
for iter=1:N_iter
H = sqrt(0.5)*(randn(nR,nT)+j*randn(nR,nT));
if nR>=nT, HH = H'*H; else HH = H*H'; end
for i=1:length(SNR_dB) %random channel generation
C(Icase,i) = C(Icase,i)+log2(real(det(I+SNR_linear(i)/nT*HH)));
end
end
end
C = C/N_iter;
figure, plot(SNR_dB,C(1,:),'b-o', SNR_dB,C(2,:),'b-<', SNR_dB,C(3,:),'b-s');
hold on, plot(SNR_dB,C(4,:),'b->', SNR_dB,C(5,:),'b-^');
xlabel('SNR[dB]'); ylabel('Ergodic Capacity bps/Hz'); set(gca,'fontsize',8);
grid on
s1='{\it N_T}=1,{\it N_R}=1'; s2='{\it N_T}=1,{\it N_R}=2';
s3='{\it N_T}=2,{\it N_R}=1'; s4='{\it N_T}=2,{\it N_R}=2';
s5='{\it N_T}=4,{\it N_R}=4';
legend(s1,s2,s3,s4,s5)
title('the ergodic channel capacity as varying the number of antennas')
Plot1. Ergodic capacity vs SNR

Observation and Interpretation


Increasing the number of receive antennas leads to logarithmic growth in
capacity( comparing {2Rx,1Tx} with {1Rx,2Tx} ), while simultaneously
increasing the number of receive and transmit antennas leads to linear
growth. Increasing the number of transmit antennas, on the other hand,
provides only a bounded increase in capacity.So a great gain in capacity is
obtained by increasing the number of antennas both in receive and transmit
sides.
MATLAB_ Program 2: Ergodic Channel Capacity for Various
Antenna Configurations when CSI known at the transmitter
using waterfilling algorithm and The pdf of the matrix "lanada"
( V in SVD decomposition) elements using SVD decomposition
% in this programe a highly scattered enviroment is considered. The
% Capacity of a MIMO channel with nt transmit antenna and nr recieve
% antenna is analyzed. The power in parallel channel (after
% decomposition) is distributed as water-filling algorithm
% The pdf of the matrix "lanada" ( V in SVD decomposition) elements is
depicted too.

clear all
close all
clc

nt_V = [1 2 3 2 4];
nr_V = [1 2 2 3 4];

N0 = 1e-4;
B = 1;
Iteration = 1e4; % must be grater than 1e2

SNR_V_db = [-10:3:20];
SNR_V = 10.^(SNR_V_db/10);

color = ['b';'r';'g';'k';'c'];
notation = ['-o';'->';'<-';'-^';'-s'];

for(k = 1 : 5)
nt = nt_V(k);
nr = nr_V(k);
for(i = 1 : length(SNR_V))
Pt = N0 * SNR_V(i);
for(j = 1 : Iteration)
H = random('rayleigh',1,nr,nt);
[S V D] = svd(H);
landas(:,j) = diag(V);
[Capacity(i,j) PowerAllo] =
WaterFilling_alg(Pt,landas(:,j),B,N0);
end
end

f1 = figure(1);
hold on
plot(SNR_V_db,mean(Capacity'),notation(k,:),'color',color(k,:))

f2 = figure(2);
hold on
[y,x] = hist(reshape(landas,[1,min(nt,nr)*Iteration]),100);
plot(x,y/Iteration,'color',color(k,:));
clear landas
end

f1 = figure(1)

legend_str = [];
for( i = 1 : length(nt_V))
legend_str =[ legend_str ;...
{['nt = ',num2str(nt_V(i)),' , nr = ',num2str(nr_V(i))]}];
end
legend(legend_str)
grid on
set(f1,'color',[1 1 1])
xlabel('SNR in dB')
ylabel('Capacity bits/s/Hz')

f2 = figure(2)
legend(legend_str)
grid on
set(f2,'color',[1 1 1])
ylabel('pdf of elements in matrix landa in svd decomposition of marix H')
Title('Ergodic capacity for differnt antenna configuration of open loop MIMO
')
Plot.2.1 Ergodic capacity vs SNR for different antenna configurations

Observation and interpretation


The capacity is always higher when the channel is known than when it is
unknown. This advantage reduces at high SNR values, because at high SNR
values, all the channels perform equally well. The graph of capacity Vs SNR
shows that the capacity of the MIMO channel increases as the number of
antennas used at both the transmitter and receiver increases. The result
obtained shows that there is an improvement in capacity of MIMO channel
when the waterfilling solution is implemented to achieve capacity
maximization is used to allocate different powers to the sub channels.
Plot 2.2 The pdf of the matrix "lanada" ( V in SVD
decomposition) elements

Observation and interpretation


Here, the power allocation with individual matrix element landas for
different antenna selections is being plotted. pdf of element of antenna in
eigen value matrix for matrix H for capacity analysis for
Uncorrelated Vs(precoders). The Power Distribution Function (PDF) of the
matrix and elements (landas), the average allocationn and water filling
allocation depends on rank of the channel matrix H and by applying SVD in
the statistical behavior of MIMO channels. The above figure is showing the
PDF of elements in matrix landa in SVD decomposition of matrix H of nt x nr
and the capacity of a MIMO channel with nt x nr nr antenna varying from 1 to
4 respectively.
MATLAB_ Program 3: Open-Loop vs. Closed-Loop MIMO Channel
Capacity
%This program performs the ergodic capacities for the
% closed-loop and open-loop systems are computed and compared.
%compares the ergodic capacities for 4_4 MIMO channels with and without
%using CSI( channel state information) at the transmitter side.
%The power in parallel channel is distributed as water-filling %algorithm

clear all, close all;


SNR_dB=[0:5:20]; SNR_linear=10.^(SNR_dB/10.);
N_iter=1000;
%%----------------- 4x4 -----------------------------
nT=4; nR=4; n=min(nT,nR); I = eye(n);
rho=0.2; sq2=sqrt(0.5);
Rtx=[1 rho rho^2 rho^3; rho 1 rho rho^2;
rho^2 rho 1 rho; rho^3 rho^2 rho 1];
rho=0.2;
Rrx=[1 rho rho^2 rho^3; rho 1 rho rho^2;
rho^2 rho 1 rho; rho^3 rho^2 rho 1];
C_44_OL=zeros(1,length(SNR_dB)); C_44_CL=zeros(1,length(SNR_dB));
for iter=1:N_iter
Hw = sq2*(randn(4,4) + j*randn(4,4));
H = Rrx^(1/2)*Hw*Rtx^(1/2); tmp = H'*H/nT; SV = svd(H'*H);
for i=1:length(SNR_dB)
%random channel generation
C_44_OL(i) = C_44_OL(i) + log2(det(I+SNR_linear(i)*tmp));
Gamma = Water_Pouringg(SV,SNR_linear(i),nT);
C_44_CL(i) =
C_44_CL(i)+log2(det(I+SNR_linear(i)/nT*diag(Gamma)*diag(SV)));
end
end
C_44_OL = real(C_44_OL)/N_iter; C_44_CL = real(C_44_CL)/N_iter;
figure, plot(SNR_dB, C_44_OL,'-o', SNR_dB, C_44_CL,'-<');
xlabel('SNR [dB]'); ylabel('Capacity in bps/Hz'); set(gca,'fontsize',10);
title('Capacity comparison between closed and open MIMO(Rayleigh channel)');
legend('Channel Unknown by transmitter','Channel Known'); grid on
Matlab Program 4 :Water_Pouring for water-pouring algorithm used
for program 3
%water-pouring power allocation algorithm
%addresses the fact that more power must be allocated to
%the mode with higher SNR

function [Gamma]=Water_Pouring(Lamda,SNR,nT)
Gamma = zeros(1,length(Lamda));
r=length(Lamda); index=[1:r];
index_temp=index;
p=1;
while p<r
irp=[1:r-p+1].';
temp= sum(1./Lamda(index_temp(irp)));
mu = nT/(r-p+1.)*(1+1/SNR*temp);
Gamma(index_temp(irp))=mu-nT./(SNR*Lamda(index_temp(irp)));
if min(Gamma(index_temp))<0
i=find(Gamma==min(Gamma)); ii=find(index_temp==i);
index_temp2=[index_temp([1:ii-1]) index_temp([ii+1:end])];
clear index_temp;
index_temp=index_temp2;
p=p+1;
clear Gamma;
else
p=r;
end
end
Gamma_t=zeros(1,length(Lamda));
Gamma_t(index_temp)=Gamma(index_temp);
%clear Gamma;
Gamma=Gamma_t;
Plot 3 Ergodic capacity of Closed vs open MIMO Systems using
waterpouring algorithm

Observation and Interpretation of closed vs open MIMO


It shows that the closed-loop system provides more capacity than the open-
loop system. However, we can see that the CSI availability does not help to
improve the channel capacity when the average SNR is extremely high. It
implies that even the lowest SNR mode is good enough to get almost the same
transmit power allocated as the highest SNR mode, when the average SNR is
extremely high. In general, we find that the relative performance of these
techniques depends on the signal-to-noise ratio and the relative values of Nt
and Nr. It is clear that the capacity is always higher when the channel is
known than when it is unknown. This advantage reduces at high SNR values,
because at high SNR values, all the channels perform equally well. The reason
for this difference is that under CSIR, the transmit power is assumed to be
equally dividied between the transmit antennas. This shows that a key part of
the eigenbeamforming technique is choosing appropriate powers for each of
the r transmitted symbols.