leony
Member level 1
I was trying to implement a MMSE equalizer in Matlab, and come across the following code on the web. The problem is, whenever I searched, I found out that they use tap size the same as the size of discrete channel model. But I have, lets say a channel of
But again I want to use a 5 tap equalizer and my unequalized signal before the equalizer has a size of 8000. In the following code is says for X (unequalized signal)
has a size of 4N+1 X 2N+1 (means 9 x 5 when the tap is 5)
I hope it is not confusing.
Thank you.
This code can be found at
**broken link removed**
Code:
h=[1 0 0 0 0 ........ 0,5 ....0 0 0 0.......0] (the size is 1x100)But again I want to use a 5 tap equalizer and my unequalized signal before the equalizer has a size of 8000. In the following code is says for X (unequalized signal)
has a size of 4N+1 X 2N+1 (means 9 x 5 when the tap is 5)
Code:
% z=xc; z=equalized output; c=tap coeff of filter; x=4N+1 X 2N+1 array;Thank you.
This code can be found at
**broken link removed**
Code:
function [c,z_equalized]=mmse(M,p2,p3,p4)
% function [filter coeff, equalized output]=min_mean_square_error(# of puls
%es total, perc in decimal ISI nearest pulses,
% perc in decimal ISI second nearest puls
%es, perc in decimal ISI third nearest pulses)
% Consider: y -> channel ISI -> x -> equalizer -> z
% z=xc; z=equalized output; c=tap coeff of filter; x=4N+1 X 2N+1 array;
%example3_6 = 1; %Set to 1 for results in example 3.6 in Sklar; set to 0 for
%general results;
if(example3_6==1)
M=7; %number of taps
h=[0.0110 0.0227 -0.1749 1.0 0.1617 -0.0558 0.0108];
%example3.6 results in Sklar; {x[k]} is backward in example;
figure; stem(h);
title('Plot of Channel ISI Impulse Response');
xlabel('Samples n');
ylabel('Amplitude');
else
h=[p4 p3 p2 1.0 p2 p3 p4]; % Channel ISI impulse response;
% If eye diagram is not open
% before equalization, then zero-forcing
% equalizer does not work well.
figure; stem(h);
title('Plot of Channel ISI Impulse Response');
xlabel('Samples n');
ylabel('Amplitude');
end
if(mod(M,2)==0) %if M is even, make it odd;
M=M+1;
end
N=(M-1)/2; %2N+1=M;
if(M<7) %length(h) = 7;
h=h(4-N:4+N); %h(4) is center tap;
elseif(M>7)
h=[zeros(1,(M-7)/2) h zeros(1,(M-7)/2)];
end
%Solve for c=filter weights or coeff. using equation c=(Rxx^-1)(Rxz);
%Tap weights or filter coeff are to be determined by transmitting a single
impulse as a training signal.
%So single impulse (delta function) convolves with channel ISI response
{h[n]} equals {x[n]} = {h[n]}.
%The for loops form the x array = 4N+1X2N+1;
for k1=1:4*N+1
for k2=1:2*N+1
index=k1-k2+1;
if (index<=length(h))&(index>0)
x_arr(k1,k2)=h(index);
else
x_arr(k1,k2)=0;
end
end
end
z=[zeros(2*N,1);1;zeros(2*N,1)]; % z = 4N+1X1 matrix;
x_tranpose=transpose(x_arr);
Rxx=x_tranpose*x_arr; %auto-correlation vector;
Rxz=x_tranpose*z; %cross-correlation vector;
c=inv(Rxx)*Rxz; %filter coeff;
z_equalized=x_arr*c; %equalized output;