lms1.m

自适应算法 lms仿真

% Demonstration of Wiener filter,LMS filter,Steep-descent algorithm
clear;
clc;
N = 10000;          %----- the length of the observation sequence
M = 2;              %----- the filter length
v = randn(1,N);     %----- white process as the AR excitation
a = poly(sign(randn(1,M)).*rand(1,M));      %----- coefficient of AR process
u = filter(1,a,v);          %-----the input sequence
d = v;              %----- the desired response
rf = xcorr(u,M,'biased');
rv = rf(M+1:2*M+1);
R = toeplitz(rv);       %----- the correlation matrix of the input
pf = xcorr(d,u,M,'biased');
pv = pf(M+1:2*M+1).';   %----- the cross-correlation vector between the input and the desired response
%----- the optimal tap weight vector for Wiener filter-----
wopt = inv(R) * pv;
[V,D] = eig(R);         %-----selection of a stable step size mu
lambda_max = max(diag(D));
mu = 0.9 * 2/lambda_max;
%----- the steepest descent learning-----
wsd = randn(M+1,1);                 %-----initial weight vector for steepest descent
total_iteration_number = 100;       %-----total iteration number         
for i=1:total_iteration_number
    wsd = wsd + mu * (pv - R*wsd);
end
%----- the LMS learning-----
wlms = randn(M+1,1);                %-----initial weight vector for LMS
uv = zeros(M+1,1);                  %-----initial input vector
mu = 0.1*2/lambda_max               %-----step size mu for LMS
for n=1:N;
    uv(2:M+1) = uv(1:M);
    uv(1) = u(n);
    
    y = wlms' * uv;
    e = d(n) - y;
    wlms = wlms + mu * uv * conj(e);
end;