www.pudn.com > aft.rar > lms_AR_pred.m

% lms_AR_pred.m - use multidimensional LMS algorithm to predict AR process 
% written for MATLAB 4.0 
% 
% Input parameters: 
%                Xi       : matrix of training/test points - each row is 
%                           considered a datum 
%                Xi0      : initial value Xi(0) of series 
%                verbose  : set to 1 for interactive processing 
%                mu       : the initial value of step size 
%                decay    : set to 1 for O(1/n) decay in mu 
 
Nout = size(Xi, 2); 
 
% length of maximum number of timesteps that can be predicted 
N = size(Xi, 1); 
 
% initialize weight matrix and associated parameters for LMS predictor 
W = zeros(Nout, Nout); 
Wo = []; 
 
% compute first iteration with Xi(0) = Xi0 
n	= 1; 
Wo 	= [Wo W]; 
xp(n, :) = Xi0 * W'; 
e(n, :)  = Xi(n, :) - xp(n, :); 
ne(n) = norm(e(n, :)); 
W = W + rp.mu * e(n, : )' * Xi0; 
 
for n = 2:N, 
 
    % save W matrix 
    Wo = [Wo W]; 
 
    % predict next sample and error 
    xp(n, :) = Xi(n-1, :) * W'; 
    e(n, :)  = Xi(n, :) - xp(n, :); 
    ne(n)    = norm(e(n, :)); 
    if (rp.verbose ~= 0) 
      disp(['time step ', int2str(n), ': mag. pred. err. = ', num2str(ne(n))]); 
    end; 
    % adapt weight matrix and step size 
    W = W + rp.mu * e(n, :)' * Xi(n-1, :); 
    if (rp.decay == 1) 
      rp.mu = rp.mu  * n/(n+1); % use O(1/n) decay rate 
    end; 
     
 end