OnlineEKRLS.m

function [preSignal,Gp,erle] = OnlineEKRLS(trainInput,trainOutput,lamda,delta)
[Np,L] = size(trainInput);
d_sum = 0;
deltad_sum = 0;

% Filter Parameters
laminv  = 1/lamda;
q = 3.8;
beta = 2.2;
alpha=1;
% Filter Initialization
predictorPara = zeros(Np,1);        % Initial value of predictor parameter

A = alpha*eye(Np);
P = delta*eye(Np);          % inverse correlation matrix
%% Do Onling Learning 
fprintf('## Do online learning, Please wait... \n'); 
%>>>>>>>>>>>> Set the waitbar - Initialization <<<<<<<<<<<<<<<<<<
wb1 = waitbar(0, 'RLS Online Training in Progress...');
for K = 1:L
     %>>>>>>>>>>>>>>>>> Display Waitbar <<<<<<<<<<<<<<<<<<<<<<
    waitbar(K/L,wb1)
    set(wb1,'name',['Progress = ' sprintf('%2.1f',K/L*100) '%']);
    %>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

    
    % Acquire chunk of data
    currentPreSequence = trainInput(:,K);
    % Error signal equation
    currentPreSample = predictorPara(:,K)'*currentPreSequence;
    preSignal(K,1) = currentPreSample;
    
    predictError(K,1) = trainOutput(K,1)-currentPreSample;  
    % Parameters for efficiency
    Pi = P*currentPreSequence; 
    % Filter gain vector update
    gain =  A*(Pi./(beta+lamda+currentPreSequence'*Pi));
    % Inverse correlation matrix update
    P = A*((P-gain*currentPreSequence'*P)*laminv)*A' + beta*q*eye(Np);
    % Filter coefficients adaption
    predictorPara(:,K+1) = A*predictorPara(:,K) + gain*predictError(K); 
    
    d_sum       = d_sum + trainOutput(K,1)^2;
    deltad_sum  = deltad_sum + (trainOutput(K,1)- preSignal(K,1))^2;
    erle(K)     = 10*log10((d_sum+eps)/(deltad_sum+eps));

end
close(wb1);

plotEnable = 1;
if plotEnable == 1
    figure;
    plot(trainOutput,'b') ; hold on; grid on;
    plot(preSignal,'r');
    plot(predictError,'g');
    title('training: teacher sequence (blue) vs predicted sequence (red)') ; 
end
%% Prediction gain
Gp = 10*log10((sum(trainOutput.^2)/(sum(predictError.^2))+0.0000001));