www.pudn.com > zxb.rar > demo_BSS.m

clear all;  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% global demo_id;  % Set rand number seeds. seed=9;randn('state',seed);rand('state',seed);  demo_id 		= 2;	% Used by get_data, set to 1 or 2, default=2.. num_sources		= 3; num_mixtures	= num_sources; num_samples 	= 5000;  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % GET DATA. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % sources = (num_samples x num_sources),   % mixtures = (num_samples x num_mixtures),  % A = (num_sources x num_sources) mixing matrix, mixtures=sources*A;  % Get zero mean, unit variance sources.  [sources mixtures A] = get_data(num_samples);  % one mixture per column.  %******************************************** figure(3); for i=1:num_sources     subplot(num_sources,2,2*i-1);     plot(sources(:,i));     axis('tight');     subplot(num_sources,2,2*i);     plot(mixtures(:,i));     axis('tight'); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % COMPUTE V AND U. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Set short and long half-lives. shf 		= 1;  lhf 		= 900000; 	 max_mask_len= 500; n			= 8; % n = num half lives to be used to make mask.  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Get masks to be used to find (x_tilde-x) and (x_bar-x) % Set mask to have -1 as first element, % and remaining elements sum to unity.  % Short-term mask. h=shf; t = n*h; lambda = 2^(-1/h); temp = [0:t-1]';  lambdas = ones(t,1)*lambda; mask = lambda.^temp; mask(1) = 0; mask = mask/sum(abs(mask));  mask(1) = -1; s_mask=mask; s_mask_len = length(s_mask);  % Long-term mask. h=lhf;t = n*h; t = min(t,max_mask_len); t=max(t,1); lambda = 2^(-1/h); temp = [0:t-1]';  lambdas = ones(t,1)*lambda; mask = lambda.^temp; mask(1) = 0; mask = mask/sum(abs(mask));  mask(1) = -1; l_mask=mask; l_mask_len = length(l_mask); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%  % Filter each column of mixtures array. S=filter(s_mask,1,mixtures); 	L=filter(l_mask,1,mixtures);  % Can also use this for L: L = zero_mean_cols(mixtures);  % Find short-term and long-term covariance matrices. U=cov(S,1);		V=cov(L,1);  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Find eigenvectors W and eigenvalues d. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% [W d]=eig(V,U); W=real(W);  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Recover source signals. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% ys = mixtures*W;  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % PLOT RESULTS %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Rescale ys to zero-mean and unit variance, for display purposes. temp=repmat(mean(ys),[num_samples 1]'); 	ys=ys-temp; temp=repmat(std(ys,1),[num_samples 1]');	ys=ys./temp; % Plot original sources ...  figure(1);clf;  nn=1000; if demo_id==1 imin=100; else imin=1; end; imin=1;imax=imin+nn-1; figure(1); for i=1:num_sources	 	subplot(num_sources,1,i);	 	plot(mixtures(imin:imax,i)/std(mixtures(imin:imax,i),1),'k.');  	axis off;  end; subplot(num_sources,1,1); title('Signal Mixtures');  % Plot corresponding recovered sources. a=[sources ys]; c=corrcoef(a);  rs=c(1:num_sources,num_sources+1:num_sources*2); fprintf('Correlations between sources and recovered signals ...\n');  abs(rs)  rs_bin=abs(rs)>0.9; figure(2); for i=1:num_sources 	subplot(num_sources,1,i); hold off; 	plot(-1+sources(imin:imax,i),'k'); hold on; 	% Find corresponding sources/ys via correlation matrix rs_bin. 	j=find(rs_bin(i,:)); s=sign(rs(i,j)); 	plot(2+s*ys(imin:imax,j),'r--'); hold off; axis off; end;  subplot(num_sources,1,1); title('Source signal (bottom trace) and recovered signal (top trace)'); figure(4); for i=1:3     subplot(3,2,2*i-1);     plot(ys(:,i));     axis('tight'); end xlabel('recover signal'); % Play sounds if demo_id==1 	fprintf('Playing mixtures ...\n');pause(2); 	for i=1:num_sources 		soundsc(mixtures(:,i));pause(1); 	end; 	 	fprintf('Playing recovered signals ...\n');pause(2); 	for i=1:num_sources 		soundsc(ys(:,i));pause(1); 	end; end; 	 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%