samer@32: function Z=wignerdata(m,X)
samer@32: % sliding window wigner distribution. feature:
samer@32: % - low pass filter because of undersampling of 
samer@32: %   autocorrelation function
samer@32: % - windowing of autocorrelation to avoid spectral leakage
samer@32: % - temporal smoothing followed by down sampling
samer@32: 
samer@32: 	N = 2*m+1; % frame size
samer@32: 
samer@32: 	[ filtB, filtA ] = butter( 5, 0.5);
samer@32: 
samer@32: 	% make sure frames are N samples long
samer@32: 
samer@32: 	% lowpass signal, get autocorrelation in wigner form,
samer@32: 	% then short term average (exponential window) by 1st
samer@32: 	% order IIR.
samer@32: 	Im=m+1:2*m+1;
samer@32: 	Ip=m+1:-1:1;
samer@32: 	Z=map(@wiglocal,select(filtA,filtB,X));
samer@32: 
samer@32: 	function z=wiglocal(m,Im,Ip,X)
samer@32: 		r=X(Ip,:).*X(Im,:);	
samer@32: 		%smoothing??
samer@32: 		smKernel = gauss_win(128,0.4)';
samer@32: 		wigR = spdiag(hgauss_win(m,0.5))*r;
samer@32: 		% must filter and subsample wigR here
samer@32: 		smRS = filter2(smKernel, wigR, 'valid');
samer@32: 		% subsample in time here?
samer@32: 
samer@32: 		% take real fft using window
samer@32: 		z = abs(fft([smR(1:m,:); smR(m+1:-1:2,:)]));
samer@32: 	end
samer@32: end
samer@32: 
samer@32: 
samer@32: 
samer@32: