% % This is a calling routine to test & check out the power spectrum & % spectrogram routines for unequal segment lengths. In addition, use it % to compare with Chronux routines when segments are of equal length. % clear all; if 0 dir = 'G:\ravi\Chrowser\Pass~ Tioga_0e927741-9673-46e5-9050-ca1d7541bf22\'; xfile = 'Pass~ Tioga_0e927741-9673-46e5-9050-ca1d7541bf22' %dir = 'G:\ravi\Chrowser\sample~ data_8ef647e3-e5ea-43a6-8c69-fb848b8db7c2\'; %xfile = 'sample~ data_8ef647e3-e5ea-43a6-8c69-fb848b8db7c2' else dir = 'Z:\xltekRawData\Wallis~ Terry_c3f44891-afa7-4fa7-a643-55c772a05241\' xfile = 'Wallis~ Terry_c3f44891-afa7-4fa7-a643-55c772a05241' end % Get header info % Channels are labelled from C1 through C127 and '' % total of 128 channels hdr = eegMex( dir, xfile); gram = 1 ; % 0=spectra, 1=coherence chronux = 0 ; % 0=no comparison with Chronux; 1=compare with chronux %nSamples = 4210; nChannels = 2; nSegments = 1; movingwin = [25, 25]; % % Spectral Parameters % params.fpass = [ 0 0.5 ]; params.pad = 2; params.err = [2 0.05]; % err(1)=0 is no err estimates; err(1)=1 is asymptotic estimates; err(1)=2 is Jacknife params.trialave = 1; params.Fs = 1; % % Tapers issues % halfBandWidth = 2.5; kCap = 2*halfBandWidth - 1; %params.tapers = [ halfBandWidth, kCap ]; params.tapers = [ halfBandWidth, 2 ]; % % Basic checks on inputs % if (gram==1) && (nChannels < 2), error( 'Coherence requires at least 2 channels...' ); end %if (nSegments==1) && (params.err(1)==2), error( 'Jacknife requires more than 1 segment'); end % % Generate segments endpoints randomly % myrandint is a 3rd party routine (from matlab site) % % Randomly generated segment end points sMarkers = reshape( sort( myrandint( 2*nSegments, 1, [ ceil(hdr.nSamples/500) : ceil(hdr.nSamples/50) ], 'noreplace' ) )', 2, nSegments )'; %sMarkers = [ ceil(hdr.nSamples/80), ceil(hdr.nSamples/65) ]; % % Randomly select a few channels % if ~chronux %chIndices = sort( myrandint( nChannels, 1, [ round(hdr.nChans/4) : round(3*hdr.nChans/4) ], 'noreplace' ) ); chIndices = [ 3 : 3+nChannels-1 ]; else %chIndices = [ 10 : 10+nChannels-1 ]; chIndices = [ 3, 7 ]; end % % Randomly generate the time series % fulldata = eegMex( dir, xfile, chIndices, [ 1 hdr.nSamples/50 1 ] ); mDiscardBits = 0; conversionFactor = ( 8711 / (2^21 - 0.5) ) * 2^mDiscardBits; fulldata{:} = fulldata{:} * conversionFactor; % % Create a data matrix with all the segments aligned one after another % totalSegmentLength = sum( sMarkers(:,2) - sMarkers(:,1) + 1 ); data = zeros( totalSegmentLength, length(chIndices) ); % preallocate to ensure contiguous memory newMarkers(1,1) = 1; newMarkers(1,2) = sMarkers(1,2) - sMarkers(1,1) + 1; data( newMarkers(1,1):newMarkers(1,2), : ) = detrend( fulldata{1}( sMarkers(1,1):sMarkers(1,2), :) ); for sg = 2:size( sMarkers, 1 ) newMarkers(sg,1) = newMarkers(sg-1,2) + 1; newMarkers(sg,2) = newMarkers(sg,1) + sMarkers(sg,2) - sMarkers(sg,1); data( newMarkers(sg,1):newMarkers(sg,2), : ) = detrend( fulldata{1}( sMarkers(sg,1):sMarkers(sg,2), :) ); end % To ensure that we check results from array indices beyond 1 if nChannels > 1 ix = sort( myrandint( 1, 2, [1:length(chIndices)], 'noreplace' ) ); % Arbitrarily pick two indices from selected channels for testing results i1=ix(1); i2=ix(2); % iC = m + (n-1)*(n-2)/2, for elements of the the coherence matrix, Cmn iC = ix(1) + (ix(2)-1)*(ix(2)-2)/2; else ix = sort( myrandint( 1, 1, [1:length(chIndices)], 'noreplace' ) ); % Arbitrarily pick 1 indices from selected channels for testing results i1=ix(1); end % % Power spectrum/spectrogram/coherence/coherogram % if gram==0 [ S, f, Serr ] = avgSpectrum( data, movingwin, params, newMarkers ); figure; plot( f, 10*log10( S(:,i1) ), 'k', f, 10*log10( Serr(2,:,i1) ), 'g--', f, 10*log10( Serr(1,:,i1)), 'g--' ); title('Avg. Routine:: Spectrum'); %figure; plot( f, 10*log10( S(:,i1) )); title('Avg. Routine:: Spectrum'); elseif gram==1 [Cmn,Phimn,Smn,Smm,f,ConfC,PhiStd,Cerr] = avgCoherence( data, movingwin, params, newMarkers ); % C(i,j) = Cmn(:,k) where k = j + (1/2)*(i-1)*(i-2) figure; plot( f, Cmn(:,iC), 'k', f, Cerr(2,:,iC), 'g--', f, Cerr(1,:,iC), 'g--' ); title('Avg. Routine:: Coherence'); ylim([0 1]) %figure; plot( f, 10*log10( Cmn(:,iC) ) ); title('Avg. Routine:: Coherence-Magnitude'); %figure; plot( f, phimn(:,iC) ); title('Avg. Routine:: Coherence-Phase'); disp( ['Confidence level for C (confC) at (1-p) level: ' num2str( ConfC(iC)) ] ); end % % Use to check against Chronux: only for equal length segments % if chronux win = floor( newMarkers(1,2) / movingwin(1) ); newMarkers(1,2) = newMarkers(1,2) - mod( newMarkers(1,2), win ); cdata = data( [1:newMarkers(1,2)], i1 ); cdata = detrend( reshape( cdata, [ newMarkers(1,2)/win, win ] ) ); cdata2 = data( [1:newMarkers(1,2)], i2 ); cdata2 = detrend( reshape( cdata2, [ newMarkers(1,2)/win, win ] ) ); params.trialave = 1; if gram==0 [ cS, cf, cSerr ] = mtspectrumc( cdata, params ); figure; plotvector( cS, cf, 'l', cSerr ); %figure; plot( cf, 10*log10( cS )); title('Chronux:: Spectrum'); figure; plot( cf, 10*log10(cSerr(1,:)), cf, 10*log10(cSerr(2,:)) ); title('Chronux Error-Bar Computations'); figure; plot( cf, 10*log10( cS ) - 10*log10( S(:,i1) )); title('Error in Spectrum = |New Routines - Chronux|'); figure; plot( cf, 10*log10(cSerr(1,:)) - 10*log10(Serr(1,:,i1)), cf, 10*log10(cSerr(2,:)) - 10*log10(Serr(2,:,i1)) );title('Error in Error-Bar Computations = |New Routines - Chronux| '); elseif gram==1 [cC,cphi,cS12,cS1,cS2,cf,cconfC,cphistd,cCerr]=coherencyc( cdata, cdata2, params ); %figure; plotvector( cC(:,1), cf, 'n', cCerr ); figure; plot( cf, cC(:,iC), 'k', cf, cCerr(2,:,iC), 'g--', cf, cCerr(1,:,iC), 'g--' ); title('Chronux:: Coherence'); ylim([0 1]) %figure; plot( cf, 10*log10( cC(:,1) ) ); title('Chronux:: Coherence-Magnitude'); figure; plot( cf, 10*log10( cC(:,1) ) - 10*log10( Cmn(:,iC) ) ); title('Error in Coherence = |New Routines - Chronux|'); % Phase may give a problem of 2pi difference... look into it. figure; plot( cf, cphi(:,1) - Phimn(:,iC) ); title('Error in Phase = |New Routines - Chronux|'); % % Note the remaining quantities do not really need to checked since % coherence = cross-spectrum/power spectra* power spectra, ie C = S12/(S1*S2) % so unlikely that S12, S1, S2 are incorrect if C is correct. if 1 figure; plot( cf, 10*log10( cS1(:,1) ) - 10*log10( Smm(:,ix(1)) ) ); title('Error in Power Spectrogram-1 = |New Routines - Chronux|'); figure; plot( cf, 10*log10( cS2(:,1) ) - 10*log10( Smm(:,ix(2)) ) ); title('Error in Power Spectrogram-2 = |New Routines - Chronux|'); end % % Error-Bars & Confidence Levels disp( ['Confidence levelfor C (confC) at (1-p) level: ' num2str( cconfC) ' (Chronux)' ] ); disp( ['Error in confidence level, confC: ' num2str( ConfC(iC) - cconfC ) ] ); %figure; plot( cf, cphistd(:,1), f, phistd(:,iC) ); title('Phase-Error-Bar Computations'); figure; plot( cf, cphistd(:,1) - PhiStd(:,iC) ); title('Error in PhiStd-1'); figure; plot( cf, cphistd(:,1) - PhiStd(:,iC) ); title('Error in PhiStd-2'); figure; plot( cf, abs(cCerr(1,:,1) - Cerr(1,:,iC)), cf, abs(cCerr(2,:,1) - Cerr(2,:,iC)) ); title('Error in Abs(Coherence)-Error-Bar Computations = |New Routines - Chronux|'); end end