[FieldTrip] Estimating Single-trial Granger Causality

[Lisandro Kaunitz]

Hi,

I am trying to compute grangerspectra for simulated single trials. Whenver I run granger causality on a single trial with 2 or more channels I got the following warning:

Warning: Matrix is singular, close to singular or badly scaled. Results may be inaccurate. RCOND = NaN. 
> In connectivity/private/sfactorization_wilson at 81
  In ft_connectivity_csd2transfer at 180
  In ft_connectivityanalysis at 400
  In example_maillingList at 51 


This warning disappears if I use the same number of trials and channels / signals in my dataset (i.e, 2 trials with 2 channels, 3 trials for  3 channels, etc). 


Is it possible to obtain granger spectra for single trials with Fieldtrip? I would like to average the spectra across single trials to run statistics and compare across conditions later on.

This is the code I am using in my simulation, if the value of cfg.ntrials is changed to 2 (both times is is declared), then the warning disappears:

nTrials = 5;

for tr = 1 : nTrials
    
    
    %% Simulate data
    
    cfg             = [];
    cfg.ntrials     = 1;
    cfg.triallength = 2; % in secs
    cfg.nsignal     = 2;
    cfg.fsample     = 200;
    cfg.method      = 'ar';
    cfg.bpfilter    = 'no'; % (or 'no')
    cfg.bpfreq      = [0 300]; % (default: [15 25])
    
    cfg.params(:,:,1) = [ 0.8  0 ; 
                          0.9  0.9];
    
    cfg.params(:,:,2) = [-0.5    0; 
                          0.9   -0.8];
    
    cfg.noisecov      = [ 0.3  0; 0 0.2];
    
    data            = ft_connectivitysimulation(cfg);    
    
    %% Non-parametric computation of the cross-spectral density matrix
    
    cfg                 = [];
    cfg.ntrials         = 1;
    cfg.triallength     = 2; % in secs
    cfg.nsignal         = 2;
    cfg.method          = 'mtmconvol';
    cfg.output          = 'fourier';
    cfg.taper           = 'hanning';
    cfg.foi             = 2:2:100;
    cfg.t_ftimwin       = 0.05 .* ones(size(4./cfg.foi', 1), 1);
    cfg.toi             = 0:0.05:2;
    
    freq                = ft_freqanalysis(cfg, data);
    
    
    %% GRANGER CAUSALITY
    
    cfg           = [];
    cfg.method    = 'granger';

    granger       = ft_connectivityanalysis(cfg, freq);
    
    sp_chan1_chan2(:,:,tr) = squeeze(granger.grangerspctrm(1,2, :, :)); %#ok
    sp_chan2_chan1(:,:,tr) = squeeze(granger.grangerspctrm(2,1, :, :)); %#ok

    
end


Thanks a lot


Cheers


 
Lisandro Kaunitz
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20131215/1ef5cd6a/attachment.html>