[FieldTrip] Granger causality with ft_connectivityanalysis

[Maxime Fauvet]

Hello everyone, 

I am Maxime Fauvet from ToNIC lab in Toulouse,
France, currently working on functional connectivity between brain and
muscles during movement through corticomuscular coherence analyses of
EEG and EMG signals. I am trying to perform a granger causality (GC)
analysis from those signals but somehow I cannot achieve to get a
complete time-frequency map of the GC. 

I used ft_freqanalysis with the
following raw data and cfg: 

cfg.method: 'wavelet'
cfg.output:
'powandcsq'
cfg.foi: 1:1:50
cfg.pad: 1000 

raw data: 2 matrices of 17
trials and 3 seconds sampled at 1KHz, one for EEG and the other for EMG
signals, structured as required in ft_freqanalysis 

After removing NaNs
from powerspectrum and crosspowerspectrum, and setting cfg.method:
'granger', I used ft_connectivityanalysis with the cfg and freq (issued
from ft_freqanalysis) arguments. 

However, I receive the following
warning message many times: 

Warning: Matrix is singular, close to
singular or badly scaled. Results may be inaccurate. RCOND = NaN. 
>In
sfactorization_wilson (line 187)
  In ft_connectivity_csd2tranfer (line
267)
  In ft_connectivityanalyis (line 464) 

The issue appears to come
from the matrix factorization function and more precisely from the chol
function, a Matlab core function. Interestingly, some points of the time
series do not trigger the warning message. 

Can someone tell me if
anything is wrong with the use of ft_connectivityanalysis (either in cfg
or data)? Any help would be appreciated. 

Best wishes, 

Maxime 

  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20200901/27ee3b78/attachment.htm>