[FieldTrip] LCMV time series reconstruction under multiple conditions (cov matrix problem)
fab.rotondi at gmail.com
Fri Oct 24 16:58:26 CEST 2014
Dear Fieldtrip community,
my name's Fabio and I'm currently investigating on MEG/EEG connectivity in
I'm dealing with a theoretical concern, related to the estimation of
virtual electrodes' time series for resting state EEG data (128 channels,
512 Hz sampling frequency) under 3 conditions.
For each subject, the recording protocol consists of three consecutive
steps: 1 minute of resting with eyes closed (EC), 1 minute resting EC with
auditory stimulation n.1, 1 minute resting EC with auditory stimulation n.2.
The final aim is to estimate functional connectivity changes in source
space under these conditions.
My question regards the computation of the covariance matrix: in order to
do this I identified 2 possible ways to proceed:
1. compute covariance matrix giving the whole 3-minute signal to
ft_timelockanalysis, and then use this matrix to invert separately each of
the three conditions;
2. compute a covariance matrix separately for each of the three conditions
(thus on 60 seconds) and then invert each condition by using its
respective covariance matrix.
As far as I know, solution n.1 is in literature the most used. I think it
is the solution named as "common filters", which assumes that there are the
same signal generators differently active in the three different
This assumption could not be valid in principle for my experiment, but I
was wondering if computing three different covariance matrices could
introduce any other bias.
Thank you so much for your help,
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the fieldtrip