Regularization of the noise covariance matrix with tSSSed data

Hi MNE experts,

I am wondering whether it is common to regularize the noise covariance
matrix for tSSSed data. Since tSSS may reduce the rank of the sensor data,
(in my case to 64), the covariance matrix is singular too. As a result,
the whitened forward gain matrix had a lot of zeros.
In this case, is it better to regularize the noise covariance matrix, or
to leave it intact such that the rank of noise matches the rank of
"signal"?
Thanks in advance!

Best,
Ying
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140717/0e207354/attachment.html

hi Ying,

try at first without regularizing. MNE should detect the rank
automatically when whitening. It's however possible that tSSS makes
rank estimation difficult due to slight head positions changes.

maybe Sheraz or Rey can share their insights too.

Alex

Hi Ying,
   Also note that if you use the command line tools, you can explicitly set the rank to a reasonable number manually by using the --noiserank input of mne_inverse_operator
or mne_do_inverse_operator. Of course, explicitly setting it can be dangerous if it does not correspond to the actual rank of the noise you have...

HTH,
Hari

Hari Bharadwaj
PhD Candidate, Biomedical Engineering,
Auditory Neuroscience Laboratory
Boston University, Boston, MA 02215

Martinos Center for Biomedical Imaging,
Massachusetts General Hospital
Charlestown, MA 02129

hari at nmr.mgh.harvard.edu
Ph: 734-883-5954

hi Ying,

Rey might pinch in more, my few cents, tsss projection are not orthgonal in
space, so they do not reduce the rank as SSS or PCA, tsss projections
reduces the variance of the first few components, since most rank
determining algoritham uses some sort of ratio of highest to lowest, after
tsss if you try to determine rank in Matlab or python they might give you a
value higher than 64.

Best rank guest should be 64 - all the PCA projection you are using.

sheraz

Thanks a lot for all the answers!
Ying