LCMV questions

✉️ Mailing List Archive (read-only)
1

External Email - Use Caution

I just started looking at the LCMV beamformer implementation in MNE and I'm
a bit confused by the use of a 2 covariance matrices.

The implementations I'm familiar with (fieldtrip and ctf tools) use a
single covariance. So to commute a differential beamformer you need to
compute an average covariance (between the active and baseline time
windows) to make the filter weights etc.

This is the approach taken in this tutorial.
http://www.fieldtriptoolbox.org/workshop/aarhus/beamformingerf/

Does the MNE implementation do this already? Are the filter weights the
same if I switch what I pass in for the data and noise covariances?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20191126/b1523406/attachment.html

2

External Email - Use Caution

Hi Luke,

the data_cov in the LCMV implementation serves the same purpose as
Fieldtrip's data covariance matrix (i.e., it is used to compute the spatial
filter). The noise covariance matrix is used to whiten the data, lead
field, and covariance matrix. This is only mandatory if you supply data
with several channel types (e.g., MEG and EEG or gradiometers and
magnetometers) and is used to deal with scaling issues that arise with
those different channel types.

Hope this helps,
Britta

3

External Email - Use Caution

Thanks Britta, that is what I was looking for.