Covariance matrices for LCMV: normalization, regularization etc

Support & Discussions
, , , ,
1

Hello everyone,
I’m getting used to the LCMV beamformer source-localisation method and I’m a bit confused about operations involving covariance matrices. I’m using this tutorial as a reference, but some points are still not clear.

I have MaxFiltered data, so the data rank is about 68–72. For reasons unknown, the empty-room recording has rank = 68, while the epoch rank varies. For example, I have an epochs object with rank = 71. Moreover, the noise covariance has been regularized:

cov = mne.cov.regularize(cov, raw_er.info, mag=0.01, grad=0.01, rank=rank)

My questions are:

(1) Is it okay to set rank={'meg': 68} for the data covariance, since the noise covariance has that rank?

active_cov = mne.compute_covariance(
    epochs,
    tmin=active_win[0],
    tmax=active_win[1],
    method="auto",
    rank={'meg': 68},
    verbose=False,
)

(2) Should I perform any kind of normalization on the data covariance, given that the noise covariance has been regularized?

P.S. I would be grateful for any information related to LCMV or hippocampal localisation with MEG.