I have very good experiences applying make_lcmv amd make_dics to MEGIN data after using SSS and ICA (resulting in a rank 69). The algorithms provide stable and interpretable results.

Thats for providing those - I do however have a question on what exactly is done in those algorithms when handling the low ranked cov matrices.

I understand that regularization by diagnonal loading is not sufficient given that large rank deficiency - and that shrinkage is applied by the algorithms? (as suggested by Engemann and Gramfort 2015).

My question is: how it the large rank defiency taken into account prior to inversion? If shrinkage is applied; which procedure is then applied?

Any help on this would be most appreciated!

Ole

Example:

filters = make_dics(epochs.info, fwd, csd_common.mean() , noise_csd=csd_noise.mean(),

reg=0.05, pick_ori=â€˜max-powerâ€™, reduce_rank=True, real_filter=True, rank=69, depth = 0)