[FieldTrip] lambda regularization using Neuromag data

[Johanna Zumer]

Dear Hyojin,

If you look inside beamformer_lcmv.m, you will see that the warning
you mention gets generated before the lambda regularization is applied
(i.e. the warning applies to your input rank 57 covariance matrix, not
the matrix that actually is inverted to generate InvCy which takes the
lambda into account).

Maybe someone else with experience with Maxfilter Neuromag data can
comment as to what level of lambda is appropriate.   But I would
suggest looking at your results from lambda at around 10% and see if
they are reasonable.   At the extreme, the 100% lambda case should
look similar to a min-norm result.

Cheers,
Johanna


2012/10/23 Hyojin Park <hjpark05 at snu.ac.kr>:
> Dear all,
>
>
>
> I’m working on LCMV beamforming using Neuromag data applied with Maxfilter.
>
> I am using 204 gradiometer sensors.
>
>
>
> I have rank deficiency problem, maybe because of Maxfilter since it reduces
> the rank.
>
> When I checked the data, the rank is 60.
>
> I removed EOG and ECG components using ICA.
>
> 2 or 3 components in sum in most subjects.
>
> So for some, the rank is 58, for some, it’s 57…
>
>
>
> When I tested with different lambda from 0% to 10%, (or above, even 100%), I
> got the same warning ('covariance matrix is rank deficient').
>
>
>
> Does anyone have suggestions for which lambda is most appropriate in this
> case? Or any other useful advice/experience for how to apply beamforming in
> combination with Maxfilter?
>
>
>
> Thank you in advance.
>
> Hyojin Park
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip