Proper SNR for source estimation with a mixed model

Dear Brainstorm Experts,
The default value for 'SNR' parameter in the "linear inverse 2018" is : snr=3.
On the other hand, we know that the SNR of deep sources are much less than the cortical ones. Therefore, shall we change (lower) this parameter when using a "mixed" model?

@Sylvain @John_Mosher?

Excellent question, one that has arisen in other conversations. For example, we hold the SNR parameter to be "3" whether we use raw data or data that we have averaged 100 times, which in theory should have a noise variance 100 times smaller.

The suggested parameter "3" comes from Matti Hamalainen's long experience with the min norm estimator, and this same parameter is used in MNE; hence, we chose to maintain the same parameter for Brainstorm, and indeed MNE and Brainstorm share nearly the same algorithm, thanks to our long collaboration between the two projects.

The data covariance used in the min norm is a theoretical data covariance that balances a theoretical signal covariance with an estimated (from baseline) or fixed noise covariance. The average eigenvalue of the whitened signal covariance is set to be nine times (3^2) of that of the whitened noise. Hence this theoretical data covariance acts more like a resolution filter than an actual data covariance matrix.

So, no, we don't usually adjust the MNE SNR value under varying levels of noise.

That said, deeper sources are harder to "resolve," and therefore some exploration of the SNR parameter in MNE may be warranted. Larger values tend to "sharpen" the resolution, but at the expense of more apparent noise in the estimate, while lower values tend to "smear" the signals.

My comments are definitely arguable, so others may offer a different opinion.

I am really thankful of your clear and complete description