Some issue for noise covariance for resting-state EEG data

Dear Brainstorm contributors,

Let me ask a question about the noise covariance when performing source estimates for resting state EEG data.

Your tutorial on noise covariance is written as "Calculate noise covariance over a long segment of resting recordings".
EEG data are usually epoched (e.g. 4 seconds per epoch) during preprocessing to remove artifacts.
However, I think the epoched data are too short of creating noise covariance.

Therefore, if we consider creating a noise covariance, should we use non-epochizing methods in the preprocessing of the resting EEG? or do you have any other recommended way to create noise covariance for resting-state EEG?

Best regards,

Hi @MasatakaWada,

You can use epoched data to compute the noise covariance matrix. Take into account that the required number of samples is N*(N+1)/2 samples, N is the number of sensors.

See the Using multiple continuous blocks point in:

Select the advanced option Diagonal noise covariance when computing sources to use only the diagonal elements of the noise covariance matrix:


Hi @Raymundo.Cassani ,

Thanks for the great advice!
I am very glad to hear that.

May I just confirm one point?
Does "number of samples" refer to the number of time points per epoch?

Best regards,

Nop, it refers to the total number of time points across all epochs: nEpochs * nSamplesInEpoch

Hi @Raymundo.Cassani,

Thank you for your clear and helpful advice.
I would be happy to do so!

Best regards,