Noise covariance for EEG data: steady-state stimulation and no rest segments

Discussions
, ,
1

Dear BST experts,

I have an infant EEG dataset with continuous steady-state visual stimulation (therefore, no baseline periods) and for most subjects no resting state segments.

What is the best way to compute noise covariance for source reconstruction in this case? I see here that for resting state data you advice to either compute noise covariance on all the data and take the diagonal or use an identity matrix. On one way it looks cleaner to use the identity matrix, but I would lose the information about the inevitable variability between different channels.

Any advice is welcome.

Thank you,

Best

Marco

2

The diagonal of the noise covariance computed using the entire recording seems a sensible option as it will account for the difference in variance for the sensors, rather than assuming it the same with the identity matrix.

1 Like
3

Thank you Raymundo, I'll proceed with that option then.

Best,

Marco