Dear all,

I am conducting the source estimation based on the resting state EEG and have some issues I am not very sure about:

( Assuming that we only have 5-min resting state EEG data and may want to use the 'Minimum norm imaging' method to reconstruct the source)

- Just as mentioned in the tutorial, it's troublesome for calculating the Noise Covariance Matrix for resting state EEG. In this case, which strategy I should choose? Using the identity matrix or other better choices. I am not very confident about that noise is homoskedastic, and equivalent on all sensors.
- When computing sources, we have two strategies <1> right-click on the data recordings and select 'Compute sources' <2> right-click on the head model and select 'Compute sources' (something like 'shared inverse model'). I wonder these two are equivalent or any differences between them? If any difference, which one is more suitable?
- From the tutorial, I observed that we always compute the sources based on the averaged recording (average across all epochs) and want to know whether it is ok to compute the sources based on the whole recordings? I guess this may be related to the linearity of MN.

I may have some technical mistakes when understanding source estimation and please point out.

Looking forward to any help and thanks in advance.

Just as mentioned in the tutorial, it's troublesome for calculating the Noise Covariance Matrix for resting state EEG. In this case, which strategy I should choose? Using the identity matrix or other better choices. I am not very confident about that noise is homoskedastic, and equivalent on all sensors.

Unfortunately we don't have more recommendations to give than the ones that are in the tutorials:

https://neuroimage.usc.edu/brainstorm/Tutorials/NoiseCovariance#Variations_on_how_to_estimate_sample_noise_covariance

You could try with both approaches and see if it leads to significant differences.

When computing sources, we have two strategies <1> right-click on the data recordings and select 'Compute sources' <2> right-click on the head model and select 'Compute sources' (something like 'shared inverse model'). I wonder these two are equivalent or any differences between them? If any difference, which one is more suitable?

There are no computational differences between these two options. The only thing it changes is how the inverse model is saved. If you have multiple data segments in the same folder always prefer the 2nd option, as it will save space on the hard drive.

From the tutorial, I observed that we always compute the sources based on the averaged recording (average across all epochs) and want to know whether it is ok to compute the sources based on the whole recordings? I guess this may be related to the linearity of MN.

It doesn't change anything. The inverse model is computed independently from the data, based only on the forward model and the noise covariance matrix.