I see, thank you very much! So, do you think that segmenting the data into 10 second epochs and simulating the sources for each one would be an appropriate approach? We intend to analyse this data with mostly window-based measures, but at some point we would like to do dynamic functional connectivity analyses, so that's why we tried to do this on the full recordings, in order to prevent discontinuities between segments.
I would also like to ask, when we did this process in the non-recommended way ("Downsample to atlas") we didn't run into this problem (or at least I don't recall this). Are these processess fundamentally different?