Quote
You mention having "5xx sampling points"---does that mean you upsampled by a factor of 5?

I did not remember the exact number of TRs, therefore I simply wrote "5xx", because I have 500 something TRs in one run. I did not upsample the time-series for different reasons, one of them because the upsampling would make them more linear. Sorry for the confusion.

Quote
Note that the resting state signal in general will not be a \"flat\" spectrum---the named noise distributions in signal processing are, like \"white noise\" (which has slope=0), purple noise, pink noise, brown noise,

Correct and thanks. Yes, I know, this is exactly part of what I am investigating. This is one of the reasons why I am so much after "correct" power spectra and their log-log plots.

Quote
As to the differences between the two---well, I am not sure. Up above, we discussed that for no tapering, linear detrending for each, etc., they 3dPeriodogram and scipy.signal.periodogram() yielded the exact same results.

Everything is kept identical between 3dPeriodogram (and all AFNI steps) and sps.periodogram() as close as possible. However, they still yield this difference. I don't know why. The only explanation I have is that for 3dPeriodogram we averaged thoursands of power spectra (one power spectrum per voxel in a huge ROI), whereas for sps.periodogram() I started with the extracted (average) time-series of the same ROI, and then computed the power spectrum in Python.
However, the dfference in the results (including the power-law slope) is so huge that it seems odd to me.

If you are also at the end of your ideas here, then I will stick with 3dPeriodogram for two reasons: first,I prefer the averaging of many power spectra across the voxels, and second, the resulting power-law slopes are “realistic”, while the ones in sps.periodogram() are not.



Edited 2 time(s). Last edit at 06/01/2022 12:27AM by Philipp.