Multitaper and Welch PSD Amplitudes differ by constant of ~40

💬 Support & Discussions
1
  • MNE version:1.5.1
  • operating system: RedHat Enterprise Linux 8.8 (Ootpa)

The Welch and Multitaper estimates yield very different amplitudes for the PSD estimate of the same EEG signal.

I have an EEG Raw object with 29001 samples at 500 Hz sampling: sample_eeg

sample_eeg.copy().compute_psd(method='welch').pick(sample_eeg.ch_names[0]).plot()

multitaper_psd
multitaper_psd982×333 23.1 KB

sample_eeg.copy().compute_psd(method='multitaper').pick(sample_eeg.ch_names[0]).plot()
welch_psd
welch_psd995×333 17.8 KB

Overlaying the figures shows that the shape is the same, by the y-axis amplitude is very different: the Welch method estimate ranages between -30 and 20 while Multitaper estimate ranges between -10 and 50. My hunch is that this difference is due to the window size of the Welch estimate being ~ 1/3 that of Multitaper estimate. This would imply that the absolute amplitudes of PSD estimates from two signals with different lengths are not comparable using the Multitaper estimate. I believe that since the Welch method uses fixed window sizes, the signal length should not affect its amplitude.

Is that correct?

EDIT: I figured out that adjusting the normalization parameter from “length” to “full” solves the amplitude issue. Would still appreciate any thoughts on why you would want “length” instead of full, and insight on setting that bandwidth parameter for multitaper.