This is FOOOF-adjacent, rather than FOOOF-internal, but related enough to put here:
FOOOF has no opinion on how you make power spectra. You make them, feed them in, and the the model does it's thing. But a FOOOF-related project would be to investigate properties of creating power spectra, including:
Are there pros & cons to different approaches for creating power spectra
Within an estimation approach, are there better or worse settings to use, given the context of submitting spectra for parameterization, for example, windowing.
A note on windowing:
Right now, we typically default to hann or hamming window. This reduces most of the edges of the time window, and emphasizes data in the middle. We should systematically explore using different windows (Tukey, in particular) that keep more of the time window (at the cost of frequency bleeding, of course). For the use cases of fooof, some frequency bleeding that helps keep more information (especially for use cases with data from short time windows) may turn out to be a reasonable trade off.
This is FOOOF-adjacent, rather than FOOOF-internal, but related enough to put here:
FOOOF has no opinion on how you make power spectra. You make them, feed them in, and the the model does it's thing. But a FOOOF-related project would be to investigate properties of creating power spectra, including:
A note on windowing: Right now, we typically default to hann or hamming window. This reduces most of the edges of the time window, and emphasizes data in the middle. We should systematically explore using different windows (Tukey, in particular) that keep more of the time window (at the cost of frequency bleeding, of course). For the use cases of fooof, some frequency bleeding that helps keep more information (especially for use cases with data from short time windows) may turn out to be a reasonable trade off.
Note: windowing notes originally brought up by @voytek, which was described here: https://github.com/voytekresearch/VoytekLabPrivate/issues/4