search

fooof-tools / fooof

Parameterizing neural power spectra into periodic & aperiodic components.
https://fooof-tools.github.io/
Apache License 2.0
363 stars 98 forks source link

Parts of the original spectrum are below the aperiodic fit #320

Open shnovik opened 8 months ago

shnovik commented 8 months ago

Hello, we're running into problems where parts of the original power spectra dip below the aperiodic fit -- I've pasted two examples below. Is there any way to prevent this from happening? Thanks!

image image

TomDonoghue commented 7 months ago

Dear @shnovik - perhaps you can elaborate on why this is an issue / undesired behavior? The aperiodic fitting is designed to find the best fitting fit to the isolated aperiodic component, and is not set to prioritize any direction of error. The PSD, as an estimate of power, can vary above or below the best fit line, but we don't think there's anything meaningful about small deviations below the fit, as you see in the examples.