ultra-low-frequency noise

edited January 9 in Cyton

I'm having difficulties with ultra-low-frequency noise, and with knowing how much I should be removing.

Here's a 30 second epoch of sleep EEG data (EEG data on the left, power spectral density on the right), with a bandpass filtering from 0Hz up to 1Hz. (The high filter is fixed at 30 Hz).

My questions are:

  • Clearly there's a large amount of noise somewhere between 0 and 1Hz. I've searched and can't find common causes of this (though I've noticed that many people bandpass filter somewhere from the 0.2 to 0.5 range, presumably to remove it). Does anyone know what causes it?
  • Where should I filter from? The data changes quite dramatically if I go from a 0.1 to a 0.5 Hz filter. Is that valid signal I'm filtering out, or noise?
  • Even when filtering out anything below 1Hz, the power spectral density graph doesn't seem right. It's showing most of the power at frequencies <= 1 Hz, while I'd eyeball this EEG data at being more in the theta, alpha and beta ranges (I've clipped the PSD graphs at 10 Hz for readability - there's almost zero power above there in any of the full graphs). This is my main issue, as I'm not sure I can trust the bandpowers currently.

In case it's useful, the code used for most of this is in this small snippet.

Comments

Sign In or Register to comment.