PSD: large, negative power values in rodent REM sleep

Processing: FS_REM_pwelch.fig…
Hello all,

I was hoping maybe someone could offer some insight on some data I have collected. I have previously worked on human EEG projects and used EEGLAB spectopo/ERSP to look at working memory performance. Presently I am transitioning into sleep research using rats and am a little confused by the PSD plots generated for my data. I wanted to compare the pwelch function with the parameters from the spectral source separation lecture with the EEGLAB spectopo function. My data contains epochs of REM sleep over the course of a 12 hour sleep period for 1 rat. Both plots generated have a db/Hz range of -90 to about -110, which is not like anything I have seen before (this range is pretty consistent across all subjects). As most axis labels for spectral analysis in the rodent literature that I have come across will say just “power” or they use relative power, I am rather confused by what I am getting when I plot the data (the range for human participants I have used looked nothing like this). Is this range to be expected with rodent EEG data, or is there potentially an issue with the data? I have attempted this with two different datasets from different projects and am still seeing this range. Any suggestions would be greatly appreciated! Thank you!


Hi Marissa. I guess this is raw power converted to dB without any baseline. The amplitude values can be quite small, then squaring them makes them really small, and taking the log makes them large-negative.

For example, if the amplitude value is 10^-5 (.00001), then power is 10^-10, and 10*log10(10^-10) = -100.

You can check by computing and plotting the “raw” power spectrum without taking the log (you should still divide by the number of points).

Thank you for your insight! Yes when I plot the raw spectra for the REM epochs (first plot; frequency range 0.5-15), and over the whole 12 hours of un-epoched data (second plot; frequency range 0-60) it seems to indicate very small amplitude. I suppose right now I’m struggling to find out how common this is in the literature. Thank you for your reply, I really appreciate itSpectra_REM

The absolute power values depend on several factors, including the electrode impedance and – importantly – the reference electrode. That’s why relative power is usually more insightful :wink:

I had been working through your neuroscience source separation lectures when I started playing with the pwelch function. Do you address calculating relative power in matlab in one of your other resources?

Not in that course, but you can check out my YT channel, in particular the “NEW ANTS” playlists. And my ANTS textbook :wink:

Rad! Thanks so much for your help :slightly_smiling_face: