Dear all

I am writing to ask a question on baseline normalization. I am performing TF analysisi via walvet algorithm on LFPs recorded in premotor cortex. First I compute the TFR spectrum of the signal (let’s say 1 sec trials), to get the power(averaged across trial) for every frequency of interest(from 8 to 180 Hz). I apply a baseline normalization so that, for each freqeuncy, separately( as I know from Mike’s book, lectures and many other references), the power in each time bin is 10*log10(frequency-specific-activity/ frequency-specific-baseline). Secondly I separe different freqeuncy bands( Alpha, Beta, Gamma, HighGamma), by average across frequencies belonging to that band. I wonder if ,to this purpose, normalizing for each band separately (instead of each frequency) could be more correct. The formula should be in this case 10*log10(Alpha-specific-activity / frequency-specific-baseline). I tried both solution

( normalizing each freq separately and then average; averaging across frequencies specific to different bands and then normalize), and despite the trend is similar, the effect size(e.g. the difference between Alpha’s increase and Gamma’s decrease) is not the same. The general goal is comparing different Frequency-bands in relation to different task parameters.Another possibility I thought about is to compute the power spectral density for all the frequencies of interest in the baseline period, average it, and then normalize the power in each time bin after the baseline(for every single frequency band) using this value. To me the second option, seems better(at least conceptually) but I would really appreciate an expert opinion.

Thank you in advance

Marta