Normalizing LFP data, window choice for psd

[ap...@nu.edu.kz]

Dear Mike and group, 

I wanted to ask about the baseline (pre stimulus period) normalization of LFP data. I wanted to plot power spectral density of percentage change from baseline across the frequencies. What is the best way to do it (time/frequency domain)? LFP can have both negative and positive values, so how should I normalize it? Is it ok to compute psd of normalized data?

Also, what is a suggested window (for multitapering) to reduce spectral leakage? I read that dpss, slepian sequences are good, how about the kaiser window with wide width?

I would appreciate any help,

Thanks,

 

[Mike X Cohen]

Hi. By normalization of the data, do you mean a linear baseline subtraction of the time series data as part of preprocessing, or are you referring to normalization of time-frequency power values? Power values can only be positive, so there is no concern about negative values during normalization. In other words, you would first compute the PSD, and then apply the baseline normalization on the time-frequency power values.

As for your second question, multitapering increases spectral leakage relative to short-time FFT or wavelet convolution. If you want to maximize spectral precision, FFT would be the best. For the short-time FFT, the precise choice of window doesn't really matter (Kaiser, Hann, Hamming, Blackmann, etc.). If you want to apply more than one taper (e.g., multitaper), then it's important to have the tapers be mutually orthogonal, which is why the Slepian tapers are used. 

Hope that helps

Mike

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.
To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimeseriesdata+unsub...@googlegroups.com.
Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

[ap...@nu.edu.kz]

I was talking about the linear baseline subtraction from LFP data. Since it can have negative values, I cant subtract mean/median. 

Regarding the 2 question, I wanted to use multitaper, so probably should use slepian.

Thanks for a quick reply,