# Doubt on wavelet decomposition

Hi Mike,

maybe my question is stupid but I am still new in this topic so I have to learn a lot.
I am trying to understand wavelet decomposition using the example data in sampleEEGData.mat that you provided in your course focusing for example on electrode 01. I started looking at the power spectrum using pwelch function as you have done.
As expected, the spectrum has a peak around 10 Hz and a flat pattern at higher frequencies.
I tried to verify if the wavelet decomposition leads to a similar result using as number of cycles 10 or 4.

As you can see, I have this strange high power at high frequencies that it’s quite unexpected. The situation is confirmed if I average over time my matrix to obtain like a “spectrum”.
My question is: why there is strange positive trend at gamma frequencies?

Matteo

ok, sorry for the continuous repost but I have understood something. I have this strange trend if I use a fixed number of cycles (4 o 10), whereas if I use a variable number of cycles ranging from 4 to 15 in my frequency range 1-35 Hz, I have a better pattern.

Why does it happen? So, is wrong using a fixed scale for a span of frequencies? I am a bit confused.
Thanks again,

Matteo

Hi Matteo. I guess this is raw power and not normalized (e.g., to a pre-stim period) power? With a constant number of cycles, the amount of energy that passes through convolution changes with frequency. Normalization is always a tricky issue. When possible, it’s best to normalize according to an appropriate baseline period. That should eliminate these effects.

You could also run convolution to higher frequencies to see if this persists. I think the sampling rate of that dataset is 256 Hz, so you could extract frequencies up 128 Hz. That would give some hints about whether the high-frequency power reflects an issue of differences in total energy vs. gamma-band activity.

Mike

1 Like

Hi Mike,

yes it’s raw. In my dataset I have resting state EEG so I would like to know how to tackle this tricky thing. What can you suggest to me?
Thank you,

Matteo