How to get rid of the 1/f in EEG data?

Hi everyone,

I am not a very experienced user in Matlab. I am doing time-frequency analyses on my EEG data, looking at low frequencies. To do so, I need to get rid of 1/f (pink noise), but I am not sure how to do it. I am using polyfit function to estimate the 1/f slope, but I don't know if I am doing it correctly, and I don't know what to do next. I am following this paper: https://www.sciencedirect.com/science/article/abs/pii/S0306452205011711

Some EEG people using Matlab could help me? I have done the log10 transformation to both the power spectrum and the frequencies of my data. My code is the following:

f = evoked_power_dB.freq(2:212);
x = f
y = mean(eeg_data,1) %mean of the 53 channels - 1x211

p = polyfit(x,y,1) 
x1= f(1:211)
yfit=polyval(p,x1)

figure;
plot(x,y,'o')
xlim([0 4])
ylim([-10 20])
hold on
plot(x1,yfit)
    
fitresult = (evoked_power.powspctrm./evoked_power.freq).^p(1); %we use the non-log-transformed data

Any suggestions/ideas on how to continue?

Thanks,

Iris

[ROHAN JAIN]

Hi Iris,

If it still remains an open question, I have a little suggestion for you.

Are you fitting 1/f line on the power spectral density or eeg data itself ? Since you are using frequency in X-axis, you should consider log transformed power in Y-axis and there fit the regression line using polyfit as you did.

Hope it helps!

Answer 1

[Spencer Chen]

polyfit() is not for estimating 1/f functions.

Something like this is more appopriate:

fout = fit(x,y,'(a/x)^b','Lower',[0 0])

Blessings,

Spencer