Power Spectrum Values

tourette95 · November 17, 2019, 7:41pm

Hi everyone!
I'm triyng to run an EEG experiment using Power Spectra with bS , but I don't understand the values I got from FFT and Welch:

When I compute Absolute power in neuroscan software I get values like: 47.576 (for a given electrode, let say O2 in Alpha Band), and I get similar values with Neuronic Software, however, when I do it in BS I get something like this: 4.53456E-15 (???!) ans similar values using PSD Welch.

I think it has to be due matlab fft computation but I'm not sure, anyone else got similar values computing PSD or absolute power??

Thanks in advance!

Francois · November 18, 2019, 11:17am

First you have an obvious reason for the 10^-16 factor difference: the units.
Brainstorm saves everything in the international system of units. EEG signals are in Volts, other programs probably save everything in microVolts. 1microV = 10^-6 V
The FFT/power power is in V^2/Hz, which is equivalent to 10^-12 microV/Hz

Then all the other parameters (duration of the estimator window, windowing...) affect the range of amplitudes you get in output. If the algorithm is not the same, you can't expect to get the same values in output.

You can check what is done on the Brainstorm side directly in the code:

github.com

brainstorm-tools/brainstorm3/blob/master/toolbox/timefreq/bst_psd.m#L78


end

% Next power of 2 from length of signal

% NFFT = 2^nextpow2(Lwin);      % Function fft() pads the signal with zeros before computing the FT

NFFT = Lwin;                    % No zero-padding: Nfft = Ntime 

% Positive frequency bins spanned by FFT

FreqVector = sfreq / 2 * linspace(0,1,NFFT/2+1);





% ===== CALCULATE FFT FOR EACH WINDOW =====

Nbad = 0;

for iWin = 1:Nwin

    % Build indices

    iTimes = (1:Lwin) + (iWin-1)*(Lwin - Loverlap);

    % Check if this segment is outside of ALL the bad segments (either entirely before or entirely after)

    if ~isempty(BadSegments) && (~all((iTimes(end) < BadSegments(1,:)) | (iTimes(1) > BadSegments(2,:))))

        disp(sprintf('BST> Skipping window #%d because it contains a bad segment.', iWin));

        Nbad = Nbad + 1;

        continue;

    end

    % Select indices

    Fwin = F(:,iTimes);

tourette95 · November 19, 2019, 5:09pm

Hi Francois!
Thanks for your reply!
I see it clear now!
But, then how can I convert these output values to be compatible with studies that use microvolts units?

should I just range these output values 12 zero to left?
(yes, my english is poor )

Thanks in advance!

Francois · November 22, 2019, 11:13am

Yes, and also make sure you have the same number of bins in the FFT (= same estimator window length) and a similar window applied (= hamming)