Calculating SEF (spectral edge frequency) of a EEG

I currently have a file with a 4 channel EEG, and I want to calculate the SEF (95%) using that raw data. I wanted to know if there is as automated way of doing so using Matlab. I also have computed data Hz x time x dB of the given raw EEG file.

ADICIONAL INFO: The SEF is a Hz by time chart where the Hz shown is the threshold frequency where 95% of the EEG power lies beneath it at a given time (I don't know which algorithm to use) though.

Thank you in advance!

[Christoph F.]

So you have a vector P of numbers of the power at certain frequencies, and you want to find the vector index n at which sum(P(1:n))/sum(P) >= 0.95?

[Francisco Dias]

Yeah that is pretty much it.

Answer 1

[Star Strider]

I would use cumtrapz to do the integration, and interp1 using the cumtrapz result and the frequency vector to find the SEF.

Example —

% First, Create Data
t = 0 : 0.01 : 1;
for f = 1 : 2 : 10
    y(f,:) = sin (2 * pi .* f .* t);
end

L = length(t);                                          % Signal Length
Ts = t(2) - t(1);                                       % Sampling Time Interval
Fs = 1/Ts;                                              % Sampling Frequency
Fn = Fs/2;                                              % Nyquist Frequency
Fv = linspace(0, 1, fix(L/2)+1)*Fn;                     % Frequency Vector
Iv = 1:length(Fv);                                      % Index Vector
Fourier = fft(y, [], 2)/L;                              % Fourier Transform (Along Rows Of Matrix)

Fouriers = sum(abs(Fourier));                           % Spectrum
IntSpectrum = cumtrapz(Fv, Fouriers(Iv));               % Numeric Integration
SEF = interp1(IntSpectrum, Fv, 0.95*IntSpectrum(end), 'linear');    % Interploate To Find ‘SEF’

figure(1)
plot(Fv, Fouriers(Iv)*2)
hold on
plot([1 1]*SEF, ylim, '--r')
hold off

[Francisco Dias]

I don't seem to be understanding the code. SEF is not a constant value!! it is a value per sample, in other words the result I need is a vector of SEF values for each time. for each sample I need to know the frequency value that holds 95% of the total power beneath it.

[Star Strider]

Consider that to be for each segment of your signal.

Segment your signal, do the short-time Fourier transform on each segment, and calculate the SEF for each part.

I would use the Signal Processing Toolbox spectrogram (link) function to do the short-time Fourier transforms on each time-windowed segment. (Choose the segments along the ‘time’ axis to do the SEF calculations on.)

[Noelia Scarinci]

I have my data, so I don't have to create it. How I do to adapt my data (9771 x 1) to a matrix with "y" dimensions?

[영진 정]

I love you bro...

thx...

[Star Strider]

@영진 정 — My pleasure!

And thank you for the vote!