Hanning Window 10%

Dear all,

I am trying to reproduce a previously described analysis and I struggle with one part of it: It says

"data epochs were subjected to a Hanning window (10%)"

While I know that I can create a Hanning window of the desired length via hann(), I am not sure what I should do about the 10%. Do you have an idea?

Thank you very much!

Best

[Bjorn Gustavsson]

Do they mean that the window-length is 10% of the epochs? (That ought to be "very peculiar" if the different epochs are of different lengths, but if they are of constant size then maybe - so you could check for that.)

[Katharina Paul]

the epochs are of constant size, yes. Do you think they just apply the Hanning window over the first 10% and then zero padd the rest?

[Bjorn Gustavsson]

Well, I guess that they apply a Hanning-window with a length corresponding to 10% of the length of the epocs. That would be a low-pass-filter with a rather wide pass-band. Does the plots of the data they use look like it might be low-pass-filtered this might be it. But since I don't know anything more than epochs, Hanning and 10% it is difficult to do more than guess...

[Srija Kethiri]

Hii Katharina,

Can you please share your script files so that we can help you better.

[David Goodmanson]

Hi Katharina,

could you explain what the task is, such as power spectral density, or something else? If it is PSD or something like it, then the general approach is to use the same window in overlapping intervals across the data. Suppose the data has 10000 points and suppose the window length is 10% of rhe data length (just a guess, a la Bjorn). If the window overlap was 50% which is fairly common, then you would multiply the window.*data(1:1000), then window.* data(501:1500), then window.*data(1001:2000) etc. Then you would separately fourier transform each of those 19 cases and take an average of, in the case of PSD, abs(fourier trans).^2 (I am skipping some details involving normalization). That's just one possibility for what might be meant.