import scipy.fftpack

F_data = np.zeros(data.shape)

for i in np.arange(data.shape[n]):#the value n is the time dimension

F_data[:,:,i] = np.fft.fftshift(np.fft.fft2(data[:,:,i] )) #position of i depend of the dimension of time

Hello Maxime,

Thank you for your answer. However, iteratevly performing 2D FFT I will get a matrix of spetial frequencies with time [Kx, Ky, t] while I am looking for wavenumber with frequency matrix [Kx, Ky, w].

I've tried:

FFT_3D = np.abs(np.fft.fftn(SignalMatrix))  #n_dimentional FFT

But how to plow it concidering Kx, Ky and w in order to have 3D surface of the signal spectrum.

fig = plt.figure()

kx = np.fft.fftfreq(SignalMatrix.shape[0])

ky = np.fft.fftfreq(SignalMatrix.shape[1])

freq = np.fft.fftfreq(SignalMatrix.shape[2])

#plt.pcolormesh(kx,ky,freq,FFT_3D[:,:,:]) #something like this...

you need to add the sample spacing in the fftfreq. But you compute the fft with spatial and temporal in the same time.

I think representing a 3D grid spaced matrix entirely will be very difficult to read. Why not perform a dimension reduction to observe the results? Or maybe a scatter plot? Or a set of spatial image that you shift in time..

Smething like that http://stackoverflow.com/questions/15582105/python-plot-stacked-image-slices

Maxime, I'd rather looking for way how I can treat my data in the way presented in following article (p.4 fig 6).

So for this 3d fft is done and projected on the 2d map, (kx, ky and w, its magnitude shown by color).

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.458.2148&rep=rep1&type=pdf

The approach presented by you is good rather for animation than for overal signal analysis in order to perform filtering at the end.

I think you just have to make a dimension reduction over the time axis or frequency axis. like in the article, a RMS value is done on the time axis. That lead to have a 2d image (kx,ky) with the rms value (on filtered signal) which give you the color.

Maxime, thank you very much! Indeed you revealed me the thing I was missing about dimension reduction. 

Hi,

can anyone help me how to compute 3D FFT in matlab.

Thanks in advance.

With regards,

Sivakandan

Andrii Kulakovskyi

Hello! I am also interested in computing a 3D FFT of a 3D matrix where one axis represents time, and the other two axes are spatial. Would you be willing to share your code of how to iteratively compute the FFT to project on a 2D map of the wavenumber and frequency?

Thank you!

Hello,

I created a small python notebook for young researches who struggle with this question. Please find enclosed.

Kind regards,

Andrii

@Andrii, many thanks for this nice notebook. However, how do I get the file "data.npy"?