the IFFT returns zeros

Hi,

I'm trying to do a deconvolution with the formula from a standard (to check if it returns the same results as the function "deconv"):

ifft( fft(h1) / fft(h2) )

fft(h1) / fft(h2) gives me a NxN matrix as expected. However, only one column is different to zeros(:,1), but it can make sense.

Then, when I try to do the inverse fast Fourier transform, Matlab often returns only zeros !

Here is the part of my code:

if true
AA=fft(a(:,4),siz_ft); 
BB=fft(a(:,3),siz_ft);
CC=AA/BB;
DD=ifft(CC,siz_ft);

    for k=1:siz_ft
        if DD(:,k)~=zeros(siz_ft,1)
        DD(1:4,k)
        end
    end
end

It means that DD returns something only 4 times over 16 different measures.

Does anyone have an idea about my problem ?

Thank you for any help !

Answer 1

[Matt J]

For starters, I think you need to be doing elementwise division.

 CC=AA./BB

Beyond that, it's not clear from your description what you're seeing. First you say your seeing all zeros and then you say "DD returns something only 4 times over 16 different measures" which I assume to mean is not all zeros.

I'm also not sure how you're able to loop over DD(:,k) without getting an error message, because in your code DD only has 1 column.

Answer 2

[Ambroise WALLYN]

Hi, thank you for your answer !

So, I already find that it works with elementwise division but I thought it was not correct...

Then, I'm sorry not being clear enough :

the code I wrote is part of a loop from 1 to 16 (where a(:,3) and a(:,4) change) and so DD returns something only 4 times over the 16 measures from the loop

Finaly, I don't have an error message since CC=AA/BB returns a matrice n-by-n

Whatever, thank you for your answer, since the element-wise division AA./BB works !

[Matt J]

OK. If my answer helped, you are encouraged to Accept-click it formally (my Answer, not your response).

[Ambroise WALLYN]

Ok, sorry. I've just done it Thank you again