MATLAB: Non-uniform FFT with Matlab

fast fourier transformfourierMATLABnfftnon-uniform fftnufft

Hi everyone,

I need to perform a non-uniform fft in 3D (to solve the heat equation using the method described here).

More precisely, I am using a uniform grid in space and a non uniform grid in the frequency domain. After several topics in this forum, I tried to use the toolbox NFFT3 which seemed great. I was able to use it on a non equispaced spatial grid but the Matlab function does not seem to allow for specifying the frequency grid points…

Does anyone have experience on NUFFT with Matlab and, if yes, how do you perform your computations ?

I have read that this function was included into FFTW (used by Matlab) : is the function included in some Mathworks product ?

Best regards,

Anthony

PS : I am also trying to use this toolbox but it somehow lacks documentation…

Best Answer

PS : I am also trying to use this toolbox but it somehow lacks documentation...

The relevant functions nufft.m and nufft_init.m in that toolbox do have help text in them. If the help text is unclear, perhaps you can email the author for clarification.

Related Solutions

MATLAB: How to generate a coneplot with a non-uniform grid

The function CONEPLOT(X,Y,Z,U,V,W,Cx,Cy,Cz) plots velocity vectors as cones at the points (Cx,Cy,Cz) in the vector field defined by U,V,W. The arrays X,Y,Z define the coordinates for U,V,W and must be monotonic and 3D plaid (as if produced by MESHGRID). You should be able to use a grid that is non-uniformly distributed, as long as it meets these requirements. The following simple example demonstrates the use of a non-uniformly distributed grid:

[x,y,z]=meshgrid([0 10 15 30 60],[0 10 30 50 60],[-10 -8 0 2 10]); % non-uniform

[u,v,w]=meshgrid([0 5 11 12 20],[0 5 13 15 20],[0 7 19 20 25]); % non-uniform
cx=1;
cy=2;
cz=3;
surf(peaks)
h1=coneplot(x,y,z,u,v,w,cx,cy,cz)
set(h1, 'FaceColor', 'blue', 'EdgeColor', 'none')

MATLAB: Question regarding FFT MATLAB demo

Hi Martin,

In theory, Fourier transform can be performed using any number of data points.

In old days, people prefer to have the number of FFT points to be power of 2 to take advantage of some computation benefit. Nowadays, at least in MATLAB, doing so no longer brings significant performance improvement so many people just use the length of data samples.

HTH

Best Answer

Related Solutions

MATLAB: How to generate a coneplot with a non-uniform grid

MATLAB: Question regarding FFT MATLAB demo

Related Question