MATLAB implementation of Non-Uniform Fast Fourier Transform in 3D based on Jeff Fessler's NUFFT package:

http://web.eecs.umich.edu/~fessler/code/index.html

The version here doesn't have as many options and so is maybe a little easier to understand.

• To install: put the @nufft_3d folder in the path. See test_nufft_3d for an example.
• Also need some files from parallel on the MATLAB path (e.g. pcgpc, pcgL1, HWT, fistaN)
• Uses the sparse matrix formulation only.
• Uses gpuSparse for single precision on GPU (if available).
• Radial kernel option (higher accuracy for the same number of convolution coefficients).
• Because sparse matrix multiply and transpose multiply are vastly different in performance, the code stores H and H' separately and uses the faster operation. No longer the case since CUDA-11.