comfort-fftw-extra: Signal processing algorithms based on FFTW and BLAS

This package adds utility functions to the package comfort-fftw. For fast processing it uses additional functions from comfort-blas and cblas-ffi. This way you benefit from machine optimized implementations of FFTW and BLAS.

We provide algorithms for:

• Compute two Fourier transforms of real data via one complex valued Fourier transform in one and two dimensions.

• Fast Cyclic and Acyclic Convolutions (i.e. chooses good data vector lengths for Fourier transform)

• Fast convolution or correlation of a short vector with a long one.

• Convolve two long vectors but keep only a short clip from the result. Can be useful for a pruned chirp transform.

• Batched convolution: One with many, Many with many.

• Streaming form of convolution of a vector with a possibly infinite stream of blocks.

• Fast blockwise Fourier transform (short-time Fourier transform)

• Compute only narrow frequency band of a spectrum faster than the whole spectrum.

• Synthesize time data from a narrow frequency band of a spectrum (aka band limited upsampling, aka interpolation).

Attention: Some functions use matrix transposition. For fast matrix transposition add a flag like this to your cabal.project:

package cblas-ffi
  flags: +openblas