ml-fft

ml-fft

fft library for the ml libraries.

Installation

$ npm install ml-fft

Import in node

var lib = require("ml-fft");
var FFT = lib.FFT;
var FFTUtils = lib.FFTUtils

1D FFT and IFFT

var n = 16; 
var nCols = n; 
FFT.init(nCols);
var re = new Array(nCols);
var im = new Array(nCols);

for(var i=0;i<nCols;i++){
   re[i]=i;
   im[i]=nCols-i-1;
}

FFT.fft(re, im);
FFT.ifft(re, im);

2D FFT and 2D IFFT

data contains the matrix. The even rows contain the real part, the odd rows contain the imaginary part.

var n = 4;
var nRows = n;
var nCols = n;
var data = new Array(nRows*nCols);
for(var i=0;i<nRows;i++){
    for(var j=0;j<nCols;j++){
        data[i*nCols+j]=i+j;
    }
}
var ftData = FFTUtils.fft2DArray(data, nCols, nCols);
var ftRows = nRows * 2;
var ftCols = nCols / 2 + 1;
var iftData =  FFTUtils.ifft2DArray(ftData, ftRows, ftCols);

Matrix-Matrix convolution.

It performs the convolution in the Fourier space(multiplication) and then makes an inverse transformation of the result. The difference in performance can be tested in the BenchMark script.

var n=1024;
var data = new Uint32Array(n*n);
for(var i=0;i<n;i++){
    for(var j=0;j<n;j++){
        data[i*n+j]=i+j;
    }
}

var kn = 21;
var kernel = new Array(kn);
for(var i=0;i<kn;i++){
  kernel[i]=new Array(kn);
  for(var j=0;j<kn;j++){
      kernel[i][j]=i+j;
  }
}

var convolutedData = convoluteFFT(data, kernel, n, n);

License

MIT