FFT-based Convolution

A fast 1D and 2D convolution.

Description

FFT-based convolution.
See readme.txt for more informations.

Please contribute if you find this software useful. Report bugs to [e-mail]

Luigi Rosa
Via Centrale 27
67042 Civita di Bagno
L'Aquila --- ITALY
email [e-mail]
mobile [phone]
http://utenti.lycos.it/matlab

• MATLAB release: MATLAB 6.5 (R13)

[Yuval Cohen]

Matlab has a function called fftfilt (in the signal processing toolbox) that does the exact same thing, only faster. This algorithm is also known as Overlap add.

[Ewing Young]

 

[Jay Liu]

fftfilt in matlab is only for 1D cases.

[Park Hanguk]

Isn't the standard implementation of the conv function in Matlab already based on fft?....

[Pedro Cati]

I suppose not. Conv performs time domain convolution. A possible improvement could be wavelet-based convolution...FFtfilt is a 1D FFT-based convolution (OLA method) and this operator in N-dimensions can be performed in N step.

[Andreas Silzle]

Comparison of calculation times
Signal length 1 second, impulse length 1 second, fs=44100, vectors (one-dimension).
conv 103.8 s
filter 79.4 s
fftfilt 0.38 s
convfft 0.92 s
Signal length 10 seconds, impulse length 1 second
fftfilt 1.28 s
convfft 14.61 s
Matlab 6.5.1
AMD Athlon XP 2000+ (1.67 GHz), 512 MB RAM

[Ila Fiete]

This is a simple code for 2-d convolution, based on the built-in matlab routine for 2-d fft. You can write it easily yourself, but this code is pretty clean and straightforward.

For large matrices, filter2 or conv2 are v. slow in matlab for large matrices. This code leads to a considerable speedup, by a factor of 5-10 on my machine, when I convolve a 500^2 matrix with a 100^2 one.

[Rameez Ahmed]

I need coding for convolution coding theory in MATLAB. Need some help!!

[Vishnuvenkatesh Dhage]

 

[Felipe G. Nievinski]

The code is bug-prone because it has lots of unnecessary duplicated portions. I'd recommend instead <http://www.mathworks.com/matlabcentral/fileexchange/24504-fft-based-convolution>;.