Well, I should say, time domain convolution, using multiplication in the frequency domain ;)
On Sun, 24/1/16, JF saintidle@yahoo.com wrote:
Subject: Fast convolution of 2 tables using FFT To: "pd list" pd-list@iem.at, "Me" saintidle@yahoo.com Date: Sunday, 24 January, 2016, 21:01
Hi all,
I decided to implement my own patch that does frequency domain convolution of two tables.
Have a look at the FFTTry02Long03.pd file and you will need the attached abstractions and iem_tab [tab_complex_mul] (in the path for vanilla or just imported with pd-extended).
It's very fast compared to iem_tab's [tab_conv] but the output doesn't exactly match, there is some error. It's small and I'm thinking it's because of the blocksize being very large to fft~ the tables all in one go. The ifft~ needs a very large division (blocksize) to normalise the gains so I'm presuming there's some number precision funniness going on here? If anyone could chime in on this?
Any feedback/improvement ideas would be greatly appreciated.
Thanks in advance, John.