FFT inside gen?

it's fair to assume that gen~ is not designed to operate a fourrier transformation. It works each sample after the other. It's more designed to be a subpatcher inside a pfft~ if you want to do that kind of things, so you may want to ask these programmers to send you a frequency domain signal indeed. this might be tricky of course...
but if you don't do that you'll have to create a buffer that fills at every vector, then as you process calculation on what is inside that buffer ; you'll have to fill the next buffer...at the same time... and knowing that the calculation can't go faster that sample rate, it is indeed not very practical....

hi vichug!
Thanks for your thoughts!

and knowing that the calculation can’t go faster that sample rate

This is not quite true right? I mean I have done for loops with multiple iterations per sample. Maybe I misunderstand you.
I'm not sure about all this, but to me it seems at the moment, that not being able to export something that can work with frequency domain signals s a big liability that I'd like to get rid of. Right now it seems to me that there is quite a market for gen exports :) so I'd also invest some time to get some homebrew FFT going.. if it's possible(in a more or less efficient way) that is. Which I'm not a hundred percent sure about.

100% possible!!!

If you use a window of N samples, which is base 2 power (2^ß=N), then you will need ONLY (N*ß)/2 complex multiplications and N*ß complex additions. in "normal" DFT you will need N^2 complex mul's! 2N/ß ratio!
You can do 2 signals simultaneously if you only use real values input (like most audio sig).
Complexity is O(ε log N) for the classic Cooley-Tukey algorithm.

For example, take N=2^12=4096 samples window, at 44100 sr (lowest freq bin is ~11Hz):
you will need ca. 50K mul's in a time duration of ca. 100ms… thats 12 mul's to calculate between samples…

so you can even add some overlapping windows and run HD Jitter patch at that time.

@VICHUG, making a buffer fft code, is exactly what the fft msp objects do in a normal patch. classic FFT algorithm (Cooley-Tukey) are sample by sample DSP, to minimise memory requirements.

Im just working on making a simple FFT in gen~ for getting high resolution spectra of speech modulation frequencies (less then 12Hz).

If you wanna work on it together it could be interesting, and we could post it here for gen~erations to come.

Just pick-up any beginners books on FFT with some basic easy ways to implement it in code (I hope you understand FORTRAN :))

Keep us updated! Cheers!

Did you three ever get anywhere with this? Was looking into it but it looks as if the FFT implementations I've seen require passing arrays as arguments which is impossible in gen.

I'm interested in this project, too. Please let us know of your progress!

I was thinking of using a buffer that was 1024*1024 samples and then the function would just write into it at varying places and return the offset and the length. I'll try it later.

-Matt

Buffer is a kind of Array. Try it out w stereo buffer to function as complex array.

I will go back to this. Completely forgot about it, and it can be very nice Object to have + great DSP tutorial...

So... here's what I think it should be but there's a really un-useful error. I have no clue how to fix it. But basically this is an exact replica of http://introcs.cs.princeton.edu/java/97data/FFT.java.html

The only difference is instead of creating and passing arrays it uses it gets passed the buffer location of the beginning of the array, the length of the array and the next "free" piece of real estate in the buffer.

I have no clue if it should work in gen but it seems pretty spot on in terms of math.

I'm pretty sure you're getting that error because gen cannot perform recursion.