🚀 go-pugleaf

I have some code that is uses the fft/ifft routines from "Numerical
Recipies in C". I've noticed that their algorithms are not always as
quick as the can be.  So, now I'm looking into hacking up some faster
fft routines and I thought I'd ask for contributions or suggestions
before I waste a lot of time on something that someone's already
hacked up.

My questions are:
  1. How much better can they get? if writing my own is only going to
     give me a tiny speed improvement, then it's not worth it. Speed
     is critical right now, not space, though that may change.
  2. Is it feasible to write a quick and dirty integer fft routine
     that takes advantage of the fact that my data are 12 or 16 bit
     integer waveforms. Is there a standard technique for doing this?
  3. Does anyone have any tips on where to look for algorithms or
     material on this sort of thing? Ideally code or pseudocode, but
     also paper or book references.
     
Sorry if this is some what remedial for this group, but I'm a novice
to this area and while I know what I want to use the routines for,
I'm quite ready to design my own fft algorithm from scratch!!

Please reply to me, I'll post a summary if anyone else turns out to
be interested in this topic.

Thanks,
/Jamie


--
1st: mazer@bek-owl.cns.caltech.edu  /-------------------------------------\
2nd: JMazer@caltech.bitnet         /"Caltech: ... science steals your soul,\
3rd: mazer@smaug.cns.caltech.edu   \  don't come here unless you have to." /
uhg: 818-356-6816                   \-------------------------------------/