[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [f-cpu] just for fun (or: maths can be useful)


philippe.trbich@free.fr wrote:
> On Mon, 17 Sep 2001 19:53:31 +0200
> Yann Guidon <whygee@f-cpu.org> wrote:
> if you want to try some speed FFT, try http:/www.fftw.org.
i know FFTW for a while but i do not need to analyze
megasamples in a row or PVM or codelets... There are many reasons
for me to not use it. it's rather a "mathematical tool", i need
a small "real case" tool.

btw, i found that i need more a DCT than a FFT : this is due
to the fact that DCT can catch one lower frequency component,
thus reducing the spectrum spread when there is a large
difference between sample 0 and sample N-1. This is why the DCT is
used in JPEG instead of FFT and i have a rather similar need.
And FFT requires windowing, DCT doesn't.

I could derive a 8-tap DCT from a 16-tap FFT (winograd if possible)
but i don't have any 16-point winograd code :-( and i don't know
yet how to derive the DCT from the 8-tap winograd code. The
possibilities i know require too much overhead (require much
more multiplies than needed).

> Cordially.
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/