Experimental Physics and
Industrial Control System

Till Straumann <[email protected]> · Tue, 21 Jun 2011 15:05:02 -0500

On 06/21/2011 12:42 PM, Michael Davidsaver wrote:

Has anyone done a routine for the aSub record (or similar) which does an
fft?

My input is complex data (two waveforms I and Q) as a function of time
(512 samples). Data updates come at ~5Hz.

I was thinking of using

http://www.fftw.org/fftw3_doc/Complex-DFTs.html#Complex-DFTs

I have used that but with 512@5Hz you could probably use a
brute-force algorithm.

FWIW: I also implemented a (IMHO) quite optimized FFT
algorithm which used AltiVec/SSE but I couldn't it to run as
fast as fftw3.

Bottom line: if you need something fast + reliable (but code
size doesn't matter too much) then it's hard to beat fftw3.

Just make sure to configure it with altivec/sse if you want
to let it use your vector co-processor.

Cheers
- Till

Suggestions?