Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!purdue!mentor.cc.purdue.edu!pur-ee!pur-phy!murphy
From: murphy@pur-phy (William J. Murphy)
Newsgroups: comp.dsp
Subject: FFT for integer data
Message-ID: <2620@pur-phy>
Date: 4 Oct 89 16:18:32 GMT
Reply-To: murphy@newton.physics.purdue.edu.UUCP (William J. Murphy)
Organization: Purdue Univ. Physics Dept., W. Lafayette, IN
Lines: 29


I have a few questions for the dsp experts and amateurs.
I acquire data from two microphones and end up with a file that will
contain N 16-bit samples of the microphone voltages.  Since the data is 
real, I am considering crunching it with an FFT routine as described by
NUMERICAL Recipes for putting one sequence in the real part of the array and
the other sequence in the complex part.  This should half the calculation
time of the FFT.  

I have already copied/programmed the FFT routines in pascal/Turbo-Pascal
on a PC, but what I am wondering is if NR's procedure is particularly
fast or slow.  I do know that there is a difference in speed between
Microsoft C and TP5.0 of about 2x for doing Mandelbrot calculations for 
the same algorithm.  

1.)  Does anyone have some comparative ratings for the NR routine four1
     and other routines in Pascal or C?

2.)  Though this shows my naivety about the details of the FFT, Is there
     any acceleration available for strictly integer data input versus
     real input to the FFT routine? i.e. is there an optimization available
     for integer data?  I assume that regardless of input you will return
     a real complex spectrum instead of an integer complex spectrum?

Thanks,


-- 
 Bill Murphy        murphy@newton.physics.purdue.edu