Re: Fast Fourier Transform Queries

> You wonder then what comes out when you set ChunkFreq:=1.

If I set ChunkFreq to an even value, then a single frequency with a
value ANAD (as near as dammit) to 2737396 appears at [ChunkFreq div 2].
If I set ChunkFreq to an odd number a smaller spike appears at
[ChunkFreq div 2] followed by reduving values at [(ChunkFreq div 2) +
1..[(ChunkFreq div 2) + 3] down to zero-ish at [(ChunkFreq div 2) + 5].

> You are not by any chance using a FFT that assumes a power of 2 as sample
> size?

As I said before - No. The preable to the code says ...

"{ ForwardFFT:
Perform a complex FFT on the data in Source, put result in Dest. This
routine
works best for Count as a power of 2, but also works usually faster
than DFT
by factoring the series. Only in cases where Count is a prime number
will this
method be identical to regular complex DFT.

The largest prime factor in Count should be less or equal to
cMaxPrimeFactor.

The remaining factors are handled by optimised partial FFT code, that
can be
found in the FFT_X procedures

Inputs:
Source: this can be any zero-based array type of TComplex
Count: The number of elements in the array."

I still don't know how a sine-max of 15000 in a 640 sample FFT comes
out (with an even ChunkFreq) as a single-frequency value of 24737395
(+/- 4). I would have expected it to be the average value of the sine
wave - AFAIR about 2/Pi * 15000 * 640 == 6111549.

Alan Lloyd

.

Relevant Pages

  • Re: Fast Fourier Transform Queries
    ... > If I set ChunkFreq to an even value, then a single frequency with a ... > If I set ChunkFreq to an odd number a smaller spike appears at ... > Perform a complex FFT on the data in Source, ... this can be any zero-based array type of TComplex ...
    (comp.lang.pascal.delphi.misc)
  • Re: Can you trim a FFT?
    ... This is not a FFT answer but have you considered digital difference ... difference filter which will respond to a single frequency. ...
    (sci.math)
  • Re: Fast approximate FFT
    ... There were a couple of articles on approximate FFTs in the IEEE ... Transactions on Signal Processing on FFT pruning. ... >>I know there is a fast method for extracting a single frequency (ie. a ... >>large amplitude components but poor estimates of the small ones. ...
    (sci.math.num-analysis)
Loading