FFT and complex numbers

Hi,

Thank you to everyone that replied to my original post. I think I understand
the issue a bit better now.

Many Thanks

RG

"rg" <rg1117@hotmail.com> wrote in message
news:cfbg5r$7se$1@news.freedom2surf.net...
> Hi all,
>
> I have been looking at some fft libraries libraries and since DSP is not
my
> main area, there are some things that I am confused about. I noticed that
> all major fft libraries, e.g. FFTW & kissfft, use complex data types in
> which to store both the input and the output of the fourier transform.
>
> I was wondering, must this always be the case, or is it possible to just
use
> array of primitive data types like floats or integers. My application for
> fft is to perform spectral analysis on sound data that is stored as 16 bit
> integers. I normally have to copy this data to the real part of the
complex
> arrays, but I was wondering, would it be reasonable to have an FFT library
> that works directly on the array of integer. The only information that I
> need is the amplitude of the various frequency bins in a given frame and
> nothing more (e.g. phase information).
>
> I hope somebody can explain this situation to me. Many Thanks for your
help
> in advance.
>
> RG
>
>

On 2004-08-11 15:20:12 +0200, Jerry Avins <jya@ieee.org> said:

> Stephan M. Bernsee wrote:
> 
>    ...
> 
>> Mark, "extreme numbers" doesn't sound right to me. "Extreme" suggests 
>> something about their value. What about "enhanced numbers" or "flexible 
>> numbers"? ;-)
> 
> How about planar numbers, as compared to linear?
> 
> Jerry

Yes, I like that. Makes it intuitively clear what it's all about...
-- 
Stephan M. Bernsee
http://www.dspdimension.com

Stephan M. Bernsee wrote:

> On 2004-08-11 05:43:51 +0200, stevenj@alum.mit.edu (Steven G. Johnson) 
> said:
> 
>> The output is still (in general) complex, but that's intrinsic to the
>> Fourier transform...even if you just want the amplitudes, as far as I
>> know there is no way to exploit this for a substantial speed gain.
> 
> 
> You could use the Fast Hartley transform. It's entirely real and conveys 
> the same information as the Fourier transform (for real inputs of 
> course). The magnitudes can be calculated from the transform output.

It takes a small amount of extra processing to recover the amplitudes.
The Fast Hartley Tranform is a minor repackaging of the specialized
varsions of the Fast Fourier Transform. (No free lunch in evidence
here!)

> If I understand the OP correctly, he is able to use floating point maths 
> for the actual transform and is just wondering whether he could use his 
> integers directly as input - that is very possible and easy to do, just 
> make sure you use floats for the computation and your magnitude output 
> to avoid qunatization issues.
> 
> Mark, "extreme numbers" doesn't sound right to me. "Extreme" suggests 
> something about their value. What about "enhanced numbers" or "flexible 
> numbers"? ;-)