Prev: How to effectively compute FFT from f_min to f_max, where f_min in not zero?
Next: physical layer A/D performance simulation
From: Afinko on 14 Jan 2010 05:57
>On Jun 1, 8:06 pm, "digintu" <digi...(a)gmail.com> wrote:
>you're missing a step or two. if you're gonna stuff two halves of
>your length 128 data into two length 64 FFTs and post-process the
>results, you have to put the even-indexed samples in one and the odd-
>index samples in another. do two 64-pt FFTs, and post process the
>results with 64 butterflies that have coefficients of
>
> W(k) = exp(j*2*pi*k/N) for 0 <= k < N/2
>
>where N is 128. this is called the Decimation-In-Time algorithm.
>
>if you want to do this in normal order and toss the "first half" in
>one FFT and the "last half" in the other FFT, that's the Decimation-In-
>Frequency algorithm and instead of post-processing the results, you
>*pre*-process what goes in with similar looking butterflies. either
>the DIT or DIF can be done in-place where the input is in normal or
>bit-reversed order and the output would then naturally end up bit-
>reversed or normal, respectively. in fact, the DIT with normal input,
>bir-reversed output, can be compared directly to the DIF with bit-
>reversed input and normal order output. a butterfly in the final
>stage (or "pass") of the DIT simply has its function inverted with the
>corresponding butterfly in the first pass of the DIF. (assume the
>first DIT alg is a "forward DFT" and the second DIF alg is set for
>"inverse DFT". "j" in the first becomes "-j" in the second.)
>
>r b-j

Hi Robert bristow-johnson,

Can you pleas give me more information about preprocessing by DIT or DIF,
I want to implement it in DSP. The code for MATLAB could be great. I need
to know what are the steps for preprocessing, and then after FFT how to
combine them.

Thanks in advance.
Afi
 | 
Pages: 1
Prev: How to effectively compute FFT from f_min to f_max, where f_min in not zero?
Next: physical layer A/D performance simulation