pruned FFT/IFFT

Hi folks,

Im working on developing a Matlab version of pruned FFT/IFFT algorithm.
But although i think that the implementation is correct, the reults are not
equal to full FFT. So, in a fast x-correlation problem this mismatch arises
out of the maximum (the time delay) is somtimes shifted.

I am at pains to search about this accuracy/mismatch but i have not found
the point. I need your knowledge and help.

Thanks in advance,

Patrick

On Jun 3, 9:05&#4294967295;am, "patrickm" <patrikiopat...@hotmail.com> wrote:
> Hi folks,
>
> Im working on developing a Matlab version of pruned FFT/IFFT algorithm.
> But although i think that the implementation is correct, the reults are not
> equal to full FFT. So, in a fast x-correlation problem this mismatch arises
> out of the maximum (the time delay) is somtimes shifted.
>
> I am at pains to search about this accuracy/mismatch but i have not found
> the point. I need your knowledge and help.
>
> Thanks in advance,
>
> Patrick

I don't use Matlab, but I do know that it has real to complex FFTs and
complex to real IFFTs.  So why would you bother to prune things
yourself when there's something similar already available?

I doubt that your algorithm is working right.  You should check
whatever points you're getting from the pruned FFT and compare them to
the same points in a 'known good' FFT program.  Then check your IFFT.
If you're doing pruned forward and inverse algorithms and then doing a
cross correlation, you've got 3 distinct places where you can make
mistakes.  Check each of the 3 parts against results that you know to
be correct.

I also don't know if you're doing the cross correlation correctly.
Matlab has a function to do that, too.

Kevin