| Re: Non-uniform FFT? [message #75665] |
Wed, 06 April 2011 08:32  |
Eric Hudson
Messages: 19
Registered: June 2006
|
Junior Member |
|
|
On Apr 5, 1:54 pm, "Kenneth P. Bowman" <k-bow...@null.edu> wrote:
> In article
> < 9cb6bd9c-a2b4-487a-b039-a9e636ba5...@c26g2000vbq.googlegroup s.com >,
> Eric Hudson <ehud...@mit.edu> wrote:
>
>> Hi,
>
>> I was wondering if anyone has implemented a non-uniform FFT algorithm
>> in IDL. We have non-regularly spaced real space data that we need to
>> Fourier transform, and it is painfully slow to do the discrete
>> transform. I have found several c algorithms online (e.g.
>> http://www-user.tu-chemnitz.de/~potts/nfft/download.php) but before
>> launching into either converting them or figuring out how to run C
>> code from within IDL thought maybe someone else had already gone to
>> the trouble.
>
>> Thanks,
>> Eric
>
> The approach could depend on just how non-uniform your data are.
>
> Do you need the whole spectrum, or do you know in advance
> which wavenumbers are of interest?
>
> You can do the DFT using least squares (regression), but that will
> be slow if you need the full spectrum.
>
> If you only need low wavenumbers, you could interpolate to
> a regular grid and then use least squares or the FFT.
>
> Ken Bowman
Hi Ken,
Thanks for the response. Unfortunately I need the whole spectrum (I
have 2D data, slightly irregularly gridded, and want the equivalent of
what you'd see if you did a 2D FFT on regularly gridded data). I had
thought of doing interpolation and then the standard FFT, which I
guess is to an extent what they are doing in these NFFT algorithms,
but it seems they are a little more clever than that, which is why I
was hoping someone had coded the NFFT routine in IDL. For now I am
just directly integrating A(r) exp(i*q*r) over the whole image for
each q, which is painfully slow because I have to loop on q (I don't
have enough memory to make the whole q*r array in one go).
Eric
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
