[Numpy-discussion] FFT usage / consistency

Charles R Harris charlesr.harris@gmail....
Tue Jul 29 11:55:27 CDT 2008


On Tue, Jul 29, 2008 at 8:56 AM, Felix Richter <felix@physik3.uni-rostock.de
> wrote:

> > I quickly copy-pasted and ran your code; it looks to me like the results
> > you calculated analytically oscillate too fast to be represented
> > discretely.  Did you try to transform different, simpler signals?  (e.g.
> a
> > Gaussian?)
> Yes, I run into the same problem.
>
> Since the oscillation frequency is given by the point around which the
> function is centered, it would be good to have it centered around zero.
> The FFT assumes the x axis to be [0..n], so how should I do this?
> The functions I have to transform later won't be symmetrical, so the trick
> abs(fftdata) is not possible.
>

You can apply a linear phase shift to the transformed data, i.e., multiply
by something of the form exp(ixn), where x depends on where you want the
center and n is the index of the transformed data point. This effectively
rotates the original data. Or you can just rotate the data. If the data is
not symmetric you are always going to have complex components. What exactly
are you trying to do? I mean, what is the original problem that you are
trying to solve by this method?

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.scipy.org/pipermail/numpy-discussion/attachments/20080729/a52ce926/attachment.html 


More information about the Numpy-discussion mailing list