Re: Can you trim a FFT?
- From: "Greg Neill" <gneillREM@xxxxxxxxxxxxxxx>
- Date: Mon, 19 May 2008 11:06:47 -0400
"Tim Little" <tim@xxxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:slrng332rl.j2s.tim@xxxxxxxxxxxxxxxxxxxxxxxxxx
On 2008-05-19, shoo <JGBurgess@xxxxxxxxx> wrote:
I would nievily like to know if there is a method of applying the
fft with the intention of only requiring a predetermind range of
frequency. hense no waste in the fft.
You can, but it won't save you very much. Every point in the spectrum
inherently depends upon every point in the data, so to compute even a
single value you definitely need O(N) operations, and the full
spectrum needs only O(N log N).
Still, if you want to do it anyway, then the Cooley-Tukey FFT
algorithm is relatively easy to adapt. E.g. to calculate the first T
data points of N in the final FFT, you also only need the first T of
the data points from the previous odd/even FFTs.
The net effect is that instead of O(N log N) computation, you do it in
O(N log T).
If you also want to throw away low frequencies, I think one of the
easier ways would be to just multiply the data by an imaginary
exponential of the appropriate frequency, shifting the spectrum so
that the lower edge of the desired band is at 0 Hz. Then do the
low-frequency transform and shift the spectrum back up again.
In your case that would be of dubious value: the difference in
computation for a 25 Hz band vs 30 Hz will be trivial.
Another approach might be to use a periodogram approach.
Try a search on the "Lomb-Scargle Periodogram", for
example. You can search the data for just the frequency
ranges you're interested in.
.
- References:
- Can you trim a FFT?
- From: shoo
- Re: Can you trim a FFT?
- From: Tim Little
- Can you trim a FFT?
- Prev by Date: Re: The quaternion group as a product
- Next by Date: Re: DEF -- Given "the empty set"
- Previous by thread: Re: Can you trim a FFT?
- Next by thread: Re: Can you trim a FFT?
- Index(es):
Relevant Pages
|
