| T O P I C R E V I E W |
| saraya1 |
Posted - 05/14/2003 : 04:00:33 AM
Hi,
I experimented with Origin's FFTs and IFFTs on N real data points with a resolution of dt seconds. Since I want to do inverse FFT I need to keep in mind that:
a) Shift Results is disabled
b) Normalize Amplitude is disabled
Well, what's interesting that
1) According to the Manual the first column of the FFT-worksheet contains the frequencies from (0 .. 1/dt*(N-1)/N) Hz.
THOUGH I am almost convinced that the frequencies rather seem to be stored in this case in some wrap around manner. The first row contains the constant contribution (f=0Hz), then follow ascending positive frequencies and then in descending manner negative frequencies. The frequency data seem to be mirrored at (N/2+1)th data point.
2) Performing a IFFT always produces a wrong time scale. I always need to multiply the column Time with the number of points N in the dataset to get the proper scaling. What's wrong here?
Probably I got something wrong..., possibly 2 times... Am I?
Marko |
| 5 L A T E S T R E P L I E S (Newest First) |
| easwar |
Posted - 05/15/2003 : 09:07:37 AM
Hi Marko,
Please send your opj to tech support so that we can take a look at this.
Easwar
OriginLab.
|
| marko |
Posted - 05/15/2003 : 07:07:14 AM
quote:
time axis stats at say 10, instead of 0, if you do FFT and then IFFT, the recreated signal has a time axis that starts at 0 - the offset of 10 gets dropped in the process.
my axis starts also at 0, so it's not that!
quote:
Is this what you see? And I see that this behaviour is same in v6.1 and in v7.0
Hmm, surprising. Perhaps should cross-check that on another installation of a O6.1...
Marko
|
| easwar |
Posted - 05/14/2003 : 12:01:45 PM
Hi Marko,
Regarding point 2, FFT computation itself does not involve the time axis - the time information is used independently by Origin to set the frequency of the FFT result by looking at the time resolution/step. And in the IFFT process, the frequency infromation is used to back-calculate the time information. In the process, it appears that we are dropping any offset in the time axis. So if your time axis stats at say 10, instead of 0, if you do FFT and then IFFT, the recreated signal has a time axis that starts at 0 - the offset of 10 gets dropped in the process.
Is this what you see? And I see that this behaviour is same in v6.1 and in v7.0
Easwar
OriginLab.
|
| saraya1 |
Posted - 05/14/2003 : 05:21:07 AM
Well, looks like I understood finally myself that the frequencies above fmax/2 do not make sense mathematically, since above that so called nyquist frequency there is no useful information. The FFT algorithm just expects there the mirrored spectrum.
Anyway, point 2 is still unclear. But it looks like this is a bug since O7P-SP1 doesn't show that behaviour in contrast to my O6P1-SP4, which just copies the frequency column in my time column instead of calculating the latter one properly itself.
Marko |
| saraya1 |
Posted - 05/14/2003 : 04:02:40 AM
Ooops, didn't know that it is possible to post without authentication. Sorry. :)
Strange that the message before appears as being written by author Japan... ;)
Marko |
|
|
