Rfft~ analysis question
RFFT analysis:
SR = 44100-Hz
BS = 8 samples
Input signal = 5512.5-Hz
Formula for calculating the amplitudes: sqrt (re^2 + im^2)
The manual calculation of the second, third and fourth pair of samples shows different values. Someone can explain why? Please see the attached pach.
Hi,
With all those numbers glued together it's hard to see what belongs where.
re:
2.0413e-054 -8.6705e-06-3.7551e-06-3.0716e-060 0 0
im:
0 1.7643e-055.3644e-075.3644e-070 0 0 0
If you write the output of [rfft~] to a table and then read the values from there, you'll notice that real index 1 has value 4 and the rest is all very small numbers in scientific notation, or zero. Very different from the numbers in your manual calculations.
By the way I am using [fft~] & co a lot these days and they seem to work fine. The only thing I didn't know yet is where [rfft~] stores the Nyquist coefficient. So now is a good opportunity to check that in your patch with [osc~ 22050]. O wow, the Nyquist coefficient doesn't appear anywhere. So if you need Nyquist coefficient, use [fft~] instead.
Katja
@Maelstorm said:
The Nyquist frequency is in the same place for both of them.
Indeed. My conclusion was too hasty, maybe I did the wrong test. So if you want to store the halfcomplex spectrum in a table, you just use tablesize=blocksize/2+1 as a minimum, and you can do all complex operations the proper way. That's perfect. Thanks for pointing this out.
Katja
Hi,
I discovered where the error is. In the manual calculation I have to use the values extracted like they are. For example: 2.0413e-054 and not 2.0413 like I did initially.
See the patch attached.
Thanks anyway,
http://www.pdpatchrepo.info/hurleur/rfft_question_resolved.pd
Oops! Looks like something went wrong!
