Note: You must be registered in order to post a reply. To register, click here. Registration is FREE!
T O P I C R E V I E W
kfauth
Posted - 02/06/2002 : 02:30:25 AM I am presently testing new routines of mine (for use within nlsf) that use forward and backward fft operations. I have had systematic deviations from my expectations and believe to have traced them down to the following:
Let us consider datasets of length 1024.
Data_A is zero everywhere except "1" in channel 512 (or any other channel)
Consider a "response" function nonzero in time space in three channels. In order to convolute the two, we give the response function in the shifted configuration:
Thus, the response function is zero everywhere except, e.g.: resp[1]= 0.8 resp[2]= 0.1 resp[3]=...=resp[1023]=0 resp[1024] = 0.1
(This is in fact a symmetric response function but I intend to use the routines with nonsymmetric ones)
Now fft-ing the two datasets from script with the appropriate settings (unchecked items normalize, shift results), complex multiplication and appropriate backward fft yields indeed a replica of the response function, however, its center (value 0.8) is shifted by one channel to the left (i.e. channel 511 instead of 512) with respect to the original data_A.
Now, this is important, since I want to use it in a fitting routine and the shift will produce systematic offsets, won't it?
Thanks for any help.
2 L A T E S T R E P L I E S (Newest First)
kfauth
Posted - 02/07/2002 : 01:16:30 AM The issue is solved. Sorry to have waisted your time. It was an error of mine in porting code. This post could well be removed from the forum. Kai
greg
Posted - 02/06/2002 : 1:24:54 PM Sorry, this is as clear as mud to me. Why not send a project and script to tech@originlab.com?