ardnivar
01-10-2004, 02:43 AM
Hi People,
What is the accuracy expected from a 2D FFt calculation?.
I require an accuracy of upto 10^(-6) in the real and imaginary parts.
I`m trying to get the 2DFFT of an exponential decay and am comparing the results with a lorentzian having the same parameters. I notice that:
1) I can get an accuracy of 10^(-4) only if i decrease the time step to an unacceptably low value. ( In order to get a reasonable resolution in the frequency domain with this time step, i need to do an 8192x8192 FFT calculation)
2) Secondly i notice that the absolute values are far more accurate.
My question is: What kind of errors could be responsible for this deviation? I`m using the numerical recipies code along with a wrapper which reads the input data and then shuffles it in increasing order and spits out the frequencies and data. As far as i can tell no precision is lost outside the numerical reciepies code. I would be really grateful for any comments or pointers as i`m really stuck.
Regards
Ravi
What is the accuracy expected from a 2D FFt calculation?.
I require an accuracy of upto 10^(-6) in the real and imaginary parts.
I`m trying to get the 2DFFT of an exponential decay and am comparing the results with a lorentzian having the same parameters. I notice that:
1) I can get an accuracy of 10^(-4) only if i decrease the time step to an unacceptably low value. ( In order to get a reasonable resolution in the frequency domain with this time step, i need to do an 8192x8192 FFT calculation)
2) Secondly i notice that the absolute values are far more accurate.
My question is: What kind of errors could be responsible for this deviation? I`m using the numerical recipies code along with a wrapper which reads the input data and then shuffles it in increasing order and spits out the frequencies and data. As far as i can tell no precision is lost outside the numerical reciepies code. I would be really grateful for any comments or pointers as i`m really stuck.
Regards
Ravi