08-15-2011 09:26 AM
Hi everyone,
I'm carrying out a FRF operation between two acquised signals, so I decided to simulate it simply first.I created two dummy sine signals with 50 Hz frequency value and "1" amplitude level for both.And I divide the second signal with a integer and took the FRF function between them, as below.
as expected for all frequencies, the result is 0,5 which comes from the ratio of sig2 to sig1.
But, when I changed the integer value to a float, 1,3 for example, it results nonsense;
Why these two processes do not match?
Thanks,
Ozgur
08-16-2011 11:10 PM
Hello Ozgur,
Does this happen with every DBL value you use there, not just 1.3? Were you able to probe the wire both after the Divide node and after the dynamic data converion node to see where the value first comes through as jibberish? I'll try to reproduce this on our end to get a feel for what is going on.
Thank you,
Deborah Y.
08-17-2011 12:01 AM - edited 08-17-2011 12:05 AM
@Deborah Y wrote:
Hello Ozgur,
Does this happen with every DBL value you use there, not just 1.3? Were you able to probe the wire both after the Divide node and after the dynamic data converion node to see where the value first comes through as jibberish? I'll try to reproduce this on our end to get a feel for what is going on.
Thank you,
Deborah Y.
Deborah,
Bet it does happen with every VALUE divisor that cannot be exactly represented as an Xbit binary !
Ozgur. the PC only knows 1's and 0's 1,33 is estimated in the math processor (Closely estimated but not EXACTLY represented) your graph shows the imprecision of representation of a real number and the Quantisiization error of the sine wave simulation
08-17-2011 01:22 AM - edited 08-17-2011 01:24 AM
Deborah, it happens every time as Jeff indicated. Thanks anyway.
Jeff, am I understanding right, you're saying that the graph shows the sum of the errors that caused by the uncertainity? And there is another thing, when the data is not simulated but acquised, there will be not a precise amplitude and frequency valuei will it the FRF block go crazy then ? 🙂
I've tried that process with several post-processing softwares; MATLAB, dbFA Suite (A vibratory daq system interface). All results different 🙂
Do you have any suggestions to solve that? Or can it be corrected?
Regards,
Ozgur
08-17-2011 07:41 AM
Is it a error I'm facing with, or else?
I've tried to get the FRF by using Auto and Cross Spectrums, the result was exactly same with the FRF block.
I'm confused a little...
09-01-2011 03:34 PM
Hello Ozgur,
I apologize for the delay in response. You are correct so far as the error Jeff described and quantization error/noise can be modeled. Pulling in a real signal vs. simulated signal should help a bit based on the signal, as well as using integers for dividing. It seems the post-processing software you've used all handle it a bit differently so for your acquired signal it may not be possible.
Regards,
Deborah Y.