Kalman tuning

Hi,
I am a newbie in need of help tuning a Kalman filter. The tuning knobs
I have are the process noise of the filter. I have measurements
statistics characterized by the sensor manufacturer and I am using
that as the measurement noise. Other than the measurements, I do not
have any other truth data. So how do I tell if the tuning is resulting
in a better filter or not. In a crude sense, I can tell a bad filter
if the filter output diverges from the measurements. But other than
that, how do I tell if the filter output is converging to the truth.
How do I tell if one set of process noise figures are better than
another?
Thanks,
- ta
Reply to
ta
Loading thread data ...
"ta" wrote
Look at the innovations (the error) of the Kalman filter. If it's white (uncorrelated), then that's one indication that the filter is working well.
Ciao,
Peter K.
Reply to
Peter J. Kootsookos
You could study the output of the estimator(variance and autocorrelation). The variance should be small and the noise white.
Niclas
Reply to
Niclas Granqvist
How exactly do I tell if the innovations is uncorrelated?
- ta
Reply to
ta
message news:...
Correlate them, and see if the answer averages to zero.
Reply to
Tim Wescott

PolyTech Forum website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.