What happens to the performance of UKF, or any KF based algorithm for that matter, if we change the step size of calculations (or of making observations)?
Hello, Laya,
Due to my experience with those filters, I'd say that it depends on the system you are analyzing. Once you have a step size which is small enough to capture the full dynamics of your system, reducing it will only lead to more calculations.
What I mean is that, if your sample period is small enough to represent the dynamics of your system, reducing it will not affect the rate of convergence of your filter.
If you want your filter to converge rapidly, you must tune the covariance matrices Q (of the process noise) and R (of the measurement noise) to do so.
I hope I could help,
Vinícius
If the filter is being used to track the dynamics of some system, the question may be addressed more intuitively in the frequency domain. The dynamics of the filter must be adjusted to be as great as, or greater, than the dynamics of the process, in terms o bandwidth. The smaller the step size, the greater the bandwidth of the filter. Of course if the process is being measured in the presence of noise, then there is a tradeoff between tracking error and error due to the noise admitted to the filter. Matching the filter step size to the dynamics of the process, experimentally, from observed data in the presence of noise is a reasonable way to proceed.
You may want to check:
https://www.researchgate.net/post/Joint-State-and-Parameter-Estimation-Kalman-Filter-UKF-EKF
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory