On stability of a class of filters for non-linear stochastic systems
share
This article considers stability properties of a broad class of commonly used filters, including the extended and unscented Kalman filters, for discrete and continuous-time stochastic dynamic systems with non-linear state dynamics and linear measurements. We show that contractivity of the filtering error process and boundedness of the error covariance matrix induce stability of the filter. The results are in the form of time-uniform mean square bounds and exponential concentration inequalities for the filtering error. As is typical of stability analysis of filters for non-linear systems, the assumptions required are stringent. However, in contrast to much of the previous analysis, we provide a number of example of model classes for which these assumptions can be verified a priori. Typical requirements are a contractive drift or sufficient inflation of the error covariance matrix and fully observed state. Numerical experiments using synthetic data are used to validate the derived error bounds.
READ FULL TEXT