Modelling Mismatch and Noise Statistics Uncertainty in Linear MMSE Estimation

Standard filtering techniques, such as Kalman, sigma-point or particle filters, assume a perfect knowledge of the system. This implies that both process and measurement functions, and the system noise statistics, are assumed known and fit the reality. Regarding the noise statistics this involves knowing not only the distributions but also their parameters. In this contribution, we explore the impact of system model mismatch and uncertain noise statistics parameters into linear minimum mean square error estimators for linear discrete state-space models. Illustrative examples are shown to support the discussion.