ABSTRACT
This chapter explores several approaches to adaptive filter tuning, mainly in the context of Kalman filtering (KF). The process of estimation and an appropriate choice of the values for the filter initialization and noise statistics in the KF is known as adaptive filtering or adaptive filter tuning. Interestingly enough, in communications signal processing, any recursive estimation/filtering itself is called adaptive filtering, which according to, is not a correct usage of the term adaptive filtering. The initial value of the process noise covariance Q is chosen to reflect the uncertainty in the dynamic system model and any un-modelled noise effects. In adaptive filtering, the unknown noise statistics are estimated online as the filter evolves using the measurements. A heuristic approach to adaptively estimate Q and R could be employed when one has access to extensive prior experimental data. Interacting multiple model was proposed as a sub-optimal hybrid estimator with a finite number of models pertaining to the different target motion modes.
