Particle filtering for nonlinear dynamic state systems with unknown noise statistics

In this paper, we provide a tutorial for the applications of cost-reference particle filter (CRPF) to problems in signal processing disciplines. CRPF works in particle filtering (PF) framework although it may not be viewed as a Bayesian approach because the estimation is not based on the expected posterior function. CRPF has an interesting feature, i.e., the information of the noise statistics is not needed in its applications as opposed to the cases of the Kalman filter and standard PF approaches that work in dynamic state systems. Therefore, it is highly effective when the noise information is not available; nevertheless, it may not show optimal performance in general. In this paper, we introduce and disseminate this useful approach that is not known to many researchers even in related fields, and show how to effectively apply to problems which we provide as examples.