Positioning by Cost Reference Particle Filters: Study of Various Implementations

In this paper, we investigate the application of different variants of a recently proposed class of sequential Monte Carlo filtering techniques to the problem of target positioning. The addressed methodology is known as cost-reference particle filtering (CRPF), and its main characteristic is that it does not use probabilistic assumptions related to the states and the observations (i.e., prior distributions of the states and noise distributions). The absence of these assumptions leads to practically more robust performance than the one achieved by conventional particle filters (their theory is based on probabilistic assumptions). We propose some modifications to the originally presented CRPF methods to obtain more efficient and computationally less demanding algorithms. The advantages of the obtained variants of CRPF are discussed and their validity is demonstrated through computer simulations