A performance evaluation of branching particle filters: case studies

Alexsandro Machado Jacob, Takashi Yoneyama

DOI: 10.14209/sbrt.2004.179

Evento: XXI Simpósio Brasileiro de Telecomunicações (SBrT2004)

Keywords: Nonlinear filtering Zakai equation Branching algorithm Monte Carlo approximation Parallel processing

Abstract

This paper presents two case studies for the performance evaluation of branching particle filters with the objective of contributing towards understanding and providing useful insight for practical implementation. A performance study based on the robustness of the estimate in relation to the number of independent simulations of three different modes of implementation of the branching particle filter was also made and the results were compared with those obtained by the extended Kalman filter. In the first case, an one-dimensional nonlinear stable process, all the configurations of the branching particle filter produced slightly better estimates than the extended Kalman one when the process was located at regions where the linear approximation was not good. For the second case, the filters were tested in an one-dimensional unstable system and the branching particle filter presented robust results when compared to the extended Kalman one during the simulation steps.

Download