Protocol-Based Particle Filtering for Nonlinear Complex Networks: Handling Non-Gaussian Noises and Measurement Censoring

In this paper, the particle filtering problem is investigated for a class of discrete-time nonlinear complex networks with stochastic perturbations under the scheduling of random access protocol. The stochastic perturbations stem from the on-off stochastic coupling, non-Gaussian noises and measurement censoring. The random occurrence of the on-off node coupling is governed by a set of Bernoulli distributed white sequences, and two kinds of measurement censoring models (i.e. dead-band-like model and saturation-like model) are characterized by the predetermined left- and right-end censoring thresholds. To alleviate data collision over the networks, the so-called random access protocol is elaborately exploited to orchestrate the process of measurement transmission. Moreover, two expressions of the modified likelihood function are established to weaken the adverse effects from the measurement censoring. Accordingly, a protocol-based filter is designed in the auxiliary particle filtering framework, where the new particles are generated from a mixture distribution and the associated weights are assigned based on the derived likelihood function. Finally, a multi-target tracking application is taken into account to demonstrate the practicability and effectiveness of the developed filtering scheme.

Automated summary

This paper investigates the particle filtering problem for a class of discrete-time nonlinear complex networks with stochastic perturbations under the scheduling of random access protocol. The stochastic perturbations include on-off stochastic coupling, non-Gaussian noises, and measurement censoring. A random access protocol is used to alleviate data collision over the networks, and two expressions of the modified likelihood function are established to weaken the adverse effects from measurement censoring. A protocol-based filter is designed in the auxiliary particle filtering framework to generate new particles and assign weights based on the derived likelihood function. The developed filtering scheme is demonstrated to be practicable and effective through a multi-target tracking application.