Estimation and Identification for Wireless Sensor Network Undergoing Uncertain Jumps

Abstract

Network control systems are usually modeled as simple channels having limited transmission bandwidth, nonnegligible communication delays and random packet dropouts, when contaminated by disturbances with Gaussian distributions. However, wireless sensor networks undergo several abrupt jumps that appear as discontinuities that can be modeled as non-Gaussian processes, for which the estimation and identification problem is challenging. We consider a stochastic state space model in which the signal is disturbed with Lévy-type processes. An approximated ensemble Kalman filter is proposed and a method for the sequential importance sampling is given in the signal estimation stage. Comparison between two filtering methods is investigated, which then leads us to a study of particle MCMC method that is adjusted to achieve the model parameter identification.