A Square-root Version Distributed Nonlinear Filter Based on Information Consensus

Abstract

This paper studies the problem of distributed nonlinear state estimation in sensor networks. To solve the problems of nonlinearity in system model, redundancy caused by consensus iterations, and naivety resulting from limited field of view (FOV) of sensors, a novel square-root version distributed nonlinear filter based on information consensus (SRDICF) is proposed. It is developed based on the information weighted consensus protocol with embedded square-root decomposition of the information matrices. The cubature rule is chosen to approximate the Gaussian weighted integral involved in the nonlinear Bayesian filtering paradigm. Then different weights are assigned to the local prior and measurement information to initialize the consensus terms. Finally, the global estimate of each node is acquired with sufficient consensus iterations. The experimental results illustrate that the SRDICF achieves higher estimation accuracy, faster convergence rate and better numerical stability in the existence of naïve nodes and information redundancy. In addition, the SRDICF attains comparable estimation accuracy to the centralized scheme with fewer consensus iterations.