Anti-eavesdropping distributed fusion filtering for multi-rate nonlinear systems with missing measurements*

In this paper, a novel anti-eavesdropping distributed fusion filtering algorithm is proposed for multi-rate nonlinear systems with missing measurements over sensor networks, where the missing measurements are characterized by Bernoulli random variables. During the information exchange of sensor nodes, it is assumed that there exists an eavesdropper. Accordingly, an active contamination strategy is introduced to ensure the confidentiality of information transmission. Meanwhile, a corresponding compensation strategy is adopted in the receiver side in order to reduce the performance loss. The main objective of this paper is to design a distributed filtering algorithm such that an upper bound on the local filtering error covariance is obtained and the filter gain optimizing the upper bound is given accordingly. In addition, the fusion filter is obtained by using the covariance intersection fusion criterion, and the monotonicity is analyzed about the upper bound on the filtering error covariance with respect to the occurrence probability of missing measurements. Finally, the effectiveness of the proposed algorithm is demonstrated through a simulation experiment.

I. Introduction

In recent years, the distributed filtering problem over sensor networks (SNs) has received extensive attention [1]. Generally speaking, according to the different performance indexes, the relevant research results can be roughly classified into the distributed recursive filtering method [2], distributed H_∞ filtering method [3] and distributed set-membership filtering method [4] and so on. Among them, the distributed recursive filtering method has attracted the increasing interest of scholars due to its wide range of applicability, simplicity in operation and real-time implementation. Notice that a single sensor may be affected by environmental conditions and its own limitations, which degrades the resulting estimation accuracy. In order to compensate for this limitation, the multi-sensor fusion technique has been proposed [5]. In particular, the covariance intersection (CI) fusion has attracted much attention with wider range of applications due to the that there is no need to calculate the filtering error cross-covariance [6].