Hybrid Distributed-Centralized Transmission Censoring for Dynamic Target State Estimation

A new information theoretic transmission censoring scheme is proposed to reduce communication costs in target tracking distributed sensor networks (SNs) while maintaining a high state estimation performance. In this scheme, at each time step, each sensor evaluates a sensor-side objective function for all the valid subsets in the SN using its locally calculated posterior target state probability density function (PDF) and its knowledge of all the other sensors' measurement models. Each sensor will then generate a vote for the locally determined best subset of sensors. The fusion center (FC) will tally the votes, select the subset with the most votes, and request the posterior state PDF from each sensor in the winning subset. Because the proposed approach utilizes both distributed (with sensor-side information) and centralized sensor management strategies, we name the proposed approach Hybrid Distributed-Centralized (HDC) censoring. Two new information theoretic sensor-side objective functions based on conditional expectation are proposed and their closed forms are derived, providing a computationally efficient HDC implementation. Numerical results for several distributed radar SN tracking examples are provided to demonstrate the effectiveness of the proposed method. Communication cost and computational complexity of the approach are also analyzed.