Comparison of optimal distributed estimation and consensus filtering

Each node in a sensor network estimates the state of a system from its sensor measurements and data from other nodes. Distributed estimation is challenging because the information of a sensor measurement can arrive at a processing node over multiple information paths. This paper presents three approaches for addressing this challenge. Optimal distributed estimation by state estimate fusion avoids double counting of information by explicitly tracing the information paths of the estimates to be fused. In consensus-based estimation, the nodes iteratively exchange their estimates with their neighbors with the goal of converging to a global consensus. In distributed measurement fusion, a node broadcasts only new measurements to other nodes, who compute the optimal estimates using Kalman filters. Numerical results compare the performance and communication requirements of the three approaches.