Decentralized information filter with non-common states, and application to sensor bias estimation

This paper addresses the problem of estimation in a sensor network with noncommon states. The problem of non-common states occurs in online sensor bias estimation. This problem is addressed by dividing the state vector at each node into two sets. The set of common states is observed by each node, while that of non-common states is node-specific. Each node has an independent Kalman filter to estimate the states associated with it, and communicates information associated with the common states to the connected nodes. The decentralized architecture based on information filter is modified to assimilate the estimated states from each node. Simulation results for a fully and strongly connected network of 20 nodes are presented to validate the proposed algorithm. The algorithm is shown to be as good as centralized Kalman filter with added advantages of distributed computation and robust architecture over centralized filter.