Distributed Kalman filtering for sensor networks with random sensor activation, delays, and packet dropouts

This paper studies a distributed Kalman filtering problem for sensor networks, where sensor nodes may suffer from measuring the target state with a random activation nature and random delayed and lost state estimates of neighbour nodes due to unreliability of communication links. A distributed Kalman filter (DKF) is proposed, where predictor compensations for delayed and lost estimates of neighbour nodes and different consensus filter gains for state estimates of different neighbour nodes are used to improve estimation accuracy. Optimal filter gains with optimal parameters are designed to obtain a local minimum upper bound of filtering error covariance matrix, where optimal filter gains include an optimal Kalman filter gain for each sensor node and optimal multi-consensus filter gains for state estimates of its neighbour nodes. Our proposed DKF has a low computational cost because the calculation of cross-covariance matrices between estimates of sensor nodes is avoided. Besides, the boundedness of the proposed DKF is analysed. Finally, an example of a target tracking system in sensor networks demonstrates effectiveness of the proposed DKF.

Automated summary

This study introduces a distributed Kalman filter (DKF) for sensor networks to improve estimation accuracy by compensating for delayed and lost estimates of neighbour nodes, and using different filter gains. The DKF has a low computational cost and its boundedness has been analyzed for effectiveness in a target tracking system.