Distributed Weighted Fusion Estimators with Random Delays and Packet Dropping

Abstract

This paper is concerned with the distributed fusion estimation in sensor networks where local estimates are sent to a fusion centre for fusion estimation, with random delays and packet dropouts. Under the linear minimum variance sense, a distributed optimal weighted fusion estimator is given for discrete time-invariant stochastic linear systems with multiple distributed sensors. The algorithm involves a weighted fusion of local predictors with different prediction steps from different sensor sources. A recursive computation of the cross-covariance matrix of prediction errors between any two local estimates is derived. We present two fusion strategies. One is to fuse the latest local estimates that reach the fusion centre at the current time. The other is to fuse the latest local estimates that reach the fusion centre at the current time and the predicted estimates of those that do not have estimates received at the current time. Further, to reduce the computation cost, only the local estimates satisfying the given precision requirement are fused because those with longer delays or consecutive packet dropouts have large estimation errors. A strategy to select local estimators for fusion is presented based on gate thresholds of time delays or the numbers of consecutive packet dropouts for all local estimators. This method can be implemented offline. Simulation for a tracking system with four sensors shows the effectiveness of the proposed approaches.