Distributed fusion in underwater sensor networks: Fusing bearing information

Abstract

In underwater sensor networks, distributed data fusion may be more efficient than centralized fusion because the limited data transmission capacity can make it difficult to collect all required sensor information at a centralized fusion centre. In this paper, we investigate three distributed fusion techniques applied to a network of passive acoustic underwater sensor nodes. We focus on the process of having a node combining its own bearing-to-target information with bearing-to-target information received from another node. In one of the techniques, we approximate the uncertainty in crossfixes in Cartesian coordinates by a Gaussian distribution with their second-order statistics derived from an exact distribution. The bearings and covariance matrixes are fed into a Kalman filter for tracking. The other methods are a particle filter using an exact distribution, and a distributed particle filter using an approximate likelihood representation. The performance of the methods is investigated on simulated data as well as on real-world data collected by seafloor sensor nodes during a Stockholm Archipelago sea trial in the trilateral collaborative project DUSN (Distributed Underwater Sensor Networks) between Canada, Norway, and Sweden.