Target localization in sensor networks: Consensus, but not so much

After measuring the distance from a target, the nodes of a wireless sensor network attempt to improve their measurements by implementing a gossip procedure. However, gossip algorithms eventually end up with full agreement about a common value that is useless for target location. Therefore, we propose a partial consensus scheme in which the number of gossip steps are carefully calibrated with the goal of maintaining a sufficient degree of diversity among nodes, while still refining on the original measurements. This way, we can approximately reconstruct the likelihood profile through the network, which is effective for the estimation task. Indeed, the proposed estimator, based on a partial agreement among the sensors, is shown to be asymptotically normal with variance close to its theoretical lower bound.