Rumor-robust distributed data fusion

We propose a novel Bayesian distributed data fusion methodology robust to the problem of rumor, i.e., of re-circulation of information accross the loops of a sensing & processing network. This problem is particularly important in mobile sensor networks where the communication graph is dynamically modified in an unpredictable manner. The approach proposed is based on the notion of Schur dominance, and looks for the less informative distribution that is more informative than the state of knowledge of both nodes participating in the fusion step, and that can result of factoring out common information from the nodes. The paper details construction of this dominating distribution for the case when the estimated entity takes values in a finite set, and relates the fusion operator proposed to existing rumor-robust methods, such as Covariance Intersection and a more recent approach based on the notion of Chernoff information. These methods are also revisited, and some of their intrinsic limitations are clearly exhibited.