A Weightedly Uniform Detectability for Sensor Networks

In this brief, we study the detectability issues in the context of distributed state estimation problems for a class of locally undetectable sensor networks. First, we introduce a novel detectability condition, i.e.,
weightedly uniform detectability
(WUD), which is a sufficient condition to prove that the error covariances of the consensus filtering are uniformly bounded even though the local sensor nodes are undetectable. Different from the existing detectability (or observability) conditions, our condition includes the interacting weights which could further optimize the lower detectability Gramian bound. Hence, a new weights selection method is derived in term of the criterion of WUD. This new rule of selecting weights provides a new framework for distributed state estimation. The advantages of this approach lead to a better performance in estimation without extra computational burden to the filtering process. Finally, an example shows the effectiveness of the proposed method.