Consensus-based distributed dynamic sensor selection in decentralised sensor networks using the posterior Cramér–Rao lower bound

Abstract

The paper considers the problem of dynamic sensor scheduling for non-linear tracking problems in distributed sensor/agent networks (AN/SN), where channel limitations restrict how many sensors can simultaneously participate in the estimation mechanism. Commonly referred to as sensor selection, the basic objective is to select a subset of sensors from available sensors that minimises the estimation error. The posterior Cramer–Rao lower bound (PCRLB) has recently been proposed as an effective sensor selection criteria. Nevertheless, existing PCRLB-based selectors are limited to centralised and hierarchical networks, and when extended to distributed architectures use approximate expressions. First, the non-conditional distributed PCRLB (dPCRLB) that considers observations and state variables to be random is used to derive the non-conditional sensor-selector. The dPCRLB expressions we derive are optimal. Unlike the non-conditional PCRLB, its conditional counterpart is a function of the past history of observations and is a more accurate representation of the system׳s performance. Second, the paper generalises the non-conditional dPCRLB sensor-selector to its conditional dPCRLB version. Both dPCRLB sensor-selectors use raw observations adding significant communication overhead. Third, the paper extends the conditional dPCRLB framework to quantised observations and develops the quantised version of the conditional dPCRLB sensor-selector. Numerical simulations verify the efficiency of our distributed dynamic sensor-selectors.