Decentralized sensor selection based on the distributed posterior Cramér-Rao lower bound

The paper considers the problem of sensor resource management for distributed, nonlinear tracking applications with the objective of dynamically activating a time-variant subset of observation nodes to optimize the network's performance. The posterior Cramér-Rao lower bound (PCRLB) is a predictive benchmark of the tracker's achievable performance and has recently been proposed as a criteria for sensor selection. Existing PCRLB-based selection techniques are, however, primarily limited to centralized and hierarchical architectures, and when extended to decentralized topologies use approximate expressions [1] for computing the PCRLB. The paper addresses this gap and proposes the distributed PCRLB (dPCRLB) as the sensor selection criteria for decentralized networks without any need for central fusion. We derive an exact expression for computing the dPCRLB and a near-optimal implementation used with the distributed particle filter tracker. Our simulations verify the efficiency of the proposed dPCRLB based sensor selection approach.