Event-Triggered Multi-Sensor Scheduling for Remote State Estimation Over Packet-Dropping Networks

We study the multi-sensor remote state estimation problem over packet-dropping networks and employ a stochastic event-triggered scheduler to conserve energy and bandwidth. Due to packet drops, the Gaussian property of the system state, commonly used in the literature, no longer holds. We prove that the state instead follows a Gaussian mixture (GM) model and develop the corresponding (optimal) minimum mean-squared error (MMSE) estimator. To tackle the exponential complexity of the optimal estimator, the optimal Gaussian approximate (OGA) estimator and its heuristic GM extension are further derived. Our simulations show that the approximate estimators perform similarly to the optimal estimator with significantly reduced computation time. Furthermore, our proposed scheduler outperforms standard event-triggered schedulers in a target-tracking scenario.