Whittle Index Policy for Dynamic Multichannel Allocation in Remote State Estimation

Abstract

In this paper, we consider dynamic channel allocation for remote state estimation of multiagent systems. For each subsystem, a sensor measures its state and transmits the data via a packet-dropping channel, which is dynamically allocated by the remote estimator. We first formulate the problem as a Markov decision process. Given the difficulty of obtaining an optimal policy of large-scale problems, we develop a suboptimal heuristic policy based on the Whittle index for the restless multiarmed bandit (RMAB) problem. The performance of the Whittle index policy is evaluated from both theoretical and practical aspects. The strong performance of Whittle index policy is illustrated by the numerical examples.