Abstract
This paper addresses beam scheduling for tracking multiple smart targets in phased array radar networks, aiming to mitigate the performance degradation in previous myopic scheduling methods and enhance the tracking performance, which is measured by a discounted cost objective related to the tracking error covariance (TEC) of the targets. The scheduling problem is formulated as a restless multi-armed bandit problem, where each bandit process is associated with a target and its TEC states evolve with different transition rules for different actions, i.e., either the target is tracked or not. However, non-linear measurement functions necessitate the inclusion of dynamic state information for updating future multi-step TEC states. To compute the marginal productivity (MP) index, the unscented sampling method is employed to predict dynamic and TEC states. Consequently, an unscented sampling-based MP (US-MP) index policy is proposed for selecting targets to track at each time step, which can be applicable to large networks with a realistic number of targets. Numerical evidence presents that the bandit model with the scalar Kalman filter satisfies sufficient conditions for indexability based upon partial conservation laws and extensive simulations validate the effectiveness of the proposed US-MP policy in practical scenarios with TEC states.