Multi-Armed Bandit Beam Alignment and Tracking for Mobile Millimeter Wave Communications

We propose a novel beam alignment and tracking algorithm for time-varying millimeter wave channels with a dynamic channel support. Millimeter wave beam alignment is challenging due to the expected large number of antennas. A multi-armed bandit training beam selection policy is used to balance exploration of the set of feasible beams. We track the channel using a synthesis of sparse Bayesian learning and Kalman filtering and smoothing. Results show our algorithm has a more rapid rate of initial beam alignment compared to other beam selection policies and, for dynamic channel support, long-term beamforming gain commensurate to omni-directional training.