Universal Dynamic Pilot Allocation for Beam Alignment Based on Multi-Armed Bandits

Multi-armed bandit (MAB) has been widely used to address beam alignment in millimeter-wave communications. However, existing works focus only on identifying the best beam and do not consider a system performance (i.e., effective data rate). To facilitate this, this letter proposes a dynamic pilot allocation framework that is universally applicable to MAB-based beam alignment regardless of the MAB's theoretical basis. In the framework, each pilot is dynamically allocated considering both expected improvement and degradation in performance resulting from additional pilot allocation. Furthermore, theoretical analysis supports the asymptotic optimality of the framework. Simulation results demonstrate that the proposed framework significantly improves the performance of a variety of MAB-based beam alignment methods.