Machine Learning-Based Beam Alignment in mmWave Networks

Abstract

In this chapter, we discuss the beam alignment (BA) problem in mmWave networks. We first formulate the BA problem as a stochastic multi-armed bandit problem, with the aim of maximizing the cumulative received signal strength in a certain period. In order to accelerate the BA process, we develop a learning algorithm named hierarchical beam alignment (HBA) algorithm. This algorithm exploits the correlation structure among beams such that the information from neighboring beams can be harnessed to find the optimal beam, instead of exhaustively searching the entire beam space. In addition, the prior knowledge on channel dynamics is incorporated in the HBA algorithm to reduce the BA latency. Theoretical analysis proves that the proposed algorithm asymptotically approaches the optimal solution. Extensive simulation results show that the proposed HBA algorithm can successfully find the optimal beam with a high probability. Meanwhile, compared to the existing BA method in IEEE 802.11ad, the proposed HBA algorithm reduces the BA latency from hundreds of milliseconds to a few milliseconds in the case of multipath channel.