Adversarial Multi-armed Bandit for mmWave Beam Alignment with One-Bit Feedback

Abstract

To exploit the large bandwidth available in the millimeter wave spectrum, highly directional beams need to be employed to compensate for the severe pathloss incurred at high frequencies. As a result, the beams of both the transmitter and the receiver must be constantly aligned. In this paper, the beam alignment (BA) problem is formulated as an adversarial multi-armed bandit (MAB) problem, yielding to a distributed BA search between the transmitter and receiver. First, we analyze the optimal codebook size for the BA that reduces the search space while insuring good performance levels. Then, we propose to use the exponential weights algorithm at both the transmitter and the receiver to match their beams. Our distributed algorithm relies on a single bit of feedback information and its performance is demonstrated via numerical results and compared with existing schemes.