Reliable Beam Tracking on High-Altitude Platform for Millimeter Wave High-Speed Railway

Due to the dominant line-of-sight (LOS) component of high-altitude platform (HAP) communications, tracking the high-speed railway (HSR) uplink channel commonly involves forming received beam based on the estimated direction-of-arrival (DOA) of the LOS path. Although DOA estimation accuracy is increased by using finer angular resolution, DOA-based beam tracking is not computationally efficient when a large number of antenna elements is involved, as typically encountered in mmWave communications. Moreover, maintaining link reliability becomes very challenging since beam misalignments arise frequently due to the high-speed user. In this paper, we propose a learning-based framework aiming to achieve reliable connectivity for HSR data transmission. The framework utilizes HAP infrastructure equipped with a uniform planar array (UPA) antenna configuration. A multi-armed bandit (MAB) beam tracking, based on the existing modified exponential weight (EXP3) algorithm, is implemented between two initial access (IA) phases of the fifth generation (5G) radio frame on the HAP. The MAB realizes online learning where the channel dynamics are learned over time and the best serving beam is decided sequentially. The proposed scheme results in outage probability (OP) approaching that of the DOA-based scheme but with complexity which only scales linearly with the codebook size and independent on the number of antenna elements. Even without the resource and computational burden associated with offline training tasks inherent in a neural network (NN), the OP of the proposed scheme increases by merely 1.2%. We also reveal that the proposed scheme prolongs the interval between two IA phases, yet with negligible OP deterioration, translating to more room available for data transmission.