Hierarchical Beamforming in Random Access Channels

Managing a massive number of terminals in a contention-based multiple access is challenging due to its intrinsic limited efficiency. For example, in the random access channel considered in LTE-A and 5G NR, Base Station (BS) is just aware of the collided and non-collided preambles. Several time-based protocols have been investigated to redistribute the overload under high terminal activity, thus avoiding the congestion. In this work, we explore the use of the spatial domain by means of a hierarchical codebook-based beamforming, where the BS selects the appropriate beams as a function of the number of non-collided and collided preambles. Since the activity and placement of terminals may be dynamic over time, the sequential selection of parameters can benefit from a reinforcement learning (RL) framework. We propose an algorithm that can exploit both domains, temporal and spatial, with the goal of reducing collisions and enhancing transmission delay. Our approach is able to efficiently learn whenever there is a non-homogeneous spatial distribution of terminals and adapt the spatial beams accordingly.