Title: Deep Reinforcement Learning-based Radio Resource Allocation and Beam Management under Location Uncertainty in 5G mmWave Networks

Abstract: Millimeter Wave (mmWave) is an important part of 5G new radio (NR), in which highly directional beams are adapted to compensate for the substantial propagation loss based on UE locations. However, the location information may have some errors such as GPS errors. In any case, some uncertainty, and localization error is unavoidable in most settings. Applying these distorted locations for clustering will increase the error of beam management. Meanwhile, the traffic demand may change dynamically in the wireless environment. Therefore, a scheme that can handle both the uncertainty of localization and dynamic radio resource allocation is needed. In this paper, we propose a UK-means-based clustering and deep reinforcement learning-based resource allocation algorithm (UK-DRL) for radio resource allocation and beam management in 5G mmWave networks. We first apply UK-means as the clustering algorithm to mitigate the localization uncertainty, then deep reinforcement learning (DRL) is adopted to dynamically allocate radio resources. Finally, we compare the UK-DRL with K-means-based clustering and DRL-based resource allocation algorithm (K-DRL), the simulations show that our proposed UK-DRL-based method achieves 150% higher throughput and 61.5% lower delay compared with K-DRL when traffic load is 4Mbps.