AI-Enabled Radio Resource Allocation in 5G for URLLC and eMBB Users

Abstract

The fifth generation (5G) network is expected to accommodate heterogeneous traffic with diverse QoS demands. In this paper, we address the coexistence of Ultra-Reliable Low-Latency communications (URLLC) and enhanced Mobile Broad-Band (eMBB) users in 5G networks. We propose an AI-enabled approach that uses a reinforcement learning-based algorithm to balance the Key Performance Indicators (KPIs) of both URLLC and eMBB users. The proposed algorithm aims to jointly optimize both latency and reliability of URLLC users as well as the throughput of eMBB users. To achieve this, the algorithm utilizes the flexibility of the time-frequency grid of 5G standard to jointly perform power and resource block allocations to users. We compare our results with two baseline algorithms; a priority-based proportional fairness algorithm with fixed power allocation (PPF) that gives priority to URLLC users and a Q-learning algorithm (LR-Q) that performs joint power and resource allocation with the objective of improving reliability and latency performance of URLLC users only. Our results show that the proposed algorithm outperforms LR-Q by 29% increase and PPF by 21 times increase in throughput. Meanwhile, less than 0.5 ms degradation in URLLC's latency at the 10−4 percentile is observed, compared to both LR-Q and PPF.