DRL-Based Dynamic MAC Scheduler Reconfiguration in O-RAN

This paper introduces reconfiguration mechanisms for a 5th Generation (5G) Medium Access Control (MAC) scheduler, which uses Deep Reinforcement Learning (DRL) and is deployed as an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$xApp$</tex> within the RAN Intelligent Controller (RIC) framework. The objective is to optimize the allocation of radio resources in 5G networks, seeking to meet QoS requirements while minimizing resource consumption. To this end, we propose the dynamic adjustment of configurable parameters in a Lyapunov-based scheduler, which addresses the challenges posed by the highly dynamic network environment, and the need to meet multiple Quality of Service (QoS) objectives. Exploiting the adaptability of DRL, our solution can effectively respond to fluctuating network conditions, thereby continuously enhancing scheduling decisions in real time. The proposal is evaluated through comprehensive simulations conducted over ns3 5G-LENA, which demonstrate notable improvements in QoS performance and resource efficiency. The observed results evince the great potential of exploiting DRL to optimize the MAC scheduling in modern wireless communication systems, offering a scalable and adaptive solution for 5G and future 6G networks.