A Fair Scheduling in 5G RAN Using Q-Learning

The emergence of resource-critical applications (e.g., autonomous vehicles, AR/VR) spurs the deformation of telecommunication networks like 5G/6G. Specifically, efficient radio resource allocation and management are the cornerstones of these applications, which we can achieve through smart spectrum scheduling. Reinforcement Learning (RL) takes the lead in this space due to the ability to dynamically adapt to the changes in environments without requiring data labelling or collecting a large volume of samples. Existing RL-based schedulings, however, fail to effectively incorporate key performance metrics in their learning environment for a trustworthy resource allocation and are also computationally intensive. Thus, we propose a Q-learning-based downlink scheduling scheme incorporating weighted reward to balance between multiple performance metrics without demanding significant resources. Extensive evaluations confirm that our algorithm can improve the cell throughput and packet delay by around 43% and 32%, respectively, compared to the state-of-the-art scheduler.