Adaptive Resource Allocation in 5G/6G Networks Using Deep Reinforcement Learning

Abstract

The 5G to the newly developed 6G conditions have brought about a highly dynamic heterogeneous traffic environment that is marked by a very dense user deployment, wide range of quality-of-service (QoS) needs, and very rapidly changing channels. Existing methods of resource allocation through convex optimization, heuristic scheduling and hard-coded rule-based mechanisms tend to fail efficiently in such non-stationary network dynamics resulting in suboptimal spectral utilisation and higher latency. In an effort to address these shortcomings, the paper will present a deep reinforcement learning (DRL) based adaptive resource allocation model that could learn the optimal power and spectrum allocation policies by being interacted with the wireless environment continuously. The allocation problem is modelled as a Markov Decision Process (MDP) in which the system state includes channel gains, the level of interference and traffic queue conditions; the action space includes dynamically allocated resources actions; and the reward maximisation includes simultaneously throughput, latency and energy efficiency. The configuration of an actorcriticbased DRL is structured in such a way that there is stable convergence and scalability creation within realistic compromises of 5G/6G systems. Based on large scale simulation efforts also reveal the proposed framework has attained as high as 22 percent reduction in aggregate throughput, 26 percent cut in mean latency and 17 percent reduction in energy consumption rate over the traditional allocation schemes. Such results confirm that DRL-based adaptive resource management is effective in next-generation wireless networks and has the potential to be used in intelligent and scalable deployments of 6G.