A Deep Reinforcement Learning-Empowered Channel Allocation for Intelligent Wireless Networks

The wireless networking community is paying more attention to reinforcement learning considering its ability to discover high-performing configurations from the observed outcomes. This paper offers a stateless Q-learning variant to take edge of spatial utilization in intelligent wireless networks (IWNs). We specifically enable IWNs to alter both the channel they utilize and their transmission power entirely depending on the throughput they experienced. We emphasize on a decentralized scenario in which learners are unable to collect data regarding neighboring nodes. The simulation results demonstrate that while the algorithm can identify the optimum action to improve aggregate throughput, as the throughput experienced by individual networks varies widely. Also, it notices that this variability due to the hostile nature of the configuration, where the performance acts sporadically well or inadequately depending on the decisions made by neighboring node. The proposed algorithm achieves high throughput of 80.29% and achieves high proportional fairness of 81.61% which is 1.32% more than that of throughput achieved when compared to ideal scenarios. Finally, this study discusses the impact of the intrinsic learning parameters of our proposed algorithm on this variations.