Deep Reinforcement Learning Aided Intelligent Access Control in Energy Harvesting Based WLAN.

The increasing number of low power communication devices in the era of Internet of Things (IoT) has resulted in challenges in maintenance and continuous energy supply. In this paper, the ambient energy harvesting is integrated with wireless local area network (WLAN), where low-power users adopt a distributed coordination strategy for their random access, namely ambient energy harvesting carrier-sense-multiple-access with collision avoidance (AEH-CSMA/CA). All users harvest ambient energy to power their data uploading. In order to increase the uplink throughput and to reduce the energy outage probability, a classic deep reinforcement learning (DRL) algorithm, namely deep Q-learning (DQL), is adopted by a smart user for intelligently adjusting its initial backoff window size during the backoff process. Simulation results demonstrate that the proposed DQL algorithm is capable of substantially improving the throughput performance, while gauranteeing the energy outage probability of the smart user lower than a certain threshold. [ABSTRACT FROM AUTHOR]