On Performance of Deep Reinforcement Learning-based Listen-Before-Talk (LBT) Scheme

Licensed assisted access (LAA) is a promising system to overcome the limited radio resource by sharing the unlicensed band with wireless local area networks (WLANs), and the listen-before-talk (LBT) scheme is a key technology for providing fairness between LAA and WLAN. Recently, deep reinforcement learning (DRL) has been investigated to improve the performance of LBT; however, such approaches assume that there is no processing delay and thus the optimal decision can be immediately done. In this paper, we evaluate the performance of the DRL-based LBT (DRL-LBT) scheme when different processing delays are considered for DRL. Evaluation results demonstrate that the throughput fairness index and the total throughput of DRL-LBT with the processing delay can be degraded up to by 9.4% and 10.0%, respectively, compared with an ideal case without any processing delay.