Deep Reinforcement Learning for the Coexistence of LAA-LTE and WiFi Systems

As a promising technique to handle the conflict between the explosive mobile traffic and scare spectrum resource, license-assisted access (LAA) has been proposed to operate the LTE network on the unlicensed band. This paper considers the LAA-LTE system coexisting with an unsaturated WiFi system. Specifically, deep reinforcement learning (DRL) is adopted to enable the LAA-LTE system to learn the traffic pattern of the WiFi system and adaptively optimize its transmission time in each frame. Different from conventional coexistence schemes, which require massive signaling exchanges between the two systems to achieve fairness, the proposed DRL-based algorithm can maximize the spectrum usage while protecting the WiFi system without such signaling requirements. Simulation results demonstrate that the proposed scheme can achieve almost the same LAA-LTE throughput and protection to the WiFi system of the genie-aided exhaustive search algorithm, which has high complexity and requires to know the WiFi information perfectly.