Cellular base station downlink power allocation using model-free reinforcement learning

Abstract

With the maturity of 5G and IoT technologies, the number of base stations(BS) and end-user equipment(UE) is expected to increase dramatically. Therefore, the problem of an optimal solution for complex resource and power allocation in cellular networks has become a research topic. This is especially under the tremendous pressure of increasing demand. The issue that quality of service(QoS) is getting tougher to satisfy the user needs. Hence, in this dissertation, the author using reinforcement learning(RL) techniques to optimize the power allocation of Base station downlink, further improve the throughput and reduce the inter-cell interference is the topic of this dissertation. Our results examine two mainstream approaches in RL i.e. Q-learning and Deep Q Network(DQN) will be used to model the system. Since the DQN is a data-driven and model-free RL technique, it can be more consistent with the issue of BS power allocation.