Deep Reinforcement Learning Method for Energy Efficient Resource Allocation in Next Generation Wireless Networks

Abstract

The next generation wireless networks (NGWNs) of a base station (BS) with ultra-high dense user equipment's deployment are studied. To extend the coverage area and increase the spectrum efficiency and energy efficiency of the ultra-high dense wireless networks, we propose a downlink full duplex cooperative non-orthogonal multiple access (NOMA) with simultaneous wireless information and power transmission (SWIPT) communication (FCNS) mode in millimeter wave (mmWave) network. All the coverage users are divided into edge user group and near relay user group according to the mmWave channel state information. To optimize the energy efficiency resource allocation problem (NP-hard) in the ultra-dense network (UDN), we design a deep reinforcement learning framework to dynamically and efficiently to get the near-optimal result compared to the traditional centralized iteration algorithm and the global search algorithm. The simulation results demonstrated that the convergence speed of our proposed algorithm was more efficient and the near optimal results were acceptable.