Energy-Efficient D2D Communications Based on Centralised Reinforcement Learning Techniques

Abstract

Device-to-Device (D2D) communication has emerged as an evolving communication technology in 5G networks, enabling a pair of user equipment units to communicate without passing through the base station. However, the introduction of a D2D link can cause interference with other cellular user links, which highlights the difficulty of guaranteeing the communication quality of the whole system. In addition, when a large number of cellular users are connected to the network through D2D devices at the same time, the circuit consumption of the mobile devices will greatly increase and affect the user experience. In this paper, we focus on improving the energy efficiency of D2D devices in a cellular network served by one base station, through the adjustment of D2D link transmission power. We propose a centralised power control algorithm based on reinforcement learning to optimise the energy utilisation, while minimising the interference on cellular users, to maintain the quality of service (QoS). Simulation results show that the proposed approach can significantly increase the system energy efficiency and maintain the cellular user QoS, compared with the benchmark algorithm.