This paper investigates the user pairing in a full-duplex (FD) communication system, aiming at maximizing the overall data rate of the system by reducing inter-user interference. The traditional user pairing methods usually suffer from high computational complexity and therefore are not suitable for practical implementation. Inspired by the recent innovation of deep reinforcement learning (DRL), we develop a low-complexity algorithm for the user pairing problem in FD networks. We first transform the problem into a Markov decision process (MDP) to facilitate the implementation of DRL. We then utilize the semi-supervised training paradigm to speed up the training process by adding some expert experience to the replay buffer. Finally, our proposal is extensively tested via numerical simulation, which demonstrates that the DRL-based user pairing algorithm can achieve a good performance with a significantly reduced computational complexity.
Deep Reinforcement Learning-Based User Pairing in Full-Duplex Communication Systems
2020-11-01
991281 byte
Conference paper
Electronic Resource
English