Deep Reinforcement Learning Based Throughput Maximization Scheme for D2D Users Underlaying NOMA-Enabled Cellular Network

Abstract

Device-to-Device (D2D) communication is a potential technology that efficiently reuses spectrum resources with CMUs in a fifth-generation (5G) underlay and even beyond the network. It improves network capacity and spectral efficiency at the cost of co-channel interference. Moreover, massive connectivity has not been fully exploited for efficient spectral efficiency usage in the existing solutions. To resolve the aforementioned issues, we combine non-orthogonal multiple access (NOMA) approaches with cellular mobile users (CMUs) in order to improve their throughput while preserving the signal-to-interference noise ratio (SINR) offered by CMUs and D2D mobile pairs (DMPs). The problem of power allocation is formulated as mixed-integer non-linear programming, which is then transformed to machine learning using the markov decision process (MDP). Then, a deep reinforcement learning (DRL) approach is proposed for solving the continuous optimisation problem in a centralised fashion. Furthermore, to achieve better performance and a faster convergence rate, the higher proximal policy optimization (PPO) scheme is employed. Numerical results reveal that the proposed algorithm outperformed state-of-the-art schemes in terms of throughput.