RIS-Assisted Multi-Cell NOMA Downlink Communication Approach Based on Deep Reinforcement Learning

In the scenario of reconfigurable intelligent surface (RIS)-assisted multi-cell non-orthogonal multiple access (NOMA) communication, cell edge users (CEUs) face severe inter-cell interference (ICI), and existing coordinated multipoint (CoMP) transmission methods for mitigating ICI suffer from high system complexity. In order to reduce the system complexity and to mitigate the ICI, this paper proposes a RIS-assisted multi-cell NOMA downlink communication method based on deep reinforcement learning (RIS-MND-DRL). This method focuses on CEUs that are located in the coverage overlap region of multiple cells, and deploys RIS at the edges of two neighboring cells to assist these CEUs. Since both the action space and the state space are continuous spaces, RIS-MND-DRL uses deep deterministic policy gradient (DDPG) algorithm to optimize the phase shift matrix of RIS thereby reducing the ICI and maximizing the sum rate of CEUs. Simulation results show that RIS-MND-DRL can achieve faster and more stable convergence than Deep Q-Network (DQN) method, and its sum rate of CEUs is significantly improved when compared with DQN method.