Resource allocation for multiple RISs assisted NOMA empowered D2D communication: A MAMP-DQN approach

Abstract

In this paper, we investigate the non-orthogonal multiple access (NOMA) empowered D2D communications system, where multiple reconfigurable intelligent surfaces (RISs) are deployed to enhance the service quality for mobile D2D groups. Meanwhile, the RISs can further improve the NOMA gains with artificially controlled channel conditions. Although multiple distributed deployment RISs can enhance the coverage for the multiple mobile D2D groups. However, the joint reflective capabilities of multiple RISs require a more complex phase shift design. Therefore, to maximize the average sum data rate of all D2D groups, a multi-agent multi-pass deep Q-network (MAMP-DQN) based framework is invoked to optimize the sub-channel assignment and power allocation of mobile D2D groups, as well as the phase shifts adjustment policy. Specifically, the actions in the formulated Markov game (MG) are modeled in the hybrid discrete–continuous structure, and the MP-DQN algorithm is invoked to deal with the hybrid action. Our simulation results indicate that the multiple RISs can significantly improve the average sum data rate of D2D groups. Moreover, the NOMA-empowered D2D groups are capable of achieving a gain of 27% in contrast with the OMA scheme in terms of the sum data rate. Finally, the sum data rate of D2D groups can be further enhanced by increasing the number of the RISs or the REs.