Reinforcement Learning Based Technique for NOMA User Pairing Enhancement in RIS Assisted HetNets

Reconfigurable intelligent surface (RIS) as a promising technology has been proposed to increase spectrum efficiency by changing the weak communication environment into a strong one, giving the surrounding environment's dynamicity as a degree of freedom compared to the static environment in the absence of the RIS. However, most of the current work addressing RIS is not focusing on resource allocation (RA) schemes. In order to increase the system spectrum efficiency, this paper designed an RA scheme for an RIS-assisted HetNet with non-orthogonal multiple access (NOMA). In particular, an optimization problem is formulated to jointly optimize the user pairing (UP) and power allocation for each user using the Reinforcement Learning (RL) Double Deep Q-Network (DDQN) model to maximize the sum rates of all small cell users subject to a number of constraints such as guaranteeing a minimum rate for each user. Simulation results show the improved performance of the proposed RL-based UP technique compared to the conventional techniques.