Reconfigurable Intelligent Surface-Aided Cognitive NOMA Networks: Performance Analysis and Deep Learning Evaluation

Abstract

This paper investigates reconfigurable intelligent surface (RIS)-aided cognitive non-orthogonal multiple access (NOMA) systems, where an RIS is deployed to serve two users under multi-primary users' constraints. Our analysis assumes imperfect channel state information and successive interference cancellation under scenarios with and without line-of-sight (LoS) link between source and users. We derive exact closed-form expressions for the outage probability, throughput, and an upper bound for the ergodic capacity (EC). To provide further insights, an asymptotic analysis is carried out by considering two power settings at the source. It is also determined the optimal data rate factors of all users that maximize the system throughput. In addition, a deep learning framework (DLF) for EC prediction is designed. Numerical results show that: i) compared to the system without LoS link, the performance of the proposed system with LoS link can significantly improve when the number of reflecting elements at the RIS increases, and ii) the proposed system has superior performance compared to its orthogonal multiple access counterpart. Furthermore, our proposed DLF exhibits the lowest root-mean-square error and low execution-time among other approaches, verifying the effectiveness of this method for future analysis.