LMA-SLN beamforming optimization for sum-rate maximization in intelligent reflecting surface assisted NOMA Systems

Abstract

Intelligent reflecting surfaces (IRS) can effectively improve the system performance of non-orthogonal multiple access (NOMA) systems. In this paper, we propose a Levenberg–Marquardt algorithm-based supervised learning network (LMA-SLN) to maximize the sum-rate of an IRS-assisted NOMA system. By decoupling the sum-rate maximization problem into the active and passive beamforming optimization sub-problems, we design an alternating optimization scheme to optimize the active and passive beamformings. Then, the LMA-SLN is trained with ideal channel state information (CSI) to obtain the optimized network parameters. Finally, the trained LMA-SLN is applied to optimize the active and passive beamformings without requiring CSI. The experimental results show that the proposed LMA-SLN scheme achieves the superior performance on improving the sum-rate.

Keywords

• Intelligent reflecting surfaces
• Non-orthogonal multiple access
• Levenberg–Marquardt algorithm
• Supervised learning networks
• Alternating optimization