Machine Learning Empowered Resource Allocation to Reconfigurable Intelligent Surface Supported MIMO-NOMA Networks

In this chapter, we have presented a study on using machine learning to optimize resource allocation in future communication systems that incorporate reconfigurable intelligent surfaces (RIS) and support multiple-input multiple-output (MIMO) and non-orthogonal multiple access (NOMA). The main objective of our work is to improve the quality of services in these systems by maximizing the sum rate. Our focus was on optimizing the transmission power vectors of the base station and the power vectors of the RIS reflector for efficient resource allocation. We achieved this by employing a first-order Taylor approximation to converge to the optimal values of these vectors. To address the non-convex nature of the problem, we developed an iterative algorithm based on the sub-gradient method. This algorithm utilized lagrange dual analysis and applied the Taylor approximation to convert the dual variables into the difference of two convex (DC) functions. For the optimal phase shifts of the RIS-supported MIMO-NOMA systems, we utilized a deep learning approach. Specifically, we employed a deep neural network fitting function to ensure power fairness among the MIMO-NOMA users. The deep learning algorithm introduced a loss function to solve the optimization problem. To evaluate the effectiveness of our proposed scheme, we conducted simulations using MATLAB programming with the CVX toolbox, which is a MATLAB-based system for convex optimization. The results demonstrated the improved ergodic capacity, spectral efficiency, low outage probability, low decoding error probability, and low computational complexity achieved by our approach. Finally, this work utilizes machine learning techniques to optimize resource allocation in RIS-supported MIMO-NOMA networks. The proposed scheme shows promising results in terms of performance metrics and computational efficiency.