Joint Power Control and User Assignment in RIS-based NOMA: A Multi-kernel Neural Network Approach
Résumé
As mobile devices and data usage continue to grow rapidly, wireless communication systems are being pushed to meet more stringent demands for ultra-low latency, high reliability, and massive connectivity. However, conventional resource allocation algorithms are increasingly unable to cope with the growing problem dimensionality and the rising complexity of the objective functions. These approaches struggle to meet the scalability demands and adapt to the dynamic nature of modern communication environments, highlighting the need for more advanced solutions, such as machine learning-based techniques. While traditional Artificial Intelligence (AI)-driven methods show promise, they often face challenges, including reliance on labeled data and difficulties in generalizing across diverse scenarios. To address these limitations, we propose a novel supervised learning framework for wireless networks that utilizes pre-trained models to tackle scalability issues. Our approach leverages a pre-trained model optimized for a relatively simple scenario and extends its patterns to a generalized scenario using a multi-kernel neural network architecture. The proposed system is applied to achieve joint power control and user assignment for a RIS-based non-orthogonal multiple access (NOMA) systems. We formally establish that the estimation error is upper bounded and does not scale with the size or dimensionality of the generalized network architecture. Simulation results demonstrate the effectiveness of our method, achieving scalable and efficient wireless network optimization.