Performance Analysis of MIMO-NOMA system under varied channel conditions using Hybrid Attention-assisted Residual Deep Neural Network

In today’s fast-changing world of wireless communication, there’s a growing need for networks that are highly reliable, super fast, and able to connect a massive number of devices. To meet these demands, combined Non-Orthogonal Multiple Access (NOMA) with Multiple Input Multiple Output (MIMO) technology provides a powerful approach to increase data capacity and utilise energy more efficiently. However, despite its potential, MIMO-NOMA faces significant challenges, particularly when the wireless environment changes rapidly or the system setup becomes overly complex. To address these issues, our research presents a novel deep learning-based framework known as the Hybrid Attention-Assisted Residual Deep Neural Network (Hyb-AttRDNN). This intelligent system is designed to enhance both data speeds and energy efficiency by learning to allocate power more effectively across users. What makes Hyb-AttRDNN unique is its architecture—it combines different types of attention mechanisms (such as channel attention, cross-attention, and self-attention) with a technique called residual learning, allowing the model focusing most important features and avoid common training issues. The model’s main goal is to handle power allocation intelligently, and it utilises deep learning’s ability to recognise complex patterns to achieve this more effectively than traditional methods. Our simulations show that Hyb-AttRDNN delivers excellent results: a sum data rate of around 106, user rate of 34, and energy efficiency score of 32. These outcomes suggest that Hyb-AttRDNN could be a significant step forward in enhancing the efficiency, speed, and power management capabilities of MIMO-NOMA systems, even in challenging wireless conditions.