Research on grouping strategy for NOMA downlink based on Pointer Network

Abstract

With the application of Non-Orthogonal Multiple Access (NOMA) technology in 5G and beyond communication systems, how to effectively group users to optimize power allocation has become a key challenge. This paper proposes a user grouping method based on a Pointer Network, which efficiently extracts user location information through embedding layers, encoder–decoder structures, and attention mechanisms, achieving the goal of precise grouping decisions and power optimization. The embedding layer maps users’ two-dimensional coordinates into a high-dimensional space, enhancing the model’s spatial awareness. The encoder–decoder structure, combined with Long Short-Term Memory (LSTM) networks and attention mechanisms, captures the spatiotemporal dependencies between users and dynamically selects the optimal path during the grouping process. Experimental results show that when users are located 300 meters from the base station, the recognition accuracy of a 4-user grouping reaches 94.85%, and that of a 6-user grouping reaches 89.3%. The method also demonstrates strong robustness under multipath fading channels and low signal-to-noise ratio conditions. Compared to random grouping methods, the proposed grouping strategy exhibits better adaptability and scalability in complex communication environments, significantly reducing power consumption, and providing new technical support for resource allocation and energy management in NOMA systems.