Hypergraph Attention Network for Spatio-Temporal Interference Modeling in Cellular Networks

Interference is an indelible dominant feature of wireless cellular networks, which has heavy impact on the transmission quality and network capacity. While much effort has been put on the construction of radio maps that intend to characterize signal distribution with complex channel propagation models, network interference modeling remains largely overlooked, leading to incomplete and inaccurate performance evaluation in network planning and optimization. In this letter, we propose an innovative data-driven approach to characterize the spatio-temporal interference pattern, or statistical interference map, of a multi-cell network, without knowing the exact complex underlying propagation environment and traffic pattern. Specifically, a hypergraph attention network (HGAT) is designed, in which the representation of nodes and edges are inherently aligned with the overall structure and operation of a wireless network, capturing the intricate dependencies between base stations, users, radio resource blocks and the received interferences. By leveraging spatio-temporal convolution, HGAT effectively encodes network dynamics and interference characteristics, enabling accurate and robust statistical prediction. Experimental results demonstrate that our approach surpasses conventional methods in accuracy, efficiency, and scalability, marking a significant step toward intelligent and efficient wireless network planning and optimization.