A Graph Neural Networks Approach to Heterogeneous Link Switching in SATIN

Space-Aerial-Terrestrial Integrated Networks (SATIN) establish a transformative paradigm for enhancing emergency response systems through multimodal disaster data acquisition and dynamic resource orchestration during catastrophic events. The transition to 6G communication infrastructure intensifies requirements for network heterogeneity and ultra-massive connectivity, while introducing critical challenges in service allocation for terrestrial users. To resolve this dichotomy, a novel satellite-assisted emergency communication architecture is proposed, where LEO satellite constellations provide resilient backhaul connectivity to support unmanned aerial vehicle networks. The core technical challenge involves formulating the link selection mechanism between user equipment and UAV nodes as a constrained nonlinear integer programming problem with backhaul capacity limitations as primary constraints. A graph neural networks-driven heterogeneous link switching(GNN-HLS) mechanism is developed to tackle this combinatorial optimization challenge. In comparison with traditional methods, the GNN-HLS mechanism effectively optimizes the user access rate with less impact on UAV energy consumption and link duration.