ISN is deeply integrating AI with communications. Addressing the bottleneck of scenario-dependent training and weak cross-scenario generalization in existing algorithms,the laboratory has unveiled the world's first topology-aware wireless resource management foundation model — Yuanshu (Origin Hub).Distilling six space-air-ground topologies, from star to mesh, and integrating interference characteristics with spatiotemporal service patterns, the model generalizes efficiently across eight typical scenarios — terrestrial heterogeneous, low-altitude 3D, satellite-terrestrial converged and emergency networks — delivering a new paradigm for intelligent wireless resource management.

ISN全国重点实验室锚定“AI+通信“深度融合，针对现有无线资源管控算法高度依赖特定训练场景、跨场景泛化能力薄弱的瓶颈，实验室面向全球首发拓扑感知无线资源管控元枢大模型。模型归纳星形、树状、网状等6类空天地组网拓扑，融合网络干扰特征与业务时空分布规律，实现对地面异构组网、低空立体组网、星地融合组网、灾害应急组网等8类典型空天地一体化场景的高效迁移泛化，为空天地多场景无线网络智能资源管控提供全新解决方案。

Model Parameters

Backbone: Qwen2.5-1.5B-Instruct

Dual topology graph encoder: hidden dimension 256, 2 layers, 12 attention heads

Topology-guided bias: injected into the first 4 layers

Task adapter: bottleneck dimension 384

Training dataset: 93,785 parameters

Model repository: Click to access