Hierarchical Intelligent Routing Under Multi-Layer Mega-Constellation Network Management Architecture

The convergence of revolutionary communication technologies and exponentially growing demand for ubiquitous connectivity has propelled low earth orbit (LEO) mega-constellation networks (MCNs) to critical infrastructure status. Addressing the dimensional scalability challenges inherent in ultra-dense topologies, multi-layer management architectures now constitute a foundational mitigation framework. However, the provision of high-quality services continues to be hindered by conventional routing schemes that lack adaptability to ultra-dense topologies, highly dynamic behavior, and cross-layer heterogeneity. To address these limitations, we propose a hierarchical deep reinforcement learning (HDRL)-driven routing mechanism that inherently aligns with the multi-layer architecture of MCNs. The proposed approach allocates specialized policies to each layer, enabling it to learn and optimize routing decisions at multiple levels of abstraction. This design decomposes complex routing tasks into manageable subtasks while fostering intelligent inter-layer coordination, thereby establishing a scalable and adaptive routing solution. Simulation results demonstrate that our proposed mechanism exhibits superior scalability and adaptability compared to existing approaches in MCNs. Furthermore, prospective future directions are discussed to guide and inspire meaningful research.