A Hierarchical Decoupling Framework for Non-Stationary Vehicular Access in Space-Terrestrial Integrated Networks

Abstract

Integrating Low Earth Orbit (LEO) satellites with terrestrial Road Side Units (RSUs) into Space-Terrestrial Integrated Networks (STIN) offers a promising architecture for continuous cooperative vehicular access. However, realizing efficient multi-vehicle coordination in such heterogeneous environments is hindered by the complex spatiotemporal coupling of distributed decisions under long-term Quality-of-Service (QoS) constraints, alongside the severe channel non-stationarity induced by the dual mobility of vehicles and satellites. To address these issues, we propose a hierarchical spatiotemporal decoupling framework that synergistically integrates Lyapunov optimization to enforce long-term time-averaged QoS requirements, Mean-Field Game (MFG) approximation to resolve high-dimensional spatial competition, and dual-timescale Contextual Thompson Sampling (CTS) to navigate unpredictable channel dynamics. Extensive simulations validate the effectiveness of our proposed framework, demonstrating significant outperformance in scheduling efficiency, latency reliability, and robust long-term constraint satisfaction.