Dynamic access strategy in integrated satellite-terrestrial networks using attention-based DRL

As the sixth-generation (6G) mobile communication network evolves toward a satellite-terrestrial integrated architecture, low Earth orbit (LEO) satellite systems have become a crucial component of integrated satellite-terrestrial networks (ISTNs) due to their cost efficiency and wide-area coverage. However, in the context of spectrum sharing, ISTNs face significant challenges, including complex CCI and dynamically varying traffic demands. To address these issues, this paper proposes a satellite-terrestrial collaborative user access selection algorithm that combines reverse matching (RM) with deep reinforcement learning (DRL). By incorporating an interference threshold within the scheduling framework, the algorithm effectively decouples the strong interdependencies between satellite and terrestrial resources. On the satellite system, a Transformer-based model employing multi-head attention is introduced to capture the spatiotemporal correlations between beams and users. On the terrestrial system, a user access scheme based on RM is proposed, which adjusts the uplink-downlink spectrum association between satellite and terrestrial systems to transform interference from BSs to satellite users into inter-user interference, thereby enhancing terrestrial communication performance. Simulation results demonstrate that the proposed algorithm significantly outperforms benchmark schemes in terms of system throughput and user satisfaction.