Adaptive and Cooperative Resource Scheduling for Satellite-Terrestrial Networks

Abstract

Satellite-terrestrial networks (STNs) consisting of satellite segment and ground segment have been regarded as a desirable solution for 6G. Efficient cooperative resource scheduling strategies, which cover the cooperation in satellite segment for data relay and the cooperation between satellite segment and ground segment for data downloading, play a pivotal role in enhancing the system performance in STNs. Since the dynamic channel condition and energy feeding greatly influence the network status, cooperative resource scheduling should be adaptive to the future environmental fluctuation. In this paper, we model the cooperative resource scheduling problem in STNs as a resource limited Markov Decision Process (MDP). Considering the fact that satellites are unaware of future environmental status, the traditional static optimization solution is infeasible. Therefore, we propose a Deep Reinforcement Learning (DRL) based Cooperative Store-and-Relay Resource Scheduling Algorithm (CSR-RSA), where inter-satellite links are utilized to coordinate with intermittent satellite-ground links for improving the transmission performance of the network. By exploiting the proposed CSR-RSA, the well-trained neural networks can be obtained to generate the adaptive and cooperative resource scheduling strategy without the knowledge of future environmental status. Simulation results verify the effectiveness of the proposed algorithm compared with traditional algorithms.