Optimization Satellite Power Control in Integrated Non-Terrestrial Networks Using Multi-Agent Deep Reinforcement Learning

Abstract

High altitude platform station (HAPS) and low Earth orbit (LEO) satellites that provide services to the ground in the line of sight (LoS) environment are gaining attention as next-generation communication networks. Additionally, integrating HAPS and satellite networks is being researched to provide high-quality communication services to terrestrial environments. However, a critical issue in integrated networks is the increased interference, which leads to degraded signal quality for users and a higher outage probability when the same frequency is used. In this paper, we propose a method to reduce outage probability for users being served by the HAPS by adjusting the transmission power of the satellite using deep reinforcement learning (DRL). This paper uses a multi-agent deep Q-network (MADQN) and an attention-based A-MADQN, which combines MADQN with an attention mechanism. By employing reinforcement learning, the optimal transmission power can be determined. The simulations illustrate confirmed a reduction in user outage probability compared to the conventional satellite transmission power for each cell.