Quantum Multi-Agent Reinforcement Learning is All You Need: Coordinated Global Access in Integrated TN/NTN Cube-Satellite Networks
Soohyun Park, Gyu Seon Kim, Zhu Han, Joongheon Kim
Abstract
This article addresses novel quantum multi-agent reinforcement learning (QMARL)-based scheduling for integrated terrestrial ground-stations and large-scale non-terrestrial cube-satellites networks to enable coordinated global access services. By utilizing quantum-based neural networks for designing QMARL, stable large-scale cube-satellite scheduling can be realized thanks to fast training with quantum-specific learning computation methods. In addition to the benefit from conventional QMARL algorithms, our proposed QMARL-based scheduler has two key characteristics: high-scalability for low-dimensional scheduling, and energy-efficient geometry-awareness. For high-scalability, projection value measure (PVM)-based measurement is proposed for reducing scheduling action dimensions into a logarithmic-scale, which is obviously helpful for stable and fast convergence during QMARL training. Furthermore, energy-efficient geometry-aware operations can be realized by considering the fair energy consumption and positions in cube-satellites, which can be used for differentiated charging strategies. Our evaluation results verify that the proposed low-dimensional geometry-aware QMARL-based scheduler achieves desired performance improvements.