Autonomous Resource Orchestration for 6G Space-Air-Ground Networks: A Self- Supervised Learning Approach

Abstract The realization of sixth generation (6G) wireless networks heavily relies on incorporating space-air-ground integrated network (SAGIN) capabilities to realize global ubiquity. However, the inherent stochasticity of non-terrestrial network (NTN) node mobility, in conjunction with terrestrial resource variability, creates a multidimensional orchestration problem. Traditional optimization and supervised learning schemes are often prohibitive due to high computational complexity and the extreme cost of data labeling. This manuscript therefore presents a novel autonomous resource orchestration scheme with the application of self-supervised learning. By learning efficient spatial-temporal patterns, the scheme supports optimized power allocation together with beamforming for RIS-assisted wireless links. Extensive simulations for the application of the scheme confirm near-optimal sum-rate performance with 40% lower data requirements when compared with standard deep reinforcement learning schemes.