A Survey of AI-Based Resource Management and QoS Modeling in 6G Space-Air-Ground Integrated Networks: A Three-Axis Taxonomy

Abstract

6G targets ultra-wide coverage together with ultra-low-latency and ultra-reliable services. To this end, Space-Air-Ground Integrated Networks (SAGINs), which integrate non-terrestrial networks (NTNs) with terrestrial networks (TNs), have emerged as a key candidate architecture. However, legacy resource management methods designed for terrestrial systems are difficult to apply directly due to high mobility and long propagation delays (and Doppler effects) of satellite/aerial platforms, dynamic topologies, and constrained onboard resources. In addition, under short-packet transmission (finite blocklength) regimes, QoS analysis must go beyond average-rate metrics and explicitly ensure latency and reliability simultaneously. This paper surveys resource management for SAGIN/TN-NTN integration through a three-axis taxonomy: (i) resource allocation/scheduling, (ii) mobility/dynamics, and (iii) statistical multi-QoS (latency-reliability) modeling. We compare representative works spanning optimization, graph deep reinforcement learning (Graph DRL), and finite-blocklength-based analyses. We also summarize virtualization/slicing and security/robustness as cross-cutting constraints, and highlight open research challenges.