Adaptive resource allocation in 6G A2G-TN integrated system: A cross-layer multi-agent PPO approach

초록

In dense Air-to-Ground (A2G) and Terrestrial Network (TN) integrated systems, severe interference and dynamic traffic conditions pose significant challenges, leading to reduced throughput and degraded communication quality. Existing resource allocation techniques, including traditional Multi-Agent Proximal Policy Optimization (MAPPO) and other multi-agent frameworks, often fail to address these challenges efficiently in highly dynamic environments. To overcome these limitations, we propose a novel Cross-Layer MAPPO (CL-MAPPO) framework that leverages dynamic Time Division Duplexing (D-TDD) and cross-layer optimization to enhance resource allocation. The framework employs centralized training and distributed execution (CTDE) to optimize uplink (UL) and downlink (DL) resource allocation while incorporating conditional activation function to reduce computational overhead by adapting policies only to significant network changes. The fusion of CTDE with cross-layer optimization in CL-MAPPO enables precise interference management and real-time adaptability in complex scenarios. Simulation results demonstrate that CL-MAPPO outperforms traditional MAPPO and other state-of-the-art multi-agent techniques, achieving an 18% improvement in throughput and a 16% reduction in interference. These findings underscore the potential of CL-MAPPO to significantly enhance the efficiency and reliability of A2G-TN integrated systems, particularly in high-traffic conditions.

키워드