Cross-Layer Coordination in Space-Terrestrial Time-Sensitive Transmission for Serving 6G Digital Twin Cities: A Multi-Agent Two-Timescale Reinforcement Learning Approach

Abstract

The emergence of 6G Digital Twin Cities (DTC) requires ultra-reliable and Time-Sensitive Transmission (TST) to synchronize physical and cyber entities. Space-Terrestrial Integrated Networks (STIN) is a pivotal technology to achieve global coverage and uninterrupted TST in 6G DTC, but its dynamic and heterogeneous nature poses noticeable challenges in capacity optimization, power allocation, and scheduling timeliness. This paper introduces a novel cross-layer coordination method in space-terrestrial TST that integrates the Rate-Splitting Multiple Access (RSMA) and an improved Age of Information (AoI) technique, dubbed cumulative Value-AoI (VAoI), to harmonize physical-layer resource allocation and data-link layer traffic scheduling. RSMA at the physical layer optimizes capacity and power by dividing public and individual traffic into common and private streams, while the cumulative VAoI at the data-link layer ensures traffic scheduling timeliness. To achieve the cross-layer coordination, we propose a multi-agent two-timescale reinforcement learning approach, where high-level agents manage long-term cumulative VAoI to enhance traffic scheduling effectiveness at the data-link layer, while low-level agents execute short-term physical-layer RSMA for capacity optimization and power allocation. The objective is to maximize network capacity while minimizing cumulative VAoI and power consumption. Simulations demonstrate that our method enhances STIN capacity, reduces transmission failures and latency, and optimizes energy efficiency during TST in 6G DTC.