Joint Offloading and Adaptive Routing for Computing-Communication Integrated Dual-Layer Satellite Networks

Abstract

With the rapid development of satellite edge computing, high-resolution remote sensing (RS) services generated by very low earth orbit (VLEO) satellites can be offloaded to computing satellites at higher orbital layers for on-orbit processing, enabling communication-computing integration (CCI) in dual-layer satellite networks (DLSNs). However, existing CCI schemes are not suitable for high-resolution RS services due to the frequent switching of cross-layer links and stochastically arriving high-volume data. To address these challenges, this paper proposes a CCI framework in DLSNs for high-resolution RS services. A joint offloading and multipath routing strategy is then proposed. For offloading, a virtual queue mechanism is introduced to predict the load of computing satellites, and a dynamically updated upper confidence bound (UCB) is employed to predict transmission rates, based on which the offloading decision is determined. For multipath routing, we propose a control coefficient to capture network congestion and data heterogeneity via multi-agent deep reinforcement learning. This coefficient is used to determine subtask partition for all tasks. Then, the traffic splitting ratios across multiple paths are allocated based on path quality, such that link setup delay, bottleneck bandwidth, and path load are explicitly incorporated. Historical routing results are fed back to the offloading algorithm for UCB updating, enabling joint optimization of offloading and routing. Simulation results indicate that our proposed strategy significantly reduces the end-to-end delay compared to baseline algorithms.