Multi-Modal Learning-Based Multi-Task Offloading Schemes for Satellite-Ground Integrated Networks

doi:10.1109/TWC.2025.3548574

Multi-Modal Learning-Based Multi-Task Offloading Schemes for Satellite-Ground Integrated Networks

2025-07-01

0

PRE

AI

George K. Karagiannidis

DOI：10.1109/TWC.2025.3548574

Original

Original request for help

Save

Abstract

En 中文

Satellite-Ground Integrated Networks (SGINs) are promising network architectures that can help reduce the load on terrestrial networks, provide mega-access capabilities and intensive task offloading functions. However, traditional resource management methods are difficult to apply directly into SGINs due to their multi-layered, heterogeneous and dynamic three-dimensional characteristics. In addition, massive multi-modal and multi-task information hinders better service performance in SGINs. Therefore, we design a multi-task integrated computation offloading model to process complex multi-modal network information, such as time-varying channel gains and dynamic Low Earth Orbit (LEO) locations, which can efficiently improve data transmission rate and privacy level. Furthermore, we propose three multi-modal based learning methods, such as centralized actor-critic (C-AC) algorithm, distributed multi-agent deep deterministic policy gradient (D-MADDPG) algorithm, and quantization-based federated learning (Q-FL) algorithm for computation-intensive, latency-critical and privacy-preserving tasks, which can further optimize the local execution or LEO offloading ratio, CPU cycle frequency and transmission power. Meanwhile, we demonstrate the quantization error upper bound between the optimal solution and the quantization scheme through massive mathematical derivations. Finally, extensive simulation results show that the proposed multi-modal based learning methods have better performance gains in terms of model convergence performance, quantization metrics, data transmission rate and number of bits processed.

Keywords:

Satellite-ground integrated networks (SGINs)

multi-task integrated computation offloading model

multi-modal based learning methods

quantization error upper bound

Multi-Modal Learning-Based Multi-Task Offloading Schemes for Satellite-Ground Integrated Networks

Abstract

Journal

Organization