Task Offloading and Resource Allocation in Satellite Terrestrial Networks: A Deep Deterministic Policy Gradient Approach

Abstract

Low Earth Orbit (LEO) satellite network, as a crucial part of 6G key technology, can provide high network coverage for mobile communications. Due to the rotation and limited resource of satellites, it still challenging to improve QoE with satellite network. To address the above issue, this paper takes advantage of the computation resources provided by terrestrial users, satellites and cloud to achieve better Quality of Experience (QoE). Firstly, we propose a novel edge computing architecture based on satellite-terrestrial integration. Secondly, we formulate the task offloading and resource allocation problem under this architecture as an optimization problem of maximizing the QoE-aware utility. Since the problem is a mixed integer nonlinear programming (MINLP) problem, we proposes a Deep Deterministic Policy Gradient (DDPG)-based Optimization Algorithm for Task Offloading and Resource Allocation (DATR). The DATR algorithm decomposes resource allocation from offloading decision to reduced complexity. The Lagrangian multiplier method is used to achieve the optimal allocation of edge node resources; The offloading decision problem is solved by the DDPG algorithm. Simulation results show that our algorithm outperforms baselines in terms of QoE indicators.