| 摘要: |
| 低轨道(Low Earth Orbit，LEO)卫星和高空平台(High Altitude Platform，HAP)成为了实现全域、全时段、全覆盖通信的关键技术。为了更好地为地面用户提供高效稳定的服务，针对HAP辅助的LEO卫星边缘计算，提出了一种终端-高空平台-低轨卫星组成的三层网络架构，任务可以在3个平台上处理，星间也可以协作实现星上负载均衡。考虑时间延迟、资源约束、建模问题的高度复杂性以及星地信道的快速衰落等问题，联合优化卸载决策、带宽与计算资源分配策略，提出了一种基于深度确定性策略梯度的任务卸载与资源分配算法，将问题建模为马尔可夫决策过程，同时对环境状态参数采用状态归一化算法进行预处理。与深度Q网络、全部卸载、无星间链路3种策略算法相比，所提出的算法在时延与能耗方面均能表现出优秀的性能。 |
| 关键词: 低轨卫星 高空平台 移动边缘计算 卸载决策 资源分配 深度强化学习 |
| DOI：10.20079/j.issn.1001-893x.241009002 |
|
| 基金项目:重庆市自然科学基金创新发展联合基金（中国星网）（CSTB2023NSCQ-LZX0114） |
|
| Edge Computing Offloading and Resource Allocation for Space-Air-Ground Collaborative Network |
| YANG Liming,ZHOU Yuqian,JIN Yufeng,ZHAO Hongjun |
| (School of Communications and Information Engineering，Chongqing University of Posts and Telecommunications，Chongqing 400065，China) |
| Abstract: |
| Low Earth orbit(LEO) satellite and high altitude platform(HAP) have become the key technologies to achieve all-domain，full-time，and full-coverage communication.In order to better provide efficient and stable services for ground users，aiming at the HAP-assisted LEO satellite edge computing，a three-layer network architecture composed of terminals，HAPs，and LEO satellites is proposed，and the tasks can be processed on the three platforms，and the satellites can also cooperate to achieve on-board load balancing.Considering the problems of time delay，resource constraints，the high complexity of the modeling problem and the rapid fading of satellite-to-ground channels，by jointly optimizing offload decisions，bandwidth and computing resource allocation strategies，a task offloading and resource allocation algorithm based on deep deterministic policy gradient is proposed，which models the problem as a Markov decision process，and preprocesses the environmental state parameters by state normalization algorithm.Compared with three strategy algorithms including deep Q network，full offloading and no inter-satellite link，the proposed algorithm shows excellent performance in terms of delay and energy consumption. |
| Key words: LEO satellite high altitude platform mobile edge computing offload strategy resource allocation deep reinforcement learning |
