Satellite remote sensing technology has promoted the emergence of a variety of earth observation (EO) tasks with Ultra HD resolution and/or high real-time requirements. Limited by the bandwidth of space-ground links and the computing power of a single satellite, on-orbit collaborative edge computing improves task processing efficiency. The task is split into subtasks and offloaded to different computing nodes. Multi-layer AI structure leads to sequential dependency among some subtasks. With the challenges of sequential constraints subtasks in dynamic satellite collaborative edge computing scenarios, satellite visible model based on spatial geometry is first proposed in this paper to characterize the communication window between satellites. And the objective function is formulated to minimize the weighted cost of system delay and energy consumption by jointly optimizing cooperative task offloading and resource allocation. This non-convex problem is further decomposed into the subtask offloading and resource allocation subproblems, which are solved by Tabu search algorithm and successive convex approximation algorithm respectively. The simulation results show that the proposed cooperative edge computing scheme reduces the latency and energy consumption weighted cost by 29.3% and 69.4%, respectively, compared with the satellite local computing scheme and the method of calculation on the ground after full download.

中文翻译：

---

卫星边缘计算网络中顺序约束任务的协同任务卸载与资源分配

卫星遥感技术推动了各种具有超高清分辨率和/或高实时性要求的地球观测（EO）任务的出现。受天地链路带宽和单颗卫星计算能力的限制，在轨协同边缘计算提高了任务处理效率。任务被拆分为子任务，并卸载到不同的计算节点。多层 AI 结构导致某些子任务之间的顺序依赖。针对动态卫星协同边缘计算场景中顺序约束子任务的挑战，该文首次提出了基于空间几何的卫星可见模型来表征卫星间通信窗口。并制定目标函数，通过共同优化协同任务卸载和资源分配，将系统延迟和能耗的加权成本降到最低。将该非凸问题进一步分解为子任务卸载子问题和资源分配子问题，分别采用塔布搜索算法和逐次凸逼近算法进行求解。仿真结果表明，与卫星本地计算方案和全下载后地面计算方法相比，所提协同边缘计算方案的时延和能耗加权成本分别降低了 29.3%和 69.4%。