星载混合异构处理器在轨重构方法

On orbit reconstruction method for hybrid heterogeneous processors

摘要:

针对有限的星载边缘计算资源,灵活方便的在轨任务重构方法有利于实现多载荷多任务对星载算力资源的有效复用。星载边缘计算处理器已发展为CPU主控制器加FPGA、GPU等协处理器模式,在轨重构已由传统升级CPU的APP转换为面向任务的异构计算升级,既要重构CPU主控制器又要重构协处理器。本文基于ZYNQ处理器实现气象卫星星载边缘计算任务重构,完成异构计算在轨重构方法验证,实现一套硬件同时满足极轨卫星和高轨卫星边缘计算需求。ZYNQ处理器由ARM主控制器和FPGA协处理器构成,ARM处理器实现气象卫星数据预处理和数据调度,FPGA协处理器实现算法硬件加速。本文在轨重构方法由CPU控制协处理器加速算子的加载和版本管控,FPGA的在轨编程上注也由CPU统一管理。

Abstract:

For the limited on-board edge computing resources, a flexible and convenient on orbit task reconfiguration method is conducive to the effective reuse of on-board computing resources by multiple loads and tasks. The on-board edge computing processor has developed into the CPU master controller plus FPGA, GPU and other coprocessor modes. In orbit refactoring has shifted from traditional CPU upgrading apps to task oriented heterogeneous computing upgrades. In this paper, the meteorological satellite on-board edge computing task reconfiguration is implemented based on ZYNQ processor, and the verification of heterogeneous computing on orbit reconfiguration method is completed. A set of hardware is implemented to meet the edge computing needs of polar orbit satellites and high orbit satellites at the same time. The in orbit reconstruction method in this article is controlled by the CPU to accelerate the loading and version control of the co processor operators, and the in orbit programming of FPGA is also managed uniformly by the CPU.