基于Jetson的异构SAR成像高性能计算实现OA

Synthetic aperture radar(SAR)is one of the important components of high-resolution imaging radar,and is crucial in military defense,resource exploration,and disaster emergency.The traditional space-based SAR workflow of issuing instructions,acquiring data,and ground-based processing can no longer meet the near-real-time requirements.Whether driven by the exponentially growing data volumes or increasingly complex mission requirements,modern space-based SAR systems must evolve toward a space-based computing paradigm with on-orbit edge processing capabilities.The heterogeneous computing architecture of central processing unit(CPU)+graphics processing unit(GPU)can effectively improve the computing performance.Based on this architecture,NVIDIA's edge computing platform,i.e.,Jetson,can effectively improve the on-orbit computing capabilities with low power consumption.Therefore,in this paper,a high-performance computing system for heterogeneous SAR imaging based on Jetson is designed,which performs imaging on orbit after obtaining echoes,and greatly shortens the information acquisition interval.The heterogeneous computing platform of CPU+GPU is used to compute the SAR data in parallel,and tens of thousands of compute unified device architecture(CUDA)cores are calculated simultaneously with the same command.Based on the chirp scaling(CS)imaging algorithm,the system can compress the imaging time of 19 432×9 288 data to 32.37 s,and the energy consumption is less than 1/4 of the workstation.It effectively verifies the comprehensive on-orbit computing performance of the Jetson platform,and provides experimental support for future on-orbit computing,cluster distributed collaborative computing,and satellite embodied intelligence.