中国空间科学技术

中国空间科学技术 ›› 2026, Vol. 46 ›› Issue (5): 58-65.doi: 10.16708/j.cnki.1000-758X.2026.0070

• 数学与航天专题 • 上一篇    下一篇

基于并行计算的月背射电干涉成像仿真分析

李毓伦，唐文林*，段然，刘伟杰，高辰，彭晓东*   

  1. 中国科学院国家空间科学中心复杂航天系统电子信息技术实验室，北京101408
  • 收稿日期:2025-11-03 修回日期:2026-01-06 录用日期:2026-03-04 发布日期:2026-07-31 出版日期:2026-09-30

Simulation and analysis of lunar far side radio interferometric imaging based on parallel computing

LI Yulun，TANG Wenlin*，DUAN Ran，LIU Weijie，GAO Chen，PENG Xiaodong*   

  1. Laboratory of Electronic Information Technology for Complex Aerospace Systems, National Space Science Center, Chinese Academy of Sciences, Beijing 101408, China
  • Received:2025-11-03 Revision received:2026-01-06 Accepted:2026-03-04 Online:2026-07-31 Published:2026-09-30

PDF (PC)

21

摘要： 作为中国太空探源科学卫星计划任务之一，鸿蒙计划拟通过多颗绕月飞行的小卫星，在月球背面实现首次高分辨率超长波巡天，打开射电天文电磁观测的最后窗口。为验证任务在实际工程条件下的成像可行性，亟需可支撑百亿级基线与亿级像素处理的端到端并行仿真框架。构建了一个基于单机并行计算的端到端仿真框架，涵盖可用基线筛选、可见度数据生成、地月遮挡判定与图像重建等环节。围绕大规模观测数据的生成与处理提出针对性的数据切分策略，并建立成像链路的工程可行性评估方法。面向全天区与局部天区两类应用场景，进一步开展了常规观测及多子星失效情形下的成像质量分析。仿真试验初步表明，低分辨率全天区成像可获得SSIM=0.9985、PSNR=57.9568，高分辨率局部天区成像可获得SSIM=0.8399、PSNR=23.9138；随着可用基线数量增加，重建图像的SSIM与PSNR指标逐渐提升，并在达到一定观测时长后趋于稳定；在多达2颗低频子星在观测后期失效的情况下，成像质量未显著下降；所提出的切分与并行策略能够在单机内存不溢出的前提下，完成可见度数据生成与大规模方程组求解。提出的端到端并行仿真框架能够有效支撑鸿蒙计划在常规观测及部分子星故障条件下的成像性能评估，具备良好的工程可落地性。该框架为后续向高性能GPU集群迁移奠定了技术基础，并有望推广至其他大规模射电干涉阵列与深空电磁成像任务。

关键词: 射电天文, 孔径综合, 干涉成像, 并行计算, 分层贝叶斯

Abstract: As one of the missions under China's Space Exploration Science Satellite Program, the Hongmeng project aims to deploy multiple small satellites in lunar orbit to conduct the first high-resolution ultra-long-wavelength sky survey on the lunar farside, opening the last observational window of the radio spectrum. To evaluate the feasibility of imaging under practical engineering conditions, an end-to-end parallel simulation framework capable of supporting tens of billions of baselines and processing hundreds of millions of pixels is required. An end-to-end simulation framework based on single-machine parallel computing is developed, covering available-baseline selection, visibility data generation, Earth-Moon occultation determination, and image reconstruction. Targeted data partitioning strategies are introduced for the generation and processing of large-scale observational data, and an engineering feasibility evaluation method is established for the full imaging chain. Imaging performance is further analyzed for both full-sky and partial-sky scenarios under nominal observations and multiple satellite failure conditions. Simulation experiments preliminarily show that low-resolution full-sky imaging achieves SSIM=0.9985 and PSNR=57.9568, while high-resolution partial-sky imaging achieves SSIM=0.8399 and PSNR=23.9138. As the number of available baselines increases, SSIM and PSNR improve steadily and become stable after sufficient observation time. When up to two low-frequency satellites fail during the later stages of observation, imaging quality doesn't degrade. The proposed data-partitioning and parallel strategies enable visibility generation and large-scale linear system solving without exceeding single-machine memory limits. The proposed end-to-end parallel simulation framework effectively supports imaging performance evaluation for the Hongmeng project under both nominal and partial sub-satellite failure conditions, demonstrating strong engineering applicability. The framework provides a solid technical foundation for future migration to high performance GPU clusters and is expected to be extendable to other large scale radio-interferometric arrays and deep space electromagnetic imaging missions.

Key words: radio astronomy, aperture synthesis, interferometric imaging, parallel computing, hierarchical Bayes