不规则问题驱动下的多维度SAR回波混合粒度并行模拟

项目名称： 不规则问题驱动下的多维度SAR回波混合粒度并行模拟

项目编号： No.61501018

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 张帆

作者单位： 北京化工大学

项目金额： 19万元

中文摘要： 随着SAR向着多维度、高分辨率、宽测绘带方向发展，新体制SAR系统设计、成像算法研究需求大场景、复杂工作模式的回波数据提供支撑，同时也给海量回波的模拟效率提出了更高的要求。在SAR回波模拟过程中，回波相干积累过程是唯一的不规则问题，难以细粒度并行化，严重影响模拟效率。因此，本课题拟在之前基于GPU的SAR回波并行模拟基础上，结合多核CPU矢量化并行方法，探索多维度SAR回波的高效模拟方法，从以下方面开展工作：1）通过并行归约和粒度配置的思路，解决不规则问题并行化问题，提高GPU并行累加速度；2）建立多层次多核CPU矢量化体系，解决相干积累过程难于矢量化的问题，并结合多维度SAR体制研究不规则问题矢量化的新方法；3）提出面向多维度SAR回波模拟的混合粒度并行模拟方法，实现多模式SAR回波的统一模拟，并可扩展至大规模异构系统进行实时计算，可望有助于多维度SAR系统的研制、成像方法研究。

中文关键词： 回波仿真；多维度SAR；不规则问题；矢量化扩展；协同计算

英文摘要： With the development of SAR system toward the multidimensional observation, high resolution and wide swath, new SAR system design and imaging processing research require the huge echo data of large scene and complex working modes, and pose a challenge for SAR raw data fast simulation. In the SAR raw data simulation, the signal coherent accumulation process is actually an irregular issue, which seriously decreases the parallel efficiency. Therefore, this study intends on the basis of previous studies and multi-core vector extension method to explore the optimization methods of irregular issue driven multidimensional SAR raw data hybrid-grained parallel simulation, and the research work is as follows: 1) Introduce the parallel reduction method and parallel granularity configuration method to solve the parallelization of irregular issue, thus increase the speed of GPU parallel accumulation. 2) Build the multi-level vectorized system to solve the vectorization problem of the coherent accumulation process. Based on the features of the multidimensional SAR system, study a new vectorization method for irregular issue. 3) Propose a novel hybrid-grained parallel method for multidimensional SAR echo simulation, which can be extended and run in the large-scale heterogeneous computing system, and is expected to contribute to the multidimensional SAR system development, imaging algorithm studies.

英文关键词： raw data simulation;multidimensional space joint-observation synthetic aperture radar;irregular issue;vector extension;collaborative computing