电大尺寸目标电磁散射和绕射高效混合异构并行计算方法

      Efficient hybrid heterogeneous parallel computing method for electromagnetic scattering and diffraction of electrically large objects

      • 摘要: 高频电磁散射机理主要包括镜面反射、边缘绕射和爬行波绕射等。针对电大尺寸复杂计算中单一高频算法散射机制不完备、计算效率与精度难以兼顾的难题,本文提出一种融合多种散射机理的建模框架与异构并行计算加速方法。该框架采用多层快速物理光学(multilevel fast physical optics, MLFPO)法分析镜面散射,结合增量长度绕射系数(incremental length diffraction coefficient, ILDC)法与截断劈ILDC(truncated wedge ILDC, TWILDC)法,分别有效计算边缘与表面爬行波绕射,构建了涉及三大高频散射机制的完备协同物理模型,实现计算精度提升。计算效率上,基于CUDA架构设计细粒度GPU并行优化策略与事件触发的异构同步调度机制,实现多算法的CPU-GPU异构协同高效执行。数值算例表明,针对电大尺寸目标,所提协同模型与多层快速多极子算法(multilevel fast multipole algorithm, MLFMA)结果吻合良好,相较单一MLFPO的平均误差降低1.82 dB,所提异构并行方案相较MLFMA加速比可达318.43,加速比突破两个数量级。该方法可为提升电大尺寸目标雷达散射截面(radar cross section, RCS)的可信度和高效计算提供坚实的技术支撑。

         

        Abstract: The main mechanisms of high-frequency scattering include specular reflection, edge diffraction, and creeping wave diffraction. To overcome limitations of single-algorithm methods for electrically large targets, which suffer from incomplete mechanisms and trade-off between efficiency and accuracy, this paper proposes a collaborative modeling framework with heterogeneous parallel acceleration. The framework integrates multilevel fast physical optics (MLFPO) for specular scattering, incremental length diffraction coefficient (ILDC) for creeping waves, and truncated wedge ILDC(TWILDC) for edge diffraction, effectively computing edge and creeping wave diffraction, and establishes a physical model covering three major scattering mechanisms, improving accuracy. For efficiency, a fine-grained GPU parallelization strategy and an event-triggered heterogeneous scheduling mechanism are designed for efficient CPU-GPU collaborative execution. Numerical examples show that the proposed model achieves good agreement with multilevel fast multipole algorithm(MLFMA), reducing mean absolute error by 1.82 dB versus standalone MLFPO. The proposed scheme delivers a maximum 318.43× speedup over MLFMA, exceeding two orders of magnitude. This method provides solid technical support for high-confidence and efficient radar cross section(RCS) prediction of electrically large targets.

         

      /

      返回文章
      返回