闫超泽,袁馨,吴比翼,等. 一种分层媒质空间格林函数的快速扫参算法[J]. 电波科学学报,xxxx,x(x): x-xx. DOI: 10.12265/j.cjors.2023316
      引用本文: 闫超泽,袁馨,吴比翼,等. 一种分层媒质空间格林函数的快速扫参算法[J]. 电波科学学报,xxxx,x(x): x-xx. DOI: 10.12265/j.cjors.2023316
      YAN C Z, YUAN X, WU B Y, et al. A fast parameter sweeping algorithm for Green’s function in layered media[J]. Chinese journal of radio science,xxxx,x(x): x-xx. (in Chinese). DOI: 10.12265/j.cjors.2023316
      Citation: YAN C Z, YUAN X, WU B Y, et al. A fast parameter sweeping algorithm for Green’s function in layered media[J]. Chinese journal of radio science,xxxx,x(x): x-xx. (in Chinese). DOI: 10.12265/j.cjors.2023316

      一种分层媒质空间格林函数的快速扫参算法

      A fast parameter sweeping algorithm for Green’s function in layered media

      • 摘要: 分层媒质空间积分方程方法是仿真射频集成电路、平面微波器件的一类高效且精确的算法,但如何高效计算分层媒质空间中的格林函数即Sommerfeld积分,仍是计算电磁学领域中一个重要技术挑战。本文提出了一种层状格林函数快速参数(频率、材料、厚度等)扫描算法。通过将Sommerfeld数值积分转化为矩阵乘法运算,然后采用GPU中的Tensor核进行硬件加速,实现了不同参数条件下Sommerfeld积分同步计算。为了验证本文方法的有效性,在一款商用NVDIA GPU中测试了多种分层媒质结构下,不同频点、材料以及分层媒质厚度下的格林函数计算。数值实验表明,在不损失精度的情况下,本文所提快速扫参方法相对利用高端Intel CPU与OpenMP并行的常规方法,加速上千倍以上。

         

        Abstract: The integral equation method in layered media is an efficient and accurate method for simulating radio frequency (RF) integrated circuits and planar microwave devices. The efficient evaluation of Green’s functions in layered media or Sommerfeld integrals is still a major challenge in the computational electromagnetics. This paper proposes a fast parameter (frequency, material, thickness, etc.) sweeping algorithm for Green’s function in layered media. By converting the numerical Sommerfeld integral into a matrix multiplication and then using the tensor core in the GPU for hardware acceleration, the Sommerfeld integral at different parameters is achieved. To verify the efficiency of this method, we evaluate Green’s function in various layered media at different frequencies, materials and layered media thicknesses on a commercial NVDIA GPU. Numerical experiments show that the proposed parameter sweeping algorithm is 3 orders of magnitude faster than the traditional method accelerated by OpenMP on a high-end CPU without losing accuracy.

         

      /

      返回文章
      返回