Eves E Eugene, Murphy Ethan K, Yakovlev Vadim V
The Ferrite Company, Inc., Hudson, NH, USA.
J Microw Power Electromagn Energy. 2007;41(4):81-94. doi: 10.1080/08327823.2006.11688572.
The paper discusses characteristics of a new modeling-based technique for determining dielectric properties of materials. Complex permittivity is found with an optimization algorithm designed to match complex S-parameters obtained from measurements and from 3D FDTD simulation. The method is developed on a two-port (waveguide-type) fixture and deals with complex reflection and transmission characteristics at the frequency of interest. A computational part is constructed as an inverse-RBF-network-based procedure that reconstructs dielectric constant and the loss factor of the sample from the FDTD modeling data sets and the measured reflection and transmission coefficients. As such, it is applicable to samples and cavities of arbitrary configurations provided that the geometry of the experimental setup is adequately represented by the FDTD model. The practical implementation of the method considered in this paper is a section of a WR975 waveguide containing a sample of a liquid in a cylindrical cutout of a rectangular Teflon cup. The method is run in two stages and employs two databases--first, built for a sparse grid on the complex permittivity plane, in order to locate a domain with an anticipated solution and, second, made as a denser grid covering the determined domain, for finding an exact location of the complex permittivity point. Numerical tests demonstrate that the computational part of the method is highly accurate even when the modeling data is represented by relatively small data sets. When working with reflection and transmission coefficients measured in an actual experimental fixture and reconstructing a low dielectric constant and the loss factor the technique may be less accurate. It is shown that the employed neural network is capable of finding complex permittivity of the sample when experimental data on the reflection and transmission coefficients are numerically dispersive (noise-contaminated). A special modeling test is proposed for validating the results; it confirms that the values of complex permittivity for several liquids (including salt water acetone and three types of alcohol) at 915 MHz are reconstructed with satisfactory accuracy.
本文讨论了一种基于建模的新型材料介电特性测定技术的特点。通过一种优化算法来寻找复介电常数，该算法旨在匹配从测量以及三维时域有限差分（FDTD）模拟中获得的复S参数。该方法是在一个双端口（波导型）夹具上开发的，用于处理感兴趣频率下的复反射和传输特性。计算部分构建为基于径向基函数（RBF）网络的反演程序，可从FDTD建模数据集以及测量的反射和传输系数中重建样品的介电常数和损耗因子。因此，只要FDTD模型能充分表征实验装置的几何结构，该方法就适用于任意构型的样品和腔体。本文所考虑方法的实际应用是WR975波导的一部分，其中在矩形聚四氟乙烯杯的圆柱形切口处装有液体样品。该方法分两个阶段运行，并使用两个数据库——首先，为复介电常数平面上的稀疏网格构建数据库，以便定位预期解所在的区域；其次，构建一个覆盖已确定区域的更密集网格数据库，用于找到复介电常数点的确切位置。数值测试表明，即使建模数据由相对较小的数据集表示，该方法的计算部分仍具有很高的准确性。当处理在实际实验夹具中测量的反射和传输系数并重建低介电常数和损耗因子时，该技术的准确性可能会降低。结果表明，当反射和传输系数的实验数据存在数值色散（受噪声污染）时，所采用的神经网络能够找到样品的复介电常数。本文提出了一种特殊的建模测试来验证结果；它证实了在915MHz频率下，几种液体（包括盐水、丙酮和三种酒精）的复介电常数能够以令人满意的精度重建。
