The paper is focused on some practical aspects of a new efficient technique for determining the dielectric properties of materials. Complex permittivity is found by an artificial neural network designed to control 3D FDTD computation of S-parameters and to process their measurements. The method is cavity-independent and applicable to samples of arbitrary configurations (as long as the geometry is adequately represented in the FDTD model). We consider a two-port approach which exploits the real and imaginary parts of the reflection and transmission coefficients at the frequency of interest and is capable of handling frequency-dependent media parameters. Numerical testing demonstrates a high accuracy of the computational part of the method (less than 2% for dielectric constant and the loss factor varying in very wide ranges). It is shown that when processing the measured 5-parameters, the developed network is capable of efficiently generalizing and reconstructing complex permittivity even from experimental data which are numerically inconsistent with the modeling data used for network training. Special modeling tests validate a satisfactory level of accuracy in permittivity reconstruction for salt water, ethylene glycol-water mixture, denatured alcohol and acetone at 915 MHz.
展开▼
