Neural networks for FDTD-backed permittivity reconstruction

Purpose – To outline different versions of a novel method for accurate and efficient determining the dielectric properties of arbitrarily shaped materials. Design/methodology/approach – Complex permittivity is found using an artificial neural network procedure designed to control a 3D FDTD computation of S-parameters and to process their measurements. Network architectures are based on multilayer perceptron and radial basis function nets. The one-port solution deals with the simulated and measured frequency responses of the reflection coefficient while the two-port approach exploits the real and imaginary parts of the reflection and transmission coefficients at the frequency of interest. Findings – High accuracy of permittivity reconstruction is demonstrated by numerical and experimental testing for dielectric samples of different configuration. Research limitations/implications – Dielectric constant and the loss factor of the studied material should be within the ranges of corresponding parameters associated with the database used for the network training. The computer model must be highly adequate to the employed experimental fixture. Practical implications – The method is cavity-independent and applicable to the sample/fixture of arbitrary configuration provided that the geometry is adequately represented in the model. The two-port version is capable of handling frequency-dependent media parameters. For materials which can take some predefined form computational cost of the method is very insignificant. Originality/value – A full-wave 3D FDTD modeling tool and the controlling neural network procedure involved in the proposed approach allow for much flexibility in practical implementation of the method.