A neural network approach to real-time dielectric characterization of materials

Artificial neural networks (ANNs),are proposed as an inversion approach in microwave measurement methods of the dielectric characteristics of materials. Reflection and transmission methods require a proper electromagnetic (EM) model of the measurement system, and solutions in terms of the material permittivity are usually in implicit form, requiring an inversion procedure that can be costly in terms of computing time. In such applications, the ANN approach is favorable in that the EM computations are performed during the off-line training phase of the network. As an example of a real-world application, the ANN approach is tested on a waveguide method for the measurement of permittivity in the X band for which a working solution is yet available, showing that the parameters of interest (permittivity and thickness of the material under measurement) can be obtained in real-time with a negligible loss of accuracy. (C) 2002 Wiley Periodicals, Inc.