Material Characterization Using Power Measurements: Miracle of Machine Learning

Abstract

This paper introduces a machine learning (ML) measurement method to retrieve the complex permittivity profile of dispersive and non-dispersive materials with high or low loss. The proposed method employs amplitude-only transmission measurements, removing the need for phase and reflection measurements. This approach relies on training a multi-layer artificial neural network (ANN) which utilizes the full-wave simulation results of a partially loaded coaxial line as its training set. A valid frequency dispersion model (Debye model) is utilized, and consequently, the model parameters are retrieved. For experimental validation, a suspended coaxial line was designed and fabricated, and the permittivities of several liquid chemicals were measured within the 0.3-3 GHz band. The retrieved results demonstrate much higher accuracy, up to roughly 30 times error reduction, for the ML approach compared to our previous work.