Microwave Permittivity Characterization Using Power Measurements and Machine Learning

Abstract

This study uses a machine learning assisted (MLA) method to determine the complex permittivity profile of high-to-low loss materials. The presented method is based on using the amplitude of the transmission response of the measurement sensor, eliminating the requirement of phase and reflection measurements. We trained a multi-layer artificial neural network (ANN) using the full-wave simulation results of a partially loaded coaxial line. Debye dispersion model parameters of the material under test (MUT) are retrieved using the ANN, and consequently, the complex permittivity profile is reconstructed. The permittivities of various liquids were measured within the 0.3 – 3 GHz band using a suspended coaxial line. The results of the MLA method exhibit much higher retrieval accuracy compared to our previous work.