Complex-Valued DNN for Broadband Dielectric Characterization of Dispersive Lossy Materials

This paper presents a broadband dielectric characterization method based on a Complex-Valued Deep Neural Network (CVNN) that allows the retrieval of permittivity and conductivity of dispersive lossy materials using ad-hoc setups. To validate the method, we numerically tested it employing a partially filled custom-made double-ridge waveguide setup, working from 0.95 to 4.2 GHz. Moreover, we include a feature importance analysis using agnostic explainable-AI (XAI) techniques. The results demonstrate the flexibility and the retrieval capabilities of the method, as well as the advantages and drawbacks in comparison with traditional techniques. We publicly release the dataset and codes to support further research.