Electromagnetic Machine Learning for Inverse Modeling Using the Spatial Singularity Expansion Method

Abstract

In this article, we propose a general technique that utilizes a unified electromagnetic and machine learning (ML) technique for inverse modeling and antenna characterization. The recently developed spatial singularity expansion method (S-SEM) is deployed to explicate the electromagnetic behavior of antennas in the form of an accurate digital signal processing (DSP) model. An ML framework is devised and combined with the S-SEM-based DSP model to design a novel inverse source modeling algorithm. The combined S-SEM-ML system departs from the state-of-the-art approaches in being capable of processing far-field data in order to jointly estimate the surface current distribution on the examined radiators, in addition to reconstructing their geometrical details. Various straight and bent wire systems are investigated, including single and multiple array configurations. A study on the impact of additive free-space noise on the estimation process is also presented. Finally, a measurement apparatus is provided to validate the proposed S-SEM-ML system.