Estimation of Electrical Characteristics of Inhomogeneous Walls Using Generative Adversarial Networks

TL;DR

This paper presents a GAN-based method to accurately estimate the dielectric profile and thickness of inhomogeneous walls from scattered electric fields, improving through-wall radar signal interpretation.

Contribution

It introduces a novel application of GANs for estimating wall electromagnetic properties from limited data, achieving high accuracy with simulated training data.

Findings

GANs estimated wall properties with up to 95% accuracy

Method works with limited training data

Effective in real-world wall scenarios

Abstract

Through-wall radars are researched and developed for the detection, localization, and tracking of human activities in indoor environments. Electromagnetic wave propagation through walls introduces refraction, attenuation, multipath, and ghost targets in the radar signatures. Estimation of wall characteristics (dielectric profile and thickness) can enable wall effects to be deconvolved from through-wall radar signatures. We propose to use generative adversarial networks (GAN) to estimate wall characteristics from narrowband scattered electric fields on the same side of the wall as the transmitter. We demonstrate that the GANs, consisting of two neural networks configured in an adversarial manner, are capable of solving the highly nonlinear regression problem with limited training data to estimate the dielectric profile and thickness of actual walls up to 95\% accuracy based on training with simulated data generated from full-wave solvers.