Machine learning-based prediction of reflection coefficients for multilayer electromagnetic absorbers

Abstract

The increasing number of wireless devices and the coexistence of multiple systems in shared environments can cause electromagnetic interference, which can be mitigated with the use of multilayer absorbers (MLAs). With MLA design, selecting suitable materials and layer configurations is crucial for achieving the desired performance with the lowest reflection coefficient (RC) and the least thickness. Traditional optimization algorithms (TOAs) are used for this purpose; however, they are computationally expensive and time consuming due to the complexity of layer stacking and thickness optimization. In this study, we propose a machine learning (ML)-based approach for estimating the RC of MLA designs in the 2-18 GHz frequency range that provides results significantly faster than traditional methods. Although TOAs only yield results after exhaustive simulations, ML enables rapid estimation of RC for any desired configuration. ML models, including Gaussian process regression, wide neural network, bagged trees, and boosted trees were trained using data generated by the bald eagle search optimization algorithm. These models were tested on previously unseen data and evaluated using mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), and R-squared (R2) metrics. The RC of a given 3-layer absorber can be predicted in approximately 3 s with high accuracy, achieving R2 = 0.9025, MAE = 0.0694, RMSE = 0.0842, and MSE = 0.0070. The proposed ML framework enables fast and accurate RC prediction in MLA design, supporting engineers and researchers in rapid prototyping and development of electromagnetic absorbers.