Unit Cell Design for Space-Fed Surfaces Via Kernel-Based Machine Learning Regression

This work explores a novel approach for the automatic design and optimization of unit cells (UCs) via kernel-based machine learning regression. Traditional UC optimization relies on brute-force full-wave simulations, which are computationally expensive and time-consuming. The proposed method uses the Least-Squares Support Vector Machines (LS-SVM) regression to build surrogate models enabling the efficient design space exploration. The optimal UC geometry obtained by the proposed optimization methodology is then validated through the complete design of a three-layer Transmitarray Antenna (TA), achieving a 32 dB peak gain at 30 GHz with approximately 50 % efficiency, a 1-dB bandwidth of 14 %, confirming its excellent radiation performance.