Dielectric Resonator Antenna Gain Enhancement via Deep Reinforcement Learning-Controlled Surface Deformation

This article presents a novel deep reinforcement learning (DRL)-controlled surface deformation approach for dielectric resonator antenna (DRA) gain enhancement. The proposed method was applied to a hemispherical dielectric resonator antenna (HDRA), which was selected as a demonstration case due to the geometric constraints that limit conventional parameterized optimization methods. Instead of manually defining a set of geometric parameters, the DRL agent autonomously explored and learned optimal surface shape modifications that maximize the gain while maintaining a compact structure. The study began with an aperture-coupled liquid HDRA, designed to operate in the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TE</i><sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">111</sub> fundamental mode at 2.45 GHz. The DRL algorithm then iteratively deformed the surface, refining the antenna geometry based on a custom reward function that prioritized gain improvement while preserving frequency stability. The optimized HDRA-based design example achieved a peak gain of 9.6 dBi with an electrical size of 0.55λ<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> × 0.55λ<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> × 0.245λ<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> and a total efficiency exceeding 90%. The gain enhancement trends and associated design considerations are discussed to provide insight into the observed performance improvements. The proposed DRL framework provided an efficient, automated optimization approach for complex DRA structures where conventional methods struggled to define a sufficient set of optimization parameters. The resulting designs exhibited compact size, high realized gain, and simple overall structures, demonstrating the effectiveness of the DRL-based method in electromagnetic optimization. The methodology can be applied to various DRA shapes and traditional antennas, offering a data-driven approach to shape-based performance enhancement.