"Explainable AI-Aided Design for Antennas and Implementation on a Leaky Wave Antenna Dataset"

"Modern antenna design is critical for advanced wireless and aerospace communication systems, yet its reliance on electromagnetic (EM) simulations\u2014often slow, resource- intensive, and limited to predefined geometries\u2014constrains rapid prototyping and innovation. In response, recent advances in machine learning (ML) have shown promise in accelerating the antenna design process. ML has been used not only to predict performance metrics but also to enable inverse design, where desired specifications drive the synthesis of optimal geometries. In this paper, we demonstrate that an ML-based approach optimizes a leaky wave antenna (LWA) design significantly faster than full-wave simulations. By training an ML model on 233 unique combinations of geometric parameters and their corresponding S11 magnitude values, we used the model to predict S11 across 327,892 parameter combinations within the defined design space. Multiple ML models were evaluated, and XGBRegressor achieved the highest accuracy at 62.71%. Applying explainable AI (XAI) to identify and refine the most influential geometric parameters improved the score to 73.43%. Our approach was validated by fabricating and testing two optimized LWA prototypes. The antennas demonstrated operational bandwidths of 11.8\u201317.2 GHz and 12.8\u201318.8 GHz, with frequency-dependent beam steering up to 80\u25e6 and realized gain ranging from 5.9 to 7.8 dBi. These results illustrate that ML-driven optimization reduces computational overhead without sacrificing accuracy, highlighting the potential of adaptive design methods for rapid and scalable antenna development."