Modern antenna design is an intrinsic undertaking. It relies heavily on expert knowledge and computational tools, especially electromagnetic (EM) simulation software. The most daunting phase is the development of a fundamental antenna topology that exhibits the assumed functionality (such as antenna’ ability to operate across a broad or multiple frequency bands) while fulfilling other conditions (e.g., sufficiently small size). The process typically takes weeks of interactive work and involves partly trial-and-error-based geometry modification and parametric studies. As a result, a limited number of possible configurations may be examined. Unsupervised topology generation constitutes an attractive alternative, although the existing methodologies are computationally expensive or intrusive (e.g., require dedicated fast solvers and/or adjoint sensitivities). This study introduces a database-aided automated antenna design technique and its experimental validation. Our approach employs a flexible parameterization incorporating adjustable elliptical patches and gaps. Comprehensive simulation data acquired for diverse arrangements of the building blocks and substrate sizes is looked up at the design stage, and the most promising architecture is tuned using the expedited gradient-based optimizer. As demonstrated, the presented approach enabled the specification-based development of high-performance antenna geometries, requiring fewer than two hundred EM simulations. For illustration, several broadband, ultrawideband, and multi-band antennas are designed, including structures optimized for minimum size. A distinctive feature of the procedure is the ability to generate wideband antennas (e.g., 5–6 GHz, 3.1–5.8 GHZ), ultra-wideband structures (3.1–10.6 GHz), and up to triple-band radiators (e.g., 2.45–5.3–7.5 GHz), or devices featuring specific dimensions (e.g., fixed width or length). Due to highly unconventional antenna geometries rendered by the proposed approach, experimental validation of the developed structures is instrumental to corroborate the relevance of the methodology. Toward this end, selected designs are prototyped, and their electrical and field characteristics are measured, showing good agreement with full-wave simulations. This underscores the suitability of the proposed technique for automated antenna development with no expert interaction required.
Koziel et al. (Wed,) studied this question.