Convergence- and Quality-Controlled Model Resolution Adjustment for Accelerated Multifidelity Design of Antenna Structures

Formal optimization methods are now standard in antenna design, driven by the growing complexity of antenna structures and the limitations of traditional approaches such as interactive parametric studies. Reliable evaluation requires electromagnetic (EM) simulation tools; however, the need for repeated simulations during optimization makes the process computationally expensive. Existing mitigation strategies include reducing the number of EM evaluations at the algorithmic level, employing surrogate models, or using variable-fidelity models—typically restricted to two levels. In this work, we propose an adaptive model management strategy that dynamically adjusts the EM model resolution during the search process, guided by convergence and design quality indicators. The algorithm begins with the lowest acceptable fidelity, and progressively increases accuracy as the design converges and performance improves. This ensures cost savings by relying on faster, less accurate models early in the search, while switching to high-fidelity simulations near the optimum to guarantee reliability. The strategy is embedded within a gradient-based procedure. Comprehensive validation with four broadband antennas and multiple benchmark methods confirms the effectiveness of the approach, yielding an average acceleration of ~62% over the baseline method, with only minor quality deterioration.