Tuning of geometry parameters is one of the essential stages of contemporary antenna design. It is necessary because available design methods, whether based on theoretical considerations or on engineering experience, are only capable of yielding initial designs that need further adjustment in order to boost the performance parameters as much as possible. Numerical optimization is also imperative for antenna re-design with respect to different operating conditions (e.g., center frequency) or material parameters (e.g., substrate permittivity/thickness). As the tuning process normally involves full-wave electromagnetic (EM) analysis, it may incur considerable computational expenses. Reducing these costs is highly desirable from the perspective of both academic research and industry. This paper proposes a simple surrogate-assisted framework for accelerated antenna parameter tuning that allows for reusing the previously acquired design data. The first (inverse) surrogate serves as a reliable predictor for generating a reasonable initial design, whereas the second (forward) model encodes antenna sensitivities at the level of so-called response features. The involvement of the response feature technology leads to a more accurate rendition of the antenna gradients, which speeds up the design refinement as compared to the forward model constructed at the level of original antenna characteristics. Our methodology is demonstrated using two microstrip antennas and compared to the previously reported warm-start optimization procedures.

中文翻译：

---

使用逆和基于特征的前向克里金代理加速天线结构的参数调整

几何参数的调整是当代天线设计的基本阶段之一。这是必要的，因为现有的设计方法，无论是基于理论考虑还是工程经验，都只能产生需要进一步调整以尽可能提高性能参数的初始设计。针对不同操作条件（例如，中心频率）或材料参数（例如，基板介电常数/厚度）的天线重新设计，数值优化也是必不可少的。由于调谐过程通常涉及全波电磁 (EM) 分析，因此可能会产生相当大的计算费用。从学术研究和工业的角度来看，降低这些成本是非常可取的。本文提出了一个简单的代理辅助框架，用于加速天线参数调谐，允许重复使用以前获取的设计数据。第一个（反向）代理作为生成合理初始设计的可靠预测器，而第二个（正向）模型在所谓的响应特征级别对天线灵敏度进行编码。响应特征技术的参与导致天线梯度的更准确再现，与在原始天线特性水平构建的前向模型相比，这加速了设计改进。我们的方法使用两个微带天线进行了演示，并与之前报道的热启动优化程序进行了比较。第一个（反向）代理作为生成合理初始设计的可靠预测器，而第二个（正向）模型在所谓的响应特征级别对天线灵敏度进行编码。响应特征技术的参与导致天线梯度的更准确再现，与在原始天线特性水平构建的前向模型相比，这加速了设计改进。我们使用两个微带天线演示了我们的方法，并与之前报道的热启动优化程序进行了比较。第一个（反向）代理作为生成合理初始设计的可靠预测器，而第二个（正向）模型在所谓的响应特征级别对天线灵敏度进行编码。响应特征技术的参与导致天线梯度的更准确再现，与在原始天线特性水平构建的前向模型相比，这加速了设计改进。我们使用两个微带天线演示了我们的方法，并与之前报道的热启动优化程序进行了比较。响应特征技术的参与导致天线梯度的更准确再现，与在原始天线特性水平构建的前向模型相比，这加速了设计改进。我们使用两个微带天线演示了我们的方法，并与之前报道的热启动优化程序进行了比较。响应特征技术的参与导致天线梯度的更准确再现，与在原始天线特性水平构建的前向模型相比，这加快了设计改进。我们使用两个微带天线演示了我们的方法，并与之前报道的热启动优化程序进行了比较。