催交
灵敏度(控制系统)
计算机科学
微波食品加热
微波工程
数学优化
功率(物理)
架空(工程)
可靠性(半导体)
电子工程
数学
工程类
电信
物理
系统工程
量子力学
操作系统
作者
Anna Pietrenko‐Dabrowska,Sławomir Kozieł
出处
期刊:IEEE Transactions on Microwave Theory and Techniques
日期:2022-09-29
卷期号:70 (11): 4765-4771
被引量:7
标识
DOI:10.1109/tmtt.2022.3207482
摘要
Geometry parameters of contemporary microwave passives have to be carefully tuned in the final stages of their design process to ensure the best possible performance. For reliability reasons, the tuning has to be carried out at the level of full-wave electromagnetic (EM) simulations. This is because traditional modeling methods are incapable of quantifying certain phenomena that may affect the operation and performance of these devices, such as cross-coupling effects. As a consequence, the designs yielded with the use of equivalent network models may only serve as starting points that need further refinement. Unfortunately, simulation-driven numerical optimization is computationally demanding even in the case of local search procedures. Thus, significant research efforts have been aimed toward identifying effective ways of expediting EM-driven optimization procedures, critical from the point of view of cost of design cycles. Among these, one may list the recently proposed multifidelity optimization frameworks. Another option for accelerating simulation-driven design procedures is sparse sensitivity updating schemes, where costly gradient estimation through finite differentiation (FD) is suppressed for selected variables. This work proposes a novel algorithm that capitalizes on both aforementioned mechanisms to reduce the optimization cost of local gradient-based parameter tuning of compact microwave components. In our approach, multifidelity optimization is further expedited by replacing expensive FD sensitivity updates with the Broyden formula for selected design variables. Verification using two microwave structures, a branch-line coupler and a power divider, demonstrates average savings of around 80% over the basic trust-region (TR) routine, with only minor degradation of the design quality.
科研通智能强力驱动
Strongly Powered by AbleSci AI