Cost-efficient variable-fidelity machine learning for globalized optimization of microwave structures.

Area of Science:

• Electrical Engineering
• Computational Electromagnetics
• Machine Learning Applications

Background:

• Microwave circuit optimization is computationally intensive.
• Existing methods often lack efficiency and reliability in global optimization.
• Accurate surrogate modeling is crucial for efficient design.

Purpose of the Study:

• To develop a cost-efficient methodology for global optimization of microwave circuits.
• To enhance the reliability and computational efficiency of the circuit design process.
• To reduce the number of required electromagnetic (EM) simulations.

Main Methods:

• Kriging-based machine learning in a reduced dimensionality space.
• Rapid global sensitivity analysis for dimensionality reduction.
• Variable-resolution electromagnetic (EM) simulations for acceleration.
• Two-stage optimization: global search followed by local tuning.

Main Results:

• The methodology achieved cost-efficient optimization for microwave components.
• Demonstrated superior performance over benchmark algorithms in reliability and efficiency.
• Average optimization cost was less than 80 EM analyses, showcasing significant computational savings.
• Validated on rat-race couplers, branch-line couplers, and a dual-band power divider.

Conclusions:

• The proposed method offers a significant advancement in cost-efficient microwave circuit optimization.
• Dimensionality reduction, variable-resolution simulations, and a two-stage approach are key to its success.
• The technique provides a reliable and computationally efficient solution for complex circuit design problems.