Optimization of microwave components using machine learning and rapid sensitivity analysis.

Area of Science:

• Electrical Engineering
• Computational Electromagnetics
• Applied Mathematics

Background:

• Optimization methods are increasingly vital for high-frequency electronics and microwave design.
• Complex passive microwave components require meticulous parameter tuning for optimal performance.
• Global optimization is often necessary but computationally expensive due to electromagnetic (EM) analysis and extensive simulations.

Purpose of the Study:

• To introduce a novel procedure for expedited globalized parameter adjustment of microwave passives.
• To address the high computational costs associated with traditional global optimization methods.
• To improve the efficiency and effectiveness of microwave component design.

Main Methods:

• A surrogate-assisted machine learning framework is employed for the optimization process.
• The search domain is restricted to important parameter space directions identified via fast global sensitivity analysis.
• Domain confinement reduces surrogate model costs and enhances predictive accuracy, complemented by local tuning.

Main Results:

• The proposed approach significantly reduces the computational expense of global optimization for microwave components.
• The dimensionality reduction technique improves the accuracy and efficiency of surrogate model establishment.
• Verification experiments confirm the approach's remarkable efficacy compared to benchmark methods.

Conclusions:

• The developed procedure offers a computationally efficient and effective solution for global parameter adjustment in microwave passive design.
• This surrogate-assisted machine learning framework overcomes limitations of traditional methods in high-dimensional spaces.
• The approach demonstrates significant advantages for optimizing complex microwave circuits, paving the way for faster design cycles.