Bayesian Optimization of Microwave Filters: a Physics-Informed Approach Using the Szegö Kernel

This work presents a data-efficient approach for optimizing microwave devices by combining Bayesian optimization with physics-informed Gaussian process models. The proposed approach models complex-valued frequency responses as a function of design parameters and frequency instead of modeling an objective function, enabling dynamic optimization of devices with intricate performance criteria. The method is validated through the optimization of a zig-zag microstrip bandpass filter to achieve the desired transmission characteristics. This work highlights the potential of Bayesian optimization to reduce computational costs in microwave device design.