Fast optimization of microwave filters using surrogate-based optimization methods

This paper investigates the use of surrogate-based optimization to optimize the behavioral response of broadband microwave filters. The proposed method makes use of an efficient infill criterion (called expected improvement) that sequentially samples the response of the device at well-chosen regions in the design space. Based on these data samples, successive global surrogate models are built that become increasingly accurate near the optimum solution. A microwave filter example confirms that this approach significantly accelerates device optimization when compared to standard gradient-based methods.