Fast EM-driven design optimization of microwave filters using adjoint sensitivity and response features

We present an algorithm for design optimization of microwave filters utilizing multi-fidelity electromagnetic (EM)-simulation models, adjoint sensitivities, and a trust-region framework as a convergence safeguard. To further speed up the design process, the optimization is performed at the level of suitably selected response features whose dependence on the designable parameters is significantly less nonlinear than that of the original responses (here, S-parameters versus frequency). Switching between the EM models of different fidelity is governed by suitably defined convergence criteria. Our approach is demonstrated by a waveguide filter example. A comprehensive numerical comparison with single- and multi-fidelity optimization algorithms (both adjoint-based) is also provided.