Variable-fidelity optimization of antennas using adjoint sensitivities

Computationally efficient and reliable procedure for antenna design using electromagnetic (EM) simulation models is proposed. Our approach exploits trust-region-based gradient search and adjoint sensitivities for fast derivative evaluation. Derivative data allows for constructing the approximation model of the antenna responses. Computational cost of the design process can be further reduced by utilizing variable-fidelity EM simulations and adaptive procedure of switching between EM antenna models of increasing accuracy. We demonstrated the proposed methodology using a dielectric resonator antenna. The overall design cost using the proposed approach corresponds to less than eight antenna simulations at fine discretization and it is 50 percent lower than direct high-fidelity model optimization (also exploiting adjoint sensitivities).