This airticle provides a review of transfer function-based (TF-based) surrogate optimization for electromagnetic (EM) design. Transfer functions (TF) represent the EM responses of passive microwave components versus frequency. With the assistance of TF, the nonlinearity of the model structure can be decreased. Parallel gradient-based EM optimization technique using TF in rational format and trust region algorithm is introduced first. Following that, we review the EM optimization using adjoint sensitivity-based neuro-TF surrogate, where the neuro-TF modeling method is in pole/residue format. The adjoint sensitivity-based neuro-TF surrogate technique can reach the optimal EM responses solution faster than the existing gradient-based surrogate optimization methods without sensitivity information. As a further advancement, we discuss the multifeature-assisted neuro-TF surrogate optimization technique. With the help of multiple feature parameters, the multifeature-assisted neuro-TF surrogate optimization has a better ability of avoiding local minima and can achive the optimal EM solution faster than the surrogate optimizations without feature assistance. Three examples are used to verify the above three methods.
"Recent Advances in Transfer Function-Based Surrogate Optimization for EM Design (Invited)," Progress In Electromagnetics Research,
Vol. 172, 61-75, 2021. doi:10.2528/PIER21110302
1. Zhang, C., J. Jin, W. Na, Q. J. Zhang, and M. Yu, "Multivalued neural network inverse modeling and applications to microwave filters," IEEE Trans. Microw. Theory Tech., Vol. 66, No. 8, 3781-3797, Aug. 2018. doi:10.1109/TMTT.2018.2841889
2. Jin, J., C. Zhang, F. Feng, W. Na, J. Ma, and Q. J. Zhang, IEEE Trans. Microw. Theory Tech., Vol. 67, No. 10, 4140-4155, Oct. 2019. doi: