Parallel EM Optimization Method for Microwave Filter Design Using Neuro-TF Surrogate and Isomorphic Orthogonal DOE Sampling

This paper introduces a parallel electromagnetic (EM) optimization method using neuro-transfer function (neuro-TF) surrogate and isomorphic orthogonal Design of Experiments (DOE) sampling for microwave filter design. A sampling strategy that combines isomorphic orthogonal DOE with parallel EM simulations is employed to generate training data for the neuro-TF surrogate. This sampling strategy ensures a scattered distribution, thereby enhancing the accuracy of the surrogate model and speeding up the optimization process. Using adaptive splitting, a poleresidue tracking technique adjusts based on neighboring sample data, improving the continuity and precision of the surrogate’s pole-residue representation. This approach results in faster and more efficient optimal solutions with fewer iterations, as demonstrated in an example of EM optimization for a microwave filter.