Adaptive Orthogonal Sampling for Automated Model Generation of Microwave Components

Automated model generation (AMG) algorithms have become popular techniques for the systematic development of artificial neural network (ANN) models for microwave components. However, existing AMG algorithms rely on a grid-based sampling strategy, which can be time-consuming and inefficient for high-dimensional microwave modeling problems. To address this challenge, we propose a novel adaptive orthogonal sampling method for AMG. Starting from a small amount of training data in orthogonal distribution, the proposed method employs stage-wise orthogonal sampling. At each stage, we propose to use interpolation techniques to evaluate the sufficiency of training data and dynamically determine the additional training samples required, as well as their optimal locations within the modeling region. The resulting sampling distribution is efficient, with more samples orthogonally distributed in nonlinear subregions and fewer in smooth subregions. This proposed orthogonal sampling method allows the AMG to use significantly fewer samples than traditional grid-based AMG, effectively improving the efficiency of ANN modeling for microwave applications. Two microwave filter modeling examples are provided for demonstration.