A robust correction model based neural network modeling framework for electromagnetic simulations and RF measurements

This paper introduces a new artificial neural networks (ANNs)-based correction-modeling approach for simulations and measurements. The proposed approach improves the accuracy of conventional neural models by reversing input-output variables in a systematic manner, while keeping the model structures simple relative to complex knowledge-based ANNs (KBNNs). The approach facilitates accurate/fast neural network modeling of practical electromagnetic (EM) structures, for which, training data is expensive. Two examples are presented to demonstrate the accuracy, efficiency, and feasibility of the proposed modeling approach. The first example is a broadband wire monopole antenna loaded by an annular dielectric ring resonator (DRR) at the antenna feed point. The second example is a metallic waveguide (WG) tube coated with inhomogeneous lossy materials for enhanced electromagnetic interference (EMI) shielding. The proposed approach is significant to RF circuit designers since it helps in building accurate models using reduced numbers of full-wave EM simulations and/or RF measurements.