A machine learning methodology for inferring network S-parameters in the presence of variability

Abstract

This paper proposes the use of Variational Autoencoders, a generative modeling technique, for the problem of inferring S-parameters of linear multiport networks in the presence of manufacturing variability. The Variational Autoencoder learns the underlying data generation process and yields a generative network that can approximately mimic the probability distribution of the training data. The generated samples can be used for subsequent statistical simulations. A post-processing step, applying Vector Fitting to the predicted S-parameters, constrains the model to a finite-order rational function form and enforces appropriate physical constraints. The method is validated through its application to a coupled micro strip transmission line.