Neural approaches for parameter extraction of microwave transistor noise models

The aim of this paper is to analyze and compare two artificial neural network based approaches for parameter extractions of microwave transistor equivalent circuits including noise. In the first approach equivalent circuit parameters are determined from the operating conditions, whereas in the second approach equivalent circuit parameters are determined directly from the measured scattering and noise parameters. In both approaches, multilayer perceptron artificial neural networks are applied. The considered extraction approaches are analyzed on an example of temperature dependent modeling of a pHEMT transistor.