Artificial neural network model for HEMTs constructed from large-signal time-domain measurements

A methodology to construct behavioural models for microwave devices from time-domain large-signal measurements has been modified by using artificial neural networks (ANNs) for the multivariate fitting functions instead of polynomials. The behavioural models for the class of devices (microwave transistors) considered can be defined by expressing the terminal currents as functions of the state variables, the embedded voltages. In this work, we show that ANNs are valuable candidates to represent these relationships. They outperform models based on multivariate polynomials, because they can better model the typical physical characteristics of the devices considered. Experimental results are quantitatively confirmed by using comparison metrics.