Prior knowledge input method in device modeling

The artificial neural networks are being used in modelling electronic elements and devices especially at microwave frequencies where non-linearity and dependence on frequency cannot be neglected. In this paper, instead of using artificial neural networks as a unique modelling device prior knowledge input method based on feed-forward artificial neural network structures as multi-layer perceptrons and wavelet-based neural networks is investigated. The benefits of prior knowledge input method over plain usage of artificial neural networks in modelling BJT is explored by comparing models obtained with and without prior knowledge input. The novelty of the paper is utilizing wavelet-based neural networks as feed-forward structure in prior knowledge input method. The training and test data used in simulations are obtained by HP 4155 parameter analyser.