Abstract

A new procedure to learn a nonlinear model together with its derivative parameters using a composite neural network is presented.So far neural networks have never been used to extract large-signal device model accounting for distortion parameters.Applying this method to FET devices leads to nonlinear models for current- voltage functions which allow improved prediction of weak and mildly device nonlinearities in the whole bias region. The resulting models have demonstrated to be suitable for both small-signal and large-signal analyses,including intermodulation distortion prediction.