Smoothing the canonical piecewise-linear model: an efficient and derivable large-signal model for MESFET/HEMT transistors

In this paper we present the smoothed piecewise-linear (SPWL) model as a useful tool in the device modeling field. The SPWL model is an extension of the well-known canonical piecewise-linear model proposed by Chua, which substitutes the abrupt absolute value function for a smoothing function (the logarithm of hyperbolic cosine), thus providing the model with several interesting properties. In particular, this function makes the model derivable, which is important to predict the intermodulation distortion behavior. Moreover, it allows one to control the smoothness of the global model by means of a single smoothing parameter. The parameters of the model are adapted to fit the nonlinear function, while the smoothing parameter is selected according to derivative constraints. The applied learning algorithm is a second-order gradient method. The proposed SPWL model is successfully applied to model a microwave HEMT transistor under optical illumination using real measurements. The model receives as input the bias voltages of the transistor, the instantaneous voltages, and the optical power and provides the drain to source current. The performance and computational burden of the SPWL model is compared with an empirical model and with some neural networks-based alternatives