NARX approach to black-box modeling of circuit elements

This paper deals with the identification of NARX (Nonlinear AutoRegression with eXtra input) models for the numerical simulation of circuits containing nonlinear dynamic elements. NARX identification, based on a sequence of input/output samples, is useful for black-box modeling and for the refinement of models of nonlinear circuit elements. In order to assess the suitability of such an approach, we apply it to a CMOS inverter gate and experiment with the main elements controlling the identification process. We obtain accurate models with relatively simple structure and observe reliable operation of the identification process, as well as a good insensitivity to the noise content of the output samples. Such results confirm that NARX identification could be a useful tool for circuit simulations.