Evaluation of Nonlinear ARX System Identification Technique on Modeling Crosstalk

Estimating crosstalk appropriately is very important in the process of mitigating electromagnetic interference. This study evaluates a black-box modeling technique named nonlinear autoregressive with exogenous inputs (NARX) on crosstalk modeling application, especially crosstalk due to random pulse width modulation. The model is developed using the input and output data from the measurement as regressor inputs. For validation, the mean squared error of this model is calculated by comparing the model output with the real measurement output. For evaluation, the model performance also was compared to a Spice-based SACAMOS model. Although less flexible than the Spice model, NARX model can represent the signal on the victim cable well with a small mean squared error value.