Abstract
Modeling the electroencephalography signal response of a cortical could provide a very highly useful tool for monitoring and/or predicting diseases such as apoplexy, spinal cord injury or progressive nervous system disorder. However, these signals are commonly low in power, feature high signal to noise ratios and are extremely nonlinear. In this paper, a nonlinear system identification parametric approach is adopted using a Wiener Box-Jenkins model structure, where a linear dynamic subsystem in cascade with nonlinearity, is followed to estimate an empirical model including colored noise model added after the static nonlinearity. Prediction Error Method-based techniques are applied to predict the model, where simple Instrumental Variable method is used for the initial estimate. Separable Least Square approach is applied to identify the model, where only the nonlinear coefficients at the predictors are solved using the Hessian iterative optimization techniques; subsequently, the linear parameters are fitted using a simple Least Square approach. Finally, MATLAB simulation examples including benchmark data are provided.