Suboptimal identification of nonlinear ARMA models using an orthogonality approach

Proposes a scheme based on orthogonal projection to identify a class of nonlinear auto-regressive, moving-average (NARMA) models. The scheme decouples the nonlinear and linear identification problems, and hence there are two steps. The first step extracts nonlinearities for each delay element within the model via conditional expectations. The second step evaluates dispersion functions to weight the nonlinear functions so that the cost is minimized. This paper focuses on the second step of the proposed scheme. The characteristics of the identification scheme are studied, and simulations are provided.