Nonlinear system identification using discrete laguerre functions

Volterra functional series approximations are well known techniques for nonlinear system identification. These structures, while quire general, suffer from difficulties in parameterising the kernels, and may require a large number of parameters for models higher than second order. Discrete Laguerre filters and other orthogonal filters have been re-considered recently for both linear and nonlinear system identification. A class of multilayer perceptrons incorporating an orthogonal filter preprocessing unit is introduced. Derivations for parameter estimation of these models are given, including a consideration of both off-line and on-line methods for estimating the Laguerre pole. The paper concludes with a brief discussion of some open issues concerning such signal models.