Reduced complexity Volterra models for nonlinear system identification (Q5957455)

The identification problem of nonlinear dynamical systems represented by a discrete-time time-invariant truncated Volterra model is considered. To reduce model complexity (the number of free model parameters), an approximate fixed pole expansion technique is developed, employing truncated orthonormal approximations of the Volterra kernels. In the approach, the specific orthonormal basis functions, in the \(z\)-domain, depend on the selection of the pole locations in a set of pre-specified transfer functions of linear stable filters. A recursive gradient-descent procedure for optimum selection of the unknown parameters and fixed pole locations in the approximate model for given input-output measurement data is proposed. The approach is illustrated by means of simulation examples.