Adaptive nonlinear prediction with state reduction

The signal subspace technique for state reduction in nonlinear Volterra series (VS) and radial basis function (RBF) predictors are examined. The concept of applying signal subspace techniques to nonlinear prediction problems was first presented by Mulgrew et al. (see IEE Colloquium on Adaptive Filters, 1991). Since then, two alternative approaches (the indirect method and the direct method) have been developed. Results are presented which demonstrate the effectiveness of these techniques when applied to the prediction of chaotic time series