System identification using discrete orthogonal functions

In this paper a novel method is described for the design of adaptive IIR filters used in system identification. the adaptive filter is implemented as a parallel connection of subsections whose transfer functions constitute a set of discrete orthogonal systems. the adaptation algorithm used, which is of the Gauss-Newton type, adjusts the parameters of these discrete orthogonal systems in order to match the desired output data of the unknown plant in a least squares sense. Owing to the orthogonality property, which ensures complete independence of subsequent sections, convergence is very rapid. Closed-form expressions for the gradient signals required to update the filter are given. Illustrative examples have shown that this method always results in much improved adaptation properties compared with the already existing approaches.