AN ADAPTIVE RECURSIVE FILTER FOR AUTOREGRESSIVE INPUTS

A novel off-line procedure for IIR system identification problems has been recently proposed and analysed under the assumption of a white in put signal. The method happens to have a single, unbiased stationary p oint under this condition. This approach is based in interpolation the ory and tries to overcome the drawbacks of equation- and output-error strategies. Here it is shown that the unicity of the stationary point is still preserved for a wider class of input signals, namely autoregr essive processes of a certain order. With coloured input signals, an S PR condition arises for the convergence of the original on-line algori thm. A modification is proposed that eliminates this SPR condition, wh ile leaving the stationary point untouched. Two different structures t hat implement the on-line algorithm are presented; the use of a lattic e structure for the implementation of the recursive part of the filter is considered as well. (C) 1998 John Wiley & Sons, Ltd.