On-line Identification of Time-varying Processes at Various Signal-to-Noise Ratios

New results in the on-line identification of time-varying dynamicprocesses are presented. The performance of the weighted least-squaresidentification algorithm, for different values of signal-to-noiseratio (SNR), is investigated with reference to artificial recordsfrom Monte Carlo tests. It is obvious that a higher SNR willresult in a good estimated model. This paper discusses whetherit stays as easy at low SNR to find the model structure accuratelyas it is at high SNR.