Abstract

An identification method for time-varying transfer function models is proposed by using wavelet-based periodograms of input/output data. The instantaneous periodograms are estimated by directly making wavelet transforms of input/output data. The time-varying frequency transfer function is identified from the ratio of input/output periodograms. It is shown that the proposed wavelet-based method is effective even when the output is contaminated with noise due to the band-pass filtering characteristics in the analyzing wavelet. In addition, it is demonstrated through simulation results that this method is more advantageous for its control application than the Blackman-Tukey based method that has already been proposed.