Frequency response function estimation via a robust wavelet de-noising method

The FRF (frequency response function) estimation can be performed by the vi bration analysis of a linear time-invariant dynamic system. Since a single FRF estimate is highly sensitive to measurement errors of input/output sign als, the mean averaging of repeatedly observed FRF estimates is employed in most of the practical applications. The main result of this work is in red ucing the number of averaging operations and enhancing estimation accuracy by using a robust wavelet de-noising method. This approach removes outliers and zero-mean Gaussian noise simultaneously and effectively while preservi ng most of the important signal features of a true FRF with a dramatically smaller number of operations as compared with the traditional mean-averagin g procedure. The robust wavelet de-noising method is based on a wavelet-rel ated median filtering and a wavelet shrinkage to reduce the effect of outli ers and zero-mean Gaussian noise respectively. The effectiveness of the pre sent FRF estimation technique is demonstrated using both simulated and expe rimental data. (C) 2001 Academic Press.