Detection and Estimation of Heteroscedastic Noise by Means of the Wavelet Transform

Abstract

The ability of the wavelet transform (WT) to detect and estimate heteroscedastic noise was determined. The WT provides information in both the time and the frequency domains which is necessary for the detection of heteroscedastic noise. By a simple F-test applied to non-overlapping intervals of wavelet coefficients differences in the variance can be detected. By fitting a smoothed version of the signal versus a smoothed estimate of the local variance the proportionality factor can be estimated.