A New Method for Variance Estimation of White Noise Corrupting a Signal

Presented in this paper is a new method for estimating the variance of white noise corrupting a signal using Donoho and Johnstone (D-J) threshold. For piecewise smooth signals with limited number of incontinuities, the new method is insensitive to outliers, i.e., the large value coefficients associated to signal incontinuities, and is of higher efficiency than the median absolute deviation (MAD)-based variance estimator. Evaluation of the new method is implemented using a simulated piecewise smooth signal and a real differential pressure signal. Comparisons with the MAD-based variance estimator are also given