Randomized Signal Classes for Evaluating the Performance of Wavelet Shrinkage Denoising Methods

Previous simulation experiments for the comparison of wavelet shrinkage denoising methods have used fixed signal classes defined by adding instances of noise to a single test signal. New simulation experiments are reported here with randomized signal classes defined by adding instances of noise to instances of randomized test signals. As expected, significantly greater variability in the performance of the denoising methods was observed. Statistically valid comparisons must be conducted with respect to this variability. Use of randomized, rather than fixed, signal classes should yield more realistic and meaningful results.∗