Removal of non-stationary noise from experimental data

In this paper we present two-step method for denoising. The noise variance is changed in random moments of time. The first step of our method consists of clustering wavelet coefficients which are computed by maximal overlap discrete wavelet transform. Then one defines of noise parts with piecewise constant variance. The second step consists of calculation of the modified thresholds for each part and removing noise on basis of these thresholds. The approach practically completely allows to remove local noise in comparison with traditional methods of denoising in which the threshold calculate on all sample of data..