Hybrid Algorithms for Denoising Electrical Waveforms Containing Steady Segments

The paper deals with an original hybrid denoising algorithm applicable to waveforms containing steady segments, polluted by white noise. The algorithm relies on 2 denoising techniques. Firstly the “average signal method” is used to get an estimation of the noise power. An average signal (AS) is computed across 4 consecutive periods of the polluted signal (SP). A “per period” evaluation of the noise (NP) is then performed by subtracting AS from each period of SP. Each NP is divided into 6 equal subintervals, getting sets of 6 estimated noise signals (NS) for each period. The values of standard deviation (SD) are computed considering the absolute values of local peaks for all NS-s. 3 metrics were used to evaluate the power of estimated noises: NS’s power (PNS), SD and the average value of NS (ANS). The minimum value (P) of PNS across all subintervals from a dataset (DS) represents the estimated power of the noise associated to DS. P is used in the 2-nd stage of the algorithm – denoising with a wavelet based thrashing tree (TT). The TT’s number of levels is determined considering the best match between the power of noises yielded by TT and P. The algorithm was additionally validated by proving that all the minimum values of PNS appear only for subintervals associated to almost minimum values of SD and almost zero values of ANS. The original hybrid algorithm yielded reliable results for all datasets.