De-noising technique based on wavelet decomposition for impulse voltage measurements and noise analysis

This paper describes a novel approach to estimate the mean-curve of impulse voltage waveforms that are recorded during impulse voltage tests. The waveforms measured in practice contain oscillations and overshoots due to contribution of different noise sources. In this sense, usage of automated signal analysis tools that determine the important parameters of the impulse waveform such as peak value, front time, tail time etc. is very useful. This paper presents a noise analysis approach that is based on multi-resolution signal decomposition and statistical analysis for high-voltage impulse measurements. As the results of this analysis, the effective noise peaks are shown at approximately frequencies of 2.3, 17, 30 and 35 MHz. Also the effect of electromagnetic disturbance is observed around 2 MHz and noise components which are higher than 10 MHz are related to digitizers in the test hall. In this study, these noise parts are separated from the mean curve using the multi-resolution wavelet analysis and then, the noise spectra are given to define the characteristic peaks. Consequently, common properties of the spectra, which are independent from the electrode system, reflect the similar peak values at the specific frequency values. Hence, this research presents a new possibility for de-noising of the measurements.