Threshold selection for wavelet de-noising of partial discharge data

UHF Partial discharge (PD) measurements taken on-site are frequently affected by noise due to external disturbances. In order to locate the discharge source, the PD-signals have to be accurately processed to calculate the arrival time in each sensor. In this paper a wavelet based de-noising technique is used to isolate the PD signals from the noise. The technique utilizes the one-dimensional discrete wavelet transform and soft and hard thresholding are compared. Crucial is the choice of threshold level, distinguishing between coefficients related with noise and those associated with the PD signal. Four threshold criteria: the Stein's unbiased risk estimator, its Heuristic variance, the Universal logarithmic threshold and the Minimax are investigated. Finally, the technique is tested on PD signals detected in a power transformer.