In order to simultaneously suppress the periodic narrow-band noise and random white noise in ultra-high frequency（UHF） partial discharge（PD） signal

by analyzing the spectrum characteristics of the UHF PD signal

we have found that the spectrum of PD pulse possesses the characteristics of reaching its maximum value at a certain frequency and decaying gradually toward the edge. The amplitude of the spectrum at its center frequency is significantly higher than that in the side-band section

and it is located at the local maximum point of the UHF signal spectrum. Thus

we proposed a novel method based on the improved Protrugram and wavelet transform to suppress the above noise in UHF PD signal. In the proposed method

firstly

the center frequency of the PD pulse is recognized in frequency domain; then

the corresponding bandwidth of the center frequency is obtained by self-adapting using kurtosis as an indicator

realizing the frequency spectrum locating of PD pulse; finally

the redundant white noise is eliminated using a wavelet de-noising method

realizing the effective suppression of multiple types of noise. The simulation and practical tests show that the proposed method has better performances not only in noise suppression but also in useful signal reserve compared with the traditional wavelet de-noising method and the ensemble empirical mode deposition（EEMD） adaptive threshold method. The research can provide reference for the de-noising of the UHF PD signal.