Sparsity-enabled ground-roll noise suppression using tunable Q-factor wavelet transform

Low-frequency oscillatory ground-roll noise is regarded as a type of main regular interference waves that obscures primary reflection information in land seismic data. Suppressing ground-roll reasonably can improve the signal-to-noise ratio of seismic data. Conventional suppression methods, such as high-pass and various f-k filtering, usually cause waveform distortions and body wave information missing owing to the simple cut-offoperation. In this abstract, sparse representation of signals based on morphological component analysis (MCA) theory is a new attenuate approach according to the oscillatory behavior of the signal rather than the scale or frequency. Tunable Q-factor wavelet transform (TQWT) with specified Q-factor is employed to represent two signals sparsely. Body waves are low-oscillatory and the corresponding wavelet transform sparse dictionary should have a low Q-factor, which is quite different from high Q-factor dictionary corresponding to ground-roll. Thus, seismic data including body waves and ground-roll can be decomposed into low-oscillatory and high-oscillatory components nonlinearly. Both synthetic and field shot data tests prove the effectiveness of this technique in the perfect preservation of waveform characteristics and frequency bandwidth of reflections.