Random Noise Attenuation for Seismic Data by Multiscale Time-Frequency Peak Filtering via Nonsubsampled Pyramid

Time-Frequency Peak Filtering (TFPF) is an effective method to eliminate pervasive random noise when seismic signals are analyzed. TFPF can give an unbiased estimation in case that the signal is linear. In general, we use pseudo Wigner-Ville distribution (PWVD) to realize the local linearity. However, there is a pair of contradiction in this method. If we choose a short window length (WL) for PWVD in the TFPF, it leads to good preservation for signal amplitude, but the denoising performance is relatively poor. So a fixed WL cannot solve the contradiction between the noise attenuation and signal preservation. To solve the problem, we adopt a nonsubsampled pyramid (NSP) to decompose the seismic data into multiscale components from low to high frequency. Then we can apply a short WL in signal-dominant scale to preserve the signal and a long WL is chosen for noise-dominant scale by the TFPF to eliminate more noise. We test the performance of our new method on both synthetic and real seismic data. According to the results, the Multiscale TFPF based on nonsubsampled pyramid can more effectively improve the signal-to-noise ratio and preserve events better than conventional TFPF.