Nonstationary pattern-based signal-noise separation using adaptive prediction-error filter

Conclusion

We introduced a new pattern-based approach for nonstationary signal-noise separation. Our method used the APEF as the pattern operator, which was suitable for characterizing the nonstationary properties of seismic data and noise in the time-space domain. After calculating the data pattern $\mathbf{D}$ and the noise pattern $\mathbf{N}$ , we could separate the signal and noise by solving a constrained least-squares problem. We adopted different algorithms to deal with the random noise and ground-roll noise separation problem. Numerical examples showed that the proposed method provided a robust signal-noise separation, even in the presence of random noise with nonstationary energy distribution and strongly curved ground roll. Multiple suppression and diffraction separation were also other applications of this method.