Seismic data decomposition into spectral components using regularized nonstationary autoregression

Conclusions

I have presented a constructive approach to decomposing seismic data into spectral components with smoothly variable frequencies and smoothly variable amplitudes. The output of the proposed algorithm is close to that of empirical model decomposition (EMD) and related techniques, such as the synchrosqueezing transform (SST), but with a more explicit control on parameters and more direct access to instantaneous-frequency and amplitude attributes. The main tool for the task is regularized nonstationary regression (RNR), which is applied twice: first to estimate local frequencies by autoregression (RNAR) and then to estimate local amplitudes. Although all examples shown in this paper use only 1D analysis, the proposed technique is also applicable to analyzing 2D or 3D variable-slope seismic events in the $f$-$x$ domain. Potential applications may include noise attenuation, data compression, and data regularization.