F-x Noise Attenuation by Empirical Mode Decomposition

This paper proposes a new frequency-offset (f-x) domain filtering technique for noise attenuation. The novelty is in the use of a nonlinear decomposition (EMD), instead of linear autoregressive prediction filtering. The motivation is to properly account for the nonlinear and nonstationary spatial properties of real data, and to handle irregular trace spacing. The so-called FX-EMD is equivalent to an adaptive, data-driven frequency-wavenumber (f-k) filter with a frequency-dependent, high-cut wavenumber filtering property. It is effective in removing both random and dipping noise, and displays good performance for irregularly spaced land data. It is also a parameter-free filtering technique in its simplest implementation.