Blind Identification of Seismic Signals

Blind Identification of Seismic Signals

Blind identification of seismic signals

Mirko van der Baan1, Christian Jutten2, Anthony Larue2, Jérôme Mars2, Dinh Tuan Pham3 and J-Michael Kendall1

(1) School of Earth and Environment, Earth Sciences, Univ. of Leeds, Leeds LS2 9JT, UK.

(2) Laboratoire des Images et des Signaux, National Polytechnic Institute of Grenoble, Grenoble, France.

(3) Institute of Informatics and applied Mathematics, Grenoble, France.

Blind deconvolution and independent component analysis (ICA) are rapidly emerging promising new technologies in the field of advanced signal processing. Blind deconvolution tries to jointly estimate the unknown source wavelet and solve the deconvolution problem. ICA tries to determine the statistically most independent signals in an instantaneous mixture of superposed signals thereby seeking to retrieve the original input signals.

Blind deconvolution and ICA distinguish themselves from conventional deconvolution and source separation techniques in that they appeal to higher-order statistics, while conventional techniques only evaluate second-order statistics (e.g., auto-/cross-correlation and variance). Examples of conventional source separation techniques include principal component analysis and singular value decomposition.

The use of higher-order statistics are both their main advantage and their principal inconvenience. Higher-order statistics reveal significantly more about the characteristics of signals than the mere application of second-order statistics. On the other hand, higher-order statistics are considerably more sensitive to the presence of noise (outliers) and often instable for short time series. Techniques for robust implementation of

ICA and blind deconvolution are therefore currently actively researched.

Potential applications of blind deconvolution and ICA in seismic data processing are manifold. Blind deconvolution actually evolved from the so-called sparse deconvolution algorithms. The latter enforce that retrieved reflection series are composed of individual spikes thus significantly boosting the resolution of seismic data. Blind deconvolution shares this feature with sparse deconvolution. In addition, it automatically estimates both the amplitude and phase characteristics of the unknown wavelet. Phase estimation is not possible with conventional algorithms based on second-order statistics (e.g., predictive deconvolution). ICA has for instance already been successfully applied to separate up-and down-going wavefields in VSP experiments with irregular receiver spacing.

Blind deconvolution and ICA therefore represent promising new avenues to increase the signal-to-noise ratio of seismic data and thereby their resolution. An enhanced resolution is critical for successful exploration of in particular new satellite fields in the North Sea and to improve confidence in volume estimates, well targets and production profiles in general.