Research On The High Fidelity Deblending Of Multi-source Blended Seismic Acquisition Wavefield

Seismic acquisition is a trade-off between economic and quality. Conventional acquisitionsurveys are designed such that the time intervals between successive shots are sufficientlylarge to avoid the interference between different source responses. However in thissituation both the image quality and measure parameters, such as azimuths and samplingdensity may not be achieve the best. Unfortunately the improvement of seismic acquisitionmethod is fall behind other technologies in seismic processing. Recently a great change ofmindset in seismic acquisition has occurred which is known as blended acquisition orsimultaneous acquisition, where different sources at different locations are shot in anoverlapping fashion. Both which significantly improved the acquisition efficiency andpotentially image quality can be significantly improved.Deblending is the procedure of recovering data as if they were acquired in theconventional survey. From the mathematics point of view, deblending can be regarded assolving an underdetermined equation such as Ax=b, yet a simple least-squares procedurecannot remove the blending noise. Focus on the2D regularized acquisition system, wecombine the blended data in common source domain and common receiver domain underthe umbrella of reciprocity theorem, In this paper we focus on the blended acquisitionsystem instead of the inversion algorithm for the purpose of solving this ill posed inversionproblem. According to our derivation in the regularized2D acquisition system, the blendeddata in common receiver domain can be connected with that in common source domain,which making this inversion problem the above mentioned equation less moreunderdetermined. When applied to a numerically blended dataset, reflection coefficients ofsingle source gather is calculated using a simple spgL1norm basic pursuit sparse inversionalgorithm at first, and then separation results are acquired via convolution. Field data testshows that our method can get high quality separation results, which verified our theoryand derivation.While dealing with blended seismic acquisition data, a simple least-squares procedure isonly able to get pseudodeblend results, where the blending noises cannot be removed. Inhigh blending factor data, the blending noises are usually several times higher than theuseful signals, which multiply the difficulty of source separation. Fortunately in pseudodeblend records, these noises are only coherent in the common source domain, butincoherent in other domain. For this character, multilevel median filter and Curveletthreshold iteration denoising are combined used in this paper, and a new source separationmethod based on the iterative denoising in different domain is introduced: Focus on thedifferent character of blendiong noises in different domain, corresponding denoisingmethod is utilized, and an iteration method is designed for the optimization. While dealingwith numerical blended real dataset, ideal results could be produced after only a fewiterations, which verify that our method can largely improve the separation quality andcalculation efficiency.Multi-source simultaneous acquisition has brought a revolution of seismic data acquisitionmethods, from which many seismic data processing methods will benefit. This acquisitionmethod is applied to the virtual source method. Compare the virtual source gathersconstructed by the conventional acquisition method and the simultaneous method we canfind that: the virtual source records builded from the separation results of blended data aremuch more closed to the synthetic ones.