Inversion deblending in a sparse transformed domain is an important approach to obtain high quality deblended data for simultaneous source acquisition. For ocean bottom node data, more than one vessel with multiple sources are used to reduce acquisition duration and increase shot density, but the increased blending fold tends to make the inversion solution unstable and less accurate. We present a new method and some tips to improve the stability and accuracy of inversion deblending for node data and demonstrate application on two different surveys.

This new method uses sub-L1 norm for regularization in the objective function, the Iterative Shrinkage-Thresholding Algorithm (ISTA) to solve for the deblended model, and time-variant local 3D FK transform to promote sparsity. Some pointers for improving deblending performance on field data are provided as well.