In the previous chapter, I showed a method for separating signal and noise with an inversion technique. In that chapter, it was assumed that the data was completely available and the only problem was the separation of signal and noise on samples organized on a regular grid. For much prestack data, although the data is generally still organized on a regular grid, at least some of the data is missing. In addition to the data not recorded, some samples are so contaminated with noise that they must be ignored, especially when using a least-squares inversion technique. A method for removing these samples was shown in chapter . This removal of bad data creates more samples that are effectively missing.
This chapter addresses the issue of predicting missing data while separating signal and noise by inversion.