Blended De-signature - A New Approach to Source Separation
Margherita Maraschini, S. Grion · Proceedings · 2017
Summary A blended dataset is nothing but an unblended dataset generated by a blended source, and if the source signature of this blended source is known, the dataset can be de-blended by deconvolution. This paper illustrates a new methodology that uses this idea to perform source separation when the source signatures of the blended sources are different and known. This methodology does not require random delays, and can be used jointly with other de-blending methods. The effectiveness of the suggested methodology, alone or combined with other de-blending steps, is demonstrated using a synthetically blended real dataset obtained combining datasets from two different sources firing on the same streamer. The results suggest that, for a similar level of residual overlap energy, the application of de-signature before other deblending steps improves the signal preservation of the whole de-blending processing.