Simultaneous CRS Parameters Search Based on a Non-linear Conjugate Gradient Method with Preconditioning

Multi-parameter methods, e.g., Common-Reflection-Surface (CRS) stacking, have become popular because they exploit increased data fold compared to conventional CMP processing, resulting in an improved signal-to-noise ratio and, thus, enhanced images. However, these methods require additional effort since they incorporate more stacking parameters than CMP processing. The simultaneous brute-force search for all parameters is still not common because the increase in computer performance is countered by an increasing amount of acquired data (the amount grows faster than Moore’s law). An optimization method can therefore serve as an alternative to the global (brute-force) approach since it reduces the number of evaluations of the coherence functional (e.g., semblance) significantly. I propose to use a non-linear conjugate gradient method to estimate the CRS attributes. The method utilizes the secant method, the Polak-Ribiere formula, and preconditioning. I applied this approach to a complex marine dataset which led to more focused stack results than using the traditional pragmatic CRS attribute search.