A Graphics Processing Unit Implementation for Time-Frequency Phase-Weighted Stacking

Stacking is an efficient approach to increase signal-to-noise ratio, which is a key issue in seismic data processing. The time-frequency domain-phase weighted stack (tf-PWS) that uses coherency of instantaneous phase as a weighting function can significantly improve the stacked signal quality of many data-sets. A graphics processing unit implementation was developed to reduce the heavy computational cost of tf-PWS. Synthetic tests suggest the speed-up factor is up to 20. Our real-data test shows that the convergence of noise cross-correlation functions can be substantially improved by tf-PWS without a computational cost increase.