Tile-Low Rank Compressed Multi-Dimensional Convolution and Its Application to Seismic Redatuming Problems

Ravasi, Matteo; Hong, Y.; Ltaief, Hatem; Keyes, David E.

Tile-Low Rank Compressed Multi-Dimensional Convolution and Its Application to Seismic Redatuming Problems

Files

EAGE_2022_MDCTLR.pdf (4.26 MB)

Type

Conference Paper

Authors

Ravasi, Matteo

Hong, Y.
Ltaief, Hatem

Keyes, David E.

KAUST Department

Applied Mathematics and Computational Science Program
Computer, Electrical and Mathematical Science and Engineering (CEMSE) Division
Earth Science and Engineering Program
Extreme Computing Research Center
King Abdullah University Of Science And Technology
King Abdullah University of Science and Technology
Office of the President
Physical Science and Engineering (PSE) Division

Date

2022

Abstract

A variety of algorithms in seismic processing and imaging rely on the repeated evaluation of a multidimensional integral of convolution (or correlation) type. This operator is notoriously expensive due to the fact that it inherently requires accessing the entire seismic reflection response to perform a batched matrix-vector (or matrix-matrix) multiplication. In this work, we propose to alleviate this memory and computational burden by leveraging data sparsity in the frequency-domain and using Tile Low-Rank (TLR) matrix approximation. We also show that a geographically aware re-arrangement of the rows and columns of the kernel of the operator can further boost the compression capabilities of the TLR algorithm with minimal impact on the quality of the processing outcome. A synthetic example of 3D Marchenko redatuming is used to validate the proposed strategies.

Citation

Ravasi, M., Hong, Y., Ltaief, H., & Keyes, D. (2022). Tile-Low Rank Compressed Multi-Dimensional Convolution and Its Application to Seismic Redatuming Problems. 83rd EAGE Annual Conference & Exhibition. https://doi.org/10.3997/2214-4609.202210253

Acknowledgements

The authors thank KAUST for funding this work. For computer time, this research used the resources of the Supercomputing Laboratory at KAUST in Thuwal, Saudi Arabia.