Next: REFERENCES
Up: R. Clapp: Data precision
Previous: FPGA
In this paper I showed that data at 32 bit floating bit precision is
not needed to form an accurate migration image. It takes only four
bits to form an acceptable zero offset image and six bits for a prestack image.
The representation of the wavefield within the migration
can also be reduced by at least a factor of two. These results
offer promise in making external hardware processing solutions, such
as FPGAs, GPUs, and the cell processor, effective in the near term.
Next: REFERENCES
Up: R. Clapp: Data precision
Previous: FPGA
Stanford Exploration Project
1/16/2007