Noniterative f-x-y streaming prediction filtering for random noise attenuation on seismic data

Discussions

The frequency-space PFs shows different characteristics from the time-space PFs that depend on the sample rate along time direction. PFs in time-space domain cannot predict the seismic events from all directions (with all kinds of slopes) because the whole trace length is seldom chosen as the filter size along the time direction. The frequency method can avoid this problem, and it may get an accuracy prediction from the complex seismic data. However, computation in frequency domain may result in artificial effect, so we designed the local smoothness constraint to reduce the effect.

We highlighted the computational efficiency of the proposed $f$ -$x$ -$y$ SPF because of not only the noniterative property but also the memory saving ability. Assuming that a conventional 3D data volume 500 (frequency) $\times$ 500 (space $x$ ) $\times$ 500 (space $y$ ) is about 1 gigabyte with complex number format, then the $f$ -$x$ -$y$ RNA with 5-sample (X) $\times$ 5-sample (Y) filter structure needs over 23 gigabytes memory space to cache the filter coefficients for data calculation, and the data exchange time is much more than that of iterative calculation. The proposed $f$ -$x$ -$y$ SPF needs only about 50 megabytes to cache filter coefficients when calculating.

Noniterative f-x-y streaming prediction filtering for random noise attenuation on seismic data