Next: Synthetic Test: Data Interpolation
Up: Theory of noise attenuation
Previous: Sparse inversion
Applying a mask in equation (6) eliminates the contribution
of the empty traces in the model space, making them invisible to the
inversion. Therefore, by simply remodeling a data panel from
the estimated model
after inversion without the mask,
the missing traces are reconstructed. Then, the interpolated data
vector
can be estimated as follows:
| ![\begin{displaymath}
{\bf d_{int}}={\bf d}+{\bf (I-M)L\hat{m}},\end{displaymath}](img20.gif) |
(9) |
where I is the identity matrix. Now, for the noise removal, we
simply (1) apply a mute
in the radon domain that isolates and
preserves the signal, and (2) transform the muted panel in the data
space as follows:
| ![\begin{displaymath}
{\bf n_{est}}={\bf MLK\hat{m}},\end{displaymath}](img22.gif) |
(10) |
where
is the estimated signal (specular reflections and impinging source).
The estimated noise
(diffracted energy and ambient noise) is
obtained by subtracting the estimated signal from the input data:
| ![\begin{displaymath}
{\bf s_{est}}={\bf d-MLK\hat{m}}.\end{displaymath}](img24.gif) |
(11) |
Note that the estimated noise and signal in equations (10) and
(11) are for the non-interpolated data. To compute the
estimated noise and signal for the interpolated data, M must be removed
in equations (10) and (11) and d must be
replaced by
in equation (11).
Next: Synthetic Test: Data Interpolation
Up: Theory of noise attenuation
Previous: Sparse inversion
Stanford Exploration Project
5/3/2005