A user has the following options to increase the whiteness of a coherency image:
Several of these options contain several individual parameters, e.g., the subvolume size is defined by 3 integer values that define the subvolume's number of samples along each axis. The user has to set about a dozen parameters. Some of these parameters have obvious limits.
I chose the pre-whitening PE filter to be a 1-D deconvolution filter in time. This should yield optimal spatial whitening, while preserving a coherent waveform along the time axis.
The second PE filter step involves three 2-D PE filters (one in each Cartesian plane). I choose these filters to be two columns wide, which limits them to the removal of a single planar event. I choose the number of rows large enough to be able to predict the data's steepest dipping feature.
PE filter coefficients are found from the data by iteratively solving a minimization problem. An increase in the number of iterations usually improves the filter's ability to predict the input data. Since the filters are usually small, I decided to use a sufficient number of iterations to find the optimal least square filter. However, limiting the number of iterations is a convenient way to decrease a filter's ability to whiten the output.
An individual subvolume has to be large enough to determine the unknown PE filter coefficients by minimizing the convolution output. The subvolume serves as an expectation volume of a locally stationary signal.
Despite these obvious limits to the individual parameters, a twelve dimensional parameter space is very hard to explore exhaustively. The parameters are probably interdependent. For example, decreasing the individual subvolume size has an effect similar to increasing the number of iterations for finding the filter; both yield a more accurate filter estimation and therefore a whiter filter output. I tested several sets of parameters in the subsequent field data examples and I did not succeed creating a satisfying coherency image. The images that I show in the next section are representative.