One of the most important parameters in our algorithm is the sampling
in the pyramid domain. We find that for interpolating
irregularly-spaced data, half of the value of given
in equation 11 is usually enough. For interpolating
aliased data, however, the sampling becomes more difficult to
establish. In this case, the sampling becomes smaller as the maximum
non-aliased frequency becomes smaller. In practice, we tend to find the
right sampling by trial and error, starting from the value we would
use for the irregularly-spaced data case. This approach is unsatisfying and
we feel that more work should be done to automate this important
parameter selection.