This step is not usually necessary for either workflow but is included for completeness.
The DREAM observing mode moves the secondary mirror in a pattern which causes adjacent bolometers to observe the same region of the sky multiple times [3,6]. The data are then described by a series of simultaneous equations which are solved via a matrix inversion (using singular value decomposition). Since the pattern traced out by the secondary mirror is known, the inverse of the matrix can be pre-calculated (a time-consuming task) for a given output grid and applied to data at the time of observation (quick).
It is strongly recommended that users become familiar with the details of DREAM  before using weights other than the default for science data processing.
The user may wish to experiment with different regridding schemes or
there may be other reasons, not known at the time of observation,
which require the inverse matrix to be re-calculated. The task for
doing this is called dreamweights, and this allows the user to
specify different output grid parameters. The calculation of the
inverse is moderately time-consuming (compared with calculating the
images) and takes of order 5-10 minutes on current hardware. The
output file from dreamweights is the weights file required by
dreamsolve. KAPPA fitsedit may be used to specify weights files
which differ in name from the default. Note that this process must be
repeated for each subarray for which data exist and that data which
are to be combined should all use the same grid.
SMURF -- the Sub-Millimetre User Reduction Facility