AGENDA
(a.) NRMDR = 0, NTrim = 10 amounts to the previous symmetric trimmed averaging that could not eliminate the anti-persistence artifacts;
(b.) NRMDR = 20, NTrim = 0 amounts to the asymmetric trimming which did succeed in eliminating anti-persistence artifacts, but which may degrade low density scans without persistence;
(c.) NRMDR = 10, NTrim = 5, which will allow some asymmetric trimming where needed but should not degrade cases where it's not needed as much as NRMDR = 20, followed by symmetric trimming of the remaining samples.
The tests also include the new saturation processing. The first run indicates that something therein is not working, as no saturated pixels were reported by dflat, whereas it is known that an extremely bright object is in the scan. J. White and J. Fowler will examine the code and interfaces to determine where the problem lies. Preliminary work in this vein has shown that the saturation threshold may not be appropriate for the 2-byte integer values involved in the raw data. These are produced at the observatory as unsigned 16-bit integers, but the 16-bit integers in Sun FORTRAN are exclusively signed, necessitating special handling at the wrap-around point of 32767. Since the design no longer involves testing for exact (integer) equality of a pixel value with a saturation threshold (because thresholding at lower values to avoid significant nonlinearity has been adopted), the testing can be done in the REAL*4 domain. R. Cutri requested that the threshold for now be 3.0e4, since the indications are that hard saturation occurs somewhere between 3.3e4 and 4.0e4, and significant nonlinearity appears to set in around 3.0e4 in each band.