The DARKBASE (“Dark Baseline”) module subtracts a constant offset from all samples of each read-out channel of its input file, rowdroop.fits. The purpose of this module is to take out a “pedestal” or baseline in the signal of all ramps, so that the output ramps extrapolate to zero signal at the start of the DCE. The module first fits a slope to the second and third samples of each ramp; the first sample is not used because we have empirically found it to often be “depressed” with respect to the rest of the ramp slope. The computed slope is then used to extrapolate to the signal at t=0 (the beginning of the DCE). In principle, such signal indicates the needed offset to apply to the ramps. In practice, such signals and resulting offsets are calculated statistically over the array, as follows:
For each read-out channel of the IRS arrays, the trimmed mean of the extrapolated t=0 signal (baseline) is calculated for all pixels (ramps) located within the un-illuminated region of the array. This region is specified by the umask.fits file, and is unique to each instrument channel. The trimmed mean uses an 80% trim fraction, meaning that, among all the values to average, a subset is constructed by first sorting the values in ascending order, and then discarding 10% of the values that are the lowest in the sorted set, and 10% of those that are the highest.
After computing the baseline for each of the four read-out channels, these baselines are respectively subtracted from all pixels (all samples of the ramps) of the corresponding read-out channels.
The computation of baselines excludes pixels masked in the pmask, and saturated samples masked in the dmask. No baseline subtraction is done for any such pixels or samples.
The output files of this module are drk.fits and drku.fits.