sbailey
commented
3 years ago I made new bias+dark models for r2, r6, and z6, which were missing from desi_spectro_calib/0.2.1 used by blanc so far, using darks from 20201202, 1209, 1210. New calibs are in /global/cfs/cdirs/desi/users/sjbailey/spectro/desi_spectro_calib/trunk
Applying this to a 300s dark on 20201218 and looking at the vertical profiles (i.e. averaged across columns for left and right amps) shows that it improves the flatness:
Note that the r6 and z6 have some noisy amplifiers with big horizontal stripey structure (but not in a manner that is amenable to row-by-row subtraction), thus resulting in the extra noise on those plots.
I then reran 20201214 exposures 67744 67765 67766 67767 67768 using the new calibrations and the pre-calculated psfnight and fiberflatnight from blanc (i.e. I didn't recalculate those). Results in /global/cfs/cdirs/desi/users/sjbailey/spectro/redux/darktest/.
It helps a bit, but there is still an offset problem between b and r. Comparing Mt Blanc with new calibs:
We're not using nightly biases yet, and that might help, but now that we're leaving VCCD on and stabilizing the CCDs before taking data, bias variations are much less than they were before and incorporating nightly biases will need to wait for the next release.
I think this is as much improvement as we'll get for Mt Blanc, but I'll discuss with @akremin and @julienguy before deleting and rerunning r2, r6, and z6 for Mt Blanc.
so perhaps we are chasing both a CCD calibration issue (mostly fixed) and pollution from a neighboring fiber (maybe an extraction bias, maybe a readout "shadow" effect). Unfortunately in this case fiber 3429 (read out before the bright star on fiber 3428) is completely masked, but we could seek out other cases of neighboring fibers to bright standards and compare offsets on the fiber before vs. after.
julienguy
Nathalie reported a discontinuity at the boundary between blue and red CCD in faint target spectra in SP6. This is traced back to an inaccurate bias+dark for r6 (and z6). We have a new model but it was not finalized when we started the Mt Blanc processing.
Most striking example:
This is fiber 3427 of SP6. It's next to a bright star in fiber 3428, but the discontinuity is seen in other neighboring fibers, like 3423 and 3424.
This is due to an over-subtraction of the sky in r6 which itself is due to a residual structured CCD background in r6 than can be fixed with a correct bias + dark model. We are in the process of refitting this bias + dark and rerunning the Mt Blanc prod with this fix for r6,z6 (but also r2).
The effect is barely visible in a single exposure, but rerunning for instance the exposures of tile 80607 for the night of 20201214 (expids: 67744 67765 67766 67767 67768) could be sufficient to detect if the problem is solved with the new bias+dark.