Open julienguy opened 2 years ago
Here is an example from a recent night where we have large bias instability in b4.
https://nightwatch.desi.lbl.gov/20211215/00114120/preproc-b4-00114120-4x.html (click prev
next
on the web page to see the time variation). The sky subtracted frames present large steps +- 10 elec/A (!) in the spectra caused by the discontinuity at the boundary between amplifiers.
plot_frame -i $DESI_SPECTRO_REDUX/daily/exposures/20211215/00114225/sframe-b4-00114225.fits --sky-fibers --rebin 100
This step is interpreted as Balmer breaks (positive or negative!) by redrock and results in pile-ups of incorrect redshift at z~0.25 for sky fibers (see https://data.desi.lbl.gov/desi/spectro/redux/daily/nightqa/20211215/skyzfiber-20211215.png ). The effect is however quite small for science fibers.
In PR #1551 , I consider subtracting this background at the preprocessing level (before extraction and sky subtraction).
At the end of the sky subtraction fit, we could consider fitting for a residual spectral background per CCD amplifier, either a constant term or a linear function of wavelength and a polynomial function of fiber (or CCD X coordinate). This should allow us to correct some of the issues related to fluctuating CCD bias.