desihub / nightwatch

DESI Nightwatch: online data quality assurance
BSD 3-Clause "New" or "Revised" License
2 stars 2 forks source link

Read noise errors sometimes triggered due to CTE #425

Open djschlegel opened 3 months ago

djschlegel commented 3 months ago

The CTE problem can trigger read noise errors in NightWatch. Recent examples from the night 07-23-2024 had two backup tiles that triggered this error on the z1-D amplifiers (exposures 245014, 245021). I suspect the read noise is computed in the overscan region, and is thrown off by the CTE when there is a bright object in fibers near the overscan. (Is this interpretation correct?)

Options?

(1) Do nothing except in documentation. This is triggering off a real problem (CTE), even if the error is not obvious. This could be added to the NightWatch documentation. (One other comment is that most of the documentation is on the "Troubleshooting" page linked off NightWatch, rather than the "Documentation" page.)

(2) Run the CTE correction code within NightWatch to correct for the CTE before measuring the read noise.

(3) Explicitly trigger on the CTE.

djschlegel commented 3 months ago

Screenshot 2024-07-25 at 7 29 19 AM

sybenzvi commented 3 months ago

@djschlegel, I'd like to try option 2 (run the CTE correction code in Nightwatch) if that is possible. I'm ignorant of where in the processing the correction is happening so this could be easier said than done. We can discuss at the next survey-ops call.

I have the impression this is not a common error -- in the night you highlighted, there are only two exposures affected unless I missed something. But, I will add some documentation to the Nightwatch troubleshooting wiki page.

schlafly commented 3 months ago

The CTE correction code relies on calibration parameters that are fit from the analysis of the flat field exposures. We would have to bring that processing over to nightwatch in order to use it. Alternatively, we could do something like use the CTE parameters from the last good night of data from the offline pipeline, which sounds easier to me but would require some extra bookkeeping. David, would you mind elaborating on what the major features in your plot are? Where is the overscan region? I'm not accustomed to looking at these raw (?) files.

sbailey commented 2 months ago

@schlafly can you also clarify whether applying the CTE corrections require the full Horne 1986 row-by-row optimal extractions, or whether that is just for the measurement of the CTE parameters? I'm concerned that their could be a substantial processing time cost to including CTE corrections in Nightwatch.

schlafly commented 2 months ago

Yes, the spectra are extracted a handful of times using the full row-by-row optimal extractions. It probably takes O(1 minute) / CCD in total.