Project

General

Profile

Regression Tests

This page describes tests that will be done on a regular basis to monitor performance of DECam and the telescope, and provide calibration files needed for DESDM that are not already produced as part of the DESDM pipeline.

Each test's description should include the data to be collected to execute the test, the analysis algorithm & code to be applied to the data, and the actions to be taken on the results, e.g. thresholds for alarms and any calibration files that get forwarded to DESDM, CP, or elsewhere.

See the "Early Years Calibration Plan" (docdb 6584) for details of some of the data needed for some of these items.

The modus operandi will be that the regression tests run at NCSA (at least for biases, flats, PTC, etc. - for the optical and tracking tests, and DECal data, this needs to be worked out). Regression tests taken outside of DES observing time will be scripted to run with proposal ID #9993 ??? so the images get sent to NCSA.

In the list below, statistics to create from the tests are italicized. Alarms are boldfaced.

Daytime tests

Bias & read noise

(Daily or weekly)
  1. Take sequence of 20 bias exposures
  2. Subtract overscan line-by-line. Calculate median overscan level of each amplifier. Alarm overscan levels more than 50 ADU ??? from nominal value for amp.
  3. Create image containing sigma-clipped pixel-by-pixel mean and variance of the bias stack.
  4. Estimate read noise of each amplifier from sample of its pixels. Alarm on read noise change more than 10% ??? from nominal for each amplifier.
  5. Bad pixel: flag any pixel with variance <50% ??? of nominal value for that amplifier.
  6. Alarm on change of more than 10 ADU ??? in bias level from nominal in any unmasked pixel.

Dark Current tests

(Monthly)
  1. Take sequence of 20 dark frames of 90s each.
  2. Subtract overscan and the mean bias level from each.
  3. Create image containing sigma-clipped pixel-by-pixel mean and variance of the dark stack.
  4. Alarm on change of more than 1 ADU/sec ??? dark current from nominal in any unmasked pixel.

Bad pixel and gain tests

(Monthly)

  1. Take sequence of 20 dome flats (in g,r, i, or z) with ~200 ADU ??? exposure
  2. Subtract overscan and bias.
  3. Normalize each exposure to have median exposure equal to central exposure.
  4. Create image of sigma-clipped pixel-by-pixel mean and variance of the normalized stack.
  5. Bad pixels: flag as bad pixels any with <50% ??? of expected read + Poisson noise.
  6. Bad pixels: flag as low-level bad pixels any with variance more than 1.1x ??? value expected read+Poisson
  7. Bad pixels: flag as high-level bad pixels any with variance in excess of 1.3x ??? the value expected. Note that this means the pixel will increase by >20% the variance expected in a sky exposure with 1000 ADU.
  8. Calculate gain of each amplifier from ratio of mean to variance (averaging over sample of unmasked pixels). Alarm for change of >0.3% ??? from nominal gain.
  9. Some statistic to flag blocked or "funky" columns???
  1. Take sequence of 20 dome flats (in g, r, i, or z) with ~3000 ADU ??? exposure.
  2. Subtract overscan and bias.
  3. Normalize each exposure to have median exposure equal to central exposure.
  4. Create image of sigma-clipped pixel-by-pixel mean and variance of the normalized stack.
  5. Bad pixels: flag as bad pixels any with <90% or >110% ??? of expected read + Poisson noise.
  6. Bad pixels: flag as high-level bad pixels any with variance in excess of 1.1x ??? the value expected.
  7. Bad pixels: flag as "cold" any pixel with <90% ??? the local mean value.
  8. Calculate gain of each amplifier from ratio of mean to variance (averaging over sample of unmasked pixels). Alarm for change of >0.2% ??? from nominal gain.
  9. Some statistic to flag blocked or "funky" columns???

CTI tests

EPER measurements???

Flat-field stability

(Daily or weekly, repeat in each grizY filters)
  1. Take 10 high-level dome flats, ~8k ADU per pixel, ~400ke total per pixel.
  2. Subtract overscan and bias
  3. Normalize exposures to the mean flux of the central exposure.
  4. Calculate sigma-clipped pixel-by-pixel mean and variance of the dome stack.
  5. Calculate the median value for each amplifier relative to a reference amplifier. Alarm for change in amplifier gain ratio of >0.3% ??? from nominal.
  6. Divide each mean dome flat by reference flat. Now look for changes from reference flat (always ignoring all bad pixels):
    1. Alarm on change of any individual pixel by more than 2% ???
    2. Block-average the ratio image into 8x8 regions (ignoring the bad pixels). Alarm on change of any region by >0.25% ???
    3. Might need to remove a linear fit from above to allow for dome illuminator / ambient light changes across the FOV. May want to average larger scales.

Linearity checks

(Monthly)
  1. Take "Photon Transfer Curve (PTC)" sequence of dome-flat exposure pairs at each of ~20 fluences (from ~100 ADU to saturation), interspersed with reference exposures at fixed time to track light-level drifts. The "official" PTC script is DECamObserver/ExposureScripts/PTC_dlt_20130108.json.
  2. Subtract overscan and bias.
  3. Normalize each exposure pair to common flux level.
  4. For each exposure pair and each amplifier, use the unmasked pixels to calculate a robust estimate of the mean and the mean-squared difference between the two exposures.
  5. Plot the variance vs mean for each amplifier. Alarm on deviations above ??? from linearity, or from the reference nonlinearity.
  6. Alarm on deviations above ??? from the reference read noise and gain calculated from intercept and slope of above fit.
  7. Use reference exposures to track light-level drifts. Then plot adjusted mean flux vs exposure time for each amplifier.
  8. Alarm on change of 0.3% or 3 ADU (whichever is larger) from the reference linearity curve for any amplifier.

Cloudy-night tests

DECal Flats

(cloudy nights, hopefully at least once per semester and after every configuration change, e.g. mirror washings). Definition of the DECal observing and analysis scripts will require collaboration with the Texas A&M & Calibration groups.

  1. Measure spectral response of all pixels through all filters.
  2. Subtract overscan and bias.
  3. Ratio to reference response maps for each frequency. Alarm on any changes above 6 sigma ? at any unmasked single pixel, and changes >0.5% ? when binned into 16x16 pixel regions.

On-sky tests

Star Flats

(Monthly or quarterly) Sequence of ~20 exposures per filter in rich star field, log-spiral dither pattern from 10" to >1 degree. Generates internal solutions for stellar response and astrometric distortions. Generates star flat calibration files and astrometric reference files. Alarm on changes beyond % / 5 milliarcsec.

Note that once the magnitudes of numerous stars in the standard-star fields have been put onto the natural DECam system, we could also use each clear evening's exposures of the standards fields - and the SN fields - to check for changes in relative response across the array.

The codes to make these astrometric and photometric references are still in development.

Reference wavefront

(Nightly) - Single out-of-focus exposure in a single filter at nominal hexapod position, to check for any dramatic changes in the wavefront.

Need to consult with Aaron Roodman to define this test and get some codes in place.

Zeropoints

(Clear nights) - standard-field exposures; alarm for >0.02 (?) mag change in zeropoint from reference value when skies are known to have been clear.

Need to consult with DESDM/Calibration (Robert Gruendl, Bob Armstrong, Douglas Tucker) to see how this is / could be implemented.

Pointing, tracking, look-up tables

(Monthly engineering nights) - Run by the CTIO staff.

  1. Exposures taken in focus in i band (???) with 30,30,30,90 s exposures at each of a grid of HA, Dec around the sky - need to avoid bright moon though.
  2. Steve Kent software produces pointing errors for each exposure. Report RMS pointing error. Alarm on individual pointing error above 30" ? or RMS > ?
  3. Kevin Reil code calculates RMS tracking errors and motion after shutter-open for each exposure. Alarm on motion >0.2" in 1st 2s after shutter open of any exposure, or RMS tracking error > 0.1" ??? in any axis.
  4. Guider RMS calculated by IH for each exposure. Alarm on guider RMS above 0.2" ??? in any axis for any exposure.
  5. AOS reports hexapod adjustment required for each exposure. Alarm for any 90s exposure being more than ? microns out of focus (dodz), more than ? um or ? arcsec out of collimation in x and y. Alarm for the derived best-focus hexapod position being more than ? from the lookup table.

Crosstalk

(Seasonal or yearly) - Observe field with a one or a few v. bright stars, measure the crosstalk functions.

Needs development. Talk with Brian Yanny, Kerstin Paech, ???