Things to measure in support of assessing 30 second integrations:
- dark frames scaled up by factor of two to accommodate muon rate at sea level vs. at Pachon altitude, to test out cosmic ray rejection algorithms
- read noise and crosstalk as a function of pixel rate. We need 550 Kpix/sec for 2 second readout so if we drop to 250 Kpix/sec it would take 4.4 seconds to read out an image.
- read noise and crosstalk vs. timing diagram choices- delays, dual slope integration times, etc
- rise time measurements at slow pixel rates
Setting aside the fact that the camera team promised an instrument that can read out in 2 seconds at around 10 e- read noise, I think we can mitigate many of the current sensor issues by going to single 30 sec exposure per visit. This would have the benefits we've discussed before:
- 40% higher read noise values can be tolerated, since sky noise goes up by sqrt(2).
- by taking 5 sec to read out instead of 2, the read noise falls by much more than the square root of that ratio, since much of it is clock feedthrough.
- efficiency goes up. The dead time per image goes from (slew_and_readout + open shutter + readout + open shutter) = 5+1+2+1=9 sec to (slew_and_readout + open ) = 6 seconds. Shutter-open efficiency goes from (30)/(30+9)=77% to (30)/(30+6)=83%, a fractional gain that is equivalent to increasing our throughput across all bands by 7.8%.
- astrometric shimmy per image from the atmosphere goes down, I think like sqrt(t) but not sure about that...
- crosstalk goes down by a factor of ten
- power dissipation is lower
- data volume is 2x lower
We do lose a magnitude of dynamic range off the bright end, but we can get that back during twilight observations. The weak lensing folks lose some dithering, but that's mitigated somewhat by having the overall SNR go up and the astrometric rubber-sheet distortions go down. Or maybe we do the 15 pairs in the r band where the read noise is subdominant anyway? We lose short-duration transients and I don't know what to do about that, apart from the r-band 15 sec plan. We lose sensitivity for the trailed NEO's but as we heard in Bremerton the solar system folks don't seem to care. There might be an impact on asteroid linkages, however, if we can't keep the cosmic ray leakage down.
SO...
I think we need to take seriously the heresy discussed in Bremerton, namely going to a single 30 sec exposure per visit. As was also discussed there, this would rely on single-image cosmic ray rejection. This leads to my question for you:
Setting aside the current DM priorities, how long would it take you to demonstrate a robust ability to do single-image CR rejection? I'm thinking the difference images will be quite helpful, or alternatively a multiframe fit that accommodates the possibility of CR events? Some real innovation here seems appropriate.