In the course of working on finalizing the DM-OCS ICD LSE-72, this issue arises for me:
What is the dependency of the Level 1 pipeline on knowledge of the pointing before it has performed an initial astrometric recovery on the imagery?
I have two specific questions about this:
By OCS-DM-COM-ICD-0031 in that ICD we know we will get an advance notice of the planned pointing 20 seconds before a visit begins, which means nearly 60 seconds before we have both snaps in hand. This advance notice is supposed to be good enough to let us retrieve warped templates, SSO predictions, and other inputs. We have not yet defined (but should) the precision required in the pointing notice to enable us to do that work. So, my question #1 is how good this information needs to be. How far can it be off from the true pointing before the advance lookup becomes useless (or at least leads to degraded AP results)? My guess is that this will be a very forgiving spec, but we should state a number just to be concrete.
Beyond the mandated advance notice, we will certainly also have available, by the time we receive the pixel data, the actual pointing coordinates passed to the TCS for the current visit (which we would normally expect to be identical to the advance notice). We can validate those against the advance notice and flag an exceptional condition if they do not match (which would result in the AP analysis having to wait for the proper reference data to be retrieved, likely miss the 60-second latency requirement, and possibly have to be rescheduled to a catch-up system to avoid producing a cascading effect on subsequent visits).
Beyond that, theoretically there is additional information available. The guider system (when active) will be looking for known guide stars from a reference catalog. Although it will generally only be used to hold the relative position of the pointing constant on the target, it will also be able to compute an estimate of the true pointing, and this could be made available through Observatory telemetry for use by DM. Because the guiders are read out and analyzed continuously (at 9Hz) during exposures, such a true-pointing estimate could be available in plenty of time to get it merged into the input data to AP. There are 8 guiders, in the corners of the field, so this true-pointing analysis would have good leverage to compute an estimate of the true rotation angle in addition to the optical axis pointing.
So question #2, in its most general terms, is: how should we initialize the pointing metadata information for the images when the science pipelines code first sees them? From the advance notice, from the final actual TCS command, and from an estimate of the true pointing from the guide system? Would we use the latter if it were available?
Currently this is a hypothetical; there is no baselined requirement for the TCS to report this information (though they will certainly have it and would easily be able to publish it). If we are interested, I would like to state that in LSE-72.
I've set a provisional due date of January 20th to support getting any answers to this question into a change request against LSE-72 which could be voted on at the February CCB meeting.