Week 23 of the Early Operations system optimization period
There were two full nights of pre-LSST observations this week. Unfortunately, the nearly full moon was passing in front of the COSMOS Deep Drilling Field that has currently deployed template coverage, and no alerts were generated on those nights.
Most of the remaining on-sky observations continued the intensive on-sky engineering work to improve the delivered image quality. The team tested two wavefront estimation approaches for the Active Optics System (AOS) this week: an updated version of the current baseline forward-modeling approach, and an alternative fully independent machine-learning model trained directly on simulated corner wavefront sensor images.
The team continued gathering observations in full-array mode with the focal plane intentionally pistoned to intra- and extra-focal positions, building a grid of measurements across a range telescope elevation and camera rotator angles. These observations are being used to study the variation of optical aberrations across the field of view, including higher-order spatial variations, and the translation between wavefront measurement at the corner wavefront sensors and the delivered image quality across the field of view. Data collected this week is intended to help examine the extent to which higher-order spatial variations of optical aberrations across the field of view can be predicted as a function of telescope pointing.
The team also began a series of tests of the AOS closed-loop feedback control. In brief, the control challenge is to optimally incorporate feedback from the corner wavefront sensors in the presence of statistical noise in the wavefront measurement and a delay between the observation used for wavefront measurement and application of the computed correction to the optical degrees of freedom. The current baseline approach uses a Proportional-Integral-Derivative (PID) controller, continuously acquiring images and applying corrective feedback at the next visit that results are available, such that multiple iterations of corrective feedback can be simultaneously “in flight” for a typical correction lag of 2 to 3 visits. One test this week used a simplified control loop that discards intermediate updates to the optical state so that corrections do not accumulate across multiple visits, with the goal of demonstrating more stable convergence, and/or motivating the pursuit of more predictive model-based control loop approaches.
The team is planning to run sustained pre-LSST observations on a few nights during the next week to get a current assessment of the performance.
The currently installed filter set is ugriz. The next filter swap to exchange u for y is planned for 21 April.