Joseph Diamond, Peraton; Jason Paul, Peraton; Weihan Yang, Peraton; Noah Combs, Peraton; Jessica Kalmanson, Peraton
https://doi.org/10.64861/GMCR9729
Keywords: sensor calibration,space domain awareness,orbit determination,automated calibration,observation residuals,space-based sensors,calibration automation,proliferated architectures,space surveillance,observation association
Abstract:
Ground-based optical sensors contributing to Space Domain Awareness (SDA) carry per-sensor systematic error in reported angles and per-sensor error in the uncertainty assigned to those angles; the first propagates into state error that additional observations do not remove, while the second corrupts the covariance rather than the state, and a corrupted covariance is what conjunction assessment consumes. That such sensors require calibration is not in question, and this paper does not argue for it. What remains difficult is performing calibration without analyst intervention, continuously, at a sensor count where periodic expert assessment does not scale.
We present first results from CalWatch, an automated calibration capability that estimates per-sensor angular bias and measurement-noise scale in right ascension and declination with no analyst in the loop, coupled directly to DarkWatch, the downstream association, orbit determination, and maneuver-detection engine that consumes its coefficients. We evaluate the pairing on six co-located optical sensors observing the geostationary belt over a twelve-day span, contrasting an automatically calibrated run against an uncalibrated baseline across 1.7 million angle observations matched pairwise between runs, with coefficients derived from one object population and scored on another.
The automated path produced and applied coefficients without intervention throughout, and the resulting downstream behavior is the substance of what we report: which measurements the chain retains, how it maintains custody, and what it detects. Our central finding, however, is structural rather than numerical. Because association gating and observation editing are expressed relative to measurement sigma, a calibration update silently changes the severity of every threshold that consumes it, so calibration and threshold retuning are a single joint operation rather than sequential ones.
Further, an accelerated-playback test construct is not neutral with respect to the metrics such a chain reports, and matched playback rate therefore belongs in the experimental controls of any comparative evaluation of an association or orbit-determination chain. One finding is larger than anything the calibration contrast produced and belongs to both runs alike: observation volume per object dominates association performance, and on objects with fewer than ten observations, more than half of all associations disagree with the identification supplied by the collecting system, a disagreement rate rather than an error rate because that source carries no quantified accuracy.
Date of Conference: September 15-18, 2026
Track: SDA Systems & Instrumentation