On-board science data quality analysis using anomaly detection for ASTHROS

ASTHROS (Astrophysics Stratospheric Telescope for High Spectral Resolution Observations at Submillimeter-wavelengths) is a high-altitude balloon mission utilizing an array of sixteen spectrometers to create high spatial resolution 3D maps of ionized nitrogen gas in galactic and extragalactic star-forming regions. During data collection, we utilize on-the-fly mapping, where the instrument continuously collects spectra while scanning over a target area. After a sweep across the target, we take a calibration spectra to correct our science data. These calibration spectra provide a baseline for how the instrument is operating at a given moment. As we collect new calibration spectra, we can compare the current calibration with a series of past calibrations to determine if our system is producing anomalous spectra. Some examples of anomalous spectra are changes in RFI spike frequency, location, or amplitudes, changes in the overall readout level, and changes in the shape of the spectra. We compare statistical and data-driven methods for detecting these anomalies and evaluate their performance to determine the best fit for the ASTHROS readout system. For data-driven methods, we compare the latent space representation of our calibration spectra with past calibrations using models like Variational AutoEncoders (VAE) and Principal Component Analysis (PCA). By comparing with a rolling window of past calibrations, we allow our system to change gradually while identifying sudden irregularities. When spectra are labeled as anomalous, they are prioritized for review so that the ground operations team can analyze and address the issue. On-board analysis is enabled by the readout system architecture which utilizes the RabbitMQ (RMQ) messaging networking. RMQ allows us to modularly build our readout system and create additional functionality, such as on-board analysis, without making modifications to the operation pipeline.