Speaker
Description
Accurate primary beam models are essential for high-fidelity calibration and imaging in modern radio astronomy, but existing approaches either lack the flexibility to capture realistic antenna patterns (analytical models) or become computationally expensive for wide-band observations (electromagnetic simulations, holographic measurements). We present the Generative Antenna Model from Bayesian Inference Technique (GAMBIT), a non-parametric Bayesian framework grounded in Information Field Theory and implemented in the NIFTy library. GAMBIT represents the beam as a single base pattern subject to three smooth, frequency-dependent geometric transformations---an isotropic scale factor and two spatial shifts---and is applied here to holographic measurements of the array-averaged MeerKAT Stokes~I beam in the L-band. The model compresses 67~MB of holographic data per antenna into just 150~KB (a 400:1 ratio) while maintaining statistical consistency with the data chi2/dof = 1, and outperforms Gaussian, Airy disc, JimBeam, Eidos, and electromagnetic simulation reference models by more than an order of magnitude in mean-squared deviation across the L-band. The Zernike decomposition of the base beam, combined with frequency-dependent geometric transformations, enables computationally efficient beam reconstruction at arbitrary spatial resolution. GAMBIT is readily adaptable to other radio telescopes, offering a general pathway toward improved calibration precision for current and future radio surveys.