Speaker
Description
We develop a simulation-based Bayesian pipeline that reconstructs the incident electric field in the proposed Giant Radio Array for Neutrino Detection (GRAND) from antenna voltages produced by neutrino-induced air showers. Direct deconvolution of the antenna response can be ill-conditioned and strongly amplify measurement noise, motivating Bayesian methods such as the Information Field Theory reconstruction of Strähnz et al. (2024). Building on their physics-based forward model, we combine a parametric description of the signal with physically motivated models for the antenna response, instrumental noise, and the Galactic radio background, and train an amortized inference model on detector simulations. Three-polarization voltage waveforms are denoised and compressed with an autoencoder, and a normalizing-flow posterior is trained to infer six electric-field waveform parameters. Our work is ongoing; by the conference we will report tests on posterior reliability with coverage diagnostics and evaluate the sensitivity of our method to low signal-to-noise simulated signals.