Data-driven beam diagnostics using betatron radiation

Abstract Plasma accelerators are a promising candidate for future compact accelerator design. Among them, the Advanced Wakefield Experiment (AWAKE) is a unique facility that accelerates electron beams using CERN’s high-energy proton beam. The plasma accelerators require novel diagnostic techniques to accurately determine their beam parameters. A method of machine learning-based betatron diagnostic is presented to predict witness beam parameters in the context of the AWAKE experiment. Betatron radiation in plasma accelerators could serve as a footprint of the accelerating electrons, and therefore potentially provide information about the beam parameters. Here, a neural network model is developed and trained using the dataset generated from extensive simulation scans. The dataset is systematically generated using particle-in-cell simulations, exploring a range of parameters close to the AWAKE Run 2 baseline, with variations in the initial emittance and the radius of the witness beam. The results demonstrate the model’s ability to predict key final witness beam parameters such as energy, energy spread, and emittance, based on measurable features of the betatron spectrum. Predictions for the beam energy and energy spread show strong agreement with the simulation data, while emittance predictions exhibit more variability, suggesting areas for further refinement.