21–26 Jun 2026
University of California, Irvine
US/Pacific timezone

Machine Learning Driven Signal Processing and Event Reconstruction for Cyclotron Radiation Emission Spectroscopy in QTNM

Not scheduled
20m
Conference Center (University of California, Irvine)

Conference Center

University of California, Irvine

Poster Neutrino Mass Poster session

Speaker

Nathan Higginbotham (University College London)

Description

Determining the absolute neutrino mass is one of the primary objectives in particle physics today. While oscillation experiments have constrained the differences between the mass eigenstates, the absolute scale is unknown. The Quantum Technologies for Neutrino Mass (QTNM) collaboration aims to address this by utilising Cyclotron Radiation Emission Spectroscopy (CRES) with an atomic tritium source. This approach leverages quantum sensors, employing Rydberg atoms in superposition for precise magnetic field mapping and quantum-noise-limited microwave detection. By trapping tritium β-decay electrons in a magnetic field and measuring the frequency of the emitted cyclotron radiation, we can perform precision spectroscopy near the tritium endpoint to probe the sub-eV mass regime.

The sensitivity required for QTNM’s goals (10 meV/c²) imposes strict demands on signal analysis. The signals are femtowatt-scale, frequency and amplitude modulated chirps buried in thermal noise, often exhibiting complex sideband structures driven by the emitted electron's motion in the magnetic trap. To characterise these signals, we have developed a comprehensive simulation pipeline that models the full electrodynamics of the QTNM spectrometer. This includes the generation of high-fidelity synthetic datasets that account for diverse trap geometries, relativistic effects, and the spectral sidebands from axial modulation.

We use these simulations to develop and validate our detection algorithms while quantifying systematic uncertainties. This work focuses on robust triggering, detection and reconstruction frameworks capable of identifying CRES tracks in high-noise environments. We explore machine learning approaches for signal classification and parameter estimation, benchmarking their detection efficiency against conventional signal processing techniques. This study aims to converge on an optimal methodology for event detection and the high-precision reconstruction of the initial cyclotron frequency - the critical parameter for neutrino mass determination.

This poster outlines the current status of these frameworks, the signal processing techniques implemented to maximise spectral resolution, and how these simulations inform the projected sensitivity of the QTNM experiment.

Author

Nathan Higginbotham (University College London)

Presentation materials