Identifying environmentally induced calibration changes in cryogenic RF axion detector systems using deep neural networks

Andrew G. Engel, T. Braine, C. Boutan

Abstract

The axion is a compelling hypothetical particle that could account for the dark matter in our universe, while simultaneously explaining why quark interactions within the neutron do not appear to give rise to an electric dipole moment. The most sensitive axion detection technique in the 1 to 10 GHz frequency range makes use of the axion-photon coupling and is called the axion haloscope. Within a high Q cavity immersed in a strong magnetic field, axions are converted to microwave photons. As searches scan up in axion mass, towards the parameter space favored by theoretical predictions, individual cavity sizes decrease in order to achieve higher frequencies. This shrinking cavity volume translates directly to a loss in signal-to-noise, motivating the plan to replace individual cavity detectors with arrays of cavities. When the transition from one to multiple cavities occurs, haloscope searches are anticipated to become much more complicated to operate: requiring N times as many measurements but also the new requirement that N detectors function in lock step. To offset this anticipated increase in detector complexity, we aim to develop new tools for diagnosing low temperature RF experiments using neural networks for pattern recognition. Current haloscope experiments monitor the scattering parameters of their RF receiver for periodically measuring cavity quality factor and coupling. However off-resonant data remains relatively useless. In this paper, we ask whether the off resonant information contained in these VNA scans could be used to diagnose equipment failures/anomalies and measure physical conditions (e.g., temperatures and ambient magnetic field strengths). We demonstrate a proof-of-concept that AI techniques can help manage the overall complexity of an axion haloscope search for operators.

Introduction: The Scaling Challenge of Axion Detection

Axion haloscopes search for theoretical dark matter particles by converting them into microwave photons within high-quality (high Q) resonant cavities immersed in powerful magnetic fields. As researchers push toward higher axion masses, the physical size of these cavities must decrease to maintain resonance at higher frequencies. This reduction in volume leads to a significant drop in signal-to-noise ratios.

Schematic of an axion haloscope detector system showing the conversion of axions to photons within a resonant cavity and the resulting power spectrum.

To compensate, future experiments will transition from single-cavity detectors to large arrays of multiple cavities working in unison. However, this shift introduces immense operational complexity. Monitoring hundreds of sensitive RF components—such as circulators and amplifiers—inside a dilution refrigerator at temperatures near absolute zero is a daunting task for human operators. Currently, Vector Network Analyzers (VNAs) are used to monitor resonant frequencies, but the vast majority of "off-resonant" data is discarded as noise.

Methodology: Mining Off-Resonant Data

This research proposes that discarded off-resonant scattering parameters (S-parameters) contain a unique "fingerprint" of the experimental environment. By treating the extraction of diagnostic information as a supervised machine learning problem, the team developed a two-step pipeline:

  1. Principal Component Analysis (PCA): Wide-band S-parameter traces, which contain thousands of frequency points, are compressed into 5 to 20 principal components. This dimensionality reduction is critical to prevent overfitting and remove noise.
  2. Artificial Neural Networks (ANNs): These reduced features are fed into a feed-forward neural network (one hidden layer of 512 nodes) to predict specific physical states or environmental variables.

Experimental Validation

The approach was tested across three distinct diagnostic scenarios:

  • Device Characterization: Distinguishing between two physically identical RF circulators.
  • Magnetic Field Regression: Predicting the position of a permanent magnet (and thus the field strength) independent of other mechanical changes in the system.
  • Cryogenic Thermometry: Predicting temperatures across different stages of a dilution refrigerator (from 50K down to the Mixing Chamber) during a multi-day cooldown.

Key Results

The results demonstrate that AI can effectively isolate overlapping physical effects and provide highly accurate diagnostics where traditional methods fail.

PCA-reduced feature distributions for two different RF circulators, demonstrating the high classification accuracy achieved by the neural network.
  • 100% Identification Accuracy: The model perfectly distinguished between identical components. Even after swapping the physical wiring, accuracy remained at 99.7%, proving the network was learning the component's unique RF signature rather than cable artifacts.
  • Precision Tracking: The network predicted magnet positions with a near-perfect correlation ($R = 0.98$).
  • Virtual Thermometry: The system predicted the Mixing Chamber temperature with an error of just $0 \pm 1.9$ K, suggesting that RF components themselves can act as secondary thermometers.

Conclusion and Future Outlook

This proof-of-concept shows that Deep Neural Networks can transform "useless" background data into a powerful diagnostic tool for complex physics experiments. By automating the monitoring of environmental conditions and equipment health, these techniques will be essential for managing the next generation of multi-cavity axion haloscopes. Future work will focus on training models on simulated data to predict real-world behavior and automating the calibration of quantum-limited amplifiers.

Summary updated

Abstract

The axion is a compelling hypothetical particle that could account for the dark matter in our universe while simultaneously solving the strong CP problem in quantum chromodynamics. The most sensitive axion detection technique demonstrated so far makes use of a high Q cavity immersed in a strong magnetic field, where axions are converted to microwave photons. This is called an axion haloscope and has primarily targeted the 1-10 GHz range. As searches scan up in axion mass, toward the parameter space favored by theoretical predictions, individual cavity sizes decrease in order to achieve higher frequencies. This shrinking cavity volume translates directly to a loss in signal-to-noise, motivating the plan to replace individual cavity detectors with arrays of cavities. When the transition from one to (N) multiple cavities occurs, haloscope searches are anticipated to become much more complicated to operate, requiring N times as many measurements but also the new requirement that N detectors operate in unison, which can be achieved by locking them to a common frequency. To offset this anticipated increase in detector complexity, we aim to develop new tools for diagnosing experiments using neural networks. Current experiments monitor scattering parameters of their receiver for periodically measuring cavity quality factor and coupling. However, off-resonant data remain relatively useless. In this paper, we ask if it is possible that off-resonant information contained in vector network analyzer scans could be used to diagnose equipment failures/anomalies and measure physical conditions (e.g., temperatures and ambient magnetic field). We demonstrate a proof-of-concept that AI techniques can help manage the complexity of an axion haloscope search for operators.
BibTeX
@article{engel2025identifying,
  author = {Andrew G. Engel and T. Braine and C. Boutan},
  title = {{Identifying environmentally induced calibration changes in cryogenic RF axion detector systems using deep neural networks}},
  journal = {Review},
  year = {2025},
  doi = {10.1063/5.0272636},
}
Figure