Identifying environmentally induced calibration changes in cryogenic RF axion detector systems using Deep Neural Networks

TL;DR

The paper tackles the challenge of diagnosing environmentally induced calibration changes in cryogenic RF axion detectors with multi-cavity scalability. It introduces a PCA-assisted neural-network framework to extract physical diagnostics from off-resonant S-parameter data collected by a vector network analyzer, leveraging the existing diagnostic data without interrupting axion searches. Through three progressively complex experiments, the authors demonstrate that wide-band S-parameters can: (i) fingerprint specific RF circulators with perfect accuracy, (ii) regress external magnetic-field proximity with high correlation, and (iii) predict cryogenic temperatures during dilution fridge cooldown with meaningful precision. These results suggest a practical path to automate health monitoring and performance forecasting in future, larger-scale haloscope arrays, potentially preserving scan efficiency while maintaining sensitivity to axion signals. The approach could be extended to more complex layouts and aided by simulated S-parameters to further reduce operator burden and improve reliability in axion searches.

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 (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 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.

Figures (8)

• Figure 1: A simplified diagram of a Axion Haloscope BoutanThesis. Axions are converted to photons via magnetic stimulation, and are resonantly enhanced by a microwave cavity. Power from the cavity is sampled via an antenna, amplified by an ultra-low noise receiver, and digitized. An axion signal would manifest as a small, persistent power excess above the thermal background energy spectrum.
• Figure 2: Binary Classification Apparatus Two circulators (bottom), labeled C0 and C1, are wired to a digital switch board such that a VNA can measure the transmission coefficient (S21) through the first and third port of each circulator depending upon the switch state. The second port of each circulator is left open and unterminated. This was done in order to have the injection signal transmit through more of the circulators' pathways, so that the resultant signal would capture more of the unique characteristics of the given circulator. The wiring is such that the ports are kept the same between the VNA and circulator pairs to minimize any differences in the S21 parameter not originating from the circulators.
• Figure 3: Binary Circulator Recognition via Scattering Params We plot the distribution of PCA-reduced features from the VNA for the training data (left) the test data (center) and a second evaluation set formed by switching the wiring paths of the physical circulators (right). The background heatmap shows the approximate prediction confidence of the NN in the PCA feature space. We note that the confidence transitions in prediction around vertical line along PCA dim0 value of 0, which approximately is the mid-point formed between class clusters in the training data; which is entirely expected.
• Figure 4: Two Degree-of-Freedom Experiment Apparatus Two stepper motors controlled by a PyBoard micro-controller are connected to separate worm-drives. The first motor controls the distance between a permanent magnet and a circulator, while the second motor controls the insertion depth of a coupling antennae to a RF cavity. These motors are controlled separately, and therefore two degrees-of-freedom are achieved. Using the VNA, we make a cavity reflection measurement through the RF circulator: Port 1 (P1) injects a signal into the first port of the circulator that reflects off the coupling antennae inside the RF cavity (this is connected to the adjacent circulator port), passing back through the circulator, and then is measured through port 2 (P2) that is connected to the final port of the circulator.
• Figure 5: Left: The percentage residual error of between the model's prediction and the true magnet's position plotted against the antenna motor position. While it is biased to over predict the true value slightly, the residuals have little covariance with the antennae position, meaning that the model is homoskedastic in performance across Antennae position. Right A second visualization of the NN performance as a point-for-point plot, where the predicted magnet position is plotted against the true magnet position. The diagonal red dashed line is parity, and the color of the scatter points are scaled with the Antennae position.