Dark Matter Axion Detection with Neural Networks at Ultra-Low Signal-to-Noise Ratio

TL;DR

This paper demonstrates that neural networks can significantly enhance the sensitivity of dark matter axion detection by reducing required integration time, especially under ultra-low signal-to-noise conditions.

Contribution

It introduces a novel application of neural networks to improve axion detection sensitivity in microwave circuits, achieving a 5000-fold reduction in integration time.

Findings

Neural networks can improve detection sensitivity by a factor of 5000.

The approach effectively distinguishes axion signals from thermal noise.

Potential to reduce measurement times or enhance sensitivity in experiments.

Abstract

We present the first analysis of Dark Matter axion detection applying neural networks for the improvement of sensitivity. The main sources of thermal noise from a typical read-out chain are simulated, constituted by resonant and amplifier noises. With this purpose, an advanced modal method employed in electromagnetic modal analysis for the design of complex microwave circuits is applied. A feedforward neural network is used for a boolean decision (there is axion or only noise), and robust results are obtained: the neural network can improve by a factor of $5\cdot 10^{3}$ the integration time needed to reach a given signal to noise ratio. This could either significantly reduce measurement times or achieve better sensitivities with the same exposure durations.