Denoising radio pulses from air showers using machine-learning methods

TL;DR

This paper presents a machine learning-based denoising method using a convolutional encoder-decoder network to improve detection of radio signals from air showers caused by ultra-high-energy cosmic particles, enhancing sensitivity.

Contribution

The study introduces a supervised convolutional neural network trained on realistic simulations to effectively denoise radio signals from air showers, a novel application in this context.

Findings

Demonstrated increased sensitivity in simulated GRAND signals after denoising.

Showed the effectiveness of the neural network in removing noise from complex radio signals.

Initial results indicate potential for improved cosmic particle detection.

Abstract

The Giant Radio Array for Neutrino Detection (GRAND) aims to detect radio signals from extensive air showers (EAS) caused by ultra-high-energy (UHE) cosmic particles. Galactic, hardware-like, and anthropogenic noise are expected to contaminate these signals. To address this problem, we propose training a supervised convolutional network known as an encoder-decoder. This network is used to learn a coded representation of the data and remove specific features from it. This denoiser is trained using high-fidelity air shower simulations specifically tailored to replicate the characteristics of signals detected by GRAND. In this contribution, we describe our machine-learning model and report initial results demonstrating the sensitivity enhancement resulting from our denoising algorithm when applied to realistically simulated GRAND signals with varying signal-to-noise ratios.