# Classification and Recovery of Radio Signals from Cosmic Ray Induced Air Showers with Deep Learning

## TL;DR

This paper demonstrates deep learning methods to classify and recover cosmic ray air shower radio signals from noisy broadband data, achieving high accuracy and energy resolution in simulated environments.

## Contribution

It introduces two deep learning approaches for classifying and cleaning radio signals from cosmic ray air showers, improving detection accuracy and signal reconstruction.

## Key findings

- 90% true positive rate for signals with SNR > 3
- 20% energy resolution without bias for 80% of signals
- Effective removal of radio frequency interference from signals

## Abstract

Radio emission from air showers enables measurements of cosmic particle kinematics and identity. The radio signals are detected in broadband Megahertz antennas among continuous background noise. We present two deep learning concepts and their performance when applied to simulated data. The first network classifies time traces as signal or background. We achieve a true positive rate of about 90% for signal-to-noise ratios larger than three with a false positive rate below 0.2%. The other network is used to clean the time trace from background and to recover the radio time trace originating from an air shower. Here we achieve a resolution in the energy contained in the trace of about 20% without a bias for $80\%$ of the traces with a signal. The obtained frequency spectrum is cleaned from signals of radio frequency interference and shows the expected shape.