Using Convolutional Neural Networks (CNNs) to Search for Cosmic-Ray Radio Events at the IceTop Enhancement Prototype Station

Cosmic rays originating from beyond the solar system generate extensive air showers upon interaction with our atmosphere. Within these air showers, charged particles are deflected by the Earth's geomagnetic field, resulting in the emission of radio signals. These radio signals can be used to infer the characteristics of the incident cosmic rays. However, the task of detecting these radio signals becomes difficult in the presence of galactic and anthropogenic noise.To mitigate the impact of background noise, we employ Convolutional Neural Networks (CNNs) in this work.

IceTop or the surface component of IceCube neutrino observatory, which is capable of detecting cosmic rays in the PeV to EeV range is planned to be enhanced in the coming years. A prototype station of the surface enhancement, consisting of eight scintillator panels and three radio antennas, was installed in 2020. We trained CNNs on 1µs long background waveforms collected from these antennas in the 60-350 MHz frequency range. For signal waveforms, we simulated radio pulses from air showers using the CoREAS Monte Carlo software.

Once trained, we utilize the CNNs to search for air-shower radio pulses. In a dataset spanning about four months, we identified 176 events; roughly three times as many radio events than with the traditional event search method. Moreover, all identified events are found to be in agreement with the standard IceTop reconstructions. This demonstrates the effectiveness of our CNN-based approach in enhancing the detection and analysis of cosmic-ray radio signals, even in the presence of challenging background noise.