CNN-based Approach for Cleaning and Identifying Radio Signals Emitted by Cosmic-Ray Air Shower.

Cosmic rays with high energies produce extensive air showers when they enter Earth's atmosphere. The charged shower particles, mostly electrons and positrons, are deflected by the Earth's magnetic field and produce radio emission. Studying this radio emission can reveal the properties, such as energy and direction, of the initial cosmic-rays. However, radio detection is often impaired by interference, such as human-made radio frequency noise and continuous Galactic background noise. In this study, we present a machine learning method based on a convolutional neural network (CNN) for the classification and reduction of background noise in radio signals. The CNNs are trained and evaluated using simulated radio signals from Monte Carlo simulations of EAS events and measured background noise from a prototype station of IceCube's surface enhancement at the South Pole. Our study shows that the use of CNNs can significantly enhance the accuracy of the arrival time and amplitude of air-shower radio pulses, which will subsequently result in a more precise reconstruction of the properties of cosmic rays.