Cosmic particles with energies millions of times greater than anything our accelerators can produce slam into the atmosphere and create a shower of secondary particles. This shower emits radio waves that antennas on the ground pick up. But the signal is weak: it's swamped by the radio noise of our galaxy and the receivers' own noise. Picking out such a signal is like trying to hear a whisper on a noisy street.
The neural network works like an interpreter who listens to intonation and parses words at the same time. It looks at the signal in both time and frequency, memorizing the characteristic patterns of radio bursts from cosmic showers. Trained on 410,000 simulated signals, it boosts clarity by 15–23 decibels, like a whisper turning into a full-voice conversation. After this cleaning, 2–3 times more antennas can accurately determine the arrival time of the signal, and thus more precisely calculate where the particle came from.
The method is especially important for the future radio telescope GRAND, which will spread antennas over an area of one hundred thousand square kilometers. The more accurate each antenna, the more reliable the whole network. And that means we'll be able to spot the distant sources of the most energetic particles in the universe, and perhaps figure out which cosmic accelerators can boost them to such monstrous energies.
The energy of one such particle is comparable to the impact of a tennis ball serve — except it's packed into a single proton.