AI-Enhanced Self-Triggering for Extensive Air Showers: Performance and FPGA Feasibility

Abstract: Cosmic-ray detection with radio antennas has traditionally depended on external triggers from particle detectors, constraining sensitivity and increasing complexity. Previous attempts at fully standalone, radio-only triggers have often failed under intense radio frequency interference, making genuine air-shower signals difficult to isolate. We present a proof-of-principle artificial intelligence-based self-triggering system that overcomes these limitations. By training a deep learning model on both real noise data and injected cosmic-ray-like pulses, we achieve an exceptionally low false-positive rate alongside high detection efficiency. Configurable operating points can suppress false positives below 0.01\% while retaining more than 88\% of genuine signals, and can even eliminate false positives entirely at a modest reduction in signal efficiency. This flexibility makes single-station cosmic-ray detection feasible without requiring external trigger inputs. Applying our approach to real-world noise conditions reduces the initial false-positive event rate by several orders of magnitude, supporting large-scale deployments. Extrapolation to dedicated hardware implementations, such as FPGAs, indicates that sub-\SI{}{\micro\second} inference times are achievable, enabling real-time autonomous triggering. These results highlight the transformative potential of artificial intelligence for enhancing radio detection sensitivity and inaugurate a new generation of fully self-triggered cosmic-ray observatories.