Speaker
Description
Radio detection of neutrinos remains the most promising technique for the detection of UHE neutrinos. Construction of large-scale radio-neutrino detectors, however, is limited by logistics; thus, optimization of the detector stations is the only way to enhance the science reach of future radio detectors. Improving the trigger efficiency also for faint signals is thus crucial. A complete digital readout chain of antennas enables the implementation of the trigger in the FPGA's logic, allowing for more flexible and advanced trigger decisions based on neural networks. In a first approach, a conventional pre-trigger reduces the data rate to 10 kHz followed by a second stage CNN-based neural network reducing the trigger rate further to 1Hz. The performance can be further improved by a continuous running CNN-based neural network that directly runs on the raw data, however, requiring a more complex model to run on the FPGA. Simulation studies suggest an enhancement in the neutrino detection rate by up to a factor of two, translating into a factor-of-two improvement of most science objectives. We setup evaluation boards in the lab to quantify the operation, power consumption, background rejection, and signal efficiency. In this contribution, we present the neural network design, its simulation performance, and initial lab tests.