NUTRIG: Development of a Novel Radio Self-Trigger for GRAND
Jelena Köhler · 2025
Read NUTRIG: Development of a Novel Radio Self-Trigger for GRAND by Jelena Köhler (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-04339.txt.
Introduction
Huge radio arrays that hunt for ghost particles from space must spot the faint flash of an air shower on their own, in real time. The NUTRIG project builds such self-trigger tricks for the GRAND array. Its method leans on the shape of the radio pulse that a shower leaves in each unit.
The team's first-level trigger matches each pulse against a stored shape, right at the unit. Lab tests report how often it fires true and how much data it lets through. A second level then joins fire times and pulse strengths across the array. That stage gives a rough read of the shower's timing, path, and wave tilt, cheaply enough to run live.
Catalog record
| CREATOR | Jelena Köhler |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-04339
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
NUTRIG: Development of a Novel Radio Self-Trigger for GRAND Jelena Köhler Abstract The NUTRIG project is dedicated to the development of advanced radio self-trigger methods for large-scale arrays such as the Giant Radio Array for Neutrino Detection (GRAND). The developed techniques are based on features of the radio emission of air showers to perform an efficient online rejection of background. We first describe a first-level trigger (FLT) template-matching method that uses the shape of transient radio pulses measured at the detection-unit level to target those induced by air showers. We present trigger efficiencies and throughput tests of the template-matching FLT in controlled laboratory conditions. Next, we describe the second-level trigger (SLT), which utilizes the measured FLT times and corresponding voltage amplitudes to construct a trigger at the array level. We present offline performances of the SLT, which performs a coarse reconstruction of timing, direction, and polarization of the air shower. [a]Pablo Correa 1 Introduction The primary goal of the Giant Radio Array for Neutrino Detection (GRAND) [ 1 , 2 ] is to observe ultra-high-energy astroparticles using radio self-trigger techniques. GRAND is currently in its prototyping phase (see [ 3 , 4 ] for details); its main pathfinder array is GRANDProto300 (GP300) located in the Chinese Gobi Desert. As of today, 65 detection units (DUs) have been commissioned at GP300. However, in its next stages (GRAND10k and GRAND200k), GRAND will consist of subarrays with 5,000–10,000 DUs each, with a sparse DU spacing of 1 km. This scale demands an autonomous, efficient, low-cost radio trigger system with minimal inter-DU data transfer. NUTRIG is the first step towards such a state-of-the-art radio self-trigger. Apart from modeling efforts of air-shower emission between 50–200 MHz [ 5 , 6 ] , the frequency band in which GRAND operates, NUTRIG consists of two major parts: • First-Level Trigger (FLT) at the DU level. The classical, so-far used DU-level trigger [ 3 ] applies a double-threshold technique designed for a nominal rate of 1 kHz at GP300. With NUTRIG, we aim to reduce the DU-level trigger rate to 100 Hz. To achieve this, we developed a template-matching method, described in this work, and an algorithm using a convolutional neural network, which we shall omit here—see [ 7 , 8 ] for more details. • Second-Level Trigger (SLT) at the array level. For GRAND prototypes, multiple DUs are required to have an FLT within a coincidence time window to result in a so-called central trigger [ 3 ] . The expected nominal rate of this array-level trigger is 10 Hz. With NUTRIG, our goal is to reduce this rate to 1 Hz by applying fast online reconstructions on FLT data. These proceedings present the latest results of the NUTRIG FLT and SLT methods, first outlined in [ 8 , 9 ] . In Section 2 , we describe novel NUTRIG-dedicated databases, consisting of recent GP300 data and up-to-date air-shower simulations. Next, in Section 3 , we present the template-matching FLT, including both its offline and online performances. Section 4 is reserved for the SLT method and its offline performance. Finally, concluding remarks are given in Section 5 . 2 Databases for NUTRIG Studies As summarized in Table 1 and detailed below, we construct several databases for our FLT and SLT studies, using both experimental data and simulations. For the FLT we select individual pulses, while for the SLT we select array-level events. Database name Data type Number of SNR FIR filter FLT-0 setting pulses/events [offline] [offline] FLT background MD pulses 6,628 4–8 Yes FLT-0-light FLT signal Sim. pulses + stat. MD 8,000 4–8 Yes FLT-0-light SLT background CD events 2,146 ( \geq 5 DUs) > 5 Yes FLT-0-optimal SLT signal Sim. events + stat. MD 5,620 ( \geq 5 DUs) > 5 Yes FLT-0-optimal Electric transformer CD pulses 1,368 > 5 No — Cosmic-ray candidates CD pulses 225 > 5 No — Table 1: Summary of the different databases used in this work, which consist of experimental GP300 data in the form of MD and CD, as well as air-shower simulations (sim.) superimposed to stationary (stat.) MD. See text for more details. 2.1 Selection of Pulses from GP300 Data As extensively described in [ 10 ] , a DU of GP300 consists, among other components, of a three-arm butterfly antenna—two dipoles along the North-South ( X ) and East-West ( Y ) axes, and a monopole along the vertical ( Z ) axis—and its front-end electronics board (FEB). The FEB contains a 14-bit analog-to-digital converter (ADC) with a sampling rate of 500 Msps as well as a Xilinx system on chip (SoC). The SoC consists of a field-programmable gate array (FPGA) that performs the GP300 double-threshold trigger, which we will name FLT-0 in this work, and a central processing unit (CPU) for event building. The NUTRIG FLT component, which we label as FLT-1, is implemented on the CPU (see Section 3.3 ). At GP300, the observed radio-frequency interference (RFI) is dominated by airplanes and the electric transformer of a mine located 15 km from the site [ 11 ] . To mitigate the high-frequency noise of this electric transformer as well as aeronautic communication lines, a finite impulse response (FIR) lowpass filter with a cutoff frequency of 115 MHz is applied by the FPGA before evaluating the FLT-0. For triggered events, 2048 ns of the unfiltered ADC traces are recorded, for each polarization, by the central data-acquisition (DAQ) system. In this work, we first construct the FLT background database by identifying pulses embedded in forced-trigger monitoring data (MD) taken in February 2025, with a total runtime of \sim 12 hours at a rate of 20 Hz. We mimic the online conditions by passing the data through the FIR filter and FLT-0 algorithm, which can trigger on the X or Y channels. Here, we use relaxed conditions on the FLT-0 parameters (FLT-0-light) that allow us to select pulses down to relatively low signal-to-noise ratios \rm SNR\gtrsim 4 . We define the SNR as maximum of the pulse divided by the root-mean-square (RMS) of the trace excluding the pulse. From the background pulses that pass the FLT-0-light, we then randomly select 2,000 traces for both \rm 4\leq SNR<5 and \rm 5\leq SNR<6 . Due to the limited MD runtime, only 355 and 74 traces are selected for \rm 6\leq SNR<7 and \rm 7\leq SNR<8 , respectively. Next, we construct the SLT background database, which consists of coincidence data (CD; runtime of \sim 30 hours) taken in April–May 2025. The CD contains events that passed the online GP300 central trigger, for which at least 4 DUs are required to yield an FLT-0 (which was exlusively applied to the Y channel) within a time window of 11 \mu s. For consistency with the SLT signal database (Section 2.2 ), we reapply the FIR filter and FLT-0 algorithm to the recorded data. In this case, we use FLT-0 settings that are optimized for a maximal signal selection efficiency while keeping the expected mean trigger rate below 100 Hz in the aforementioned MD (FLT-0-optimal). We select 2,146 background events where at least 5 coincident DUs pass the FLT-0-optimal with an \rm SNR>5 . Finally, for an FLT-1 case study presented in Section 3.2