The trigger design for AdvCam

TL;DR

The paper presents AdvCam, a silicon-photomultiplier camera for the CTAO LSTs, and a fully digital, multi-level trigger design intended to lower the gamma-ray energy threshold. It details a comprehensive simulation pipeline (Corsika and sim_telarray) to model showers, NSB, and electronics, and proposes several Local Level-2 triggers (DBSCAN, 3D hexagonal convolution via TDSCAN, CNN) plus a Topo-Stereo coordination and a Level-3 software trigger using CTLearn/OpenVINO. The key contributions are the demonstration that $E_{thr}$ can be reduced to $13$ GeV, the exploration of FPGA-friendly L2 algorithms with sub-$0}s latencies, and the validation of a CNN-based L3 path for online background rejection and data-rate control. The results indicate substantial gains in low-energy sensitivity and effective background suppression, with practical implications for DAQ bandwidth and online analysis in CTAO.

Abstract

The AdvCam is a next-generation camera for the Large-Sized Telescopes of the Cherenkov Telescope Array Observatory, based on silicon photomultipliers. Its fully digital readout system enables the design of new, sophisticated trigger logic. The Large-Sized Telescopes aim to cover the low-energy range of the cosmic gamma-ray spectrum, with a threshold starting at about 20 GeV, using the existing photomultiplier tube camera. The AdvCam, along with the new trigger logic, as shown by simulations, lowers the detectable energy threshold to 13 GeV. The proposed trigger logic has a multilevel structure. The first level involves fast coincidences among small pixel regions at a rate of approximately 1 GHz, while the second level processes all camera pixels within an approximately 10-nanosecond time window. Different families of machine learning algorithms optimized for FPGAs form the second-level trigger. In this work, we consider two main approaches: Deep Neural Networks and Density-Based Spatial Clustering of Applications with Noise, both running with latencies below 1 microsecond at a 1 MHz rate. This work provides a detailed description of the trigger chain and its performance, as studied through simulation.

Figures (7)

• Figure 1: (Left): A close-up view of the PMT and SiPM-based cameras. The PMTs are round and have a diameter of $\sim 4$ cm. The hexagonal outline around it shows the light guide interfaced to the PMT. The SiPMs and their light guides have a hexagonal shape. The flat-to-flat distance of the SiPM is approximately 2.2 cm. (Right): Pixel grouping patterns referred to as : flower and super-flower. The flower pattern consists of a central pixel, or seed (shown in black), with 6 surrounding pixels (total of 7 channels). The super-flower pattern consists of a central flower surrounded by 6 flowers (total of 49 channels).
• Figure 2: Trigger Flow Diagram.
• Figure 3: The 12 GeV on-axis gamma ray with a core located 180 m from the telescope. We plot the instantaneous ADC value of the digitized waveform by selecting the time frame with the maximum signal. (First left): The nominal NSB rate is 268 MHz/channel, and the nominal electronic noise 3.94 ADC counts RMS. (Second left): The NSB rate is set to 0, and the electronic noise is set to 0.1 RMS ADC counts. The fine red lines indicate the contours of the flower-like pattern where the digital sum exceeds the threshold, superimposed over the entire readout interval. (First right): Only spatially coincident flowers are shown. (Second right): Only flowers that are spatially and temporally coincident are shown. Note: simulations with a 268 MHz/channel NSB rate and an electronic noise level of 3.94 ADC are used in the analysis, with visualizations superimposed on noise-free waveforms.
• Figure 4: TDSCAN algorithm and FPGA implementation. (Left): Kernel sizes used for convolution. (Center): Hexagonal 3D convolution over two spatial coordinates and one temporal dimension (frames N-1, N, N+1 ...), producing the output frame. (Right): Resource usage and timing at 350 MHz target frequency.
• Figure 5: The differences in camera signal barycenters for proton events between LST-1 and other LSTs: LST-2 (left), LST-3 (center), and LST-4 (right).