DigiCam: Digital System for SST-1M Camera

• DigiCam is a fully digital read-out and trigger system for the SST-1M camera, integrating 1296 SiPM pixels and FPGA-based processing.
• It digitizes signals at 250 MHz, aggregates trigger primitives, and buffers data with a configurable, dead-time-free architecture.
• Commissioning results validate DigiCam’s effectiveness in stereo operation and adaptation to high night-sky background conditions.

DigiCam is the fully digital read-out and trigger system of the SST-1M camera, developed for the single-mirror small-size Cherenkov telescope prototype in the Cherenkov Telescope Array context. In the SST-1M instrument, DigiCam is mounted behind a compact Photo-Detector Plane (PDP) composed of 1296 hexagonal silicon photomultiplier pixels, digitizes shaped signals with 250 MHz analog-to-digital converters, executes FPGA-based triggering, buffers data, and sends triggered events to the camera server. The design literature presents it as a compact, configurable, dead-time-free architecture intended for long-lived operation and high night-sky background conditions, while the commissioning literature documents its use on the two SST-1M telescopes operating at the Ondřejov Observatory (Rajda et al., 2015, Schioppa et al., 2015, Tavernier et al., 2024).

1. Instrumental setting within SST-1M

SST-1M is a single-mirror small-size Cherenkov telescope prototype following the Davies-Cotton concept. The optical system uses 18 hexagonal mirror facets, each 78 cm across, forming a spherical surface with curvature radius 11.2 m; the mirror area is 9.42 m$^2$ without shadowing, the focal length is 5.6 m, the field of view is 9$^\circ$, the optical PSF is $< 0.25^\circ$, specifically $0.09^\circ$ on-axis and $0.21^\circ$ at $4^\circ$ off-axis, and the optical time spread at the focal plane is $< 0.84$ ns. The telescope aim is sensitivity in the $1$ to $300$ TeV range. Within this system, DigiCam is the focal-plane camera electronics that converts the Cherenkov signal collected by the SiPM-based PDP into trigger decisions and event data suitable for later reconstruction (Tavernier et al., 2024).

The commissioning papers describe two SST-1M telescopes installed at the Ondřejov Observatory in the Czech Republic in 2022 and currently in commissioning. One summary reports a telescope separation of 152.5 m, while another reports 155 m; this discrepancy reflects the wording of the two commissioning accounts rather than a reconciled specification. DigiCam is central to stereo operation because event coincidence is coordinated at array level, first by offline timestamp matching in the earlier commissioning campaign and later through the CTA software array trigger, SWAT, using trigger timestamps with nanosecond precision synchronized by a White Rabbit server (Tavernier et al., 2024, Tavernier et al., 2024).

2. Physical architecture and electronics

The SST-1M camera mechanically separates the PDP from DigiCam. The integrated camera is enclosed in an IP65-compliant aluminum box with a Borofloat window and a six-petal shutter, while DigiCam is mounted at the back of the camera. The digital system is partitioned into three identical minicrates, each serving one third of the PDP. Each minicrate contains nine digitizer boards and one trigger board interconnected through a custom backplane; one trigger board acts as master and the other two as slaves. In total, the camera uses 27 digitizer boards and 3 trigger boards for 1296 channels (Schioppa et al., 2015, Rajda et al., 2015).

Each digitizer board processes 48 channels. The 2015 electronics paper specifies multi-channel 12-bit, 250 MS/s flash converters, with converter types tested including the Analog Devices AD9239 and the Intersil ISLA222S25, and Xilinx Virtex-7 FPGAs on both ADC and trigger boards. Digitized samples are continuously written into ring buffers up to 1024 samples deep, corresponding to about $4\,\mu\mathrm{s}$. The trigger boards aggregate local and neighboring trigger information, buffer data in DDR3 memory, and send event data over 10 Gb Ethernet to the camera server; inter-crate communication uses InfiniBand CXP-based links in a triangular topology. The camera paper also gives a total camera power of about 2 kW, with about 500 W for the PDP and about 1.2 kW for DigiCam, and describes heat-pipe cooling coupled to water cooling lines (Rajda et al., 2015, Schioppa et al., 2015).

A recurrent point in the design literature is that DigiCam should not be identified with the sensor plane alone. The PDP is optimized for photon collection and front-end conditioning, whereas DigiCam is the digitization, buffering, triggering, and event-building subsystem. This separation underlies the maintainability and firmware reconfigurability emphasized in the early design papers (Rajda et al., 2015).

3. PDP interface, signal path, and trigger logic

The PDP comprises 1296 hexagonal pixels. In the 2015 camera paper, these are arranged into 108 modules of 12 pixels each. Each pixel uses a hollow hexagonal Winston cone feeding a custom hexagonal SiPM sensor; the sensor comprises 36,840 square microcells with 50 $^\circ$0m pitch on a 93.56 mm$^\circ$1 area and is segmented into four channels to reduce effective capacitance and increase bandwidth. A PreAmp board sums the four SiPM channels into one analog output per pixel, while a Slow Control Board provides per-sensor bias voltage, routes analog outputs to DigiCam, and performs temperature-compensated gain stabilization using an NTC probe integrated in each sensor, a 2 Hz update rate, and 6.69 mV bias-voltage precision (Schioppa et al., 2015).

The front-end is DC-coupled, enabling baseline monitoring and direct night-sky-background estimation. The design literature quotes expected night-sky background of about 30 MHz per pixel on dark nights and SiPM dark noise of about 5–10 MHz. In the commissioning papers, the operational consequence is explicit: the SiPM-based camera can operate under high NSB by increasing trigger thresholds rather than implementing sensor-protection strategies, and per-pixel trigger thresholds can be adjusted dynamically (Schioppa et al., 2015, Tavernier et al., 2024).

DigiCam trigger formation is fully digital. The design papers describe a two-level hierarchy: L0 trigger primitives are computed on digitizer boards from local three-pixel sets, or “triplets,” using multiplicity and/or clipped-sum logic; L1 is computed on the crate trigger board by summing L0 values over 7-triplet or 19-triplet hexagonal patches and applying thresholds. The implemented 7-triplet neighborhood trigger algorithm used only about 2% of trigger-FPGA resources and ran at twice the required processing speed in the 2015 implementation. Because local and neighboring trigger paths have different transport latencies, the system applies a fixed delay alignment of about 256 ns, corresponding to 64 samples at 250 MS/s, so that L0 data presented to the L1 engine correspond to the same sampling point (Rajda et al., 2015, Prandini et al., 2015).

The commissioning literature documents a chronological transition in stereo operation. In the ICRC 2023 analysis, stereo trigger was not yet implemented and stereo events were reconstructed offline by matching local timestamps after image cleaning. The later technical summary describes stereo triggering via SWAT operating on nanosecond-precision timestamps from White Rabbit synchronization. This suggests a progression from offline stereoscopy toward integrated array-level triggering during commissioning (Tavernier et al., 2024, Tavernier et al., 2024).

4. Calibration strategy and charge reconstruction

The calibration chain converts DigiCam waveforms into calibrated pixel charges in photoelectrons and per-pixel timing. The 2015 calibration paper describes pedestal subtraction, gain calibration, flat-field equalization, NSB-dependent gain corrections, cross-talk measurement, and time alignment; the commissioning paper updates this with the procedures actually used in on-sky analysis, notably dark runs at the beginning and end of each observation night and baseline subtraction followed by fixed-window waveform integration (Prandini et al., 2015, Tavernier et al., 2024).

For physics runs in the commissioning analysis, per-pixel waveforms are baseline-subtracted and integrated over a 28 ns window starting 12 ns before the local pulse maximum. For dark runs used in gain and crosstalk calibration, multiple photo-electron spectra are formed from integrated ADC counts over 15 successive samples randomly selected within the readout window; at 250 MHz this corresponds to 60 ns. The dark-run multiple photo-electron spectrum is modeled by a Generalized Poisson including dark pulses and optical crosstalk,

$^\circ$2

with the measured ADC spectrum given by

$^\circ$3

The crosstalk-corrected conversion factor is

$^\circ$4

applied after baseline or pedestal subtraction (Tavernier et al., 2024).

The earlier calibration paper formulates the charge extraction at waveform level as

$^\circ$5

followed by conversion to photoelectrons through the single-photoelectron scale and NSB-dependent corrections. It also describes continuous gain stabilization through temperature-compensated bias control and the use of dark count, pedestal, and flasher events during normal operation (Prandini et al., 2015).

Muon-ring calibration provides an independent optical and camera-performance cross-check. In the commissioning analysis, muon selection uses tailcuts of 5 p.e. and 4 p.e. with at least two neighboring pixels, a completeness requirement of at least 8 angular bins out of 12 with at least 7 p.e., and rejection of events with charge $^\circ$6 p.e. outside selected ring pixels. The surviving muon-ring rate is about 0.1 Hz. The intensity of muon rings versus ring radius is linear in both data and Monte Carlo, but the Monte Carlo optical efficiency is 10–20% higher than data, likely due to mirror reflectivity modeling. The D80 is $^\circ$7 mm (stat) for SST-1M-1 and $^\circ$8 mm (stat) for SST-1M-2, consistent with star-based PSF estimates of 9.6 mm and 10.8 mm (Tavernier et al., 2024).

5. Reconstruction pipeline and observatory software integration

Offline analysis is performed with the ctapipe-based sst1mpipe pipeline. In the 2024 summaries, this pipeline performs SiPM response calibration, event image cleaning, gamma/hadron separation, and energy and direction reconstruction using random forests trained on Monte Carlo simulations. It produces DL3 data in the Gamma-ray Astronomy Data Format, and higher-level products such as sky maps and spectra are produced with gammapy (Tavernier et al., 2024, Tavernier et al., 2024).

The commissioning analyses also specify concrete image-cleaning and sky-mapping choices. For Crab analysis in the ICRC 2023 paper, tailcuts thresholds are 8 p.e. and 4 p.e. with a two-neighbor requirement, followed by a gammaness cut of 0.8 and a $^\circ$9 cut of 0.02 deg$< 0.25^\circ$0, corresponding to $< 0.25^\circ$1 for 68% containment at $< 0.25^\circ$2 zenith. In the later technical summary, sky maps use the gammapy ring background method with ring inner radius $< 0.25^\circ$3 and outer radius $< 0.25^\circ$4, and maps are convolved with a disk kernel of $< 0.25^\circ$5 radius (Tavernier et al., 2024, Tavernier et al., 2024).

At telescope-control level, DigiCam is represented by a dedicated ACS component. The SST-1M control software follows the ALMA Common Software framework, with subsystem communication exposed through OPC UA. A DigiCam-specific OPC UA bridge wraps the bespoke control and monitoring protocol, exposing telemetry and control points to the DigiCam ACS component and to the engineering web-GUI. The telescope master connects to DigiCam in the Ready state and commands it to start triggering and data acquisition in the Observing state, after PDP high voltages are up and the camera lid is open. Commissioning tests exercised continuous readout of DigiCam telemetry and successful data acquisition with an artificial light source (Sliusar et al., 2017).

6. Commissioning results, limitations, and evolving status

The commissioning literature presents DigiCam as an operational system undergoing refinement rather than a frozen design. In the winter 2023/2024 Crab campaign, wobble observations at $< 0.25^\circ$6 and $< 0.25^\circ$7 offsets yielded a total dataset of 22.03 hours of stereo observation after quality selection, with a gamma-ray excess of 176.1 events and a Li & Ma significance of $< 0.25^\circ$8. The same paper notes that the conclusion mentions 22.5 hours of stereo observation in aggregate, whereas earlier sections state 22.03 hours after run quality selection. The $< 0.25^\circ$9 distribution of the gamma-ray excess was fitted with a Gaussian PSF and compared to Monte Carlo weighted to the Crab spectrum for the same livetime, although no numerical fit parameters were reported (Tavernier et al., 2024).

An earlier commissioning analysis, based on a shorter stereo dataset of 2 h 38 m at zenith $0.09^\circ$0–$0.09^\circ$1, reported an excess of 18 events with Li–Ma significance $0.09^\circ$2, compared with a Monte Carlo expectation of $0.09^\circ$3 events for the same time. This earlier result is consistent with the later, larger Crab dataset and illustrates the maturation of calibration and reconstruction during commissioning (Tavernier et al., 2024).

Beyond the Crab Nebula, the later technical summary reports Markarian 421 observations on 13 March 2024 with livetime of about 3.3 hours, detection significance $0.09^\circ$4, excess observed up to 7 TeV, and flux close to the Crab, as reported in Astronomer’s Telegram #16533. An additional about 5.0 hours on 17 March 2024 produced a significance map. The authors state that the commissioning datasets are being used to tune Monte Carlo and pipeline components, and that the implemented calibration and reconstruction processes are validated by the Crab detections (Tavernier et al., 2024).

The same papers also define current limitations. The commissioning analysis identifies 20–30 pixels in SST-1M-2 with lack of gain and notes ongoing maintenance. It reports that mirror-reflectivity modeling makes Monte Carlo optical efficiency 10–20% higher than data, implying possible energy-threshold bias in the current analysis. The 2024 papers do not specify several lower-level DigiCam parameters that were prominent in the 2015 electronics papers, including detailed trigger hierarchy beyond the broad FPGA description, latency, dead time, buffer depth, absolute trigger thresholds, per-event data size, throughput, and bandwidth. Next objectives include exploiting the wide field of view for extended sources and improving responsiveness to alerts and targets of opportunity (Tavernier et al., 2024, Tavernier et al., 2024).

DigiCam therefore occupies a specific place in SST-1M: it is neither only a camera housing nor only a trigger board, but the integrated digital system that links the SiPM-based focal plane to Cherenkov event reconstruction. The 2015 papers established a fully digital, firmware-reconfigurable, dead-time-free architecture; the 2024 commissioning papers show that this architecture is functioning on sky, with stereo observations, validated calibration chains, and preliminary astrophysical detections, while still undergoing optimization in trigger coordination, Monte Carlo tuning, and hardware maintenance (Rajda et al., 2015, Tavernier et al., 2024).