Argus Array: Multiplexed Telescope System

• Argus Array is a large-scale optical telescope array composed of hundreds to thousands of small-aperture, wide-field telescopes arranged in a hemispherical structure.
• It employs synchronized tracking and advanced tiling algorithms to achieve rapid, continuous, and deep all-sky surveys for transient phenomena.
• The system integrates commercial components with GPU-accelerated pipelines to enable cost-effective, scalable, and real-time data processing.

The Argus Array denotes a new class of large, multiplexed optical telescope systems structured as tightly-packed arrays of hundreds to thousands of wide-field, small-aperture telescopes. Its most prominent realization is the Argus Optical Array, a 900-telescope, arcsecond-resolution, 5-meter effective-aperture facility constructed with commercial off-the-shelf telescopes, CMOS detectors, and a novel hemispherical “pseudofocal” mounting architecture. Argus Arrays are optimized for simultaneous high-cadence and deep all-sky surveys, offering the ability to capture synoptic, multi-color imaging at sub-second cadence across nearly half the sky, at a small fraction of the cost and operational complexity of traditional large single-mirror telescopes. This approach enables large-scale, continuous monitoring for fast transients, microlensing, multi-messenger events, and a broad portfolio of time-domain science cases.

1. Optical and Mechanical Architecture

The Argus Optical Array is constructed from 900 commercial Planewave Schmidt–Cassegrain astrographs (203 mm aperture, f/2.8), each delivering a 9 deg$^2$ FoV and arcsecond-level resolution across a $61$ MPix back-illuminated Sony IMX455 CMOS sensor (1.38″ pixel$^{-1}$). The telescopes are rigidly mounted on the interior of a $R_c\approx6$ m hemispherical bowl structure (diameter ≈10 m), forming an “inside-out” dome encompassing nearly half the sky at any instant (Law et al., 2022).

Each telescope’s optical axis is normal to its local sphere, such that all boresights intersect at the geometric center of the array, the “pseudofocal point.” This allows all beams to exit through a single, non-moving window situated along the virtual polar axis. The entire hemisphere is suspended from dual pivots, defining a virtual equatorial mount: a top ball-joint bears the structural load, while the lower precision pivot determines orientation. Sidereal tracking is achieved by rotating the hemisphere about its polar axis with a single linear actuator, enabling exposure durations of $\sim$9–15 minutes between rapid “ratchet” adjustments. Only this actuator, the entire rotating hemisphere, and per-telescope flip-shutters operate as moving elements at night, minimizing the complexity compared to conventional distributed mount arrays (Corbett et al., 2022, Law et al., 2022, Law et al., 2021).

2. Sky Packing, Coverage, and Mounting Strategies

Optimal packing of the Argus Array’s FoVs onto the celestial hemisphere is achieved via a “stripes of constant declination” tiling: rectangular FoVs are arranged in declination bands, with small ($\sim$1%) overlaps in right ascension and declination to eliminate holes, and stripe spacings adapted near the poles to mitigate spherical projection gaps. For $N\approx900$ telescopes, per-camera FoV_x ≈ $2.45^\circ$, FoV_y ≈ $3.67^\circ$, yielding $\gtrsim7850$ deg$^2$ of instantaneous sky coverage and overall overlap fraction of $\sim$3.4%. Geometric algorithms are used to evaluate coverage, overlap statistics, and to dynamically assign pointings to subarrays (“pattern constant” approach) when implementing multi-dome architectures (Galliher et al., 2022).

Traditional individual-mount telescope arrays require thousands of active mechanisms; the Argus bowl array enables all telescopes to share tracking, and reduces external moving parts to a minimum. Subarray division (e.g., $N=7$ domes of $\sim$130 telescopes) is possible, with careful sky-tiling ensuring each subarray sweeps the full nightly sky, but at the cost of extra infrastructure (Galliher et al., 2022).

3. Detector, System Performance, and Data Generation

Each telescope utilizes a $61.8$ MPix sCMOS sensor (Sony IMX455): pixel scale 1.38″/pix, read noise $\sim$1.7 $e^-$ at 0.4 s, $>90\%$ quantum efficiency at 500 nm, dark current $<$0.01 $e^-$/s/pix at $-5^\circ$C, and low dead time ($\sim$80 $\mu$s). The combined mosaic yields $54.9$ GPix per exposure, at 16-bit depth (122 MB/image/telescope).

Survey cadence modes include:

• High-speed: 1 s integrations over 20–47% of the sky to $m_g\approx16.1$
• Standard: 9 min sidereal tracking, 48 s “ratchet” (slew), alternating $g$ and $g{+}r$ filters
• Coadding stacks: Up to 5-night deep coadds to $m_g\approx23.6$ over nearly half the sky (Law et al., 2021, Corbett et al., 2022, Law et al., 2022).

Data rates are extreme: at 1 s cadence, the system reaches $\sim$11 Tbps ($4.3$ PB/night); at baseline 30 s cadence, $367$ Gbps ($145$ TB/night). This requires a fully hierarchical data processing system (Argus-HDPS), with ~19 Camera Command (CC) nodes each managing $\sim$50 cameras, local SSD cache ($>$50 TB/node), and GPU-accelerated real-time processing. The rollout has been validated in a $9$-telescope technology demonstrator (Corbett et al., 2022).

4. Data Processing, Reduction Pipelines, and Alert Generation

The real-time control and calibration pipeline (Argus-HDPS) executes the following steps per exposure:

• GPU-based frame calibration (dark, flat), background estimation, and source detection (<1–7 ms total)
• Astrometric solution on CPU (100–190 ms)
• Reprojection to HEALPix sky tiles (NSIDE=256, 13.7′ tiles)
• Image segmentation, storage as FITS-like cubes
• Matched-filter coaddition (Zackay et al. 2017): 15 min coadds reach $m_g\approx19.8$
• Fast difference imaging: direct subtraction (20 ms/tile at 1 s), ZOGY (1400 ms/tile at 30 s+)
• Daytime batch jobs: deep coadds, SysRem detrending, precision light curves (Corbett et al., 2022)

The instrument supports near-linear compute scaling with additional CC nodes, enabling future expansion. All components communicate asynchronously (HTTP APIs, Apache Plasma shared memory), supporting modular infrastructure. Real-time transient alerts and coadded images are released publicly via Kafka/ANTARES brokers (Corbett et al., 2022).

5. Survey Performance and Science Applications

The Argus Array enables:

• Continuous photometric monitoring of $10^7$–$10^8$ sources per exposure at sub-percent ($\sim$7 mmag) RMS (Corbett et al., 2022)
• Discovery of fast, rare, faint, and rapid transients, including prompt optical counterparts to FRBs, gravitational-wave events, supernovae, microlensing, and stellar flares
• Sub-second mapping of $\sim$8,000 deg$^2$, unique among ground-based facilities
• High-cadence temporal coverage (minute–hourly coadds to $m_g\sim$22–22.7, 5-night stacks to $m_g\sim$23.6 over 47–48% sky)
• Asteroid, NEO, KBO, and other Solar System body characterization with dense light curves ($10^5$–$10^6$ epochs)
• All-sky movies: multi-band, million-epoch photometric databases for time-domain astronomy (Law et al., 2021, Law et al., 2022)

Through its high étendue ($E_{\rm Argus}\sim2.3\times10^5$ m$^2$ deg$^2$; $\sim$100–1000$\times$ Rubin LSST for A$\cdot\Omega$), Argus provides discovery space for ultra-fast and rare phenomena, albeit with per-source depth–exposure compromises for full-field versus coadded depths (Law et al., 2021).

6. Limitations, Reliability, and Cost Structure

Major limitations arise from:

• Local storage and GPU memory requirements per node
• CPU-bottlenecked astrometry and I/O for full real-time reduction at the highest data rates
• Management of massive file counts ($\sim$786,000 HEALPix tiles per exposure set)
• Window ghosting and dome seeing, mitigated via AR coatings, insulation, and air management (Law et al., 2022, Corbett et al., 2022)

However, the cost structure is highly favorable: the full 900-telescope system is priced at %%%%52$-5^\circ$53%%%%6.1$M, dome/mount$3.0$M, compute/storage$2.1$M, and personnel/software$5.0$$M. Key factors are the abdication of individual telescope mounts and dome mechanics in favor of a single pivoting, inside-out structure (<a href="/papers/2107.00664" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Law et al., 2021</a>).</p> <h2 class='paper-heading' id='broader-array-concepts-and-related-systems'>7. Broader Array Concepts and Related Systems</h2> <p>Variants and related Argus-named arrays exist:</p> <ul> <li>ARGUS (Astronomical taRGets detection framework for Unified telescopes): a deep-learning-based, multi-WFSAT, real-time transient detection system implementing onboard Faster-R-CNN detectors and AdaBoost ensemble fusion (<a href="/papers/2011.10407" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Jia et al., 2020</a>)</li> <li>Argus (JAX state-space filtering for PTA): a Python/JAX Kalman-filter-based Bayesian inference package for gravitational-wave searches in <a href="https://www.emergentmind.com/topics/pulsar-timing-array-pta" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">pulsar timing array</a> data, achieving$$\mathcal{O}(N)$ scaling for likelihood evaluation (Kimpson et al., 13 Oct 2025)

Argus cross-camera video analytics: a distributed smart camera collaboration system for multi-camera spatio-temporal video analysis, not directly related to the astronomical Argus Array (Min et al., 2024)

8. Future Prospects and Upgrades

Prospective improvements include expansion to wider FoV via new AR coatings, multi-glazed windows for further thermal decoupling, and denser 2D packing to increase fraction-of-sky coverage. The pathway to larger aperture arrays (higher N, larger individual telescopes), alternate mounting strategies, and their use as prototypes for future massive, low-cost, array-based synoptic facilities is under investigation (Law et al., 2022, Law et al., 2021).

The Argus Array therefore defines a paradigmatic shift in survey telescope architecture toward large-scale multiplexing, optimized for time-domain sky monitoring and scalable, real-time data reduction, while preserving high photometric precision and operational efficiency at manageable cost (Corbett et al., 2022, Law et al., 2022, Law et al., 2021).