SONG: Global Robotic Stellar Observatory

Updated 19 November 2025

SONG is a global network of 1-m robotic telescopes designed to deliver high-precision radial-velocity and imaging data for asteroseismology and exoplanet detection.
It integrates advanced instrumentation such as high-resolution echelle spectrographs and lucky imaging cameras with automated scheduling to achieve sub-meter-per-second precision and diffraction-limited imaging.
The network achieves continuous, homogeneous temporal coverage with >80% duty cycle and robust fault-handling, significantly advancing stellar structure analysis and exoplanet demographics.

The Stellar Observations Network Group (SONG) is a global, distributed facility of 1-m class, fully robotic telescopes designed to deliver high-precision radial-velocity (RV) time series and high-angular-resolution imaging for asteroseismology, exoplanet detection, and microlensing science. Leveraging identical instrumentation, automated scheduling, and seamless data flow, SONG achieves networked, continuous coverage of bright targets at photometric and spectroscopic precision previously unattainable from single sites, enabling unprecedented constraints on stellar structure and exoplanet demographics (Andersen et al., 2019, Andersen et al., 2019, Creevey et al., 2011).

1. Network Architecture, Sites, and Infrastructure

SONG is conceived as a longitudinally distributed network of eight 1-m telescopes—four per hemisphere—to provide continuous, homogeneous temporal coverage for time-domain stellar science (Creevey et al., 2011, Uytterhoeven et al., 2011, Tian et al., 2016). Operating at primary nodes including Teide Observatory (Tenerife, 28°N), Delingha (Qinghai–Tibetan Plateau, 32°N), and Mt Kent (Australia), each node features:

A 1-m primary mirror (Zerodur, 5 cm thick), active optics (30 Shack–Hartmann actuators), and an alt-azimuth mount (slew rate up to 20° s⁻¹, <5″ blind pointing) (Andersen et al., 2019, Creevey et al., 2011).
Fully robotic operation: weather stations, all-sky cameras, rain/wind/humidity sensors, robotic opening/closure, UPS-backed power (Andersen et al., 2019).
Dome enclosure (∼5 m Ash-type) for rapid thermal control and minimal gradients (Creevey et al., 2011).
Robust network protocols: database replication via Slony-I, secure remote access, data archiving to central “SODA” servers (Andersen et al., 2019).
Key site metrics (for Delingha): ∼248 usable nights yr⁻¹ (fᵤ=0.68), median seeing 1.58″, median sky brightness 21.5 mag arcsec⁻², operational reliability ≥95% (Tian et al., 2016).

Global deployment enables >80% mean duty cycle (higher with full southern augmentation), suppression of daily spectral aliases, and tailored node parameters via per-site configuration (Andersen et al., 2019, Arentoft et al., 2013).

2. Instrumentation: High-Stability Spectrographs and Imaging Systems

Each SONG node is equipped with:

Coudé-fed, stabilized, high-resolution echelle spectrograph: λ=4400–6900 Å, resolving power R=60 000–180 000 (typical R≈90–120 000) (Andersen et al., 2019, Uytterhoeven et al., 2011, Creevey et al., 2011).
- Iodine absorption cell calibration for meter-per-second RV precision (single-exposure σ_v ≲1–3 m s⁻¹ for V≲6 in 5–10 min) (Creevey et al., 2011).
- Temperature regulation (±0.01 K), vacuum enclosures (where implemented), and active optics ensure long-term wavelength stability (Andersen et al., 2019).
- Andor Ikon-L 2k × 2k or equivalent CCDs, low read noise, fast parallel readout.
Lucky Imaging Cameras: Dual-band EMCCD system (512 or 1k² sensor, 0.09–0.1″/pix), enabling high-cadence (10–35 Hz) imaging, diffraction-limited (FWHM ≲0.2–0.25″) in optimal conditions (Skottfelt et al., 2014, Creevey et al., 2011).
- Dichroic splitting (λ<655 nm to “visual”, >655 nm to “red”) permits simultaneous two-color photometry.
- Odin software for simultaneous, real-time reduction and monitoring, supporting autonomous operation (Skottfelt et al., 2014).
Optional auxiliary ports: Atmospheric dispersion correction, image derotation, and additional filter wheels (Skottfelt et al., 2014).
Instrument Switching: Movable M3 tertiary allows rapid (≤1 min) transition between spectrograph and imager with minimal overhead (Andersen et al., 2019).

End-to-end throughput in spectroscopy typically reaches 10–15% at 5500 Å; spatial resolution in imaging approaches the telescope diffraction limit for the best-seeing selection (Andersen et al., 2019, Skottfelt et al., 2014).

3. Scheduling, Robotic Operation, and Data Flow

SONG’s operational workflow is mediated by a sophisticated trio of software subsystems: Conductor, Scheduler, and Monitor (Andersen et al., 2019):

Conductor: Python service, querying SODA PostgreSQL archive at ∼10 s cadence to pull active targets, applying an explicit priority ranking for each scheduling class (time-critical, standards, large programs, periodical, filler, backup), using equations such as: $p_{\text{periodical}} = (\Delta t/\Delta t_{\text{base}})\times 100$ , $p_{\text{filler}} = p_{\text{pre}} + f/(|h-50^{\circ}| + 1)$ .
Scheduler/Monitor: On-site daemons execute ORs (Observing Requests), check real-time weather, perform instrument control, and autonomously handle safety closures. Integration of environmental data ensures immediate adaptation to changing conditions (clouds, gusts, humidity, alarms).
Database and Replication: Observational requests, status, and metadata are maintained centrally in SODA, mirrored at each node. Slony-I provides transactional replication; data delivery utilizes GlusterFS and rsync for bulk transfer (Andersen et al., 2019).
Autonomous Fault Handling: On failures, the Monitor initiates safe shutdown and notifies staff. Operations resume automatically on recovery (Andersen et al., 2019).

This system achieves a measured on-sky efficiency of 56–63%, with per-target overhead ≈107 s (slew, acquisition, readout), and achieves >78% integration efficiency even in highly segmented nights (Andersen et al., 2019, Andersen et al., 2019).

4. Science Drivers: Asteroseismology and Exoplanet Detection

SONG targets two central science objectives:

Asteroseismology: High-precision, continuous RV time series for solar-type and evolved stars enable detection and modeling of p-mode oscillations, mixed modes, and rotational splittings—delivering mass, radius, age, internal rotation, and convection-zone boundaries to percent-level precision (Arentoft et al., 2013, Grundahl et al., 2017, Knudstrup et al., 2023, Beck et al., 2020, Frandsen et al., 2018, Malla et al., 2020).
- Critical observables: ν_max (frequency of maximum power), Δν (large frequency separation), both extracted via robust power spectrum fitting, autocorrelation, and global optimization (Andersen et al., 2019).
- Scaling relations (with calibrated solar zero-points):
- $\nu_{\max} \simeq \nu_{\max, \odot} \left(\frac{M}{M_\odot}\right)\left(\frac{R}{R_\odot}\right)^{-2} \left(\frac{T_{\rm eff}}{T_{\rm eff,\odot}}\right)^{-1/2}$
- $\Delta\nu \simeq \Delta\nu_\odot \left(\frac{M}{M_\odot}\right)^{1/2} \left(\frac{R}{R_\odot}\right)^{-3/2}$ (Andersen et al., 2019).
- Comprehensive, dual-site and multi-site networks are required to resolve daily spectral aliases, reach sub-μHz frequency precision, and access higher-degree modes (ℓ=3), as shown in simulations and application to β Aql, γ Cep, 46 LMi, μ Her, Aldebaran (Kjeldsen et al., 31 May 2025, Knudstrup et al., 2023, Frandsen et al., 2018, Grundahl et al., 2017, Beck et al., 2020).
- SONG–space (e.g., TESS–BRITE) synergy enables cross-validation, amplitude ratio and phase measurements for oscillation diagnostics (Kjeldsen et al., 31 May 2025, Beck et al., 2020).
Exoplanet Science: Simultaneous RV and high-resolution photometry allow detection of hot Jupiters and lower-mass exoplanets, transit timing, microlensing planets, and improvement of planet-host fundamental parameters. High-cadence photometry from lucky imaging contributes to microlensing and transit searches, supporting discovery of low-mass and free-floating objects in the Galactic bulge (Skottfelt et al., 2014, Creevey et al., 2011).

5. Performance, Calibration, and Scientific Output

Key metrics benchmarked on the Tenerife prototype and subsequent network sites:

Metric	Value / Description	Source
Spectral resolution (R)	60,000–180,000 (typically 90,000–120,000)	(Creevey et al., 2011, Andersen et al., 2019)
Single-exposure RV precision (bright stars)	~1–3 m s⁻¹ (V≲6), ≤1 m s⁻¹ for V<3	(Creevey et al., 2011, Andersen et al., 2019)
Typical asteroseismic S/N (per-minute)	>40 for Sun-as-star campaigns	(Andersen et al., 2019)
Duty cycle (node-level)	≥58–65% (year, single node); >80% global	(Andersen et al., 2019, Arentoft et al., 2013)
On-sky efficiency	56–63% (proportion of integration to available)	(Andersen et al., 2019, Andersen et al., 2019)
Imaging resolution	0.2–0.25″ achievable in lucky imaging	(Skottfelt et al., 2014)
Nights >90% integration efficiency	Routinely achieved for single large campaigns	(Andersen et al., 2019)
Per-target overhead (slew+acq+readout)	∼107 s	(Andersen et al., 2019)

Spectroscopic reduction employs the iSONG pipeline and pyodine (highly automated), with rigorous calibration of instrumental response and atmospheric effects via standard stars, rigorous drift-monitoring, and injection of ThAr or I2 calibrations.

Empirical validation demonstrates agreement at the <0.1 μHz level for solar oscillations (ν_max, Δν), few μHz for subgiant and red giant campaigns, and parameter precisions of 1–4% in mass/radius, <0.02 dex in log g, and ~10–15% in age in seismic modeling for well-studied benchmark stars (Andersen et al., 2019, Grundahl et al., 2017, Frandsen et al., 2018).

6. Data Analysis, Calibration, and Scaling Relations

For all asteroseismic and photometric applications, SONG employs a rigorously calibrated analysis path:

Reference solar parameters (SONG): νmax,⊙ = 3141 ± 12 μHz, Δν⊙ = 134.98 ± 0.04 μHz, radial mode amplitude A_⊙ = 16.6 ± 0.4 cm s⁻¹ (Andersen et al., 2019).
Parameter extraction:
- Gaussian or Lorentzian envelope fits for ν_max.
- Autocorrelation for Δν.
- Mode “peak bagging” for individual frequencies, including rotational multiplets and mixed modes (Grundahl et al., 2017, Kjeldsen et al., 31 May 2025).
All scaling-relations should adopt SONG’s internal solar zero-points for consistency (Andersen et al., 2019).
For amplitude ratio and phase studies: simultaneous RV and photometric measurements, with amplitude ratio R ∝ T_eff⁻¹ and direct phase shift measurement (β Aql: –113 ± 7° for RV leads photometry) (Kjeldsen et al., 31 May 2025).

Uncertainties are propagated via analytic logarithmic differentials or Monte Carlo methods, with fractional errors on ν_max, Δν, T_eff, and solar references explicitly included in the mass and radius error budget (Andersen et al., 2019).

7. Network Expansion and Future Directions

SONG’s roadmap emphasizes (Andersen et al., 2019, Andersen et al., 2019, Arentoft et al., 2013):

Scaling to full eight-node global network: Each successor node (hardware identical or homologous, e.g., Mt Kent, Delingha, Tenerife) deploys its own Conductor instance, referencing the central SODA database. Campaigns can be scheduled globally or node-local, supporting continuous >90% duty cycles.
Adaptive scheduling and rapid-response: Dynamic OR injection for transient events (e.g., TESS/GRB), sub-minute interruption/restart capability.
Failover resilience: Local cache fallback (mini-Conductor), backup target list for cut WAN links, modular hardware/software for rapid recovery after failures.
Automation upgrades: Further reduction in required human interaction via fully autonomous pipeline, precise environmental monitoring, and centralized control.
Synergy with space missions: Cross-calibration and joint analyses with TESS, BRITE, and PLATO augment the scientific reach, improving mass, radius, and convection-parameter estimation, and enabling the removal of oscillation “jitter” in exoplanet RV searches (Kjeldsen et al., 31 May 2025).

The anticipated steady expansion and integration of new sites is expected to yield alias-free, continuous, and uniform time series for a large sample of bright stars, rivaling or exceeding the data quality of existing and planned space missions for many asteroseismic and exoplanet science cases (Arentoft et al., 2013, Kjeldsen et al., 31 May 2025).