BINGO: BAO from Neutral Gas Observations

• The paper introduces BINGO, a dedicated radio intensity mapping experiment that detects BAO via the redshifted 21-cm HI signal across z≈0.13–0.45.
• It employs an innovative crossed-Dragone optical design, pseudo-differencing receivers, and advanced component separation techniques to minimize systematics.
• Simulations and Fisher forecasts demonstrate that BINGO can deliver competitive constraints on dark energy and cosmological parameters as a pathfinder for future surveys.

Baryon Acoustic Oscillations from Integrated Neutral Gas Observations (BINGO) is a purpose-built, single-dish radio intensity-mapping experiment designed to make the first detection of Baryon Acoustic Oscillations (BAO) at radio wavelengths. By measuring the large-scale fluctuations in redshifted 21-cm emission from neutral hydrogen (HI) at $0.13 < z < 0.45$, BINGO aims to constrain the expansion history of the universe and the equation of state of dark energy independently of optical galaxy surveys. BINGO’s architecture, methodology, and data-analysis pipelines are optimized for robust systematics control, precise calibration, and competitive cosmological parameter inference within the $\Lambda$CDM and $w_0w_a$CDM models.

1. Instrumentation, Optical Design, and Survey Strategy

BINGO’s optical system comprises two fixed, off-axis reflectors (primary: $\sim$40 m paraboloid; secondary: $\sim$36 m hyperboloid), following a crossed-Dragone configuration to achieve low aberrations, minimal sidelobes ($\lesssim-25$ dB), and polarization leakage $\lesssim-40$ dB. The focal plane consists of 28 corrugated conical feedhorns ($\sim$1.7–1.9 m diameter, $\sim$4.3 m length), each with dual-polarization, producing Gaussian beams with FWHM $\theta_\mathrm{FWHM} \approx 40'$ at 1 GHz across a $\Lambda$0 instantaneous field of view. All horns are stationary; the survey operates in drift-scan mode, covering a fixed $\Lambda$1 declination strip centered at $\Lambda$2, yielding $\Lambda$35300–5400 deg$\Lambda$4 sky area (Wuensche et al., 2018, Wuensche et al., 2021).

Receiver architecture is based on room-temperature correlation radiometers with pseudo-differencing via “magic-tee” hybrids and InP HEMT low-noise amplifiers, achieving a system temperature $\Lambda$5–$\Lambda$6 K. Laboratory measurements confirm insertion loss $\Lambda$7 dB, cross-polarization $\Lambda$8 dB, and return loss $\Lambda$9 dB across 980–1260 MHz (Wuensche et al., 2019, Wuensche et al., 2021).

Frequency coverage of 980–1260 MHz ($w_0w_a$0) is divided into 30 bins ($w_0w_a$1 MHz). Each $w_0w_a$2 beam pixel in a $w_0w_a$3 MHz channel achieves $w_0w_a$41 day dwell time per year, reaching r.m.s.\ map noise $w_0w_a$584–102 $w_0w_a$6K for one year, 28 horns, and 60% duty cycle (Wuensche et al., 2021, Motta et al., 15 Jan 2026).

Drift-scan mapping ensures uniform declination coverage as the sky passes through the field-of-view. The focal-plane layout is optimized for uniform exposure by annual shifts in horn elevation, and the survey area is refined by masking regions close to the Galactic plane. Site selection surveys at Serra do Urubu (Paraíba, Brazil) found RFI contamination $w_0w_a$7 dBm ($w_0w_a$8 mK, well below the HI signal) (Peel et al., 2018).

2. Intensity Mapping Formalism and BAO Extraction

BINGO exploits the 21-cm intensity mapping (“IM”) technique, integrating HI emission over large voxels without resolving individual galaxies. The mean differential HI brightness temperature at redshift $w_0w_a$9 is modeled as: $\sim$0 where $\sim$1 is the cosmic HI density fraction, $\sim$2 is the dimensionless Hubble parameter, and $\sim$3 is the Hubble rate (Wuensche et al., 2018, Costa et al., 2021).

The 3D HI power spectrum is: $\sim$4 with $\sim$5 the HI bias and $\sim$6 the underlying matter power spectrum. $\sim$7 denotes the thermal noise, determined by the radiometer equation and voxel integration time (Wuensche et al., 2018, Costa et al., 2021). BAO manifest as sinusoidal modulations (“wiggles”) in $\sim$8 at $\sim$9–$\sim$0; their angular analogue $\sim$1 is used for tomographic bins. The volume-averaged distance $\sim$2 and the BAO acoustic scale are quantified via: $\sim$3 The fractional error on the acoustic scale, from Fisher forecasts, is $\sim$4 after 3–5 years (Wuensche et al., 2018, Dickinson, 2014).

3. Data Analysis, Simulations, and Component Separation

The analysis pipeline incorporates realistic sky realizations, time-ordered data (TOD) generation (including Galactic foregrounds, CMB, point sources, and atmospheric emission), Gaussian beam convolution, and instrument noise. Map-making employs naive binning, destriping, and maximum-likelihood solutions.

Intensity-mapping foregrounds (dominant: synchrotron, free-free; secondary: AME, extragalactic point sources) are spectrally smooth. Component separation uses blind algorithms—FastICA, GNILC, and GMCA—tuned for the BINGO frequency domain (Marins et al., 2022, Liccardo et al., 2021). Simulations indicate that three non-physical foreground templates suffice for effective separation, with FastICA favored for computational efficiency at the current stage. GNILC leverages spatial/frequency localization (needlet basis) and an adaptive subspace criterion (AIC), while GMCA employs sparsity in wavelet space. All methods achieve robust BAO reconstruction with signal residuals $\sim$5, improving with integration time and horn count (Liccardo et al., 2021).

End-to-end mission simulations (lognormal HI signal based on Planck cosmology) confirm angular power-spectrum recovery to $\sim$6 up to $\sim$7 and demonstrate that realistic systematics (beam, thermal noise, residual foregrounds) can be mitigated to tolerable levels (Motta et al., 15 Jan 2026).

4. Cosmological Forecasts and Parameter Constraints

The BINGO 21-cm power spectrum provides sensitivity to the expansion history, neutrino mass, and non-standard cosmologies via BAO and full APS measurements. Fisher-matrix and Bayesian joint analyses combining BINGO with Planck (TT/TE/EE) yield significant improvements:

• For $\sim$8CDM, $\sim$9 marginalized uncertainties (Planck+BINGO vs Planck):
  • $\lesssim-25$0: 0.23 km s$\lesssim-25$1 Mpc$\lesssim-25$2 (63% reduction)
  • $\lesssim-25$3, $\lesssim-25$4: 58–61% reduction (Motta et al., 15 Jan 2026)
• For $\lesssim-25$5CDM:
  • $\lesssim-25$6, $\lesssim-25$7 (Motta et al., 15 Jan 2026)
  • $\lesssim-25$8, $\lesssim-25$9 with BAO-only+Planck for CPL parametrization (Ostergaard et al., 2024)
• Full power-spectrum analyses using HI APS outperform BAO-only approaches, but the BAO signature is less susceptible to bias from uncertain parameters or foreground residuals (Ostergaard et al., 2024).

BINGO independently measures $\lesssim-40$0 and $\lesssim-40$1, constraining the HI mass function and reducing systematics in cosmological inference (Motta et al., 15 Jan 2026).

The inclusion of redshift-space distortions (RSD) and cross-shell $\lesssim-40$2 increases parameter sensitivity and robustness; finer binning up to $\lesssim-40$3–128 improves tomographic leverage until limited by noise/systematics (Costa et al., 2021).

5. Systematics: Mitigation and Calibration

Systematic error control is central to the BINGO design:

• RFI Mitigation: Site selection at Serra do Urubu provides $\lesssim-40$4 dB terrain screening and an RFI floor $\lesssim-40$5 mK. Continuous RFI monitoring and digital filtering are implemented (Peel et al., 2018).
• Gain Stability and $\lesssim-40$6 Noise: Pseudo-correlation receivers, noise diodes (injected $\lesssim-40$71 K signals), and frequent celestial calibrator transits constrain gain drifts to $\lesssim-40$8 per integration and achieve a knee frequency $\lesssim-40$9 mHz (Liccardo et al., 2021, Wuensche et al., 2021).
• Polarization Leakage: Corrugated horns, circular polarizers, and OMTs deliver cross-polarization isolation $\sim$0 dB, minimizing $\sim$1 leakage and controlling foreground contamination (Wuensche et al., 2019).
• Component Separation Residuals: Simulated cleaning efficacy with GNILC, FastICA, GMCA is $\sim$2 error in APS at BAO scales, with $\sim$30.5% bias in acoustic scale $\sim$4 (Liccardo et al., 2021).

Noise and foreground residuals are incorporated into the Bayesian inference and Fisher-matrix framework, ensuring robust error propagation to final cosmological estimates (Motta et al., 15 Jan 2026, Costa et al., 2021).

6. Extended Science, Ancillary Capabilities, and Future Upgrades

In addition to cosmology, BINGO enables extragalactic HI science, Galactic emission mapping, and transient detection:

• Fast Radio Bursts (FRBs): The BINGO-ABDUS extension introduces phased-array feeds and outlier stations for real-time FRB detection and localization ($\sim$5), with forecasted rates of up to $\sim$6180 year$\sim$7 depending on array size (Abdalla et al., 2023).
• Pulsar Science: Drift scan enables daily coverage of dozens of southern-sky pulsars, supporting timing-array efforts and polarimetric studies (Abdalla et al., 2021).
• Pathfinding for SKA: BINGO’s instrument and analysis pipeline serve as validation for next-generation SKA intensity-mapping experiments.

The ABDUS upgrade involves focal-plane phased arrays to enhance survey speed and real-time RFI excision, and outrigger interferometric stations with baselines up to 20 km for high-precision localization of transient phenomena (Abdalla et al., 2023).

7. Comparison with Contemporaneous Experiments and Outlook

Standalone BINGO constraints are less stringent than future surveys like SKA1-MID ($\sim$8, $\sim$9 with Planck), but are competitive with optical BAO/dark energy experiments (DESI, Euclid) and other HI IM projects in the same redshift range (Yohana et al., 2019, Costa et al., 2021). BINGO’s key strength is the independence of its systematics and redshift leverage ($\sim$0–$\sim$1) compared to optical and higher-$\sim$2 HI projects (e.g., CHIME, HIRAX).

Multi-year, multi-horn operation, coupled with robust calibration and systematics control, positions BINGO as the first experiment likely to deliver a statistically significant ($\sim$3) HI BAO detection at radio wavelengths, with forecasted fractional distance-scale precision $\sim$4 over its redshift interval (Wuensche et al., 2018, Liccardo et al., 2021, Ostergaard et al., 2024). These results break key degeneracies with CMB-alone data and deliver independent constraints on dark energy at the percent level (Motta et al., 15 Jan 2026, Ostergaard et al., 2024).

---

References:

(Wuensche et al., 2018, Wuensche et al., 2021, Liccardo et al., 2021, Costa et al., 2021, Abdalla et al., 2021, Marins et al., 2022, Wuensche et al., 2019, Peel et al., 2018, Motta et al., 15 Jan 2026, Dickinson, 2014, Yohana et al., 2019, Abdalla et al., 2023, Ostergaard et al., 2024, Olivari et al., 2017).