Modelling beam uncertainties in global 21cm experiments through Bayesian data analysis

Dominic Anstey, Eloy de Lera Acedo, Will Handley2023 · DOI

TL;DR

Detecting the faint 21cm signal from the early universe is like trying to hear a whisper in a hurricane; local radio "noise" is five orders of magnitude louder than the signal itself. This paper demonstrates that even tiny errors in antenna modeling can create artifacts that mimic the signal, but these can be overcome by using Bayesian analysis to "learn" the antenna's behavior and the cosmic signal simultaneously.

What Problem Does This Paper Address?

The "Global 21cm" signal is a holy grail for cosmologists, as it carries data from the Cosmic Dawn—the era when the first stars and galaxies began to light up the universe. However, capturing this signal is notoriously difficult for two main reasons:

  1. Bright Foregrounds: Radio emissions from our own galaxy and other local sources are up to 100,000 times brighter than the signal researchers are looking for.
  2. Beam Chromaticity: The "beam" (the sensitivity pattern of the antenna) changes depending on the frequency of the radio waves. If the antenna isn't modeled perfectly, these frequency-dependent variations couple with the bright foregrounds to create "systematics"—fake signals that look almost exactly like the real cosmological data.

The authors address the fact that "realistic" construction errors in antennas (small physical deviations from the design) can lead to incorrect beam models, which in turn lead to failed signal detections.

Methods

The researchers propose moving away from using a fixed, "perfect" theoretical beam model. Instead, they employ Bayesian data analysis to perform a joint fit.

  • Parameterized Beam Modeling: Rather than assuming the beam is exactly as designed, they describe the beam using a set of parameters that can vary.
  • Joint Inference: The software fits the beam parameters, the foreground noise, and the cosmological signal all at the same time.
  • Time-Dependency Exploitation: Because the Earth rotates, the foreground radio sources move across the sky while the antenna beam stays fixed relative to the ground. The authors suggest using this time-varying data to help the model distinguish between features caused by the sky (foregrounds) and features caused by the instrument (the beam).

Datasets

The paper utilizes simulated datasets to validate their approach:

  • Simulated Dipole Antenna: They modeled a dipole antenna with "realistic construction errors" to see how much a slight physical imperfection ruins the data.
  • REACH Framework: The work is situated within the context of the REACH (Radio Experiment for the Analysis of Cosmic Hydrogen) experiment, using the Bayesian framework previously developed for this project.

Key Insights and Ideas

  • The "Aha" Moment on Modeling: The most significant insight is that the antenna beam shouldn't be treated as a "known constant." In high-precision cosmology, the instrument itself is an "unknown" that must be solved for alongside the science.
  • Degeneracy Breaking: The paper highlights that beam errors and foregrounds are often "degenerate"—meaning different combinations of the two can produce the same observed data. By introducing time-dependent observations, they can "break" these degeneracies because the foregrounds change over time while the beam's internal errors do not.
  • Robustness to Imperfection: The results suggest that we don't necessarily need to build a "perfect" antenna; we just need a modeling framework flexible enough to account for the imperfections we inevitably build in.

Key Results

The authors present a stark comparison in Figure 1:

  • Without Joint Fitting: When using a standard model that doesn't account for construction errors (Beam A vs. Beam B), the recovered signal is wildly inaccurate, failing to match the "True Signal" (the red line in the left plot deviates significantly from the target).
  • With Joint Fitting: When the beam is fitted simultaneously with the data (the right plot), the reconstructed signal (blue shaded regions) successfully recovers the "True Signal" (green line) within statistical confidence intervals ($1\sigma$ to $3\sigma$). This proves that the Bayesian approach can "self-correct" for hardware uncertainties.

Limitations and Future Work

While the paper proves the concept, some challenges remain:

  • Computational Complexity: Jointly fitting beams, foregrounds, and signals is computationally expensive, especially as the number of parameters increases.
  • Complex Beams: The study focused on a relatively simple dipole antenna. Future work will need to determine if this method scales to more complex antenna geometries used in other global 21cm experiments.
  • Real-world Data: The next major step is applying this parameterized Bayesian fit to actual data collected by the REACH radiometer in the field, where environmental factors (like ground reflections) add further layers of complexity.

Summary updated

BibTeX
@misc{anstey2023modelling,
  author = {Dominic Anstey and Eloy de Lera Acedo and Will Handley},
  title = {{Modelling beam uncertainties in global 21cm experiments through Bayesian data analysis}},
  year = {2023},
  doi = {10.46620/ursigass.2023.0819.xrjy9189},
}
Figure