Foreground removal and 21 cm signal estimates: comparing different blind methods for the BINGO Telescope

TL;DR

This paper assesses three blind foreground-removal algorithms—FastICA, GMCA, and GNILC—for recovering the redshifted 21 cm signal in the BINGO intensity mapping pipeline. It uses a simulation-based debiasing framework and jackknife errors to compare reconstruction quality across mixing-dimension choices and numbers of realizations, finding no strong methodological preference but identifying $n_{\mathrm{s}}=3$ as a favorable choice and showing that 50–400 simulations yield equivalent debiased HI spectra. FastICA is validated in a map-making context with HIDE TODs, achieving robust HI power-spectrum recovery for five years of data, particularly at $z\lesssim0.39$ and $\ell$ in the range where signal dominates over noise, with an overall $SNR$ of $204$ and $\chi^2 \approx 1.8$. The results demonstrate that blind foreground separation can yield statistically consistent HI reconstructions in this idealized setting, supporting BAO-focused cosmology with BINGO while highlighting edge effects in the highest-redshift channel and the importance of simulation-based debiasing. These findings inform practical choices for foreground cleaning trade-offs, computational cost, and data-analysis pipelines for upcoming 21 cm IM surveys.

Abstract

The BINGO radiotelescope will observe hydrogen distribution using Intensity Mapping (IM) to analyze the Dark Energy paradigm through Baryon Acoustic Oscillations. The target signal is contaminated by unwanted signals and instrumental noise, making accurate estimations essential for characterizing the 21 cm signal. In this study, we evaluated the performance of three blind foreground-removing algorithms - FastICA, GNILC, and GMCA - on the BINGO pipeline. Each method used different approaches to estimate foreground contributions, and we also investigated how the number of simulations for debiasing affects estimation quality. Our findings indicate that using 50 or 400 simulations yields equivalent results at this stage of analysis. All algorithms produced statistically consistent estimates of the 21 cm signal. We used FastICA for estimating and debiasing the HI spectra from five years of observations, which yielded reliable results, although the first channel was affected by edge effects from the mixing matrix. The overall signal-to-noise ratio was 204, and the chi-squared value was 1.8.

Figures (11)

• Figure 1: Intensity maps in the BINGO sky region. The maps are in antenna temperature and represent: observed (top left), synchrotron (top middle), free-free (top right), radio point sources (middle left), AME (center), CMB (middle right), HI (bottom left), thermal noise for 1-year mapping at 70 K system temperature (bottom center), and HI + thermal noise (bottom right). The gray region represents the masked part of the Galaxy.
• Figure 2: Angular power spectrum masked (solid line) and unmasked (dashed line) within the BINGO coverage region for all foreground components assumed in this work: CMB (gray), AME (orange), free-free (red), FRPS (green), and synchrotron (blue).
• Figure 3: Effective $\chi^2$ (i.e. $\chi^2$ per channel) for mixing-matrix dimensions from 2 to 4, comparing (left) FastICA and (right) GMCA. In each panel, the smaller plot is a zoom-in of the main plot to highlight the behaviour of the $n_{\mathrm{s}}=3$ curve. Error bars are jackknife uncertainties from 400 realizations. The black and grey dashed lines are for reference only.
• Figure 4: Effective $\chi^2$ per channel for the three algorithms. The results are for five numbers of simulations from 25 (yellow) to 400 (black).
• Figure 5: The plots are the $\chi^2$ (Eq. \ref{['eq: xi2']}) heat map of the three algorithms. The horizontal axis corresponds to the channel, and the vertical one to the estimated H i angular power spectrum corresponding to a specific multipole. The redder values concerning H i angular power spectrum are harder to estimate.