[Paper Review] The Effect of Foreground Mitigation Strategy on EoR Window Recovery

Emma Chapman, Saleem Zaroubi+4 more|arXiv (Cornell University)|Aug 20, 2014

Radio Astronomy Observations and Technology2 references4 citations

TL;DR

This paper compares foreground avoidance and foreground removal strategies for recovering the Epoch of Reionization (EoR) 21-cm signal in interferometric radio surveys. Using simulations with LOFAR- and SKA-like noise levels, it finds that while foreground avoidance preserves signal in high-k_perp modes, foreground removal recovers significantly more signal at low k_los, especially in low-noise regimes, and remains robust under moderate foreground variations.

ABSTRACT

The removal of the Galactic and extragalactic foregrounds remains a major challenge for those wishing to make a detection of the Epoch of Reionization 21-cm signal. Multiple methods of modelling these foregrounds with varying levels of assumption have been trialled and shown promising recoveries on simulated data. Recently however there has been increased discussion of using the expected shape of the foregrounds in Fourier space to define an EoR window free of foreground contamination. By carrying out analysis within this window only, one can avoid the foregrounds and any statistical bias they might introduce by instead removing these foregrounds. In this paper we discuss the advantages and disadvantages of both foreground removal and foreground avoidance. We create a series of simulations with noise levels in line with both current and future experiments and compare the recovered statistical cosmological signal from foreground avoidance and a simplified, frequency independent foreground removal model. We find that while, for current generation experiments, foreground avoidance enables a better recovery at $k_{perp} > 0.6 \mathrm{Mpc}^{-1}$, foreground removal is able to recover significantly more signal at small $k_{los}$ for both current and future experiments. We also relax the assumption that the foregrounds are smooth by introducing a Gaussian random factor along the line-of-sight and then also spatially. We find that both methods perform well for foreground models with line-of-sight and spatial variations around $0.1\%$ however at levels larger than this foregrounds removal shows a greater signal recovery.

Motivation & Objective

• To evaluate the trade-offs between foreground avoidance (using an EoR window in k-space) and foreground removal (via blind source separation) for detecting the 21-cm signal during the Epoch of Reionization.
• To assess how noise levels (LOFAR-like, SKA-like, and no noise) affect the performance of both methods in recovering cosmological power spectra.
• To investigate the impact of realistic foreground variations—both along the line-of-sight (LOS) and spatially—on the robustness of each method.
• To determine whether foreground avoidance or removal provides better signal-to-noise recovery across different k-space scales, particularly in the context of future high-sensitivity experiments like the SKA.
• To examine the sensitivity of both methods to deviations from smooth foreground models, including 0.1% and 1% random variations in LOS and spatial dimensions.

Proposed method

• Simulates visibility data for a LOFAR-like interferometric array with realistic instrumental and sky systematics, including noise levels representative of current (LOFAR) and future (SKA) experiments.
• Applies a simplified, frequency-independent foreground removal model using the Group-Matched Component Analysis (GMCA) algorithm to separate foregrounds from the cosmological signal.
• Defines an EoR window in 2D cylindrical k-space (k_perp, k_los) based on instrument bandwidth, baseline distribution, and frequency resolution, excluding regions dominated by foregrounds.
• Computes the cylindrical power spectrum of the recovered signal within the EoR window and compares it to the full-sky power spectrum after foreground removal.
• Introduces controlled Gaussian random variations (0.1% and 1%) in both line-of-sight and spatial dimensions to test robustness against non-smooth foregrounds.
• Evaluates performance using signal-to-noise ratio (S/N) and power spectrum recovery across k-space, particularly focusing on k_perp > 0.6 Mpc⁻¹ and low k_los regions.

Experimental results

Research questions

• RQ1 How does foreground avoidance compare to foreground removal in terms of signal recovery at high k_perp (>0.6 Mpc⁻¹) for current-generation experiments?
• RQ2 To what extent does foreground removal recover more cosmological signal at low k_los compared to foreground avoidance, especially under low-noise conditions?
• RQ3 How do line-of-sight and spatial variations in foregrounds (up to 1%) affect the performance of both foreground avoidance and removal methods?
• RQ4 Does the EoR window remain a viable strategy when foregrounds exhibit non-smooth structures, such as 1% LOS variations?
• RQ5 How does the performance of GMCA-based foreground removal degrade under increasing noise or foreground complexity, and does it overfit at SKA-like noise levels?

Key findings

• For current-generation experiments (LOFAR-like noise), foreground avoidance recovers more signal at k_perp > 0.6 Mpc⁻¹ than foreground removal, due to reduced fitting errors in high-k_perp modes.
• Foreground removal recovers significantly more signal at low k_los for both current and future experiments, particularly in low-noise scenarios where fitting accuracy improves.
• At 0.1% LOS and spatial variations, both methods perform well, with no significant degradation in signal recovery, indicating robustness to small-scale foreground fluctuations.
• With 1% LOS-only variation, foreground avoidance becomes ineffective as the EoR window is obscured, while GMCA-based removal still recovers the cosmological signal reasonably well.
• For 1% spatial and LOS variations, GMCA's performance degrades but remains superior to foreground avoidance, which shows better performance under combined variations than under LOS-only variation, likely due to chance cancellation of LOS effects.
• In the no-noise scenario, GMCA recovers the same range of k_perp scales as foreground avoidance, suggesting that in ideal conditions, both methods can achieve comparable high-k_perp recovery.