Cosmic Dawn and Epoch of Reionization Foreground Removal with the SKA

TL;DR

This paper evaluates foreground removal techniques for SKA's observations of the Cosmic Dawn and Epoch of Reionization, demonstrating that current methods can effectively recover the cosmological signal despite instrumental and foreground complexities.

Contribution

It provides a comprehensive comparison of parametric and non-parametric foreground removal methods using realistic SKA simulations, highlighting their effectiveness and challenges.

Findings

Foregrounds can be removed with high accuracy using current methods.

Instrumental effects like frequency-dependent PSF complicate signal recovery.

Splitting observations into smaller bandwidth segments improves results.

Abstract

The exceptional sensitivity of the SKA will allow observations of the Cosmic Dawn and Epoch of Reionization (CD/EoR) in unprecedented detail, both spectrally and spatially. This wealth of information is buried under Galactic and extragalactic foregrounds, which must be removed accurately and precisely in order to reveal the cosmological signal. This problem has been addressed already for the previous generation of radio telescopes, but the application to SKA is different in many aspects. In this chapter we summarise the contributions to the field of foreground removal in the context of high redshift and high sensitivity 21-cm measurements. We use a state-of-the-art simulation of the SKA Phase 1 observations complete with cosmological signal, foregrounds and frequency-dependent instrumental effects to test both parametric and non-parametric foreground removal methods. We compare the recovered cosmological signal using several different statistics and explore one of the most exciting possibilities with the SKA --- imaging of the ionized bubbles. We find that with current methods it is possible to remove the foregrounds with great accuracy and to get impressive power spectra and images of the cosmological signal. The frequency-dependent PSF of the instrument complicates this recovery, so we resort to splitting the observation bandwidth into smaller segments, each of a common resolution. If the foregrounds are allowed a random variation from the smooth power law along the line of sight, methods exploiting the smoothness of foregrounds or a parametrization of their behaviour are challenged much more than non-parametric ones. However, we show that correction techniques can be implemented to restore the performances of parametric approaches, as long as the first-order approximation of a power law stands.