The Epoch of Reionization Window: II. Statistical Methods for Foreground Wedge Reduction

For there to be a successful measurement of the 21 cm Epoch of Reionization (EoR) power spectrum, it is crucial that strong foreground contaminants be robustly suppressed. These foregrounds come from a variety of sources (such as Galactic synchrotron emission and extragalactic point sources), but almost all share the property of being spectrally smooth, and when viewed through the chromatic response of an interferometer, occupy a signature "wedge" region in cylindrical $k_\perp k_\parallel$ Fourier space. The complement of the foreground wedge is termed the "EoR window", and is expected to be mostly foreground-free, allowing clean measurements of the power spectrum. This paper is a sequel to a previous paper that established a rigorous mathematical framework for describing the foreground wedge and the EoR window. Here, we use our framework to explore statistical methods by which the EoR window can be enlarged, thereby increasing the sensitivity of a power spectrum measurement. We adapt the FKP approximation (commonly used in galaxy surveys) for 21 cm cosmology, and also compare the optimal quadratic estimator to simpler estimators that ignore covariances between different Fourier modes. The optimal quadratic estimator is found to suppress foregrounds by an extra factor of $\sim 10^5$ in power at the peripheries of the EoR window, boosting the detection of the cosmological signal from $12\sigma$ to $50\sigma$ at the midpoint of reionization in our fiducial models. If numerical issues can be finessed, decorrelation techniques allow the EoR window to be further enlarged, enabling measurements to be made deep within the foreground wedge. These techniques do not assume that foreground are Gaussian-distributed, and we additionally prove that a final round of foreground subtraction can be performed after decorrelation in a way that is guaranteed to have no cosmological signal loss.
Comment: 19 pages, 8 figures. v2: Updated to match published PRD version. Fixed typos, added figure showing fiducial theoretical model, appendix about Fisher matrices, and detection significance figures in response to referee comments. Main conclusions unchanged