Improved Foreground Modelling for Bayesian 21 cm Power Spectrum Estimation with BayesEoR

Estimating the power spectrum (PS) of the redshifted 21 cm signal from the Epoch of Reionization (EoR) has proven difficult due to the presence of bright astrophysical foregrounds (FGs). Existing techniques to mitigate FGs during PS estimation under-utilize the data and do not properly account for the covariance between the observed EoR and FG signals. We are developing a Bayesian framework, BayesEoR, that jointly models the instrument, FGs, and 21 cm signal and marginalizes over their uncertainties to enable extraction of statistically robust and unbiased estimates of the EoR PS. In this paper, we present a brief overview of our approach to Bayesian PS estimation and the results of a recent study in which we used BayesEoR to analyze a set of detailed simulations containing mock EoR and realistic FG signals. Due to computational constraints, the forward model of the instrument in BayesEoR, which involves application of the instrument primary beam in the image domain, is restricted to model a subset of the sky. We find that, when the sky emission outside of the modelled domain is downweighted by the beam at the level of the dynamic range between the EoR and FGs, BayesEoR can model visibilities which see the whole sky using only a subset of the sky. We also present several techniques that can be used to mitigate FG contamination during EoR PS estimation while ameliorating computational costs.