Sky-averaged 21-cm signal extraction using multiple antennas with an SVD framework: the REACH case

TL;DR

This paper introduces a novel SVD-based framework for extracting the sky-averaged 21-cm signal by fitting multiple antennas and time slices simultaneously, significantly reducing bias and uncertainty in the presence of complex foregrounds.

Contribution

It develops a multi-antenna, multi-time fitting strategy using SVD to better separate the 21-cm signal from foregrounds, improving extraction accuracy over single-antenna methods.

Findings

Simultaneous multi-antenna fitting reduces bias and uncertainty by 2-3 times.

Using multiple time slices improves signal extraction accuracy.

The method performs well with both simple and realistic foreground models.

Abstract

In a sky-averaged 21-cm signal experiment, the uncertainty on the extracted signal depends mainly on the covariance between the foreground and 21-cm signal models. In this paper, we construct these models using the modes of variation obtained from the Singular Value Decomposition of a set of simulated foreground and 21-cm signals. We present a strategy to reduce this overlap between the 21-cm and foreground modes by simultaneously fitting the spectra from multiple different antennas, which can be used in combination with the method of utilizing the time dependence of foregrounds while fitting multiple drift scan spectra. To demonstrate this idea, we consider two different foreground models (i) a simple foreground model, where we assume a constant spectral index over the sky, and (ii) a more realistic foreground model, with a spatial variation of the spectral index. For the simple foreground model, with just a single antenna design, we are able to extract the signal with good accuracy if we simultaneously fit the data from multiple time slices. The 21-cm signal extraction is further improved when we simultaneously fit the data from different antennas as well. This improvement becomes more pronounced while using the more realistic mock observations generated from the detailed foreground model. We find that even if we fit multiple time slices, the recovered signal is biased and inaccurate for a single antenna. However, simultaneously fitting the data from different antennas reduces the bias and the uncertainty by a factor of 2-3 on the extracted 21-cm signal.