Simulations for single-dish intensity mapping experiments

TL;DR

This paper evaluates simulation-based methods for removing foreground contamination in single-dish HI intensity mapping experiments, demonstrating that principal component analysis effectively recovers the HI signal with manageable leakage.

Contribution

It introduces a simulation framework including noise and foreground models and compares two foreground cleaning methods, highlighting PCA's robustness and limitations.

Findings

PCA can remove foreground contamination down to thermal noise levels.

Parametric fitting residuals remain contaminated by foregrounds and 1/f noise.

Foreground cleaning effectiveness depends on spectral smoothness of foregrounds and noise.

Abstract

HI intensity mapping is an emerging tool to probe dark energy. Observations of the redshifted HI signal will be contaminated by instrumental noise, atmospheric and Galactic foregrounds. The latter is expected to be four orders of magnitude brighter than the HI emission we wish to detect. We present a simulation of single-dish observations including an instrumental noise model with 1/f and white noise, and sky emission with a diffuse Galactic foreground and HI emission. We consider two foreground cleaning methods: spectral parametric fitting and principal component analysis. For a smooth frequency spectrum of the foreground and instrumental effects, we find that the parametric fitting method provides residuals that are still contaminated by foreground and 1/f noise, but the principal component analysis can remove this contamination down to the thermal noise level. This method is robust for a range of different models of foreground and noise, and so constitutes a promising way to recover the HI signal from the data. However, it induces a leakage of the cosmological signal into the subtracted foreground of around 5%. The efficiency of the component separation methods depends heavily on the smoothness of the frequency spectrum of the foreground and the 1/f noise. We find that as, long as the spectral variations over the band are slow compared to the channel width, the foreground cleaning method still works.