Impact of calibration errors on CMB component separation using FastICA and ILC

TL;DR

This paper compares FastICA and ILC methods for CMB component separation under calibration errors, revealing that ILC can produce inaccurate results in high signal-to-noise scenarios due to partial cancellation of the CMB signal.

Contribution

It demonstrates the impact of calibration errors on the accuracy of ILC and emphasizes the need for careful calibration error estimation in component separation methods.

Findings

ILC can lead to inaccurate CMB maps with calibration errors

FastICA is more robust to calibration errors than ILC

Calibration errors can cause partial cancellation of the CMB signal in ILC

Abstract

The separation of emissions from different astrophysical processes is an important step towards the understanding of observational data. This topic of component separation is of particular importance in the observation of the relic Cosmic Microwave Background Radiation, as performed by the WMAP satellite and the more recent Planck mission, launched May 14th, 2009 from Kourou and currently taking data. When performing any sort of component separation, some assumptions about the components must be used. One assumption that many techniques typically use is knowledge of the frequency scaling of one or more components. This assumption may be broken in the presence of calibration errors. Here we compare, in the context of imperfect calibration, the recovery of a clean map of emission of the Cosmic Microwave Background from observational data with two methods: FastICA (which makes no assumption of the frequency scaling of the components), and an `Internal Linear Combination' (ILC), which explicitly extracts a component with a given frequency scaling. We find that even in the presence of small calibration errors with a Planck-style mission, the ILC method can lead to inaccurate CMB reconstruction in the high signal-to-noise regime, because of partial cancellation of the CMB emission in the recovered map. While there is no indication that the failure of the ILC will translate to other foreground cleaning or component separation techniques, we propose that all methods which assume knowledge of the frequency scaling of one or more components be careful to estimate the effects of calibration errors.