Separating cosmological B modes from foregrounds in cosmic microwave background polarization observations

The detection and characterization of the B mode of Cosmic Microwave Background (CMB) polarization anisotropies will not be possible without a high precision removal of the foreground contamination present in the microwave band. In this work we study the relevance of the component separation technique based on the Independent Component Analysis (ICA) for this purpose and investigate its performance in the context of a limited sky coverage observation and from the viewpoint of a precise recovery of the B-mode power spectrum. We focus on the low Galactic emission sky patch centered at 40 degrees in right ascension and -45 in declination, corresponding to the target of several operating and planned CMB experiments and which, in many respects, adequately represents a typical “clean” high latitude sky. We consider two combinations of low and high frequencies: 40, 90 GHz and 150, 350 GHz, where the main diffuse polarized Galactic signal is dominated by the synchrotron and thermal dust emissions, respectively. The foreground templates have been simulated in accordance with the existing observations in the radio and infrared bands, as well as the Wilkinson Microwave Anisotropy Probe (WMAP) and Archeops data. With the help of a parallel implementation of the FastICA code we explore a substantial parameter space varying Gaussian pixel noise level, observed sky area and the amplitude of the foreground emission, and conduct large Monte Carlo simulations to evaluate the errors and biases induced in the reconstruction for different choices of these variables. We identify a large subspace of this parameter space for which the quality of the CMB reconstruction is excellent. That includes the cases when the B mode CMB signal is up to a few times weaker than the foreground contamination and the noise amplitude comparable with the total CMB polarized emission. In these cases the error induced by the separation process is found to be comparable to or lower than the one from the cosmic variance and instrumental noise, and the biases much smaller than that. This performance is a result of the high spatial resolution and the high level of statistical independence between background and foreground emissions, as anticipated from the foreground models and the available observations. We also point out and discuss limiting cases of noise and foreground amplitude, when the ICA approach fails. Although our analysis is limited by the absence of systematics in the simulated data, these results indicate that these component separation techniques could play a crucial role in the forthcoming experiments aiming at the detection of B modes in the CMB polarization.