Foreground analysis of the Wilkinson Microwave Anisotropy Probe 3-yr data with FASTICA

We present an analysis of the foreground emission present in the Wilkinson Microwave Anisotropy Probe (WMAP) 3-yr data as determined by the method of Independent Component Analysis. The WMAP data averaged at each frequency are used together with the standard foreground emission templates as inputs to the FASTICA algorithm. The returned coefficients can be interpreted as coupling coefficients between the WMAP data and foreground templates. These results are then used to infer the spectral behaviour for three foreground components - synchrotron, anomalous dust-correlated emission and free-free. For the first two components, we find values consistent with previous results although slightly steeper. We confirm the inconsistency in the scaling between the Hα template and free-free emission at K and Ka bands where an electron temperature of ~4000 K is indicated. We also see evidence of significantly flatter spectral behaviour to higher frequencies than expected theoretically and previously noted by Dobler et al., but only when analysing the Kp2 sky coverage. We further apply FASTICA `iteratively', using data pre-cleaned using foreground templates scaled to the WMAP frequencies by coupling coefficients determined by a prior FASTICA analysis. This multifrequency analysis allows us to determine the presence of residual foreground emission not traced by the templates. We confirm the existence of a component spatially distributed along the Galactic plane and particularly enhanced near the centre (the `WMAP haze'). This emission is less extended when using the WMAP K-Ka data as the synchrotron template confirming that it can be considered a better template for foreground cleaning of the WMAP data. However, its use complicates the physical interpretation of the nature of the foreground emission and residuals since it contains a mixture of several, physically distinct emission mechanisms. The good agreement between the extracted cosmic microwave background (CMB) component and previous results, as well as the low amplitude of residual foreground emission make FASTICA a viable tool to infer foreground emission properties, via template fitting, and the CMB amplitude.