Diffuse source separation in CMB observations

Spectacular advances in the understanding of the Big-Bang model of cosmology have been due to increasingly accurate observations of the properties of the Cosmic Microwave Background (CMB). The detector sensitivities of modern experiments have permitted to measure fluctuations of the CMB temperature with such a sensitivity that the contamination of the data by astrophysical foreground radiations, rather than by instrumental noise, is becoming the major source of limitation. This will be the case, in particular, for the upcoming observations by the Planck mission, to be launched by ESA in 2008 [Lamarre et al. (2003), Mandolesi et al. (2000), Lamarre et al. (2000)], as well as for next generation instruments dedicated to the observation of CMB polarisation. In this context, the development of data analysis methods dedicated to identifying and separating foreground contamination from CMB observations is of the utmost importance for future CMB experiments. In many astrophysical observations indeed, and in particular in the context of CMB experiments, signals and images contain contributions from several components or sources . Some of these sources may be of particular interest (CMB or other astrophysical emission), some may be unwanted (noise). Obviously, components cannot be properly studied in data sets in which they appear only as a mixture. Component separation consists, for each of them, in isolating the emission from all the other components present in the data, in the best possible way. It should be noted that what “best” means depends on what the isolated data will be used for. Very often, one tries to obtain, for each component, an estimated map (or a set of maps at different frequencies) minimising the total error variance, i.e. minimising