Abstract

Topological defects arise from early phase transitions in cosmological scenarios predicted by theories for the unification of the fundamental interactions. The current status of observation leaves room for the existence of such defects. The issue of their detection thus represents today a central question in cosmology.

In this talk we present the first version of an efficient algorithm for the denoising of a non-Gaussian string signal in the standard Gaussian CMB signal seen as additive noise. A steerable wavelet decomposition allows one to probe the signal at various analysis scales and orientations in terms of wavelet coefficients. The wavelet coefficients of the string signal are estimated by a Bayesian least square denoising procedure relying on a string model trained on simulations, and on a preliminary assessment of the string tension. The string signal itself is reconstructed from its identified coefficients and the string network can then be mapped as the magnitude of gradient of the denoised signal. After denoising, the kurtosis of the magnitude of gradient can be used as a statistic to define a hypothesis test for the existence of strings.

We discuss the performance of the algorithm in the perspective of forthcoming arcminute CMB experiments. Further improvements are also discussed, both to provide fine constraints on the string tension through denoising, and to deal efficiently with secondary CMB anisotropies.