Improvement of the performance of a classical matched filter by an independent component analysis preprocessing

Current gravitational wave searches for compact binaries coalescence are done using a bank of templates (matched filters) on each running detector. Given a network of interferometers, we propose to use a denoising strategy based on an independent component analysis which considers two interferometers at a time and then to use a standard matched filter on the processed data. We show that this method allows to lower the level of noise and increases the signal-to-nose ratio at the output of the matched filter.