A large fraction of spectra delivered by Gaia will come from faint stars and will therefore be particularly noisy, degrading the accuracy of RV measurement as well as other important parameters. In such cases several numerical techniques can be employed to help reduce systematic offsets and improve the overall accuracy. They include the principal component analysis, support vector machines, artificial neural networks, and wavelet transforms. As part of our contribution to the RAVE survey (www.rave-survey.org) we already used the first three methods during building of the automated pipeline to extract the RAVE survey data. Classical works to start from when using the wavelet transforms are those of Fligge and Solanki (1997, A&AS, 124, 579 and 1998, ASP Conf. Ser., 154, 833). Here we propose to study noise removal techniques in line with the needs and decisions of CU6. There are several ways on how to derive the radial and rotational velocity from an observed spectrum which is already extracted, wavelength calibrated and made as noise-free as possible (e.g. by using the techniques mentioned above). Most possibilities are already taken care of in the scheme of tasks established by the CU6. There is however one RV & V_rot derivation scheme decided to be studied by the CU6 but which has no group allocated yet. So we propose here to test and pursue this remaining idea. We will use a binary mask scheme, starting from its very successful uses for the Coravel (Baranne, Mayor & Poncet 1979, Vistas Astron., 23, 279) and the Elodie (Baranne et al., 1996, A&AS, 119, 373) instruments. Clearly the software implementation of the mask, as described in the latter paper, will be preferred. As the first upgrade we shall increase the number of template masks from one or two to the whole set built from templates covering the whole parameter space.