The paper considers the robust nonparametric estimation problems for an additive regression model in discrete time. Based on the nonparametric filtering methods developed in Pchelintsev et al. (2022a), new estimation procedure is proposed, for which it is shown that the convergence rate of robust risks, up to a logarithmically increasing coefficient, coincides with the parametric one, i.e. the robust superefficient property is provided. The Pinsker constant for the Sobolev ellipse with exponentially increasing coefficients is calculated. The developed methods are applied to the robust signal processing observed in multi-path connection channels under non-Gaussian noises. Numerical studies were carried out using Monte - Carlo simulations, which confirmed the obtained theoretical results for the constructed estimation procedures and their significant advantage over maximum likelihood estimates.
