26
112

Deconvolution with unknown error distribution

Abstract

We consider the problem of estimating a density fXf_X using a sample Y1,...,YnY_1,...,Y_n from fY=fXfϵf_Y=f_X\star f_{\epsilon}, where fϵf_{\epsilon} is an unknown density. We assume that an additional sample ϵ1,...,ϵm\epsilon_1,...,\epsilon_m from fϵf_{\epsilon} is observed. Estimators of fXf_X and its derivatives are constructed by using nonparametric estimators of fYf_Y and fϵf_{\epsilon} and by applying a spectral cut-off in the Fourier domain. We derive the rate of convergence of the estimators in case of a known and unknown error density fϵf_{\epsilon}, where it is assumed that fXf_X satisfies a polynomial, logarithmic or general source condition. It is shown that the proposed estimators are asymptotically optimal in a minimax sense in the models with known or unknown error density, if the density fXf_X belongs to a Sobolev space H\mathbhpH_{\mathbh p} and fϵf_{\epsilon} is ordinary smooth or supersmooth.

View on arXiv
Comments on this paper