Blind deconvolution of multivariate signals : a deflation approach

摘要

1 IntroductionIn s number of applications, a certain scalar stationary signal y = (y(n))n∈Z(the observation) is the output of an unknown filter of transfer function h(z) driven by an independent identically distributed (i.i.d.) sequence w= (w(n))n∈Z representing the useful signal. It is often desirable to be able to adaptively reconstruct the signal w from the observation y, or equivalently, to estimate the transfer function h-1(z). This is particularly easy when h(z) is known to be minimum-phase (or maximumphase), since w coincides with the forward (or backward)innovation of y. However, this restrictive assumption does not hold in certain important applications, such as channel equalization in Telecommunications. In that case, the above question is referred to as the blind