In the second part of this paper we focus on our main problem: the construction of an optimal
estimation procedure to reconstruct the unknown signal at a given fixed point. This or similar problems
received attention of many authors. Among others let us emphasize the following works: Lepski (1990),
Lepski and Spokoiny (1997), Goldenshluger and Nemirovski (1997), Tsybakov (1998), Klemela and
Tsybakov (2001), Chichignoud (2012). Recently, Goldenshluger and Lepski (2008) proposed for this
problem a universal pointwise selection rule. Considering various collections of kernels, the associated
estimation procedures can adapt to the smoothness or the structure of the underlying function. In
particular, it can be proved that there exists an adaptive estimation procedure that achieves the minimax