Oversampling and reconstruction functions with compact support (Q1014540)

\textit{M. Unser} and \textit{A. Adroubi} [Numer. Funct. Anal. Optimization 15, No. 1--2, 1--21 (1994; Zbl 0794.41024) and IEEE Trans. Circuits Syst., II, Analog Digit. Signal Process. 45, No. 8, 959--969 (1998; Zbl 0998.94519)] have shown that if \(\{\varphi(t-n); n \in \mathbb{Z}\}\) is a Riesz sequence in \(L^2(\mathbb{R})\) and \(V_{\varphi}\) denotes the closure of its linear span then, under suitable conditions, any function \(f \in V_{\varphi}\) can be recovered from a sequence of samples of a suitable operator \({\mathcal L}f\). In fact, they proved that \[ f(t) = \sum_{n \in \mathbb{Z}} {\mathcal L}f(n) S(t-n), \] where \(\{S(t-n); n \in \mathbb{Z}\}\) is a Riesz basis of \(V_{\varphi}\). Even when the generator \(\varphi\) has compact support, rarely the same happens with the reconstruction function \(S\). This paper overcomes this difficulty by using oversampling techniques to show the existence of sampling formulas having reconstruction functions with compact support. Under suitable hypotheses, a necessary and sufficient condition is given in terms of the Smith canonical form of a polynomial matrix (the so--called modulation matrix). The obtained sampling formulas provide approximation schemes for functions in the Sobolev spaces \(W^m_{\infty}(\mathbb{R})\) in the uniform norm. The results in the paper are illustrated with examples in shift--invariant spaces generated by quadratic \(B\)--splines.