On numerical realizations of Shannon's sampling theorem (Q6568384)

The paper studies the numerical properties of various summability methods for the Shannon sampling series\N\[\N\sum_{k\in\mathbb Z}f\left(\frac kL\right)\operatorname{sinc}(L\pi t - \pi k),\N\]\Nwhere \(L\ge N > 0\) and the function \(f\) belongs to the Paley-Wiener space\N\[\N\mathcal B_{N/2}(\mathbb R) := \left\{ f \in L^2(R) : \operatorname{supp} \widehat f \subseteq [-\frac N2, \frac N2]\right\}.\N\]\NThe authors improve previously known estimates for the norm of the partial sum operator\N\[\NS_Tf = \sum_{k=-T}^T f\left(\frac kL\right)\operatorname{sinc}(L\pi t - \pi k).\N\]\NThe estimates are used to show quantifiably poor convergence of the partial sums and their deterministic non-robustness with respect to the additive noise in the samples (even though the sums are stochastically robust).\N\NThe authors also compare the numerical properties of the regularized series that use different types of mollifying windows in the time or frequency domains. Among the methods considered, the best numerical results are obtained by oversampling and regularization in the time domain using a sinh-type window function or a continuous Kaiser-Bessel window function.