Truncation, amplitude, and jitter errors on \(\mathbb R\) for sampling series derivatives (Q630499): Difference between revisions

This work analyzes errors for derivatives of bandlimited functions, including truncation errors for sampling series expansions, amplitude errors arising when true samples values are replaced by approximate sample values, and jitter errors that arise when grid sample locations are replaced by perturbed grid samples. In each case, the authors reference a typical known result for each type of error for a given bandlimited signal \(f\), and then proceed in an analogous fashion to derive corresponding derivative estimates, either pointwise or in \(L^2\)-norm. The work is organized such that technical estimates needed, in some cases, for several different error estimates, are presented first. Then a number of truncation error estimates are given, followed by amplitude error estimates and jitter error estimates. The authors denote by \(B^2_\sigma\) the Paley-Wiener space of \(L^2\)-convergent sampling series of the form \[ f(t) =\sum_{n=-\infty}^\infty f\Bigl(\frac{n\pi}{\sigma}\Bigr) \, S_n(t) \] where \(S_n(t)=\text{sinc} (\sigma t - n\pi)\) and \(\text{sinc} (t)=\sin{t}\, / t\). The \(r\)th derivative of the \(N\)th truncation of \(f\) is \(f_N^{(r)}(t) =\sum_{n=-N}^N f(n\pi/\sigma) S_n^{(r)} (t)\) and the associated truncation is \(T_{r,N} (t) = f^{(r)} (t) -f_{N}^{(r)}(t)\). A typical error estimate (Theorem 3.1 here) states that if \(f\in B^2_\sigma\) and \(|t|<N\pi/\sigma\) then for even values of \(r\), \[ |T_{2r,N}(t)|\leq \frac{1}{\pi}{\sigma^{2r+1/2} (2r)! \|f\|_2 } E_r(t) \] where, for \(m(t)=\max \{1/(N\pi-\sigma t),\, 1/(N\pi+\sigma t)\}\), \[ E_r(t)\leq|\sin (\sigma t)|\sum_{j=0}^{r-1} m^{2j+1/2}(t)+|\cos (\sigma t)|\sum_{j=0}^{r-1} m^{2j+3/2}(t) \, , \] with a similar estimates for cases in which \(r\) is odd. This particular estimate generalizes the \(r=0\) case due to \textit{H. S. Piper jun.}, [IEEE Trans. Inf. Theory 21, 482--484 (1975; Zbl 0304.94007)].