Generalized sampling in shift-invariant spaces with multiple stable generators (Q2382686)

The Whittaker-Shannon-Kotel'nikov Sampling Theorem states that if a function \(f\) is band-limited to \( [-\sigma ,\sigma ]\), i.e., it is representable as \[ f(t)= \int_{-\sigma}^{\sigma} e^{-ixt}g(x)\,dx \qquad (t\in\mathbb R) ,\tag{1.1} \] for some function \(g\in L^2(-\sigma,\sigma)\), then \(f\) can be reconstructed from its samples, \(f(k\pi / \sigma).\) The construction formula is \[ f(t)=\sum_{k=-\infty}^{\infty} f\left( t_k \right) \frac{ \sin{\sigma( t-t_k)}}{\sigma( t-t_k)} \qquad (t\in\mathbb R), \tag{1.2} \] where \(t_k= {k\pi}/{\sigma}.\) If we set \[ \phi (t)=\frac{\sin \sigma t}{\sigma t},\quad t\neq 0, \text{ and } \phi(0)=1, \] the above relation can be written in the form \[ f(t)=\sum_{k=-\infty}^{\infty}f(t_k) \phi (t-t_k), \] which shows that the space of bandlimited functions is shift-invariant. Shift-invariant spaces, which have become the focus of many research papers in recent years because of their connection with wavelets and multiresolution analyses, are spaces of the form \[ V_\phi=\overline{\text{span}}\left\{ \phi(t-n),\; n\in \mathbb Z \right\}, \phi \in L^2(\mathbb R). \] More generally, shift-invariant spaces with \(\ell\) generators, which arose in connection with multiwavelets, are given by \[ V_{\Phi}=\overline{\text{span}}\left\{ \phi_1 (t-n),\dots, \phi_\ell (t-n),\; n\in \mathbb Z , \phi_k \in L^2(\mathbb R), k=1,2, \dots, \ell \right\} , \] where \(\Phi =\left\{ \phi_1, \dots, \phi_\ell \right\}.\) The authors derive a stable generalized sampling formula in shift-invariant spaces with \(\ell\) generators, where the samples are not point-values of the function, but rather some filtered values thereof, namely values of the form \(\left\{ \left({\mathcal L}_j f\right)(rn)\right\},\) where \(n\in \mathbb Z, j=1, 2, \dots, s ,\) \(r\) is the sampling period, and \({\mathcal L}_j\) is a linear, time-invariant filter. Two types of filters are considered: \[ \left({\mathcal L}f\right)(t)=\int_{-\infty}^\infty f(x)h(t-x) \, dx , \] and \[ \left({\mathcal L}f\right)(t)=\sum_{m=0}^N c_m \delta^{(m)}(t+d_m), \] where \(c_m\) and \(d_m\) are constants. First, it is shown that to have a viable sampling formula, we must have \(r\ell\leq s .\) The main result of the paper is to show that for any \(f\in V_\Phi ,\) we have \[ f(t)=\sum_{n\in \mathbb Z} \sum_{j=1}^s \left({\mathcal L}_jf\right)(rn)S_{j,n}(t),\quad \text{ in } L^2(\mathbb R), \] where \(\left\{S_{j,n}(t)=S_j(t-rn)\right\}\) with \(n\in \mathbb Z, j=1,2, \dots ,s\) is a frame in \(V_\Phi ,\) while if \(s=r\ell ,\) \(\left\{S_{j,n}(t)=S_j(t-rn)\right\}\) is a Riesz basis. The last section discusses oversampling, i.e., when \(r\ell < s\), and the generators and the impulse response of the filters \({\mathcal L}_j\) have compact support. An example with two generators using the Hermite cubic splines is given. Overall the paper is well written and very interesting.