A matrix pencil approach to the existence of compactly supported reconstruction functions in average sampling (Q636245): Difference between revisions

The existence of a sampling formula involving samples of some convolution operator is studied in the case of reconstruction functions with compact support. A shift-invariant space \(V_{\varphi}\) in \(L^2({\mathbb R})\) with stable generator \(\varphi\) is considered, i.e. \[ V_{\varphi}=\{f(t)=\sum_{n\in{\mathbb Z}}a_n\varphi(t-n) : l^2({\mathbb Z})\}, \] where the sequence \(\{\varphi(\cdot-n)\}_{n\in\mathbb Z}\) is a Riesz basis for \(V_{\varphi}\). Linear time invariant (LTI) systems \({\mathcal L}\) are considered, which are defined on \(V_{\varphi}\) and have one of the following forms: \[ \begin{aligned} ({\mathcal L}f)(t)&=(f\star h)(t)=\int_{-\infty}^{\infty}f(x)h(t-x)dx,\quad t\in{\mathbb R},\\ ({\mathcal L}f)(t)&=f(t+d),\quad t\in{\mathbb R},\;\text{for some constant}\;d\in{\mathbb R}.\end{aligned} \] The oversampling technique is used with the sampling period \(T=r/s<1\) for fixed positive integers \(r\) and \(s\). The studied problem is equivalent to the recovery of \(f\in V_{\varphi}\) from the sequence of samples \(\{{\mathcal L}_jf(rn)\}_{n\in{\mathbb Z}},\;j=1,2,\dots,s\), where the LTI systems \({\mathcal L}_j\) are defined by \(({\mathcal L}_jf)=({\mathcal L}f)[t+(j-1)r/s],\;t\in{\mathbb R}\). The generalized sampling theory gives some conditions for the existence of a frame \(\{S_j(\cdot-n)\}_{n\in\mathbb Z},\;j=1,2,\dots,s\), such that for any \(f\in V_{\varphi}\) the following recovery formula holds: \[ f(t)=\sum_{n\in{\mathbb Z}}\sum_{j=1}^{s}({\mathcal L}_jf)(rn)S_j(t-rn)\;\text{in}\;L^2({\mathbb R}). \] It is noticed that compactly supported reconstruction functions \(S_f\) involve low computational complexity and it avoids truncation errors. Assuming that the generator \(\varphi\) and \({\mathcal L}{\varphi}\) are compactly supported, the problem of finding reconstruction functions \(S_j\) with compact support reduces to finding a polynomial left inverse of a polynomial matrix \(G(z)\) associated to the sampling problem. Using the structure of this matrix, the problem is transformed into one which is solved by the mean of the Kronecker canonical form of a matrix pencil associated to the matrix \(G(z)\). A necessary and sufficient condition is derived for the existence of compactly supported reconstruction functions. The matrix pencil approach is used to provide a method for computing a polynomial left inverse of \(G(z)\) in the case of minimum oversampling rate (i.e. \(s=r+1\)). It is shown that this inverse leads to reconstruction functions with minimal support. An example involving as generator \(\varphi\) the quadratic B-spline \(N_3(t)\) illustrates the proposed approach.