Convolution, average sampling, and a Calderon resolution of the identity for shift-invariant spaces (Q818492)

The authors consider subspaces of \( L^2 (\mathbb R^d) \) of the form \[ V^2 (\Phi ) = \biggl\{\sum_{j \in \mathbb Z^d} D(j)^T \Phi (\cdot -j ) : D \in (l^2)^{(r)}\biggr\} \] for a vector function \( \Phi = (\phi_1, \ldots , \phi_r)^T \in (L^2)^{(r)}\) and \( D = (d_1 , \ldots , d_r)^T \) a vector sequence such that \( d_i := \{d_i (j) \}_{j \in \mathbb Z^d}\in l^2\). It is required, that the Gramian \( G_{\Phi} (\xi ) := \sum_{k \in \mathbb Z^d} \widehat \Phi (\xi + k) \overline{\widehat \phi(\xi + k)}^T \) is bounded, i.e., \( G_\phi (\xi) \leq MI \) a.e. \( \xi \in \mathbb R^d\). The considered convolution problem consists of convolving a function \( f \in V^2 (\Phi )\) with \( s \) functions \( \psi_1, \ldots , \psi_s \) resulting in a vector \((f \ast \psi_1, \ldots , f \ast \psi_s)^T \in (L^2)^{(s)}\), and it is required that any \(f \in V^2 (\Phi ) \) can be recovered from its convolution vector. The authors characterize convolution operators on \( V^2 (\Phi ) \) that have bounded inverses and those that have inverses being not necessarily bounded. Further, a restriction on the convolution operators and the generator is given such that the existence of an inverse already implies its boundedness. Moreover, convolution operators on \( V^2 (\Phi ) \) are characterized, such that critical uniform sampling after convolution is sufficient for stable reconstruction. Connections between critical uniform sampling, existence of a reconstruction and stability are discussed. Finally, the relation between irregular sampling and reconstruction after convolution as well as the convolution-deconvolution problem is considered. The theory is demonstrated by several examples.