Weighted sampling and reconstruction in weighted reproducing kernel spaces (Q961075): Difference between revisions

A weighted reproducing kernel subspace \(V_v\) of \(L^p_v\) is a subspace such that \(|f(x)|\leq C_x\|f\|_{p,v}\) where \(\|f\|_{p,v}=\|fv\|_p\) and \(\|\cdot\|_p\) denotes the \(L^p\)-norm on \(\mathbb{R}^d\). Such a space can be identified with an image of a continuous idempotent mapping \(T\) from \(L^p_v\) to itself (\(1\leq p\leq \infty\)) expressed in terms of an integral kernel \(K(x,y)\) such that \(Tf(x)=\int K(x,y)\, f(y)\, dy\). Here the weight \(v\) is assumed to be \(w\)-moderate, i.e., \(v(x+y)\leq v(x) w(y)\) (\(x,y\in\mathbb{R}^d\)) for some submultiplicative weight \(w\), i.e., \(w(x+y)\leq w(x)w(y)\). Let \(\omega_\delta (K)(x,y)=\sup_{|s|,|t|<\delta}|K(x+s,y+t)-K(x,y)|\) denote the bivariate modulus of continuity. Let \(1\leq p\leq \infty\), and \(K\) be a reproducting kernel for \(L^p_v\) such that \[ \sup_x \|K(x,x-\cdot)\|_{1,w}^{1-1/p} \,\sup_y \|K(\cdot+y,y)\|_{1,w}^{1/p} <\infty \] and such that the corresponding product with \(K\) replaced by \(\omega_\delta(K)\) is at most \(1/M\) for \(\delta=(\delta_0+a)/2\), where \(0<\delta_0,a<\infty\) are given. Let \(\Gamma\) be finite pairwise disjoint union of \(\delta_0\)-separated subsets of \(\mathbb{R}^d\) (i.e., \(\|s-t\|\geq\delta_0\) if \(s\neq t\) are in one of the subsets) and let \(\Psi=\{\psi_\gamma\}_{\gamma\in\Gamma}\) be a family of weighted sampling functionals with the property that \(\|\psi_\gamma\|_{L^1}\leq M\) for all \(\gamma\in\Gamma\), \(\int \psi_\gamma=1\) for all \(\gamma\), and \(\psi_\gamma\) is supported in \(\gamma+[-a/2,a/2]^d\) for each \(\gamma\in\Gamma\). The first main result states that, under these hypotheses, any signal in \(V_v\) can be reconstructed from its {\textit{average samples}} \(\langle g,\, \psi_\gamma\rangle\), \(\gamma\in\Gamma\) with the following inequalities applying: \[ \begin{multlined} \Bigl(1-{1\over M}\Bigr) \Bigl( {A_\Gamma(\delta_0)\over \delta_0^d}\Bigr)^{1/p} \sup_{|z|<\delta/2 } w(z)^{-1} \|g\|_{p,v}\leq \|\{\langle g,\psi_\gamma\}_{\gamma\in\Gamma}\|_{\ell^p_v}\leq \Bigl(M+{1\over M}\Bigr)\\ \Bigl( {B_\Gamma(\delta_0)\over \delta_0^d}\Bigr)^{1/p} \sup_{|z|<\delta/2} w(z)^{-1} \|g\|_{p,v}\end{multlined} \] Here, \(A_{\Gamma}(\delta)\) and \(B_{\Gamma}(\delta)\) are the infima and suprema of the quantities \(\sum_{\gamma\in\Gamma} \chi_{\gamma+[-\delta/2,\delta/2]^d}(x)\). An iterative algorithm for reconstructing any \(f\in V_v\) from its coefficients \(\langle g,\, \psi_\gamma\rangle\) is provided and reconstruction error is shown to decay exponentially in the number of iterations.