Optimal kernels for a general sampling theorem (Q1066385): Difference between revisions

In this paper there is a first systematic approach using linear combinations of central B-splines of different degree in order to establish general optimal sampling theorems delivering arbitrarily high orders of approximation and involving only a minimal number of samples, and that for practical engineering purposes. The main advantage is that the kernel of the sampling series is now piecewise algebraic with finite support. The main ingredients are Jackson-de La Vallée Poussin kernels (i.e., powers of the sinc-function in the language of signal processing): appropriate linear combinations of these kernels deliver rapid approximations to the rectangular (window) function. Their Fourier transforms, in turn, are central B-splines - with all of their properties. This construction is possible by a comprehensive application of central factorial numbers enabling a simple Taylor series expansion of any power of sin x. (The general theory will be treated in a forthcoming paper.) The remaining part uses Vandermonde determinants. To state a simple result in formula form: with \(M_ n(x)\) denoting the spline of degree n the almost trivial combination \(\Phi_{4,5}(x):=5M_ 4(x)-4M_ 5(x),\) \(\sup p=[-5/2,5/2]\) represents an optimal kernel in the sampling theorem with only 5 samples being needed and order of approximation \(O(\omega^{-4})\), \(\omega\) \(\to \infty\), 1/\(\omega\) being the distance between the sampling points, i.e., \(\| \sum^{2}_{k=- 2}f(k/\omega)\Phi_{4,5}(\omega \cdot -k)-f(\cdot)\|_ C\leq M/\omega^ 4\) provided the signal \(f\in C^{(4)}({\mathbb{R}})\).