Approximation and compression of scattered data by meshless multiscale decompositions (Q942151)

Starting from sensor network applications, the authors introduce a class of multiscale decompositions for scattered data. In order to realize a significant reduction in network power expenditure, it is necessary to apply first a distributed multiscale transform to node measurements and then, to use threshold-based compression techniques on the transform coefficients. The goal of the multiscale transform is to concentrate the relevant information into a small set of numerically significant coefficients, similar to wavelet transforms. The aim of the paper is to provide a mathematical analysis for a meshless multiscale protocol for sensor networks that inherits the attractive compression properties. Consequently, the authors devise a thinning procedure that organizes the points in a multiscale hierarchy. The decomposition is also based on a local least-squares polynomial fitting prediction operator. Finally, the authors prove that the multiscale coefficients satisfy the same decay properties as the classical wavelet coefficients when the analyzed function has some local smoothness properties, so the compression capability yields.