Search of weights in problem of weighted finite-rank time-series approximation

Abstract

The problem of weighted finite-rank time-series approximation is considered for signal estimation. Finding of weights, which lead to improvement of estimate accuracy, is examined. An effective method for numerical search of weights is constructed and proved. The theory of quadratic optimization is used. For effective algorithm, the problem of quadratic optimization with large number of linear constraints reduced to a sequence of problems with smaller number constraints, which is completed by a stopping criterion. For the algorithm deduction, equivalence of different statements of the original optimization problem is proved. A numerical simulation is performed to approve the algorithm efficiency and improvement of the estimate accuracy. Refs 10. Figs 2. Table 1.