Data compression-A covariance analysis approach

The performance of linear discrete-time filters operating on preprocessed data is addressed in this paper. The preprocessing, or data compression, consists of reducing a sequence of N measurements of a system output vector, zk, to a single vector, ZLR, of manageable dimension. (The subscript LR denotes "low rate" data.) Results reported include techniques which can be applied to two important problems. First, the techniques can be used in a procedure for developing a complete data compression/filtering system. Applications include on-line filtering as well as approximate, economical, off-line data-processing. In the second application, the techniques can be used to develop a low-rate model which approximates an actual high-rate filter. This model is useful for performance analysis applications in which repeated exact covariance simulations of the high-rate processing would be infeasible due to computational expense.