Test Input Evaluation for Optimal Adaptive Filtering

摘要

Average statistical divergence is proposed as a figure of merit for ranking test inputs used in identifying the unknown parameters of a system. Average divergence is a concept taken from communication theory; it can be computed a priori from a recursion relation derived in this paper. Two closed-form analytic examples are presented. The development in the paper is for linear multistage processes and is applicable to on-line nonstationary adaptive filtering problems. The average divergence of a general multistage process that has unknown parameters can be calculated recursively. (Author)