Robust, Distributed, and Adaptive Quickest Detection Procedures.
Abstract
This dissertation focuses on sequential techniques for detecting a change, or disorder, in the statistics of a random process. First, the minimax robust quickest detector is derived for the case when the underlying noise models are only partially known. It is shown that when the robust processor is used, the minimax asymptotic performance measure is equal to the Kullback-Leibler divergence, and that the least favorable densities are those that minimize this quantity. The robust quickest detector is also determined for the weak signal case, and we show an equivalence between the performance measure, the classical efficacy, and Fisher's information. Performance curves are given to show the gain available when robustness is built into the procedure. The robust quickest detector is also derived under mean and covariance uncertainty for a multivariate Gaussian noise process. It is shown that the robust processor is exactly the robust discrete-time matched filter, which has been studied previously. Expressions for the asymptotic performance are derived, and particular solutions are presented for several uncertainty classes. Performance curves are provided to illustrate the tradeoffs when there is a mismatch between the assumed and actual levels of uncertainty. The applicability of the robust procedure to non-Gaussian noise is also discussed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1995
- Accession Number
- ADA309507
Entities
People
- R. W. Crow
- S. C. Schwartz
Organizations
- Princeton University