Nonlocal Methods for Signal Detection and Estimation in the Dependent Nonstationary Environment
Abstract
We have obtained a number of results pertaining to signal detection and estimation, where the underlying random processes are imperfectly known and often possess dependency and/or nonstationarity. Our results heavily emphasize nonlocal methods, that is, methods which allow an imperfectly known distribution to vary substantially and not simply be modeled as local to a nominal. Much of this work features robustness, but we also include research involving nonparametric algorithms. Our results include the design and analysis of the classically robust saddlepoint detector for nominally Laplace noise, development of quantitative nonlocal robustness measures for signal detection, parameter estimation, and the estimation of a random variable (all with dependent data), development of a 'user friendly' concept of average nonlocal robustness (a vast improvement over 'worst case' or 'least favorable' approaches), and an analysis of the stability of the false alarm rate of a classical 'nonparametric' detector (an analysis which uses nonlocal techniques). This work underscores that traditional algorithms, while useful, are limited by their design assumptions and can offer disappointing performance when presented with realistic data which reflects imperfectly known random processes possessing dependency and/or nonstationarity. Our quantitative results not only shed light on how bad the situation can be, but how to compensate for it with improved design procedures.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 30, 1993
- Accession Number
- ADA278472
Entities
People
- Don Halverson
Organizations
- Texas A&M University