Sequential Partition Detectors with Applications.
Abstract
The main theme of this dissertation is to extend the theory of nonparametric partition detectors to include sequential operation and dependent sampling. In particular, the problem of sequentially testing the hypothesis, H0, against a composite alternative, H1, when the functional form of the underlying noise disstribution is unknown, is considered. To that end, the observation space in R1 is partitioned into m intervals based on knowledge of only a finite number of quantiles of the noise distribution. The ease of implementation and robustness of the m-interval detectors make the sequential partition detectors (SPD) based on m-interval partitioning particularly attractive. In the first part, a formulation for the SPD is given and then test statistics are constructed, based on the Lehmann and shift of the mean alternatives. Next, three dependent sequential partition detector structures are introduced. First, the theory of SPD is extended to include q-dependent processes. Second, efficiency expressions are extended to include dependent sequential partition detectors (DSPD) and it is shown that rapid sampling can improve efficiency based on time to detection. Third, two new dependent sampling techniques, bivariate sequential partition detector (BSPD) and Markov sequential partition detector (MSPD), are introduced to overcome the need to know the thresholds. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 04, 1976
- Accession Number
- ADA027830
Entities
People
- Roger F. Dwyer
Organizations
- Naval Underwater Systems Center