Robust sequential detection using arrays

Abstract

A major problem in designing sequential detectors for array applications under varied temporally correlated noise distributions is the generalization of Wald's fundamental identity to include dependence. The approach taken is to first obtain a robust sequential detector structure for the array by partitioning the observations into known probability regions obtained from a learning sequency of independent samples from the unknown noise field. Assuming the sampling rate is increased so that the temporal samples are correlated, the data is accumulated in blocks of no samples. The block information is then processed sequentially. Using Wald's fundamental identity the probability of detection and the average sample size is obtained. It is shown that for weak signals the thresholds needed to insure the desired type I and type II errors depend upon the correlation function of the noise. However, once these thresholds are known improved performance in terms of reduced time to detection can be obtained. Data is presented which compares the performance of an array of M hydrophones for various correlation functions assuming Gaussian noise. [Work supported by Naval Material Command.]