Analysis of Some Undersea Noise with Applications to Detection

摘要

This paper treats a signal detection problem using real noise records. The authors have had access to two medium-sized (each about 100000 samples) data sets, each from a non-Gaussian marine environment. The first, collected near a merchant shipping lane, has lower-than-Gaussian probability of high-amplitude observations (light tails). This appears to be due to the presence of a powerful narrowband component; when this is removed by an adaptive notch filter, the tails become heavier. The second, collected near a shrimp bed, has extremely heavy tails and a kurtosis (normalized fourth moment) of 82. Excellent results are achieved through the use of a Johnson noise model fit. Keywords: Merchant noise; Shrimp noise; Johnson noise; Nonlinear detectors. (jhd)