Adaptive sonar detection performance prediction in an uncertain ocean

Abstract

This paper addresses the problem of predicting detection performance when the signal wavefront is uncertain and the noise field directionality is unknown. Passive sonar detection in this scenario typically involves robust adaptive beamforming with limited training data. The classical sonar equation, however, assumes the noise field and signal wavefront are known exactly. In this paper, we use the statistics of the generalized likelihood ratio test (GLRT) for the composite hypothesis of a multirank signal in Gaussian noise with unknown covariance matrix to evaluate the detection threshold (DT) as function of ocean uncertainty and number of noise training snapshots. Further, the trade-off between array gain (AG) and detection threshold (DT) is studied as a function of training sample size in a dynamic interference environment. Detection performance of the GLRT is characterized in terms of bounds on the middle 80th percentile of classical passive sonar figure of merit (FOM) and range-of-the-day (RD) over an ensemble of ocean environments. Different classes of environments including downward refracting and upward refracting scenarios are examined with particular attention to the Florida Straits region. Example performance prediction bounds are presented using real horizontal noise field and environmental data. [Work supported by ONR.]