Robust sequential detection of weak signals in undefined noise using acoustical arrays

Abstract

In detection problems where the noise distribution may be unknown, employing nonlinear elements that precede the standard detector often significantly improves system performance in non-Gaussian noise. However, in order to optimize the system to the noise conditions and, thereby, prevent degradation under Gaussian noise, the nonlinear elements should possess an inherent capability for adapting to the changing noise conditions. This paper presents a discrete formulation based on m-interval partitioning for sequentially detecting weak signals in undefined noise. Partitioning requires knowledge of a small number of quantiles and related functions from the unknown noise distribution. The resultant detector is easily implemented and is adaptable to slowly changing noise conditions. Two examples are given that clearly show improved performance, over a wide class of noise distributions, for an array employing m-interval partitioning compared with arrays employing linear correlators and sign detectors.