A Computational Evaluation of New Detection Algorithms

Abstract

Abstract : This paper summarizes results of a computational study of two new signal detection algorithms. The new algorithms have the potential for significantly improving existing detection methods when the signal-plus-noise process is broadband (stationary or nonstationary), especially when it is nonGaussian. They require no assumptions on the statistical properties of the signal-plus-noise process; instead, they require that the drift function of a diffusion be known or estimated. When this function is known, the new discrete- time, algorithms are approximations to the likelihood ratio for the continuous- time data under some reasonable assumptions on the data characteristics. These assumptions include that of Gaussian noise, although the computational results indicate that good performance can be obtained when the noise is not Gaussian. The study included comparisons with several reference algorithms, using both simulated and passive sonar data. The new methods gave superior performance despite the use of a very rudimentary procedure for estimating the drift function. NonGaussian signal detection, Likelihood ratios, Passive sonar, Sonar signal processing.