Stochastic signal detection in nearly-Gaussian noise using moment detectors

The detection of stochastic signals in non- but nearly-Gaussian noise with an unknown probability density function is investigated. The approach here is based on a Gram-Charlier series expansion on the noise pdf and the use of its statistical moments. A case study which uses the epsilon-contaminated mixture-noise model is presented to investigate some technical issues associated with the Gram-Charlier series including convergence and other irregularities. Examining the receiver operating curves, the proposed detector based on series expansion is found to be less sensitive to model deviations than the Neyman-Pearson optimal detectors.