Distributed detection in dependent nonGaussian noise

Finding optimum distributed detection schemes is a difficult mathematical problem which has received very little attention. Cases with dependent non-Gaussian impulsive noise are of particular interest and have not yet been studied. A two-sensor known-signal detection problem is considered where additive impulsive noise, which is dependent from sensor to sensor, corrupts the observations. The noise is modeled as a mixture of Gaussian distributions, a typical model for impulsive noise. A criterion of Bayes risk is adopted for cases with fixed fusion rules. The optimum sensor tests are shown to be different from the best isolated sensor tests (likelihood ratio tests) in several cases. Further, a methodology for predicting the form of the optimum sensor tests is presented