DETECTION OF NON-GAUSSIAN PROCESSES IN NON GAUSSIAN NOISE

The detection of stochastic processes in noise is considered, under the assumption that neither the signal nor the noise need be Gaussian. The detector structure is found in terms of the semiinvariants of the signal and noise processes. The general detector structure is extremely complicated, but a threshold form may be obtained. For symmetric processes with zero mean and independent sampling, the energy detector is obtained. Error probabilities are computed for the energy detector with non-Gaussian signal process andor non Gaussian noise. It is shown that large degradations in sensitivity occur if the noise is highly impulsive in character, but the non-Gaussian character of the signal process is found to have very little effect on the detector sensitivity.