Neural detectors for signals in non-Gaussian noise

The authors demonstrate that a neural network can be trained for the purpose of detecting a known signal corrupted by additive Gaussian as well as non-Gaussian noise of the impulse type. It is shown that, in the presence of Gaussian noise, the performance of a properly trained neural network is very similar to that of the optimum matched filter detector. In the presence of non-Gaussian noise, however, neural detectors are shown to perform better than both the matched filter and locally optimum detectors.<>