N.M. Novikova, V.G. Lyalikova

The signals detection in random noise is a most important problem of the theory and technics of communication and radar systems. This problem is decided with statistical methods, but the neural networks are quite capable of doing it. The articles aim is comparison of characteristic neural networks and statistical (parametric and non-parametric) of signals detectors. Bayesian method, maximum-likelihood method has been used for modeling parametric detectors. Sign method, sign-rank method has been used for modeling non-parametric detectors. Algorithms of education Hamming neural network, Kohonen neural network and RBF neural network have been used for modeling signals neural detectors. Quasideterministic signal in gaussian noise and chaotic pulse noise has been used for computing experiment realization. The detection probability and false-alarm probability have been received in the computing experiments. The comparative analysis of the experimental results has been realized. In this article has shown, that RBF neural network has provided with such characteristics of signals detection as Bayesian algorithm in the presence gaussian noise, and in the presence chaotic pulse noise.