S. V. Zimina

In the article the results with take into account weight vector jitter of statistical analysis of neural networks characteristics with gradient algorithms tuning and accepting broadband signals are shown. In the statistical analysis process are considered the two algorithms - discrete gradient algorithm with repeated linear constraints and algorithm of recurrent inversion of correlation matrix of input signals. Both algorithms were adapted to tuning neural network. Expressions are received for correlation functions and powers of output signals neural network lays. It is shown, that and take into account weight vector jitter, and take into account broadband input signals leads to distortions of the signal neural networks lays. Take into account weight vector jitter leads to occurrence new terms in the formulas of correlation functions and powers of output signals neural network lays. Take into account broadband input signals adds to these statistical characteristics additional multipliers, which to deform allocating of the neural network signal. These distortions the it is more, than it is more lays contains the neural network.