Neural Network Models for Approximation of One-Dimensional Signals

The paper suggests a new solution for the problem of analyzing one-dimensional digital signals by implementing a neural network approach. The proposed neural network approach allows improving the quality of approximation by simplifying structural identifying, whereas only the first hidden layer of artificial neural network models is used. The same approach allows reducing the parametric identification computational complexity and ensuring very good scalability through the usage of batch training mode of artificial neural network models. Finally, it is capable of describing nonlinear dependencies through the usage of artificial neural network models, thus achieving great accuracy through the usage of local approximation. The proposed method and models make it possible to extend the application scope of application of methods of approximation of one-dimensional digital signals based on artificial neural networks, which contributes to the efficiency of intelligent systems for special and general purposes.