Amplitude noise level prediction in signals with noises by statistics extraction and CFNN techniques.

The predictive analysis about the amplitude levels of noises on the form of transmitted signals in communications and electronics is a task to reduce the effect of their influence on the form of transmitted signals. The article proposes an approach combining procedures under Descriptive Analysis of registered noise signals with imposed noise, extraction of selected statistical indices and synthesis of models for prediction of noise variations through the concept of Artificial Intelligence. Research objects are Sine, Square, Triangle and Sawtooth waveforms with influence of Uniform White Noise (UWN) and Periodic Random Noise (PRN). As a neural analysis for predictive analysis, the applicability of Cascade-Forward Neural Networks (CFNNs) in Levenberg-Marquardt learning algorithm was investigated. Baseline criteria such as Mean-Squared Error, Network Performance, Correlation Coefficients, Residuals variance between Observed and Predicted Noise values were assessed in the selection processes of CFNNs for predictive analysis. [ABSTRACT FROM AUTHOR]