Noise Impact Recognition in Fiber-Optic Construct Materials by Machine Learning Approach

The research examines one of the main problems of ensuring the performance of the transmission medium in the structure of optical routes. The main aspect is consist of the establishment of the noise impact types occurring in the construct materials of optical transmission lines. Gaussian White Noise and Periodic Random Noise in composite Glass and Plastic Optic Fibers were introduced as research objects. Regarding the identification processes of the target noises, an approach based on Machine Learning Methods was created. Different Machine Learning approaches were analyzed in STATISTICA and MATLAB platforms. With the highest confidence degree according to the achieved quantitative and qualitative assessment, Support Vector Machine, k-Nearest Neighbors, Adaptive Neuro-Fuzzy Interface Systems, Cascade-Forward Neural Networks and Probabilistic Neural Networks. The final synthesis of classification models is tied to the obtained maximum indices of Noise recognition, respectively 85.000 %, 87.100 %, 97.900 %, 99.06166 %, 99.700 % and 100.000 % using the Machine Leaning methods defined.