Classification of the temperature-dependent gain of an erbium-doped fiber amplifier by using data mining methods

Abstract

In this study, the experimental results of an erbium-doped fiber amplifier (EDFA) system designed and experimentally installed in the C-band were classified by using data mining methods. The non-linearity of the EDFA gain with temperature is a problem for EDFA designers. For this reason, the wavelength range to be used in the design, the temperature values and change in the input power values of signals coming to the design are not linear with respect to the output power. Data mining method based on different models are proposed to solve this problem. The data set used contains four variables: temperature, wavelength, input power, and output power. The success results of at WEKA software (Waikato environment for knowledge analysis) were compared with 6 different algorithms. As a result of the analysis made, the C4.5 algorithm, which is the decision tree classification technique, obtained 98.4 % close results in experimental results. When the results are evaluated, data mining algorithms can be constructed and validated using larger data sets. This will help the fiber optic data to be quickly understood and easy to design for optical designers in practical applications.