Machine learning-based prediction of attenuation in optical fiber communication systems

Attenuation is a vital limiting factor in optics fibre communication systems which has a direct impact on the transmission distance and signal quality. Conventional studies are mostly based on analytical and simulation approach and has no predictive and optimisation capability. In this paper, an optimization-based analysis of attenuation is presented as well as a machine learning (ML) prediction framework. Key parameters such as transmission distance, wavelength, number of joints, converters, etc. Connector’s role is considered. Sensitivity analysis to determine the dominant attenuation factors is carried out whereas an optimized fiber configuration is found given a definite constraint for attenuation. Furthermore, supervised ML regression models are used for predicting the attenuation with greater accuracy. Simulation results demonstrate that the proposed approach reduces prediction error from ±8 % to ±2.5 %, making it suitable for intelligent optical network planning.