Abstract
Abstract
With continuous growth in modulation formats, the requirement for autonomous devices is becoming more important than ever. Predicting error vector magnitude (EVM) of m-ary quadrature amplitude modulation (mQAM) are intricate issue for the effective design of transmission systems. Existing estimation techniques have survived through repetitive processes that are frequently computationally expensive, and time-consuming. Recently deep learning approaches demonstrated good performance as useful computational tools, offering a different way for accelerating such mQAM simulations. This paper introduces an artificial neural network (ANN) architecture that aims to forecast the EVM of the popular modulation forms including 18 Gbaud 8QAM, 14 Gbaud 16QAM, and 10 Gbaud 64QAM under different transmission conditions. Amplitude histograms (AHs) are produced from constellation diagrams obtained with varying launch power, laser linewidth, OSNR, and transmission distance by an offline preprocessing flow. The fully trained framework exhibits superior performance in terms of computing cost compared to the simulation experiments. The overall execution time of the ANN-based modeling method is approximately 234 s as opposed to more than 23000 s when employing the simulation technique, resulting in a 99% reduction in computation time. As a result, this technology opens the door to quick, all-encompassing techniques for characterizing and analyzing optical fiber problems.