Probing fiber parameters and predicting pulse evolution by using machine learning algorithms

Inhalt:
Ultrashort pulses evolve into seemingly chaotic complex signals in nonlinear systems. These signals carry a large amount of information of the transmission system. Through information extraction and data analysis of these chaotic signals, fiber information can be extracted. However, such nonlinear evolution is computationally demanding and difficult to achieve using traditional methods to extract effective information. In this paper, a method for extracting effective information of optical fiber based on machine learning is proposed, which can be used to realize optical fiber parametric detection. The mean square error detected by this method can be controlled below 1.19%. Further, we use the machine learning method to simulate the nonlinear evolution of the optical signal in the pulse, which can solve the difficulty of optimizing the pulse propagation requires a large number of numerical simulations, and the prediction error of evolution can be controlled below 0.37%.