Research On Optical Performance Monitoring Technology In Optical Communication System Based On Neural Network

With the emergence of more emerging services,the demand for data traffic has increased exponentially.The optical network has also been continuously upgraded,and is gradually developing towards the direction of dynamic,transparent and high-capacity.In high-speed optical networks,the impact of impairment factors such as chromatic dispersion on system performance has been aggravated.In order to compensate the damage of the signal at the receiving end,it is necessary to monitor the impairment such as chromatic dispersion in advance to achieve better compensation effect.At the same time,the modulation format of the signal needs to be obtained in advance to select the appropriate compensation algorithm for the compensation of many parameters.In this paper,a neural network is used to complete the task of optical performance monitoring.The input data of the neural network is the asynchronous amplitude histogram,and the optimal monitoring result is obtained by optimizing the structure of the neural network.The main contents of this paper are as follows:(1)The optical fiber communication systems with six modulation formats were built by simulation,and the data sets of optical signals with six modulation formats were obtained.A three-layer fully connected neural network was built to realize the classification task of six modulation formats and the asynchronous amplitude histogram was used as the input vector of the neural network.(2)The multi-layer neural network was built to complete the simultaneous monitoring task of chromatic dispersion and optical signal-to-noise ratio(OSNR)in the simulation system.Then the influence of the number of network layers on the monitoring effect was verified.Based on the neural network and asynchronous amplitude histogram,the monitoring range of chromatic dispersion was studied,and the influence of the number of bins on the effect of chromatic dispersion monitoring was also verified.(3)In order to improve the calculation speed of the neural network,an opto-electronic neural network based on a few-mode fiber was designed,and the multiplication and summation operations of the neural network can be realized on the light.Using the parallel computing capability of the few-mode fiber,the length of the fiber used to stretch the optical pulse can be reduced.The performance of the electrical neural network and the opto-electronic neural network was verified using the handwritten digit recognition data set.The classification accuracy of the two neural networks is basically the same,but the opto-electronic neural network has a faster calculation speed.