Research On Parameter Estimation Of Optical Fiber Physical Bayer Based On Machine Learning

With the increasing popularity of the Internet and the Internet of Things,as well as the introduction and use of various network technologies and online applications,the demand for network bandwidth is increasing.The demand for digital information in the future will exceed the maximum capacity of the current optical communication system.This poses a huge challenge to the optical fiber transmission technology as a backbone network.On the one hand,it is necessary to ensure high reliability,on the other hand,it must ensure large capacity.In order to balance these two points,it is necessary to use suitable optical performance monitoring(Optical Performance Monitoring,OPM)technology to monitor the physical layer parameters of optical fiber transmission to achieve flexible resource allocation and early abnormal diagnosis.Traditional OPM technology is becoming more and more difficult to meet the demand under the current demand,which prompts people to find new technologies to support OPM.Artificial Intelligence(AI)technology is an emerging technology that has first-class function fitting and parameter prediction capabilities.The use of AI technology is expected to find new ways to utilize existing resources to meet future challenges in larger and more complex optical communication systems.With the efforts of scientific researchers,the OPM technology and AI have been successfully combined.The combination of the two enables OPM to break through the traditional technical barriers.It can not only realize highprecision parameter prediction,but also realize multi-parameter joint prediction at the same time.Multi-level cascade networks and multi-task learning(MTL)networks are the first choices people use for multi-parameter joint monitoring.The method of combining manual feature extraction and multi-parameter monitoring network can realize multiparameter monitoring,and usually does not require additional hardware facilities.This reduces the cost of parameter monitoring to a certain extent and will gradually become the first choice for parameter monitoring.Coherent optical communication can utilize the signal's amplitude,phase and polarization state information,which can greatly increase the signal transmission capacity.Using the method of coherent detection,the coherent receiver has extremely high sensitivity,so that the signal can be transmitted over a long distance.In order to explore the application of deep learning in coherent optical communication,this paper improves and researches the two-parameter joint monitoring method based on deep learning for modulation format identification(MFI)and OSNR estimation in polarization multiplexed coherent optical communication systems.The cascade network and its improvement plan are discussed through simulation and experiment.The improvement plan is cut from two perspectives.One is to reduce the error transmission of the cascade network by introducing an abnormality detection algorithm to the cascade network,thereby improving the monitoring accuracy of the cascade network.The second is to use migration learning to simplify the complex architecture of the multi-level cascaded network,so that the cascaded network simplifies the structure without losing accuracy.Both of these two methods can overcome the shortcomings of the cascade structure to a certain extent,and make the cascade structure more accurate and simpler.In order to show that the improved cascade structure is an effective structure,the method of adaptive multitask learning and the original cascade structure is also used for parameter monitoring for comparison and explanation.