Research On Construction Technology Of Multicast Light-Tree Based On Deep Learning

With the development of Internet,Internet of things,and 5G industry,multicast services,such as scientific computing,ultra-high-definition TV delivery,online games,video conferencing,and data backup,are gaining popularity and momentum.Point to multi-point transmission is a typical feature of multicast services,which transmit the same data from one source to multiple destination users.Multicast service can be transmitted in IP layer and optical layer.With the development of optical network technology,optical network effectively supports the growing demand of data transmission and communication.In order to reduce network cost and spectrum resource consumption,multicast light-tree is often used to accommodate multicast service in optical layer.The way of light-tree can effectively reduce the consumption of transceivers and spectrum resources than using light-path to accommodate multicast service.In addition,multicast flow aggregation based on a certain strategy can reduce the number of light-tree construction.However,as the routing and spectrum allocation of multicast service is NP-complete,it is difficult to find the optimal solution in the case of large scale networks and high traffic volume.At the same time,due to the signal-to-noise ratio and the availability of the multicast light-tree is limited by modulation format,transmission distance,light-splitting,number of users,and a series of factors.So it is difficult to make an accurate model to unify these variables,which makes it difficult to predict the signal-to-noise ratio and the availability of multicast light-trees.Therefore,how to efficiently construct the light-trees in optical network is an important problem to be solved.This thesis focuses on the construction of the multicast light-trees.The optimization of multicast light-trees routing and spectrum allocation,the difficulty of availability prediction,and the complexity of optical signal noise ratio(OSNR)measurement are discussed.This thesis proposes a routing and spectrum allocation method based on multicast flow aggregation.And the strong awareness and learn ability of deep learning is applied to predict the OSNR and availability of light-trees,and which is introduced in the optical network architecture to direct the routing and resources allocation of multicast light-trees.Specific research work is as follows.(1)To reduce light-tree building cost and improve the utilization rate of network resources,this thesis proposes a modulation-level-aware multicast flow aggregation scheme in elastic optical networks,which also considers the data source and user sets of multicast services.The highest available modulation-levels based multicast flow aggregation(HMLA-MFA)scheme and the lowest available modulation-levels based multicast flow aggregation(LMLA-MFA)scheme are designed,according to the constraint relation between the modulation format,transmission distance and the number of users.Two heuristic algorithms are developed to realize it respectively.The simulation results show that the scheme can reduce the number of light-trees construction and transceivers with acceptable spectrum consumption and blocking rate.(2)Because the quality of transmission(QoT)of a light-tree is influenced by a variety of physical layer impairments,the traditional modeling method is complex and time-consuming.Optical network is a dynamic network,so the QoT of the light-tree is difficult to calculate accurately.In order to improve the success rate of light tree construction,this paper proposes a deep learning-based light-tree signal-to-noise ratio prediction method and a deep learning-based light-tree availability prediction method.The performance of the proposed method is analyzed on the test platform.Numerical results show that the accuracy of the deep learning based light-tree signal-to-noise ratio prediction method is about 95%.The accuracy of the deep learning based availability prediction method is more than 98%.These two methods provide a quick decision making method for the light tree construction.(3)With the development of Internet,Internet of things and 5G,the volume of communication business has increased significantly,and the service quality requirements of each business have become higher and higher,which has put forward higher requirements for the performance of optical network.In this paper,deep learning and Software define optical network are combined to develop an optical network architecture.The light-tree signal-to-noise ratio and availability prediction based on deep learning are integrated into the optical network architecture.The function realization of each module is studied in detail.