Research On Network Time Reference Scheme And Time Scale Algorithm

Time is regarded as one of the seven important physical quantities of the International Organization for Standardization,and high-precision time is crucial for basic science,industrial production,and economic growth.In recent years,with the continuous development of modern informatization and networking,important fields such as mobile communication networks,power networks,and financial networks have higher requirements for network time,which further improves the requirements for reliability,accuracy,and stability of network time sources.At present,network time relies on external GNSS(Global Navigation Satellite System,GNSS)signals or optic signals obtained from UTC(Coordinated Universal Time)time laboratories as time reference sources.GNSS signals are susceptible to interference and the coverage is limited.The dedicated optical fiber network connecting UTC laboratories is limited by cost and distance.Different clocks or time-using devices within the network use master-slave time synchronization to achieve a hierarchical transfer of reference time.Nevertheless,this mode of obtaining and transferring time references has external dependencies and single-point failure problems.In view of the above problems,this thesis designs a software-defined network time reference scheme based on the existing time synchronization network combined with SDN(Software Defined Network)technology.To address the fundamental issue of time scale computation in time reference establishment,a neural network-based algorithm incorporating the concept of multi-clock combination from time-frequency theory and a deep learning model has been devised.The research work and main progress of this thesis are as follows:(1)A software-defined network time reference scheme is proposed.The network time synchronization technology and the ITU-T(ITU Telecommunication Standardization Sector)network reference master clock system are investigated,and on this basis,the SDN paradigm is introduced to design the network time reference scheme.In the scheme,a software-defined network time reference architecture is first proposed,which decouples network data forwarding from time synchronization and reference generation functions.In this network,the data plane is responsible for transmitting time information,while the control plane’s network controller orchestrates network functions.The synchronization controller manages the network time synchronization process and enables the generation of a time reference.The plan also clarifies the establishment mechanism of the network time reference,which includes time synchronization and data collection,reference generation,and deviation distribution.The scheme also focuses on a reference generation method.Based on clock data preprocessing and clock screening,the time scale algorithm is used to realize the calculation of the reference time difference of multiple clock combinations.(2)A neural network-based multi-clock combined time scale algorithm is proposed.Combining the multi-clock combination scenario and the neural network model,the clock weight and prediction method in the time reference calculation process are analyzed in detail,and the historical data is fully used to predict the time difference of the clock,and the weight is adaptively corrected by network training,thereby designing the ETSP(Ensemble Time Scale Perceptron)algorithm.In this thesis,experiments are designed based on the simulation data and the measured data respectively.The results show that the ETSP algorithm can realize the time reference calculation of multi-clock combination under different data sets,showing certain applicability.At the same time,the algorithm can significantly improve the accuracy and stability of the time reference compared with other algorithms.(3)The verification experiment of the network time reference scheme is designed.According to the synchronization accuracy of the existing PTP(Precision Time Protocol)protocol,four network time error scenarios are designed,and the time reference is generated by using the ALGOS algorithm,the Kalman filter method,and the ETSP algorithm.The experiment first confirmed the feasibility of the scheme.Through the analysis of the experimental results,the following conclusions were drawn:the performance of the network time reference obtained by combining multiple clocks is related to the network performance and algorithm type.In the same network performance scenario,the ETSP algorithm can synthesize a network time reference with better performance than other algorithms.