Research On Distributed Time Synchronization For Wireless Sensor Networks

Wireless sensor networks(WSN)is a multi-hop ad hoc distributed network with low-power,low-cost and low-rate.As a crucial technical support for WSN application,time synchronization is able to build a consistent concept of time,and has been widely used in data acquisition,location service,and low-power dormancy,but it has been greatly constrained by the network node in many aspects,such as computing capability,communication resources and energy,and also faces the constraints of network structure on the performance of algorithms.Due to their respectively significant advantages in large diameter and dynamic WSN,flooding time synchronization and average consistency time synchronization have become essential research hotspots in distributed WSN time synchronization.However,influenced by transmission delay,parameter estimation model and synchronization framework,they still face major challenges:the continuous decrease of synchronization accuracy with the increase of network diameter in flooding time synchronization caused by accumulation of multi-hop errors;the slow synchronization convergence rate in average consensus-based time synchronization caused by iteration;and the adverse effect of transmission delay on estimation accuracy caused by the one-way broadcast model for estimating clock offset,which is a common problem in both methods.With the aspects of parameter estimation model,error factors and convergence characteristics,both of theory and error regularities analysis on existing time synchronization methods and are carried out in this paper.Aiming for the existing technical shortcomings and problems,new parameter estimation methods and time synchronization algorithms are proposed.The advantages and limitations of the proposed algorithm in synchronization accuracy and convergence speed are verified by experiments,and the influence of network structure on the performance of the algorithm is also analyzed.The main work in this paper:1.Study on the characteristics of random delay distribution and the error factor of parameter estimation is done.The expression of abnormal delay values is introduced into the existing Gaussian random delay model by analyzing the influence of multi-interrupt tasks on random delay distribution;and the corresponding error models of two-way message exchange model and one-way broadcast model are deduced by studying the error factors of clock offset estimation in both models.2.The proposed clock skew estimation method and clock offset estimation method based on Maximum Likelihood Estimation.Aiming at the time synchronization framework of unidirectional broadcast,the clock skew maximum likelihood estimation is developed based on Gaussian stochastic delay model,with the corresponding Cramer-Rao lower bound;and the clock offset maximum likelihood estimation is also developed with the corresponding error estimation model.3.Rapid-flooding Multi-broadcast Time Synchronization(RMTS)is proposed in this paper,and a innovate implementation of RMTS with real-time delay compensation is obtained by constructing a clock skew estimation model with joint clock drift in unidirectional broadcast.The theory and regularity of multi-hop error accumulation in flood time synchronization are analyzed,and the technical routes of existing flood time synchronization algorithms and optimized RMTS are all described with their respective multi-hop error accumulation models.4.The convergence theory of average consistency are evolved by studying its convergence characteristics,and the multi-hop average consistency time synchronization(MACTS)is proposed based on the conclusion that the greater algebraic connectivity leads to faster average consistency convergence.The implementations of MACTS based on message forwarding and parameter sharing are designed respectively without changing the topological structure,meanwhile corresponding qualitative analysis of error in the algorithm is carried out.Comparison experiments and analysis are carried out for verification:1)Comparison experiments on proposed clock skew maximum likelihood estimation with clock skew estimation based on linear regression and direct estimation shows that the proposed method has smaller estimation error.2)Comparison experiments in large-diameter linear network on proposed RMTS with FTSP,FCS A and PulseSync verify that the improved accuracy of multi-hop synchronization in RMTS.3)Comparison experiments in line(Grid)network on proposed MACTS algorithm with ATS show that the synchronization accuracy and convergence speed of MACTS are better than ATS,and the synchronization performance of MACTS is less affected by the network shape,which can achieve accurate time synchronization with fast convergence in large-scale WSN.4)By comparing and analyzing the impact of network structure on RMTS and MACTS,the practicability and limitations for different application scenarios are elaborated.