Message Passing Based Localization Algorithm For Wireless Sensor Networks

With the development of new generation of information technology such as Internet of Things(IoT),wireless sensor networks(WSNs)will gradually penetrate into various industries and application fields,influencing all aspects of people's life.WSNs are bridges connecting the human world and the material world.Embedded in various “things”,sensors sense and monitor the states of the “things” and transmit the collected information to the processing terminals via Internet to realize the monitoring and management in real-time.The information collected by the sensors is meaningful only when combined with location information which makes the administrators clear enough about what-when-where the “things” happened.So,positioning is the one of the key technologies in location-based applications of WSNs.In this thesis,node self-localization and target tracking algorithms are proposed for WSNs based on message passing algorithms(MPAs).The main contents of the thesis are as follows:1.For the static WSNs with fixed nodes,a distributed cooperative node localization algorithm based on variational message passing(VMP)algorithm is proposed.Considering the nonlinear ranging model,the existing sum-product algorithm over a wireless network(SPAWN)based on belief propagation(BP)uses a set of particles to represent messages,but the computational complexity and communication overhead are both unaffordable.The Gaussian VMP algorithm approximates the beliefs with Gaussian functions by minimizing the Kullback-Leibler divergence(KLD)to reduce the communication overhead while the computational complexity is still a problem.In the proposed algorithm,we develop an approximate method by exploiting second-order Taylor series expansion to obtain first two orders statistics for variables to approximate non-Gaussian beliefs into Gaussian forms.The simulation results show that the proposed algorithm performs close to the SPAWN algorithm and the Gaussian VMP algorithm,while there are significant reduction in terms of computational complexity and communication overhead.2.For the dynamic WSNs in which the velocity vectors of nodes can be measured,a distributed cooperative node localization algorithm based on BP and VMP is presented.According to the linear state transition model and the nonlinear ranging model,BP and VMP algorithms are used to calculate the predicted messages and the cooperative messages,respectively.Accordingly,we take advantages of the high precision of BP and the low complexity of VMP.The simulation results show that the proposed algorithm has better accuracy and convergence than the maximum likelihood estimation method.And when the variance of the priori is relatively small,its performance is close to that of the SPAWN algorithm while the computational complexity and the communication overhead are both significantly reduced.3.For the dynamic WSNs with uncertain motion model,a distributed cooperative node localization algorithm based on adaptive prediction and VMP is developed.According to the inertia of the motion,preliminary location prediction based on the correlation of trajectories within a short time period is regarded as the priori information of node location.And then exploiting the ranging measurements of the neighbor nodes,the further location estimation is obtained though VMP-based localization algorithm.The simulation results show that the proposed algorithm is more adaptive and can achieve better performance than algorithms based on instant prediction,linear prediction or square prediction.4.For the dynamic WSNs with moving nodes and targets,a joint location and target tracking algorithm is provided.The positioning process is divided into two phases: joint positioning and target trajectory smoothing.In the joint positioning phase,the agents locate themselves using the observation information between the neighboring nodes and the detectable targets.And the consensus algorithm is employed to realize distributed positioning of the targets.In the target trajectory smoothing phase,current and previous position estimations are updated by performing the forward-and-backward smoothing algorithm.The simulation results show that the proposed algorithm is better than the separate cooperative self-localization and target localization algorithms,and the trajectory smoothing algorithm further improves the performance of target tracking.