Research Of Localization And Tracking In Wireless Sensor Networks

Wireless sensor networks is a new information acquisition and processing technology, and is integrated from the development of wireless communication, embedded system, micro-electro-mechanical system and low power design techniques. Wireless Location is one of the key technologies in wireless sensor networks. The contents of three aspects in wireless location are measurement, localization and tracking.The accuracy of the traditional localization techniques based on ranging is always low, and the algorithm convergence is poor. The basic particle filter for tracking depends on the complex motion model, which has a great influence on the precision. More over,the computational complexity of the traditional DOA estimation algorithms are always too high to engineering realization.Aiming at these problems, this dissertation focuses on three problems: more accuracy location method, more efficient particle filter without the dependency of motion model, and low complexity DOA estimation algorithm. Based on the analysis of the existing algorithms, the main works and innovations in this dissertation are summarized as follow:A more accurate localization method base on ranging is proposed. To overcome the performance degradation caused by the low SNR and the non-Gaussian noise distribution,the trust region algorithm based on the conic model is employed for iterative localization.The cone model is used for approximating the objective function of localization, and the optimization process is converted into a series of optimization sub-problems. Through the simulation and experiment with the Zigbee nodes in a meeting room, it has been proved to be efficient even in low SNR scenario. This method is not sensitive to the geometric distribution of anchor nodes, which means that it has higher robustness and stronger adaptability in dynamic environment. Further more, a fingerprint method combined with inertial navigation tracking solution is proposed, and a tracking experiment has been conducted for proving the validity of this method.An improved resampling method and a particle filter based on trust region algorithm are proposed. By introducing the genetic crossover and immune mutation mechanism, the particles are processed with arithmetic crossover and mutation operators for improving the diversity, and the particle degradation and particle poverty problem have been eased.The method with improved resampling can handle the jumping of system state for higher performance of tracking. In the improvement of particle filter, the trust region algorithm is employed for predicting the change of system state, and an iterative correction method based on the sequential importance sampling in the neighbourhood of the state been predicted is introduced. Without the inaccuracy of the motion model, the method above provides higher performance in the tracking of moving target even with small number of particles.Two improved DOA parameter estimation methods for coherently distributed(CD)sources are proposed. Using the approximate rotational invariance property between two subarrays in a L antenna array and the propagator method, the first method avoids spectrum-peak searching and the eigendecomposition of the high-dimensional sample covariance matrix. It has significantly reduced the computational cost compared with the existing methods, and is also a robust estimator which does not depend on the angular distribution shape of CD sources. The second method represents the generalized steering vectors of CD source on overcomplete dictionaries subject to sparse constraint in subspace fitting method, the subspace fitting problem is transformed into a sparse reconstruction problem. Then the sparse reconstruction problem can be solved and optimized by the second order cone programming(SOCP) framework. This method has better resolution performance, especially in small number of snapshots.