Research On Noise Adaptive Filtering Technique In Distributed Sensor Network

Recently,with the rapid development of the large-scale sensor networks,distributed and collaborative techniques have been widely used in various kinds of collaborative localization and tracking applications,e.g.,multirobot collaborative target tracking,collaborative target tracking in radar network and collaborative localization in wireless sensor network etc.Aiming at the practical problems of nonlinear time-varying characteristics of target and sensors,the unknown noise statistics and even the abnormal data,the existing noise adaptive estimation methods are not well suited due to the limited amounts of data.Therefore,the research on noise adaptive filtering design in distributed sensor network is very important.Due to the demands of collaborative localization and target tracking under unknown noise statistics in distributed sensor network in 973 and NSFC projects,the distributed estimation of noise statistics,abnormal data detection and sensor scheduling methods are emphatically studied to improve the estimation accuracy,real-time performance and the robustness to abnormal data on the basis of summary of the domestic and overseas existing techniques in noise adaptive estimation,distributed estimation,abnormal detection and sensor scheduling area.Firstly,to solve state estimation problem in linear time-invariant(LTI)system with unknown process and measurement noise covariance,the distributed auto-covariance least squares algorithm is developed.The algorithm includes noise covariance distributed estimation and information sharing design parts.The theoretic accuracy analysis and the simulation results of target tracking under static and mobile sensor networks show the superiority of the proposed algorithm to the conventional algorithms in terms of the estimation accuracy and the realtime performance.The experiment of multirobot collaborative target tracking is carried out to show the improvement of the proposed algorithm in robustness to abnormal data.Secondly,to solve state estimation problem in nonlinear system with unknown process noise covariance,the weighted-optimization based distributed Kalman filtering(WODKF)algorithm is developed.The algorithm includes the design and the optimization procedure of the weighted cost function,the sensor selection,abnormal data detection and the reconstruction of the filters parts.The algorithm is also combined with the local probability data association method to solve the data association problem in multi-target tracking case.The simulation results of 1D signal,2D unicycle and multi unicycles tracking cases show the superiority of the proposed algorithm to the conventional algorithms in terms of the estimation accuracy and the realtime performance.The experiment of high speed target tracking in radar network is carried out to show the improvement of the proposed algorithm in detection ability of abnormal data and the tracking accuracy in initial time.Thirdly,to solve state estimation problem in linear time-varying system with unknown non-Gaussian measurement noise distribution under three conventional distributed network topology,i.e.,increment,diffusion and consensus,the fast variational Bayesian based distributed adaptive filtering(FVBDF)algorithm is developed to improve the performance of the estimation accuracy and robustness to outliers with acceptable computation and communication capacity.The algorithm includes the distributed estimation of the noise statistics approximated by Studentt distribution and the sensor selection based on Cramér-Rao lower bound.The simulation results under increment,diffusion and consensus networks show the superiority of the proposed algorithm to the conventional algorithms in terms of the estimation accuracy and the real-time performance.Fourthly,to solve the problem of the sensitiveness to distance outliers due to nonline-of-sight environment and multipath fading channel in cooperative target localization application in distributed sensor networks,an outlier robust trilateration algorithm is developed based on the previous noise adaptive algorithms and sensor selection method.The algorithm includes the intersection determination principle and distance compensation method,which solves the problems of two intersection points and no intersection points of anchor circles.We design the indoor UWB localization experiment to show the good performance of the proposed method in terms of location accuracy and robustness to distance outliers.