Research On RFID Positioning Algorithm Based On Calman Filtering

By sending out radio frequency signal to detect the targets, radio frequencyidentification (RFID) technology can extract a lot of information related to the targets.By using RFID technology, the reader first sends out radio frequency signal, thenreceives the signal reflected from or transmitted by the tags, the information of the tagswill be obtained after processing at the reader side. This obtained information, byfurther processing, can be used to locate the tags. Among various RFID positiontechnologies, the position technologies based on artificial network and Kalman filter(KF) gain a lot of attention. Inspired by these facts, this paper researches on RFIDposition technologies with focus on artificial network and Kalman filter. The main workof the paper includes the follows:First, the paper introduces the basic theory of RFID and gives out the basic ideas ofRFID position technologies. According to different application principals, some RFIDposition systems are introduced, including Active Badge system, Cricket system,RADAR system and LANMARC system.Secondly, the paper gives out analysis the position technologies based onback-propagation (BP) networks and radial basis function (RBF) networks. Thedrawbacks of these technologies, such as the large number of training data, the largernumber of points in the middle layer and the dependence of adaptive capability on thetraining, have been addressed.Then, the paper analyzes some key factors in RFID position based on extendKalman filter (EKF), including: the effect of the non-optimal and non-convergence, theeffect of start and stop position, the effect of non-Gaussian noise. Based on the analysis,the paper proposes a modified location method with a modified KF. In this modifiedmethod, we first approximate the probability density function of the samples to lead tothe samples of Gaussian distribution, then we get the precise second order estimation ofthe mean and variance such that the detection is not sense to non-linear estimation.Simulation results verify that the modified method has good convergence and theestimation is closer to the real position than that of KF.At the end, we conclude the work and discuss the future development.