| The GPS is a mature outdoor positioning technology,but its positioning accuracy is not high in indoor environments where satellite signals are partially blocked.Various indoor positioning techniques have been proposed based on the situation.Indoor positioning technology includes active positioning and passive positioning.Compared with active positioning technology,passive positioning target does not need to carry equipment and can meet the needs of emerging markets.At present,due to the dense deployment of WIFI in many indoor environments,and its application restrictions are small,it can be positioned without being noticed.So,indoor passive positioning technology based on WIFI is studied in the thesis.The CSI can provide channel information more detailed than RSSI.This information can improve the accuracy of indoor passive positioning technology,and CSI indicators can now be obtained on some civilian network cards,not just dedicated devices,so it can be applied in many scenarios.In this paper,CSI with multiple-input multiple-output(MIMO)technology can be used as an indicator of indoor passive positioning.Positioning technology uses CSI space and frequency diversity to improve indoor passive positioning accuracy is designed.Firstly,a new CSI preprocessing scheme to solve the problem of abnormal values in CSI samples collected under multipath and environmental interference was proposed.The noise was reduced by density clustering of CSI original data,and the subcarrier with the largest contribution was extracted.Then,the weighted average processing of each set of CSI amplitudes is performed,and the processed CSI amplitudes and geographic location coordinates from different APs are used as fingerprint features to construct a location fingerprint database.In the online real-time positioning stage,the position of the measured object is estimated according to the Euclidean distance priority selection method and the WKNN algorithm.Finally,by analyzing the CSI data value,the mathematical linear transformation method is used to extract reliable phase information,therefore,the amount of available information for indoor positioning was increased.At the same time,in order to overcome the problem that the positioning accuracy is degraded due to the expiration of the fingerprint database,a rigorously designed artificial neural network update scheme is used to update the fingerprint database,and the multi-layer perceptron(MLP)network structure is used for nonlinear relationship mapping.BP algorithm training further improves the performance of the system.The influence of multiple parameters,number of antennas,size of training grid unit and different positioning algorithms on indoor positioning performance are simulated.The results show that when the size of training grid unit is 0.5m*0.5m and the number of antennas is 3,better precision can be obtained compared with other indoor passive positioning technologies of the same type. |
