Design And Implementation Of CSI Fingerprint Location Based On Convolutional Neural Network

Indoor position technology shows commercial value with the increasing demand for locationbased services.At present,Wi-Fi-based indoor positioning technology is widely used because of the advantages of widely deployed facilities.Compared with Received Signal Strength Indication(RSSI),Channel State Information(CSI)provides more refined the channel state information of the system,which is less affected by the multipath effect,and the signal characteristics are more stable.Therefore,the fingerprint location system based on CSI has become a research hotspot.At the same time,with the prevalence of machine learning,the use of neural networks for CSI fingerprint recognition has also been increasingly studied in indoor positioning,such as the Con Fi system.To resolve the problem of the poor generalization ability of neural network-based CSI fingerprint localization method,and the significant location error of unlearned location,a CSI fingerprint positioning method(Re-Fi system)based on the residual convolutional neural network and the best data enhancement strategy—Full Antenna Cross(FAC)strategy are proposed.First,the CSI subcarrier information and time domain information of different channels are analyzed.By changing the organization of CSI feature images,the accuracy of the positioning system is improved.After a comparative analysis of different data enhancement strategies,the FAC strategy is determined to be the best data enhancement strategy.Then the residual convolutional neural network is applied to the CSI fingerprint positioning model,the structure of the residual learning unit is analyzed,and the pre-activated structure with a better positioning effect is selected through experiments.Moreover,the size of the convolution kernel is selected,and the network parameters with better classification effects are selected through experiments to be suitable for indoor positioning application scenarios.The experimental results show that the FAC strategy improves the positioning accuracy of the Con Fi system by 14% in the scenario of 16m×7m and the test position of a non-training position.Compared with Con Fi system and D-CSI system,the Re-Fi system and FAC strategy reduce the average fixed error by 0.96 m and 1.02 m respectively,and improve the positioning accuracy by about25%.Completed the design index and improved the positioning accuracy.