Research On Wireless Location Algorithm Based On Graph Convolutional Network

With the development of wireless communication technology,the demand for location-based services also increases and the location-based services can be conveniently developed through wireless positioning technology.Therefore,the research on wireless positioning technology is very important.In recent years,deep learning has been widely used in various fields.Among them,the emergence of graph convolutional networks solves the problem of structure of non-Euclidean space.Since the mobile communication network is a natural graph structure(unstructured),data analysis and processing of problems in mobile communication can be carried out through the graph convolutional network.Therefore,This thesis applies graph convolutional network to mobile communication network scenarios to realize the inference of target user location,the main tasks are as follows:First of all,this thesis introduces common positioning methods,these methods usually require the target user to interact with more than 3 base stations to estimate the user's location.This thesis designs a new wireless positioning method based on MAC(Media Access Control)layer and physical layer data,using the known information of each base station side and the information reported by users to the base station,and combining the process and results of base station resource scheduling.Through the graph convolutional network model for end-to-end learning to complete the inference of the target user's location.Secondly,this thesis explains how to construct a mobile communication network as graph structure data.The specific method is to treat the signal link information between the user and the serving base station as a node,and treat the co-channel interference between users in the coverage of different base stations as the edge of the nodes.Then introduced the graph neural network model designed in this thesis,named it GCNWL(Graph Convolutional Network Wireless Locating)model.It is composed of graph convolutional layer and fully connected layer,the input of the first layer of graph convolution is the graph data after data processing,which is specifically expressed as the feature matrix and adjacency matrix corresponding to the graph data.The output of the last layer of graph convolution is used as the input of the fully connected layer after dimension conversion.Finally,the fully connected layer will output the location vector of all target users.The dimension of the vector is the number of users,and the elements in the vector are the relative angles between each user and its serving base station.Since the distance between the user and the serving base station is easy to measure,if the relative angle can be known,the specific location of the user can be calculated.Finally,this thesis designs a graph attention model(GAM)based on the graph attention mechanism,which can give greater weight to the concerned nodes.The weight coefficient between nodes reflects the intensity of co-channel interference between users,because the intensity of co-channel interference has a certain relationship with the user location distribution.The difference between the GAM model and the traditional graph convolutional network is that the graph convolutional network uses the Laplacian matrix with adjacency information when performing the convolution process,while the GAM model uses the matrix which is calculated by Laplacian matrix and the attention matrix;the attention weight between nodes will not change with the number of layers of the graph neural network,which is also the difference from the general graph attention mechanism.Finally,replace the GCN(Graph Convolutional Network)layer in the GCNWL model with GAM,and the attention matrix is added to the input of GAM so that the difference between the central node and different adjacent nodes can be obtained to improve the positioning performance.