Channel state information (CSI) provides a fine-grained description of the signal propagation process, which has attracted extensive attention in the field of indoor positioning. The CSI signals collected by different fingerprint points have a high degree of discrimination due to the influence of multi-path effects. This multi-path effect is reflected in the correlation between subcarriers and antennas. However, in mining such correlations, previous methods are difficult to aggregate non-adjacent features, resulting in insufficient multi-path information extraction. In addition, the existence of the multi-path effect makes the relationship between the original CSI signal and the distance not obvious, and it is easy to cause mismatching of long-distance points. Therefore, this paper proposes an indoor localization algorithm that combines the multi-head self-attention mechanism and effective CSI (MHSA-EC). This algorithm is used to solve the problem where it is difficult for traditional algorithms to effectively aggregate long-distance CSI features and mismatches of long-distance points. This paper verifies the stability and accuracy of MHSA-EC positioning through a large number of experiments. The average positioning error of MHSA-EC is 0.71 m in the comprehensive office and 0.64 m in the laboratory.

Channel state information (CSI) provides a fine-grained description of the signal propagation process, which has attracted extensive attention in the field of indoor positioning. However, considering the influence of environment and hardware, the phase of CSI is distorted in most cases. It is difficult to extract effective location features in multiple scenes only through the determined artificial experience model. Graph neural network has performed well in many fields in recent years, but there is still a lot of room to explore in the field of indoor positioning. In this paper, a phase feature extraction network based on multi-dimensional correlation is proposed, named Cooperation-Graph Convolution Network (C-GCN). The purpose of C-GCN is to extract new features of multiple correlation and to mine the relationship between antenna and subcarrier as much as possible. C-GCN is composed of convolution layer and graph convolution layer. In the graph convolution layer, C-GCN regards each subcarrier of each antenna as a node in the graph network, constructs the connection by the correlation between the antenna and the subcarrier, and aggregates the node vectors by graph convolution. In the convolution layer, there is a natural corresponding structure between data packets, C-GCN extracts the fluctuation with convolution in Euclidean space. C-GCN combines these two layers, and applies end-to-end supervised training to obtain effective features. Extensive experiments are conducted in typical indoor environments to verify the superior performance of C-GCN in restraining error tailing. The average positioning error of C-GCN is 1.29 m in comprehensive office and 1.71 m in garage. Combined with the amplitude feature, the average positioning error is 0.99 m in comprehensive office and 1.14 m in garage.