CSI Fingerprint and GCN Based Indoor Localization Using Graph Structures Fusion

Since the channel state information (CSI) can provide a fine-grained description of signal propagation process, in this paper, a new CSI fingerprint and graph convolutional network (GCN) based indoor localization algorithm is proposed. First, the amplitude and phase information of CSI measurement is extracted for amplitude and phase based CSI gray image construction, respectively. And then, the corresponding CSI RGB images are obtained by the image render techniques. Next, a pixel value comparison based rule is used to form the graph structure of each channel of the amplitude and phase based CSI image. The front-end fusion is used to form the final graph structure. At last, the GCN model is used to training the graph structure. Some Experiments are carried out to verify the performance of the proposed algorithm.