An Indoor Positioning Method Based on CSI by Using Features Optimization Mechanism With LSTM

In recent years, the research on indoor positioning has received extensive attention, especially the positioning method without portable devices. In this paper, we propose a positioning method by optimizing the channel state information (CSI) amplitude and phase data feature ratio. In the off-line training stage, for each experimental scenario, we select different proportions of amplitude and phase data to form different data sets, and train with LSTM neural network. By comparing the results of the training, the model under the optimal feature ratio is obtained. In the on-line localization phase, the model predicts the regression of the test points by calling the prediction function and outputs the results. Experiments are conducted in open environment and complex laboratory environment to evaluate the performance of the method, and we compare this method with the current state-of-the-art indoor positioning solutions, such as: DeepFi, FILA, RNN and EC-SVM. Experimental results are presented to confirm that our method can effectively improve the accuracy of indoor positioning.