Indoor Positioning by CSI Amplitude and Neural Networks

Requirements for positioning accuracy indoors are constantly increasing. However, there is no single solution for organizing such positioning, and each individual project uses its own practical implementations. The paper uses channel state information at the physical layer to determine the distance between radio devices. In particular, the amplitude of the signal over 56 subcarriers is used. The task of determining the distance is positioned as a classification task. To solve it, a neural network with fully connected hidden layers is used. To train the neural network, an experiment was conducted to collect information in a technical laboratory measuring
${5.5\mathrm{x}4}$
meters. The resulting dataset is divided into a training set, a validation set, and a test set. The results of the experiments showed the accuracy in determining the distance between radio devices in the range of 99.47 – 99.75% on the training set. While on the test sample, the accuracy was 96.2 – 98.8%. The distances obtained as a result of the proposed algorithm can be used for indoor positioning using the trilateration method.