Dynamic Phase Calibration Method for CSI-based Indoor Positioning

The demand for location-based services (LBS) increases significantly with the development of smart devices. Their built-in WiFi capability makes WiFi-based approaches essential for a range of indoor positioning applications. In such LBS systems, accessing received signal strength indicator (RSSI) and finer-grained channel state information (CSI) is enabled by modifying commodity WiFi devices. Additionally, multipleinput and multiple-output (MIMO) and orthogonal frequencydivision multiplexing (OFDM) provide the spatial and frequency diversity to build the fingerprint database with CSI. However, due to hardware and environmental impacts, such systems suffer from phase errors and fingerprint noise. In this paper, a novel phase calibration method is proposed to reduce the fingerprint noise and improve the accuracy of CSI-based indoor positioning systems. The CSI phase of each subcarrier is extracted from the WiFi access points in a multi-antenna wireless network. First, the phase offset is calculated through the conventional method that uses a linear transformation to remove phase errors. Then, a dynamic phase calibration method is introduced to compensate for the phase offset by tracking the anomalous phase difference between each CSI sample and neighboring subcarrier. Finally, a machine learning algorithm is trained to estimate the target position. The performance of the proposed algorithm is investigated by evaluating the prediction rate from a margin of error (MoE) model and calculating the average distance error between the predicted grid and ground truth. Experimental results show the dynamic phase calibration method outperforms the conventional linear transformation calibration method by a higher prediction rate and improves the average position accuracy.