WiFi-based Environment Adaptive Positioning with Transferable Fingerprint Features

Channel state information (CSI)-based fingerprint approach for WiFi-based indoor localization has attracted a lot of attention recently. The fine-grained CSI represents the location-dependent channel characteristics more effectively than the coarse-grained received signal strength indicator (RSSI). However, the CSI fingerprints can deviate drastically with environmental variations. Consequently, the CSI-based fingerprint positioning models need to be adapted for different environments and/or updated over time, which is time-consuming and labor-intensive in practice. In this paper, an environment-adaptive positioning system is proposed to transfer the fingerprint features that significantly reduce reconstructing the fingerprint database. A CSI extraction platform is developed based on the modified OpenWrt firmware, enabling access to CSI measurements on commodity WiFi devices. To transfer the fingerprint features, a domain adaptation approach is proposed to reconstruct the CSI fingerprint database from the existing fingerprints with a limited number of new measurements. Experiments are conducted in several real-world test sites, including the laboratory and the lounge, with environmental change. The results show that the performance of the proposed system is promising in terms of localization accuracy and adaptation efficiency.