Abstract:

Existing fingerprint-based indoor localization uses either fine-grained channel state information (CSI) from the physical layer or coarse-grained received signal strength indicator (RSSI) measurements from the MAC layer. In this paper, we propose to use an intermediate channel measurement — spatial beam signal-to-noise ratios (SNRs) that are inherently available during the beam training phase as defined in the IEEE 802.11ad standard — to construct the feature space for location-and orientation-dependent fingerprinting database. We build a 60-GHz experimental platform consisting of three access points and one client using commercial-off-the-shelf routers and collect realworld beam SNR measurements in an office environment during regular office hours. Both position/orientation classification and coordinate estimation are considered using classic machine learning approaches. Comprehensive performance evaluation using real-world beam SNRs demonstrates that the classification accuracy is 99.8% if the location is only interested, while the accuracy is 98.6% for simultaneous position-and-orientations classification. Direct coordinate estimation gives an average root-mean-square error of 17.52 cm and 95% of all coordinate estimates are less than 26.90 cm away from corresponding true locations. This concept directly applies to other mmWave band (e.g., 5G) devices where beam training is also required.