Machine Learning for Location and Orientation Fingerprinting in MIMO WLANs

Abstract

To detect the position and the orientation of a mobile device within a Wireless Local Area Network (WLAN) covered by multiple access points (APs), the intrinsic properties of multiple-input multiple-output (MIMO) channels are used linking the received signal strength indicators (RSSIs) to the distance and exploiting the received signal correlation structures. Location and orientation fingerprinting is a map based positioning solution that stores for a given orientation past measurements of RSSIs at known reference/grid points in a database that is later used to localize a mobile device at an unknown location and with unknown orientation to the closest reference point. This paper focuses on processing the RSSI data vectors from multiple receiving antennas on a downlink by applying the core tools of Machine Learning (ML) classification methods to evaluate the effects of MIMO RSSI meta-data when capturing 802.11n/ac packets using commodity hardware. Specifically, the paper provides insights into the design of the overall location fingerprinting system operating with new WiFi physical link layer protocols. To verify the operation of the proposed system, experimental results are presented to investigate the impact of different factors, like the number of receive antennas, affecting the estimation accuracy for the location and the orientation of mobile user.