Building Optimal Radio-Frequency Signal Maps

Indoor localization is a key enabler for pervasive computing and network optimization. WLAN positioning systems typically rely on fingerprints of Received Signal-Strength (RSS) measures from multiple access points (AP). In a typical use case scenario, the assumption is that motion information is not available and that the localization algorithm relies solely on matching the most recent RSS readings with accurate RSS spatial maps. And yet, the traditional way of building such signal maps, by manually collecting repeated RSS measurements at predefined locations in a building, is labor intensive and time consuming. Alternative approaches rely on robot-based automation or on crowd-sourcing that can be enhanced using WiFi-based simultaneous localization and mapping. Typically, these approaches produce time-stamped trajectories along with time-stamped RSS and enable the robot or human to move freely about the building while collecting the RSS. The remaining question is how to combine both sources of information to produce a robust signal map, i.e., a collection of Radio-Frequency (RF) fingerprint cells, each cell describing some statistics on the RSS coming from multiple access points (AP) and indexed by a spatial location within the building. In this paper, we explore several strategies for building optimal signal maps from RSS collected along robotic or pedestrian trajectories. We compare 1) the use of naïve fixed-size grids, 2) clustering the RSS top-down into cells of increasingly smaller size using decision trees and 3) iteratively aggregating smaller-size fingerprint cells based on the probability of pair-wise confusion between fingerprint cells. We study the trade-off between the spatial extent of RSS fingerprint cells and the differentiability of the RSS distribution in each cell, as well as their impact on localization accuracy. We evaluate these strategies on a large dataset of WiFi RSS collected by an autonomous robot exploring a large multi-floor office environment and investigate which signal map building strategies are optimal for localization accuracy.