Optimizing radio map for WLAN fingerprinting

In this paper, questions related to design of WLAN radio map for fingerprinting based positioning were investigated. The experiment results show that with histogram based algorithms, the positioning accuracy improves as the number of histogram bins increases until the number of bins reaches eight. With the number of bins lower than this, the uneven bin distribution with separate bin for missing samples gives better accuracy than even bin widths. If the calibration data contains samples from several measurement directions, it is beneficial to combine them into one fingerprint, as this decreases the size of the radio map and gives at least the same accuracy as having separate fingerprints for different measurement directions. In the experiments, the combination of the signals from correlating sources before the computation of the radio map and position estimate decreases the positioning accuracy only by 1–2 m, but decreases significantly the size of the radio map. The best method for signal combinations depends on the bin configuration and position estimation algorithm.