Indoor Localization based on 5G Positioning Reference Signal
No Thumbnail Available
Sähkötekniikan korkeakoulu | Master's thesis
CCIS - Master’s Programme in Computer, Communication and Information Sciences (TS2013)
AbstractIndoor localization is a challenging task due to signal propagating with various reflections, diffraction, and fading. The complex environment deteriorates the performance of geometric localization methods. A higher positioning accuracy in three dimensions is an important enabler for smart industry 4.0 technologies. This thesis addresses localization in an indoor factory environment and develops a system-level simulator based on 5G positioning reference signal. Three localization methods are considered, classical geometry localization based on time difference of arrival (TDoA) measurements from multiple base stations (BSs), and the location is estimated by the Gauss-Newton algorithm. Two data-driven approaches are considered: a maximum a posteriori (MAP) estimator and k-nearest neighbor (KNN) fingerprinting. The MAP estimator is used for localization in non-line of sight (NLoS) environments. The distance bias due to the NLoS environment is modeled by a Gaussian mixture (GM) distribution. The GM parameters are estimated from an offline data set, containing physical locations and time of arrivals (ToAs) from multiple BSs. For KNN, an offline data set is created, and different channel state information (CSI) features are considered. The geometric localization shows satisfactory localization accuracies only in LoS scenarios but hardly convergences in NLoS. At the online phase, the MAP estimator shows poor performance due to the wide spread of the NLoS bias in indoor factory scenarios. The KNN fingerprinting shows better performance in both LoS and NLoS scenarios at the online phase. The performance of different algorithms is evaluated in terms of 50% and 80% percentiles of the localization error.
Thesis advisorAl-Tous, Hanan
indoor localization, LoS and NLoS, TDoA, geometric localization, data-driven, KNN fingerprinting