Thumbnail Image


Publication or External Link






In this dissertation, ROLAX location determination system in 4G networks is presented. ROLAX provides two primary solutions for the location determination in the 4G networks. First, it provides techniques to detect the error-prone wireless conditions in geometric approaches of Time of Arrival (ToA) and Time Difference of Arrival (TDoA). ROLAX provides techniques for a Mobile Station (MS) to determine the Dominant Line-of-Sight Path (DLP) condition given the measurements of the downlink signals from the Base Station (BS). Second, robust RF fingerprinting techniques for the 4G networks are designed. The causes for the signal measurement variation are identified, and the system is designed taking those into account, leading to a significant improvement in accuracy.

ROLAX is organized in two phases: offline and online phases. During the offline phase, the radiomap is constructed by wardriving. In order to provide the portability of the techniques, standard radio measurements such as Received Signal Strength Indication (RSSI) and Carrier to Interference Noise Ratio(CINR) are used in constructing the radiomap. During the online phase, a MS performs the DLP condition test for each BS it can observe. If the number of the BSs under DLP is small, the MS attempts to determine its location by using the RF fingerprinting.

In ROLAX, the DLP condition is determined from the RSSI, CINR, and RTD (Round Trip Delay) measurements. Features generated from the RSSI difference between two antennas of the MS were also used. The features, including the variance, the level crossing rate, the correlation between the RSSI and RTD, and Kullback-Leibler Divergence, were successfully used in detecting the DLP condition. We note that, compared to using a single feature, appropriately combined multiple features lead to a very accurate DLP condition detection. A number of pattern matching techniques are evaluated for the purpose of the DLP condition detection. Artificial neural networks, instance-based learning, and Rotation Forest are particularly used in the DLP detection. When the Rotation Forest is used, a detection accuracy of 94.8% was achieved in the live 4G networks. It has been noted that features designed in the DLP detection can be useful in the RF fingerprinting.

In ROLAX, in addition to the DLP detection features, mean of RSSI and mean of CINR are used to create unique RF fingerprints. ROLAX RF fingerprinting techniques include: (1) a number of gridding techniques, including overlapped gridding; (2) an automatic radiomap generation technique by the Delaunay triangulation-based interpolation; (3) the filtering of measurements based upon the power-capture relationship between BSs; and (4) algorithms dealing with the missing data.

In this work, software was developed using the interfaces provided by Beceem/Broadcom chip-set based software. Signals were collected from both the home network (MAXWell 4G network) and the foreign network (Clear 4G network). By combining the techniques in ROLAX, a distance error in the order of 4 meters was achieved in the live 4G networks.