Experimental Assessment of Indoors UWB Range Error Mitigation Techniques
Harris Perakis, School of Rural and Surveying Engineering, National Technical University of Athens (NTUA), Greece; Andrea Masiero, interdepartmental Research Center of Geomatics, University of Padova, Italy; Jelena Gabela, Department of Electrical and Electronic Engineering, The University of Melbourne, Australia; Vassilis Gikas, School of Rural and Surveying Engineering, NTUA, Greece; Guenther Retscher, Department of Geodesy and Geoinformation, TU Wien-Vienna University of Technology, Austria; Charles Toth, Department of Civil, Environmental and Geodetic Engineering, The Ohio State University, USA; Salil Goel, Department of Civil Engineering, Indian Institute of Technology, India; Allison Kealy, Department of Geospatial Science, RMIT University, Australia; Zoltán Koppányi, Leica Geosystems, Switzerland; Wioleta Blaszczak-Bak, Institute of Geodesy of the University of Warmia and Mazury, Poland; Yan Li, Department of Electrical and Electronic Engineering, The University of Melbourne, Australia; Dorota Grejner-Brzezinska, College of Engineering, The Ohio State University, USA
Location: Pavilion Ballroom East
Alternate Number 2
Contemporary Radio Frequency (RF) technologies are increasingly becoming a standard for numerous indoor Location Based Services (LBS) that necessitate high positioning accuracy, availability and robustness. Ultra-Wide Band (UWB) systems are capable of providing positioning accuracy in the order of decimeters in unobstructed (Line of Sight - LoS) environments. However, in order to achieve a high-quality positioning solution in general, thorough range calibration is required due to the various error sources that may contaminate raw UWB range observables. This study deals with the range calibration problem of UWB technologies, and evaluates alternative range error calibration techniques and models based on a series of field data collected in different conditions during a joint experimental campaign of FIG Working Group 5.5 and IAG Working Group 4.1.1 on Multi-sensor Systems at The Ohio State University (OSU) in October 2017.
The data collection was performed using two UWB systems: the Time Domain® and the Pozyx proprietary ranging modules deployed on purpose built mounting platform for collecting ranges between a number of static (anchor) and mobile (rover) nodes. The experimental campaign took place within an indoor facility of the Department of Civil, Environmental and Geodetic Engineering, OSU, in which the locations of the anchor points and reference checkpoints, used for the validation of the positioning algorithm, were accurately surveyed before-hand. To support range error calibration and validation analyses, separate sessions were carried out for the collection of static ranging data for each checkpoint. Moreover, in order to further expand investigations and to assess the impact of the environmental conditions on range computations and the positioning solution, extensive datasets were collected for a group of four pedestrians moving in the study area, including both LoS and Non-LoS ranges. In particular, the datasets were collected in a realistic scenario, without restricting the area only to personnel carrying out the tests.
The ranging error mitigation techniques examined in this study can be grouped in two categories:
(a) a “standardized” error calibration technique utilizing static data collected prior to the kinematic positioning campaign, and (b) an “on-the-fly” error calibration technique using part of the kinematic data as a “learning” data-set and a different part as the validation data-set.
(a) This first approach requires consistent data collection at checkpoints distributed throughout the test area. Two error models are generated and implemented for range calibration. For the first model, the error correction values are estimated as a function of distance between UWB devices, while for the second one the correction values are estimated as a function of their 2D position.
(b) The second approach does not require separate data collection for the generation of the calibration models, and thus provides a much faster deployment, given that the trajectory includes certain stop-and-go sections at pre-surveyed checkpoints. The calibration models generated using this approach can be limited by the area covered by the “learning” dataset; however, it provides scale-up capabilities easing future model updates in a crowd-sensing fashion.
The two empirical approaches are evaluated both in a quantitative (i.e., with respect to their accuracy against the reference checkpoints locations) as well as qualitative manner (i.e., with respect to their functionality, efficiency and generalization standard). Our investigations suggest the existence of error-prone areas along the trajectory paths, due to either the weak anchor geometry or the presence of NLoS measurements. We focus on these areas for providing a systematic analysis of NLoS UWB range inaccuracies , and potential mitigation methods to support the overall positioning solution, firstly comparing approaches for the detection of NLoS ranges, and secondly for the scope of investigating suitable techniques for compensating their effects. Preliminary examinations undertaken with subsets of the collected data reveal an overall improvement in the positioning solution in the order of 40%-80% depending on operational conditions and observation geometry.