Robust Wi-Fi Assisted GNSS Positioning in Urban Canyons

S. Ramakrishnan, D.W. Waters, J. Balakrishnan

Abstract:	Robust Wi-Fi Assisted GNSS Positioning in Urban Canyons Recently wifi based positioning has drawn a lot of attention due to its potential ability to improve GNSS performance in difficult urban canyon environments. Primarily there have been two approaches to wifi positioning – either multilateration based, which converts Received Signal Strength (RSS) measurements from access points (APs) to a distance and then solves for the user location or finger-printing based [1], which creates a database of Access Point (AP) and RSS observed at every point in the region of interest. The multilateration based approach needs exact AP locations while the finger printing based approach needs large amounts of data to be stored and accessed to estimate the user location, which makes it unsuitable for a low cost implementation. Furthermore, the finger printing based approach is difficult to keep up-to-date as new APs may be added or old APs may be removed or there may be permanent changes in the channel due to addition/removal of blocking artifacts like furniture, walls etc. In practical scenarios it is more realistic to use a database of AP location estimates, which are typically built through wardriving or crowd-sourcing. AP location estimation has been well studied [4] but typically a known channel model is assumed resulting in in-correct AP locations in real-life scenarios. One problem in this case is the presence of mobile APs, which may be reported from multiple locations, in some cases even from different cities. This causes the AP location estimates to be in significant error and can cause huge errors in estimated user location. Further since the AP database is built based on reported GNSS locations which may themselves be in error, the AP location estimate can have moderate errors of the order of ~ 100m. In this paper, we first present an outdoor Wi-Fi positioning algorithm that is robust to the presence of these outlier APs and then a robust blending of the Wi-Fi positions with GNSS. Our algorithm builds on earlier work such as [3] in that we use a weighted centroid algorithm as the building block for outlier detection. We detect outliers using a simple recursive weighted centroid algorithm. Furthermore, we pass such instantaneous Wi-Fi positions through a Kalman filter to extract the user velocity and user location. The user velocity estimate allows us to calculate position predictions which are also used to reject outlier APs. This enables a robust Wi-Fi positioning algorithm even in areas with low AP density. Also the Kalman filtering reduces position scatter and improves the accuracy and availability of Wi-Fi position estimates even in low AP density regions. We detect vehicular and pedestrian scenarios autonomously and adapt the Kalman Filter bandwidth appropriately thereby resulting in a good performance under both conditions. Results from multiple urban canyon tests show that the performance of the Wifi-only positioning algorithm with real life AP database is robust. CEP95 improves by about 20-30m due to the Kalman filtering and the overall solution has CEP95 of the order of 50-70m in most scenarios. GNSS and Wi-Fi positioning systems provide two independent methods to compute position. These two lend themselves well into a loosely-coupled blending architecture since the GNSS measurements are in the pseudorange domain and Wi-Fi measurements are in the Received signal strength domain. While there have been some work on directly combining wifi and GNSS measurements [2], they assume perfect AP location and RSSI – distance model information, both of which are unrealizable in practice. In urban canyons, GNSS based positioning can occasionally have large errors even when it is available and blending Wi-Fi positions with GNSS positions can help mitigate this error. A simple weighted combining of Wi-Fi and GNSS position estimates would not be effective for multiple reasons. First, a weighted combination is non-optimal because the Wi-Fi and GNSS positions are not independent over time. Second, there may be dead-zones where both the GNSS and Wi-Fi position estimates are always poor. Third, often GNSS deviations become large over time if left uncorrected. Using Wi-Fi position estimates to prevent this problem is an important part of blending GNSS and Wi-Fi position estimates. A weighted combining of Wi-Fi and GNSS positions does not improve this important use case. To address these shortcomings, we propose the use of three approaches to blending. First, is constrained blending where the Wi-Fi position estimate is used to correct the GNSS position without moving it outside GNSS’s uncertainty region. This ensures that bad AP locations in the data-base do not cause large errors in the blended position. Second, selective blending, meaning the Wi-Fi positions are used for blending only when GNSS positioning needs help. For example, when the number of Satellites is low or the GNSS position uncertainty is high or when we detect that the user is in a deep urban canyon. Third, feedback blending – the GNSS position filter is corrected using the blended position estimate. This ensures that GNSS position errors are corrected early on thereby preventing large drifts in GNSS positions. Extensive field results from urban canyons in Chicago, Dallas and Toronto of the proposed solution show a significant performance improvement over GNSS only positioning as well as a weighted wifi and GNSS blending solution. In some difficult scenarios the CEP95 improved from 100-150 m for GPS only positioning to 50-70 m for our proposed GPS + wifi blended solution. References: 1. Binghao Li, et.al , “On outdoor positioning with wifi”, Journal of Global Positioning Systems, 2008, Vol 7 2. Zirari, Soumaya; Canalda, P.; Spies, F., "WiFi GPS based combined positioning algorithm," Wireless Communications, Networking and Information Security (WCNIS), 2010 IEEE International Conference. 3. Guo, H., Chen, Q., Yu, M., Siddharth, S., "Weighted Centroid Localization Algorithm Based on ZigBee for Indoor Positioning," Proceedings of the 2011 International Technical Meeting of The Institute of Navigation, San Diego, CA, 4. Thompson, R.J.R., Cetin, E., Dempster, A.G., "Unknown Source Localization using RSS in Open Areas in the Presence of Ground Reflections," Proceedings of IEEE/ION PLANS 2012, Myrtle Beach, South Carolina.
Published in:	Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013) September 16 - 20, 2013 Nashville Convention Center, Nashville, Tennessee Nashville, TN
Pages:	2058 - 2065
Cite this article:	Ramakrishnan, S., Waters, D.W., Balakrishnan, J., "Robust Wi-Fi Assisted GNSS Positioning in Urban Canyons," Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013), Nashville, TN, September 2013, pp. 2058-2065.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In