1. 1 An Algorithmic and Systematic Approach for Improving Robustness of TOA-based Localization Yongcai Wang, Lei Song Institute for Interdisciplinary Information Sciences (IIIS), Tsinghua University, Beijing, China in EUC2013, Nov.13, 2013

2. 2 Positioning by Time of Arrival (TOA) of ultrasound Advantages:  Accurate: Centi-meter level positioning accuracy  Safe: user-imperceptible  Low cost: US transducers are cheap (around 10 RMBs). Context ultrasoundRF TOA1 0 t ultrasoundRF TOA3 0 t ultrasoundRF TOA2 0 t

3. 3 Challenges ： Sensitive to Environment Challenge 1: Non-Line-of-Sight Impacts 1.Non-Line-of-Sight (NLOS) paths, caused by furniture, doors, moving people may lead to large positioning error.  NLOS is generally inevitable

4. 4 Challenges (2): Miss of Synchronization 2.Background RF signal from WiFi, microwave oven, etc may collide the synchronization RF signal, leading to positioning error. T1T1 T2T2 TOA 1 TOA 2 Interference from background RF is also inevitable R1R1 R2R2

5. 5 Our Work We show NLOS outlier detection problem is NP-hard. We developed COFFEE, an iterative clustering, voting and filtering algorithm to detect NLOS distances. First-Falling-Edge robust time synchronization A prototype of Dragon system which implements COFFEE and First-Falling-Edge time synchronization.

6. 6 1. NLOS Outlier Detection Problem N beacons with known coordinates N beacons take N distance measurements: m of the distances are NLOS outliers: m<N/2 NLOS detection Problem: To detect the m outliers among the N distances.

7. 7 Conventional Approach: 1. Geometrical method Outlier detection by Triangular Inequality [zhao2008]. Graph embeddability and rigidity. [Jian 2010]  High computation cost.  May fail to detect outlier when normal ranging distances have noises. Coarse-grained, may fail to detect the outlier

8. 8 Conventional Approaches 2: Least Trimmed Square Method [Pireto2009] is a subset of distance measurements Enumerate D s to find the set with the minimum positioning residue. N distances can generate at most O(2 N ) subsets. Searching all sets needs high computation cost Problem: Method:

9. 9 Our Approach: Clustering and Filtering (COFFEE)  Distance measurements Assign doubting weight Potential positions Density-based clustering Core cluster Position outlier 1 1 1 2 1 Filter outlier distance Delete outlier positions COFFEE: Conduct clustering and filtering iteratively on the bipartite graph

10. 10 N=15 m=3 N=15 m=3

11. 11 Algorithm Properties Convergence speed  Detect m distance outliers in m iterations. Complexity  N 4 logN

12. 12 COFFEE performs best in positioning accuracy improvement Coffee provides the best accuracy

13. 13 COFFEE is Robust to the number of the distance outliers  Positioning error of COFFEE is small until m=8

14. 14 2. First-Falling-Edge Time Synchronization A hardware type design  Using a sync-line to connect all the receivers (beacons) All receivers can be synchronized only if one receiver detects the synchronization RF signal.

15. 15 Robust Synchronization By First-Falling-Edge When probability of missing RF signal is 1% It helps more receivers to provide correct ranging, which improves the positioning accuracy.

16. 16 3. Prototype System (Dragon)

17. 17 Deployment of Dragon System

18. 18 Experiment Results Without sync-lineWith sync-line Positioning failure probability

19. 19 Indoor Positioning Error

20. 20 Reason of Positioning Error in Dragon The ranging error caused by angle

21. 21 Conclusion We proposed COFFEE, an efficient clustering and filtering algorithm for NLOS outlier detection.  Accurate  Low complexity  Robust to the number of distance outliers. First-Falling-Edge time synchronization improves the time synchronization probability effectively. We developed a prototype of Dragon system, which verified the effectiveness of above designs.

22. 22 Thanks a lot For your patience Visit my homepage for further information http://iiis.tsinghua.edu.cn/~yongcai