Locating Sensors in the Wild: Pursuit of Ranging Quality Wei Xi, Yuan He, Yunhao Liu, Jizhong Zhao, Lufeng Mo, Zheng Yang, Jiliang Wang, Xiangyang Li
2 Outline Motivation Observation on GreenOrbs Design of CDL Evaluation Ongoing work of GreenOrbs
3 GreenOrbs
4 Existing approaches (1) GPS Problems with tree covers Range-Based Approaches –TOA, TDOA, AOA Require extra hardware support Expensive in manufactory cost and energy consumption –RSSI-based Based on the log-normal shadowing model Inaccurate due to channel noise, interference, attenuation, reflection, and environmental dynamics
5 Existing approaches (2) Range-Free Approaches –Rely on connectivity measurements –The accuracy is affected by node density and network conditions RSD (SenSys ’ 09) –Regulated signature distance SISR (MobiCom ’ 09) –Merely differentiate good and bad links DV-Hop
6 Outline Motivation Observation on GreenOrbs Design of CDL Evaluation Ongoing work of GreenOrbs
7 Two-folded ranging quality 1. Irregular 2. Dynamic 3. Susceptible to the environment 4. Ubiquitous diverse errors Node location accuracy & range measurement accuracy Fine-grained differentiation is necessary!
8 Outline Motivation Observation on GreenOrbs Design of CDL Evaluation Ongoing work of GreenOrbs
9 1.Range-free localization: virtual-hop 2.Local filtration: two types of matching 3.Calibration: weighted robust estimation Design of CDL
10 DV-Hop r When non-uniform deployment is present, nodes with equal hop- counts often have different distances to the landmark(s).
11 Virtual-hop localization For a node, its number of previous-hop or next-hop neighbors reflects the relative distance from the node to its parent node.
12 Virtual-hop vs. DV-hop Compared with DV-hop, Virtual-hop reduces the localization errors by 10%~99%.
13 Local filtration (1) Indiscriminate calibration probably reduces localization accuracy.
14 Local filtration (2) Bad nodes exhibit more mismatches Neighborhood hop-count matching –Compare the real hop-distance with the one calculated using estimated node coordinates (a) A good node with one bad neighbor (b) A bad node with six good neighbors
15 Local filtration (3) Neighborhood sequence matching NANA BCDEFG SASA SA’SA’ NANA BCDEFG SASA SA’SA’ Compare RSSI sequence with estimated distance sequence Matching degree
16 According to the matching degree, we sort nodes into three classes Good Bad Undetermined Local filtration (4)
17 The basic objective function in LSE RQAC –Weight good nodes by good neighbors –Differentiates links with different ranging qualities Ranging-Quality Aware Calibration
18 Outline Motivation Observation on GreenOrbs Design of CDL Evaluation Ongoing work of GreenOrbs
19 Evaluation Setup –Experiments 100 GreenOrbs nodes (4 landmarks) –Simulations Randomly deploy 200~1000 nodes A 500*500m 2 square region Transmission range: 30m
20 Comparison AlgorithmError (m) DV-HOP8.7 MDS-MAP5.9 SISR5.6 CDL2.9
21 CDF of localization errors
22 Efficiency of iteration The number of good nodes quickly increases as iterations go on.
23 Humidity has a positive impact on the localization accuracy of all the four approaches. Impact of environmental factors
24 Increasing node density or landmarks yields better localization accuracy. Impact of system parameters
25 Summary of CDL GreenOrbs –A most challenging scenario of WSN localization Our belief: ranging quality is two-folded –The location accuracy of the reference nodes –The accuracy of range measurements Combined and Differentiated Localization –VH localization addresses non-uniform deployment –Filtration picks good nodes with good location accuracy –RQAC emphasizes the contribution of the best range measurements
26 Outline Motivation Observation on GreenOrbs Design of CDL Evaluation Ongoing work of GreenOrbs
27 Ongoing work of GreenOrbs New applications –Carbon sink/emissions measurements –Forest fire risk prediction Research on WSN management
28 Thanks! Q & A