Locating Sensors in the Forest: A Case Study in GreenOrbs Tsung-Yun Cheng 20120611 Outline • • • • • Introduction Preliminary experiments System design Performance evaluation Discussion Introduction • GreenOrbs – http://www.greenorbs.org/ – one of the world’s largest wireless sensor networks – monitor the forest condition • • • • Temperature Humidity Illumination Carbon dioxide Introduction • GreenOrbs – Potential application • Canopy closure calculation • Climate change observation • Search and rescue in the forest – need the location information of sensor nodes – Environmental noise • • • • Illustrates Temperature Humidity canopy closure Introduction • Environmental aware localization (EARL) – Joint Neighbor Distance (JND) • measure the distance between sensor nodes – Neighbor node relation verification technology • identifies nodes with good location accuracy – Two-phases location calibration • Rectify the node locations – Implementation • 20% better than existing work Preliminary experiments I Preliminary experiments I • Consider the relationship between RSSI and three parameters in the forest – Temperature (0.0613) – humidity (0.0907) – Illumination (0.1325) • The relationship is quite hard to capture – Taking temperature, humidity, illumination and RSSI into account, it is quite difficult to estimate the distance between nodes Preliminary experiments II • Experiment in different environments – Grass, Woods, Forest • Exam the RSSI value in different power level – Put two nodes in three environments – Distance = 10 meter • Exam the reachability of RSSI – One anchor nodes in the center – 10 nodes are deployed around in every 5 meters, ranging from 5 meters to 50 meters Preliminary experiments II Preliminary experiments II Preliminary experiments II • Exam the RSSI value in different power level – The variance is large – E.g., Figure 5(a) the Grass case: • -40dBm to -35dBm when the power level is 4 • -29dBm could range from the 9th to 14th RF power level • Exam the reachability of RSSI – When the RF power level increases, anchor node could reach more neighboring nodes – In the forest, many curves share the similar RSSI according to the different power level • After checking the location, they are in the same area Preliminary experiments • RSSI is quite susceptible to environment – the distance cannot be well computed directly • RSSI sensing results just can be used as an indicator for the relative “near-far” relationship System design • Determine Neighbor relationship – A near-far ordering relationship • • • • Obtained by RF power scanning Neighboring sequence e.g., {G, C, E, B, F, D} One-hop neighbors – Doesn’t show: • how far the distance • direction System design • Joint Neighbor Distance (JND) – estimate the distance of each pair of nodes – – = Neighbor Count of Xj with respect to Xi – E.g.: • NC(A, B) = 4 • NC(B, A) = 5 • JND(A,B) = 7 System design • Calculate the coordinate using JND – relative distance turns to the smallest accumulated JND – Choose some landmark nodes • Known position • Calculate JND-unit – Compute the distance to the landmark nodes • • Trilateration by least square estimation System design • Testbed in the wood – 50 nodes, 4 landmark nodes – 1.3 meter above the ground System design • Testbed in the wood – The boundary nodes have smaller neighbor nodes – Also the nodes near the obstacles System design • Calibration – Empirically, nodes with small neighbor count will lead to the great error of locations, e.g., boundary nodes – When RF power level increases, the transmission radius increases none-linearly when obstacles exist • more than one neighboring nodes may be added into the neighbor sequence at same level System design • Calibration for boundary nodes – Detection • Select every nodes as root to establish a tree • Leafs is the possible boundary nodes • Pi larger than certain threshold => boundary node – Calibrated neighbor count • Virtual NC: • j is the nearest neighbor • CNC = Max{VNC, NC} System design • Calibration for good nodes & bad nodes – Get correct neighbor sequence: Two-step process • Group the neighbor nodes according to the appearance of the RF level e.g., ((A), (G), (B, C, E), (D, F)) • In the same group, RSSI value is measured to get a precise neighbor nodes sequence – Get a JND scheme sequence • Use JND localization scheme to compute the distance – Compare the two sequence System design • Calibration for good nodes & bad nodes – comparison • Longest common subsequence • good nodes > bad nodes • Set a certain threshold – Calibrate bad nodes • Don’t mention… System design • Calibration for good nodes & bad nodes – Calibrate good nodes: Reverse-localization • Iteratively choose four of good nodes as the landmark nodes, compute the location of four original landmark nodes • Find the four goods nodes with minimum error • calibrate the location of good nodes using 8 landmarks nodes Performance Evaluation • Previous testbed (in the woods) – Compare to other works: DV-Hop, CDL – Mean error • EARL: 5m, CDL: 9m, DV-Hop: Large (~=18m) Performance Evaluation • Testbed in the forest – 230 nodes, 4 landmark nodes – Mean error • EARL: 9m, CDL: 12m, DV-Hop: Large(~=20m) Performance Evaluation • more landmark nodes help improve the localization accuracy Discussion • Network is highly affected by the complex environment factors • Environmental aware localization scheme ,EARL, takes the joint neighbor count to measure the distance between two nodes and compute the location of nodes • EARL outperforms existing approaches in terms high accuracy and efficiency Discussion • Writing – Structure is weird • Content – Don’t mention how they come up with these approaches and why they adopt these methods – Don’t mention how to calibrate the bad nodes Q&A ~ Thank you for your attention ~