my Master 4

Location-sensing using the IEEE 802.11 Infrastructure and the Peer-to-peer Paradigm for Mobile Computing Applications: 

Location-sensing using the IEEE 802.11 Infrastructure and the Peer-to-peer Paradigm for Mobile Computing Applications Anastasia Katranidou Supervisor: Maria Papadopouli Master Thesis, University of Crete andamp; FORTH-ICS, Hellas 20 February 2006

Overview: 

Overview Location-sensing Motivation Proposed system - CLS Evaluation of CLS Conclusions Future work

Pervasive computing century: 

Pervasive computing century Pervasive computing enhances computer use by making many computers available throughout the physical environment but effectively invisible to the user

Why is location-sensing important ?: 

Why is location-sensing important ? Navigation systems Locating people andamp; objects Wireless routing Smart spaces Supporting location-based applications transportation industry medical community security entertainment industry emergency situations

Location-sensing properties: 

Location-sensing properties Metric (signal strength, AoA, ToA, TDoA) Techniques (triangulation, proximity, scene analysis) Multiple modalities (RF, ultrasound, infrared) Limitations andamp; dependencies (e.g., infrastructure vs. ad-hoc) Localized or remote computation Physical vs. symbolic location Absolute vs. relative location Scalability Cost Specialized hardware Privacy

Related work: 

Related work

Motivation: 

Motivation Build a location-sensing system for mobile computing applications that can provide position estimates: using the available communication infrastructure within a few meters accuracy without the need of specialized hardware and extensive training operating on indoors and outdoors environments Use peer-to-peer paradigm knowledge of the environment and mobility

Design goals: 

Design goals Robust to tolerate network failures, disconnections, delays due to host mobility Extensible to incorporate application-dependent semantics or external information (e.g., floorplan, signal strength map) Computationally inexpensive Scalable Use of cooperation of the devices and information sharing No need for extensive training and specialized hardware Suitable for indoor and outdoor environments

Thesis: 

Thesis Implementation of the Cooperative Location System (CLS) Extension of the CLS design signal strength map information about the environment (e.g., floorplan) heuristics based on confidence intervals Extensive performance analysis range error density of hosts mobility Empirical study of the range error in FORTH-ICS

Cooperative Location System (CLS): 

Cooperative Location System (CLS) Communication Protocol Each host estimates its distance from neighboring peers refines its estimations iteratively as it receives new positioning information from peers Voting algorithm accumulates and evaluates the received positioning information Grid-representation of the terrain

Communication protocol: 

CLS beacon neighbor discovery protocol with single-hop broadcast beacons respond to beacons with positioning information (positioning entry andamp; SS) CLS entry set of information (positioning entry andamp; distance estimation) that a host maintains for a neighboring host CLS update messages dissemination of CLS entries CLS table all the received CLS entries CLS table of host u Positioning entry Distance estimation CLS entries Communication protocol

Voting algorithm: 

Voting algorithm Grid for host u (unknown position) Corresponds to the terrain Peer A has positioned itself Positive votes from peer A The value of a cell = sum of the accumulated votes The higher the value of a cell, the more hosts agree that this cell is likely position of the host Peer B has positioned itself Positive votes from peer B Negative vote from peer C

Voting algorithm termination: 

Voting algorithm termination Set of cells with maximal values defines possible position A cell is a possible position If the num of votes in a cell is above ST and the num of cells with max value below LECT terminate the iteration process report the centroid of the set as the host position u

Evaluation of CLS: 

Evaluation of CLS Impact of several parameters on the accuracy ST and LECT thresholds range error density of hosts and landmarks Simulation testbed 100x100 square units in size Randomly placed nodes (10 landmarks + 90 nodes) in the terrain Location andamp; range errors as % of the transmission range (R=20 m)

Impact of range error : 

Impact of range error avg connectivity degree = 10 avg connectivity degree = 12

Impact of connectivity degree & percentage of landmarks: 

Impact of connectivity degree andamp; percentage of landmarks For low connectivity degree or few landmarks the location error is bad For 10% or more landmarks and connectivity degree of at least 7 the location error is reduced considerably 5% range error

Extension of CLS: 

Extension of CLS Incorporation of: signal strength map information about the environment (e.g., floorplan) confidence intervals topological information pedestrian speed

Signal strength map: 

Signal strength map Training phase: each cell andamp; every AP 60 measured SS values 1 signal strength (SS) value / sec 95% - confidence intervals Estimation phase: SS measurements in 45 cells if LBi[c] ≤ ŝi ≤ UBi[c] cell c accumulates vote from APi final position: centroid of cells with maximal values

CLS with signal strength map: 

CLS with signal strength map 95% - confidence intervals no CLS: 80% hosts ≤ 2 m extended CLS: 80% hosts ≤ 1 m

Impact of mobility: 

Impact of mobility Movement paths Speed Frequency of CLS runs Simulation setting 10 landmarks, 10 mobile and 80 stationary nodes transmission range (R) = 20 m range error = 10% R

Impact of movement paths: 

Impact of movement paths Simulation setting 10 different scenarios max speed = 2 m/s time= 100 sec Mean location error [%R] Simulation time (sec)

Impact of the speed: 

Impact of the speed Simulation setting 6 times the same scenario fixed initial and destination position of each node at each run time = 100 sec Simulation time (sec) location error [%R]

Impact of the frequency of CLS runs: 

Impact of the frequency of CLS runs Simulation setting 6 times the same scenario every 120, 60, 40, 30, 20 sec CLS run = 1, 2, 3, 4, 6 times speed = 2 m/s time = 120 sec Tradeoff accuracy vs. overhead message exchanges computations Simulation time (sec) location error [%R]

Evaluation of the extended CLS under mobility: 

Evaluation of the extended CLS under mobility Incorporation of: topological information signal strength map pedestrian speed Simulation setting 5 landmarks, 30 mobile and 15 stationary nodes speed = 1m/s R = 20 m range error = 10% R sim time = 120 sec CLS every 10 sec

Use of topological information: 

Use of topological information mobile node cannot: walk through walls enter in some forbidden areas negative weights CLS under mobility: 80% of hosts ≤ 90% location error (%R) CLS andamp; topological information: 80% of hosts ≤ 60% location error (%R)

Use of signal strength map: 

Use of signal strength map CLS andamp; topological information andamp; SS map: 80% of hosts ≤ 30% location error (%R)

Use of the pedestrian speed: 

Use of the pedestrian speed pedestrian speed: 1 m/s time instance t1: at point X after t sec: at any point of a disc centered at X with radius equal to t meters CLS andamp; topological information andamp; SS map andamp; pedestrian speed: 80% of hosts ≤ 13% location error (%R)

Estimation of Range Error in FORTH-ICS: 

Estimation of Range Error in FORTH-ICS 50x50 cells, 5 APs For each cell we took 60 SS values 95% confidence intervals (CI) for each cell c and the respective APs i Range errori[c] = max{|d(i,c) - d(i,c’)|},  c' such that: CIi[c]∩CIi[c’] ≠ Ø 90% cells ≤ 4 meters range error (10% R) Maximum range error due to the topology ≤ 9.4 meters

Conclusions: 

Conclusions Evaluation and extension of the CLS algorithm 80% of hosts ≤ 0.8 m estimations from peers give better accuracy than SS measurements Evaluation of CLS under mobility 80% of hosts ≤ 2.6 m great impact of frequency of CLS runs Comparison with related work static RADAR: 80% ≤ 4.5 m mobile RADAR: 80% ≤ 5 m

Future work: 

Future work Incorporate heterogeneous devices (e.g, RF tags, sensors) to enhance the accuracy Employ theoretical framework (e.g., particle filters) to support the grid-based voting algorithm and mobility models Provide guidelines for tuning the weight votes of hosts Use more sophisticated radio propagation model

Publications: 

Publications Under preparation for submission to the Mobile Computing and Communications Review (MC2R) journal

Location-sensing using the IEEE 802.11 Infrastructure and the Peer-to-peer Paradigm for Mobile Computing Applications: 

Location-sensing using the IEEE 802.11 Infrastructure and the Peer-to-peer Paradigm for Mobile Computing Applications Anastasia Katranidou Supervisor: Maria Papadopouli Master Thesis, University of Crete andamp; FORTH-ICS, Hellas 20 February 2006 THANK YOU!

APPENDIX: 

APPENDIX Appendix

RADAR vs. CLS: 

RADAR vs. CLS RADAR: 3 APs 90% hosts ≤ 6 m sampling density: 1 sample every 13.9 m2 Extended static CLS: 5 APs 90% hosts ≤ 2 m sampling density: 1 sample every 14.8 m2

Ladd et al. vs. CLS: 

Ladd et al. vs. CLS Static localization Ladd et al. 9 APs 77% of hosts ≤ 1.5 m Extended static CLS 5 APs 77% of hosts ≤ 0.8 m Static fusion Ladd et al. 9 APs 64% of hosts ≤ 1 m Extended mobile CLS 5 APs 45% of hosts ≤ 1 m

Savarese et al. vs. CLS: 

Savarese et al. vs. CLS Savarese et al. better with very small connectivity degree (4) or less than 5 landmarks Extended static CLS better with connectivity degree of at least 8 and 10% or more landmarks

Impact of ST and LECT thresholds: 

Impact of ST and LECT thresholds Terminate the iteration process ST: the num of votes in a cell must be above it LECT: the num of cells with max value must be below it LECT Host h defined solely from host g not acceptable: the possible cells of host h correspond to a ring Host h defined from host g and k 1 case: not acceptable 2 case: location errormax = √ [Dmax2 – (Dmin + e)2 ] Host h defined from host g, k and m Possible area: (2· ε +1)2 location errormax: √[(2· ε +1)2 / 2] ST eventually each host will receive votes from every landmark and every other host (CLS updates) wall_landmarks +wall_hosts

ST and LECT: 

ST and LECT Simulation setting 10 landmarks and 90 nodes avg connectivity degree = 10 range error = 10% R Best values ST = 800 LECT = 5

Interpolation methods: 

Interpolation methods Cubic interpolation Least squares Linear interpolation

Impact of connectivity degree under mobility: 

Impact of connectivity degree under mobility Simulation setting 5 landmarks 30 mobile nodes 15 stationary nodes Simulation setting 5 landmarks 5 mobile nodes 5 stationary nodes

Grid size: 

Grid size 100x100: reasonable choice

Message exchanges: 

Message exchanges

Movement example: 

Movement example Random waypoint model Max speed Pause time