Document

Report
Empirical-based Analysis of a Cooperative
Location-Sensing System
Maria Papadopouli 1,2
K. Vandikas1 L. Kriara1,2 T. Papakonstantinou1 A. Katranidou1 H. Baltzakis1
1
Institute of Computer Science, Foundation for Research & Technology-Hellas (FORTH)
2 Department
of Computer Science, University of Crete
http://www.ics.forth.gr/mobile/
This research was partially supported by EU with a Marie Curie IRG and the Greek General
Secretariat for Research and Technology.
1
2
Overview






Motivation
Taxonomy of location-sensing systems
Collaborative Location Sensing (CLS)
Performance analysis
Conclusions
Future work
3
Motivation

Emergence of location-based services in several areas
 transportation & entertainment industries
 emergency situations
 assistive technology
→ Location-sensing is critical for the support of location-based
services
4
Taxonomy of location-sensing systems







Modalities
Dependence on & use of specialized infrastructure & hardware
 Radio (Radar, Ubisense, Ekahau),
Position
and
coordination
 Infrared
(Active
Badge) system description
Cost,
accuracy
& precision requirements
 Ultrasonic
(Cricket)
 Bluetooth
Localized
or remote computations
 Vision
(EasyLiving) classification or recognition
Device
identification,
 Physical contact with pressure (smart floor) or touch sensors
Models & algorithms for estimating distances, orientation &
position
5
Cooperative Location-Sensing (CLS)
Enables a device to determine its location in self-organizing
manner using the peer-to-peer paradigm
 Employs a grid-based representation of the physical space
→ can incorporate contextual information to improve its estimates
 Uses a probabilistic-based framework
 Each cell of the grid has a value that indicates likelihood that
the local device is in that cell
 These values are computed iteratively using distance
between peers and position predictions

6
Classifying CLS
Modalities
 Dependence on & use of specialized infrastructure & hardware
Radio and/or Bluetooth
 Position
and
coordination
system description
Can
beforextended
to incorporate
other
type of modalities
 No
need
specialized
hardware
or
infrastructure
 Cost,
accuracy
precision
requirements
 Can
use only &
IEEE802.11
APs,
if necessary
Grid
representation
of the space
Objective:or
0.5remote
to
2.5 m
(90%)
 Transformation
Localized
computations
to/from
any
coordination system
Computations
be grid
performed
remotely
or at the device
a cellcan
in the
 Position:
Device identification,
classification
or recognition
depending
on
the device
capabilities
 Does&not
perform
any
these functionalities
 Models
algorithms
for of
estimating
distances, orientation &
position
 Statistical analysis and particle filters on signal strength
measurements collected from packets exchanged with other
peers

7
Example of voting process (1/2)

Accumulation of votes on grid cells of host at different time steps
8
Example of voting process (2/2)
Most likely position
Peers A, B, C have positioned themselves
Host A
x
x
x
Host C votes
Host B votes
9
Voting algorithm
1.
2.
3.
4.
5.
6.
Initialize the values of the cells in the grid of the local device
Gather position information from peers
Record measurements from these received messages
Transform this information to probability of being at a
certain cell of its local grid
Add this probability to the existing value that this cell had
from previous steps
Assess if the maximal value of the cells in the grid is
sufficient high to indicate the position of the device
10
Example of training & run-time signature comparison
Signal-strength measurements per AP
AP1
cell
AP2
Training-phase signature
comparison
weight of that cell
Run-time signature
11
Position estimation (at peer A)
Landmarks vote
1.
2.
3.
4.
Initialize the values of the cells in the grid of the local device
Training phase: Build a signal-strength map of the space
(training-phase signatures)
Run-time phase: Build signal-strength signature of the current
position
Compare the run-time and training phase signatures
Non-landmark peers vote
5.
For each new peer that sends its position estimation (e.g., peer B)
I.
Position B on the local grid of A based on B’s estimation
II.
Determine their distance based on signal-strength signature
III.
Infer likely positions of A
IV.
Update the value of the cells accordingly
6.
12
Assess maximal weight of the cells, accept or reject the solution
Signature based on confidence interval
of signal-strength values
Weight of cell c assigned as:
total number of APs
training confidence interval of i-th AP
run-time confidence interval of i-th AP
13
Example of
confidence interval-based comparison
Signal-strength measurements per AP
[ T-, T+ ] confidence interval
based on signal strength
measurements from an AP
cell
T+1
T1-
T+2
T2-
AP1
AP2
T+1
T1- T+2
T2- …
Training-phase signature
weight of that cell
R1+ R1- R+ R2
2
…
Run-time signature
14
Distance estimation between two peers
entries of training set
ith distance from training set
confidence interval of the run-time measurements
confidence interval of the i-th entry
in the training set
15
Signature based on percentiles of the
signal-strength values
number of percentiles
jth percentile of ith cell in training set
samples in training set
jth run-time percentile
16
Particle filter-based framework
step 1
for L = 1, … , P
(L-th particle)
Transition:
Draw new sample xk(L) , P( xk(L) | xk-1(L) )
Compute weight wk(L) of xk(L), wk(L) = wk-1(L)* P( yk | xk(L) ),
where yk measurement vector: signal strength values
end loop
Normalize weights
Resample
Goto step 1
17
Performance evaluation


Performance analysis of CLS via simulations [percom’04]
Empirical-based measurements in different areas
 Various criteria for comparing the training phase and run-time
signatures
 Particle-filter model
 Impact of the number of signal strength measurements
 Impact of the number of APs and peers
 CLS vs. Ekahau
18
Testbed description




Area 7m x 12m @ Telecommunication and Networks Lab (in FORTH)
Each cell of 50cm x 50cm
Total 11 IEEE802.11 APs in the area
3.5 APs, on average @ any cell
19
CLS variations
features
Criteria
variations
CLS
confidence interval
CLS-p2p
confidence interval
CLS-percentiles
percentiles
CLS-particles
particle-filter
Only APs Peers Distance
vote
vote
computed
20
Similarities between CLS & Ekahau v3.1



Use IEEE802.11 infrastructure
Create map with callibration data
Compare trainning & run-time measurements
21
Ekahau vs. CLS
no peers
only APs participate
additional measurements
Percentiles capture more information
about the distribution of signal strength
22
Impact of number of APs
One AP off
23
Impact of peers
One extra peer
24
Use of Bluetooth instead of IEEE802.11
25
Conclusions

The density of landmarks and peers has a dominant
impact on positioning

Experiments were repeated at the lab in FORTH and in a
conference room @ ACM Mobicom


median location error 1.8 m
Incorporation of Bluetooth measurements to improve
performance

median location error 1.4 m
26
Discussion & future work (1/2)



Reduce training, management & calibration overhead
Easily detect changes of the environment
 density and movement of users or objects
 new/rogue APs
 Inaccurate information & measurements
Singular spectrum analysis of signal strength
 Distinguish the deterministic and noisy components
 Construct training and run-time signatures based on the
deterministic part
27
Discussion & future work (2/2)

Incorporate heuristics


about hotspot
areas, user presence and mobility
I
information, and topological information of the
area (e.g., existence of walls)
Experiment with other wireless technologies

Sensors, cameras, and RF tags
28
UNC/FORTH Archive

Online repository of models, tools, and traces

Packet header, SNMP, SYSLOG, signal quality
http://netserver.ics.forth.gr/datatraces/
 Free login/ password to access it
Joint effort of Mobile Computing Groups @ FORTH & UNC

[email protected]
Thank You! Any questions?
29
Multimedia Travel Journal Tool


Novel p2p location-based application for
visitors
Allow multimedia file sharing among mobile
users
30
31
32
Simulations
33
Simulations

Simulation setting (ns-2)





For low connectivity degree or few
landmarks


10 landmarks
90 stationary nodes
avg connectivity degree = 10
transmission range (R) = 20m
the location error is bad
For 10% or more landmarks and
connectivity degree of at least 7

the location error is reduced considerably
34
Bluetooth estimation experiments
35
Bluetooth-only estimation
validation experiments
36
Joint IEEE802.11 & Bluetooth estimation experiments
37
Joint IEEE802.11 & Bluetooth estimation experiments
impact of modalities - performance analysis
38
Modality comparison
39

similar documents