Distance and Time Based Node Selection for Probabilistic Coverage in People-Centric Sensing

Distance and Time Based Node
Selection for Probabilistic Coverage
in People-Centric Sensing
Asaad Ahmed1, Keiichi Yasumoto1, Yukiko Yamauchi2, and Minoru Ito1
1 Nara Institute of Science and Technology (NAIST), Japan
2 Kyushu University, Japan
1

Motivation Scenario
2
 People Centric Sensing (PCS)
 People with mobile devices play a role of mobile
sensors and sense data in urban district
 Local torrential rain prediction
 Difficult to predict it from satellite photos (sudden, very small area)
 Weathernews company uses PCS to predict 1 hour before
occurrence
AoI
Company
1) Send a request to people in
Area of Interest (AoI)
2) Upload photos to
the server
3) Predict occurrence of local
torrential rain with collected photos
How to select mobile nodes for sensing/uploading data?
The preliminary solution is selecting all mobile nodes in AoI
Cost for communication and incentive fees paid to the
selected nodes will be high!

Challenges for PCS
3
Coverage difficulties in PCS
 Mobility of people is uncontrollable
 Selecting all mobile nodes in AoI induces large costs in
network and server (when # nodes in AoI is large)
 Sensing should be completed by deadline (e.g., 1 hour)
Challenges
 Predict mobile node’s future locations
 Minimize overall cost (network, server, incentive fees, etc) by
selecting a minimal set of mobile nodes that meet the
required coverage within specified time constraint

Outline
4
 Motivation Scenario
 Related work
 (a , T)-Coverage Problem
 Proposed Algorithms
 Performance Evaluation
 Conclusion

Related Work
5
SensorPlanet platform[8]
 Enables collection of sensor data in large/heterogeneous scale
 Establishes central repository for sharing the collected data
CarTel [9]
 Provides urban sensing information such as traffic conditions
 Based on car-mounted communication platform exploiting open
WiFi access points
CitySense [10]
 Provides a static sensor mesh for urban sensing data
Bubble-sensing [12]
Allows mobile users to affix bubble task at area of interest so
that they receive sensed data in delay tolerant manner

Related Work (Cont.)
6
Main purpose of existing works
 Information collection
Unconsidered issues
 Probabilistic nature of coverage in PCS
 Sensing coverage of a relatively wide area
 On-demand query with a time deadline
Overall cost for network, server, incentive fees, etc
Our contribution is to provide solutions to the
above issues

Outline
7
 Related work
 (a, T)-Coverage Problem
 Conclusion

Assumptions – service area
8
 A mobile user moves on a road network over service area A
There are multiple sensing locations on each road with
uniform spacing D
Road network is represented by a connected graph G (V, E )
V : the set of vertices
E : the set of edges
D

Assumptions – network & node mobility
9
 There is a server/cloud in the Internet that
executes node selection algorithm
 Each node can communicate with server from any
location of service area A via 3G network
 All nodes move on G according to the same
probabilistic model, where moving probability at
each sensing location of A is given by matrix P
 Time progresses discretely and nodes move from
one vertex to one of its neighbors in unit of time

(a, T )- coverage definition
 AoI is (a, T)-covered by the set of nodes U
⇔ probability that every sensing location in AoI is visited by
some nodes of U in time interval T is equal to/larger than a
AoI
10

Problem Formulation
11
AoI
q(α, T , SType)
Given
AoI as a set of sensing locations
A set of mobile nodes U in AoI
A query with
A required coverage ratio α
A specified interval T
Sensing type Stype
Objective
Finding a minimal set of mobile nodes U’
Subject to
AoI is (α, T )-covered

Outline
12
 Related work
 Conclusion

Basic Idea
13
 Problem: How to select a minimal set of mobile nodes to
meet the required coverage α within a time interval T
(1) Compute probability that each mobile node visits each
sensing location in AoI from its initial location
(2) Select nodes one by one until sum of probabilities that
the selected nodes visit each location is equal to/larger than
a
AoI
u1
Sum of prob.: 0.375
u1visits within T=2 at prob.
0.5×0.25=0.125
u2 visits within T=2 at
prob. of 0.25
u2

Probability Calculation
14
 Probability that a node u visits location x at time t
 Obtained by multiplying probability matrix t times
 Probability that the set of nodes U visit location x in T
 1 minus the probability that any nodes do not visit x during 1≤t≤ T

Prob(u,t,x0
u
,x) [vector(x0
u
)  Pt
]x
N×N matrix
representing
moving probability
between two locs
N-item vector
(0…1…0)
N: # locs in A
SetProb(x,U,T) 1 (1Prob(u,t,x0
u
,x))
1tT

uU

Initial location of u
reducible to
the area within
distance T/2
from AoI

Selection Strategy
15
We want to select the minimal set U’ satisfying:
 For every location x in AoI, SetProb(x, U’, T) ≥ a
 Deriving the optimal solution  NP-hard
 Heuristic in selecting nodes
 Random greedy selection: many redundant nodes can be selected
 Better selecting nodes that are not likely to visit the same locations
 Proposed selection strategies:
 Select nodes whose initial mutual distance is large
 Select nodes whose fist expected meeting time is late

Inter-Location Based algorithm (ILB)
16
 ILB selects nodes one by one until AoI is (a, T )-covered
So that the distance between any pair of nodes  dth (threshold)
 How to decide dth value
 Depends on T and α
 Intuitively, it should be larger as T increases and/or α decreases
u1
u2
u3
u4
dth  min(
T
a  dmax
,dmax),
dmax  max
u, u U
{du, u }
u5
dth=4

Inter-Meeting Time Based algorithm (IMTB)
17
 IMTB selects nodes one by one until AoI is (a, T )-covered
So that expected meeting time between any pair of nodes  mtth (threshold)
 How to decide mtth value
 Depends on T and a
 Intuitively, it should be larger as T increases and/or a decreases
u1
u2
u3
u4
Expected meeting time
mtth  min(
T
a  mtmax
,mtmax),
mtmax  max
u, u U
{mtu, u : mtu, u  }

Outline
18
 Related work
 Conclusion

Simulation Setting
19
Filed Size: 500 x 500 m2
# of nodes: 100 (25-200 nodes cases can be found in the paper)
Node speed: 1 meter/second
Area of Interest
 AoI-Size: 0.01, 0.25, 0.45, 0.5, 0.65, 0.85 of the whole field
Query Deadline T = 2, 4, 6, …, 20 [units of time], 1 unit = 50 sec
Required Coverage a : 0.5 (0.2-0.9 cases in the paper)
Each experiment was evaluated 5 times and averaged

Evaluation Scenarios
20
 Equal Moving Probabilities Scenario
Grid-based map with equal moving
probability (i.e., 0.25 for
up, down, left, right direction)
 Realistic Scenario
For city map near Osaka
station, Japan, generated a realistic
mobility trace with MobiREAL [19] from
actually observed pedestrians density on
roads
Determined probability matrix P based on
the map and the generated trace A specific city map near
Osaka station in Japan

Performance Metrics
21
 # of selected nodes
 Achieved coverage
 The ratio of the number of sensing locations visited by at least
one node to the total number of sensing locations in AoI
Achieved coverage =
Total # of locations
# of covered locations

Equal Moving Probabilities Scenario (1)
different AoI-Size22
N =100, α = 0.5 , T = 8
0
10
20
30
40
50
60
70
80
90
0.01 0.25 0.45 0.5 0.65 0.85
#SelectedNodes
AoI-Size
ILB IMTB
# all nodes in AoI
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.01 0.25 0.45 0.5 0.65 0.85
AchievedCoveredRatio
AoI-Size
ILB IMTB
•# all nodes rapidly increases as AoI size grows
•ILB and IMTB suppress # selected nodes to a great extent
•IMTB is better when AoI-Size ≥ 0.45

Equal Moving Probabilities Scenario (2)
different query deadlines, T
AoI-Size = 0.5, N =100, α = 0.5
0
5
10
15
20
25
30
35
40
45
50
2 4 6 8 10 12 14 16 18 20
#SelectedNodes
Total # steps, T
ILB IMTB
# all nodes in AoI
0.1
0.2
0.3
0.4
0.5
0.6
0.7
2 4 6 8 10 12 14 16 18 20
Total # steps, T
ILB IMTB
23
•ILB and IMTB suppress # selected nodes to a great extent
•ILB and IMTB adjust # selected nodes depending on T value
•IMTB is better when T ≥ 6

Realistic Scenario
different query deadlines, T24
AoI-Size = 0.5, N =100, α = 0.5
0
10
20
30
40
50
60
70
2 4 6 8 10 12 14 16 18 20
#SelectedNodes
Total # of steps, T
ILB IMTB
# all nodes in AoI
0.1
0.2
0.3
0.4
0.5
0.6
0.7
2 4 6 8 10 12 14 16 18 20
Total # steps, T
ILB IMTB
•# selected nodes is larger than equal probabilities case (still good)
•IMTB is better independently of T

Conclusion
25
 We formulated (α, T)-coverage problem in PCS
 We proposed two heuristic algorithms: ILB and IMTB
 Proposed algorithms achieved (α, T)-coverage with good
accuracy for variety of values of α , T, # nodes, AoI size
 IMTB selects a smaller number of nodes without
deteriorating coverage accuracy
 Future work
Update mechanism: removing useless nodes and adding some
extra nodes that more contribute coverage during period T
Extending selection area: selecting nodes inside and outside AoI
when insufficient number of nodes exists inside AoI

Distance and Time Based Node Selection for Probabilistic Coverage in People-Centric Sensing

More Related Content

What's hot (20)

Similar to Distance and Time Based Node Selection for Probabilistic Coverage in People-Centric Sensing (20)

More from Ubi NAIST (20)

Recently uploaded (20)

Distance and Time Based Node Selection for Probabilistic Coverage in People-Centric Sensing