Top-Down Clustering Based Self-Organization of ...Self-Organization of Collaborative Wireless Sensor...

Top-Down Clustering Based

Self-Organization of Collaborative

Wireless Sensor Networks

M.Sc Final ExamM.Sc Final Exam

Dilum BandaraDepartment of Electrical and Computer Engineering

July 21, 2008

Graduate Committee

� Prof. Anura P. Jayasumana (Advisor)� Prof. V. Chandrasekar� Dr. Daniel F. Massey

� This work is supported in part by the

grant from Environmental Sciences

Division, Army Research Office

Overview

� Motivation

� Virtual Sensor Networks (VSNs)

� Cluster & cluster tree formation algorithm

� Simple Hierarchical Clustering (SHC)

� Hop-ahead Hierarchical Clustering (HHC)� Hop-ahead Hierarchical Clustering (HHC)

� Routing

� Hierarchical addressing scheme

� Cross-links based routing

� Circular path based routing

� Forming Virtual Sensor Networks

� Secure backbone design

2MSc Final Exam - 7/21/2008

Wireless Sensor Networks (WSNs)

� Sensing of physical world at a far greater temporal &

spatial granularity

� Novel applications

� Habitat monitoring, earthquake monitoring, disaster response, eldercare, battlefield intelligence, etc. eldercare, battlefield intelligence, etc.

� Sensor nodes are resource constrained

� Battery powered, low processing, memory, & transmission power

� In most cases

� Randomly deployed, location unaware, no time synchronization,

unreliable, & dense networks


Motivation – Collaborative WSNse-SENSE project [1]

Underwater Acoustic Sensor Networks project [2]

Plume tracking

[1] www.ist-esense.org[2] http://www.ece.gatech.edu/research/labs/bwn/UWASN/work.html 4

Virtual Sensor Networks (VSNs)

� An emerging concept that

� supports collaborative, resource efficient, & multipurpose WSNs

� involve dynamically varying subset of sensor & users

� Provide protocol support for

� formation, usage, adaptation, & maintenance� formation, usage, adaptation, & maintenance

� Realization of VSNs require

� some structure within the network

� many-to-many communication

� VSN management functions

� The solutions should be independent of

� neighbourhood information, location awareness, network topology, time synchronization, etc.


Contributions

� Clustering algorithm

� 2 clustering schemes� Cluster tree formation� Performance analysis

� Routing

� Hierarchical addressing scheme� 3 routing mechanisms� Performance analysis

� Virtual Sensor Networks

� Formation� Communication



Generic Top-Down Cluster & cluster

tree formation (GTC) algorithm

C

C2

C4

C1


C3

GTC - Cluster tree formation

� Cluster tree is formed by keeping track of parent & child relationships

C2

C4 C1

C1

C3

C4C3C2

C10C8C7C6C5C9


GTC algorithm

Form_Cluster(NIDCH, CIDCH, delay, nCCHs, hopsmax, TTLmax, depth)

Wait(delay)

TTL ← TTLmax

Broadcast_Cluster(NIDCH, CIDCH, hopsmax, TTLmax, TTL, depth)

ack_list ← Receive_ACK(NIDchild, hops, p1, p2, timeoutACK)

IF(ack_list = NULL)

Join_Cluster()Join_Cluster()

FOR i = 1 TO nCCHs

CCHi ← Select_Candidate_CHs(ack_list)

CIDi ← Select_next_CID(i)

delayi ← Select_delay(i)

depthi ← depth + 1

Request_Form_Cluster(CCHi, CIDi, delayi, nCCHs, hopsmax, TTLmax, depthi)


GTC algorithm (cont.)

Join_Cluster()Listen_Broadcast_Cluster(NIDCH, CIDCH, hopsmax, TTLmax, TTL, depth)

TTL ← TTL – 1, hops ← TTLmax – TTL

IF (hops ≤ hopsmax AND my_CID = 0)

my_CID ← CIDCH, my_CH ← NIDCH, my_depth ← depth + 1

Send_ACK(my_NID, hops, p1, p2)

IF(TTL > 0)IF(TTL > 0)

Wait(Random(backoff_time))

Forward_Broadcast_Cluster(NIDCH, CIDCH, hopsmax,, TTLmax, TTL, depth)

IF(hops ≤ hopsmax)

Exit()

ELSE

IF(Wait_Listen_Neighbors(Random(backoff_time) = FALSE)

Send_ACK(my_NID, hops, p1, p2)

IF(Listen_Form_Cluster(CCH, CID, delay, nCCHs, hopsmax, TTLmax, depth, timeout) = TRUE)

Form_Cluster(my_NID, CID, delay, nCCHs, hopsmax, TTLmax, depth)

Exit()

Join_Cluster()


GTC algorithm (cont.)

� SHC – Simple Hierarchical Clustering

� HHC – Hop-ahead Hierarchical Clustering

CID 1

73

10

CID 3CID 7

CID 10

SHC clustershopsmax = TTLmax = 1

Similar to IEEE 802.15.4 cluster tree

HHC clustersTTLmax = 2×hopsmax + 1

1

CID 1

5

4

3

2

9

8

6

CID 2

CID 4

CID 5

CID 6CID 9

CID 8


Clusters

� HHC

� 5000 nodes

� Grid - 201×201

� Grid spacing – 5m

� R = -20dBm

� Root node in the middle12

Cluster tree

13

Performance analysisC

ircu

lari

ty

20

30

40

50

60

70

80

90

SHC

HHC

RHHC

Hexagonal

Nu

mb

er

of

clu

ste

r h

ead

s

60

80

100

120

140

160

SHC

HHC

RHHC

Hexagonal

Transmission power (dBm)

-20 -18 -16 -14 -12 -10

Nu

mb

er

of clu

ste

rs

0

300

600

900

1200

1500

1800

2100

SHC

HHC

RHHC

Hexagonal



-20 -18 -16 -14 -12 -10

0

10

Depth

0 2 4 6 8 10 12 14 16N

um

be

r o

f clu

ste

r h

ead

s0

20

40

60

� HHC� Uniform clusters

� Better circularity

� Lower number of clusters

� Lower depth

� Message complexity O(n)

Optimization phase –Handling disconnected nodes & optimizing cluster tree

PT = -12dBm, 5000 nodes


Overview

� Motivation



� Hop-ahead Hierarchical Clustering (HHC)

� Routing� Routing







Cluster tree based routing17

� Need some sort of an addressing scheme to route


Hierarchical addressingA(0)

D(20)C(10)B(00) F(40) G(50)E(30)

18

� Single point of failure at root node

I

(100)

H

(000)

K

(110)

L

(210)

J

(010)

M

(020)

N

(030)

O

(050)

Q

(0100)

P

(0210)


Cross-links based routingA(0)

D(20)C(10)B(00) F(40) G(50)E(30)

19

� Routing through cross links

� Reduce burden on the root node

� Use hierarchical addresses

I

(100)

H

(000)

K

(110)

L

(210)

J

(010)

M

(020)

N

(030)

O

(050)

Q

(0100)

P

(0210)


� Send message from U to K� Hierarchical routing - 5 hops� Cross links - 5 hops� Circular path - 4 hops

Circular path based routing

20

� Circular path� Connects clusters at the same

depth� Reduce workload on root node� Use hierarchical addresses

MSc Final Exam - 7/21/2008

Circular path based routing – Analytical model

][

][

][][

dEr

E

rangeontransmissi

distancehopperenergyE

hopshopperenergyE

transmittoenergyEcostE

T ×=

×=

×=

=

� R – Radius of sensor field

� D – Radius of circular link

� r – Transmission range of a node

� r1 – distance to source node

� r2 – distance to destination node

� ET – energy to send a message

21

][

rrdwhere

dEr

+=

×=

θθθ ddrdrrrprrddE 212121 ),,(),,(][ ∫∫∫=


Analytical model (cont.)


Analytical model (cont.)

4

5

2

332

3

4][

R

D

R

DDRdE

πππ−+−=

4

4

2

2 592][

R

D

R

D

dD

dE

πππ−+−= 43

2

3

82

43

2

3

82

4

2

2

4

2

2

R

D

R

D P(C)

R

D

R

D P(R)

πππ

πππ

π

+−=

−+−

=

D/R0.0 0.2 0.4 0.6 0.8 1.0

Pro

ba

bili

ty

0.0

0.2

0.4

0.6

0.8

1.0

P(R)

P(C)

RD 509.0=∴

23

• P(R) – Probability of going through root node

• P(C) – Probability of going through circular path

Routing - Performance analysis

14

16

18

Distance (m)

0 100 200 300 400 500

De

pth

0

2

4

6

8

10

12


PT = -20dBm, 5000 nodes

Performance analysis (cont.)N

um

be

r o

f m

essa

ge

s

10000

15000

20000

25000

30000

35000

40000

Tree only

Tree + Cross-links

Tree + Circular path

Ene

rgy p

er

messag

e (

µJ)

2.5

2.6

2.7

2.8

2.9

Depth at which circular path is formed

0 0-1 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9

Num

be

r o

f m

essa

ges

15000

20000

25000

30000

35000

40000


� Cross links helps to deliver more messages

� Circular path delivers even more messages

PT = -20dBm


-20 -18 -16 -14 -12 -10

Nu

mbe

r o

f m

essa

ge

s

0

5000

10000

5000 nodes, circular sensor field

Depth at which circular path is formed

0 0-1 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9

2.4

Overview

� Motivation



� Hop-ahead Hierarchical Clustering (HHC)

� Routing� Routing







Plume tracking

MSc Final Exam - 7/21/2008 27

Forming VSNs

� Messages are guaranteed to meet at root node� Forms a virtual tree

� More efficient & reliable than rumor & ant routingMSc Final Exam - 7/21/2008 28

Forming VSN –Virtual tree that connects VSN members


Single event region Three event regions

VSN - Performance analysis

Num

ber

of

messag

es

6000

8000

10000

12000

14000

16000

18000

20000

Region 1

Region 1+2+3

distributed

Num

ber

of m

ultic

ast

messages

500

1000

1500

2000

2500

3000

3500

Region1

Region 1+2+3

Distributed

Routing scheme

Tree Only Tree+Cross-links Tree+Circular path

Nu

mber

of

messag

es

10000

15000

20000

25000

30000

35000

40000

45000

50000

Region 1

Region 1+2+3

Distributed

30


-20 -18 -16 -14 -12 -10

2000

4000


-20 -18 -16 -14 -12 -10

Num

ber

of m

ultic

ast

messages

0

500

PT = -20dBm

Secure backbone design

� Can support

� dynamic key distribution

� secure upper layer functions

� Secure VSNs need dynamic key assignment

� GTC algorithm can be extended to build a secure backbone


-20 -18 -16 -14 -12

Circu

lari

ty

40

45

50

55

60

65

70

75

80

85

HHC

HHC + RBMCD

HHC + DKR

HHC + RBMCD-Nei

HHC + DKR-Nei

Hexagonal

Number of compromized nodes

2 4 6 8 10

Num

ber

of

CH

s a

ffecte

d

0

10

20

30

40

50

60

70

80 HHC + RBMCD - Direct

HHC + RBMCD - Indirect

HHC + DKR - Direct

HHC + DKR - Indirect

� GTC algorithm can be extended to build a secure backbone

� No significant changes are required

� Retains most of its desirable characteristics


Future directions

� Test performance of our algorithms on a rigorous

simulation platform such as TOSSIM

� Use actual data from a tank based testbed

� VSN management functions

� VSN dynamics

� e.g., migrating, disappearing, merging, & splitting plumes

� Identifying multiple networks

� Connecting them

� Routing algorithms that are not tied to the cluster tree

� Dynamic key distribution with & across VSNs


Summary

� A mechanism to form VSNs� Connect nodes observing the same phenomenon

� Inter-VSN & intra-VSN communication

� A cluster & cluster tree formation algorithm� Hop-ahead Hierarchical Clustering (HHC)� Hop-ahead Hierarchical Clustering (HHC)

� More uniform & circular clusters

� Cluster tree with lower depth

� Properties are comparable with hexagonal packing

� Cluster tree based routing


� Cross-links & circular path based routing schemes increase network lifetime at least by a factor of 2


Related publications

� Conference Papers� H. M. N. D. Bandara and A. P. Jayasumana, “An enhanced top-down cluster and

cluster tree formation algorithm for wireless sensor networks”, 2nd International

Conference on Industrial and Information Systems (ICIIS 2007), Sri Lanka, Aug.

2007.

� H. M. N. D. Bandara, A. P. Jayasumana, and I. Ray, “Key pre-distribution based

secure backbone design for wireless sensor networks”, 3rd Internationalsecure backbone design for wireless sensor networks”, 3 International

Workshop on Practical Issues in Building Sensor Network Applications

(SenseApp 08), Oct. 2008, to be published.

� Posters� H. M. N. D. Bandara, A. P. Jayasumana, T. H. Illangasekare, and Qi Han, “A

wireless sensor network based system for underground chemical plume

tracking,” CSU Ventures, Fort Collins, CO, Feb. 2008.

� First place - ISTeC Student Research Poster Contest


Questions ?

Thank You…Thank You…

Top-Down Clustering Based Self-Organization of ...Self-Organization of Collaborative Wireless Sensor...

Documents

Transcript of Top-Down Clustering Based Self-Organization of ...Self-Organization of Collaborative Wireless Sensor...