Post on 26-Mar-2020
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
3MSc Final Exam - 7/21/2008
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.
5MSc Final Exam - 7/21/2008
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
� Secure backbone design
6MSc Final Exam - 7/21/2008
Generic Top-Down Cluster & cluster
tree formation (GTC) algorithm
C
C2
C4
C1
7MSc Final Exam - 7/21/2008
C3
GTC - Cluster tree formation
� Cluster tree is formed by keeping track of parent & child relationships
C2
C4 C1
C1
C3
C4C3C2
C10C8C7C6C5C9
8MSc Final Exam - 7/21/2008
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)
9MSc Final Exam - 7/21/2008
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()
10MSc Final Exam - 7/21/2008
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
11MSc Final Exam - 7/21/2008
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
14MSc Final Exam - 7/21/2008
Transmission power (dBm)
-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
15MSc Final Exam - 7/21/2008
Overview
� Motivation
� Cluster & cluster tree formation algorithm
� Simple Hierarchical Clustering (SHC)
� Hop-ahead Hierarchical Clustering (HHC)
� Routing� Routing
� Hierarchical addressing scheme
� Cross-links based routing
� Circular path based routing
� Forming Virtual Sensor Networks
� Secure backbone design
16MSc Final Exam - 7/21/2008
Cluster tree based routing17
� Need some sort of an addressing scheme to route
17MSc Final Exam - 7/21/2008
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)
18MSc Final Exam - 7/21/2008
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)
19MSc Final Exam - 7/21/2008
� 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 ),,(),,(][ ∫∫∫=
21MSc Final Exam - 7/21/2008
Analytical model (cont.)
22MSc Final Exam - 7/21/2008
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
24MSc Final Exam - 7/21/2008
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
25MSc Final Exam - 7/21/2008
� Cross links helps to deliver more messages
� Circular path delivers even more messages
PT = -20dBm
Transmission power (dBm)
-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
� Cluster & cluster tree formation algorithm
� Simple Hierarchical Clustering (SHC)
� Hop-ahead Hierarchical Clustering (HHC)
� Routing� Routing
� Hierarchical addressing scheme
� Cross-links based routing
� Circular path based routing
� Forming Virtual Sensor Networks
� Secure backbone design
26MSc Final Exam - 7/21/2008
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
29MSc Final Exam - 7/21/2008
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
Transmission power (dBm)
-20 -18 -16 -14 -12 -10
2000
4000
Transmission power (dBm)
-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
Transmission power (dBm)
-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
31MSc Final Exam - 7/21/2008
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
32MSc Final Exam - 7/21/2008
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
� Hierarchical addressing scheme
� Cross-links & circular path based routing schemes increase network lifetime at least by a factor of 2
33MSc Final Exam - 7/21/2008
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
34MSc Final Exam - 7/21/2008
Questions ?
Thank You…Thank You…