Wireless Ad Hoc / Sensor Networks:
Energy Efficiency and Cooperativeness
Xiang

Yang Li
Illinois Institute of Technology
xli@cs.iit.edu
2
Acknowledgment
Colleagues
Ophir Frieder, Sanjiv Kapoor, Peng

Jun Wan, Gruia
Calinescu, Ming

Yang Kao, Zheng Sun, Xiaowen
Chu,….
PhD Students
Yu Wang, WeiZhao Wang, WenZhan Song, Kousha
Moaveninejad, Chih

Wei Yi
Support
NSF CCR 0311174, 0342259
3
Organization
Achievement Summary
Research on Wireless Networks
Students Supervising, Supervised
Services
Research
Wireless networks
Energy efficiency
Cooperative issues
Algorithm design and analysis
Computational geometry
Algorithm mechanism design
Conclusion
4
Research on Wireless Networks
Published papers (
since joined IIT at 2000
)
Journals:
31
(20 published, 11 accepted)
10 IEEE Transactions, 8 ACM Journals
Referred Conferences:
57
2 ACM MobiCom, 4 ACM MobiHoc, 5 IEEE INFOCOM, 1 ACM
SODA….
Best paper awards
COCOON 2001
IEEE HICSS 35 (2002)
ACM MobiCom 2005
one of three best paper candidates
–
other 2 from MIT
Funding
NSF for Wireless CDMA assignment (co

PI)
NSF for workshop on Algorithms in Wireless Networks
5
Students
Students Supervised
Yu Wang (PhD 2004,
Assistant professor
at CS, UNCC)
WenZhan Song (PhD 2005,
Assistant professor
at CS, WSU)
Ovidiu Cristea (MS 2004), Mihai Moldovan (MS 2005)
Students Supervising
Kousha Moaveninejad (PhD expected 2006)
Weizhao Wang (PhD expected 2006)
Ashraf Nusairat (PhD, 2004

?)
Yanwei Wu (PhD, 2005

?)
QiZhong Hu (MS)
Thesis Committee
A number of PhD and MS students
6
Services
To the discipline
Guest editor of
ACM MONET
,
IEEE JSAC
Editor:
Ad hoc & Sensor Wireless Networks
: An
International Journal
TPC member of a number of conferences, e.g.,
ACM MobiHoc 2005, IEEE INFOCOM 2005, IEEE ICCCN
2005, IEEE MASS 2005, IEEE RTSS 2004
Invited review of
NSF proposals
Articles for numerous well

known journals
Gave more than
15
invited colloquiums worldwide
HongKong, China, Mexico, USA
Invited
tutorial
at ACM MobiHoc
7
Services
To the university, department
Graduate student admission (2000

present)
Graduate Study Committee (2002

present)
Undergraduate Study Comm. (2000

2002)
Undergraduate CAMRAS Award Interviewer
Sophomore Leadership Retreat
…
8
Research
Main research area
Wireless networks
Energy efficiency
Efficient distributed algorithm design
Cooperative issues
Algorithm Design and Analysis
Algorithm mechanism design
Computational geometry
High quality mesh generation
9
Organization
Achievement Summary
Research on Wireless Networks
Students Supervising, Supervised
Services
Research
Wireless networks
Energy efficiency
Cooperative issues
Algorithm design and analysis
Computational geometry
Algorithm mechanism design
Conclusion
10
Wireless Ad Hoc Network
No wired infrastructure
Self

organized
All nodes act as routers
Broadcasted signal
Powered by battery
(majority)
Mobile (maybe)
Potential Multi

hop routes
11
Energy Efficiency at Routing
Many routing protocols proposed
Metric Based Routing
DSR, AODV, ….
Location Based Routing
GPSR, GFG, AFR,….
Content Based Routing
Which links to use
Shorter links more stable, thus less retransmission
Save energy possibly
12
Location Based Routing
Each node forwards message to “
best
”
neighbor
E.g., “
best
”
closest
to target
t
s
13
Greedy Routing?
Fails to deliver
t
?
s
w
What should node w do?
14
Get out of local minimum
Find a planar graph
Gabriel Graph, for example
Face Routing or Right Hand Rule
t
?
s
w
15
Topology Control
Topology control is to select some nodes and/or
some available links as candidates for routing
Backbone
based structures select some nodes
Mainly used for broadcast, multicast
Typically assume that node’s power
fixed
—
thus minimize the number of backbone nodes (MCDS)
Flat
structures select some links, e.g., GG,LMST
Used for unicast, or broadcast
Typically assume that node’s power
adjustable

thus minimize the total power (so called low

weight), or
power to connect any pair of devices (so called spanner)
16
Backbone Structure
Select some nodes
Form a backbone (
Connected Dominating Set
)
each other node is connected to some node in backbone
Backbone needs to be connected
Our efficient distributed methods
Using only O(
n
) total messages, find a backbone at
most 12 times optimum
Proved to be power spanner (
fixed or adjustable
)
Published at
IEEE ICDCS’02
, then
IEEE TPDS’03
17
Flat: What we want to achieve?
Build a
single
structure
efficiently
with a
number of nice properties:
Power efficient Unicast (
majority operations
)
Power efficient broadcast (
widely used in WSN
)
Bounded node degree (
logical, physical
)
Planar structure (
support greedy routing
)
Separated neighbors (
directional antenna, reduce signal
interference
)
All these properties are achieved in a single
structure
After a sequence of results
tradeoffs
18
What do we mean by “efficiently”?
Best scenario
Localized method (
run in constant rounds
) to build
such structure
Each node u quickly determines which links uv to keep
locally
Our achievement
A semi

localized method with total communication
cost
O(n log n) bits
with wireless broadcast model
Worst case still
O(n)
rounds
19
Our Network Model
A set
V
of n wireless nodes in 2D region
All nodes with
same
transmission power (
fixed power
)
Ideal case,
It induces a unit disk graph
UDG
Two nodes are connected directly if distance at most one unit
Each node knows the position of its one

hop
neighbors
Localization techniques assumed already in place
20
Adjustable Power Model
Power needed to support a link uv is
proportional to
This model
Only accounts for emission power
Good only if long range communication, or
techniques are used to reduce the receiving power
uv
u
v
21
Priory Arts: Some Structures
RNG
GG
Yao
MST
22
Priori Arts
published
Topology
Planar
Unicast
Spanner
Low
weight
Degree
Bound
Comm.
Cost
INFOCOM 01
YAO+GG
Yes
Yes
NO
NO
~O(n)
PODC 01
CBTC
No
Yes
No
Yes
~O(n)
MobiHoc 01
RDG
Yes
Yes
No
No
~ O(n
2
)
ICCCN 02
Yao’
No
Yes
No
No
~O(n)
INFOCOM 03,
LMST
Yes
No
No
6
~ n
MobiCom 04
FLSS
No
Yes
No
No
~ O(n)
Not completed here, due to space limit
23
Our Results
published
Topology
Planar
Unicast
Spanner
Low
weight
Degree
Bound
Comm.
Cost
ICCCN 01
RNG
Yes
No
No
No
n
GG
Yes
Yes
No
No
n
Yao
No
Yes
No
7
(2K+1)n
INFOCOM 02,
TPDS
LDel
Yes
Yes
No
No
~60n
INFOCOM 04,
TPDS
LMST2
Yes
No
Yes
6
~700 n
MONET
IMRG
Yes
No
Yes
6
~7n
DialM 03,
MONET
BPS
Yes
Yes
No
27
~700n
MobiHoc 04,
MONET
OrdYaoGG
Yes
Yes
No
12
24n
SYaoGG
Yes
Yes
No
9
3n
MobiCom 05,
LS
Q
䝇
奥Y
奥Y
奥Y
9
12n
24
Power Efficient Unicast Structure
Assume GG has been constructed. All nodes
marked unprocessed initially.
Once a node u has smallest ID among
unprocessed
neighbors, then:
If it has
processed
neighbors, then it
keeps the nearest
processed
neighbor and
delete other links
conflicted
with this
Otherwise, it selects the nearest
unprocessed
neighbor and delete the
conflicted
links and repeat till all nodes
are processed
Let S
q
GG be the final structure
Q

region
u
v
q
w
25
Structure Illustration
u
b
a
c
d
e
f
g
h
i
j
k
l
m
n
o
p
q
r
processed
unprocessed
26
Properties
We can prove that the resulting topology is
Planar
Power efficient for unicast
Bounded logical node degree
Neighbor
q

separation
What we miss is (
counter example omitted
)
Power efficient for broadcast (
low weight
)
27
Energy Efficient Broadcast
s
1
v
2
v
4
v
3
v
i
v
sv
s
p
i
max
)
(
u
u
p
T
p
)
(
)
(
Any broadcast can be viewed as an arborescence rooted at s
28
Priori Arts On Efficient Broadcast
Several Structures Proposed
MST, BIP, SPT, RNG, etc.,
Theoretically Good
–
INFOCOM, WINET
MST, BIP: within constant of optimum
But,
not localized, or even not efficient in a distributed
way
Not efficient for unicast
)
(
MST
uv
c
OPT
MST
uv
29
Broadcast
–
Low

weight “Optimal”
A structure is called
low

weighted
if its total link
length is within O(1) of MST
Proved previously: (
INFOCOM’03, TPDS’04
)
Given any low

weighted structure H, the total power
consumption for broadcast is asymptotically
best
among
all locally
constructed structures
Proposed several
localized
methods with O(n) messages
that construct a low

weighted structure

(
TPDS’04,
WINET’05
)
30
Add Low

Weight Property
Our previous approach
Given a structure, such as RNG, LMST
x
y
u
v
Any
node x removes the longest link of
any
quadrilateral xyvu
31
Not Efficient for Unicast Anymore
1
u
1
v
2
u
2
v
n
u
n
v
3
u
3
v
4
u
4
v
May break connectivity for
graph S
Q
䝇G捯湳瑲c捴敤灲敶楯畳汹
32
Our New Unified Structure
Build S
q
GG graph
Each node x collects 2

hop links E(x) in S
q
GG
Node x picks an incident link xy
with
smallest
ID
(xy, maxID(x,y), minID(x,y))
If exits uv such that
xy>max(uv,3ux,3vy)
removes link xy from E(x)
Otherwise
keeps link xy
forever
Let LS
q
GG be the final structure
x
y
u
v
33
Properties
We can prove that the resulting topology
Planar
Power efficient for unicast
Bounded logical node degree
Neighbor
q

separation
Power efficient for broadcast (
low weight
)
Can be constructed efficiently using O(n) messages
34
Expected Interference
Interference
The physical degree of node u
I(u)=7
x
y
u
v
w
35
Random Deployment
When nodes are of Poisson distribution
The maximum node interference is at
least
O(log n)
for
any connected
structure almost surely
Proof omitted
Thus our structures also are bad in terms of
the maximum node interference
36
Random Deployment
When nodes are of Poisson distribution
The average node interference is only
O(1)
for the following structures
RNG, GG, LMST, S
q
GG, LS
q
GG,…..,
See our ACM MobiCom 2005 paper for more details
about these structures and proofs
37
Other Results
Determine Transmission Range (
MobiHoc’03
)
So the induced graph has some properties almost surely for certain random
distribution
The critical range for
connectivity
and
k

connectivity
OVSF/CDMA Code Assignment (
DialM’03, COCOON’05
)
PTAS for IS, VC in some graphs
To maximize the bottleneck and throughput
Build CDS Efficiently (
ICDCS’02, TPDS’03
)
Linear messages, 12

approximation
Some Geometry Results
New structure: Local Delaunay Graph (
INFOCOM’02, WADS’04
)
Spanning ratios of Beta

Skeleton (
CCCG
)
High Quality mesh generation (
STOC’00, SODA’01
)
38
Organization
Achievement Summary
Research on Wireless Networks
Students Supervising, Supervised
Services
Research
Wireless networks
Energy efficiency
Cooperative issues
Algorithm design and analysis
Computational geometry
Algorithm mechanism design
Conclusion
39
New Dimension
Previously,
Efficient topology control
Time, space, communication efficiency
Assumption
Participants act as instructed
Not always true
Faulty ones
Fault

tolerant computing
Malicious ones
Security, and Trusted
computing
Selfish ones
Truthful computing
40
How to deal with selfish nodes?
Reputation based methods
Nodes are rated by peers
Detecting/punishing/avoiding
Pay each node its declared cost
Node will manipulate its declared “cost” to increase
its profit
May reach a stable point: no node will unilaterally
change its declared cost

Nash Equilibrium
Pay each node some payment
Node maximizes its profit when it reports cost
truthfully

Dominant Strategy
So relieve nodes from manipulating declared cost
41
Non

Cooperative Networks
SP 1
SP 2
SP 3
3
5
7
4
4.5
4.8
4.9
6
42
Non

Cooperative Networks
Network Agent
Selfish
: Only interested in its own benefit instead of system
performance
Rational
: Do what will
maximize
its own benefit
Non

Cooperative Networks
A set of n agents which are selfish and rational
For each agent, it has a set of strategies
Algorithm mechanism design
Mechanism M=(O,P)
O determines who to be selected
P determines how much to pay the agents
43
Unicast
0
v
9
v
4
v
1
v
8
7
v
2
v
6
v
5
v
8
v
6
7
7
9
5
1
7
3
v
Node v
k
costs c
k
to relay
(private knowledge)
Each node v
k
is asked to
report a cost
d
k
Find the
least cost path
from node v
0
to node v
9
based on reported costs
d
Compute a payment
p
k
for
node v
k
based on
d
Objective: Find a payment
p
k
(
d
) so
node maximizes utility when
d
k
=
c
k
44
Truthful Unicast Scheme
Output O
Least cost path from s to t, by LCP(
s, t
, G)
Payment to a relay node v
k
(
VCG
mechanism:
2
nd
price auction
)
Remove it and its incident links
Compute the shortest path from s to t
The payment to v
k
is
Otherwise the payment is 0
Present a centralized method with time
O(
m+n log n
)
to compute
payment to all nodes
Clearly asymptotically optimum
IEEE Transaction on Mobile Computing, 2005
)
,
,
(
)
\
,
,
(
G
t
s
LCP
v
G
t
s
LCP
d
p
k
k
k

45
0
q
3
4
7
5
9
2
3
q
1
q
1
v
5
v
2
v
2
q
4
v
6
v
3
v
1
Multicast
K receiving nodes R and a source
Node v
k
costs c
k
to relay (private
knowledge)
Each node v
k
is asked to report a
cost
d
k
Find the
minimum cost tree
spanning all receivers and source
node based on reported costs
d
Compute a payment
p
k
for node v
k
based on
d
Objective: Find a payment
p
k
(
d
) so
node maximizes utility when
d
k
=
c
k
46
Structure (node or link or both)
Calculate all shortest paths from source
node to receivers
Combine these shortest paths
The structure is a tree called Least Cost
Path Tree (LCPT)
Payment Scheme
Calculate the payment for node v
k
based
on every LCP containing v
k
Choosing the maximum of these
payments as the final payment
0
q
3
4
7
3
9
2
2
q
1
q
1
v
5
v
2
v
3
q
4
v
6
v
3
v
1
6
6
9
3
2
0
2

q
q
p
4
6
7
3
3
0
2

q
q
p
6
)
,
max(
2
0
2
0
2
2
2
q
q
q
q
p
p
p
7
v
LCPT Based Mechanism
47
Other Structures
VCG Mechanism generally does
not
work
Since finding minimum cost spanning tree is NP

hard.
Virtual Minimum Spanning tree
Construct the virtual complete graph
K(G)
Nodes are receivers, plus source node
Edges are LCP between two end

points
Find the MST on
K(G)
,
say
V
MST(G)
All agents on VMST(G) are selected
General link weighted Steiner Tree
NP

Hard, constant approximation methods exist
Efficient computing of payments
General Node weighted Steiner Tree
NP

Hard, best approximation ratio
O(ln k)
Efficient computing of payments
See our ACM MobiCom 2004 paper for more details
Multicast Cost/Payment
Sharing

cooperative games
49
General “Cost” Sharing
Given a set of players N
The cost of C(S) for every is known
The cost is cohesive: C(S+T)<= C(S)+C(T)
Fair
Cost Sharing
For all players
Budget balance
:
For every subset of players S:
Core:
Cross

monotone:
)
(
N
i
N
S
)
(
)
(
1
N
c
N
n
i
i
)
(
)
(
S
c
N
S
i
i
)
(
S
i
T
S
T
S
i
i
),
(
)
(
50
Multicast Cost Sharing(fixed tree)
Given a structure for multicast
The cost of each relay agent is known
A
fixed
tree from the source to all receivers
Share the cost among receivers
Budget balance, core, Cross

monotone
Methods:
Equally share for downstream receivers (ELDS)
Comcast
$10
$10
Alice
Bob
Digital Classic
$20
51
Cost Sharing (no fixed tree)
All receivers must get the data
Find an efficient tree as output
Share the cost of tree among receivers
fairly
?
Various concepts of fair: core, etc

Core:

Budget balance
“
core
”
Tight bound
No core allocation can recover more than fraction of cost
Conjecture
: A core allocation can recover fraction of cost
)
(
)
(
1
N
C
x
N
C
n
i
i
)
(
S
OPT
x
S
i
i
n
ln
1
n
ln
1
See STACS’05 for more details
52
Cost Sharing (no fixed tree)
Cross monotonic

Core:

Budget balance
“
Core
”
Cross monotone
Tight bound
No CM

Core allocation can recover more than
fraction of cost
of Shapley value on LCPT can recover fraction
of cost and being a CM

Core!
)
(
)
(
1
N
C
x
N
C
n
i
i
)
(
S
OPT
x
S
i
i
n
1
n
1
n
1
See STACS’05, INFOCOM’05 for more details
53
Multicast Payment Sharing
Multicast payment sharing (
IEEE INFOCOM 2005)
Given a mechanism M=(O,P)
Example: Truthful Payment for LCPT
How much each receiver should pay?
Fair
Payment Sharing Scheme:
Budget balance
: the payment is all agents is recovered
Cross

monotonic
: more receivers, less sharing
No negative transfer
: The sharing is positive
No free rider
: sharing of each receiver is within some bound of
what it has to pay in its unicast
54
Recall LCPT Payment
Payment for agent
v
k
is max
q
i
P
k
UNI
(s,q
i
).
1
p
3
p
2
p
v
k
3
2
1
p
p
p
Payment
v
k
to is
p
3
55
Simple Sharing Not Works
Fair Sharing: ELSD?
Comcast
Alice
Bob
Digital classic with
HBO $60
$30
$30
Digital Classic
$20
Digital Classic
+ HBO $60
Digital classic
$20
56
Illustration of Fair Sharing
Comcast
Alice
Bob
Digital Classic
$20
Digital Classic +
HBO $60
Digital Classic with
HBO $60
$20
$40.00
$10
$10
57
Sharing LCPT Payment
Payment for agent e
k
is max
q
i
P
k
UNI
(s,q
i
).
1
p
3
p
2
p
v
k
3
1
p
Each shares
2
1
2
p
p

q
2
and q
3
each shares
q
1
q
2
q
3
2
3
p
p

q
3
shares
58
Properties
No negative transfer
Budget balance
Cross

monotonic
No

free rider
Dummy:
sharing is its cost if marginal payment = payment of
unicast
Symmetry:
shared payments are same if two are interchangeable
59
Other Results
Algorithm mechanism design
General framework for binary demand games (
ACM
EC’05
)
AMD and cost sharing for set cover games
(
STACS’05, TCS’05
)
Sets or elements are agents
DiffServ Multicast (
AAIM’05, COCOON’05
)
AMD design and payment sharing
Nash equilibrium
Nash equilibrium and AMD for unicast and multicast
(
ISAAC’05
)
60
Questions and Comments
Enter the password to open this PDF file:
File name:

File size:

Title:

Author:

Subject:

Keywords:

Creation Date:

Modification Date:

Creator:

PDF Producer:

PDF Version:

Page Count:

Preparing document for printing…
0%
Comments 0
Log in to post a comment