SELF ORGANIZED MOBILE NETWORKS PROJECT TOPIC: ROUTING TOWARDS A MOBILE SINK

foamyflumpMobile - Wireless

Nov 21, 2013 (3 years and 6 months ago)

52 views









SELF ORGANIZED MOBILE NETWORKS



PROJECT

TOPIC:


ROUTING TOWAR
D
S
A

MOBILE SINK



Prepared By Eda Baykan

07.02.2005




Supervisors

Jun Luo, Jean
-
Pierre Hubaux, Jacques Panchard






































































A
B
STRACT


In this paper

w
e propose a routing protocol

that
aims to maximize
network lifetime in
wireless sensor networks where sources are static and the sink is
mobile
.
During the design of
the protocol, it is assumed that the sink moves i
n a predefined pat
h in a non
-
predefined
manner and data is disseminated throughout the network in many to one manner.

What non
-
predefined manner m
eans is,
the sink can move with dif
ferent speeds and delays. In this
paper, network life time is taken as the first time when on
e of the nodes runs out of energy.

Minimizing number of transmissions by incorporating link connectivity status

and
distance to mobile sink

into routing decision
as well as enabling energy

balancing among the
nodes, are the key

points of our the routing
protocol.
[
3] proposed to use the first and second
criteria in order to have a reliable routing. Our contributi
on is

to take

into consider number of
in danger nodes

in addition to link quality and distance to sink during next node selection
.
We
did high le
vel simulations in order to see whether our proposal outperformed [3].


Keywords

Sensor Networks,
Network Lifetime,

Multihop
Routing, Link Connectivity, Energy

Balancing
, Mobile Sink


1.
INTRODUCTION


Sensor networks have many applications areas such as mi
l
itary and

environmental
monitoring.
Their popularity increases day by day both in academia and industry. However
bottleneck of these low cost sensors is their limited batteries. Furthermore once deployed in
the observation area, it is very difficult to re
charge them.

That’s
why

making

network life
time

longer is an important issue that deserves to be explored.


Routing protocols for wireless sensor networks must take into consideration metrics
such as reliability,

throughput, latency
, storage requirement
,

o
verhead

and
network lifetime
.
We

compare
recent
routing proposals
[1], [2], [3]
that deal with network lifetime problem in
terms
of
above defined metrics.

After clarifying the pros and cons of t
hem
,
we base our
routing protocol on [3].
What is lacking in t
hese routing proposals is
energy
balancing.

In
order to make the network live more, nodes that have more energy must do work on the
behalf of others that have
less energy.
In other words
,

r
outing decisions must take care of
remaining energies of nodes.

The

problem then reduces
to
determine the remaining energies
of nodes in an online manner.
Authors of [4] proposed an online routing protocol

that
takes
the remaining energies of nodes

as cost metric

and runs Dijkstra according to that
. In that
approach nodes

transmit messages in order to announce their remaining energies
.
However
this approach

causes nodes to

consume additional energy. In order to alleviate t
his problem, in
our proposal t
his information

is put

in
to

the

route messages
.

In our routing protocol
,

each node is assigned an
initial packet processing capacity
.
When a node transmits and receives packets for itself and for relaying on the behalf of other
nodes, it reduces its capacity by the amount of processed packets. If capacity becomes lower
than t
he predef
ined threshold, node becomes in
danger node.

In this paper one of the
assumptions is, during the transmission all nodes that are in transmission range of the sender
node
,

consume energy although the packet is not destined to them or transmission
i
s
not
successful.
When the transmission is not successful, transmitter node stops sending that
packet after trying three times.

In
forwarding
decisions
main question is how to choose next node
from the neighbor
nodes.
W
e have three criteria: link quality,
distance to base station and number of nodes that
are in danger.

The desired solution is to choose a node that

has high link quality, small
distance to sink and has small number of in danger nodes on the
shortest
path to sink.

High
link quality mean
s small

number of transmissions.

Small distance to sink again means less
number of
transmission
s because
number of hop
s will decrease and nodes will consume less
energy during the transmission.

We assigned weights to each criteria where distance to sink
has the h
ighest weight
.


Section 2 presents recent
proposals that deal with network lifetime problem

in wireless
sensor networks
.

In

Section 3, detailed comparison of routing proposal
s

that are mentioned in
Section 2 is given.

Our proposal,
Energy Balanced & Link
Status Aware Routing
is described
in Section 4
. Results
of high

level Matlab

simulations a
re given in Section 5. Finally
,

s
ection
6

concludes this paper.


2.
RELATED WORK


Many routing protocols have been proposed to deal with

energy conservation problem
i
n
sensor networks. However most of them focus on scenario where the sink, node that is
interested in getting data from

other sensor nodes
, is static.

[5]

shows

that when the sink is
mobile, sensor networks
have
higher network lifetime.

Some recent work

suc
h as
[1],

[2] and
[3]
has investigated the p
roblem of r
outing to a sink
consid
ering
network
lifetime
.

These

papers propose protocols to minimize

number of transmissions
.


2.1
University

of
Southern California

(USC)

Proposal


[1]

proposes a learning
-
based a
pproach to route data to a mobile sink
that
moves in a
predefined

path
silently. The sink does not make
any
queries. When source nodes sens
e
unusual events, they send information

to the mobile sink. In fact
[1]
focuses on locating the
moving sink effici
ent
ly when it moves silently

in a non
-
predefined manner
in a

predefined
path.

The aim of making the sink silent is
,

to
reduce the energy
consumed during the
broadcast of query.

In a distributed manner sensor nodes learn an efficient way to route data
to the s
ink. This learning process is done through positive and negative reinforcements
according to the goodness values
calculated
at the nodes close to the path of the sink.
Goodness value is nothing but the probability density function that indicates how far th
e sink
is from the node who is in the vicinity of the path of the sink.
T
our period

of the sink

is
divided into domains.

When a node has data to be forwarded, the probability of selecting
next
node
is determined according to its possibility bei
ng on a good

path to the sink in

that time
domain.
D
efinition of good path is open to discussion.

A
uthors

of [1]
refer to good path as the
path that leads to the sink
.

They ignore the scenarios where their chosen good path would
cause
one of the nodes die at an early
phase
.

Another issue that has ambiguity about [1] is the
assumption that nodes close to the path of the sink can somehow detect the presence of the
sink.

This assumption is not realistic because
if
the sink does not sen
d any query how i
t

can be
detected

by

nearby nodes
.

Furthermore each node must know the period of travel of the sink.

How nodes learn

this

period
is not clearly stated in the paper.


2.2 UCLA
Proposal



[2]

proposes

to use

controlled mobility
in order to improve network performance.
Sink (
bas
e station) is static however there is an autonomous mobile router that goes as close
as possible to the source

to collect data
.

After getting data from static source nodes m
obile
router goes to the

base station for submitting
. Mobile router has no energy l
imit

because of
mobility
. In other words
it can

be recharged easily

at the base station
.

Main aim in [2] is to
enable dat
a transmission over fewer hops in order to reduce
the number of
packets
transmitted. Consequently nodes will consume less energy becaus
e
of less transmission.

Another benefit of decreasing number of

hops is the reduction
in the probability of error.

When

less error occurs during the transmission, number of retransmission will decrease

consequently
.
Nodes can save the energy that would be
used for retransmission.

In [2] source
nodes wait to send data until mobile router comes as close as possible to them. The
disadvantage is

latency i
ntroduced with this approach. For

real time applications this proposal
won’t work
well
.


2.3 Berkeley
Propo
sal


2.3.1 Link Connectivity

In

order to achieve reliability and minimiz
e total number of transmissions
[3]

proposes
that

link

quality between nodes should be considered during routing decisions
.

Li
nk quality is
the percent of packets that arrived undamage
d on a link.

Quality of links can be determined by
capturing link connectivity statistics dynamically.
With this approach lossy and dynamic
nature of wireless sensor networks is take
n

into consideration in the
design of the routin
g
protocol. Link quality c
an be estimated

b
y listening to incoming
data
massages, even the ones
that are not sent to them
.

This approach does not cause nodes to consume additional energy
because all nodes in the transmission range consume energy even they are not the destination
no
des
.
For estimating link quality, window mean exponentially moving average is used.
Link
estimator computes




Packets

Received


in

t



Max
(
Packets Expected in t
,

Packets Received in t
)




Figure 1



E
ach node in the net
work ke
eps a neighborhood table
,
which has constan
t and limited
size. Routing decisions are made according to this table
.

Main aim is to keep nodes that have
high link quality in the neighborhood table
.

Link estimation is used to determine which nodes
should be c
onsidered as neighbors.
Sometimes
communication with
close nodes is at

poor
quality and
with nodes at higher distance is at good

quality.
From

Figure 1 how link quality
varies

with distance can be seen
. G
raph
in Figure
1 is

taken after an empirical observa
tion.

At
a
power level of 50
, one
node is transmittin
g and the others are listening.

As it can be seen
from figure 1 i
n the transitional region, some nodes at feet 25 have better link quality than the
ones at 15 feet.

In addition to distance, link quality
also changes with time.

Neighborhood table management policy is used to determine the nodes for which to
keep statistics since neighborhood table is constant.

For this purpose frequency algorithm is
used. Each time a packet comes from a source, if it is i
n the table its counter value is increased
by 1.If it is not in the table, counter values of existing entries are decreased by 1.

The
neighbor that has counter equal to 0 is dropped from the table and new one is added in to the
table.

Authors of [1] showed

that with frequency algorithm, %30 of nodes in the
neighborhood table has link quality greater than %75.


2.3.2 Berkeley Routing Protocol



[3]

bases its routing protocol
on

distance vector routing. However in the paper there
are several offers for cost
metric. One of them is to use the expected number of transmissions
as cost metric. Another approach
proposed by
authors of [3]
is
to apply shortest path
routing
for the nodes t
hat have high quality.

By incorporating link quality into routing decision, [3]
deals with the scenario in which long path with less transmission is better
,

in terms of energy
consumption,
than shorter path with many transmissions.



Neighborhood table manager decides which nodes to keep in the neighbor table
according to the frequ
ency algorithm and
reception
link quality that is calculated by link
estimator
.

The next node that the data will be forwarded is named as parent. Parent sel
ection
is done according to distan
c
e to sink

and send link quality
.
However as stated before link
es
timator calculates reception link quality.

That’s why
reception quality for
parent
node is
included in
route
message. When
parent
node gets the route

mes
sage, it will put the value of
reception quality in
to

send link quality
field of
the node from it recei
ved the message.



Route
Message


Number of hop
s to the sink


Mac Address

Of Parent


Reception link quality of Parent


Data
Message


Sequence Number


Mac Address

Of
Source Node


Mac Address of Destination Node


Neighborhood Table


Mac Address

of Ne
ighbor


Number of
hop
s to the
sink


Reception link

quality



Send link
quality


Child Flag



C
hild flag is used to avoid cycles. The packet is not forwarded to the neighbor from whom it
is received since the child flag is set.


Berkeley
proposal was des
igned for stationary networks but it is adaptable to mobile
sink scenario as well.

When the sink becomes mobile, it is necessary to transmit its new
location throughout the network.

This can be achieved by increasing the frequency
of route

message transmis
sion.

Since route messages inclu
de the field

number of hops to the sink
,

each
static source
node will be aware of the sink’s new location.


3. COMPARISON OF EXISTING APPROACHES


3.1 Storage Requirement

In
USC
proposal
, all nodes assign weights to their nei
ghbors

for each time domain.
As
stated in Section 2
, these weights are positively or negatively reinforced by the goodness
value
calculated by
nodes that are in direct communication with the sink.

Nodes keep entries
for their neighbors for each time domain
. If number of time domains increase, the possibility
of locating the moving sink
accurately
increases. However the tradeoff is cache size.

Cache
size increases as number of time domain increases.


UCLA propo
sal is based on directed diffusion where the si
nk broadcasts interest
messages.

Source nodes store interest messages in their cache and forward data according to
these interests.

Interest packets are smaller when compared to entries in neighborhood table

that’s why UCLA pr
otocol has no problem for stor
age.

In Berkeley proposal
,

neighborhood table has constant space.

If number of neighbors
is larger than the table size, neighborhood table manager

enables

nodes that

send packet
frequently
,

to stay in the table.



3
.2

Latency


In
USC
approach, as the sin
k moves nodes continue to transmit messages. That’s why
proposal does not put additional delay.


However in UCLA approach, nodes wait until
router come

into their transmission
range for transmitting data.

When router goes out of range, they stop transmitti
ng.

For real
time applications and urgent data transmission, this protocol is weak.


Berkeley approach does not add additional delay since nodes continue
transmitting

over multihop routing.


3.3

Reliability


If a protocol
is reliable
, it will

definitely

d
eliver the message to the destination.

Authors of [1] did not design the protocol to be reliable. Their main aim was to locate the
moving sink efficiently.

Moreover, they ignored the scenarios where reinforcement packets
can be lost.

In [2],

transmission o
ccurs

over

minimum number of hops, when source node and
router are at closest distance.

Transmission over small number of hops makes the protocol
reliable

to some extent
.
Furthermore when number of transmission decreases, the probability
of successful data

transmission will increase.
However the protocol does not take any
precaution against the possibility of ACK losses. Authors of [2] ignore the scenario where
ACK is lost and
source
nodes stop transmitting because they assume that router went
away
although

it did not.


Berkeley
protocol
is able to
adapt to changing

conditions of the network. It gives

emphasis to reducing number of
transmissions by considering link quality and distance to sink
during next node selection.

With this approach
,

authors of

[3] a
imed to minimize number of
transmission, including retransmissions.




3.4

Throughput


There is a

possibility
of tradeoff

between reliability and throughout. A reliable
protocol’s throughput can be very low. If the number of successful transmission is low,

a lot
of retransmissions

will occur in order to have a successful transmission.

T
his

approach

automatically will decrease throughput.

However if the transmission occurs successfully in
first attempts protocol
will be

reliable and
have
a good throughput

at

the same time.


In
USC
proposal, there is no metric they explicitly used. That’s why from the paper it
is not possible to comment on throughput of the protocol.


Throughput
of [
2] is better when compared to others
because;

transmission performs
over minim
um number of hops.

Berkeley approach puts more weight on to the number of hops to the sink when
compared to link quality during routing decision.
That’s why t
ransmission occurs over small
number of hops
in Berkeley protocol too.

This approach increases th
e possibility of successful
transmission and throughput.


3.5

Overhead


In USC Proposal, reinforcement packets add additional overhead to the system. If
number of time domains increases, number of reinforcement packets will increase because
reinforcement m
essages are created for each time domain.

In UCLA proposal, static nodes continue to transmit messages to the mobile router
while it is in their transmission range. Mobile router sends ACK packet to the node from
which it received data. Nodes stop sending
data to router if they do not receive these ACK
packets which add overhead to the system.


In Berkeley proposal there is no overhead. All relevant information is included in the
route message.



3.6

Energy Cost

& Network Lifetime


USC
p
roposal suggests tha
t sink must be silent in order to save energy

that is
consumed for initial broadcast of the query
.

In order to locate the sink and give routing
decisions accord
ing to the location of the sink, their protocol uses positive and negative
reinforcements.

Howev
er authors of [1] ignored the energy cost for
transmitting

these
reinforcements.

The main point of
USC
protoc
ol is to locate the silent sink.
In order to
achieve this, more weight is given to the nodes

(moles)

closer to the base station.
Inherent to
multih
op
routing closest

nodes to the sink
consume

more energy. With this protoc
ol moles

do
more work since all n
odes choose paths that lead to them
because
they are
assigned higher
weights

by reinforcements
.


Authors
of [
2] claim that

their proposal

achieves a
n increase in network lifetime. The
reason is the reduction in the number of packets.

As described in section 2, mobile router
goes
as close as possible to the source node

before
transmission
. In other words transmission
occurs over less number of hops.

UC
LA proposal

is
the extension of the Directed Diffusion
paradigm to the mobile sink scenario.

The main idea in Directed Diffusion is, when a node
initiates interest in data with certain attributes, the interest is transmitted in the network. This
process se
ts up gradients in the network to the initiator of the node (sink) who is interested in
the data. Set of gradients form the return path to the initiator. Directed Diffusion aims to
reduce energy consumption with application aware nodes that do caching and
aggregation.
However the initial route discovery phase is very expensive in terms of energy consumption.

In [2],

s
tatic nodes should send their data directly to the sink if possible or send to the closest
node that can communicate with the sink. According
to

their proposal, as shown in F
igure

2
,
node A should not waste energy for relaying B’s message if B is able to communicate with
the sink directly.
However they ignored the scenario where
B will run out of energy while
sending directly on the other hand i
f A relayed the message, there is a possibility that both B
and A live.

Another issue is nodes that are in transmission range of A and B consume energy
too although the packets are not destined to them. UCLA protocol does not handle this issue
either.




Figure 2



Berkeley approach
suggest
s

saving

energy by minimizing expected number of
transmissions.

In order to achieve this, they added link quality criteria

to routing dec
is
i
on.

By
this way they are considering the scenarios where a longer path

with fewer transmissions will
make the system consume less energy when
compared to a shorter path
that has bad link
quality and
has
a lot
of retransmissions
.


4.
ENERGY BALANCED &

LINK STATUS
AWARE ROUTING




We based our routing protocol on Berkeley

appr
oach because it enables reliability,
energy efficiency as well as constant storage space.

Link estimation for selecting neighbors,
neighborhood table management policy for keeping table size constant
, considering link
quality and

distance to sink
is extend
ed from

Berkeley

proposal
.

As mentioned in Section 2.3.3 Berkeley proposal was designed for stationary
networks
. However it is adaptable to mobile base station

scenario as well. When the sink
becomes mobile, it is necessary to transmit its new location th
roughout the network. This can
be achieved by increasing the frequency of route message transmission. Since route messages
include the field number of hops to the sink, each static source node will be aware of the
sink’s new location.

Our contribution is a
dding
number of nodes that are in danger field to
route
messages
and
neighborhood
table.

By adding this field into
route
messages, our proposal does not
consume additional energy for transmission of

remaining energy capacity
.
Main aim of
routing protocol p
roposed in this paper is to make network lifetime longer by preventing
choosing
paths

that

have

higher number of nodes that are in danger of running out of energy.

We tried to achieve energy balancing among nodes.

Each node has an initial packet processing

capacity. Each tim
e it transmits or receives
packet
, its capacity decreases by 1.
Furthermore capacity of nodes that are in the transmission
range decrease by 1 too.
When the capacity falls
below a predetermined threshold, the node
becomes in danger node.

After a node becomes in danger, while transmitting route messages
it increments the number of nodes in danger field by 1.

With this approach in a distributed
manner nodes can become

aware of the remaining energy levels

of
other
nodes in the
network.



When

base station

moves to a new locat
ion, it
makes static nodes that are closest to it
start transmitting route messages. With this approach

static nodes become aware of t
he new
location of the sink. This announcement is very important since while forwarding
data
,

nodes
consider the smallest distance to sink as well as link quality and number of in danger nodes on
the shortest path

to sink
.


Each node that transmits

route
message updates node_id field with its id

and increases
the number of hops field by its d
istance to the node from which it received the route message.
If its capacity is below threshold, it increases number of nodes in danger field by 1.

In the
system
,

node_ids are MAC addresses of nodes. Number

of hops to sink means the minimum
number of hops

required to reach to sink from that node. Number of nodes in danger field
shows

how many nodes are under the threshold capacity on the shortest path toward sink.


Route
M
essage


Number of hop
s
to the sink


Node_id of

Parent


Reception Quality

Of Parent


Number of nodes that are
in danger




Neighborhood Table


Node_id


Number of hop
s
to the sink


Reception
link

quality



Send link
quality


Child Flag



Number of nodes
that are in
danger


The challenging part is how to choose next node from the neighbo
r table.

We assigned
weights to fields.

For each neighbor

i

in the table
α X(i) + β Y(i) + γ Z(i)

is calculated. The
node that makes above equation
minimum

is selected as
next
node

to forward data
.

X(i) =
shortest
distance
from n
ode i

to sink.
(hop count)

Y
(i) = Node i’s link quality

Z(i) = Number of in danger nodes on

shortest

path from sink to node i

α = 0.5

β = 0.2

γ=0.3

We put more weight on distance to sink

by
following the approach of Berkeley authors.

In
order to see whether Link Status and Energy Ba
lanced Routing made network lifetime longer,
we did high level simulations.


5. SIMULATIONS



We performed simulations with a high level simulator
written
in Matlab.

Our routing
proposal is based on Berkeley approach in which
,

a neighbor node can be one of

the nodes
which is far away

as well

as
one of
the nodes that are close in distance
.

D
uring the simulations
we made simplifications.

In order to define connectivity between nodes we based our
simulations on binary model
. In other words, for a specified tra
nsmission range
,

nodes that are
in that range are defined as neighbors.
We

ignored MAC

effects during

the simulation.

Nodes are deployed within a circle of R = 10 units.
They
are randomly scattered as a
Poisson Process with density 0.1.
Each node sends s
ame number of packets in 1 second.

Transmission and reception
consumes

same energy.

If node can not send packet successfully,
it will stop trying to send that packet after 3
attempts. During transmission process, nodes that
are in transmission range of th
e sender, neighbors, also consume energy although the massage
is not destined to them or transmission was not successful.

Mobile sink moves around t
he circle

with predefined
intervals.

For all sink locations,
there is many to one transmission. In other wo
rds all static nodes send packets to the base
station.
Initially

each node is assigned
a predefined
packet processing capacity. Each time it
receives or transmits a packet, this capacity is decreased by 1. If the receiver is the destined
next node, its
cap
acity
decreases by 2 since it also has the responsibility
of forwarding
. This
forwarding process continues until the node at which sink

located is

reached.
If the capacity
of the node falls below the threshold, it announces itself as in danger node
. While
sending

route messages
, it increases the field number in danger by 1.

First we did simulations for Berkeley proposal where next node selection is done
according to neighbor nodes’ distance to sink and link quality.

X(i) = Node i’s

shortest

distance to sink
.

Y(i) = Node i’s link quality

α = 0.8

β = 0.2

for all i ε neighbor nodes

c
hoose
i

which has
minimum

α
X(i) +
β Y(i)

as parent(next node)

Then we did simulations for our proposal
,

where next node selection is done according
to neig
hbor nodes’ distance to
sink, link quality and

number of in danger nodes on the
shortest
path between
neighbor

nod and sink

X(i) = Node i’s distance to sink.

Y(i) = Node i’s link quality

Z(i) = Number of in danger nodes on path from sink to node i

α = 0.5

β = 0.2

γ=0.3

for all i

ε neighbor nodes

choose i which has minimum α X(i) + β Y(i)

+

γ Z(i)

as parent(next node)


5.1Simulation Results


Density
of nodes

#of
nodes

Radius
of circle

Transmission

range

Initial
capacit
y

Farmer
moves with
this angle
around the
circle

Th.

Berkeley

Network

Lifetime

Our

P
roposal

Network

Lifetime

0.1

20

10

4

100

40

30

151

148

0.1

20

10

4

100

30

30

146

150

0.1

30

10

4

100

40

30

172

160

0.1

30

10

4

100

30

30

203

180


Th. field refers to

predefined threshold.


6. CONCLUSION


We expected to get higher

results for network lifetime for our proposal. However

as it
can be seen from
section

5.1,
our results are very close to the Berkeley results.

Because of
computational power limit of the pc’s on which we worked, we simulated for only 20 and 30
nodes.

If w
e were able to do more simulations, maybe we could get more clarifying results.

Another issue is, we did simulations only for one weight combination. We gave
highest
weight to the

field

distance to sink.

It is possible that we could get better results if w
e
assigned different weights in each trial.

One of the possible reasons that made our protocol
d
oes

not outperform Berkeley proposal is, Berkeley protocol
achieve
s

load balancing
among
nodes to some extent.
If all
neighbor
nodes that have same number of ho
ps to

sink, link
quality is the criteria for next node selection. If current next node’s link quality decreases, in
the next selection

of next node,

an
other node t
hat has higher link quality

than the current one

will be chosen. This approach enables load
balancing among nodes.

During
selection of
next node,
we are consid
ering candidate next node’s shortest
distance to sink and calculating number of nodes that are in danger on the shortest path.
Potential weakness of our protocol arises from this approach. Perhaps next node is not using
the shortest path during next forwar
ding. However we are counting the number of nodes that
are in danger on the shortest path. What could be done as a future improvement is
,

to count
the danger nodes in the path which next node chooses according
to α

X(i) + β Y(i) + γ Z(i)

equation.




7.
R
EFERENCES


[1]

P.

Baruah, R. Urgaonkar, and B.Krishnamachari,


Learning Enforced Time Domain
Routing to Mobile S
inks in Wireless Sensor Fields,”

i
n
Proc. of 1st IEEE EmNetS
-
I
, 2004.

[2]

A. Kansal,
A
. Somasundara
, D. Jea, M.
Srivastava, a
nd D.
Estrin,

"Intelligent Fluid
I
nfrastru
cture for Embedded Networkings,"

in
Proc. of 2nd ACM MobiSys
, 2004.


[3]

A Woo, T. Tong, D. Culler, “
Taming the Underlying Challenges of Reliable Multihop

Routing in Sensor Networks
,”

in

SenSys’03,

2003

[4]

K.
Kar, M. Kodial
am, T. V. Lakshman, and L.

Tassiulas, "Routing for Network Capacity
Maximization in Ener
gy
-
constrained Ad
-
hoc Networks"

in
Proc. of 22nd IEEE INFOCOM,
2003

[5]

J.

Hubaux and J. Luo,
“Joint Mobility and Routing

for Lifetime Elongation in Wireless
Sensor Networks
,”
in

IEEE INFOCOM,
2005