A New TDMA Based Sensor Network for Military Monitoring (MIL-MON)

eggplantcinnabarMobile - Wireless

Nov 21, 2013 (3 years and 4 months ago)

66 views


1



Abstract

Wireless sensor networks

(WSN)

is a new network
family that enables to create smart environments. Although WSN
has many application areas, military a
pplications of WSN are
very interesting. In this paper, a new TDMA based sensor
network for military monitoring

(MIL
-
MON) is proposed. MIL
-
MON is developed to operate in large areas for acceptable
lifetime periods. In order to realize MIL
-
MON, distributed
time
scheduling mechanism, topology construction algorithm and
rescheduling algorithm is proposed. Simulation results have
shown that MIL
-
MON
can operate in large areas, in acceptable
lifetime and delay constraints.


Index Terms

sensor networks, TDMA, mili
tary monitoring.

I.

I
NTRODUCTION

n recent years, with the pace of the developing micro
-
electro
-
mechanical systems (MEMS) technology, it has
been possible to integrate battery
-
operated sensor, computing
power and low power wireless communication components
i
nto one small size device. A sensor node collects data from
the environment continuously. The data collected by only one
node are nearly useless, but the collaborative work of
thousands of these nodes can be used to collect process and
send the data about
the environment. The network that is
composed of these wireless sensors is called wireless sensor
network (WSN). The potential applications of wireless sensor
networks are highly varied. Environmental monitoring,
condition based maintenance, smart spaces,
military, precision
agriculture, transportation, inventory tracking are just some of
the sample application areas [1]. One of the most common
application areas of sensor networks is military. In this paper,
a new TDMA based military monitoring sensor netwo
rk
system (
MIL
-
MON
) is proposed.

Because of the unattended structure of the sensor nodes, the
scarcest resource in sensor networks is power. Power
consumption can be divided into three domains
,

as
sensing,
communication, and data processing
domains and d
om
inant
factor in energy consumption for sensor nodes is
communication

[2].
. Not only transmission but also receiving
is the main cause of energy waste. The easiest way to reduce

Ilker Bekmezci is with Bogazici University Computer Engineering
Department. He is working in Turkish Air Force Academy

(e
-
mail:
i.bekmezci
@
hho.edu.tr
).

Fatih ALAGOZ is with Bogazici University Computer Engineering
Department. (e
-
mail: fatih.a
lagoz@boun.edu.tr)


energy consumption is to turn the radio off, when it is not used
[3]. Fixed all
ocation methods, TDMA or FDMA is extremely
suitable for this kind of network. There are some sensor
networks based on TDMA such as LEACH [4], SMACS [5],
two
-
tiered architecture [6], PACT [7].

LEACH is a self
-
organizing, adaptive clustering protocol
that u
ses randomization to distribute the energy load evenly
among the sensors in the network [4]. Although LEACH can
reduce power consumption, there is a problem with the
assumptions of LEACH. LEACH assumes that each node can
hear each other. So LEACH is not su
itable for using in large
areas. SMACS is another sensor network that uses TDMA. In
fact SMACS uses TDMA in addition to FDMA. After a series
of handshaking signals, neighbor nodes can agree on a
frequency and time pair to construct a link. SMACS produces
a

scalable and reliable flat network. However, SMACS needs
FDMA as well as TDMA, but sensor nodes are so tiny and
limited that current sensor nodes cannot meet the requirements
of SMACS. PACT uses Unifying Slot Assignment Protocol
(USAP) which is a TDMA sch
eduling scheme for on demand
ad hoc networks. USAP is adopted for sensor networks in
PACT. However, USAP is originally developed for ad hoc
networks and PACT is not fully successful in power
consumption for sensor networks.

Another TDMA based
sensor networ
k proposal is two
-
tiered structural health
monitoring wireless sensor network architecture. According to
this structure, there are some fixed cluster heads and sensor
nodes. Sensor nodes are clustered around cluster heads. This
network is designed to use f
or monitoring buildings. It can not
be used in large areas.

In this paper, a new TDMA based sensor network system,
which can be used for military monitoring systems in a
relatively large area, is presented. The coverage of the network
can be in the order
of kilometer squares. In order to realize
MIL
-
MON
, time synchronization, time slot assignment,
topology construction algorithms are developed. In order to
enhance the network in terms of delay and power usage,
rescheduling and data indicator slot mechanism
s are proposed.

The organization of the paper is as follows. In Section 2,
preliminary works on TDMA based sensor networks is
presented. In Section 3, the basic mechanisms and
enhancements for TDMA based sensor network will be
proposed. In Section 4, the
performance results of newly
proposed algorithms will be outlined. Section 5 states the
conclusion and future work.

A New
TDMA Based Sensor Network for
Military Monitoring

(
MIL
-
MON
)

İ
lker

Bekmezci
, Fatih
Alagöz

I


2

II.

PRELIMINARY

WORK

TDMA is a technique to share the medium among multiple
users. Time frames are divided into time slots and each slot is
assi
gned to a user. In this way, when a user wants to access to
the medium, it uses the medium in its own slot. TDMA has
many advantages for sensor networks. With the help of
TDMA, node can know exactly when it should use its radio
circuit and sensor node can
turn off its radio circuit to save
more power. Node uses transmitter or receiver only when it is
needed. This is why TDMA is very attractive for sensor
network applications. There are some sensor network systems
based on TDMA, such as LEACH, SMACS, PACT, t
wo
-
tiered
building monitoring system.

LEACH is a self
-
organizing, adaptive clustering protocol
that uses randomization to distribute the energy load evenly
among the sensors in the network [
4
]. In LEACH, the nodes
organize themselves into local clusters, a
round one node acting

as the local base station or
cluster
-
head
. A cluster head is
responsible for collecting data from its sensor nodes and
sending the data to a central sink. In this case, cluster heads
use more energy than the regular nodes. In order to

prevent
quick power drain of cluster heads, cluster heads are changed
periodically. The complete algorithm of LEACH is as follows:


While (there is at least one node with power) do


Cluster_Head_Selection_and_Advertisement


Cluster_Construction


Create
_and_Send
_
Time_Scheduling


While (data transmission period is not exceeded)


Data_Transmission


Wend

Wend


Comparison study shows that LEACH achieves between 7x
and 8x reduction in energy compared with direct
communication and between 4x and 8x reduc
tion in energy
compared with MTE routing. In addition to reducing energy
dissipation, LEACH successfully distributes energy
-
usage
among the nodes in the network such that the nodes die
randomly and at essentially in the same rate [
8
].

Although LEACH achie
ves power consumption decrease
dramatically, the assumption of LEACH system is not realistic.
According to LEACH, each node can receive signals from all
the other nodes. However, it is not possible in every
environment, especially in larger areas. If senso
r network is
designed to operate for relatively large areas as in
MIL
-
MON
,
every node cannot receive all signals.

Another unrealistic assumption of LEACH is about the
existence of data. LEACH assumes that every node has data to
send every time. However, i
n most of the time, there is no data
to send in sensor network. For example,
MIL
-
MON

is
designed for intrusion detection and the existing of an intruder
is not likely to occur most of the time.

Another disadvantage of LEACH is the need for changing
cluste
r heads. Cluster heads are responsible for collecting data
from the sensor nodes and relay its data to central sink.
Because of these duties, cluster heads have to spend more
power than regular nodes. The solution of this problem is to
change cluster heads

periodically. Unfortunately, the solution
of quick power drain of cluster heads is a cause for spending
power. Periodic cluster head selection and reorganization of
network for the new cluster heads may consume considerable
energy.

Two tiered architecture

for structural health monitoring is
another sensor network model that uses TDMA technique.
This sensor network architecture is designed for monitoring
structural health of buildings. It is also based on clustered
approach. Sensor nodes are clustered and c
luster heads are
special nodes called local site masters (LSM). LSMs are
deployed one by one manually and the power source of LSMs
is the power source of the building. So, there is no power
limitation on LSM. LSMs collect the data from their sensor
nodes.
LSMs construct a higher level of network and send the
collected data to a center. According to analysis of two tiered
sensor network, the expected lifetime of network is about 18
months

[6]
. It is very impressive and acceptable. However,
because of LSM lim
itations, the application of this network in
open areas is extremely difficult. There must be lots of LSM
units and these LSM units should be deployed manually.
Moreover, LSMs must have unlimited source power. Because
of these limitations, it is not suitab
le to use this system as
military monitoring system.

SMACS is a TDMA based sensor network that produces flat
topology. In fact, SMACS uses TDMA and FDMA at the same
time. After deployment of sensor nodes, they wake up at
random times and they start to lis
ten to a certain frequency.
This frequency can be called as establishment frequency.
After listening to the establishment frequency for a random
period of time, if it can not receive any invitation signal from
other nodes, it transmits a short invitation
packet that contains
some basic data about the node. This packet is called TYPE1
message. If a node can receive a TYPE1 message from other
nodes and if it wants to establish a link, it responds TYPE1
that it will be an invitee. This response is TYPE2 messa
ge.
There may be more than one TYPE2 messages for a TYPE1
message. At that particular time, the node that transmits
TYPE1 message has to decide which node it should choose.
After its decision, it sends a response to TYPE2 that includes
data about the node
that is selected. This is TYPE3 message.
At the end, TYPE4 is sent by the invitee. In TYPE3 and
TYPE4 phases, nodes agree on a certain schedule to
communicate. This schedule is a time and frequency pair. In
this case, the collision probability of schedules

can be
minimized.

As times goes on, there will be some subnets in the wireless
sensor networks. These subnets unite with each other, when
new links are established. At the end, all sensor nodes
construct a connected network.

SMACS is an infrastructure bu
ilding protocol that forms a
flat topology for sensor networks. It is a distributed protocol

3

which enables a collection of nodes to discover their neighbors
and establish transmission/reception schedules, with
out any
need for master node [5
]. Although flat

topology has some
advantages, it has also some difficulties. Flat topology requires
a separate network layer, while cluster based approach has
implicit network layer in itself. However, the most serious
handicap of SMACS is that there is no sensor node th
at can
support the requirements of SMACS. According to SMACS,
node must support frequency multiplexer and there is no such
a node up to now. A sensor network should be applicable on
the existing node models.

PACT (Power Aware Clustered TDMA) is a TDMA bas
ed
sensor network that uses passive clustering. Its TDMA
scheduling scheme is based on Unifying Slot Assignment
Protocol (USAP) which is a time scheduling algorithm for on
demand ad hoc networks. PACT can be considered as a sensor
network version of USAP i
n terms of time scheduling. This is
why it is not fully optimized for power consumption. In PACT,
every node must listen to all the control slots of a time frame.
The number of control slot must be equal to time slot number
in a time frame and it means eve
ry node must spend
considerable amount of power in every time frame. Another
important question about PACT is time synchronization. Every
TDMA based system must synchronize time and PACT does
not propose a system for time synchronization and it does not
ta
ke into account time synchronization overhead.


III.

MIL
-
MON

S
YSTEM
M
ECHANISMS

A.


Overview

MIL
-
MON

is proposed to monitor a relatively large area
against intruders and send the data about intruders to sink as
soon as possible. The main design considerations of


MIL
-
MON

are to be able to operate in large areas, to minimize
pow
er consumption, to reduce delay.

In order to satisfy design considerations,
MIL
-
MON

is
designed as a TDMA based multihop sensor network. The sink
of
MIL
-
MON

is supposed to have
a
high
range transmitter so
that every node can receive its broadcast signal.

MIL
-
MON

includes global time synchronization, distributed
time slot assignment, topology construction mechanisms. In
addition to these basic mechanisms, there are data slot
indication
mechanism to save more power and rescheduling
mechanism to reduce delay. Before introducing the
mechanisms of
MIL
-
MON
, a sample application and basic
assumptions of
MIL
-
MON

is presented.

B.

Sample Application and Basic Assumptions

The potential applications o
f wireless sensor networks are
highly varied. Environmental monitoring, condition based
maintenance, smart spaces, military, precision agriculture,
transportation, factory instrumentation, inventory tracking are
just some
of the sample application areas
. B
ehaviors of one
specific sensor node can be totally different from the others
even under the same conditions. In this case, it is clear that
protocols that are developed for a WSN may not be optimal
for another WSN [
3
]. Sensor network systems should be
app
lication specific [
4
]. A sample application scenario can
help to understand the assumptions and mechanisms of

MIL
-
MON
.

In a typical military monitoring scenario, large numbers of
unattended, limited powered sensor nodes are deployed near
the borders o
f a base. Sensor nodes organize themselves, so
that, when sensor nodes detect an intruder, they send their data
to a sink. In this way, soldiers can defend against the intruders.
In most of the cases, sink is a PDA or a laptop.


Here are the assumptions of

the sensor network system:

1)

Sensor nodes will be immobile. Mobile cases can be
investigated for further analysis.

2)

Power consumption model of sensor node is the same as
described in
[9]
. This model is one of the mostly used
models in sensor network simulati
on analysis researches.

3)

Radio channel is symmetric.

4)

Sink node’s power source theoretically infinite.

The assumptions, which have been mentioned above, are
valid for most of the sensor networks. There are some
additional
assumptions specifically for
MIL
-
MO
N
.

Sink node has high range transmitter as well as low range
transmitter. In this way, sink can use its low range transmitter
to communicate with its neighbor nodes and it can send
broadcasts for all nodes. All the sensor nodes can receive
broadcasts of t
he sink.

Sink node has GPS. In this way, c
lock drift of sink is near
zero
[10]
.

C.

Basic Mechanisms

There are some basic mechanisms for operating
MIL
-
MON

properly. These are time synchronization, distributed time
scheduling and topology construction mechanism
s.

1)

Time Synchronization

According to assumption of this sensor network, every node
in the system can receive the signals of the sink. Sink transmits
a broadcast signal to sensor nodes at the beginning of each
time frame. These broadcast signals synchronize

the network.
This time synchronization scheme is very similar to LEACH,
however there are a few differences based on the differences of

system assumptions of
MIL
-
MON

and LEACH. The number
of cluster in LEACH is a piori in the system. Each cluster head
syn
chronizes

its own cluster.
MIL
-
MON

has only one cluster
and the head of cluster is the sink. Only the sink
synchronizes
the whole sensor nodes with its broadcast signal.

2)

Distributed Time Scheduling Mechanism (DTSM)

Almost all sensor network architectures t
hat use TDMA
produce its time schedule centrally. Cluster head collects the
data about its sensor nodes and produces the time schedule of
its cluster. Time schedule is sent to nodes by the cl
uster head.
The main assumption

of this system
is that w
hen clust
er head
transmits a s
ignal, sensor nodes can receive or vice versa.

However, in
MIL
-
MON
, although the signal of the sink can
be received by all sensor nodes, the sink can not receive the

4

signals of all sensor nodes. In this case, sensor nodes can not
send
their join request signal. If distributed time scheduling
algorithm is used, there is no need to send a signal from sensor
node to the sink directly.

In centralized time scheduling algorithm, cluster head
collects all the requests from sensor nodes in a c
ertain
protocol. In most of the systems, this protocol is contention
based. If the number of sinks is limited and number of sensor
nodes for each sink is very high, the traffic from sensor nodes
to sink will be heavily loaded. Contention based protocols ca
n
not achieve high efficiency under heavy traffic. This can be a
serious problem for power sensitive systems. However, in
distributed time scheduling algorithms, there is no need for
communication between nodes and the sink directly. This
leads to power sa
ving.

According to DTSM, after sensor nodes are deployed, every
node selects a random time slot as its own slot. It listens to all
time slots in the first time frame and only in its own slot, it
transmits a special signal. If it receives a jammed signal,

it
means there is a collision at that particular slot. The node
collects all the collision slots. In the next time frame, it
transmits a signal at the collision slots. In the same time frame,
it listens to its own slot. If it receives a signal, it means i
t has
the same slot with another node’s slot so that their signals are
jammed. In this case, it sleeps. If it does not receive any signal
at its own slot, it can use that slot and it continues to operate.

This protocol is simple and consumes low power. Ho
wever,
it does not result in a complete solution. Some nodes have to
sleep. Fortunately, most of the time, sensor nodes are deployed
densely and the non
-
existence of a small number of nodes can
be tolerated.

3)

Topology Construction

After distributed time sc
heduling, the second step for the
proposed sensor network is topology construction. Many
TDMA based sensor network systems use star topology. All
the nodes are directly connected to cluster head. However, in
large areas, multi hop structure consumes less p
ower than one
hop. There is a trade off between multihop and one hop
structures. If the distance is less than 30

m
eters
, direct
transmission is more efficient than multihop transmission.
However if it is longer than 30

m
eters

mu
ltihop strategy can
save ene
rgy

[11]
.

If sensor network is aimed to cover a large area, average
distance between sensor nodes and the sink is supposed to be
much longer than 30m. In this case, multi hop strategy is an
essential part of
MIL
-
MON

architecture.

A distributed topology co
nstruction algorithm is proposed to
support multi hop strategy. Time scheduling algorithm is also
distributed. In this way, the only scalability constraint of the
system is the transmission range of the sink.

The basic structure of the algorithm lies on h
andshaking
signals. After getting a proper time slot, sensor nodes listen to
their neighbors to catch hop number advertisement signal. At
the beginning, only the sink sends hop number advertisement
signal. Hop number of the sink is zero. If a sensor node c
an
catch advertisement signal or signals, it chooses one of them as
its predecessor. The simplest choice is the closest predecessor
candidate. In the next time frame, it sends its own
advertisement signal. If it receives hop number as
h

from its
predecesso
r, it advertises its own hop number as
h+1
. In the
same signal, it also sends the node number of its predecessor.
This signal is received by the predecessor. If the predecessor
receives a hop advertisement signal with its node number, it
understands that t
he owner of the signal has become its child.
In the third time frame, it listens to its all neighbors to learn
which nodes became its children. Because in the third time
frame, the children of the sensor nodes advertise their hop
number and predecessor nod
e number.

Figure
1

shows a sample run of the distributed topology
construction algorithm. Suppose that A has caught an
advertisement from
X
with hop number
h
. It advertises its own
advertisement signal with the parameters
X

and
h+1
. B catches
this signal
and other advertisement signals, if any. If A is the
closest node among the nodes that have sent advertisement
signal, B sends its advertisement signal with parameters A and
h+2
.

If maximum hop number of the network is
H
, constructing
the topology takes
H
+1

time frame for the whole network. It is
fast, simple and scalable. It consumes low power and it results
in minimum hop paths, which is a good approximation for
minimum power consumption.

4)

Rescheduling

Although power consumption is generally the most
imp
ortant design issue in sensor networks, there are some
other design considerations like delay. In some applications,
delay between time of event and time of data arrival to the sink
is not crucial. However, if sensor network is designed for
military purpos
es, delay can be very important. In a military
monitoring system, the existence of
intruder

should be
reported as soon as possible.

If sensor node is
h

hop away from sink, and one time frame
is t seconds, the worst delay is
t*h

and
t*h

delay can not be
ac
ceptable in some cases. Suppose that
t

is one second and
sensor node that has sensed the arrival of
intruder

is 7 hops
away. If the average distance between neighbor nodes is 20 m.,
intruder

is 140
-
150 m. away to the sink and the sink can know
the existenc
e of
intruder

after 7 seconds which can be enough
for the
intruder
to
harm

the soldier that holds the sink.

Reducing delay is possible by the help of assigning time
slots carefully. The rule is that smaller hop numbered nodes
should get higher slot numbers
. For example, if a sensor node,

A sends adv(X,h+1)

A receives adv(A,h+
2
) and
understands that B is its child.

A
receives
adv(Y,h) from X

B receives adv(X,h+1)

B sends adv(A,h+
2) and

advertises that A is its
predecessor.


Fi
g. 1.
A sample run of topology construction

algorithm.



5

say A is 3 hops away and its slot number is 320, slot number
of another node with 2 hop number, say B, should be greater
than 320. In this case, node A can send its message in 320
th

slot and node B can receive at 320
th

slot
. In the same time
frame, node B can send the data of node A to its predecessor.

In order to realize to reschedule, time frame is divided into
m
sub time frames. If the whole time frame has
n
slots, a sub
time frame has
n/m
slots. The slot number assigne
d to a node
with hop number
h
, must be in
(m
-
((h
-
1) mod m))
th

sub time
frame. According to this formula, the nodes with 1 hop number
will have a slot number from
m
th

sub time frame. The nodes
with 2 hop number will have a slot number from
m
-
1
th

sub
time
frame. In this way, the slot number of consecutive hop
numbered nodes will belong to consecutive sub time frames.
Figure
2

shows the structure of this approach.

In order to assign proper time slots, node must know its hop
distance to the sink. In this case
, rescheduling algorithm can
not be run before distributed time scheduling algorithm.
Rescheduling should be placed between DTSM and topology
construction.

After running distributed time scheduling algorithm, non
sleeping nodes has valid time slot. Nodes l
isten to its
neighbors to receive an advertisement signal which includes
the hop number and node number of the sender. When a node
receives a hop advertisement signal, rescheduling is started.
The advertisement signal includes the hop number of the
sender.

In this way, the receiver of the advertisement signal
can calculate its hop number. In addition to this, node can get
the number of sub time frames
m

from the sink’s
synchronization signal.
T
he node can calculate the sub frame
number with the formula
(m
-
(
(h
-
1) mod m))
. It selects a
random time slot in the
(m
-
((h
-
1) mod m))
th
sub time frame and
sends a broadcast signal that includes the information about
the selected time slot. The node sends this information in its
original time slot that was assigned with

DTSM mechanism.
All the neighbors of the node listen to its signal and check
whether there is collision about its new slot number or not. If
there is a collision, the nodes that catch the problem send a
signal in their own time slots. The node that tries
to reschedule
listens to its all neighbors in the next time frame. If it gets a
signal from the others, it sleeps. If it does not get any signal
related with its new time slot, it means there is no problem and
it starts to use its new slot and continues wi
th topology
construction procedure. Figure
3

shows the signal handshaking
required for rescheduling. The figure starts with the node that
has got a slot with DTSM mechanism.


An example helps to understand rescheduling clearer. Let us
assume that the node
s in Figure
4

are one hop away from
its
consecutives. In this particular network, time frame has 300
time slots. The first slot is reserved for time synchronization
broadcast of the sink. Let us assume that DTS_SN algorithm
has been run and the assigned t
i
me slots are as in the F
igure
4(
a).


According to rescheduling algorithm, the sink transmits a
broadcast signal with hop number 0. A receives this and
calculates its sub time frame and selects a random time slot in
that particular sub time frame. In this
example, A must get a
time slot between 300 and 201. Let us assume that A selects
250
th

time slot. It advertises its new time slot in 82
nd

time slot
(its current time slot. The sink and node B listens to 82
nd

time
slot for new time slot of A. If the sink a
nd the nodes with hop
number 2 listen to all its neighbors and collect the newly
requested time slots. If they determine a collision for new time
slots, they send a warning message in the slots that request the
new time slots. In this example, neither the
sink nor B receives
a collision for the new time slot of A. Node A does not
receives a warning signal at 82
nd

time slot
. In this case, new
time slot
number
of A becomes 250. After rescheduling
algorithm, topology construction mechanism is run with the
new
time slots. So, A sends an advertisement signal with hop
number 1. The same procedure is run for all nodes. An
example resulting network is presented in Figure
4
(b).

The relay of an event from D to the sink takes 737 time slots
for non
-
rescheduled network
in Figure
4(a)
. However, it takes
only 305 time slots for the rescheduled network in Figure
4(b)
.



Sub time frame numbers

Hop numbers

(m
-
1)
th
……. 1
st


m
th

one time frame divided into

m sub time frames

2

……. m

1


Fig.
2
.
Rescheduling structure. There are m sub time frames in one time
frame, and each sub time frame has n/m slot
s, if total number of slots in time
frame is n.


If it gets a signal from
neighbors, it sleeps.

If it does not get, it
changes its time slot and
continue
s with topology
construction.

If there is collision, it
sends a signal to A.

If there is no collision, it
remains silent.

Receives new time slot
number and checks
whether there is
collision.

Calculates its sub time frame,
selects a time slot randomly an
d
sends its time slot number to its
neighbors.

Gets an adv. that
includes hop number
of sender.

Neighbors of A

Node A


Fig. 3. Rescheduling mechanism.


Sink
,
(
2
)


A, (
82
)

B, (9
1)


C, (1
98
)

D, (
2
45)






(a)


Sink
,
(
2
)


A, (250)

B, (142)


C, (45
)

D, (245)






(b)

Fig. 4.

Example network and time slots (a) Before rescheduling
.

(b) After
rescheduling
.


6

IV.

P
ERFORMANCE
R
ESULTS

It is very difficult to analytically model the interactions of
sensor nodes, even for the limited number of nodes. In order to
investiga
te performance results of MIL
-
MON, simulation
method is used. Although there are some network simulators
that can simulate wireless networks, there is no
built
-
in sensor
network module
.

In this paper, a new simulator has developed
to simulate MIL
-
MON.

Simulation p
arameters are listed in Table1.

Energy
consumption model is the same as
in
[9]. Energy of a node is
assumed to be supplied with 15 mg. Ni
-
Cd battery which can
support 2 J [12].

Performance results are discussed in three domains. These
are perfo
rmance of DTSM, network lifetime and delay.

A.

Performance of DTSM

DTSM is a distributed time scheduling mechanism to assign
proper time slots to sensor nodes. However, DTSM can not
assign time slots for every nodes. The nodes that can not get
time slot sleep

and do not join into network. The ratio of
sleeping nodes is an indicator of performance of DTSM.
MIL
-
MON is simulated for 1000 slotted time frames. 1000 slotted
MIL
-
MON is investigated for non
-
rescheduled, rescheduled
with 5 sub
-
slotted, 10 sub
-
slotted a
nd 20 sub
-
slotted versions.

Figure
5

shows percentage of sleeping nodes for different
number of nodes

and different number of slots
.

Percentage of
sleeping nodes increases linearly with the
increasing of
number of nodes

and slots
.
The increasing rate is
higher for
MIL
-
MON with 500 time slot. Sleeping node rate is
acceptable for 1000 and 1500 time slots. However, sleeping
node rate is high for 500 time slots.

The same cases are investigated with non
-
rescheduled 1000
time slots, resheduled with 5 sub
-
slots,

10 sub
-
slots and 20
sub
-
slots. The results are presented

in Figure 6
.

Figure 6

shows that rescheduling increases sleeping node
ratio linearly.

B.

Network Lifetime

Network lifetime is one of the most common performance
metrics of sensor networks.
However, ne
twork lifetime
definitions vary. In this paper, network lifetime is the time that
the first sensor node exhausts its energy. Network lifetime of
MIL
-
MON is discussed with the system load. In this
experiment, number of nodes is 2000, number of slots is 1000
.
The results are in Figure
7
.

In Figure 7
, system load is the number of events in one
second. If there is no event, the first node exhausts after more
than 35000 seconds. Network lifetime decreases with the
increasing of system load. MIL
-
MON is designed
to be used
for military purposes and the attack of
intruder
is not so
common in the battlefield.
MIL
-
MON can operate for more
than 10 hours. MIL
-
MON can be used for securing the
environment especially for
one
night operations.

0
1
2
3
4
5
6
7
8
9
10
2
2,5
3
3,5
4
4,5
5
Number of nodes (x1000)
% of sleeping nodes
1000
1000-5
1000-10
1000-20

Fig. 6
.
Sleeping node ratio for different number of nodes with rescheduling.

0
2
4
6
8
10
12
2
2,5
3
3,5
4
4,5
5
Number of nodes (x1000)
% of sleeping nodes
500
1000
1500
Fig.
5
. Sleeping

node ratio for different number of nodes.

0
5000
10000
15000
20000
25000
30000
35000
40000
0
0,25
0,50
1
System Load
Time (s)
Fig. 7
.
Network lifetime for different system loads.

TABLE

I

SIMULATION PARAMETER
S

Parameter

Default values

Power needed for radio
electronics

per bit

50 nJ

Power for receiving

per bit

50 nJ

Power
for transmitting

per bit

50nJ + 10pJ*d*d

(d is distance)

Max. range of node
s

30 m.

Power in one node

2 J

Simulation area diameter

1000 m.

Position of the sink

Center of the area.

Time for one

time slot

1 ms.

Bit

rate

1
Mbps.



7

C.

Delay

Delay is very important

for military applications. Long
lifetime is not enough, if it gives information too late. MIL
-
MON uses TDMA and constructs a tree. It means delay of
events is a function of distance from event to sink. Delay is
investigated for different distance regions.

Network area is
divided
into

10 regions. The first region is 10m. distant from
the sink. The second is between 10
-
20m. distant from the sink
,
and so on
. Figure
8

shows average delay of events for different
time slot numbers in these regions.


When distanc
e between the sink and event increases, delay
increases dramatically. MIL
-
MON is an intruder detection
system and early alarm is critical. MIL
-
MON should give
information as soon as possible, even if event is far from the
sink. Even 500 slot is not accepta
ble. If
intruder

is 500m.
distant and if MIL
-
MON gives information after 5 seconds, it
may be too late.

Rescheduling is proposed to reduce delay. The same
experiment is repeated with 1000 slot non
-
rescheduled, 1000
slot rescheduled with 5, 10 and 20 sub
-
sl
ot MIL
-
MON. T
he
results are in Figure 9.




Rescheduling improves delay very successfully. Delay in
rescheduled with 20 sub
-
slot is 18 times smaller than non
-
rescheduled MIL
-
MON
, especially in the 10
th

region. (the
border of the sensor network area.)


V.

C
ONC
LUSION

In recent years, sensor network is one of the hottest topics in
wireless communication area. In this paper, a new TDMA
based
sensor networks for
military monitoring (MIL
-
MON) is
proposed. The most important design objectives of MIL
-
MON
are to prolo
ng network lifetime, to reduce delay and to be able
to operate in large areas.

In order to realize MIL
-
MON, distributed time scheduling,
topology construction and rescheduling mechanisms are
studied
. Simulation
results have

shown that network lifetime
of M
IL
-
MON is
long enough to be

used as military
monitoring
.

Delay problem sourced form TDMA structure
can be handled by rescheduling mechanism and delay is
reduced by up to 18 times.

When a MIL
-
MON node exhausts, the network system is
broken in this version o
f MIL
-
MON. However, sensor nodes
are prone to failure and sensor network should be able to fix
itself, when a node fails. As future work, a maintenance
algorithm will be developed.

R
EFERENCES

[1] Deborah Estrin, Lewis Girod, Greg Pottie, Mani Srivastava,

“Instrumenting the World with Wireless Sensor Networks”,
International
Conference on Acoustics, Speech, and Signal Processing (ICASSP 2001),
Salt
Lake City, Utah, May 2001.


[2] I. F. Akyildiz and W. Su and Y
. San
karasubramaniam and E. Cayirci.

Wireless Sensor Networks: A Survey
.

Computer Networks” (Amsterdam,
Netherlands: 1999),

38(4):393
-
422, 2002
.


[3] E. Shih, S.
-
H. Cho, N. Ickes, R.

Min, A. Sinha, A. Wang, and A.
Chandrakasan, “Physical layer driven algorithm and protocol design for
energy
-
efficient wireless sensor networks”, In Proc. ACM MOBICOM, pages
272
--
286, Rome, Italy, July 2001.


[4] W. Heinzelman, A. Chandrakasan, and H. Ba
lakrishnan, “An
Application
-
Specific Protocol Architecture for Wireless Microsensor
Networks”, IEEE Transactions on Wireless Communications, Vol. 1, No. 4,
October 2002, pp. 660
-
670.


[5] K. Sohrabi, J. Gao, V. Ailawadhi, and G. J. Pottie, “Protocols for s
elf
-
organization of a wireless sensor network”, IEEE Personal Communications,
vol. 7, pp. 16
-

27, October 2000.


[6]
. Kottapalli, A. S. Kiremidjian, J. P. Lynch, E. C., T. W. Kenny, K. H.
Law, Y. Lei,

Two
-
tiered wireless sensor network architecture for
structural
health
”,

SPIE'03.


[7]
Guangyu Pei

and Charles Chien
,


Low power TDMA in large wireless
sensor networks

, MILCOM 2001
-

IEEE Military Communications
Conference
,
no.
1
,
October

2001
,
pp. 347
-

351


[8] W. Heinzelman, “
Application
-
Specific Protocol Architectures for
Wireless Networks”,

PhD Thesis,
Massachusetts Institute of Technology
,
June 2000.


[9]

Wendi Heinzelman, Anantha Chandrakasan, and Hari Balakrishnan,
“Energ
y
-
Efficient Communication Protocols for Wireless Microsensor
Networks”, Proc. Hawaaian Int'l Conf. on Systems Science, January 2000.

0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
1
2
3
4
5
6
7
8
9
10
Region number
Delay (ms)
Normal 500
Normal 1000
Normal 1500
Fig. 8
.
Delay for different regions for different number of slots.

0
2000
4000
6000
8000
10000
12000
14000
1
2
3
4
5
6
7
8
9
10
Number of region
Delay (ms)
Normal 1000
Re 1000-5
Re 1000-10
Re 1000-20
Fig. 9
.
Delay for different regions for different number of slots.


8


[10]

J. Mannermaa, K. Kalliom
aki, T. Mansten, and S. Turunen,


Timing
perfo
rmance of various GPS receivers”

In
Proceedin
gs of the 1999 Joint
Meeting of the European Frequency and Time Forum and the IEEE
International Frequency Control Symposium
, pages 287

290, April 1999.


[11]

Rex Min and Anantha Chandrakasan, “Energy
-
Efficient Communication
for Ad
-
Hoc Wireless Sensor Netw
orks”,
35th Asilomar Conference on
Signals, Systems, and Computers
, vol. 1, November 2001, pp. 139
-
143.


[12] J. Frieman.

Portable Computer Power Sources


, In Proceedings of the
Ninth Annual Battery Conference on Application and Advances, pp. 152
-
158,
J
anuary 1994.