An Adaptive Coordinated Medium Access Control
for Wireless Sensor Networks
Jing
Ai, Jingfei Kong,
Damla
Turgut
Electrical and Computer Engineering
University of Central Florida
Orlando, FL
32816
Email: {jingai, jfkong, turgut} @cs.ucf.edu
Abstract-
In this paper, we have developed Adaptive Coor-
dinated Medium Access Control (AC-MAC), a contention-based
Medium Access Control protocol for wireless sensor networks.
To
handle the load variations in some real-time sensor applications,
ACMAC introduces the adaptive duty cycle scheme within
the framework of sensor-MAC (S-MAC). The novelty
of
our
protocol is that it improves latency and throughput under a wide
range
of traffic loads while remaining
as
energy-efficient as
S-
MAC. We illustrate such optimized trade-offs of AC-MAC via
extensive simulations performed over wireless sensor networks.
Our
simulation
results
show that AC-MAC is
as
energy-efficient
as S-MAC while its latency and throughput are always trying
to follow the classic
IEEE
802.11
MAC (no duty cycle), which
outperform the S-MAC (fixed duty cycle), specially under the
heavy load.
I.
INTRODUCTION
Recently the convergence of
micro-electro-mechanical
sys-
tem (MEMS) technology, wireless communications and digital
electronics has enabled the development of small size, low
cost, low power sensor nodes that are capable of sensing,
computing and communicating untethered in short-distances
[
11.
Compared with traditional sensors without communication
capability, these nodes facilitate random deployment inside
or
close to the target with collaboration among them to fulfill
tasks in a remote environment
or
even inaccessible terrain.
These features ensure a wide range of promising applications
for
wireless sensor networks based on collaborative efforts
of a large number of these nodes: battlefield surveillance,
environment monitoring, vehicle tracking, smart environment
and
so
on. There are some important constraints for realization
of wireless sensor network applications. One of the most
important considerations is the requirement for low power
consumption since sensor nodes carry limited power sources
and they are often expected to work in a remote
or
inhos-
pitable environment without attendance during their lifetime.
Therefore, protocols for wireless sensor networks must focus
on energy-efficiency.
The primary consumers of energy in wireless networks are
the RF communications
[2],
which are very costly compared
to other electrical hardware functions including instruction
execution and
so
on, the efforts of energy-minimization tech-
niques are focused on the following approaches:
0
Coordinating the transmission range dynamically to
nique is termed as
“power control”
[3], [4].
Reduced
transmission power also promotes the spatial reuse by
maintaining connectivity at the lowest level; however,
it lengthens the routes which can affect the end-to-end
reliability of flows.
Reducing the volume
of
data that needs to be trans-
mitted:
Instead of end-to-end traffic model in other
networks, it is possible to intelligently combine
or
aggregate data into small set of information and make
contribution towards energy savings
[5], [6].
Further-
more, by combining these unreliable readings, it can
produce more accurate signal by improving the common
signal and reducing the noise.
Powering down the nodes’ antennas during times
of
in-
activity:
Such inactive states can be explored in different
ways: from the space perspective, there are topology
control techniques
[7], [8],
which trade network density
for energy savings while preserving the data forwarding
capacity of the network. Furthermore, by alleviating the
restriction
of
network capacity preservation,
[9]
trades
off the extensive energy savings
for
an increased latency
to set up a multi-hop path; from the time perspective,
there are TDMA-based schemes, which allow nodes to
go
to
sleep
or
standby during particular time-slots
of
a predetermined schedule
[lo].
Moreover, some vari-
ant techniques
([ll], [12],
[13],
[14]),
combined with
TDMA-based scheduling, enable nodes to shut down
their antennas when they are not involved in any traffic
in the following slotted time periods.
The above up-to-date designs for energy-efficient protocols
in wireless sensor networks reveal that to be
adaptive
is the
key to achieve efficient usage of energy, which usually has
two forms:
1)
Topology adaptive:
It
employs the minimum number of
nodes and power levels to construct the communication
system which consumes the lowest energy.
2)
TraDc adaptive:
It requires the communication system
to be available only when there is traffic.
Furthermore, these two different mechanisms are orthogonal
and can be combined to achieve the energy savings to the
largest extent.
reduce fhe per packet energy consumption:
This tech-
In this paper, focusing on the MAC layer, we only explore
0-7803-8623-W04/$20.00 02004
IEEE 21
4
the latter principle
to
present AC-MAC protocol for wireless
sensor networks. Following the basic periodic listen and sleep
scheduling proposed by S-MAC [13], AC-MAC introduces
adaptive duty cycle technique to adapt its behavior to the
traffic loads. Compared to its predecessor S-MAC, which
achieves energy-efficiency by sacrificing other important at-
tributes, such as throughput and latency and
so
on, AC-MAC is
designed
to
provide more optimized trade-offs among energy,
throughput and latency.
Our
simulation results also validate
that it is at least as energy-efficient as S-MAC while offering
the changing latency and throughput under various traffic
loads. Such a traffic-adaptive attribution is very useful
to
some
real-time sensor applications, such as target-tracking, in which
the amount of instantaneous traffic load depends on the speed
of the moving target. It is better to meet the energy-efficiency,
latency as well as throughput simultaneously under the re-
quirement of acquiring such dynamic tracking information
accurately and timely in some scenarios.
The reminder of the paper is organized
as
follows. Section
I1
describes the detailed design of AC-MAC protocol. Section
111
presents the simulation study which contains simulation
metrics and simulation results. Section
IV
reviews related work
in the literature with reference
to
energy-efficiency schemes at
the MAC layer for wireless sensor networks. Finally, section
V
concludes the paper and outlines directions for ongoing
research.
11.
AC-MAC
PROTOCOL
DESIGN
A. Motivation
Energy-efficiency is the first design consideration for wire-
less sensor networks. Narrowed down into MAC layer, the
efforts for efficient usage of energy are mainly focused on
turning down the radios of nodes as much as possible
to
reduce the energy waste for their idle states. The reasons are
as follows:
1) An idle receive circuit can consume almost as much
energy as an active transmitter under current wireless
interface design.
2)
The traffic model of a wireless sensor network largely
depends on its application. However, as to a quite repre-
sentative application in wireless sensor networks, event
tracking, which has widespread use in applications such
as security surveillance and wildlife habitat monitoring,
nodes will remain largely inactive for long period of
time but then suddenly become active when something
is detected.
S-MAC [13] is one of the early works in this area. As
a signal-frequency contention-based protocol, it inherits the
good scalability from the family
of
CSMA MAC protocols.
The synchronization between the sender and the receiver,
viewed as the first level, is achieved by the regular RTSKTS
mechanism; to reduce the energy wasted by idle listening,
S-
MAC introduces the periodic listen and sleep scheduling, the
second level synchronization, by virtual clustering technique
during the startup phase. Therefore, nodes will only be active
Fig.
I.
The periodic listen
and
sleep cycle for
S-MAC.
in relatively short listen intervals for communication while be
inactive in relatively long sleep intervals for energy saving.
However, the fixed parameters for S-MAC, such as listen
interval, sleep interval and
so
on, provide a limited adaptability
to a range
of
traffic loads, since throughput is limited by the
fraction of the listen interval and latency may be increased if
a message arrives during the sleep interval. As a result, these
parameters must be carefully tuned for
a
specific application,
otherwise it will incur large latency and low throughput which
may not be tolerated by the application despite of efficient
energy consumption.
As an optimization option for S-MAC,
Adaptive
Listening
[14]
is
also proposed by the same authors to reduce latency
at the price of some energy consumption. Although one
half reduced latency is achieved theoretically by providing
one more communication chance within a frame which is
composed by a listen interval and the following sleep interval,
it is still limited by its inflexibility infrastructure.
Thus, take S-MAC as a basis, the design goal of AC-MAC
is
to
provide a traffic-adaptive framework
for
energy-efficient
MAC protocol in wireless sensor networks, in which a third-
level synchronization depending on traffic conditions will be
introduced.
B.
Synchronization and Virtual Clustering
Adopting the framework of periodic listen and sleep
scheduling from S-MAC, AC-MAC inherits its scalability and
energy-efficiency. In the following, synchronization and the
scheme to achieve such synchronization during the startup
phase are described respectively.
Once synchronization is achieved among nodes, time is
divided into a fairly large number of frames. Every frame has
two intervals: a listen interval and a sleep interval. During
the sleep interval, most of the nodes turn off their radios
to preserve energy, while few nodes keep awake to continue
the data transmission negotiated in the previous listen interval
until they finish. During the listen interval, all nodes wake
up
and contend for the shared media. Once seized the channel,
the nodes start the transmissions which usually span to the
next coming sleep interval while most of the other nodes turn
off their radios due to the collision avoidance mechanism by
receiving the RTS
or
CTS packet not targeted for them. The
scheme of such periodic listen and sleep is shown in Figure
1.
In addition, the listen interval can be further divided into
SYNC interval and RTSKTS intervals. SYNC interval is
used for maintaining the synchronization among nodes by
exchanging SYNC packets, and RTSKTS interval
is
used for
medium sensing and collision avoidance. In summary, as seen
21
5
from the above Figure
1,
the timing relation among various
intervals can be formalized
as
follows:
and the duty cycle which can indicate the percentage of energy
saving is defined
as:
However, it must be noted that before entering into the
periodic listen and sleep cycles, it needs time for the startup
phase. Such
a
synchronization scheme can be termed as
virtual
clustering technique,
which urges nodes to form clusters with
the same schedule without enforcing this schedule to all nodes
in the network. The algorithm is executed in the following
way:
When a node comes to life, it starts listening while waiting.
If it does not hear anything for
a
certain amount of time, it
chooses
a
frame schedule and transmits
a
SYNC packet, which
contains the time until the next frame starts. If the node hears
a
SYNC packet from another node during the startup period,
it follows the schedule in that SYNC packet and transmits its
own SYNC accordingly.
Nodes retransmit their SYNC packets after some time
durations. Nodes must
also
listen for
a
complete frame spo-
radically,
so
that they can detect the existence of different
schedules. This
allows
new and mobile nodes to adapt to an
existing group. If
a
node has
a
schedule and hears
a
SYNC
with
a
different schedule from another node, it must adopt
both schedules. It must also transmit
a
SYNC with its own
schedule to the other node and let the other node know about
the presence of another schedule. Adopting both schedules
means that the node will have an activation event at the start
of both frames.
Nodes must start a data transmission only at the start of their
own active time. During that time, both neighbors with the
same schedule, and neighbors that have adopted the schedule
as
extra are awake. If
a
node would start transmission at the
start of
a
neighbor's frame, it might be transmitting
to
another
sleeping neighbor. Note that this scheme makes it possible that
broadcasts only need to be transmitted once.
C.
Adaptive
Duty
Cycle
Dynamically coordinating the duty cycles is the key scheme
of
AC-MAC in order to be adaptive to
a
wide range of traffic
loads. The premise is not to violate the established listen
and sleep scheduling among nodes since the realization of
synchronization among nodes needs
a
very high cost during
the initial startup phase. Currently, AC-MAC uses the number
of packets queued
at
the MAC layer
as
an indication of the
traffic load. Based on the traffic information, AC-MAC can
provide an adaptive number of chances
for
communications
.....E**?
%2!!
.............
.....
l%W<~":~>~&
Fig.
2.
The periodic listen and sleep cycle
for
AC-MAC.
For reason
of
possible high traffic,
AC-MAC
provides
2
chances
of
communication within
corresponding one cycle interval
of
S-MAC.
within one basic cycle time for S-MAC. The scheme of
adaptive duty cycle in AC-MAC is shown in Figure
2.
The decisions of how many reduced duty cycles needed to
fill the interval
of
the former basic cycle are first made by
individual node distributed in the following way: the number
of packets queued at MAC layer for node
i,
say
N;,
can be
mapped into
a
right value,
Ri.
It can be expressed
as:
Ri
=
f
(Ni)
(4)
where
f
(.)
is an application-specific function which expresses
the desire of
a
node for transmission. Each new reduced duty
cycle composes
a
fixed RTS/CTS interval and
a
varying sleep
interval depending on
Ri.
Thus, if we know
R;,
the reduced
duty cycle can be computed
as:
-
A
Tframe
=
TRTSJCTS
+
Geep
(5)
-
TRTSICTS
+
Tsleep
-
Ri
The range of the value
of
Ri
is limited by the timing relation
within one cycle time. Obviously, the lower limit of
Ri
should
be
1
to enable AC-MAC to keep the same behavior
as
S-MAC.
As for the upper limit
of
Ri,
with the increase of the value
R;,
the sleep interval in
a
reduced cycle time is decreasing since
the RTSlCTS interval is fixed. Without collision between the
data transmission and the next following RTS/CTS interval,
the shortest sleep interval should cover the maximum time
for
data transmission.
If
the lime for data transmission during one
cycle time is no more than
Tdora,
the upper limit of
R;
can be
derived
as
follows:
(7)
In order to keep the basic schedule framework of S-MAC,
the
Ri
value is determined only at every beginning of the basic
cycle time by every node
i.
Next, node will try to announce
its
R
values via RTS/CTS packet. It is natural
to
let the node
with the biggest
R
value occupy the transmission media for the
optimized latency and throughput metrics. This can be done
21
6
TABLE
I
PARAMETERS
OF
ACMAC
IMPLEMENTATION
ON
NS-2
f(Njj
{or node i
g(R;)
for
node i
Radio bandwidth
Control packet length
Data packet length
MAC header length
Contention window for SYNC (SYNC-CW)
Contention window
for
data (DATA-CW)
Slot
Time
DIFS
SlFS
Transmission power
Receiving power
Sleep Dower
min(Nj, Rimax)
DATA-CW
-
3
*
R;
20kbps
10 bytes
up
to
250
bytes
10
bytes
31 slots
63 slots
1
ms
lOms
5ms
36mW
14.4mW
15uw
by setting a smaller contention window
(cw)
for the node with
the bigger
R
value:
cwi
=
g(R;)
where
g(.)
is monotonously decreasing function and is also
application-specific. However, such a mechanism will increase
the probabilities of collisions when there is a heavy load of
traffic among the local nodes.
As
a result, there should be a
trade-off by choosing an appropriate
g(.)
depending on the
network topology information.
In the following, when one node
i
has seized the chance to
send out its RTS packet, it will append the
R;
value to the RTS
packet,
so
each node within its one-hop away may learn the
R;
value by equation
(5).
As
for the intended receiver, upon
receiving the RTS packet, it will append the
R;
value in its
CTS
packet to propagate such new duty cycle to the nodes
within its one-hop away. It should be noted that one node will
accept only one
R
value within one basic cycle time. Thus,
all
the nodes, either within one-hop away of the transmitter
or
the receiver, may follow the same new duty cycle within
one basic cycle time for
S-MAC.
Therefore, the adaptive duty
cycle algorithm is performed on per basic cycle time basis.
111.
SIMULATION
STUDY
The purpose of
our
simulation is to demonstrate the ef-
fectiveness of the proposed
AC-MAC
protocol, measure its
performance, and compare it with its predecessor
S-MAC
as
well
as
the IEEE 802.1
1
MAC
protocol which obtains the best
latency and throughput performance metrics; however, without
energy-efficient design goals in mind.
The implementations of both
S-MAC
and IEEE
802.11
have already been included in the newest version of ns-2
[I51
distribution. For convenience of comparing the three protocols,
we develop the
AC-MAC
in
the same simulation platform.
The important simulation parameters used are presented in
the table
I.
Moreover, there is no appropriate routing protocol
implemented in current ns-2 version to cooperate
S-MAC
and
our
AC-MAC
for a wireless sensor network. Thus, we also
develop an “IdeaAgent” for routing, which can compute the
optimized routing based on the global information. Thus,
our
.
Fig. 3.
Topology: ten-hop linear network with one source and one sink.
Fig.
4.
Average end-to-end message latency over 10-hop network.
simulation results directly reflect the performance metrics of
the
MAC
layer.
Our
simulation model is described as follows.
For
com-
parison purpose, we use the same topology which was used
to evaluate the performance of
S-MAC
in
[14].
It is a linear
network with eleven nodes, as shown in Figure
3.
The nodes
are configured
in
the way that it can be reached only by its
neighbor nodes. The first node is the source and the last node
is the sink.
We vary the traffic load by changing the packet inter-amval
time on the source node. The packet inter-arrival time changes
from
Is
to
15s.
Under each traffic condition, the simulation
is independently carried out for
10
times. In each simulation,
the source node sends
50
messages each with
50
bytes and
there is no fragmentation on any messages.
In
our
simulations, three different
MAC
protocols, the IEEE
802.11, the
S-MAC
with
10%
duty cycle, and the
AC-MAC
with 10% duty cycle are compared. Since the periodic listen
time is 160ms and
10%
duty cycle corresponds to a frame
length of 1.60s.
A.
Measurement
of
End-to-End Latency
Since
S-MAC
trades off latency for energy-savings, it is
expected to have very long latency in such multihop network
scenario due to the periodic sleep on each node. By introduc-
ing the adaptive duty cycle scheme, the sleep delay for a sender
waiting for the wakeup of a receiver will be largely reduced
at each hop in
AC-MAC,
so
that the end-to-end latency
is decreased.
As
for IEEE 802.11
MAC
protocol, without
periodic sleep, the end-to-end latency is mainly contributed
by carrier sense delay, backoff delay and transmission delay
which are inherent in contention-based
MAC
protocols. Figure
4
shows the measurement .,f average end-to-end message
latency.
I,...
,.
,
.I
.....
;
Meap
n1er-amv*
paDqs1
‘0
12
3
4
5
6
I8
9
1011
12131415
Fig.
5.
Average end-to-end throughput over IO-hop network.
It can be seen that the end-to-end latency of IEEE 802.11
MAC
protocol increases linearly with the decrease of traffic.
load, since the message delivery delay always dominates the
end-to-end latency in these scenarios;
as
to
S-MAC,
in high
traffic load where the packet inter-arrival time is less than
10s.
its end-to-end latency does not change obviously and remains
large compared to the other two protocols, whereas in low
traffic loads where the packet inter-arrival time is larger than
IOs,
the latency smootljly increases and approaches that of
IEEE 802.1
1
MAC
protocol with the decrease of traffic load;
as
to
AC-MAC,
due to the adaptive duty cycle scheme, its end-
to-end latency always follows the IEEE 802.1 1
MAC
protocol
and a bit larger than it under
a
wide range of traffic loads.
B.
Measurement
of
End-to-End Throughput
Since
S-MAC
trades off throughput for energy-savings, it
is expected to have very low throughput in such multihop net-
work scenario due to very limited chances of data transmission
for
nodes.
By
introducing the adaptive duty cycle scheme,
AC-MAC
can provide multiple chances for data transmission
within the basic cycle time for
S-MAC
depending on the load
of traffic,
so
that the end-to-end throughput is increased.
As
for IEEE 802.11
MAC
protocol without periodic sleep, data
transmissions are performed on contention-based. Figure
5
shows the measurement of the average end-to-end throughput.
It
can be seen that the throughput does not change much
within
a
wide range of traffic loads since the throughput is
restricted by its fixed duty cycle whereas the throughput of
AC-MAC
and IEEE 802.1
1
MAC
protocol keeps the track of
the message delivery rate.
To
further compare
AC-MAC-
and
S-MAC,
we find that in
high traffic loads where the packet inter-arrival time is less than
lOs,
the messages are queued in the
MAC
layer. Due to the
adaptive duty cycle scheme,
AC-MAC
achieves much higher
end-to-end throughput than
S-MAC.
In low traffic loads where
the packet inter-arrival time is larger than
IOs,
the messages
are not queued and the adaptive duty cycle scheme will not
be performed.
In
this case,
AC-MAC
obtains almost the same
throughput
as
S-MAC.
I
Fig.
6.
Average energy consumed per bits over 10-hop network.
C.
Measurement
of
Energy Consumption
t
We use EPB (energy per useful bit) to measure the energy
consumption. The actual time is counted from the transmission
of the first useful bit.
Due to the periodic listen and sleep scheduline adopted by
both
S-MAC
and
AC-MAC,
EPB
of
these two
MAC
protocols
are obviously less than that of IEEE 802.11
MAC
protocol
which is designed without energy-efficiency consideration.
From Figure
6,
it can be seen that EPB for
all
MAC
protocols
is increasing with the decreasing of traffic load since they
spend more time for idle listening when the arrival of the
messages are less frequent.
Furthermore,
AC-MAC
is almost
as
energy-efficient
as
S-
MAC
in terms
of
EPB. In high traffic load where the packet
inter-arrival time is less than
IOs,
EPB for
AC-MAC
is even
sightly lower than
S-MAC.
It can be explained
as
multiple data
transmissions in a basic cycle time in
AC-MAC
save more
overhead of
SYNC
packets than only one data transmission
in one cycle time in
S-MAC.
In low traffic loads where the
packet inter-arrival time is larger than
IOs,
EPB for
AC-MAC
is
almost the same
as
that of
S-MAC
since the
R
value
in
AC-MAC
will always be
0.
Iv.
RELATED
WORK
MAC
protocols for wireless sensor networks are designed
to be energy-efficient which is determined by the restricted
resources of sensor nodes. The basic scheme is to enable
nodes which are not involved in the communication to keep
their radios off
as
long
as
possible thereby to meet the
need of energy constrained environments. Therefore,
a
highly
efficient
synchronization
scheme between the communication
pairs becomes critical. It assures the nodes to be synchronized
as soon
as
possible once they wake up for the communication
purpose.
Bluetooth
[I61
is
a
TDMA-based
MAC
protocol. Its
TDMA
mechanism has a natural advantage of energy conservation
compared to contention-based protocols. Since the duty cycle
of the radio
is
reduced and there is no contention-introduced
overhead and collisions. Synchronization is easily achieved
based on reservation and scheduling. However, using
TDMA
21
8