Wireless Sensor Network, 2009, 1, 407-416
doi:10.4236/wsn.2009.15049 Published Online December 2009 (http://www.scirp.org/journal/wsn).
407
An Energy-Efficient MAC Protocol for Ad Hoc Networks
Yongsheng SHI, T. Aaron GULLIVER
Department of El ectri cal and C om puter Engineering Universit y of Victori a
P.O. Box 3055 STN CSC Victoria, B.C. V8W 3P6 CANADA
Received July 21, 2009; revised August 26, 2009; accepted August 28, 2009
Abstract
A mobile ad hoc network (MANET) is a collection of nodes equipped with wireless communications and a
networking capability without central network control. Nodes in a MANET are free to move and organize
themselves in an arbitrary fashion. Energy-efficient design is a significant challenge due to the characteristics
of MANETs such as distributed control, constantly changing network topology, and mobile users with lim-
ited power supply. The IEEE 802.11 MAC protocol includes a power saving mechanism, but it has many
limitations. A new energy-efficient MAC protocol (EE-MAC) is proposed in this paper. It is shown that
EE-MAC performs better than IEEE 802.11 power saving mode and exceeds IEEE 802.11 with respect to
balancing network throughput and energy savings.
Keywords: Energy-Efficient, MAC Protocol, IEEE 802.11, Ad Hoc Networks
1. Introduction
Energy efficiency is a major challenge in wireless net-
works. In order to facilitate untethered communication,
most wireless network devices are portable and battery-
powered and thus operate on an extremely constrained
energy budget. However, progress in battery technology
shows that only small improvements in battery capacity
can be expected in the near future [1]. Furthermore, since
recharging or replacing batteries is costly or, under some
circumstance, impossible, it is desirable to keep the en-
ergy-dissipation level of devices as low as possible.
A mobile ad hoc network is a collection of two or
more nodes equipped with wireless communications and
networking capabilities without central network control,
i.e. an infrastructure-less mobile network. Energy-efficient
design in MANETs is more important and challenging
than with other wireless networks. First, due to the ab-
sence of an infrastructure, mobile nodes in a n ad hoc net-
work must act as routers and participate in the process of
forwarding packets. Therefore, traffic loads in MANETs
are heavier than in other wireless networks with fixed
access points or base stations and thus MANETs have
more energy consumption. Second, energy-efficient de-
sign needs to consider the trade-offs between different
network performance criteria. For example, routing pro-
tocols usually try to find a shortest path from sources to
destinations. It is likely that some nodes will over-serve
the network and th eir energy will be drained quickly, and
thus cause the network to be partitioned. Therefore sim-
ple solutions that only consider power constraints may
cause a severe performance degradation. Third, no cen-
tralized control implies that energy-efficient management
in MANETs must be done in a distributed and coopera-
tive manner, which is difficult to achiev e.
At the wireless interface, energy consumption in idle
mode is only slightly less than transmit mode and almost
equal to receive mode [2]. Therefore, it is desirable to
build a network protocol that maximizes the time the
device is in sleep mode (the wireless interface turned off),
and also maximizes the number of wireless devices in
sleep mode. Many protocols have been proposed to deal
with this challenge [3–6].
In this paper, a new energy-efficient MAC protocol,
EE-MAC, is proposed. The design is based on the fact
that most applications of ad hoc networks are data-
driven, which means that the sole purpose of forming
an ad hoc ne twork is to co llec t and d isperse data . Hence,
keeping all network nodes awake is costly and unnec-
essary when some nodes do not have traffic to carry.
The proposed protocol conserves energy by turning off
the radios of specific nodes in the network. The goal is
to reduce energy consumption without significantly
reducing network performance. EE-MAC is based on
IEEE 802.11 and its power saving mode, and can pro-
vide useful information to the network layer for route
discovery.
The rest of this paper is organized as follows. Section
2 introduces related work and gives an overview of cur-
rent energy-efficient protocols for MANETs. Section 3
introduces IEEE 802.11 pow er saving mode (PSM). Sec-

Y. S. SHI ET AL. 408
tion 4 describes the proposed protocol, EE-MAC. In Sec-
tion 5, performance results are given and EE-MAC is
compared to 802.11 and 802.11 PSM. Finally, some
conclusions are given in Section 6.
2. Related Work
Energy-efficient protocol design is a cross-layer issue
and usually spans the network layer and MAC layer.
These two layers have different approaches to dealing
with power management. At the network layer, en-
ergy-efficient routing is a very active research topic. The
aim is to choose routes for unicast sessions so as to
maximize the overall network lifetime. Essentially, the
design principle of energy-efficient routing is to equally
balance energy expenditure among network nodes rather
than directly reduce power consumption at each node.
On the other hand, the MAC layer approach is to turn off
the device network interface when it does not have any
traffic. Thus, a design combining routing and MAC con-
siderations is appropriate for energy-efficient protocols.
We discuss some of the proposed solutions in the re-
mainder of this section.
Local energy-aware routing (LEAR) [4] is an en-
ergy-efficient routing protocol that does not consider the
MAC layer, while the dynamic power saving mechanism
(DPSM) [3] and the on-demand power management [5]
protocols are MAC layer approaches. Geographic adap-
tive fidelity (GAF) [6] is a cross-layer design, but it
needs geographic position devices to provide location
information.
LEAR is based on the dynamic source routing (DSR)
protocol, where route discovery requires flooding of
route-request messages. The basic idea of LEAR is to
consider the willingness of each mobile node to partici-
pate in the routing and forwarding of data packets on
behalf of others. This is based on the local information of
a mobile node. When a routing path is being established,
each mobile node relies on information on remaining
battery power to decide whether or not to participate in
the selection process of a route path. When a node’s re-
maining battery power is higher than a certain threshold,
route-request messages are forwarded and the node joins
in the route path selection process; otherwise, the mes-
sage is discarded. Thus, all intermediate nodes along the
route path have sufficient power and the first arriving
route message is considered to have followed an en-
ergy-efficient as well as a reasonably short path. If any of
the intermediate nodes drop the route-request message,
which means no nodes are willing to join the route path,
the source will not receive a single reply even though a
route may exist. To prevent this, the source node will
resend the same route request message with a lower
threshold.
Observing that the fixed beacon interval in IEEE
802.11 PSM wastes energy, DPSM uses adaptively
changed ad hoc traffic indication messages (ATIMs).
Coupled with a separate DATA window, DPSM can
control the transition to the low-power state in the middle
of a beacon interval. Therefore, a node is allowed to en-
ter sleep mode after completing any transmissions that
are explicitly announced in the ATIM window, and a
longer sleep mode time is achieved.
On-demand power management for ad hoc networks
bases power management decisions on traffic in the net-
work. The key idea is that transitio ns from power-saving
mode to active mode are triggered by communication
events instead of the established beacon interval used in
IEEE 802.11 PSM. On the other hand, transitions from
active mode to power-saving mode are determined by a
soft-state timer which is refreshed by the same commu-
nication events that trigg er a transition to active mode. A
node uses HELLO messages to track its neighbor’s
power management state to decide whether or not to send
packets to them.
The GAF protocol identifies redundant nodes with re-
spect to routing and turns them off without sacrificing
routing fidelity. Each node uses location information
based on GPS to associate itself with a virtual grid,
where nodes in a particular grid square are redundant
with respect to forwarding packets. One master node in
each grid stays awake to route packets. With GAF, nodes
can be in three states, sleep, discov er or active. Initially a
node is in the discover state and exchanges discovery
messages including grid IDs to find other nodes within
the same grid. A node becomes a master if it does not
hear any discovery messages for a given period of time.
If more than one node can become a master, the one with
the longest expected lifetime becomes the master and
handles the routing for that grid square.
3. An Overview of IEEE 802.11 Power
Saving Mode
Power management can achieve great savings in infra-
structure networks. All traffic for mobile stations must
go through access points, so they are ideal locations to
buffer traffic. However, in ad hoc networks, far more of
the burden is placed on the sender to ensure that the re-
ceiver is active or awake. Receivers must also be more
available and cannot sleep for as long as in infrastructure
networks.
Power management in IEEE 802.11 power saving
mode (PSM) is based on traffic indication messages.
Nodes use ATIMs to notify other nodes to prepare to
receive data. All nodes have to wake up periodically to
listen for ATIMs and check whether they have packets to
receive.
In PSM [7,8], time is divided into beacon intervals and
each beacon interval starts with an ATIM window. This

Y. S. SHI ET AL.409
window is the period during which nodes must remain
active and no stations are permitted to power down their
wireless interface. The ATIM window size is a parameter
that can be adjusted. Setting it to 0 means no power man-
agement is used. There are four possibilities for a nod e in
terms of ATIMs: the node has transmitted an ATIM,
received an ATIM, neither transmitted nor received, or
both transmitted and received. Nodes that transmit ATIM
frames do not sleep because this indicates an intent to
transmit buffered traffic. Nodes to which an ATIM is
addressed must also keep awake so they can receive data
packets from the ATIM sender. A node that both trans-
mits and receives of course needs to be active. Thus,
only those nodes that neither transmit nor receive an
ATIM can go to sleep after the ATIM window. Figure 1
illustrates the basic PSM operations. Nod es A and B have
advertised packets in the ATIM window by sending
ATIMs and receiving ATIM-ACKs, both of which are
subject to the DCF rules described earlier. Therefore
nodes A and B remain awake for the rest of the beacon
interval. The transmission of data packets from nodes A
and B takes place during the beacon interval. The node
that has no packets to transmit can go into sleep mode at
the end of the ATIM window if it does not receive an
ATIM during the window. In Figure 1, node C enters
sleep mode after the ATIM window, thus saving energy.
All sleeping nodes wake up again at the start of the next
beacon interval.
The beacon and ATIM window sizes can affect the
performance of PSM. Since no data packets are trans-
mitted in the ATIM window, overhead in terms of en-
ergy consumption and bandwidth is incurred. If we use a
small ATIM window to improve energy savings, there
may not be enough time to advertise all buffered data
packets. Conversely, using a large ATIM window may
unnecessarily waste bandwidth and not leave enough
time to transmit buffered data. Moreover, PSM also suf-
fers from long packet delivery latency: for each hop that
a packet traverses, the packet is expected to be delayed
Figure 1. IEEE 802.11 PSM operation.
for at least a beacon period. PSM was originally de-
signed for single-hop networks, which means all nodes in
the network are fully connected. However, ad hoc net-
works are usually multi-hop networks, and thus PSM is
not an ideal solution.
4. The Proposed EE-MAC Protocol
The key idea of EE-MAC is to elect master nodes from
all nodes in the network. Master nodes stay awake all the
time and act as a virtual backbone to route packets in the
ad hoc network. Other nodes, called slave nodes, remain
in an energy-efficient mode and wake up periodically to
check whether they have packets to receive. To be fair, a
rotation mechanism between masters and slaves is used.
EE-MAC uses some features of PSM, such as periodi-
cally waking up at the beginning of the beacon interval.
EE-MAC can provide knowledge and guidance to the
route lookup process, because only master nodes can be
selected along a routing path. On the other hand,
EE-MAC requires a mechanism to awaken a sleeping
node when packet delivery is imminent. This is usually
handled by low-level mechanisms at the MAC or physi-
cal layers. In EE-MAC, if a node has been asleep for a
while, packets addressed to it are not lost but are stored
at one of its upstream nodes, usually a master. When the
node awakens, the buffered data is sent to it (this is a
PSM feature which is used in our protocol).
4.1. Design Criteria
We consider the following design criteria.
 The protocol must ensure enough master nodes are
elected to build the backbone of the network so that
every node has at least one master in its vicinity. A col-
lection of masters can be described as a connected
dominating set (CDS). All nodes are either a member of
the CDS or a direct neighbor of at least one of the mem-
bers of the CDS. Nodes in the CDS serve as the routing
backbone and remain active all the time. All other nod es
are slave nodes and can choose to sleep. Since slave
nodes do not join in the process of route discovery or
packet forwarding, network connectivity is decreased. To
prevent a dramatic decrease in throughput, an acceptable
set of masters is required to maintain global connectivity
with some redund ancy.
 The master node election algorithm is based on lo-
cal information, which is a distributed approach. Each
node only employs local information to determine
whether it will become a master. Due to the characteris-
tics of distributed management in ad hoc networks and
the two essential requirements, low overhead and fast
convergence, the algorithm for finding a CDS should be
localized. The election algorithm is given in the next
section.

Y. S. SHI ET AL. 410
 The algorithm must have a fair way to rotate mas-
ters and slaves in order to ensure th at nodes equa lly share
the job of providing global connectivity. Over-using
some critical nodes will severely decrease the network
lifetime. Thus, if alternative nodes appear, masters can
step down and give the new nodes a chance to serve as
masters to balance node energy consumption.
4.2. Master Election and Forming a Connected
Dominating Set
To form a CDS, many researchers have proposed solu-
tions [9–11]. In this paper, we use the algorithm in [12]
modified for the energy sav ing condition.
Given a simple graph , where V is a set of
nodes and E is a set of links, a link from u to v is denoted
by a pair (u, v). According to [12], a set is a
dominating set of G if every node vV is con-
nected by at least one node
(,)GVE
'uV
'VV 
'V

. For example, in Fig-
ure 2, the node sets u, v in a and u, v in b are dominatin g
sets of the corresponding graphs. If all nodes in a domi-
nating set are connected together, it forms a CDS.
To quickly elect masters in an ad hoc network, we use
the following steps:
1) Initially assign the marker F to each node u in V.
2) Each node u exchanges its neighbor set N(u) with
all its neighbors.
3) u changes its marker to T if there exist nodes v and
w such that and (, , but(,(,)wu E)uv E)wv E

.
The T-marked nodes form a connected dominating set
and become masters, while the F-marked nodes become
slaves. However, we may not need all T-marked nodes
elected to act as the backbone of the network because
there are redundancies in this set. We say a node is cov-
ered if its neighbors can reach each other directly or via
other connected T-marked nodes. We establish a rule to
reduce the number of masters based on the idea that if a
node is covered by no more than k connected T-marked
nodes, we can change the marker of this node to F. In
general, assuming that 

'12
, ,...,
k
Vvvvk is the node set of
a connected subgraph in G' and if '
()( )
k
N
uNV in G,
then u can change its marker from T to F. This rule
Figure 2. Examples of connected dominating sets.
can be simply described as: if every pair of neighbors of
a T-marked node can be connected directly or via no
more than k other connected T-marked nodes, this node
is marked as F.
Two more issues need to be considered, node connec-
tivity and nod e energy. We deno te the connectivity level
of a node i as . Let
i
CL i
N
be the number of neighbors
of node i and be the number of pairs of nodes among
these neighbors that can be connected via i if i becomes a
T-marked node. Clearly, , and define the
maximum as . The energy level of node i
can be expressed as
i
C
CL 
2
Ni
i
C



/
irii
E
0
2
Ni



E
/
ii
C
EL

, where Eri is the re-
maining node energy and Ei is the initial node energy.
Finally, the node id, idi, will be considered if the two
factors given above are identical.
Overall, the rule to reduce redundant T-marked nodes
is as follows:
Assuming


'12
,,,
ki
Vvv vv
k
is the node set of a
connected subgraph in G', the marker of u is changed to
F if one of the following conditions holds:
1) in G, and for any node
'
()( )
k
Nu NV'
ik
vV

,
'
()() ()
iki
NvNV vNu .
2) in G, and for some nodes
'
()( )
k
Nu NV
'
ik
V
1,...,vv

,
'
11
., )(,..., )()
ik i
vNV vvNu ( ,..Nv
 12 min{ ,,...,}
ui
ELEL ELEL

or
 12 min{ ,,...,}
ui
CLCL CLCL

if 12 min{ ,,...,}
ui
ELEL ELEL

or
 12 min{ ,,...,}
ui
idid idid

if 12 min{ ,,...,}
ui
ELEL ELEL

and 12 min{ ,,...,}
ui
CLCL CLCL

After connected dominating set selection and reduc-
tion, all T-marked nodes will become masters and the
other nodes will become slaves. We use periodically
broadcasted HELLO messages to make each node in the
network aware of its neighbors’ status, including whether
or not they are masters, their current masters and their
current neighbors. Using a small value for k will increase
network connectivity but there will be many redundant
masters which will consume more energy. Conversely, a
large value for k will save energy but decrease the ro-
bustness of the network. In addition, a large k will usu-
ally require more frequent HELLO messages to collect
information. To balance the energy savings and network
throughput, we use k = 3 in this paper.
As mentioned above, rotation of masters and slaves is
an important design requirement. The rotation of masters
and slaves is done to allow every possible node to have a
chance to become a master, and let current masters

Y. S. SHI ET AL.411
change their role to save energy. Each master periodi-
cally checks if it should withdraw as a master. The con-
ditions to trigger a withdrawal are essentially the same as
for CDS reduction given above. However, in order to
balance the network load, we force some masters to quit
even if the conditions to withdraw are not met. After a
node has served as a master for some period of time or if
its energy level is below a certain value of ELi and the
average of its neighbors, it will withdraw even if there
are no masters nearby. The only exception is if some
neighbors can only be connected to the network via that
node.
4.3. Features of EE-MAC
In EE-MAC, since masters do not operate in power sav-
ing mode and can forward packets all the time, the
packet delivery ratio and packet delay can be improved
greatly compared to PSM. In this section, we present the
important features of EE-M A C .
4.3.1. Entering Sleep Mode Earlier
In the original PSM, a node with packets to transmit will
send an ATIM frame to the destination, and both source
and destination will stay awake in that beacon interval,
no matter how many packets need to be transmitted.
While this approach has its advantages, it may result in
much higher energy consumption than necessary. For
example, if a source only has one packet pending, they
have to waste the whole beacon period to deal with this
packet. To avoid this, we add the number of data packets
remaining at the sender to every data packet sent to the
destination. This information allows the destination to
know when it has received all pending packets for it.
When the source or destination have sent or received all
their packets, they can enter sleep mode until the begin-
ning of the next beacon interval.
4.3.2. Priority Processing of Packets to Slaves
When nodes are trying to send packets, they first deal
with those to be sent to slave nodes. After transmitting
all packets to slave nodes, packets between masters can
be sent. By using this method, slaves can be in sleep
mode as long as possible.
4.3.3. Prolonging the Sleep Period for Slaves
In EE-MAC, most packets are forwarded by masters and
packet routing via slaves is kept to a minimum. To take
advantage of this, each slave uses history information to
decide their sleep time. When a node observes two con-
secutive beacon intervals without any packets addressed
to it, it will decide to sleep through the next beacon in-
terval. The corresponding master must store this infor-
mation since failure to get an ACK does not guarantee a
broken link. If the master does not know a slave’s situa-
tion, it just buffers the packets to that slave. Only when
the master does not hear from a neighboring slave for
two consecutive beacon intervals does it discard these
packets.
4.3.4. Additi o nal MAC Layer Control
Nodes in an ad hoc network may move randomly. Thus,
to quickly adapt to network topology changes, a node
informs its neighbors of its status, master or slave, by
using the power management bit in the MAC header.
Since the MAC header can be heard anywhere in the
network, including RTS/CTS packets, this information
will help neighbors to know each other’s situation.
5. Performance Results
5.1. Simulation Environment
Our conclusions are based on the results gathered by
extensive simulation of a network model which imple-
ments EE-MAC. For the simulations, we used Network
Simulator-2 (NS-2) [13,14]. NS-2 is a popular package
which has been widely used in mobile ad hoc network
studies. For comparison with EE-MAC, we also imple-
mented IEEE 802.11 and its PSM mode.
We consider 25, 50 and 75 nodes moving in a square
area of 500m×500m, 750m×750m and 1000m×1000m
based on a mobility model called random waypoint [15].
Initially, each node chooses a random position in the area,
chooses a random destination, chooses a speed at random
uniformly distributed between 0m/s and 10m/s, and
moves towards the destination at the chosen speed. The
node then pauses for a period of time before repeating
the same process. Longer pause times reflect lower node
mobility and shorter pause times reflect higher mobility.
Simulations were performed for 400 seconds, so a 400
second pause time means no node mobility.
The nodes have 2 Mbps bandwidth and 250m radio
range. Each source node generates a Constant-Bit-Rate
(CBR) flow to the d estination with 256 byte packets. We
vary the number of sources and the number of packets
sent per second to change the network load. A network
load of 10% means that the total bit rate of all traffic
sources is 2×10% = 0.2 Mbps. DSR [16] is used as the
routing protocol. For the energy model, we use the data
shown in Table 1. All performance results shown in this
paper are an aver age of 10 runs.
We use the following metrics to evaluate network
performance:
 Data packet delivery ratio: The data packet deliv-
ery ratio is the ratio of the number of packets generated at
Table 1. Power Consumption Model [2]
Transmit ModeReceive Mode Idle Mode Sleep Mode
1400mW 1020mW 890mW 70mW

Y. S. SHI ET AL. 412
the sources to the number of packets received by the des-
tinations. This metric reflects the network throughput.
One of our goals is to design an energy-efficient MAC
protocol which can improve energy consumption without
suffering a significant capacity loss. Thus, this metric is
useful to measure any degradation in network through-
put.
 End-to-end delay: This metric not only includes
the delays due to data propagation and transfer, but also
those caused by buffering, queuing and retransmitting
data packets.
 Energy efficiency: We define energy efficiency as
Energy efficiency= Total bits transmitted
Total energy consumed
where the total bits transmitted is calculated using appli-
cation layer data packets only and total energy consump-
tion is the sum of the energy consumption in the nodes
during the simulation time. The unit of energy efficiency
is bit/Joule and the greater the number of bits per Joule,
the better the energy efficiency achieved.
5.2. Performance Eval uation
We now present our simulation results. The figures in
this section show three curves labeled 802.11, PSM and
EE-MAC. The curves labeled 802.11 correspond to the
IEEE 802.11 pr otocol without u sing power saving mode.
The curves labeled PSM indicate the IEEE 802.11 pro-
tocol with power saving mode. The curves labeled
EE-MAC represent the protocol proposed in this paper.
5.2.1. Impact of the Network Load
From the simulation results, we observe that network
load has a significant impact on all three protocols.
However, we show that varying the network load affects
these protocols differently in terms of our performance
metrics.
In Figures 3 and 4 we show the packet delivery ratio
under different network loads from 10% to 40%. When
the network load is low (10%), 802.11 performs a little
better than EE-MAC while EE-MAC provides a signifi-
cant improvement over PSM. As the network load in-
creases to 40%, all three protocols become worse due to
the higher collision rate. However, the performance dif-
ferences between 802.11 and EE-MAC, and EE-MAC
and PSM also increase, which means heavier traffic has
more impact on EE-MAC than 802.11 because under a
heavy network load, the master election algorithm oper-
ates more frequently to rotate masters and slaves. Among
the three protocols, PSM always performs worst. PSM
drops significantly more packets than the others because
of the existence of a fixed ATIM window, which wastes
bandwidth. When the traffic is high, it is possible that the
Figure 3. Packet delivery ratio with 50 nodes and 10 sources
in an area of 750m×750m.
Figure 4. Packet delivery ratio with 75 nodes and 10 sources
in an area of 750m×750m.
ATIM window is not long enough to advertise all pend-
ing packets, or the buffered data packets cannot all be
sent out during a beacon interval. On the other hand,
EE-MAC has the advantage of masters which never enter
sleep mode, so traffic between masters does not need to
be advertised. Coupled with the fact that most of the
network traffic is data traffic between masters, EE-MAC
can use a shorter ATIM window than PSM and thus pro-
vide better performance than PSM. EE-MAC is worse
than 802.11 because it still uses an ATIM window in
every beacon interval which wastes some bandwidth.
Moreover, the overhead of the master election algorithm
and using fewer nodes to forward packets also decreases
the packet delivery ratio.

Y. S. SHI ET AL.413
In Figures 5 and 6 we present the average packet delay.
Again, 802.11 performs the best among the three tech-
niques, and as the network load becomes heavier this
advantage increases. EE-MAC is not much worse than
802.11, but is far superior to PSM. PSM suffers from
long packet delays mainly because of its mechanism of
receiving-buffering-advertising-sending. Thus, each hop
in a PSM network corresponds to the length of the bea-
con interval. In addition, if the network load is high,
some packets have to be buffered up to 3 beacon inter-
vals before being sent out. Note that packets are dropped
if they have been kept in the buffer for 3 beacon intervals.
These factors cause PSM to have poor packet delay per-
formance. Similarly, the overhead due to master elections,
using ATIM windows, and fewer routing nodes,
Figure 5. Average packet delay with 50 nodes and 10 sources
in an area of 750m×750m.
Figure 6. Average packet delay with 75 nodes and 10 sources
in an area of 750m×750m.
results in EE-MAC having higher packet delays than
802.11.
In Figures 7 and 8, the metric of most interest in this
paper, energy efficiency, is presented. The results show
that EE-MAC performs best among all protocols. This is
because EE-MAC allows slave nodes to enter sleep
mode when no packets are addressed to them, but there
always exist awake nodes (masters) to forward packets.
Furthermore, EE-MAC can tell slaves to enter sleep
mode once they have finished receiving all data ad-
dressed to them in a beacon interval. These benefits allow
EE-MAC to nicely balance energy consumption and
packet delivery ratio, resulting in much better energy
efficiency. PSM does perform better than 802.11 and is
comparable to EE-MAC under light network load conditions.
Figure 7. Energy efficiency with 50 nodes and 10 sources in
an area of 750m×750m.
Figure 8. Energy efficiency with 75 nodes and 10 sources in
an area of 750m×750m.

Y. S. SHI ET AL. 414
As the ne twork load in creas es, PS M becomes worse v ery
quickly due to high data packet loss. Moreover, more
nodes need to participate in packet forwarding under a
heavy network load, which means more nodes must stay
awake all the time, causing high energy consumption.
Comparing energy efficiency between EE-MAC and
802.11 under different network loads is somewhat com-
plicated because it is related to both network throughput
and energy consumption. Since the difference in power
consumption among transmit, receive and idle modes is
not significant, the energy savings achieved is highly
dependent on the network node density, the ratio of time
in sleep mode to other modes, and the ratio of masters to
slaves. EE-MAC gains by reducing the number of awake
nodes. In some cases, 802.11 can outperform EE-MAC.
In Figure 9, the performance is given for 50 nodes, 20
sources and 20 packets/s, and 75 nodes, 5 sources and 20
packets/s. With 50 nodes, EE-MAC is sometimes worse
than 802.11 because at least 20 of the 50 nodes in the
network can never enter sleep mode. Conversely, with 75
nodes and 5 sources, EE-MAC is approximately 3 times
better than 802.11. Figure 9 also indicates that with 5
sources and 75 nodes, PSM is slightly better than
EE-MAC because in this situation, the cost of maintain-
ing a CDS is higher than the advantages it brings. As the
network load increases, EE-MAC will improve relative
to PSM.
5.2.2. Impact of Mobility
From the results shown, it is clear that high mobility de-
creases the performance of all three protocols. Overall,
mobility has a greater impact on EE-MAC than the other
two protocols. The reason is that with high mobility, the
network topology changes rapidly and links between
nodes can break often. Thus, the master election algo-
rithm has to operate frequently, which introduces more
overhead than with low mobility. Although mobility im-
pacts EE-MAC in terms of packet delivery ratio, it still
performs better than PSM. In terms of energy efficiency,
PSM performs very badly because under high mobility,
frequent route discovery messages cause a node to stay
awake much of the time.
5.2.3. Varyi ng Node Density
Clearly, high density can significantly improve network
performance with all three protocols. They will have
more options to choose a better route, and if a route
breaks, it is easier and quicker to find another one. As
mentioned above, EE-MAC relies more on node density
to enhance its performance than the other protocols be-
cause if the number of sources is constant, with high
node density, only a small fraction of the nodes need to
be elected as masters and most nodes can remain in
power saving-mode. This will result in significant energy
savings. Furthermore, with high node density, the impact
of mobility on EE-MAC is reduced. In other words, the
higher the node density, the better EE-MAC performs.
5.2.4. Chan ging Network Area
Reducing the network area from 750m×750m to 500m×
500m results in increased packet delivery ratio, de-
creased average packet delay and increased energy effi-
ciency for all three protocols. In a smaller network area,
the advantages of EE-MAC are not as prominent because
the weaknesses of the other two protocols are reduced.
Simulation results show that in an 500m×500m area,
most routes are 2–4 hops long, while in an 750m×750m
area, routes are often 4–7 hops long, so the routing over-
head and packet delay are much less in small networks.
The results show that EE-MAC only provides a slight
benefit in energy efficiency and is not as superior to
PSM as in a 750m×750m area. The performance with the
network area increased to 1000m×1000m was also
evaluated. Not only is the node density decreased, but
also forming a CDS requires more nodes in general and
the CDS can be broken more easily. These factors cause
a degradation in performance with EE-MAC, especially
in a high mobility network. As node mobility increases,
the packet delivery ratio and energy efficiency with
EE-MAC is reduced more compared to 802.11 and PSM.
5.2.5. Static N e tw ork
Figures 10 and 11 show the performance under static
network conditions. We fix the number of sources at 10
and vary the CBR to change the network load. Packet
delivery ratio and energy efficiency are given corre-
sponding to different network loads. It is clear that as the
network load increases, the packet delivery ratio of PSM
drops much mor e quickly than with EE-MAC and 802.11.
The decreased difference between EE-MAC and 802.11
with 50 nodes, compared to that with 75 nodes, shows
that EE-MAC benefits more from a higher node density.
Figure 9. Energy efficiency with 50 nodes and 20 sources,
and 75 nodes and 5 sources, in an area of 750m×750m.

Y. S. SHI ET AL.415
Figure 10. Packet delivery ratio with 50 and 75 nodes, 10
sources and 5% to 50% network load without mobility in
an area of 750m×750m.
Figure 11. Energy efficiency with 50 and 75 nodes, 10
sources and 5% to 50% network load without mobility in
an area of 750m×750m.
Note that in terms of energy-efficiency, under certain
conditions, PSM performs slightly better than EE-MAC
for two reasons. Firs t, PSM has good network th ro ughpu t
under a light network load. Second, EE-MAC needs an
almost constant number of nodes to form a CDS even
when the network load is very light and thus has con-
stantly awake nodes with little traffic through them.
6. Conclusions
This paper presented EE-MAC, an energy-efficient MAC
protocol for mobile ad hoc networks. The goal was to
reduce ener g y consu mption in an ad hoc networ k withou t
significantly reducing network performance. The key
i d ea of EE-MAC is to elect some nodes to form a connected
dominating set and use this as a virtual backbone to route
packets, while other network nodes, called slaves, stay in
power-saving mode. EE-MAC is a cross-layer design
which spans the network layer and the MAC layer.
The performance of EE-MAC was evaluated using the
NS-2 network simulator, and compared to IEEE 802.11
with and without power saving mode. The results show
that IEEE 802.11 p erforms better than EE-MAC in terms
of packet delivery ratio and average packet delay. How-
ever, EE-MAC exceeds IEEE 802.11 in energy effi-
ciency and is much better than PSM in overall terms. The
network load has a great impact on the behavior of
EE-MAC. Under a light network load, EE-MAC is only
slightly worse than IEEE 802.11, but as the network load
increases, the difference in performance between EE-
MAC and IEEE 802.11 increases because EE-MAC
needs to rotate masters and slaves more frequently with
high traffic and EE-MAC still uses the ATIM window.
The results also show that the higher the node density,
the better EE-MAC performs. In summary, a mid-sized
network with relatively high node density is the best en-
vironment to utilize EE-MAC.
7. References
[1] S. Sheng, A. Chandrakasan, and R. W. Brodersen, “A
portable multimedia terminal,” IEEE Communications
Magazine, Vol. 30, No. 12, pp. 64–75, December 1992.
[2] L. M. Feeney and M. Nilsson, “Investigating the energy
consumption of a wireless network interface in an ad hoc
networking environment,” Proceedings of IEEE INFO-
COM, pp. 1548–1557, April 2001.
[3] E. S. Jung and N. H. Vaidya, “An energy efficient MAC
protocol for wireless LANs,” Proceedings of IEEE IN-
FOCOM, pp. 1756–1764, June 2002.
[4] K. Woo, C. Yu, D. Lee, H. Y. Youn, and B. Lee,
“Non-blocking, localized routing algorithm for balanced
energy consumption in mobile ad hoc networks,” Pro-
ceedings of International Symposium in Modeling,
Analysis and Simulation of Computer and Telecommu-
nication Systems, pp. 117–124, August 2001.
[5] R. Zheng and R. Kravets, “On-demand power manage-
ment for ad hoc networks,” Proceedings of IEEE INFO-
COM, pp. 481–491, March 2003.
[6] Y. Xu, J. Heideman n, and D. E st rin, “ Geog raphy-i nfo rme d
energy conservation for ad hoc routing,” Proceedings of
ACM International Conference on Mobile Computing
and Networking, pp. 70–84, 2001.
[7] Supplement to IEEE Standard for Information Technol-
ogy-Telecommunications and Information Exchange be-
tween Systems-Local and Metropolitan Area Net-
works-Specific Requirements-Part 11: Wireless LAN
Medium Access Control (MAC) and Physical Layer

Y. S. SHI ET AL. 416
(PHY) Specifications: High-speed Physical Layer Exten-
sion in 2.4 GHz Band, IEEE Std 802.11b, 2000.
[8] M. S. Gast, 802.11 Wireless Networks: The Definitive
Guide, O’Reilly & Associates, 2002.
[9] P. J. Wan, K. M. Alzoubi, and O. Frieder, “Distributed
construction of connected dominating set in wireless ad
hoc networks,” Mobile Networks and Applications, Vol.
9, No. 2, pp. 141–149, April 2004.
[10] Y. P. Chen and A. L. Liestman, “Approximating mini-
mum size weakly-connected dominating sets for cluster-
ing mobile ad hoc networks,” Proceedings of ACM In-
ternational Symposium on Mobile Ad Hoc Networking
and Computing, pp. 165–172, June 2002.
[11] J. Wu and H. Li, “On calculating connected dominating
set for efficient routing in ad hoc wireless networks,”
Proceedings of ACM International Workshop on Discrete
Algorithms and Methods for Mobile Computing and
Communication, pp. 7–14, August 1999.
[12] J. Wu and H. Li, “Domination and its applications in ad
hoc wireless networks with unidirectional links,” Pro-
ceedings of IEEE International Conference on Parallel
Processing, pp. 189–197, August 2000.
[13] UCB LBNL VINT Group, Network simulator (V. 2),
http://www.isi.edu/nsnam/ns.
[14] K. Fall and K. Varadhan, Network Simulator Notes and
Documentation, The VINT Project, UC Berkeley, USC/
ISI, LBL and Xerox PARC, 1999,
http:// www.isi.edu/nsnam/ns.
[15] J. Broth, D. Maltz, D. Johnson, Y. C. Hu, and J. Jetcheva,
“A performance comparison of multi-hop wireless ad hoc
network routing protocols,” Proceedings of International
Conference on Mobile Computing and Networking, pp.
85–97, October 1998.
[16] D. Johnson, D. Maltz, and Y. C. Hu, “The dynamic
source routing protocol for mobile ad hoc networks
(DSR),” IETF draft, July 2004,
http://www.ietf.org/ rfc/rfc3684.txt.