1

A Power Saving MAC Protocol

for Wireless Networks


Eun-Sun Jung and Nitin H. Vaidya

Dept. of Computer Science

Texas A&M University, College Station
Email: [email protected]


Dept. of Electrical and Computer Engineering,
and Coordinated Science Laboratory
University of Illinois at Urbana-Champaign
Email: [email protected]

Technical Report
July 2002

Abstract—This paper presents a protocol which improves I. I NTRODUCTION

on the power saving mechanism in the IEEE 802.11 Dis-
tributed Coordination Function (DCF). In the power saving
ATTERY power is one of the critical resources in
mechanism (PSM) for DCF, all nodes are synchronized by
beacons. In each beacon interval, there is a fixed time in-
terval called the ATIM window where every node has to be
B wireless networks. Due to limited battery power,
various energy efficient protocols have been proposed to
awake. During the ATIM window, a source node informs a reduce energy consumption. Various approaches are pro-
destination node about a pending packet by transmitting an posed for different protocol layers, including work on bat-
ATIM frame. When the destination node receives an ATIM tery management [1], [4], [13], power control [5], [11],
frame, it replies with an ATIM-ACK. Both the source and
[12], and energy-efficient protocols using directional an-
destination nodes stay awake for the remaining beacon in-
terval. The source can transmit data after the ATIM win- tennas [16], [21].
dow finishes. A node that does not have traffic to send or Since the wireless network interface consumes a sig-
receive can enter the doze state after its ATIM window fin-
nificant amount of energy, a large body of research has
ishes.
focused on reducing energy consumption. A power sav-
During the ATIM window in PSM, no data transmission ing mode is often used to reduce energy consumption by
is allowed. Thus, the available bandwidth in PSM is reduced putting the wireless network interface into a doze state.
according to the ATIM window size. Also, energy is con-
sumed in transmitting and receiving ATIM and ATIM-ACK In this paper, we propose a new power saving MAC
frames. This paper proposes a protocol that removes the protocol using the power saving mode. We consider two
overhead of the ATIM window and uses the bandwidth for states for the wireless network interface. Specifically, a
data transmission. Simulation results show that removing wireless network interface can be in either the awake or
the ATIM window gives better aggregate throughput and
doze states. In the awake state, there are three different
energy saving.
modes, transmit, receive, and idle, and each consumes a
different amount of energy. In the doze state, the wireless
network interface consumes much less energy as com-
pared to the awake state. However, there exists transition
This research is supported in part by National Science Foundation delay and additional energy consumption when a node
grants 99-76626 and 01-96410. changes its state from doze to awake (or vice versa). For
 2

instance, [9] and [6], [14] report 250  s and 800  s for the width is wasted. In this paper, we propose a new power
transition time, respectively. saving scheme which removes the ATIM window and uses
the bandwidth for data transmission.
The IEEE 802.11 standard [19] specifies two medium
access control protocols – PCF (Point Coordination Func- The rest of the paper is organized as follows. Sec-
tion) for a centralized protocol and DCF (Distributed Co- tion II reviews the related work. Section III presents our
ordination Function) for a fully distributed protocol. Both proposed protocols. Section IV describes our simulation
protocols support a power saving mechanism (hereafter model and discusses the simulation results. Section V
referred as PSM) which requires nodes in the network to concludes the paper.
be synchronized by periodic beacon transmissions. In this
paper we only focus on PSM in DCF.
II. R ELATED W ORK
Fig. 1 illustrates the PSM in DCF. As the figure indi-
cates, time is divided into beacon intervals in PSM. At the Simulation results for the power saving mechanisms of
beginning of each beacon interval, there exists a specific two wireless LAN standards, IEEE 802.11 and HIPER-
time interval, called the ATIM window (Ad-hoc Traffic LAN, are presented in [22]. It shows the sizes of a beacon
Indication Message window), where every node is awake. interval and an ATIM window in IEEE 802.11 have a sig-
When a node has a packet to transmit, it first transmits an nificant impact on throughput and energy consumption.
ATIM frame to the destination node during the ATIM win- As shown in [22], a fixed ATIM window size in
dow. When the destination node receives the ATIM frame, IEEE 802.11 cannot perform well in all situations. The
it replies with an ATIM-ACK. After the ATIM and ATIM- optimal ATIM window size depends on various factors. A
ACK handshake, both the source and the destination will mechanism to choose the ATIM window size dynamically
stay awake for the remaining beacon interval to perform is proposed in [8]. As observed in [8], the power saving
the data transmission. A node that has not transmitted or mechanism in IEEE 802.11 does not provide much energy
received an ATIM frame during the ATIM window may savings because nodes have to stay awake for a whole bea-
enter the doze state after finishing its ATIM window. con interval even if they have few packets to transmit. The
protocol in [8] allows nodes to power off their network
A ATIM DATA
interface during the beacon interval whenever they finish
beacon ATIM window
announced packet transmissions, thus improving energy
B savings.
ATIM−ACK ACK beacon

C
The synchronization of beacon intervals when using
ATIM window
Dozing
ATIM window DCF can be difficult in multi-hop wireless networks.
beacon interval beacon interval Some solutions are proposed in [20].
Fig. 1. Power saving mechanism (PSM) for DCF: Node A announces In PAMAS [15], each node uses two separate channels,
a buffered packet for B using an ATIM frame. Node B replies by
sending an ATIM-ACK, and both A and B stay awake during the entire
one for control and the other for data packet transmissions.
beacon interval. The actual data transmission from A to B is completed By using the control channel, a node determines when and
during the beacon interval. Since C does not have any packet to send for how long to power off the wireless network interface.
or receive, it dozes after the ATIM window.
Similar to PAMAS, S-MAC [24] allows nodes to sleep
during neighbors’ transmissions. Nodes enter the doze
As shown in [8], [22], the performance of PSM is sig-
state after hearing an RTS or CTS destined for a neigh-
nificantly affected by the size of the ATIM window. The
bor. S-MAC is designed for wireless sensor networks. To
optimal ATIM window size is closely related to the net-
reduces contention latency, long messages are fragmented
work load, the size of beacon interval, etc. As mentioned
into many smaller fragments, then transmitted in bursts.
earlier, during the ATIM window all nodes are awake
and only ATIM and ATIM-ACK frames can be trans- Span [3] utilizes the IEEE 802.11 power saving mecha-
mitted. Real data transmission can only occur after the nism. Span elects coordinators, which periodically rotate
ATIM window. Overhead in energy consumption is in- their roles. The coordinators stay awake and forward traf-
curred for transmitting or receiving additional ATIM and fic for active connections. Non-coordinators follow the
ATIM-ACK frames, and there is overhead in time due to power saving mechanism of IEEE 802.11 DCF. Nodes
the ATIM window. For example, if the beacon interval is buffer the packets for dozing destinations and announce
100 ms and the ATIM window size is 20 ms, 20% of band- these packets during the ATIM window as in DCF. SPAN
 3

introduces a new advertised traffic window following an the beginning of time, and also after each successful trans-
ATIM window. During this advertised traffic window, the mission of a data packet by S. Now, if the transmission
announced packets and the packets for the coordinators medium is not idle, S waits until it becomes idle. Then,
can be transmitted. After this window, only the packets for while the medium is idle, the backoff counter is decre-
the coordinators can be transmitted, and non-coordinators mented by 1 after each “slot” time 2 . Eventually, when the
can enter the doze state if they do not have traffic to send backoff counter reaches 0, S transmits an RTS packet for
or receive. the intended destination D. When D receives the RTS, if
D can communicate with S at the present time, D replies
The protocol proposed in [25] also uses the power
by sending a CTS to node S. Node S, on receipt of the
saving mechanism of IEEE 802.11. However, unlike
CTS, sends DATA to D. After receiving DATA success-
IEEE 802.11, the protocol in [25] uses information from
fully, D sends an acknowledgment (ACK) to S. Now, it is
the network layer to reduce packet delivery latency. When
possible that two nodes may choose their backoff counters
a node receives routing packets, such as route request,
such that they both transmit their RTS packets simultane-
route reply, etc., the node will stay awake for a predefined
ously, causing a collision between the RTS packets. In
time duration, which is much longer than a beacon inter-
this case, node S will not receive a CTS. Absence of the
val. Nodes involved in packet forwarding will be awake
CTS is taken as an indication of congestion, and node S
for a longer time, so that the end-to-end latency is reduced.
doubles its contention window size  , picks a new back-
GAF [23] uses location information, provided by GPS, off counter uniformly distributed over   , and repeats
to form “virtual grids”. All nodes in the same grid are the above procedure.
equivalent in terms of traffic forwarding. GAF guarantees
IEEE 802.11 DCF also incorporates physical and “vir-
that one node in each grid stays awake in order to forward
tual” carrier sensing. The details are omitted here, how-
traffic.
ever, note that the proposed protocol borrows the physical
Bluetooth [2] is designed for a low-cost and low-power and virtual carrier sensing mechanism from 802.11 DCF.
wireless network. Bluetooth devices are usually organized
into so called piconets, which consist of one master and
up to 7 slave devices. Bluetooth provides three different B. Time Synchronization
low power modes (sniff, hold and park) to reduce energy
consumption. Energy efficiency in Bluetooth is studied The proposed NPSM uses periodic beacon trans-
in [7], [26]. missions to synchronize nodes in the network, as in
IEEE 802.11 PSM. In PSM for IEEE 802.11 DCF, time
is synchronized in a distributed manner. Each node main-
III. P ROPOSED P OWER S AVING M ECHANISM tains a local timer and transmits a beacon, which contains
a timestamp of local timer. When a node receives a beacon
We now present the proposed power saving mechanism, from a neighbor and its beacon frame has not been trans-
referred to as NPSM (New PSM) hearafter. Since NPSM mitted, it cancels its beacon transmission. The node will
is similar to the IEEE 802.11 MAC protocol, we first de- update its local timer, if the timestamp of the received bea-
scribe how IEEE 802.11 works. con frame is more recent than the value of its own timer.
Reference [20] proposes other solutions to achieve syn-
chronization in multi-hop networks, which could also be
A. IEEE 802.11 MAC Protocol used with NPSM.

The DCF in IEEE 802.11 uses an exchange of

RTS (Ready to Send) and CTS (Clear to Send) packets C. Removing the ATIM window
between the sender and receiver prior to transmission of a
data packet 1 . NPSM removes the ATIM window from IEEE 802.11
PSM in order to reduce control overhead. As mentioned
When a node S wants to transmit a packet to a node earlier, time is divided into beacon intervals in NPSM.
D it chooses a “backoff” counter uniformly distributed in At the start of a beacon interval, every node enters an
the interval   , where  is size of the so-called con- awake state for a specified duration called DATA window.
tention window.  at node S is reset to a value  at 
 Before counting down the backoff counter, a node waits for a dura-
For small data packet the RTS-CTS exchange may be omitted. tion of DIFS time.
 4

The DATA window can be considered analogous to the TABLE I

ATIM window in PSM since every node is awake dur- T HE C OUNTER I NCLUDED IN E ACH PACKET
ing the DATA window. However, the difference is that Packet type Counters included in the packet
we do not transmit ATIM or ATIM-ACK frames during DATA ?>A@CBED GFIHJD.K.FL! and # &)(*&,+.-
the DATA window. Instead, nodes transmit data packets RTS ?>A@CBED GFIHJD.K.FL! + #'&)(*&,+.-
during the DATA window without any ATIM or ATIM- CTS, ACK ?BCK.M4N  @C! + # &,(*&,+.-
ACK transmission. The purpose of the ATIM window in
PSM is to announce pending packets to destination nodes.
NPSM has a different way to achieve the same function.
The basic idea is whenever a node transmits a packet to a indicates the minimum data transmission that the node 
destination it includes the number of pending packets in will perform while staying awake. 13240.! is reset to zero
the packet. We describe the details next. at the beginning of each beacon interval for all  . Node 6
knows that node  will stay awake as long as the counter,
1O24 "! is greater than zero. A node uses 13240.! to decide if
D. Announcing pending packets it can enter the doze state after the DATA window finishes,
as described next.
In IEEE 802.11 PSM, the purpose of the ATIM win-
dow is to announce the existence of pending packets. To Note that instead of including the number of pending
achieve the same goal, in NPSM, each node maintains packets for a destination, ?>A@CBED GFIHJD.K.FL! , an alternative
counters to indicate the number of pending packets to is to include the list of destinations to whom the source
transmit or receive. has pending packets (and the amount of data pending to
them). We have not evaluated this alternative yet. The ap-
The following counters are maintained by each node  : proach evaluated here may introduce a short term unfair-
 "! : the number of packets pending at this node ness since nodes that have not received any packet may
(i.e., node  ) for node  . enter the doze state. However, this is also the case in
$#% "! : the number of packets destined for node  IEEE 802.11 PSM – if a node has not received an ATIM or
known (to  ) to be pending at node  . ATIM-ACK, it will enter the doze state. In this paper, we
$#'&)(*&,+.- : sum of #/0.! over all neighbors  of node do not include the list of destinations for all the pending
 . # &)(*&,+.- is the total number of packets destined for packets. This is an issue for future work.
node  known to be pending at all its neighbors.
$13240.! : the number of packets that the neighbor node
 needs to transmit or receive. Node  learns 13240.!
E. Transition to doze state
by overhearing packet transmission from node  , as
described below. When the current DATA window expires, a node de-
The above counters are included in DATA, RTS, CTS, cides whether it should extend the DATA window (and
and ACK packets, as listed in Table I, and explained be- stay awake longer) or go to the doze state. The node de-
low. cides to extend the DATA window if any of the following
conditions are satisfied.
 DATA: When node  transmits a DATA packet to
 As explained earlier, a node maintains an estimate
node 5 it includes  5 ! and # &,(*&)+.- 3 . When node 5 re-
ceives the DATA packet from node  it updates #% "! . of the number of packets it needs to receive from its
(Update to #/0.! also changes # &)(*&,+.- at node 5 .) neighbors. If the node has not finished receiving all
 RTS: An RTS from node  to node 5 includes  5 ! + the packets (i.e., #'&)(*&,+.- is greater than zero) it will
# &)(*&,+.- . stay awake longer.
 CTS and ACK: A CTS and an ACK from node 5 to  In NPSM, a node can infer the neighbors’ state (doze
node  includes  "! + # &,(*&)+.- . or awake) from overheard information. For exam-
ple, node 6 can assume that node  will be awake
Whenever any neighbor node 6 overhears an RTS, CTS, for packet transmission if 1O24 "! maintained by 6 is
DATA, or ACK packet from node  to node 5 , node 6 up- greater than zero. When node 6 has a packet to trans-
dates 1324 "! to  5 ! + #'&,(*&)+.- included in the packet. 13240.! mit to a neighbor  , and 13240.! is greater than zero, 6
7 8:9,; 9,< = will remain awake and try to transmit packets to the
in DATA packet is necessary only when the RTS-CTS hand-
shake is omitted for small DATA packet. destination.
 5

In our simulation of NPSM, the beacon interval and the 3) Average packet delay over all flows in the network
initial DATA window are set to 100 ms and 20 ms, respec- Packet delay is calculated as the difference between
tively. The DATA window size is increased in increments the time when MAC layer receives a packet from the
of 5 ms. When the increased DATA window expires, the upper layer and the time when a node receives an
same process happens as when the initial DATA window ACK from the receiver for the packet. We calculate
expired. This process is repeated until the next beacon in- average delay over all flows in the network.
terval is started. Thus, if the network is highly loaded, it
is possible the node does not enter the doze state at all. A. Simulation Model
Fig. 2 illustrates how NPSM works. Suppose node A We used ns-2 [18] for our simulations. Each simulation
has packets to send to node B. Node A transmits pack- was performed for a duration of 20 seconds. The channel
ets during the DATA window without using any ATIM bit rate is 11 Mbps and the transmission range is 250 P .
frames. Since node A has not finished all packet trans- Different network sizes are simulated, as described later.
missions after the original DATA window, both A and B In each scenario, a source node generates and transmits
will stay awake for 5 ms longer. Since node A includes constant-bit rate traffic. The packet size of each flow is
the number of packets to transmit to node B within its first fixed at 512 bytes unless otherwise specified. (We also
DATA packet, B knows it has not received all the packets performed some simulations varying packet size.)
from A. When the increased DATA window expires, and
if all packet transmissions have finished, both A and B For the energy model, we assume the wireless network
can go to the doze state, thus saving energy. Since node interface consumes 1.65 W, 1.4 W, 1.15 W, and 0.045 W
C does not have any packet to transmit or receive it enters in the transmit, receive, and idle modes and the doze state,
the doze state when the initial DATA window expires. respectively [10], [17].
20 ms 5 ms 20 ms We use 800  s as the doze-to-awake transition time [6],
A DATA DATA
[14], which is a more conservative than the 250  s esti-
beacon Dozing DATA window mate in [9]. During this transition time, a node will con-
sume twice as much power as the idle mode (i.e., 2.3 W).
B
beacon ACK ACK Dozing beacon
The initial energy for each node is 1000 joules so nodes
C do not run out of energy during the simulations. All the
Dozing
DATA window DATA window
simulation results are averages over 30 runs.
Beacon interval (100 ms) Beacon interval

Fig. 2. NPSM does not have the ATIM window in order to reduce
The beacon interval is set to 100 ms for both PSM and
the overhead of transmitting extra control packets (ATIM and ATIM- NPSM. The ATIM window size for PSM and the initial
ACK). This gives more bandwidth for data transmissions as compared DATA window for NPSM are set to 20 ms (recall that
to PSM. the DATA window is changed dynamically). Simulations
were performed in both a wireless LAN and a multi-hop
network as described below.
IV. P ERFORMANCE E VALUATION 1) Wireless LAN Scenario: Simulated network sizes
are 20, 40, and 60 nodes for a wireless LAN. By a wireless
We have simulated the proposed NPSM and the PSM
LAN, we mean all nodes are within each other’s transmis-
in IEEE 802.11, as well as the IEEE 802.11 MAC with-
sion range. In each scenario for 20 and 40 node networks,
out using power saving mode. Three metrics are used to
half the nodes are source nodes and the other half are des-
evaluate the MAC protocols.
tination nodes. Thus, there are 10 flows in a 20 node net-
1) Aggregate throughput over all flows in the network work.
Since the throughput may be degraded by using a
For a 60 node network, we simulate a scenario where
power saving protocol, we measure the aggregate
one source transmits packets to multiple destinations.
throughput.
Twenty sources transmit packets to 40 other destination
2) Total data delivered per unit of energy consumption
nodes – each source node transmits two flows.
(or, Kbits delivered per joule)
This is calculated as the total data delivered by all We varied the total network load to observe the ef-
the flows divided by the total amount of energy con- fect of network load on aggregate throughput and en-
sumption over all the nodes. ergy consumption. Simulated network loads are 10%,
 6

20%, 30%, 40%, and 50%, measured as a fraction of Fig. 4 shows the total data delivered per joule
the channel bit rate of 11 Mbps. For example, at a net- (Kbits/joule) for a wireless LAN with various network
work load of 10%, the total bit rate of all traffic sources is loads. NPSM performs the best among all schemes. Since
Q)R3STRR%UVRWQ,R Mbps. Each traffic source has the same bit we measure the total data delivered per joule, the poor ag-
rate. Thus, with a total load of 10%, and 10 traffic sources, gregate throughput of PSM (see Fig. 3) results in lower
each traffic source has a rate of 0.11 Mbps. total data delivered per joule. As observed in [8], the en-
ergy saving of PSM is poor when the network load is high.
2) Multi-hop Network Scenario: For a multi-hop net-

As mentioned earlier, PSM specifies that nodes have to be
work, 50 nodes are randomly placed in a 1000 x 1000 P
awake for the whole beacon interval even if they have few
area. Ten source and ten destination nodes are randomly
packets to transmit. This leads to less dozing time, re-
chosen. Note that a source or destination node can also be
sulting in less energy saving. Moreover, every node has
an intermediate node that forwards traffic for other nodes.
to be awake during the ATIM window in PSM, transmit-
The average route length of the flows is 4 hops with a
ting extra ATIM and ATIM-ACK frames. This introduces
range of 2 to 6 hops. Each traffic source generates a data
extra energy consumption. In Fig. 4(a), PSM performs
rate of 10, 20, 30, 40, or 50 Kbps.
slightly better than 802.11. However, when the network
load is high, PSM does not achieve energy savings as
B. Simulation Results seen in Fig. 4(b) and (c) and also degrades the aggregate
throughput as seen in Fig. 3(b) and (c). In NPSM, nodes
We now present our simulation results. We show the do not use the ATIM window, and nodes can go to the
simulation results for the wireless LAN case first, fol- doze state during a beacon interval if they do not have any
lowed by the simulation results for a multi-hop network. traffic. Therefore, NPSM performs much better than PSM
The graphs in this paper show three curves labeled as or 802.11 in Fig. 4(a), (b), and (c). Note that the sim-
802.11, PSM and NPSM. The curve labeled as 802.11 ulation results for the scenario with two flows per source
corresponds to IEEE 802.11 DCF without using the node (Fig. 4(c)) are similar to those of one flow per source
power saving mode. The curve labeled as PSM indicates (Fig. 4(a) and (b)). In Fig. 4(c), the data delivered per
IEEE 802.11 DCF with PSM. The curve labeled as NPSM joule for PSM is slightly worse than IEEE 802.11 due to
is for the scheme proposed in this paper. its poor throughput (see Fig. 3(c)).

B.1 Wireless LAN: varying the network load Fig. 5(a), (b), and (c) show the average packet delay
over all flows in a 20, 40, and 60 node network, respec-
Fig. 3 shows the aggregate throughput (in Kbps) vary- tively. IEEE 802.11 performs the best among all schemes
ing the network load using different schemes in a wireless because it does not use power saving mode. The delay
LAN. When the network load is low, for instance at the for PSM is longer than that of 802.11, but shorter than
network load of 10%, all schemes perform similarly for that of NPSM. This is because with moderate and high
20, 40, and 60 node networks in Fig. 3(a), (b), and (c), loads, nodes are awake most of time (no dozing) in PSM.
respectively. However, as the network load increases, the (Recall that in PSM, a node has to be awake for the whole
aggregate throughput of PSM degrades severely. This is beacon interval even if it has a few packets to transmit
mainly due to the overhead of the ATIM window. In PSM, or receive.) NPSM gives the longest delay due to longer
20% of the channel bandwidth is used for the ATIM win- dozing time (yielding more energy saving, as seen in
dow where only ATIM and ATIM-ACK transmissions are Fig. 4). As the network load increases, the average packet
allowed. Therefore, there is less time for actual data trans- delay also increases in NPSM. When a destination node is
mission using PSM. The aggregate throughput of NPSM in the doze state, packets at a source node have to stay in
is also lower than IEEE 802.11 without PSM, but the a buffer. This increases the packet delay in NPSM. Note
degradation is not as significant as PSM. Since NPSM that the average packet delay does not include the delay
does not have the ATIM window, more bandwidth can be for lost packets. The simulation results for the scenario
used for data transmissions. However, like PSM, NPSM with two flows per source node (Fig. 5(c)) are similar to
also has extra beacon transmissions for synchronization. those of one flow per source scenarios (Fig. 5(a) and (b)).
This is why the the aggregate throughput of NPSM is
slightly lower than that of 802.11 in Fig. 3(a), (b), and B.2 Wireless LAN: varying packet size
(c). Note that the simulation results for the scenario with
two flows per source node (Fig. 3(c)) are similar to those Fig. 6 and 7 show the aggregate throughput varying
for one flow per source (Fig. 3(a) and (b)). packet sizes in a wireless LAN with the network load of
 7

Aggregate Throughput (10 flows) Aggregate Throughput (20 flows) Aggregate Throughput (40 flows)
3000 3000
Aggregate Throughput (Kbps)

Aggregate Throughput (Kbps)

Aggregate Throughput (Kbps)

2000
2500 2500

2000 2000 1500

X 1500
X 1500
X
1000
1000 1000
802.11 802.11 500 802.11
500 500
PSM PSM PSM
NPSM NPSM NPSM
0 0 0
10 20 30 40 50 10 20 30 40 50 10 20 30 40 50
Network Load (%) Network Load (%) Network Load (%)

(a) 20 node network (b) 40 node network (c) 60 node network

Fig. 3. Aggregate throughput: wireless LAN with fixed packet size

Kbits delivered per Joule (10 flows) Kbits delivered per Joule (20 flows) Kbits delivered per Joule (40 flows)
Kbits delivered per Joule (Kbits/Joule)

Kbits delivered per Joule (Kbits/Joule)

Kbits delivered per Joule (Kbits/Joule)

160 100 40
140 35
80
120 30
100 60 25
80 20
60 40 15
40 10
802.11 20 802.11 802.11
20 PSM PSM 5 PSM
NPSM NPSM NPSM
0 0 0
10 20 30 40 50 10 20 30 40 50 10 20 30 40 50
Network Load (%) Network Load (%) Network Load (%)

(a) 20 node network (b) 40 node network (c) 60 node network

Fig. 4. Total data delivered per joule (Kbits/Joule): wireless LAN with fixed packet size

Average Packet Delay (10 flows) Average Packet Delay (20 flows) Average Packet Delay (40 flows)

802.11 802.11
Average Packet Delay (msec)

Average Packet Delay (msec)

Average Packet Delay (msec)

PSM PSM 100

100
100 NPSM NPSM

Y Y 10
Y 10
10 802.11
PSM
NPSM

1 1 1

10 20 30 40 50 10 20 30 40 50 10 20 30 40 50
Network Load (%) Network Load (%) Network Load (%)

(a) 20 node network (b) 40 node network (c) 60 node network

Fig. 5. Average packet delay: wireless LAN with fixed packet sizes (vertical axis uses log scale)

20% and 40%, respectively. The corresponding total data aggregate throughput, and this does not change with the
delivered per joule is shown in Fig. 8 and 9, respectively. packet size.
Simulated packet sizes are 256, 512, and 1024 bytes.
For NPSM with the packet size of 1024 bytes in Fig. 6,
Since the RTS/CTS overhead per packet is identical re- the aggregate throughput of NPSM is slightly lower than
gardless of the packet size, as the packet size increases that of 802.11 or PSM. However, with a high network load
in Fig. 6 and 7, the aggregate throughput of all schemes in Fig. 7, NPSM performs much better than PSM due to
also increases. The aggregate throughput of PSM is lower the ATIM overhead of PSM.
than 802.11 or NPSM, especially when the network load
In NPSM, as the packet size increases, the number of
is higher (please refer Fig. 7). As we explained in Fig. 3,
packets transmitted during the DATA window may be
the overhead of the ATIM window in PSM results in lower
reduced because a large packet takes more time to be
 8

Aggregate Throughput (10 flows) Aggregate Throughput (20 flows) Aggregate Throughput (40 flows)
2500 2500 1800
Aggregate Throughput (Kbps)

Aggregate Throughput (Kbps)

Aggregate Throughput (Kbps)

1600
2000 2000 1400
1200
X 1500
X 1500
X 1000
800
1000 1000
600
500 802.11 500 802.11 400 802.11
PSM PSM 200 PSM
NPSM NPSM NPSM
0 0 0
256 512 1024 256 512 1024 256 512 1024
Packet Size (Bytes) Packet Size (Bytes) Packet Size (Bytes)

(a) 20 node network (b) 40 node network (c) 60 node network

Fig. 6. Aggregate throughput: wireless LAN with different packet sizes (network load of 20%)

Aggregate Throughput (10 flows) Aggregate Throughput (20 flows) Aggregate Throughput (40 flows)
4500 4500 3500
802.11 802.11 802.11
Aggregate Throughput (Kbps)

Aggregate Throughput (Kbps)

Aggregate Throughput (Kbps)

4000 PSM 4000 PSM 3000 PSM
3500 NPSM 3500 NPSM NPSM
2500
3000 3000
X 2500 X 2500 X 2000
2000 2000 1500
1500 1500
1000
1000 1000
500 500 500

0 0 0
256 512 1024 256 512 1024 256 512 1024
Packet Size (Bytes) Packet Size (Bytes) Packet Size (Bytes)

(a) 20 node network (b) 40 node network (c) 60 node network

Fig. 7. Aggregate throughput: wireless LAN with different packet sizes (network load of 40%)

transmitted. If the DATA window size is too small, the are similar to those for the wireless LAN. However, since
aggregate throughput of NPSM may degrade with a large a packet travels four hops (on average) to reach a desti-
packet size – specifically, if there is severe congestion nation in our scenario, there is more energy consumption
during the DATA window, only a few nodes can transmit by intermediate nodes that forward packets to destination
data and the other nodes will go to the doze state. nodes.
However, this yields longer dozing period in NPSM.
Fig. 10 shows the simulation results for the multi-hop
Therefore, when the data delivered per joule (Kbits/joule)
network. Similar to the simulation results in Fig. 3, all
is compared (see Fig. 8 and 9), NPSM performs the best
schemes perform comparably when the network load is
among all schemes. Particularly, with the packet size of
low in Fig. 10(a). However, as the network load increases,
1024 bytes, the gap between NPSM and 802.11 or PSM is
the aggregate throughput with PSM and NPSM is lower
much greater compared a packet size of 256 or 512 bytes.
than 802.11. However, NPSM performs better than PSM
PSM performs better than 802.11 with a large packet size
since it does not have the ATIM window overhead.
and a low network load as seen in Fig. 8. However, with a
small packet size in Fig. 8 or with a high network load in NPSM performs better than PSM and 802.11 with re-
Fig. 9, PSM does not save energy as compared to 802.11 spect to the total data delivered per joule in Fig. 10(b).
because of the overhead of the RTS/CTS exchange as In contrast with the wireless LAN (Fig. 3), PSM gener-
well as the overhead of the ATIM window (explained in ally performs better than 802.11 in a multi-hop network in
Fig. 4). Note the curves for PSM and 802.11 in Fig. 9(b) Fig. 10(b). This is because in a multi-hop network, nodes
are overlapped. that do not have any packet to forward can go to the doze
state, resulting in energy savings.
B.3 Multi-hop Network: varying the network load
Fig. 10(c) shows the average packet delay over all
We now present the simulation results for a multi-hop flows in the network. In Fig. 10(c), 802.11 without power
network. The simulation results for the multi-hop network saving mode performs the best among all schemes. As
 9

Kbits delivered per Joule (10 flows) Kbits delivered per Joule (20 flows) Kbits delivered per Joule (40 flows)
Kbits delivered per Joule (Kbits/Joule)

Kbits delivered per Joule (Kbits/Joule)

Kbits delivered per Joule (Kbits/Joule)

140
200 802.11 802.11 802.11
PSM PSM 50 PSM
120
NPSM NPSM NPSM
150 100 40

80 30
100
60
20
40
50
10
20

0 0 0
256 512 1024 256 512 1024 256 512 1024
Packet Size (Bytes) Packet Size (Bytes) Packet Size (Bytes)

(a) 20 node network (b) 40 node network (c) 60 node network

Fig. 8. Total data delivered per joule: wireless LAN with different packet sizes (network load of 20%)

Kbits delivered per Joule (10 flows) Kbits delivered per Joule (20 flows) Kbits delivered per Joule (40 flows)
Kbits delivered per Joule (Kbits/Joule)

Kbits delivered per Joule (Kbits/Joule)

Kbits delivered per Joule (Kbits/Joule)

140
802.11 802.11 802.11
PSM PSM 50 PSM
200 120
NPSM NPSM NPSM
100 40
150
80 30
100 60
20
40
50
10
20

0 0 0
256 512 1024 256 512 1024 256 512 1024
Packet Size (Bytes) Packet Size (Bytes) Packet Size (Bytes)

(a) 20 node network (b) 40 node network (c) 60 node network

Fig. 9. Total data delivered per joule: wireless LAN with different packet sizes (network load of 40%) – the curves for PSM and 802.11 are
overlapped in (b).

Aggregate Throughput (10 flows) Kbits delivered per Joule (10 flows) Average Packet Delay (10 flows)
Kbits delivered per Joule (Kbits/Joule)

500 10
802.11 802.11
Aggregate Throughput (Kbps)

Average Packet Delay (msec)

450 PSM PSM

400 NPSM 8 NPSM
350 10
X 300
250
6
Y 802.11
PSM
200 4 NPSM
150
100 2
1
50
0 0
10 20 30 40 50 10 20 30 40 50 10 20 30 40 50
Data Rate per Flow (Kbps) Data Rate per Flow (Kbps) Data Rate per Flow (Kbps)

(a) Aggregate throughput (b) Total data delivered per joule (c) Average packet delay
(Kbits/Joule)
Fig. 10. Multi-hop network: 50 nodes with 10 flows (fixed packet size). Vertical axis uses log scale in (c).

the network load increases the delay for NPSM gets B.4 Multi-hop Network: varying packet size
longer. In Fig. 10(c), the packet delay for PSM is slightly
Fig. 11 shows the simulation result for a multi-hop net-
shorter than NPSM, but the aggregate throughput of PSM
work with various packet sizes. Each flow generates traf-
is worse than that of NPSM (see Fig. 10(a)); recall that
fic at the rate of 50 Kbps. In Fig. 11(a), NPSM gener-
lost packets are not considered for packet delay. There
ally performs better than PSM. For the total data deliv-
is a trade-off between energy savings and packet delay.
ered per joule in Fig. 11(b), PSM with the packet size of
NPSM gives a long dozing time (more energy savings, in
256 bytes performs worse than 802.11 due to the low ag-
Fig. 10(b)) with the cost of a longer packet delay.
gregate throughput (Fig. 11(a)). NPSM performs the best
for reasons explained in the wireless LAN case.
 10

Aggregate Throughput (10 flows) Kbits delivered per Joule (10 flows)

Kbits delivered per Joule (Kbits/Joule)

500
Aggregate Throughput (Kbps)

12
400
10

Z 300 8

6
200
4
100 802.11 802.11
PSM 2 PSM
NPSM NPSM
0 0
256 512 1024 256 512 1024
Packet Size (Bytes) Packet Size (Bytes)
(a) Aggregate throughput (b) Total data delivered per joule (Kbits/Joule)
Fig. 11. Multi-hop network: 50 nodes with 10 flows (50 Kbps per flow, with different packet sizes)

V. C ONCLUSION [6] P. J. M. Havinga and G. J. M. Smit, “Energy-efficient TDMA

Medium Access Control Protocol Scheduling,” in Proc. Asian
International Mobile Computing Conference (AMOC 2000),
We have presented a new power saving MAC protocol, November 2000.
NPSM. The aggregate throughput of IEEE 802.11 PSM [7] P. Johansson, R. Kapoor, M. Kazantzidis, and M. Gerla, “Ren-
degrades as the network load increases due to the time dezvous Scheduling in Bluetooth Scatternets,” in ICC 2002,
overhead of the ATIM window. Also, in PSM, extra en- vol. 1, pp. 318–324, April 2002.
[8] E.-S. Jung and N. H. Vaidya, “An Energy Efficient MAC Proto-
ergy is consumed by transmitting ATIM and ATIM-ACK
col for Wireless LANs,” in INFOCOM 2002, June 2002.
frames during the ATIM window. [9] A. Kamerman and L. Monteban, “WaveLAN-II: A High-
Performance Wireless LAN for the Unilicensed Band,” Bell Labs
The NPSM removes the ATIM window overhead from Technical Journal, vol. 2, summer 1997.
PSM in IEEE 802.11 in order to increase channel capacity [10] Lucent, IEEE802.11 WaveLAN PC Card - User’s Guide, p.A-1.
for data transmission and reduce the energy consumption. [11] J. P. Monks, V. Bharghavan, and W. mei W. Hwu, “A Power Con-
Removing the ATIM window makes more sense when the trolled Multiple Access Protocol for Wireless Packet Networks,”
in INFOCOM 2001, April 2001.
channel bandwidth is high, where a packet transmission
[12] S. Narayanaswamy, V. Kawadia, R. S. Sreenivas, and P. R. Ku-
occurs quickly. Simulation results confirm that NPSM mar, “Power Control in Ad-Hoc Networks: Theory, Architecture,
gives better aggregate throughput and energy savings as Algorithm and Implementation of the COMPOW protocol,” in
compared to PSM. Since nodes can enter the doze state in European Wireless 2002, February 2002.
NPSM, the average packet delay of NPSM is longer than [13] P. Nuggehalli, V. Srinivasan, and R. R. Rao, “Delay Constrained
Energy Efficient Transmission Strategies for Wireless Devices,”
IEEE 802.11 without using power saving mode. NPSM in INFOCOM 2002, June 2002.
saves energy at the cost of increasing the packet delay. [14] T. Simunic, H. Vikalo, P. Glynn, and G. D. Micheli, “Energy Ef-
ficient Design of Portable Wireless Systems,” in International
Symposium on Low Power Electronics and Design (ISLPED),
July 2000.
R EFERENCES [15] S. Singh, M. Woo, and C. S. Raghavendra, “Power-aware Rout-
ing in Mobile Ad Hoc Networks,” in MOBICOM ’98, October
[1] M. Adamou and S. Sarkar, “A Framework for Optimal Battery 1998.
Management for Wireless Nodes,” in INFOCOM 2002, June [16] A. Spyropoulos and C. Raghavendra, “Energy Efficient Commu-
2002. nications in Ad Hoc Networks Using Directional Antennas,” in
[2] C. Bisdikian, “An Overview of the Bluetooth Wireless Technol- INFOCOM 2002, June 2002.
ogy,” IEEE Communications Magazine, vol. 39, pp. 86–94, De- [17] M. Stemm and R. H. Katz, “Measuring and Reducing Energy
cember 2001. Consumption of Network Interfaces in Hand-Held Devices,” IE-
[3] B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris, ICE Transactions on Communications, special Issue on Mobile
“Span: An Energy-Efficient Coordination Algorithm for Topol- Computing, vol. E80-B, no. 8, pp. 1125–31, 1997.
ogy Maintenance in Ad Hoc Wireless Networks,” in MOBICOM [18] The CMU Monarch Project., “The CMU Monarch Project’s
2001, July 2001. Wireless and Mobility Extensions to NS.”
[4] C. F. Chiasserini and R. R. Rao, “Energy Efficient Battery Man- [19] The editors of IEEE 802.11, Wireless LAN Medium Access Con-
agement,” in INFOCOM 2000, March 2000. trol (MAC) and Physical Layer (PHY) Specification, 1997.
[5] J. Gomez, A. T. Campbell, M. Naghshineh, and C. Bisdikian, [20] Y.-C. Tseng, C.-S. Hsu, and T.-Y. Hsieh, “Power-Saving Proto-
“Conserving Transmission Power in Wireless Ad Hoc Net- cols for IEEE 802.11-Based Multi-Hop Ad Hoc Networks,” in
works,” in ICNP’01, November 2001. INFOCOM 2002, June 2002.
 11

[21] J. E. Wieselthier, G. Nguyen, and A. Ephremides, “Energy-

Limited Wireless Networking with Directional Antennas: The
Case of Session-Based Multicasting,” in INFOCOM 2002, June
2002.
[22] H. Woesner, J.-P. Ebert, M. Schlager, and A. Wolisz, “Power-
Saving Mechanisms in Emerging Standards for Wireless LANs:
The MAC Level Perspective,” IEEE Personal Communications,
June 1998.
[23] Y. Xu, J. Heidemann, and D. Estrin, “Geography-informed En-
ergy Conservation for Ad Hoc Routing,” in MOBICOM 2001,
July 2001.
[24] W. Ye, J. Heidemann, and D. Estrin, “An Energy-Efficient MAC
Protocol for Wireless Sensor Networks,” in INFOCOM 2002,
June 2002.
[25] R. Zheng and R. Kravets, “On-demand Power Management for
Ad Hoc Network,” Tech. Rep. UIUCDCS-R-2002-2280, Depart-
ment of Computer Science, University of Illinois at Urbana-
Champaign, 2002.
[26] H. Zhu, G. Cao, G. Kesidis, and C. Das, “An Adaptive Power-
Conserving Service Discipline for Bluetooth,” in ICC 2002,
vol. 1, pp. 303–307, April 2002.