Вы находитесь на странице: 1из 9

Performance analysis of an improved power-saving

medium access protocol for IEEE 802.11 point


coordination function WLAN
G.A. Safdar and W.G. Scanlon
Abstract: Wireless enabled portable devices must operate with the highest possible energy efciency
while still maintaining a minimum level and quality of service to meet the users expectations. The
authors analyse the performance of a new pointer-based medium access control protocol that was
designed to signicantly improve the energy efciency of user terminals in wireless local area
networks. The new protocol, pointer controlled slot allocation and resynchronisation protocol
(PCSAR), is based on the existing IEEE 802.11 point coordination function (PCF) standard.
PCSAR reduces energy consumption by removing the need for power saving stations to remain
awake and listen to the channel. Using OPNET, simulations were performed under symmetric
channel loading conditions to compare the performance of PCSAR with the infrastructure power
saving mode of IEEE 802.11, PCF-PS. The simulation results demonstrate a signicant
improvement in energy efciency without signicant reduction in performance when using
PCSAR. For a wireless network consisting of an access point and 8 stations in power saving mode,
the energy saving was up to 31% while using PCSAR instead of PCF-PS, depending upon frame
error rate and load. The results also show that PCSAR offers signicantly reduced uplink access
delay over PCF-PS while modestly improving uplink throughput.

Introduction

Energy consumption is a major performance metric for


wireless communication networks that support portable
devices. The capabilities of portable computing devices are
often limited by the size and the lifetime of the batteries that
power them, so minimising the energy usage of every
component in a mobile system is an important design goal.
Consequently there has been an increasing interest in
investigating energy efcient wireless communication protocols. In particular, medium access control (MAC)
protocols can have a signicant impact on energy consumption since both transmitting and receiving are costly
in energy terms. There are two main types of energy
consumption with regard to network operations [1]:
communication-related, and computational-related. Communication-related energy consumption is dependent on
the state of the node, e.g. energy consumed by the mobile
device in the doze or sleep state is less than the energy
consumed in the transmit or receive states. In turn, the state
of each node in the network is determined by the MAC
protocol, trafc loading and channel conditions. Computational energy consumption is related to processing aspects:
programmable logic, CPU, memory and to some extent
disk storage and other components. Hence, there may be a
r The Institution of Engineering and Technology 2006
IEE Proceedings online no. 20045093
doi:10.1049/ip-com:20045093
Paper rst received 28th June 2004 and in nal revised form 22nd February
2006
The authors are with the School of Electrical and Electronic Engineering, The
Queens University of Belfast, Ashby Building, Stranmillis Road, Belfast,
Northern Ireland BT9 5AH, UK
E-mail: w.scanlon@ee.qub.ac.uk
IEE Proc.-Commun., Vol. 153, No. 5, October 2006

trade-off between communication and computation energy


consumption since techniques that strive to achieve lower
communication consumption may result in higher computational consumption and vice versa. The system architecture and philosophy also have a role to play in energy
efciency. For example, in thin client and terminal services
applications portable terminals are used only for input and
display processing reducing overall energy consumption [2].
Previous research has considered energy conservation at
various layers of the protocol stack, including work on
routing [3], MAC [4] and transport control protocols [5].
Regardless of the target standard (i.e. IEEE 802.11) a
number of general principles have been established. For
example, collisions should be eliminated as much as possible
within the MAC layer [6], since they result in retransmissions and retransmissions lead to unnecessary power
consumption. Likewise, transmission power management
is known to improve energy saving in wireless networks and
also helps to avoid collisions. Jung and Vaidya [7] presented
a MAC protocol that saves energy by allowing stations to
vary their transmit power level on a per-frame basis. Ebert
et al. [8] described a protocol that has a frame length
dependent power control mechanism to ensure minimum
energy expenditure for each transmitted bit of information,
while Qiao et al. [9] used physical rate adaptation and
transmit power control.
Jung et al. [10] presented a new power saving mechanism
(NPSM) where energy was saved by removal of the ATIM
(ad hoc trafc indication map) window to reduce the control
overhead and to reuse the saved bandwidth for data
transmission. Higher energy savings can also be achieved by
controlling the on and off timings of a stations RF
transceiver. For example, Singh et al. [11] introduced a
power aware multiple access protocol with signalling
(PAMAS) where energy was conserved by intelligently
697

powering off stations that are not actively transmitting or


receiving packets. Feeney [12] presented a quality-of-service
aware power saving protocol in which each station
establishes a periodic sleep-awake cycle. The relative phase
difference between sleep-awake cycles is estimated by
the stations wishing to communicate and adjustment of
these phase relationships helps to avoid contention among
stations leading to increased effective bandwidth.
Improved scheduling algorithms also help in achieving
lower energy consumption. A centralised scheduling
mechanism that computes the system transmission schedule
at the access point (AP) or base station is more energy
efcient than a distributed algorithm where each mobile
station computes the schedule independently. The mobile
should be allocated contiguous slots by the scheduler for
transmission or reception to reduce turn around time
values. Chen et al. [13] provide a low power scheduling
algorithm that considers contiguous allocation and aggregate packet requests. Information about the stations
battery power level can be used at the scheduler in an AP
for energy saving by allowing trafc from the low power
stations to be transmitted sooner in order to avoid frame
loss due to depletion of their power reserves [14, 15]. Energy
saving can also be obtained at the link layer by taking into
account the channel conditions, so that the transmissions
may be avoided when the channel conditions are poor.
Zorzi et al. [16] presented a low power error control that
slows down data transmission when degraded channel
conditions are encountered. Error control schemes that
combine automatic repeat request (ARQ) and forward
error correction (FEC) mechanisms can also be used to
conserve energy [17]. Many systems require the receiver at
each mobile station to remain active at all times for channel
listening. This results in signicant energy wastage even
when clear channel assessment is not required and
techniques such as address listening are used. For example,
in [18] a pointer-based MAC protocol was used to create a
form of reservation controlled channel access that avoided
address listening in a centralised control network.
In this paper, the performance of a new protocol, pointer
controlled slot allocation and resynchronisation (PCSAR),
is compared with the standard point coordination function
(PCF) power saving mode (PCF-PS) in IEEE 802.11. PCF
is a centralised MAC protocol for infrastructure networks
and PCSAR is a combination of the pointer-based channel
access mechanism of [18] and PCF-PS. PCSAR achieves a
further reduction in energy consumption over PCF-PS by
increasing the opportunities for mobile stations to power
down their transceiver components.

the next station in its polling list after the PCF-inter frame
space (PIFS). This allows the AP to retain control of the
medium (Fig. 1).
PCSAR is a contention-free protocol where each station
is assigned at least one DL transmission opportunity within
the contention-free period (CFP), normally followed
immediately by a corresponding UL opportunity (Fig. 1).
PCSAR requires a time-slotted architecture so UL and DL
frames are restricted to a maximum length with frame
assembly and disassembly needed at both the AP and the
stations. In PCSAR, a slot is dened as the time needed to
transmit two maximum length MAC frames (one DL and
one UL) plus two SIFS. Like PCF-PS, PCSAR is based on
a superframe consisting of CFP and a contention period
(CP) and the transmission mode is time division duplex. In
PCSAR, the CP is used only for association with the AP.
During the CFP, the AP takes control of the medium and
starts contention-free services after the PIFS interval. The
AP broadcasts a PCSAR beacon management frame that is
otherwise identical to the IEEE 802.11 PCF-PS beacon
frame but with the addition of a series of next slot pointers
(NSP) and a resynchronisation slot pointer (RSP). These
new pointers are explained below. After the beacon frame,
the AP can send a Data Poll; a Data Poll Ack or a
Poll Ack frame to each station. Each station then replies
with a Data Ack or a Null Ack frame.
In PCF-PS, the AP periodically announces the trafc
buffered for stations by transmission of a trafc indication
map (TIM) within the broadcast beacon management
frame. Any PS station that is contention-free pollable
(CF-pollable) and has data buffered at the AP is not
allowed to resume sleeping. All such PS stations must stay
awake, listening to the channel until the buffered frame
(MAC protocol data unit MPDU) is retrieved from the
AP or the CFP ends. The AP transmits any pending frames
to the stations whose TIM bits are set in ascending order

Super Frame Repetition Interval

Super Frame Repetition Interval


CFP

CP

CFP

CP

PIFS
B

PCF

DCF

SIFS

SIFS

PCF

DCF

PIFS

SIFS

PIFS

PCF-PS and PCSAR protocols

Centralised infrastructure networks provide the greatest


opportunity for energy saving in IEEE 802.11 wireless
networks [19]. In PCF-PS, all downlink (DL) trafc
destined for power saving (PS) mobile stations in a wireless
network is buffered at the AP, allowing stations to save
energy by remaining in sleep mode. The AP transmits
the buffered frames between the beacon and the contention
free end (CF-END) frame using the short inter frame
space (SIFS) with either Data Poll or Data Poll Ack,
assuming error-free conditions. Each station then replies
after a SIFS with either a Data Ack or a Null Ack
frame depending upon its uplink (UL) queue status. If
a station fails to receive DL packets due to channel
errors then there is no UL transmission from the station to
the AP. In the absence of a UL transmission the AP
will perform a back off operation and sends Data Poll to
698

D+
POLL

D+
POLL

D+
POLL

CF _
END

U+
ACK

U+
ACK
No

SIFS

SIFS

Response
to D + POLL
PCF-PS Frame Transfer

SIFS

SIFS

SIFS

PIFS
B

D+
POLL

D+
POLL

CF _
END

U+
ACK

SIFS

U+
ACK

SIFS

PCSAR Frame Transfer

Fig. 1

PCSAR and PCF-PS superframe structure


IEE Proc.-Commun., Vol. 153, No. 5, October 2006

according to their IDs. Figure 2a shows the timing diagram


for 4 stations operating under PCF-PS, all having frames
buffered at the AP.
As with PCF-PS, an AP featuring PCSAR sets the bits in
the TIM eld element and assigns an initial transmission
slot for those stations that have frames buffered by setting
the appropriate NSP in the beacon frame. Stations then
only wake up in their assigned slot to receive the buffered
trafc from the AP and to respond with a UL frame. An
NSP is also added to every DL frame, so that additional
DL frames can be assigned to the station if required within
the current superframe (Fig. 2b). The assignment of
additional slots to each station leads to a very exible
system capable of supporting frame-by-frame adaptive
scheduling that can respond to quality of service, load
balancing or channel error conditions and constraints.
The PCSAR DL frame is identical to the PCF-PS DL
one but with the addition of two pointer elds, NSP and
RSP respectively. The length of both RSP and NSP elds is
settled according to the maximum number of slots available
in the CFP (7 bits for 128 slots in this paper). In PCSAR,
every station wakes to receive the broadcast beacon
management frame and the RSP then indicates the number
of remaining slots before the transmission of the next
beacon after the CP. The RSP is also included as part of
every DL frame transmitted by the AP. This is an important
aspect of PCSAR as it facilitates error recovery and allows
unsynchronised stations to simply awake and listen to any
DL frame before powering down to wait for the next
beacon. The RSP is updated for each DL frame, counting

down to the next registration beacon. The RSP is also used


by the stations wishing to associate with the AP, as the
value of RSP informs them of the number of remaining
slots before the start of the CP. Although the implementation considered in this paper is symmetric (with one UL
frame following each DL frame), PCSAR can support
adaptive asymmetrical operation in the DL direction. By
setting the NSP to 0 in a DL frame, the AP can indicate
that instead of being allowed to make a UL transmission
after the SIFS, the station must wait to receive a further
DL frame. Therefore, the AP (and the scheduler employed)
has full control over the degree of asymmetry for
each individual station and this can change within each
superframe.
Under PCF-PS, all stations settle a listen interval with the
AP at the time of association, which is expressed in units of
beacon intervals. Stations wake up at these intervals to
receive the broadcast beacons from the AP. Stations with
frames buffered at the AP are not allowed to resume
sleeping until the buffered data has been retrieved from the
AP or until the end of the CFP. The listen interval value is
critical; if the listen interval is too large, then the AP may
drop the frames for a particular station due to the aging
factor, whereas if the listen interval value is too small, then
the station energy consumption will increase. In PCSAR
there is no listen interval negotiation with the AP. Also in
PCSAR, a station goes back to sleep as soon as it receives
the beacon frame, even if it still has frames buffered at the
AP; a station only wakes at the specic slot scheduled for its
transaction with the AP.

2.1 Analytical analysis of PCSAR and


PCF-PS downlink
The steady state performance of PCSAR under error-free
conditions was analytically compared with PCF-PS
using queuing theory for different trafc loads and 8
mobile stations. In PCSAR, the service time probability
distribution is modelled as general since during the CFP
the queues are serviced with the minimum service time
(TService Min PCSAR ), while once per superframe the presence of
the beacon, CF_END and CP portion will cause the service
time to increase to its maximum value (TService Max PCSAR ).
Therefore, M/G/1 was used to analyse PCSAR, assuming
exponential trafc arrivals.
Based on the earlier description of the PCSAR protocol,
TService Max PCSAR is calculated as follows:
TService Max PCSAR
bTBeacon 2TSIFS  x TDL TUL  x TCF
2y

END

TCP c
1

Fig. 2

Timing diagrams

a PCF-PS
b PCSAR
IEE Proc.-Commun., Vol. 153, No. 5, October 2006

Here, x is the number of slots in the superframe and y is the


number of stations. For all of the results in this paper,
x 128 and y 8 and TDL TUL . Using the values of
TBeacon , TSIFS , TDL , TCF END and TCP given in Table 1,
TService Max PCSAR was 17.51 ms. However, within the CFP
the NSP mechanism ensures that queues can be serviced
more than once depending on the available slots.
This marginally reduces the service time from
TService Max PCSAR to TService Min PCSAR (17.44 ms without
TBeacon ; TCF END and TCP ). Therefore, assuming roundrobin scheduling and with x 128 and y 8, each queue
will be serviced 7 times at TService Min PCSAR and once at
TService Max PCSAR during the xed CFP. Hence, the mean
service time, Ts, remains constant for all channel loading
values at 17.47 ms. The corresponding standard deviation
(computed using the biased or n method provided within
699

Parameter

Symbol

Value

TSIFS

10 ms

20 ms

SIFS time
PIFS

Slot time + SIFS Time

Data rate

11 Mbit/s

Layer three mean


packet size (error-free)

16 500 bit

Layer three mean


packet size (with errors)

11 000 bit

Layer two payload


(error-free)

11 008 bit

Layer two payload


(with errors)

7 008 bit

MAC frame time


(error-free)

Throughput (Mbps)

Slot time

0.05

0.4

0.04

0.3

0.03

0.2

0.02

0.1

0.01
0
20

40

60

80

100

% Loading

Fig. 3 Analytical analysis of PCSAR and PCF-PS DL


performance

54  slot time

TCP

MAC frame time + SIFS time

Beacon time

TBeacon

3  slot time

CF-END time

TCF-END

3  slot time
2  slot time

PS-POLL time

Microsoft Excel) of PCSAR service time sTs is 0.02315 ms.


Using the useful payload (Table 1) and TService Max PCSAR , the
maximum PCSAR DL throughput SMax PCSAR was calculated as 0.62866 Mbit/s per station.
According to M/G/1 queuing theory, the average waiting
time in the queue including processing time (the residence
time) is given by [20]:
rTs A
2
Wq Ts
1  r
where mean arrival rate l 1=TInterArrival , m 1=Ts , the
utilisation factor r l=m and A 1=21 sTs =Ts 2 ,
while TInterArrival is directly calculated from the required
channel loading.
For PCF-PS, the CFP length is determined by the DL
queue status. Therefore, the average value of PCF-PS
service time Ts will directly depend upon channel loading
and the maximum value will occur at the maximum CFP
length since the probability that all the stations in the
network will have frames queued at the AP is very high
1 at increased channel loading values. The PCF-PS
maximum service time, TService PCF PS Max , depends on the
number of active stations (y):
TBeacon 2TSIFS TDL TUL
 y TCF

END

TCP

Using (3) with y 8 and the other parameters as given in


Table 1, TService PCF PS Max was calculated as 18.65 ms. Given
the direct interdependency of service time, load and CFP
length, we can assume that the mean PCF-PS service time is
directly proportional to trafc load, i.e. it is constant for
each individual value of load considered in this analysis,
justifying a deterministic distribution. The error associated
with this assumption is greatest at low values of load.
Hence, M/D/1 queuing theory was used for the analysis of
PCF-PS for each specic value of trafc load. The M/D/1
residence time is given by:
1
r
4
Wq
m 2m1  r
700

0.5

CP

PCF PS Max

0.06

36  slot time

MAC frame time


(with errors)

TService

PCSAR Acc Delay


PCF-PS Acc Delay

0.6

TDL

PCSAR Tput
PCF-PS Tput

Access Delay (Sec)

Table 1: Main queuing model and simulation parameters

Figure 3 shows the DL throughput and access delay values


for both PCSAR and PCF-PS obtained by queuing
analysis. Both access delay and throughput increase with
increased loading for PCSAR and PCF-PS. The results
show that PCF-PS DL access delay is lower than for
PCSAR, particularly at light trafc load values, whereas
PCSAR throughput is marginally higher than PCF-PS,
particularly at higher loading.
3

Simulation details

Using discrete event simulation, PCSAR was compared


with PCF-PS for an IEEE 802.11b network of 8 stations
and an AP with symmetrical DL and UL trafc conditions.
Several different scenarios were simulated both with and
without channel errors. Simulation models for both PCFPS and PCSAR were developed using OPNET and the key
simulation parameters are given in Table 1. Results were
obtained by simulating the network for a period of 60 s. The
AP handled all DL trafc and each station had an
independent UL trafc source. Network layer data was
simulated by using bursty trafc sources with an exponential inter-arrival rate and a uniformly distributed packet
size. Error-free simulations were performed at a range of
symmetrical trafc loading values, whereas simulations with
channel errors were performed at a 60% trafc load for
different values of frame error rate (FER). In all cases the
trafc load was specied relative to the useful data capacity
of the network. The useful data capacity was determined by
considering the possible network layer data transfer
between the transmission of the two consecutive beacon
frames. In the error-free simulation scenarios a maximum
MAC layer payload value of 1376 bytes was used, giving a
useful data capacity of 10.054 Mbit/s. The corresponding
100% load for each of the 8 stations was 628.37 kbit/s in
both DL and UL directions. To avoid having an overly
excessive number of retransmissions under channel error
conditions, the MAC layer payload was reduced to an
arbitrary but xed maximum of 876 bytes (useful data
capacity was 9.554 Mbit/s, 597.125 kbit/s per trafc source).
In both PCSAR and PCF-PS simulations the AP uses
pure round robin scheduling and starts the CFP by taking
control of the medium after PIFS and broadcasting a
beacon management frame. The AP maintains an independent per-destination DL queue for each station, while each
station maintains its own UL queues. Segmentation and
reassembly was used in both UL and DL due to the bursty
nature of the trafc generated by the trafc sources. In all of
IEE Proc.-Commun., Vol. 153, No. 5, October 2006

PCSAR Tput

PCSAR Acc Delay

PCF-PS Tput

PCF-PS Acc Delay

0.8

0.6

3
Throughput

0.4

2
Access Delay

0.2

Results and discussion

Access Delay (Sec)

Throughput (Mbps)

the simulations in this paper, the length of the CFP was


variable for PCF-PS as it depended on the status of the
TIM (i.e. DL trafc) at the time of the beacon transmission.
In the PCSAR simulations, those stations without frames
buffered at the AP were still assigned a transaction slot
using the NSP in the beacon frame. Therefore, the PCSAR
CFP was xed at 128 slots (280.29 ms for error-free and
187.77 ms with channel errors). This gives an opportunity
for the station to send back any UL frames that might have
been buffered at the station, increasing UL throughput and
decreasing UL frame access delay.

0
0

10

20

30

40

50

60

70

80

100

90

% Loading

Error-free throughput and access delay

PCSAR Tput

PCSAR Acc Delay

PCF-PS Tput

PCF-PS Acc Delay

Fig. 5

0.6

3
Throughput

0.4

2
Access Delay

0.2

throughput (up to 8% higher at loads below saturation).


Furthermore, the UL mean access delay results are much
better for PCSAR than for PCF-PS at lower values of
channel loading. The delay values for PCF-PS monotonically decreased with channel loading up to a value of 70%,
after which queue saturation caused an exponential rise,
similar to the DL case. With PCF-PS, the requirement that
UL trafc must wait until the station is polled (i.e. the
stations TIM bit is set due to DL trafc being buffered at
the AP) means that UL trafc is unduly delayed in
symmetrical light loading conditions. As the loading
increases, the higher levels of DL trafc increases the
number of UL transmission opportunities that a station
receives leading to a reduction in UL access delay.

4.2 Error-free energy consumption and


transmission efficiency
A Prism 2.5, 3.3 V, IEEE 802.11b network interface card
was used as a reference for energy consumption calculations: a stations operating current was 290 mA for
transmitting, 205 mA for receiving and 62 mA for doze or
sleep mode. Figure 6 shows the average station energy
consumption versus channel loading for both PCSAR and
PCF-PS. PCSAR energy consumption was always lower
than PCF-PS, due to almost zero channel listening time for
PCSAR, where every station only wakes up in its own
assigned transaction slot. With PCF-PS the energy
consumption dramatically rose at high load due to
increased channel listening time. Averaged over all loading
conditions, a PCSAR station required 25% less energy than
it would have with PCF-PS, while the energy saving at
saturation point (70% loading) was more than 40%.

10

20

30

40

50

60

70

80

90

400
300
200
100
0

100

% Loading

Fig. 4

Error-free DL throughput and mean access delay

IEE Proc.-Commun., Vol. 153, No. 5, October 2006

PCF-PS

500

0
0

Error-free UL throughput and mean access delay

PCSAR

0.8

Access Delay (Sec)

Throughput (Mbps)

Error-free results were obtained at channel loading values


ranging from 10% to 90% in steps of 10% and were used to
compare the overall per station throughput and mean
access delay averaged over all 8 stations. Normalised
throughput (Fig. 4) was almost identical for both PCSAR
and PCF-PS and was found to increase linearly with
channel loading until the network saturates around 70%
offered load. This saturation was caused by segmentation of
the bursty trafc sources necessarily producing unlled
frames (with high overhead) and is the reason why the
simulated DL throughput (Fig. 4) is consistently higher
than analytical throughput (Fig. 3) at all loading values.
The additional overhead needed for the PCSAR pointers
(63 bits for the beacon and 14 bits for each DL frame) was
low and had no signicant effect on DL throughput.
Figure 4 also compares the mean access delay calculated
across all 8 stations. The DL access delay trend was similar
for both protocols, but PCSAR delay was always around
65% higher than PCF-PS at a load of 60% or less.
However, queue saturation caused an exponential growth in
delay at 70% load and above and PCSAR performed better
under these conditions with PCF-PS having at least 41%
higher delay. At light and medium loads the advantage of
PCF-PS is that the CFP duration is directly proportional to
the number of stations with trafc queued at the AP, while
in this work we assumed that PCSAR had a xed duration
CFP. At a channel loading of 60% or lower, the DL access
delay values obtained by simulation were consistent with
the analytical results, while beyond 60% load, queue
saturation effects preclude a meaningful comparison.
Figure 5 shows the UL throughput and mean access
delay results for the same error-free simulations. Unlike
PCF-PS, in PCSAR stations are always assigned a UL
transmission opportunity even in the absence of any data
buffered at the AP, resulting in slightly higher UL

Energy Consumption (mW)

4.1

10

20

30

40

50

60

70

80

90

100

% Loading

Fig. 6

Error-free average station energy consumption


701

PCF-PS

DL Re - Tx Frames

1000

800

600

Frames

Transmission Efficiency
(Kbits/mJ)

PCSAR
5

2
1

400
200

0
0

10

20

30

40

50

60

70

80

90

100

0.0001

0.001

% Loading

0.01

0.1

FER

Error-free transmission efficiency

Fig. 9

Due to the slightly higher average throughput and lower


average energy consumption, PCSAR also offers better
transmission efciency than PCF-PS (Fig. 7). For both
PCSAR and PCF-PS, transmission efciency increased with
channel loading. However, for PCF-PS the transmission
efciency decreased after 60% of channel loading due to the
sharp increase in energy consumption shown in Fig. 6. On
average, PCSAR stations were 29% more transmission
efcient than PCF-PS ones.

PCSAR DL retransmission and CF-POLL frames

PCSAR UL Re - Tx Frames
PCF-PS UL Re - Tx Frames

PCSAR Null Frames


PCF-PS Null Frames

1000
800

Frames

Fig. 7

CF-POLL Frames

600
400

Uplink
Retransmissions

200

4.3 Throughput with uncorrelated channel


errors
An uncorrelated channel error model was used to generate
results for PCSAR and PCF-PS performance under nonideal transmission conditions. Results were obtained for
values of frame error rate (FER) ranging from 10  4 to
10  1 at a symmetric load of 60% to avoid queue saturation
effects. The success of each packet was determined using a
uniform pseudorandom number generated from a subtractive algorithm [21]. DL throughput, which includes
frame retransmissions, was found to be almost constant at
492 kbit/s for both PCF-PS and PCSAR across all values of
FER considered. Figure 8 shows how FER affected the
number of retransmissions and in turn, a stations channel
listening and off time for both PCF-PS and PCSAR.
For PCSAR, a CF-POLL frame is sent to the station in
the absence of a DL data frame. Thus at higher values of
the FER the bandwidth initially taken by CF-POLL frames
is released and is taken by data frames (due to retransmissions Fig. 9), which results in a constant DL throughput
for PCSAR.
UL throughput for PCSAR and PCF-PS also remained
fairly constant (at 464 kbit/s for PCF-PS and 489 kbit/s for
PCSAR DL Re - Tx Frames

PCF-PS DL Re - Tx Frames

PCSAR CH - LTN Time

PCF-PS CH - LTN Time

PCSAR Off Time

PCF-PS Off Time

0.001

0.01

0.1

FER

Fig. 10

UL retransmission and NULL frames

PCSAR, an improvement of 6.3%) while the FER was


below 0.05. This was due to the bandwidth that was taken
by NULL frames being released and reallocated to UL
retransmissions for both PCSAR and PCF-PS cases
(Fig. 10). PCSAR has a higher number of NULL frames
than PCF-PS because PCSAR stations are always assigned
a DL slot even in the absence of data at the AP and they
have an opportunity to send back either UL data packets or
NULL packets to the AP.

4.4 Access delay with uncorrelated


channel errors
Figure 11 shows the mean access delay calculated across all
8 stations under channel error conditions. As expected,

PCSAR_DL

PCSAR_UL

PCF-PS_DL

PCF-PS_UL

0.001

0.01

10

500

40
300
30
200
20
100

Time (Sec)

50

Access Delay (Sec)

60

400

Frames

0
0.0001

0.1

10

0.01
0

0.0001

0.001

0.01

FER

Fig. 8
702

0.0001

0.1

DL retransmission frames and channel listening/off time

0.1

FER

Fig. 11
errors

DL and UL mean access delay with uncorrelated channel

IEE Proc.-Commun., Vol. 153, No. 5, October 2006

Cumulative Distribution Function


Plot (CDF)

PCSAR 0.0001

PCSAR-60
PCF-PS-60

Energy Consumption (mW)

PCF- PS 0.0001

PCF- PS 0.1

0.8
PCSAR 0.1

0.6

0.4

0.2

500
400

60 % Loading

300
200

30 % Loading

100
0
0.0001

0.001

0
0

0.1

0.2

0.3

0.4

Fig. 14
errors

DL access delay CDF at FER of 10  4 and 10  1

Fig. 12

Cumulative Distribution Function


Plot(CDF)

PCSAR 0.0001

1
PCSAR 0.1

0.8
PCF- PS 0.0001

0.6
PCF- PS 0.1

0.4
0.2
0
0

Access Delay (Sec)

Fig. 13

UL access delay CDF at FER of 10  4 and 10  1

PCSAR DL access delay was signicantly higher (60%


averaged across all values of FER) than PCF-PS. However,
PCF-PS DL access delay increased at a much higher rate
than PCSAR for FER values greater than 0.05, while at
FER 0.1, PCF-PS DL delay is almost the same as for
PCSAR. Furthermore, even under channel error conditions,
PCSAR mean UL access delay results were on average 90%
lower than with PCF-PS.
Figures 12 and 13 shows the cumulative distribution
function (CDF) plots of the access delay at 60% offered
load for FER 10  4 and 10  1. The DL CDFs (Fig. 12)
illustrate that at low FER, PCF-PS performed better than
PCSAR, while at high FER there was no clear benet with
PCF-PS. The same was true for DL delay variance, where
at low error rates PCSAR variance was higher than for
PCF-PS (1.95 ms2 versus 0.39 ms2 at FER 10  4) but this
was not the case at higher error rates with 5.73 ms2 for
PCSAR compared to 6.10 ms2 for PCF-PS at FER 10  1.
The difference between PCSAR and PCF-PS UL CDFs
(Fig. 13) was more signicant. Although the UL load is
below saturation, the PCF-PS delay values extend up to
over 7.5 s compared with 250 ms for PCSAR. This highlights the fundamental limitation of the PCF-PS where
UL trafc is effectively given a lower priority even under
symmetric load conditions. In terms of UL delay variance,
PCSAR was signicantly better than PCF-PS, with 2.20 ms2
at FER 10  4 and 5.10 ms2 at FER 10  1 versus
284 ms2 and 4360 ms2, respectively, for PCF-PS.

The energy consumption was calculated for both protocols


under uncorrelated channel error conditions (Fig. 14) at
IEE Proc.-Commun., Vol. 153, No. 5, October 2006

0.1

Station energy consumption with uncorrelated channel

light (30%) and moderate (60%) channel loading, avoiding


saturation effects. Regardless of the FER, PCSAR station
energy consumption was lower than for PCF-PS, but this
advantage was less signicant at the lower load. At 60%
loading, PCSAR requires 25% less energy than PCF-PS
when averaged over all FER values, whereas at higher
values of FER the difference in energy consumption also
increases, e.g. at FER 10  1, PCF-PS required 31% more
energy then PCSAR. For the 30% case, the best saving (at
FER 10  1) was 15%. For higher levels of channel
loading (e.g. 70% and above), energy consumption under
channel errors would remain unchanged because frame
retransmissions would have no effect as trafc queues
would be already saturated (Fig. 6).
5

Conclusions

A new low power MAC protocol (PCSAR) for IEEE


802.11 infrastructure networks has been presented and its
steady state performance was compared with PCF-PS. The
proposed protocol is an improved version of PCF-PS that
uses pointers to increase the degree of control exercised by
the AP. PCSAR reduces energy consumption by removing
the need for power saving stations to remain awake and
listen to the channel during the contention free period until
their transaction with the AP is completed. Discrete event
simulation under symmetrical channel loading conditions
was used to assess the performance of PCSAR for a
network of 8 power saving stations and an AP. The results
presented demonstrate that PCSAR has consistently
improved performance over PCF-PS in terms of energy
consumption and transmission efciency. Uplink throughput and delay performance was much better with PCSAR,
regardless of trafc or channel conditions. However,
downlink performance of PCSAR was relatively poor,
particularly under light trafc, low error conditions.
PCSAR has the potential to be an extremely effective low
power protocol for wireless data networks. However, it
remains that further work should address the downlink
performance issues possibly by the use of a variable
contention free period duration within each superframe.
Additional benets may also be achieved by making use of
the pointer-based medium access control mechanism to
accommodate asymmetric operation and improved quality
of service performance through intelligent frame scheduling.
6

4.5 Energy consumption with uncorrelated


channel errors

0.01

FER

0.5

Access Delay (Sec)

PCSAR-30
PCF-PS-30

Acknowledgment

G.A. Safdar would like to thank the Commonwealth


Scholarship Commission for providing him with a Commonwealth Fellowship in relation to this work.
703

References

1 Jones, C.E., Sivalingam, K., Agrawal, P., and Chen, J.: A survey of
energy efcient protocols for wireless networks, Wirel. Netw., 2001, 7,
pp. 343358
2 Hadjiyiannis, G., Chandrakasan, A., and Devdas, S.: A low power
low bandwidth protocol for remote wireless terminals, Wirel. Netw.,
1998, 4, pp. 315
3 Xu, Y., Heidemann, J., and Estrin, D.: Geography informed energy
conservation for ad hoc routing. Proc. ACM Int. Conf. on Mobile
Computing and Networking, Rome, Italy, July 2001, pp. 7084
4 Chockalingarn, A., and Zorzi, M.: Energy efciency of media access
protocols for mobile data networks, IEEE Trans. Commun., 1998, 46,
pp. 14181421
5 Karvets, R., and Krishnan, P.: Application driven power management for mobile communication, Wirel. Netw., 2000, 6, pp. 263277
6 Sivalingam, K., Chen, J., Agrawal, P., and Srivastava, M.: Design
and Analysis of low-power access protocols for wireless and mobile
ATM networks, Wirel. Netw., 2000, 6, pp. 7387
7 Jung, E.S., and Vaidya, N.: A power control MAC protocol for
ad hoc networks. ACM Int. Conf. on Computing on Mobile
Computing and Networking, Mobicom, 2002, pp. 3647
8 Ebert, J.P., Stremmel, B., Wiederholt, E., and Wolisz, A.: An energy
efcient power control approach for wireless LANs, J. Commun.
Netw., 2000, 2, pp. 197206
9 Qiao, D., Choi, S., Soomro, A., and Shin, K.: Energy efcient PCF
operation of IEEE 802.11a WLANs via transmit power control, Int.
J. Comput. Telecommun. Netw., 2003, 42, pp. 3954
10 Jung, E.S., and Vaidya, N.: A power saving MAC protocol for
wireless networks. Technical Report, Texas A&M University, 2002

704

11 Singh, S., and Raghavendra, C.: Power aware multiple access


protocol with signalling for ad hoc networks, ACM SIGCOMM
Comput. Commun. Rev., 1998, 28, pp. 526
12 Feeney, L.: A QoS aware power save protocol for wireless
ad hoc networks. Technical Report, Swedish Institute of Technology,
2002
13 Chen, J.C., Sivalingam, K., Agrawal, P., and Acharya, R.: Scheduling
multimedia services for a low power MAC in wireless and
mobile ATM networks, IEEE Trans. Multimed., 1999, 1, (2),
pp. 187201
14 Kishore, S., Chen, J., Sivalingam, K., and Agrawal, P.: Adaptive
power control and scheduling algorithms based on battery power level
for CDMA wireless networks. IEEE ICUPC, 1998, pp. 967971
15 Price, C.: Power aware scheduling algorithms for wireless networks.
Masters thesis, Washington State University, 2000
16 Zorzi, M., and Rao, R.: Error control and energy consumption in
communications for nomadic computing, IEEE Trans. Comput.,
1997, 46, (3), pp. 279289
17 Lettieri, P., Fragouli, C., and Srivastava, M.: Low power error
control for wireless links. ACM/IEEE Int. Conf. on Mobile
Computing and Networking, 1997, pp. 139150
18 Chui, T., and Scanlon, W.: A novel MAC protocol for power efcient
short range wireless networking. IEEE Int. Conf. on Wireless LANs
and Home Networks, 2001, pp. 187196
19 Gast, M., and Gast, M.S.: 802.11 wireless networks: the denitive
guide (OReilly Networking, 2002)
20 Stallings, W.: High speed networks and Internets: performance and
quality of service (Prentice-Hall, 2002), Chap. 8
21 Kahaner, D., Moler, C., and Nash, S.: Numerical methods and
software (Prentice Hall, Englewood Cliffs, NJ, 1989), Chap. 10

IEE Proc.-Commun., Vol. 153, No. 5, October 2006

Вам также может понравиться