Wireless adapter sleep scheduling based on video streaming video?

Wireless adapter sleep scheduling based on video
QoE: How to improve battery life when watching
streaming video?
M´arton Csernai, Andr´as Guly´as
Budapest University of Technology and Economics
Department of Telecommunications and Media Informatics
1117 Budapest, Magyar tud´osok k¨or´utja 2. Hungary
Email: {csernai,gulyas}@tmit.bme.hu
Abstract—Power management is a critical issue for mobile
devices, where a significant portion of the total system energy
can be often consumed by heavy wireless network activity. In
this paper we present a framework that reduces the energy
consumption of wireless mobile devices during streaming video
content over Wireless LAN networks by utilizing the power
saving mechanisms defined in the IEEE 802.11e standard. The
framework allows video applications to directly optimize the
overall energy efficiency by controlling the sleep cycles of wireless
network adapters based on video QoE. The framework employs
a simple QoE estimation algorithm that estimates the user level
video quality, and adjusts the sleep intervals of the wireless
adapter to maintain video quality while maximizing power
efficiency. As an important part of our paper the building and
configuration steps of a working testbed system is carefully
described, which demonstrates the availability and efficiency of
the U-APSD power saving technology at hand. A preliminary
evaluation of the proposed framework supports that by enabling
the WMM Power Saving feature 20-30% of total system energy
can be saved even in case of asynchronous video traffic.
Index Terms—wireless LAN, IEEE 802.11e, U-APSD, QoE,
power management
Recently the Wireless LAN technology has spread to a
growing number of mobile phones and with the advent of
the tablets there are more and more battery limited wireless
mobile devices connecting to the Internet. The majority of user
initiated internet traffic is due to multimedia content (youtube,
facebook, video chat), which imposes a great load on mobile
devices with limited energy resources and the improvement of
battery life is a widely researched critical issue.
Many standards that define different mechanisms regarding
power management of the wireless network adapter, such as
IEEE 802.11e, have been around for years. However, to the
best of our knowledge there are only a few complete integrated
solutions that utilize the standardized technologies [1]. It is
conveniently believed that U-APSD mechanism defined in
802.11e is best exploited in symmetrical traffic scenarios, such
as VoIP, as the U-APSD triggering mechanism requires uplink
data traffic at well defined time intervals. In most cases, even
if the device supports the technology, the U-APSD feature is
disabled by default and even professionals on forums advise to
do so. This is due to the fact that even if the feature is enabled
and configured properly, the user can experience problems
with the wireless connection (i.e. in case of asynchronous
traffic the lack of trigger packets sent uplink to the access point
blocks the traffic flow) and there are hardly any applications
supporting the technology. We aim to fill the gap by proposing
a general application framework that allows mobile devices to
employ the foregoing power saving mechanism.
In this paper we investigate the possibilities to reduce
the energy consumption of wireless mobile devices in case
of streaming video content over Wireless LAN (WLAN)
networks. The scope of our research extends to find the tradeoff between the quality of the video perceived by the end user
and the energy efficiency of the wireless network adapter in
the mobile device. As our main contribution we propose a
general framework that allows video applications to directly
optimize the power management in wireless network adapters
by adaptively adjusting the sleep cycles of the network adapter
based on the estimated measure of the video quality of user
experience (QoE). The proposed framework is the combination
of a simple QoE estimation algorithm and a packet sending
mechanism, which can be widely implemented in various
video player applications in the form of a network capable
video player plugin. We found that it is an interesting challenge to be able to determine, which devices implement and
support the required power saving protocols, let alone their
precise configuration. As an important part of our paper the
building and configuration steps of a working testbed system
is carefully described, which demonstrates the availability and
efficiency of the power saving technology at hand.
There are numerous related proposals which optimize energy efficiency in mobile devices by using the sleep scheduling
mechanisms in wireless network adapters [2], [3]. Fewer
papers are concerned with energy optimization while streaming video content to wireless mobile devices. Lee et al. [4]
propose a client-server architecture, that reduces the unnecessary transmission of bitstreams by using Scalable Video
Coding on the server side and manages the power states of
the network adapter based on the video buffer occupancy of
the client. Mur et al. [5] propose an architectural change in
978-1-4577-0638-7 /11/$26.00 ©2011 IEEE
the power saving protocol by designing an adaptive power
saving mechanism that provides QoS to mobile devices using
information available only at the MAC layer. Adams et al. [6]
also propose an adaptive algorithm that defines different power
saving levels for the wireless adapter, and switches between
the states according to the power level of the battery of the
device utilizing the Legacy Power Save protocol. Although all
the proposals above achieve some boost in energy efficiency
during video streaming, none of them includes a user level
quality of experience feedback regarding the video content
played on the mobile device and most of them requires some
level of architectural change.
The rest of the paper is structured as follows. Section II
gives a brief overview of the power saving technologies in
wireless networks looked at from our applications perspective.
In Section III the proposed framework for power saving based
on user level quality of experience is presented. We carefully
describe in Section IV the building and configuration of our
testbed system. Section V contains a preliminary evaluation
of the proposed framework, which is based upon specific
measurements conducted on the testbed.
Power saving technologies in wireless networks describe
and implement different mechanisms for devices to efficiently
reduce their power consumption. There are many ways for
a device to achieve energy efficiency during wireless communication, like adaptively adjusting the transmitter power of
the wireless adapter [7], dynamically switching between single
and multiple antenna use [8], and also by scheduling the sleep
cycles of mobile devices [9].
In the case of our wireless video streaming scenario the most
suitable method is the sleep scheduling based power saving
technologies, since it is possible to download chunks of the
video stream in advance if bandwidth allows, and while video
is played from the buffer of the device, its wireless controller
can be put to low energy consuming sleep mode.
A. Power management in IEEE 802.11
Different power saving mechanisms have been built into
the IEEE 802.11 standard [10] over time and most of them are
types of sleep scheduling of the wireless network adapter. The
basic principle behind all sleep scheduling mechanisms is that
lots of energy is wasted listening on the wireless channel while
there is nothing to receive or no data is being send. The sleep
scheduler puts the wireless network adapter from time to time
into low powered doze state, when there is no traffic between
the access point (AP) and the mobile station. In this state the
device is not able to transmit or receive packets. Also, the
scheduler periodically wakes up the network adapter, so it can
check if there is traffic intended for the mobile device. The AP
keeps track of the power state of the stations by investigating
the Power Management bits within the Frame Control field of
the received frames and buffers packets destined for a mobile
station while it is in doze state.
The most basic power management scheme is available from
the original version of the IEEE 802.11 standard, which is now
often referred to as Legacy Power Save. Stations in legacy
power saving mode wake up for every beacon frame that is
sent by the AP, and learn about buffered packets addressed to
them by examining the Traffic Indication Map (TIM) field of
the beacon frame. The stations can request the buffered packets
one at a time by sending PS-Poll packets to the AP as long as
there are remaining buffered data indicated within the More
Data bit in the frame control field. Note that this mechanism
introduces a delay of downlink packets in the order of the
beacon interval (its typical value is 100 ms) which is critical
in most real-time applications and also the mechanism is very
inefficient because the station needs to send one PS-Poll packet
for every buffered packet [11].
B. U-APSD Protocol
The IEEE 802.11e extension defines an enhanced power
saving mechanism called Unscheduled Automatic Power Save
Delivery (U-APSD) and this amendment has already been
incorporated into IEEE 802.11-2007. The main idea behind
the U-APSD design is that the power saving stations can
indicate the instants, when they are awake, with data packets
sent uplink to the AP. When the AP receives a trigger packet
from a station, it responses with the packets that were buffered
while the station was in doze state. A trigger packet sent by a
station instantiates an unscheduled Service Period (SP). During
this SP the AP sends buffered packets to the station and marks
the last buffered packet with a special End Of Service Period
(EOSP) bit within the QoS Control field of the QoS frame1 .
The station is awake only for the duration of the SP then it goes
back to doze state. The maximum length of an SP can be set
by the station at association time with the Max SP Length bits.
If there are more buffered data even after the SP has ended,
this will be indicated by the AP in the More Data bit of the last
packet of the SP. In this case the station triggers new SPs with
QoS Null trigger packets until all buffered data is received. For
a more detailed overview of the U-APSD protocol the reader
is referred to [9].
The 802.11e standard defines four different access categories (AC) which correspond to four different transmit
queues, which allow devices to prioritize their traffic over
the wireless channel. The four ACs, ordered by descending
priority, are defined for voice, video, best effort and background traffic. For a packet to go to a specific transmit queue,
the DSCP2 field of the IP packet has to be marked with a
value reflecting the respective AC. The corresponding DSCP
value depends on the actual implementation of the U-APSD
protocol, and may vary among devices. Also, trigger packets
with different DSCP values will trigger different corresponding
AC queues.
1 In a WMM WLAN session QoS MAC frames are sent instead of normal
MAC frames. This type of frames contains an additional 2 byte QoS Control
field [12].
2 Differentiated Services Code Point
The U-APSD power saving protocol is defined in the IEEE
802.11 standard but it is only optional for manufacturers to
implement it. The market demand for Voice over IP capable
WLAN mobile devices drove the Wi-Fi Alliance to issue a WiR
Fi Multimedia Power Save certificate system. The WMM
Power Save Certificate [13] is based on a subset of the
IEEE 802.11e standard. The Alliance certificates products with
WMM-PS that support the U-APSD power save protocol.
It is noteworthy that the new IEEE 802.11n standard
contains an enhancement to the U-APSD protocol called
Unscheduled Power Save Multi-Poll (U-PSMP), which add
some minor modifications to the power management protocol
but doesn’t alter the U-APSD mechanism presented here.
We propose a framework which is able to estimate the
quality of video content in mobile devices, and adaptively
cycles the sleep modes of the wireless adapter based on the
video quality of experience (QoE). IEEE 802.11e and WMM
Power Save technologies permit manufacturers and software
developers to incorporate a power saving mechanism into the
mobile devices’ energy management scheme through software
and driver components specified according to standards. Our
main contribution is a software framework which extends the
WMM Power Save technology to streaming video applications. During video transmission a QoE estimation algorithm is
run on the video content in the mobile device, and the U-APSD
trigger packet sending interval is based upon the estimated
QoE measure.
A. QoE estimation
The estimation of the user experience, which is a measure
of the quality of the content directly perceived by users,
gives an important guideline to enhance standard video content
distribution techniques with network traffic optimization and
hence energy saving features. The system is based on the so
called no-reference image quality analysis. The no-reference
method estimates user experience according to the displayed
image properties and is very flexible and efficient since it
doesn’t need any reference video signal to compare the output
with. The idea of the estimation architecture is to search for
specific distortions like blockiness, blurring, etc. in the output
video. Typical distortions caused by network transmission
errors and delays can be effectively detected by image processing methods. There are many such adequate QoE estimation
algorithms, and we analyzed the method of T. Vlachos [14]
and the BABU method [15]. Both of the algorithms calculate
the QoE measure based on the blockiness of the decoded
video image, which is the most typical distortion in packet
switched networks. The Vlachos method requires slightly
more computational power and produces more effective QoE
estimates, while the BABU method places itself into a more
energy efficient but slightly less reliable part of the trade-off
space. Our analysis shows that both of them are suitable for
video QoE estimation in today’s mobile device crowd. The
measurements with the presented algorithms may be achieved
in real time, which enables the video source to tune the UAPSD triggering parameters accordingly.
B. U-APSD triggering based on video QoE
Our mechanism adjusts the U-APSD power save protocol
triggering interval based on the above mentioned estimated
video QoE measure. The schematic view of our proposed
framework can be seen in Figure 1. The mechanism gets input
in the form of decoded video frames. The QoE estimation
algorithm is then run on the decoded video data, and it determines the quality of the video by processing the video frame
image. The different algorithms provide the QoE measures
in different domains which has to be scaled accordingly to
provide input for the trigger interval setting mechanism with
fine enough resolution. This measure can also be scaled to
the 5-grade Mean Opinion Score (MOS) system. Depending
on the QoE measure the U-APSD triggering component will
send out trigger packets on the WLAN channel. When the QoE
estimate show bad video quality, the triggering component
raises the frequency of the trigger packets. On the other hand,
if the video quality is flawless, then it can lower the trigger
frequency, thus enabling the wireless adapter of the mobile
device to be in sleep mode for longer intervals.
The trigger packets has to be a regular IP packet with DSCP
field set according to the 802.11e QoS video AC. This ensures
that in response to this packet the AP sends all packets to the
mobile station, which were previously buffered at the AP while
the network adapter of the mobile station was in sleep mode.
When the transmission of the buffered packets is over, the AP
marks the last packet with the More Data bit set to 0, and the
network adapter goes back to sleep mode.
Fig. 1: Schematic diagram of the U-APSD triggering mechanism based on estimated video QoE.
We emphasize that for our framework to be fully functional
the video stream has to be configured also on the server side
according to the WMM Power Save standard. This means that
the IP packets of the video stream only have to be flagged
with the appropriate DSCP value, so that the WLAN AP
directs them into the video AC queue, and buffers them as long
as the mobile device is in sleep mode, and no other actions
are required from the server (encoding, adaptivity etc.). The
method is transparent from the server side which eases the
introducion of the system independently from video content
providers. Note that the right DSCP value may depend on the
implementation of the AP firmware.
Due to the architecture of our framework some limitations
of the system arise. Besides the wireless channel, transmission
error or delay in other segments of the network can also
cause video quality degradation, which cannot be identified by
our mechanism. In response, our mechanism tries to improve
the video quality by setting higher value for the trigger
packet sending frequency, which however is not going to help
improving the video quality and will render the device to
wake up more often, hence degrading its energy efficiency.
Although providing high quality video content to end users
can be considered far gone, the resolution of this issue is out
of our scope and we look at it as future improvement to our
Setting the desired QoE level, the framework adaptively
adjust the sleep scheduling intervals so that the desired QoE
level is realized. Wireless channels with different throughput
parameters (signal strength, saturation/congestion of the wireless channel), different energy efficiency while achieving the
same QoE level.
We built a testbed which contains key resources of our
proposed framework. It is a simplified setting compared to the
framework, which is chosen so that the power saving efficiency
of our framework could be finely tuned and measured.
Wireless Access Point: Our testbed contains two HTC
Dream mobile devices with Android 1.6 operating system, a
Linksys WRT54G WLAN WiFi AP running DD-WRT v24
SP1 [16]. These devices are chosen since they are all WMM
Power Save certified wireless LAN devices. The configuration
in the Wireless/Advanced settings section of the AP is set so
that it enables WMM support, and the device indicates this
by carrying the Vendor Specific WME3 Information Element
(IE) in its Beacon, Probe Response and Association Response
management frames with WME Parameter Element set to 128
[12]. In Figure 2 there is a Wireshark [17] screenshot showing
the WME IE within a Beacon frame and the meaning of the
corresponding bits. This vendor specific tag sent by the AP
lets mobile stations in the BSS to discover at association time
whether or not they can use the power saving abilities defined
in WMM Power Save. It contains every important information
regarding the WMM Power Save mechanism broken down for
the four different AC types. Such an association between an
AP and station is called a WMM Association.
Mobile devices: One of the mobile stations is also configured through its WLAN adapter driver configuration file
[18] so that it enables the same power saving technology. In
most cases, even if the device supports the technology, WMM
Power Save feature is disabled by default. This can be due
is a former abbreviation for WMM, they are used interchangeably
Fig. 2: Wireshark screenshot of the WME Information Element
within a Beacon frame sent by the AP and the meaning of
parameter bits.
to the fact that if the feature is enabled but not configured
correctly, the user can experience problems with the wireless
connection (i.e the lack of trigger packets sent uplink can block
all traffic to the mobile device). It is also noteworthy that we
might have to dig deep into the device’s configuration files to
enable the feature. For our framework to be functional we set
the device to use U-APSD protocol for its video and voice
Access Categories (AC VI and AC VO, respectively). Now,
when the mobile device learns from the WLAN management
frames that the AP supports WMM Power Save, it sends back
the appropriate IE in its Association Request management
frame, where it indicates for which ACs it enables U-APSD
(see Figure 3). If the Association Request frame contained the
WME IE, then the AP replies with an Association Response
which carries also a WME Information Element.
Fig. 3: Wireshark screenshot of the WME Information Element
within an Association Request frame sent by the station and
the meaning of parameter bits.
We implemented an Android [19] application for the mobile
device so that it can send trigger packets at a given interval.
The trigger frame is a simple UDP packet with the 1 byte IP
TOS field set to 160 in decimal representation. The exact value
is defined in the configuration file of the wireless adapter as the
DSCP code corresponding to the video AC [18]. The packet
is then sent through the video AC from the station to the AP,
which in turn will trigger the buffered packets belonging to
the video AC queue to be flushed to the mobile station. This
way the sleep time interval of the station can be quantitatively
adjusted, and also the frequency of the video data transfer can
be easily set, which after all effects the quality of the video.
Video Server: On the server side QuickTime Streaming
Server [20] is used running over Mac OS X from which RTP
streaming video data is sent to the mobile devices through the
AP. The video stream is configured to be easily playable by
the mobile devices (25 frames per second, 200 kbps data rate,
8 key frame per second). A network proxy running iptables
is also set up in the video transmission flow, which tags the
Fig. 4: Testbed configuration
IP TOS field of each video RTP packets to 0xb8 (184d ). This
also puts the video traffic into the video AC on the WLAN
AP which will buffer these packet until the mobile device is
in sleep mode.
Power Usage Monitoring: Our trigger packet sending application is implemented to include the mobile device’s current
power consumption measure in mA in the trigger packet,
which is in turn sent to a monitoring Linux server connected
to one of the LAN ports of the AP, where the measurements
data are logged and evaluated in real time.
We checked the wireless traffic during video streaming
with Wireshark to see if WMM Power Save was properly
configured. According to the 802.11 level network traffic we
could verify that the AP started to send the buffered data
packets to the station only when the device sent a special
trigger packet to the AP. In response, the AP sent all buffered
packets4 to the station.
Fig. 5: Power consumption in mobile devices configured with
U-APSD vs. Legacy Power Save Mode.
show that the device power consumption is higher in case of
higher video bit rates for identical MOS values and if we
enable a slightly degraded video quality the device can save
some 5-7% additional energy.
For the evaluation of the proposed framework we conducted
measurements on our testbed. The measurements aim at determining the power saving capability in the mobile devices
introduced by our framework. Also by measuring the different power consumption levels which correspond to different
subjective QoE measures, we can quantify the available power
saving possibility that can be achieved by defining the desired
QoE level in the mobile device.
As a reference measurement we compare the U-APSD
power saving mechanism with Legacy Power Save Mode.
Figure 5 shows the power consumption of the two identical
mobile devices (brightness of the displays set to the same
setting, etc.), one with WMM Power Save enabled and the
other configured with Legacy PS. The device with U-APSD
enabled can save up to 20-30% compared to the other device.
We measured how the video quality of experience grades
affect the power consumption of the device. We manually
adjusted the trigger interval to achieve the same subjective
Mean Opinion Score value with different video bit rates. The
relationship between QoE measures of the video and power
consumption of the device can be seen in Figure 6. The results
4 Unfortunately the Max Service Period length could not be set in the driver
configuration of the mobile station and the default setting was 2. This rendered
the device to send a new trigger frame to the AP after 2 received data packets
until the More Data bit of the received packet indicated that there are still
buffered packets left at the AP. If the Max SP length is set so the device gets
all buffered packets in reply to a single trigger packet, there can be some
further optimization in the energy efficiency.
Fig. 6: MOS - Mean Opinion Score vs. Power consumption
for different bit rate videos.
As a reference the corresponding trigger values for different
bit rates and different MOS values can be seen in Table I.
The varying trigger intervals imply that there is some space
for the proposed framework to level out the differences in the
wireless transmission channel and the varying bit rates of the
streaming video by constantly optimizing the energy efficiency
of the wireless network adapter.
TABLE I: Trigger interval values that correspond to the MOS
grades while streaming videos with different bit rates.
Video bitrate
100 kbit/s
200 kbit/s
500 kbit/s
0-10 ms
0-5 ms
0-5 ms
10-1000 ms
5-100 ms
5-100 ms
1000-3000 ms
100-1000 ms
100-1000 ms
In this paper we presented a framework that optimizes the
energy efficiency of wireless mobile devices during video
streaming. The proposed mechanism estimates the quality of
experience of the video image and adaptively adjusts the sleep
cycles of the power saving capable network adapters. The
QoE estimation is done by specific algorithms and their output
measure controls the triggering interval of the underlying UAPSD power saving protocol.
We evaluated our framework by conducting measurements
Power Save certified WLAN
on a testbed containing WMM
devices. Our measurements support the claim that by utilizing
the technology up to 30% of the total power consumption can
be saved, and we show that this is achievable even during
asynchronous wireless traffic scenarios.
Our general conclusion is that significant amount of power
saving can be easily achievable with the right software
components by utilizing readily available power management
technologies. This paper presents an out-of-the-box power
management framework that provides some key insights for
the development and implementation of energy efficient video
streaming applications.
The research leading to these results has received funding
from the ARTEMIS Joint Undertaking under grant agreement
n◦ 100029 and from the Hungarian National Office for Research and Technology (NKTH).
[1] “T-Mobile [email protected] review at CNET,” CNET. [Online]. Available: http://reviews.cnet.com/cell-phone-and-smart/t-mobilehotspot-home/1707-6448 7-32486222.html
[2] F. R. Dogar, P. Steenkiste, and K. Papagiannaki, “Catnap: exploiting
high bandwidth wireless interfaces to save energy for mobile devices,”
in Proceedings of the 8th international conference on Mobile systems,
applications, and services, 2010, pp. 107–122.
[3] J. Liu and L. Zhong, “Micro power management of active 802.11
interfaces,” in Proceeding of the 6th international conference on Mobile
systems, applications, and services. ACM, 2008, pp. 146–159.
[4] S. Lee, J. Koo, and K. Chung, “Content-aware rate control to improve
the energy efficiency in mobile IPTV services,” IJCSNS, vol. 10, no. 5,
p. 52, 2010.
[5] D. C. Mur, X. Perez-Costa, and S. S. Ribes, “An adaptive solution for
wireless LAN distributed power saving modes,” Computer Networks,
vol. 53, no. 18, pp. 3011–3030, 2009.
[6] J. Adams and G. Muntean, “Power save adaptation algorithm
for multimedia streaming to mobile devices,” in 2007 IEEE
International Conference on Portable Information Devices,
Orlando, FL, USA, 2007, pp. 1–5. [Online]. Available:
[7] A. Sheth and R. Han, “Adaptive power control and selective radio
activation for low-power infrastructure-mode 802.11 lans,” Distributed
Computing Systems Workshops, International Conference on, p. 812,
[8] M. Z. Siam, M. Krunz, A. Muqattash, and S. Cui, “Adaptive multiantenna power control in wireless networks,” in Proceedings of the
2006 international conference on Wireless communications and mobile
computing, ser. IWCMC ’06. New York, NY, USA: ACM, 2006, pp.
[9] D. C. Mur and X. Perez-Costa, “IEEE 802.11E QoS and power saving
features overview and analysis of combined performance,” IEEE Wireless Communications, vol. 17, no. 4, pp. 88–96, 2010.
[10] IEEE 802.11 WG, “IEEE Standard for Local and Metropolitan Networks
Part 11: Wireless LAN Medium Access Control (MAC) and Physical
Layer (PHY) Specifications,” IEEE 802.11-2007.
[11] X. Perez-Costa, D. Camps-Mur, and T. Sashihara, “Analysis of the
integration of IEEE 802.11 e capabilities in battery limited mobile
devices,” IEEE Wireless Communications, vol. 12, no. 6, pp. 26–32,
[12] D. R. L. Praphul Chandra, Wi-Fi telephony: challenges and solutions
for voice over WLANs. Newnes; 1st edition, 2006.
[13] Wi-Fi Alliance, “Wi-Fi CERTIFIED for WMM Power Save,” 2005.
[Online]. Available: http://www.wi-fi.org/white papers/whitepaper120505-wmmpowersave/
[14] T. Vlachos, “Detection of blocking artifacts in compressed video,”
Electronics Letters, vol. 36, no. 13, pp. 1106–1108, 2002.
[15] R. Babu, A. Bopardikar, A. Perkis, and O. Hillestad, “No-Reference
metrics for video streaming applications,” in International Workshop on
Packet Video, 2004.
[16] DD-WRT website. [Online]. Available: http://www.dd-wrt.com/
[17] “Wireshark packet analyzer webpage,” Wireshark. [Online]. Available:
[18] “Wl1271 ini file description and parameters user’s guide,” Texas Instruments. [Online]. Available: http://focus.ti.com/lit/ug/sprugt8/sprugt8.pdf
[19] “Android
[20] “Quicktime streaming server,” Apple. [Online]. Available: