Battery Energy Consumption Footprint of Embedded Multimedia Systems
Jiucai Zhang, Song Ci, and Xueyi Wang Department of Computer and Electronics Engineering University of Nebraska-Lincoln NE 68182, USA Vczhang,email@example.com!.edu, [email protected]
Abstract-The rapid demand of mobile multimedia causes a grant challenge on energy and environment. However, little research has been done on the carbon footprint of embedded multimedia system. To fill this gap, this paper presents a comprehensive architecture to measure the carbon footprint of
of raw video data, it has to be effectively compressed. Otherwise, tremendous energy will be needed for video transmission. Therefore, many battery powered systems adopted computation intensive and energy consuming video
systems. The measurement results of H.264 under various
compression algorithms and international standards such as H.263, MPEG-4, and H.264  for efficient video compres
system operation modes provide an insightful guideline to
sion. Typically, these algorithms consume about
H.264 codec, which has been widely used in mobile multimedia
design and optimize system parameters to minimize the carbon footprint of various embedded multimedia systems. Keywords·carbon footprint; H.264; battery; multimedia;
60% '" 80%
of the total energy of battery powered mobile systems . Furthermore, battery is not an ideal energy source. The nonlinear battery effects may cause carbon footprint consumption of battery to vary with discharging current of battery.
The rapid development of semiconductor, computer, and,
Although battery powered embedded multimedia devices
communication technologies enables mobile devices such
with H.264 codec have been widely used, little research
as Smartphones, PDAs, laptops having computation and
has been done on the analysis of the carbon footprint of
communication capability to support multimedia applica
H.264 codec with consideration of nonlinear battery effects.
tions such as multimedia streaming, VoIP, and mobile TV.
To bridge this gap, in this paper, we propose an architecture
Recently, demand for these multimedia devices has soared
to track carbon footprint in H.264 codec under various oper
worldwide, which causes fast growing energy cost. For in
ation modes. The architecture determines the computational
stance, according to a recent survey, worldwide smartphone
complexity profile of H.264 codec based on the number
sales have been increasing by 29 percent per year to 180
of basic computational operations required by a codec to
million units in 2009 . Using these smartphones for a
perform the key subfunctions and the frequency of use of
year on average consumes 759.78 billion joules of energy. To
each of the required subfunctions. Then, carbon footprint
put this in carbon footprint perspective, these smartphones
model of battery can convert computational complexity to
20.16Gt of CO2
a year. Therefore, energy consumption
carbon footprint of the H.264 codec. The carbon footprint
of these devices has significant energy and environmental
measurement provides a way to understand its absolute
energy-efficiency of H.264 codec. The quantified energy
On the front of global warming issues and energy security,
efficiency can be used to reduce overall energy consumption
many researchers are engaged in identifying strategies to reduce their climate impacts. So does the field of battery
by knowing the theoretical bound of energy efficiency.
powered embedded multimedia systems. In such systems,
The measurement can also be used as an objective tool to compare the energy-efficiency performance of various
live videos are captured and compressed on battery-powered
embedded devices, and then the compressed video is trans
The paper is organized as follows. Section II analyzes the
mitted to remote users through wired/wireless channels.
carbon footprint of mobile multimedia system. Section III
To enable low carbon footprint, it is expected that video
proposes the architecture for carbon footprint measurement
communications over battery powered mobile multimedia
of H.264 codec. Section IV measures carbon footprint of
devices can respond quickly to the challenge of energy
H.264 codec under various operation configurations. The paper is concluded in Section V.
consumption. Understanding energy consumption in mobile multimedia devices is the key for efficient energy management of the next-generation mobile multimedia. In general, energy con sumption of mobile multimedia is mainly on computation, communication, and 110 devices . Due to large volume 978·1·4244· 7614-5/10/$26.00 ©2010 IEEE
II. T HE CARBON FOOTPRINT OF MOBILE MULTIMEDIA SYSTEM
A general architecture of mobile multimedia system is shown in Figure 1. Raw video data are compressed by video
o (I) <
-:::::0 3 (I) o �. -(") (I) (I)
Architecture of mobile multimedia systems
0.3 0.4 0.5 0.6 0.7 0.8 0.9 Current (normorlized to 1 C 41.3mA) =
Battery carbon footprint vs. discharge current rate.
codec such as H.264, H.263, and MPEG-4, and then the compressed data packets are passed through the network protocol stack and are transmitted over a communication
codec. The number of function calls is measured by using
channel. To estimate the overall carbon footprint of a mobile
program profiling technology under different configuration
multimedia device, it is important to understand that there
of the codec. By mapping these computational requirements
are at least three major factors in carbon footprint: I) com
to a given processing capability of hardware platform, an
putational complexity of video codec including algorithm complexity, bit rate, and distortion (or perception quality)
estimate of the codec's carbon footprint can be determined for the platform. The methodology of energy profiling is
, 2) communications and 3) nonlinear battery effects. The
shown in Figure 2. Program profiling involves running a
computational complexity can be measured by the approx
version of a H.264 codec program in which instrumentation
imate number of operations required to execute the codec
code has been incorporated to determine how much time
under a specific configuration. The optimal configuration
different components of the H.264 codec program will take.
of video codec may have significantly lower computational complexity than a less optimal one for a given hardware
First, it determines how much CPU time was spent for each
The program profiling generates two forms of information.
platform, despite the fact that the two configurations may
of the function in the H.264 codec. Second, it keeps tracking
produce identical perceptual quality (or video distortion).
of how many times each function gets called, categorized by
A higher compression ratio will result in a lower bit rate,
which function performs the call. Both forms of information
but resulting in higher computational complexity and bigger
can be useful. The timing information gives a sense of the
carbon footprint, and vice versa. Furthermore, battery is not
relative importance of the different functions in determining
an ideal energy source, and the energy efficiency of battery
the overall run time. The calling information allows us to understand the dynamic behavior of the H.264 codec.
varies with battery discharge current rates. Battery tends to lead to bigger carbon footprint at a high discharge current rate. Therefore, achieving a smaller carbon footprint needs a systematic consideration of both video source adaptation and network transmission adaptation under a given battery characteristics. A. Carbon Footprint Profiles of H.264 Codec
B. Nonlinear Battery Carbon Footprint
Mobile multimedia devices are usually powered by Nickel-cadmium and Lithium-ion battery. Due to nonlinear battery effects, such as current effect, recovery effect, aging effects, and self-discharging, no battery chemistry can obtain
columbic efficiency . Generally, the efficiency will
To derive an accurate estimate of the carbon footprint
drop as the current is increased because the rate dependent
of H.264 codec , we should first discuss the complex
component of electrode over-potentials and the formation of
ity of H.264 codec. H.264 codec contains a number of
concentration gradients will consume energy . In order to
optional coding features that may have significant effects
effectively design and utilize the battery for dynamic power
on carbon footprint of codec. In this paper, we take a
management, deep understanding of how carbon footprint
systematic approach to quantifying the carbon footprint
varies with current is highly desirable. In this paper, we have
of a H.264/AV C baseline codec. The objective here is
set up a testbed to study this relationship. The experiment
to determine the number of basic operations required by
results of a Bellcore PLION cell  by using Arbin Battery Testing Instrument BT2000 are presented in Figure 3. In this
the codec to perform each of the key functions of the
Others Program profile Flat profile: Each
sample counts as
74.87 22.59 1.06 0.65
326.71 425.29 429.90 432.72
326.71 3303048305 98.58 2415105 4.61 102372150 2.82 41055625
p.CO O. 00 0.00 O. 00
0.00 O. 00 0.00 0.00
computeSAO FullPelBlockHot.ionsearch HadamsrdSA04x4 computeSATD
Times that the function
Times that the function
Energy profiling of encoder
Energy profiling architecture
figure, each point is obtained in the following method. First,
a low-resistance shunt and measured the voltage across the
we charge a fresh battery with constant current and constant
shunt using a Arbin Battery Testing Instrument BT2000 .
voltage to full, which means that the battery is charged to
At the same time, battery voltage can be directly measured
full with same consumed energy and carbon footprint. After
by one of the BT2000 channels. The measured current and
minutes, this battery is discharged
voltage can be used to calculate consumed power, energy,
from the full capacity to exhaustion at discharge current rate
and capacity of the H.264. In this experiment, a fresh battery
battery is rested for
X.C (Here, lC 25°C. The carbon
All discharges took place at
is firstly charged to the full capacity via the constant current
footprint is defined as the ratio between
and constant voltage (CCCV ). Then, the battery is rested
total carbon footprint for charging and consumed power for
for minutes. Finally, the battery is used to power Imote 2.
discharging . The experiment results are shown in Figure 3.
The consumed energy in amperes hours can be translated
From this figure, we can conclude that the carbon footprint
into carbon footprint using the carbon footprint of battery
of a Bellcore PLION cell increases with increment of the
as shown in Figure 3. For an embedded multimedia system,
discharge currents due to nonlinear battery current effects.
we usually use H.264 codec to compress videos
This can be observed by the fact that the enhancement of
the carbon footprint for discharging a full charged battery
then partition the raw video data into
will change from about 0.59gjWh at the discharge current rate of O.IC to about 0.674gjWh at the discharge current of
which provides a useful guideline for carbon
footprint-aware green computing. III. T HE ARCHITECTURE FOR CARBON FOOTPRINT MEASUREMENT
To track carbon footprint of the H.264 codec, we have
(176 x 144
frames per second) captured from a camera. We
the aforementioned framework, we can obtain the carbon footprint of H.264 codec at various operation configurations. Then, we compute the number of cycles required to execute a particular subfunction on a chosen hardware platform. This intermediate result may be derived by mapping the subfunction operations on the hardware, applying single instruction multiple data parallelism and distributing the required operation as efficiently as possible among the exe
implemented H.264 codec on an embedded platform Imote 2
cution subunits on the hardware. Furthermore, the number of
with Linux 2.6.29 Kernel and an Xscale PXA processor ,
CPU cycles is derived by multiplying the intermediate result
which is powered by a 3.7V, 2600mAh lithium-ion battery
by the frequency with which the subfunction was called.
HE18650. The architecture of carbon footprint measurement
Based on the carbon footprint of the battery and complexity
is shown in Figure 4. Based on this setting, the carbon
profile of H.264 codec, carbon footprint of H.264 Codec is
footprint of the H.264 codec is measured in the following procedures. First, we redirected the device current through
Profile of power & energy
I ,,,)·-,oo,,,,,,, _,.oo"n".
... 0,00621731' .. 0 013 149/ + 0.62514
- ...." . ---- CiJTent °0
The system architecture of carbon footprint measurement.
IV. CARBON FOOTPRINT MEASUREMENT OF EMBEDDED MULTIMEDIA SYSTEMS
0N S2. C
.� c 0 .0
Due to various configurations of H.264 codec, it is very
. , . . .. .. . � ... ' ... . . . . . ..
difficult to get the carbon footprint of H.264 under all possible modes. The main carbon footprint is consumed by
... -. . ... ...
the module of motion estimation. The two main parameters
.... , .
that affect computational complexity of motion estimation are maximum search range and number of inter reference frames . Therefore, in this section, we will focus on the
carbon footprint measurement of H.264 under various search
ranges and different numbers of inter reference frames.
Figure S. ranges
By analyzing the run-time profiling data of codec subfunc tions generated by varying the number of maximum search range, we can obtain the carbon footprint of H.264 codec
Max search range
as a function of maximum search range and quantization
Carbon footprint of H.264 at various values of maximum search
step size in Figure 5. It is worthwhile to emphasize that maximum search range has critical effect on the carbon footprint of H.264 codec. Reducing maximum search range can dramatically reduce the carbon footprint of the codec. The carbon footprint measurement of H.264 under various numbers of reference frames and quantification steps is shown in Figure 6. Carbon footprint of H.264 varies with
ON S2. C
number of inter reference frames and quantization step size.
In the application scenarios where carbon is very critical,
reducing the number of reference number of video can lower the carbon footprint. On the other hand, the closer the reference frame is, the higher the frame correlation will be, except for occluded, and uncovered objects. Therefore,
.8 0.64 Co U 0.62 60
we should precede the blocking matching process from the nearest reference frame to the farthest reference frame. Thus,
a lot of carbon footprint is saved. Both experimental results provide a quantitative solution for carbon-footprint aware
Inter reference number
computing. V. CONCLUSIONS
Figure 6. fames
Carbon footprint of H.264 at various numbers of inter reference
In this paper, we have developed a comprehensive frame work to measure carbon footprint of H.264 codec with
consideration of nonlinear battery effects. The measurement of carbon footprint of H.264 codec will benefit us to enhance energy efficiency in multimedia computing and performance optimization. The measurement results will enable designers to reduce energy consumption in data compression. The carbon footprint results can be adopted by various mobile multimedia devices, providing a design guideline for carbon footprint-aware computing. ACKNOWLEDGMENT
This work was partially supported by NSF under Grant No. 0708460 and No. 0801736. REFERENCES  D. Y ilmaz and M. Yildiz,"Analysis of the mobile phone effect on the heart rate variability by the calculation of correlation dimension," in 14th National Biomedical Engineering Meet ing, may 2009,pp. 1 -4.  C.-J. Lian, S.-Y. Chien, C. ping Lin, P.-c. Tseng, and L.
G. Chen, "Power-Aware Multimedia: Concepts and Design Perspectives," pp. 26-34, 2007.  M. Horowitz, A. Joch, F.
Kossentini, and A. Hallapuro,
"H.264/avc baseline profile decoder complexity analysis," IEEE Transactions on Circuits and Systems for Video Tech nology, vol. 13,no. 7,pp. 707-716,2003.  Y.-W. Huang, c.-y. Chen, C.-H. Tsai, C.-F. Shen, and L.
G. Chen, "Survey on block matching motion estimation algorithms and architectures with new results," Journal of VLSI Signal Processing Systems archive, vol. 42, no. 3, pp. 297-320,2006.  D. Marpe,T. Wiegand,and G. J. Sullivan,"The h.264/mpeg4 advanced video coding standard and its applications," IEEE Image Communications Magazine, vol. 44, no. 8, pp. 134143,2006.  R. Rao,S. V rudhula, and D. N.Rakhmatov, "Battery Mod eling for Energy-Aware System Design," IEEE Computer Society, vol. 36,pp. 77-87,Dec. 2003.  P. Rong and M. Pedram," An Analytical Model for Predicting the Remaining Battery Capacity of Lithium-Ion Batteries," IEEE Journal Very Large Scale Integrated, vol. 14,pp. 441451,May 2006.  V. Iverson, J. McVeigh, and B. Reese, " Real-time h.24-avc codec on intel architectures," in International Conference on Image Processing, vol. 2,oct. 2004,pp. 757 - 760 Vol.2.  J. Zhang,S. Ci, H. Sharif, and M. Alahmad, "An enhanced circuit-based model for single-cell battery," in IEEE Con ference on the Applied Power Electronics Conference and Exposition, Plam Sprints,CA,Faburary 2010.  B. I. Richardson and Y. Zhao, "Efficient mode selection for h.264 complexity reduction in a bayesian framework," in IEEE International Conference on Image Processing, 2006, pp. 1-30.