2012 IEEE 75th Vehicular Technology Conference (VTC2012-Spring)

A Measurement Based Energy Model for

IEEE 802.16e Mobile WiMAX Devices
Bjoern Dusza, Christoph Ide and Christian Wietfeld
Communication Networks Institute
TU Dortmund University
44227 Dortmund, Germany
e-mail: {Bjoern.Dusza, Christoph.Ide, Christian.Wietfeld}@tu-dortmund.de

AbstractThe operational time of modern smart-phones with

one filling of the accumulator is one of the most important
performance parameter for the customers of new devices. This
is the reason why extensive research has been performed in the
fields of battery design and system optimization with regard on
energy consumption. Due to the fact that the radio part of a
smart-phone is one of the major energy consumers, a special focus
has to be set on different approaches and strategies how to reduce
the energy that has to be spent for the transmission of a certain
amount of data. However, before new energy saving protocols can
be efficiently developed one needs to have detailed knowledge on
the different factors which impact the energy efficiency.
In this paper, we present a measurement based energy model for
IEEE 802.16e conform Mobile WiMAX devices. For this purpose,
extensive measurements of the energy consumption of a Mobile
WiMAX USB Stick have been performed for different system
parameterizations, different data sizes and different application
data rates. From the results of the measurements analytic models
have been derived which allow for the calculation of the energy
that has to be spent per successfully submitted bit. These analytic
formulas can now be integrated in system level simulators for the
evaluation of the energy efficiency of newly designed protocols.

I. I NTRODUCTION
The operational time for which a mobile phone can be used
prior it has to be recharged is one of the most important
performance parameter for the consumers of new smart-phones
[1]. This leads to the major challenge that on the one hand new
smart-phones with fast Central Processing Units (CPU), bright
displays, lots of sensors and many different communication
interfaces need more and more energy and on the other hand
the battery technology is not developing as fast as it would be
necessary. This is the reason why energy efficiency is one of
the most important design targets if novel smart phones based
on Long Term Evolution (LTE) or Mobile WiMAX have to be
designed. However, before extensive simulations on the energy
consumption of new systems, protocols or algorithms can be
performed, detailed energy models are needed as a basis for the
performance evaluation. One of the most critical performance
indicators in this context is the energy that has to be spent for
the successful submission of 1 bit.
In this paper, we present a detailed measurement based energy model for IEEE 802.16e Mobile WiMAX. The model
describes the influence of the transmission power, the data
size (e.g. in case of file transfer) and the application data rate
(e.g. for streaming applications such as Voice over IP (VoIP))

iPerf Server

Gateway PC

Evaluation PC

Ethernet

USB
Probe
USB

Base Station Emulator

RF

DUT

iPerf Client

iPerf generated Uplink Traffic

Fig. 1.

Measurement Setup for Detailed Power Measurements

on the energy consumption per bit in closed formulas. For

the derivation of the model extensive laboratory measurements
have been performed based on a Mobile WiMAX base station
emulator and a commercially available Universal Serial Bus
(USB) enabled Device Under Test (DUT). For the measurement of the energy consumption, the current as well as the
voltage at the input of the USB stick were sampled during
an iPerf generated uplink data submission for different system
parameterizations (see Fig. 1). The models show that all of
the considered parameter have a major impact on the energy
consumption per bit and therefore should be incorporated by
future energy efficiency considerations. Exemplary concrete
results presented in this paper are therefore the minimum
amount of data that has to be collected for an energy efficient transmission as well as a proposal for optimized data
aggregation schemes in wireless sensor networks.
II. R ELATED W ORK
Modeling the energy consumption of modern smart-phones
is gaining importance during the last few years. This is due
to the fact that cellular phones are no longer only phones but
complete portable computer with lots of additional functions.
This comes along with the fact that the power consumption
is significantly increasing. In [2], an approach is presented
which allows for the automated construction of energy models
for Android based smart-phones. Instead of measuring the
actual current drain from the battery, the method described
here bases on the battery voltage sensor and knowledge about
the battery discharge voltage curve. For the determination of

2012 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media,
including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to
servers or lists, or reuse of any copyrighted component of this work in other works.
Link to published version: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6239855

TABLE I
M OBILE W I MAX S YSTEM PARAMETRIZATION
300 ms between CQI Reports

Parameter
Carrier Frequency [GHz]
Channel Bandwidth [MHz]
UL Tx-power per Subcarrier
Fast Fourier Transform (FFT) Size
UL Modulation Scheme
UL Coding Rate
Coding Type
Duplexing Scheme
Carrier Interleaving Scheme
MIMO Type
Map Repetition Factors
HARQ
Maximum Transfer Unit (MTU)

Value
2.5405 GHz
10
-6 dBm
1024
QPSK
1/2
Convolutional Turbo Code (CTC)
Time Division Duplex (TDD)
Full Usage of Subchannels (FUSC)
1 x 1 (SISO)
0 (No Repetition)
Disabled
1368 Bytes

the energy consumption of different components of the device,

the different parts are successively activated and deactivated
by a training program which allows for the monitoring of the
battery voltage change for each activation step. From that,
power coefficients for a generic energy model are derived.
Although the model covers the power consumption of the
overall device, the radio parts for WiFi and 3G are only
modeled by means of three states each (Idle, Forward Access
Channel (FACH) and Dedicated Transport Channel (DCH) for
3G). A comparable approach can also be found in [3] where
only two states (Idle and Active) for the energy consumption
of the 3G radio interface are assumed.
A much more detailed investigation of the energy consumption
of GSM, UMTS and WiFi has been performed in [4]. Here,
the impact of different locations and different times are taken
into consideration as well as user mobility. For the evaluation
of the energy consumption the measured value is divided into
a ramp-part, a data-part and a tail-part. The measurements
show that for short data bursts the setup of the data channel
as well as the termination causes significant costs in terms
of energy consumption. Based on this knowledge a new
protocol is proposed which optimizes the scheduling for a
minimum energy consumption. It is worth noting that in [4] the
assumption is used that the energy consumption is proportional
to the length of transmission and the transmit power level. The
results presented in this paper will show that this is not true
under any circumstances for Mobile WiMAX.
Very detailed investigations on the energy consumption of
WiFi systems have been performed in [5]. Here a measurement based approach was used to map the current power
consumption to corresponding protocol events. Beside this,
energy efficiency analyses have been performed in terms of
energy per submitted bit.
III. M EASUREMENT S ETUP
The measurement setup used for the determination of the
USB sticks energy consumption can be seen in Fig. 1. The
actual bidirectional Mobile WiMAX link between the Base
Station Emulator (BSE) and the DUT is realized by means of
an RF cable to avoid external influences on the radio link.

PCQI = 929 mW

Periodic Submission of
Channel Quality Reports

PIDLE = 880 mW

(a) Idle State with Channel Quality Transmission

5 ms Frame with PF RAME = 1.32 W
tUL = 1.28 ms

tDL = 3.72 ms

PDL = 0.9 W
PUL
= 2.52 W

(b) Power Distribution for WiMAX Frame (DL/UL = 35/12)

Fig. 2.

Mobile WiMAX Power Consumption Measurement

The BSE as the central component of the setup acts as a

base station and therefore creates a standard conform radio
cell towards the DUT. In the context of energy consumption
measurements the main role of the BSE is the forwarding of
the transmitted uplink data to the server as well as the submission of Medium Access Control (MAC) messages related
to power control in the downlink. These messages are needed
due to the fact that for the measurements a pure closed loop
power control is applied which means that the uplink transmit
power is solely set by the base station. This allows for full
control of the measurements. The actual parametrization of
the Mobile WiMAX link can be seen in Tab. I. The iPerf
server, which acts as counterpart for the client PC, is connected
to the BSE via Ethernet. Beside this, the BSE used for the
investigations makes use of an additional gateway PC for
end-to-end application testing. Some of the MAC mechanisms
needed for this kind of test are executed on this external PC
which is also connected to the BSE via Ethernet. The actual
power measurement is performed by a measurement probe that
is placed between the DUT and the client PC. Therefore, the
energy consumed by the DUT has to pass the probe where it is
measured in terms of electric current and voltage. From this,
the current power consumption is calculated at a frequency of
100 kHz and transmitted to an evaluation PC via USB. Here
the data is stored for a detailed evaluation such as for example
the determination of the consumed energy in a certain period
of time. From this, it is for example possible to very precisely
measure the energy that is consumed for the transmission of
single symbols, bursts and frames as it can be seen in Fig. 2.
IV. R ESULTS
In this section the results of the energy consumption measurements are described in detail. A general overview on

the course of the actual power over time can be seen from
Fig. 2. Fig. 2(a) illustrates the power consumption over time
for a Mobile WiMAX device that is connected to a base
station but does not transmit or receive any data. Therefore a
continuous average power consumption of PIDLE = 880 mW
is needed while every 300 ms a channel quality (CQI) report
is submitted in the uplink. For this submission period the
average power increases to PCQI = 929 mW . Fig. 2(b) shows
a zoom to the power measurements for one Time Division
Duplex (TDD) frame. One can clearly see the different parts
of the burst for the uplink and the downlink where the time
ratio tDL /tUL corresponds to the defined TDD downlink to
uplink ratio of 35/12. While the overall average power for
the submission of one burst is PF RAME = 1.32 W only
PDL = 0.9 W are consumed during the reception phase while
PUL = 2.52 W are needed for the submission of data in the
uplink (assuming an uplink Tx-power of 6 dBm per OFDM
subcarrier as reported to the BS and Full Usage of Subchannels
(FUSC)).
A. Impact of the Transmit Power on the Energy Efficiency
The influence of the uplink Tx-power per OFDM subcarrier
(SC) on the energy needed for the successful submission of
one bit was determined by submitting an iPerf generated data
stream at the maximum possible data rate DRMax for the
different Downlink (DL) to Uplink (UL) ratios . During the
transmission, the average power consumption was measured
by using the setup described in section III. Afterwards the
consumed energy E for the submission of one bit was calculated by
RT
1
T 0 P (t)dt
(1)
E[J/Bit] =
RT
1
T 0 DRMax (t)dt

for a defined measurement period T of 30 seconds where

P denotes the actually measured power and DRMax represents the maximum available data rate for a given system parametrization. For the concrete measurement assuming
10 M Hz channel bandwidth and FUSC the values for the
transmission power per subcarrier can be converted to the
overall transmission power by adding 27.5 dB (560 used
x 106

data subcarrier). The results of the measurements can be

seen from the solid lines in Fig. 3. The determined figures
show that the overall power consumption can be divided in
a fixed processing power that does only depend on the DL
to UL ratio and a Tx-power dependent component that
is independent of and added to the processing power.
Polynomial fitting has been performed in a least square sense
to find a polynomial that models the non-linear increase of
the Tx-power dependent component. From that, we found that
a second order polynomial is a very good approximation and
the overall transmit power dependent energy consumption per
bit can by analytically described as
(
C( )
, PTx PTx,Min
E(PTx ) =
2
PTx + PTx + + C( ) , PTx > PTx,Min
(2)
with the constants as shown in Tab. II. The results show
that decreasing the Tx-power allows for massive energy savings. Therefore, the aim of protocol designers should be to
spent only as much energy as necessary to meet the actual
applications QoS requirements. A study on the relationship
between the Signal to Noise and Interference Ratio (SNIR)
and different QoS targets can be found in [6]. Nevertheless,
under extremely good channel conditions it does not make
sense to transmit at a Tx-power below 25 dBm per subcarrier. This is due to the fact that for very low Tx-power
values the processing power is dominating the overall energy
consumption. In this case, the transmit power of 25 dBm per
subcarrier should, from an energy efficiency point of view, be
used to ensure an optimum signal quality even if lower power
would be sufficient for a successful transmission. Nevertheless,
interference avoidance considerations could lead to a further
decreases of the Tx-power.
B. Impact of the App. Data Rate on the Energy Efficiency
Due to the fact that typical applications of broadband
wireless networks include real-time applications such as VoIP
or video streaming the energy efficiency in terms of energy
consumed for the successful submission of one bit has been
determined for different application data rates. From an energy
efficiency perspective the continuous submission of data with a
data rate below the maximum is something quite different than

Energy Consumption per Bit (E) [J/Bit]

1.5
DL/UL Radio Independent
Dynamic Transmitting Power
Plus Fixed Processing Power

1.4

DL/UL Radio Dependent

Fixed Processing Power

TABLE II
PARAMETER S ET F OR T X - POWER D EPENDENT E NERGY E FFICIENCY

1.3
1.2
: 35/12

1.1
1
0.9

: 32/15

0.8
: 29/18

0.7
0.6

: 26/21

Original Measurement
Interpolation by Energy Model

0.5
-10

-15

-20

-25

-30

-35

Tx-power per Subcarrier (SC) of the MS [dBm]

Fig. 3.

Original Measurement vs. Analytical Energy Model

-40

Parameter

Value

1.0325 106

4.6000 105

5.2250 104

C( = 35/12)

1.1084 103

C( = 32/15)

8.3530 104

C( = 29/18)

6.9160 104

C( = 26/21)

5.9940 104

PTx,M in

25dBm/SC

where DRApp denotes the application data rate. Although

one can see from Fig. 4 that the impact of the DL to UL
ratio is negligible for different application data rates, the
zoom to the maximum data rate area shows an interesting
detail: For application data rates below the dependent
maximum (e.g. in case of VoIP) the best choice for reducing
the energy consumption is to apply a DL/UL ratio with less
UL resources. Nevertheless the highest overall efficiency
can be achieved by means of = 26/21, but only if the
maximum available data rate is occupied by the application.
From the results one can conclude that the maximum
energy efficiency can be achieved in case of a maximum
utilization of the link. For the example of wireless sensor
networks this would mean that not every node should be
maintaining a Mobile WiMAX link but only one aggregation
node which collects the data via an energy efficient local
network such as Zigbee. This aggregation node would than
be able to optimally utilize the link and therefore minimize
the overall energy consumption.
C. Impact of the File Size on the Energy Efficiency

10

Energy Consumption per Bit (E) [J/Bit]

Energy Consumption per Bit (E) [J/Bit]

For the determination of the energy efficiency in case of

an uplink file transfer, files with different sizes have been

: 32/15

10

: 26/21

103
Application Data Rate (DRApp ) [kByte/s]

9.7 104
0.94
DRApp

106
101

102

103

Application Data Rate (DRApp ) [kByte/s]

Fig. 4.

0.17
PS

Saturation Area
with Constant
Energy per Bit

Transition
Area

103
Const =
E(PTX = 6 dBm)
(see Eq. 2)

104
105
10

Log. Linear Area

with 1/Datasize Characteristic

107
101

102

103

104

105

106

107

108

Size of Submitted Data (P S) [Byte]

Fig. 5. Energy Efficiency for Different Amounts of Data @ -6 dBm Tx-power

transfered. In contrast to the previously described models

for this investigation the energy consumption per bit was
calculated as
R TP S
P (t)dt
(4)
E[J/Bit] = 0
PS
where TP S determines the time that is needed to submit a file
of P S bits. The results plotted in Fig. 5 show that the energy
consumption per bit has to be modeled for three different
areas. For a packet size of up to 20 kByte the shape is
linear in a double logarithmic plot which again leads to a 1/x
relationship. For packet sizes above 900 kByte the energy
consumption per bit is constant and can be calculated based
on the UL/DL ratio and the Tx-power applying equation
(2). For the transition area between 20 kByte and 900 kByte
either the analytical expression for lower packet sizes or higher
packet sizes can be applied but increased errors have to be
expected. The overall energy model for the packet size P S
dependent energy consumption can be analytically expresses
as:

0.17
P S < 20kByte
E(P S) = P S
(5)
Const P S > 900kByte

: 29/18

105

E(DRApp ) =

E(PS) =

102

From the results one can conclude that from an energy

efficiency perspective the optimal Mobile WiMAX packet
scheduler is one that waits until at least 900 kByte of data
are awaiting submission before it requests the necessary bandwidth from the base station.

: 35/12
6

101
Energy Consumption per Bit (E) [J/Bit]

the submission of one big file (e.g. file transfer). Therefore,

measurements comparable to those described in the previous
section have been performed for variable application data rates
and a fixed uplink Tx-power of 6 dBm/SC. The results
of the investigations can be seen in Fig. 4. The linear shape
of the energy per bit figure in a double logarithmic plot and
the negative slope lead to a 1/x relationship between the
application data rate and the energy efficiency. We found that
the energy consumed for one bit can be analytically modeled
as
9.7 104
E(DRApp ) =
(3)
0.94
DRApp

Energy Model for Different Application Data Rates

D. Generality for Different Chip-Sets and Frequencies

Due to the fact that all of the previously described measurements have been performed based on the same device, an
additional validation measurement has been performed using a
Mobile WiMAX USB stick from an other manufacturer which
is using a different chip-set. Furthermore, the stick applied
for the validation works at a center frequency of 3.5 GHz
instead of 2.5 GHz. Fig. 6 shows the plot of the energy
efficiency as a function of the data size as an exemplary
measurement using this second device. One can see from
the comparison with Fig. 5, that the overall behavior of the
device is comparable to the 2.5 GHz device especially for

Energy Consumption per Bit (E) [J/Bit]

102
Transition
Area

Log. Linear Area

103

V. C ONCLUSIONS

Saturation Area
with Constant
Energy per Bit

104
3.5 GHz Lab Model
0.016

E(PS3.5GHz ) =
105

PS0.9

Measurement with
real BS at 3.5 GHz

106
101

102

103

104

105

106

107

108

Size of Submitted Data (PS) [Byte]

Fig. 6. Model Validation Measurement with different Devices in the Lab

and Outdoor

large packet sizes. Beside this, the dimensions of the different

areas are comparable for both devices. Nevertheless, there are
also differences between the energy efficiencies of the two
different DUTs. The most important one is the significantly
decreased energy consumption per bit of the 3.5 GHz stick
for small data sizes. The reason for this behavior is a tail
energy that is consumed by the 2.5 GHz device after the
termination of the actual data transfer. This behavior is not
observable for the 3.5 GHz device which leads to a decreased
energy consumption especially for small data sizes. Finally,
an additional measurement has been performed using a real
base station instead of the emulator (see Fig. 7). The results
are shown in Fig. 6 (dotted line). One can see, that for larger
packet sizes the energy efficiency equals the efficiency of the
other sticks but for small packets the stick used for the real
world measurements is significantly more efficient than the
other devices. One possible explanation would be, that the
device used for the real world validation is the newest revision
of the stick used for the 3.5 GHz lab test and proceedings in
terms of energy efficiency have been made. Beside this, it has
to be noted that for the real world measurement no control
of the system parameters such as Tx-power and DL/UL ratio
have been possible. Therefore, it is hard to explain the reasons
for the differences in detail. Nevertheless, one can summarize
that the derived energy models in general are valid for different
USB based Mobile WiMAX devices although device specific
offsets may have to be applied for different chip-sets and
frequency bands.

d = 400m

Mobile WiMAX
Basestation

Fig. 7. Mobile WiMAX Link for Validation Measurement, (c) 2011 Google

In this paper we have presented a detailed energy model for

IEEE 802.16e conform Mobile WiMAX devices. Therefore,
extensive measurements of the energy consumption per bit
have been performed for different uplink Tx-power, different
application data rates and different packet sizes. From all of the
measurements closed analytical expressions have been derived
which allow for the application of the models in for example
system level simulators. From the results the following direct
conclusions can be derived for the design of an energy efficient
Mobile WiMAX based network:
The transmit power should be as low as possible. Nevertheless it should, from an energy efficiency perspective,
not be below -25 dBm per subcarrier because for very
low transmission power no further energy savings are
possible.
In wireless sensor networks as few nodes as possible
should maintain a Mobile WiMAX connection due to
their low application data rate. Instead of this, aggregation
nodes should be used which collect the data via for
example Zigbee. Applying this architecture allows for
a higher utilization of the Mobile WiMAX link and
therefore an optimized energy efficiency.
If non real time data has to be transmitted it is useful to
collect at least 900 kByte of data before uplink resources
are requested by the scheduler. For lower packet sizes the
energy efficiency is significantly decreasing.
ACKNOWLEDGMENT
Part of the work on this paper has been supported by
Deutsche Forschungsgemeinschaft (DFG) within the Collaborative Research Center SFB 876 Providing Information by
Resource-Constrained Analysis, projects A4 and B4.
R EFERENCES
[1] L. Bloom, R. Eardley, E. Geelhoed, M. Manahan, and P. Ranganathan,
Investigating the Relationship Between Battery Life and User Acceptance
of Dynamic, Energy-Aware Interfaces on Handhelds, in Proc. Mobile
HCI, 2004, pp.13-24.
[2] L. Zhang, B. Tiwana, Z. Qian, Z. Wang, R. Dick, Z. Mao and L.
Yang, Accurate Online Power Estimation and Automatic Battery Behavior
Based Power Model Generation for Smartphones, in Proc. of the 8th
IEEE/ACM/IFIP International Conference on Hardware/Software Codesign and System Synthesis, p. 105-114, ACM, Scottsdale, USA, 2010
[3] M. B. Kjaergaard, J. Langdal, T. Godsk, and T. Toftkjaer, EnTracked:
Energy-Efficient Robust Position Tracking for Mobile Devices, in Proc.
of the 7th international conference on Mobile systems, applications and
services, p. 221-234, ACM, Krakow, Poland, 2009
[4] N. Balasubramanian, A. Balasubramanian, and A. Venkataramani, Energy
Consumption in Mobile Phones: A Measurement Study and Implications
for Network Applications, in Proc. of the 9th ACM SIGCOMM conference on Internet measurement, p. 280-293, ACM, Chicago, USA, 2009
[5] A. Rice and S. Hay, Decomposing Power Measurements for Mobile
devices, in Prof. of the IEEE International Conference on Pervasive Computing and Communications (PerCom), p. 70 -78, Mannheim, Germany,
2010
[6] C. Ide, B. Dusza, C. Wietfeld, Mobile WiMAX Performance Measurements
with Focus on Different QoS Targets, in Proc. of the 18th IEEE International Workshop on Local and Metropolitan Area Networks (LANMAN),
Chapel Hill, USA, 2011