Content uploaded by Weihua Zhuang
Author content
All content in this area was uploaded by Weihua Zhuang on Nov 02, 2015
Content may be subject to copyright.
1
A Survey on Green Mobile Networking:
From The Perspectives of Network Operators and
Mobile Users
Muhammad Ismail, Member, IEEE, Weihua Zhuang, Fellow, IEEE, Erchin Serpedin, Fellow, IEEE, and
Khalid Qaraqe, Senior Member, IEEE
Abstract—Efficient usage of energy in wireless networks rep-
resents a major concern in academia and industry, mainly
because of environmental, financial, and quality-of-experience
considerations. Various solutions have been proposed to enable
efficient energy usage in wireless networks, and these approaches
are referred to as green wireless communications and networking.
In this survey, we mainly focus on energy efficient techniques
in base stations and mobile terminals as they constitute the
major sources of energy consumption in wireless access networks,
from the operator and user perspectives, respectively. Unlike
the existing articles and surveys, we aim to present a unified
treatment of green solutions and analytical models for both
network operators and mobile users. Such a unified treatment
will help in the future to develop green solutions that enable an
improved and balanced efficient usage of energy by operators
and end users.
Index Terms—Energy efficiency, green communications, power
consumption modeling, resource on-off switching, scheduling
techniques, traffic modeling.
I. INT ROD UC TI ON
During the past decade, there has been an increasing
demand for wireless communication services, which have
extended beyond telephony services to include video streaming
and data applications [1]. This trend has been accompanied by
a wide deployment of wireless access networks. In general,
a wireless access network is defined as a wireless system
that uses radio base stations (BSs) or access points (APs)
to interface mobile terminals (MTs) with the core network
or the Internet [2]. Hence, a wireless access network is
mainly composed of BSs/APs, core network, and MTs [3].
The BSs/APs are responsible for radio resource management
and user mobility management, and provide access to the
Internet. The core network serves as a backbone network with
Internet connectivity and provides data services [3]. Currently,
MTs are equipped with processing and display capabilities
that enable them to provide not just voice services but also
video streaming and data applications. In addition, MTs have
M. Ismail and K. Qaraqe are with the Department of Electrical and
Computer Engineering, Texas A&M University at Qatar, Doha, Qatar, e-
mail:{m.ismail, khalid.qaraqe}@qatar.tamu.edu. Part of this work was done
while the first author was with the Department of Electrical and Computer
Engineering, University of Waterloo, Waterloo, Canada.
W. Zhuang is with the Department of Electrical and Computer Engineering,
University of Waterloo, Waterloo, Canada, e-mail: wzhuang@uwaterloo.ca.
E. Serpedin is with the Department of Electrical and Computer Engi-
neering, Texas A&M University, College Station, TX 77840, USA, email:
serpedin@ece.tamu.edu.
multiple radio interfaces and mobile users can enjoy single-
network and/or multi-homing services [4] - [6].
A. The Need for Green Communications
The BS is the main source of energy consumption in the
wireless access network, from the operator side [2]. It has
been estimated that more than 57% of the operator total energy
consumption is in the BS [2], [7], [8]. In total, there are about
3million BSs worldwide that consume 4.5GW of power [9].
From the user side, it has been estimated that there are around
3billion MTs in the world with power consumption of 0.2−
0.4GW [10]. The high energy consumption of wireless access
networks has promoted increased environmental and financial
concerns for both service operators and users, and quality-of-
experience (QoE) considerations for the mobile users.
From an environmental perspective, the telecommunications
industry contributes 2% of the total CO2emissions worldwide,
and this percentage is expected to increase to 4% by 2020 [11].
In addition, the expected lifetime of rechargeable batteries is
around 2−3years and manifests in 25000 tons of disposed
batteries per year, which triggers environmental concerns (and
financial considerations for mobile users as well) [12]. More-
over, the high energy consumption of BSs and MTs results
in high heat dissipation and electronic pollution [13]. From a
financial perspective, a significant portion of annual operating
expenses of a service provider are energy costs [14], [15].
It has been estimated that the cost of energy bills of service
providers range from 18% (in mature markets in Europe) to
32% (in India) of their operational expenditure (OPEX) [16],
[17]. For cellular networks outside the power grid, the energy
expenses reach up to 50% of the OPEX [18], [19]. Finally,
from a user QoE perspective, it has been shown that more
than 60% of mobile users complain about their limited battery
capacity [20]. The gap between the demand for energy and the
MT offered battery capacity is increasing exponentially with
time [21]. Hence, the MT operational time between battery
charging has become a significant factor in the user perceived
quality-of-service (QoS) [22].
Due to the aforementioned concerns, there has been an
increasing demand for energy efficient solutions in wireless
access networks. The research works carried out in this
direction are referred to as green solutions. The term green
emphasizes the environmental dimension of the proposed
solutions. Therefore, it is not sufficient to present a cost
2
effective solution if it is not friendly to the environment.
For instance, it is not acceptable within the green paradigm
to have a cost effective electricity demand schedule for a
network operator that relies on different electricity retailers,
in a liberated electricity market, without ensuring that the
proposed solution is also environmental friendly in terms of
the resulting CO2emissions [23]. The green wireless com-
munications and networking paradigm aims to reduce energy
consumption of communication devices while taking into
account the environmental impacts of the proposed solutions.
B. Technical Contributions
In the current literature, there exist several surveys that
present different approaches for energy efficient communi-
cations and networking solutions for wired networks [24],
optical networks [25], wireless networks [17], [26] - [28],
and mobile users [29] - [31]. The existing surveys investigate
energy efficiency either from a network operator perspective
[24] - [28] or a mobile user perspective [29] - [31]. However,
green solutions should balance energy efficiency between
network operators and mobile users. There exists a trade-
off in energy saving between service providers and mobile
users, yet the existing research does not account for it. For
instance, consider the BS on-off switching approach that is
proposed in literature to save energy for network operators
in a low call traffic load condition. This approach can lead
to higher energy consumption for MTs in the uplink due to
a larger transmission distance. In this case, such a solution
only shifts the energy consumption burden from the network
operator to the mobile users, leading to battery drain for MTs
at a faster rate. Therfore, it is necessary to develop a better
understanding of green solutions for both network operators
and mobile users, in order to develop energy efficient solutions
that account for and balance the trade-off in energy saving
between network operators and mobile users. In addition,
call traffic dynamics and power consumption modeling play
a significant role in developing effective green solutions,
but have not been carefully studied in literature. Moreover,
other performance metrics conflicting with achieving energy
efficiency should also be considered. The contributions of this
survey are next briefly summarized1:
•A presentation of various power consumption models
for BSs and MTs to capture transmission power, circuit
power, and reception power consumption.
•A detailed description of different energy efficiency def-
initions proposed in literature, under different call traffic
load conditions for network operators and under different
networking settings for mobile users (e.g., single-user or
multi-user with or without fairness considerations).
•A review of different models proposed for call traffic
load dynamics that take into account the spatial and
temporal fluctuations at different scales (long-term, short-
term, flow-level, and packet-level).
•An overview of the performance metrics conflicting with
achieving energy efficiency in wireless communications
1In this survey, we mainly focus on direct communications (i.e., no multi-
hop or relaying techniques) for licensed networks and users.
and networking.
•A unified treatment of energy efficient solutions for
network operators and mobile users. Specifically, we
classify the energy efficient solutions based on the call
traffic load condition into low and high call traffic load
solutions. Using such a classification, we discuss the
similar approaches adopted by network operators and
mobile users to save energy for BSs and MTs.
•Identifying future research directions that help to develop
effective green solutions, which can balance energy sav-
ing among network operators and mobile users.
C. Organization
Throughout this study, we aim to present a complete picture
of energy efficient (green) models and solutions for BSs and
MTs that enable new approaches to balance the energy saving
for both network operators and mobile users. In order to
develop/analyze a green communication solution, an appro-
priate definition of energy efficiency for network operators
and mobile users should be introduced. Such a definition
relies on power consumption, throughput, and traffic load
models for network operators and mobile users. Moreover, the
green communication solution should satisfy some conflicting
performance metrics. Hence, the first part of the paper is
dedicated to energy efficiency definitions and power consump-
tion, throughput, and traffic load models for network operators
and mobile users, along with conflicting performance metrics.
After introducing the background concepts, the second part
of the paper focuses on state-of-the-art green communication
solutions and analytical models for network operators and
mobile users at different traffic load conditions. Finally, we
discuss the impact of green communication solutions from
the perspectives of network operators and mobile users, re-
spectively, aiming to balance the existing trade-offs.
The rest of this paper is organized as follows. In Section II,
different definitions are presented to describe energy efficiency
in wireless networks at different traffic load conditions, along
with throughput and power consumption models, from the
network operator and mobile user perspectives. Section III
discusses different models to capture the temporal and spatial
fluctuations in traffic load, along with some performance
metrics that conflict with the general objective of achieving
energy efficiency in wireless networks. Green solutions and
analytical models are reviewed for network operators and
mobile users at low and high traffic load conditions in Sections
IV and V, respectively. Finally, future research directions and
conclusions are given in Sections VI and VII, respectively.
II. MODELING OF EN ER GY EFFI CI EN CY I N WIRELESS
NET WORKS
In this section, we present different definitions that have
been proposed in the literature to assess the energy efficiency
of wireless networks from the network operator and mobile
user perspective, respectively. As an important component of
energy efciency denitions, we rst present different throughput
and power consumption models for BSs and MTs.
3
A. Throughput Models
In what follows, we first introduce the concepts of aggregate
network capacity Cn, area spectral efficiency Tn, and the user
achieved data rate Rm. These concepts are necessary in the
energy efficiency definitions that we will later present.
To measure the network aggregate capacity Cn, Shannon
formula is used [23]
Cn=Bnlog2det(I+P H H†),(1)
where Bnrepresents network ntotal bandwidth, Idenotes
the identity matrix, Pstands for the transmission power
matrix of every BS of network nto every MT min service,
His the channel matrix between each BS of network n
and MT m, and †denotes the transposition operation. The
channel matrix Hmight account for fast fading, noise, and
interference affecting the radio transmission. Cnin (1) has
the unit of bits per second. On the other hand, in a low call
traffic load condition, the area spectral efficiency Tnis used
in the energy efficiency definition rather than the aggregate
capacity given in (1) [15], as will be shown. The area spectral
efficiency measures the network throughput while considering
the coverage probability. Let P(γx→u> ζ)denote the success
probability of the signal-to-noise ratio (SNR) γof a BS at
location xreceived by an MT at location u, satisfying a certain
QoS threshold ζ. The coverage probability Pn(ζ)is obtained
by averaging the success probability over the propagation
range to location u. Hence, for network nwith BS density
λn, the area spectral efficiency Tnis given by
Tn=λnPn(ζ) log2(1 + ζ)(2)
where Tnis measured over a unit area.
In literature, two key definitions have been proposed to
assess the data rate Rmachieved by MT m. The first definition
is based on the knowledge of the instantaneous channel state
information (CSI) [13], [32] - [34], and takes the form
Rm=Bmlog2(1 + γm
Γ),(3)
where Bmdenotes the allocated bandwidth on the uplink to
MT m,γmis the SNR of MT mreceived at the destination,
and Γis the SNR gap between the channel capacity and
a practical coding and modulation scheme. For Shannon
formula, Γ = 1. However, the instantaneous CSI requires a
feedback from each MT to the serving BS, which results in a
large overhead. Hence, statistical CSI can be used to reduce
the amount of overhead. In this case, Rmis described in a
statistical average sense [35], i.e.,
Rm=EH[Bmlog2(1 + γm
Γ)],(4)
where EHdenotes the expectation over channel state H, which
affects the SNR γm.Rmin (3) and (4) is measured in bits per
second.
B. Power Consumption Models
Different models have been proposed in literature to repre-
sent power consumption for the network, Pn, and MTs, Pm,
as shown in Table I.
Power amplifier and
feeder
65%
Power supply
7%
Signal processing
(digital and analog)
10%
Air conditioning
18%
Fig. 1. Percentage of power consumption at different components of a large-
cell BS [9].
The network power consumption Pncan be modeled as the
aggregate power consumption of the network BSs. Practical
measurements of power consumption at BSs are reported in
[7], [9], [19], [36]. Figure 1 shows the percentage of power
consumption at different components of a large-cell BS. On
the other hand, Table II shows the power consumption profile
for a femto-cell BS. By comparing the values in Figure 1
and Table II, we observe that a femto-cell BS consumes
most of the power in the signal processing part as opposed
to a large-cell BS (65.6% and 10% for femto and large-cell
BSs, respectively). In addition, the RF transmission/reception
energy consumption part in a femto-cell BS is almost half
of that of a large-cell BS, with only 19.6% of the power
consumed in the femto-cell BS power amplifier as opposed
to 65% in a large-cell BS.
Let Pbdenote the power consumption for BS b. In literature,
different models are used to represent Pb. The first one is an
ideal load dependent power model [37]. This model assumes
that the BS consists only of energy proportional devices and
hence, it assumes no power consumption in an idle state, i.e.,
Pb=ρPtb,(5)
where ρstands for the system traffic load density and Ptb
denotes the BS transmitted power. However, such a model is
not realistic as the power consumption of some components
shown in Figure 1 does not scale with the call traffic load. A
more sophisticated model captures the power consumption of
different BS components, and assumes the expression [38]
Pb=
Ptb
ξ(1−σfeed)+PRF +PBB
(1 −σDC)(1 −σMS )(1 −σcool),(6)
where PRF is the radio frequency (RF) power consumption,
PBB represents the baseband unit power consumption, ξde-
notes the power amplifier efficiency, and σfeed,σDC,σMS ,
and σcool stand for the losses incurred by the antenna feeder,
DC - DC power supply, main supply, and active cooling,
respectively. For simplicity, the model (6) can be further
approximated by a linear model [11], [15], [18], [23], [37],
[38]. In such a linear model, two components are introduced
to represent Pb. The first component is a fixed power term
that captures the power consumption at the power supply,
cooling, backhaul, and other circuits, and is denoted by Pf.
The second component is proportional to the traffic load. The
linear (affine) model is described by
Pb= ∆pPtb +Pf,(7)
4
TABLE I
ASU MMA RY OF DI FFE REN T POW ER M ODE LS I N LIT ER ATURE .
BS
Model
Comments
References
Operation only
Large
-
cell
Ideal
The BS consumes no power
when idle, i.e., the BS consists
only of energy proportional
devices.
[37]
Realistic
The model captures the BS
traffic load independent power
consumption.
[11], [15],
[18
], [23],
[37], [38]
Femto-cell
Load
independent
The BS power consumption
does not depend on the offered
traffic load.
[39]
Load
dependent
The BS power consumption
relies on traffic load, packet
size, and has an idle part.
[40]
Operation and Embodied
Besides the operation power, it
accounts for the consumed
energy in BS manufacturing
and maintenance.
[16]
MT
Transmission power
only
Without power
amplifier efficiency
The models does not account
for the transmitter power
amplifier efficiency.
[34], [42
],
[43]
With power
amplifier efficiency
The model accounts for the
transmitter power amplifier
efficiency.
[13], [32],
[44], [45]
Including circuit
power
Constant
The circuit power consumption
is given by a constant term
independent of the bandwidth
and data rate.
[13], [20],
[32], [33],
[35], [46]
Bandwidth scale
The circuit power consumption
scales with the MT assigned
bandwidth.
[47]
Data rate scale
The circuit power consumption
scales with the MT achieved
data rate.
[45]
Including reception power
Besides the transmitter and
circuit power consumption, the
model also accounts for the
receiver power consumption.
[13
], [47]
TABLE II
POWE R CO NSU MPT IO N PRO FILE F OR A F EMT O-CE LL BS [36].
Hardware component
Power Consumption (W)
Percentage (%)
Microprocessor
Associated memory
Backhaul circuitry
1.7
0.5
0.5
26.4
FPGA
Associated memory
Other hardware functions
2
0.5
1.5
39.2
RF transmitter
RF receiver
RF power amplifier
1
0.5
2
34.3
5
where ∆pis the slope of the load dependent power consump-
tion.
For a femto-cell BS, the power consumption model [39] is
expressed as
Pb=Pmp +PFPGA +Ptx +Pamp,(8)
where Pmp,PFPGA ,Ptx, and Pamp denote the power con-
sumption of the microprocessor, FPGA, transmitter, and power
amplifier. While the power consumption model in (8) accounts
for most of the components in Table II, it does not exhibit
any dependence on the call traffic load. Experimental results
in [40] have illustrated the dependence of the femto-cell BS
power consumption on the offered load and the data packet
size. Hence, the power consumption model for a femto-cell
BS is given by [40]
Pb=Pb(q, s) + Pf,(9)
where Pb(q, s)and Pfdenote the BS power consumption that
relies on the traffic load q(expressed in Mbps) and packet
size s(expressed in bytes) and the idle power consumption
component, respectively.
The models (5) - (9) focus mainly on the BS operation
power, which is expressed in watts. In a more general model,
the power consumption is described in terms of the BS oper-
ating energy, Eo, and embodied energy, Ee, which represents
30−40% of the BS total energy consumption [16]. The embod-
ied energy accounts for the energy consumed by all processes
associated with the manufacturing and maintenance of the BS,
and is calculated as 75 GJ over the BS lifetime [16]. It consists
of two components, the first refers to the initial embodied
energy Eei and comprises the energy used to acquire and
process raw materials, manufacture components, and assemble
and install all BS components, and it is accounted for only
once in the initial BS manufacturing. The second component
stands for the maintenance embodied energy Eem and includes
the energy associated with maintaining, repairing, and replac-
ing the materials and components of the BS throughout its
lifetime. Hence, the BS total energy consumption, in joules,
throughout its lifetime is expressed as
Eb=Ee+Eo= (Eei +Eem) + Eo,(10)
where Eem =PemTlifetime and Pem and Tlifetime denote
the BS maintenance power and lifetime, respectively, and
Eo=PoTlifetime, where Pois defined in terms of the BS
operating power given in (5) - (9). The model in (10) is useful
for measuring the BS total power consumption in the network
design stage, for instance, while designing a multi-tier wireless
network, as will be discussed.
Practical measurements of power consumption at MTs are
summarized based on different experiments in [29]. Besides
the results reported in [29], we present experimental results
from [41] and [31], in Tables III and IV, respectively, to
discuss the models to be presented later. Different models have
been proposed in literature for the MT power consumption Pm.
In the simplest model, Pmrepresents the MT transmission
power Ptm [34], [42], [43]. When the effect of the power
amplifier efficiency is considered, the MT power consumption,
in watts, is expressed as [13], [32], [44], [45]
Pm=Ptm
ζm
,(11)
where ζmdenotes the power amplifier efficiency for MT m,
ζm∈(0,1]. With such a power consumption model, for a data
call, the minimum energy consumption is attained by using
the modulation of the lowest order while satisfying the QoS
constraints (e.g., time delay) [33]. However, in practice, the
MT circuit power should be captured in the power consump-
tion model Pm. Three different models have been proposed
to capture the MT circuit power Pcm. In the first model,
the circuit power consumption is modeled as a constant,
independent of the achieved data rate Rm[13], [20], [32],
[33], [35], [46]. With constant circuit power consumption,
transmitting with the lowest modulation order is no longer the
best strategy as the energy consumption is proportional to the
transmission duration [33]. The constant power consumption
model, however, does not reflect the effect of transmission
bandwidth and data rate on the MT circuit power consumption.
From Table III, it is evident that different radio interfaces
consume different circuit power, which for one reason is due
to the different operating bandwidth. To account for the effect
of the allocated bandwidth, the circuit power consumption is
modeled via two components [47]. The first component refers
to the digital circuit power consumption, which is modeled as
a linear function of the transmission bandwidth (as bandwidth
increases, more computations and baseband processing are
required), i.e.,
Pbm =Pref
b+αBm
Bref
,(12)
where Pref
bdenotes the reference digital circuit power con-
sumption, in watt, for a reference bandwidth Bref and αis
a proportionality constant. The second component captures
the power consumption of the radio frequency (RF) chain,
represented by a constant to account for power consumption
in the digital-to-analog converter, RF filter, local oscillator, and
mixer. However, the model in (12) does not account for the
effect of transmission data rate on the power consumption,
as indicated in Table IV. To account for the transmission
data rate, the circuit power consumption is modeled as a
linear function of the achieved data rate, under the assumption
that the clock frequency of the MT digital chips scales with
the achieved data rate [45]. Therefore, the circuit power
consumption is expressed as
Pcm =µ+βRm,(13)
where µand βare two appropriately chosen constants, mea-
sured in watt and watt per bit per second, respectively. Besides
the transmission and circuit power modeling in Pm, a constant
term is considered to represent the MT receiver circuit power
consumption [13], [47].
It is worth mentioning that Rmand Pmcan be defined as the
sum of corresponding terms over multiple subcarriers assigned
to MT mfor orthogonal frequency division multiple access
(OFDMA) networks [13], [32] - [34], [42], [43] or the sum
6
TABLE III
MT POWER CONSUMPTION FOR DIFFERENT TECHNOLOGIES [41].
Technology
Action
Power (mw)
WiFi
IEEE 802.11
(Infrastructure mode
)
In connection
868
In disconnection
135
Idle
58
Idle in power save mode
26
Downloading at 4.5 Mbps
1450
WiFi
IEEE 802.11
(ad hoc mode)
Sending at 700 kbps
1629
Receiving
1375
Idle
979
2G
Downloading at 44 kbps
500
Handover to 3G
1389
3G
Downloading at 1 Mbps
1400
Handover to 2G
591
TABLE IV
MT P OWE R CON SUM PT ION F OR D IFFE RE NT DATA RATE S OF AUD IO S TRE AM ING A ND D OWNL OAD IN G A 200 MB FIL E USI NG WIFI[31].
Bit Rate
(kbps)
Nokia E-71
(mW)
Nexus S
(mW)
Samsung Galaxy S3
(mW)
128
192
256
File download
990
1004
1007
1092
350
390
390
998
419
440
452
1012
over multiple radio interfaces for MT mwith multi-homing
capability [20], [46].
C. Energy Efficiency Models
Based on the throughput and power consumption models
discussed before, we next present different energy efficiency
definitions for networks and MTs in this subsection. A list de-
scribing the energy efficiency definitions proposed in literature
is presented in Table V.
A generic definition that can be used as a measure of energy
efficiency is referred to as energy consumption gain, and it is
defined as the ratio of the energy consumed by a base system
(BS or MT) to the energy consumed by the system under
test, assuming the same conditions [7], [48]. Formally, this is
expressed by
η=Ebase −Etest
Ebase
(14)
where Ebase and Etest are measured in joules. The definition
in (14) is a relative definition that can be used in any call
traffic load condition. Next, we present the absolute energy
efficiency definitions.
In a low call traffic load, the mobile user demands do not
require that the network operates at its full power. Therefore,
one way to measure energy efficiency for network operators
at a low call traffic load condition is by means of the ratio
between the network output power (energy) and the total input
power (energy) [2], [19], i.e., the energy efficiency ηnfor
network nis expressed as
ηn=Pt
Pn
,(15)
where Ptand Pndenote the network output power (i.e., the
power of the RF transmitted signal) and input (consumed)
power, respectively. Hence, ηnis unitless. In addition, due
to the low service demands, it is not necessary to guarantee
that the network achieves a full coverage. It is sufficient to
satisfy an acceptable coverage probability. As the definition in
(15) does not reflect the achieved network coverage, another
definition of energy efficiency measures the power consumed
to cover a certain area [15], [19]. Hence, energy efficiency can
be defined as [15]
ηn=Tn
Pn
.(16)
In (16), ηnhas unit of watt−1.
From an operator perspective, in a high call traffic load
condition, energy efficiency is defined as the ratio of the
aggregate network capacity to the total power consumed by
the entire network [2], [9]. Therefore, in a high call traffic
load condition, energy efficiency of network nisexpressed as
ηn=Cn
Pn
,(17)
in bit per second per watt.
For mobile users, energy efficiency is expressed as a mea-
sure of the maximum amount of bits that can be delivered per
7
TABLE V
ASU MMA RY OF DI FFE REN T EN ERG Y EFFI CIE NC Y DEFI NI TIO NS I N LIT ER ATURE .
BS/MT
Model
Comments
References
Energy consumption gain
A ratio of the energy consumed
by a base system to the energy
consumed by the system under
test. It is a relative measure that
can be used at any traffic load.
[7], [48]
BS
Low traffic load
Output – input
power
A ratio of network output to input
power.
[2], [19]
Area spectral
efficiency
–
input
power
The definition measures the
power consumed for certain area
coverage. It is used at low traffic
load.
[15], [19]
High traffic load
Network capacity –
input power
A ratio of the aggregate network
capacity to the total power
consumed by the network.
[2]
, [9]
MT
Single-
user
system
Without error
consideration
A ratio of throughput to power
consumption.
[13], [20],
[32], [42
],
[46]
With error
consideration
A ratio of goodput to power
consumption.
[42], [43]
Multi-
user
system
Without fairness
consideration
It can be sum rate of all MTs to
total power consumption or sum
of energy efficiency for
individual MTs.
[33] –
[35],
[43], [44]
With fairness
consideration
It is a geometric mean of energy
efficiencies of all MTs.
[33]
joule of energy consumed [13], [20], [32], [42], [46]. This can
be described, for mobile user m, as
ηm=Rm∆T
∆Em
=Rm
∆Em/∆T=Rm
Pm
,(18)
where ∆Emstands for the consumed energy during time
interval ∆Tby MT m. However, the definition in (18) does not
account for the energy consumed for correct reception of data.
Another definition measures the net number of information bits
that are transmitted without error per joule [42], [43], and it
is expressed as
ηm=Rmf(γm)
Pm
,(19)
where f(γm)stands for the packet transmission success rate
for a given SNR γmfor MT m. The definition in (19)
assumes a ratio of the goodput to power consumption, as
compared to the definition in (18) which assumes the ratio
of throughput to power consumption. In general, the packet
transmission success rate follows an S-shaped (sigmoidal)
function, exhibiting an increasing trend with respect to γm.
In addition, f(γm)approaches zero as γmapproaches zero
and f(γm)approaches unity as γmapproaches infinity [42],
[43]. The unit of ηmin (18) and (19) is bit per second per
watt.
The definitions in (18) and (19) have been proposed for a
single-user scenario [13], [20], [32], [42], [46]. A multi-user
system is considered in various scenarios due to the shared
bandwidth [33] or interference [34] caused by simultaneous
transmissions. In a multi-user scenario, energy efficiency is
defined as the ratio between the sum rate of all MTs to the
total power consumption [44]
ηtotal =mRm
mPm
.(20)
The definition in (20) treats all mobile users as a single unit,
and takes into account only the total achieved throughput and
power consumption. In order to model the system as a set
of distinct mobile users, another definition is used, which
represents the total energy efficiency expressed as the sum
of energy efficiency for individual MTs [33] - [35], [43]
ηtotal =
m
ηm.(21)
The unit of ηtotal in (20) and (21) is bit per second per watt.
However, the definitions in (20) and (21) provide no fairness
guarantee for energy efficiency among different MTs. Con-
sequently, some MTs might present high energy efficiencies
while others might exhibit low energy efficiencies very close
to zero. The geometric mean of energy efficiencies of all MTs
promotes fairness among users [33], and is expressed as
ηtotal =
m
log(ηm).(22)
Unlike (20) and (21), ηtotal in (22) is unitless.
An energy efficiency definition and an associate green
solution must be chosen in accordance with an appropriate
traffic load model. Hence, in the next section, we discuss the
different traffic models proposed in literature.
8
III. TRA FFIC MO DE LI NG A ND PE RFORMANCE METRICS
Different green solutions can be employed at different
call traffic load conditions. In addition, some green solutions
exploit the temporal and spatial fluctuations in the call traffic
load to save energy. For instance, in order to determine the
switch off duration of a BS or MT, as will be discussed
later, traffic models are used to probabilistically determine
the idle period duration. Furthermore, performance evaluation
of the proposed green solutions should be carried out using
an appropriate traffic model. Hence, it is necessary to gain a
better understanding of different traffic load models proposed
in literature. After specifying an appropriate energy efficiency
definition and a traffic load model, the next step is to define
the performance metrics that may conflict with the objective
of enhancing the system energy efficiency. Therefore, before
we discuss various green networking solutions, in this section,
we present traffic models for BSs and MTs. Then, we discuss
performance metrics that are considered for developing green
networking solutions.
Traffic modeling can be classified in two categories, as
shown in Table VI. The static model assumes a fixed set of
MTs, M, that communicate with a fixed set of BSs, B[20],
[33], [34], [43], [44], [46], [49] - [51]. The static model does
not capture the mobility of MTs in terms of their arrivals and
departures. Also, the static model does not consider the call-
level or packet-level dynamics in terms of call duration, packet
arrival, etc. On the other side, the dynamic model captures
the spatial and temporal fluctuations of the traffic load, and is
discussed in more details next.
A. Traffic Spatial Fluctuation Models
It has been shown that traffic is quite diverse even among
closely located BSs [52]. Therefore, different models have
been proposed to capture the spatial fluctuations in call traffic
load [15], [37], [53].
One approach to model traffic spatial fluctuations is by
defining a location-based traffic load density [37]. In this
case, a geographic region is served by a set Bof BSs. The
geographic region is partitioned into a set of locations. In
a location x, the file transfer requests arrive following an
inhomogeneous Poisson point process (PPP) with an arrival
rate per unit area λ(x). The file sizes are independently
distributed with mean 1/µ(x)at the location. Hence, the traffic
load density is expressed as ϱ(x) = λ(x)/µ(x)<∞, and it
is used as a measure to capture the spatial traffic variability.
While the above approach uses a predefined set of BSs,
B, with specific locations, an alternative approach defines the
locations of available BSs using stochastic geometry theory
[15]. In this case, the BSs of network nare located according
to a homogeneous PPP, Θn, with intensity λnin the Euclidean
plane. MTs are distributed according to a different independent
stationary point process with intensity λm. With a stationary
PPP Θn, the distribution of the distance between an MT and
its serving BS, Dm, is the same regardless of the MT exact
location. The probability density function (pdf) of Dmis given
by [15]
fDm(d) = 2πλndexp−λnπd2, d > 0.(23)
The above two models capture the traffic spatial variability
among different cells. To model the spatial distribution of MTs
within a given cell i, a finite-state Markov chain (FSMC)
model is used [53]. This model categorizes the MTs into
Sclasses according to the radius of cell i. Assuming there
are MMTs in cell i, an SMspatial location distribution
is considered within the cell. Hence, the FSMC presents
L={L1, . . . , LSM}states. The state transition probability
Pr{Li(t+ 1) = vi|Li(t) = ui}is the probability of spatial
distribution of the MTs within cell iat time slot t+1 being vi
given that it was uiat time slot t, where ui={ui,1, . . . , ui,M }
and vi={vi,1, . . . , vi,M }. Using such a model, the dynamic
fluctuations in the number of MTs in different regions within
the cell can be captured.
B. Traffic Temporal Fluctuation Models
Traffic temporal fluctuations can be observed over two
different time scales [11], [38]. The first scale is a long-term
traffic fluctuation model that captures the traffic variations over
days of the week. This model is very useful in investigating
energy efficient solutions for network operators as it captures
both high and low call traffic load conditions. The second
scale is a short-term traffic fluctuation that models the call
(packet) arrivals and departures of the MTs. This model plays
an important role in investigating energy efficient resource
allocation schemes for MTs and BSs. Next, we will describe
the two scales in more details.
1) Long-term Traffic Fluctuations: Real traffic traces show
a sinusoidal traffic profile in each cell [14], [54]. Traffic during
day time (11 am - 9 pm) is much higher than that during
night time (10 pm - 9 am) [52], [14]. Moreover, the traffic
profile during weekends and holidays, even during the peak
hours, is much lower than that of a normal week day [14].
During a weekday, the traffic profile is 10% less than its
peak value 30% of the time, increasing to 43% of the time
during weekends [14]. Such a behavior can be modeled using
an activity parameter ψ(t)to specify the percentage of active
subscribers over time t[38]. Hence, if the population density
is pusers per km2, the number of operators is N, each being
able to carry 1/N of the total traffic volume, and the fraction
of subscribers is skwith an average data rate rkfor terminal
type k(e.g., smart phone and tablet), then the traffic demand,
in bit per second per km2, is given by
A(t) = p
Nψ(t)
k
skrk.(24)
It has been shown that the traffic load difference between
two consecutive days for 70% of the BSs is less than 20%
[52]. Therefore, the long-term fluctuations in call traffic load
are estimated from the historical mobile traffic statistics (i.e.,
the activity parameter ψ(t)and the average data rate rkcan
be inferred from historical data).
2) Short-term Traffic Fluctuations: The short-term traf-
fic fluctuation models can be classified into two categories,
namely call (flow)-level and packet-level models. The call
(flow)-level models are useful in investigating green resource
scheduling algorithms at the BSs and MTs in a high call traffic
9
TABLE VI
ASU MMA RY OF D IFFE REN T TR AFFIC M OD ELS .
Model
Comments
References
Static
It does not capture the MT
mobility
and the
traffic
dynamics.
[20], [33],
[34], [43],
[44], [46],
[49] – [51]
Dynamic
Spatial
Regional traffic load density
It defines a location-based
traffic load density.
[37]
Stochastic geometry
BSs and MTs are located
according to homogeneous
Poisson point process.
[15]
FSMC
It models the spatial
distribution of MTs within a
cell.
[53]
Temporal
Long-scale
The model captures traffic
fluctuations over the days of
the week.
[14], [38],
[52], [54]
Short
scale
Flow-
level
Poisson-
exponential
It models call arrivals as
Poisson process and call
departures as exponential
distribution.
[11],
[55] –
[57]
FSMC
The number of calls within a
cell is represented by a state
in a Markov chain.
[53]
Packet-
level
Infinite
buffer
It models the number of
backlogged packets in an
MT buffer with infinite
capacity.
[58]
Finite buffer
It models the number of
backlogged packets in an
MT buffer with finite
capacity.
[59]
load. One approach that can be used for myopic resource
allocation solutions models the call arrivals using a Poisson
process with rate λaand call durations with an exponential
distribution [11], [55] - [57]. For dynamic resource allocation
solutions, traffic dynamics in terms of call arrivals and depar-
tures are modeled using an FSMC [53]. The number of calls
in a given cell iis represented by an M-state Markov chain,
with state set M={0,1, . . . , M −1}. The state transition
probability Pr{Mi(t+ 1) = hi|Mi(t) = gi}is the probability
of having hiMTs within cell iat time slot t+ 1 given that
there were giMTs at time slot t, where hi,gi∈ M.
Packet-level traffic models are useful in investigating green
resource solutions (on-off switching) at the BSs and MTs, in
a low call traffic load condition, through modeling the BS/MT
buffer dynamics in terms of packet arrival and transmission
[58], [59]. For instance, for an infinite buffer size, the MT
buffer dynamics are represented by
om(t+1) = max{om(t)+am(t+1)−zm(t), am(t+1)},(25)
where om(t),am(t), and zm(t)denote the numbers of back-
logged packets in the buffer, arriving packets, and transmitted
packets, for MT min time slot t, respectively. For a buffer
with finite size O, the MT buffer dynamics can be expressed
as
om(t+ 1) = min{om(t) + am(t+ 1) −zm(t), O}.(26)
The models (25) and (26) are used to design and evaluate
optimal on-off switching schemes for radio interfaces of MTs
to achieve energy efficient communications in low call traffic
load conditions, a topic which will be addressed later herein
survey.
C. Performance Metrics
Improving energy efficiency in wireless networks may con-
flict with other performance metrics. Therefore, there is a trend
on improving energy efficiency while not violating the target
performance metric. Such performance metrics can be divided
into two main categories. The first category deals with the
quality of the ongoing application and hence is referred to
as application quality requirements. The second deals with
connection establishment and hence is referred to as admission
quality requirements. These are summarized as follows.
1) Application Quality Requirements:
•SNR [52], [53]: Improving energy efficiency reduces the
transmission power which reduces the overall received
SNR. The receiver may not be able to decode the trans-
mitted signal. Hence, a minimum SNR should be satisfied
while improving energy efficiency.
•Data rate [20], [46], [50], [51], [53], [56]: With a re-
duced transmission power level, transmission data rate
is reduced. For some applications, a minimum required
data rate should be achieved [46], [52], [53] or a constant
required data rate should be satisfied [20], [50].
•Delay time [59]: A reduced data rate can lead to a
violation of the required delay deadline for delay sensitive
10
applications. An equivalent representation to ensure a
minimum required data rate is not to violate a maximum
delay bound for data transmission.
•Video quality [60], [61]: For video streaming applica-
tions, using a lower transmission rate for improved energy
efficiency can result in video packets missing their delay
deadlines. Hence, the resulting video quality is degraded.
It is required to maintain the achieved video quality
higher than a target value.
2) Admission Quality Requirements:
•Call blocking probability [11], [23]: Improving energy
efficiency for wireless networks can be achieved by
switching-off some BSs with a low call traffic load,
as explained in the next section. This can lead to an
increased call blocking probability. Hence, it is required
to maintain the call blocking probability below a certain
threshold.
•Coverage probability [15]: BSs can improve their energy
efficiency by reducing their transmission power. However,
this may result in failure in service coverage. It is
required to maintain a target performance level in terms of
coverage probability Pn(ζ), as explained in the previous
section.
Given the background provided in the previous sections,
in the next section we will present state-of-the-art green
communication solutions and analytical models for network
operators and mobile users at different traffic load conditions.
In green wireless networks, the proposed solutions/models to
enhance/analyze energy efficiency can be divided into two cat-
egories based on the call traffic load condition. At a low and/or
bursty call traffic load, resource on-off switching techniques
are adopted, while scheduling techniques are employed at a
high and/or continuous call traffic load. These are discussed
in detail in the next two sections.
IV. GRE EN SOLUTIONS AND ANALYT IC AL MO DE LS AT
LOW AN D/O R BUR ST Y CAL L TRA FFIC LOAD
In this case, on-off switching of radio resources is adopted
to enhance energy efficiency, as shown in Table VII. Network
operators can employ on-off switching mechanisms for their
BSs at a low call traffic load. Similarly, MTs can switch on-off
their radio interfaces in a bursty traffic condition. The related
research issues and modeling techniques are discussed in the
following.
A. BS On-Off Switching
In network planning, the cell size and capacity are in general
designed based on the peak call traffic load. As discussed in
Section III, the call traffic load exhibits significant spatial and
temporal fluctuations. At a low call traffic load, the network
is over-provisioned which results in energy waste. It has been
argued that switching off some of the available radio resources
(e.g., radio transceivers of BSs) at a low call traffic load
can save energy and maintain acceptable performance metrics.
However, studies have shown that, when a BS is active, the
energy consumption of processing circuits and air conditioner
amounts to 60% of the BS total power consumption (which
is represented by the fixed power component in (7)) [51].
Hence, an effective approach for energy saving at a low
call traffic load is to switch off some of the network BSs
while satisfying the required performance metrics. Switching
BSs on and off according to call traffic load conditions is
referred to as dynamic planning [11]. Two important issues
must be addressed while designing an effective BS switching
mechanism, namely, user association and BS operation.
Switching BSs on and off is coupled with the user asso-
ciation problem. In order to switch off some BSs, the call
traffic load should be first concentrated in a few BSs, which is
achieved through user association. Newly incoming MTs have
to be associated with a subset of active BSs, and MTs already
in service should perform handover when the serving BSs are
switching off. One can identify two research directions related
to the MT association problem. The first direction deals with
developing new user association mechanisms [37], [51], [62]
- [64], while the second one focuses on deriving analytical
models to assess the performance of different association
mechanisms [55]. In developing a MT association mechanism,
two approaches can be adopted to meet the MT required
QoS while concentrating traffic load in a few BSs. The
first approach adopts an objective function that minimizes
the networks’ energy consumption while satisfying the user
required QoS constraints, while the second approach aims
to balance the trade-off between flow level performance for
MTs (e.g., data rate or delay) and energy consumption of the
network [37]. In the later case, the problem is a multi-objective
one with a weighting factor. When the weighting factor equals
zero, the MT association is determined based on the flow
level performance and, as the weighting factor increases, the
MT association decision focuses more on the network power
consumption performance. As the weighting factor goes to
infinity, the MT connects to the BS that maximizes the network
energy efficiency performance in bits per joule. The MT
association mechanism can be implemented in a centralized
or a decentralized architecture [51]. Both architectures aim
to concentrate MTs in a few BSs while satisfying the data
rate requirements of MTs and bandwidth limitations of BSs.
In the centralized mechanism, a central controller performs
MT association based on global network information that is
related to channel conditions and user requirements. On the
other hand, an MT locally selects the BS with the highest call
traffic load that can serve its required data rate in the iterative
decentralized mechanism. One challenge with designing such
a mechanism is related to computational complexity, due to
the binary nature of the decision variables related to the
BS on-off switching, and hence the mixed-integer nature of
the optimization problem. Therefore, greedy algorithms are
mainly adopted to reach a good switching decision [37], [51].
In designing such algorithms, a decision criterion should be
defined. For instance, in a user-BS distance decision criterion,
the greedy algorithm tends to switch off the BSs with the
longest user-BS distance to improve energy efficiency of the
network [62]. The rationale behind such a decision criterion is
that the longer the user-BS distance, the greater the transmis-
sion power required to meet the target service quality of the
11
TABLE VII
ASU MMA RY OF G REE N SOL UT ION S AN D ANA LYTI CAL M OD ELS AT LO W AND /OR B URS TY CA LL T RAFFI C LOA D.
Solution/Analytical Model
Comments
References
BS on-off
switching
User association
This phase concentrates the
MTs in a few BSs to enable
switching off other BSs.
[37], [51],
[55], [62],
[63], [64],
BS operation
This phase specifies which
BSs should be turned off and
how.
[11], [51],
[65] -
[69]
MT radio
interface
on-
off
switching
With
downlink
traffic
Without
traffic
shaping
An MT switches its radio
interface if no data packets
are available for the MT at the
BS.
[48], [70],
[71]
With traffic
shaping
Traffic shaping at the MT or
BS is introduced to enable
longer idle duration for the
MT.
[72], [73
],
[74]
With uplink traffic
Besides radio interface on-off
switching, an MT controls the
transmission power and
modulation and coding.
[58], [75]
With bi-
directional traffic
This case deal with presence
of both uplink and downlink
traffic while switching on and
off the MT radio interface.
[76]
users. The network-impact notion is introduced in [63] as a key
decision criterion, which quantifies the effect of switching off a
given BS on the network performance. Specifically, switching
off a given BS results in additional load increments into the
neighboring BSs. Besides, switching off a BS can result in
a positive impact on the neighboring BSs due to a reduced
inter-cell interference. By quantifying the two aforementioned
effects, the network-impact criterion modifies the switching off
decision as a BS selection problem, aiming at finding the BSs
that when switched off leads to the highest network-impact
[63]. In addition, an important problem associated with BS
on-off switching is related to coverage holes. Hence, another
decision metric is related to avoiding coverage holes. In [64],
it is shown that finding the optimal set of BSs that minimizes
the network power consumption while avoiding coverage holes
is closely related to the minimum-weight disk cover problem,
which is known to be an NP-hard problem and hence a greedy
algorithm is proposed to switch off BSs while maintaining
network coverage in polynomial time complexity. To assess the
performance of different MT association mechanisms, queue-
ing models are used [55]. Specifically, the MT association
process in the overlapped coverage of different BSs is modeled
as a customer joining a queue with V=|B||M| servers,
where |B| and |M| denote the number of BSs with overlapped
coverage and the maximum number of MTs accommodated in
each BS, respectively. Consider a two-BS scenario with three
service areas. In service areas 1 and 2, an MT is served by the
BS covering that area. In service area 3, an MT can be served
by either of BSs with overlapped coverage. A BS is switched
off and hence its corresponding |M| servers are shut down, if
no MT is assigned to it. Using the queueing model, analytical
expressions are derived for call blocking probability, average
number of MTs assigned to each BS, and average power and
energy consumed by the network operator to serve one MT
[55]. The model can be approximated to account for the case
with multiple-BS overlapped coverage.
Based on the MT association phase, the BS operation
decision is specified. Hence, BSs with a concentrated call
traffic load become active, while light loaded BSs are switched
off. The BS operation problem deals with three concerns,
namely accommodating future traffic demands, determining
BS wake-up instants for switched off BSs, and finally how
to implement the BS on-off switching decisions. For the first
concern, it should be noted that the BS operation decision lasts
for a long duration (i.e., several hours), as frequent BS on
and off switching is not desirable due to the increased energy
consumption in the BS start-up phase [11] and the unavailable
service for the off cells during the decision computation phase
[51]. As a result, the BS operation decision should address the
future call traffic load either by reserving some resources to
account for the future demands [51] or by exploiting the histor-
ical call traffic load pattern [11]. In [65], an online stochastic
game theoretic algorithm is proposed, where neighboring BSs
communicate with each other to predict their traffic profiles,
which eventually will lead to optimal switching decisions and
result in minimum network energy consumption. As for the
second concern dealt with in the BS operation problem, it
should be noted that switching off some cells is executed given
that active BSs extend their coverage areas to provide service
for the cells with inactive BSs. As the call traffic load of the
inactive cells increases beyond the capacity limitation of the
active BSs, some of the inactive BSs are switched on. Hence,
12
(a) M-based scheme (b) V-based scheme: single vacation
(c) V-based scheme: multiple vacation
Fig. 2. BS wake-up schemes [66]. In the M-based scheme, the BS is switched off in an idle condition and it wakes up when Musers arrive at the system.
On the other hand, for the V-based scheme, the BS remains sleep for a period of vacation time before waking up. In a single vacation case, the BS remains
awake after the vacation period even if there is no call request to serve, while in a multiple vacation case, the BS goes back to sleep if it wakes up and finds
no call request to serve.
in addition to specifying which BSs to be switched off, another
equally important research direction aims to determine the
wake up instants for switched off BSs. For instance, two wake
up schemes are presented in [66], namely number M-based
and vacation time V-based schemes, as shown in Figure 2. One
limitation with the M-based scheme is the requirement that
the BS needs to continuously monitor the user request arrivals,
which translates into an advantage for the V-based scheme.
For femto-cell BSs with overlapped coverage with macro-cell
BSs, three wake up modes are presented in literature, namely
BS controlled, MT controlled, and network controlled modes
[67]. In the BS controlled mode, the femto-cell BS performs
continuous sensing for user activity for wake up, while in
the MT controlled mode, the MT sends wake up messages
for a sleeping femto BS. Finally, in the network controlled
mode, the core network controls the femto BS operation
through wake up messages over the backhaul link. The three
different modes of operation yield different performance in
terms of BS and MT energy consumption and signaling
overhead. Specifically, the BS controlled mode results in less
energy saving for the BS, the MT controlled mode increases
energy consumption for the MT, and the network controlled
mode results in additional signalling overhead [67]. MDP-
based optimal wake up schemes are proposed in [68] for
network operated femto BSs overlapping with a macro BS.
To wake up the right femto BSs, which serve the extra traffic
demand and still result in efficient energy usage, information
regarding call traffic load and user localization within the
macro cell is required. In absence of the traffic localization
information, the femto BS wake up problem can be formulated
as a partially observable MDP [68]. The last issue dealt with
in the BS operation problem deals with switching off mode
entrance and exit, which are two important design stages in
implementing the BS operation decision [69]. The switching
off mode entrance stage should specify how the transition
from the on state to the off state is implemented. If a BS
is switched off very fast, the corresponding MTs may not be
able to execute successfully their handover procedures and
their calls will be dropped. One reason is a strong received
signal from the BS that the MT is associated with, which
prevents the MT from hearing signals from nearby BSs. Hence,
if the BS that an MT is connected to is suddenly switched
off, the MT will not be able to synchronize and associate with
another active BS. Another reason is the maximum number of
handovers that can occur simultaneously towards a new BS,
due to the limited signaling channel capacity. As a result, a
progressive switching off operation can be used, an operation
that is referred to as BS wilting [69], as shown in Figure 3a.
During this process, the MTs associated with the wilting BS
initiate a handover process to the neighboring BSs and the BS
switching off operation is suspended if the handover process
of MTs is unsuccessful. On the other hand, the switching off
mode exit specifies how the transition from the off state to
the on state is implemented. A BS that is switched on too fast
can generate a strong interference to MTs in service. Hence,
a progressive switch on process can be used, and such an
approach is referred to as BS blossoming [69], as shown in
Figure 3b.
B. MT Radio Interface On-Off Switching
Similar to BS on-off switching, an MT with a low or bursty
traffic load can switch off from time to time its radio interface
to save energy. Designing an appropriate on-off switching
schedule for the MT radio interface varies according to
whether the MT establishes communications on the downlink
[48], [70] - [74], uplink [58], [75], or both links [76].
13
Fig. 4. Modeling of MT on-off switching as a server with repeated vacations [48]. The model is similar to the BS V-based scheme with multiple vacations.
BS transmission power (W)
t
(a) BS wilting
BS transmission power (W)
t
(b) BS blossoming
Fig. 3. BS switching off mode entrance and exit [69]. In BS wilting, the BS
transmit power is progressively halved until the BS is off. In BS blossoming,
the BS transmit power is progressively doubled until the BS is on.
For downlink communications, the MT radio interface on-
off switching mechanism can be classified into two categories
based on whether it implements a traffic shaping technique
or not [48], [70] - [74]. In the absence of traffic shaping
techniques, an MT (with a low or bursty traffic load) switches
off its radio interface if no data packets are available for the
MT at the serving BS. Hence, the MT establishes a switching
on-off schedule that specifies the switching off intervals and
switching on instants. At a switching on instant, the MT listens
to its serving BS to check if there are any packets available
for it. If no packets are available, the MT assumes a switching
off interval; otherwise, the MT keeps its radio interface active
to receive the available packets. During the MT switch off
interval, the incoming packets are buffered at the BS until the
MT next switch on instant. While a long switch off interval
can enhance the energy savings for the MT, it increases the
buffering delay of the packet at the BS until it is received by
the MT. In addition, incoming data packets for the MT may be
discarded in case of a buffer overflow at the BS. Furthermore,
unnecessarily switching on the MT to check packet availability
at the BS buffer results in MT energy losses. Hence, the main
research challenge in this case is how to design a switching
schedule for the MT radio interface that maximizes its energy
saving while reducing the buffering delay of packets available
at the BS. One approach is to model the MT radio interface as
a server that assumes repeated vacations [48], [70], as shown
in Figure 4. Hence, analytical expressions can be derived
for the expected number of switching off intervals until a
packet is available for the MT at the BS. Using the analytical
expressions, myopic optimization problems can be formulated
to minimize the MT energy consumption rate while achieving
acceptable performance in terms of the message response
time, which is defined as the time interval from arrival time
of an arbitrary message at the BS to the time it leaves the
system (BS) after service completion [48]. Besides myopic
optimization techniques, dynamic programming can be used to
design a switching off schedule that minimizes a cost function
consisting of a weighted sum of the energy consumed for radio
interface on-off switching and a target performance metric
(e.g., the buffering delay at the BS for the MT when switched
off) [70]. In addition to queueing models coupled with myopic
and dynamic optimization techniques, a Llyod-max algorithm
can be used to design a schedule that specifies the switching on
instants for the MT radio interface [71]. One limitation with
the aforementioned works is that, if the packet inter-arrival
time of the application is too small, the MT cannot switch off
its radio interface to provide acceptable QoS performance. In
addition, the MT consumes a significant amount of energy to
switch on its radio interface. Further, every time the MT finds a
single packet available at the BS buffer, an interruption signal
is triggered by the MT radio interface to activate the MT data
bus and central processing unit (CPU). If the MT experiences
frequent interrupts, it will not be able to enter a deep sleep state
and only a small amount of energy will be saved. Therefore,
traffic shaping techniques are introduced to enable a longer
14
idle duration for an MT2. Such a traffic shaping technique can
be implemented by the MT itself, where the MT buffers the
incoming data packets for a short period at its radio interface,
without activating its data bus and CPU, and then releases the
data packets as a burst so as to reduce the interruption trigger
events and to save more energy [72]. For transmission control
protocol (TCP) applications, an alternative approach can be
triggered by the MT, where the MT forces the BS to send
data packets in bursts and can enjoy a longer idle duration by
announcing a zero congestion window size. Hence, the data
packets are buffered at the BS for a longer period, until the
MT announces an appropriate window size to allow the BS
to release data packets in bursts [73]. While the above traffic
shaping research deals with a single-user scenario, the main
objective in a multi-user environment is to schedule the on-off
switching of radio interfaces for different MTs so as to satisfy
their target QoS and save energy by switching off the MT radio
interfaces for a long enough time [74]. An MT stores sufficient
data at its buffer to satisfy its QoS and switches off its radio
interface to save energy while the BS serves another MT. The
MT switches on its radio interface only when no sufficient
data is available at its buffer to satisfy the QoS requirement.
For uplink communications, besides adapting the physical
layer parameters such as controlling the transmission power
and modulation/coding schemes, an MT can switch on and
off its radio interface to further save energy. In [75], it is
shown that different parameters such as the packet arrival rate
and packet delay constraint affect the practicality of adopting
such a switching approach. Specifically, it is practical to
employ an on-off switching mechanism for energy saving at
MTs for small packet arrival rate and/or large packet delay
constraint. In such scenarios, the research challenge is how
to jointly adapt the power control, modulation and coding
schemes, and switching on and off the MT radio interface
to save energy in presence of stochastic traffic and channel
conditions (i.e., no a-priori knowledge of traffic arrivals and
channel conditions). In this case, an MT can choose to switch
off its radio interface and hold data packets in its buffer to
transmit them as a burst in better channel conditions. Besides
saving energy, the transmission mechanism should satisfy the
target QoS in terms of data packet delay and should avoid an
overflow event at the MT buffer. A Markov decision process
(MDP) problem can be formulated to control the data packet
transmission throughput (and hence the amount of buffered
data packets), the achieved bit error probability, and the MT
radio interface state (switch on or off) so as to balance energy
saving with QoS guarantee (i.e., minimizing data packet delay
and avoiding buffer overflow) [58].
A general model for MT radio interface on-off switching
is captured in the context of bi-directional communications
[76]. In such a scenario, no BS buffering delay is experienced
by incoming downlink traffic during uplink transmission, as
the MT radio interface is already switched on. Hence, a
finite general Markovian background process can be used
to model the uplink activity and downlink traffic so as to
2The idle duration in this context represents the interval during which an
MT is not receiving any data packets.
derive analytical expressions for the buffer occupancy and
downlink packet delay statistics [76]. Such expressions can be
useful in developing an efficient on-off switching mechanism
for the MT radio interface for both uplink and downlink
communications.
C. Discussion
Based on the above review, BS on-off switching aims to
exploit spatial and temporal fluctuations in call traffic load to
achieve energy saving. As a result, using static call traffic mod-
els for switching schedule design (i.e., to determine decisions
on switch-off and wake-up instants) and/or performance evalu-
ation, as in [64], is not realistic. Instead, the call traffic models
should reflect a joint spatial and long-term temporal fluctuation
behavior, as in [11] and [62]. Traffic models that capture joint
spatial and short-term temporal call-level fluctuations, such as
[37] and [51], are not capable of assessing the daily switching
schedule performance due to a time varying traffic demand.
Furthermore, traffic models that capture only long-term (as
in [62] and [65]) or short-term (as in [55]) temporal call-level
fluctuations fail to exploit the spatial dimension of the problem
and stand unrealistic for performance evaluation in large-scale
networks with multiple BS sites. For BS power consumption
models, both static and dynamic components, as in (7) and (9),
should be accounted for, which is the case for the algorithms
developed in [11], [37], and [64]. Power consumption models
under the assumption of constant transmission power, as in
[51], [55], [63], and [65], neglect the scaling of transmission
power with the call traffic load, which is unrealistic. Overall,
the reported solutions in Section IV.A. aim to minimize the
network energy consumption, which is somehow similar in
concept to maximizing the energy consumption gain given
in (14). However, such an expression does not assess the
network gain (in terms of transmitted power as in (15) or
network coverage as in (16)) versus the incurred cost (in
terms of the network consumed power). The reported solutions
minimize the network energy consumption while satisfying a
target performance metric. For BS on-off switching solutions,
the target performance metrics are based on admission quality
requirements, as in [11], [51], [55], [64]. Few works account
for application quality requirements, as in [37]. In practice, a
solution should satisfy both admission and application quality
requirements, as in [62] and [63], to better serve the users
required QoS.
On the other hand, MTs can save energy by switching
off their radio interfaces during idle periods of bursty traffic.
Hence, static traffic models for a fixed number of backlogged
data packets ready for transmission, as in [75], are not re-
alistic to determine the MT idle period and hence will not
help in developing practical sleep schedules for the MTs.
Instead, practical traffic models should capture the packet-
level short-term temporal fluctuations, as in [48], [58], [70],
[71], [73], [74], and [76]. While some solutions account for
both active and idle power consumption values, as in [48],
[70], [71], and [73], and reception power consumption, as
in [74] and [76], these solutions do not include the circuit
power consumption component of the MTs. Both transmission
15
and circuit power consumptions should be accounted for as
in [58] and [75]. However, such models assume fixed circuit
power consumption and neglect the dynamic circuit power
component as given in (12) and (13). The reported solutions
in Section IV.B. minimize the MT energy consumption while
satisfying application quality performance metrics. However,
such a modeling approach overlooks the network capacity
limitations, e.g., in terms of available bandwidth, which may
lead to call blocking. Hence, the proposed solutions should aim
to satisfy both application and admission quality requirements.
V. GR EE N SOLUTIONS AND ANALYTICAL MODELS AT
HIG H AN D/O R CONTINUOUS CALL TRA FFIC LOAD
Energy efficient scheduling techniques are adopted to sat-
isfy the required QoS at reduced energy consumption when
switching on-off techniques are infeasible due to a high and/or
continuous call traffic load. Various scheduling techniques are
proposed in literature for network operators and mobile users,
and can be divided into four categories, as shown in Table VIII.
The categories include scheduling for single-network access,
multi-homing access, small size cells, and scheduling with
different energy supplies, which are discussed next in more
details.
A. Scheduling for Single-network Access
In this case, a mobile user connects to a single wireless
access network at a time. Two system models are adopted in
literature for single-network access. The first model assumes
that a single network covers a given geographical region,
which can be referred to as a homogenous wireless medium.
The second model deals with the availability of multiple
networks with overlapped coverage in the geographical region,
which is referred to as a heterogeneous wireless medium.
For the homogeneous wireless medium, the network operator
aims to assign its resources to MTs so as to reduce the
total power consumption of its BSs. Such an objective can
be achieved by minimizing the transmission power while
providing acceptable QoS performance, a technique that is
referred to as margin adaptive strategy [9]. An approach
to implement the margin adaptive strategy is via a score-
based scheduler. For instance, in an OFDMA system, the BS
calculates a score for every resource block qto be assigned
to MT m[77]. The calculated score ensures that the BS
would consume the least transmission power by assigning
resource block qto MT m. Moreover, the score promotes fair
resource allocation among MTs, as a penalty function can be
included based on the number of already allocated resource
blocks for MT m. A low score indicates a more desirable
resource block. Fairness consideration is also investigated in
[78] following a proportional rate constraint, which ensures
that each user eventually obtains a specific proportion of the
system throughput. Admission control policies can also be
employed to implement a margin adaptive strategy, where
a new session is admitted into the system as long as the
sub-frame energy in an OFDMA-based BS is kept below a
certain threshold [79]. Moreover, a margin adaptive strategy
can be implemented through a discrete rate adaptation policy
that controls the transmission rate and power according to
channel conditions, so as to maximize the achieved energy
efficiency while satisfying a bit-error-rate constraint [80].
Similarly, a channel driven rate and power adaptation strategy
can be achieved by jointly adapting modulation and coding
schemes (MCS) and transmission power to optimize the trade-
off between goodput and energy efficiency [81]. In addition,
a margin adaptive strategy can be implemented by scheduling
resources among MTs based on their traffic delay tolerance
[54]. Specifically, delay tolerant traffic (e.g., video and data)
can be served in an opportunistic way during periods of good
channel conditions (i.e., soft real time service). One limitation
with the margin adaptive strategy is the requirement of CSI to
allocate the transmitted power, which requires using pilot sym-
bols. These pilot symbols will incur some energy consumption.
Two approaches can be used for pilot energy assignment [9],
namely constant single pilot energy and constant total pilot
energy. In the former approach, each pilot keeps the same
energy level independent of the number of pilot symbols.
Hence, the larger the number of pilot symbols is, the more
accurate CSI is available, yet the higher energy consumption
is. The later approach assigns a constant energy to all pilots,
resulting in reduced energy per pilot for a larger number of
pilots, which can lead to inaccurate CSI. On the other hand,
in a heterogeneous wireless medium, energy can be saved by
assigning MTs to the BSs that reduce energy consumption for
a set of operators with BSs of overlapped coverage [49]. In
addition, each BS in such a heterogeneous environment may
choose between two modes of operation, i.e., point-to-point or
point-to-multi-point. Hence, the problem can be decomposed
into two sub-problems, one for BS selection and the other for
BS mode choice. While the work in [49] controls transmission
power only through BS operation mode selection, a joint BS
selection and power control mechanism is proposed in [82],
which aims to associate MTs to BSs with overlapped coverage
while minimizing the BS transmission powers to reduce the
interference among different links. Furthermore, offloading
techniques can be adopted to enhance energy efficiency in a
heterogeneous wireless medium. Specifically, through mobility
prediction and using the pre-fetching feature, data traffic can
be offloaded from cellular networks to WiFi hotspots or femto-
cells [83]. Hence, delay tolerant traffic can be downloaded
when mobile users are close to the WiFi access point or femto-
cell instead of using the macro-cell [84]. Overall, offloading
can be either network or user driven [85]. Various factors affect
the energy efficiency performance in terms of user mobility,
backhaul throughput, data size, and WiFi and/or femto-cell
densities [83].
Similarly, MTs can save energy by appropriate resource
scheduling on the uplink, based on the network multiple
access scheme. Various energy efficient mechanisms are pro-
posed for OFDMA-based networks [33], [34], [43], [57].
The mechanisms mainly enhance energy efficiency through
subcarrier allocation, power control, and joint subcarrier al-
location and power control [43]. Both centralized and de-
centralized architectures can be adopted to implement the
mechanisms [33], [34]. In a centralized architecture, the BS
in each cell jointly performs subcarrier allocation, modulation
16
TABLE VIII
ASUMMARY OF GREEN SOLUTIONS AND ANALYTICAL MODELS AT HIGH AND/OR CO NTI NUO US C ALL T RA FFIC LO AD.
Solution/Analytical Model
Comments
References
Single-
network
BS
Margin adaptive strategy
It minimizes the
transmission power while
providing an acceptable
QoS.
[9], [54],
[77] -
[81]
User association in
Heterogeneous wireless
medium
It assigns the MTs to the
BSs, with coverage overlap,
that reduce energy
consumption for a set of
operators.
[49],
[82] –
[85]
MT
OFDMA
network
Sub-carrier
allocation
It enhances energy
efficiency through joint sub-
carrier allocation and power
control.
[33], [34], [43]
Carrier
aggregation
It employs both PCC and
SCC carrier components for
energy saving.
[57]
TDMA network
The MT energy efficiency is
maximized in TDMA
through opportunistic
transmission.
[59]
Multi-
homing
BS
Network Cooperation
The MT receives required
data rate from multiple BSs
simultaneously. The BSs
coordinate their transmitted
power for energy saving.
[50], [53]
MT
BS selection and power
allocation
The MT specifies a set of
BSs for uplink transmission
and determines the allocated
transmission power for each
radio interface.
[20]
, [46]
Small Cells
It divides the cell into
several tiers of smaller cells
to reduce transmission range
for BSs and MTs.
[36],
[56], [
86],
[87]
Multiple
Energy
Sources
BS
Multiple retailers
The network operator
decides how much
electricity to procure from
each retailer.
[23], [88
]
On-
grid and green
energy sources
The objective is to
maximize the utilization of
green energy and saves the
on-grid energy.
[52]
Complementary
renewable sources
The BSs are powered using
only renewable sources.
[89], [90],
[92] – [97]
MT
Multiple batteries
It employs the recovery
effect of batteries.
[98]
17
order adaptation, and power control for MTs. In a distributed
mechanism, given a subcarrier assignment, an MT adjusts
its modulation order and transmission power to optimize its
own energy efficiency. In a multi-cell environment, multi-cell
interference should be taken into account via energy efficient
uplink resource allocation scheduling [34], [43]. In addition
to subcarrier allocation and power control, energy efficiency
is maximized for OFDMA-based networks through dynamic
carrier aggregation [57]. In general, while an MT served by
all carrier components will enjoy an enhanced throughput,
its energy consumption also increases. Following a dynamic
carrier aggregation technique, an MT is assigned to the queue
of a given carrier component, which is referred to as primary
carrier component (PCC). Whenever the queue of a carrier
component is empty, it helps other carrier components through
aggregation and therefore, it is referred to as supplementary
carrier component (SCC). Two mechanisms can be adopted
for SCC assignment [57]. The first mechanism aggregates all
SCCs to support the PCC with the longest queue. The second
mechanism orders PCCs according to queue length and SCCs
are circularly allocated to the ordered PCCs in a round-robin
fashion. For time division multiple access (TDMA)-based
networks, energy efficiency is maximized for a set of MTs
by opportunistic transmission [59]. Specifically, a scheduler
is designed at the BS to select an MT for transmission and
to determine its transmission rate. The problem complexity
is reduced by decomposing it into two sub-problems. The
first is a user scheduling sub-problem which selects an MT
opportunistically for transmission, based on channel conditions
and backlog information. The second sub-problem determines
the transmission rate for the selected MT to minimize the
transmission power by transmitting packets in queue such that
the average delay constraint is satisfied with equality.
B. Scheduling for Multi-homing Access
Recently, the wireless communication medium has become
a heterogeneous environment with overlapped coverage due to
different networks. In such a networking environment, MTs
are equipped with multiple radio interfaces. Through multi-
homing capability, an MT can maintain multiple simultaneous
associations with different networks. Besides enhancing the
achieved data rate through bandwidth aggregation, multi-
homing service can enhance energy efficiency for network
operators and mobile users. This is because an MT experiences
different channel conditions and bandwidth capabilities over
its different radio interfaces.
Different network operators can reduce the transmission
power of their BSs by supporting multi-homing services. The
motivation behind employing multi-homing to enhance energy
efficiency can be explained using the power-rate curve, which
can be divided into two regions [50]. In the first region,
power consumption increases slowly with the growth of data
rate, while in the second region power consumption increases
dramatically with data rate. Hence, a multi-homing threshold,
Rb, of data rate can be determined to start multi-homing trans-
mission if the required data rate is larger than Rb[50]. The
multi-homing threshold is based on the ratio of channel gain
between the MT and BSs of different networks. In addition, the
optimal transmission data rate from each BS can be specified
to maximize the energy efficiency of the networks. Moreover,
cooperating BSs can control their transmission power using a
semi-Markov decision process (SMDP) to minimize the total
BS power consumption under a target QoS constraint at the
MTs [53].
Similarly, MTs can enhance their energy efficiency through
multi-homing service. In this case, an MT determines which
and how many BSs will be selected for multi-homing, based on
the required data rate and the channel parameters of available
BSs [20]. To reduce the complexity, the problem can be
decomposed into two sub-problems. The first sub-problem
specifies which BSs will be selected for multi-homing and
the second sub-problem determines the optimal transmission
rate from each selected BS. For a constant data rate service,
energy efficiency maximization is equivalent to MT total
power consumption minimization. Different from [20], the
work in [46] deals with energy efficiency maximization for a
variable data rate using multi-homing service through power
allocation.
C. Scheduling With Small Size Cells
A small cell has a radio coverage of tens to a few hundreds
of meters (e.g., pico and femto cells) [36]. As a result,
the division of a macro-cell into several tiers of smaller
cells replaces a long range transmission with a short range
transmission due to the close proximity between small cell BSs
and MTs [86]. It is expected that the power consumption of a
small cell will be approximately 5watts by 2020 [87]. Hence,
an improved energy efficiency can be achieved. In [86], an
expression of the possible power gain G(J)resulting from cell
splitting into Jsmaller cells is provided. It is shown that, for
an ideal free space propagation channel model, the achieved
gain satisfies G(J)<1, and hence cell splitting should not be
implemented. On the other hand, in a non-ideal propagation
environment, G(J)>1and it increases with the number
Jof small cells3. Different configurations are presented in
literature for small cell deployment, as shown in Figure 5.
It is shown in [87] that the cell-on-edge deployment results
in a significant reduction in network energy consumption, as
compared to the uniformly distributed configuration, due to
lower transmit power for cell edge users.
The main challenge of cell splitting is the associated inter-
tier interference. This is mainly due to the limited radio
resources. Hence, the radio resources of the macro-BS are
shared among the small cells. Multi-cell processing can be
employed to mitigate interference [86]. Hence, multiple BSs
within a cluster exchange CSI and users’ data to support MTs
and eliminate interference. Based on the gathered information,
beam-forming techniques are used to minimize the total trans-
mit power while satisfying a certain signal-to-interference plus
3The BS power consumption model in [86] does not capture the BS
embodied energy as in [16]. When BS embodied energy is considered, there
is a limit on the number of small cells that can be included to enhance energy
efficiency.
18
Cell-on-edge
(a) Cell-on-edge deployment
Small cell
(b) Uniformly distributed deployment
Fig. 5. Configurations for small cell deployment [87]. Cell-on-edge deploy-
ment distributes the small cells around the edge of a macro cell. On the other
hand, in uniform distribution, small cells are uniformly distributed across the
macro cell.
noise ratio (SINR) for different MTs. In addition to multi-
cell processing (and in presence of both co-tier and cross-
tier interference), admission control with QoS guarantee can
play a vital role in mitigating interference, where a joint
resource allocation mechanism can be employed among multi-
tier networks [56] .
D. Scheduling With Multiple Energy Sources
Various scheduling techniques have been proposed to deal
with the presence of multiple energy sources [23], [52], [88]
- [98]. The main objective in these works is to jointly control
transmission power and select the energy source that mini-
mizes the total energy consumption. For network operators,
multiple energy sources deal with the availability of different
electricity retailers [23], [88], on-grid and green (renewable)
energy [52], and different (complementary) renewable sources
[89] - [97]. For MTs, multiple energy sources deal with the
availability of multiple batteries [98].
In an electricity market liberalization model, electricity re-
tailers compete with each other and aim to achieve the highest
individual profits by adjusting the electricity price offered to
users in different regions [23]. Electricity prices offered by
different retailers change frequently to reflect variations in the
cost of energy supply, which is referred to as real time pricing.
Given a set of electricity retailers, a Stackelberg game can be
formulated, where each retailer provides its real time price,
to maximize its profit, to the network operator which decides
how much electricity to procure from each retailer to power
on its BSs and achieve the lowest call blocking with the least
cost [23]. In [88], the optimal amount of energy to be procured
from each retailer is determined using evolutionary algorithms
(Genetic Algorithm and Particle Swarm Optimization), which
due to the random nature of the evolution process is shown
to outperform the deterministic algorithm developed in [23].
In addition to the presence of multiple electricity retailers,
it is argued that the BSs of future cellular networks will be
powered by both on-grid and green (renewable) energy (e.g.,
solar energy) [52]. With such a hybrid energy system, the
objective is to optimize energy utilization in such networks
by maximizing the utilization of green energy and saving on-
grid energy. Network designers are faced with two central
issues [52]: 1) how to optimize the green energy usage at
different time slots to accommodate the temporal dynamics
of the green (solar) energy generation and the mobile traffic,
and 2) how to accommodate the spatial dynamics of the mobile
traffic with the objective of maximizing the utilization of green
energy by balancing the green energy consumption among
BSs through cell size adjustment. While the aforementioned
works deal with the presence of on-grid energy, the long
term objective is to power BSs in appropriate locations using
only a combination of complementary renewable sources (e.g.,
wind in winter and solar in summer) [89]. Furthermore, power
cooperation enables different BSs (networks) to share (trade)
their green power with each other whenever possible for a
sustainable and energy efficient network operation [90]. In uti-
lizing renewable energy sources, renewable energy generation
and storage should be investigated. Since renewable energy
sources are intermittent, energy storage is used to address this
limitation. Hence, the harvested energy is stored in a battery
with finite capacity before it is used for transmission [91],
[92]. In this context, the energy replenishment process and
the storage constraints of the rechargeable batteries need to
be taken into account while designing efficient transmission
strategies [93]. Two constraints should be accounted for at the
energy harvesting battery [94]. The first ensures that the energy
drawn from the battery is at most equal to the energy stored
at the battery, which is referred to as the causality constraint.
The second constraint ensures that the energy level at the
battery does not exceed a maximum level to avoid battery
energy overflow. Hence, storage sizing is very important to
guarantee a sustainable energy at a reduced cost. In addition,
BSs have to adapt their data transmission to the availability
of energy at a particular instant [95], [96]. Therefore, more
studies are needed to minimize the overall power consumption
of BSs, through on-off switching at a low call traffic load or
scheduling and node cooperation [97] at a high call traffic
load, to reduce the required generation potential and storage
capacity. A very important aspect of green communications
is to consider the environmental dimension of the proposed
solution. Hence, while selecting an appropriate energy source
(i.e., electricity retailer and/or renewable energy source), it is
necessary to guarantee that the CO2emission cost is below a
target level. The CO2emission cost, in kg/hr, related to the
19
Fig. 6. A green hybrid solution. The hybrid solution uses a combination of renewable and grid energy sources. Complementary renewable energy sources
can be used. If the power grid is absent, i.e., the BS is not connected to the power grid and hence controller 2 does not exist, the BS is powered only through
renewable sources. An energy harvesting battery is used to overcome the intermittent nature of renewable energy sources.
BS power consumption Pbcan be expressed as [23]
I(Pb) = αP 2
b+βPb,(27)
where αand βare constants that depend on the pollutant level
of the electricity retailer.
For MTs, under a pulsed discharge profile, the battery is able
to recover some charges during the interruptions of the drained
current (i.e., no transmission periods). Hence, an improved
battery performance can be achieved. This phenomena is
referred to as the recovery effect. To promote the recovery
effect and enhance the battery performance, a package of
multiple batteries can be used and a scheduling policy can
be developed to efficiently distribute the discharge demand
among the multiple batteries connected in parallel [98].
E. Discussion
The majority of research works that investigate green com-
munication solutions at a high traffic load employ static traffic
models for resource scheduling and performance evaluation,
as in [9], [23], [33], [34], [43], [49], [50], [77], [78], [82],
[86], and [88]. Very few works use traffic models that reflect
long-term (as in [79] and [87]) or short-term (as in [54],
[56], and [57] for call-level and [59] for packet-level) tem-
poral fluctuations. Also, few works use traffic models that
capture spatial fluctuations in traffic load, as in [52] and [53].
Spatial and temporal traffic models should be employed for
performance evaluation of green resource scheduling solutions.
Spatial traffic models are useful in evaluating the algorithm
performance in large-scale networks, while temporal models
are important to investigate the associated signaling overhead,
which may jeopardize the energy saving benefits, if high
overhead is expected. In addition, many works account only
for transmission power consumption as in [9], [43], [49], [50],
[54], [56], [57], [79], [82], [86], and [87]. Both transmission
and circuit power consumption should be accounted for, as in
[23], [33], [34], [52], [53], [59], [77], [78], and [88]. However,
the aforementioned models do not account for dynamic circuit
power consumption, as in (12) and (13). Also, BS transmission
power consumption should scale with the traffic load as
expressed in (7) and (9). Furthermore, for small-cell and multi-
tier deployment, both operation and embodied energy should
be accounted for as in (10). Accounting only for operation
power consumption in such scenarios can be misleading.
While some works aim to minimize the energy consumption,
the work in [77] is to maximize an energy consumption gain
expression similar to (14). Moreover, the works in [33], [34],
[43], [50], [57], and [78] aim to maximize an energy efficiency
expression similar to (17), (18), or (19). Such an expression
provides a better indication of the performance in terms of
the achieved gain (in terms of resulting data rate) versus the
incurred cost (in terms of the energy consumed). Almost all
reported solutions aim to minimize energy consumption or
maximize energy efficiency, while maintaining a satisfactory
performance in terms of application quality requirements. The
works in [23] and [88] target admission quality requirements.
In practice, an effective solution should satisfy both admission
and application quality requirements, as in [56].
VI. FU TU RE RE SE AR CH
The existing research works mainly focus on enhancing
energy efficiency either of network operators or mobile users.
However, a green solution implemented at the network opera-
tor side can lead to high energy consumption at the mobile user
side, and vice-versa. Hence, green solutions should capture the
trade-off in energy efficiency among network operators and
mobile users and should be jointly designed to balance such
a trade-off.
For instance, the BS on-off switching mechanism involves
two phases, namely user association and BS operation. Fo-
cusing only on energy efficiency of the network operator,
a BS on-off mechanism can lead to an energy inefficient
user association from the mobile user perspective. Specifically,
it can lead to MTs being associated in the uplink with a
far away BS in order to switch off a nearby BS. This will
result in energy depletion for the MTs and hence dropped
services. Thus, a BS on-off switching mechanism should
capture the trade-off in the achieved energy efficiency for
the network operator and mobile users, and should aim at
20
balancing them. MTs should be associated with BSs that can
balance energy saving for both network operators and mobile
users. The existing research, however, focuses on balancing
energy consumption performance of a BS with the flow-level
performance at the MT, e.g., [37]. Instead, the multi-objective
function in [37] should aim to balance energy saving for BSs
and MTs while satisfying the MT required QoS. As a result,
BS switching off decision criteria in literature, such as user-
BS distance [62], call traffic load [51], network impact [63],
and network coverage holes [64], should be revised. Specif-
ically, the switching off criterion should include, besides the
aforementioned metrics, an MT energy consumption metric.
Similarly, the existing mechanisms present only the call traffic
load as a wake up criterion [68]. The switching mechanisms
should capture the degradation in energy consumption for MTs
and include it as a BS wake up decision metric. Furthermore,
MTs suffer from inter-cell interference. An uplink scheduling
scheme at MTs performs power allocation while dealing with
the inter-cell interference negative effect. However, inter-cell
interference can be affected by the BS on-off switching deci-
sion. Such a dependence can be modeled in the user received
SINR using a BS activity parameter, which equals to one if
the BS is on, and zero otherwise. In addition, the BS on-off
switching decision should promote energy saving at MTs by
switching off cells that result in the highest interference during
a low call traffic load condition. Moreover, the analytical
models used in literature, e.g., the queuing model in [55],
mainly assess the network energy saving performance for a
given mechanism. Such models should be extended to assess
the energy saving performance for both network operators and
mobile users.
Similarly, the existing MT radio interface on-off switching
mechanisms focus mainly on the energy saving performance
at the MT without capturing the impact of the energy saving
mechanisms implemented at the BSs. Specifically, the down-
link mechanisms allow an MT to switch off its radio interface
for a given interval while dealing with only the buffer delay
and/or overflow at the BS, e.g., [48], [70] - [74]. However,
the impact of BS on-off switching is not considered. If the
serving BS is switched off during the MT sleep interval, the
MT connection will be dropped and the buffered data will
be lost. Hence, the MT radio interface switching schedule
design needs to be revised. For instance, in [48], the MT
switching on is triggered upon a packet arrival at the BS.
Such a model should be extended to account for the BS
switching off decision as an additional switching on trigger
for the MT radio interface. Moreover, the existing switching
off design metrics focus on balancing energy consumption at
the MT with the buffer delay at the BS [70]. An extension
is required to account for the BS energy consumption due
to a delayed switching off decision for the BS while waiting
for the MT to wake up. Furthermore, network operators can
save energy at BSs by scheduling delay tolerant applications
(e.g., data and video) opportunistically in the presence of
good channel conditions. MT radio interface on-off scheduling
should take account of the delay at the BS due to both MT
inactivity and BS opportunistic scheduling of traffic. The radio
interface on-off scheduling at an MT and the opportunistic
traffic scheduling at the BS should balance energy efficiency
for both network operators and mobile users, while satisfying
the target performance metrics. Opportunistic scheduling can
also be used for energy saving at MTs. However, such an
approach does not always work in practical scenarios (e.g., a
stationary user suffering from a slow fading channel), which
is not the case for BSs due to spatial user diversity. For the
MT energy saving mechanisms at the uplink, power control
and radio interface on-off switching mechanisms account in
their design only for the channel and traffic dynamics [58]. In
addition, the BS on-off switching dynamics should be captured
while designing an energy saving mechanism.
Furthermore, the energy efficient resource scheduling mech-
anisms at a heterogeneous wireless medium assign MTs to the
BSs which reduce energy consumption for network operators
[49], [82]. Such mechanisms mainly deal with downlink re-
source scheduling. However, no investigation is performed for
MTs with bidirectional traffic, e.g., for video call applications.
In this case, two approaches can be implemented for energy
saving at both network operators and mobile users. The first
relies on single-network access, where the MT is associated
with the BS that balances energy saving for the network
operators and the mobile users. On the other hand, the second
approach employs multi-homing where the MT connects on
the uplink to the BS that promotes energy saving for the mo-
bile user while the MT connects on the downlink to the BS that
promotes energy saving for the network operators. Moreover,
the potentials of the heterogeneous wireless medium should
be better exploited to enhance energy saving. In addition, for
multi-homing service, as MTs connect to multiple networks
simultaneously, radio resources at different radio interfaces
can be properly scheduled to enhance energy efficiency. How-
ever, existing research works focus only on power allocation
schemes at the MT different radio interfaces to save energy in
different channel conditions. Given the bandwidth capabilities
of different networks, cross-layer designs that incorporate joint
bandwidth and power allocation can lead to an improved
energy efficiency.
In addition, the existing opportunistic scheduling mech-
anisms focus on energy saving for network operators [54]
or MTs [59]. However, for MTs with bidirectional traffic,
opportunistic scheduling should be implemented such that the
time slot for uplink and downlink transmission can balance
energy saving for both network operators and mobile users.
Finally, for radio resource scheduling in BSs powered by
renewable energy sources, the existing research focuses mainly
on downlink delay tolerant applications [52], [88]. Hence,
BSs aim to schedule data transmissions at time slots when
energy is available. However, when MT radio interface on-off
scheduling is implemented, the BSs need to account for the
MT sleep interval, which may conflict with the BS energy
limitation due to the finite size of the energy harvesting
buffer at the BS and might lead to buffer overflow. Hence,
the resource scheduling mechanism should balance energy
availability at the BS with energy saving at the MT.
21
VII. CONCLUSION
Due to environmental, financial, and QoE considerations,
there has been a great emphasis on the need to develop
energy efficient solutions for wireless communications and
networking. Such solutions are referred to as green solutions.
In this survey, we have reviewed the existing research activities
dedicated to green wireless communications and network-
ing solutions, from the network operator and mobile user
perspectives. The first step towards developing an effective
green solution is to identify the call traffic load condition,
based on which an appropriate definition of energy efficiency
can be proposed, as discussed in Section II. The second
step is to use proper models for power consumption and for
call traffic loads that account for both spatial and temporal
fluctuations, as discussed in Sections II and III, respectively.
Besides improving energy efficiency, certain performance met-
rics should be satisfied, as discussed in Section III, which
are determined based on the target application, e.g., voice,
data, or video services. Given the call traffic load condition,
different green solutions and analytical models can be adopted,
as presented in Sections IV and V. At a low call traffic load
condition, on-off switching of radio devices (e.g., BSs for
network operators and MT radio interfaces for mobile users)
can improve the performance of energy consumption. Radio
resource scheduling techniques have been proposed for a high
call traffic load condition. Despite the various efforts proposed
to analyze and design effective green solutions, many open
issues remain to be further investigated. As future research,
green solutions should capture the tradeoff in energy efficiency
among network operators and mobile users and should be
designed to balance such a trade-off.
VIII. ACK NOWLEDGMENT
This research work was financially supported by Natural
Sciences and Engineering Research Council (NSERC) of
Canada. In addition, parts of this paper, specifically Sections
I, IV, V, and VI, were made possible by NPRP grant NPRP 4-
1293-2-513 from the Qatar National Research Fund (a member
of Qatar Foundation). The statements made herein are solely
the responsibility of the authors.
REF ER EN CE S
[1] M. Ismail and W. Zhuang, “Cooperative networking in a heterogeneous
wireless medium,” Springer Briefs in Computer Science, New York, April
2013.
[2] T. Chen, Y. Yang, H. Zhang, H. Kim, and K. Horneman, “Network energy
saving technologies for green wireless access networks,” IEEE Wireless
Commun., vol. 8, no. 5, pp. 30-38, Oct. 2011.
[3] D. Cavalcanti, D. P, Agrawal, C. Cordeiro, B. Xie, and A. Kumar “Issues
in integrating cellular networks, WLANs, and MANETs: a futuristic
heterogeneous wireless network,” IEEE Wireless Commun. Magazine, vol.
12, no. 3, pp. 30-41, 2005.
[4] M. Ismail and W. Zhuang, “A distributed multi-service resource allocation
algorithm in heterogeneous wireless access medium,” IEEE J. Select.
Areas Commun., vol. 30, no. 2, pp. 425-432, Feb. 2012.
[5] M. Ismail, A. Abdrabou, and W. Zhuang, “Cooperative decentralized
resource allocation in heterogeneous wireless access medium,” IEEE
Trans. Wireless Commun., vol. 12, no. 2, pp. 714-724, Feb. 2013.
[6] M. Ismail and W. Zhuang, “Decentralized radio resource allocation for
single-network and multi-homing services in cooperative heterogeneous
wireless access medium,” IEEE Trans. Wireless Commun., vol. 11, no.
11, pp. 4085-4095, Nov. 2012.
[7] C. Han, T. Harrold, S. Armour, and I. Krikidis, “Green radio: radio
techniques to enable energy-efficient wireless networks,” IEEE Commun.
Magazine, vol. 49, no. 6, pp. 46-54, June 2011.
[8] Y. Chen, S. Zhang, S. Xu, and G. Y. Li, “Fundamental trade-offs on green
wireless networks,” IEEE Commun. Magazine, vol. 49, no. 6, pp. 30-37,
June 2011.
[9] H. Bogucka and A. Conti, “Degrees of freedom for energy savings in
practical adaptive wireless systems,” IEEE Commun. Magazine, vol. 49,
no. 6, pp. 38-45, June 2011.
[10] S. Mclaughlin, P. M. Grant, J. S. Thompson, and H. Haas, “Techniques
for improving cellular radio base station energy efficiency,” IEEE Wireless
Commun., vol. 18, no. 5, pp. 10-17, Oct. 2011.
[11] M. Ismail and W. Zhuang, “Network cooperation for energy saving in
green radio communications,” IEEE Wireless Commun., vol. 18, no. 5,
pp. 76-81, Oct. 2011.
[12] K. Pentikousis, “In search of energy-efficient mobile networking,” IEEE
Commun. Magazine, vol. 48, no. 1, pp. 95-103, Jan. 2010.
[13] G. Miao, “Energy-efficient uplink multi-user MIMO,” IEEE Trans.
Wireless Commun., vol. 12, no. 5, pp. 2302-2313, May 2013.
[14] E. Oh, B. Krishnamachari, X. Liu, and Z. Niu, “Toward dynamic energy-
efficient operation of cellular network infrastructure,” IEEE Commun.
Magazine, vol. 49, no. 6, pp. 56-61, June 2011.
[15] Y. S. Soh, T. Q. S. Quek, M. Kountouris, and H. Shin, “Energy efficient
heterogeneous cellular networks,” IEEE Journal Selected Areas Commun.,
vol. 31, no. 5, pp. 840-850, May 2013.
[16] I. Humar, X. Ge, L. Xiang, M. Jo, M. Chen, and J. Zhang, “Rethinking
energy efficiency models of cellular networks with embodied energy,”
IEEE Network, vol. 25, no. 2, pp. 40-49, April 2011.
[17] D. Feng, C. Jiang, G. Lim, L. J. Cimini, G. Feng, and G. Y. Li, “A survey
of energy-efficient wireless communications,” IEEE Commun. Surveys
Tutorials, vol. 15, no. 1, pp. 167-178, 2013.
[18] S. Tombaz, A. Vastberg, and J. Zander, “Energy- and cost-efficient ultra-
high-capacity wireless access,” IEEE Wireless Commun., vol. 18, no. 5,
pp. 18-24, Oct. 2011.
[19] L. M. Correia, D. Zeller, O. Blume, and D. Ferling, “Challenges and
enabling technologies for energy aware mobile radio networks,” IEEE
Commun. Magazine, vol. 48, no. 11, pp. 66-72, Nov. 2010.
[20] X. Ma, M. Sheng, Y. Zhang, and X. Wang, “Concurrent transmission
for energy efficiency in heterogeneous wireless networks,” IEEE Trans.
Wireless Commun., under review.
[21] G. Miao, N. Himayat, Y. Li, and A. Swami, “Cross-Layer optimization
for energy-efficient wireless communications: a survey,” Wiley J. Wireless
Commun. and Mobile Computing, vol. 9, pp. 529-542, March 2009.
[22] M. Ismail, W. Zhuang, and S. Elhedhli, “Energy and content aware
multi-homing video transmission in heterogeneous networks,” IEEE
Trans. Wireless Commun., vol. 12, no. 7, pp. 3600-3610, July 2013.
[23] S. Bu, F. R. Yu, Y. Cai, and X. P. Liu, “When the smart grid
meets energy-efficient communications: green wireless cellular networks
powered by the smart grid,” IEEE Trans. Wireless Commun., vol. 11, no.
8, pp. 3014-3024, Aug. 2012.
[24] A. P. Bianzino, C. Chaudet, D. Rossi, and J. L. Rougier, “A survey on
green networking research,” IEEE Commun. Surveys Tutorials, vol. 14,
no. 1, pp. 3-20, 2012.
[25] Y. Zhang, P. Chowdhury, M. Tornatore, and B. Mukherjee, “Energy ef-
ficiency in telecom optical networks,” IEEE Commun. Surveys Tutorials,
vol. 12, no. 4, pp. 441-458, 2010.
[26] R. Bolla, R. Bruschi, F. Davoli, and F. Cucchietti, “Energy efficiency in
the future Internet: a survey of existing approaches and trends in energy-
aware fixed network infrastructures,” IEEE Commun. Surveys Tutorials,
vol. 13, no. 2, pp. 223-244, 2011.
[27] J. B. Rao and A. O. Fapojuwo, “A survey of energy efficient resource
management techniques for multicell cellular networks,” IEEE Commun.
Surveys Tutorials, to appear.
[28] L. Suarez, L. Nuaymi, and J. M. Bonnin, “An overview and classification
of research approaches in green wireless networks,” EURASIP Journal on
Wireless Communications and Networking, vol. 142, pp. 1-18, April 2012.
[29] N. Vallina and J. Crowcroft, “Energy management techniques in modern
mobile handsets,” IEEE Commun. Surveys Tutorials, vol. 15, no. 1, pp.
179-198, 2013.
[30] M. Kennedy, A. Ksentini, Y. H. Aoul, and G. M. Muntean, “Adaptive
energy optimization in multimedia-centric wireless devices: a survey,”
IEEE Commun. Surveys Tutorials, vol. 15, no. 2, pp. 768-786, 2013.
[31] M. A. Hoque, M. Siekkinen, and J. K. Nurminen, “Energy efficient
multimedia streaming to mobile devices - a survey,” IEEE Commun.
Surveys Tutorials, to appear.
22
[32] G. Miao, N. Himayat, and G. Y. Li, “Energy-efficient link adaptation
in frequency-selective channels,” IEEE Journal Selected Areas Commun.,
vol. 58, no. 2, pp. 545-554, Feb. 2010.
[33] G. Miao, N. Himayat, G. Y. Li, an S. Talwar, “Low-complexity energy-
efficient scheduling for uplink OFDMA,” IEEE Trans. Commun., vol. 60,
no. 1, pp. 112-120, Jan. 2012.
[34] C. Y. Ho and C. Y. Huang, “Non-cooperative multi-cell resource
allocation and modulation adaptation for maximizing energy efficiency
in uplink OFDMA cellular networks,” IEEE Wireless Commun. Letters,
vol. 1, no. 5, pp. 420-423, Oct. 2012.
[35] Y. Rui, Q. T. Zhang, L. Deng, P. Cheng, and M. Li, “Mode selection
and power optimization for energy efficiency in uplink virtual MIMO
systems,” IEEE Journal Selected Areas Commun., vol. 31, no. 5, pp.
926-936, May 2013.
[36] I. Ashraf, F. Boccardi, and L. Ho, “Sleep mode techniques for small
cell deployments,” IEEE Commun. Magazine, vol. 49, no. 8, pp. 72-79,
Aug. 2011.
[37] K. Son, H. Kim, Y. Yi, and B. Krishnamachari, “Base station operation
and user association mechanisms for energy-delay tradeoffs in green
cellular networks,” IEEE Journal Selected Areas Commun., vol. 29, no.
8, pp. 1525-1536, Sep. 2011.
[38] G. Auer, V. Giannini, C. Desset, and I. Godor, “How much energy is
needed to run a wireless network?” IEEE Wireless Commun., vol. 18, no.
5, pp. 40-49, Oct. 2011.
[39] M. Deruyck, D. D. Vulder, W. Joseph, and L. Martens, “Modelling
the power consumption in femtocell networks” WCNC’12 Workshop on
Future Green Communications, pp. 30-35, April 2012.
[40] R. Riggio and D. J. Leith, “A measurement-based model of energy
consumption in femtocells” Wireless Days 2012 IFIP, pp. 1-5, Nov. 2012.
[41] G. P. Perrucci, F. H. P. Fitzek, and J. Widmer, “Survey on energy
consumption entities on the smartphone platform” IEEE VTC’11-Spring,
pp. 1-6, May 2012.
[42] F. Meshkati, H. V. Poor, S. C. Schwartz, “Energy-efficient resource
allocation in wireless networks,” IEEE Signal Processing Magazine, vol.
24, no. 3, pp. 58-86, May 2007.
[43] S. , G. Colavolpe, D. Saturnino, and A. Zapone, “Potential games
for energy-efficient power control and subcarrier allocation in uplink
multicell OFDMA systems,” IEEE Journal Selected Areas in Signal
Processing, vol. 6, no. 2, pp. 89-103, April 2012.
[44] O. Onireti, F. Heliot, and M. A. Imran, “On the energy efficiency-spectral
efficiency trade-off in the uplink of CoMP system,” IEEE Trans. Wireless
Commun., vol. 11, no. 2, pp. 556-561, Feb. 2012.
[45] C. Isheden and G. P. Fettweis, “Energy-efficient multi-carrier link
adaptation with sum rate-dependent circuit power,” Proc. Globecom’10,
pp. 1-6, Dec. 2010.
[46] O. Galinina, S. Andreev, A. Turlikov, and Y. Koucheryavy, “Optimizing
energy efficiency of a multi-radio mobile device in heterogeneous beyond-
4G networks,” Performance Evaluation, under review.
[47] G. Lim and L. J. Cimini, “Energy-efficient cooperative relaying in
heterogeneous radio access networks,” IEEE Wireless Commun. Letters,
vol. 1, no. 5, pp. 476-479, May 2013.
[48] A. P. Azad, “Analysis and optimization of sleeping mode in WiMAX
via stochastic decomposition techniques,” IEEE Journal Selected Areas
Commun., vol. 29, no. 8, pp. 1630-1640, Sep. 2011.
[49] K. Ying, H. Yu, and H. Luo, “Inter-RAT energy saving for multicast
services,” IEEE Commun. Letters, vol. 17, no. 5, pp. 900-903, May 2013.
[50] X. Ma, M. Sheng, and Y. Zhang, “Green communications with network
cooperation: a concurrent transmission approach,” IEEE Commun. Let-
ters, vol. 16, no. 12, pp. 1952-1955, Dec. 2012.
[51] Z. Niu, Y. Wu, J. Gong, and Z. Yang, “Cell zooming for cost-efficient
green cellular networks,” IEEE Commun. Magazine, vol. 48, no. 11, pp.
74-79, Nov. 2010.
[52] T. Han and N. Ansari, “On optimizing green energy utilization for
cellular networks with hybrid energy supplies,” IEEE Trans. Wireless
Commun., to appear.
[53] X. Zhang and P. Wang, “Optimal trade-off between power saving and
QoS provisioning for multicell cooperation networks,” IEEE Wireless
Commun., vol. 20, no. 1, pp. 90-96, Feb. 2013.
[54] Z. Niu, “TANGO: traffic-aware network planning and green operation,”
IEEE Wireless Commun., vol. 18, no. 5, pp. 25-29, Oct. 2011.
[55] A. da Silva, M. Meo, and M. Marsan, “Energy-performance trade-off
in dense WLANs: a queuing study,” Computer Networks, vol. 56, pp.
2522-2537, March 2012.
[56] L. B. Le, D. Niyato, E. Hossain, D. I. Kim, and D. T. Hoang, “QoS-
aware and energy-efficient resource management in OFDMA femtocells,”
IEEE Trans. Wireless Commun., vol. 12, no. 1, pp. 180-194, Jan. 2013.
[57] F. Liu, K. Zheng, W. Xiang, and H. Zhao, “Design and performance
analysis of an energy-efficient uplink carrier aggregation scheme,” IEEE
Journal Selected Areas Commun., to appear.
[58] N. Mastronarde and M. van der Schaar, “Fast reinforcement learning for
energy-efficient wireless communication,” IEEE Trans. Signal Processing,
vol. 59, no. 12, pp. 6262-6267, Dec. 2011.
[59] H. Kwon and B. G. Lee, “Energy-efficient scheduling with delay
constraints in time-varying uplink channels,” Journal of Commun. and
Networks, vol. 10, no. 1, pp. 28-37, March 2008.
[60] X. Lu, E. Erkip, Y. Wang, and D. Goodman, “Power efficient multimedia
communication over wireless channels,” IEEE Journal Selected Areas
Commun., vol. 21, no. 10, pp. 1738-1751, Dec. 2003.
[61] T. H. Lan and A. H. Tewfik, “A resource management strategy in wire-
less multimedia communications-total power saving in mobile terminals
with a guaranteed QoS,” IEEE Trans. Multimedia, vol. 5, no. 2, pp. 267-
281, June 2003.
[62] T. Han and N. Ansari, “On greening cellular networks via multicell
cooperation,” IEEE Wireless Commun., vol. 20, no. 1, pp. 82-89, Feb.
2013.
[63] E. Oh, K. Son, and B. Krishnamachari, “Dynamic base station switching-
on/off strategies for green cellular networks,” IEEE Trans. Wireless
Commun., vol. 12, no. 5, pp. 2126-2136, May 2013.
[64] C. Y. Chang, W. Liao, H. Y. Hsieh, and D. S. Shiu, “On optimal cell
activation for coverage preservation in green cellular networks,” IEEE
Trans. Mobile Computing, to appear.
[65] N. Saxena, B. J. R. Sahu, and Y.S. Han, “Traffic-aware energy optimiza-
tion in green LTE cellular systems,” IEEE Commun. Letters, vol. 18, no.
1, pp. 38-41, Jan. 2014.
[66] J. Wu, S. Zhou, and Z. Niu, “Traffc-aware base station sleeping
control and power matching for energy-delay tradeoffs in green cellular
networks,” IEEE Trans. Wireless Commun., vol. 12, no. 8, pp. 4196-4209,
Aug. 2013.
[67] S. Navaratnarajah, A. Saeed, M. Dianati, and M. A. Imran, “Energy
efficiency in heterogeneous wireless access networks,” IEEE Wireless
Commun., vol. 20, no. 5, pp. 37-43, Oct. 2013.
[68] L. Saker, S. E. Alayoubi, R. Combes, and T. Chahed, “Optimal control
of wake up mechanisms of femtocells in heterogeneous networks,” IEEE
Journal Selected Areas Commun., vol. 30, no. 3, pp. 664-672, April 2012.
[69] A. Conte, A. Feki, L. Chiaraviglio, D. Ciullo, M. Meo, and
M. A. Marsan, “Cell wilting and blossoming for energy efficiency,” IEEE
Wireless Commun., vol. 18, no. 5, pp. 50-57, Oct. 2011.
[70] A. P. Azad, S. Alouf, E. Altman, V. Borkar, and G. S. Paschos, “Optimal
control of sleep periods for wireless terminals,” IEEE Journal Selected
Areas Commun., vol. 29, no. 8, pp. 1605-1617, Sep. 2011.
[71] A. Agarwal and A. K. Jagannatham, “Optimal wake-up scheduling for
PSM delay minimization in mobile wireless networks,” IEEE Wireless
Commun. Letters, to appear.
[72] R. Wang, J. Tsai, C. Maciocco, T. Y. C. Tai, and J. Wu, “Reducing
power consumption for mobile platforms via adaptive traffic coalescing,”
IEEE Journal Selected Areas Commun., vol. 29, no. 8, pp. 1618-1629,
Sep. 2011.
[73] H. Yan, S. A. Watterson, D. K. Lowenthal, K. Li, R. Krishnan, and
L. L. Peterson, “Client-centered, energy-efficient wireless communication
on IEEE 802.11b networks,” IEEE Trans. Mobile Computing, vol. 5, no.
11, pp. 1575-1590, Nov. 2006.
[74] H. Zhu and G. Cao, “On supporting power-efficient streaming applica-
tions in wireless environments,” IEEE Trans. Mobile Computing, vol. 4,
no. 4, pp. 391-403, Aug. 2005.
[75] Y. Jin, J. Xu, and L. Qiu, “Energy-efficient scheduling with individual
packet delay constraints and non-ideal circuit power,” Journal Commun.
and Networks, vol. 16, no. 1, pp. 36-44, Feb. 2014.
[76] K. D. Turck, S. D. Vuyst, D. Fiems, S. Wittevrongel, and H. Bruneel,
“Performance analysis of sleep mode mechanisms in the presence of
bidirectional traffic,” Computer Networks, vol. 56, pp. 2494-2505, March
2012.
[77] S. Videv, J. S. Thompson, H. Haas, and P. M. Grant, “Resource
allocation for energy efficient cellular systems,” EURASIP Journal on
Wireless Communications and Networking, vol. 181, pp. 1-15, May 2012.
[78] Z. Ren, S. Chen, B. Hu, and W. Ma, “Energy-efficient resource allocation
in downlink OFDM wireless systems with proportional rate constraints,”
IEEE Trans. Vehicular Technology, to appear.
[79] Y. L. Chung, “Rate-and-power control based energy-saving transmis-
sions in OFDMA-based multicarrier base stations,” IEEE Systems Jour-
nal, to appear.
[80] L. Chen, Y. Yang, X. Chen, and G. Wei, “Energy-efficient link adaptation
on rayleigh fading channel for OSTBC MIMO system with imperfect
23
CSIT,” IEEE Trans. Vehicular Technology, vol. 62, no. 4, pp. 1577-1585,
May 2013.
[81] L. C. Wang, W. C. Liu, A. Chen, and K. N. Yen, “Joint rate and
power adaptation for wireless local area networks in generalized nakagami
fading channels,” IEEE Trans. Vehicular Technology, vol. 58, no. 3, pp.
1375-1386, March 2009.
[82] L. P. Qian, Y. J. Zhang, Y. Wu, and J. Chen, “Joint base station
association and power control via benders decomposition,” IEEE Trans.
Wireless Commun., vol. 12, no. 4, pp. 1651-1665, April 2013.
[83] V. A. Siris, M. Anagnostopoulou, “Performance and energy efficiency of
mobile data offloading with mobility prediction and prefetching,” Proc.
IEEE WoWMoM’13, pp. 1-6, June 2013.
[84] N. Ristanovic, J. Y. Le Boudec, A. Chaintreau, and V. Erramilli, “Energy
efficient offloading of 3G networks,” Proc. IEEE MASS’11, pp. 202-211,
Oct. 2011.
[85] S. Tarkoma, M. Siekkinen, E. Lagerspetz, and Y. Xiao, “Smartphone
energy consumption: modeling and optimization,” Cambridge University
Press, 2014.
[86] T. A. Le, S. Nasseri, A. Z. Esfahani, M. R. Nakhai, and A. Mills,
“Power-efficient downlink transmission in multicell networks with limited
wireless backhaul,” IEEE Wireless Commun., vol. 18, no. 5, pp. 82-88,
Oct. 2011.
[87] M. Z. Shaker, K. A. Qaraqe, M. S. Alouini, E. Serpedin, and M. A. Im-
ran, “Green heterogeneous small-cell networks: toward reducing the
CO2emissions of mobile communications industry using uplink power
adaptation,” IEEE Wireless Commun., vol. 51, no. 6, pp. 52-61, June
2013.
[88] H. Ghazzai, E. Yaacoub, M. S. Alouini, and A. A. Dayya, “Optimized
smart grid energy procurement for LTE networks using evolutionary
algorithms,” IEEE Trans. Vehicular Technology, to appear.
[89] C. McGuire, M. R. Brew, F. Darbari, G. Bolton, A. McMahon,
D. H. Crawford, S. Weiss, and R. W. Stewart, “HopScotch-a low-power
renewable energy base station network for rural broadband access,”
EURASIP Journal on Wireless Communications and Networking, vol.
112, pp. 1-12, March 2012.
[90] T. Han and N. Ansari, “Powering mobile networks with green energy,”
IEEE Wireless Commun., vol. 21, no. 1, pp. 90-96, Feb. 2014.
[91] B. Devillers and D. Gunduz, “A general framework for the optimization
of energy harvesting communication systems with battery imperfections,”
Journal of Commun. and Networks, vol. 14, no. 2, pp. 130-139, April
2012.
[92] O. Orhan, D. Gunduz, and E. Erkip, “Throughput maximization for an
energy harvesting communication system with processing cost,” Proc.
IEEE Information Theory Workshop, pp. 84-88, Sept. 2012.
[93] K. Tutuncuoglu and A. Yener, “Optimum transmission policies for
battery limited energy harvesting nodes,” IEEE Trans. Wireless Commun.,
vol. 11, no. 3, pp. 1180-1189, Feb. 2012.
[94] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient resource allo-
cation in OFDMA systems with hybrid energy harvesting base station,”
IEEE Trans. Wireless Commun., vol. 12, no. 7, pp. 3412-3427, July 2013.
[95] K. Tutuncuoglu and A. Yener, “Sum-rate optimal power policies for
energy harvesting transmitters in an interference channel,” Journal of
Commun. and Networks, vol. 14, no. 2, pp. 151-161, April 2012.
[96] O. Ozel, K. Tutuncuoglu, J. Yang, S. Ulukus, and A. Yener, “Transmis-
sion with energy harvesting nodes in fading wireless channels: optimal
policies,” IEEE Journal Selected Areas Commun., vol. 29, no. 8, pp.
1732-1743, Sept. 2011.
[97] B. Gurakan, O. Ozel, J. Yang, and S. Ulukus, “Energy cooperation in
energy harvesting communications,” IEEE Trans. Commun., submitted.
[98] C. F. Chiasserini and R. R. Rao, “Energy efficient battery management,”
IEEE Journal Selected Areas Commun., vol. 19, no. 7, pp. 1235-1245,
July 2001.
Muhammad Ismail (S’10, M’13) is an assistant
research scientist in the Electrical and Computer
Engineering department in Texas A&M University at
Qatar. He received his PhD degree in Electrical and
Computer Engineering from University of Waterloo,
Canada in 2013, and MSc. and BSc. degrees in
Electrical Engineering (Electronics and Communi-
cations) from Ain Shams University, Cairo, Egypt in
2007 and 2009, respectively. Dr. Ismail research in-
terests include distributed resource allocation, green
wireless networks, cooperative networking, smart
grid, and biomedical signal processing. He is a co-recipient of the best paper
award in IEEE ICC 2014. He served as a TPC member in the ICWMC in 2010
- 2014 and IEEE ICC 2014 and 2015. He served in the IEEE INFOCOM 2014
organizing committee as a web chair. He joined the International Journal On
Advances in Networks and Services editorial board since January 2012. He
was an editorial assistant for the IEEE Transactions on Vehicular Technology
in the period January 2011 - July 2013. He has been a technical reviewer for
several IEEE conferences and journals.
Weihua Zhuang (M’93,SM’01,F’08) has been with
the Department of Electrical and Computer Engi-
neering, University of Waterloo, Canada, since 1993,
where she is a Professor and a Tier I Canada Re-
search Chair in Wireless Communication Networks.
Her current research focuses on resource allocation
and QoS provisioning in wireless networks, and on
smart grid. She is a co-recipient of several best paper
awards from IEEE conferences. She received the
Outstanding Performance Award 4 times since 2005
from the University of Waterloo and the Premier’s
Research Excellence Award in 2001 from the Ontario Government. Dr. Zhuang
was the Editor-in-Chief of IEEE Transactions on Vehicular Technology (2007-
2013), and the Technical Program Symposia Chair of the IEEE Globecom
2011. She is a Fellow of the IEEE, a Fellow of the Canadian Academy of
Engineering (CAE), a Fellow of the Engineering Institute of Canada (EIC),
and an elected member in the Board of Governors and VP Mobile Radio of
the IEEE Vehicular Technology Society. She was an IEEE Communications
Society Distinguished Lecturer (2008-2011).
Erchin Serpedin (F’13) received the specialization
degree in signal processing and transmission of
information from Ecole Superieure DElectricite (SU-
PELEC), Paris, France, in 1992, the M.Sc. degree
from the Georgia Institute of Technology, Atlanta, in
1992, and the Ph.D. degree in electrical engineering
from the University of Virginia, Charlottesville, in
January 1999. He is currently a professor in the
Department of Electrical and Computer Engineering
at Texas AM University, College Station. He is the
author of 2 research monographs, 1 textbook, 90
journal papers and 150 conference papers, and serves currently as associate
editor for IEEE TRANSACTIONS ON COMMUNICATIONS, Signal Pro-
cessing (Elsevier), EURASIP Journal on Advances in Signal Processing, Phys-
ical Communication (Elsevier), and EURASIP Journal on Bioinformatics and
Systems Biology. His research interests include statistical signal processing,
information theory, bioinformatics, and genomics. He is an IEEE Fellow.
Khalid Qaraqe (M97,SM00) was born in Bethle-
hem. He received the B.Sc. degree in EE from the
University of Technology, Baghdad in 1986, with
honors. He received the M.Sc. degree in EE from the
University of Jordan, Jordan, in 1989, and he earned
his Ph.D. degree in EE from Texas AM University,
College Station, TX, in 1997. From 1989 to 2004
Dr. Qaraqe has held a variety positions in many
companies and he has over 12 years of experience in
the telecommunication industry. He has worked for
Qualcomm, Enad Design Systems, Cadence Design
Systems/Tality Corporation, STC, SBC and Ericsson. He has also worked on
numerous GSM, CDMA, WCDMA projects and has experience in product
development, design, deployments, testing and integration. Dr. Qaraqe joined
the department of Electrical Engineering of Texas AM University at Qatar, in
July 2004, where he is now professor. Dr. Qaraqe research interests include
communication theory and its application to design and performance, analysis
of cellular systems and indoor communication systems. Particular interests are
in the development of 3G UMTS, cognitive radio systems, broadband wireless
communications and diversity techniques.
