Energy-Efficient Base Stations Sleep Mode Techniques
Energy-Efficient Base Stations Sleep Mode Techniques
Energy-Efficient Base Stations Sleep Mode Techniques
Abstract—Due to global climate change as well as economic recently, the design objectives for cellular networks, since their
concern of network operators, energy consumption of the infras- introduction in the late 1970s, have been maximum through-
tructure of cellular networks, or “Green Cellular Networking,” put, spectral efficiency and meeting Quality of Service (QoS)
has become a popular research topic. While energy saving can
be achieved by adopting renewable energy resources or improving requirements rather than energy conservation. Another mo-
design of certain hardware (e.g., power amplifier) to make it more tive for the need to find ways to conserve energy in cellular
energy-efficient, the cost of purchasing, replacing, and installing networks is increasing energy bills for telecommunications
new equipment (including manpower, transportation, disruption service providers. Due to the introduction and popularization
to normal operation, as well as associated energy and direct cost) of smart phones and tablets, which provide services that in-
is often prohibitive. By comparison, approaches that work on the
operating protocols of the system do not require changes to current volve exchange of large volume of data traffic and motivates
network architecture, making them far less costly and easier new high-speed (and certainly more energy consuming) mobile
for testing and implementation. In this survey, we first present network standards such as 4G LTE. As a result, expenses related
facts and figures that highlight the importance of green mobile to energy consumption now comprise a large proportion of
networking and then review existing green cellular networking operating cost for service providers [6].
research with particular focus on techniques that incorporate the
concept of the “sleep mode” in base stations. It takes advantage of It is now widely acknowledged that cellular communication
changing traffic patterns on daily or weekly basis and selectively networks will have greater economic and ecological impact in
switches some lightly loaded base stations to low energy consump- the coming years [4], [5]. This issue has been recognized as
tion modes. As base stations are responsible for the large amount a matter for both the planet and the wallet. Seeing this, an
of energy consumed in cellular networks, these approaches have innovative new research discipline called “green cellular net-
the potential to save a significant amount of energy, as shown
in various studies. However, it is noticed that certain simplifying works,” concentrating on environmental influences of cellular
assumptions made in the published papers introduce inaccuracies. networks, has been formed and attracted many researchers. The
This review will discuss these assumptions, particularly, an as- term “green” is originally a nickname of dedicated efforts to
sumption that ignores the effect of traffic-load-dependent factors reducing unnecessary green house gases (e.g., CO2 ) emissions
on energy consumption. We show here that considering this effect from industries. For mobile operators in particular, another
may lead to noticeably lower benefit than in models that ignore this
effect. Finally, potential future research directions are discussed. motivation and objective of “green” approaches is to gain extra
commercial benefits, mainly by reducing operating expense
Index Terms—Cellular network, base station, energy consump- related to energy cost [5], [7].
tion, energy efficiency, green networking, sleep mode.
There are various distinctive approaches to reduce energy
consumptions in a mobile cellular network. Approaches in
I. I NTRODUCTION
previous research can be broadly classified into the following
TABLE I
C OMPARISON OF G REEN C ELLULAR N ETWORK A PPROACHES
from such renewable resources to complement existing electric- tion of techniques for sleep mode in cellular networks. After
operated infrastructure, would probably be the long-term en- that, specific research and applications of sleep mode in vari-
vironmental solution for the mobile cellular network industry. ous network standards are discussed in Section V and perfor-
Especially for those areas without mature network infrastruc- mance comparisons of approaches are presented in Section VI.
ture, deploying energy harvesting networks would be ideal. For Potential areas for improvement is discussed in Section VII
developed countries with completed infrastructure, however, and insightful remarks on future directions are explored in
the same question of embodied and replacement cost arises Section VIII. Finally, the survey is concluded in Section IX.
as the component-based approaches. While service migrates
from the obsolete electric-operated BSs to the new energy
harvesting BSs, it is technically challenging to preserve fault- II. FACTS AND F IGURES
tolerance and data security without any service interruption. A. Objectives of Traditional Mobile Networks Design
The advantages and limitations of each approach, are sum-
marized in Table I. Previously, mobile networks, or wireless communication
Generally speaking, green cellular network is a relatively networks in general, have been designed with the objective of
new area of research. Most of existing publications are based optimizing coverage, capacity, spectral efficiency or throughput
on ideal models. The fundamental aim, as its name implies, is [52], [53]. Clearly, it does not necessarily maximize energy
to make cellular networks “greener” by reducing total power efficiency. Also, traditional facilities were mostly designed to
consumption through various approaches described above. For endure peak load and extreme conditions. Many of them are
more survey information on the entire field of green technolo- even dimensioned with redundancy, providing extra capacity to
gies in wireless communication networks, the reader is referred possible peak load, in order to allow for unexpected events.
to [5], [24], [29], [32], [34], [49] and [50]. As a result, the system is significantly under-utilized during
In this survey, we focus on the sleep mode techniques in non-peak hours, creating an opportunity for possible energy
BSs, and provide more details beyond the coverage of previous saving. It is worth noting that traditional design objectives are
surveys. As discussed above, sleep mode techniques do not potentially contradictory to green ones, which makes green
require upgrade of equipment, therefore they have the benefit networking an interesting and technically challenging research
of low implementation cost since replacement of hardware is field. Therefore, a new networking paradigm is urgently needed
avoided. In surveying the literature, we have observed that so that existing networks will maintain the same level of
studies on the topic of applications of sleep mode techniques QoS while reducing the amount of energy consumed in the
to mobile networks made different assumptions on system future [6], [7].
and power models, e.g., the effect of traffic load on energy
consumptions. We discuss these inconsistencies in the paper
B. Energy Consumption of Mobile Networks Today
and demonstrate that the benefit of sleep mode techniques is
significantly affected by the assumptions. It was estimated that ICT roughly accounted for about 10%
The remainder of this article is organized as follows. of global electricity consumption and up to 4% of global carbon
Section II provides recent facts and figures that motivate green dioxide emissions (around 1 billion tons, approximately equal
cellular networking research. Then, in Section III energy ef- to that of aviation industry and one fourth of emission by cars
ficiency metrics of interest with respect to BS sleep mode worldwide) as of early 2013 [54]. ICT’s share in global carbon
techniques in cellular networks are discussed. Next, Section IV emissions were expected to grow every year, and double to 4%
introduces the potential of savings, feasibility and the founda- by the year 2020 [55]. Another figure shows that by the end of
806 IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015
2012, the amount of carbon dioxide emissions from BS towers impact in overall energy consumption [12], [58]. Relevant
alone has reached 78 million tons, which is approximately data (Fig. 2) show that the number of BSs worldwide has
equivalent to emissions from 15 million cars, or 150 000 round- approximately doubled from 2007 to 2012, and the number of
trip flights between Paris and New York [55]. BSs today has reached more than 4 million [51], [59]. When
The prevalence of smart phones and tablets accessing cellular cellular networks extend to remote districts, or developing or
network remarkably contributes to the increasing energy con- undeveloped regions, off-grid BSs need to be deployed as no
sumption. Smart phones were introduced around 2000. How- electrical grids are available nearby. Compared to their on-grid
ever, it was the success of mobile operating systems such as counterparts, off-grid BSs may cost ten times more to run, since
iOS, Android and Windows Phone about a decade later that they generally depend on fuel, which is a costly and unreliable
finally helped them take over traditional feature phones. Tablet power source [5], [25], [59]. On the other hand, hydrocarbon
computers became popular almost at the same time, marked by energy, one of primary conventional energy resources that
the release of the iPad by Apple Inc. With the help of higher provides 85% of primary energy usage in the United States and
data transmission rate in 3G and 4G (and 5G in the future) cellu- releases large amounts of greenhouse gases when combusted,
lar networks, smart phones and tablets enable users to perform is proved not sustainable and expected to be depleted in the
much more tasks than ever before using cellular networks, in- foreseeable future [60].
cluding, but not limited to, streaming videos, downloading and The BSs serving small cells are known to be much less
reading e-books, and social networking. As a consequence, both energy consuming as compared to those serving macro cells.
the number of mobile subscribers (4.5 billion in 2012, estimated They have been deployed at densely populated area or edges of
7.6 billion by 2020) and the amount of data traffic requested existing macro cells to improve spectral and energy efficiency.
by each subscriber (on average 10 GB per subscriber per year However, due to the tremendous deployment of small cells in
in 2012, estimated 82 GB per subscriber per year by 2020) the foreseeable future, they will consume around 4.4 TWh of
have increased explosively [56]. Also, more bursty and dynamic power by the year 2020, constituting an extra 5% of energy
mobile data and video traffic have replaced mobile voice as consumption of the conventional macro cell network [37], [61].
the dominant traffic in cellular networks. These factors lead On the other hand, while the penetration rate of mobile
to significant increase in energy consumption. Manner et al. phones in developed areas has exceeded 100%, the rising
[57] showed that, in order to provide the same level of coverage, trend in developing areas is rapid and still far from saturation
an LTE network has to consume about 60 times amount of [62]. For example, China, the biggest developing market for
energy as compared to a 2G network. telecommunications, had a 90% mobile phone penetration rate,
More BSs, data centers and other network equipment are but only less than 30% of population had access to 3G or
required to support the growth in mobile traffic. Since BSs more advanced networks, as of October 2013 [63]. All of these
consume more than half of the total energy in a typical cellular indicate great potential for further growth of mobile data traffic
network, the increase in the number of BSs has a significant and thus, energy consumption.
WU et al.: BASE-STATIONS SLEEP-MODE TECHNIQUES IN GREEN CELLULAR NETWORKS 807
C. Improvement on Energy Efficiency in User Equipment In addition, given the concern regarding carbon dioxide
There have been significant improvements during the last two emission, regulatory units, non-for-profit organizations and en-
decades in carbon footprint per mobile subscriber. In the early vironmental advocates have together initiated projects to reduce
1990s, an average mobile subscriber would be responsible for energy consumption in cellular networks, or at least slow down
100 kg of carbon dioxide emissions per year. This figure had the increasing trend. Nordhaus [76] demonstrated that reducing
been reduced to one quarter, namely 25 kg per subscriber by the the global greenhouse gas emissions by a factor of 1/3 will
mid-2000s. However, since the number of mobile subscribers generate economic benefit higher than the investment required
has dramatically increased, the total amount of carbon footprint to accomplish such reduction. Notable collaborative projects
is still rising despite of the reduction in footprint per subscriber that aim to reduce energy consumption in mobile networks in-
[64]. Meanwhile, the increasing number of subscribers also clude 3GPP [77], EARTH [56], OPERA-Net [78], C2POWER
causes total data volume of wireless networks to increase [79], eWin [80], and TREND [81]. They proposed advanced
approximately by a factor of ten every five years, which is asso- technologies such as effective utilization of spectrums, innova-
ciated with 16% to 20% increase in energy consumption [65]. tive component designs, energy efficient network architectures,
The energy consumption in a cellphone, including battery energy efficient routing protocols, node selection protocols, and
chargers and user equipment (UE, any mobile device used clustering techniques. The reported potential saving is up to
by end users to communicate), had been reduced from 32Wh 50% as compared to the baseline of today’s network system.
per day in the early 1990s to 0.83 Wh per day in 2008, It is also worth notice that while we have reached the 4G
with a saving of more than 97%. The achievement in energy era, 2G (GSM, GPRS, CDMAone, IS-95, EDGE) and 3G
saving in UE has made the energy consumption negligible (WCDMA, HSPA, UMTS, EV-DO, etc.) networks are expected
as compared to that in BSs. Nowadays, the main motivation to coexist in the coming years due to their mature architectures,
for further improving energy efficiency in cellphones is not business models and lower price to subscribe. Besides, the older
ecological or economic impacts, but longer battery life and thus generations of networks could always act as the “backup” net-
better user experience [50], [66]–[68]. For example, extensive work when higher standard networks are down or temporarily
research has been carried out on energy efficiency of data and not available. Therefore, energy saving technologies on 2G and
power consuming applications such as social networking and 3G networks still have large impacts on the improvement of
multimedia streaming, to improve battery life of UE [67], [69], overall energy efficiency of cellular networks today.
[70]. A comprehensive survey of energy efficient techniques on
UE can be found in [50]. E. Summary
In this section, we have presented the background, achieve-
ments and concerns associated with green cellular network
D. Problems Faced by Mobile Network Operators research. The increasing trend of energy consumption in mobile
networks will probably continue in the foreseeable future. It
While energy efficiency in UE has been significantly im- will adversely affect our civilization in many ways unless
proved, the same work on BSs, which consumes the most adequate measures are taken.
amount of energy in mobile network, has been lagged behind. Generally, energy efficiency of UE now draws interest from
As a result, mobile operators are charged even more with cellphone manufacturers for the sake of longer battery life [82],
skyrocketing number of BSs. while network operators and policy makers are more involved in
The fact that energy cost comprises a large proportion of total reducing energy consumption in BSs due to economic and eco-
cost justifies efforts from telecommunication service providers logical reasons. As the motivation, objectives and techniques
to reduce energy consumption in order to improve their bottom for the two sides are drastically different from each other, this
line. Figures show that their energy bills are now comparable paper will mainly focus on energy conservation in BSs.
to their personnel costs for network operations, which range
from 18 percent (in the EU) to 32 percent (in India) of their III. R ELEVANT E NERGY E FFICIENCY M ETRICS
total operating expense (OPEX) [64], [71]. In Germany, the
electricity bill for mobile network operators is more than Given various proposals to improve energy efficiency of
200 million Euros per year [3]. It is predicted that the revenue cellular networks, a framework to evaluate the performance is
for global mobile network industry will start to shrink from essential to assess and compare different schemes [65], [83].
the year 2018 [72]. Apart from direct economic benefits, envi- In this regard, energy efficiency metrics measuring at different
ronmental and marketing reasons (better corporative image can levels have been proposed by both academia and industry.
boost sales) are other driving forces for telecom providers to
take the green initiative. Evidently, several telecom providers A. Classical Energy Efficiency Metrics
such as PCCW and Vodafone have been taking measures A classical and widely-used metric to evaluate the energy ef-
to reduce their energy-related operating cost [73]–[75]. An ficiency of telecommunication networks is Bit-per-Joule [84]. It
analysis shows that total energy per unit traffic declined by represents the system throughput (amount of information trans-
approximately 20% and energy per connection declined by mitted) for unit-energy consumption. It is still referenced in
5% from 2009 to 2010, indicating that the industry is making recent relevant studies [85]–[87] because of its simplicity. The
significant efforts and progress towards the goals of energy reciprocal of Bit-per-Joule measurement, namely energy con-
saving [58]. sumption per delivered information bit, is referred as Energy
808 IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015
Consumption Ratio (ECR) [88]. A close variant to the Bit-per- Another metric on the node level is the Energy Consumption
Joule metric is Bit-per-Joule-Hz. Instead of system throughput Index (ECI) [91] given by:
(bits/s), Bit-per-Joule-Hz uses spectral efficiency (bits/s/Hz) as psite
the performance measurement. It is equal to the ratio of spectral ECI = (3)
KP I
efficiency over the total power consumption [89].
External energy benchmarks, such as JouleSort ([90], which where psite refers to total input power of the site (e.g., BS)
measures energy consumption of a component or system), have and KP I (key performance indicator) could be either coverage
also been proposed to assess the scale of improvements in area or throughput. The ECI measures the efficiency of power
energy efficiency of telecommunication networks. However, utilization for a BS. Lower values of ECI indicate better
as the energy saving in ICT draws global attention from re- energy efficiency. Energy saving achieved by sleep mode is
searchers, tailor-made energy efficiency metrics are needed to taken into consideration in psite as the input power decreases
assess the improvements provided by various approaches. if sleep mode is adopted. A specific example is Energy Con-
More recently proposed energy efficiency metrics provide sumption Ratio (ECR) discussed in [95] (different from the
effective means for understanding, measuring, reporting, de- ECR discussed in previous text, which is the reciprocal of bit-
signing objectives and evaluating the performance of energy per-joule), in which KP I is the peak data throughput. Another
efficiency of components, systems and networks. Energy effi- similar example is presented in [21], in which energy efficiency
ciency metrics on component (such as power amplifier), node is defined as spectral efficiency per unit of energy consumption.
(BS or BS site) and system levels in mobile cellular networks At the system level, energy efficiency is generally mea-
have been extensively covered [5], [64], [65], [91], [92]. In gen- sured in average power consumption per user or per unit area.
eral, a good metric should be standardized to compare energy An example is the performance indicators (P I), proposed by
consumption of different units in the same class and provide European Telecommunications Standards Institute (ETSI) in
directions for possible research and development targets. They [96], defined by
should accurately reflect the energy efficiency and evaluate
the performances fairly and objectively, but should not be too Total coverage area
P Irural = (4)
complicated that they are difficult to understand and derive. Power consumption
and
B. Energy Efficiency Metrics on BS Sleep Mode Techniques
Number of users in peak hour
For sleep mode techniques in particular, the fraction of time P Iurban = (5)
that the component or BS spends in sleep mode (Tsleeping ) over Power consumption
a certain period (Ttotal ) is usually adopted as an approximate where performance in rural areas is measured in average power
saving estimation at component or node level, which can be consumption per coverage area because the density of sub-
expressed as: scribers is typically low, while average power consumption per
Tsleeping user is thought to be a more accurate measurement for urban
Savings from sleep mode = . (1)
Ttotal areas where density of subscribers is high. Higher values of P I
indicate better energy efficiency. The performance indicator P I
Assuming static energy consumption in active mode, zero
and its close variants can be found in a number of studies on
energy consumption in sleep mode and zero cost for switching
sleep mode, e.g., [38].
operation, the rough approximation of (1) is used in various
Similar to P I, the Area Power Consumption (AP C) mea-
studies (e.g., [93], [94]). A somewhat more realistic modifica-
sures the power consumption in a considered area [95]. The
tion considers active mode and sleep mode, each consumes a
area can be coverage of a certain BS (node level) or the whole
fixed amount of energy, and the average power consumption
network (system level). It is measured by the ratio of power
during a certain period is obtained by:
consumption to the area, i.e.,
Ptotal = fsleep psleep + factive pactive (2) Power consumption
AP C = [W/km2 ]. (6)
where fsleep and factive refer to the fraction of time that the Area
component or BS is in sleep or active state, while psleep and
A new metric integrating ECR and AP C is proposed in
pactive are the power consumption in sleep mode and active
[95]. Taking both power consumption per unit and requested
mode, respectively [40]. The exact values of psleep and pactive
capacity into account, it is expressed as:
depends on specific implementation. For example, if the entire
BS is treated as the unit to be switched, psleep will be minimal Power consumption
as it only accounted for power consumption of the signal γ= [W·km−2 ·bps−1 ]. (7)
Requested capacity·Coverage area
processing unit for transmitting pivot signals, while pactive
can be described as the sum of fixed power consumption (in- For both AP C and γ in (7), lower values indicate better energy
cluding air conditioning, signal processing and power supply) efficiency.
and traffic-dependent power consumption. Details of the two For small cell network deployment, energy consumption gain
elements comprising pactive will be discussed in detail later in (ECG) quantifies the gain in energy saving in radio access
Section VII. network by deploying smaller cells [88]. It is defined by the
WU et al.: BASE-STATIONS SLEEP-MODE TECHNIQUES IN GREEN CELLULAR NETWORKS 809
quotient of energy consumption in large cell deployment di- C. Discussions and Summary
vided by that in small cell deployment, namely
There have been a number of proposals on various energy effi-
Elargecell ciency metrics as discussed above. The various metrics actually
ECG = . (8)
Esmallcell reflect the fact that energy efficiency is a relatively subjective
concept. It depends on the specific model (e.g., homogeneous
An interesting metric, called absolute energy efficiency met-
macro cell or heterogeneous cells), environment (e.g., rural or ur-
ric, was proposed to incorporate the cost in carbon footprint
ban) and information available that which metric should be used.
along with energy consumption [97]. Apart from energy con-
The first question arise here is the level of measurement.
sumption and throughput, absolute temperature is taken into
Normally, BS sleep mode techniques involve turning off one
account for its role in carbon emission. A logarithmic example
or more BSs, therefore component level metric might not be
called dB is defined by
appropriate in most cases. The choice between node and system
Power/Bit Rate level metrics depends on specific area of interest. System level
dB = (9)
ln2(kT ) metrics are more appropriate when the overall performance and
energy consumption of the network is concerned, while node
where k is the Boltzmann constant and T is the absolute
level metrics provide useful insights for certain parts of the sys-
temperature measured in Kelvin. It can be applied at a device,
tem covered by a single BS (e.g., most densely populated area).
node or system level. However, this metric is not yet mature, Another concern for choosing metrics is the point of view.
as the logarithmic relationship of [97] still requires stronger If measured from the service point of view, metrics should
justification, and no follow up research can be found so far. evaluate power or energy consumption against QoS parameters
In addition to savings by sleep mode, various costs related such as throughput, capacity, or blocking probability. Suitable
to switching between active and sleeping modes, such as cost metrics which are discussed previously include ECI, absolute
of exchanging load update messages between BSs, cost of energy temperature and ECR (in [95]). On the other hand,
handover between BSs, and cost of collecting traffic load when taking a deployment perspective of view, parameters
information, are also considered in a few studies (e.g., [98]). such as coverage area, number of users under coverage, and
However, they are still measured separately so far. A unified number of BSs are better measurements and thus should be re-
metric may better assess the overall performance of different flected in the metrics used. Therefore, metrics such as P I area
approaches. power consumption, GASE and AGE are considered better
Zhang et al. [99] proposed a new performance metric to candidates. Finally, while classical metrics such as bit-per-joule
evaluate energy efficiency as well as spectral efficiency of measures absolute energy consumption, metrics comparing
cellular networks. The metric, termed as Generalized Area energy consumption with and without green approaches would
Spectral Efficiency (GASE, denoted by η), is equal to ergodic provide insights on relative savings. Fraction of sleeping time
capacity (denoted by C) divided by the size of affected area of and ECG fulfill such requirements.
transmission (denoted by A), namely While heterogeneous deployment is becoming more and
C more popular in cellular networks, energy-efficient techniques
η= . (10) including sleep mode in such network have also been inten-
A
sively studied. Metrics focused on networks with heteroge-
It is somehow similar to (Γ) in Eqn. (7), where the measurement neous deployments thus also attracted lots of attention. These
involves capacity and the size of coverage area. The authors tailor-made metrics, including ECG and AGE discussed ear-
argued that the proposed metric provided a more comprehen- lier in the section, quantify the savings in macro and micro or
sive evaluation of spectral and energy efficiencies of wireless femto cells separately.
networks, by taking more elements of the network such as Most of energy efficiency metrics discussed above can be
relays and secondary transmitters into account. In cellular net- used in parallel with each other, if relevant data are available.
works, secondary transmitters refer to unlicensed users offered However, the results may not be consistent in all cases. It should
opportunistic access to the network. be noticed that energy gain in one component or cell of interest
Tabassum et al. [100] proposed another area-based energy may be the result of energy loss in another component or
efficiency measurement for two-tier heterogeneous cellular net- neighboring cells. Therefore, when there are conflicting results
works called area green efficiency (AGE). The two-tier network among measurements of different levels, system level metrics
is formed by a macro cell in the center and a number of femto should prevail over node and component level metrics, and node
cells distributed around the edge of the reference macro cell. level metrics should dominate component level metrics [25].
AGE is defined as: A potential area for improvement in terms of measuring
Pm + Pn metrics is that QoS degradation should be taken into account.
AGE = . (11) Intuitively, while BSs are turned off for the sake of energy
π(Rm + Rn )2
saving, total capacity, throughput and coverage of the network
The numerator of (11) is the sum of aggregated power savings by will be negatively affected as a result of less active BSs. Mean-
sleep mode or other green technologies in the macro cell (Pm , while, blocking probability would increase. If the gain in energy
1st tier) and in femto cells (Pn , 2nd tier). The denominator is the saving is not enough to offset the loss due to significantly
total coverage area of macro cell and the femto cells, with Rm worsening user experience, it is pointless in implementing
being the radius of macro cell and Rn the radius of femto cell. such schemes.
810 IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015
TABLE II
E NERGY E FFICIENCY M ETRICS
Possible negative impacts brought by sleep mode on delay It is actually important because one of practical objectives of
and throughput have been studied in [101]. As discussed, for green technologies is to reduce carbon footprint, and energy
example, in [102] and [103], this trade-off can be adjusted by sources have different carbon emission level. For example, a
tuning some parameters such as guard interval and hysteresis certain amount of energy generated by solar source should
time in implementation. The energy–delay tradeoff is further be considered “greener” than the same amount generated by
investigated in [104], in which impacts of BS control policies diesel. A metric incorporating this factor could possibly evalu-
including close-down time, the number of waiting customers ate the overall environmental effect of certain implementations.
for the BS to reactivate, and delay bound were discussed. It is Table II summarizes the energy efficiency metrics discussed
shown that energy is a decreasing function of delay in most in this section. More general and comprehensive energy metrics
cases. Therefore, it is necessary for mobile operators to find the can be found in the referenced articles.
optimal balance between QoS preservation and energy saving.
In the EARTH project, the concept of utility function for energy
IV. E NERGY S AVING P OTENTIAL AND T ECHNICAL
saving techniques has been proposed to assess the trade-off
F EASIBILITY OF BS S LEEPING
between system performance and user experience [91]. One of
the proposed utility functions is: Before implementing sleep mode in BSs, it is important to
estimate its potential savings, as well as to establish its technical
U = 100 + (αE ΔUE ) + (αQ ΔUQ ) (12)
feasibility. One example of important technical concerns is
where 100 is the utility of the reference system (before energy whether there will be any coverage holes when some BSs are
saving technologies are applied), αE and αQ are the weights of turned into sleep mode. In this section, we will discuss the
value of energy and QoS, respectively, and ΔUE and ΔUQ are energy saving potential and technical feasibility of BS sleep
the energy and QoS deviations of the assessed system from the mode techniques.
reference system, respectively. It is still an open question how
to determine the most suitable values for αE and αQ , as well as
A. Energy Consumption Breakdown in BSs
which specific parameter or combination of parameters should
represent QoS (blocking probability, throughput, coverage or There are three key components in a typical mobile cel-
others), for different markets, thus providing network operators lular network: (1) UE for end users to access the network,
with explicit objectives. A similar but more complicated system (2) network switching subsystem to route calls and data, and
cost function considering flow-level performance and system (3) BSs (sometimes referred to as BS subsystem or access
energy consumption is proposed in [105]. network) for commuting mobile traffic and signaling between
Another shortcoming of existing metrics is that they do not the previous two components. As discussed, BSs consume the
measure the “greenness” based on the source of power supply. largest proportion of energy in mobile cellular networks.
WU et al.: BASE-STATIONS SLEEP-MODE TECHNIQUES IN GREEN CELLULAR NETWORKS 811
Li et al. [161] also proposed a traffic forecast scheme based and 3GPP TS 36.902 for cases and solutions) [41], [115]–
on transferred learning expertise from historical periods or [117], is intended to be gradually implemented in BSs along
neighboring regions, by formulating the traffic variation as a with the 4G standards including LTE and WiMAX. It adds
Markov decision process. The proposed scheme is then utilised automatic network management and intelligence features to
to minimize the energy consumption of cellular radio access the system and thus reduces costs, improves performance and
network with the help of BS sleeping operations. increases flexibility of the cellular system through network
optimization and reconfiguration processes. SON enables the
C. Techniques Enabling Sleep Mode in BSs BSs to adjust their own configuration when necessary without
human intervention, thus more operations such as timed sleep
In order for the sleep mode techniques to function, BSs mode, user location prediction and reverse channel sensing are
usually need to cooperate with each other. A BS controller possible in the system [41], [117]. Sleep mode in BSs is one of
(BSC) facilitates exchange of traffic information between BSs. the various applications of SON, where BSs are enabled to act
If certain BSs are selected to sleep, the sleeping BSs release collectively to save energy by redistributing traffic and sharing
their channel resources to active neighbors, while active BSs traffic information among BSs.
make use of obtained resources to cooperatively provide exten- There have been many publications on energy savings in
sive coverage to the mobile users located in the service areas LTE networks based on SON. For example, associations of BSs
of nearby sleeping BSs. When doing so, QoS requirements powered on or off can form collective network elements based
such as outage probability need to be carefully monitored on SON [98]. Cell sizes and on/off states can also be optimally
as the SNR (signal-to-noise ratio) changes with the distance adjusted [41]. In LTE, SON-based regulation schemes dynam-
between the mobile user and its serving BS, and more blocking ically minimize the number of active sectors in each evolved
events might occur due to insufficient available capacity in Node B (eNB, equivalent to a BS), and thus enable efficient
active BSs [114]. energy consumption [117]. SON-based traffic load redistribut-
User association, as the word implies, means associating ing strategies can also complement and support BS sleep mode
mobile end users with BSs in an energy efficient way without techniques by intelligently balancing or concentrating traffic at
the coordination from BSCs. Users originally connecting to appropriate times, resulting in more BSs available to be turned
BSs that went asleep need to be associated with new active BSs. off [118].
This process is required to ensure the QoS does not degrade Cell zooming or cell breathing is a similar concept to SON
significantly during BS sleeping operations. User association is but it provides higher level of flexibility. As introduced in
determined at a much faster time-scale than that of any dynamic [19], [119]–[121], cell zooming is a network layer technique
BS operations, which makes them two separated problems adaptively adjusting the cell size according to traffic conditions
with potential aggregated gains. It has also been noticed that by adjusting antenna tilt angles, height, or transmit power. It is
simply associating a user to the closest BS may be sub-optimal
much simpler than switching on/off BSs from an implementa-
when traffic distribution is inhomogeneous, because the closest
tion perspective [121]. It can be applied to balance the traffic
BS, if it is located in a low traffic area, may be preferable
load and reduce the energy consumption. When the traffic load
to be turned off [105]. Therefore, optimal user association,
in a certain cell increases, the cell will zoom in to reduce the
based on locations of users and BSs, average or instantaneous
coverage area and therefore avoid possible congestion. The
received signal quality (SINR) as well as traffic load, is an
“service hole” will be taken care of by the neighboring cells
essential condition for sleep mode schemes to be advantageous
with less traffic, which are supposed to zoom out. A cell
[21], [105]. Significant effort has been made on optimizing
zooming server, which can be implemented virtually at the
user association, for example, Dufkova et al. [101] formulate
gateway, or distributed in the BSs, controls the procedure of cell
the problem as an integer linear program in a users-cells
zooming. It sets the zooming parameters based on the traffic
affinity graph.
load distribution, user requirements, as well as channel state
Tabassum et al. [22] investigated the user association process
information. In fact, zooming in the cell coverage to 0 is equal
in BS sleeping operations. As compared to the conventional
to switch off the entire BS. Therefore, cell zooming can be
user association scheme based on maximum instantaneous
perceived as a generalization of BS sleep mode.
received signal power (MRSP), the authors proposed a new
Heterogeneous deployment is now common in cellular net-
user association mechanism based on maximum mean channel
works. The ultra-dense small cells are very energy consuming if
access probability (MMAP). In the proposed scheme, any user
not managed properly. Fortunately, the underlying macro cells
originally associated to a cell that went asleep chooses its
are able to provide coverage when the micro cells go to sleep
new associated cell based on maximum probability that it can
[122]. BS sleeping in heterogeneous cellular networks is thus
obtain a channel in the new cell. This probability depends
both feasible and desirable. More details of BS sleep mode in
on a number of factors, including traffic load in active BSs
heterogeneous networks will be covered in Section V.
prior to the sleeping operation, receiving signal power strength,
and cumulative interference power from neighboring BSs. The
authors showed that the proposed scheme and the conventional
D. Summary
scheme both have pros and cons in different scenarios, and
suggested that a hybrid approach of the two schemes may The fact that BSs have been designed to serve peak traffic
achieve best performance in terms of spectral and energy leads to wastage of energy during low traffic hours. Sleep mode
efficiency. operations exploit the opportunity by turning lightly loaded BSs
A technique called self-organizing network (SON), intro- to sleep and to save fixed part of energy consumption, e.g., air
duced in the 3GPP standard (see 3GPP TS 32.521 for definition conditioning. Traffic prediction and estimation techniques help
WU et al.: BASE-STATIONS SLEEP-MODE TECHNIQUES IN GREEN CELLULAR NETWORKS 813
TABLE III
S UMMARY OF T ECHNIQUES E NABLING BS S LEEPING
power saving protocols based on sleep mode of transceivers A stochastic analytical framework for BS sleeping in LTE
have been proposed for 4G standards. networks with OFDMA as the physical layer transmission
There have been proposals to lower power consumption of technology is presented in [127]. The authors extended the
LTE networks by exploiting DTX and DRX schemes to switch theoretical model quantifying the key metrics of the outage
certain energy consuming components in UE or BSs into sleep probability (e.g., SINR) and the average user capacity to ac-
mode in idle periods [59]. Basically, both DTX and DRX count for the effect of BS sleeping, such that optimal en-
reduce transceiver duty cycle while it is in active operation, ergy saving can be obtained while outage probability is held
but no packet is being transmitted. DRX focuses on the uplink constant. Furthermore, the authors proposed a modified non-
transmission and power of UE [69], [70], while DTX works on singular path loss model appropriate for small distance between
the downlink and thus it is relevant to energy consumption of the user equipment and the BS, which is more realistic for
BSs. Here we focus on DTX. micro cells.
Frenger et al. [17] introduced and discussed the feasibility of Dual connectivity is another new feature in LTE networks
the so-called cell DTX (where certain number of cells in a site that could provide new insights of sleep mode research [136].
are set to operate in DTX mode) in LTE by only transmitting It enables a mobile devoice to be simultaneously connected to
mandatory synchronization signals in a downlink radio frame, BSs in different tiers, for example, a macro cell and a micro
leaving six out of ten sub-frames empty. In this way, the cell. While the macro cells take the responsibility for the frame
radio transmitter can be turned off. They also compared the control, sending pilot signals and low-rate transmission, the
performance of the traditional cell micro DTX scheme and micro cell can focus on high rate transmissions. This dual
an enhanced cell DTX scheme. The enhanced scheme was connectivity allows longer sleep periods for micro cells as they
claimed to achieve 89% savings in a realistic traffic scenario do not need to wake up every time to transmit control messages.
compared to the scenario without cell DTX while the micro cell It is shown that such functionality separation scheme could
DTX provided an energy reduction of 61%. In a similar study, increase energy savings of sleep mode operations by more than
Wang et al. [133] proposed a novel time-domain sleep mode a third [137].
design in BSs. By optimally selecting the number of active Research on sleep mode in another 4G standard, WiMAX
subframes in each frame according to the traffic load, the energy (IEEE 812.16 family) is also widely available. Jang et al.
efficiency in LTE networks can be improved. The authors [138] introduced sleep mode for mobile subscriber stations
presented that an energy reduction of up to 90% can be obtained and relevant power saving strategies. WiMAX enables a sleep
at low traffic load. window size dynamically changes adaptively to traffic condi-
In [98], the authors considered the energy saving re- tions. If no traffic is destined to a sleep BS during its sleeping
configuration based on traffic conditions applied as an SON interval, the interval (window size) will be increased in the next
function in radio access networks. They further proposed the active-sleeping cycle, and vice versa. The authors stated that
concept of “energy partitions,” that is, to form associations of by optimally select initial, maximum and minimum window
powered on and off BSs by a collective decision of network sizes according to different traffic types, remarkable energy
elements. In a more recent work, Ghosh et al. [134] take saving can be achieved without a significant increase in delay.
traffic dependent energy consumption and penetration loss into Li et al. [44] also compared the performance of the periodical
consideration in an LTE configuration. The authors integrate discontinuous transmission (PDTX) scheme proposed in IEEE
sleep/active modes operations with optimization in antenna 812.16m with other novel proposed sleep mode schemes.
variables such as tilt, height, vertical beam-width and transmit
power, subject to constraints in the Signal to Interference and
C. BS Sleep Mode in General Cellular Networks
Noise Ratio (SINR), spectral efficiency and user throughput.
Bousia et al. [135] proposed another power saving algo- In contrast to DTX and DRX in 4G, the entire device or
rithm in LTE. The feature of their scheme is to switch off BS must be perceived as the smallest unit to be turned on
eNBs not only based on traffic load, but also according to or off in 2G and 3G networks. This has been studied more
the average distance of its associated users. The authors ar- extensively. The basic idea is when some BSs are switched
gue that greater average distance leads to greater transmission off, radio coverage and service provisioning are assumed to be
power, therefore this factor has to be taken into account when taken care of by the stations that remain active. In this way,
deciding which eNBs are to be switched off. The proposed service is guaranteed to be available over the entire service area
scheme outperformed the random switching off scheme in at all times.
terms of both the absolute energy saving and bit-per-joule Saker et al. [139] first proposed an energy-aware system
measurement. selection scheme that splits the mobile traffic between 2G and
CoMP (coordinated multi-point transmission) is another fea- 3G systems optimally, which can save up to 10% of total
ture of LTE standard which can also be utilized for sleep mode energy consumption while satisfying QoS requirements. They
applications. It eliminates the need for increasing transmission further proposed a sleep mode for either 2G or 3G systems. It
power of active BSs to maintain the coverage of sleeping turned out that in low to medium traffic hours, large amount of
BSs. An optimized approach based on CoMP is presented in energy saving can be achieved without significant degradation
[131] such that the amount of power saving less extra power in QoS. In a later study by the same research group [140],
consumed in backhaul and signal processing is maximized, by they developed a generic framework for applying sleep mode
selecting an optimized set of points for coordination. to mobile cellular networks. They proposed two schemes. The
WU et al.: BASE-STATIONS SLEEP-MODE TECHNIQUES IN GREEN CELLULAR NETWORKS 815
first scheme is a dynamic one where BSs are put to sleep or per day versus several progressive switch-offs (switching off
waken up based on the instantaneous number of users in the cell certain number of BSs at a time in increasing order of load)
(the time scale of sleep/wake corresponds to minutes), while per day. They also argued that when the number of switch-off
the second one is semi-static where the resources need to stay configurations per day increases, the complexity in operation
in a mode for some tens of minutes, or even for hours, in will also increase. By analyzing homogeneous networks and
order to minimize the sleep/wake commands. The numerical heterogeneous networks, as well as a case study given by a
results showed that the dynamic scheme led to much more realistic cell deployment, they reached the conclusion that the
energy saving in high traffic periods, while the performance extra energy saving gained by multiple switch-offs over single
of the two schemes was comparable in low traffic periods. switch off is only marginal. Thus they recommended limited
The same group of authors also discussed practical issues effort on the side of network management would be beneficial
for sleep mode implementation in BSs [102]. A guard period enough in terms of energy saving.
was proposed in order to avoid calls being blocked when the Niu et al. [119] discussed the performance evaluation of
resources are being activated (takes around 3–5 seconds). They centralized and distributed cell zooming algorithms and also
also introduced the hysteresis time, which keeps resources on compared them with static configurations. The results showed
for an additional period of time compared to the ideal sleep that the centralized algorithm performs superior to the dis-
mode case, where unnecessary sleep/wake switches happen tributed algorithm (ignoring the overhead), while both of them
due to minor traffic variations. Based on their simulation and outperform the static configuration. Son et al. [105] proposed
analysis, it is concluded that both guard interval and hysteresis and verified a framework for BS energy saving that encom-
time would provide better QoS while reduce the gain in energy passes user association and BS operation. A simple greedy
savings. Elayoubi et al. [103] studied practical implementation turning on and off algorithm is proved to perform close to the
issues of sleep mode operations in an HSDPA network with a optimal solution. Badic et al. [88] discussed the potential of
Markovian model. The authors derived an optimal switching energy saving by reducing cell size and including sleep mode
policy where energy saving is maximized while QoS is not feature in an HSDPA radio access network. Solely reducing the
degraded by solving a set of balance equations. cell size would reduce the ECR but does not change ECG,
BS sleep mode techniques have also been applied on real which can be however improved by powering-off unused cells.
mobile traffic profile to evaluate saving. Oh et al. [141] studied It is concluded in the paper that under constant user density, the
dynamically switching off scheme of redundant BSs during ECG is linearly correlated to the number of cells within a given
periods of low traffic, by analyzing temporal and spatial traces service area when BS sleep mode techniques are enabled.
of real cellular traffic in a part of Manchester, United Kingdom. Guo et al. [87] proposed three strategies for BS sleeping
The authors estimated that, by sharing and cooperating traffic on a queuing model. The three strategies differ in how BSs
between BSs, between 32 and 60 kWh of absolute power could detect incoming customers while sleeping. The authors noted
be saved for an area of roughly 12 square km in Manchester. that a number of parameters including delay constraint, BS
The energy savings can be further translated to 200 to 375 setup time and pre-determined time length for sleeping would
metric tons of annual carbon dioxide emission or about $42 000 affect the performances of different strategies. Particularly, if
to $78 000 on the bill for the owners of the BSs. They also sniffing cost is high enough, more complicated strategies may
suggested in the paper that cooperation between operators not necessarily outperform simple strategy even if optimal
would be even more profitable, particularly in metropolitan switching time is achieved.
areas where dense deployments are required for every operator. Network planning may influence the effect of BS sleep mode
Marsan et al. [93] discussed different switching-off schemes techniques. To achieve optimal savings, the locations of BSs
in cellular access networks. They compared different fixed must be planned carefully. Parameters including user density,
switching schemes, each of which switches off a different coverage area of single BS, inter-BS distance, number of active
fraction of cells when the traffic falls below a certain threshold. BSs and energy consumption are interlinked. Too high density
The results in the paper indicated that for different traffic of BSs would result in waste of energy while too low density
profiles, the optimal configuration is also different. If the rate simply would not suffice. Wu et al. [143] discussed energy
of change in traffic is high, then it is more desirable to turn efficiency planning of BSs in cellular networks that would in-
off a larger proportion of cells for shorter period of time rather crease potential energy savings. Notably, an analytical method
than turn off a smaller proportion of cells for longer period of is presented in the work to approximate real performance in
time. In a following study [142], the same authors discussed the various scenarios.
possibility of cooperation between multiple mobile operators,
where the users of the sleeping operator roam to the network of
D. BS Sleep Mode and Heterogeneous Network Deployment
the active provider without violating QoS constraints. Several
strategies, including balanced switch-off frequencies, balanced As discussed in Section I, heterogeneous network deploy-
roaming costs, balanced energy savings and maximum saved ment schemes were originally designed to improve the spectral
energy are discussed. It is shown that remarkable savings are efficiency in cellular networks by offloading traffic from classi-
possible if operators are willing to cooperate. The same group cal macro cells, and may lead to increase in energy consumption
of authors developed analytical models to identify optimal because of the extra cells deployed. Nevertheless, by intro-
fixed BS switch-off times as a function of the daily traffic ducing sleep mode in BSs, heterogeneous cellular networks
pattern [94]. They compared the cases of only one switch-off can now outperform traditional macro-cell-only counterparts
816 IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015
in terms of energy efficiency. During peak traffic hours, more an increasing function of energy consumption and a decreasing
energy-efficient smaller cells can be deployed, replacing some function of target throughput, a QoS measure. The switching
of macro cells in the macro-cell-only network. Then, those operation is added to the state space as a new dimension.
smaller cells are turned to sleep during light traffic hours when Apart from the straightforward case that BSs have complete
remaining macro cells are capable of maintaining throughput information of its associated traffic, optimal solutions have also
and coverage [129]. been found for partial traffic information (based on Partially
For example, Ashraf et al. [18] studied the application of Observable MDP) and delayed information (by transforming
dynamic sleep mode in BSs with pico-cell deployments. Het- their MDPs into equivalent MDPs without delay).
erogeneous network planning can improve the coverage of the In another study, Li et al. [44] proposed a clustering based
cellular network, but will likely result in even more severe over- power saving algorithm for self-organized sleep mode in femto-
provisioning and thus consume more energy if the cells are cell networks. In the cluster construction process, the leader of
unable to adapt to traffic load. The solution proposed by the each cluster is first elected based on the sum of received pilot
authors is to introduce the dynamic sleep and wake modes in the signal power and the distribution density, and then other femto
pico-cells. The result shows that the network with both macro- BSs determines whether they are to be attached to the leader
cells and pico-cells, where dynamic sleep mode algorithm is on the basis of the pilot signal power. The leader and members
applied in pico-cells, consumes less amount of energy than the in the same cluster will then exchange information collected by
network with only macro-cells. Cai et al. [128] proposed an a sniffer installed with each femto BS. The member will only
energy model for heterogeneous cellular network and a cross- turn on the pilot transmission and the processing if notified by
layer optimization method. Several pico cells (lower layer) the leader and the received signal energy rise above a certain
are in the coverage area of one macro cell (upper layer). The threshold within a predefined period of time. The proposed
problem to solve is how to associate users to the group of macro scheme is shown to outperform other three main traditional
cell and pico cells, so that energy consumption is minimized schemes.
after lightly-loaded pico cells turned to sleep. Similar models The locations of BSs, either macro or micro, in heteroge-
are also presented in [95] and [129]. neous networks are often assumed to follow a Poisson Point
Soh et al. [21] particularly addressed inter-tier interference Process (PPP) [127], [144]. In this regard, capacity extension
among multi-tier heterogeneous cells. In their work, macro by additional micro cells (increase density of BSs) and en-
BSs are modeled by a Poisson Point Process (PPP) while ergy saving by BS sleeping (decrease density of BSs) can be
users are distributed according to different stationary point pro- generalized into a single optimization problem on BS density
cesses. A Bernoulli trail-based random sleeping technique and based on the cost per micro BS. Cao et al. [144] illustrated
a strategic sleeping technique aiming at maintaining coverage numerical calculation to obtain the optimal BS density for both
are examined for both homogeneous and heterogeneous cases. homogeneous and heterogeneous networks.
The strategic sleeping, based on activity of macro cell users, is Huang et al. [122] considered a scenario where users in
designed to maintain or improve coverage probability of users macro cells and micro cells have different traffic patterns. They
as in the non-sleeping case. Numerical results show that random assume that micro cells serve hotspots with higher traffic vol-
sleeping harms rather than benefits energy efficiency of the ume. The authors investigated three energy saving approaches
system, as energy saved from sleeping mode is not sufficient to including micro cell BS sleeping and expansion/shrinking cov-
compensate for the decrease in coverage and data throughput. erage of micro cells (similar to cell zooming). The coverage
Strategic sleeping, on the other hand, actually improves both and power consumption of macro cells are held constant. It is
coverage and energy efficiency. The authors also showed that shown that each approach is effective under different traffic
gain in energy efficiency saturate as the density of smaller cells conditions. The crucial factor affecting the performances of
reaches a certain level. different approaches is traffic rate ratio, namely the ratio of
Wildemeersch et al. [126] investigated how small cell access traffic rate per unit area in hotspots to that in non-hotspots.
points (SAPs) can play a role enhancing energy efficiency of
heterogeneous cellular networks. Sleep mode of SAPs actually
E. BS Sleep Mode and Cooperative Relays
corresponds to the trade-off between energy consumption and
false alarm rate. The authors note that bursty transmissions The mobility of end users in cellular network can be also
from macro-cell traffic, due to mobility of users, makes duty exploited to provide coverage when some of BSs are in the
cycling of sleep mode in smaller cells more complicated. PPP sleep mode. As an alternative and complimentary method,
is also used to model the locations of SAPs and macro BSs. migrating traffic by moving nodes in cellular networks is more
Saker et al. [125] proposed a similar sleeping strategy that cost effective than increasing radiation power from neighboring
femto cells are switched off when the cell itself is not heavily active BSs, since typical relays usually consume negligible
loaded and the macro cell can serve the overall traffic without energy when compared to BSs. Another benefit of adopting
deteriorating the QoS. Based mainly on queuing theories, the relays is to postpone communication to future time instances
work utilizes a Continuous Time Markov Decision Process when an effective contact is available for message forwarding.
(CTMDP), in which states represent load status of each BS. For delay-insensitive traffic, the compromise in delivery time is
Every user brings a certain load to its connected BS. Each worthwhile for savings in power consumption. Relay selection
possible action for the state and transition probabilities is strategies involving which, when and where to relay need to be
assigned a value of rewards/cost. The cost function is defined as optimized to minimize overall energy consumption.
WU et al.: BASE-STATIONS SLEEP-MODE TECHNIQUES IN GREEN CELLULAR NETWORKS 817
Cao et al. [110] introduced wireless relays in addition to niques exclusively available in one particular standard, e.g.,
sleep mode in BSs. It is shown that, if the energy cost of DTX. Traffic load is always the major concern when deciding
wireless relays is small enough relative to that of BSs, the which BSs to switch off. Accordingly, a method that has been
system with wireless relays will outperform the one with pure found to be successful for one standard, may be considered for
BS cooperation strategy. Kolios et al. [145] considered mov- other standards.
ing vehicles as relaying nodes and propose a mathematical Heterogeneous deployment is expected to be more popular
programming formulation for finding the optimal forwarding in cellular networks due to denser traffic in hotspots. There-
policies that minimize the weighted sum of energy, delivery fore, research on integrating BS sleeping with heterogeneous
delay and fixed cost of operating the BSs. The authors argue deployment deserves to receive more attention. Applying green
that when the traffic is more delay-tolerant, it is more beneficial techniques such as BS sleeping, heterogeneous deployment,
to exploit the relays to migrating traffic between active and cooperative relay and energy harvesting could take advantage
sleeping cells. of each technique, e.g., heterogeneous deployment for higher
spectral efficiency and cooperative relay for even further reduc-
tion of power consumption.
F. BS Sleep Mode and Renewable Energy Resources
Renewable energy sources such as solar and wind are not
available at all times, constrained by natural condition. There- VI. P ERFORMANCE C OMPARISON
fore, it is essential to distinguish them from traditional energy
As we discussed in the previous section, there are many
sources when applying BS sleep mode techniques on a network
schemes, protocols and algorithms proposed to reduce energy
powered by both sources. When there is a lack of power supply,
consumption in cellular networks by implementing sleep mode
BSs operated by renewable energy resources could possibly not
or equivalent in BSs. While we have illustrated the performance
be able to support its associated traffic and might be forced
comparison between BS sleeping and other green technolo-
to switch off. To overcome this possible shortage, energy
gies in Section I, in this section we will compare the perfor-
harvesting technique is proposed to exploit renewable energy
mances of the different approaches within the scope of BS
sources [68], [130]. Spare energy from renewable resources is
sleeping.
stored and used when input power from the source does not
suffice. Furthermore, stored energy could be even used to main-
tain the coverage of sleeping conventional BSs, thus reduce
A. Fixed Switch Schemes vs. Dynamic Switch Schemes
(more expensive and unsustainable) energy consumption from
traditional sources. There have been several comparisons between fixed
The design principle for energy harvesting is to save energy schemes, where the number of switches in a certain period of
in conventional BSs, and minimize possible energy outage in time is fixed, and dynamic schemes, where the system switches
renewable energy operated BSs. In this work [130], a greedy as many times as necessary based on real-time or predicted
way of utilizing stored energy has been proved to be inefficient traffic information [94], [95], [119], [125], [140]. The results
and could lead to frequent outages. Instead, dynamic program- are generally in favor of dynamic schemes or of increasing fre-
ming algorithms based on the number of battery states in each quency of switching operations in fixed schemes. In [119], both
BS have been discussed to optimize network performance, the centralized and distributed dynamic algorithms outperform
evaluated by a weighted combination of blocking probability the static switching-off scheme (1/2 or 1/3 off). In [140], the
and energy consumption. semi static sleep mode prevails in lower traffic hours while the
dynamic sleep mode saves more energy under medium to high
traffic. In [110], the authors argues that performance of fixed
G. Discussions and Summary
and dynamic schemes depends partly on the traffic load of the
BS sleep mode techniques have a variety of potential appli- system. They show that in the coverage-limited (low traffic)
cations in modern cellular networks. Certain applications have region, offline fixed algorithms would be more suitable due to
been adopted by mobile operators. Shutting down the entire the constraint in coverage requirement, while in the energy-
BS during low traffic hours is commonly seen in industry. For efficient (medium traffic) region, online dynamic algorithms
example, the flagship network operator in the Caribbean region, are superior because the energy saving performance largely
Digicel Jamaica, has reported a energy saving of 23%, which is depends on instant traffic load.
equivalent to a carbon footprint deduction of 1.9kt or a cost According to [94], the amount of energy saving increases (al-
reduction of $1.4 million by adopting sleeping techniques in though only marginally) as the number of switching operations
BSs [146]. per day increases. Dynamic schemes generally require more
Newer techniques such as DTX in LTE, are currently more switching operations as compared to fixed schemes, especially
seen in research papers than actual operations. Nevertheless, the when the traffic pattern is unpredictable and highly fluctuated.
feasibility has been proven and the technology foundations for Therefore, another trade-off to consider is between more ab-
implementation are ready. Therefore, large potential remains solute energy saving in sleep mode and the cost of switching
untapped as market share of LTE grows in the future years. operations, which includes but is not limited to overhead,
The research methodologies are not significantly different transient time, delay constraint, extra power for monitoring and
among different network standards, with the exception of tech- switching, and impact to the operational lifetime of BSs.
818 IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015
TABLE V
C OMPARISON OF BS S LEEPING S CHEMES
TABLE VI
C OMPARISON OF M ODEL A SSUMPTIONS AND R EAL S CENARIOS
TABLE VII
Fig. 7. 49-cell hexagonal configuration network model with wrapped-around PARAMETERS FOR 49-C ELL N ETWORK S IMULATION
design [156].
challenges for research on energy saving approaches including of BS, and where BSs are densely deployed. Also, the larger
sleep mode. Methods based on spectrum management [158] the fixed proportion of energy consumption in BSs is (the case
and renewable energy resources [157] have been proposed. in macro BSs), the larger amount of saving can be possibly
For BS sleep mode, it is acknowledged the reduced symbol attained.
durations and duty cycle could increase the flexibility in imple- In summary, the sleep mode technologies as well as the
mentation [159]. However, no specific research has been done whole green cellular network are promising areas of research.
at the time of writing to the best of our knowledge. We believe It will probably remain a popular research topic for the coming
that sleep mode in 5G cellular networks could be a research years, since there are bright prospective as well as issues
area with many innovative ideas in the foreseeable future. waiting to be solved.
On the other hand, as mentioned before, it is possible to
adopt different green technologies together in cellular net-
works. Aggregated savings could probably be achieved because R EFERENCES
the technologies are not conflicting. Meanwhile, by implement- [1] GreenTouch. [Online]. Available: http://www.greentouch.org/
ing other techniques such as adjustments in antenna tilt and [2] Int. Telecommun. Union, “ITU Green Standards Week,” Geneva,
Switzerland, Sep. 2011. [Online]. Available: http://www.itu.int/ITU-T/
height, it is possible to offset the degradation in QoS due to climatechange/gsw/201102/index.html
sleeping BSs [134]. Incorporating different technologies may [3] G. Fettweis and E. Zimmermann, “ICT energy consumption—Trends
require significant effort by researchers with a wide range of and challenges,” in Proc. 11th Int. Symp. Wireless Pers. Multimedia
Commun., Lapland, Finland, Sep. 2008, pp. 1–4.
backgrounds, but the benefit will justify the effort. For example, [4] O. Blume, H. Eckhardt, S. Klein, E. Kuehn, and W. M. Wajda, “Energy
Chiaraviglio et al. [153] considered three network planning savings in mobile networks based on adaptation to traffic statistics,” Bell
strategies: the first one minimizes the number of transmitters Labs Tech. J., vol. 15, no. 2, pp. 77–94, Sep. 2010.
[5] Z. Hasan, H. Boostanimehr, and V. K. Bhargava, “Green cellular net-
(TX), the second one minimizes the power consumption (MP), works: A survey, some research issues and challenges,” IEEE Commun.
while the last one is the hybrid strategy (H) combining the pre- Surveys Tuts., vol. 13, no. 4, pp. 524–540, 2011.
vious two strategies. While the TX strategy effectively lowers [6] Y. Zhang, P. Chowdhury, M. Tornatore, and B. Mukherjee, “Energy
efficiency in telecom optical networks,” IEEE Commun. Surveys Tuts.,
the capital expenditure (CAPEX) for mobile operator in the vol. 12, no. 4, pp. 441–458, 2010.
BS planning stage, the MP strategy minimizes the operating [7] A. P. Bianzino, C. Chaudet, D. Rossi, and J. Rougier, “A survey of
expense (OPEX) by turning under-utilized BSs into sleep mode green networking research,” IEEE Commun. Surveys Tuts., vol. 14, no. 1,
pp. 3–20, 2012.
to reduce energy consumption. The result shows energy savings [8] D. Kimball et al., “High-efficiency WCDMA envelope tracking base-
between 11% and 28% are achievable even for already energy- station amplifier implemented with GaAs HVHBTs,” IEEE J. Solid-State
efficiency-optimized BSs deployment. Therefore, by adopting Circuits, vol. 44, no. 10, pp. 2629–2639, Oct. 2009.
[9] K. Cho, J. Kim, and S. Stapleton, “A highly efficient Doherty feedfor-
the H strategy, the total expenditure of the mobile operator ward linear power amplifier for W-CDMA base station applications,”
could be significantly reduced. IEEE Trans. Microw. Theory Techn., vol. 53, no. 1, pp. 292–300,
Jan. 2005.
[10] O. Arnold, F. Richter, G. Fettweis, and O. Blume, “Power consump-
IX. C ONCLUDING R EMARKS tion modelling of different base station types in heterogeneous cellular
networks,” in Proc. Future Netw. Mobile Summit, Dresden, Germany,
This paper has surveyed recent advancement in the research Jun. 2010, pp. 1–8.
[11] Y. Wei, J. Staudinger, and M. Miller, “High efficiency linear GaAs
of BS sleep mode techniques in cellular networks, which is MMIC amplifier for wireless base station and femto cell applica-
one of the major methods to reduce total energy consumption. tions,” in Proc. IEEE Topical Conf. PAWR, Tempe, AZ, USA, Jan. 2012,
Increasing energy consumption has been a major concern and pp. 49–52.
[12] C. Han et al., “Green radio: Radio techniques to enable energy-efficient
many studies on energy saving strategies in networking and wireless networks,” IEEE Commun. Mag., vol. 49, no. 6, pp. 46–54,
telecommunication sectors have been published after the first Jun. 2011.
work concerning the energy issue in the mainstream networking [13] H. Claussen, L. Ho, and F. Pivit, “Effects of joint macrocell and resi-
dential picocell deployment on the network energy efficiency,” in Proc.
area published in 2003, which refers to the greening of the IEEE 19th Int. Symp. PIMRC, Sep. 2008, pp. 1–6.
Internet [160]. The fact that cellular networks were designed [14] TriQuint, “RF Power Amplifier Module: TGA2450-SM.” [Online].
with the primary goal of maximum throughput and coverage Available: http://store.triquint.com/ProductDetail/TGA2450-SM-
TriQuint-Semiconductor-Inc/472292/
indicates significant gains are possible by effective power man- [15] J. Wu, S. Zhou, and Z. Niu, “Traffic-aware base station sleeping control
agement schemes. It is now widely accepted that the energy and power matching for energy–delay tradeoffs in green cellular net-
consumption reduction in cellular networks will bring both works,” IEEE Trans. Wireless Commun., vol. 12, no. 8, pp. 4196–4209,
Aug. 2013.
operating profit for mobile operators and environmental benefit [16] J. Louhi, “Energy efficiency of modern cellular base stations,” in Proc.
for the planet. The importance of this area has been therefore 29th INTELEC, Sep./Oct. 2007, pp. 475–476.
quickly realized and it is now a popular topic because of its huge [17] P. Frenger, P. Moberg, J. Malmodin, Y. Jading, and I. Godor, “Reducing
energy consumption in LTE with cell DTX,” in Proc. IEEE 73rd VTC
environmental, economic, financial and social impacts. Various Spring, Linkoping, Sweden, May 2011, pp. 1–5.
technologies to achieve the goal, as well as the broad scope of [18] I. Ashraf, F. Boccardi, and L. Ho, “Sleep mode techniques for small
knowledge required at different stage of implementation, make cell deployments,” IEEE Commun. Mag., vol. 49, no. 8, pp. 72–79,
Aug. 2011.
it open to researchers with diverse background. [19] G. Micallef, P. Mogensen, and H. Sch, “Cell size breathing and possi-
For BS sleep mode techniques, the main saving comes from bilities to introduce cell sleep mode,” in Proc. EW Conf., Lucca, Italy,
the fixed part of energy consumption in under-utilized BSs. The Apr. 2010, pp. 111–115.
[20] M. A. Marsan, L. Chiaraviglio, D. Ciullo, and M. Meo, “On the ef-
potential for energy saving is large in areas where the traffic is fectiveness of single and multiple base station sleep modes in cellular
highly variant and relative low as compared with full capacity networks,” Comput. Netw., vol. 57, no. 17, pp. 3276–3290, Dec. 2013.
WU et al.: BASE-STATIONS SLEEP-MODE TECHNIQUES IN GREEN CELLULAR NETWORKS 823
[21] Y. S. Soh, T. Quek, M. Kountouris, and H. Shin, “Energy efficient [47] Y. Chan, “China’s Huawei to supply solar-powered base sta-
heterogeneous cellular networks,” IEEE J. Sel. Areas Commun., vol. 31, tions to Bangladesh,” Business Green, London, U.K., Aug. 2009.
no. 5, pp. 840–850, May 2013. [Online]. Available: http://www.businessgreen.com/bg/news/1802444/
[22] H. Tabassum, U. Siddique, E. Hossain, and M. Hossain, “Downlink per- chinas-huawei-supply-solar-powered-base-stations-bangladesh
formance of cellular systems with base station sleeping, user association, [48] C. Okoye, “Airtel base stations to be solar powered,” Daily Times Nige-
scheduling,” IEEE Trans. Wireless Commun., vol. 13, no. 10, pp. 5752– ria, Lagos, Nigeria, Dec. 2011. [Online]. Available: http://dailytimes.
5767, Oct. 2014. com.ng/article/airtel-base-stations-be-solar-powered
[23] S. Cui, A. Goldsmith, and A. Bahai, “Energy-efficiency of MIMO and [49] X. Wang, A. V. Vasilakos, M. Chen, Y. Liu, and T. T. Kwon, “A survey
cooperative MIMO techniques in sensor networks,” IEEE J. Sel. Areas of green mobile networks: Opportunities and challenges,” Mobile Netw.
Commun., vol. 22, no. 6, pp. 1089–1098, Aug. 2004. Appl., vol. 17, no. 1, pp. 4–20, Feb. 2012.
[24] G. Li et al., “Energy-efficient wireless communications: Tutorial, survey [50] M. Ismail, W. Zhuang, E. Serpedin, K. Qaraqe, “A survey on green mo-
and open issues,” IEEE Wireless Commun., vol. 18, no. 6, pp. 28–35, bile networking: From the perspectives of network operators and mobile
Dec. 2011. users,” IEEE Commun. Surveys Tuts, to be published. [Online]. Avail-
[25] M. Butt et al., “On the energy–bandwidth trade-off in green wireless able: http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6949027
networks: System level results,” in Proc. 1st IEEE ICCC, Beijing, China, [51] GSMA, “Community power: Using mobile to extend the grid,”
Aug. 2012, pp. 91–95. London, U.K., Jan. 2010. [Online]. Available: http://www.gsma.com/
[26] Y. Zou, J. Zhu, and R. Zhang, “Exploiting network cooperation in mobilefordevelopment/wp-content/uploads/2012/05/Community-
green wireless communication,” IEEE Trans. Commun., vol. 61, no. 3, Power-Using-Mobile-to-Extend-the-Grid-January-2010.pdf
pp. 999–1010, Mar. 2013. [52] D. Everitt, “Traffic engineering of the radio interface for cellular mobile
[27] C. Xiong, G. Li, S. Zhang, Y. Chen, and S. Xu, “Energy- and spectral- networks,” Proc. IEEE, vol. 82, no. 9, pp. 1371–1382, Sep. 1994.
efficiency tradeoff in downlink OFDMA networks,” IEEE Trans. [53] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “NeXt
Wireless Commun., vol. 10, no. 11, pp. 3874–3886, Nov. 2011. generation/dynamic spectrum access/cognitive radio wireless networks:
[28] C. Xiong, G. Y. Li, S. Zhang, Y. Chen, and S. Xu, “Energy-efficient re- A survey,” Comput. Netw., vol. 50, no. 13, pp. 2127–2159, Sep. 2006.
source allocation in OFDMA networks,” IEEE Trans. Commun., vol. 60, [Online]. Available: http://www.sciencedirect.com/science/article/pii/
no. 12, pp. 3767–3778, Dec. 2012. S1389128606001009
[29] D. Feng et al., “A survey of energy-efficient wireless communications,” [54] Eur. Commission Release, “Digital agenda: Global tech sector measures
IEEE Commun. Surveys Tuts., vol. 15, no. 1, pp. 167–178, 2013. its carbon footprint,” Brussels, Belgium, Mar. 2013. [Online]. Available:
[30] A. Attar, H. Li, and V. C. M. Leung, “Green last mile: How fiber- http://europa.eu/rapid/press-release_IP-13-231_en.htm
connected massively distributed antenna systems can save energy,” IEEE [55] T. Langedem, “Reducing the carbon footprint of ICT devices, platforms
Wireless Commun., vol. 18, no. 5, pp. 66–74, Oct. 2011. and networks,” GreenTouch, Amsterdam, The Netherlands, Nov. 2012.
[31] Y. Zou, Y.-D. Yao, and B. Zheng, “Opportunistic distributed space–time [Online]. Available: http://www.greentouch.org/uploads/documents/
coding for decode-and-forward cooperation systems,” IEEE Trans. Van%20Landegem%20GeSI%20Reducing%20Carbon%20Footprint%
Signal Process., vol. 60, no. 4, pp. 1766–1781, Apr. 2012. 20May%202012.pdf
[32] H. ElSawy, E. Hossain, and M. Haenggi, “Stochastic geometry for mod- [56] Energy Aware Radio and neTwork tecHnologies (EARTH). [Online].
eling, analysis, design of multi-tier and cognitive cellular wireless net- Available: https://www.ict-earth.eu/
works: A survey,” IEEE Commun. Surveys Tuts., vol. 15, no. 3, pp. 996– [57] J. Manner, M. Luoma, J. Ott, and J. Hamalainen, “Mobile networks un-
1019, 2013. plugged,” in Proc. 1st Int. Conf. Energy-Efficient Comput. Netw., 2010,
[33] M. Li, P. Li, X. Huang, Y. Fang, and S. Glisic, “Energy consumption pp. 71–74.
optimization for multihop cognitive cellular networks,” IEEE Trans. [58] GSMA, “Mobile’s green manifesto,” London, U.K., 2012. [Online].
Mobile Comput., vol. 14, no. 2, pp. 358–372, Feb. 2015. Available: http://www.gsma.com/publicpolicy/wp-content/uploads/
[34] Y. Chen, S. Zhang, S. Xu, and G. Y. Li, “Fundamental trade-offs on green 2012/06/Green-Manifesto-2012.pdf
wireless networks,” IEEE Commun. Mag., vol. 49, no. 6, pp. 30–37, [59] L. Correia, “Challenges and enabling technologies for energy aware
Jun. 2011. mobile radio networks,” IEEE Commun. Mag., vol. 48, no. 11, pp. 66–
[35] G. He, S. Zhang, Y. Chen, and S. Xu, “Fundamental tradeoffs and 72, Nov. 2010.
evaluation methodology for future green wireless networks,” in Proc. [60] Energy Information Administration Brochures, “Greenhouse gases, cli-
1st IEEE ICCC, 2012, pp. 74–78. mate change, energy,” May 2008. [Online]. Available: http://www.eia.
[36] V. Chandrasekhar, J. Andrews, and A. Gatherer, “Femtocell networks: gov/oiaf/1605/ggccebro/chapter1.html
A survey,” IEEE Commun. Mag., vol. 46, no. 9, pp. 59–67, Sep. 2008. [61] M. Shakir et al., “Green heterogeneous small-cell networks: Toward
[37] A. R. Ekti, M. Z. Shakir, E. Serpedin, and K. A. Qaraqe, “Downlink reducing the CO2 emissions of mobile communications industry using
power consumption of HetNets based on the probabilistic traffic model uplink power adaptation,” IEEE Commun. Mag., vol. 51, no. 6, pp. 52–
of mobile users,” in Proc. IEEE 24th Int. Symp. PIMRC, Sep. 2013, 61, Jun. 2013.
pp. 2797–2802. [62] Int. Telecommun. Union, “Key global telecom indicators for the world
[38] F. Richter, A. Tehske, and G. Fettweis, “Energy efficiency aspects of base telecommunication service sector,” Geneva, Switzerland, Nov. 2011.
station deployment strategies for cellular networks,” in Proc. IEEE 70th [Online]. Available http://www.itu.int/ITU-D/ict/statistics/at_glance/
VTC Fall, Sep. 2009, pp. 1–5. KeyTelecom.html
[39] W. Guo and T. O’Farrell, “Green cellular network: Deployment solu- [63] MobiThinking, “Global mobile statistics.” [Online]. Available: http://
tions, sensitivity and tradeoffs,” in Proc. WiAd, London, U.K., Jun. 2011, mobithinking.com/mobile-marketing-tools/latest-mobile-stats
pp. 42–47. [64] G. Auer et al., “Enablers for energy efficient wireless networks,” in Proc.
[40] H. Claussen, I. Ashraf, and L. Ho, “Dynamic idle mode procedures for IEEE 72nd VTC-Fall, Sep. 2010, pp. 1–5.
femtocells,” Bell Labs Tech. J., vol. 15, no. 2, pp. 95–116, Sep. 2010. [65] T. Chen, H. Kin, and Y. Yang, “Energy efficiency metrics for green
[41] S. Bhaumik, G. Narlikar, S. Chattopadhyay, and S. Kanugovi, “Breathe wireless communications,” in Proc. Int. Conf. WCSP, Oct. 2010, pp. 1–6.
to stay cool: Adjusting cell sizes to reduce energy consumption,” in Proc. [66] M. Etoh, T. Ohya, and Y. Nakayama, “Energy consumption issues on mo-
1st ACM SIGCOMM Workshop Green Netw., Aug./Sep. 2010, pp. 41–46. bile network systems,” in Proc. Int. SAINT, Turku, Jul. 2008, pp. 365–368.
[42] C. Li, J. Zhang, and K. Letaief, “Energy efficiency analysis of small cell [67] M. Hoque, M. Siekkinen, and J. Nurminen, “Energy efficient multimedia
networks,” in Proc. IEEE ICC, 2013, pp. 4404–4408. streaming to mobile devices 2014: A survey,” IEEE Commun. Surveys
[43] A. Mukherjee, S. Bhattacherjee, S. Pal, and D. De, “Femtocell based Tuts., vol. 16, no. 1, pp. 579–597, 2014.
green power consumption methods for mobile network,” Comput. Netw., [68] H. Chen, Y. Li, J. Rebelatto, B. Filho, and B. Vucetic, “Harvest-
vol. 57, no. 1, pp. 162–178, Jan. 2013. [Online]. Available: http://www. then-cooperate: Wireless-powered cooperative communications,” IEEE
sciencedirect.com/science/article/pii/S1389128612003295 Trans. Signal Process., vol. 63, no. 7, pp. 1700–1711, Apr. 2015.
[44] W. Li, W. Zheng, Y. Xue, and X. Wen, “Clustering based power saving [69] A. Koc, S. Jha, R. Vannithamby, and M. TORLAK, “Device power
algorithm for self-organized sleep mode in femtocell networks,” in Proc. saving and latency optimization in LTE-A networks through DRX con-
15th Int. Symp. WPMC, 2012, pp. 379–383. figuration,” IEEE Trans. Wireless Commun., vol. 13, no. 5, pp. 2614–
[45] I. H. Rowlands, P. Parker, and D. Scott, “Consumer perceptions of “green 2625, May 2014.
power”,” J. Consum. Marketing, vol. 19, no. 2, pp. 112–129, 2002. [70] M. Gupta, S. Jha, A. Koc, and R. Vannithamby, “Energy impact of
[46] Y.-K. Chia, S. Sun, and R. Zhang, “Energy cooperation in cellular net- emerging mobile Internet applications on LTE networks: Issues and
works with renewable powered base stations,” in Proc. IEEE WCNC, solutions,” IEEE Commun. Mag., vol. 51, no. 2, pp. 90–97, Feb. 2013.
Apr. 2013, pp. 2542–2547.
824 IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015
[71] I. Humar et al., “Rethinking energy efficiency models of cellular net- [97] M. Parker, S. Nagraj, and S. Walker, “Absolute energy efficiency
works with embodied energy,” IEEE Netw., vol. 25, no. 2, pp. 40–49, metric for carbon footprint resource management and network optimi-
Mar./Apr. 2011. sation,” in Proc. Int. Conf. Clean Elect. Power, Capri, Italy, Jun. 2009,
[72] S. Taylor, R. Klemann, I. Verjee, and P. Ford, “Digging for the new pp. 111–116.
mobile gold: The next generation of mobile monetization,” White Paper, [98] K. Samdanis, D. Kutscher, and M. Brunner, “Self-organized energy
May 2014. [Online]. Available: ttp://www.cisco.com/c/en/us/solutions/ efficient cellular networks,” in Proc. 21st Annu. IEEE Int. Symp.
collateral/service-provider/mobile-Internet/white-paper-c11-731757. Pers., Indoor Mobile Radio Commun., Istanbul, Turkey, Sep. 2010,
pdf pp. 1665–1670.
[73] PCCW, “Environmental policy.” [Online]. Available: http://www.pccw. [99] L. Zhang, H.-C. Yang, and M. Hasna, “Generalized area spectral effi-
com/About+PCCW/Corporate+Social+Responsibility/Environment/ ciency: An effective performance metric for green wireless communica-
Environmental+Policy tions,” IEEE Trans. Commun., vol. 62, no. 2, pp. 747–757, Feb. 2014.
[74] Vodafone, “Carbon and energy.” [Online]. Available: http:// [100] H. Tabassum, M. Shakir, and M. Alouini, “Area green efficiency (AGE)
www.vodafone.co.uk/our-responsibilities/our-environment/carbon-and- of two tier heterogeneous cellular networks,” in Proc. IEEE GC Wkshps,
energy/ Dec. 2012, pp. 529–534.
[75] A. J. Fehske, G. Fettweis, J. Malmodin, and G. Biczok, “The [101] K. Dufkova, M. Bjelica, B. Moon, L. Kencl, and J.-Y. Le Boudec,
global footprint of mobile communications: The ecological and eco- “Energy savings for cellular network with evaluation of impact on data
nomic perspective,” IEEE Commun. Mag., vol. 49, no. 8, pp. 55–62, traffic performance,” in Proc. EW Conf., Apr. 2010, pp. 916–923.
Aug. 2011. [102] L. Saker and S. Elayoubi, “Sleep mode implementation issues in green
[76] W. Nordhaus, “To slow or not to slow: The economics of the greenhouse base stations,” in Proc. IEEE 21st Int. Symp. PIMRC, Istanbul, Turkey,
effect,” Econ. J., vol. 101, no. 407, pp. 920–937, Jul. 1991. Sep. 2010, pp. 1683–1688.
[77] 3rd Generation Partnership Project (3GPP). [Online]. Available: http:// [103] S. Elayoubi, L. Saker, and T. Chahed, “Optimal control for base station
www.3gpp.org/ sleep mode in energy efficient radio access networks,” in Proc. IEEE
[78] OPREA-Net, “Optimising power efficiency in mobile radio networks INFOCOM, Shanghai, China, Apr. 2011, pp. 106–110.
project.” [Online]. Available: http://opera-net.org/ [104] Z. Niu, J. Zhang, X. Guo, and S. Zhou, “On energy–delay tradeoff in
[79] Cognitive Radio and Cooperative Strategies for POWER saving in multi- base station sleep mode operation,” in Proc. IEEE ICCS, Singapore,
standard wireless devices (C2POWER). [Online]. Available: http://www. Nov. 2012, pp. 235–239.
ict-c2power.eu/ [105] K. Son, H. Kim, Y. Yi, and B. Krishnamachari, “Base station operation
[80] Wireless@KTH, “eWin: Energy-efficient wireless networking.” [On- and user association mechanisms for energy–delay tradeoffs in green cel-
line]. Available: http://www.wireless.kth.se/research/projects/19-ewin lular networks,” IEEE J. Sel. Areas Commun., vol. 29, no. 8, pp. 1525–
[81] Towards Real Energy-efficient Network Design (TREND). [Online]. 1536, Sep. 2011.
Available: http://www.fp7-trend.eu/ [106] K. Pentikosis, “In search of energy-efficient mobile networking,” IEEE
[82] C. Bontu, “DRX mechanism for power saving in LTE,” IEEE Commun. Commun. Mag., vol. 48, no. 1, pp. 95–103, Jan. 2010.
Mag., vol. 47, no. 6, pp. 48–55, Jun. 2009. [107] Y. Ko, K. Lee, and T. Nandagopal, “CAPS: A peer data sharing sys-
[83] “The energy consumption rating initiative, energy efficiency for network tem for load mitigation in cellular data networks,” in Proc. IEEE
equipment: Two steps beyond greenwashing (white paper),” Aug. 2008. GLOBECOM, Dec. 2003, vol. 2, pp. 656–660.
[Online]. Available: http://www.ecrinitiative.org/pdfs/ECR-TSBG_1_0. [108] Alcatel-Lucent, “9900 wireless network guardian,” White paper,
pdf Aug. 2012. [Online]. Available: http://www3.alcatel-lucent.com/wps/
[84] H. Kwon and T. Birdsall, “Channel capacity in bits per joule,” IEEE J. DocumentStreamerServlet?LMSG_CABINET=Docs_and_Resource_
Ocean. Eng., vol. 11, no. 1, pp. 97–99, Jan. 1986. Ctr&LMSG_CONTENT_FILE=White_PapersGuardian_EN_Tech_
[85] V. Rodoplu and T. Meng, “Bits-per-joule capacity of energy-limited WhitePaper.pdf
wireless networks,” IEEE Trans. Wireless Commun., vol. 6, no. 3, [109] C. Peng, S.-B. Lee, S. Lu, H. Luo, and H. Li, “Traffic-driven power
pp. 857–865, Mar. 2007. saving in operational 3G cellular networks,” in Proc. 17th Annu. Int.
[86] A. J. Fehske, P. Marsch, and P. Fettweis, “Bit per joule efficiency of co- Conf. Mobile Comput. Netw., Las Vegas, NV, USA, 2011, pp. 121–132.
operating base stations in cellular networks,” in Proc. IEEE GC Wkshps, [110] D. Cao, S. Zhou, C. Zhang, and Z. Niu, “Energy saving performance
Dresden, Germany, Dec. 2010, pp. 1406–1411. comparison of coordinated multi-point transmission and wireless relay-
[87] X. Guo, S. Zhou, Z. Niu, and P. Kumar, “Optimal wake-up mechanism ing,” in Proc. IEEE GLOBECOM, Miami, FL, USA, Dec. 2010, pp. 1–5.
for single base station with sleep mode,” in Proc. 25th ITC, Sep. 2013, [111] L. Suarez, L. Nuaymi, and J. Bonnin, “An overview and classification of
pp. 1–8. research approaches in green wireless networks,” EURASIP J. Wireless
[88] B. Badic, T. O’Farrell, P. Loskot, and J. He, “Energy efficient radio Commun. Netw., vol. 142, pp. 1–18, Apr. 2012.
access architectures for green radio: Large versus small cell size deploy- [112] Y. Zhang and A. Arvidsson, “Understanding the characteristics of cel-
ment,” in Proc. IEEE 70th VTC-Fall, Anchorage, AK, USA, Sep. 2009, lular data traffic,” in Proc. ACM SIGCOMM Workshop Cellular Netw.,
pp. 1–5. Oper., Challenges, Future Des., 2012, pp. 13–18.
[89] C. He, B. Sheng, P. Zhu, and X. You, “Energy efficiency and spectral effi- [113] G. Wang, C. Guo, S. Wang, and C. Feng, “A traffic prediction based
ciency tradeoff in downlink distributed antenna systems,” IEEE Wireless sleeping mechanism with low complexity in femtocell networks,” in
Commun. Lett., vol. 1, no. 3, pp. 153–156, Jun. 2012. Proc. IEEE ICC, Jun. 2013, pp. 560–565.
[90] S. Rivoire, M. Shah, P. Ranganthan, and C. Kozyrakis, “JouleSort: [114] F. Han, Z. Safar, W. Lin, Y. Chen, and K. Liu, “Energy-efficient cellular
A balanced energy efficiency benchmark,” in Proc. SIGMOD, Beijing, network operation via base station cooperation,” in Proc. IEEE ICC,
China, Jun. 2007, pp. 365–376. Ottawa, ON, Canada, Jun. 2012, pp. 4374–4378.
[91] EARTH, “EARTH Deliverable D2.4 Aware Radio Netw. Technol., [115] 3rd Generation Partnership Project, “3GPP Specification: 3GPP TS
Most suitable efficiency metrics and utility functions,” Jan. 2012. 32.521,” Telecommunication management; Self-Organizing Networks
[Online]. Available: https://bscw.ict-earth.eu/pub/bscw.cgi/d70454/ (SON) Policy Network Resource Model (NRM) Integration Reference
EARTH_WP2_D2.4.pdf Point (IRP); requirements, Cedex, France, 2012. [Online]. Available:
[92] A. P. Bianzino, A. K. Raju, and D. Rossi, “Apple-to-apple: A frame- http://www.3gpp.org/ftp/Specs/html-info/32521.htm
work analysis for energy-efficiency in networks,” ACM SIGMETRICS [116] Nokia-Siemens Networks, “Self-Organizing Network (SON): In-
Perform. Eval. Rev., vol. 38, no. 3, pp. 81–85, Dec. 2010. troducing the Nokia Siemens SON Suite—An efficient, future-
[93] M. A. Marsan, L. Chiaraviglio, D. Ciullo, and M. Meo, “Optimal energy platform for SON, Espoo, Finland. [Online]. Available: http://www.
savings in cellular access networks,” in Proc. IEEE ICC Workshops, nokiasiemensnetworks.com
Dresden, Germany, Jun. 2009, pp. 1–5. [117] M. Hossain, K. Munasinghe, and A. Jamalipour, “Toward self-
[94] M. Marsan, L. Chiaraviglio, D. Ciullo, and M. Meo, “Multiple daily organizing sectorization of LTE eNBs for energy efficient network op-
base station switch-offs in cellular networks,” in Proc. 4th ICCE, Hue, eration under QoS constraints,” in Proc. IEEE WCNC, Shanghai, China,
Vietnam, Aug. 2012, pp. 245–250. Apr. 2013, pp. 1279–1284.
[95] S. Morosi, P. Piunti, and E. Del Re, “Improving cellular network energy [118] L. Xu, Y. Chen, and Y. Gao, “Self-organizing load balancing for relay
efficiency by joint management of sleep mode and transmission power,” based cellular networks,” in Proc. IEEE 10th Int. Conf. CIT, Bradford,
in Proc. 24th TIWDC—Green ICT, Sep. 2013, pp. 1–6. U.K., Jun. 2010, pp. 791–796.
[96] Eur. Telecommun. Standard Inst., “Environmental engineering (EE) en- [119] Z. Niu, Y. Wu, J. Gong, and Z. Yang, “Cell zooming for cost-efficient
ergy efficiency of wireless access network equipment,” Sophia-Antipolis green cellular networks,” IEEE Commun. Mag., vol. 48, no. 11, pp. 74–
Cedex, France, ETSI TS 102 706-v 1.1.1, Aug. 2009. 79, Nov. 2010.
WU et al.: BASE-STATIONS SLEEP-MODE TECHNIQUES IN GREEN CELLULAR NETWORKS 825
[120] A. Baumgartner and T. Bauschert, “Greening cellular radio access net- works,” IEEE Trans. Wireless Commun., vol. 12, no. 9, pp. 4350–4362,
works: A numerical method for the selection of detachable base stations Sep. 2013.
in low traffic scenarios,” in Proc. 24th TIWDC—Green ICT, Sep. 2013, [145] P. Kolios, V. Friderikos, and K. Papadaki, “Switching off low utilization
pp. 1–6. base stations via store carry and forward relaying,” in Proc. IEEE 21st
[121] R. Balasubramaniam, S. Nagaraj, M. Sarkar, C. Paolini, and P. Khaitan, Int. Symp. PIMRC Workshops, Sep. 2010, pp. 312–316.
“Cell zooming for power efficient base station operation,” in Proc. 9th [146] GSMA, “Sleep mode for mobile base stations cuts Jamaican op-
IWCMC, Jul. 2013, pp. 556–560. erators’ carbon footprint,” London, U.K., Feb. 2013. [Online].
[122] K. Huang and V. Lau, “Enabling wireless power transfer in cellular net- Available: http://www.gsma.com/publicpolicy/sleep-mode-for-mobile-
works: Architecture, modeling and deployment,” IEEE Trans. Wireless base-stations-cuts-jamaican-operators-carbon
Commun., vol. 13, no. 2, pp. 902–912, Feb. 2014. [147] J. Gong, S. Zhou, Z. Niu, and P. Yang, “Traffic-aware base station sleep-
[123] S.-R. Cho and W. Choi, “Energy-efficient repulsive cell activation for ing in dense cellular networks,” in Proc. 18th IWQoS, Beijing, China,
heterogeneous cellular networks,” IEEE J. Sel. Areas Commun., vol. 31, Jun. 2010, pp. 1–2.
no. 5, pp. 870–882, May 2013. [148] M. Marsan, L. Chiaraviglio, D. Ciullo, and M. Meo, “Switch-off tran-
[124] U. Paul, A. Subramanian, M. Buddhikot, and S. Das, “Understanding sients in cellular access networks with sleep modes,” in Proc. IEEE ICC,
traffic dynamics in cellular data networks,” in Proc. IEEE INFOCOM, Kyoto, Japan, Jun. 2011, pp. 1–6.
Apr. 2011, pp. 882–890. [149] Z. Niu, “TANGO: Traffic-aware network planning and green operation,”
[125] L. Saker, S.-E. Elayoubi, R. Combes, and T. Chahed, “Optimal control IEEE Wireless Commun., vol. 18, no. 5, pp. 25–29, Oct. 2011.
of wake up mechanisms of femtocells in heterogeneous networks,” IEEE [150] Y. Zhang et al., “An overview of energy-efficient base station man-
J. Sel. Areas Commun., vol. 30, no. 3, pp. 664–672, Apr. 2012. agement techniques,” in Proc. 24th TIWDC—Green ICT, Sep. 2013,
[126] M. Wildemeersch, T. Quek, C. Slump, and A. Rabbachin, “Cognitive pp. 1–6.
small cell networks: Energy efficiency and trade-offs,” IEEE Trans. [151] S. Zhou, J. Gong, Z. Yang, Z. Niu, and P. Yang, “Green mobile ac-
Commun., vol. 61, no. 9, pp. 4016–4029, Sep. 2013. cess network with dynamic base station energy saving,” in Proc. ACM
[127] D. Tsilimantos, J.-M. Gorce, and E. Altman, “Stochastic analysis of MobiCom, Beijing, China, Sep. 2009, pp. 1–2.
energy savings with sleep mode in OFDMA wireless networks,” in Proc. [152] A. Conte, “Power consumption of base stations,” in Proc. TREND
IEEE INFOCOM, Apr. 2013, pp. 1097–1105. Plenary Meet., Feb. 2012, pp. 1–19.
[128] S. Cai, L. Xiao, H. Yang, J. Wang, and S. Zhou, “A cross-layer op- [153] L. Chiaraviglio, D. Ciullo, G. Koutitas, M. Meo, and L. Tassiulas,
timization of the joint macro- and picocell deployment with sleep “Energy-efficient planning and management of cellular networks,”
mode for green communications,” in Proc. 22nd WOCC, May 2013, in Proc. 9th Annu. Conf. WONS, Courmayeur, Italy, Feb. 2012,
pp. 225–230. pp. 159–166.
[129] P. Dini, M. Miozzo, N. Bui, and N. Baldo, “A model to analyze the [154] A. J. Fehske, F. Richter, and G. Fettweis, “Energy efficiency improve-
energy savings of base station sleep mode in LTE HetNets,” in Proc. ments through micro sites in cellular mobile radio networks,” in Proc.
IEEE GreenCom, IEEE iThings/CPSCom, Aug. 2013, pp. 1375–1380. IEEE GreenComm, Honolulu, HI, USA, Dec. 2009, pp. 1–5.
[130] S. Zhou, J. Gong, and Z. Niu, “Sleep control for base stations pow- [155] G. Auer et al., “Cellular energy efficiency evaluation framework,” in
ered by heterogeneous energy sources,” in Proc. ICTC, Oct. 2013, Proc. IEEE 73rd VTC Spring, Yokohama, Japan, May 2011, pp. 1–6.
pp. 666–670. [156] M. Zukerman, “Introduction to queueing theory and stochastic tele-
[131] G. Cili, H. Yanikomeroglu, and F. Yu, “Cell switch off technique com- traffic models,” 2014. [Online]. Available: http://www.ee.cityu.edu.hk/
bined with coordinated multi-point (CoMP) transmission for energy effi- zukerman/classnotes.pdf
ciency in beyond-LTE cellular networks,” in Proc. IEEE ICC, Jun. 2012, [157] L.-C. Wang and S. Rangapillai, “A survey on green 5G cellular net-
pp. 5931–5935. works,” in Proc. Int. Conf. SPCOM, Jul. 2012, pp. 1–5.
[132] R. Kumar and J. Gurugubelli, “How green the LTE technology can be?” [158] R. Hu and Y. Qian, “An energy efficient and spectrum efficient wireless
in Proc. 2nd Int. Conf. Wireless VITAE, Chennai, India, Mar. 2011, heterogeneous network framework for 5G systems,” IEEE Commun.
pp. 1–5. Mag., vol. 52, no. 5, pp. 94–101, May 2014.
[133] R. Wang, J. Thompson, and H. Haas, “A novel time-domain sleep [159] T. Levanen, J. Pirskanen, T. Koskela, J. Talvitie, and M. Valkama, “Radio
mode design for energy-efficient LTE,” in Proc. 4th ISCCSP, Limassol, interface evolution towards 5G and enhanced local area communica-
Cyprus, Mar. 2010, pp. 1–4. tions,” IEEE Access, vol. 2, pp. 1005–1029, Sep. 2014.
[134] P. Ghosh, S. Das, S. Naravaram, and P. Chandhar, “Energy saving in [160] M. Gupta and S. Singh, “Greening of the Internet,” in Proc. ACM
OFDMA cellular systems using base-station sleep mode: 3GPP-LTE a SIGCOMM, Karlsruhe, Germany, Aug. 2003, pp. 19–26.
case study,” in Proc. NCC, Kharagpur, India, Feb. 2012, pp. 1–5. [161] R. Li, Z. Zhao, X. Chen, J. Palicot, and H. Zhang, “TACT: A Transfer
[135] A. Bousia, A. Antonopoulos, L. Alonso, and C. Verikoukis, “‘Green’ Actor-Critic learning framework for energy saving in cellular radio sc-
distance-aware base station sleeping algorithm in LTE-Advanced,” in cess networks,” IEEE Trans. Wireless Commun., vol. 13, no. 4, pp. 2000–
Proc. IEEE ICC, 2012, pp. 1–5. 2011, Apr. 2014.
[136] D. Astely, E. Dahlman, G. Fodor, S. Parkvall, and J. Sachs, “LTE release
12 and beyond,” IEEE Commun. Mag., vol. 51, no. 7, pp. 154–160,
Jul. 2013. Jingjin Wu received the B.Eng. degree (with first-
[137] A. Zakrzewska, D. Lopez-Perez, S. Kucera, and H. Claussen, “Dual class honors) in information engineering (with fi-
connectivity in LTE HetNets with split control- and user-plane,” in Proc. nance minor) in 2011 from City University of
IEEE GC Wkshps, Dec. 2013, pp. 391–396. Hong Kong, Kowloon, Hong Kong, where he is
[138] J. Jang, K. Han, and S. Choi, “Adaptive power saving strategies for IEEE currently working toward the Ph.D. degree with the
802.16e mobile broadband wireless access,” in Proc. APCC, Busan, Department of Electronic Engineering. His current
Korea, Dec. 2006, pp. 1–5. research focuses on performance evaluation and en-
[139] L. Saker, S. Elayoubi, and H. Scheck, “System selection and sleep mode ergy conservation of cellular networks.
for energy saving in cooperative 2G/3G networks,” in Proc. IEEE 70th
VTC-Fall, Anchorage, AK, USA, Jan. 2009, pp. 1–5.
[140] L. Saker, S. Elayoubi, and T. Chahed, “Minimizing energy consumption
via sleep mode in green base station,” in Proc. IEEE WCNC, Sydney,
NSW, Australia, Apr. 2010, pp. 1–6.
[141] E. Oh, B. Krishnamachari, X. Liu, and Z. Niu, “Toward dynamic energy-
Yujing Zhang received the B.Eng. degree (with
efficient operation of cellular network infrastructure,” IEEE Commun.
Mag., vol. 49, no. 6, pp. 56–61, Jun. 2011. first-class honors) in information engineering (with
business intelligence minor) from City University
[142] M. A. Marsan and M. Meo, “Energy efficient management of two cellu-
of Hong Kong, Kowloon, Hong Kong, in 2012 and
lar access networks,” ACM SIGMETRICS Perform. Eval. Rev., vol. 37,
the M.S. degree in information networking from
no. 4, pp. 69–73, Mar. 2010.
[143] Y. Wu, G. He, S. Zhang, Y. Chen, and S. Xu, “Energy efficient coverage Carnegie Mellon University, Pittsburgh, PA, USA,
in 2014. She is currently a Software Engineer with
planning in cellular networks with sleep mode,” in Proc. IEEE 24th Int.
Facebook, Menlo Park, CA, USA. Her current re-
Symp. PIMRC, Sep. 2013, pp. 2586–2590.
[144] D. Cao, S. Zhou, and Z. Niu, “Optimal combination of base sta- search interests include telecommunication networks
and networked and distributed computer systems.
tion densities for energy-efficient two-tier heterogeneous cellular net-
826 IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015