SPECIAL SECTION ON GREEN COMMUNICATIONS AND NETWORKING FOR 5G WIRELESS

Received July 4, 2016, accepted July 21, 2016, date of publication August 10, 2016, date of current version September 28, 2016.
Digital Object Identifier 10.1109/ACCESS.2016.2598813

Survey of Strategies for Switching Off Base

Stations in Heterogeneous Networks
for Greener 5G Systems
FENGXIA HAN1 , SHENGJIE ZHAO2 , (Senior Member, IEEE), LU ZHANG3 ,
AND JINSONG WU4 , (Senior Member, IEEE)
1 TongjiUniversity, Shanghai 201804, China
2 KeyLaboratory of Embedded System and Service Computing, Ministry of Education, Tongji University, Shanghai 200092, China
3 Nokia Shanghai Bell Company, Ltd., Shanghai 201206, China
4 Department of Electrical Engineering, Universidad de Chile, Santiago 8370451, Chile

Corresponding author: S. Zhao ([email protected]).

This work was supported in part by the National Basic Research Program of China (973 Program) under Grant 2014CB340404 and in part
by the National Science Foundation of China under Grant 61471267.

ABSTRACT For heterogeneous network, which has been viewed as one pioneering technology for making
cellular networks be evolved into 5G systems, reducing energy consumption by dynamically switching
off base stations (BSs) has attracted increasing attention recently. With aiming at optimization on energy
saving only or another energy-related performance tradeoffs, several BS switch-off strategies have been
proposed from different design perspectives, such as random, distance-aware, load-aware, and auction-
based strategies. Furthermore, work has been done to consider joint design for BS switch-off strategy and
another strategies, such as user association, resource allocation, and physical-layer interference cancellation
strategies. Finally, there have been research results about this topic in emerging cloud radio access networks.
In this paper, we take an overview on these technologies and present the state of the art on each aspect. Some
challenges that need to be solved in this research filed for future work are also described.

INDEX TERMS Energy consumption, BS switch-off (or called sleeping) strategy, heterogeneous networks,
cloud radio access networks, greener 5G systems.

I. INTRODUCTION reduced latencies, and higher uplink and downlink data rate
Recently, there has been an explosion in mobile data due to end users.
to the popularization of smart phones and tablets. Since the On the other hand, processing an explosively increased
traffic load on conventional cellular networks is predicted amount of mobile data traffic in 5G systems will also bring
to be increased by 1000 times in the next 10 years [1], the ever-increasing energy consumption and carbon footprint to
fifth generation (5G) of mobile communication networks the mobile communication industry. In particular, the whole
will have to deal with an augmented amount of mobile data information and communication technology (ICT) industry
traffic, which is expected to reach 24.3 exabytes per month has been estimated to contribute to about 2% of global
by 2019 [2], [3]. Also, the major technical challenges of CO2 emissions, to which the mobile communication industry
5G systems include 0.1 ∼ 1 Gbps user experienced data rate, contributes 15-20% [8]. With increasing awareness of the
tens of Gbps peak data rate, tens of Tbps/km2 traffic volume potential harmful impact on environment and the depletion of
density, ms-level end-to-end latency, 1 million connections non-renewable energy sources, establishing greener mobile
per squared kilometer, etc [4]. One promising approach to communication networks has become an economic issue
tackle the challenge is to exploit the heterogeneous networks and a big challenge for sustainable development [9], [10].
(HetNets), where traditional macro base stations (MBSs) In particular, 100 times energy efficiency improvement has
are deployed to provide the coverage for large areas while been proposed as another technical challenge in the design of
overlaid lower-powered small base stations (SBSs) are used 5G systems [4].
to cover relatively small areas (e.g., hot spots) [5]–[7]. Specifically, according to some surveys on energy con-
Being constructed with different tiers of small cells and sumption (e.g., see [11], [12]), 80% of energy consumption
large cells, HetNets can deliver the increased bandwidths, in mobile communication networks is due to the operation

2169-3536 2016 IEEE. Translations and content mining are permitted for academic research only.
VOLUME 4, 2016 Personal use is also permitted, but republication/redistribution requires IEEE permission. 4959
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 F. Han et al.: Survey of Strategies for Switching Off BSs

of BSs. Further, based on the results from laboratory tests A. OPTIMIZATION ON ENERGY OR POWER
done by China Mobile Communications Corporation, a BS CONSUMPTION
consumes 100% energy at the state with the maximum traffic The most popular optimization objective is energy or power
load and about 50%-60% energy at the state with zero traffic consumption. Early research in [19] has applied a general
load, while the energy consumption of a BS can be reduced energy model to HetNets and evaluated the energy saving
to 40% if it is switched off (i.e., in the sleeping state). There- in an urban scenario. However, it was not considered in [19]
fore, an effective way to achieve energy saving in mobile that the energy consumed by a BS is generally not constant
communication networks is to dynamically switch off BSs, but variable with the traffic load. For example, when a SBS
especially for scenarios with low traffic load where less BSs is switched off, the coverage-overlapped MBS will take over
can meet the traffic needs of all user equipments (UEs) [13]. the traffic load of this SBS so that its energy consumption is
A traditional BS consists of baseband unit (BBU) for increased. Illustrated by Fig. 1 [20], the energy model used
baseband signal processing and remote radio head (RRH) in [20] gets rid of the above limitation by assuming that the
for transmitting/receiving radio signals. When a traditional energy/power consumption of a MBS is linearly varied with
BS is switched off, BBU and RRH of this BS would be its traffic load while the energy/power consumption of a SBS
switched off together. In contrast, in cloud radio access net- is constant, thus allowing to perform a more practical analysis
works (CRANs) which would be investigated and pursued in for the performance of BS switch-off strategy in HetNets.
5G systems, BBUs of several traditional BSs are centralized
in a single location and the corresponding BBU resources
are sliced via virtualization technologies, while RRHs are
left at cell sites. With this kind of system architecture, the
switch-off operation for RRHs and virtual BBUs could be
done separately, through combination with flexible resource
allocation on virtual BBUs.
Many works (e.g., see summary works in [14]–[16])
have reported the switch-off (or called sleeping) mech-
anism for traditional BSs in HetNets and verified the
resulting benefit in terms of energy saving with realis-
tic data. Further, there have also existed research efforts
(e.g., see [17], [18]) for BS switch-off mechanisms in het-
erogeneous CRANs (H-CRANs), where the system archi-
tecture of CRAN is applied to HetNets. Motivated by these
research works and the goal of building greener 5G systems
with HetNets being deployed, in this paper, the state of the
art in strategies for design on switching off BSs are intro-
duced. Some challenging issues for future work in the related
research areas are also discussed. FIGURE 1. Power consumption of MBS (denoted as PM ) and Power
consumption of SBS (denoted as PS ) as a function of the traffic load [20].
The rest of this paper is organized as follows. Section II will (PM = a × load + WM where a is linear coefficient and WM is the basic
take an overview on the choice for optimization objective and circuit power consumption of a MBS when there is no traffic load,
PS = WS where WS is a constant independent of traffic load.)
constraint of the optimization problem used for designing BS
switch-off strategy, and analyze their characteristics. Then,
section III will describe specific BS switch-off strategies Further, in [21], the power consumption of a SBS is mod-
which are designed for HetNets according to different crite- eled in a more realistic way, which can be found as below in
ria. In Section IV, with focusing on HetNets, we will discuss equation (1):
joint design for BS switch-off strategy and other strategies 
P0 + 1m PTm , active mode
which are also important for 5G systems. In Section V, PSBS = (1)
Psleep , sleep mode
the BS switch-off strategies designed for either CRANs or
H-CRANs will be reviewed. Finally, conclusions and future where P0 is the basic circuit power consumption of a SBS in
work are presented in Section VI. the active mode, 1m is the proportionality coefficient of load
dependent power consumption, PTm is the transmission power
II. OPTIMIZATION OBJECTIVES AND CONSTRAINTS of a SBS, and Psleep denotes the power consumption when
FOR DESIGNING BS SWITCH-OFF STRATEGY SBS is in the sleep mode (i.e., being switched off).
Appropriate performance evaluation metrics are of primary
importance in the design on BS switch-off strategy, because B. OPTIMIZATION ON TRADEOFF BETWEEN ENERGY
they are directly related to the choice for the optimization CONSUMPTION AND QoS INDICATORS
objective and constraint of the corresponding optimization Besides considering the optimization on only energy or power
problem. consumption for designing BS switch-off strategy, efforts

4960 VOLUME 4, 2016

F. Han et al.: Survey of Strategies for Switching Off BSs

have been made to optimize the tradeoff between energy of Ptotal and D. As shown in [27], based on utilizing Markov
consumption and data throughput. With focusing on HetNets state transfer diagram to derive the proper expression for
where SBSs have multi-level sleeping modes, the tradeoff traffic delay, Fig. 2 illustrates the energy-delay tradeoff for
between energy consumption and data throughput is quanti- different average distances between the MBS and SBSs with
fied in [22] using an analytical model, where a metric named a given traffic load of SBS.
‘‘energy efficiency (EE)’’ and with the unit of bits per joule
is derived as below:
R
EE = (2)
P
where R (with the unit of bits per second) and P (with the
unit of Watt) denote overall throughput and overall power
consumption, respectively. In [22] and [23], EE maximization
has been considered in HetNets to design the optimal sleeping
strategy for MBSs (in [23]) or for SBSs (in [22]). Further-
more, in [24], in order to perform joint design on BS sleeping
strategy and power control strategy for self-powered SBSs, an
optimization problem is formulated to maximize the channel
capacity (i.e., the theoretic upper bound of data throughput)
with using energy consumption and energy storage limit in
the self-powered battery as optimization constraints.
Inspired by using the tradeoff between energy consump-
tion and data throughput as the optimization objective, FIGURE 2. Energy-delay tradeoff for different densities of SBS
people have considered another energy-related tradeoffs in deployment (denoted by average distances between the MBS and SBSs,
DMBS-SBS ) when the average traffic load of SBSs is given as ρ = 0.30,
the design on BS switch-off strategy (e.g., see [25]–[29]). where ρ = λ/µ with λ being the total arrive rate of flows and µ being the
Those could be the tradeoffs between energy/power con- serving rate in the modeled Markov process [27].
sumption and quality-of-service (QoS) indicators except for
data throughput (such as traffic delay, coverage probability, In addition, to design BS switch-off strategy, some research
outage probability). For example, when a SBS is turned off, works also consider the tradeoffs between energy/power con-
UEs originally served by this SBS have to wait until this SBS sumption and QoS indicators related to successful probability
wakes up or need to be re-associated to the MBS or another of transmissions over wireless links (e.g., see [28], [29]). With
SBSs, leading to longer service delay. considering a homogeneous network, the work in [28] con-
There have been research efforts in considering energy- ducts the design for BS switch-off strategy through consid-
delay tradeoff (EDT) in the design on BS switch-off strat- ering to reduce the power consumption of BSs and minimize
egy (e.g., see [25]–[27]). With setting the average system the increment of sum power of UEs, where a constraint on
power consumption as optimization objective and using the outage probability is used. In [29], with assuming that either
mean delay as optimization constraint, the work in [25] per- MBSs or SBSs in the considered HetNet could be switched
forms design for control parameters of switching off SBSs off, the QoS indicator like coverage probability is used as the
(e.g., SBS operation time, user number in queue). Further, optimization objective, in order to perform joint design on
the authors of [26] build a more complicated multi-objective the control parameters for a given BS switch-off strategy and
optimization problem where the mean power consumption some given user association strategies.
and the mean traffic delay are two optimization objectives,
and exploit the explicit relationship between these two opti- C. OPTIMIZATION ON ANOTHER
mization objectives to derive the optimal control parameters ENERGY-RELATED TRADEOFF
of sleeping operation for SBSs (including close-down time, The tradeoff between energy/power consumption and
sleeping period, and set-up time). In [27], similar as defin- energy/power cost caused by the transition between sleeping
ing EE to represent the energy-throughput tradeoff in one mode and active mode is considered in [30]. Specifically,
mathematical expression, the EDT is quantified as below for similar as the way used in equation (3) to quantify the EDT,
minimization: this tradeoff is represented as below for minimization in [30]:
ϕ(N ) = Ptotal + ηD (3) upp
f (vk , vlow
k )=P
net
+ ξ Pcost (4)
where N is the threshold used to switch off or awaken a SBS where SBS k will be switched on when the number of its
upp
in terms of the traffic flows which could be served by the served users reaches vk (i.e., the upper threshold) and will
SBS, Ptotal denotes average system power consumption and switch off when the user number is less than vlow
k (i.e., the
is a function of N , D denotes average delay and is a function lower threshold), Pnet denotes network power consumption,
of N , and η is the weight that controls the relative proportion Pcost denotes mode transition power consumption, and ξ is the

VOLUME 4, 2016 4961

 F. Han et al.: Survey of Strategies for Switching Off BSs

TABLE 1. Optimization objectives and constraints for designing BS switch-off strategy.

parameter that balances the tradeoff between Pnet and Pcost . of operators to SBSs of the third parties during peak hours.
Further, in [30], the average number of users served by active When considering one mobile network operator (MNO) and
SBSs and the average data rate for users served by active multiple third parties who can rent out their SBS resource,
SBSs are used as optimization constraints. the works in [31] and [32] perform design on BS switch-off
Also, a novel tradeoff between energy/power consumption strategy by setting the total financial cost of the MNO and all
and some certain financial cost or benefit could be consid- third parties as the optimization objective. In [33], under the
ered (e.g., see [31]–[33]), which is applicable for the emerg- situation that multiple MNOs and one single third party are
ing situation that operator(s) lease capacity and bandwidth considered, besides traditional energy consumption, another
resources from one or multiple third parties who deploy two finance-related optimization objectives are considered,
lower-powered SBSs to enable offloading traffic from BSs including the profit of the third party and profit of each MNO.

4962 VOLUME 4, 2016

F. Han et al.: Survey of Strategies for Switching Off BSs

D. SUMMARY AND FUTURE WORK improvement of approximately 30% in EE with using random
The comparative summary for the literature reviewed in this sleeping strategy.
section can be seen in Table 1. Although the random switch-off strategy takes less opera-
Summary: Further, the four main findings from the above tional cost and low computational complexity; however, this
discussions are: strategy lack adaptability to varying conditions in realistic
1) Energy/power consumption is the most popular compo- mobile networks (such as traffic load, MBS-SBS distance or
nent when choosing optimization objective and constraint of BS-UE distance).
the optimization problem used for designing BS switch-off
strategy. B. DISTANCE-AWARE STRATEGY
2) Considering the tradeoffs between energy/power con- Inspired by the work in [38] which designs more intelligent
sumption and QoS indicators (such as data throughput, BS switch-off strategy based on the distance between UEs
service delay, coverage probability, outage probability) will and their associated BS for a network without HetNet deploy-
generate more comprehensive and flexible choices on opti- ment, an optimal distance-based switch-off algorithm for
mization objective and constraint, which will make the SBSs is proposed in [39] for HetNets. In [39], the operation
resulted design on BS switch-off strategy be more effective. modes of the SBSs (i.e. switch-off or active) are decided
3) The transition between sleeping mode and active mode according to their distances to the MBS; specifically, the
for a BS would increase the operational cost, for example, SBSs closer to the MBS are turned off gradually to minimize
additional energy/power consumption is required to switch the total power consumption in HetNets.
on or off the hardware. It is a more realistic way to consider Based on firstly utilizing one well-designed random
this kind of mode transition cost in the design on BS switch- switch-off strategy for SBSs (as described in Section III-A)
off strategy. to determine the desired number of SBSs in different modes
4) The involvement of various entities which have dif- (such as active and sleeping), a distance-aware switch-off
ferent financial goals generates a new tradeoff between strategy for SBSs is further proposed in [22] for HetNets. For
energy/power consumption and some certain financial cost this strategy, the number of UEs in the coverage of each SBS
or profit. and the distance from the nearest UE to each inactive SBS are
Future Work: In the future, besides the energy-related considered for a static traffic model, while the number of UEs
tradeoffs already investigated, the infrastructure cost or back- in the coverage of each SBS and the dynamics of the distance
haul overhead can be further taken into account in the design from the nearest UE to each inactive SBS (in terms of velocity
on BS switch-off strategy in HetNets. and moving direction) are considered for a dynamic traffic
model. Simulation results in [22] show that the improvement
of approximately 45% in EE can be achieved by using this
III. OVERVIEW OF SPECIFIC BS SWITCH-OFF
distance-aware sleeping strategy.
STRATEGIES IN HETNETS
It needs be realized that, due to some certain reasons (such
In HetNets, BS switch-off strategies could be designed
as technology limits or designing principles), the information
either for SBSs or for the coverage-overlapped MBS. Of
about BS-UE distances may not be able to be accurately or
course, switching off MBS may have non-negligible negative
easily obtained. Therefore, it is meaningful to seek for more
impact on the network coverage. Moreover, when compared
feasible switching methods.
with the power consumption of a MBS which significantly
increases with its traffic load, the power consumption of a C. LOAD-AWARE STRATEGY
SBS is relatively flat and is not very closely related to its
The distribution of traffic load at different geographical areas
load, so that the aggregated power consumption of SBSs
would not be uniform in realistic networks. Fortunately, such
could be larger than the MBS. Thus, more efforts have
spatial traffic load fluctuation creates the opportunity to save
been put by people into the design on switch-off strategies
energy significantly, by switching off the underutilized SBSs.
for SBSs.
In [40], a load-aware approach is used to switch off a SBS
when its traffic load is below a certain threshold for a certain
A. RANDOM STRATEGY period. It is observed in [40] that, by incorporating load
As we all know, the easiest strategy for switching off BSs balancing between different types of BSs into the design on
is to switch off each BS independently at a certain probabil- switch-off strategies for SBSs, EE improvements up to 68%
ity, which has been investigated without considering HetNet for low traffic load and up to 33% for medium traffic load
deployment (e.g., see [34]–[36]). can be achieved in HetNets. In [41], a heuristic algorithm is
The work in [37] focuses on random BS switch-off strategy developed for determining which BS to be switched on or off,
in HetNets, where the optimal switch-off probability for MBS based on the number of UEs that each BS in the considered
is derived based on minimizing BS energy consumption. HetNet can serve. However, it requires the knowledge on the
In contrast, in [22], random switch-off strategy for SBSs is load and location of all UEs, making this strategy infeasible
designed based on maximizing EE with the constraint for in realistic networks. In [42], with taking user mobility into
coverage probability. Simulation results in [22] indicate the account, an optimal load-aware strategy is derived as below:

VOLUME 4, 2016 4963

 F. Han et al.: Survey of Strategies for Switching Off BSs

when the load is not high, the MBS can alone handle the arrival be modeled as a Poisson process. The corresponding
traffic and all the SBSs could be switched off; as the load periodic sinusoidal traffic profile could be further obtained
increases, one or more SBS(s) will be switched on depending easily.
on the estimated load and localization info of traffic. Then, The coexistence of multiple MNOs in the same geo-
the behavior of this strategy is investigated for the case with graphic area, which has not been considered in another works,
complete traffic localization information and the case with motivates the research work in [46], where a traffic load
partially known or delayed traffic localization information. based switch-off strategy is proposed for BSs (including
The traffic load will also have fluctuation over time. As for MBSs and SBSs) with the help of cooperation between
time-varying traffic load, if we consider all UEs in a given MNOs, and the QoS of re-associated UEs is guaranteed
urban area, the distribution of these UEs might fluctuate by incorporating the roaming-based infrastructure sharing
during one day, resulting in various traffic states. The sim- among the rival MNOs. Furthermore, many aforementioned
plest BS switch-off strategy is proposed in [43] based on works have assumed that there exists a centralized entity
only traffic statistics over time, where a SBS is switched off in the network. This centralized entity will be used to
based on a fixed timer and this timer is manually config- collect the required information over the whole network
ured for a statistical cycle when the traffic load is very low for implementing the designed BS switch-off strategy, thus
(e.g., during some time periods at night). More adaptive causing tremendous communication overhead. With focus-
sleeping strategies based on dynamic traffic monitoring have ing on decentralized implementation, the authors of [47]
been discussed in [44] and [45] for HetNets. In [44], based propose a practical BS switch-off strategy, in which the
on minimizing the cumulative energy consumption which is exchanged information for coordination between neighboring
obtained by considering time-varying traffic load, the switch- BSs (including the load management information) complies
off strategy for a SBS is derived for complete information with the X2 interface defined in the 3GPP LTE-Advanced
case and incomplete information case, respectively. With the standard. In particular, the always active MBS (called cov-
purpose of adapting the necessary network resource to the erage BS in [47]) will be in charge of switching off SBSs
actual traffic demand at the moment, a heuristic algorithm and (called capacity BSs in [47]) in the order of their traffic
a progressive algorithm are introduced in [45] to dynamically load.
switch off the unnecessary SBSs in HetNets. Both algorithms
can well track the traffic variation over time. In [45], it is
assumed that the daily traffic profile of the whole HetNet is D. AUCTION-BASED STRATEGY
the same and repeats periodically, which can be approximated To avoid significantly increased capital and operational
by a sinusoidal-like periodic behavior. As shown in [45], expenditures (CapEx and OpEx) resulted by densely
Fig. 3 illustrates the traffic curve when assuming the statis- deployed SBS while meeting the traffic demands in an eco-
tic traffic information is available and making the service nomical way, MNOs may consider leasing SBS resources
from one or multiple third parties. Focusing on the case with
one MNO and multiple third parties, authors in [31] and [32]
propose an auction-based offloading scheme for implement-
ing the switch-off of the MNO’s BSs. In particular, the
corresponding reverse combinatorial auction problem is for-
mulated using the integer linear programming model, which
provides the optimal allocation for the auction, namely the
SBSs to be rented out and the mobile data traffic that can be
offloaded. Of course, the works in [31] and [32] assume the
capacity of SBSs rented by third parties is sufficient to serve
the offloading demand of the MNO. However, it is shown
in [48] that the presence of multiple MNOs in the same area
will make the situation become more complicated, since the
SBSs may not be able to fully support the offloading demand
of all MNOs.
For the case with multiple MNOs and one single third
party, when each MNO proposes a single bidding value for
its requested capacity (which is based on the maximum pre-
diction about the traffic that it is willing to offload), this
FIGURE 3. Average daily network traffic variations of UE service arrival
rate [45]. Here, the service arrival is modeled as a Poisson process with prediction may not always correspond to the actual traffic
intensity ρ(t ) = Ab [1 + sin( πt
12
+ Bπ )]b + C where A denotes the volume, which will lead to decreased benefit for the operators
2
parameter that controls the amplitude of the traffic profile, B means the and inefficient use of the third party’s SBS resources. Moti-
phase of traffic curve which regulates the position of the peak, C denotes
the minimum traffic intensity in the network, b ∈ {1, 3} is used to vated by this issue, the authors of [33] propose to let each
modulate the gradient of traffic profile curve. of multiple MNOs submit a set of bidding values to the third

4964 VOLUME 4, 2016

F. Han et al.: Survey of Strategies for Switching Off BSs

TABLE 2. Overview of specific BS switch-off strategies in HetNets.

party, which request different capacity resources provided by interests of the involved parties are also considered in [33],
SBSs of the third party. Further, in [33], an auction-based so as to the optimal solution, which maximizes the financial
switch-off strategy is designed for BSs of MNOs, where each profits of both the third party and the MNOs and minimizes
of multiple MNOs will quantify its tolerance about requesting the network energy consumption simultaneously, is provided
less SBS resources with different bidding values. Conflicting by solving a multi-objective optimization problem.

VOLUME 4, 2016 4965

 F. Han et al.: Survey of Strategies for Switching Off BSs

E. SUMMARY AND FUTURE WORK moreover, the first constraint represents that the load of each
The comparative summary for the literature reviewed in this BS should be no more than the maximum number of available
section can be seen in Table 2. resource, and the second constraint indicates that every user
Summary: Further, the main findings of this section are can only access one BS at one time. Solving the above opti-
described as follows: mization problem requires high computational complexity of
1) The design of specific BS switch-off strategy in HetNets o(2N +M ) for completing exhaustive searching. In [49], based
can be done based on different criteria/perspectives. However, on transforming this optimization problem to be a classical
the design which can better adapt to the dynamics of network 0/1 knapsack problem, a dynamic programming algorithm is
conditions would achieve more energy saving and better net- used to get the solution. In [50], also aiming at minimizing
work performance. the total power consumption in HetNets, the joint design
2) The design could be performed with focusing on apply- for BS switch-off strategy and user association strategy is
ing the switch-off strategy to only SBSs, or only MBSs, or formulated as a form of the minimum set cover problem. One
both types of BSs in HetNets. characteristic of the work in [50] is that the simulations are
Future Work: Although lots of works have been done for conducted based on the system-level simulation platform of
the design on specific BS switch-off strategy in HetNets, LTE (which is a commercial mobile communication system);
enhancement and challenges remain to be investigated in the this make the feasibility of the solution be verified more
future. For instance, it is meaningful to investigate the designs strictly.
done by appropriately exploiting the combination of different The joint design considered in this subsection has also
criteria/perspectives. For example, combination of random been investigated when taking the maximization on EE as the
and load criteria, combination of load and distance criteria, optimization objective (e.g., see [51]–[53]). In [51], the opti-
etc. mization problem is transformed to be a linear programming
problem, and a series of Lagrangian dual methods are used
IV. OVERVIEW OF JOINT DESIGN FOR BS SWITCH-OFF to obtain the solution when assuming a central controller is
STRATEGY AND OTHER STRATEGIES IN HETNETS available to collect the information about the whole network.
In this section, we will focus on the joint design for BS In addition, to get a more feasible solution without the need of
switch-off strategy and another strategies which are also a central controller, a user repeated bidding game is utilized
important for 5G systems, such as user association strategy, in [51] to find out the solution of the original EE maximiza-
resource allocation strategy, physical-layer (PHY-layer) inter- tion problem. In contrast, in [52], the joint design is done
ference cancellation strategy. based on providing a heuristic solution for the formulated
EE maximization problem. The key design principle is to
A. JOINT DESIGN FOR BS SWITCH-OFF AND consider the impact of switching off of a SBS on the network,
USER ASSOCIATION in terms of ensuring additional load increments on its neigh-
In the aforementioned works, the user association strategy is boring BSs (including the MBS and neighboring SBSs) can
designed based on the given design of BS switch-off strategy. be handled well with meeting the required QoS. Specifically,
Here, we consider joint design for BS switch-off strategy for any one given load state, switch off SBSs one by one and
and user association strategy. The work in [49] formulates update user association iteratively till the outage constraint
the following combinatorial optimization problem to perform cannot be satisfied. In [53], the EE maximization problem is
this joint design: also solved by using a heuristic algorithm. In particular, the
X X SBS with the lowest EE value is the most possible candidate
min Ptotal = bj ai,j pi,j for being switched off; if the system EE is improved after
0≤j≤N 1≤i≤M this SBS is deactivated and its served users are re-associated,
M
X the algorithm is repeated to successively switch off the SBS
s.t. total
ni,j ai,j ≤ NRB , ∀j with lowest EE value at each round, till there is no gain in
i=1 EE anymore.
XN Unlike other works which perform joint design in terms of
ai,j = 1, ∀i both methodology and related control parameters, the work
j=0 in [29] aims at the joint design on only control parameters,

1,
P when the methodologies for both BS switch-off strategy and
bj = Pi ai,j 6 = 0 (5) user association strategy have been given.
0, i ai,j = 0

In equation (5), ai,j = 1 denotes that the i-th user is attached B. JOINT DESIGN FOR BS SWITCH-OFF AND
to the j-th BS while ai,j = 0 denotes the opposite case, RESOURCE ALLOCATION
bj = 1 denotes the ‘‘switch-on’’ status of the j-th BS while Due to aggressive frequency reuse, serious inter-tier inter-
bj = 0 denotes the ‘‘switch-off’’ status of the j-th BS, ference observed in HetNets has raised a major chal-
ni,j denotes the minimum number of resource blocks for the lenge to the optimization on system capacity. Both of
i-th user to meet the QoS constraint of minimum data rate; wireless resource allocation and BS switch-off have been

4966 VOLUME 4, 2016

F. Han et al.: Survey of Strategies for Switching Off BSs

TABLE 3. Overview of joint design for BS switch-off strategy and other strategies in HetNets.

regarded as useful mechanisms for achieving interfer- Nets. Many relevant works have been done in the literature
ence reduction and/or EE improvement. Performing joint (e.g., see [24], [54]–[56]).
design for these two kinds of strategies will be benefi- To perform joint design for BS switch-off strategy
cial to further enhancing the performance of future Het- and resource allocation strategy, in [54], an optimization

VOLUME 4, 2016 4967

 F. Han et al.: Survey of Strategies for Switching Off BSs

problem is formulated in terms of minimizing total energy considered in terms of effective signal-to-interference-plus-
consumption of all BSs in HetNets subject to some certain noise ratio (SINR) at the receiver side. Furthermore, in [59],
user QoS constraint. This problem is then decoupled into the the amount of cross-tier interference is utilized to perform the
following two sub-problems: switch-off strategy for MBSs joint deign considered in this subsection. Specifically, the
in the considered HetNet, user association and resource allo- design of pre-coding used in beamforming technology will
cation; then, the algorithm for solving each sub-problem is determine the amount of cross-tier interference, while switch-
developed. ing on or off a SBS is decided by the amount of the interfer-
A threshold-based optimization problem is formulated ence generated by or to the MBS.
in [55] to jointly optimize switch-off strategy for SBSs and
spectrum resource allocation. It is found in [55] that the D. SUMMARY AND FUTURE WORK
lightly-loaded SBSs should be switched off first when their The comparative summary for the literature reviewed in this
sleeping probabilities are low, while the most heavily-loaded section can be seen in Table 3.
ones should go into sleep when their sleeping probabilities Summary: The main finding is that the joint design for
are high. The work in [56] focuses on jointly designing partial BS switch-off strategy and another strategies in HetNets will
spectrum reuse (PSR) and switch-off strategy for BSs (includ- simultaneously improve the energy saving and enhance the
ing MBSs and SBSs). With assuming the sleeping/active overall network performance, to get greener and also more
probability of each MBS is the same and the sleeping/active robust 5G systems.
probability of each SBS is the same, in [56], the first decom- Future Work: In the future, the effort could be put into the
posed sub-problem of maximizing the coverage probability joint design for BS switch-off strategy and another advanced
of all BSs is solved by deriving the relationship between the HetNet-related strategies, such as inter-site carrier aggre-
optimal PSR factor and the ratio of MBSs’ active probability gation (CA) between MBS and SBS [60], dual connec-
to SBSs’ active probability, and the solving for the second tivity (DC) between MBS and SBS [61], downlink-uplink
sub-problem of minimizing the total network energy cost is decoupled biased user association [62], [63], etc.
then followed.
Furthermore, if transmit power is also viewed as one kind V. OVERVIEW OF BS SWITCH-OFF STRATEGIES IN
of wireless resource, the joint design for BS switch-off strat- CLOUD RADIO ACCESS NETWORKS
egy and transmit power allocation strategy is investigated When CRAN is considered, the amount of research works
in [24] for self-powered SBSs. for BS switch-off strategies has not been large and most of
them focus on CRAN without HetNet deployment. Thus, in
C. JOINT DESIGN FOR BS SWITCH-OFF AND PHY-LAYER this section, we will firstly review BS switch-off strategies
INTERFERENCE CANCELLATION in CRAN without HetNet deployment; then, the emerging
Due to their potential for achieving higher link reliability research direction of designing BS switch-off strategies for
and data rates, multiple-input multiple-output (MIMO) tech- H-CRAN will be discussed.
niques utilizing multiple antennas at the transmitter and/or
receiver have emerged as a milestone of modern wireless A. BS SWITCH-OFF STRATEGIES IN CLOUD RAN
communications. As one of MIMO techniques, beamform- WITHOUT HETNET DEPLOYMENT
ing is the most popular PHY-layer interference cancellation In CRAN, baseband signal processing for multiple distributed
technique used in commercial mobile communication sys- RRHs is performed in the BBU pool (or called cloud), where
tems, for mitigating intra-cell multi-user and/or multi-stream the computing and storage resources are configured as several
interference and inter-cell multi-BS interference. When virtual BBUs. Depending on the computing and storage capa-
beamforming is exploited to combat inter-cell multi-BS inter- bility given to each virtual BBU and the maximum amount
ference, the resulted transmission scheme is also called coor- of baseband processing tasks needed for each RRH, either
dinated multi-point (CoMP) transmission. Some works have multiple RRHs could be supported by one virtual BBU or one
been done to investigate joint design for BS switch-off strat- RRH at most needs to be served by multiple virtual BBUs.
egy and beamforming (e.g., see [57]–[59]). In the following, for the sake of simplicity, we just use the
The joint deign considered in this subsection is done in [57] term ‘‘BBU’’ to represent ‘‘virtual BBU’’.
by formulating a signal-to-noise ratio (SNR) maximization
problem, which tries to find the optimal beamforming pattern 1) STRATEGY VIA SWITCHING OFF BBUs
of SBSs with setting the maximal number of SBSs which can In the BBU pool, the number of used BBUs can be reduced
be active as one optimization constraint. In [58], an optimized by matching a right amount of BBU resources with the cor-
solution about choosing which BSs (including MBSs and responding RRHs according to the traffic load. Some works
SBSs) to be turned on is obtained by either minimizing the have been done to investigate how to appropriately perform
transmission power consumption resulted by beamforming the ‘‘BBU-to-RRH’’ mapping and switch off underutilized
transmissions or maximizing the sum rate achieved by beam- BBUs (e.g., see [64]–[67]).
forming transmissions. When utilizing transmission power With the objective of minimizing network power con-
minimization to do the joint design, a QoS constraint is sumption, in [64], the ‘‘BBU-to-RRH’’ mapping problem is

4968 VOLUME 4, 2016

F. Han et al.: Survey of Strategies for Switching Off BSs

modelled as the typical bin packing problem when assuming

the traffic load has been given. In particular, each BBU is
modeled as a bin and the baseband signal processing tasks
which come from the corresponding RRHs are modeled as
items to be packed. After properly packing these items into
some BBUs, the number of used BBUs needs to be mini-
mized. Since the modeled optimization problem is NP-hard,
three heuristic algorithms are proposed in [64] to generate
good but not necessarily optimal ‘‘BBU-to-RRH’’ mapping.
In [65], based on more practically modeling energy con-
sumption for housing facilities (e.g., cooling facility), internal
switch in central office at which the BBU pool is located, cen-
tralized BBUs, optical line terminal in central office, RRHs,
and optical network units at cell sites, a network energy con-
sumption minimization problem is formulated to minimize
the number of active BBUs. Then, the problem is solved for
static traffic load and dynamic traffic load, respectively. For FIGURE 4. Flexible constructions for the network architecture of H-CRANs.
the case of static (i.e., given) traffic load, the problem is more
difficult to solve when compared with the problem modeled
in [64], because each BBU is modeled as two sub-bins to re-optimizing the coordinated beamforming for the remain-
pack two kinds of items including user processing and cell ing active RRHs. Further, to reduce the complexity, an equiva-
processing. lent sparsity-based representation of the focused optimization
Focusing on the case that multiple RRHs could be sup- problem is proposed, where the group sparsity pattern of the
ported by one BBU in the BBU pool, a traffic load balancing aggregated beamforming vector indicates those RRHs which
based strategy is designed in [66], to assign a minimum can be switched off. The work in [69] is an extension of [68],
number of active BBUs to the RRHs so that power saving where the considered set of UEs is divided into multiple non-
is maximized. In this strategy, if the resource usage of one overlapping multicast subsets so that the formulated sparsity-
BBU reaches the upper limit, partial traffic of this BBU will based optimization problem is more difficult to solve. With
be offloaded to another light-loaded or sleeping BBU; in QoS constraints on potentially a large number of served UEs,
contrast, an underutilized BBU will be switched off after its the focused power consumption minimization problem may
traffic is completely offloaded to another suitable BBU. Sim- be unable to find feasible solutions. While this practical
ilar as [66], through proper scheduling and timely offload- concern is ignored in [68], the work in [69] proposes to
ing/consolidation for traffic load required to be processed by reformulate the joint design problem to be maximizing the
active BBUs, one heuristic strategy is proposed in [67] to number of supported users through user admission control.
dynamically minimize the number of active BBUs. In addi- The work in [70] considers the same joint design as that
tion, in [67], a testbed is constructed for concept verification focused in [68] and [69] and also formulates an optimization
and practical performance evaluation. problem in terms of minimizing the total power consumption
of RRHs and the corresponding fronthaul links. However,
2) STRATEGY VIA SWITCHING OFF RRHs AND unlike [68] and [69] where SINRs of all the served UEs
FRONTHAUL LINKS are used as QoS constraints, traffic delay is used in [70] for
In CRAN, with the centralized processing at the BBU setting optimization constraint on QoS. More importantly, in
pool, the transport links between RRHs and the BBU pool contrast to [68] and [69], in [70], the power consumption min-
(i.e., fronthaul links) need to provide high-capacity connec- imization is formulated as a stochastic optimization problem,
tions, so that the power consumption over fronthaul links based on considering random traffic arrivals and time-varying
becomes enormous and cannot be ignored. Thus, to reduce channel conditions.
power consumption of the entire network for CRAN, it
is also crucial to switch off some RRHs and the corre- B. BS SWITCH-OFF STRATEGIES IN
sponding fronthaul links based on the data traffic require- HETEROGENEOUS CLOUD RAN
ments. There have been some relevant works in the literature When transforming a traditional HetNet into a H-CRAN,
(e.g., see [68]–[70]). in theory, an extreme situation is to make all of traditional
In [68], the joint design on selection of active RRHs and SBSs and MBSs in the considered HetNet be transformed
coordinated beamforming among active RRHs is done, with into one or more pools of virtual BBUs plus the correspond-
the objective of minimizing the total power consumption of ing RRHs. Of course, this kind of cloudification could be
RRHs and the corresponding fronthaul links. This optimiza- flexibly done for only part of traditional SBSs and part of
tion problem is firstly solved by one greedy algorithm, which traditional MBSs. One situation, which is preferred by most
iteratively chooses one RRH to switch off at each step while of the exiting literature, is to do the cloudification for all of

VOLUME 4, 2016 4969

 F. Han et al.: Survey of Strategies for Switching Off BSs

TABLE 4. Overview of BS switch-off strategies in cloud radio access networks.

traditional SBSs but for none of traditional MBSs. To the to illustrate flexible constructions for the network architecture
best of the authors’ knowledge, none of the existing literature of H-CRANs.
has clearly represented the network architecture of H-CRANs It can be seen from Fig. 4 that, if every large-scale
like this. Further, the authors of this survey paper use Fig. 4 coverage area in a H-CRAN is constructed like Coverage

4970 VOLUME 4, 2016

F. Han et al.: Survey of Strategies for Switching Off BSs

Area Type I, the extreme situation with full cloudification strategies which are important for 5G systems, and research
occurs. Of course, the mixed deployment of Coverage Area results about this topic in emerging CRANs/H-CRANs were
Types I, II, and III is also possible for a H-CRAN. Further, introduced. For readers to easily find interesting references
one currently preferred situation is to make each large-scale which have been presented in this paper, Tables 1, 2, 3, 4 give
coverage area be constructed like Coverage Area Type III. the comparative summary for each of the above four aspects,
In this case, while being used to guarantee the backward respectively.
compatibility with the existing cellular systems, traditional Existing research works show that well-designed BS
MBSs could also be used to deliver all the control signaling switch-off strategies (either with joint design considering
in a H-CRAN so that the capacity and delay constraints on also another strategies or not) could significantly reduce
fronthaul links are alleviated. the energy consumption of the entire network and improve
For H-CRANs, there has been little literature to dis- overall network performance. Nevertheless, current research
cuss the switch-off strategy for ‘‘virtual BBUs in the BBU results are still quite preliminary; enhancement and chal-
pool’’ and/or ‘‘RRHs plus the corresponding fronthaul links’’. lenges remain to be investigated in the future.
Further, in quite a few related works which we can find As for the specific research directions in the future, besides
(e.g., see [17], [18]), the switch-off strategy is just simply what have been presented in Section II-D, Section III-E,
mentioned at a concept level. In [17], with focusing on a Section IV-D, and Section V-C, the design under more
H-CRAN constructed by deploying all large-scale coverage advanced HetNet architectures in 5G systems need to
areas with only the way of Coverage Area Type III shown be considered too. Specifically, with the introduction of
in Fig. 4, it is briefly mentioned that the number of activated Millimeter wave (mmWave) frequencies in 5G systems, the
RRHs should be adaptive to traffic volume so that the EE desired deployment is a mixed sub-6GHz and mmWave
performance of the focused H-CRAN could be improved. HetNet deployment where mmWave BSs act as small cells
In [18], for a H-CRAN which is constructed by deploying and sub-6GHz BSs act as MBSs [71], [72]; a further inclusion
all large-scale coverage areas with only the way of Coverage of sub-6GHz SBSs also needs to be considered. Clearly, new
Area Type II shown in Fig. 4, with using the concept-level design issues would exist when applying the BS switch-off
statement, it is pointed out that the dynamic energy saving mechanism to this kind of novel 5G HetNets.
could be achieved via minimizing the number of active RRHs
and traditional MBSs/SBSs while ensuring the desired QoS. REFERENCES
[1] The 1000x Data challenge, Qualcomm, San Diego, CA, USA, 2013.
C. SUMMARY AND FUTURE WORK [2] Cisco Visual Networking Index: Global Mobile Data Traffic Forecast
The comparative summary for the literature reviewed in this Update, 2013–2018, Cisco, San Jose, CA, USA, 2014.
[3] Cisco Visual Networking Index: Global Mobile Data Traffic Forecast
section can be seen in Table 4. Update, 2014–2019, Cisco, San Jose, CA, USA,, 2015.
Summary: The main findings of this section can be found [4] ‘‘5G wireless technology architecture,’’ IMT-2020(5G) Promotion Group,
as below: White Paper, 2015.
[5] A. Cimmino et al., ‘‘The role of small cell technology in future smart
1) In CRANs or H-CRANs, for the cloudified BSs, the city applications,’’ Trans. Emerg. Telecommun. Technol., vol. 25, no. 1,
design of switch-off strategy would be done through making pp. 11–20, Jan. 2014.
either ‘‘virtual BBUs in the BBU pool’’ or ‘‘RRHs plus the [6] J. G. Andrews et al., ‘‘What will 5G be?’’ IEEE J. Sel. Areas Commun.,
corresponding fronthaul links’’ enter into the sleeping mode. vol. 32, no. 6, pp. 1065–1082, Jun. 2014.
[7] N. Bhushan et al., ‘‘Network densification: The dominant theme for wire-
2) With the help of centralized baseband signal processing less evolution into 5G,’’ IEEE Commun. Mag., vol. 52, no. 2, pp. 82–89,
in the BBU pool, CoMP transmission among RRHs would Feb. 2014.
be implemented more easily; therefore, it is meaningful to [8] J. B. Rao and A. O. Fapojuwo, ‘‘A survey of energy efficient resource
management techniques for multicell cellular networks,’’ IEEE Commun.
perform joint design for RRH switch-off strategy and coordi- Surveys Tuts., vol. 16, no. 1, pp. 154–180, 1st Quart., 2014.
nated beamforming among active RRHs. [9] I. Chih-Lin, C. Rowell, S. Han, Z. Xu, G. Li, and Z. Pan, ‘‘Toward
Future Work: Much more research efforts would be needed green and soft: A 5G perspective,’’ IEEE Commun. Mag., vol. 52, no. 2,
pp. 66–73, Feb. 2014.
to investigate the switch-off strategy for cloudified BSs
[10] C.-X. Wang et al., ‘‘Cellular architecture and key technologies for 5G
in H-CRANs, although some switch-off strategies already wireless communication networks,’’ IEEE Commun. Mag., vol. 52, no. 2,
designed in CRANs could be directly applied to H-CRANs. pp. 122–130, Feb. 2014.
Further, for the cloudified BSs, it will be quite attractive to [11] A. De Domenico, E. C. Strinati, and A. Capone, ‘‘Enabling green cellular
networks: A survey and outlook,’’ Comput. Commun., vol. 37, pp. 5–24,
investigate how to simultaneously minimize the number of Jan. 2014.
active virtual BBUs and the number of active RRHs. [12] Y. Li, H. Celebi, M. Daneshmand, C. Wang, and W. Zhao, ‘‘Energy-
efficient femtocell networks: Challenges and opportunities,’’ IEEE Wire-
less Commun., vol. 20, no. 6, pp. 99–105, Dec. 2013.
VI. CONCLUSIONS AND FUTURE WORK
[13] D. Feng, C. Jiang, G. Lim, L. J. Cimini, Jr., G. Feng, and G. Y. Li,
In this paper, we discussed the state-of-the-art research on BS ‘‘A survey of energy-efficient wireless communications,’’ IEEE Commun.
switch-off design for greener 5G systems with the deploy- Surveys Tuts., vol. 15, no. 1, pp. 167–178, 1st Quart., 2013.
ment of HetNets. Optimization objectives and constraints [14] X. Chen, J. Wu, Y. Cai, H. Zhang, and T. Chen, ‘‘Energy-efficiency
oriented traffic offloading in wireless networks: A brief survey and a
for designing BS switch-off strategy, specific BS switch-off learning approach for heterogeneous cellular networks,’’ IEEE J. Sel. Areas
strategies, joint design for BS switch-off strategy and another Commun., vol. 33, no. 4, pp. 627–640, Apr. 2015.

VOLUME 4, 2016 4971

 F. Han et al.: Survey of Strategies for Switching Off BSs

[15] R. L. Cavalcante, S. Stanczak, M. Schubert, A. Eisenblaetter, and U. Türke, [40] A. A. Abdulkafi, T. S. Kiong, D. Chieng, A. Ting, and J. Koh, ‘‘Energy
‘‘Toward energy-efficient 5G wireless communications technologies: efficiency improvements in heterogeneous network through traffic load
Tools for decoupling the scaling of networks from the growth of operating balancing and sleep mode mechanisms,’’ Wireless Pers. Commun., vol. 75,
power,’’ IEEE Signal Process. Mag., vol. 31, no. 6, pp. 24–34, Nov. 2014. no. 4, pp. 2151–2164, Apr. 2014.
[16] G. Wu, C. Yang, S. Li, and G. Y. Li, ‘‘Recent advances in energy-efficient [41] W. Vereecken et al., ‘‘Evaluation of the potential for energy saving in
networks and their application in 5G systems,’’ IEEE Wireless Commun., macrocell and femtocell networks using a heuristic introducing sleep
vol. 22, no. 2, pp. 145–151, Apr. 2015. modes in base stations,’’ EURASIP J. Wireless Commun. Netw., vol. 2012,
[17] M. Peng, Y. Li, J. Jiang, J. Li, and C. Wang, ‘‘Heterogeneous cloud radio no. 1, pp. 1–14, Dec. 2012.
access networks: A new perspective for enhancing spectral and energy effi- [42] L. Saker, S.-E. Elayoubi, R. Combes, and T. Chahed, ‘‘Optimal control of
ciencies,’’ IEEE Wireless Commun., vol. 21, no. 6, pp. 126–135, Dec. 2014. wake up mechanisms of femtocells in heterogeneous networks,’’ IEEE J.
[18] M. Peng, Y. Li, Z. Zhao, and C. Wang, ‘‘System architecture and key Sel. Areas Commun., vol. 30, no. 3, pp. 664–672, Apr. 2012.
technologies for 5G heterogeneous cloud radio access networks,’’ IEEE [43] S. Navaratnarajah, A. Saeed, M. Dianati, and M. A. Imran, ‘‘Energy
Netw., vol. 29, no. 2, pp. 6–14, Mar./Apr. 2015. efficiency in heterogeneous wireless access networks,’’ IEEE Wireless
[19] M. A. Marsan, L. Chiaraviglio, D. Ciullo, and M. Meo, ‘‘Multiple daily Commun., vol. 20, no. 5, pp. 37–43, Oct. 2013.
base station switch-offs in cellular networks,’’ in Proc. IEEE Int. Conf. [44] X. Gan et al., ‘‘Energy efficient switch policy for small cells,’’ China
Commun. Electron. (ICCE), Aug. 2012, pp. 245–250. Commun., vol. 12, no. 1, pp. 78–88, Jan. 2015.
[20] P. Dini, M. Miozzo, N. Bui, and N. Baldo, ‘‘A model to analyze the energy
[45] J. Wu, S. Jin, L. Jiang, and G. Wang, ‘‘Dynamic switching off algorithms
savings of base station sleep mode in LTE HetNets,’’ in Proc. IEEE Int.
for pico base stations in heterogeneous cellular networks,’’ EURASIP J.
Conf. Green Comput. Commun. (GreenCom), Aug. 2013, pp. 1375–1380.
Wireless Commun. Netw., vol. 2015, no. 1, pp. 1–18, Dec. 2015.
[21] G. Auer et al., ‘‘How much energy is needed to run a wireless network?’’
[46] M. Oikonomakou, A. Antonopoulos, L. Alonso, and C. Verikoukis,
IEEE Trans. Wireless Commun., vol. 18, no. 5, pp. 40–49, Oct. 2011.
‘‘Cooperative base station switching off in multi-operator shared hetero-
[22] C. Liu, B. Natarajan, and H. Xia, ‘‘Small cell base station sleep strategies
geneous network,’’ in Proc. IEEE Global Commun. Conf. (GLOBECOM),
for energy efficiency,’’ IEEE Commun. Veh. Technol., vol. 65, no. 3,
Dec. 2015, pp. 1–6.
pp. 1652–1661, Mar. 2016.
[23] Y. S. Soh, T. Q. S. Quek, M. Kountouris, and H. Shin, ‘‘Energy efficient [47] E. Yaacoub, ‘‘A practical approach for base station on/off switching in
heterogeneous cellular networks,’’ IEEE J. Sel. Areas Commun., vol. 31, green LTE-A HetNets,’’ in Proc. IEEE Int. Conf. Wireless Mobile Comput.,
no. 5, pp. 840–850, May 2013. Netw. Commun. (WiMob), Oct. 2014, pp. 159–164.
[24] C.-Y. Chang, K.-L. Ho, W. Liao, and D.-S. Shiu, ‘‘Capacity maximiza- [48] M. A. Marsan and M. Meo, ‘‘Network sharing and its energy benefits:
tion of energy-harvesting small cells with dynamic sleep mode operation A study of European mobile network operators,’’ in Proc. IEEE Global
in heterogeneous networks,’’ in Proc. IEEE Int. Conf. Commun. (ICC), Commun. Conf. (GLOBECOM), Dec. 2013, pp. 2561–2567.
Jun. 2014, pp. 2690–2694. [49] G. Wu, L. Dong, Z. Qin, and Z. Xu, ‘‘Dynamic programming-based
[25] X. Guo, Z. Niu, S. Zhou, and P. R. Kumar, ‘‘Delay-constrained energy- pico base station sleep mode control in heterogeneous networks,’’ Int. J.
optimal base station sleeping control,’’ IEEE J. Sel. Areas Commun., Commun. Syst., to be published, doi: 10.1002/dac.2967.
vol. 34, no. 5, pp. 1073–1085, May 2016. [50] J. He et al., ‘‘Energy efficient BSs switching in heterogeneous net-
[26] Z. Niu, X. Guo, S. Zhou, and P. R. Kumar, ‘‘Characterizing energy-delay works: An operators perspective,’’ in Proc. IEEE Wireless Commun. Netw.
tradeoff in hyper-cellular networks with base station sleeping control,’’ Conf. (WCNC), 2016.
IEEE J. Sel. Areas Commun., vol. 33, no. 4, pp. 641–650, Apr. 2015. [51] M. Feng, S. Mao, and T. Jiang, ‘‘BOOST: Base station on-off switching
[27] B. Liu, M. Zhao, W. Zhou, J. Zhu, and P. Dong, ‘‘Flow-level-delay con- strategy for energy efficient massive MIMO HetNets,’’ in Proc. IEEE Int.
straint small cell sleeping with macro base station cooperation for energy Conf. Comput. Commun. (INFOCOM), Apr. 2016, pp. 1–9.
saving in HetNet,’’ in Proc. IEEE Veh. Technol. Conf. (VTC), Sep. 2015, [52] L. Zhou et al., ‘‘Green small cell planning in smart cities under
pp. 1–5. dynamic traffic demand,’’ in Proc. IEEE Int. Conf. Comput. Commun.
[28] I. Aydin, H. Yanikomeroglu, and U. Aygolu, ‘‘User-aware cell switch- Workshops (INFOCOM), Apr./May 2015, pp. 618–623.
off algorithms,’’ in Proc. Int. Wireless Commun. Mobile Comput. [53] G. K. Tran, H. Shimodaira, R. E. Rezagah, K. Sakaguchi, and K. Araki,
Conf. (IWCMC), Aug. 2015, pp. 1236–1241. ‘‘Dynamic cell activation and user association for green 5G heterogeneous
[29] S. Song, Y. Chang, X. Wang, and D. Yang, ‘‘Coverage and energy modeling cellular networks,’’ in Proc. IEEE Annu. Int. Symp. Pers., Indoor, Mobile
of HetNet under base station on-off model,’’ ETRI J., vol. 37, no. 3, Radio Commun. (PIMRC), Aug./Sep. 2015, pp. 2364–2368.
pp. 450–459, Jun. 2015. [54] P. Ren and M. Tao, ‘‘A decentralized sleep mechanism in heterogeneous
[30] G. Yu, Q. Chen, and R. Yin, ‘‘Dual-threshold sleep mode control scheme cellular networks with QoS constraints,’’ IEEE Wireless Commun. Lett.,
for small cells,’’ IET Commun., vol. 8, no. 11, pp. 2008–2016, Jul. 2014. vol. 3, no. 5, pp. 509–512, Oct. 2014.
[31] S. Paris, F. Martignon, I. Filippini, and L. Chen, ‘‘A bandwidth trading mar- [55] S. Zhang, J. Wu, J. Gong, S. Zhou, and Z. Niu, ‘‘Energy-optimal probabilis-
ketplace for mobile data offloading,’’ in Proc. IEEE INFOCOM, Apr. 2013, tic base station sleeping under a separation network architecture,’’ in Proc.
pp. 430–434. IEEE Global Commun. Conf. (GLOBECOM), Dec. 2014, pp. 4239–4244.
[32] W. Dong et al., ‘‘iDEAL: Incentivized dynamic cellular offloading via auc-
[56] X. Chai, Y. Li, Y. Lv, and Z. Zhang, ‘‘Joint spectrum-sharing and base-
tions,’’ IEEE/ACM Trans. Netw., vol. 22, no. 4, pp. 1271–1284, Aug. 2014.
station-sleep model for improving energy efficiency of HetNet,’’ in Proc.
[33] A. Bousia, E. Kartsakli, A. Antonopoulos, L. Alonso, and C. Verikoukis,
IEEE Int. Conf. Commun. (ICC), Jun. 2015, pp. 1851–1856.
‘‘Multiobjective auction-based switching off scheme in heterogeneous
[57] G. Bartoli, L. Carla, R. Fantacci, and D. Marabissi, ‘‘Efficient multicast
networks to bid or not to bid?’’ IEEE Trans. Veh. Technol., to be published.
beamforming in HetNet with an opportunistic cells activation,’’ in Proc.
[34] E. Altman et al., ‘‘Stochastic geometric models for green networking,’’
IEEE Globecom Workshops (GC Wkshps), Dec. 2015, pp. 1–6.
IEEE Access, vol. 3, pp. 2465–2474, Nov. 2015.
[35] D. Tsilimantos, J.-M. Gorce, and E. Altman, ‘‘Stochastic analysis of energy [58] W.-C. Liao, M. Hong, Y.-F. Liu, and Z.-Q. Luo, ‘‘Base station activa-
savings with sleep mode in OFDMA wireless networks,’’ in Proc. IEEE Int. tion and linear transceiver design for optimal resource management in
Conf. Comput. Commun. (INFOCOM), Apr. 2013, pp. 1097–1105. heterogeneous networks,’’ IEEE Trans. Signal Process., vol. 62, no. 15,
[36] Y. S. Soh, T. Q. Quek, and M. Kountouris, ‘‘Dynamic sleep mode strate- pp. 3939–3952, Aug. 2014.
gies in energy efficient cellular networks,’’ in Proc. IEEE Int. Conf. [59] A. Adhikary, H. S. Dhillon, and G. Caire, ‘‘Massive-MIMO meets HetNet:
Commun. (ICC), Jun. 2013, pp. 3131–3136. Interference coordination through spatial blanking,’’ IEEE J. Sel. Areas
[37] J. Peng, P. Hong, and K. Xue, ‘‘Stochastic analysis of optimal base station Commun., vol. 33, no. 6, pp. 1171–1186, Jun. 2015.
energy saving in cellular networks with sleep mode,’’ IEEE Commun. Lett., [60] ‘‘LTE-advanced carrier aggregation optimization,’’ Nokia, Espoo, Finland,
vol. 18, no. 4, pp. 612–615, Apr. 2014. White Paper, 2015.
[38] A. Bousia, A. Antonopoulos, L. Alonso, and C. Verikoukis, ‘‘‘Green’ [61] H. Wang, C. Rosa, and K. I. Pedersen, ‘‘Dual connectivity for LTE-
distance-aware base station sleeping algorithm in LTE-advanced,’’ in Proc. Advanced heterogeneous networks,’’ Wireless Netw., vol. 22, no. 4,
IEEE Int. Conf. Commun. (ICC), Jun. 2012, pp. 1347–1351. pp. 1315–1328, May 2016.
[39] S. Cai, Y. Che, L. Duan, J. Wang, S. Zhou, and R. Zhang, ‘‘Green 5G [62] H. Elshaer, F. Boccardi, M. Dohler, and R. Irmer, ‘‘Downlink and uplink
heterogeneous networks through dynamic small-cell operation,’’ IEEE J. decoupling: A disruptive architectural design for 5G networks,’’ in Proc.
Sel. Areas Commun., vol. 34, no. 5, pp. 1103–1115, May 2016. IEEE Global Commun. Conf. (GLOBECOM), Dec. 2014, pp. 1798–1803.

4972 VOLUME 4, 2016

F. Han et al.: Survey of Strategies for Switching Off BSs

[63] F. Boccardi et al., ‘‘Why to decouple the uplink and downlink in cel- LU ZHANG received the Ph.D degree in 2009
lular networks and how to do it,’’ IEEE Commun. Mag., vol. 54, no. 3, from the Department of Electrical and Computer
pp. 110–117, Mar. 2016. Engineering, University of Delaware, USA. And,
[64] T. Sigwele, A. S. Alam, P. Pillai, and Y. F. Hu, ‘‘Evaluating energy-efficient he received the M.E. and B.E. degrees in 2002
cloud radio access networks for 5G,’’ in Proc. IEEE Int. Conf. Data Sci. and 1999, respectively, from the Department of
Data Intensive Syst., Dec. 2015, pp. 362–367. Electronic Engineering, Beijing Institute of Tech-
[65] X. Wang et al., ‘‘Energy-efficient virtual base station formation in optical- nology, China. He is currently a Product Line
access-enabled cloud-RAN,’’ IEEE J. Sel. Areas Commun., vol. 34, no. 5,
Manager for 5G products at the Nokia Shanghai
pp. 1130–1139, May 2016.
Bell Co., Ltd. (which is the former Alcatel-Lucent
[66] M. Khan, R. Alhumaima, and H. Al-Raweshidy, ‘‘Reducing energy con-
sumption by dynamic resource allocation in C-RAN,’’ in Proc. Eur. Conf. Shanghai Bell Co., Ltd.). Since 2009, he joined
Netw. Commun. (EuCNC), Jun./Jul. 2015, pp. 169–174. Alcatel-Lucent Shanghai Bell Co., Ltd., where he was a Research Scientist
[67] Z. Kong, J. Gong, C.-Z. Xu, K. Wang, and J. Rao, ‘‘EBase: A base- at the Bell Labs China from 2009 to 2012 and began to act as a Product Line
band unit cluster testbed to improve energy-efficiency for cloud radio Manager since 2012. He was awarded as Distinguished Member of Technical
access network,’’ in Proc. IEEE Int. Conf. Commun. (ICC), Jun. 2013, Staff (title for life) in 2015, and awarded as Member of the Alcatel-Lucent
pp. 4222–4227. Technical Academy in 2011. From November 2008 to March 2009, he was
[68] Y. Shi, J. Zhang, and K. B. Letaief, ‘‘Group sparse beamforming for a full-time Research Intern at the Mitsubishi Electric Research Laboratories,
green cloud-RAN,’’ IEEE Trans. Wireless Commun., vol. 13, no. 5, Cambridge, USA. From January 2004 to August 2004, he was a System
pp. 2809–2823, May 2014. Engineer at the Siemens Ltd., China. From April 2002 to January 2004, he
[69] Y. Shi, J. Cheng, J. Zhang, B. Bai, W. Chen, and K. B. Letaief, ‘‘Smoothed was a R&D Engineer at the Beijing Research Institute, Huawei Technologies
lp-minimization for green cloud-RAN with user admission control,’’ IEEE Co., Ltd. His research interests include emerging techniques in 5G systems,
J. Sel. Areas Commun., vol. 34, no. 4, pp. 1022–1036, Apr. 2016. interference management for small cells and/or heterogeneous networks
[70] J. Li, J. Wu, M. Peng, and P. Zhang, ‘‘Queue-aware energy-efficient (HetNet), coordinate multi-point (CoMP) transmission technologies in LTE-
joint remote radio head activation and beamforming in cloud radio access
Advanced systems, carrier aggregation technologies in LTE-Advanced sys-
networks,’’ IEEE Trans. Wireless Commun., vol. 15, no. 6, pp. 3880–3894,
tems, theoretical analyses for multi-cell co-channel interference in MIMO-
Jun. 2016.
[71] Z. Pi and F. Khan, ‘‘An introduction to millimeter-wave mobile broadband based cellular networks, decentralized cooperative relaying techniques in
systems,’’ IEEE Commun. Mag., vol. 49, no. 6, pp. 101–107, Jun. 2011. adhoc networks, space-time coding in MIMO systems, etc.
[72] J. Park, S.-L. Kim, and J. Zander, ‘‘Tractable resource management with
uplink decoupled millimeter-wave overlay in ultra-dense cellular net-
works,’’ IEEE Trans. Wireless Commun., to be published.

FENGXIA HAN is a Ph. D. student with Tongji

University since 2015. She holds a bachelors
degree from University of Electronics Science and
Technology of China in the area of Wireless Com-
munications. Her research interests include wire-
less big data, massive MIMO and heterogeneous
networks. JINSONG WU (SM’11) is the Founder and
Founding Chair of IEEE Technical Commit-
tee on Green Communications and Computing
(TCGCC), which was established in 2011 as an
SHENGJIE ZHAO (SM’09) received the official Technical Subcommittee (TSCGCC) and
B.S.degree in electrical engineering from the elevated as TCGCC in 2013. He is also the
University of Science and Technology of China Co-Founder and Founding Vice-Chair of IEEE
(USTC), Hefei, China, in 1988; the M.S. degree in Technical SubCommittee on Big Data (TSCBD),
electrical and computer engineering from China established in January 2015, He is Series Editor
Aerospace Institute, Beijing, China, in 1991; and of IEEE Journal on Selected Areas in Communi-
the Ph.D. degree in electrical and computer engi- cations Series on Green Communications and Networking, Editor of IEEE
neering from Texas A & M University, College Communications Surveys & Tutorials, Associate Editor of IEEE Systems
Station, TX, USA, in 2004. He is a Professor Journal, Associate Editor of IEEE Access. He is the founder and Series
with the College of Electronics and Information Editor of IEEE Series on Green Communication and Computing Networks
Engineering, Tongji University, Shanghai, China. In previous postings, he in IEEE Communications Magazine. He has served as a co-leading Guest
conducted research at Lucent Technologies, Whippany, NJ, USA, and China Editor of Special Issue on Green Communications, Computing, and Systems
Aerospace Science and Industry Corporation, Beijing, China. His research in IEEE Systems Journal, and Associate Editor of Special Section on Big
interests include big data, wireless communications, image processing, and Data for Green Communications and Computing in IEEE Access. He was
signal processing. Dr. Zhao is a PI of 4th Topic of Project of Big Data and the leading Editor and a co-author of the comprehensive book, entitled Green
the National Basic Research Program of China (973Program), and was a PI Communications: Theoretical Fundamentals, Algorithms, and Applications,
of TD-LTE-Advanced System Research and Development, National Science published by CRC Press in September 2012.
and Technology Specific Projects of China. He is a Fellow of the Thousand
Talents Program of China and is a Member of Sigma Xi..

VOLUME 4, 2016 4973