Received April 16, 2018, accepted May 16, 2018, date of publication May 28, 2018, date of current

version June 20, 2018.

Digital Object Identifier 10.1109/ACCESS.2018.2841044

An Interference Contribution Rate Based Small

Cells On/Off Switching Algorithm for 5G Dense
Heterogeneous Networks
BIN SHEN , ZHENZHU LEI , XIAOGE HUANG , AND QIANBIN CHEN
School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400-065, China
Corresponding author: Bin Shen (shenbin@cqupt.edu.cn)
This work was supported by the National Natural Science Foundation of China under Grant 61571073.

ABSTRACT Dense heterogeneous network serves as one of the most promising technologies for the
upcoming 5G network. However, with a larger number of densely deployed small cells, the problem of
co-channel interference and power consumption of 5G small cell base stations (also known as S-gNBs)
becomes more severe. In order to improve the power efficiency and alleviate the co-channel interference,
S-gNBs in the network need to be dynamically switched on or off via appropriate criteria or strategies. This
paper proposes an interference contribution rate (ICR) based small cell on/off switching algorithm. With
the incorporation of network adjacency matrix (NAM), the proposed algorithm circumvents the complicated
computation of the ICRs of the S-gNBs in the network and hence yields less computational complexity.
In addition, the proposed algorithm identifies the S-gNBs that need to be switched on or off based on the
NAM and the serving signal strength measurements of user equipment’s and therefore involves less signaling
information in the procedure of switching on/off decision. Simulation results verify that the proposed
algorithm effectively diminishes the co-channel interference among the small cells, ameliorates the network
power efficiency and total data rate, and meanwhile maintains a lower traffic loss in the network.

INDEX TERMS HetNets, small cell on/off switching, power efficiency, interference contribution rate.

I. INTRODUCTION spots area, in order to effectuate seamless coverage and suffi-

With the explosive growth of mobile data applications and cient network capacity, intensively deploying a large number
services in recent years, mobile communications are facing of S-gNBs straightforwardly becomes an efficacious method,
greater challenges than ever before in increasing the net- i.e. deploying a large number of low power small cell
work capacity, boosting the data rate, and enhancing the facilities, such as micro base stations, Femto cells, Pico
energy efficiency, and hence imposing tremendous pressure cells or relay nodes, achieves load diversion and improves the
on the existing mobile communication networks. To meet network capacity [4], [5].
the fast-growing demand of mobile services, the fifth gen- Although the HetNet with dense small cell deployment
eration (5G) of mobile communication has become a hot exhibits the above advantages, it introduces a number of prob-
topic in academia and industry. Many new solutions and tech- lems and technical challenges, such as serious co-frequency
nologies have emerged, such as massive MIMO, millimeter interference among small cells [6], S-gNBs’ unnecessarily
wave (mmWave), and dense cellular HetNet [1]–[3]. Massive huge power consumption [7], tedious collaborative oper-
MIMO technology effectively raises the transmission rate ations of small cells [8] and etc. In [9], the transitional
and spectrum efficiency through space multiplexing. A large behaviors from noise-limited regime to dense interference-
number of under-utilized mmWave frequency bands provide limited regime in dense small cell networks are revealed and
potential gigahertz transmission bandwidth. Different from the factors that affect the performance trend are analyzed
massive MIMO and mmWave whose benefits mainly come in detail. In the actual network, the number of UEs to be
from the physical layer, the HetNet with densely deployed served by the small cells is varying with time. When only a
small cells significantly helps increase the system capacity few UEs in the network are active, a large number of small
and mobile data rates over the network layer. In the hot cells may have less or even no traffic load in their cover-

2169-3536
2018 IEEE. Translations and content mining are permitted for academic research only.
VOLUME 6, 2018 Personal use is also permitted, but republication/redistribution requires IEEE permission. 29757
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

age area [10]. Meanwhile, if all the S-gNBs in the network the basis of power-awareness is proposed. The S-gNBs deter-
are fully active all the time, a large amount of S-gNBs’ mines whether to turn off one or several S-gNBs according
power is therefore wasted, resulting in a sharp drop in their to the amount of power consumption required for support-
power efficiency, and unfortunately at the same time causing ing UE communication. Specifically, the power consump-
serious interference among the small cells [11]. In order to tion values corresponding to the S-gNBs are arranged in
heighten power efficiency of the S-gNBs and reduce the descending order, and the ones with larger power consump-
co-frequency interference among the small cells, small cell tion are preferred to be turned off. However, the scheme only
on/off switching technology is under concern. According considers the power consumption of the S-gNBs for signal
to [12], the macro-cell base stations and the S-gNBs are transmission, ignoring the system traffic load loss caused
different in their power consumptions with different traffic by small cells that are switched off unnecessarily. Further-
loads. The power consumption of macro-cell base stations more, since the measurement of power consumption is greatly
increases exponentially with its traffic load in terms of the affected by the radio environment, it is difficult to obtain
number of UEs served, given that each UE has a constant data accurate measure in real time. These factors limit the effec-
rate requirement [13], while the S-gNBs power consumption tiveness of the algorithms. Different from the above methods,
is almost independent of its load and even constant for any Samarakoon et al. [23] proposed dynamic clustering based
load [14], [15]. Consequently, the small cell on/off switching on/off switching strategies, where the small cell on/off con-
technology helps save a lot of S-gNBs’ power and reduce co- trol operation is divided into two steps. In the first step,
channel interference among the small cells under the premise the cooperation and clustering of the small cells is performed,
of ensuring the stability of the system throughput [16], [17]. in which the small cells are partitioned into small cell clus-
In small cell on/off switching mechanism, the S-gNBs have ters by certain means. Then, the small cell on/off switching
two operation modes, i.e. active/sleep (On/Off) mode. In the algorithm is carried out in each cluster in the second step. The
sleep mode, the S-gNBs enter a dormant state and only peri- scheme achieves an on/off switching strategy that maximizes
odically send cell discovery signals (DS) for the purpose of the power efficiency of the S-gNBs based on the energy loss
potential UE access. equations for the small cell clusters under some constraints.
Existing literature has investigated the challenges of ame- However, this scheme encounters an NP hard problem for
liorating S-gNBs’ power efficiency under various scenarios. which seeking the optimal solution would trigger a compli-
In [18], a unified framework, in which the user associa- cated process and result in heavy signaling overhead, and the
tion strategies based on the maximum instantaneous received small cell on/off switching operations might be unnecessarily
power and the maximum average received power can be over-frequent.
studied, assuming log-normal shadowing, rayleigh fading With an aim to enhance the power efficiency of S-gNBs
and incorporating probabilistic non-line-of-sight (NLoS) and and mitigate the co-frequency interference among the small
line-of-sight (LoS) transmissions. In [19]and [20], a load- cells with less signaling overheads and measurement pro-
aware small cell on/off switching algorithm is proposed, cesses, a centralized small cell on/off switching algorithm
in which the network controller determines whether a small based on interference contribution ratio (ICR) is proposed
cell should enter the sleep state according to its load factor. in this paper. In the proposed algorithm, each UE in the
When the traffic load in the nearby small cells is increasing, network periodically reports the RSRP (reference signal
the network controller will send a command to the last closed received power) and cell ID of its serving S-gNB to the central
small cell, and ask it to enter the active state again. Although controller (or some specific macro basestation).1 The central
the algorithm can lessen the power consumption of S-gNBs, controller evaluates the total traffic volume of each S-gNB
there still exists a problem of measurement inaccuracy. For and the number of UEs it serves, based on the information
example, when the load comes from another small cell in off collected from the UEs. At the same time, the central con-
state instead of from the last closed small cell, turning on the troller also calculates the interference strength generated by
last closed small cell is obviously not the best option. With the each S-gNB to its neighboring small cells according to the
same idea, a distance-awareness basestation on/off switching total number of UEs served in each small cell and the network
scheme is proposed in [21]. Firstly, each S-gNB estimates adjacency matrix (NAM). Then, the decision and operation
its distances to their surrounding UEs. Then, the average of small cell on/off switching are performed according to
distance value for each small cell are calculated and arranged the ICR of each small cell and the distribution of traffic
in descending order. Finally, if resources available to the load in the network. Simulation results illustrate that the pro-
neighboring small cell can satisfy the UE requirements in its posed algorithm greatly increases the power efficiency of the
coverage, turning off the small cell with the largest average S-gNBs and lowers the amount of traffic load loss resulted
distance would be a highly reasonable option and it may from the dormant small cells.
only cause a tolerable QoS degradation. Since the scheme
needs to calculate the distances from all UEs to each S-gNB, 1 It is generally assumed in this paper that the central controller is a func-

the signaling overhead is apparently large, and chances are tional entity in the heterogeneous network where the macro basestation(s)
may spontaneously serve as the central controller(s) to fulfill the purpose of
that the benefit of small cell on/off switching may not be small cell on/off switching control. The central controller and the S-gNBs are
noticeable. In [22], a small cell on/off switching algorithm on also assumed to be able to exchange information over the X2 or Xn interface.

29758 VOLUME 6, 2018

B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

A. CONTRIBUTION of small cells that should be switched on and the subset

The main contributions of this paper are summarized as of the ones that should be turned off within the current
follows: switching period can be obtained respectively.
The rest of the paper is organized as follows. The sys-
• Definition of ICR. In the HetNets, we define the ICR tem model and the problem formulation are presented
of each small cell and design it as a critical trigger- in Section II and the ICR based small cells on/off switching
ing parameter of on/off switching to effectively initiate algorithm is given in Section III. Section IV provides simula-
small cell switching operations. For a specific small cell, tion results and conclusions are drawn in Section V.
the RSRP sum of all active UEs is defined as its target
signal strength, and the RSRP sum of other non-serving II. SYSTEM MODEL
UEs in the network is defined as its interference strength. In this section, the network scenario, power consump-
Then, the ratio of the interference signal strength to the tion model for S-gNBs, and the problem formulation are
target signal strength is regarded as a simple definition presented.
of ICR in this paper. In order to prevent a large num-
ber of UEs from handover operations caused by small
cell on/off switching, we refine the definition of ICR.
In the new method of ICR calculation, the ICR of a
small cell decreases as the number of UEs it is serving
increases.
• Calculation of interference strength for small cells.
With an aim to obtain the ICR value of each small
cell, each UE in the network needs to measure the
RSRP values corresponding to all the other small cells
in addition to its own serving small cell. In order to
simplify the calculation of ICR and decrease the amount
of UE’s measurements reported to the central controller,
the concept of NAM is proposed and designed in this
paper. We can calculate the ICR of each small cell in
the network in a quick and simple way according to FIGURE 1. A general depiction of HetNet scenario with dense deployment
the NAM and the information reported form the UEs. of small cells.
For the refined ICR calculation, each UE only needs
to feed back the RSRP value and cell ID of its serving
small cell. In other words, the central controller will no A. NETWORK MODEL
longer unnecessarily demand that each UE reports the We consider a dense heterogeneous network where the mech-
RSRP values and cell IDs of all its non-serving small anism of small cell on/off switching is employed, as shown
cells, and therefore tremendously reduces the amount of in Figure 1. In the network, the S-gNBs (or small cells) are
information feedback. denoted by a set S = {b1 , b2 , · · · , bM }, which satisfies that
• ICR based small cell on/off algorithm. In order to kSk0 = M (k.k0 indicates the set cardinality),2 and the
effectually perform small cell on/off switching, an ICR positions of S-gNBs are subject to Poisson’s point distribution
based on/off switching algorithm is designed and a within the macro-cell coverage area. The macro cell set is 9
dynamic small cell ICR threshold is defined in this and satisfies: k9k0 = N . In the network, the macro cells
paper. Specifically, in each on/off switching cycle, the occupy the low frequency band (< 2.5GHz) and provide
small cell subset, suppose it is denoted by A, that needs wide area coverage and UE location tracking, whereas the
to be turned on or off within the current switching cycle small cells operate in high frequency band (> 3.5GHz) to
is obtained according to the condition of traffic load in achieve stable high-speed data transmission and increase the
each small cell. An iterative decision process is proposed network capacity. The UE set in the network can be expressed
for each small cell in subset A. Firstly, in each iteration, as following:
the ICR value of each small cell in subset A and the small U = {k1 , k2 , · · · , kkUk0 }, kUk0 = L (1)
cell ICR threshold value of current iteration are calcu-
lated by using the NAM and the measurements of UEs. where L is the total number of UEs in the network. Assuming
Then, the maximum ICR value of small cell in A is that the subset of UEs served by S-gNB bi is Ubi , we also
compared with the ICR threshold in current iteration. obtain the total number of UEs in the network as
If the condition is satisfied, the S-gNB corresponding to
kUb1 k0 + kUb2 k0 + · · · + kUbM k0 = L (2)
the maximum ICR value will be turned off; otherwise,
the current iteration process stops and the next iteration 2 It is assumed that the sets and subsets in this paper take the vector form
begins. After all the iterations are completed, the subset when their set cardinalities are evaluated via k.k0 .

VOLUME 6, 2018 29759

 B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

It is assumed that each UE is equipped with a single and the total data rate of S-gNB bi can be calculated as:
antenna. In the network, the small cells and the macro cells X
establish connections and maintain information exchange λbi = λbi ,kl (9)
(including UE location, mobility, traffic load, and etc.) kl ∈Ubi
through an X2 air interface [24]. The S-gNBs are responsible
for collecting relevant measurements and information from B. PROBLEM FORMULATION
the UEs in the network and sharing them with the central It is assumed in addition that the macro cells are always
controller. The central controller obtains the decision on small active to maintain wide area coverage and track the UEs in
cell on/off switching by processing the measurements and the network, and the S-gNBs will perform on/off switching
information reported from the UEs and controls the on/off operations. Therefore, the S-gNBs have two operation modes,
switching operations of each S-gNB. namely On and Off. Since the macro base stations are always
Denote vbi , bi ∈ S as the indicators corresponding to the on, we only consider the power consumption of the S-gNBs
S-gNBs’ on/off status, i.e., vbi = 1 indicates that S-gNB bi is in this paper. The literature [26], [27] illustrate that the power
on, and vbi = 0 means that S-gNB bi is off. Accordingly, consumption corresponding to S-gNB bi can be divided into
we define the indicator set as V = {vb1 , vb2 , · · · , vbM }. two parts: basic power Pon tr
bi and transmit power Pbi , where
on
Pbi is the standby power consumption of the active S-gNB bi
If UE kl is connected to S-gNB bi , the service data rate
λbi ,kl obtained by UE kl from S-gNB bi can be calculated, resulted from components such as air conditioner and power
according to the Shannon formula, as: supply. When S-gNB bi is working in active mode, its total
power consumption PTotalbi can be expressed as:
λbi ,kl = Bkl log2 (1 + ϒbi ,kl ) (3)
PTotal
bi = Pon tr
bi + Pbi (10)
where Bkl is the bandwidth allocated by S-gNB bi for UE kl ,
ϒbi ,kl is the signal to interference plus noise ratio (SINR) of
X δPmax
Ptrbi = Lkl ,bi (11)
UE kl , which is defined in [25] as: Bbi η
kl ∈Ubi
vbi Pbi Gbi ,kl
ϒbi ,kl , P (4) where η is the efficiency of the basestation power amplifier,
vbj Pbj Gbj ,kl + N0 Bkl δ is the slope of the load-dependent power consumption of the
bj 6=bi ,bj ∈S
S-gNBs, and Pmax and Bbi denote the maximum transmission
where Gbi ,kl is the channel gain from S-gNB bi to UE kl power and maximum number of RBs for S-gNB bi , respec-
including the path loss and shadowing effects, Pbi is the tively. When S-gNB bi is off, the transmit power consumption
transmission power of S-gNB bi , and N0 is the noise power can be ignored, and the corresponding power consumption
spectral density. In order to guarantee the QoS of UEs at a can be reduced to Poff = ρPon bi , 0 < ρ < 1. Hence, the power
certain moment, S-gNB bi needs to allocate a certain amount consumption profile of S-gNB bi can be modeled as:
of time-frequency resource to its UEs according to the data
rate requirement of the UEs connected to itself. Accordingly, PTotal
bi = αbi Ptrbi vbi + (1 − ρ)Pon
bi vbi + ρPbi
on
(12)
the system load of S-gNB bi can be defined as in [26] as:
X k where αbi denotes the power consumption coefficient for
Lbi = l
dλmin /λ∗kl ,bi e (5) feeders and power amplifier of S-gNB bi . Based on the above
kl ∈Ubi analysis, we simplify this problem to the one of network
power efficiency maximization by jointly optimizing the net-
Lkl ,bi = dλkmin
l
/λ∗kl ,bi e (6) work accessing procedure and the S-gNBs on/off switching
λ∗bi ,kl = BRB log2 (1 + ϒbi ,kl ) (7) (i.e., determining the S-gNBs on/off indicator set V). We for-
mulate the S-gNBs power minimization problem as in [25] as
where Lbi represents the system load of S-gNB bi , λkminl
is the
vbi λbi ,kl
P P
minimum data rate requirement of UE kl , BRB denotes the
RB (resource block) bandwidth (e.g., 180KHz), Lkl ,bi denotes bi ∈S kl ∈Ubi
max
PTotal
P
the number of occupied RBs when UE kl is associated with bi
S-gNB bi (i.e. the number of time-frequency resource blocks bi ∈S
needed to satisfy the lowest bit rate of UE kl ), λ∗bi ,kl is the s.t. vbi ∈ {0, 1}, ∀bi ∈ S (13)
service rate of UE kl from S-gNB bi on one RB, kl ∈ Ubi
indicates that UE kl is served by S-gNB bi , and d·e stands for where the minimization of the total power consumption of all
the ceil operator. the S-gNBs in the network is actually a combination problem.
According to the above analysis, the total load of S-gNB In other words, the selection (or identification) of suitable
bi can be expressed as the total number of RBs required by S-gNBs among all the small cells to be switched on or off
all the UEs served by itself. Consequently, Eq. (3) can be is a combination problem that enables the whole network
expressed as: to consume the minimum power but it essentially requires
extremely high computational complexity in exhaustive com-
λbi ,kl = BRB Lkl ,bi vbi log2 (1 + ϒbi ,kl ) (8) binational search for the optimal solution.

29760 VOLUME 6, 2018

B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

III. SMALL CELLS ON/OFF ALGORITHM BASED ON ICR In order to prevent a large number of UEs from performing
In order to lessen the co-channel interference between small handover caused by small cell on/off switching, the calcula-
cells and ameliorate the power efficiency of S-gNBs, a small tion formula for the ICR of S-gNB bi is amended as:
cell on/off switching algorithm based on ICR is proposed.
Firstly, the ICRs of small cells in each switching cycle are 0bi = Rbi /(Rbi bi ) (17)
calculated by using the NAM and the UEs’ RSRP values of
their serving S-gNBs, and then the decision on small cells where bi = 12 ln(||Ubi ||0 ) is the correction factor. It can
to be switched on or off within the current switching cycle be seen from Eq. (17) that if the number of UEs served by
is obtained. The algorithm considers the small cells’ traffic S-gNB bi is getting large, the smaller the value of 0bi will be
load, the UE QoS, and the inter-cell interference simulta- calculated via Eq. (17), and the probability that S-gNB bi is
neously. Compared with the existing on/off switching algo- turned off is accordingly lower.
rithms based on the traffic load of small cells, the proposed
algorithm can effectively improve the power efficiency of S- B. CALCULATION OF INTERFERENCE STRENGTH
gNBs, restrain the interference among small cells and escalate VALUES FOR SMALL CELLS
the throughput of the network with less overhead, while From Eq. (14) and (15), for the purpose of obtaining the inter-
the traffic load loss resulted from small cell off-switching ference signal strength of each S-gNB, each UE in the net-
operations remains at a very low level. work needs to measure the RSRP values corresponding to all
the S-gNBs except its own serving S-gNB. This unfortunately
gives rise to a very large amount of power consumption of the
UEs, and therefore increases the computational delay on the
UE side and reduces the timeliness of small cell switching
operations as well. Moreover, in each switching cycle, each
UE needs to report a large amount of measurement results to
its serving S-gNB and the central controller, which undoubt-
edly increases the amount of UE feedback information and
consumes a large amount of time-frequency resources. For
this reason, we adopt an approximate calculation method,
FIGURE 2. The interference signal strength and target signal strength
of S-gNB b1 . as shown in Figure 3. Assuming that there are two S-gNBs
in the network, in which S-gNB b1 serves two UEs, S-gNB
b2 serves four UEs, and all UEs are randomly distributed.
A. ICR DEFINITION According to the approximation principle, the total interfer-
This section gives the physical definitions for the target signal ence signal strength generated by S-gNB b1 to the UEs served
strength Rbi and the interference signal strength Rbi of S-gNB by b2 can be expressed as:
bi , respectively. As shown in Figure 2, it is assumed that the
network is composed of two small cells, i.e. S-gNB b1 and Rb1 = Pb1 Gb1 ,k1 + Pb1 Gb1 ,k2 + Pb1 Gb1 ,k3 + Pb1 Gb1 ,k4
b2 . We define Rb1 as the sum of received signal strengths ∼
= 4Pb1 Gb1 ,b2 (18)
(i.e. RSRP) of all the UEs served by S-gNB b1 . The sum of
received signal strengths of S-gNB b1 for its non-served UEs where Gb1 ,b2 is the channel gain from S-gNB b1 to b2 . From
(i.e. UEs served by S-gNB b2 ) indicates the value of Rb1 . Eq. (18), in a general heterogeneous network, the interfer-
Accordingly, the target signal strength of S-gNB bi can be ence signal strength Rbi generated by S-gNB bi to other
generally defined as S-gNB bj ∈ S \ bi can be approximated by the number of
UEs served by bj , the path loss between bi and bj , and the
X X
Rbi , 3bi ,kl = Pbi Gbi ,kl (14)
kl ∈Ubi kl ∈Ubi
transmit power of bi . Consequently, the Rbi of S-gNB bi can
be approximated as:
where 3bi ,kl is the receive signal strength of UE kl from
S-gNB bi . X
Rbi = kUbj k0 Gbi ,bj Pbi (19)
Similarly, the Rbi of S-gNB bi can be defined as:
X X bj ∈S\bi
Rbi = 3bi ,kl = Pbi Gbi ,kl (15)
kl ∈U\Ubi kl ∈U\Ubi In order to further simplify the calculation of ICR,
we incorporate the NAM concept, as shown in Figure 4. It is
where A \ B means to delete the elements belonging to set B
assumed that there are five S-gNBs, namely A, B, C, D, and
in set A.
E in the network, where nA is the number of UEs served by
From the above analysis, the ICR corresponding to S-gNB
S-gNB A, and GAB is the channel gain between S-gNB
bi is defined as 0bi , which can be calculated as:
A and B. Based on the graph theory, the corresponding adja-
0bi , Rbi /Rbi (16) cency matrix of the graph (in this paper, this matrix is defined

VOLUME 6, 2018 29761

 B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

FIGURE 3. The approximation process for the calculation of interference signal strength.

as the network adjacency matrix) can be written as: From the above analysis, the interference signal strength of

0 GAB GAC GAD GAE
 each S-gNB to other active small cells in the network can be
GBA 0 GBC GBD GBE  generally expressed as:
 
H0 =  G
 CA G CB 0 GCD GCE   (20) R = [Rb1 , Rb2 , · · · , RbM ] = (H ∗ n)T ∗ P (23)
GDA GDB GDC 0 GDE 
GEA GEB GEC GED 0 n = [kUb1 k0 , kUb2 k0 , · · · , kUbM k0 ] T
(24)
 
0 Gb1 ,b2 ··· Gb1 ,bM
where H0 is the NAM of the network shown in Figure 4.  Gb ,b
 2 1 0 ··· Gb2 ,bM 

 Gb ,b
3 1 Gb3 ,b2 ··· Gb3 ,bM 
H=  ···
 (25)
 ··· ··· ···  
GbM −1 ,b1 GbM −1 ,b2 · · · GbM −1 ,bM 
Gb ,b GbM ,b2 ··· 0
 M 1 
Pb1 0 ··· 0 0
 0 Pb2 · · · 0 0 
 
P=  · · · · · · · · · · · · · ·· 
 (26)
 0 0 · · · PbM −1 0 
0 0 ··· 0 PbM

where R represents a vector formed by the interference

strength value of each small cell in the network, H is the
NAM, n = [nb1 , nb2 , · · · , nbM ]T denotes a vector formed by
the number of active UEs in each small cell, nbi is the number
of UEs served by S-gNB bi , and P represents a diagonal
matrix formed by the transmit power of each S-gNB in the
network. Setting Pb1 = Pb2 = · · · = PbM = Pt , we can
FIGURE 4. A simple network topology with five S-gNBs.
rewrite Eq. (23) in a more simplified form as
Then, the total interference signal strength vector of the R = [Rb1 , Rb2 , · · · , RbM ] = (H ∗ n ∗ Pt )T (27)
five S-gNBs can be given by
Since the geographical positions of all the S-gNBs are gen-
R0 = [RA , RB , RC , RD , RE ] = (H0 ∗ n0 )T ∗ P0 (21)
  erally fixed in practical network, the channel gains between
PA 0 0 0 0
0 the S-gNBs can be considered to be constant or less volatile.
PB 0 0 0
  Furthermore, for a heterogeneous network, once the S-gNBs
P0 = 
 0 0 P C 0 0 (22)
0
 are readily deployed, we can consider its corresponding NAM
0 0 PD 0
is determined accordingly. Therefore, once the network topol-
0 0 0 0 PE
ogy is given, the NAM can be measured in the off-line
where the vector R0 is formed by the interference signal phase in advance and stored in the central controller. When
strength values of S-gNB A, B, C, D, and E, P0 is a diag- the central controller performs small cell on/off switching
onal matrix formed by the transmit powers corresponding to operations, active UEs in the network only need to report their
S-gNB A, B, C, D, and E, n0 = [nA , nB , nC , nD , nE ]T repre- serving small cell IDs and the corresponding RSRP values.
sents a vector formed by the number of active UEs in small
cell A, B, C, D, and E respectively, and RX (X ∈ Sexample = C. DEFINITION OF DYNAMIC ICR THRESHOLD
{A, B, C, D, E}) is the sum of interference signal strength of In order to effectively obtain the decisions of small cells to
S-gNB X to the UEs served by S-gNBs in Sexample \X . be switched on or off, we define a dynamic ICR threshold

29762 VOLUME 6, 2018

B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

calculated as: UE periodically reports the routine data including the RSRP
P
Pbj Gbi ,bj of its serving S-gNB and the corresponding cell ID to the
1 X bj ∈Son \bi central controller. During a certain switching period, when a
0th = (28) S-gNB is in the off state, UEs in the proximity may determine
kSon k0 Pbi Gbi bi
bi ∈Son whether there is a serving S-gNB by performing small cell
where Gbi is the channel gain from S-gNB bi to its coverage discovery and RSRP measurement with the small cell dis-
edge, and Son is the set of active S-gNBs in the network. covery signal broadcast by the dormant S-gNB. Specifically,
By Eq. (28), we can easily calculate the ICR threshold the timing of on/off switching process is shown in Figure 6.
under current network conditions. Within the small cell on/off We assume that the RSRP values of UEs served by S-gNB
switching cycle, by comparing the ICR value of a given bi are saved in the vector rbi , and the vector n is composed
S-gNB, that may need on/off switching operation, with the of the number of UEs being served in each small cell. In the
threshold value 0th , the small cell on/off switching decision proposed algorithm, within the γ th switching cycle, S-gNB
can be obtained. bi must be switched on for the following two cases:
γ
• The traffic load of S-gNB bi satisfies: Lbi > Lth , where
γ
Lbi is traffic load of bi in the γ th switching cycle, and
Lth is the traffic load threshold.
γ γ γ
• The vector rbi satisfies: max(rbi ) > rth , where rbi is
a vector formed by RSRP values of the UEs served
by S-gNB bi , rth represents the received signal strength
threshold of the UEs.
According to the above analysis, the specific steps of the
proposed ICR-based small cell switching on/off algorithm are
as follows:
Step 1: During the γ th small cell switching cycle, the cen-
γ
tral controller first obtains the set Sact containing the S-gNBs
γ
that must be turned on and the set Soff composed of the
S-gNBs that must be turned off according to the measure-
ments and information collected from the UEs and S-gNBs.
It is worth noting that the measurement and information
γ
include the vector rbi = [r1 , r2 , · · · , rnbi ]T formed by
FIGURE 5. Calculation of the ICR threshold.
RSRP values of the UEs served by S-gNB bi and the vector
nγ = [nb1 , nb2 , · · · , nbM ]T formed by the number of
A simple network including three small cells is given as an active UEs in each small cell, satisfying Eq. (29) and (30)
example to describe Eq. (28). As shown in Figure 5, assuming respectively as:
that S-gNB b1 , b2 and b3 are all on in the network. The steps
γ γ γ
for calculating the ICR threshold under current network oper- max(rbi ) > rth or Lbi > Lth , bi ∈ Sact (29)
ating condition are as follows: Firstly, the ICR values of b1 , b2 γ
nbj = 0, bj ∈ Soff (30)
and b3 are calculated via Eq. (17). Taking b1 as an example,
Pb1 Gb1 , b2 +Pb1 Gb1 , b3
the corresponding ICR value is: 0b1 = Pb1 Gb1 b1 . where the value of rth is set as 10 ∗ Pt Gbi in this paper.
Similarly, the ICR value 0b2 and 0b3 corresponding to S-gNB The value of rth can be adjusted according to the practi-
b2 and b3 can be calculated respectively. Then, averaging cally allowed tolerance of traffic loss caused by the S-gNB
the ICR values corresponding to small cells b1 , b2 and b3 , switching-off operations in the network. Furthermore, we can
γ γ γ
we can obtain the ICR threshold of the current network as obtain the subset SD = S \ (Sact ∪ Soff ) consisting of
0b +0b +0b S-gNBs that will be decided whether to perform the on/off
0th = 1 32 3 .
switching operation or not. The specific algorithm flow is
D. ICR BASED SMALL CELL ON/OFF ALGORITHM
given in Algorithm 1.
Step 2: Use Eq. (14) to calculate the target signal strength
It is apparently reasonable to assume that the NAM of a γ
of each S-gNB in the small cell subset SD within the γ th
certain network has been obtained and saved to the central
switching cycle. Let the set of the target signal strength values
controller in advance. The UEs in the network are supposed γ γ
corresponding to the S-gNBs in subset SD be denoted by RD ,
to be equipped with dual-link capability. In other words, each γ
Rs represent the target signal strength corresponding to the
UE establishes connections with its serving S-gNB and the γ
sth S-gNB in the subset SD , and we obtain
central controller at the same time.3 In the network, each
γ γ
3 If direct connection to the central controller is not feasible for the UEs, Rγs = krSγ (s) k1 , s = 1, 2, · · · , ||SD ||0 (31)
D
we assume that the S-gNBs can serve as the relay nodes between the central γ γ γ γ
controller and the UEs. RD = [R1 , R2 , · · · , R||Sγ || ] (32)
D 0

VOLUME 6, 2018 29763

 B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

FIGURE 6. The process of small cell On/Off switching.

Algorithm 1 Traffic Load Based Classification of Small Algorithm 2 ICR Based Small Cell On/Off Switching
γ γ γ
Cells
γ γ γ γ
Require: SD ; nγ ; Son ; Z = ||SD ||0 ;
Require: H; rbi , bi ∈ S; Soff = ∅; Sact = ∅; SD = ∅ γ
1: Utilizing Eq. (34) and (35) to obtain RD ;
1: for i = 0 to M do 2: for q = 0 to Z do
γ γ
2: if max(rbi ) > rth or Lbi > Lth then 3:
γ
Obtaining the 0q by P Eq. (36), (37) and (38);
γ γ
3: Sact = [Sact , bi ] γ
Pt Gbi ,bj
q bj ∈Son \bi
4: else if nbi = 0 then 4: 0th = 1
γ
P
;
γ γ Pt Gbi bi
5: Soff = [Soff , bi ]; kSon k0 γ
bi ∈Son
γ
6: end if 5: b∗ = argmax(0q );
γ ∗ q
7: end for 6: if 0q (b ) > 0th then
γ γ γ γ γ γ γ γ
Ensure: SD = S \ (Sact ∪ Soff ); Son = S \ Soff ; 7: Soff = [Soff , SD (b∗ )];
γ γ γ γ
8: nq+1 (b∗ ) = 0; Son = Son \ SD (b∗ );
9: update: nγ ;
γ γ 10: else
where SD (s) represents the sth S-gNB in SD , and || ∗ ||1
represents the sum of the absolute values of all elements in 11: Break;
its input vector. 12: end if
γ 13: end for
Step 3: Calculate the interference signal strength value Rbj γ γ
γ q Ensure: Son ; Soff ;
of S-gNB bj (bj ∈ SD ) and the ICR threshold value 0th in
current iteration (assuming it is the qth iteration) via Eq. (33),
(34), and (35):
γ
P
Pt Gbi ,bj where 0q is the vector composed of the ICR value of each
γ γ
1 X bj ∈Sγon \bi S-gNB in subset SD ,  q is the vector formed by the ICR
q γ γ
0th = γ (33) correction coefficient of each S-gNB in subset SD , and nq is
kSon k0 γ Pt Gbi bi
bi ∈Son a vector formed by the number of UEs served by each S-gNB
γ γ
γ γ
R = H ∗ n ∗ Pt = [Rb1 , Rb2 , · · · , RbM ]T
γ γ
(34) in set SD at the qth iteration. The operators in ‘‘A . ∗ B’’ and
γ γ γ γ ‘‘A./B’’ respectively indicate that the entries with the same
RD = [RSγ (1) , RSγ (2) , · · · , RSγ (Z ) ]T (35) index in A and B are element-wise multiplied and divided.
D D D
γ
γ Step 5: Based on Step 4, use 0q to iteratively obtain the
where Son is composed of S-gNBs whose status is ‘‘on’’ in γ
γ on/off switching decision for the S-gNBs in SD . In the qth iter-
the γ th small cell switching cycle, R is the vector formed γ ∗
ation, when the maximum ICR value 0q (b ) corresponding
by the interference strength value of each small cell in small γ γ q
γ to the small cell subset SD satisfies: 0q (b∗ ) > 0th , we decide
cell set S, RD is the vector formed by the interference strength γ ∗
γ γ to turn off the S-gNB SD (b ) and obtain the updates as:
value of each S-gNB in SD , and Z = ||SD ||0 is the number of
γ γ
elements in subset SD . nq+1 (b∗ ) = 0 (39)
Step 4: The ICR value corresponding to each S-gNB in the γ γ γ
γ Son = Son \ SD (b∗ ) (40)
subset SD is calculated by Eq. (36), (37) and (38) in current
qth iteration: Finally, the algorithm returns to Step 3 to further handle the
γ γ
0qγ = 2RD ./((RD ) . ∗  γq ) (36) rest S-gNBs; otherwise, it just exits the on/off switching deci-
γ γ
sion iteration. In this way, the small cell subset Soff and Son
 γq = [ln(nγq (1)), ln(nγq (2)), · · · , ln(nγq (Z ))] (37) can be obtained. Details of the ICR based on/off switching
nγq = [n γ
SD (1) ,n γ
SD (2) ,··· ,n γ
SD (Z ) ] (38) algorithm are given in Algorithm 2.

29764 VOLUME 6, 2018

B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

In the γ th switching cycle, small cell switching opera- • Traffic evaluation based on/off scheme: In each
tions are performed according to the obtained S-gNB subsets switching cycle, the system determines a specific set
γ γ
Son and Soff . To be more specific, for the S-gNBs in the subset of small cells with smaller traffic load in the network
γ
Son , the original state of a S-gNB will be maintained if its state and performs small cell switching-off operation over the
is ‘‘on’’ in the (γ −1)th switching cycle; otherwise, the S-gNB small cells in the set. The rest small cells are kept active.
will be activated from the previous ‘‘off’’ state. Similarly, for • ICR based On/Off algorithm (the proposed scheme):
γ
the S-gNBs in the subset Soff , the original state of a S-gNB is The S-gNBs perform on/off switching operations based
maintained if its state is ‘‘off’’ in the (γ −1)th switching cycle; on the ICR and traffic load condition of the small cells.
otherwise, it needs to be switched off. The entire flowchart of In each switching cycle, each UE in the network period-
the proposed algorithm is illustrated in Figure 7. ically reports the RSRP and cell ID of its serving S-gNB
to the central controller. The total traffic load of each
IV. SIMULATION AND ANALYSIS small cell can be obtained according to the information
A. SIMULATION SETTING AND PARAMETERS collected from the UEs being served within that small
To verify the proposed small cell on/off switching algorithm, cell. Meanwhile, the ICR value of each small cell is
a dense heterogeneous network composed of three macro calculated via the number of UEs served by itself and
cells (||9||0 = 3) and 250 small cells (kSk0 = 250) is the network adjacency matrix. With the ICR value of
considered. All small cells are located within the coverage each small cell and traffic load distribution, the small
area of the macro cells, and the locations of the S-gNBs are cell on/off switching algorithm proposed in this paper is
subject to the Poisson point distribution. In order to guarantee performed.
the network coverage area is not varying, it is assumed that
the macro base stations are always powered on. The UEs in C. SIMULATION RESULTS AND ANALYSIS
the network are distributed randomly. The path-loss models Figure 8 shows the relationship between the number of
used in simulation are [28]: S-gNBs being turned off and the number of UEs in the net-
work. As can be seen from the figure, the number of S-gNBs
PL 1 = 128.1 + 37.6log10 (R) (41) that can be turned off in the random on/off scheme and the
PL 2 = 140.1 + 36.7log10 (R) (42) traffic evaluation based on/off scheme is independent of the
number of UEs in the network and fixed at a predetermined
where Eq. (41) corresponds to distance dependent path loss switching-off ratio. When the number of UEs in the network
from Macro base stations to UEs and Eq. (42) corresponds to is kept at a low or middle level, the proposed ICR algorithm is
that from S-gNBs to UEs. Hereby R is the distance between capable of switching off a larger portion of the S-gNBs, com-
the transmitter and receiver in kilometers. Detailed simulation pared with the other two schemes with different switching-off
parameters are the same as [25] and [29] and given in Table 1. ratios. This essentially helps attain higher power efficiency.
However, in order to ensure that most UEs can be effectively
TABLE 1. Simulation parameters. connected with their appropriate serving S-gNBs when the
number of UEs in the network is increasing, the number of
S-gNBs that can be turned off decreases significantly for the
ICR based On/Off algorithm.
The power consumption shown in Figure 9 is defined as
the overall power consumed by all the S-gNBs in the net-
work, which is calculated via the S-gNBs power consump-
tion model introduced in Section 2.2. In order to verify the
power efficiency of the proposed scheme, we evaluate the
S-gNBs power consumption corresponding to the aforemen-
tioned four schemes respectively. Since the small cell on/off
switching mechanism is incorporated in the network, it can be
clearly seen that much more S-gNBs power can be saved in
B. SCHEMES UNDER COMPARISON the on/off switching schemes than the conventional scheme.
To facilitate comparison and analysis, we make the following In particular, with the increase of the proportion of the
scheme notations: S-gNBs being switched off (for instance from 20% to 30%),
• Conventional scheme: All S-gNBs remain active in the the number of S-gNBs that are turned off in corresponding
network without incorporating on/off switching mecha- schemes also increased. The power consumption of S-gNBs
nism. has thus decreased remarkably in the random on/off scheme
• Random On/Off scheme: During each small cell and the traffic evaluation based on/off scheme. As the num-
switching on/off cycle, the central controller randomly ber of UEs increases, the total S-gNBs power consumption
selects and determines a certain percentage of small cells in the random on/off scheme and traffic evaluation based
to be turned off and the other small cells remain active. on/off scheme increases slightly, due to the increasing power

VOLUME 6, 2018 29765

 B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

FIGURE 7. The entire flowchart of the proposed algorithm.

FIGURE 8. The relationship between the number of S-gNBs turned off FIGURE 9. The relationship between the total power consumption of
and the number of UEs in the network. S-gNBs and the number of UEs in the network.

consumption of more connections provided by the S-gNBs in The normalized traffic load losses due to small cell on/off
the network. In contrast, for the proposed scheme, the number switching is given in Figure 10. The traffic load losses of
of S-gNBs that can be turned off will decrease when the the random on/off scheme and the traffic evaluation based
UE number is becoming larger, which further results in that on/off scheme are obviously much higher than that of the
the power consumption of S-gNBs is enlarged significantly. ICR based on/off algorithm, when the allowed maximum
In the case of a small number of UEs in the network, since a ratio of small cells being turned off is 20%, 30% and 40%
large number of small cells have been turned off, the proposed respectively. As straightforwardly expected, the traffic eval-
scheme has apparently more prominent advantages compared uation based on/off scheme outperforms the random on/off
with the other three schemes in saving power consumption of scheme. When the number of UEs increases in the network,
the S-gNBs. the traffic loss of the proposed scheme basically remains at

29766 VOLUME 6, 2018

B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

FIGURE 11. The cumulative distribution function of SINR values of UEs in

the conventional scheme and the ICR based On/Off algorithm.

FIGURE 10. The relationship between the amount of normalized traffic

loss of small cells and the number of UEs in the network.

a much lower level, compared with the other two schemes.

For the random on/off scheme and traffic evaluation based
on/off scheme, the traffic loss of small cells has increased in
a relatively sharp manner when the number of UEs increases,
and in each switching cycle, the larger the ratio of small
cells being switched off is allowed, the more the traffic loss
becomes. For the ICR based on/off algorithm, the small cells
are turned off according to the traffic loads of S-gNBs and
the ICR value of each small cell, which more potentially
considers the problem that UEs cannot establish connection
with the small cells to be switched off. Therefore, in each
switching cycle, the proposed scheme cannot only effectively
turn off some S-gNBs, but also ensure that most UEs can FIGURE 12. The relationship between the total network rate and the
effectively establish connections with their appropriate small number of UEs in the network.

cells, causing only slight traffic loss. On the contrary, in the

random on/off scheme, a certain proportion of S-gNBs are
randomly switched off. As the number of UEs in the network more severe. However, in the conventional scheme, since all
increases, due to the fact that their serving S-gNBs may have the S-gNBs in the network are always on, the SINR of UEs
been turned off already, where probably some UEs cannot does not change with the number of UEs in the network at all
successfully handover to or reselect the neighboring small and it remains at a low level all the time, compared with the
cells, a large traffic loss will undoubtedly occur. As for the proposed scheme.
traffic evaluation based on/off scheme, it merely considers The relationship between the total network data rate and the
the amount of traffic load in small cells in a simple way, and number of UEs is shown in Figure 12. In the ICR based on/off
hence it is straightforward that when the number of UEs is algorithm, the total network data rate is obviously higher
large, the traffic loss of small cells proportionally rises. than those of the other three schemes. In addition, the traffic
Figure 11 illustrates the cumulative distribution function of evaluation based on/off scheme achieves higher total data rate
SINR values of UEs. It can be seen that with different number than the conventional scheme and the random on/off scheme
of UEs, the SINR value of UEs with the proposed algorithm has the lowest total data rate. For the traffic evaluation based
is always higher than that of the conventional scheme, indi- on/off scheme and the ICR based on/off algorithm, small cell
cating that the proposed algorithm can effectively suppress on/off switching are verified to effectively improve the SINR
the inter-cell interference. As the number of UEs increases, of UEs in the network. Meanwhile, the reduction of traffic
in order that most UEs can effectively establish connections load loss caused by small cell on/off switching is relatively
with the S-gNBs, the number of small cells satisfying the con- higher than that of the random on/off scheme. Due to the
dition of being turned off decreases in the proposed scheme, fact that a certain proportion of S-gNBs are switched off
and hence the interferences between small cells become periodically in time and randomly among the small cells in

VOLUME 6, 2018 29767

 B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

the random on/off scheme, the small cell switching on/off the traffic evaluation based on/off scheme is lower than that
operation gives rise to a higher amount of traffic losses and of the ICR based on/off algorithm, resulting in a decrease of
results in a decline in the overall data rate compared with the total system rate and a lower power efficiency of S-gNBs.
the conventional scheme. Different from the traffic evaluation In summary, compared with the other three schemes,
based on/off scheme, the number of S-gNBs that can be the proposed algorithm can not only perform small cell on/off
switched off in the proposed scheme in each switching cycle switching operations effectively, but also ensure that the loss
is determined according to the traffic load distribution of the of traffic load in small cells is maintained at a low level and
network and the ICR of each small cell. As a result, the per- the co-channel interference among the small cells is sup-
formance of the ICR based on/off algorithm is better than pressed to some extent. In addition, in each switching cycle,
that of the other three schemes. In addition, as the number the central controller can quickly and effectively obtain the
of UEs increases, the proposed scheme has more apparent ICR value of each small cell with the NAM, which demands
advantages in terms of reducing the traffic losses than the less UE measurements and hence accelerate the small cell
other three schemes. on/off switching operations.

V. CONCLUSIONS
Considering the power consumption of S-gNBs and the
co-channel interference among the small cells in dense
5G HetNets, we propose an ICR based small cell on/off
switching algorithm. The proposed algorithm designs and
employs an innovative concept of ICR as a trigger parameter
for small cell on/off switching. To decrease the total amount
of UEs’ measurements, the network adjacency matrix and
an ICR calculation method are incorporated, where the ICR
value of each small cell can be obtained in a simple and
quick way. Aided by the network adjacency matrix, the deci-
sion on selecting suitable small cells for on/off switching is
obtained via an iterative identification procedure. Compared
with the conventional small cell on/off switching strategies,
the proposed algorithm improves the power efficiency of
S-gNBs and reduce the co-channel interference between
small cells with lower traffic loss of the small cells. At the
same time, the signaling information between the UEs and
FIGURE 13. Comparisons of power efficiency in the network.
S-gNBs involved in the small cell switching mechanism is
greatly reduced. These technical merits enable the proposed
In Figure 13, the power efficiency comparison between algorithm to serve as a practical candidate solution for power-
the four schemes is given. As can be seen, the random efficient dense deployment of S-gNBs in 5G network.
on/off scheme, the traffic evaluation based on/off scheme, and
the ICR based on/off algorithm can effectively improve the REFERENCES
power efficiency of S-gNBs, compared with the conventional [1] T. E. Bogale and L. B. Le, ‘‘Massive MIMO and mmWave for 5G wireless
HetNet: Potential benefits and challenges,’’ IEEE Veh. Technol. Mag.,
scheme. This is due to the fact that, S-gNBs are always on in vol. 11, no. 1, pp. 64–75, Mar. 2016.
the conventional scheme, the power consumption of S-gNBs [2] S. Hur et al., ‘‘Proposal on millimeter-wave channel modeling for 5G cel-
is therefore relatively large and the co-channel interference lular system,’’ IEEE J. Sel. Topics Signal Process., vol. 10, no. 3,
pp. 454–469, Apr. 2016.
of the small cells cannot be manipulated. Compared with [3] X. Liu, Z. Pan, N. Liu, and X. You, ‘‘Downlink SINR and rate distribution
the other three schemes, the proposed scheme has the best of ultra-dense HetNets with burst traffic,’’ China Commun., vol. 13, no. 9,
power efficiency performance. Since some S-gNBs necessary pp. 24–32, Sep. 2016.
[4] X. Ge, S. Tu, G. Mao, and C. X. Wang, ‘‘5G ultra-dense cellular networks,’’
to be switched off can be identified effectively and the SINR IEEE Trans. Wireless Commun., vol. 23, no. 1, pp. 72–79, Feb. 2016.
of UEs’ received signal can be improved prominently with [5] M. Peng, Y. Li, Z. Zhao, and C. Wang, ‘‘System architecture and key
smaller traffic load loss, the total network date rate of the technologies for 5G heterogeneous cloud radio access networks,’’ IEEE
Netw., vol. 29, no. 2, pp. 6–14, Mar./Apr. 2015.
proposed scheme is larger than the other three schemes. [6] E. Hossain, M. Rasti, H. Tabassum, and A. Abdelnasser, ‘‘Evolu-
Although a fixed portion of S-gNBs can be turned off in each tion toward 5G multi-tier cellular wireless networks: An interference
cycle of the random on/off scheme and the traffic evaluation management perspective,’’ IEEE Wireless Commun., vol. 21, no. 3,
pp. 118–127, Jun. 2014.
based on/off scheme, the traffic load loss caused by small [7] C. Liu, B. Natarajan, and H. Xia, ‘‘Small cell base station sleep strate-
cells being switched off is larger than that of the ICR based gies for energy efficiency,’’ IEEE Trans. Veh. Technol., vol. 65, no. 3,
on/off algorithm on account that many UEs cannot estab- pp. 1652–1661, Mar. 2016.
[8] J. Xu and R. Zhang, ‘‘Cooperative energy trading in CoMP systems
lish effective connections with some small cells. Moreover, powered by smart grids,’’ IEEE Trans. Veh. Technol., vol. 65, no. 4,
the SINR value of the UEs in the random on/off scheme and pp. 2142–2153, Apr. 2016.

29768 VOLUME 6, 2018

B. Shen et al.: ICR Based Small Cells On/Off Switching Algorithm for 5G Dense HetNet

[9] B. Yang, G. Mao, M. Ding, X. Ge, and X. Tao, ‘‘Dense small cell networks: BIN SHEN received the M.Sc. degree in com-
From noise-limited to dense interference-limited,’’ IEEE Trans. Veh. munication engineering from the University of
Technol., vol. 67, no. 5, pp. 4262–4277, May 2018. Electronic Science and Technology of China
[10] L. Tang, W. Wang, Y. Wang, and Q. Chen, ‘‘An energy-saving algorithm in 2005 and the Ph.D. degree in communica-
with joint user association, clustering, and ON/OFF strategies in dense het- tion engineering from Inha University, South
erogeneous networks,’’ IEEE Access, vol. 5, pp. 12988–13000, Jul. 2017. Korea, in 2010. He is currently a Professor with
[11] M. Feng, S. Mao, and T. Jiang, ‘‘Base station ON-OFF switching in 5G the School of Communication and Information
wireless networks: Approaches and challenges,’’ IEEE Wireless Commun.,
Engineering, Chongqing University of Posts and
vol. 24, no. 4, pp. 46–54, Aug. 2017.
Telecommunications. His research interests are
[12] S. Cai, Y. Che, L. Duan, J. Wang, S. Zhou, and R. Zhang, ‘‘Green 5G
heterogeneous networks through dynamic small-cell operation,’’ IEEE in statistical signal processing, MIMO systems,
J. Sel. Areas Commun., vol. 34, no. 5, pp. 1103–1115, May 2016. and cognitive radios.
[13] S. Luo, R. Zhang, and T. J. Lim, ‘‘Optimal power and range adaptation
for green broadcasting,’’ IEEE Trans. Wireless Commun., vol. 12, no. 9,
pp. 4592–4603, Sep. 2013.
[14] S. Buzzi, C.-L. I, T. E. Klein, H. V. Poor, C. Yang, and A. Zappone,
‘‘A survey of energy-efficient techniques for 5G networks and challenges
ahead,’’ IEEE J. Sel. Areas Commun., vol. 34, no. 4, pp. 697–709,
Apr. 2016. ZHENZHU LEI received the B.S. degree from
[15] W. Vereecken et al., ‘‘Evaluation of the potential for energy saving in
Henan Polytechnic University, Henan, China,
macrocell and femtocell networks using a heuristic introducing sleep
in 2015. He is currently pursuing the M.S. degree
modes in base stations,’’ EURASIP J. Wireless Commun. Netw., vol. 2012,
no. 1, p. 170, Dec. 2012. in information and communication engineer-
[16] A. Kumar and C. Rosenberg, ‘‘Energy and throughput trade-offs in cellular ing with the Wireless Transmission Laboratory,
networks using base station switching,’’ IEEE Trans. Mobile Comput., Chongqing University of Posts and Telecommu-
vol. 15, no. 2, pp. 364–376, Feb. 2016. nications, Chongqing, China. His main research
[17] Y. L. Che, L. Duan, and R. Zhang, ‘‘Dynamic base station operation in interest is dense heterogeneous networks.
large-scale green cellular networks,’’ IEEE J. Sel. Areas Commun., vol. 34,
no. 12, pp. 3127–3141, Dec. 2016.
[18] B. Yang, G. Mao, X. Ge, M. Ding, and X. Yang, ‘‘On the energy-
efficient deployment for ultra-dense heterogeneous networks with NLoS
and LoS transmissions,’’ IEEE Trans. Green Commun. Netw., vol. 2, no. 2,
pp. 369–384, Jun. 2017.
[19] S. O. Elbassiouny, A. Elhamy, and A. S. Ibrahim, ‘‘Traffic-aware user
association technique for dynamic ON/OFF switching of small cells,’’ in
Proc. IEEE Wireless Commun. Netw. Conf., Mar. 2015, pp. 866–871. XIAOGE HUANG received the Ph.D. degree
[20] Z. Li, D. Grace, and P. Mitchell, ‘‘Traffic-aware cell management for green (Hons.) from the Group of Information and
ultradense small-cell networks,’’ IEEE Trans. Veh. Technol., vol. 66, no. 3, Communication Systems, Institute of Robotics
pp. 2600–2614, Mar. 2017. and Information and Communication Technolo-
[21] A. Bousia, A. Antonopoulos, L. Alonso, and C. Verikoukis, ‘‘‘Green’ gies, University of Valencia, Spain. In 2013, she
distance-aware base station sleeping algorithm in LTE-Advanced,’’ in joined the Group of Wireless Communication
Proc. IEEE Int. Conf. Commun. (ICC), Jun. 2012, pp. 1347–1351. Technology, Chongqing University of Posts and
[22] T. Elshabrawy and R. Mourad, ‘‘Power-aware ON/OFF switching strate- telecommunications, as an Associate Professor.
gies of eNodeB for green LTE networks,’’ in Proc. IEEE Int. Conf. New Her research interests include convex optimiza-
Technol., Mobility Secur., Mar. 2014, pp. 1–4.
tion, centralized and decentralized power alloca-
[23] S. Samarakoon, M. Bennis, W. Saad, and M. Latva-Aho, ‘‘Dynamic clus-
tion strategies, game theory, and cognitive radio networks.
tering and ON/OFF strategies for wireless small cell networks,’’ IEEE
Trans. Wireless Commun., vol. 15, no. 3, pp. 2164–2178, Mar. 2016.
[24] S. Deb, P. Monogioudis, J. Miernik, and J. P. Seymour, ‘‘Algorithms for
enhanced inter-cell interference coordination (eICIC) in LTE HetNets,’’
IEEE/ACM Trans. Netw., vol. 22, no. 1, pp. 137–150, Feb. 2014.
[25] N. Yu, Y. Miao, L. Mu, H. Du, H. Huang, and X. Jia, ‘‘Minimizing energy
cost by dynamic switching ON/OFF base stations in cellular networks,’’
IEEE Trans. Wireless Commun., vol. 15, no. 11, pp. 7457–7469, Nov. 2016. QIANBIN CHEN received the Ph.D. degree in
[26] L. Dong, G. Wu, Z. Xu, and S. Li, ‘‘Energy efficient pico base station communication and information systems from the
switching-ON/OFF in heterogeneous cellular network with minimum rate University of Electronic Science and Technology
requirement,’’ in Proc. IEEE 6th Int. Conf. Wireless Commun. Signal
of China, Chengdu, China, in 2002. He is cur-
Process., Oct. 2014, pp. 1–6.
rently a Professor with the School of Communi-
[27] H. Ghazzai, M. J. Farooq, A. Alsharoa, E. Yaacoub, A. Kadri, and
M.-S. Alouini, ‘‘Green networking in cellular HetNets: A unified radio cation and Information Engineering, Chongqing
resource management framework with base station ON/OFF switching,’’ University of Posts and Telecommunications, and
IEEE Trans. Veh. Technol., vol. 66, no. 7, pp. 5879–5893, Jul. 2016. the Director of the Chongqing Key Laboratory
[28] Further Advancements of E-UTRA: Physical Layer Aspects, of Mobile Communication Technology. He has
document 3GPP TS 36.814, Feb. 2009. authored or co-authored over 100 papers in jour-
[29] L. Li, M. Peng, C. Yang, and Y. Wu, ‘‘Optimization of base-station density nals and peer-reviewed conference proceedings, and has co-authored seven
for high energy-efficient cellular networks with sleeping strategies,’’ IEEE books. He holds 47 granted national patents.
Trans. Wireless Commun., vol. 65, no. 9, pp. 7501–7514, Sep. 2016.

VOLUME 6, 2018 29769