4832 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO.

7, JULY 2017

Joint Energy-Spectral-Efficiency Optimization

of CoMP and BS Deployment in Dense
Large-Scale Cellular Networks
Guogang Zhao, Sheng Chen, Fellow, IEEE, Liqiang Zhao, Member, IEEE, and Lajos Hanzo, Fellow, IEEE

Abstract— In this paper, the energy-spectral efficiency (ESE) resources for supporting the substantial increase of mobile
benefiting from the joint optimization of coordinated multi- devices involving high-rate applications in the fifth genera-
point (CoMP) transmission and base station (BS) deployment tion (5G) cellular network [1]–[3]. It is widely recognized
is evaluated in the context of dense large-scale cellular network.
We first derive a closed-form network ESE expression for a large- that a promising technique of increasing the network capacity
scale CoMP-enhanced network, which allows us to quantify the is that of shrinking the coverage area of base stations (BSs)
influence of key network parameters on the achievable network with the spectral resource spatially reused as densely as possi-
ESE, including the BS density and the cooperation activation ble [2]–[5]. However, the tremendous increase in the number
probability, characterized by a CoMP activation factor as well of BSs and mobile devices will inevitably lead to fast-rising
as the users’ behaviors, such as their geographical mobile-traffic
intensity and average user rate. With the aid of this tractable energy cost and severe interference caused by the massive
ESE expression and for a given BS density, we next formulate number of BSs deployed in self-organized patterns. Hence, 5G
a cellular-scenario-aware CoMP activation optimization problem must also aim for an improved energy efficiency (EE) [2]–[8].
while considering the users’ outage probability as constraints to A substantial amount of energy is consumed by the BSs
maximize the network’s ESE. We then jointly optimize the CoMP both for signal transmission by the power amplifiers and for
activation factor and the BS density to maximize the network
ESE, again under the constraint of the users’ outage probability. other related power consumptions such as the air conditioners
Our simulation results confirm the accuracy of our analysis and [7]. Thus, the network’s energy dissipation is predominantly
verify the impact of several key parameters on the network determined by the BS density and by how BSs operate. The
ESE. Finally, the ESE improvement of our proposed strategies term ‘BS deployment’ in this context refers to the positioning
is evaluated under diverse scenarios, which provides valuable of the BSs and how the deployed BSs operate, i.e., switched
insight into the joint CoMP and BS deployment optimization in
dense large-scale cellular networks. on or off. Traditional BS deployment is designed for worst-
case ubiquitous coverage scenarios, which often results in the
Index Terms— Large-scale dense cellular networks, energy- under-utilization of the BSs whilst consuming considerable
spectral efficiency, base station deployment, coordinated multi-
point (CoMP), large-scale users’ behaviors. energy to maintain cellular coverage even under the scenario of
only supposing few users [5]. Therefore, it is vital to conserve
energy by exploiting both the temporal and geographic fluc-
I. I NTRODUCTION tuations of the users’ behaviors, such as the network’s mobile
traffic intensity and the users’ required service rates, in dense
W ITH the spectral efficiency of radio links approaching
the theoretical limit, it is a quite challenge to increase
the overall system capacity in the face of limited spectral
cellular networks.
Considering the fact that cell-edge users face the most
critical quality of service (QoS) guarantee problem due to
Manuscript received September 15, 2016; revised March 12, 2017; accepted the serious inter-cell interference encountered, coordinated
May 3, 2017. Date of publication May 16, 2017; date of current version multi-point (CoMP) transmission is particularly useful since
July 10, 2017. This work was supported in part by the National Natural Sci-
ence Foundation of China under Grant 61372070, in part by the Hong Kong, it is capable of converting interference into a desirable signal
Macao and Taiwan Science and Technology Cooperation Special Project under and therefore improve the resource utilization [9]. However,
Grant 2014DFT10320, in part by the 111 Project under Grant B08038, and in in order to perform CoMP transmissions, flawless information
part by the China Scholarship Council under Grant 201506960042. The work
of L. Hanzo was supported by the Advanced Fellow Grant from the European exchange between the BSs is necessary which requires safe
Research Council. The associate editor coordinating the review of this and high-capacity backhaul links. Therefore, the joint signal
paper and approving it for publication was T. Taleb. (Corresponding author: processing and transmission of CoMP also increases the
Lajos Hanzo.)
G. Zhao and L. Zhao are with the State Key Laboratory of Integrated Service resource costs [5], [9]. Hence, it is critically important to
Networks, Xidian University, Xian 710071, China (e-mail: [email protected]; address how to appropriately support the cell-edge users by
[email protected]). CoMP transmissions according to the specific cellular system
S. Chen is with the School of Electronics and Computer Science, University
of Southampton, Southampton SO17 1BJ, U.K., and also with King Abdulaziz conditions in order to strike the best balance between the
University, Jeddah 21589, Saudi Arabia (e-mail: [email protected]). performance gain achieved and the resource costs increased.
L. Hanzo is with the School of Electronics and Computer Sci-
ence, University of Southampton, Southampton SO17 1BJ, U.K. (e-mail: A. Related Works
[email protected]). A number of studies have been focused on the energy reduc-
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. tion in cellular networks by exploiting both the temporal and
Digital Object Identifier 10.1109/TWC.2017.2703306 spatial traffic variations [10]–[12]. For instance, the authors of
1536-1276 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 ZHAO et al.: JOINT ESE OPTIMIZATION OF CoMP AND BS DEPLOYMENT 4833

[10] solved an energy-saving dynamic BS switching on/off not valid for the EE analysis and optimization of large-scale
problem by analyzing a limited number of BSs located in CoMP-enhanced cellular networks.
a certain area. By contrast, Wu et al. [11] investigated the Nonetheless, a few studies dedicated to CoMP-enhanced
impact of the average delay of services on the total power large-scale randomly deployed cellular networks have also
consumption for a single BS site. Furthermore, by introduc- emerged in the literature [35]–[39]. In particular, the authors
ing an EE metric, the impacts of key network parameters, of [38] integrated a multi-BS CoMP mechanism with a large-
including the number of BSs in a given area, power amplifier scale cellular network and improved the users’ coverage prob-
with delay and mobile traffic demand, are evaluated based on ability by carefully choosing a suitable CoMP cluster. Sakr and
system-level simulation [13]–[15]. Additionally, the work [16] Hossain [37] proposed a cross-tier CoMP scheme by enabling
maximizes the EE by joint power and subcarrier allocation both macro- and small-BS cooperation based on the users’
for a single-cell OFDMA network with multiple relays. An spatial positions to improve the performance both in terms of
energy-efficient carrier aggregation scheme is considered in outage probability and the achievable data rate. By quantifying
[17], while large-scale MIMO based systems are discussed the impact of the mobile-traffic intensity and BS density,
in [18]. Similarly, the designs of [19]–[21] are either for single- Cao et al. [35] analyze the energy savings gained from inter-
cell or fixed networks. However, these existing algorithms are BS cooperation diversity under an average outage constraint.
only based on local system performance metrics, but ignore However, these existing studies all assume a constant power
large-scale mobile-traffic variations, such as the temporal and consumption for each BS and they do not take into account
geographic fluctuations of the users’ behaviors. Therefore, the impact of inter-BS cooperation on the BS’s energy and
these methods cannot be readily applied to large-scale densely spectrum utilization as well as EE performance.
deployed cellular networks. It is widely recognized that beneficially exploiting the users’
The distribution of BSs is typically modeled by a Poisson large-scale behaviors and the inter-BS cooperation diversity
point process (PPP), which has been shown to be a tractable in large-scale randomly deployed cellular networks to jointly
and accurate model for characterizing large-scale randomly optimize CoMP and the BS deployment is challenging, which
deployed networks [22]–[24]. However, these existing con- has not been carried out in the existing literature.
tributions mainly focus on user-centric performance analysis,
such as the coverage probability [23] and the users’ average
throughput [22], [24], but they ignore the network’s EE. In the B. Our Contributions
literature, there exist a few studies addressing the challenging In this paper, we formulate a holistic energy and spectrum
problem of energy efficient BS deployment strategies and efficiency (ESE) evaluation framework for dense large-scale
network EE analysis [25]–[27]. These studies all recognize that CoMP-enhanced homogeneous cellular networks. In conduct-
the optimal BS deployment strategy in a randomly deployed ing the joint analysis and optimization of CoMP transmission
dense network is equivalent to minimizing the BS density sub- and BS deployment with a particular focus on the users’ large-
ject to a coverage probability or outage constraint. However, scale behaviors, which includes their mobile-traffic intensity
the optimization criteria adopted in these studies are based on and the users’ average data rates as well as the spatial and
a constant power consumption for each BS and they do not temporal distributions of the mobile traffic associated with
take into account the important features of the users’ large- the CoMP and No-CoMP modes, respectively, we aim for
scale behaviors, in particular, the network’s mobile-traffic addressing the following two fundamental questions: i) Is it
intensity. By contrast, the recent study in [28] has formulated possible to operate a large-scale network in a ‘best’ CoMP
an optimal mobile-traffic-aware BS deployment strategy to state? ii) How much quantitative performance gain can we
maximize the network’s EE while meeting the users’ outage obtain by the joint optimization of CoMP and BS deployment?
probability requirements, for dense large-scale homogeneous Our contributions are summarized as follows.
cellular networks. Furthermore Ge et al. [29] analyzed both 1) CoMP-Enhanced ESE Modeling: We first introduce a
the SE and EE of dense homogeneous cellular networks by CoMP activation factor to characterize the inter-BS cooper-
considering the effects of the call arrival rate and by proposing ation in a dense large-scale cellular network, which is suffi-
a Markov-chain access model. A similar Gauss-Markov mobile ciently flexible to control the fraction of cell-edge users relying
model was utilized for analyzing small cell backhaul networks on the CoMP mode. Then the downlink (DL) transmit power
in [30]. illuminating the coverage area of a typical BS is modeled
More recently, several investigations [31]–[34] have studied by deriving the average DL transmit power for a typical
the benefits of CoMP for the network EE. Heliot et al. [31] No-CoMP user equipment (UE) and CoMP UE. Instead of
introduced an idealized CoMP-enabled cellular system and unrealistically assuming a constant DL BS power consumption
evaluate its EE, which shows that the potential improvement as in most existing works [25]–[27], [35]–[39], the quantitative
of CoMP in terms of EE is limited by the energy cost of impact of both the users’ large-scale behaviors, as well as the
cooperative processing and backhauling. Under an idealized inter-BS cooperation degree and large-scale BS deployment
circular cell and with only a few BSs, the study [33] compares parameters are statistically characterized. The network’s ESE
the network EE for two CoMP schemes, with the BS’s cover- is defined and derived, together with a comprehensive insight
age area imposed as a constraint. Both careful BS-positioning into how the geographic mobile-traffic intensity, the average
and CoMP are utilized in [34] to jointly improve the spectral users’ rate, the BS density, the CoMP activation factor and
efficiency (SE) and EE. However, these existing results are other cellular parameters influence the network’s ESE.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 4834 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

2) Tractable CoMP-Enhanced ESE Analysis: We derive the the total number of UEs in cell Vi . We now introduce the
optimum CoMP activation factor with the aid of our closed- CoMP activation factor 0 < ρ < 1 to quantify the inter-cell
form ESE expression, given the other system parameters, cooperation degree in a large-scale CoMP-enhanced cellular
the BS density, the geographical mobile-traffic intensity and network. Specifically, let the distance between a typical UE
the average user rate. Our result shows that there always exists located in Vi , i.e., u i ∈ u,i , and its serving BS bi be r1 and
an unique CoMP activation factor maximizing the network’s further denote r2 as the distance from u i to its second-nearest
ESE under a specific inter-BS spectrum balance constraint. BS, where obviously r1 < r2 . According to [36], [37], whether
Similarly, by exploring the mathematical relationship of the UE u i operates either in the No-CoMP or in the CoMP mode
ESE as a function of the BS density, we demonstrate that is decided according to the following rule
there exists only a single BS density, which optimizes the ESE.
u i operates in No-CoMP mode: if r1 ≤ ρr2 ,
The closed-form nature of this ESE expression thus makes our
search for network ESE-based optimal solutions particularly u i operates in CoMP mode: if r1 > ρr2 .
efficient, which enables us to design optimal solutions for Physically, the value of ρ is inversely proportional to the aver-
dense large-scale cellular networks. age percentage of users in a cell relying on the CoMP mode
3) Design Strategies: We first design an optimal cellular- and, therefore, it specifies the degree of inter-cell cooperation.
scenario-aware inter-BS cooperation strategy with aid of the Clearly, the CoMP activation factor must obey 0 < ρ < 1.
above-mentioned efficient ESE analysis tool, which maximizes Otherwise, there will be no users ever relying on the CoMP
the network’s ESE while meeting a specific outage constraint. mode. A further restriction on ρ is imposed by the specific
Based on this network ESE analysis tool, we further jointly consideration of resource assignment, as discussed in the next
optimize the CoMP activation factor and the BS density, again subsection.
under the outage constraint, to propose a mobile-traffic-aware
joint CoMP and BS deployment strategy to find the global
C. CoMP Resource Assignment
ESE maximum value. We demonstrate that our proposed joint
design strategy significantly outperforms the stand-alone inter- The system adopts orthogonal frequency-division multiple
BS cooperation strategy in terms of achievable ESE. access (OFDMA) combined with CoMP-enhanced fractional
frequency reuse. The total system bandwidth of B [Hz] for bi
i ,
is divided into two parts: the interior-subbands, denoted as BIn
II. S YSTEM M ODEL
which can only be occupied by the UEs in u,i , and the outer-
A. Network Layout i
subbands, denoted by BOut , which can only be used by the
A large-scale CoMP-enhanced DL cellular system is mod- CoMP UEs in the adjacent cells. Obviously, B = BIn i
+ BOut
i
.
eled, where a large number of randomly deployed BSs, col- According to Lemma 2 of [36], for a typical UE u i ∈ Vi ,
lected in the set b = {bi }, are distributed according to a the following conditions hold
homogeneous PPP with a density of λb in the Euclidean plane    
Pr r1 ≤ ρr2 = ρ 2 and Pr r1 > ρr2 = 1 − ρ 2 .
R2 . Similarly, the UEs denoted by the set u are spatially
i i
scattered in R2 according to another independent homoge- Therefore, we have the expected values of BOut and BIn for a
neous PPP with a density of λu . All the UEs are classified typical cell Vi as follows:
into two modes, No-CoMP and CoMP modes. A UE in No-  i     i 
E BOut = 1 − ρ 2 B and E BIn = ρ 2 B,
CoMP mode only associates with the geographically nearest
BS, while a CoMP UE will be simultaneously served by the where E[·] denotes the expectation operator. Each cell  i should

nearest BS and the second-nearest BS for the CoMP mode. ensure
 i  an adequate resource balance between E B In and
Since the geographical mobile-traffic intensity is proportional E BOut . Specifically, each cell should assign more resources
to the UEs’ spatial density λu given the average user-rate to serve its users than the resources that it schedules for inter-
requirement, we will use λu to equivalently represent the cell corporation, because the primary responsibility of a cell
mobile-traffic intensity. We may then characterize a network- is for the users within its own coverage area. This naturally
ing scenario by its BS density and mobile-traffic intensity. For leads to 1 − √ρ 2 B < ρ 2 B, or the flexible region of ρ should
modeling practicality, we assume that λu /λb ≤ 100 is ensured satisfy ρ ∈ 2/2, 1 .
for the total number of available resource blocks (RBs) in each The operation of the CoMP-aided cellular network is illus-
BS. We may further assume that all the UEs and BSs only use i
trated in Fig. 1. Specifically, BIn is equally partitioned into Ni
a single antenna to transmit or receive. subbands, each having a bandwidth of BIn i
/Ni . The j -th No-
(WoC) (WoC)
CoMP UE u i, j ∈ u,i or the k-th CoMP UE u i,k(WC) ∈
B. CoMP Mode Selection (WC)
u,i will randomly access one of the subbands with a
The set of BSs b forms a Voronoi tessellation model where specific service rate requirement denoted by Ri, j or Ri,k ,
(WC)
the cell Vi denotes the coverage area of the typical BS bi . The respectively. In particular, CoMP UE u i,k associated with
set of UEs located in Vi is denoted by u,i , which satisfies the subband Bi,k ∈ BIn is simultaneously served by its second-
i
u,i = u,i(WoC) ∪ u,i(WC) and u,i(WoC) ∩ u,i(WC) = ∅, where Sec in a subband, which belongs to the outer-
nearest BS bi,k
u,i(WoC) and u,i(WC) are the sets of No-CoMP and CoMP Sec . In this way, the cell-edge CoMP-UEs are
subbands of bi,k
UEs associated with BS bi , respectively, and |u,i | = Ni is guaranteed to reuse the same frequency for CoMP operation

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 ZHAO et al.: JOINT ESE OPTIMIZATION OF CoMP AND BS DEPLOYMENT 4835

requirement, namely,

Pi,Rxj
Ri, j = Bi, j log2 1 + , (1)
Ii, j

where Pi,Rxj = Pi,Txj L ij h ij in which L ij is the large-scale path-

loss gain between BS bi and u (WoC) i, j , while h ij is the gain of the
(WoC)
fast-fading channel2 linking BS bi to u i, j . The worst-case
interference Ii, j is given by
 
Ii, j = PInsb L ij h ij , (2)
bi  ∈{b \bi }

where PInsb = Pm /Ni is the transmit power of the interference

Fig. 1. Operation of the CoMP-aided cellular network.
from all the other BSs in the subband Bi, j , and Pm is the
maximum BS transmit power in BIn i . Since Rayleigh distrib-

uted fast-fading is assumed, all the fast-fading channel gains

[40], [41]. Thus, the strongest interference in subband Bi,k
follow the independent exponential distribution with a unity
becomes useful information for u i,k(WC) . In the same way, BS mean, i.e., h ij ∼ exp (1). Unless noted otherwise, the effect
i , which is divided equally into a number of outer-
bi uses BOut of the channel noise is ignored, since the interference power
subbands, to serve the CoMP UEs in the adjacent cells. More experienced by a UE is far higher than the noise power in an
specifically, the BS bi can use a subband of BOut i to assist

interference-limited scenario.
an adjacent BS bi  , where i = i , by cooperatively serving a Similarly, a typical CoMP UE u i,k
(WC)
located in Vi is
CoMP UE in Vi  . associated with the interior subband Bi,k of BIn i
as well as
Unlike the existing contributions [36]–[39], which ignore Sec
with the outer subband Bi,k from BS bi,k . Hence, the transmit
the benefits of resource scheduling in a large-scale CoMP- Tx and P Tx of b and b Sec to u (WC) must satisfy
enhanced cellular system, our modeling specifically charac- powers Pi,k,1 i,k,2 i i,k i,k
terizes the explicit relationship between the spectral-domain the associated rate-requirement, namely,
scheduling and the degree of inter-cell cooperation. The  Rx + P Rx
Pi,k,1 i,k,2
CoMP-enhanced resource assignment in our work is consistent Ri,k = Bi,k log2 1 + , (3)
with the CoMP-based completely dynamic frequency planning Ii,k
scheme of [41], where the cells having more CoMP links will
assign more resources for CoMP operation, and vice versa. Rx = P Tx L i1 h i1 and P Rx = P Tx L i2 h i2 in which
where Pi,k,1 i,k,1 k k i,k,2 i,k,2 k k
Note that the overhead of CoMP-based inter-cell interaction L ik1 and h ik1 are the path-loss and fast-fading gains of the
is increased compared to the fixed allocation based scheme of channel linking bi to u (WC)
i,k while L ik2 and h ik2 are the gains
[40], which however limits the capacity of the interface X2 Sec to
of path-loss and fast-fading for the channel linking bi,k
(i.e., the interface between the eNodeBs in LTE) and is not (WC)
adopted in this paper. In our work, we adopt non-coherent joint u i,k . In order to obtain the maximum achievable CoMP
Rx = P Rx , i.e., b and b Sec
diversity [37], [38], we assume Pi,k,1
transmission (NC-JT) in the physical layer for the CoMP-aided i,k,2 i i,k
(WC)
network [36]–[39].1 contribute equally to u i,k in terms of its rate requirement.3
The worst-case interference Ii,k is given by

D. CoMP-Enhanced Power Consumption Modeling Ii,k =   PInsb L i  h i  . (4)

bi  ∈ b \ bi ,bi,k
Sec k k
A QoS-based DL transmit power control mechanism is
adopted in every BS for each subband based on the current The aggregate DL transmit power illuminating the cell area
statistic channel state information (CSI) over the duration of Vi is defined as the summation of the transmit powers allocated
a scheduling interval. Specifically, a typical No-CoMP UE i
in both the interior-subbands BIn and all the outer-subbands
u (WoC)
i, j associated with an interior subband Bi, j is assigned
the right amount of transmit power Pi,Txj , to meet its rate- 2 The instantaneous CSI h i is used here only for the convenience of
j
representing the instantaneous DL transmit power. As it will be shown
1 There exist two approaches to implement physical-layer transmission: in Section III, what we need is the average DL transmit power. Hence,
the scheduling in fact relies on E[h ij ], not h ij .
1) Coordinated scheduling/interference avoidance, where the cooperating BSs 3 There exists a trade off between CoMP diversity and total transmit power.
adopt transmission precoding, such as zero forcing beamforming or maximum
Rx = P Rx , bSec has to transmit to u (WC) at a higher power
To achieve Pi,k,1
ratio transmission, to avoid imposing interference on the users in the adjacent i,k,2 i,k i,k
cells. This first approach requires instantaneous CSI, which may be difficult than bi . To save the total transmit power Pi,k,1Tx + P Tx while maintaining
i,k,2
to acquire but it is generally optimal. 2) NC-JT, where both the serving BS Sec . In fact,
the required rate Ri,k , bi should transmit at a higher power than bi,k
and another BS simultaneously transmit their data to the user, employing a Tx
non-coherent transmission technique. An advantage of this second approach the total transmit power is minimized by setting Pi,k,2 = 0 and putting all
is that only the statistical CSI, not the instantaneous one is required. the necessary power on Pi,k,1 Tx . But this completely loses the CoMP diversity.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 4836 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

TABLE I E. Performance Metrics

L IST OF S YMBOLS
We continue by defining the network-level ESE metric as
Throughput
ηESE = [bps/(W · Hz)], (6)
Power × Bandwidth
where ‘Throughput’ refers to the sum of the users’ average
service rates per unit area, and ‘Power’ is the sum of the
power consumptions per unit area, while ‘Bandwidth’ is the
total system bandwidth. When deriving the ESE, we can adopt
the general power consumption model for each BS specified
by the standards organizations [4]. Next we define the outage
probability4 for a typical No-CoMP UE u (WoC)i, j as
 
i, j = Pr Pi, j > PIn ,
Q out Tx sb
(7)
(WoC)
which is the probability that the service rate of u i, j cannot
be guaranteed under the maximum BS transmit power PInsb
of the current subband. Note that the outage probability of a
CoMP UE will be lower than that of a typical No-CoMP UE,
which is served simultaneously by two BSs.
Our main focus is the ESE evaluation for a large-scale
CoMP-enhanced cellular network and, therefore, the outage
probabilities of No-CoMP UEs are used as constraints. Table I
summarized the symbols used throughout our discussions.

III. A NALYSIS OF J OINT CoMP AND BS D EPLOYMENT

FOR L ARGE -S CALE D ENSE C ELLULAR N ETWORKS
This section characterizes the ESE for CoMP-enhanced
cellular networks as a function of the average DL transmit
power, for both typical No-CoMP and CoMP UEs, as well as
the average aggregate DL transmit power in a typical cell.

A. Average DL Transmit Powers of No-CoMP and

CoMP UEs
We first focus on a typical CoMP UE’s DL transmit power.
Proposition 1: In a large-scale CoMP-enhanced cellular
network having BS density λb , mobile-traffic intensity λu and
CoMP activation factor ρ, the average DL transmit power of
BS bi to a typical CoMP UE u (WC) i,k requiring a service rate
i
Ri,k and associated with the interior subband Bi,k of BIn is
given by
for serving the CoMP UEs in Vi , given by   4Pm 1 − ρ α+2  λb  Ri,k 
Tx
E Pi,k,1 =   2 Bi,k
− 1 , (8)
i
Pcell = Pi,Txj + Tx
Pi,k,1 α 2 − 4 λu
(WoC) (WoC) (WC) (WC)
u i, j ∈u,i u i,k ∈u,i
    where α > 2 is the pathloss exponent.
No-CoMP UEs CoMP UEs: interior-subbands Proof:
 Tx  See Appendix A.
+ (WC) (WC)
Tx
Pi,k,2 , (5) E Pi,k,1 includes a factor of 1−ρ α+2 , which decreases as ρ
u i,k ∈u,i
  increases. Increasing ρ implicitly leads  to an increase of Bi,k ,
which causes a reduction in E Pi,k,1 Tx . Therefore, increasing
CoMP UEs: outer-subbands
the CoMP activation factor decreases the average DL transmit
which depends on many factors but it is predominately influ- power for a typical CoMP UE u (WC)i,k from its associated BS bi .
enced by the cooperation activation degree, the mobile-traffic
load and the average user rate. Our formulation of Pcell i
4 Replacing P Rx = P Tx L i h i in (1) with the maximum possible received
specifically takes into account the impact of the mobile-traffic i, j i, j j j 
sb L i h i , we have the capacity R max = B log 1+ P sb L i h i /I

power PIn j j i, j i, j 2 In j j i, j .
intensity λu . This is in contrast to the existing contributions The outage probability is defined as the probability that the capacity cannot
 max
[25]–[27] which ignore the effects of mobile-traffic by assum- meet the requirement of the service rate Ri, j , namely, Q outi, j = Pr Ri, j <

ing a constant transmit power. Ri, j . Clearly and obviously, this is equivalent to (7).

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 ZHAO et al.: JOINT ESE OPTIMIZATION OF CoMP AND BS DEPLOYMENT 4837

 Tx 
On the other hand, although E Pi,k,1 contains a factor of λ1u , B. Average Aggregate DL Transmit Power in a Typical Cell
it increases as the mobile-traffic intensity λu increases. This If we denote the average DL transmit power transmitted to
is because 2 Ri,k /Bi,k increases exponentially with λu . a UE in all its associated subbands with the UE location of Z
Proposition 2: In a large-scale CoMP-enhanced cellular in R2 , then the average aggregate DL power transmitted to a
network having BS density λb , mobile-traffic intensity λu and typical cell Vi having the coverage area of Ai can be expressed
CoMP activation factor ρ, the average DL transmit power of as
Sec to a typical CoMP UE u (WC) requiring a service rate
BS bi,k  
i,k  i 
Ri,k and associated with the outer subband Bi,k from bi,k Sec is E Pcell = PZ · I {Z ∈ Vi } dZd Ai , (13)
given by Ai Z∈Vi

  2Pm 1 − ρ 2  λb  Ri,k  where I {·} is the indicator function, which is equal to 1, when
Tx
E Pi,k,2 = 2 Bi,k − 1 . (9) the condition inside the bracket is satisfied and 0 otherwise.
(α − 2)λu We consider a pair of rate allocation conditions between
Proof: See Appendix B.  Tx  No-CoMP and CoMP UEs in a typical cell Vi as follows.
From Propositions
 1 and 2, it can be seen that both E Pi,k,2
Tx
1) Dual-rate (DR) mode, where all the No-CoMP UEs have an
and E Pi,k,1 have the same trend with respect to the CoMP
 Tx  identical rate of R, while all the CoMP UEs have an identical
factor ρ. But the impact of ρ on E Pi,k,1 is larger than on
 Tx  rate of 2R. 2) Single-rate (SR) mode, where all the No-CoMP
E Pi,k,2 . Moreover, the following corollary may be inferred. UEs and CoMP UEs have an identical rate of R.
Corollary 1: In a large-scale CoMP-enhanced cellular net- Proposition 4: Given the BS density λb , mobile-traffic
work having BS density λb , mobile-traffic intensity λu and intensity λu and CoMP activation  i factor ρ, the averaged
CoMP activation factor ρ, the average total DL transmit power aggregate DL transmit power E Pcell in the SR mode with a
for a typical CoMP UE u (WC)i,k , requiring a service rate Ri,k cell area Vi is given by
i as well
and associated with the interior subband Bi,k of BIn   C (ρ) (K λb ) K
Sec
as the outer subband Bi,k from BS bi,k , is given by SR
E Pcell =     K − C (ρ), (14)
R
      K λb − 2 ρ 2 B − 1 λu
Tx Tx Tx
E Pi,k = E Pi,k,1 + E Pi,k,2
   Ri,k  where we have
4Pm λb 1 − ρ α+2 1 − ρ2    
= + 2 Bi,k
−1 . 1 − ρ α+2 1 − ρ 2 4Pm 1 − ρ 2 8Pm ρ α+4
(α − 2)λu α+2 2 C (ρ) = + +  2  ,
(10) α+2 2 (α − 2) α −4
We continue by analyzing a typical No-CoMP UE’s DL (15)
transmit power. and K = 3.75.
Proposition 3: In a large-scale CoMP-enhanced cellular
Proof: See Appendix D.
network having BS density λb , mobile-traffic intensity λu For the DR case, we have Proposition 5. The proof of
and CoMP-factor ρ, the average DL transmit power of BS Proposition 5 is similar to that for Proposition 4.
bi transmitted to a typical No-CoMP UE u (WoC) i, j requiring a Proposition 5: Given the BS density λb , mobile-traffic
service rate Ri, j and associated with the interior subband Bi, j intensity λu and CoMP-factor ρ, the averaged aggregate DL
i
of BIn is given by  i 
transmit power E Pcell in the DR mode with a cell area Vi
 R is given by
  8P ρ α+2 λ i, j
 
E Pi,Txj =  2
m

b
2 Bi, j − 1 . (11)   4Pm 1 − ρ α+2 1 − ρ2  
α − 4 λu E Pcell =
DR
+ 1 − ρ2
(α − 2) α+2 2
Proof: See Appendix C. 
From  (11), it can be seen that the relationship between (K λb ) K
×    K − 1
E Pi,Txj and the CoMP factor ρ is complicated, because ρ α+2 K λb − 22R/ρ B − 1 λu
2

increases, while 2 Ri, j /Bi, j decreases upon increasing ρ.

ρ α+4
By considering both the No-CoMP and CoMP cases, the fol- + 2 
lowing corollary is inferred. α −4

Corollary 2: The average DL power transmitted to a typical 8Pm (K λb ) K
UE u i in cell Vi requesting a rate Ri, j is given by ×     K − 8Pm . (16)
K λb − 2 R/ρ B − 1 λu
2

     
Tx
E PUE = Pr(r1 > ρ 2r2 )E Pi,k
Tx
+ Pr(r1 < ρ 2r2 )E Pi,Txj C. Network ESE Evaluation
   
4Pm 1 − ρ 2 λb 1 − ρ α+2 1 − ρ2 The powers in the network-level ESE metric (6) for the SR
= +
(α − 2)λu α+2 2 and DR cases can be expressed respectively by
 Ri,k   R  
8Pm ρ α+4 λb i, j
× 2 Bi,k − 1 + 2 2 Bi, j − 1 . PowerSR = βλb E PcellSR
+ λb POM , (17)
(α − 4)λu  
(12) PowerDR = βλb E PcellDR
+ λb POM , (18)

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 4838 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

where β is the power amplifier efficiency, and POM is the BS’s A. Cellular-Scenario-Aware Cooperation Activation Degree
non-transmission related power consumption, including base- Optimization Under Outage Performance Constraint
band processing, battery backup and cooling. The throughputs We formulate a cellular-scenario-aware cooperation activa-
in the DR and SR modes are given respectively by tion optimization problem while satisfying all users’ QoS con-
  straints. Specifically, we maximize the network ESE metric by
ThroughputDR = ρ 2 Rλu + 2R 1 − ρ 2 λu , (19) choosing a suitable CoMP activation factor, while guarantee-
ThroughputSR = Rλu . (20) ing the users’ outage performance. The following proposition
indicates the existence of the unique optimal CoMP degree
Based on the definition of (6) and on the above-mentioned solution for the associated unconstrained optimization.
analytical results, the following proposition is inferential. Proposition 7: Given λb and λu , there exists an unique
Proposition 6: The ESE of a large-scale CoMP-enhanced optimal cellular-scenario-aware CoMP activation factor ρ  ,
cellular network, measured in bit/Hz/Joule, for the BS density which maximizes the network’s ESE metric ηESE SR of (21), and

λb , mobile-traffic intensity λu and CoMP factor ρ, can be it can be obtained numerically by solving (24) and (25), as
expressed in the closed-form expressions of (21) for the SR shown at the bottom of this page.
mode and (22) and (23) for the DR mode, respectively, at the Proof: See Appendix E.
bottom of this page. We continue by analyzing the outage probability constraint.
Since the outage probability for a CoMP UE is much lower
than that of a No-CoMP UE, we only focus our attention on
IV. O PTIMIZATION OF J OINT CoMP AND the outage performance of a typical No-CoMP UE.
BS D EPLOYMENT Proposition 8: Given λb , λu and ρ, the outage probability
for a typical No-CoMP UE u (WoC) i, j requesting rate Ri, j in
In this section, we apply the analytical results from the pre-
subband Bi, j and conditioned on r1 is given by
vious section to formulate two energy-spectral efficient design  
strategies for large-scale CoMP-enhanced cellular networks. Q out 2
i, j [r1 ] = 1 − exp −πr1 λb τ , (26)
We will only focus our attention on the SR case, since the  +∞
DR case is similar and it is therefore omitted. where τ = T 2/α T −2/α 1+u1α/2 du and T = 2 Ri, j /Bi, j − 1.

  Rλu
ηESE
SR
ρ, λb , λu , R = ⎛ ⎞ (21)
 K
⎜ B K λb ⎟
βλb C (ρ) ⎜
⎝ 
R K − B⎟
⎠ + λb B POM
K λb − 2 ρ2 B −1 λu
 
  ρ 2 Rλu + 2 R 1 − ρ 2 λu
ηESE ρ, λb , λu , R =
DR   (22)
D ρ, λb , λu , R
with
⎛ ⎞
 ⎜   K
  4 Pm 1 − ρ α+2 1 − ρ2 ⎜ 1 − ρ 2 K λb  ⎟
2 ⎟
D ρ, λb , λu , R = βλb B + ⎜  2 R  K − 1 − ρ ⎟
(α − 2) α+2 2 ⎝ ⎠
K λb − 2 ρ 2 B − 1 λu
⎛ ⎞
 K
ρ α+4 ⎜⎜ 8 Pm K λb ⎟
⎟
+ βλb B  2  ⎜  R   K − 8 Pm ⎟ + λb B POM (23)
α −4 ⎝ ⎠
K λb − 2 ρ 2 B − 1 λu

  

R

  2 RK K +1 λbK λu 2 (ρ  )2 B C ρ  ln 2
ρ =     −K   R   K +1 (24)

 C ρ   B(K λ ) K K λ − 2 (ρ R)2 B − 1 λ
3

b b u −B K λb − 2 (ρ  )2 B − 1 λu

with

   
1 − (α + 2)ρ α+1 1 − 2ρ 4Pm (1 − ρ 2 ) 1 − ρ α+2 1 − ρ2 4Pm (1 − 2ρ) 8Pm (α + 4)ρ α+3
C (ρ) = + + + + (25)
α+2 2 (α − 2) α+2 2 (α − 2) (α 2 − 4)

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 ZHAO et al.: JOINT ESE OPTIMIZATION OF CoMP AND BS DEPLOYMENT 4839

 2

βC (ρ) 2 R/ρ B − 1 λu
λb =       K +1    K +1
(29)
2 B −1 2 2
2 R/ρ λu 2 R/ρ B −1 λu 2 R/ρ B −1 λu
βC (ρ) 1 − K λb
− βC (ρ) 1 − K λb
+ B PO M 1 − K λb

Proof: See Appendix F. B. Mobile-Traffic-Aware Joint CoMP and BS Deployment

Utilizing E [Ni ] = λλub and the average cell radius √πλ
1
, Optimization With Outage Constraint
b
the following corollary is obtained. When BSs are controllable and they can be switched on/off,
Corollary 3: Given λb , λu and ρ, the average outage proba- we can jointly optimize the CoMP activation factor ρ and BS
bility for a typical No-CoMP UE requesting rate R in subband density λb to maximize the network’s ESE metric for the given
ρ 2 Bλb
λu is given by mobile-traffic intensity and average user rate, while satisfying
all the users’ QoS constraints. The following proposition
 
Q out ρ; λb , λu = 1 − exp (−τ ), (27) focuses on the existence and uniqueness of the unconstrained
optimal BS density solution that maximizes the network’s ESE
!  +∞ metric, for the given ρ and λu .
where T = 2λu R λb ρ B − 1 in τ = T 2/α T −2/α 1+u1α/2 du.
2
Proposition 9: Given ρ and λu , there exists a unique opti-
We are now ready to formulate the outage-constrained mal BS-density λb that maximizes the network’s ESE metric
energy-spectral efficient cellular-scenario-aware CoMP degree ηESE
SR and it can be obtained numerically by solving (29),
optimization problem (ESE-SACoMP) which is: max ηESE SR
shown at the top of this page.
ρ
subject to the constraint Q out (ρ; λb , λu ) ≤ εout , where 0 < Proof: See Appendix G.
εout < 1 is the tolerable outage probability. More specifically, Combining Propositions 7 and 9 leads to the following
the ESE-SACoMP is formulated as Proposition.
! Proposition 10: For a large-scale CoMP-enhanced cellular
λu R Bλb network with mobile-traffic intensity λu , there exists an unique
max  K ,
ρ globally optimal
K λb βC (ρ)  joint CoMP activation factor and BS density
 
R K − βC (ρ) + POM solution ρ  , λb that maximizes the network’s ESE metric
K λb − 2 ρ 2 B −1 λu ηESE
SR , which can be obtained by jointly solving the simultane-

ous equations of (24) and (29) with respect to ρ and λb .

s.t. 1 − exp(τ ) ≤ εout ,
√ The following proposition can be proved in a similar way.
2/2 ≤ ρ ≤ 1. (28) Proposition 11: For a large-scale CoMP-enhanced cellular
network with BS density λb , there exists an unique globally
We continue to derive the optimal CoMP activation joint optimalCoMPactivation factor and mobile-traffic inten-
factor solution of the ESE-SACoMP, denoted as ρout  .
sity solution ρ  , λu that maximizes the network’s ESE metric
Note that Q out (ρ; λb , λu ) is a monotonically increasing ηESE
SR .
function of τ and τ increases as ρ decreases, which We are ready to introduce the outage-constrained energy-
implies that Q out (ρ; λb , λu ) is a monotonically decreasing spectral efficient mobile-traffic-aware joint CoMP and BS
function of ρ. Therefore,
 the outage-constrained feasible  density optimization (ESE-JCoMPBD). λu, the solution
√  Given
region of ρ is ρ ∈ max Q −1 out,1 (ε out ; λb , λu ), 2/2 , 1 , of the ESE-JCoMPBD, denoted as ρout , λb,out , is obtained
where Q −1 out,1 (·; λb , λu ) denotes the inverse function of
by solving the following optimization problem:
Q out (ρ; λb , λu ) with respect to the first argument ρ, and   λu R
√ 
ρout , λb,out = arg max   SR  , (30)
max Q −1 out,1 (εout ; λb , λu ), 2/2 < 1 always holds, other- cell + POM
ρ,λb Bλb E P
wise the ESE-SACoMP optimization (28) has no solution.  √ 
From the proof ρ ∈ max 2/2, Q −1
out,1 (εout ; λb , λu ) , 1 ,
√ of Proposition
 7, we know that in the
region of ρ ∈ 
2/2, ρ , ηESE is a monotonically increasing
SR (31)
  SR  
function of ρ, while in the region of ρ ∈ ρ  , 1 , ηESE is a λb ∈ Q −1
out,2 (εout ; ρ, λu ), +∞ , (32)
monotonically decreasing function of ρ. Therefore, the outage-
constrained optimal CoMP activation factor solution ρout  of
where Q −1 out,2 (·; ρ, λu ) denotes the inverse function of
the ESE-SACoMP can be obtained as shown in the following Q out (ρ; λb , λu ) with respect to the second argument λb , while
theorem, and the proof is straightforward.
   (31) and (32) define the outage-constrained flexible regions
Theorem 1: Set Q −1 
out,1 εout ; λb , λu = ρ . If εout < for ρ and λb , respectively. Note that (32) reflects the fact that
 −1 −1 
εout , i.e., Q out,1 (εout ; λb , λu ) > Q out,1 εout ; λb , λu = ρ  , Q out (ρ; λb , λu ) is
 a monotonically
 decreasing function of λb .
the outage-constrained optimal CoMP activation factor is By setting Q out ρ  ; λb , λu = εout  , the outage-constrained

given by ρout  = Q −1 
out,1 (εout ;λb , λu ). If  εout > εout ,
joint optimal CoMP activation factor and BS density  solution
−1 −1   of the ESE-JCoMPBD, denoted by ρout  , λ
i.e., Q out,1 (εout ; λb , λu ) < Q out,1 εout ; λb , λu = ρ , then the b,out , is readily
solution of the ESE-SACoMP (28) is given by ρout  = ρ. obtained as follows.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 4840 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

TABLE II
PARAMETERS OF S IMULATED N ETWORK

 
Case 1: When εout > εout  , ρ  , λ
out b,out is located in the
Fig. 2. Networks ESE ηESE as the function of CoMP factor ρ and mobile-
outage-constrained flexible region specified by (31) and (32). traffic intensity λu , given bandwidth resource B = 20 MHz, average user rate
Therefore, we have the following  theorem.   R = 0.2 Mbits/s and BS density λb = 5 BS/km2 .
  , λ  

Theorem 2: If εout > εout , ρout b,out = ρ , λb .
Case 2: When εout ≤ εout  , the unconstrained joint where ARCbase and ARCopt denote the ARCs required by the
optimal baseline design and our optimized design, respectively. Unless
    CoMP activation factor and BS density solution otherwise stated, we set α = 4. We will only consider the SR
ρ , λb is not located within the outage-constrained flex-
ible region. Since outage-constrained joint optimal CoMP
SR = η
mode and, therefore, we will use the notation ηESE ESE .
activation factor and BS The results for the DR mode are similar and thus omitted here.
 density solutions
 must satisfy the
outage constraint Q out ρout ; λb,out , λu = εout , by using λb =
Q −1
out,2 (εout ; ρ, λu ), we can transform the ESE-JCoMPBD into A. Key System Performance Indicators
the following single-variable optimization problem
Fig. 2 portrays the network’s ESE ηESE (ρ; λb , λu ) as the
λu R/B Q −1out,2 (εout ; ρ, λu ) function of the CoMP activation factor ρ and mobile-traffic
√ max    , intensity λu , given λb = 5 BS/km2 , B = 20 MHz and R =
−1
2/2≤ρ≤1 E Pcell ρ; Q out,2 (εout ; ρ, λu ), λu + POM
SR
0.2 Mbits/s. It can be seen from Fig. 2 that given λu , there
(33) exists an unique optimal
 point ρ  . Moreover, in the region
  √ 
SR
where we have explicitly expressed E Pcell as the function of of ρ ∈ 2/2, ρ , ηESE (ρ; λb , λu ) is a monotonically
 
ρ, λb and λu . The one-variable optimization (33) can be solved increasing function of ρ, while in the region of ρ ∈ ρ  , 1 ,
efficiently, using for example the binary search algorithm [42]. the ESE is a monotonically decreasing function of ρ. Similarly,
We have the following obvious theorem, which specifies the given ρ, there exists a unique optimal λu that maximizes the
outage-constrained joint optimal CoMP  activation factor and ESE. These observations confirm our analysis presented in
the BS density solution ρout  , λ Section III. It can also be seen from Fig. 2 that the optimal
b,out .

Theorem 3: If εout ≤ εout , the solution of the optimization λu is an increasing function of ρ. Furthermore, the optimal ρ 
problem (33) yields ρout  , while λ  increases as λu increases, which implies that CoMP plays a
b,out is given by λb,out =
−1  
 more important role in improving the network ESE, when the
Q out,2 εout ; ρout , λu .
mobile-traffic intensity is high. Intuitively, increasing λu may
result in an increasing demand for invoking CoMP by the UEs
V. N UMERICAL AND S IMULATION R ESULTS
to save transmit power, particularly for cell-edge UEs.
The parameters of our simulated CoMP-enhanced cellular In Fig. 3, we depict the network’s ESE ηESE (ρ; λb , λu ) as
network are listed in Table II. We will use both numerical the function of ρ and λb , given λu = 200 user/km2 , B =
and simulation results to characterize the ESE metric as the 20 MHz and R = 0.2 Mbits/s. It can be seen from Fig. 3 that
key system performance indicator and to verify the accu- given λb , there exists an unique optimal ρ  that maximizes
racy of our analysis. Furthermore, we will compare the area the network’s ESE, as proved in Proposition 7, while given
resource consumptions (ARCs) required by our proposed ESE- ρ, there exists a unique optimal λb that maximizes the ESE,
SACoMP and ESE-JCoMPBD to that of the practical baseline as proved in Proposition 9. More significantly, there exists a
cellular designs [43]–[45], where ARC is defined in terms unique globally optimal joint solution (ρ  , λb ) that maximizes
of power × bandwidth (i.e., the denominator of the network the ESE metric, as indicated in Proposition 10. Furthermore,
ESE metric) divided by the network’s area, and it is measured λb is a monotonically decreasing function of ρ, and ρ  is also
in W · Hz/m2 . In particular, we will use the resource saving a monotonically decreasing function of λb . It is worth noting
ratio (RSR) to quantify the gain of our optimized design over that a much higher ESE can be obtained, when λb is relatively
the baseline design of [43]–[45], which is defined as small, such as λb = 2 BS/km2 , thus indicates that CoMP is
ARCbase − ARCopt a preferred means of improving the ESE in a sparse cellular
RSR = , (34) system. Moreover, for the case of full cooperation we have
ARCbase

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 ZHAO et al.: JOINT ESE OPTIMIZATION OF CoMP AND BS DEPLOYMENT 4841

Fig. 5. Theoretical and simulated network ESE ηESE as the function of

Fig. 3. Networks ESE ηESE as the function of CoMP factor ρ and BS density
CoMP factor ρ given R = 0.2 Mbits/s, λb = 5 BS/km2 , λu = 400 user/km2
λb , given bandwidth resource B = 20 MHz, average user rate R = 0.2 Mbits/s
with different bandwidth resources B.
and mobile-traffic intensity λu = 200 user/km2 .

Fig. 4. Theoretical and simulated network ESE ηESE as the function of Fig. 6. Theoretical average DL transmit power for typical cell as the function
CoMP factor ρ given B = 20 MHz, λb = 5 BS/km2 , λu = 400 user/km2 with of CoMP factor ρ, given R = 0.1 Mbits/s with different λb and λu .
different average user rates R.

√ the system bandwidth is reduced. Moreover, a larger system

ρ = 2/2 ≈ 0.7. Then the ESE performance becomes lower bandwidth reduces the CoMP cost and thus increases the
than that of ρ = 1 since CoMP consumes extra resources. CoMP activation probability, i.e., the maximum ESE shifts
Fig. 4 depicts the network’s ESE ηESE (ρ; λb , λu ) as the to the left corresponding to a reduced ρ. The accuracy of our
function of ρ, given B = 20 MHz, λb = 5 BS/km2 and theoretical derivation is also verified in Fig. 5 with the aid of
λu = 400 user/km2 with different average user rates R. The our simulation results.
accuracy of our theoretical derivation is verified in Fig. 4, since Additionally,
 the average aggregate DL transmit power
SR for a typical cell is plotted in Fig. 6 as the function
the theoretical values agree well with the simulation results. E Pcell
Observe from Fig. 4 that the higher R, the fewer cell-edge of ρ, given R = 0.1 Mbits/s with different λb and λu .
UEs are involved in the CoMP mode, i.e., the maximum ESE From Fig. 6, it can be seen  SR  given λu , increasing λb
that
shifts to the right corresponding to an increasing ρ. This is leads to the reduction of E Pcell since the coverage area is
because transmitting the signals at a higher data rate from reduced,
 SR  while given λb , increasing λu leads to the increase of
the secondary BS may consume excessive power and this E Pcell . Furthermore with the increase  of CoMP
 cooperation,
discourages the CoMP activation. Interestingly, we note that in i.e., when ρ is decreased from 1, E Pcell SR firstly decreases,

the case of full CoMP cooperation, i.e., ρ ≈ 0.7, the highest since a fraction of UEs switches to the CoMP mode which
CoMP-based ESE performance is attained at the moderate user consumes less transmit power due to benefiting from a reduced
rate of R = 0.2 Mbits/s in Fig. 4. interference. But when the probability of CoMP coopera-
Fig. 5 characterizes ηESE (ρ; λb , λu ) as the function of ρ, tion continues to increase towards full cooperation, i.e, ρ
given λb = 5 BS/km2 , λu = 400 user/km2 and R = 0.2 Mbits/s approaches 0.7, the increased transmit power of the secondary
for different bandwidth resources B. It can be seen that BSs dominates the total power consumption, leading to the
better ESE performance can be obtained by CoMP, when increase of E PcellSR .

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 4842 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

Fig. 7. Resource saving ratios achieved by the proposed ESE- Fig. 8. Resource saving ratios achieved by the proposed ESE-
SACoMP and ESE-JCoMPBD as the functions of mobile-traffic intensity λu , SACoMP and ESE-JCoMPBD as the functions of mobile-traffic intensity λu ,
given B = 20 MHz and three different values of R. given B = 20 MHz, R = 0.3 Mbits/s and three different values of α.

B. Evaluation of Resource Saving Ratio

We now evaluate the RSR for our proposed design strate-
gies, namely for ESE-SACoMP and ESE-JCoMPBD, over the
baseline design of [43]–[45] with a typical cellular scenario
of λb = 4 BS/km2 . More specifically, for our proposed ESE-
SACoMP design that optimizes the CoMP degree ρ, all the
network parameters are identical to those used for the baseline
design, including the BS density of λb = 4 BS/km2 . The
baseline design of [43]–[45] is not CoMP-enhanced. For our
proposed ESE-JCoMPBD design that jointly optimizes the
CoMP activation degree ρ and BS density λb , except for
λb which is optimized, all the other network parameters are
identical to those used for the baseline design. Since the
ARCs required by our optimized designs are smaller than that
of the baseline design, the values of RSR are all positive,
which indicates that CoMP optimization as well as joint CoMP
and BS deployment optimization are highly effective means
of conserving system resources. Moreover, the results will Fig. 9. Resource saving ratios achieved by the proposed ESE-SACoMP
and ESE-JCoMPBD as the functions of mobile-traffic intensity λu , given
confirm that the ESE-JCoMPBD significantly outperforms the R = 0.3 Mbits/s and three different values of B.
ESE-SACoMP in terms of conserving system resources.
In Fig. 7, the RSR is illustrated as a function of mobile-
traffic intensity λu , given B = 20 MHz and three different
values of R. As expected, the RSR achieved by the ESE- and three different values of the pathloss exponent α. Again,
JCoMPBD is significantly higher than that of the ESE- the joint CoMP and BS deployment optimization significantly
SACoMP. In particular, for a relatively high rate of R = enhances the achievable RSR over the CoMP optimization
0.35 Mbits/s, when the mobile-traffic intensity is relatively alone. In particular, with a small path-loss α = 2.5 and a high
high, i.e., near λu = 5 × 10−4 users/m2 , the RSR achieved by mobile-traffic intensity of λu = 5 × 10−4 users/m2 , the ESE-
the ESE-JCoMPBD is significantly higher than that attained JCoMPBD considerably outperforms the ESE-SACoMP, while
by the ESE-SACoMP. On the other hand, for a relatively low for a large α = 3.5 and a low mobile-traffic intensity of
rate of R = 0.2 Mbits/s, when the mobile-traffic intensity is λu = 2×10−4 users/m2 , the ESE-JCoMPBD also outperforms
relatively low, near λu = 2×10−4 users/m2 , the RSR achieved the ESE-SACoMP considerably.
by the ESE-JCoMPBD is also considerably higher than that Fig. 9 depicts the RSR achieved by the ESE-JCoMPBD over
attained by the ESE-SACoMP. These observations indicate that the ESE-SACoMP as the functions of network mobile-traffic
in addition to CoMP, the BS deployment is extremely effective intensity λu , given R = 0.3 Mb/s and three different values of
in terms of saving system resources. the system bandwidth B. Again, the results of Fig. 9 confirm
Fig. 8 compares the RSRs achieved by the ESE-JCoMPBD that joint CoMP and BS deployment optimization is extremely
and ESE-SACoMP as the functions of λu , given R = 0.3 Mb/s effective in terms of conserving system resources.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 ZHAO et al.: JOINT ESE OPTIMIZATION OF CoMP AND BS DEPLOYMENT 4843

VI. C ONCLUSIONS where

 
In this paper, the network energy-spectral efficiency of    2
N(s) = −πr22 1 − exp − shr2−α + π(sh)2/α 1− ,0
large-scale CoMP-enhanced cellular networks has been mod- α
eled, which provides an accurate network ESE expression as  
2
the function of the key cellular system parameters, including − π(sh)2/α 1 − , shr2−α , (38)
α
the CoMP activation factor, BS-density, mobile-traffic inten-
sity, system bandwidth and average data rate requirement. h is an exponentially distributed random variable with a unity
We have carried out a joint analysis of CoMP and BS mean, and (a, z) is the upper incomplete Gamma function
 ∞
deployment , which specifically considers the impact of the
network mobile-traffic intensity and other key cellular system (a, z) = exp(−t)t a−1 dt. (39)
z
parameters on the network ESE. Most significantly, we have
designed two optimal strategies for maximizing the network’s Substituting the result of M Ii,k (s) into (35), we obtain
   Ri,k 
ESE metric. The first design is based on cellular-scenario- "
Tx " Pm λb r22−α πλb r1α
aware CoMP optimization under an outage constraint, while E Pi,k,1 r1 , r2 = 2 Bi,k − 1 . (40)
(α − 2)λu
the second design performs mobile-traffic-aware joint COMP
and BS deployment optimization under an outage constraint. Furthermore, since the joint distribution of r1 and r2 is fr1 ,r2 =
Our numerical and simulations results have verified the accu- (2πλb )2 r1r2 exp −πλb r22 , we have
racy of our analysis and demonstrated that system resources  
Tx
can be conserved by our proposed two design strategies. The E Pi,k,1
 +∞  r1 /ρ  
joint CoMP and BS deployment optimization has been shown
Tx "
"
to be particularly effective in conserving system resources. Our = E Pi,k,1 r1 , r2 fr1 ,r2 dr2 dr1
0 r
study has therefore offered valuable insights and guidelines to  1  r2
X k,1 +∞ 3−α  
mobile operators for exploiting the CoMP and BS deployment = r2 (2πλb )3 exp − πλb r22 dr2 r1α+1 dr1
techniques in future dense large-scale cellular networks. 2 0 ρr2
   +∞
X k,1 1 − ρ 2  
A PPENDIX = r22 (2πλb )3 r25 exp − πλb dr2 , (41)
2(α + 2) 0
A. Proof of Proposition 1  
where X k,1 = Pm λb 2 Ri,k /Bi,k − 1 /(α − 2)λu . Completing the
Proof: The average DL power transmitted from BS bi to
(WC) last integration in (41) leads to (8).
a typical CoMP UE u i,k requiring rate Ri,k and associated
i
with the interior subband Bi,k of BIn , conditioned on r1 , r2
and interferer set b \(bi , bi,k ), is given by
Sec B. Proof of Proposition 2
 "   Proof: The average DL power transmitted from BS bi,k Sec
Tx "
E Pi,k,1 r1 , r2 , b \ bi , bi,k
Sec
(WC)
to a typical CoMP UE u i,k requiring rate Ri,k and associated
#  $ Sec , conditioned on r , r
Ri,k " with the outer subband Bi,k of BS bi,k 1 2
I i,k "
= E 2 Bi,k − 1 "r 1 , r 2 and interferer set  b \ (b i , b Sec ), is expressed as
i,k
2 h ik1 L ik1   
"
#   " $ Tx "
E Pi,k,2 r1 , r2 , b \ bi , bi,k Sec
∂ Ri,k
ln M Ii,k (s) " " "
" #   " $
= E 2 Bi,k − 1 " "r1 , r2 ,
∂s 2r1 −α " ∂ Ri,k
ln M Ii,k (s) "" "
"
s=0 = E 2 Bi,k
−1 " "r1 , r2 . (42)
(35) ∂s 2h ik2 L ik2 "s=0
where α > 2 is the pathloss exponent and M Ii,k (s) denotes Similar to the proof of Proposition 1, we have
the moment generating function (MGF) of Ii,k . The first  "  πλ r 2 P λ  Ri,k 
Tx " b 2 m b
equality is obtained by applying the rate formula Ri,k of (3) E Pi,k,2 r1 , r2 = 2 Bi,k
−1 , (43)
Rx = P Rx , while the second equality
in conjunction with Pi,k,1 (α − 2)λu
i,k,2
is based on the well-known property of  MGF
 as well as on and therefore
the pathloss formula L ik1 = r1−α and E h ik1 = 1.  
Tx
E Pi,k,2
The MGF of the interference I˜i,k in the annulus with the
 +∞  r1 /ρ  
radius ranging from r2 to rbn with rbn > r2 is given by Tx "
"
   = E Pi,k,2 r1 , r2 fr1 ,r2 dr2 dr1
M I˜i,k (s) = E exp −s I˜i,k 0
 1
r
 r2
%  rbn    & X k,2 +∞ 2  3  
sb −α = r2 2πλb r2 exp − πλb r2 dr2 2
r1 dr1
= exp 2πλb 1 − exp −s PIn x x dx . 2 0 ρr2
r2    +∞
X k,2 1 − ρ 2  3  
(36) = r25 2πλb exp − πλb r22 dr2 , (44)
4 0
Based on [46], we have  R /B !
where X k,2 = Pm λb 2 i,k i,k − 1 (α − 2)λu . Carrying out
M Ii,k (s) = lim M I˜i,k (s) = exp (−λb E [N(s)]), (37) the last integration in (44) leads to (9).
rbn →∞

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 4844 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

 +∞
C. Proof of Proposition 3 (x) = 0 t x−1 exp (−t)dt. By substituting this PDF into
Proof: The average DL power transmitted from BS bi (48), the average DL aggregate transmit power is given by
(WoC)  
to a typical No-CoMP UE u i, j requiring rate Ri, j and SR
E Pcell
in subband Bi, j , conditioned on r1 , r2 and the interferer set  
b \bi , is given by  +∞ K K λ K A K −1 E P Ai
b i SR
  =   d Ai = −C (ρ)
E Pi,Txj |r1 , r2 , b \bi 0 exp K λb Ai (K )
#  R " $  +∞
K K C (ρ)λbK AiK −1
∂ i, j ln M Ii, j (s) "" " +   R    d Ai .
= E 2 Bi, j
−1 " "r1 , r2 . (45)
∂s h ij L ij " 0
exp K λb − 2 ρ 2 B − 1 λu Ai (K )
s=0

Similar to the proofs of Propositions 1 and 2, we have (49)

 R
  2πλ r α r 2−α P λ i, j Carrying out the last integration in (49) leads to (14).
b m b
E Pi,Txj |r1 , r2 = 1 2
2 Bi, j − 1 , (46)
(α − 2)λu
E. Proof of Proposition 7
and therefore
  Proof: The network’s ESE in the SR case is defined by
E Pi,Txj λu R
 +∞  +∞  ηESE
SR
=   . (50)
 Bβλb E Pcell +
SR Bλb POM
= E Pi,Txj |r1 , r2 fr1 ,r2 dr2 dr1
0 r1 /ρ SR with respect to ρ is given by
The derivative of ηESE
 +∞  ρr2
 
= Xj Tx 3−α
r2 (2πλb ) exp − πλb r2 dr2
3 2
r1α+1 dr1 F (ρ)
0 0  

X Tx ρ α+2  +∞   ∂ E Pcell
SR
j   λu R β
= r22 (2πλb )3 r22 exp − πλb r22 r2 dr2 , (47) ∂ηESE
SR
∂ρ
α+2 0 =−   SR  2 . (51)
 ∂ρ λb B β E Pcell + POM
where = Pm λb
X Tx
j 2 Ri, j /Bi, j
− 1 /(α − 2)λu . Carrying out
the last integration in (47) leads to (11). ∂ηSR
Hence the sign of ∂ρ ESE
is equal to the sign of −F(ρ) in (51).
Specifically, F(ρ) is given by
D. Proof of Proposition 4 ⎛ ⎞

Proof: For BS bi having the cell coverage area Ai and ∂C (ρ) ⎜

⎜ (K λb ) K ⎟
⎟
serving Ni UEs, the probability mass function (PMF) of Ni is F(ρ) = β ⎜  R   K − 1⎟
∂ρ ⎝ ⎠
given by FAi (Ni ) = (λu Ai ) Ni exp (−λu Ai ) /Ni ! with Ni ≥ 0 K λb − 2 ρ 2 B − 1 λu
[46]. By applying Corollary 1 and Proposition 3, the averaged ⎛ ⎞
aggregate DL transmit power of a typical cell Vi in the SR R
⎜ 2 ln 2λu K (K λb ) K 2 ρ 2 B 3R ⎟
mode can be expressed as ⎜ ρ B ⎟
− βC (ρ) ⎜   R   K +1 ⎟. (52)
  ⎝ ⎠
A
E PSRi K λb − 2 ρ 2 B − 1 λu
⎛ ⎞
∞       After some algebraic
= ⎝ Tx
E Pi,k + E Pi, j ⎠ FAi Ni
Tx   manipulations, it can be shown that
∂ F (ρ) ∂ 2 E Pcell
SR
Ni =1 k j ∂ρ = β ∂ρ 2
> 0. This implies that F(ρ) is a
∞   λu Ai  Ni monotonically
" increasing function
" of ρ. Furthermore, we have
= Tx
Ni E PUE exp (−λu Ai ) F(ρ)"ρ= √2 < 0 and F(ρ)"ρ=1 > 0. Therefore, there must
Ni ! 2
√   
Ni =1
⎛ ⎞ exist a unique ρ  ∈ 22 , 1 which yields F ρ  = 0, that
∞  N ∞ 
  an unique optimal ρ that maximizes
2
2 Ni R/ρ B λu Ai i (λu i )
A Ni is, there always exists
= C (ρ) ⎝ − ⎠ ηESE (ρ). Setting F ρ = 0 yields (24).
SR
exp (λu Ai ) Ni ! exp (λu Ai ) Ni !
Ni =0 Ni =0
  R   
= C (ρ) exp 2 ρ 2 B − 1 λu Ai − exp (−λu Ai ) F. Proof of Proposition 8
Proof: Based on the definition (7), the outage probability
− C (ρ) (1 − exp (−λu Ai ))
  R    for a typical No-CoMP UE u (WoC) i, j conditioned on the BS-UE
= C (ρ) exp 2 ρ 2 B − 1 λu Ai − 1 , (48) distance r1 is given by
 
Ni  Ri, j /Bi, j 
α
where C (ρ) is given in (15). According to [47], the approxi- Q out
i, j [r 1 ] = 1 − Pr h i
j ≥ 2 − 1 r 1 I i, j
P
mate probability distribution function (PDF) of Ai is f (Ai ) = " m
K K λbK AiK −1 exp (−K λb Ai ) / (K ), where K = 3.75 and = 1 − L Ii, j (s)"s= Ni T r α , (53)
Pm 1

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 ZHAO et al.: JOINT ESE OPTIMIZATION OF CoMP AND BS DEPLOYMENT 4845

where L Ii, j (s) is the Laplace transform

  of Ii, j and T = [6] D. Astely, E. Dahlman, G. Fodor, S. Parkvall, and J. Sachs, “LTE release
/B Pm i  Ni 12 and beyond,” IEEE Commun. Mag., vol. 51, no. 7, pp. 154–160,
2 R i, j i, j − 1. Since Ni h j ∼ exp Pm , based on Appendix B Jul. 2013.
of [22], we have [7] C.-L. I, C. Rowell, S. Han, Z. Xu, G. Li, and Z. Pan, “Toward green and
  soft: A 5G perspective,” IEEE Commun. Mag., vol. 52, no. 2, pp. 66–73,
Ni α Feb. 2014.
L Ii, j Tr
Pm 1 [8] A. Asadi, V. Sciancalepore, and V. Mancuso, “On the efficient utiliza-
   Ni tion of radio resources in extremely dense wireless networks,” IEEE
∞ Commun. Mag., vol. 53, no. 1, pp. 126–132, Jan. 2015.
Pm
= exp −2πλb 1− Ni Ni α −α
v dv [9] S. Yang, X. Xu, D. Alanis, S. X. Ng, and L. Hanzo, “Is the low-
r1 Pm + Pm T r1 v complexity mobile-relay-aided FFR-DAS capable of outperforming the
  high-complexity CoMP?” IEEE Trans. Veh. Technol., vol. 65, no. 4,
= exp −πr12 λb τ , (54) pp. 2154–2169, Apr. 2016.
∞ [10] K. Son, H. Kim, Y. Yi, and B. Krishnamachari, “Base station oper-
where τ = T 2/α T −2/α 1+u1α/2 du. This completes the proof. ation and user association mechanisms for energy-delay tradeoffs in
green cellular networks,” IEEE J. Sel. Areas Commun., vol. 29, no. 8,
pp. 1525–1536, Sep. 2011.
G. Proof of Proposition 9 [11] J. Wu, S. Zhou, and Z. Niu, “Traffic-aware base station sleeping
Proof: By setting x = λλub and taking the derivative of control and power matching for energy-delay tradeoffs in green cel-
lular networks,” IEEE Trans. Wireless Commun., vol. 12, no. 8,
ηESE
SR with respect to x, we have pp. 4196–4209, Aug. 2013.
[12] Z. Yang and Z. Niu, “Energy saving in cellular networks by dynamic

     RS–BS association and BS switching,” IEEE Trans. Veh. Technol.,
∂ E Pcell SR   vol. 62, no. 9, pp. 4602–4614, Nov. 2013.
R −βx − β E Pcell
SR
− POM [13] K. Klessig, A. J. Fehske, and G. P. Fettweis, “Energy efficiency gains in
∂ηESE
SR ∂x interference-limited heterogeneous cellular mobile radio networks with
=   SR  2 , (55) random micro site deployment,” in Proc. 34th IEEE Sarnoff Symp.,
∂x B βx E Pcell + x POM Princeton, NJ, USA, May 2011, pp. 1–6.
[14] M. M. A. Hossain, K. Koufos, and R. Jäntti, “Energy efficient deploy-
with the formula of  expressed as ment of HetNets: Impact of power amplifier and delay,” in Proc. WCNC,
Shanghai, China, Apr. 2013, pp. 778–782.
−βC (ρ)
=     K + βC (ρ) [15] F. Richter, A. J. Fehske, and G. P. Fettweis, “Energy efficiency aspects of
1 − 2 R/ρ B − 1 /K x
2 base station deployment strategies for cellular networks,” in Proc. IEEE
 2
 70th Veh. Technol. Conf. Fall (VTC), Anchorage, AK, USA, Sep. 2009,
βC (ρ) 2 R/ρ B − 1 pp. 1–5.
+    K +1 − B POM , (56)
[16] K. T. K. Cheung, S. Yang, and L. Hanzo, “Achieving maximum
energy-efficiency in multi-relay OFDMA cellular networks: A frac-
1 − 2 R/ρ B − 1 /K x
2
x tional programming approach,” IEEE Trans. Commun., vol. 61, no. 8,
which satisfies pp. 2746–2757, Jul. 2013.
⎧ [17] F. Liu, K. Zheng, W. Xiang, and H. Zhao, “Design and performance
⎪
⎪ lim  = 0+ − B POM < 0, analysis of an energy-efficient uplink carrier aggregation scheme,” IEEE
⎨x→+∞ J. Sel. Areas Commun., vol. 32, no. 2, pp. 197–207, Feb. 2014.
lim   = +∞ + βC (ρ) − B PO M > 0. [18] K. Zheng, L. Zhao, J. Mei, B. Shao, W. Xiang, and L. Hanzo, “Survey
⎪
⎪
 +
⎩ 2 R/Bρ 2 −1 of large-scale MIMO systems,” IEEE Commun. Surveys Tuts., vol. 17,
x→ K
no. 3, pp. 1738–1760, 3rd. Quart., 2015.
(57) [19] T. Abrão, L. D. H. Sampaio, S. Yang, K. T. K. Cheung, P. J. E. Jeszensky,
and L. Hanzo, “Energy efficient OFDMA networks maintaining statis-
∂
It may be readily verify that < 0, which implies that
∂x
tical QoS guarantees for delay-sensitive traffic,” IEEE Access, vol. 4,
pp. 774–791, Feb. 2016.
 is a monotonically decreasing function of x. From the [20] K. T. K. Cheung, S. Yang, and L. Hanzo, “Distributed energy spec-
equality (57), therefore, there must exist a unique x  to enable tral efficiency optimization for partial/full interference alignment in
 = 0 and hence to make ∂ηESE SR /∂ x = 0. Thus this unique multi-user multi-relay multi-cell MIMO systems,” IEEE Trans. Signal
  Process., vol. 64, no. 4, pp. 882–896, Feb. 2016.
x = λb /λu maximizes ηESE . Setting  = 0 yields (29), and
SR
[21] K. T. K. Cheung, S. Yang, and L. Hanzo, “Spectral and energy spectral
this completes the proof. efficiency optimization of joint transmit and receive beamforming based
multi-relay MIMO-OFDMA cellular networks,” IEEE Trans. Wireless
R EFERENCES Commun., vol. 13, no. 11, pp. 6147–6165, Nov. 2014.
[22] J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach to
[1] S. Chen and J. Zhao, “The requirements, challenges, and technologies coverage and rate in cellular networks,” IEEE Trans. Commun., vol. 59,
for 5G of terrestrial mobile telecommunication,” IEEE Commun. Mag., no. 11, pp. 3122–3134, Nov. 2011.
vol. 52, no. 5, pp. 36–43, May 2014. [23] H. Zhang, S. Chen, L. Feng, Y. Xie, and L. Hanzo, “A universal
[2] X. Ge, S. Tu, G. Mao, and C. X. Wang, “5G ultra-dense cellular approach to coverage probability and throughput analysis for cellular
networks,” IEEE Trans. Wireless Commun., vol. 23, no. 1, pp. 72–79, networks,” IEEE Trans. Veh. Technol., vol. 64, no. 9, pp. 4245–4256,
Feb. 2016. Sep. 2015.
[3] S. F. Yunas, M. Valkama, and J. Niemelä, “Spectral and energy efficiency [24] H. S. Dhillon, R. K. Ganti, F. Baccelli, and J. G. Andrews, “Mod-
of ultra-dense networks under different deployment strategies,” IEEE eling and analysis of K-tier downlink heterogeneous cellular net-
Commun. Mag., vol. 53, no. 1, pp. 90–100, Jan. 2015. works,” IEEE J. Sel. Areas Commun., vol. 30, no. 3, pp. 550–560,
[4] I. Hwang, B. Song, and S. S. Soliman, “A holistic view on hyper-dense Apr. 2012.
heterogeneous and small cell networks,” IEEE Commun. Mag., vol. 51, [25] Y. S. Soh, T. Q. S. Quek, M. Kountouris, and H. Shin, “Energy efficient
no. 6, pp. 20–27, Jun. 2013. heterogeneous cellular networks,” IEEE J. Sel. Areas Commun., vol. 31,
[5] H. A. U. Mustafa, M. A. Imran, M. Z. Shakir, A. Imran, and R. Tafazolli, no. 5, pp. 840–850, May 2013.
“Separation framework: An enabler for cooperative and D2D communi- [26] J. Peng, P. Hong, and K. Xue, “Energy-aware cellular deployment
cation for future 5G networks,” IEEE Commun. Surveys Tuts., vol. 18, strategy under coverage performance constraints,” IEEE Trans. Wireless
no. 1, pp. 419–445, 1st Quart., 2016. Commun., vol. 14, no. 1, pp. 69–80, Jan. 2015.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 4846 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

[27] S.-R. Cho and W. Choi, “Energy-efficient repulsive cell activation for Guogang Zhao received the B.S. degree in
heterogeneous cellular networks,” IEEE J. Sel. Areas Commun., vol. 31, communications engineering from People’s Lib-
no. 5, pp. 870–882, May 2013. eration Army Information Engineering University,
[28] G. Zhao, S. Chen, L. Zhao, and L. Hanzo, “A tele-traffic-aware Zhengzhou, China. He is currently pursuing the
optimal base-station deployment strategy for energy-efficient large- Ph.D. degree with the Department of Telecommuni-
scale cellular networks,” IEEE Access, vol. 4, pp. 2083–2095, cation Engineering, Xidian University, China. Since
Mar. 2016. 2015, he has been a visiting Ph.D. student with the
[29] X. Ge, B. Yang, J. Ye, G. Mao, C.-X. Wang, and T. Han, “Spatial spec- School of Electronics and Computer Science, Uni-
trum and energy efficiency of random cellular networks,” IEEE Trans. versity of Southampton, Southampton, U.K., with
Commun., vol. 63, no. 3, pp. 1019–1030, Mar. 2015. Prof. L. Hanzo and Prof. S. Chen. His research inter-
ests focus on 5G wireless communication, network
[30] X. Ge, S. Tu, T. Han, and Q. Li, “Energy efficiency of small cell
planning, and ultra-dense cellular network.
backhaul networks based on gauss-Markov mobile models,” IET Netw.,
vol. 4, no. 2, pp. 158–167, 2015.
[31] F. Heliot, M. A. Imran, and R. Tafazolli, “Energy efficiency analysis
of idealized coordinated multi-point communication system,” in Proc.
IEEE 73rd Veh. Technol. Conf. (VTC), Yokohama, Japan, May 2011, Sheng Chen (M’90–SM’97–F’08) received the
pp. 1–5. B.Eng. degree in control engineering from the
[32] K. M. S. Huq, S. Mumtaz, J. Bachmatiuk, J. Rodriguez, X. Wang, East China Petroleum Institute, Dongying, China,
and R. L. Aguiar, “Green HetNet CoMP: Energy efficiency analy- in 1982, the Ph.D. degree in control engineering
sis and optimization,” IEEE Trans. Veh. Technol., vol. 64, no. 10, from the City, University of London, London, U.K.,
pp. 4670–4683, Oct. 2015. in 1986, and the D.Sc. degree from the University
[33] Z. Liu, Y. Zhou, X. Han, L. Tian, J. Hu, and J. Shi, “Energy efficiency of Southampton, Southampton, U.K.
of CoMP-based cellular networks with guaranteed coverage,” in Proc. From 1986 to 1999, he held research and acad-
WCNC, Shanghai, China, Apr. 2013, pp. 2034–2039. emic appointments with the Universities of Sheffield,
Edinburgh, and Portsmouth, in U.K. Since 1999,
[34] S. Han, C. Yang, and A. F. Molisch, “Spectrum and energy efficient
he has been with the School of Electronics and
cooperative base station doze,” IEEE J. Sel. Areas Commun., vol. 32,
Computer Science, University of Southampton, where he is currently a
no. 2, pp. 285–296, Feb. 2014.
Professor in intelligent systems and signal processing. He has authored over
[35] D. Cao, S. Zhou, C. Zhang, and Z. Niu, “Energy saving perfor- 550 research papers. His research interests include adaptive signal processing,
mance comparison of coordinated multi-point transmission and wireless wireless communications, the modeling and identification of nonlinear sys-
relaying,” in Proc. IEEE Global Telecommun. Conf. (GLOBECOM), tems, neural network and machine learning, intelligent control system design,
Dec. 2010, pp. 1–5. and evolutionary computation methods and optimization.
[36] F. Baccelli and A. Giovanidis, “A stochastic geometry framework for Dr. Chen is a fellow of the United Kingdom Royal Academy of Engineering,
analyzing pairwise-cooperative cellular networks,” IEEE Trans. Wireless a fellow of the IET, a Distinguished Adjunct Professor with King Abdulaziz
Commun., vol. 14, no. 2, pp. 794–808, Feb. 2015. University, Jeddah, Saudi Arabia, and an ISI Highly Cited Researcher in
[37] A. H. Sakr and E. Hossain, “Location-aware cross-tier coordi- engineering in 2004.
nated multipoint transmission in two-tier cellular networks,” IEEE
Trans. Wireless Commun., vol. 13, no. 11, pp. 6311–6325,
Nov. 2014.
[38] V. Garcia, Y. Zhou, and J. Shi, “Coordinated multipoint transmis- Liqiang Zhao received the B.Eng. degree in electri-
sion in dense cellular networks with user-centric adaptive cluster- cal engineering from Shanghai Jiaotong University,
ing,” IEEE Trans. Wireless Commun., vol. 13, no. 8, pp. 4297–4308, China, in 1992, and the M.Sc. degree in commu-
Aug. 2014. nications and information systems and the Ph.D.
[39] G. Nigam, P. Minero, and M. Haenggi, “Coordinated multipoint degree in information and communications engineer-
joint transmission in heterogeneous networks,” IEEE Trans. Commun., ing from Xidian University, China, in 2000 and
vol. 62, no. 11, pp. 4134–4146, Nov. 2014. 2003, respectively. In 2008, he was given an award
[40] Discussions on CoMP SU-MIMO, document R1-090613, 3GPP TSG by the Program for New Century Excellent Talents
RAN WG1 Meeting #56, Samsung, Feb. 2009. in University, Ministry of Education, China.
[41] Further Discussion of Frequency Plan Scheme on Comp-SUMIMO, From 1992 to 2005, he was a Senior Research
document R1-091415, 3GPP TSG-RAN WG1 #56bis, Potevio, Engineer with the 20th Research Institute, Chinese
Mar. 2009. Electronics Technology Group Corporation, China, where he was involved in
[42] L. F. Williams, Jr., “A modification to the half-interval search (binary mobile communication systems and spread spectrum communications. From
search) method,” in Proc. 14th Annu. Southeast Regional Conf., 2005 to 2007, he was an Associate Professor with the State Key Laboratory
Apr. 1976, pp. 95–101. of Integrated Service Networks, Xidian University, China, where he was
involved in WiMAX, WLAN, and wireless sensor network. He was appointed
[43] W. Guo, S. Wang, X. Chu, J. Zhang, J. Chen, and H. Song, “Automated
a Marie Curie Research Fellow with the Center for Wireless Network Design,
small-cell deployment for heterogeneous cellular networks,” IEEE Com-
University of Bedfordshire, in 2007, to conduct research in the GAWIND
mun. Mag., vol. 51, no. 5, pp. 46–53, May 2013.
project funded under EU FP6 HRM program, where his activities focused
[44] Y. Zhou et al., “Large-scale spatial distribution identification of base on automatic wireless broadband access network planning and optimization.
stations in cellular networks,” IEEE Access, vol. 3, pp. 2987–2999, Since 2008, he has been with Xidian University. His current research focuses
Dec. 2015. on broadband wireless communications and space communications. He has
[45] S. Tombaz, A. Vastberg, and J. Zander, “Energy- and cost-efficient ultra- authored or co-authored over 70 publications in authorized academic periodi-
high-capacity wireless access,” IEEE Wireless Commun., vol. 18, no. 5, cals both in China and abroad and in international science conferences, 20 of
pp. 18–24, Oct. 2011. which are retrieved in SCI, and over 50 of them are EI indexed. He has also
[46] S. Srinivasa, “Modeling interference in uniformly random wireless applied for five national invention patents. He has hosted/participated in many
networks: Theory and applications,” Ph.D. dissertation, Dept. Elect. vertical research projects, such as the National Natural Science Foundation,
Eng., Univ. Notre Dame, Dame, IN, USA, 2007. 863 Program and the National Science and Technology Major Projects, China,
[47] J.-S. Ferenc and Z. Néda, “On the size distribution of Poisson Voronoi as well as many horizontal scientific research projects, including the EU
cells,” Phys. A, Statist. Mech. Appl., vol. 385, no. 2, pp. 518–526, FP6, FP7 Plans for International Cooperation and Exchange Projects, and
Nov. 2007. the Huawei Fund.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.
 ZHAO et al.: JOINT ESE OPTIMIZATION OF CoMP AND BS DEPLOYMENT 4847

Lajos Hanzo (M’89–SM’91–F’04) FREng, FIEEE, in excess of 10 000 pages, published 1600+ research contributions at IEEE
FIET, Fellow of EURASIP, DSc received his degree Xplore, acted both as TPC and General Chair of IEEE conferences, presented
in electronics in 1976 and his doctorate in 1983. keynote lectures and has been awarded a number of distinctions. Currently
In 2009 he was awarded an honorary doctorate by he is directing a 60-strong academic research team, working on a range of
the Technical University of Budapest and in 2015 research projects in the field of wireless multimedia communications spon-
by the University of Edinburgh. In 2016 he was sored by industry, the Engineering and Physical Sciences Research Council
admitted to the Hungarian Academy of Science. (EPSRC) UK, the European Research Council’s Advanced Fellow Grant and
During his 40-year career in telecommunications he the Royal Society’s Wolfson Research Merit Award. He is an enthusiastic
has held various research and academic posts in supporter of industrial and academic liaison and he offers a range of industrial
Hungary, Germany and the UK. Since 1986 he has courses. He is also a Governor of the IEEE VTS. During 2008–2012 he
been with the School of Electronics and Computer was the EditorinChief of the IEEE Press and a Chaired Professor also at
Science, University of Southampton, UK, where he holds the chair in telecom- Tsinghua University, Beijing. For further information on research in progress
munications. He has successfully supervised 111 PhD students, coauthored and associated publications please refer to http://wwwmobile.ecs.soton.ac.uk.
18 John Wiley/IEEE Press books on mobile radio communications totalling Lajos has 30 000+ citations and an H-index of 68.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on March 29,2022 at 12:24:00 UTC from IEEE Xplore. Restrictions apply.