Sustainable Cities and Society 63 (2020) 102445

Contents lists available at ScienceDirect

Sustainable Cities and Society

journal homepage: www.elsevier.com/locate/scs

Optimizing the ultra-dense 5G base stations in urban outdoor areas:

Coupling GIS and heuristic optimization
Qi Wang a, Xiang Zhao a, *, Zhantian Lv a, Xiaoya Ma b, c, Ruitian Zhang a, Yifan Lin a
a
School of Resource and Environmental Science, Wuhan University, Wuhan, 430079, China
b
School of Geoscience, Yangtze University, Wuhan, 430010, China
c
Key Laboratory of Urban Land Resources Monitoring and Simulation, Ministry of Natural Resources, Shenzhen, Guangdong, 518034, China

A R T I C L E I N F O A B S T R A C T

Keywords: Due to the high propagation loss and blockage-sensitive characteristics of millimeter waves (mmWaves), con­
5G structing fifth-generation (5G) cellular networks involves deploying ultra-dense base stations (BSs) to achieve
Base station satisfactory communication service coverage. However, ultra-densely deployed BSs are associated with
GIS
extremely high construction and operation costs for 5G cellular networks. Reducing the construction cost and
Optimization
Network planning
decreasing the energy consumption of BSs under the premise of ensuring the quality and coverage of services
have become major challenges for the rollout of 5G technology. Essentially, the location optimization of 5G BSs
can be regarded as a type of maximum coverage location problem (MCLP). Hence, this study coupled geographic
information system (GIS) and a heuristic optimization algorithm to spatially explicit simulate the propagation of
5G signals and to optimize the service coverage of 5G BSs. The developed model was applied to search for the
optimal solutions in 5G cellular network planning for an urban outdoor area in Wuhan, China. The optimal
solutions and comparative experiments demonstrate that the proposed model can provide reasonable and robust
results to support 5G cellular network planning. Therefore, this approach can help address the cost and energy
consumption challenges faced in constructing 5G infrastructures and facilitate the rollout of 5G technology.

1. Introduction required for the construction of 5G cellular networks. For instance, as

To meet the rapidly growing demand for wireless communication in
the coming decade, in recent years, fifth-generation (5G) communica­
tion technology has been developed. As the latest generation of cellular
communication technology, 5G provides a significantly larger data ca­
pacity, a higher speed and extremely low latency in wireless data
transformation compared to previous technologies. Furthermore, 5G is
expected to revolutionize the development of many emerging technol­
ogies (Khan, Pi, & Rajagopal, 2012; Pi & Khan, 2011), such as the
Internet of Things (IoT), self-driving cars, virtual reality (VR) and arti­
ficial intelligence (AI). Moreover, 5G is regarded as one of the most
essential drivers of clean economic growth in the Industry 4.0 era
(Adam, 2019). Consequently, governments around the world, especially
in developed countries, have attached great importance to the devel­
opment and rollout of 5G technology.
Although 5G technology has been developed and matured in labo­
ratories for a long time, the rollout of 5G still faces some challenges
(Jingyi, 2019). The first challenge is the huge capital investment
Yizhong Li (the former minister of the Ministry of Industry and Infor­
mation Technology of China) estimated (Jingyi, 2019), the construction
of 5G cellular networks in China might last 6-7 years and cost 170-200
billion US dollars to achieve nationwide service coverage. To cover the
same area as traditional cellular networks (2G, 3G, and 4G), the number
of 5G base stations (BSs) could be tripled (Wang et al., 2014). Further­
more, Ge, Tu, Mao, Wang, and Han, (2016) suggested that to achieve
seamless coverage services, the density of 5G BSs would reach 40-50
BSs/km2. Another challenge for the rollout of 5G is posed by concerns
about power consumption. In the pre-5G era, information and commu­
nication technology already consumes 10% of energy worldwide
(Usama & Erol-Kantarci, 2019). The environmental and economic costs
associated with the energy consumption of BSs cannot be ignored (Buzzi
et al., 2016). It is estimated that the energy consumption of each 5G BS is
approximately 2-3 times that of a 4G BS (I et al., 2014). Considering the
ultra-dense deployment characteristics of 5G BSs, to achieve seamless
service coverage in the same area, the total energy consumption of 5G
BSs might be 9 times or more than that of 4G BSs.

* Corresponding author.
E-mail address: zhaoxiang@whu.edu.cn (X. Zhao).

https://doi.org/10.1016/j.scs.2020.102445
Received 6 February 2020; Received in revised form 30 June 2020; Accepted 9 August 2020
Available online 15 August 2020
2210-6707/© 2020 Elsevier Ltd. All rights reserved.
Q. Wang et al.
Clearly, optimizing the deployment of 5G cellular networks will play
an essential role in addressing these challenges related to the rollout of
5G (Buzzi et al., 2016) because optimization can significantly reduce the
number of BSs required to cover the same area. Not only is the location
optimization of 5G BSs an effective way to downscale the energy con­
sumption of cellular networks, but it also helps reduce the construction
cost of 5G cellular networks and helps improve service coverage.
Although previous studies have developed many optimization models to
solve the BS location optimization problems in 2G/3G/4G cellular
network planning, a robust and spatially explicit optimization model
that considers the propagation characteristics of 5G signals for the
location optimization of 5G BSs is still lacking.
The objective of this study is to develop a location optimization
model to support the planning of ultra-dense 5G BSs in urban outdoor
areas and to help address the cost challenges facing 5G. In this study, we
couple geographic information system (GIS) and a heuristic algorithm to
search for the optimal locations of each BS in a 5G network. The spatial
modelling and visualization approaches provided by GIS will be used to
simulate the signal propagation and service coverage of 5G BSs in urban
outdoor areas. An artificial immune system (AIS) algorithm, which is a
robust heuristic optimization algorithm, will be employed to optimize
the service coverage of 5G cellular networks. The AIS algorithms were
developed based on the principles of natural immune systems (de Castro
& Timmis, 2003), and AIS algorithms have been widely used in spatial
optimization (Ma & Zhao, 2015; Zhao, Ma, Tang, & Liu, 2019). Previous
studies have shown that AIS algorithms can avoid local optima and
provide superior convergence characteristics for solving optimization
problems (Huang, Liu, Li, Liang, & He, 2013; Shang, Jiao, Liu, & Ma,
2012).
The remainder of this paper is organized as follows: Section 2 pre­
sents a short introduction on 5G communication technology and a brief
review of location optimization models for cellular network planning.
The formulation and modelling of 5G BS optimization problems are
defined in Section 3. Section 4 describes the location optimization al­
gorithm based on GIS and the AIS algorithm. Section 5 designs experi­
ments and discusses the results. Section 6 concludes and presents future
work.
2. Background
2.1. 5G and 5G cellular network deployment
5G communication technology uses a high-frequency millimeter
wave (mmWave) to carry huge amounts of data over a short distance
(Bai & Heath, 2015). Due to the strong absorption effect of the atmo­
sphere on mmWaves and the extremely high penetration loss, the service
coverage radius of a 5G BS is much shorter than that of previous tech­
nologies (such as 2G, 3G and 4G). Most of the service/coverage radii of
5G BSs are between 100 and 300 meters (Maccartney, Zhang, Nie, &
Rappaport, 2013; Sulyman et al., 2014).
In addition, the densely distributed buildings in urban areas limit the
propagation and coverage of 5G signals. According to Pi and Khan
(2011), even for buildings made of common materials such as bricks and
concrete, the penetration loss of mmWaves in urban areas is excessively
high, which can lead to a significant isolation of indoor and outdoor
areas (Palizban, Szyszkowicz, & Yanikomeroglu, 2017). This penetra­
tion loss phenomenon in 5G propagation is called the line-of-sight (LOS)
effect (Al-Falahy & Alani, 2019; Maccartney et al., 2013; Qamar et al.,
2019). Thus, mmWaves, which have high penetration and path loss (Lee
et al., 2018; Maccartney et al., 2013), are susceptible to interference (Bai
& Heath, 2015) and are very sensitive to environmental blockages
(Sulyman et al., 2014). The LOS coverage in the demand area must be
considered to ensure that the quality of service (QoS) of 5G is adequate
(Rappaport et al., 2013).
Clearly, the propagation characteristics of mmWaves have led to
huge investment cost challenges for the rollout of 5G. To achieve
2
Sustainable Cities and Society 63 (2020) 102445
satisfactory service coverage, 5G telecom operators must deploy ultra-
dense BSs in urban areas. In recent years, using energy-saving technol­
ogies (Buzzi et al., 2016) and upgrading 4G BSs to 5G BSs (Sharma et al.,
2018) to reduce the construction and operation costs of 5G cellular
networks have also gained a considerable amount of attention. BS
energy-saving technology aims to reduce the energy cost in operating
cellular networks by adjusting the operating status of BSs or by
providing cleaner energy such as solar or wind energy for BSs (Usama &
Erol-Kantarci, 2019). Furthermore, dynamic spectrum sharing (DSS)
(Sharma et al., 2018) and non-standalone (NSA) networking technolo­
gies mainly focus on reducing the construction cost of 5G cellular net­
works by upgrading existing 4G BSs to support 5G communication
(Salah & Rinne, 2018; Yaovaja & Klunngien, 2019).
The first rollout of 5G networks usually involves NSA deployments
that can help telecom operators maximize the use of installed 4G BSs, be
the first to launch 5G, and gain technology and market leadership
(Ekstrom, 2019). Nevertheless, the most effective way to reduce the
construction cost of 5G cellular networks is to cover the demand area
with the minimum number of BSs deployed. According to a market
survey conducted by China Mobile, which is China’s largest mobile
communication provider, the total investment for deploying a new 5G
BS in China is presently approximately 53,000 US dollars (Fan, Diao,
Liu, & Wang, 2019). On the other hand, although upgrading 4G BSs to
5G BSs can reduce the construction cost of 5G cellular networks, there
are not enough existing BSs to meet the demand of 5G service coverage.
In other words, we still have to choose new places for most of the BSs in
5G cellular network planning to be deployed.
2.2. A brief overview of spatial optimization models for cellular network
planning
Service coverage is considered an important indicator of the QoS of
cellular networks (Yildirim, Kalayci, & Ugur, 2008). In this sense, the
location optimization of BSs in cellular networks can be regarded as a
typical maximum coverage location problem (MCLP) (R. Church &
ReVelle, 1974). Previous studies have found that heuristic algorithms
are efficient in and robust for solving MCLPs (Ismail & Abd Manaf, 2010;
Rabbani, Heidari, Farrokhi-Asl, & Rahimi, 2018). Therefore, in recent
decades, a variety of heuristic algorithms, such as genetic algorithms
(GAs) (Chen, Yang, & Su, 2014; Guo & Mu, 2003; Qin, Guan, & Xie,
2004), AIS algorithms (Wang & Shen, 2017), and firefly algorithms
(Eldrandaly, Abdel-Basset, & Abdel-Fatah, 2019), have been used to
improve the spatial coverage and signal quality of 2G/3G/4G cellular
networks.
One of the greatest challenges encountered by existing heuristic-
based models in solving 5G BS location optimization lies in the simu­
lation of LOS propagation and the coverage of 5G signals in urban en­
vironments. Most of the existing location optimization models for
cellular network planning were developed for 2G/3G/4G networks
(Krishnamachari & Wicker, 2004). The penetration loss of 2G/3G/4G
signals in urban areas is far lower than that of 5G signals (Maccartney
et al., 2013). Consequently, the location optimization models for pre­
vious generations of cellular networks did not have to simulate the LOS
propagation of signals.
In recent studies, to simulate the LOS propagation and coverage of
5G signals, GIS has been employed to support optimization and to
conduct intuitive visual analyses for the planning of 5G networks. For
instance, Szyszkowicz, Lou, & Yanikomeroglu (2016) used GIS and a
computational geometry approach to optimize the locations of
below-rooftop wall-mounted 5G BS in urban outdoor areas with the
support of GIS (Palizban et al., 2017; Szyszkowicz et al., 2016). How­
ever, the computational geometry approach is not an appropriate
method for global optimization. On the other hand, recent studies have
also developed some spatially-implicit heuristic algorithm-based models
to support 5G cellular network planning in some special 5G application
scenarios (Cicek, Gultekin, Tavli, & Yanikomeroglu, 2019; Jimenez,
Q. Wang et al.
Chiaraviglio, Amorosi, & Blefari-Melazzi, 2018; Lyu, Zeng, Zhang, &
Lim, 2017). For example, unmanned aerial vehicles (UAVs) carrying 5G
BSs can effectively reduce the penetration loss of 5G signals in urban
areas. Hence, these optimization models for UAV BS deployment do not
simulate the LOS propagation of 5G signals.
Determining how to optimize the locations of BSs to cover the service
demand area with the minimum number of 5G BSs deployed, according
to the spatial distribution of buildings in urban outdoor areas, still re­
quires further study. Simulating the LOS propagation and service
coverage of each BS is indispensable to better support decision-making
in 5G network planning (Galiotto, Pratas, Doyle, & Marchetti, 2017;
Li, Shi, Wu, Liu, & Zhang, 2018; Tian et al., 2018; Zhang & Hu, 2018).
3. Location optimization of Ultra-dense 5G BS in urban outdoor
areas
3.1. LOS service coverage model
Most of the coverage demand for 5G services in urban outdoor areas
is distributed in a space not exceeding 10 meters above the ground.
Therefore, the service demand can be represented in a two-dimensional
(2D) spatial framework. Furthermore, although three-dimensional (3D)
spatial analysis can be performed with some commercial GIS software
packages such as ArcGIS, these analyses are extremely time consuming
and difficult to integrate into heuristic optimization algorithms. Hence,
we use 2D GIS to simulate the LOS propagation and service coverage of
5G BSs.
Although the propagation loss of mmWaves in urban outdoor areas is
affected by many environmental factors (temperature, humidity,
weather, etc.), these factors can be regarded as homogeneous in a local
urban area. In addition, the penetration loss of mmWaves between
densely distributed buildings is undoubtedly the most important factor
that influences the coverage of 5G services in urban areas (Al-Dabbagh,
Al-Aboody, & Al-Raweshidy, 2017; Lu, Hsu, Chen, & Lee, 2018; Rap­
paport et al., 2017; Wang et al., 2014). In other words, the blockage of
buildings is the main cause of the spatial heterogeneity of 5G signal
propagation. Therefore, the simulation of 5G signal propagation and the
spatial optimization of 5G BSs in urban outdoor areas should focus on
the spatial heterogeneity of 5G service coverage. The model for simu­
lating the LOS propagation of 5G signals in urban outdoor areas is shown
Fig. 1. Line-of-sight service coverage model for 5G BSs in urban outdoor areas.
3
Sustainable Cities and Society 63 (2020) 102445
in Fig. 1.
As Fig. 1 illustrates, the modeling of LOS coverage includes the
following steps. (1) Discretizing a continuous service space. To meet the
requirements of LOS analysis and reduce the search space of optimiza­
tion algorithms, the continuous service space in the real world is dis­
cretized using a systematic sampling method. As a result, the continuous
service space is represented by a series of regular grid points. Each point
represents a square area in the real world. The smaller the sampling
interval d is, the closer the grid points are to the real continuous space
and the greater the number of computations required. Both the demand
for 5G services and the candidate locations for BSs are represented by
these points in the discretized space. In addition, the areas where the
deployment of 5G BSs is opposed by landowners or where the land-use
condition is not suitable for BS deployment (large lakes, etc.) can be
excluded from the candidate location set using the overlay analysis
provided by GIS. (2) Modeling of buildings in GIS. The spatial distri­
bution and the outlines of buildings are modeled as 2D vector polygons
in this study. (3) Simulating the LOS service coverage. For a given
maximum effective service radius R of a 5G BS, the service area of a BS
can be described as a circle. In addition, this study focuses only on the
LOS coverage of 5G signals. We assume that only the points that have
LOS communication with a BS and that fall in the service radius of a BS
can be effectively covered by 5G services (black squares in Fig. 1). The
points located a distance greater than R from a BS (gray solid circles in
Fig. 1) or the points falling in the service area without LOS coverage
(gray solid squares in Fig. 1) cannot receive effective 5G signals.
3.2. Formulation of the 5G BS optimization problem
The location optimization approach for 5G BSs aims to cover the
service demand area with the minimum number of BSs or to maximize
the service coverage area of a given number of BSs. To solve this typical
coverage problem, an MCLP model was employed for the location
optimization of 5G BSs. MCLP models aim to maximize the service
coverage of a given number of facilities or fixed investments (Church &
ReVelle, 1974; Church, Stoms, & Davis, 1996; Yao, Zhang, & Murray,
2019). The proposed location optimization model for 5G BSs extends the
MCLP model by considering the spatial heterogeneity of 5G mmWave
propagation in urban outdoor environments. Accordingly, the location
optimization of 5G BSs in urban outdoor areas can be formulated as
Q. Wang et al. Sustainable Cities and Society 63 (2020) 102445

follows: includes the following steps (de Castro & Von Zuben, 2002). (1)
Encoding. A feasible solution to an optimization problem is encoded as
∑
M
/
Maximize SC = Di × 100 M (1) the artificial individuals (antibodies) of the algorithm. Each decision
i=1 variable (location of a BS) in the solution is represented as virtual units
{ (genes) in the AIS algorithm. (2) Initialization. The initialization process
Di =
1, if Di ∈ ϕ
(2) randomly generates the initial population with T antibodies for p deci­
0, otherwise sion variables, where T is an integer used to denote the size of the
population.
ϕ = {∀Ck ∈ S : (dik ≤ R ∧ Di ∈ LOSk )} (3) (3) Evaluation. The affinity of each antibody in the population is
evaluated using the objective function defined in formulas 1-5. The
∑
N
Subject to Cj = p (4) higher the affinity level is, the better the solution represented by the
j=1 antibody. As Fig. 3b illustrates, in this step, the buffer and LOS analysis
approaches provided by GIS were integrated to assess the service
{
1, if Cj ∈ S coverage of each antibody (solution). (4) Determining whether the al­
Cj = (5)
0, otherwise gorithm has converged or whether the algorithm has reached the
maximum number of iterations. If one of these conditions is met, then
where SC is the proportion of demand points covered by LOS 5G signals; the iteration will be terminated, and the optimization process will stop;
D and C denote the sets of demand points and candidate locations for BSs otherwise, the algorithm will proceed to the next step. (5) Cloning. In
in the entire study area, respectively; and M and N represent the the cloning step, antibodies with high affinity levels will be selected and
numbers of elements in D and C, respectively. p is a constant given by cloned to produce a new generation of antibodies. (6) Hyper-mutation.
decision makers, and it represents the number of BSs to be deployed in The gene values in antibodies are altered randomly to produce new in­
the area. S is a subset of C and represents an alternative solution for 5G dividuals for the next generation and to obtain better solutions. Steps 3-6
BS planning; the number of BSs in S equals p. LOSk is the set of demand above are repeated until the termination condition is met. The details of
points with LOS coverage provided by the k-th BS in S. R is the maximum the AIS method of optimization can be found in previous studies (Deb,
distance at which the demand points can obtain effective 5G service Pratap, Agarwal, & Meyarivan, 2002). Once the algorithm converges or
from a BS. dik is the distance of the i-th demand point from the k-th terminates, the maximum service coverage of the p BSs in the optimal
selected BS in S. Thus, ϕ is the set of LOS-covered demand points that solution is assessed. If the service coverage constraint in the study area
can obtain effective services from the 5G cellular networks in the area. has not been satisfied, then the value of p is increased gradually, and a
Once the locations of BSs are determined, the set of ϕ can be obtained by subsequent round of optimization occurs based on steps 1-6. Otherwise,
using buffer analysis and LOS analysis in GIS. the optimal solutions we obtained can be used to support 5G cellular
The MCLP model seeks to maximize the coverage area of p BSs. network planning.
Usually, the value of p is determined by prior knowledge according to
the area of the study area and the distribution of buildings. To achieve
satisfactory coverage of the study area, the value of p is gradually
adjusted by assessing the maximum demand points covered by p BSs
(Yao et al., 2019). We can use the following equation to determine the
initial value of p:
/
pinitial = A (6)
π × R2

where A is the area of the whole study area and π is the mathematical
constant that is approximately equal to 3.14159. Thus, we can solve the
optimization problem by starting with p = pinitial and then determine
whether the value of SC meets the expected service standards. If not,
then p is gradually increased until the service coverage is satisfactory.

4. Coupling AIS algorithm with GIS for location optimization

The major task in the location optimization of 5G BSs consists of

selecting the optimal locations for p BSs from N candidate locations to
maximize the service coverage of the corresponding cellular networks.
Clearly, the optimization of p locations is an NP-hard problem that
cannot be solved in polynomial time. Previous studies have already
shown that heuristic algorithms are effective in solving these types of
NP-hard optimization problems (Lakshminarasimman, Baskar,
Alphones, & Iruthayarajan, 2011; Seyhan, Snyder, & Zhang, 2018).
Therefore, we coupled an AIS algorithm with GIS to search for the
optimal deployment solutions for 5G BSs. The encoding scheme and
mutation operations of the algorithm were improved to integrate the
data model and spatial analysis methods in GIS.

4.1. AIS algorithm for optimization

In recent decades, many AIS algorithms have been developed based

Fig. 2. Framework for the GIS-based AIS algorithm for 5G BS optimization.
on different immune mechanisms and theories for various purposes. As
Fig. 2 shows, a classical AIS algorithm for optimization typically

4
Q. Wang et al.
Fig. 3. Improved spatially explicit immune operators for 5G BS optimization.
4.2. AIS algorithm coupled with GIS
To perform spatially explicit simulation and optimization, the
encoding scheme and hyper-mutation operations in the classical AIS
algorithm were improved to integrate the data model and spatial anal­
ysis methods provided by GIS. The principles of the improved immune
operators are illustrated in Fig. 3.
4.2.1. Encoding and initialization
In heuristic algorithms, encoding often aims to build a mapping
relationship between the feasible solution in the real world and in­
dividuals (antibodies) in the virtual model. As Figs. 1 and 3a demon­
strate, the demand points and candidate locations are modeled using a
vector data model provided by GIS. Thus, for any alternative solution for
5G BS optimization, we can use an array with p elements to represent the
antibodies in the virtual optimization model. Each BS in the solution is
mapped as a gene of the antibodies. Once a candidate location is selected
as a candidate BS, the corresponding object ID (OID) in the candidate
location dataset will be stored in a gene. To perform service coverage
assessment, the geographical coordinates of these locations are also
stored in genes.
The initialization operation is designed based on the encoding
scheme of the algorithm. In this study, we used a stochastic approach to
generate the first generation of antibodies. The steps involved in
generating an antibody are as follows: (1) set j=1, and randomly select a
candidate location c from set C that contains N candidate locations; the
OID and coordinates of c will be stored in the first gene of the antibody.
(2) Remove the selected c from set C, and set j = j + 1 and N = N-j. Steps
1-2 are repeated until j equals p. Thus, we can obtain a randomly pro­
duced antibody. The initialization algorithm runs T times, and we can
obtain the first-generation population with T antibodies for the AIS
algorithm.
4.2.2. Hyper-mutation
Hyper-mutation is process commonly used to produce new anti­
bodies in AIS algorithms. As shown in Fig. 3c, for a given mutation rate
Pm , the hyper-mutation operation involving an antibody includes the
following steps. (1) For each gene g in the antibody, a float number Pr
between 0.0-1.0 is randomly generated; (2) if Pr < Pm , then randomly
select a candidate location from C, that is not present in the current gene
set of the antibody (solution); otherwise, the mutation operation will be
skipped for gene g. (3) Replace the value of g with the new location
selected in the previous step. All the genes in the antibody are traversed
to perform the steps above, and at that point, the mutation operation is
completed for the selected antibody.
4.3. Model implementation
The proposed model was developed using the C++ programming
language. The GDAL/OGR (http://www.gdal.org) open source library
was used for spatial data access and management in our model. In
addition, the spatial analysis functions were implemented using the
GEOS (https://geos.osgeo.org/doxygen/) open source geometry engine
5
Sustainable Cities and Society 63 (2020) 102445
library. All the vector data, including the demand points, candidate lo­
cations, buildings and so on, were stored in a file geodatabase. The tests
and the experiments in this study were performed on a multicore
workstation with a Windows operating system installed; it had an Intel
Xeon E5-1620 4-core 3.5GHz CPU and 8 GB memory.
5. Case study
5.1. Study area and data
We selected several blocks from Wuhan as the study area (see Fig. 4).
Wuhan is the capital city of Hubei Province, China, and is located in
central China. The population of the city is 11 million people. As of the
end of 2018, the GDP of Wuhan had reached 212 billion US dollars.
According to the development plan of the municipal government, by the
end of 2020, over 20,000 5G BSs will be deployed in the downtown area
of Wuhan. The study area was selected from the downtown area of
Wuhan. The total area of the study area and the total area of buildings in
the study area are 356.1 ha and 64.5 ha, respectively. Thus, the building
density in the study area is approximately 0.18.
The data we used to perform the optimization are listed in Table 1.
All data were processed and stored in a file geodatabase. The study area
was discretized using 25 × 25 m grid points. The service demand in the
study area and the candidate locations for BSs were selected from these
points. In addition, the points covered by building polygons were
eliminated since we focused only on the outdoor areas in this study.
Thus, we obtained 4,693 demand points/candidate locations for BSs in
the study area.
5.2. Optimal results
To cover the study area with the minimum number of BSs, the model
that we developed in this study was used to optimize the locations for
the 5G BSs, and a series of optimal solutions was obtained in this
experiment (denoted as E1). We set the initial value of p equal to 50, and
then, p was gradually increased to 200 in steps of 10. The population size
of the algorithm was set to 100, and the maximum number of iterations
was set to 500. The mutation rate was adjusted according to the number
of BSs deployed, which equals 1.0 divided by p. The selection rate of the
algorithm was set to 0.15; thus, only the top 15 antibodies were selected
and cloned to produce a new generation of antibodies. Each antibody
selected was cloned four times to perform the hypermutation operation.
As a type of stochastic algorithm, our model was repeated 20 times for
each value of p to ensure that the algorithm converged. In addition, we
assumed that the effective service coverage radius of each 5G BS was
200 meters (Palizban et al., 2017). The service coverage and the optimal
BS deployment solutions that we obtained are shown in Figs. 5 and 6,
respectively.
As Fig. 5 illustrates, when 50 BSs are deployed in the study area, the
maximum service coverage of the BSs is only 75.06%. As the number of
BSs increases, the service coverage also increases. When p increases to
160, 95.88% of the demand points are covered by LOS services, and the
density of BSs in the study area is approximately 45 BSs/km2, which is
Q. Wang et al.
Fig. 4. Map of the study area.
consistent with the estimation of Ge et al. (40-50 BSs/km2) (Ge et al.,
2016). According to public reports, the service coverage target of most
telecom operators in China is approximately 95%. Therefore, the in­
vestment in 5G BS deployment can end when the service coverage in the
area reaches 95%. Moreover, when p > 160, increasing the number of
BSs does not significantly improve the service coverage of 5G cellular
networks. For example, when p increases from 160 to 200, the effective
6
Sustainable Cities and Society 63 (2020) 102445
service coverage increases from 95.88% to 97.99%. Therefore, although
the number of BSs increases by 40 (accounting for 25% of the BSs), the
effective coverage in the study area increases by only 2.11%.
Fig. 6 presents the service coverage and the simulated signal strength
of each optimal solution in a spatially explicit way. As shown in Fig. 6,
when p = 50, a large area cannot receive LOS 5G signals due to building
blockages. As the number of BSs deployed in the area increases, the
Q. Wang et al.
Table 1
Attributes of the data used for 5G BS optimization in the case study.
Data Source Format Geometry Spatial reference
Roads Open Street
Map
Line
Projection: Gauss-
ESRI File
Buildings AliMap Polygon Kruger Datum:
Geodatabase
Existing Telecom
Point
WGS1984
4G BSs Operator
Fig. 5. Optimal service coverage for different numbers of BSs in the study area.
number of dead signal areas in the study area decreases. Another
interesting result is that as p increases, more BSs are deployed in the
southern and northwestern parts of the study area where buildings are
densely distributed. In other words, more BSs are required in areas with
densely distributed buildings to improve service coverage. Fig. 6 also
illustrates that GIS is an intuitive and effective tool for simulating the
effective service coverage of 5G cellular networks. Nevertheless, in this
study, we focus on the impact of the building distribution on the opti­
mization of 5G BSs. Hence, the signal strength simulated in Fig. 6 is
based on the assumption that the loss of 5G signals in the atmosphere is
linearly related to the propagation distance.
It is extremely difficult to effectively transmit LOS 5G signals to all
demand points or areas due to building blockages in urban areas. The
points or areas distributed near some narrow lanes and corridors and
corners of buildings have difficulty receiving LOS 5G signals. This sit­
uation may occur in some areas with densely distributed buildings. For
instance, although there are many BSs near points 2, 3, 4, 5, 6, 9 and 10
in Fig. 7, these points (or areas) cannot receive effective LOS signals due
to the building blockages around them. Fig. 7 also demonstrates that it is
difficult to receive effective signals in areas completely surrounded by
buildings (e.g., points 1, 7 and 8). Therefore, the deployment of BSs in
these areas needs to be assessed from the perspectives of cost and
demand.
5.3. Impact of LOS coverage on 5G BS deployment
For many common materials, the penetration loss of mmWaves is so
severe that previous studies emphasized the impact of the LOS coverage
of 5G services (Palizban et al., 2017; Wang et al., 2014; Wei, Hu, Qian, &
Wu, 2014). We designed a comparative experiment (denoted as E2) to
further assess the necessity of considering the influence of buildings on
5G BS deployment. The only difference between experiments E1 and E2
is that experiment E2 does not consider the LOS effect in the propagation
of 5G signals. Therefore, as long as a base station can be found within
200 meters of a demand point, the demand point is considered to be
covered by services. We started experiment E2 with p = 29, and the
value was increased in increments of 1. The mutation rate used in E2 was
also set to 1.0/p. The other data and parameters used in E2 were the
same as those in E1. The service coverage of the optimal solutions ob­
tained by both E1 and E2 are shown in Fig. 8. In addition, the optimal
solutions that we obtained in E2 were assessed with an approach that
considers the LOS effect to analyze the impact of this effect on 5G BS
deployment.
7
Sustainable Cities and Society 63 (2020) 102445
As shown in Fig. 8, we need only 48 BSs to achieve seamless service
coverage in the study area in the scenario that does not consider the LOS
effect. If the service coverage target is set to 95%, then only 33 BSs are
needed. On the other hand, when the LOS coverage model is applied to
the E2 solutions, the actual service coverage of the optimal solutions
obtained in E2 significantly decreases. For example, in the case of
considering the LOS effect, when p = 50, the service coverage of the
optimal BS deployment solution obtained by E2 is only 66.08%. As Fig. 5
shows, when p = 50, we can obtain an optimal solution with a service
coverage of approximately 75.06%. Overall, Fig. 8 indicates that the
LOS effect caused by building blockages in urban areas has a significant
impact on the deployment of 5G BSs. The service coverage and simu­
lated signal strength of some optimal solutions obtained in experiment
E2 are shown in Fig. 9.
As shown in Fig. 9, by assuming that 5G signals can penetrate
buildings or that there are no buildings in the study area, we can achieve
seamless service coverage with a small number of BSs. However, the
reality is that it is difficult for mmWaves to penetrate buildings. Due to
the blockage effect of buildings on mmWaves, most of the dead signal
zones are distributed in areas with densely distributed buildings.
Clearly, it is important to consider the LOS propagation effect of
mmWaves in the spatial planning of 5G BSs, especially in urban areas
with high building densities.
5.4. Considering the existing 4G BSs
As we introduced in Section 2, most of the first 5G cellular networks
being rolled out are based on NSA architecture. Hence, we designed
another decision-making scenario in which the telecom operator con­
structs a 5G cellular network in the study area based on NSA networking.
There are 32 existing 4G BSs in the study area (see Fig. 4). To achieve
satisfactory service coverage, operators must deploy additional new 5G
BSs. Because the service coverage of the existing 4G BSs is fixed, we
exclude points that are covered by these BSs from the demand point set.
Then, the locations for newly deployed 5G BSs to cover the remaining
demand points in the study area are optimized using the model devel­
oped in this study.
In this scenario, when the number of newly deployed 5G BSs is 140,
the service coverage of the entire study area reaches 95.47%. Thus, the
satisfactory service coverage goal for the entire study area is achieved
using 172 BSs, including 140 new BSs and 32 existing BSs (Fig. 10). In
comparison, we used 160 BSs to cover 95.88% of the demand points in
the entire study area in experiment E1. Because the existing 4G BSs are
deployed without considering the propagation characteristics of 5G
signals, compared with standalone (SA) 5G networking solutions, tele­
com operators might need more BSs to cover the same area using NSA
networking solutions. Nevertheless, the number of newly deployed 5G
BSs required to cover the same demand area is indeed reduced using
NSA networking technology. Therefore, NSA networking technology is
still attractive for telecom operators to quickly roll out 5G services,
especially in urban areas where 4G BSs have already been densely
deployed.
5.5. Sensitivity analysis and performance evaluation
The discretization of the continuous space and the determination of
candidate locations for BSs are the two most important uncertainties in
our model. To evaluate the robustness and performance of the model, we
conducted three comparisons. The first analysis examined the effects of
different sampling intervals in spatial discretization on the optimization
results (denoted as E3). The second analysis was performed to determine
whether using prior knowledge to select candidate locations for BSs can
help improve the quality of the optimal solutions (denoted as E4). The
third analysis aimed to evaluate the performance of the AIS algorithm in
global optimization.
Q. Wang et al.
Fig. 6. Optimal 5G BS deployment solutions for different numbers of BSs.
5.6. Discretization of the continuous space: the sampling interval
In experiment E1, the sampling interval (d) we used to discretize the
study area was 25 m. To perform the sensitivity analysis, we discretized
the study area with sampling intervals of 20 meters, 15 meters, and 10
meters. As a result, we obtained three sets of points with different spatial
resolutions to model the same service demand and candidate locations
for BSs in the study area. These points were also taken as the candidate
locations for BSs. The data and parameters used in Experiment E3 were
the same as those used in E1. The results of these experiments are shown
in Fig. 11.
As shown in Fig. 11, in most cases, as the sampling interval de­
creases, the maximum service coverage that we obtained for the same
number of BSs slightly decreases (no more than 1.0%). For example,
when p = 200 and d = 10 m, the maximum proportion of demand points
covered by LOS 5G signals is approximately 97.25%, which is 0.74% less
than that when p = 200 and d = 25 m. Notably, as the sampling interval
increases, the number of points distributed in narrow corridors and
8
Sustainable Cities and Society 63 (2020) 102445
surrounded by buildings increases in the discretized point set. The
optimization results obtained with a small sampling distance are more
accurate than those obtained with a large sampling distance; however,
more computational resources (or time) are required to search for the
optimal results. Overall, the influence of the sampling interval used in
spatial discretization on the optimization result is not significant.
5.7. Candidate locations of BSs: systematic sampling or prior knowledge
determination
Most human activities occur near or in buildings. Fig. 12 shows the
population density of the study area, which was estimated using the
location big data set from Tencent (Yao et al., 2017). Buildings also have
a considerable influence on the propagation of 5G signals. Therefore,
wall-mounted 5G BSs can effectively improve the service coverage near
and inside buildings (Palizban et al., 2017). On the other hand, many
human outdoor activities are clustered on roads or near roads. In addi­
tion, BSs deployed along roads are vital for some 5G applications, such
Q. Wang et al.
Fig. 7. Demand points where LOS signals are difficult to receive.
Fig. 8. Influence of the LOS effect on the service coverage of 5G BSs.
as self-driving cars. Hence, 5G BSs can also be deployed along roads
(Sakaguchi et al., 2017). This prior knowledge can be used to guide the
deployment of 5G BSs. Consequently, it is necessary to analyze the effect
of using prior knowledge to select candidate locations on the optimi­
zation results.
Based on the approach above, the steps involved in using prior
knowledge to select the candidate locations of BSs are as follows: (1) the
inflection points on building outlines (e.g., corners) are selected as
candidate locations for BSs. (2) Sampling is performed at 20-m intervals
along roads to obtain candidate locations. (3) Road intersections, which
usually have high LOS values, are also selected as candidate locations for
5G BSs. (4) For areas without buildings and roads, the same systematic
sampling approach is used to obtain candidate locations. The sampling
intervals used in these areas range from 25 m to 50 m. Sampling distance
d was set to 50 m in sparsely populated areas and 25 m in densely
populated areas. The points obtained in steps 1-4 were merged; thus, we
obtained a candidate location set with 16,623 points (see Fig. 12). Then,
these points were used to replace the candidate locations determined
using systematic sampling in experiment E1. The other data and pa­
rameters used in experiment E4 were the same as those used in E1. The
service coverage of the optimal solutions obtained in E4 is shown in
Fig. 13.
Surprisingly, the solutions that we obtained from the prior-
knowledge-selected candidate locations did not display better perfor­
mance than those determined using the systematic sampling approach.
9
Sustainable Cities and Society 63 (2020) 102445
In some cases, the performance of the prior knowledge method was even
worse than that of the systematic sampling approach. For example,
when p = 50, p = 70, p = 80, p = 120 and p = 180, the service coverage
level of the optimization results that we obtained using the prior-
knowledge-selected candidate locations was slightly lower than that of
the results obtained from the systematic sampling approach (see
Fig. 13). Therefore, we conclude that using prior knowledge to select
candidates for the locations of 5G BSs is not necessary for the proposed
model. Overall, the location optimization model for 5G BS planning that
we developed in this study is robust and reliable.
5.8. Performance analysis of the optimization algorithm: the AIS
algorithm vs. GA
We choose a GA as the control algorithm to evaluate the global
optimization ability of the AIS algorithm. GAs are well-known and
widely used heuristic algorithms, and they have been considered an
effective tool for solving MCLPs (Tong, Murray, & Xiao, 2009). Unlike
the AIS algorithm, GAs use crossover and mutation operations to
conduct local searches.
To compare the performance of the AIS algorithm and the GA, the
mutation operation used in the GA is the same as that used in the AIS
algorithm that we developed in this study. The crossover operation in
the GA that we used in this study is an improved single-point crossover
algorithm. More details about the GA and crossover algorithm can be
found in the study by Tong et al. (2009)). The parameter settings of the
GA and the AIS algorithm are presented in Table 2. Because the GA and
the AIS algorithm use different search strategies, the computation time
for each iteration of the two algorithms is also very different. To
compare the performance of the two algorithms in a more reasonable
manner, we let them run for the same computation time. The conver­
gence curves of the two algorithms are shown in Fig. 14.
As shown in Fig. 14, after running for 3958 seconds, both the AIS
algorithm and GA converged. The service coverage of the best solution
that the AIS algorithm obtained is 95.88%, while that obtained by the
GA is only 93.36%. The convergence curves also indicate that the AIS
algorithm shows better performance than the GA within the same
computation time. Therefore, we conclude that the optimization model
that we developed in this study has good performance in global
Q. Wang et al. Sustainable Cities and Society 63 (2020) 102445

Fig. 9. Service coverage and simulated signal strength of the optimal solutions obtained in experiment E2.

10
Q. Wang et al.
Fig. 10. Optimal solution for 5G network deployment considering the existing
4G BSs.
optimization.
6. Conclusions
The development of 5G technology is critical to many emerging
technologies. 5G technology uses mmWaves to achieve high-speed, low-
latency and large-capacity wireless communication. However, the high
propagation and penetration loss of mmWaves make the effective
coverage of 5G BSs extremely limited. It is difficult for mmWaves to
penetrate buildings in urban areas; thus, more BSs must be deployed in
areas with densely distributed buildings to achieve satisfactory service
coverage. The ultra-dense deployment of 5G BSs in urban outdoor areas
requires considerable investments and will greatly increase energy
consumption. Determining how to optimize the locations of BSs ac­
cording to the distribution of buildings in urban areas to achieve satis­
factory coverage with the minimum number of BSs will play an essential
role in the rollout of 5G technology.
In this study, we developed a GIS-based optimization model to sup­
port 5G cellular network planning in urban outdoor areas. First, we
Fig. 11. Effects of different sampling intervals on the service coverage of the optimization results.
11
Sustainable Cities and Society 63 (2020) 102445
employed GIS to simulate the LOS propagation of 5G signals in urban
outdoor areas in a spatially explicit way. Second, the optimization
objective of the ultra-dense 5G BSs in urban outdoor areas was formu­
lated based on the MCLP model. Third, the spatial analysis technologies
provided by GIS were combined with an AIS algorithm to search for
optimal 5G BS deployment solutions. Overall, one of the most important
contributions of our model is that the blockage effect of buildings on
mmWave propagation was simulated using a spatially explicit LOS
coverage model with the support of GIS and was integrated into the
heuristic optimization model.
The optimization results that we obtained with the proposed model
in the case study demonstrate that the model provides valuable solutions
to support ultra-dense 5G BS planning in urban outdoor areas. The
experimental results indicate that to achieve service coverage greater
than 95% in the study area, the density of BSs in this area cannot be
lower than 45 BSs/km2, which is consistent with the estimates of Ge
et al. (2016) (40-50 BSs/km2) and Palizban et al. (2017) (40-60
BSs/km2). The case study also demonstrates that areas with high
building densities require more BSs to achieve satisfactory service
coverage. We conclude that the density of BSs required is positively
related to the building density of the area. In addition, although the
spatial distribution of the signal strength of each 5G BS deployment
solution was simulated based on the assumption that the loss of
mmWaves is linearly related to the propagation distance, the results
suggest that GIS is a useful and essential tool for 5G cellular network
planning.
To analyze the impact of the LOS effect on 5G BS planning, a
comparative experiment was conducted. The results show that it is
essential to consider the LOS effect in location optimization models. As
shown in Figs. 8 and 9, ignoring the LOS propagation effect of mmWave
in urban areas might lead to dead signal zones. Additionally, due to the
effects of building blockages on mmWaves, the number of BSs required
to achieve satisfactory service coverage was approximately 5 times that
when the LOS effect was not considered in the study area.
The modeling of the service demand in the study area and the
determination of candidate locations for 5G BSs are the two main un­
certainties in our model. Hence, two additional experiments were con­
ducted to perform a sensitivity analysis of the model. The results of the
sensitivity analysis suggest that the sampling interval used to discretize
the continuous demand space does have an impact on the quality of the
optimization results; however, this effect is not significant. Additionally,
the candidate locations of the BSs in our model were determined using a
systematic sampling approach. Hence, we determined whether prior
knowledge, which has been widely used in 5G cellular network planning
Q. Wang et al. Sustainable Cities and Society 63 (2020) 102445

Fig. 12. Candidate locations for 5G BSs selected using prior knowledge.

Fig. 13. Impact of different methods for candidate location selection on the service coverage of the optimization results.

in previous studies, can help improve the modeling results. However, the
results shown in Fig. 13 suggest that including prior knowledge did not
improve the modeling results. In addition, the comparison between the
AIS algorithm and the GA shows that our model has good performance in
global optimization. Overall, the results of the sensitivity analysis and
performance evaluation indicate that the optimization model that we
developed in this study is a useful tool for generating alternatives for 5G
cellular network planning in urban outdoor areas.
Nevertheless, some limitations still exist in this study. Future works
can focus on integrating a mmWave propagation path loss model
(Maccartney et al., 2013). This study simulated the LOS coverage of 5G
BSs based on the assumption that the loss of mmWaves is linearly related
to the propagation distance. Therefore, to accurately simulate the real
strength of 5G signals and to improve the optimization results, it is
necessary to integrate a mmWave propagation model into the optimi­
zation model. Determining whether there is a correlation between
building density and the BS density required to achieve satisfactory
service coverage is another interesting problem that requires further

12
Q. Wang et al. Sustainable Cities and Society 63 (2020) 102445

Table 2 References
Parameter settings of the GA and the AIS algorithm.
Adam, I. (2019). Future of AI & 5G IV: Driving Cleaner Economic Growth & Employment.
Parameter AIS GA
Retrieved December 20, 2019, from Linkedin website: https://www.linkedin.com
p 160 160 /pulse/future-ai-5g-iv-driving-cleaner-economic-growth-employment-adam/.
Population size 100 100 Al-Dabbagh, R. K., Al-Aboody, N. A., & Al-Raweshidy, H. S. (2017). A simplified path loss
Mutation rate 1.0/160 0.1 model for investigating diffraction and specular reflection impact on millimetre
wave propagation. In Proceedings of the 2017 8th International Conference on the
Crossover rate — 0.8
Network of the Future, NOF 2017 (pp. 153–155). https://doi.org/10.1109/
Clone rate 4.0 —
NOF.2017.8251241. Retrieved from.
Selection rate 0.15 — Al-Falahy, N., & Alani, O. (2019). Coverage and capacity improvement of millimetre
Computation time 3958s 3958s wave 5G network using distributed base station architecture. IET Networks, 8(4),
Number of iterations 500 4798 246–255. https://doi.org/10.1049/iet-net.2018.5059.
Bai, T., & Heath, R. W. (2015). Coverage and rate analysis for millimeter-wave cellular
networks. IEEE Transactions on Wireless Communications, 14(2), 1100–1114. https://
doi.org/10.1109/TWC.2014.2364267.
Buzzi, S., I, C.-L., Klein, T. E., Poor, H. V., Yang, C., & Zappone, A. (2016). A survey of
energy-efficient techniques for 5G networks and challenges ahead. IEEE Journal on
Selected Areas in Communications, 34(4), 697–709. https://doi.org/10.1109/
JSAC.2016.2550338.
Chen, Z., Yang, X., & Su, Z. (2014). Multi-target sites selection method based on genetic
algorithmin TD-LTE system. Science Technology and Engineering, 14(7), 29-33+44.
Church, R. L., Stoms, D. M., & Davis, F. W. (1996). Reserve selection as a maximal
covering location problem. Biological Conservation, 76(2), 105–112.
Church, R., & ReVelle, C. R. (1974). The maximal covering location problem. Papers in
Regional Science, 32(1), 101–118.
Cicek, C. T., Gultekin, H., Tavli, B., & Yanikomeroglu, H. (2019). UAV base station
location optimization for next generation wireless networks: Overview and future
research directions. 2019 1st International Conference on Unmanned Vehicle Systems-
Oman (UVS).
Fig. 14. Convergence curves of the GA and the AIS algorithm in the compar­ de Castro, L. N., & Timmis, J. I. (2003). Artificial immune systems as a novel soft
ative experiment. computing paradigm. Soft Computing, 7(8), 526–544. https://doi.org/10.1007/
S00500-002-0237-z.
de Castro, L. N., & Von Zuben, F. J. (2002). Learning and optimization using the clonal
study. We concluded that the density of BSs required to be deployed is selection principle. IEEE Transactions on Evolutionary Computation, 6(3), 239–251.
positively related to the building density in the study area. In future https://doi.org/10.1109/tevc.2002.1011539.
Deb, K., Pratap, A., Agarwal, S., & Meyarivan, T. (2002). A fast and elitist multiobjective
research, we will try to build a quantitative model to describe the genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation, 6(2),
relationship among building density, the shape complexity of buildings, 182–197. https://doi.org/10.1109/4235.996017.
and the BS density required to achieve satisfactory service coverage. Ekstrom, H. (2019). Non-standalone and Standalone: two standards-based paths to 5G.
Retrieved May 31, 2020, from https://www.ericsson.com/en/blog/2019/7/standal
Thus, we can provide more useful information to support 5G cellular one-and-non-standalone-5g-nr-two-5g-tracks.
network planning with this approach. For example, telecom operators Eldrandaly, K., Abdel-Basset, M., & Abdel-Fatah, L. (2019). Grid quorum-based spatial
can accurately estimate the minimum number of BSs required to cover coverage in mobile wireless sensor networks using nature-inspired firefly algorithm.
Expert Systems, 36(4), 1–18. https://doi.org/10.1111/exsy.12421.
an area based on the density and shape of the buildings in the area. The Fan, C., Diao, Z., Liu, W., & Wang, Z. (2019). Challenges and breakthroughs in China’s
cost of BS deployment and the energy consumed by the 5G cellular current 5G networks construction. Communications World, 31(11), 33–35.
network in the area can also be accurately estimated. Galiotto, C., Pratas, N. K., Doyle, L., & Marchetti, N. (2017). Effect of LOS/NLOS
propagation on 5G ultra-dense networks. Computer Networks, 120, 126–140. https://
doi.org/10.1016/j.comnet.2017.04.012. Retrieved from.
Declaration of interests Ge, X., Tu, S., Mao, G., Wang, C.-X., & Han, T. (2016). 5G ultra-dense cellular networks.
IEEE Wireless Communications, 23(1), 72–79.
Guo, T., & Mu, C. (2003). Radio network design using adaptive-migration parallel
The authors declare that they have no known competing financial
genetic algorithms. Systems Engineering-Theory & Practice, 2, 57-61+110.
interests or personal relationships that could have appeared to influence Huang, K. N., Liu, X. P., Li, X., Liang, J. Y., & He, S. J. (2013). An improved artificial
the work reported in this paper. immune system for seeking the Pareto front of land-use allocation problem in large
areas. International Journal of Geographical Information Science, 27(5), 922–946.
https://doi.org/10.1080/13658816.2012.730147.
Submission declaration and verification I, C.-L., Rowell, C., Han, S., Xu, Z., Li, G., & Pan, Z. (2014). Toward green and soft: A 5G
perspective. IEEE Communications Magazine, 52(2), 66–73.
Submission of an article implies that the work described has not been Ismail, W. Z. W., & Abd Manaf, S. (2010). Study on Coverage in Wireless Sensor Network
using Grid Based Strategy and Particle Swarm Optimization. In F. Z. Rohani,
published previously, that it is not under consideration for publication S. A. Ahmad, A. C. Soh, H. Jaafar, M. H. M. Zaini, M. H. M. Sidek, & M. Shaufiq
elsewhere, that its publication is approved by all authors and tacitly or (Eds.), Proceedings of the 2010 IEEE Asia Pacific Conference on Circuit And System
explicitly by the responsible authorities where the work was carried out, (APCCAS) (pp. 1175–1178).
Jimenez, J. G., Chiaraviglio, L., Amorosi, L., & Blefari-Melazzi, N. (2018). Multi-period
and that, if accepted, it will not be published elsewhere in the same mission planning of UAVs for 5G coverage in rural areas: A heuristic approach.
form, in English or in any other language, including electronically Proceedings of the 2018 9th International Conference on the Network of the Future, NOF
without the written consent of the copyright-holder. 2018, 52–59. https://doi.org/10.1109/NOF.2018.8598123. Retrieved from.
Jingyi, C. (2019). China to have full 5G network coverage within 7 years. December 18.
Retrieved from. Global Times https://www.globaltimes.cn/content/1173988.shtml.
Acknowledgements Khan, F., Pi, Z., & Rajagopal, S. (2012). Millimeter-wave mobile broadband with large
scale spatial processing for 5G mobile communication. 2012 50th Annual Allerton
Conference on Communication, Control, and Computing, Allerton 2012, 1517–1523.
This research was supported by the National Natural Science Foun­
https://doi.org/10.1109/Allerton.2012.6483399.
dation of China (Grant No. 41971336), the Open Fund of the Key Lab­ Krishnamachari, B., & Wicker, S. (2004). In Lipo Wang (Ed.), Base Station Location
oratory of Urban Land Resources Monitoring and Simulation, Ministry of Optimization in Cellular Wireless Networks using Heuristic Search Algorithms BT - Soft
Natural Resources (Grant No. KF-2018-03-033) and the National key Computing in Communications. https://doi.org/10.1007/978-3-540-45090-0_10.
Lakshminarasimman, N., Baskar, S., Alphones, A., & Iruthayarajan, M. W. (2011).
research and development program (Grant No. 2018YFD1100801). Evolutionary multiobjective optimization of cellular base station locations using
modified NSGA-II. Wireless Networks, 17(3), 597–609. https://doi.org/10.1007/
s11276-010-0299-2.
Lee, J., Tejedor, E., Ranta-Aho, K., Wang, H., Lee, K.-T., Semaan, E., … Jung, S. (2018).
Spectrum for 5G: Global status, challenges, and enabling technologies. IEEE
Communications Magazine, 56(3), 12–18. https://doi.org/10.1109/
MCOM.2018.1700818. Retrieved from.

13
Q. Wang et al.
Li, S., Shi, L.-Y., Wu, L.-S., Liu, Y., & Zhang, Y. P. (2018). 45-Degree polarized microstrip
grid arrays for millimeter-wave micro base station. 2018 Joint IEEE International
Symposium on Electromagnetic Compatibility and 2018 IEEE Asia-Pacific Symposium on
Electromagnetic Compatibility (EMC/APEMC), 1045–1049.
Lu, M. J., Hsu, S. C., Chen, P. C., & Lee, W. Y. (2018). Improving the sustainability of
integrated transportation system with bike-sharing: A spatial agent-based approach.
Sustainable Cities and Society, 41, 44–51. https://doi.org/10.1016/j.scs.2018.05.023.
Lyu, J., Zeng, Y., Zhang, R., & Lim, T. J. (2017). Placement optimization of UAV-mounted
mobile base stations. IEEE Communications Letters, 21(3), 604–607. https://doi.org/
10.1109/LCOMM.2016.2633248.
Ma, X. Y., & Zhao, X. (2015). Land use allocation based on a multi-objective artificial
immune optimization model: An application in Anlu County, China. Sustainability, 7
(11), 15632–15651. https://doi.org/10.3390/su71115632.
Maccartney, G. R., Zhang, J., Nie, S., & Rappaport, T. S. (2013). Path loss models for 5G
millimeter wave propagation channels in urban microcells. GLOBECOM - IEEE Global
Telecommunications Conference, 3948–3953. https://doi.org/10.1109/
GLOCOM.2013.6831690. Retrieved from.
Palizban, N., Szyszkowicz, S., & Yanikomeroglu, H. (2017). Automation of millimeter
wave network planning for outdoor coverage in dense urban areas using wall-
mounted base stations. IEEE Wireless Communications Letters, 6(2), 206–209. https://
doi.org/10.1109/LWC.2017.2659732.
Pi, Z., & Khan, F. (2011). An introduction to millimeter-wave mobile broadband systems.
IEEE Communications Magazine, 49(6), 101–107. https://doi.org/10.1109/
MCOM.2011.5783993.
Qamar, F., Hindia, N., Dimyati, K., Noordin, K. A., Majed, M. B., Rahman, T. A., …
Amiri, I. S. (2019). Investigation of future 5g-iot millimeter-wave network
performance at 38 GHz for urban microcell outdoor environment. Electronics
(Switzerland), 8(5). https://doi.org/10.3390/electronics8050495.
Qin, H., Guan, L., & Xie, S. (2004). An improved genetic algorithm for transceiver
placement problem in radio network design. Computer Engineering and Applications,
15, 72-73+191.
Rabbani, M., Heidari, R., Farrokhi-Asl, H., & Rahimi, N. (2018). Using metaheuristic
algorithms to solve a multi-objective industrial hazardous waste location-routing
problem considering incompatible waste types. Journal of Cleaner Production, 170,
227–241. https://doi.org/10.1016/j.jclepro.2017.09.029.
Rappaport, T. S., Gutierrez, F., Jr., Ben-Dor, E., Murdock, J. N., Qiao, Y., & Tamir, J. I.
(2013). Broadband millimeter-wave propagation measurements and models using
adaptive-beam antennas for outdoor urban cellular communications. IEEE
Transactions on Antennas and Propagation, 61(4, 1, SI), 1850–1859. https://doi.org/
10.1109/TAP.2012.2235056.
Rappaport, T. S., Xing, Y., MacCartney, G. R., Jr., Molisch, A. F., Mellios, E., & Zhang, J.
(2017). Overview of millimeter wave communications for fifth-generation (5G)
wireless networks-with a focus on propagation models. IEEE Transactions on Antennas
and Propagation, 65(12, 1), 6213–6230. https://doi.org/10.1109/
TAP.2017.2734243.
Sakaguchi, K., Haustein, T., Barbarossa, S., Strinati, E. C., Clemente, A., Destino, G., …
Heath, R. W. (2017). Where, when, and how mmwave is used in 5G and beyond.
IEICE Transactions on Electronics, E100.C(10), 790–808. https://doi.org/10.1587/
transele.E100.C.790.
Salah, F., & Rinne, M. (2018). Performance Analysis of User Plane Connectivity in the 5G
Non-Standalone Deployment. 2018 IEEE Global Communications Conference
(GLOBECOM), 1–6.
Seyhan, T. H., Snyder, V. L., & Zhang, Y. (2018). A new heuristic formulation for a
competitive maximal covering location problem. Transportation Science, 52(5),
1156–1173. https://doi.org/10.1287/trsc.2017.0769.
14
Sustainable Cities and Society 63 (2020) 102445
Shang, R. H., Jiao, L. C., Liu, F., & Ma, W. P. (2012). A novel immune clonal algorithm for
MO problems. IEEE Transactions on Evolutionary Computation, 16(1), 35–50. https://
doi.org/10.1109/tevc.2010.2046328.
Sharma, S. K., Bogale, T. E., Le, L. B., Chatzinotas, S., Wang, X., & Ottersten, B. (2018).
Dynamic spectrum sharing in 5G wireless networks with full-duplex technology:
Recent advances and research challenges. IEEE Communications Surveys and Tutorials,
20(1), 674–707. https://doi.org/10.1109/COMST.2017.2773628.
Sulyman, A., Nassar, A., Samimi, M., Maccartney, G., Rappaport, T., & Alsanie, A.
(2014). Radio propagation path loss models for 5G cellular networks in the 28 GHZ
and 38 GHZ millimeter-wave bands. IEEE Communications Magazine, 52(9), 78–86.
https://doi.org/10.1109/MCOM.2014.6894456. Retrieved from.
Szyszkowicz, S. S., Lou, A., & Yanikomeroglu, H. (2016). Automated placement of
individual millimeter-wave wall-mounted base stations for line-of-sight coverage of
outdoor urban areas. IEEE Wireless Communications Letters, 5(3), 316–319. https://
doi.org/10.1109/LWC.2016.2552168.
Tian, G., Li, J., Hou, F., Zhang, W., Guo, X., Cao, L., … Wan, L. (2018). Design and
implementation of a compact 3-D stacked RF front-end module for micro base
station. IEEE Transactions on Components Packaging and Manufacturing Technology, 8
(11), 1967–1978. https://doi.org/10.1109/TCPMT.2018.2842084.
Tong, D., Murray, A., & Xiao, N. (2009). Heuristics in spatial analysis: A genetic
algorithm for coverage maximization. Annals of the Association of American
Geographers, 99(4), 698–711. https://doi.org/10.1080/00045600903120594.
Usama, M., & Erol-Kantarci, M. (2019). A Survey on recent trends and open issues in
energy efficiency of 5G. Sensors, 19(14). https://doi.org/10.3390/s19143126.
Wang, C. X., Haider, F., Gao, X., You, X. H., Yang, Y., Yuan, D., … Hepsaydir, E. (2014).
Cellular architecture and key technologies for 5G wireless communication networks.
IEEE Communications Magazine, 52(2), 122–130. https://doi.org/10.1109/
MCOM.2014.6736752.
Wang, L., & Shen, J. F. (2017). The challenge of spatial plan coordination in urban China:
The case of Suzhou City. Urban Policy and Research, 35(2), 180–198. https://doi.org/
10.1080/08111146.2016.1159553.
Wei, L., Hu, R., Qian, Y., & Wu, G. (2014). Key elements to enable millimeter wave
communications for 5G wireless systems. IEEE Wireless Communications, 21(6),
136–143. https://doi.org/10.1109/MWC.2014.7000981.
Yao, J., Zhang, X., & Murray, A. T. (2019). Location optimization of urban fire stations:
Access and service coverage. Computers, Environment and Urban Systems, 73
(November 2017), 184–190. https://doi.org/10.1016/j.
compenvurbsys.2018.10.006.
Yao, Y., Liu, X., Li, X., Zhang, J., Liang, Z., Mai, K., … Zhang, Y. (2017). Mapping fine-
scale population distributions at the building level by integrating multisource
geospatial big data. International Journal of Geographical Information Science, 31(6),
1220–1244. https://doi.org/10.1080/13658816.2017.1290252.
Yaovaja, K., & Klunngien, J. (2019). Teleoperation of an industrial robot using a non-
standalone 5G mobile network. 2019 IEEE Eurasia Conference on IOT, Communication
and Engineering (ECICE), 320–323.
Yildirim, K. S., Kalayci, T. E., & Ugur, A. (2008). Optimizing coverage in a k-covered and
connected sensor network using genetic algorithms. In D. P. Dimitrov, V. Mladenov,
S. Jordanova, & N. Mastorakis (Eds.), Advanced topics on evolutionary computing (pp.
21–26).
Zhang, Z., & Hu, R. Q. (2018). Dense cellular network analysis with LoS/NLoS
propagation and bounded path loss model. IEEE Communications Letters, 22(11),
2386–2389. https://doi.org/10.1109/LCOMM.2018.2850815. Retrieved from.
Zhao, X., Ma, X., Tang, W., & Liu, D. (2019). An adaptive agent-based optimization
model for spatial planning: A case study of Anyue County, China. Sustainable Cities
and Society, 51, Article 101733. https://doi.org/10.1016/j.scs.2019.101733.