5G MMWAVE SMALL CELL NETWORKS: ARCHITECTURE, SELF ORGANIZATION, AND MANAGEMENT

Capacity Enhancement for 5G Networks

Using MmWave Aerial Base Stations:
Self-Organizing Architecture and Approach
Peng Yu, Wenjing Li, Fanqin Zhou, Lei Feng, Mengjun Yin, Shaoyong Guo, Zhipeng Gao, and Xuesong Qiu

Abstract fore, it is capable of improving the mmWave chan-

nel conditions and achieving larger capacity [3, 4].
In 5G networks, an mAeBS can overcome As a result, mAeBS becomes a viable CE solution in
on-ground constraints and enable rapid and flexi- 5G networks.
ble deployment, providing an ideal solution for CE The mAeBS refers to an AeBS with 5G
in hotspot areas. However, how to control multiple mmWave communication technologies, which
mAeBSs for efficient CE is a challenging issue, and may be an unmanned aerial vehicle (UAV), air-
we refer to it as mACE. To tame the trivial manage- ship, or drone. Since a large number of human
ment practices involved in mACE, we resort to the activities occur within 200 m above the ground,
SON methodology and propose a self-organizing wireless channel characteristics in this range are
architecture. A closed-loop management process is strongly related to the environment of ground sur-
correspondingly designed for it, which consists of face. Thus, the research on mAeBSs has created
four phases: analyzing, planning, devaluating, and much interest in academia and industry. The feasi-
executing. Thereafter, solutions to the key technical bility of introducing the mmWave communication
issues, including hotspot identification and mAeBS system in AeBSs is pointed out in [5], and some
deployment, are investigated. The effects of the technical problems in deploying an mmWave base
overall mACE approach are intuitively illustrated in station on a high-speed unmanned aerial vehicle
a small-scale example case. A few more challeng- to serve ground users are analyzed in [6], and the
ing issues related to mACE are also identified for solutions are put forward as well. Xu et al. [7] ana-
future research. lyzed the air-to-ground propagation characteristics
of mmWave signals, and a method for precoding
Introduction the air-to-ground beam signal assisted by the user
According to the latest Cisco forecast [1], by location was proposed. However, few studies have
2020, mobile data will be 10 times more than in been found focusing on the management architec-
2014. Due to the increased bandwidth demand of ture and procedures for mAeBS application sce-
mobile users, the aggregation of a small number narios.
of high-bandwidth users would put a large amount Although exact mAeBS research is barely
of traffic load in cellular networks, taking up a lot observable in released industrial projects, some
of resources and affecting the admission and ser- AeBS-related projects have really drawn much
vice of other users. Moreover, the limited capaci- attention from the public, academia, and indus-
ty of traditional cellular cells makes it hard for the try. The most famous AeBS project is the well-
redundant traffic load to be mitigated via conven- known LOON by Alphabet using high-altitude
tional approaches, such as mobility load balanc- balloons. Similar actions are also observed in other
ing. This requires fifth generation (5G) networks to well-known companies such as Facebook, which
be capable of performing capacity enhancement launched the internet.org program for Internet
(CE) flexibly and rapidly to address urgent service access in remote rural areas. From the standards
need in traffic bursting areas correspondingly. Hav- perspective, New Radio (NR) Non-Standalone
ing strong maneuverability, the aerial base station (NSA) 5G specifications (Release 15) were official-
(AeBS) provides an efficient way to achieve this ly approved at Third Generation Partnership Proj-
goal [2]. Millimeter-wave (mmWave) communi- ect (3GPP RAN#78) by the end of 2017, which
cation boasts abundant spectrum resources and introduced advanced technologies in 5G NR com-
high directional gain, making it ideal for providing patible with mmWave. Considering that the appli-
wireless access to bandwidth-demanding users. cations of AeBS are in a burgeoning stage, these
The mmWave aerial base station (mAeBS), formed developments also facilitate the application of 5G
by combining AeBS with 5G mmWave technolo- technologies in the mAeBS.
gy, will possess the advantage of sufficient spec- Figure 1 presents an example scenario where
trum resources in mmWave band, and at the same CE via deploying mAeBSs is performed. As illus-
time overcome its disadvantage of proneness to trated in the figure, random aggregation of users,
obstruction by flexibly adjusting its position. There- caused by crowd commuting or sudden events,

Digital Object Identifier: The authors are with Beijing University of Posts and Telecommunications (BUPT); Zhipeng Gao is with International School of BUPT;
10.1109/MWC.2018.1700393 Wenjing Li is the corresponding author.

58 1536-1284/18/$25.00 © 2018 IEEE IEEE Wireless Communications • August 2018

Authorized licensed use limited to: Auckland University of Technology. Downloaded on June 06,2020 at 20:27:59 UTC from IEEE Xplore. Restrictions apply.
puts high traffic load in mobile networks, produc-
mAeBS
ing three hotspots after work hours in the urban
area. In the stadium, a hotspot appears because a
large number of users gather and share high-defi-
nition multimedia with friends or family. People
commute from an office area to a residential area
while enjoying high-definition episodes, producing mAeBS
the other two hotspots. The aggregation of those
users will impose high traffic load to specific cellu-
lar cells and generate hotspots. For these scenarios, LPN
we propose to deploy an mAeBS at each traffic
hotspot for flexible network capacity improvement
mAeBS
rather than to deploy denser terrestrial low-power
nodes.
Base station
However, the mAeBS itself cannot deduce
comprehensive information about hotspots in
the networks, and the directional beam transceiv- Coverage of
LPN
er would consume much time and resources to Coverage of
Overload
acquire local user information in its vicinity. In addi- mAeBC
tion, conflicts between mAeBSs should be carefully Non-line-of-sight Coverage of
avoided, positions call for optimization, and radio macro base station
mmW beam link
parameters need adjustment. It raises a challenging
issue to manage and control multiple mAeBSs for
efficient CE. We refer to it as mACE. In the process
of solving mACE, manual processing is obvious-
ly no longer feasible for the rapid response to CE
demands. Luckily, a self-organizing network (SON)
based on the autonomic management paradigm
enables a promising solution.
To reduce the operation and maintenance
expenditure but improve the efficiency, the LTE net-
work introduced the concept of SON, which is to
make a network manage itself on its own through
utilizing self-configuration, self-optimization, and
self-healing, three key functions, and implementing
four closed-loop procedures: analyzing, planning,
evaluating, and executing. To manage the highly FIGURE 1. Illustration of capacity enhancement research scenario of mAeBS
heterogeneous and integrated 5G networks, the deployment.
Horizon2020 project SELFNET (2015-2018) con-
tinues to promote SON research. 3GPP started
SON standardization in Release 8 for LTE, and of mAeBSs, the factors affecting the capacity of
many more use cases have been added in Release mAeBSs should be analyzed, which also facilitates
15 to enhance SON for 5G networks. However, the design of mAeBS 3D deployment strategy.
how to design a reasonable management frame-
work based on SON and devise appropriate opti- SON Architecture and Approaches for Capacity
mization algorithms to form an overall approach Enhancement Using mAeBS
becomes the key for mACE. After presenting the
background, the rest of the article first explains The use case design of SON generally includes
the principles in SON use case design, together the following aspects: determine the manage-
with other key research issues and the state of the ment architecture, define the management pro-
art related to mACE. Then we present our mAeBS cess, identify functions in each stage, and design
deployment approach for capacity enhancement in key self-optimization algorithms. The self-orga-
SON enabled networks, and demonstrate the load nization degree of the mobile communication
alleviation effect and statistical results of user data network varies with operators and specific cases.
rate performance. Finally, some future research In general, a specific SON use case needs to be
directions of mACE are discussed. determined by an operator according to its own
network conditions [8]. At present, the SON man-
Key Research Issues and the State of the Art agement architecture includes three types: cen-
The application of the mAeBS in 5G networks tralized, distributed, and hybrid [9]. The hybrid
for capacity enhancement still faces many chal- SON can not only take the global network sta-
lenges. First of all, the management of mAeBSs tus into full consideration, but also fully use the
and the mACE procedures call for support from autonomous optimization capability of the net-
5G network architecture, where SON provides work locally. Thereby, hybrid SON can improve
a common framework to enable self-organiz- the efficiency and performance of autonomous
ing of mAeBSs and self-execution of mACE. To management, and is commonly used in SON
accurately perform mACE in overloaded areas, use case design [8]. In order to ensure the stable
we should know the areas with extreme traffic effect of self-optimization, SON requires a closed-
load, a procedure to which we refer as hotspot loop management process. According to [10],
detection. In order to achieve the highest mACE this process generally includes four stages: moni-
performance by deriving the optimal 3D positions toring, analyzing, executing, and evaluating.

IEEE Wireless Communications • August 2018 59

Authorized licensed use limited to: Auckland University of Technology. Downloaded on June 06,2020 at 20:27:59 UTC from IEEE Xplore. Restrictions apply.
Under proper data After confirming the mACE management archi- nas is a prerequisite before communication. In a
mining methods, such tecture, process, and functions, we need to study single time slot, beam alignment is performed to
as K-means clustering, the optimization algorithm specifically for the two establish or maintain a beam link. The alignment
we can get the cluster key issues of hotspot detection and location selec- process takes up a time interval during which
tion of mAeBSs. no data transmission is supported. According to
center point. Due to the classical two-stage beam alignment method,
the improvement of Hotspot Detection coarse sector-level scanning is performed in the
satellite positioning The deployment of micro base stations is an first stage, and the time can be ignored. In the
accuracy, the popular- important way to enhance capacity for high-densi- second stage, the beam is aligned within the sec-
ization of terminal GPS ty areas. We hope that mAeBSs can be deployed tor. The time used is directly related to the beam-
flexibly at the highest density areas to improve the width, which means a narrower beam will cause
and the advanced data load condition. The users’ highest density position larger alignment time overhead. In order to effec-
analysis techniques, tends to be the center of a hotspot. There has tively reduce the impact of alignment time over-
user location clustering been a lot research on wireless access network head on throughput, on one hand, we can find
analysis will become users’ hotspot detection. The existing approaches the optimal trade-off between beam width and
the preferred approach can be divided into three parts: beam gain. On the other hand, we can adopt a
• Hotspot detection utilizing special devices hierarchical scanning scheme with more than two
for hotspot detection. such as pseudo picocell [11]. These devices levels to reduce the temporal overhead.
can obtain users’ location information within its Beam Scheduling: Due to the highly direction-
coverage in real time, and complete the user al transmission of mmWave, users from different
density assessment. directions can be well differentiated using differ-
• Hotspot detection based on user group mobili- ent beams. Thus, multiple users can be served
ty analysis. Hotspots are often generated by the by mAeBSs with different beams in the same
composition of quantities of active users’ ser- time slot. This is often referred to as space-divi-
vice, such as the procession of a crowd and the sion multiple access (SDMA) or beam-division
people in a driving bus. The hotspot location multiple access (BDMA). Theoretically, when the
can be obtained through analyzing the mobility BS transceiver is equipped with N RF chains, and
of these special groups through signal-to-inter- each user is equipped with a single transceiver
ference ratio (SIR) and other indirect indicators. RF chain, the overall multi-user capacity can be
• Hotspot detection based on user location clus- increased to N times when using SDMA. But
tering analysis. With current GPS signaling, the a key issue of SDMA is how to group users so
trilateration measurement, the uplink/downlink that different users from different groups can be
time difference of arrival, and so on, we can served by an mAeBS at the same time without
obtain mobile user location with high accuracy. interfering with each other. A simple but practical
Under proper data mining methods, such as strategy is to group users based on their directions
K-means clustering, we can get the cluster center to the mAeBS. Users of overlapped directions are
point. Due to the improvement of satellite posi- grouped together, and only allow users from dif-
tioning accuracy, the popularization of terminal ferent groups to be served simultaneously. Anoth-
GPS, and advanced data analysis techniques, user er intuitive strategy is to use a spectrum resource
location clustering analysis will become the pre- allocation method, which uses one beam to serve
ferred approach for hotspot detection. multiple users in the same group of users in differ-
ent spectrum sub-bands. However, as user posi-
Capacity Analysis of mAeBS tions change over time, both types of strategies
Different from sub-6 GHz spectrum, mmWave need additional mechanisms to dynamically man-
communication makes the mAeBS have altered age and maintain the user groups.
capacity properties, which are primarily affected
by mmWave air-to-ground wireless channel, align- 3D Deployment
ment delay in directional beam transmission, and The 3D deployment problems of the AeBS can
beam scheduling among multiple users. be divided into maximum coverage area deploy-
Channel Capacity of mAeBS: Due to the high ment, maximum admitted user capacity deploy-
penetration loss of mmWave signal and its prone- ment, and maximum achievable throughput
ness to blockage by obstacles, the line-of-sight deployment according to their objectives. The
(LoS) and non-LoS (NLoS) paths of mmWave have maximum coverage area deployment does not
large differences in transmission loss compared to consider user distributions, but finds a position for
the sub-6 GHz band. On one hand, this is because the AeBS with the largest coverage having signal
the mmWave signal loses a lot of energy when strength above a specific threshold. Therefore, the
reflected in the object surface; on the other hand, height in the 3D space becomes the main factor
the NLoS path is longer than the LoS path. There- that affects the maximum coverage, and analyses
fore, mmWave signal through the NLoS path has show that the corresponding height exists. For
a larger attenuation loss than that through the LoS maximum admitted user capacity deployment, the
path. For terrestrial mmWave communications sys- specific spatial users’ distribution should be con-
tems, because of buildings and other obstructions, sidered, and the 2D deployment location should
a large number of users do not have an LoS path be identified as well. It should be noted that the
from the on-ground BS to their location. However, 2D location and height can be separately consid-
the use of air-to-ground transmission by mAeBSs ered and successively optimized. The maximum
can significantly improve NLoS transmission due to admitted user capacity deployment needs to con-
the increased depression angle. sider not only the user’s location distribution but
Beam Alignment: Since an mAeBS uses the also the differences in user traffic requirements,
mmWave beam for data transmission, aligning the as well as the limitation of system resources,
main lobe direction of transmit and receive anten- signal interference, and other factors. Then the

60 IEEE Wireless Communications • August 2018

Authorized licensed use limited to: Auckland University of Technology. Downloaded on June 06,2020 at 20:27:59 UTC from IEEE Xplore. Restrictions apply.
 Distributed AB-SON 5G CoreNet
Executing
OAM AB-SON
Abnormality detected? AB-SON
BS-SON Analyzing Executing
No Yes
Receive instructions Send
Data collection instructions
SON coordinator
Quit
CE Quit CE State
Instruction and send evaluation Yes Generate Generate ‘3D
type? out alarms Down: Deployment instructions No
No ‘quit CE’ deployment’
Perform CE Up: States and user data Down: Optimization instructions Need CE? In CE? instructions instructions
Up: Node and user data Big-data
Perform CE at hotspot area Yes driven
knowledge No Yes
Regional user data collection mAeBS Regional hotspot data base Enough gains
and CE performance evaluation detection by CE?

Yes CE performance LPN 3D deployment Evaluate performance

below threshold? mAeBS HPN scheme generation indicators
No LPN
Self-adjustment for 3D Planning Evaluating
deployment and report data
Evaluating

FIGURE 2. Management architecture of mAeBS 5G network for capacity enhancement.

total throughput and spectrum efficiency of the Evaluating Phase: To ensure the efficacy of the
AeBS can be maximized through optimization generated CE scheme in last phase, several key
of deployment. As there are few studies on the performance indicators should be evaluated, such
deployment methods for the third one, this article as edge user data rate. If enough gains can be
mainly focuses on this issue. attained, AB-SON will go to the executing phase
to generate 3D deployment instructions; other-
Self-Organizing Architecture and Procedures wise, it will go back to the analyzing phase and
In this article, we consider the following mAeBS continue the loop.
and on-ground cellular base station collabora- Executing Phase: AB-SON generates 3D
tion in 5G networks. The on-ground cellular base deployment or quit CE instructions according to
station provides seamless coverage with some the results produced in the earlier phases. These
mAeBSs in each cell. Each mAeBS utilizes mas- instructions should be informative enough so that
sive multiple-input multiple-output (MIMO) and mAeBSs can respond to them directly. After send-
beamforming techniques to shape the mmWave ing these instructions, AB-SON goes back to the
signal into very narrow beams, selectively provid- analyzing phase and starts a new loop.
ing access to users. As the mAeBS uses cable and In addition, because of the variation in real-
fiber for energy supply and backhaul transmission, istic network environments and specific user
we do not consider the power and return trans- distribution, the instructed position for each
mission on the mAeBS energy and capacity con- mAeBS can only serve as its initial deployment
straints. The management architecture is shown position. In order to achieve optimal CE perfor-
in Fig. 2. mance, after arriving at the instructed position,
The deployment management of the mAeBS is mAeBSs should perform automatic optimiza-
managed in a hybrid manner through the distribut- tions more than just to find a better deployment
ed AB-SONs located in individual mAeBSs and the position but also to optimize the radio param-
centralized AeBS SON (AB-SON) located in the eters and resource allocations. This process
operation administration and maintenance (OAM) should be under the guidance of the distributed
system. Still, a SON coordinator exists in the OAM AB-SON, which is also responsible for abnor-
system to coordinate a variety of SON functions mality detection and interaction with central-
including normal on-ground BS-SON. This man- ized AB-SON as well.
agement mechanism is mainly supported by the
following four phases.
Analyzing Phase: AB-SON mainly collects and
Proposed mAeBS Deployment Approach for
analyzes user location data and base station load Capacity Enhancement
information. By evaluating network states with big This section introduces the related methods,
data techniques, AB-SON is able to determine together with the specific effects applied to the
whether to perform CE. If yes, it turns to the plan- sample scenario shown in Fig. 1. There are three
ning phase; otherwise, it will stay at the current hotspots in this scenario. The network needs to
stage, and if the network has already been in CE, analyze hotspot locations and user distribution
AB-SON should send “quit CE” instructions to features, and determine the best deployment
mAeBSs. location by maximizing ergodic capacity. The sim-
Planning Phase: Effective data analyzing meth- ulation settings are the same as in our previous
ods are required to deal with the massive user article [12]. Performing analysis with the Gaussian
position data efficiently so as to detect potential mixture model (GMM) on mobile user data from
hotspots efficiently. If any hotspot is detected, a network operator in cooperation, we are able to
AB-SON assesses each hotspot to deduce its user identify hotspots and the corresponding user den-
density and center position, which will be used in sities, which enable the derivation of per-mAeBS
the 3D deployment scheme generation step. coverage and ergodic capacity.

IEEE Wireless Communications • August 2018 61

Authorized licensed use limited to: Auckland University of Technology. Downloaded on June 06,2020 at 20:27:59 UTC from IEEE Xplore. Restrictions apply.
 2.5

1.5

0.5

0
0 0.5 1 0 0.5 1 0 0.5 1
Distance (km)

FIGURE 3. The hotspot identification process using the GEM method. It first tries one Gaussian component, that is, k = 1, then k = 2,
and terminates at k = 3.

Hotspot Region Detection and 2D Position Selection ate achievable data rate performance at the sys-
tem level rather than at the channel level. In this
Hotspots formed by user aggregation can be iden- work, ergodic capacity is defined as the expec-
tified by clustering algorithm. Based on the GMM, tation of Shannon capacity over signal-to-inter-
each hotspot can be matched with a Gaussian ference-plus-noise ratio (SINR), or signal-to-noise
exponent. The problem becomes training the ratio (SNR) in noise limited cases. Hence, the key
model for best similarity degree. Compared to to derive ergodic capacity is getting to know the
other clustering algorithms, a GMM-based user probability density function (pdf) of SINR. Among
clustering algorithm can identify the number of all factors, user distribution impacts pdf of SINR
hotspots automatically with existing low-complexi- more than was expected. This is because the high-
ty algorithms, such as greedy expectation-maximi- er aggregation degree of a user around a BS will
zation (GEM). On the other hand, the acquired produce larger expected achievable data rate.
Gaussian component parameters depict the main Luckily, each Gaussian exponent in the optimized
characteristics of a hotspot, for example, its cen- GMM reflects the spatial distribution of users in
ter position and the aggregation degree of users. the corresponding hotspot area. As an mAeBS has
These hotspot center positions can serve as the large capacity, here we only consider deploying
initial 2D position for mAeBSs, and the user aggre- one mAeBS at the center of each hotspot area.
gation degree information can be further utilized The ergodic capacity of an mAeBS in the cov-
to derive optimal altitude for each mAeBS. ered area can be deduced parameterized over
In this article, the GEM algorithm is used to auto- the altitude of the mAeBS, to indicate the wireless
matically find the GMM structure, the number of capacity of the mAeBS at a certain altitude.
Gaussian components, and the parameters for each Through numerical calculation techniques, we
Gaussian component. The algorithm principle is as fol- can plot the ergodic capacity against different alti-
lows. Assume that in the k + 1th iteration the existing tudes of the mAeBS, as shown in Fig. 4a. It can be
mixed structure of k Gaussian components, denoted seen that there is a peak for ergodic capacity with
as fk(x), and new Gaussian component, N(x|q), are the increase of altitude, and the peak value and
weighted summed up to form new Gaussian mixture corresponding altitude are related to the network
fk+1(x), and the weights of fk(x) and N(x|q) are both environment. In a normal urban environment, an
greater than 0, whose sum is equal to 1. In the follow- mAeBS can serve users at higher altitude to obtain
ing training process, only the weight and parameters larger capacity, while for dense urban and high-
of the new Gaussian component are trained, which rise urban areas, serious blockage of dense and tall
is called a partial EM algorithm, which is later fed to buildings severely limit the coverage of an mAeBS,
the standard EM algorithm as an initial parameter, and the mAeBS will get closer to users to reach
and the EM algorithm retrains the GMM model again maximum throughput. Moreover, we also illustrate
until the similarity degree does not increase. Similarly, the effect of user aggregation degree on maximum
as the number of Gaussian components approaches ergodic capacity and the corresponding altitude.
the actual number of hotspots, a new Gaussian com- As shown in Fig. 4b, the optimal altitude of the
ponent does not improve the similarity degree and mAeBS gets lower as the user aggregation degree
the overall algorithm ends. The whole detection pro- increases. This is because in a hotspot with higher
cess for the given three hotspots scenario is shown user aggregation degree, the mAeBS can fully uti-
in Fig. 3. lize its resources to reach its capacity limit, without
bothering to climb to higher altitude to cover a
Altitude Selection via Effective Capacity Assessment larger area and serve more users.
From the network operator’s viewpoint, the main
purpose of capacity evaluation is to know the Effects of mAeBS Deployment
average data rate performance of a user within a Using an mAeBS to enhance the wireless capac-
certain network area. To achieve this goal, [13] ity in a traffic hotspot area, the ability of the
first proposed to use ergodic capacity to evalu- network to deal with wireless services in the

62 IEEE Wireless Communications • August 2018

Authorized licensed use limited to: Auckland University of Technology. Downloaded on June 06,2020 at 20:27:59 UTC from IEEE Xplore. Restrictions apply.
 25 180 24
Normal urban Optimal LAP altitude
Dense urban Ergodic capacity
High rise urban
20
160 23

Optimal LAP altitude (h) (m)

Ergodic capacity (b/s/Hz)

Ergodic capacity [bps/Hz]

15

140 22

10

120 21
5

0 100 20
0 100 200 300 400 500 600 700 1 2 3 4 5 6 7 8 9 10
MAeBS altitude (h) (m) Deviation σ2r (m)
(a) (b)

FIGURE 4. Altitude of mAeBS and user density affect ergodic capacity: a) the influence of the altitude of the mAeBS on the ergodic
capacity under different physical environments; b) the influence of user aggregation degree on the optimal altitude of the mAeBS
and the maximum ergodic capacity.

Ratio of users with data rate above threshold

2 1.0 2 1.0 1
1.5 0.9 1.5 0.9 0.9 CE
0.8 0.8 0.8 NoCE
Normalized load density

Normalized load density

1 1
0.7 0.7 0.7
0.5 0.5 0.6
Y-axis[km]

Y-axis[km]

0.6 0.6
0 0 0.5
0.5 0.5
-0.5 -0.5 0.4
0.4 0.4
0.3
-1 0.3 -1 0.3 0.2
-1.5 0.2 -1.5 0.2 0.1
-2 0.1 -2 0.1 0
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 3
X-axis[km] X-axis[km] RATE threshold (Mbps)
(a) (b) (c)

FIGURE 5. Effect of deploying mAeBS for capacity enhancement in traffic hotspots: a) heat map of load density before deploying
mAeBSs; b) heat map of load density after deploying mAeBS; c) user data rate performance before and after deploying mAeBSs.

enhanced area will be significantly improved,

and the redundant traffic load in the hotspot
Challenge and Future Work
area will be effectively alleviated. To intuitive- As an emerging area, the application of mAeBSs,
ly reflect the regional network load changes in together with the optimization schemes of
the area, we introduce the system load density mAeBSs on deployment position, mobility control,
[14] to measure the load pressure imposed on resource allocation, and so on, for certain objec-
the network at a certain location. This indicator tives call for more research efforts. We list some
is defined as the ratio of averaged user traffic of the issues that need better solutions in future
demand within a unit area to network service work.
capability at the measured location. As the traffic Backhaul Transmission for mAeBSs: The total
volume in the network is almost constant before throughput of mAeBSs is getting considerably
and after capacity enhancement, but the service large so as to serve more bandwidth-demanding
capacity of the enhanced area is significantly users. The backhaul link capacity becomes an
improved, the load density within the area will unignorable factor constraining mAeBS through-
be significantly reduced. The heat maps of load put. Equipping an mmWave communication sys-
density in the wireless access network before tem, a high-capacity beam link becomes a natural
and after capacity enhancement are shown in choice for backhaul transmission. However, as a
Figs. 5a and 5b. It can be seen that deploying power and antenna sharing system, the resource
mAeBSs in hotspot areas will significantly alle- allocation scheme should be carefully designed to
viate the load pressure. Figure 5c illustrates the reach an optimal trade-off between backhaul link
ratio of users whose data rate is above a cer- and access capacity.
tain threshold. For instance, about 100 percent Multiple mAeBSs in the Same Hotspot:
of users have data rates larger than 0.5 Mb/s With the constant growth of mobile data traffic,
before an mAeBS is deployed, while almost all the capacity enhancement for a single hotspot
users have data rates larger than 1 Mb/s, which with excessive traffic is probably not attainable
means the worst case data rate is improved from by deploying only one mAeBS. Thus, a reason-
0.5 Mb/s to 1 Mb/s. Users generally have much able number of mAeBSs together with the joint
higher data rates. deployment of multiple mAeBSs, backhaul net-

IEEE Wireless Communications • August 2018 63

Authorized licensed use limited to: Auckland University of Technology. Downloaded on June 06,2020 at 20:27:59 UTC from IEEE Xplore. Restrictions apply.
With the widespread work design, and user admission control in the [3] M. Agiwal, A. Roy, and N. Saxena, “Next Generation 5G
Wireless Networks: A Comprehensive Survey,” IEEE Com-
application of 5G same hotspot call for good investigations. mun. Surveys & Tutorials, vol. 18, no. 3, 2016, pp. 1617–55.
mmWave technology, Mobility Control of mAeBS: The mobility and [4] Y. Niu et al., “A Survey of Millimeter Wave Communications
increasingly more state change of users will lead to the variation (mmWave) for 5G: Opportunities and Challenges,” Wireless
of hotspots. It is necessary to study advanced Networks, vol. 21, no. 8, Nov. 2015, pp. 2657–76.
terrestrial mmWave [5] K. Sakaguchi et al., “Where, When, and How mmWave Is
deployment and mobility management policy of Used in 5G and Beyond,” arXiv:1704.08131 [cs], Apr. 2017.
base stations will be mAeBSs under time-varying spatial traffic distri- [6] Z. Xiao, P. Xia, and X. G. Xia, “Enabling UAV Cellular with Mil-
deployed in cellular bution. On the other hand, the mobility of users limeter-Wave Communication: Potentials and Approaches,”
and mAeBSs will result in failure of the radio link. IEEE Commun. Mag., vol. 54, no. 5, May 2016, pp. 66–73.
networks. How to [7] Y. Xu et al., “Three-Dimension Massive MIMO for Air-to-
realize the cooperation A proper admission and handover control mecha- Ground Transmission: Location-Assisted Precoding and
nism will provide good insurance for the quality of Impact of AoD Uncertainty,” IEEE Access, vol. 5, 2017, pp.
and coordination of the service offered by mAeBSs. 15,582–96.
air-and-ground inte- Energy Supply and Consumption: The mobili- [8] H. Zhang et al., “Self-Organization in Disasterresilient Het-
erogeneous Small Cell Networks,” IEEE Network, vol. 30, no.
grated mmWave base ty of an mAeBS relies on portable energy supply. 2, Mar. 2016, pp. 116–21.
stations is an important Hence, an mAeBS is mainly powered by its own [9] M. Peng et al., “Self-Configuration and Self-Optimization in
issue in the future 5G battery, while its mechanical part is really energy LTE-Advanced Heterogeneous Networks,” IEEE Commun.
exhausting compared to the PF part. Thus, sustain- Mag., vol. 51, no. 5, May 2013, pp. 36–45.
networks. [10] P. Yu et al., “Energy-Saving Management Mechanism Based
able energy supply and energy harvesting technol- on Hybrid Energy Supplies for LTE Heterogeneous Net-
ogies can be taken as potential solutions. works,” Mobile Info. Systems, vol. 2016, 2016.
Cooperation and Coordination with Terrestri- [11] L. Ewe, R. Moedinger, and H. Bakker, “Mobile User
al mmWave Nodes: With the widespread appli- Hotspot Detection in LTE Networks by Moving Pseudo Pico
Cells,” Proc. Euro. Wireless 2016, 2016, pp. 1–6.
cation of 5G mmWave technology, increasingly [12] T. Zhang et al., “Capacity Enhancement for Next Genera-
more terrestrial mmWave BSs will be deployed in tion Mobile Networks Using Mmwave Aerial Base Station,”
cellular networks. How to realize the cooperation Proc. IEEE GLOBECOM 2017, Dec 2017, pp. 1–6.
and coordination of air-and-ground integrated [13] H. Dhillon et al., “Modeling and Analysis of K-Tier Downlink
Heterogeneous Cellular Networks,” IEEE JSAC, vol. 30, no.
mmWave BSs is an important issue in future 5G 3, Apr. 2012, pp. 550–60.
networks. [14] H. Kim et al., “Distributed Alpha-Optimal User Association
Application of Intelligent Approaches: Modern and Cell Load Balancing in Wireless Networks,” IEEE/ACM
learning approaches, such as deep reinforcement Trans. Networking, vol. 20, no. 1, Feb. 2012, pp. 177–90.
learning, have proven themselves effective in solving
complex problems, thus enabling new opportunities
Biographies
Peng Yu obtained his B.Eng. and Ph.D. degrees from Beijing
in designing strategies for systematic optimization of University of Posts and Telecommunications (BUPT), China, in
mAeBS networking in 5G networks. 2008 and 2013, respectively. He is an associate professor in the
State Key Laboratory of Networking and Switching Technology,
Conclusion BUPT. His research interests are autonomic management and
hybrid energy allocation in GreenNet.
In this article, we investigate the issue of capacity
enhancement using mAeBSs in SON-enabled 5G Wenjing Li is a professor at BUPT and serves as a director in
networks. Control of mAeBSs to perform effective the Key Laboratory of Network Management research center.
Meanwhile, she is the leader of TC7/WG1 in the China Commu-
capacity enhancement turns out to be a challeng- nications Standards Association (CCSA). Her research interests
ing issue. To tame the trivial management process, are wireless network management and automatic healing in
we put forward a self-organizing management SON.
architecture and design closed-loop self-optimiz-
Fanqin Zhou obtained his B.Eng. degree from BUPT in 2012.
ing procedures for it. During these procedures, He is working toward a Ph.D. degree at BUPT. His research
hotspot detection and 3D deployment of mAeBSs interests include resource allocation and mobility load balancing
are the key problems, and we use GMM-based in multi-RAT heterogeneous networks.
clustering and ergodic capacity maximization to
Lei Feng obtained his B.Eng. and Ph.D. degrees in communica-
identify the proper 3D positioning of mAeBSs. tion and information systems from BUPT in 2009 and 2015. He
The use case shows that the proposed approach is a lecturer at present in the State Key Laboratory of Network-
can significantly alleviate the load pressure in ing and Switching Technology, BUPT. His research interests are
hotspot areas, and users can obtain considerable resource management in wireless networks and smart grid.
throughput gain. We identify a few challenging Mengjun Yin obtained her B.Eng. degree from BUPT in 2013.
issues related to mAeBS-involved network optimi- She is working toward a Ph.D. degree at BUPT. Her research
zation for future research at last. interests are wireless network management and automatic heal-
ing in SON.
Acknowledgment Shaoyong Guo is a lecturer at the State Key Laboratory of Net-
This work is supported by the National Science working and Switching Technology, BUPT. His current research
and Technology Major Project of the Minis- interests include smart grid and network management and termi-
try of Science and Technology of China (No. nal management.
2018ZX030110004). Zhipeng Gao is a professor at BUPT and serves as the vice-pres-
ident of the International School of BUPT. He has presided over
References a series of key research projects on network and service man-
[1] Cisco, “Cisco Visual Networking Index: Global Mobile Data agement, including projects supported by the National Natural
Traffic Forecast Update, 20162021 White Paper”; https:// Science Foundation and 863 Programs. He has received eight
www.cisco.com/c/en/us/solutions/collateral/servicepro- provincial scientific and technical awards.
vider/visual-networking-index-vni/mobile-white-paper-c11-
520862.html Xuesong Qiu is a professor at BUPT and serves as the vice-pres-
[2] S. Kandeepan et al., “Aerialterrestrial Communications: Ter- ident of the Institute of Network Technology (INT) of BUPT. He
restrial Cooperation and Energy-Efficient Transmissions to has hosted a series of state research projects on network man-
Aerial Base Stations,” IEEE Trans. Aerospace and Electronic agement. He has earned more than 13 China state-level and
Systems, vol. 50, no. 4, Oct. 2014, pp. 2715–35. provincial and ministerial-level science and technology prizes.

64 IEEE Wireless Communications • August 2018

Authorized licensed use limited to: Auckland University of Technology. Downloaded on June 06,2020 at 20:27:59 UTC from IEEE Xplore. Restrictions apply.