Ultra-Dense Networks: A Survey

This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2016.2571730, IEEE
Communications Surveys & Tutorials
DRAFT 1
Ultra-Dense Networks: A Survey

Mahmoud Kamel, Student Member, IEEE, Walaa Hamouda, Senior Member, IEEE, and Amr Youssef, Senior
Member, IEEE
Abstract—The exponential growth and availability of data in TABLE I: L IST OF ACRONYMS

all forms is the main booster to the continuing evolution in the AE Antenna Element
communications industry. The popularization of traffic-intensive ASE Area Spectral Efficiency
applications including high definition video, 3D visualization, BBU Base Band Unit
augmented reality, wearable devices, and cloud computing defines BPP Binomial Point Process
a new era of mobile communications. The immense amount of BS Base Station
traffic generated by today’s customers requires a paradigm shift CDMA Code-Division Multiple Access
in all aspects of mobile networks. Ultra Dense Network (UDN) CoMP Coordinated MultiPoint
CSG Closed Subscriber Group
is one of the leading ideas in this racetrack. In UDNs, the
CSI Channel State Information
access nodes and/or the number of communication links per D2D Device-to-Device
unit area are densified. In this paper, we provide a survey- DAS Distributed Antenna System
style introduction to dense small cell networks. Moreover, we DC Dual Connectivity
summarize and compare some of the recent achievements and DSL Digital Subscriber Line
research findings. We discuss the modelling techniques and HetNet Heterogeneous Networks
the performance metrics widely used to model problems in HCPP Hard Core Point Process
UDN. Also, we present the enabling technologies for network IoT Internet of Things
densification in order to understand the state-of-the-art. We KPI Key Performance Indicator
LOS Line-Of-Sight
consider many research directions in this survey, namely, user
LTE Long-Term Evolution
association, interference management, energy efficiency, spec- MFG Mean-Field Game
trum sharing, resource management, scheduling, backhauling, M2M Machine-to-Machine
propagation modelling, and the economics of UDN deployment. MIMO Multiple-Input Multiple-Output
Finally, we discuss the challenges and open problems to the mmWaves millimeter Waves
researchers in the field or newcomers who aim to conduct MTC Machine-Type Communication
research in this interesting and active area of research. OFDMA Orthogonal Frequency-Division Multiple Access
ONF Open Networking Foundation
Index Terms—UDN, Network Densification, Small Cells, PCP Poisson Cluster Process
Cloud-RAN, DAS, D2D, IoT, Massive-MIMO, mmWaves, 5G. PPP Poisson Point Process
QoS Quality of Service
QoE Quality of Experience
RAN Radio Access Network
I. I NTRODUCTION RAT Radio Access Technology
RE Range Expansion
T HE evolution of mobile devices and applications in the

current decade have drawn a new picture for wireless
networks [1]. Where the concept of Ultra-Dense Network
RRH
RRM
RSS
SDN
Remote Radio Head
Radio Resource Management
Received Signal Strength
Software-Defined Network
(UDN) represents a new paradigm shift in future networks [2]– SDWN Software-Defined Wireless Network
SINR Signal to Interference-plus-Noise Ratio
[4]. The basic idea is to get the access nodes as close as TDD Time-Division Duplex
possible to the end users. The realization of this is simply done UDN Ultra-Dense Network
by the dense deployment of small cells in the hotspots where
immense traffic is generated. These small cells are access
nodes with small transmission power, and hence, a small Cisco, in the Cisco Virtual Networking Index, predicts a 10
coverage. As illustrated in Figure 1, the cells are deployed fold increase in the mobile data worldwide for the period 2014-
by the customers in their premises, or by the operators in 2019 [5]. With the scarce spectrum resources, one vital and
the streets (e.g., on lampposts, trees, and walls) and hotspots long-term solution is to increase the reuse per unit area of the
(e.g., airports, metro/train stations, and markets). Thus, UDN existing spectrum. This spectrum reuse is a two fold gain; the
deployment scenarios introduce a different coverage environ- spectrum is increasingly reused, which significantly improves
ment where any given user would be in close proximity to the network capacity, and the link to the end user becomes
many cells. shorter which improves the link quality.
The densification of wireless networks is motivated by the Obviously, the network densification cannot continue end-
traffic trends measured in the current decade. For instance, lessly. There would be a fundamental limit for network densi-
fication which requires extensive research for realistic system
M. Kamel and W. Hamouda are with the Department of Electrical and
Computer Engineering, Concordia University, Montreal, QC, H3G 1M8, models to capture the reality of dense networks. Andrews et
Canada email: (mah kame@encs.concordia.ca ; hamouda@ece.concordia.ca). al. addressed the question of what are the fundamental limits
A. Youssef is with the Concordia Institute for Information Systems En- of network densification [6]. However, some idealizations of
gineering, Concordia University, Montreal, QC H3G 1M8, Canada (e-mail:
youssef@ciise.concordia.ca). the model need further investigations for a complete under-
Manuscript received November 26, 2015; revised November 26, 2015. standing of these fundamental limits.
1553-877X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2016.2571730, IEEE
DRAFT 2
gaps and the unanswered questions, we review the published

research in different disciplines assuming a UDN environment
as defined in Section II. Henceforth, we use the terms dense
network, hyper-dense network, extremely-dense network, and
ultra-dense network (abbr. UDN) interchangeably to refer to
the same concept.
Modelling techniques and key performance metrics are
discussed in Section III. Then the enabling technologies and
the driving factors of network densification are discussed in
Section IV. In Section V, a discussion of the recent work
in user association considering a UDN network is presented.
Different interference management schemes are discussed and
compared in Section VI. The backhauling alternatives and
challenges in UDN context are discussed in Section VII. The
energy efficiency of UDN deployments is discussed in Section
VIII. In Section IX, we consider another set of research
disciplines which are relevant to the progress of UDN research
Fig. 1: The network is densified by deploying small cells status. This set includes small cell discovery, spectrum sharing;
indoors in buildings and stores, and outdoors on trees, lamp- resource management; and scheduling, propagation modelling,
posts, and building walls. Small cell networks coexist with and economics of UDN deployment. Finally, the challenges
macrocells, either in the same spectrum or on a dedicated facing the development of UDN and the open problems are
carrier. highlighted in Section X. The conclusions and summary of the
survey are presented in Section XI. Table I lists the acronyms
used throughout the survey and Figure 2 provides a summary
A. State-of-the-Art and Contributions of the organization of the survey.
To the best of our knowledge, this survey is the first to II. W HAT IS A UDN?
address the ultra-dense network research status in different dis-
In this section, we provide a basic background to understand
ciplines. López-Pérez et al. [7] conduct extensive simulations
UDNs. We summarize the different definitions of UDN in
to highlight the gains of three paradigm concepts, namely, net-
literature, and we discuss the fundamental features of UDN
work densification, exploiting high frequency bands, and the
as compared to traditional networks. Moreover, two different
development of high spectrum efficiency techniques on the net-
classification schemes are presented to distinguish various
work throughput, energy efficiency, and signal-to-interference-
densification approaches.
plus-noise ratio (SINR) distributions. Moreover, they develop
a common understanding of the network densification without
emphasis on reviewing the research status of this paradigm A. Definitions of UDN
shift. On the other hand, Liu et al. [8] consider reviewing the Ultra-Dense Networks can be defined as those networks
state-of-the-art of user association in different scopes, specif- where there are more cells than active users [7], [11]–[14].
ically Heterogeneous Networks (HetNets), massive multiple- In other words, λb λu , where λb is the density of access
input multiple-output (massive-MIMO) networks, millimeter points, and λu is the density of users. Another definition of
waves (mmWaves) networks, and energy harvesting networks UDN was solely given in terms of the cell density, irrespective
[9]. Finally, Gotsis et al. [10] give a general overview of of the users density. Ding et al. [15] provided a quantitative
UDN without explicit definition of ultra-dense networks or the measure of the density at which a network can be considered
evolution of cellular networks towards UDN. Generally, they ultra-dense (≥ 103 cells/km2 ). In fact, the first definition
provide some background without discussing of the research converges to the second given that the active users density
status of different disciplines in the UDN context. considered in dense urban scenarios is upper bounded by about
Different from [7]–[10], in this survey, we present the state- 600 active users/km2 [16].
of-the-art in different research directions in the context of Generally, the small cells in UDN can be classified into
ultra-dense networks. We review the conducted research in fully-functioning base stations (BSs) (picocells and femtocells)
many relevant active research disciplines. Also, we discuss and macro-extension access points (relays and Remote Radio
the challenges facing the research progress as well as the open Heads (RRHs)). The fully-functioning BS is capable of per-
problems that require serious investigations in order to answer forming all the functions of a macrocell with a lower power
the raised questions and to give insights. Consequently, this in a smaller coverage area. Specifically, the fully-function BS
shed lights on the way to the successful deployment of UDNs. performs all the functions of the entire protocol stack [17]. On
In this survey, we review the recent works in the UDN the other hand, a macro-extension access node is an extension
literature. The research work in network densification is still for the macrocell to effectively extend the signal coverage,
in its infancy, and hence there is much work to be done in and it performs all or some of the PHY layer functions
different research directions. In order to identify the research only. Moreover, the small cells feature different capabilities,
DRAFT 3
Stochastic Geometry SDN User Association
DAS Interference
Modeling Game Theory
Techniques Cloud-RAN Backhauling
Section III-B
mmWaves Energy Efficiency
massive-MIMO SC Discovery
Enabling Recent Challenges
Introduction What is a UDN? Technologies D2D Achievements Spectrum Sharing & Open Problems
Section I Section II Section IV Section V-IX Section X
Multi-RAT Scheduling
ASE: Area Spectral Efficiency Coverage Probability
RRM: Radio Resources Management Rate Coverage Proactive Caching RRM
SC: Small Cell Performance
SDN: Software-Defined Network Spectral Efficiency Propagation
Metrics
DAS: Distributed Antenna System Section III-A ASE
RAN: Radio Access Network Economy
Network Throughput
RAT: Radio Access Technology
D2D: Device-to-Device Energy Efficiency
MIMO: Multi-Input Multi-Output
Fairness
Fig. 2: The organization of the survey.
transmission powers, coverage, and deployment scenarios [18], for coverage extension of the macrocell and can be used
[19]. Table II summarizes the features of different small cell as a centralized densification alternative as compared to
types. In what follows we explain the different types of small the distributed densification performed by the picocells
cells [20]: or femtocells.
• Picocells are small BSs which are installed by operators It is important to highlight that indoor small cells (femtocells)
to cover a small coverage area in a range of one hundred operate in three different access modes: open, closed, and
meters. Usually picocells are deployed in hotspots (indoor hybrid. In open access mode, all subscribers of a given
or outdoor) to serve tens of active users by offloading operator can access the node, while in closed access mode
their traffic from the macrocell. The transmission power the access is restricted to a closed subscriber group (CSG). In
of picocells is typically up to 33 dBm, and they are hybrid mode, all subscribers can connect to the femtocell with
mainly deployed for capacity purposes. The backhauling the priority always given to the subscribers of the CSG. The
of picocells is similar to that of the macrocells (fiber deployment of small cells with regular macrocells is termed
or microwave links) in order to provide ideal high- in the literature and standards as HetNet. HetNets in general
bandwidth low-latency links. represents a paradigm shift from homogeneous networks [22].
• Femtocells are user-deployed indoor BSs which are in- As depicted in Figure 3, UDN serves as another evolution
stalled to cover indoor spots (homes, offices, and meeting from HetNets.
rooms) in order to serve a small set of users. The
transmission power of femtocells typically is less than B. Fundamental Features of UDN
20 dBm and the coverage range is in the order of tens In order to understand the current state of the research
of meters. Thus, a femtocell provides a large indoor activities in UDN, the differences between dense networks
signal strength for the home users where most of the and traditional networks need to be highlighted. These fun-
data traffic is generated. The femtocells can be connected damental differences of UDN from traditional networks can
to the network via any of the consumers’ broadband be summarized as follows:
connections such as Digital Subscriber Line (DSL), cable, 1) Many small cells are in the vicinity of a given user. The
or fiber. network access nodes in UDN environments are low-
• Relays are operator deployed access points which are power small cells with a small footprint, or in other
usually deployed for coverage purposes to cover the words, with a small coverage area. Accordingly, the
dead zones and to improve the edge performance of the inter-site distance would be in the range of meters or
macrocells. They transmit the users data back and forth tens of meters. This defines a different wireless coverage
from and to the macrocell, featuring what is considered environment where many small cells would be in a very
as wireless backhaul. Both relays and picocells have the close distance to the users.
same coverage and transmit power, but they mainly differ 2) Idle mode capabilities are of a great interest. Due to the
in three properties. First, a picocell is a fully-functioning high density of small cells, many small cells would be
BS while the relay is an extension for the macrocell. inactive. This motivates the idle mode concept, where
Second, picocells are deployed for capacity, but relays inactive small cells are turned off to partially or fully
are deployed for coverage. Finally, the backhaul of the mitigate their interference [23].
picocell is an ideal backhaul while the relay backhaul is 3) Drastic interference between neighboring cells is a lim-
a wireless in-band or out-of-band backhaul. iting factor. Close proximity of the small cells to each
• RRHs are RF units which are deployed in order to extend other in UDN environments generates high interference.
the coverage of a central BS to a remote geographic Hence, strict interference management schemes are un-
location. RRHs are connected to the central BS via high avoidable to mitigate the interference of neighboring
speed fiber or microwave links [21]. They are deployed cells [24]–[28].
DRAFT 4
TABLE II: D IFFERENT T YPES OF S MALL C ELLS

Type of small cell Deployment Scenario Coverage Power Access Scenario Backhaul
indoor (≤ 100 mW)
Picocells (fully-functioning) indoor/outdoor (planned) up to 100 meters Open Access Ideal
outdoor (0.25 – 2 W)
Femtocells (fully-functioning) indoor (unplanned) 10 – 30 meters ≤ 100 mW Open/Closed/Hybrid Access Non-Ideal
indoor (≤ 100 mW) Wireless
Relays (macro-extension) indoor/outdoor (planned) up to 100 meters Open Access
outdoor (0.25 – 2 W) (in-band/out-of-band)
RRHs (macro-extension) outdoor (planned) up to 100 meters outdoor (0.25 – 2 W) Open Access Ideal
Deployment of different Densification

types of wireless access node of small cells
propagation models, the backhauling alternatives for the dif-
ferent densification schemes, and the performance evaluation
Homogeneous
Network HetNet UDN of the corresponding densification scheme.
Vertical Densification: Deployment of

Fig. 3: The traditional homogeneous network is evolved to
Vertical Densification
indoor small cells in high-rise buildings in offices,
HetNet, and in turn the UDN is a densified HetNet. meeting rooms, food courts, and the interior
of the building.
4) Innovative frequency reuse techniques are required. In

traditional cellular networks, the spectrum is reused at
the level of a cluster of cells where a reuse pattern is
repeated in each cluster. This reuse scheme touches its Horiz
onta
limit in code-division multiple access (CDMA) and or- l Dens
if icatio
thogonal frequency-division multiple access (OFDMA) n
systems where the spectrum is reused in each cell, Horizontal Densification: Deployment of
outdoor small cells in the streets on trees,
i.e., the frequency reuse factor is one. In UDN envi- lamp posts, utility poles, and building walls.
ronments, there would be a need for a paradigm shift in
the frequency reuse concept.
Fig. 4: Vertical densification versus horizontal densification in
5) Backhauling in UDN environments is challenging. It
small cell networks.
might be difficult for operators to guarantee an ideal
high-speed low-delay backhaul for each small cell. Also,
the backhaul of a small cell might be the bottleneck of Consequently, the investigation of UDNs requires different
its capacity, where the backhaul capacity would limit the modelling for the two densification schemes, namely vertical
air-interface capacity [7], [29]–[33]. and horizontal. In this manner, the three-dimensional (3D)
6) High probability of Line-of-Sight (LOS) transmissions. modelling of the vertically densified networks is inescapable
In UDN, the distance between BSs and users is small to reflect the reality of such networks. Only a few papers
enough to have a high probability of LOS transmis- addressed the 3D modelling of cellular networks [36], [37]
sions [15], [34], [35]. Accordingly, the need for a over the past decade. Recently, the modelling of wireless
different propagation modelling becomes a necessity. networks in 3D has been addressed in the work of Gupta
Moreover, the propagation modelling in UDN should et al. considering a UDN context [34]. Furthermore, the
also consider a Rician channel model for multi-path modelling and analysis of coverage in 3D cellular networks is
fading due to the dominant LOS component in the conducted by Pan and Zhu [38]. Consequently, this gives rise
received signal. to the consideration of 3D-MIMO [39] which might be used
as a backhauling alternative in vertically densified networks.
Also, this in turn motivates the modelling and analysis of 3D
C. Horizontal and Vertical Densification fading channels [40]. Moreover, the performance of network
densification in the vertical plane is different from that of the
The densification of wireless networks takes place mainly horizontal plane. In vertical densification, the area spectral
by the deployment of increasing number of small cells. Ac- efficiency for the same area improves in high-rise buildings
cordingly, the network densification comes in two different with more floors. This suggests incorporating the clutter of the
flavors, either horizontal densification or vertical densification. buildings into the performance evaluation of such densification
In horizontal densification, the access nodes are densified in schemes.
the horizontal plane, e.g., in the streets or hotspots. On the
other hand, the vertical densification evolves in the elevation
plane where the customers deploy BSs in their apartments, D. Centralized and Distributed Densification
offices, meeting rooms, and buildings interior. Figure 4 depicts In centralized densification, a central entity takes the coor-
vertical and horizontal network densification. The aforemen- dination role between different elements in order to boost the
tioned classification highlights three major aspects of the network performance. This central entity can be a BS equipped
densification schemes, specifically, the modelling of the den- with a massive-MIMO system [41], a BS extended with a Dis-
sification scheme (vertical/horizontal) to consider appropriate tributed Antenna System (DAS) [42], or a cloud of base band
DRAFT 5
processing units in a Cloud- Radio Access Network (Cloud- coverage, which is defined as the probability that the
RAN) [43]. On the other hand, the distributed densification achievable rate of an arbitrary user is above a certain
of wireless nodes in small cell networks or the distributed minimum. Conversely, the rate outage is the probability
densification of wireless links in Device-to-Device (D2D) that the achievable rate of an arbitrary user falls below
requires scalable algorithms for the collaboration amongst a certain threshold. It is known that the rate distribu-
different nodes. tion and SINR distribution are strongly correlated in
Densification of wireless networks can be realized either macrocell homogeneous networks [22]. Conversely, this
by the deployment of increasing number of access nodes or is not the case in HetNets and then as well in UDN. In
by the densification of the number of links per unit area. In small cell networks, not only the SINR determines the
the first approach, the densification of access nodes can be achievable rate, but also the backhaul capabilities and
achieved in a distributed manner by the deployment of small the load of individual cells.
cells (picocells and femtocells) or via a centralized scheme 3) Average Spectral Efficiency: The average number of
using DAS or Cloud-RAN. Also, in the second theme, the transmitted bits per second per unit bandwidth represents
increasing of the number of links per unit area is realized the efficiency of the spectrum. The efficiency of the
either in a distributed way in D2D communication, or by a spectrum is a crucial performance metric in 5G net-
centralized massive-MIMO deployment. works due to the scarcity of spectrum along with the
high data rate requirements [68]. Also, the cell spectral
III. M ODELING T ECHNIQUES AND P ERFORMANCE efficiency is another form of this metric to measure the
M ETRICS performance of a single cell.
Different techniques and performance metrics are used in 4) Area Spectral Efficiency: Densification of cellular net-
modelling of the problems in UDNs. In this section, we focus works increases the reuse of spectrum per unit area.
on two of the most commonly used modelling techniques, Thus, the ASE is an important metric to quantify the
namely, stochastic geometry [44] and game theory [45]. Also, performance of UDN. ASE is defined as the average
we define the key performance indicators (KPIs) that are achievable data rate per unit bandwidth per unit area
usually used in quantifying the performance of the proposed [69].
techniques and solutions such as coverage and outage probabil- 5) Network Throughput: The network throughput is an-
ity, rate coverage, spectral efficiency, area spectral efficiency other metric to quantify the performance of UDN and is
(ASE), network throughput, energy efficiency, and fairness. defined as the average number of successfully transmit-
Moreover, we present some use cases of the discussed tech- ted bits per sec. per Hz. per unit area [55]. This metric
niques to model some problems in UDNs, these applications considers the success probability in the evaluation of the
give a solid background to address similar problems while ASE of a given network with a certain BS density and
avoiding the pitfalls of modelling such problems. is defined as
A. Performance Metrics Rn = λs pa (1 − pout )Ro , (1)

In this section, we explain the set of commonly used per-
formance metrics in the modelling of UDN problems. These where λs is the BSs density, pa is the probability
metrics are basically related to either the signal to interference- of active BSs, pout is the outage probability, and
plus-noise ratio (SINR) (e.g., coverage (success) probability Ro , log2 (1 + γ̂) is the link capacity with a signal-
and outage probability) or the rate (e.g., rate coverage, average to-interference ratio (SINR) threshold γ̂.
spectral efficiency, and ASE). 6) Energy Efficiency: The ratio of the network throughput
1) Coverage/Success Probability and Outage Proba- or the ASE to the power consumption per unit area is
bility/SINR Distribution: The coverage probability is defined as the energy efficiency [55], [70]. The energy
defined as the probability that the SINR of a randomly efficiency metric is a performance indicator that mea-
selected user is above a certain threshold. In other words, sures the benefit-cost ratio by comparing the achievable
the link quality is good enough to proceed to a successful rate to the energy costs to achieve this rate [70].
connection. The coverage probability is also termed as 7) Fairness: A crucial performance indicator to the evalu-
the success probability. On the other hand, the outage ation of a given cell association, scheduling, or resource
probability or the SINR distribution is the probability management scheme is the fairness between different
that the SINR of an arbitrary user falls below a minimum users. The fairness index measures how likely a given
threshold. A given user is considered in outage if the resource allocation scheme is fair. Jain’s fairness index
SINR of the link to the serving BS is not enough for [71] is a widely-adopted index that computes the fairness
a successful connection. The coverage probability, the of a set of user rates or resource allocations [8] and is
success probability, the outage probability, and the SINR given by
distribution quantify the quality of the link between the PN
user and the serving BS. ( n=1 rn )2
J (r1 , · · · , rN ) , PN (2)
2) Rate Coverage and Rate Outage: In small cell net- N n=1 rn2
works, a better coverage performance metric is the rate
DRAFT 6
TABLE III: M ODELING T ECHNIQUES AND P ERFORMANCE M ETRICS

Research Direction References Modeling Technique Performance Metrics
[46] Spectral Efficiency, ASE
User Association Stochastic Geometry
[47] Coverage Probability
[24], [25] Stochastic Geometry Success Probability
[26] Simulation-based Throughput, SINR Distribution
[27] Simulation-based Throughput
[28] Simulation-based Spectral Efficiency, Throughput
[48] Simulation-based Average Cell Throughput
Interference Management
[49] Game Theory Spectral Efficiency, Energy Efficiency, ASE
[50] Game Theory Throughput, SINR Distribution
[51] Game Theory Throughput, Coverage Probability
[52] Game Theory Throughput
[53] Simulation-based Achievable Rate per Cell
[29] Stochastic Geometry Backhaul Delay, Expected Network Delay, Backhaul Cost
[30] Simulation-based Spectral Efficiency
Backhauling [31] Simulation-based Sum Capacity
[32] Simulation-based Backhaul Throughput, Energy Efficiency
[33] Simulation-based Throughput, Fairness
[54] Stochastic Geometry Energy Efficiency
[55] Stochastic Geometry Network Throughput, Energy Efficiency
[56] Simulation-based ASE, Energy Efficiency, Cell Spectral Efficiency
Energy Efficiency
[57] Stochastic Geometry Cell Average Spectral Efficiency, Area Throughput, Energy Efficiency
[58], [59] Game Theory Energy Efficiency, Outage Probability
[60], [61] Game Theory Energy Efficiency, Spectral Efficiency
[62] Simulation-based Detection Probability
Small Cell Discovery
[63] Simulation-based Scan Power, Small Cell Connected Time
[12] Stochastic Geometry Rate Outage Probability
[64] Simulation-based Minimum Separation Distance
Spectrum Sharing, RRM, and Scheduling [65] Simulation-based Throughput
[66] Simulation-based Throughput
[67] Game Theory Throughput
[15] Stochastic Geometry Coverage Probability, ASE
Propagation Modeling [34] Stochastic Geometry Coverage Probability, Throughput
[35] Stochastic Geometry Outage Probability, ASE, Energy Efficiency
[13] Stochastic Geometry Outage Probability
Economics of UDN
[14] Stochastic Geometry Spectral Efficiency, Profit
where N is the number of users, and rn is the rate of randomness in their placement. Hence, the positions of
the n-th user. the small cells can be modelled as points in two or three-
The aforementioned modelling techniques and performance dimensional Euclidean space, which is termed as a point
metrics are widely used in the literature of UDN. Table process (PP). Stochastic geometry stems as a best-fit tool
III summarizes the corresponding modelling technique and to study such random network environments [44], [72],
metrics used in each considered discipline in this survey. [73]. Many results have been reported using stochas-
It is however important to note that the context of dense net- tic geometry in the literature of traditional networks
works to a large extent is different from traditional networks. and HetNets [74]–[81]. Additionally, the application
Thus, the need for other metrics to quantify the performance of stochastic geometry is expected to meet substantial
of UDN is a significant requirement. Andrews et al. [68] success in dense environments (e.g., see [12]–[14], [24],
introduced a new metric, the BS densification gain ρ, which is [25], [29], [46], [47], [54], [55], [57] and the references
defined as the ratio between the ratio of the rates corresponding therein).
to two different BS densities and the ratio of the corresponding The Poisson point process (PPP), is a point process
BS densities, i.e., model where the number of BSs in a given area A
R2 /R1 R2 λ1 in two-dimensional space or in a given volume V in
ρ= = (3) three-dimensional space has a Poisson distribution with
λ2 /λ1 R1 λ2
mean λb A or λb V , respectively, where the parameter λb
where R1 is the rate corresponding to a BS density of λ1 , represents the density of the BSs per unit area or unit
while R2 is the rate if the density is increased to λ2 . In other volume.
words, this measure quantifies the payoff ratio in terms of rate In Figure 5, we illustrate an example of the Voronoi
relative to the cost ratio in terms of BS density. tessellation of a PPP realization of a dense small cell
network. We consider an area of 20m × 20m, where the
B. Modelling Techniques small cells are marked with triangles, and the typical
1) Stochastic Geometry: The stochastic modelling of the user is located at the origin and is marked by a circle.
spatial distribution of small cells has achieved signifi- The Poisson Voronoi (PV) cell is defined by the edges
cant results in the literature (e.g., see [44], [72]). The marking the random coverage area of a given cell. The
unplanned deployment of the access nodes reflects the probability density function of the size of a typical PV
DRAFT 7
10 from stochastic geometry. Besides Poisson point process

(PPP), there exist many other point processes that match
different applications. Binomial point process (BPP),
5
Hard core point process (HCPP), and Poisson cluster
process (PCP) are a few examples of the stochastic
y (meters)
0 geometry toolbox [44]

2) Game Theory: Game theory is an applied mathematical
discipline which consists of a set of tools and techniques
-5 to solve conflict of interest problems among a set of
rational entities [83]. Game-theoretic approaches are
originally developed to solve problems in economics.
-10
-10 -5 0 5 10 However, the applications of game theory have found
x (meters) a significant success in many other disciplines such as
Fig. 5: The Voronoi tessellation of a PPP realization of a dense biology, political science, computer science and engi-
small cell network with density λb = 0.1 cells/m2 in an area neering.
of 20m × 20m, the small cells are marked with a triangle, and A game is characterized in terms of three main elements,
a typical user at the origin is marked with a circle. A Voronoi namely, players, actions, and payoffs (preferences) [84].
cell is the area enclosed by dotted lines around a given small Players are the rational decision makers, they usually
cell. have a conflict of interest, and a decision of one player
may harm other players. A decision of a player is an
action which is chosen by the player from a set of actions
cell is derived via Monte Carlo method [82], and is given to maximize his/her payoff or benefit. The set of actions
by: might be finite, countable, or infinite. The preference of
3d+1 an action to a given player is an ordering for the actions
( 3d+1
2 )
2 3d−1 3d+1
d
fX (x) = x 2 e−( )
2 x (4) in terms of payoffs.
Γ( 3d+1
2 ) Two different categories of games are widely studied
where X is a random variable that denotes the size of and used to model problems in the literature; non-
a typical PV cell normalized by the reciprocal of the cooperative games [84] and cooperative games [85].
density of the small cells λs , and d is the dimensionality The players in non-cooperative games have no com-
of the problem, i.e., d = 1, 2,R3 indicates 1D, 2D, munication means so that they cannot agree on how
∞ to play the game to achieve a common benefit. Con-
and 3D respectively. Γ(x) = 0 e−t tx−1 dt is the
gamma function. Considering the 2D case in 2D network versely, in cooperative games the players can bargain
models, Equation (4) is reduced to: or form coalitions to improve their respective positions
in playing the game and to achieve a better payoff. In
(3.5)3.5 2.5 −(3.5x) this way, we can identify two different subcategories of
fX (x) = x e (5)
Γ(3.5) cooperative games, namely, bargaining games [86] and
The concept of a typical user in stochastic geometry coalition formation games [85].
refers to a user residing at the origin where the properties In UDN, there exist many research areas where game
of the point process (PP) can be computed. In other theory is a potential tool to design effective algorithms.
words, the typical user is assumed to be a representative Interference management, discovery of small cells, asso-
of all users. Most of the properties of a wireless network ciation of users, and handover of mobile subscribers are
can be expressed as a function of the distance between amongst the areas where tools from game theory might
the network nodes and the typical user. Hence, the find successful applications. The conflicting interests
distance from the typical user to the nth nearest BS amongst the large number of nodes in a dense network
is of a special importance in stochastic modelling of encourages the development of distributed solutions to
wireless networks. Let fn (r) denote the probability maximize a network-wide utility, and in this situation
density function of the distance r to the nth nearest small game theoretic approaches stem as a toolbox where
cell from the typical user that resides at the origin. Then, many existing or new models can fit the problem in
fn (r) is given by [44]: hand. Further specific examples for the use of game
theory to model problems in UDN will be explained
in Sections V-IX.
2 2
fn (r) = (λs π)n r2n−1 e(−λs πr ) (6) Needless to say that the scalability of game theoretic
Γ(n) approaches with a large number of players is debatable
The close proximity of access nodes in a dense net- [87]. This is an interesting issue to consider in UDNs
work requires researchers to propose and investigate a where the number of small cells involved in a game
class of distributed clustering techniques to exploit the might be large. Mean-Field Games (MFGs) represent a
cooperation of cells in the cluster. The investigation candidate modelling technique in such scenario where a
of such structures requires more advanced techniques player takes an action based on the average of the effect
DRAFT 8
of other players’ actions rather than the individual effects The extreme densification of small cells in UDNs requires
[58], [59], [88]. an agile network architecture in order to overcome the lim-
It is equally important to mention that many problems in itations of UDNs. Densification of wireless networks has
UDN can be modelled as optimization problems, where vari- many limitations such as the bottlenecked backhaul, the drastic
ous optimization techniques can be exploited to find an optimal interference, the increasing signaling overhead due to mobility,
solution. The utility maximization under practical constraints and the high operational costs due to energy consumption
are widely used in different network densification contexts [97], [98]. These limitations, amongst others, promote the
and for various objectives. In this survey, we discuss many employment of SDN architecture. The SDN architecture can
proposals and investigations where an optimization technique significantly improve the performance of dense networks, and
is employed to understand the role of the considered param- make it possible to overcome their limitations with elegant
eters and optimization variables. Considering UDN scenarios, solutions. The flexibility and adaptability of SDN satisfy the
many problems in interference management [49], backhauling requirements of a successful dense network, where the re-
[29], [31]–[33], energy efficiency [70], spectrum sharing [89], sources, both wireless and backhaul resources, are provisioned
and the economics of UDN deployment [14] are formulated based on the traffic volumes. In other words, the dense wireless
as binary linear problems [90], [91], mixed binary nonlinear network would adapt itself via dynamic reconfiguration to the
problems [92], or convex problems [93]. actual instantaneous loads of the users.
In UDN, the scalability of the optimization algorithms is CROWD (Connectivity management for eneRgy Optimised
important in terms of overhead, complexity, and convergence. Wireless Dense networks) [97] is a two-tier architecture that is
The decoupling of the investigated problems via clustering introduced to address the limitations of UDN, specifically, the
into a family of localized optimization problems might result backhaul, the mobility management, and the energy consump-
in reducing the complexity. To this end, various distributed tion. CROWD aims at the design of protocols and algorithms
and centralized optimization techniques can be exploited. The for ultra-dense networks. The CROWD architecture proposes
joint optimization of multiple objectives in UDN deployments local controllers to take fast decisions on a fine granularity
is another important issue for further investigation. The joint basis, and regional controllers to optimize the network oper-
optimization of energy efficiency, spectrum efficiency, and ation globally. The local controllers oversee a district of BSs
BS density while considering QoS requirements significantly either cellular or ad hoc access points. They have functions
impacts the design decisions in UDN deployments. such as monitoring and filtering, network discovery, power
control setting, access selection setting, scheduling policy
control, and WiFi parameter setting. The regional controllers
IV. E NABLING T ECHNOLOGIES AND D RIVING FACTORS on the other side have global functions in a large region
To cope with the strict requirements and expected immense comprised of many districts such as long-term clustering, long-
amounts of traffic, 5G networks are likely to be a mixture term adaptation of radio parameters, and traffic-proportional
of different technologies. There is a common understanding backhaul reconfiguration [98].
that the orchestration of a number of different technologies WiSEED (Wireless Software-basEd architecture for Ex-
would govern the road to a successful next generation [68]. tremely Dense networks) [99] is another management archi-
In this section, we discuss the most relevant technologies to tecture proposed to exploit the benefits of SDN in extremely
network densification in order to give essential background dense networks. The architecture jointly manages a set of
for the understanding of the current research efforts and the operational services, namely, routing, mobility, and spectrum
challenges that need to be addressed. usage. To explain, these operational services can be seen as
a software that runs on an SDN controller. The management
provided by WiSEED aims at satisfying key requirements for
A. Software-Defined Network
the future dense networks, scalability, resilience, and energy
In 5G networks, one expects a network that is dynamic, efficiency. The authors in [100] introduce another architec-
manageable, cost-effective, and adaptable. These requirements ture to serve the machine-type communication (MTC) traffic
cannot be all in one network unless it is programmable. generated by indoor devices in homes, offices, hospitals, and
Software-Defined Networking (SDN) [94] emerges as a perfect markets. The architecture suggests the use of indoor small cells
solution in this regard, where the network control functions to handle this traffic, which significantly reduces congestion
are decoupled from the packet forwarding functions. SDN is and overloading of both the radio access network and the
a paradigm shift in networking that offers several ground- core network. SDWN (Software-Defined Wireless Network)
breaking aspects for the next generation networks, flow-based [101] is another architecture which is proposed to extend
forwarding decisions, centralized control of the network oper- and generalize the concept of SDN to the wireless networks.
ation, reconfigurability and programmability through software Niephaus et al. [101] discussed the unique challenges of the
applications, to name a few [94]. One more interesting aspect SDN concepts when it is implemented in wireless networks.
of SDN is the vendor neutrality; the specifications of SDN
is open and standard-based. OpenFlow is a standard interface
first introduced by Stanford University [95] then developed by B. Distributed Antenna Systems
Open Networking Foundation (ONF) to realize the function- A DAS is an architecture where many antenna elements
ality of SDN [96]. (AEs) are geographically distributed and connect to a central
DRAFT 9
BS in order to shorten the distance to the end users [42]. available fiber resources [106]. This suggests that Cloud-RAN
The distributed AEs provide coverage to the nearby users can be used in special scenarios where fiber resources can be
and they are connected to the central BS via high-speed provisioned in a cost effective manner.
low-latency links, fiber or microwave. The reduced distance Unlike dense small cell networks, which are distributed in
between the user and the AE creates a more uniform coverage, nature, Cloud-RAN can be considered as a centralized dense
which in turn reduces the required transmission power in both network where the densification is performed via the dense
the uplink and the downlink. Consequently, a uniform high deployment of RRHs. Although Cloud-RAN and dense small
capacity can be offered to the users in the downlink [102] or networks can be seen as competitors, the coexistence of small
the uplink [103]. However, DASs in general aim at improving cell dense networks and Cloud-RAN is more likely to occur.
the coverage first and then the capacity. Specifically, in some scenarios densification via small cells
Relays, small cells (picocells and femtocells), and DAS are would be better as in the case of homes or office buildings
competing technologies in terms of network densification [42], where fiber resources are not available. In other cases, the
[104]. DAS offers a cheap centralized densification solution densification using cloud-RAN can provide better experience
where the deployment is performed by the operator, and to the end users in crowded areas such as shopping malls,
hence a full coordination is achievable. In small cell networks, concerts, and stadiums [106].
femtocells are usually installed by the subscribers to improve
the coverage and capacity in residential areas, and the picocells D. mmWaves Networks
are installed by the operators in hotspots. Thus, in small cell
The spectrum crunch has emerged as a serious problem in
networks the coordination is likely to be distributed. Compared
this decade. The underutilized spectrum in the millimeter band
to relays, DAS transmit the user signals to the BS via fiber
stems as a potential candidate to avoid the consequences [109],
links, while relays use the wireless spectrum either in-band or
[110]. The spectrum at the 3 − 300 GHz band has a vast
out-of-band.
amount of unused or lightly used spectrum which attracts the
research interests to investigate the propagation properties and
C. Cloud-RAN the device manufacturing technology challenges in this band
[111]–[118]. The propagation characteristics in this spectrum
Cloud-RAN, as the name suggests, stems as an application
differentiate the mmWaves cellular communication from tra-
for the cloud computing in the wireless area [105]. Cloud-
ditional cellular communication. mmWaves signals suffer high
RAN can be seen as a network where the base band resources
penetration losses. Accordingly, it is sensitive to blockage ef-
of many BSs are pooled while the RF processing is left
fects than signals in lower spectrum. This promotes the use of
in the remote coverage area [43]. The RF unit of a BS is
mmWaves BSs for outdoor coverage which leaves the indoor
deployed to provide the coverage via what is called remote
coverage for other means such as mmWaves femtocells or
radio heads (RRHs) where the antenna units are installed.
mmWaves WiFi solutions [119]. Due to the short wavelengths,
The base band resources of a collection of BSs are then
a high antenna gain can be obtained by the use of antenna
pooled in a centralized base band unit (BBU) to be shared
arrays at the transmitter and the receiver [116]. This gives
amongst the BSs [106]. The RRH stands as an interface
rise to the implementation of beamforming with high gains
between the radio and the BBU to perform the main tasks
to compensate for the propagation losses. Consequently, the
of the RF such as filtering, power amplification, digital to
inter-cell interference is reduced which significantly improves
analog conversion (DAC), and analog to digital conversion
the network performance [111].
(ADC). Cloud-RAN can be considered as a dynamic DAS
Moreover, mmWaves networks consist of BSs that transmit
[56], where traditional DASs are static in nature as they cannot
and receive in the mmWaves band and cover a certain geo-
be reconfigured to adapt to the traffic fluctuations in hotspots.
graphical area. To ensure adequate coverage, a dense deploy-
The densification of cellular networks via Cloud-RAN can be
ment of the the mmWaves BSs is required [117]. However,
realized by the deployment of as many RRHs as required to
the backhauling of the dense mmWaves networks is costly
carry the generated traffic by the users [107].
due to the large number of BSs. This suggests that some of
In Cloud-RAN, the pooling of base band processing re-
the mmWaves BSs can connect to the backhaul via other BSs
sources comes with many advantages [108] such as the low
[110].
deployment costs of radio access network via sharing of
resources, the low operation costs via energy saving, the
improvement of the system performance by facilitating coor- E. Massive-MIMO
dinated processing of signals, the adaptation of the signal pro- The deployment of a large number of antennas in the BS is
cessing resources to the actual traffic loads, the optimization usually referred to in the literature as Massive-MIMO [120].
of resource utilization by the dynamic allocation of resources, Other terminology for the same technology includes large-
and the coexistence of multiple standards that can be simply scale antenna systems [121], very large MIMO [122], and
implemented in the central entity. However, these advantages full-dimension MIMO [123]. The excess number of antennas
do not come for free, the RRHs need fast fiber or microwave deployed in a two-dimensional grid [41] helps to focus the
links in order to connect to the central BBU which might transmitted power towards the end users by the accurate beam-
not be easy to achieve. In other words, Cloud-RAN is an forming in both horizontal and vertical planes. In massive-
appealing solution for the operator with free or cheap already MIMO systems, hundreds of antennas are deployed to serve
DRAFT 10
many users at the same time in the same frequency resource G. Multi-RAT
[124]. In UDNs, massive-MIMO emerges as a wireless back- Considerable attention has been given by the operators to the
hauling alternative [30] where many small cells can connect offloading of the cellular traffic to WiFi access points in the un-
to a macrocell to transmit the traffic back to the network via licensed bands or to the small cells in the licensed bands [135].
wireless backhaul links. The competition over the last decade between different Radio
Many challenges are facing the realization of the theoretical Access Technologies (RAT), namely the cellular technology
concepts behind massive-MIMO, the most prominent is the and WiFi, has came to an end and the cooperation stems as
channel knowledge [125]. Since the number of the channel a vital solution [136]. Multi-RAT refers to the coordination
measurements and the resources needed to feedback these between different RATs to provide a high quality service to
measurements both are functions of the number of antennas, the end users. The associated strict performance requirements
the feasibility of perfect channel knowledge in large antenna of the 5G [137] necessitate the exploitation of every single
grids is questionable. A deployment alternative to overcome resource. Correspondingly, the abundance of WiFi nodes with
this situation is to exploit the channel reciprocity in time divi- comparable rates to the current cellular technology inspires
sion duplex (TDD) systems [126]. Thus, the uplink measure- many collaboration techniques where the delay intolerance
ments performed by the cell would provide information about traffic can be offloaded to the WiFi layer [138].
the downlink channel status. The deployment of massive- The RAT selection algorithms and the offloading mecha-
MIMO allows for the use of inexpensive hardware compo- nisms are of a key importance to the fruition of this paradigm
nents in the BSs and mobile equipments replacing the high- [139], [140]. Also, the mobility of the flows across cellular
cost hardware. The orchestration of such massive number of and WiFi access points is challenging, however active in-
antenna elements is challenging in terms of the computational vestigations are in progress [136], [141], [142]. Moreover,
complexity, the development of distributed algorithms, and the the splitting of data across multiple flows in Multi-RAT has
synchronization of such large number of antenna elements. emerged as another challenge. Furthermore, the simultaneous
However, the energy efficiency of massive-MIMO systems and connection to access nodes of different RAT stems as a viable
the throughput scaling makes them a potential competitor to offloading alternative, while different flows with different
small cell networks [127], [128]. Quality of Service (QoS) requirements are carried by the
cellular cells or the WiFi access nodes [143].
F. Device-to-Device Communication (D2D) H. Proactive Caching

Proactive caching is the predictive storing of popular content
D2D communication is a new paradigm in cellular com- in the BSs or the user equipment (UE) to serve the user
munication where a direct connection between two nearby demand to this content in peak traffic loads [144]. The storing
devices can be established with a minimal intervention of the of such content occurs in off-peak periods to alleviate the load
network [129], [130]. The direct flow of data between devices on the wireless and backhaul resources [145]. Advances in
in close proximity significantly improves the efficient use of context-awareness, social networks, storage, secure communi-
resources through densified local reuse of spectrum [2], [131]. cations, and D2D communications have a great influence on
D2D communication falls into two different categories, stand- the potential gain of content caching, and the efficient use of
alone D2D and network assisted D2D [130]. In standalone resources in general [146]–[150].
D2D communication, the connected devices organize the inter- The design of caching schemes to be implemented in dense
communication in ad hoc manner without the intervention of networks requires the understanding of the spatial and social
the network. In the network assisted D2D communication, structure of such networks [144]. Successful prediction of the
the signaling required to establish a data flow between two popular content has a two-fold gain, efficient use of storage re-
devices is initiated in the underlying cell, while the data sources at the network-edge, and achieving the potential gains
transmission flows directly from one device to the other. In of caching. However, the fundamental limits of the predictive
D2D, the network densification can be achieved by increasing caching need to be investigated to assess the role of different
the number of links per unit area where each D2D connection parameters and to promote effective design techniques [147],
reuses the available spectrum. [151]. Moreover, the secure transmission of such content is
D2D communications are challenging in many aspects. The an important aspect to be considered, especially in case of
complex process of discovering optimal peers for pairing in a caching at UE [149].
data flow must be efficient [132]. Also, a given device might
use a D2D link or a cellular link to accomplish certain data
transmission. This kind of decisions is challenging and re- I. Internet of Things (IoT)
quires further investigations [133]. Additionally, the allocation IoT refers to a network of devices with Internet connectivity
of radio resources to the D2D users is an interesting research to communicate directly without human-intervention in order
topic. This allocation might occur in centralized, distributed, to provide smart services to users [152]. The connectivity of
or hybrid manners [134]. Other challenges to the proliferation such devices eases the collaborative decision making through
of D2D communication include power control, interference processing of real-time data and the improved access to
management, and security. information. The achievements in smart sensors, wireless
DRAFT 11
communication technologies [153], context-aware computing, ABSF: Almost-Blank Subframe

cloud technologies, and Internet protocols can be considered No MUEs are scheduled in ABSFs
MUE: Macro User Equipment
as the building blocks of IoT architecture [154]. Also, the PUE : Pico User Equipment time
research and standardization efforts in machine-to-machine
(M2M) communication plays a significant role in the devel-
opment of the state-of-the-art of IoT [155].
Applications in health [156], smart security, and smart cities
found their way to the market and realizes the potential
benefits of this technology [157]. In addition, many other
applications of IoT can be enumerated such as agriculture,
industry, natural resources (water, forests, etc.) monitoring,
transport system design, and military applications [158].
Many challenges are facing the fruition of IoT technology,
namely, addressing and identification of the connected devices, Fig. 6: More users are biased to connect to the small cells via
the standardization of the technology for optimal interoperabil- the range expansion, the users in the expanded area suffer
ity, the privacy of the information, the security of the sensors from high interference from the macrocell which requires
or actuators, the provisioning of wireless backhaul and storage coordination between the macrocell and the small cell to
resources for billions of connected devices, and the energy ef- mitigate this interference. ABSF is one of the time-domain
ficient deployment of edge-devices and communication nodes techniques for such coordination.
[158].
Network densification is considered as an enabler for the
in HetNets follows a load-based association rule, where the
successful diffusion of IoT services and application in the soci-
users are biased to connect to the nearest small cell to offload
ety. The availability of nearby communication links alleviates
their traffic. The small cells are usually lightly-loaded due to
the strict requirements on the wireless network infrastructure.
the limited coverage area, and hence, the association of a given
The deployment of indoor small cells (picocells and femto-
user to the nearest small cell gives the user a higher data rate
cells) stems as an excellent solution for providing connectivity
privilege. The biasing of users to small cells is performed via
to home smart devices and offloads a great portion of the traffic
the virtual expansion of their coverage area, which is termed
from the macro-cellular network.
as range expansion (RE) [160]. A small cell received signal
strength or quality is weighted by a range expansion bias to
V. U SER A SSOCIATION IN UDN attract more users. Consequently, the users associate to the
The association of users to cells is crucial to the optimal small cell with maximum biased signal strength or quality.
operation of cellular networks. The user association (or cell The biased association of a user located at y to a cell in
association) is the selection of a BS amongst several BSs to a multi-tier network via range expansion can be determined
serve a given user. In traditional cellular networks, the user by the downlink received power from BSs measured at the
association is initiated by the user based on the measurement user side. A user is associated to a tier-k BS if it provides
of received signal strength (RSS) of neighbouring cells. A user the maximum biased received power, this can be formulated
always connects to the cell with the maximum RSS (max-RSS as [80]:
association). Correspondingly, another association rule, based
on the signal quality rather than the signal strength, is used −α −α
to associate users to cells. In the signal quality association, a Pk Bk min |xk,i − y| ≥ Pj Bj min |xk,l − y| , ∀j
i l
user is connected to the cell with the largest SINR, which is
(7)
termed as max-SINR association.
where Bk is the bias indicating the RE bias towards tier-
A. Range Expansion k, Pk is the transmit power of BSs in tier-k, and xk,i is
the position of cell i in tier-k. Figure 6 illustrates the range
In HetNets, where there are different types of access nodes expansion concept and the biasing of users to connect to small
with different transmission power and coverage, the associa- cells rather than macrocells. It is also noteworthy to mention
tion based on the signal strength or the signal quality is far that the use of the range expansion to offload traffic to small
from the optimal operation of the network. The deployment cells might not work in dense networks. The range expansion
of small cells in HetNets is intended to offload the traffic is mainly used in co-channel deployment where the macrocells
from macrocells to small cells. However, with the large gap and the small cells share the same spectrum. However, the
in the transmission power (macrocells: 46 dBm, picocells: 33 densification of small cells requires orthogonal deployment to
dBm, and femtocells: 20 dBm), this offload will not happen, the macrocells to mitigate the cross-tier interference.
yielding to the inefficiency of small cells deployment. A user
finds the signal received from a macrocell stronger than that
of a small cell and thus associates to the macrocell [159]. B. Dual Connectivity
Recall that a UDN is a densified HetNet, and hence, the same Another emerging association alternative is the Dual Con-
situation applies in UDN. Consequently, the user association nectivity (DC) [161]–[163]. In DC mode, the user connects to
DRAFT 12
TABLE IV: S IMULATION PARAMETERS OF M ULTIPLE A S -

SOCIATION IN UDN S [46]
Parameter Value User Neighborhood

The downlink of a dense small cell network
Network Model environment where the small cell density λs
is much larger than the users density λu
r3
Idle Mode Capabilities
A BS in the idle mode is turned off to γ
fully mitigate its interference.
Antenna System SISO r1
Pathloss Exponent 4 r2
Fading Channel Model Rayleigh
Mul Cell
Multicell Size 1, 5, and 10
Small cells Power 100 mwatts
User Density λu 100 − 1000 users / Km2
Small cells Density λs 103 − 106 cells / Km2
a macrocell and a set of small cells simultaneously as depicted

in Figure 7. DC mode ensures robustness of the mobility Fig. 8: The user neighborhood can be defined as the set of
management of UEs since they are connected to two cell tier, cells which are in close proximity to a given user, where
macrocells tier and small cells tier. The main usage of DC is this proximity is characterized by the average received signal
the splitting of the user traffic flows to satisfy the user data strength threshold. Accordingly, the MultiCell is a subset of
rate requirements and to distribute the load of users amongst the user neighborhood, thus the user connects to the nearest
a set of small cells in its neighborhood. More interestingly, M cells to form its MultiCell.
the signalling plane and the data plane might be split in the
DC mode. The user signalling is transmitted to the macrocell,
while the traffic is offloaded to one or more small cells. The simulation parameters of the system model are listed in
Table IV. Figure 9 depicts the average spectral efficiency ver-
sus the small cell density (λs ) for the connection of a typical
user to the first, second, and third nearest small cells. The
results show that the average spectral efficiency monotonically
increases with higher small cell density, however the gains are
diminishing. The difference between the spectral efficiency of
the link to the first and second nearest small cell, or the second
and third nearest small cell is independent of the cell density.
Data Secondary Also, the loss in the spectral efficiency due to connecting to
Primary Cell
Signalling Cell farther cells is decreasing with the index of the small cell (first,
second, third, etc.).
We use the ASE as the performance metric to account for
the network capacity. Recall that the ASE is defined as the
Fig. 7: In dual connectivity, a user connects to a (primary) number of transmitted bits per second per Hz. per unit area.
macrocell and one or more (secondary) small cells. The In the context of multiple association, ASE is interpreted as
connection to the primary cell is either signaling or data the sum of the average ergodic rate (i.e., R̄j ) in a MultiCell
connection, while the connection to the secondary cell is data
connection only.
25
Average spectral efficiency (bps/Hz)
Analysis 1st nearest cell nd

2 nearest cell
20 Simulation
C. Multiple Association
The case might be different in UDN where a given user is 15
in the vicinity of many small cells. Hence, the cell selection

mechanism involves a more complicated decision making pro- 10
3rd nearest cell
cess. In [46], we proposed a multiple association mechanism
5
where a user connects to more than one cell to form what
we termed a Multicell. The user connects to the nearest M 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
cells and distributes the traffic amongst them to overcome λs cells/m2
the backhaul and the maximum achievable rate of individual
cells, thus aggregating higher data rate. Figure 8 illustrates the Fig. 9: Average spectral efficiency for different small cell den-
Multicell concept. sity of the connections of the typical user with the first, second,
In the aforementioned investigation [46], we simulate the and third nearest cells (Ps = 20 dBm, λu = 300 users/km2 )
proposed model to assess the accuracy of the analytical results. [46].
DRAFT 13
Area spectral efficiency (bps/Hz/km2 ) ×104 ×104
Area spectral efficiency (bps/Hz/km2 )

4
Multicell Size M = 1
4 Multicell Size M = 1
Multicell Size M = 5
Multicell Size M = 10 Multicell Size M = 5
3 Multicell Size M = 10
3
2
2
1 1
0 0
0 0.1 0.2 0.3 0.4 0.5 100 200 300 400 500 600 700 800 900 1000
2
λs cells/m λu users/km2
Fig. 10: Area spectral efficiency versus small cell density for Fig. 11: Area spectral efficiency versus user density for
different MultiCell size (Ps = 20 dBm, λu = 300 users/km2 ) different MultiCell size (Ps = 20 dBm, λs = 0.05 cells/m2 )
[46]. [46].
multiplied by the number of Multicells per unit area. Since network where they claim that the existing result [11] is not
each user forms a MultiCell by multiple association to many accurate in a general setting, or in other words its accuracy is
cells in its vicinity, the MultiCell density is the same as the conditioned on the considered association scheme (e.g., nearest
users density (i.e., λu ). Hence, the ASE is defined as cell association).
M
X VI. I NTERFERENCE M ANAGEMENT IN UDN
ASE , λu R̄j . (8) The interference management is challenging in densified
j=1 networks. Various types of small cells are deployed with large
Figure 10 illustrates the area spectral efficiency versus small densities to provide the users with very high rate connections.
cells density for different MultiCell size. The results show The use of inter-cell coordination to mitigate the interfer-
a higher ASE for higher MultiCell sizes. This is due to the ence requires increasing signalling overhead due to the large
increase of the number of connections per unit area which number of deployed small cells. Thus, distributed control is
significantly improves the area spectral efficiency. Moreover, preferred to mitigate the interference in UDN.
the ASE in case of single association (M = 1) is invariant for
higher small cells density. This is intuitive since the number of A. Interference Coordination Domains
connections does not change with higher small cells density,
The interference coordination amongst interfering cells
and the gain in spectral efficiency is diminishing. However, in
takes place in the frequency domain, time domain, space
multiple association, the ASE is increasing with the small cells
domain, power domain, or a mix of them. In the frequency
density. This can be explained by recalling that higher small
domain, interference mitigation is done through the use of
cells density brings the cells closer to the user. Accordingly,
orthogonal frequency channels, either by the static allocation
the link quality to the nearest M cells improves significantly
of these orthogonal channels to different cells or by dynamic
with higher densities of the small cells.
allocation [164]. The interference coordination in time domain
The effect of higher user density on the ASE is shown
exploits the blanking of sub-frames. Almost Blank Sub-Frame
in Figure 11 where the area spectral efficiency increases
(ABSF) is a successful proposal that is studied extensively
with the user density. In multiple association, the number
and standardized in LTE Rel10 [165]. In a different front,
of connections increases linearly with the user density. As a
the advances in MIMO systems give rise to the wide use of
result, the ASE improves significantly with higher user density
spatial interference coordination [166]. Also, the power control
for larger MultiCell sizes.
is another method to coordinate the inter-cell interference
Liu et al. [8] comprehensively surveyed the user association
especially in the uplink direction [167]. Figure 12 provides
schemes in 5G networks. They considered the state-of-the-
a summary of the different interference mitigation techniques.
art association mechanisms in three paradigms of the 5G
networks: HetNets, massive MIMO networks, and mmWaves
networks. Lie et al. delved into the aspects of user associ- B. Idle Mode Capabilities
ation in 5G networks where they discussed different mod- In UDN, the interference mitigation stems as a challeng-
elling techniques, performance metrics, and network topology ing issue [26], [27]. That is many BSs become dominant
models. Liu and Wang [47] investigated a general random interferers to each other, and the coordination between them
cell association scheme to study the fundamental correlation requires sophisticated mechanisms. On the positive side, in
between cell association and void cell probability (AKA idle the dense deployment of small cells, it is more likely to find
mode probability). The findings in [47] reveal accurate bounds idle cells i.e., without connected users. Thus, it is desirable
for the idle mode probability in a PPP modelled cellular to turn off such cells to partially or completely mitigate
DRAFT 14
TABLE V: C OMPARISON OF D IFFERENT I NTERFERENCE M ANAGEMENT S TUDIES

Coordination Involved Signalling Antenna
Ref. Direction Model Control Clustering
Domain Interferers Overhead System
Stochastic BSs within
[25] Downlink Space Distributed N/A Low MISO (M × 1)
Geometry the Nulling Range
[26] Downlink Simulation Frequency Static Planning N/A All BSs - MIMO (2 × 2)
Frequency
[27] Downlink Simulation Distributed Non-overlapped Dominant Interferer Low SISO
and Time
First and second
[28] Downlink Simulation Frequency Centralized N/A Moderate MIMO (2 × 2)
dominant interferers
[48] Downlink Simulation Space Distributed N/A Strongest interferers - MIMO (M × M )
[49] Downlink Game Theory Frequency Distributed N/A All interferers Low SISO
[50] Downlink Game Theory Frequency Distributed Overlapped All interferers Low SISO
[51] Downlink Game Theory Power Distributed Non-overlapped All interferers in a cluster - SISO
Distributed
[52] Downlink Game Theory Frequency Non-overlapped All interferers in a cluster Low SISO
and Centralized
[53] Uplink Simulation Power Distributed N/A All interferers Low SIMO (1 × M )
T1 T2 T3 interference coordination. They considered four interference

T1 T2 T3
scenarios, and they highlighted the role of the dominant
interferer (DI). On the other hand, Polignano et al. [26]
addressed the inter-cell interference dilemma in dense outdoor
small cell networks. The impairment to the user experience
Different Frequencies Different Time Slots due to inter-cell interference is investigated to evaluate the
conditions that make the interference coordination preferred
to the universal frequency reuse. In a different direction, an
enhanced dynamic cell muting scheme (eDCM) is proposed by
Wang et al. in [28] where the authors exploit the Coordinated
MultiPoint (CoMP) framework of LTE Rel-11 to develop a
dynamic muting mechanism to mute some resources at certain
small cells for the benefit of other small cells. In their work,
the inter-cell coordination function to generate the muting
Different Space patterns considers a benefit metric for individual users and a
f1 f2
proportional fairness scheduling amongst all users. Moreover,
Fig. 12: Alternatives of inter-cell interference coordination.
the enhanced muting technique considered not only the first
dominant interferer, but also the second dominant interference
their interference to the neighbouring cell. This is so called source.
idle mode capabilities of small cells [23]. The activation of In a MIMO setting, Tavares et al. [48] utilized MIMO
smart idle mode capabilities in dense small cell networks is spatial multiplexing to mitigate the inter-cell interference in
another interference mitigation alternative. In that, a small cell small cell networks. Rather than the traditional frequency
supported with idle mode functionality can reduce its power reuse planning (FRP), they proposed maximum rank planning
consumption to the minimum (device power consumption) (MRP). In their proposal, the reduction of spatial multiplexing
while no users is connected to it. streams increases the probability of higher degree-of-freedom
Ashraf et al. [168] discussed three different techniques to for the interference rejection receivers to reject the dominant
control the idle (sleep) mode of small cells, namely, small interference. In a different setting, Li et al. investigated a
cell driven, core network driven, and UE driven. In the first user-centric inter-cell interference nulling (ucICIN) scheme
approach, the small cells are equipped with a low-power sniffer [24], [25]. In their work, the authors derived an approximate
capability in order to allow for the detection of active users expression for the success probability of a typical user. In this
in its coverage area. In this event, the small cell is activated model, a user requests the suppression of the interference of
or deactivated depending on the presence of active users. In some neighbouring BSs based on their relative distance to the
the core network driven technique, turning the small cell on serving BS. Moreover, the simulation results confirmed the
or off is controlled by wake-up control messages which are accuracy of the approximation to some extent. Furthermore,
sent from the core network via the backhaul. Finally, in the the authors studied the effect of limited channel state infor-
UE driven approach the UE can broadcast wake-up signals in mation (CSI) on the performance of the interference nulling
order to activate the small cells in its vicinity. technique.
Using game theory concepts, Al-Zahrani et al. [49] ex-
C. Research Status and Findings ploited game-theoretic approaches to investigate the cross-tier
Various downlink interference mitigation techniques are and co-tier interference management. To overcome the curse of
investigated by Soret et al. [27] where two algorithms are dimensionality in case of dense networks, the authors modelled
proposed, a time domain and a frequency domain small cell their problem using mean-field game theory. In mean-field
DRAFT 15
game theory [87], a player takes an action based on the average TABLE VI: A LTERNATIVES OF S MALL C ELLS BACKHAUL
of the effect of other players’ actions rather than the individual [169]
effects. In another setting, Liu et al. [50] exploited game Backhaul Backhaul Latency
Throughput
theoretic approaches combined with graph-coloring algorithms Category Technology (one way)
Non-Ideal Fiber Access 1 10-30ms 10 Mbps-10 Gbps
to model a joint CoMP clustering and inter-cell resource Non-Ideal Fiber Access 2 5-10ms 100-1000 Mbps
allocation for interference mitigation. A scalable algorithm is Non-Ideal Fiber Access 3 2-5ms 50 Mbps-10 Gbps
proposed to account for the large number of cells in a dense Non-Ideal DSL Access 15-60ms 10-100 Mbps
network. The distributed two-step algorithm is evaluated and Non-Ideal Cable 25-35ms 10-100 Mbps
10-100 Mbps
potential performance improvements are concluded. A power Non-Ideal Wireless Backhaul 5-35ms typical, maybe up
allocation algorithm is proposed by Yuehong et al. [51] to to Gbps range
coordinate the interference especially for the benefit of edge Ideal Fiber Access 4 less than 2.5 us Up to 10Gbps
users. Non-cooperative game theory is applied to find the
Nash Equilibrium for the power allocations in the downlink Core Network
of a dense network. A non-cooperative game is formulated
by Sun et al. [52] to investigate the role of cluster-based ling through
spectrum allocation and CoMP transmission to mitigate the e MIMO or
sever interference in UDNs. In this investigation, the authors Server Waves links
consider the load condition of the cells to associate the users
to the best cell in order to optimize the network performance. eal Backhaul
In the uplink direction, Cho et al. [53] considered a power
control scheme designed for interference management in a Ideal Backhaul
time-division duplex (TDD) setting where the individual users Cable link
tune their transmission power to keep a preset interference
threshold to other BSs. Consequently, each BS schedules the Internet
users having the best normalized channel gains according to
Backhauling through Relay Backhaul
the corresponding transmission power.
Internet connec ons
It is important to note however, that the investigation of DSL link
effective clustering techniques is crucial to the implementation
of efficient interference management schemes in the scenario
Fig. 13: Alternatives of backhauling in UDN.
of UDN. A learned lesson from the above studies is that
the scalability of a solution depends greatly on the clustering
technique such that the signalling overhead between collab- One of the main challenges for the dense deployment of
orating nodes remains within practical limits. A comparison small cells is backhauling [7]. The promised radio interface
for the above-surveyed interference management techniques is capacity of the small cells might be bottlenecked by the wired
provided in Table V. or wireless backhaul capacity. To explain, the association of
a user to a single small cell limits its maximum achievable
VII. BACKHAULING IN UDN data rate to the backhaul capacity of this cell. Moreover, the
The backhauling is the transmission of data from a BS back cloud-computing trend and the bandwidth-hungry applications
to the core network, either by a direct link to the core network accelerate the need to even higher data rates than what could
or via Internet connections. The backhauling of dense small be offered by a single cell. This motivates us to propose the
cell networks has emerged as a bottleneck of their successful multiple association scheme as a solution to distribute the
deployment. The increasing number of deployed small cells traffic load of the user to multiple small cells in the user’s
and the lack of ideal backhaul links would be limiting factors neighborhood [46].
of the network densification gains.
B. Research Status and Findings
A. Backhauling Alternatives Wang et al. surveyed the backhauling solutions for 5G small
Different backhaul technologies with different capabilities cells from the perspective of radio resource management [170].
are available to small cell networks [30]. The backhauling They discussed the relation between the emergent backhauling
technology is either wired or wireless. Moreover, the wired solutions and some radio resource management (RRM) issues
backhaul can be categorized as ideal with very high throughput including, but not limited to, cell association, interference
and very low latency, or non-ideal with moderate throughput management, scheduling, and inter-cell coordination.
and latency. Different from the wired backhaul, the wireless In [32], Ge et al. studied the throughput and energy effi-
backhaul is always non-ideal. However, it might be the feasible ciency of 5G wireless backhaul networks. Specifically, they
solution in hyper-dense networks. Figure 13 depicts the various adopted two traffic models, namely centralized backhauling
backhauling alternatives in small cell deployment scenarios. model and distributed backhauling model. In the centralized
For the ease of comparison, the technical details of different model, a given macrocell aggregates the traffic of the small
backhauling technologies are summarized in Table VI. cells in its coverage area assuming an ideal backhaul link
DRAFT 16
between the macrocell and each small cell. This assumption is A. Research Status and Findings
rather ideal in UDN environments since the large number of
small cells in the coverage area of a macrocell makes it almost Björnson et al., by the aid of a stochastic geometry model,
impossible to provide the small cell tier with ideal backhaul studied the energy efficient deployment of dense small cell
links. In the distributed model, a small cell which is connected networks [54], [70]. In their work, they considered the uplink
to the core network via fiber-to-the-cell (FTTC) link collects of a multi-cell multi-user MIMO network. The closed-form
the traffic of the small cells in its vicinity through mmWaves expressions obtained via solving the energy efficiency maxi-
communication links. mization problems shed lights on the role of all the considered
parameters and optimization variables. In a different setting,
The wireless backhauling, although non-ideal, emerged as
Li et al. [55] modelled the downlink of a dense multi-
a viable solution for the backhauling in dense small cell
transmission antennae small cell network to quantify the per-
networks. Amin et al. [33] studied the performance of self-
formance in terms of ASE and energy efficiency. A tractable
backhauled small cells. In their model, a Long Term Evolu-
expression for the outage probability is derived via stochastic
tion (LTE) macrocell backhauls High Speed Packet Access
geometry approach and then exploited in the computation of
(HSPA) small cells. A massive MIMO backhauling solutions
the ASE and the energy efficiency. In this, Liu et al. assessed
were investigated in [30], [31]. Also Chen et al. [29] proposed
the effect of deploying more BSs and more transmit antennas
a hierarchical network model to investigate the backhauling in
per BS on the aforementioned performance metrics. Moreover,
small cell networks, where they considered both wired and
the optimal BS density and the optimal number of antennas
wireless backhaul. They derived analytical expressions for the
per BS to optimize the energy efficiency is computed.
backhaul delay and the average delay seen by a typical user
considering two scenarios, namely, static (i.e., no mobility A comparison between three network densification scenarios
case), and extreme mobility case. is conducted by Yunas et al. [56]. These scenarios are the
densification of the outdoor macrocell tier, the densification of
the indoor femtocell tier, and the densification of DASs. The
VIII. E NERGY E FFICIENCY IN UDN results of the study confirmed that the resources efficiency in
terms of spectrum and energy is much higher in the second and
In this section, we consider the energy efficiency of network third densification strategies, compared to the first scenario.
densification. In that, the global warming phenomenon attracts Consequently, the rule of thumb in network densification is
significant attention to the efficient use of energy in commu- to densify the small cells, not the macrocells. Moreover, they
nication networks, especially wireless networks [171], [172]. found out that the densification of indoor femtocells not only
That is the power consumption by a BS falls in two categories, improves the indoor network capacity, but also improves the
the first is the node power consumption and the second is outdoor network capacity in case of open access mode (i.e., the
the communication power consumption. In particular, the node indoor femtocells are open to serve outdoor users in their
power consumption is due to signal processing, cooling, and vicinity). Finally, the results obtained verified the efficacy of
battery backups as well. On the other hand, the communication the dense deployment of dynamic DAS in terms of resource
power consumption is the transmitted power to achieve a efficiency.
certain coverage. Consequently, the ratio between the total Another stochastic geometry analytical study considered the
network throughput and the total network power consumption performance of UDN in terms of energy efficiency along
defines the energy efficiency of the network. Hence, the energy with cell average spectral efficiency and area throughput
efficiency of the network with the units of bits/joule signifies [57]. Ren et al. [57] concluded that the three performance
to what extent a given procedure, algorithm, or technique is indicators cannot be optimized simultaneously, and a tradeoff
more efficient in the sense of energy consumption than others. is unavoidable to meet the required network performance.
This energy efficiency ηEE is defined as [173] Using game-theoretic approach, Samarakoon et al. [58],
[59] investigated a joint power control and user scheduling
T in dense scenario to optimize the energy efficiency. They
ηEE = (9) formulated a dynamic stochastic game and analyzed the mean-
P
field equilibrium. In another venue, Yang [60] studied three
where T is the effective throughput of the network measured important performance metrics in a dense network setting:
in bits per second, and P is the total power consumption of spectrum, energy, and cost efficiencies. The author formulated
the network measured in watts. a Nash-product form of the corresponding utility function,
In dense small cell networks, the energy efficiency is a vital and analyzed the tradeoff equilibrium amongst the considered
consideration. The immense number of small cells, despite metrics
the small transmit power of each, would consume a massive In order to compare the findings for different studies, we
energy. The environmental impact of this energy consumption generated the results in Figure 14 based on a normalized
is an interesting factor to consider in the deployment of dense simulation setup for the investigations in [55] and [57]. In
networks [174]. On the positive side, most of the consumed this simulation setup, we consider a path loss exponent of
power in small cells is used to provide coverage to the 4, a SINR threshold of 1, and UE density of 1000 UE/km2 .
subscribers. Thus, there is no need for cooling which consumes Both studies consider downlink dense network models, the
a large portion of the power budget of a given macrocell. first study [55] investigates the effect of multiple transmit
DRAFT 17
0.25
(eICIC) [20], require fast and efficient discovery of small
×10-3 cells. In that direction, Prasad et al. [62] evaluated disparate
Energy Efficiency (bit/J/Hz)
Analytical (M = 1) [43]
0.2
6 Analytical (M = 3) [43] cell discovery mechanisms especially designed for energy-
Analytical (M = 5) [43]
4
Simulation (M = 1)[43]
efficient detection of small cells. A graph coloring based
0.15
2 Simulation (M = 1)[43]
Simulation (M = 1)[43]
scheme for small cell discovery is proposed and evaluated by
0 0.005 0.01 0.015 Simulation (Femto) [45]
Simulation (Pico) [45]
Shuai et al. [63]. In this scheme, the small cells in the same
0.1 Simulation (Micro) [45] vicinity are clustered into disjoint groups and each cluster
takes a turn to transmit the synchronization sequence. Only
0.05 minimal changes to the conventional synchronization scheme
are required, and thus guaranteeing backward compatibility.
0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
The scheme improves the detection probability of small cells.
BS density λb (BS/m2 )
Fig. 14: Comparison between different energy efficiency stud- B. Spectrum Sharing, Resource Management, and Scheduling
ies.
The dense deployment of indoor and outdoor small cells
requires the provisioning of a new spectrum to alleviate the
antenna in a multi-input single-output (MISO) system along interference. The spectrum sharing thus stems as a viable
with the BS density on the energy efficiency of the network. solution. In spectrum sharing, the UDN cells are allocated a
On the other hand, the second study [57] considers the energy spectrum as secondary users in a cognitive network regime.
efficiency of the densification of three different BS types Another alternative, is the inter-network spectrum sharing
namely, microcells, picocells, and femtocells in a single-in where the spectrum is shared amongst multiple operators
single-out (SISO) setting where each BS and each UE is [177]. Another key aspect is the multiple access and resource
equipped with a single antenna. management in dense small cells. In spite of the small proba-
Figure 14 shows the network energy efficiency with dif- bility of having multiple users in the coverage area of a small
ferent BS density. Two important conclusions can be drawn cell in a dense network, still there would be a chance that
from theses results. Firstly, the densification of small cells many users are served by a small cell in a given hotspot.
is more energy-efficient than the densification of macrocells. The authors in [64] studied the spectrum sharing for UDN in
Evidently, the densification of macrocells is not efficient from the radar bands. They modelled a primary/secondary spectrum
neither the cost nor the operation point of view. Secondly, the sharing scenario, where the primary system is the radar system
densification of the indoor femtocells or the outdoor picocells and the secondary system is the dense small cell network.
is highly efficient in terms of energy efficiency, and there They developed deployment regulations, namely, area power
would be a tradeoff between the maximization of EE and regulation and deployment location regulation, and studied
the optimization of other parameters particularly, the cell its effectiveness in different environments. Different from the
throughput, the ASE, and the coverage probability. former cognitive radio regime, Teng et al. [65] considered the
inter-network spectrum sharing, in particular the co-primary
IX. OTHER A DVANCEMENTS IN UDN spectrum sharing. In co-primary spectrum sharing, two or
more operators pool their licenses to achieve flexible spectrum
In the previous sections, we reviewed the state-of-the-art
sharing amongst their network nodes which is co-located but
research in four other directions, specifically, user association,
with only relative displacement.
backhauling, interference management, and energy efficiency.
Stefanatos and Alexiou [12] studied the effect of multiple
These research directions are of a great importance to the
access and the density of BSs on the performance of a dense
proliferation of UDNs, thus a serious and extensive research
network scenario. They also derived a lower bound for the
has to be conducted to complement the existing results with
optimal number of bandwidth partitions and a closed-form
more findings and insights of the corresponding problems. In
upper bound for the BS density to guarantee an asymptotically
this section we discuss other research directions which also
small rate outage probability.
are considered very relevant to the advancement of UDNs. In
In terms of resources management, Jafari et al. [66] studied
the following subsections, a discussion is presented for these
the performance of different scheduling techniques. In partic-
research directions particularly, small cell discovery, spectrum
ular, they compared the performance of proportional fair (PF)
sharing, RRM, and scheduling, propagation modeling, and the
scheduler and round robin (RR) scheduler. The key aspect
economy of UDN deployment.
of their model is to consider the LOS transmission which
is more probable in dense networks. Furthermore, Chen et
A. Small Cell Discovery al. [67] investigated a distributed spectrum resource allocation
The fast discovery of small cells in dense networks is and proposed a learning algorithm. The algorithm is proven
another emerging research direction. Due to the large number to converge to Nash equilibrium and the performance results
of small cells in the vicinity of a user, there is a need for asserts that co-tier and cross-tier interference is mitigated.
optimized cell discovery mechanisms. Many features in UDN, The throughput performance of the proposed system model is
such as idle mode capabilities [23], CoMP [175], load balanc- investigated and potential performance improvements verified
ing [176], and enhanced Inter-cell Interference Cancellation by simulations.
DRAFT 18
C. Propagation Modelling network and studied the performance in terms of coverage

The propagation modelling is a vital part of the investiga- probability and throughput.
tion of wireless communication problems. Since the network
densification brings the access point closer to the users, the D. Economy of UDN Deployment
LOS transmission components become most probable. Hence, The densification of access nodes is shown to have a
the study of dense networks requires a different propagation significant impact on the performance of cellular networks.
model where the LOS transmission is considered. Also, in However, the network cost is an important factor to consider.
dense indoor networks, the cells are deployed in buildings with The cost of network densification can be categorized to
many floors, thus a three-dimensional propagation environ- capital expenditures (CAPEX) and operating expenses (OPEX)
ment should be considered. This suggests two major modifica- [14]. The CAPEX includes all deployment costs, e.g., access
tions to the traditional propagation modelling in macrocellular nodes hardware, backhauling, and core network equipments.
networks; the consideration of dual-slope or multi-slope path The OPEX costs include energy, backhaul transmission, site
loss model while the path loss exponent becomes a function of rentals, and operation and maintenance (OAM). The sharing
the distance to the user, and employing Rician fading channel of these costs between the operators and users is one of the
instead of the simple Rayleigh fading model [66]. main advantage of UDN since the deployment of small cell
In order to acquire accurate modelling for the propagation BSs in homes, offices, and markets is usually done by the
environment in dense networks, the consideration of dual-slope users while the deployment in the hotspots and outdoors is
or multi-slope becomes inevitable. In dual-slope propagation done by operators.
model, the propagation path loss l(r) is expressed using two In small cell networks, the CAPEX and OPEX are much
different path loss exponents, and is given by [34] less than the corresponding ones in traditional networks.
( The access points are much cheaper, and the backhauling is
r−α0 , for r ≤ Rc
l(r) = (10) made easier by the customer’s DSL or cable modem [19].
ηr−α1 , for r ≥ Rc Also, the transmission power of each node is small and the
where Rc is the critical distance, α0 and α1 are the close- OAM functions are automated via self organizing network
in path loss exponent and the long-range path loss exponent, functionality (SON) [178]. However, the densification of small
respectively. Also, it is assumed that 0 ≤ α0 ≤ α1 , and η = cell networks increases the energy consumption costs and the
Rcα1 −α0 is a constant to provide continuity. In the same way, backhaul cabling costs in case of outdoor deployments. Thus,
multi-slope path loss model can be expressed in terms of many the network density should be optimized to make the best of
path loss exponents indexed with n ∈ {1, 2, · · · , N } [15]: this paradigm.
(
−αL
Lee et al. [13] are the first to address the economy aspects of
lnL (r) = AL
nr
n, for LOS deploying many BSs in a cellular network. Using a stochastic
ln (r) = N L N L −αN L (11)
ln (r) = An r n , for NLOS model for the downlink of a dense cellular network, they
quantified the outage probability and found its relation to
where lnL (r) and lnN L (r) are the n-th piece of the path
the BS density. More importantly, they derived the optimal
loss function for LOS transmission and NLOS transmissions,
BS density to minimize a multi-objective cost function that
respectively. ALn and An
NL
are the path losses at a reference
accounts for the network deployment economy. The cost
distance r = 1, αn and αnN L are the path loss exponents for
L
function considers the hardware costs, backhaul cabling, and
the LOS and the NLOS cases, respectively.
energy consumption, i.e., a mix of OPEX and CAPEX. The
Ding et al. investigated a dense small cell network consid-
derived result shows a tradeoff between the performance gain
ering LOS transmission in [15]. Despite their assumption of
in terms of the outage probability and the network cost.
a fully-loaded network λu ≥ λb , the study reveals the impact
In [14], Park et al. modelled a problem to investigate
of network densification on the coverage probability which
the relation between the downlink average spectral efficiency
consequently affects the ASE. Furthermore, they presented a
and the BS density in a dense network setting. Furthermore,
clear definition for what is a UDN in terms of the access
they derived a closed-form for the optimal BS density and
point density and suggested a minimum density for dense
utilized spectrum to maximize the operator profits. Park et
networks to provide a linear ASE with the increasing BS
al. also considered a profit function that takes into account
density. In another study [35], Galiotto et al. further studied the
the user demand along with the access points and spectrum
effect of LOS and Non-Line of Sight (NLOS) transmissions
operating costs. Additionally, they provided guidelines for
on the performance of UDN. Different from [15], the authors
the operator to make the best profit by investing either in
in [35] considered both fully-loaded and partially-loaded net-
network densification or acquiring more spectrum based on
work setups. Furthermore, they suggested a frequency reuse
user demands.
mechanism to reduce the impact of network densification in
both the coverage probability and the area spectral efficiency.
Also, they studied the energy efficiency in the provided UDN X. C HALLENGES AND O PEN P ROBLEMS
setup under the assumption of LOS transmission. The significant impact of network densification on wire-
Progressively, Gupta et al. focused on the modelling of less network performance in terms of coverage, throughput,
three-dimensional dense networks [34]. In their work, the au- and area spectral efficiency has attracted the attention of
thors considered a vertically stacked horizontally infinite dense researchers and practitioners. Network densification amongst
DRAFT 19
other proposed techniques such as mmWaves and massive the study of the effect of the wireless backhauling on
MIMO are the main players in shaping the 5G wireless net- the user experience in a dense network environment.
works. In this survey, we considered the dense deployment of Another open problem is the consideration of realistic
small cell BSs and studied the state-of-the-art of the research traffic distributions and user distributions to evaluate the
work conducted in this area. performance of wireless backhauling networks in UDN.
In this section, we consider the challenges facing the suc- • Interference Management Interference management is
cessful deployment of dense networks, and hence the open of a predominant influence on the operation of a dense
problems for extensive research. In [7], the authors listed the network. Imagine a wireless network with immense num-
main challenges facing UDN, and they explained the relevance ber of cells that operate in co-channel scenario. Undoubt-
of each challenge to the practical deployment of dense small edly, the interference could be the limiting factor on the
cell BSs. Different from this, we delve into the research fruition of such dense network dispelling the densifica-
challenges that have not been addressed and we discuss the tion gains. The coordinated interference management is
open problems that require further investigations. challenging with such a large number of neighboring
cells. The curse of dimensionality arose uniquely in dense
• User Association The user association has been studied networks while considering collaborative interference
extensively in traditional networks and HetNets. How- management. The consideration of idle mode capabilities
ever, in the scenario of dense networks, there are unique in modelling interference problems in UDN would be
challenges that need to be considered and accurately another interesting problem. The performance evaluation
investigated. The drastic interference between the nearby of proactive turning off of lightly-loaded dominant inter-
cells due to the LOS components requires proposing ferers could yield interesting results. The reduced distance
of novel association rules to exploit the idle mode between the cells in the vicinity of the same user makes
capabilities of the small cell BSs. The backhauling is the interferes as strong as the servers due to the LOS
another interesting factor that must be considered while components, and this uniquely challenges the interference
associating a user to a cell. Another challenge to the management in dense networks. A multi-domain interfer-
association of users to cells in a dense network is the ence management is another interesting problem, where
mobility management, where fast users would generate the interference management is performed in frequency,
many handover events if they are associated to cells with time, space domains simultaneously. Also, the consider-
small coverage area. Effective collaborative-based solu- ation of realistic user and traffic distributions although
tions are required to account for these unique challenges. beneficial, but still an open problem to consider in dense
Another important open problem is the applicability of networks.
range expansion in dense scenarios where the interference • Energy Efficiency The power consumption plays a main
would be a limiting factor. The common understanding is role in specifying the OPEX of a dense network. In spite
that 5G networks would be a mix of many radio access of the small footprint of a small cell, the aggregate con-
technologies (RATs) such as cellular networks, WiFi, sumed power of a large number of such cells is immense.
and mmWaves networks. This introduces another research Energy efficiency refers to the number of transmitted bits
venue which is the multiple association of a user to many per unit energy. Thus, increasing the energy efficiency
cells in individual or in different RATs. To explain, a conflicts with the link quality, and hence the QoS. The
user might connect to many small cells in a cellular maximization of energy efficiency considering the user
network, or to a cell in a cellular network, to a WiFi experience is an interesting model to be investigated in
access point, and to a mmWaves cell simultaneously. In UDN. Another setting which has a great impact on the
[46], we investigated a multiple association setting to successful deployment of dense networks is the energy
lay out a basic mathematical model and to understand efficient wireless backhauling. Thus considering a joint
the insights of such user association advancement. The backhaul-aware energy efficient association of users to
backhaul-aware association of users are considered in cells in a dense network would yield interesting results.
HetNets [179], but in UDN, the study of backhaul- • Small Cell Discovery The detection of cells in close
aware association is still an open problem. Moreover, the proximity of a given user in a cellular network is crucial
consideration of Quality-of-Service (QoS), and Quality- to the optimal operation of the network. However, this
of-Experience (QoE) in dense networks is missing in the becomes more important and much harder in a densified
current published work, although it is very relevant to the network. Many small cells are in the vicinity of a user
admission of a given user to a given cell. and the efficient detection of them is not an easy task.
• Backhauling The backhauling is identified as the bot- The main challenge in this context is how to manage the
tleneck for the wide deployment of dense networks. reuse of synchronization signals in neighbouring small
The provisioning of ideal backhaul to all small cells cells, which are in the interference range of each other,
in a dense network is challenging. Accordingly, the in order to ease the cell discovery task. Optimization of
wireless backhauling emerges as a viable alternative. cell discovery in terms of time and energy-efficiency is
There are many wireless backhauling techniques includ- an open problem in UDN scenarios. Moreover, the ex-
ing mmWaves links, relays links, and massive-MIMO ploitation of location data and fingerprints in optimizing
backhaul links. Certainly, one of the open problems is small cell discovery [62] is an interesting direction to
DRAFT 20
investigate. [2] N. Bhushan, J. Li, D. Malladi, R. Gilmore, D. Brenner, A. Damn-

• Propagation Modelling Another open research direction janovic, R. T. Sukhavasi, C. Patel, and S. Geirhofer, “Network den-
sification: The dominant theme for wireless evolution into 5G,” IEEE
in the study of dense networks is the consideration of 3D Commun. Mag., vol. 52, no. 2, pp. 82–89, Feb. 2014.
channel modelling. Also, the consideration of multi-slope [3] I. Hwang, B. Song, and S. Soliman, “A holistic view on hyper-dense
path loss models requires further investigation in different heterogeneous and small cell networks,” IEEE Commun. Mag., vol. 51,
no. 6, pp. 20–27, Jun. 2013.
densification contexts. The modelling of channel fading [4] J. Zander and P. Mahonen, “Riding the data tsunami in the cloud:
to account for the propagation characteristics in UDN Myths and challenges in future wireless access,” IEEE Commun. Mag.,
suggests the use of Rician fading model which requires vol. 51, no. 3, pp. 145–151, Mar. 2013.
further rigorous investigation. [5] Cisco, “Cisco visual network index: Global mobile traffic forecast
update 2014-2019,” White Paper, May 2015.
[6] J. G. Andrews, X. Zhang, G. D. Durgin, and A. K. Gupta,
“Are we approaching the fundamental limits of wireless network
XI. C ONCLUSION densification?” CoRR, vol. abs/1512.00413, 2015. [Online]. Available:
http://arxiv.org/abs/1512.00413
The race towards 5G networks is fueled by the forecasts [7] D. David López-Pérez, M. Ding, H. Claussen, and A. Jafari, “Towards
of the imminent traffic. Network densification is a prospective 1 Gbps/UE in cellular systems: Understanding ultra-dense small cell
deployments,” IEEE Commun. Surveys Tuts., vol. 17, no. 4, pp. 2078
winner in this race and is expected to provide the required – 2101, Fourth quarter 2015.
capacity and performance for the 2020 networks. In this [8] D. Liu, L. Wang, Y. Chen, M. Elkashlan, K. K. Wong, R. Schober, and
survey, we presented an introduction to ultra-dense networks. L. Hanzo, “User association in 5G networks: A survey and an outlook,”
Also, we reviewed the recent achievement in different research IEEE Commun. Surveys Tuts., vol. PP, no. 99, pp. 1–1, 2016.
[9] X. Lu, P. Wang, D. Niyato, D. I. Kim, and Z. Han, “Wireless networks
directions to develop an understanding of the state-of-the- with RF energy harvesting: A contemporary survey,” IEEE Commun.
art. More research activities are required to account for the Surveys Tuts., vol. 17, no. 2, pp. 757–789, Second quarter 2015.
fundamental differences of the UDN from traditional and [10] A. Gotsis, S. Stefanatos, and A. Alexiou, “Ultra dense networks: The
new wireless frontier for enabling 5G access,” IEEE Veh. Technol.
HetNet cellular networks. Mag., vol. PP, no. 99, pp. 1–1, 2016.
Moreover, further in-depth investigations are required to [11] S. M. Yu and S.-L. Kim, “Downlink capacity and base station density
layout a concrete understanding of the control parameters to in cellular networks,” in Proc. 11th Int. Symp. Modeling Optimization
in Mobile, Ad Hoc Wireless Networks (WiOpt), 2013, Tsukuba, Japan,
optimize the operation of a dense network. The tradeoffs be- May 2013, pp. 119–124.
tween different performance metrics such as the area spectral [12] S. Stefanatos and A. Alexiou, “Access point density and bandwidth
efficiency, the energy efficiency, the coverage/outage probabil- partitioning in ultra dense wireless networks,” IEEE Trans. Commun.,
ity, and the network throughput require the understanding of vol. 62, no. 9, pp. 3376–3384, Sept. 2014.
[13] S. Lee and K. Huang, “Coverage and economy of cellular networks
the individual effect of the controlling parameters specifically with many base stations,” IEEE Commun. Lett., vol. 16, no. 7, pp.
the density of BSs, the number of antennae per BS, the 1038–1040, Jul. 2012.
transmission power, and the idle mode capabilities. [14] J. Park, S.-L. Kim, and J. Zander, “Asymptotic behavior of ultra-dense
cellular networks and its economic impact,” in Proc. IEEE Global
The fundamental differences in UDNs encourage the con- Communications Conf. (GLOBECOM), 2014, Austin, TX, Dec. 2014,
sideration of relevant and more appropriate propagation mod- pp. 4941–4946.
elling techniques. The decreasing distance between adjacent [15] M. Ding, D. Lopez-Perez, G. Mao, P. Wang, and Z. Lin, “Will the
area spectral efficiency monotonically grow as small cells go dense?”
cells increases the probability of LOS transmissions, and thus in Proc. IEEE Global Communications Conf. (GLOBECOM), 2015,
the need for multi-slope propagation model (e.g., different path San Diego, CA, Dec. 2015, pp. 1–7.
loss exponent at different distances). Accordingly, and as a [16] J. Deissner and G. Fettweis, “A study on hierarchical cellular structures
with inter-layer reuse in an enhanced GSM radio network,” in Proc.
consequence for the same reason, the commonly used fading IEEE Int. Workshop on Mobile Multimedia Communications (MoMuC
channel model in traditional networks (Rayleigh model) is no ’99) , 1999, San Diego, CA, Nov. 1999, pp. 243–251.
more accurate in the context of UDN. The consideration of [17] J. Boccuzzi and M. Ruggiero, Femtocells: design and applications.
Rician and Nakagami-m fading models become a necessity McGraw-Hill, 2011.
[18] J. Andrews, H. Claussen, M. Dohler, S. Rangan, and M. Reed,
in dense networks and densified cells equipped with MIMO, “Femtocells: Past, present, and future,” IEEE J. Sel. Areas Commun.,
respectively. vol. 30, no. 3, pp. 497–508, Apr. 2012.
Another challenge in the accurate modelling of UDNs is [19] A. Damnjanovic, J. Montojo, Y. Wei, T. Ji, T. Luo, M. Vajapeyam,
T. Yoo, O. Song, and D. Malladi, “A survey on 3GPP heterogeneous
the consideration of vertical densification where the small networks,” IEEE Wireless Commun., vol. 18, no. 3, pp. 10–21, Jun.
cell BSs are densified in the elevation plane. In this context, 2011.
the 3D modelling of the dense cellular network is crucial to [20] D. David López-Pérez, I. Guvenc, G. de la Roche, M. Kountouris,
T. Quek, and J. Zhang, “Enhanced intercell interference coordina-
the performance evaluation of such densification alternative. tion challenges in heterogeneous networks,” IEEE Wireless Commun.,
Numerous challenges and open problems are presented and vol. 18, no. 3, pp. 22–30, Jun. 2011.
discussed in this survey in different disciplines to identify the [21] M. Peng, Y. Li, Z. Zhao, and C. Wang, “System architecture and key
research gaps for researchers and practitioners in this active technologies for 5G heterogeneous cloud radio access networks,” IEEE
Network, vol. 29, no. 2, pp. 6–14, Mar. 2015.
area. [22] J. Andrews, “Seven ways that HetNets are a cellular paradigm shift,”
IEEE Commun. Mag., vol. 51, no. 3, pp. 136–144, Mar. 2013.
[23] H. Claussen, I. Ashraf, and L. T. Ho, “Dynamic idle mode procedures
R EFERENCES for femtocells,” Bell Labs Technical Journal, vol. 15, no. 2, pp. 95–116,
2010.
[1] P. Pirinen, “A brief overview of 5G research activities,” in Proc. 1st Int. [24] C. Li, J. Zhang, and K. Letaief, “User-centric intercell interference
Conf. on 5G for Ubiquitous Connectivity (5GU), 2014, Akaslompolo, coordination in small cell networks,” in Proc. IEEE Int. Conf. Com-
Finland, Nov. 2014, pp. 17–22. munications (ICC), 2014, Sydney, NSW, Jun. 2014, pp. 5747–5752.
DRAFT 21
[25] C. Li, J. Zhang, M. Haenggi, and K. Letaief, “User-centric intercell [49] A. Y. Al-Zahrani, F. R. Yu, and M. Huang, “A joint cross-layer and
interference nulling for downlink small cell networks,” IEEE Trans. colayer interference management scheme in hyperdense heterogeneous
Commun., vol. 63, no. 4, pp. 1419–1431, Apr. 2015. networks using mean-field game theory,” IEEE Trans. Veh. Technol.,
[26] M. Polignano, P. Mogensen, P. Fotiadis, L. Chavarria, I. Viering, and vol. 65, no. 3, pp. 1522–1535, Mar. 2016.
P. Zanier, “The inter-cell interference dilemma in dense outdoor small [50] L. Liu, V. Garcia, L. Tian, Z. Pan, and J. Shi, “Joint clustering and inter-
cell deployment,” in Proc. IEEE 79th Vehicular Technology Conf. (VTC cell resource allocation for CoMP in ultra dense cellular networks,” in
Spring), 2014, Seoul, Korea(South), May 2014, pp. 1–5. Proc. IEEE Int. Conf. Communications (ICC), 2015, London, UK, Jun.
[27] B. Soret, K. Pedersen, N. Jrgensen, and V. Fernandez-Lopez, “Interfer- 2015, pp. 2560–2564.
ence coordination for dense wireless networks,” IEEE Commun. Mag., [51] G. Yuehong, C. Lei, Z. Xin, Z. Yajun, and Z. Yunfei, “Enhanced
vol. 53, no. 1, pp. 102–109, Jan. 2015. power allocation scheme in ultra-dense small cell network,” China
[28] X. Wang, E. Visotsky, and A. Ghosh, “Dynamic cell muting for ultra Communications, vol. 13, no. 2, pp. 21–29, Feb. 2016.
dense indoor small cell deployment scenario,” in Proc. ICC Workshops, [52] Y. Sun, Y. Chang, M. Hu, and B. Wang, “A cluster-based hybrid access
London, UK, Jun. 2015, pp. 148–153. strategy using non-cooperative game theory for ultra-dense HetNet,”
[29] D. Chen, T. Quek, and M. Kountouris, “Backhauling in heterogeneous in in Proc. IEEE High Performance Computing and Communications
cellular networks: Modeling and tradeoffs,” IEEE Trans. Wireless (HPCC), 2015, New York, USA, Aug. 2015, pp. 14–19.
Commun., vol. 14, no. 6, pp. 3194–3206, Jun. 2015. [53] M.-J. Cho, T.-W. Ban, B. C. Jung, and H. J. Yang, “A distributed
[30] Y.-N. R. Li, H. Xiao, J. Li, and H. Wu, “Wireless backhaul of dense scheduling with interference-aware power control for ultra-dense net-
small cell networks with high dimension MIMO,” in IEEE Globecom works,” in Proc. IEEE Int. Conf. Communications (ICC), 2015, Lon-
Workshops (GC Wkshps), 2014, Austin, TX, Dec. 2014, pp. 1254–1259. don, UK, Jun. 2015, pp. 1661–1666.
[31] B. Li, D. Zhu, and P. Liang, “Small cell in-band wireless backhaul in [54] E. Björnson, L. Sanguinetti, and M. Kountouris, “Energy-efficient
massive MIMO systems: A cooperation of next-generation techniques,” future wireless networks: A marriage between massive MIMO and
IEEE Trans. Wireless Commun., vol. 14, no. 12, pp. 7057–7069, Dec. small cells,” in Proc. IEEE 16th Int. Workshop on Signal Processing
2015. Advances in Wireless Communications (SPAWC), 2015, Stockholm,
[32] X. Ge, H. Cheng, M. Guizani, and T. Han, “5G wireless backhaul Sweden, Jun. 2015, pp. 211–215.
networks: challenges and research advances,” IEEE Network, vol. 28, [55] C. Li, J. Zhang, and K. Letaief, “Throughput and energy efficiency
no. 6, pp. 6–11, Nov. 2014. analysis of small cell networks with multi-antenna base stations,” IEEE
[33] P. Amin, N. Kibret, E. Mutafungwa, B. Haile, J. Hamalainen, and Trans. Wireless Commun., vol. 13, no. 5, pp. 2505–2517, May 2014.
J. Nurminen, “Performance study for off-grid self-backhauled small [56] S. Yunas, M. Valkama, and J. Niemela, “Spectral and energy efficiency
cells in dense informal settlements,” in Proc. IEEE 25th Annu. Symp. of ultra-dense networks under different deployment strategies,” IEEE
Personal, Indoor, and Mobile Radio Communication (PIMRC), 2014, Commun. Mag., vol. 53, no. 1, pp. 90–100, Jan. 2015.
Washington DC, Sept. 2014, pp. 1652–1657. [57] Q. Ren, J. Fan, X. Luo, Z. Xu, and Y. Chen, “Analysis of spectral
[34] A. Gupta, X. Zhang, and J. Andrews, “SINR and throughput scaling and energy efficiency in ultra-dense network,” in Proc. IEEE Int. Conf.
in ultradense urban cellular networks,” IEEE Wireless Commun. Lett., Communication Workshop (ICCW), 2015, London, UK, Jun. 2015, pp.
vol. 4, no. 6, pp. 605–608, Dec. 2015. 2812–2817.
[35] C. Galiotto, N. K. Pratas, L. Doyle, and N. Marchetti, [58] S. Samarakoon, M. Bennis, W. Saad, M. Debbah, and M. Latva-
“Effect of LOS/NLOS propagation on ultra-dense networks,” aho, “Ultra dense small cell networks: Turning density into energy
CoRR, vol. abs/1507.01757, 2015. [Online]. Available: efficiency,” IEEE J. Sel. Areas Commun., vol. PP, no. 99, pp. 1–1,
http://arxiv.org/abs/1507.01757 Mar. 2016.
[36] J. Carle, J.-F. Myoupo, and D. Seme, “A basis for 3D cellular [59] ——, “Energy-efficient resource management in ultra dense small cell
networks,” in Proc. 15th Int. Conf. Information Networking, 2001., networks: A mean-field approach,” in Proc. IEEE Global Communi-
Beppu City, Oita, Jan. 2001, pp. 631–636. cations Conf. (GLOBECOM), 2015, San Diego, CA, Dec. 2015, pp.
[37] C. Decayeux and D. Seme, “A new model for 3D cellular mobile 1–6.
networks,” in Proc. 3rd Int. Workshop on Parallel and Distributed [60] C. Yang, “A unified design of spectrum, energy, and cost efficient
Computing, Jul. 2004, pp. 22–28. ultra-dense small cell networks,” in Proc. Int. Conf. on Wireless
[38] Z. Pan and Q. Zhu, “Modeling and analysis of coverage in 3D cellular Communications Signal Processing (WCSP), 2015, Nanjing, China,
networks,” IEEE Commun. Lett., vol. 19, no. 5, pp. 831–834, May Oct. 2015, pp. 1–4.
2015. [61] C. Yang, J. Li, and M. Guizani, “Cooperation for spectral and energy
[39] K. Zheng, L. Zhao, J. Mei, B. Shao, W. Xiang, and L. Hanzo, “Survey efficiency in ultra-dense small cell networks,” IEEE Wireless Commun.,
of large-scale MIMO systems,” IEEE Commun. Surveys Tuts., vol. 17, vol. 23, no. 1, pp. 64–71, Feb. 2016.
no. 3, pp. 1738–1760, Third quarter 2015. [62] A. Prasad, O. Tirkkonen, P. Lunden, O. Yilmaz, L. Dalsgaard, and
[40] J. Zhang, C. Pan, F. Pei, G. Liu, and X. Cheng, “Three-dimensional C. Wijting, “Energy-efficient inter-frequency small cell discovery tech-
fading channel models: A survey of elevation angle research,” IEEE niques for LTE-advanced heterogeneous network deployments,” IEEE
Commun. Mag., vol. 52, no. 6, pp. 218–226, Jun. 2014. Commun. Mag., vol. 51, no. 5, pp. 72–81, May 2013.
[41] S. Akoum and J. Acharya, “Full-dimensional MIMO for future cellular [63] P. Shuai, T. En, J. Huilin, P. Zhiwen, L. Nan, and Y. Xiaohu, “An
networks,” in in Proc. IEEE Radio and Wireless Symp. (RWS), 2014, improved graph coloring based small cell discovery scheme in LTE
Newport Beach, CA, Jan. 2014, pp. 1–3. hyper-dense networks,” in Proc. IEEE Wireless Communications and
[42] R. Heath, S. Peters, Y. Wang, and J. Zhang, “A current perspective Networking Conf. Workshops (WCNCW), 2015, New Orleans, LA, Mar.
on distributed antenna systems for the downlink of cellular systems,” 2015, pp. 17–22.
IEEE Commun. Mag., vol. 51, no. 4, pp. 161–167, Apr. 2013. [64] E. Obregon, K. W. Sung, and J. Zander, “On the sharing opportunities
[43] J. Wu, Z. Zhang, Y. Hong, and Y. Wen, “Cloud radio access network for ultra-dense networks in the radar bands,” in Proc. IEEE Int. Symp.
(C-RAN): a primer,” IEEE Network, vol. 29, no. 1, pp. 35–41, Jan. on Dynamic Spectrum Access Networks (DYSPAN), 2014, McLean, VA,
2015. Apr. 2014, pp. 215–223.
[44] M. Haenggi, Stochastic Geometry for Wireless Networks. Cambridge [65] Y. Teng, Y. Wang, and K. Horneman, “Co-primary spectrum sharing for
Univ. Press, 2012. denser networks in local area,” in Proc. 9th Int. Conf. on Cognitive Ra-
[45] K. Brown and Y. Shoham, Essentials of Game Theory. Morgan and dio Oriented Wireless Networks and Communications (CROWNCOM),
Claypool Publishers, 2008. 2014, Oulu, Finland, Jun. 2014, pp. 120–124.
[46] M. Kamel, W. Hamouda, and A. Youssef, “Multiple association in ultra- [66] A. H. Jafari, D. Lopez-Perez, M. Ding, and J. Zhang, “Study on
dense networks,” to appear in Proc. IEEE Int. Conf. Communications scheduling techniques for ultra dense small cell networks,” in Proc.
(ICC), 2016, Kuala Lumpur, Malaysia, May 2016. IEEE 82nd Vehicular Technology Conf. (VTC Fall), 2015, Boston, MA,
[47] C.-H. Liu and L.-C. Wang, “Random cell association and void proba- Sept. 2015, pp. 1–6.
bility in poisson-distributed cellular networks,” in Proc. IEEE Int. Conf. [67] J. Chen, Z. Gao, and Q. Zhao, “Load-aware dynamic spectrum access
Communications (ICC), 2015, London, UK, Jun. 2015, pp. 2816–2821. in ultra-dense small cell networks,” in Proc. Int. Conf. on Wireless
[48] F. Tavares, G. Berardinelli, N. Mahmood, T. Sorensen, and P. Mo- Communications Signal Processing (WCSP), 2015, Nanjing, China,
gensen, “Inter-cell interference management using maximum rank Oct. 2015, pp. 1–5.
planning in 5G small cell networks,” in Proc. 11th Int. Symp. Wireless [68] J. Andrews, S. Buzzi, W. Choi, S. Hanly, A. Lozano, A. Soong, and
Communications Systems (ISWCS), 2014, Barcelona, Spain, Aug. 2014, J. Zhang, “What will 5G be?” IEEE J. Sel. Areas Commun., vol. 32,
pp. 628–632. no. 6, pp. 1065–1082, Jun. 2014.
DRAFT 22
[69] L. Zhang, H.-C. Yang, and M. Hasna, “Generalized area spectral [93] S. Boyd and L. Vandenberghe, Convex optimization. Cambridge
efficiency: An effective performance metric for green wireless com- university press, 2004.
munications,” IEEE Trans. Commun., vol. 62, no. 2, pp. 747–757, Feb. [94] D. Kreutz, F. Ramos, P. Esteves Verissimo, C. Esteve Rothenberg,
2014. S. Azodolmolky, and S. Uhlig, “Software-defined networking: A com-
[70] E. Bjrnson, L. Sanguinetti, and M. Kountouris, “Deploying dense prehensive survey,” Proceedings of the IEEE, vol. 103, no. 1, pp. 14–76,
networks for maximal energy efficiency: Small cells meet massive Jan. 2015.
MIMO,” IEEE J. Sel. Areas Commun., vol. 34, no. 4, pp. 832–847, [95] N. McKeown, T. Anderson, H. Balakrishnan, G. Parulkar, L. Peterson,
Apr. 2016. J. Rexford, S. Shenker, and J. Turner, “Openflow: enabling innovation
[71] R. Jain, D.-M. Chiu, and W. R. Hawe, A quantitative measure of in campus networks,” ACM SIGCOMM Computer Communication
fairness and discrimination for resource allocation in shared computer Review, vol. 38, no. 2, pp. 69–74, 2008.
system. Eastern Research Laboratory, Digital Equipment Corporation [96] Open Networking Foundation, https://www.opennetworking.org, ac-
Hudson, MA, 1984, vol. 38. cessed: 2016-03-24.
[72] H. ElSawy, E. Hossain, and M. Haenggi, “Stochastic geometry for [97] H. Ali-Ahmad, C. Cicconetti, A. De La Oliva, M. Draxler, R. Gupta,
modeling, analysis, and design of multi-tier and cognitive cellular V. Mancuso, L. Roullet, and V. Sciancalepore, “CROWD: An SDN
wireless networks: A survey,” IEEE Commun. Surveys Tuts., vol. 15, approach for DenseNets,” in Proc. Second European Workshop on
no. 3, pp. 996–1019, Third quarter 2013. Software Defined Networks (EWSDN), 2013, Berlin, Oct. 2013, pp.
[73] D. Stoyan, W. Kendall, and J. Mecke, Stochastic Geometry and Its 25–31.
Applications. John Wiley and Sons, 1995. [98] H. Ali-Ahmad, C. Cicconetti, A. De La Oliva, V. Mancuso,
[74] J. Andrews, F. Baccelli, and R. Ganti, “A tractable approach to coverage M. Reddy Sama, P. Seite, and S. Shanmugalingam, “An SDN-based
and rate in cellular networks,” IEEE Trans. Commun., vol. 59, no. 11, network architecture for extremely dense wireless networks,” in Proc.
pp. 3122–3134, Nov. 2011. IEEE SDN for Future Networks and Services (SDN4FNS), 2013, Nov.
[75] H. Dhillon, R. Ganti, F. Baccelli, and J. Andrews, “Modeling and 2013, pp. 1–7.
analysis of K-tier downlink heterogeneous cellular networks,” IEEE [99] V. C. Borges, K. V. Cardoso, E. Cerqueira, M. Nogueira, and A. San-
J. Sel. Areas Commun., vol. 30, no. 3, pp. 550–560, Apr. 2012. tos, “Aspirations, challenges, and open issues for software-based 5G
[76] H. Dhillon, M. Kountouris, and J. Andrews, “Downlink MIMO Het- networks in extremely dense and heterogeneous scenarios,” EURASIP
Nets: Modeling, ordering results and performance analysis,” IEEE Journal on Wireless Communications and Networking, vol. 2015, no. 1,
Trans. Wireless Commun., vol. 12, no. 10, pp. 5208–5222, Oct. 2013. pp. 1–13, 2015.
[77] C. Li, J. Zhang, J. G. Andrews, and K. B. Letaief, “Success probability [100] M. Condoluci, M. Dohler, G. Araniti, A. Molinaro, and K. Zheng,
and area spectral efficiency in multiuser MIMO HetNets,” IEEE Trans. “Toward 5G densenets: architectural advances for effective machine-
Commun., vol. 64, no. 4, pp. 1544–1556, Apr. 2016. type communications over femtocells,” IEEE Commun. Mag., vol. 53,
[78] S. Singh and J. Andrews, “Joint resource partitioning and offloading no. 1, pp. 134–141, Jan. 2015.
in heterogeneous cellular networks,” IEEE Trans. Wireless Commun., [101] C. Niephaus, G. Ghinea, O. G. Aliu, S. Hadzic, and M. Kretschmer,
vol. 13, no. 2, pp. 888–901, Feb. 2014. “SDN in the wireless context - towards full programmability of wireless
[79] S. Singh, F. Baccelli, and J. Andrews, “On association cells in random network elements,” in Proc. 1st IEEE Conf. on Network Softwarization
heterogeneous networks,” IEEE Wireless Commun. Lett., vol. 3, no. 1, (NetSoft), 2015, London, Apr. 2015, pp. 1–6.
pp. 70–73, Feb. 2014. [102] Y. Wang, B. Luo, L. C. Ong, and S. W. Leong, “Fiber-connected mas-
[80] S. Singh, H. Dhillon, and J. Andrews, “Offloading in heterogeneous sively distributed antenna systems,” in Proc. 13th Int. Conf. on Optical
networks: Modeling, analysis, and design insights,” IEEE Trans. Wire- Communications and Networks (ICOCN), 2014, Suzhou, China, Nov.
less Commun., vol. 12, no. 5, pp. 2484–2497, May 2013. 2014, pp. 1–3.
[81] S. Singh, X. Zhang, and J. Andrews, “Joint rate and SINR coverage [103] L. Dai, “An uplink capacity analysis of the distributed antenna system
analysis for decoupled uplink-downlink biased cell associations in (DAS): From cellular DAS to DAS with virtual cells,” IEEE Trans.
HetNets,” IEEE Trans. Wireless Commun., vol. 14, no. 10, pp. 5360– Wireless Commun., vol. 13, no. 5, pp. 2717–2731, May 2014.
5373, Oct 2015. [104] H. Zhu, “Performance comparison between distributed antenna and
[82] J.-S. Ferenc and Z. Néda, “On the size distribution of poisson voronoi microcellular systems,” IEEE J. Sel. Areas Commun., vol. 29, no. 6,
cells,” Physica A: Statistical Mechanics and its Applications, vol. 385, pp. 1151–1163, Jun. 2011.
no. 2, pp. 518–526, 2007. [105] D. Sabella, P. Rost, Y. Sheng, E. Pateromichelakis, U. Salim, P. Guitton-
[83] K. Akkarajitsakul, E. Hossain, D. Niyato, and D. I. Kim, “Game Ouhamou, M. D. Girolamo, and G. Giuliani, “RAN as a service:
theoretic approaches for multiple access in wireless networks: A Challenges of designing a flexible RAN architecture in a cloud-based
survey,” IEEE Commun. Surveys Tuts., vol. 13, no. 3, pp. 372–395, heterogeneous mobile network,” in Future Network and Mobile Summit
Third quarter 2011. (FutureNetworkSummit), 2013, Jul. 2013, pp. 1–8.
[84] A. B. MacKenzie and L. A. DaSilva, “Game theory for wireless [106] A. Checko, H. L. Christiansen, Y. Yan, L. Scolari, G. Kardaras, M. S.
engineers,” Synthesis Lectures on Communications, vol. 1, no. 1, pp. Berger, and L. Dittmann, “Cloud RAN for mobile networks 2014: A
1–86, 2006. technology overview,” IEEE Commun. Surveys Tuts., vol. 17, no. 1, pp.
[85] W. Saad, Z. Han, M. Debbah, A. Hjorungnes, and T. Basar, “Coalitional 405–426, First quarter 2015.
game theory for communication networks,” IEEE Signal Process. Mag., [107] C. Liu, K. Sundaresan, M. Jiang, S. Rangarajan, and G. K. Chang,
vol. 26, no. 5, pp. 77–97, Sept. 2009. “The case for re-configurable backhaul in cloud-RAN based small
[86] E. Altman, T. Boulogne, R. El-Azouzi, T. Jiménez, and L. Wynter, cell networks,” in Proc. IEEE Conf. on Computer Communications
“A survey on networking games in telecommunications,” Computers (INFOCOM), 2013, Turin, Italy, Apr. 2013, pp. 1124–1132.
& Operations Research, vol. 33, no. 2, pp. 286–311, 2006. [108] M. Hadzialic, B. Dosenovic, M. Dzaferagic, and J. Musovic, “Cloud-
[87] O. Guéant, J.-M. Lasry, and P.-L. Lions, “Mean field games and RAN: Innovative radio access network architecture,” in ELMAR, 2013
applications,” in Paris-Princeton lectures on mathematical finance 55th International Symposium, Sept. 2013, pp. 115–120.
2010. Springer, 2011, pp. 205–266. [109] T. S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. N.
[88] M. Manjrekar, V. Ramaswamy, and S. Shakkottai, “A mean field game Wong, J. K. Schulz, M. Samimi, and F. Gutierrez, “Millimeter wave
approach to scheduling in cellular systems,” in in Proc. IEEE Conf. mobile communications for 5G cellular: It will work!” IEEE Access,
on Computer Communications (INFOCOM), 2014, Toronto, ON, Apr. vol. 1, pp. 335–349, May 2013.
2014, pp. 1554–1562. [110] Z. Pi and F. Khan, “An introduction to millimeter-wave mobile broad-
[89] Q. C. Li, G. Wu, and R. Q. Hu, “Analytical study on network spectrum band systems,” IEEE Commun. Mag., vol. 49, no. 6, pp. 101–107, Jun.
efficiency of ultra dense networks,” in Proc. IEEE 24th Inter. Symp. on 2011.
Personal Indoor and Mobile Radio Communications (PIMRC), 2013, [111] M. R. Akdeniz, Y. Liu, S. Rangan, and E. Erkip, “Millimeter wave pic-
London, Sept. 2013, pp. 2764–2768. ocellular system evaluation for urban deployments,” in IEEE Globecom
[90] M. X. Cheng, Y. Li, and D.-Z. Du, Combinatorial optimization in Workshops (GC Wkshps), 2013, Dec. 2013, pp. 105–110.
communication networks. Springer Science & Business Media, 2006, [112] H. Shokri-Ghadikolaei and C. Fischione, “The transitional behavior of
vol. 18. interference in millimeter wave networks and its impact on medium
[91] A. Schrijver, Theory of linear and integer programming. John Wiley access control,” IEEE Trans. Commun., vol. 64, no. 2, pp. 723–740,
& Sons, 1998. Feb. 2016.
[92] C. A. Floudas, Nonlinear and mixed-integer optimization: fundamentals [113] R. Ford, F. Gmez-Cuba, M. Mezzavilla, and S. Rangan, “Dynamic
and applications. Oxford University Press on Demand, 1995. time-domain duplexing for self-backhauled millimeter wave cellular
DRAFT 23
networks,” in Proc. IEEE Int. Conf. on Communication Workshop [134] P. Mach, Z. Becvar, and T. Vanek, “In-band device-to-device commu-
(ICCW), 2015, London, Jun. 2015, pp. 13–18. nication in OFDMA cellular networks: A survey and challenges,” IEEE
[114] U. Madhow, “Broadband millimeter wave networks: Architectures and Commun. Surveys Tuts., vol. 17, no. 4, pp. 1885–1922, Fourth quarter
applications,” in 2008 2nd Int. Symp. on Advanced Networks and 2015.
Telecommunication Systems (ANTS), 2008., Mumbai, India, Dec. 2008, [135] M. Bennis, M. Simsek, A. Czylwik, W. Saad, S. Valentin, and M. Deb-
pp. 1–3. bah, “When cellular meets WiFi in wireless small cell networks,” IEEE
[115] H. Shokri-Ghadikolaei, Y. Xu, L. Gkatzikis, and C. Fischione, “User Commun. Mag., vol. 51, no. 6, pp. 44–50, Jun. 2013.
association and the alignment-throughput tradeoff in millimeter wave [136] Y. Kojima, J. Suga, T. Kawasaki, M. Okuda, and R. Takechi, “LTE-
networks,” in Proc. IEEE 1st Int. Forum on Research and Technologies WiFi link aggregation at femtocell base station,” in Proc. of World
for Society and Industry Leveraging a better tomorrow (RTSI), 2015, Telecommunications Congress (WTC) 2014, Berlin, Germany, Jun.
Turin, Italy, Sept. 2015, pp. 100–105. 2014, pp. 1–6.
[116] T. Bai and R. W. Heath, “Coverage and rate analysis for millimeter- [137] O. Galinina, A. Pyattaev, S. Andreev, M. Dohler, and Y. Koucheryavy,
wave cellular networks,” IEEE Trans. Wireless Commun., vol. 14, no. 2, “5G Multi-RAT LTE-WiFi ultra-dense small cells: Performance dy-
pp. 1100–1114, Feb. 2015. namics, architecture, and trends,” IEEE J. Sel. Areas Commun., vol. 33,
[117] R. Baldemair, T. Irnich, K. Balachandran, E. Dahlman, G. Mildh, no. 6, pp. 1224–1240, Jun. 2015.
Y. Seln, S. Parkvall, M. Meyer, and A. Osseiran, “Ultra-dense networks [138] W. Ni and I. B. Collings, “A new adaptive small-cell architecture,”
in millimeter-wave frequencies,” IEEE Commun. Mag., vol. 53, no. 1, IEEE J. Sel. Areas Commun., vol. 31, no. 5, pp. 829–839, May 2013.
pp. 202–208, Jan. 2015. [139] A. Orsino, G. Araniti, A. Molinaro, and A. Iera, “Effective RAT
[118] M. D. Renzo, “Stochastic geometry modeling and analysis of multi-tier selection approach for 5G dense wireless networks,” in Proc. IEEE 81st
millimeter wave cellular networks,” IEEE Trans. Wireless Commun., Vehicular Technology Conf. (VTC Spring), 2015, Glasgow, Scotland,
vol. 14, no. 9, pp. 5038–5057, Sept. 2015. May 2015, pp. 1–5.
[119] E. Perahia, C. Cordeiro, M. Park, and L. L. Yang, “IEEE 802.11ad: [140] S. Andreev, M. Gerasimenko, O. Galinina, Y. Koucheryavy, N. Hi-
Defining the next generation multi-Gbps Wi-Fi,” in Proc. IEEE 7th mayat, S. P. Yeh, and S. Talwar, “Intelligent access network selection in
Consumer Communications and Networking Conf. (CCNC), 2010, Las converged multi-radio heterogeneous networks,” IEEE Trans. Wireless
Vegas, NV, Jan. 2010, pp. 1–5. Commun., vol. 21, no. 6, pp. 86–96, Dec. 2014.
[120] E. G. Larsson, O. Edfors, F. Tufvesson, and T. L. Marzetta, “Massive [141] J. L. Tomici and P. Chitrapu, “Multi-RAT traffic offloading solutions for
MIMO for next generation wireless systems,” IEEE Commun. Mag., the bandwidth crunch problem,” in Proc. IEEE Long Island Systems,
vol. 52, no. 2, pp. 186–195, Feb. 2014. Applications and Technology Conf. (LISAT), 2011, Farmingdale, NY,
[121] S. Han, C. l. I, Z. Xu, and C. Rowell, “Large-scale antenna systems May 2011, pp. 1–6.
with hybrid analog and digital beamforming for millimeter wave 5G,” [142] O. Galinina, S. Andreev, M. Gerasimenko, Y. Koucheryavy, N. Hi-
IEEE Commun. Mag., vol. 53, no. 1, pp. 186–194, Jan. 2015. mayat, S. P. Yeh, and S. Talwar, “Capturing spatial randomness
[122] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, “Energy and spectral of heterogeneous cellular/WLAN deployments with dynamic traffic,”
efficiency of very large multiuser MIMO systems,” IEEE Trans. Com- IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp. 1083–1099, Jun. 2014.
mun., vol. 61, no. 4, pp. 1436–1449, Apr. 2013. [143] 3GPP TS 23.261, v10.1.0, “IP flow mobility and seamless wireless
[123] Y. Kim, H. Ji, J. Lee, Y. H. Nam, B. L. Ng, I. Tzanidis, Y. Li, and local area network (WLAN) offload,” Tech. Rep., Sept. 2010.
J. Zhang, “Full dimension MIMO (FD-MIMO): The next evolution
[144] E. Bastug, M. Bennis, and M. Debbah, “Living on the edge: The role
of MIMO in LTE systems,” IEEE Trans. Wireless Commun., vol. 21,
of proactive caching in 5G wireless networks,” IEEE Commun. Mag.,
no. 3, pp. 92–100, Jun. 2014.
vol. 52, no. 8, pp. 82–89, Aug. 2014.
[124] Z. Gao, L. Dai, C. Yuen, and Z. Wang, “Asymptotic orthogonality anal-
[145] M. Gregori, J. Gomez-Vilardebo, J. Matamoros, and D. Gunduz,
ysis of time-domain sparse massive MIMO channels,” IEEE Commun.
“Wireless content caching for small cell and D2D networks,” IEEE
Lett., vol. 19, no. 10, pp. 1826–1829, Oct. 2015.
J. Sel. Areas Commun., vol. PP, no. 99, pp. 1–1, 2016.
[125] O. Elijah, C. Leow, T. Rahman, S. Nunoo, and S. Iliya, “A compre-
[146] J. Hachem, N. Karamchandani, and S. Diggavi, “Content caching and
hensive survey of pilot contamination in massive MIMO - 5G system,”
delivery over heterogeneous wireless networks,” in Proc. IEEE Conf. on
IEEE Commun. Surveys Tuts., vol. PP, no. 99, pp. 1–1, 2015.
Computer Communications (INFOCOM), 2015, Kowloon, Hong Kong,
[126] B. Panzner, W. Zirwas, S. Dierks, M. Lauridsen, P. Mogensen, K. Pa-
Apr. 2015, pp. 756–764.
jukoski, and D. Miao, “Deployment and implementation strategies for
massive MIMO in 5G,” in IEEE Globecom Workshops (GC Wkshps), [147] M. Ji, M. F. Wong, A. M. Tulino, J. Llorca, G. Caire, M. Effros, and
2014, Dec. 2014, pp. 346–351. M. Langberg, “On the fundamental limits of caching in combination
networks,” in Proc. IEEE 16th Int. Workshop on Signal Processing
[127] W. Liu, S. Han, C. Yang, and C. Sun, “Massive MIMO or small
Advances in Wireless Communications (SPAWC), 2015, Stockholm,
cell network: Who is more energy efficient?” in Proc. IEEE Wireless
Sweden, Jun. 2015, pp. 695–699.
Communications and Networking Conf. Workshops (WCNCW), 2013,
Shanghai, China, Apr. 2013, pp. 24–29. [148] M. A. Maddah-Ali and U. Niesen, “Fundamental limits of caching,”
[128] W. Liu, S. Han, and C. Yang, “Energy efficiency comparison of massive IEEE Trans. Inf. Theory, vol. 60, no. 5, pp. 2856–2867, May 2014.
MIMO and small cell network,” in Proc. IEEE Global Conf. on Signal [149] A. Sengupta, R. Tandon, and T. C. Clancy, “Fundamental limits of
and Information Processing (GlobalSIP), 2014, Atlanta, GA, Dec. caching with secure delivery,” IEEE Trans. Inf. Forensics Security,
2014, pp. 617–621. vol. 10, no. 2, pp. 355–370, Feb. 2015.
[129] J. Liu, N. Kato, J. Ma, and N. Kadowaki, “Device-to-device communi- [150] O. Semiari, W. Saad, S. Valentin, M. Bennis, and H. V. Poor, “Context-
cation in LTE-Advanced networks: A survey,” IEEE Commun. Surveys aware small cell networks: How social metrics improve wireless
Tuts., vol. 17, no. 4, pp. 1923–1940, Fourth quarter 2015. resource allocation,” IEEE Trans. Wireless Commun., vol. 14, no. 11,
[130] R. Alkurd, R. M. Shubair, and I. Abualhaol, “Survey on device-to- pp. 5927–5940, Nov. 2015.
device communications: Challenges and design issues,” in IEEE 12th [151] M. Ji, G. Caire, and A. F. Molisch, “Fundamental limits of caching in
Int. New Circuits and Systems Conf. (NEWCAS), 2014, Trois-Rivieres, wireless D2D networks,” IEEE Trans. Inf. Theory, vol. 62, no. 2, pp.
QC, Jun. 2014, pp. 361–364. 849–869, Feb. 2016.
[131] B. Romanous, N. Bitar, A. Imran, and H. Refai, “Network densification: [152] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and
Challenges and opportunities in enabling 5G,” in Proc. IEEE 20th M. Ayyash, “Internet of Things: A survey on enabling technologies,
International Workshop on Computer Aided Modelling and Design protocols, and applications,” IEEE Commun. Surveys Tuts., vol. 17,
of Communication Links and Networks (CAMAD), 2015, Guildford, no. 4, pp. 2347–2376, Fourth quarter 2015.
England, Sept. 2015, pp. 129–134. [153] D. Porcino and W. Hirt, “Ultra-wideband radio technology: Potential
[132] A. Laya, K. Wang, A. A. Widaa, J. Alonso-Zarate, J. Markendahl, and challenges ahead,” IEEE Commun. Mag., vol. 41, no. 7, pp. 66–74,
and L. Alonso, “Device-to-Device communications and small cells: Jul. 2003.
enabling spectrum reuse for dense networks,” IEEE Trans. Wireless [154] L. Atzori, A. Iera, and G. Morabito, “The Internet of Things: A survey,”
Commun., vol. 21, no. 4, pp. 98–105, Aug. 2014. Computer networks, vol. 54, no. 15, pp. 2787–2805, 2010.
[133] X. Chen, R. Q. Hu, and Y. Qian, “Coverage study of dense device-to- [155] S. Severi, F. Sottile, G. Abreu, C. Pastrone, M. Spirito, and F. Berens,
device communications underlaying cellular networks,” in Proc. IEEE “M2M technologies: Enablers for a pervasive Internet of Things,” in
Global Communications Conf. (GLOBECOM), 2014, Austin, TX, Dec. Proc. European Conf. on Networks and Communications (EuCNC),
2014, pp. 4353–4358. 2014, Bologna, Italy, Jun. 2014, pp. 1–5.
DRAFT 24
[156] R. R. Tyagi, K.-D. Lee, F. Aurzada, S. Kim, and M. Reisslein, “Efficient [179] J. Olmos, R. Ferrus, and H. Galeana-Zapien, “Analytical modeling
delivery of frequent small data for U-healthcare applications over LTE- and performance evaluation of cell selection algorithms for mobile
Advanced networks,” in Proc. of the 2nd ACM international workshop networks with backhaul capacity constraints,” IEEE Trans. Wireless
on Pervasive Wireless Healthcare, Hilton Head, SC, Jun. 2012, pp. Commun., vol. 12, no. 12, pp. 6011–6023, Dec. 2013.
27–32.
[157] K. M., S. M., and R. M. Banakar, “Evolution of IoT in smart vehicles:
An overview,” in Proc. IEEE Int. Conf. on Green Computing and
Internet of Things (ICGCIoT), 2015, Noida, India, Oct. 2015, pp. 804–
809.
[158] R. Khan, S. U. Khan, R. Zaheer, and S. Khan, “Future Internet:
The Internet of Things architecture, possible applications and key
challenges,” in Proc. IEEE 10th Int. Conf. on Frontiers of Information
Technology (FIT), 2012, Islamabad, Pakistan, Dec. 2012, pp. 257–260.
[159] J. Andrews, S. Singh, Q. Ye, X. Lin, and H. Dhillon, “An overview
of load balancing in HetNets: Old myths and open problems,” IEEE
Wireless Commun., vol. 21, no. 2, pp. 18–25, Apr. 2014.
[160] D. David López-Pérez, X. Chu, and I. Guvenc, “On the expanded
region of picocells in heterogeneous networks,” IEEE J. Sel. Areas
Commun., vol. 6, no. 3, pp. 281–294, Jun. 2012.
[161] S. Ahmad and D. Datla, “Distributed power allocations in heteroge-
neous networks with dual connectivity using backhaul state informa-
tion,” IEEE Trans. Wireless Commun., vol. 14, no. 8, pp. 4574–4581,
Aug 2015.
[162] J. Zhang, Q. Zeng, T. Mahmoodi, A. Georgakopoulos, and P. De-
mestichas, “LTE small cell enhancement by dual connectivity,” White
Paper, Nov. 2014.
[163] S. Jha, K. Sivanesan, R. Vannithamby, and A. Koc, “Dual connectivity
in LTE small cell networks,” in IEEE Globecom Workshops (GC
Wkshps), 2014, Austin, TX, Dec. 2014, pp. 1205–1210.
[164] A. Hamza, S. Khalifa, H. Hamza, and K. Elsayed, “A survey on inter-
cell interference coordination techniques in OFDMA-based cellular
networks,” IEEE Commun. Surveys Tuts., vol. 15, no. 4, pp. 1642–
1670, Fourth quarter 2013.
[165] M. Kamel and K. Elsayed, “Performance evaluation of a coordinated
time-domain eICIC framework based on ABSF in heterogeneous LTE-
Advanced networks,” in Proc. IEEE Global Communications Conf.
(GLOBECOM), 2012, Budapest, Hungary, Dec 2012, pp. 5326–5331.
[166] A. Adhikary, H. Dhillon, and G. Caire, “Massive-MIMO meets HetNet:
Interference coordination through spatial blanking,” IEEE J. Sel. Areas
Commun., vol. 33, no. 6, pp. 1171–1186, Jun. 2015.
[167] T. M. Ho, N. Tran, C. Do, S. Ahsan Kazmi, E.-N. Huh, and C. S.
Hong, “Power control for interference management and QoS guarantee
in heterogeneous networks,” IEEE Commun. Lett., vol. 19, no. 8, pp.
1402–1405, Aug 2015.
[168] I. Ashraf, F. Boccardi, and L. Ho, “Sleep mode techniques for small
cell deployments,” IEEE Commun. Mag., vol. 49, no. 8, pp. 72–79,
Aug. 2011.
[169] 3GPP TR 36.932, v12.1.0, “Scenarios and requirements for small cell
enhancements for E-UTRA and E-UTRAN ,” Tech. Rep., Oct. 2014.
[170] N. Wang, E. Hossain, and V. Bhargava, “Backhauling 5G small cells:
A radio resource management perspective,” IEEE Wireless Commun.,
vol. 22, no. 5, pp. 41–49, Oct. 2015.
[171] X. Wang, A. V. Vasilakos, M. Chen, Y. Liu, and T. T. Kwon, “A
survey of green mobile networks: Opportunities and challenges,” Mob.
Netw. Appl., vol. 17, no. 1, pp. 4–20, Feb. 2012. [Online]. Available:
http://dx.doi.org/10.1007/s11036-011-0316-4
[172] G. Auer, V. Giannini, C. Desset, I. Godor, P. Skillermark, M. Olsson,
M. Imran, D. Sabella, M. Gonzalez, O. Blume, and A. Fehske, “How
much energy is needed to run a wireless network?” IEEE Wireless
Commun., vol. 18, no. 5, pp. 40–49, Oct. 2011.
[173] T. Chen, H. Kim, and Y. Yang, “Energy efficiency metrics for green
wireless communications,” in Proc. Int. Conf. Wireless Communica-
tions and Signal Processing (WCSP), 2010, Suzhou, China, Oct. 2010,
pp. 1–6.
[174] GreenTouch, “Reducing the net energy consumption in communica-
tions networks by up to 90% by 2020,” Technical Report, Jun. 2013.
[175] Q. Li, R. Hu, Y. Qian, and G. Wu, “Cooperative communications
for wireless networks: techniques and applications in LTE-advanced
systems,” IEEE Wireless Commun., vol. 19, no. 2, Apr. 2012.
[176] S. Singh, “Load balancing in heterogeneous cellular networks,” Ph.D.
dissertation, The University of Texas at Austin, 2014.
[177] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “Next
generation/dynamic spectrum access/cognitive radio wireless networks:
a survey,” Computer Networks, vol. 50, no. 13, pp. 2127–2159, 2006.
[178] O. Aliu, A. Imran, M. Imran, and B. Evans, “A survey of self
organisation in future cellular networks,” IEEE Commun. Surveys Tuts.,
vol. 15, no. 1, pp. 336–361, First quarter 2013.

Ultra-Dense Networks: A Survey

Uploaded by

Copyright:

Available Formats

Ultra-Dense Networks: A Survey

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ultra-Dense Networks: A Survey

Uploaded by

Copyright:

Available Formats

This article has been accepted for publication in a future issue of this journal, but has not been

Ultra-Dense Networks: A Survey

Abstract—The exponential growth and availability of data in TABLE I: L IST OF ACRONYMS

T HE evolution of mobile devices and applications in the

gaps and the unanswered questions, we review the published

Stochastic Geometry SDN User Association

Fig. 2: The organization of the survey.

TABLE II: D IFFERENT T YPES OF S MALL C ELLS

Deployment of different Densification

Vertical Densification: Deployment of

4) Innovative frequency reuse techniques are required. In

A. Performance Metrics Rn = λs pa (1 − pout )Ro , (1)

TABLE III: M ODELING T ECHNIQUES AND P ERFORMANCE M ETRICS

10 from stochastic geometry. Besides Poisson point process

0 geometry toolbox [44]

F. Device-to-Device Communication (D2D) H. Proactive Caching

communication technologies [153], context-aware computing, ABSF: Almost-Blank Subframe

TABLE IV: S IMULATION PARAMETERS OF M ULTIPLE A S -

Parameter Value User Neighborhood

a macrocell and a set of small cells simultaneously as depicted

Analysis 1st nearest cell nd

in the vicinity of many small cells. Hence, the cell selection

Area spectral efficiency (bps/Hz/km2 ) ×104 ×104

Area spectral efficiency (bps/Hz/km2 )

TABLE V: C OMPARISON OF D IFFERENT I NTERFERENCE M ANAGEMENT S TUDIES

T1 T2 T3 interference coordination. They considered four interference

C. Propagation Modelling network and studied the performance in terms of coverage

investigate. [2] N. Bhushan, J. Li, D. Malladi, R. Gilmore, D. Brenner, A. Damn-

You might also like