Computer Networks 106 (2016) 17–48

Contents lists available at ScienceDirect

Computer Networks
journal homepage: www.elsevier.com/locate/comnet

5G roadmap: 10 key enabling technologies

Ian F. Akyildiz, Shuai Nie∗, Shih-Chun Lin, Manoj Chandrasekaran
Broadband Wireless Networking Laboratory, School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA

a r t i c l e i n f o a b s t r a c t

Article history: The fifth generation (5G) mobile communication networks will require a major paradigm shift to sat-
Received 14 April 2016 isfy the increasing demand for higher data rates, lower network latencies, better energy efficiency, and
Revised 5 June 2016
reliable ubiquitous connectivity. With prediction of the advent of 5G systems in the near future, many
Accepted 10 June 2016
efforts and revolutionary ideas have been proposed and explored around the world. The major technolog-
Available online 14 June 2016
ical breakthroughs that will bring renaissance to wireless communication networks include (1) a wireless
Keywords: software-defined network, (2) network function virtualization, (3) millimeter wave spectrum, (4) massive
5G MIMO, (5) network ultra-densification, (6) big data and mobile cloud computing, (7) scalable Internet
Wireless software-defined network (WSDN) of Things, (8) device-to-device connectivity with high mobility, (9) green communications, and (10) new
Network function virtualization (NFV) radio access techniques. In this paper, the state-of-the-art and the potentials of these ten enabling tech-
Millimeter wave nologies are extensively surveyed. Furthermore, the challenges and limitations for each technology are
Massive MIMO
treated in depth, while the possible solutions are highlighted.
Heterogeneous networks (HetNets)
Device-to-device communications (D2D) © 2016 Elsevier B.V. All rights reserved.
Internet of Things (IoT)
Big data
Mobile cloud computing
Green communications
Radio access techniques

1. Introduction [3]. Moreover, small cells and ultra-dense network will be deployed
to make the network more flexible, and to provide more users
1.1. Motivation towards 5G with network connectivity at anywhere, anytime. Additionally, new
wireless access technologies should be backward-compatible with
Over the past decades, the demand for higher data rates has the existing solutions to obtain the optimal network performance
been continuously growing to satisfy consumers’ desire for a with faster data rates, which ultimately lead to the driving forces
faster, safer, and smarter wireless network. The industry of wire- toward the fifth generation (5G) wireless communication networks
less communications has experienced a great evolution in efforts [4]. In this paper, we investigate 10 enabling technologies, namely,
of boosting system performance on data rates, from analog to wireless software-defined networking, network function virtual-
digital system, from circuit-switched to packet-switched network, ization, millimeter wave communications, massive MIMO, ultra-
from a bulky handheld cellular phone to a smartwatch on wrist densification, mobile cloud computing, Internet of Things, device-
[1,2]. However, a significant paradigm shift is required to further to-device communications, green communications, and radio ac-
strengthen the wireless communication networks, since current cess techniques, as shown in Fig. 1. Although some of the key
wireless systems are facing a bottleneck in spectrum resources technologies have been researched for LTE-Advanced networks, the
which makes it difficult to enhance performance in the limited exponential increase in data rates is propelling a major wireless
available bandwidth. The industry has predicted that in order to network architecture paradigm shift toward the wireless software-
reach the network-level capability of serving more users with defining networking (WSDN) and network function virtualization
higher data rates, more spectrum resources are required to be al- (NFV), which will fundamentally help reconfigure and solve the
located for the next generation of wireless communication net- open problems in the 5G networks. Note that there is no partic-
works, and current spectrum needs to be utilized more efficiently ular priority order for the technologies in this paper.

∗
1.2. Organization of this paper
Corresponding author.
E-mail addresses: [email protected] (I.F. Akyildiz), [email protected],
[email protected] (S. Nie), [email protected] (S.-C. Lin), As the research process of 5G communication system will re-
[email protected] (M. Chandrasekaran). main active and keep growing in the coming years, this paper

http://dx.doi.org/10.1016/j.comnet.2016.06.010
1389-1286/© 2016 Elsevier B.V. All rights reserved.
18 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

tablets and laptops which are accessible to wireless networks, in-

creasing the data rate of cellular network is becoming an inevitable
market driving force. Although the current maximum data rates
can support HD video streaming which requires 8–15 Mbps, there
are applications like ultra-HD 4K video streaming, high definition
gaming, and 3D contents, which require even higher data rates at
around 25 Mbps to provide a satisfactory experience to users.
With these emerging applications demanding higher data rates,
5G networks are expected to have the peak data rate of around 10
Gbps which is a 100-fold improvement over current 4G networks
[6]. Besides increasing the maximum data rate, the cell-edge data
rate, as the worst case data rate users experience, should also be
improved to 100 Mbps, which is a 100 times improvement over 4G
networks at cell edge. The maximum data rate is an optimum esti-
Fig. 1. The 10 key enabling technologies for 5G. mate that a user can experience. In fact, the affects of intercell in-
terference and transmission loss make the maximum value hardly
achievable. Therefore edge data rate level becomes more impor-
aims to motivate readers to foresee the revolutionary techniques tant from the perspective of network engineering, as this data rate
that will shape the next generation wireless communication net- must support around 95% of users connected to the network. An-
work. The paper is organized in the following structure. The re- other metric based on data rate that characterizes the network is
quirements for 5G are elaborated in Section 2, which covers the the area capacity, which specifies the total data rate the network
demands of high data rates, low latency, cost-efficient energy con- can serve per unit area. According to its definition, the unit of area
sumption, high scalability, improved connectivity, and high net- capacity is normally bits per second per unit area. This metric is
work security. Based on the system requirements and user de- expected to increase 100 times in 5G compared to 4G network.
mands that the 5G wireless networks are expected to satisfy, the This demand for increase in data rate can be met by techniques
10 key enabling solutions are explained in detail in Sections 3– such as millimeter wave communications, massive MIMO systems,
12, where we cover their strengths and challenges, as well as the and wireless software-defined networking, etc.
open problems needed to be resolved before deployed in the next
generation wireless networks. The global research activities on 5G
are summarized in Section 13. In Section 14, we conclude this
paper. 2.2. Low latency

The roundtrip latency of data plane in the LTE network is

2. Requirements for 5G
around 15 milliseconds (ms) [7]. However, for the recently emerg-
ing applications such as tactile Internet, virtual reality, and multi-
In conventional cellular networks, mobile phones were practi-
player gaming that 5G networks are expected to support, the la-
cally the only type of device expected to be supported. With the
tency should be upgraded to an order of magnitude faster than
proliferation of Internet and its numerous applications, there was
current network, at around 1 ms [8]. For instance, tactile Internet
the problem of handling several classes of traffic to meet the dif-
is a recently developed application where the wireless network is
ferent QoS requirements of diverse applications like video stream-
used for real-time control applications [9]. The latency required for
ing, data, VoIP calls, etc. A similar situation is arising now with
such applications is determined by the typical interaction for steer-
the need to support several types of devices and applications with
ing and control of real and virtual objects without creating cyber-
drastically varying QoS requirement to provide better experience
sickness. The expected latency that would make these applications
to the user. Unlike previous generations of cellular networks, 5G
feasible is around 1 ms [9]. Although current smartphones have
cellular network is envisioned to support a multitude of devices
touch screens as the main interface, future devices will integrate
and applications like smartwatches, autonomous vehicles, Inter-
various other interfaces like haptic, visual and auditory input and
net of Things (IoT), and tactile Internet. In particular, according to
feedback, which will provide a new way of interacting with online
the International Telecommunication Union (ITU), there are three
environment for applications in virtual reality, healthcare, gaming,
types of service scenarios to be supported in 5G, which are mobile
and sports, etc. These applications require real-time interactions
broadband services, massive machine type communications, and
with the user and any delay in the system will cause degradation
ultra-reliable and low latency communications, respectively [5].
to the user experience, thus latency is a crucial factor in 5G.
The various types of devices and application scenarios need more
Another application that 5G networks are expected to support
sophisticated networks that not only can support high throughput,
is the machine type communications (MTC) where the devices
but also provide low latency in data delivery, efficient energy con-
communicate with each other automatically [10]. This type of com-
sumption scheme, high scalability to accommodate a large number
munication also requires extremely low latency for applications
of devices, and ubiquitous connectivity for users. In this section,
like vehicle-to-vehicle communication. The METIS project proposes
we describe these requirements and explain how they can be met
that traffic efficiency and safety should be a typical application test
by potential solutions.
case where the latency is critical in system evaluation. One typical
scenario presented in METIS is intelligent traffic systems in which
2.1. High data rates vehicles require timely exchange of data to avoid accidents [11].
The requirement of low latency will also improve the user expe-
The metric of data rate has been the most important evaluation rience for currently existing applications such as multi-user gam-
factor over generations of wireless communication networks. With ing. This demand in latency reduction requires technological inno-
the advent of mobile Internet and services such as high-definition vation in waveform design as well as a flexible architecture in the
(HD) video streaming, pervasive video and video sharing, virtual higher layers of the network, which can be addressed by wireless
reality available on mobile phones, as well as the proliferation of software-defined networking.
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 19

2.3. Low energy consumption With the increase in density of the base stations and the number
of devices connected, as well as the introduction of femtocells and
The 5G networks are expected to support the IoT devices picocells, the number of handovers that the base station should
[12] which are basically some sensors that gather information handle will increase by at least two orders of magnitude. To sup-
about an environment and transmit it to a central server. These port this demand, novel handover algorithms and techniques that
devices are mostly low-power, low-cost devices with lifespans as provide improved coverage in cell edge areas are required. Another
long as several years. Since these devices are not always connected related issue is the authentication and privacy concerns related to
to the base station and are only switched on occasionally, their the handover [16]. The delay to contact the authentication server
battery life cannot afford the process of synchronization with the for each handover will be hundreds of milliseconds which would
base station every time, as the synchronization step costs more en- be intolerable for 5G applications. Also, given the use of higher fre-
ergy than that of actual data transmission. This specific case in IoT quency bands in millimeter wave, the transmission range of sig-
requires that the radio access technique for 5G support loose or nals is greatly reduced. Hence, maintaining connectivity becomes
no synchronization. Moreover, this type of service also puts con- a great challenge for 5G. For mission-critical services, the require-
straints on the computational power for decoding, the length of ments on high reliability as well as connectivity should always be
header, and packet forwarding scheme, etc. guaranteed.
With the increasing number of connected smart devices, the
number of base stations required to support these devices will also 2.6. Improved security
escalate. Because of the deployment of small cells, the base station
will be densified. This foreseeable trend demands the base stations The security aspect of wireless network recently attracts high
to be energy efficient since even a small improvement in energy attention, especially after 2015, when the applications of mobile
efficiency will translate to huge energy savings in large scale. payments and digital wallet became popular [17]. In retrospect of
the previous generations of systems, the general purpose of se-
2.4. High scalability curity is to protect basic connectivity and maintain user privacy.
However, since the 5G system will ultimately face the dramati-
To support increasing amount of mobile devices that connect to cally increasing data traffic in the entire network, the requirement
the wireless network and communicate with each other, network of security of 5G should not only be limited to providing trust-
scalability becomes an important factor in design of the next gen- worthy connectivity to users, but also improving the security on
eration wireless communications network. The increase in number the whole network, addressing concerns on authentication, autho-
of devices is further aggravated by the myriad of IoT devices and rization as well as accounting, developing novel encryption pro-
vehicle-to-vehicle communication technologies that are expected tocols, and safeguarding cloud computing and management activ-
to surge in the 5G cellular network. Fueled by this smart equip- ities. For example, the security concerns are increasing since the
ment proliferation, it is expected that the number of devices con- introduction of near field communication (NFC) technique, which
nected to the cellular network will grow to 50 billion by 2020 [13]. not only enables close proximity data transmission, but also may
Consequently, a highly scalable network that can efficiently accom- cause identity leaks. The 4G networks were not able to develop a
modate this upsurge in number of devices is required. High scal- unified standard to protect users’ personal information, which will
ability is also critical to performance of current and emerging ap- be fully addressed in the 5G networks.
plications, such as the IoT services, autonomous vehicles, etc. In In particular, as the IoT will come to its prime time in 5G net-
the case of autonomous vehicles, prompt communications among works, the processes of authentication, authorization, and account-
them at high traffic densities necessitate the scalability of cellular ing (AAA) for interconnected devices should be granted with fine-
network [14]. grained protecting mechanisms. Network operators, device manu-
In fact, the scalability of a network requires full-scale upgrade facturers, as well as standardization bodies should work together
in all network layers. On the physical layer, there should be enough on designing services, products, and protocols that can substan-
frequency spectrum resources to support high volume of signaling tially protect the users subscribed to the 5G wireless network.
and data transmission. The network infrastructure should also be Most importantly, as the Internet has become one of the indis-
able to control the transmitted power adaptively for channel es- pensable infrastructures of the society similar to power grids, there
timation and to minimize interference. Scalability also influences should be enhanced federal regulations on liabilities and obliga-
the design of efficient media access control (MAC) protocols to tions on wireless network security.
accommodate large amount of connected devices. Scheduling and The aforementioned requirements are summarized in Table 1,
multicasting protocols based on geographic locations of users can with their specifications and the associated enabling solutions.
greatly reduce the latency and increase the spectral efficiency. At The 10 enabling technologies will collaborate to shape the 5G
the network and transport layers, high scalability requires net- network architecture. The radio access network architecture will
works to deploy intelligent routing algorithms for huge groups of be enhanced with the deployment of wireless software-defined
users to provide fast and reliable connections. For users with high networking, network function virtualization, ultra-densification,
mobility, the vehicular network should also satisfy the scalability device-to-device communications, millimeter wave, massive MIMO,
by providing efficient and reliable handoffs along the directions and new radio access techniques, while the core network will
of the moving routes, as well as designing dynamic routing algo- evolve with key roles played by wireless software-defined net-
rithms based on the user movement prediction [15]. The scalability working, network function virtualization, Internet of Things, green
in networks will be achieved by changes in all layers from radio communications, big data and mobile cloud computing. In the fol-
access to core networks, using wireless software defined networks lowing sections, all the key technologies are elaborated in detail.
and network function virtualization.
3. Wireless software-defined networking
2.5. Improved connectivity and reliability
One of major challenges for 5G communication networks is
Apart from the aforementioned requirements, coverage and the design of flexible network architectures, which can be realized
handover efficiency should also be improved for a better user ex- through the software-defined networking (SDN) paradigm [18]. As
perience, particularly when millimeter wave spectrum is exploited. SDN has emerged primarily for data center networks and for the
20 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

Table 1
Key requirements and enabling solutions for 5G wireless networks.

Requirements Specifications Enabling solutions

High data rates 10 Gbps peak data rate; Millimeter wave communications;
100 Mbps cell edge data rate; Massive MIMO;
Enhancing mobile broadband services. Ultra-densification.
Reduced latency 1 ms end-to-end latency D2D communications;
Big data and mobile cloud computing.
Low energy 10 0 0 times decrease in energy consumption per bit; Ultra-densification;
Enhancing massive machine type communications. D2D communications;
Green communications.
High scalability Accommodating 50 billion devices Massive MIMO;
Wireless software-defined networking;
Mobile cloud computing.
High connectivity Improving connectivity for cell edge users Ultra-densification;
D2D communications;
Wireless software-defined networking.
High security Standardization on authentication, authorization and accounting Wireless software-defined networking;
Big data and mobile cloud computing.

Table 2
Existing WSDN comparison [33].

WSDN Architecture Network scalability Community

SoftAir [20] SD-RAN & SD-CN High Academia/industry

SoftCell [21] SD-CN Low Academia/industry
Cloud-RAN [22] SD-RAN Low Academia/industry
ProgRAN [23] SD-RAN Low Industry
SK Telecom [24] SD-RAN & SD-CN Low Industry
DOCOMO [25] SD-RAN Low Industry
CONTENT [26] SD-RAN & SD-CN Low European Union
OpenRoads [27] SD-WiFi; programmable flow tables High Academia/industry
OpenRadio [28] Programmable data plane Low Academia
CloudMAC [29] SD-MAC in WANs Low Academia
Odin [30] SD-MAC in WANs Low Academia
ADRENALINE [31] SD-CN Low Industry
SoftRAN [32] SD-RAN Low Academia/industry

next-generation Internet [18,19], its main ideas are (i) to separate fic classifier, and carrier scheduling, maximizing the entire system
the data plane from the control plane and (ii) to introduce novel performance.
network control functionalities based on network abstraction. To- To the best of our knowledge, SoftAir is the first comprehensive
wards this, by utilizing the concept of SDN, we have proposed a solution suite for 5G cellular systems that accelerates the innova-
new architecture for wireless SDN (WSDN), called SoftAir [20], as tions for both hardware forwarding infrastructure and software al-
well as its management tools. gorithms, enables efficient and adaptive resource sharing, achieves
In the literature, several studies [20–32] have explored inte- maximum spectrum efficiency, encourages the convergence of het-
grating SDN with 5G network from academic and/or industrial erogeneous networks, and enhances energy efficiency. In the fol-
perspectives. The comparison of these WSDN solutions is sum- lowing subsections, we first briefly overview the key concepts of
marized in Table 2, and more details can be found in [33]. It is both SoftAir architecture and its detailed management tools that
worth to note that several common crucial problems exist in all both serve as the foundation for WSDN design in 5G systems.
of these solutions (except SoftAir [20]), which significantly de- Then, we introduce critical open problems in WSDN domain. As
grade the network scalability. Specifically, first, none of these de- WSDN and NFV are the fundamental building blocks of the en-
signs have a complete architectural solution for 5G systems, but abling technologies in 5G systems, rest of the 10 key technolo-
only cover partial designs such as software-defined core network gies in this paper closely relates to these two technologies, and
(SD-CN), software-defined radio access network (SD-RAN), etc. On the connections will be explicitly explained later in the respective
the other hand, SoftAir has scalable software-defined planning that sections.
brings ubiquitous software-defined design across core and access
networks and has seamless OpenFlow incorporation that enables
unified centralized management for the control plane. Second, pri- 3.1. WSDN architecture
ori arts, such as in the well-known Cloud-RAN [22], have lim-
ited scalability due to their coarse-grained fronthaul network de- The overall architecture of SoftAir for WSDN in 5G systems is
coupling, as I-Q data needs to be transmitted across processing shown in Fig. 2, which composes of a data plane and a control
servers. This will place a huge burden on the optical transport plane. The data plane, which includes SD-RANs and a SD-CN, is an
network, especially when dealing with advanced wireless tech- open, programmable, and virtualizable forwarding infrastructure.
nologies. Instead, SoftAir provides fine-grained fronthaul network The control plane mainly consists of two components, which are
composition that eliminates the fronthaul bottleneck and remark- network management tools and customized applications from ser-
ably increases system capacity. Last but not the least, while miss- vice providers or virtual network operators. Three key elements of
ing in existing solutions, SoftAir equips a complete management scalable SoftAir architecture are introduced as follows, which serve
kit, including control traffic balancing, network virtualization, traf- as the fundamental building block for general WSDN.
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 21

Fig. 2. Network architecture of SoftAir [20].

Scalable software-defined planning: To increase network scal- dardized interfaces, such as CPRI (Common Public Radio Interface)
ability, SoftAir decouples control and data planes for both SD-RANs or OBSAI (Open Base Station Architecture Initiative) interface.
and the SD-CN. As shown in Fig. 2, a SD-RAN consists of software- While the distributed RAN architecture has recently received
defined base stations (SD-BSs) that jointly form a baseband server significant attention for WSDN from both industry and academia
(BBS) and connect with numerous remote radio heads (RRHs). (e.g., Cloud-RAN [22]), it faces two fundamental limitations. First,
Moreover, by interconnecting physical links, the SD-CN contains they cannot achieve scalable PHY/MAC-layer function virtualiza-
software-defined switches (SD-Switches) that have flow tables to tion. Second, they do not support NFV as the SD-CN. One major
route traffic. The control logic of SD-RANs (e.g., PHY/MAC/network consequence of realizing all physical layer functions in the remote
functions) and the SD-CN (e.g., network management and opti- server is that whole I-Q plane data needs to be transmitted to the
mization tools) is implemented in software on general purpose server, placing a huge burden on the fronthaul network. Towards
computers, network servers, and remote data centers as high- this, SoftAir adopts a new fine-grained fronthaul network decom-
performance controllers. Specifically, the network operating sys- position architecture by leaving partial baseband processing at the
tem in the control plane collects information about network states RRH (e.g., modulation/demodulation), while implementing the re-
such as device status, link utilization, link delays, etc. The infor- maining baseband functions (e.g., source coding and MAC) at the
mation collected is used to create a high-level network abstrac- BBS. Hence, the decomposition not only preserves sufficient flexi-
tion that helps the computation of optimum forwarding rules and bility offered by the distributed RAN architecture, but also elimi-
efficient resource allocations. These control policies are then fed nates the possible fronthaul network bottleneck. In short, with re-
back to SD-BSs and SD-Switches for execution through standard duced data rate requirements, SoftAir offers excellent cooperative
(southbound) interfaces, e.g., OpenFlow [34] and SNMP (Simple gain and evolvability by allowing the aggregation of a large num-
Network Management Protocol). Also, the control plane (or con- ber of technology-evolving RRHs at BBS through the diverse, cost-
trollers) provides the network abstraction to applications, such as efficient, CPRI-supported fronthaul network topologies, and over
mobility management, security, load balancing, etc., running upon different fronthaul mediums.
it through (northbound) interfaces. Hence, SoftAir establishes a Seamless OpenFlow incorporation: Unlike existing distributed
ubiquitous software-defined planning from access networks, the RAN architecture [22], SoftAir incorporates OpenFlow protocol into
core network, to the Internet, effectively realizing multi-controller SD-BSs as shown in Fig. 2. Specifically, SD-RANs implement an
scenarios and optimum managements for large-scale wireless OpenFlow interface for each SD-BS by utilizing Open vSwitch (OVS)
systems. [35], which is an OpenFlow-capable software switch that can easily
Fine-grained fronthaul network decomposition: Inspired by be realized in BBS. With OVS, each SD-BS can interpret, exchange,
the concept of NFV, SoftAir supports fine-grained fronthaul net- and respond to OpenFlow protocol messages. Equipping SD-BSs
work decomposition through dedicated base station NFV. As in with OpenFlow capabilities provides a unified interface to control
shown Fig. 2, to enhance network flexibility, SoftAir simultaneously and manage base stations with different wireless standards, thus
realizes the physical-, MAC-, and network-layer function virtual- leading to a multi-technology converged RAN that allows smooth
ization for RAN and thus forms SD-RAN. Here, the proposed SD- transitions among different radio access technologies. Adopting the
RAN follows a distributed RAN architecture, and each SD-BS is split common OpenFlow interface on both SD-Switches and SD-BSs also
into hardware-only radio heads and software-implemented base- promises the transparent interconnections between SD-CNs and
band units. These RRHs are connected to the baseband units on SD-RANs and allows the unified management over the entire Soft-
BBS through fronthaul network (fiber or microwave) using stan- Air.
22 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

3.2. WSDN management tools the scaling of SD-RAN, as SD-BSs will have uneven and widely
varying load differences. Therefore, to overcome the problem of
To enable promising features and to maximize the overall sys- WSDN scalability, three perspectives should be considered. First,
tem performance, SoftAir supports cloud orchestration that auto- based on the network topology, traffic flow statistics, application
mates the configuration, coordination, and management of soft- QoS requirements, and peak flow initiation rate, a capacity analysis
ware and software interactions in the cloud environment [20]. This of a single controller should be addressed in terms of throughput
is the first time that a complete management kit is investigated in and latency. Next, an analysis and comparison of centralized and
detail for WSDN. distributed architectures for WSDN controllers should be studied
Mobility-aware control traffic balancing: Rather than exploit- to include signaling overhead characterization and the dimension-
ing costly out-band control from a separate control channel, the in- ing for each architecture. Third, in the distributed case, the opti-
band signaling is favored and adopted gradually in practical SDN mal number and location of SDN controllers should be determined
implementation. To this end, each SD-Switch or SD-BS (i.e., BBS) with regard to the requirements of network state synchronization
needs to send control traffic, such as route setup requests for new as well as the corresponding consistency performance. A possible
flows and real-time network congestion status, to the controller(s). solution towards this can be found in [36], in which an on-demand
Then, based on the real-time control messages and unique mobile deployment of local controllers is proposed based on network
traffic features from mobile users, the controller optimizes the best dynamics.
routes for data flows according to dynamically changing traffic pat- Control traffic latency: Another critical problem with WSDN
terns and flow QoS requirements and sets up the flow tables of SD- is the control traffic latency, which possibly prohibits online and
switch or SD-BS along the optimal path via certain secure protocols adaptive traffic engineering. In particular, as SD-Switches or SD-BSs
(e.g., OpenFlow [34]), thus enabling highly efficient data transmis- have to contact controllers for routing information before forward-
sions and superior link utilization. ing data traffic, the increased latency from control traffic should
Resource-efficient network virtualization: To support be addressed carefully, especially for delay-sensitive applications.
Infrastructure-as-a-Service (IaaS), network virtualization capacity The latency resulted from OpenFlow switch processing is roughly
is critical to enable a wide range of emerging applications, provide 10 ms, which measures the time of a packet being sent until ac-
network services at the edge, as well as achieve network slicing. knowledged by an unloaded controller. Another type of latency oc-
While wireless network resources are limited, resource-efficient curs in ring protection, which is around 50 ms, in order to provide
wireless network virtualization is highly desirable, and SoftAir the time of target restoration of a SONET ring. The highest latency,
equips two levels of hypervisors. Network hypervisor focuses which is shared-mesh restoration, takes about 200 ms under the
on high-level resource management, which determines how to circumstances of voice calls dropping or ATM circuit rerouting be-
distribute non-conflicting network resource blocks among virtual ing triggered [37]. To overcome the difficulties, two issues should
network operators based on their demands; wireless/switch hyper- be considered: mobile feature extraction of SD-RANs and timely
visor is a low-level resource scheduler that enforces or executes control traffic balancing. First, to provide mobility-aware design,
the resource management policies determined by the network hy- based on users’ movement and transmission behaviors from ap-
pervisor by employing a variety of wireless resource dimensioning plications, the mobile traffic distribution can be well established to
or wireless scheduling. These hypervisors aim to slice network capture the unique mobile feature of traffic in SD-RAN. The derived
resources and provide differentiated services to different slices. traffic distribution thus serves as an accurate mobile traffic model
Distributed and collaborative traffic classifier: SoftAir enjoys and will be used in the control traffic balancing design. Second, the
great capability to significantly improve spectrum efficiency from most important requirement in the design of control traffic balanc-
a designated distributed, collaborative traffic classifier at SD-BS. ing is its computation time. To this end, in [37], the control traffic
This classifier collaborates with a global traffic learner at the net- balancing problem is first formulated as a nonlinear optimization
work controller(s) to achieve fast, fine-grained, and accurate traf- that finds the optimal control traffic forwarding paths and mini-
fic classification. Specifically, it identifies applications, QoS require- mizes the average control traffic delay.
ments, and stochastic features associated with each traffic flow and Infrastructure-as-a-Service (IaaS) with programmable data
thus preserves satisfaction to mobile users even with highly bursty plane: Given that 5G network will support different types of
mobile data traffic from diverse network applications and QoS emerging applications and services, such as machine-type commu-
demands. nication, smart grid, mobile virtual network operators, and over-
Dynamic carrier scheduling: With the aid of the proposed the-top content services like Netflix, it is essential for WSDN to
fine-grained fronthaul network decomposition, SoftAir enables dy- offer IaaS for all kinds of services complying various service level
namic carrier scheduling that exploits the baseband pool from BBS agreements. Also, as servers that run the control plane software
and allocates different carrier resources to RRHs in an energy- of WSDN will be located in a third-party cloud computing envi-
efficient manner. In particular, this scheduling scheme computes ronment, service providers should facilitate IaaS to meet this geo-
resource requirements with respect to different users and loads. separation need as well. Moreover, while WSDN comes with the
Then, it allocates carriers dynamically to satisfy the given require- promise of flexibility and configurability, this could be fully ex-
ments with minimum energy consumption and thus facilitates ploited only when SD-Switches and SD-BSs can be programmed
green communication in 5G SoftAir. to execute control policies. In order to accomplish this, both SD-
Switches and SD-BSs must provide a rich set of interfaces for the
3.3. Open problems controllers to program. Facing these challenges, resource-efficient
hypervisors [38] should be established to slice the infrastruc-
While SDN is a promising technology for future 5G networks, ture and to efficiently bring virtual infrastructures. Also, new ar-
WSDN is still in the stage of infancy. In the following, we list chitectures that speedup switches and base stations need to be
several essential open problems that need to be addressed before developed [39].
WSDN could be readily adopted. Software-defined mobility management and green commu-
WSDN scalability: While several SDN solutions have been pro- nication: In 5G wireless systems, mobility management faces
posed recently for WLANs in an enterprise or a university, scaling significant new challenges, including minimum signaling over-
WSDN with respect to SD-CN is a huge challenge that needs to be head and QoS guarantees during handoff rerouting. To address
addressed. Furthermore, the advent of IoT devices also exacerbates these challenges, a new software-defined mobility management
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 23

framework that ensures proper interaction with legacy protocols

and systems should be proposed to leverage the multi-service
multi-technology multi-network converged WSDN for low-cost,
QoS guaranteed, and seamless mobility services. Moreover, from
dynamic carrier scheduling, green communication can be easily
realized through network managements of WSDN. Enabling the
WSDN controller for resource allocations requires the knowledge of
link loads and channel characteristics, specifically the delay char-
acteristics, loss rate, and stability. Given that these parameters are
affected by multi-path fading, extensive measurements of all these
parameters being sent to WSDN controller will cause a surge in
control traffic and need to be examined [40].
Reliable, efficient traffic engineering: To support various and
stringent end-to-end QoS requirements of real-time applications si-
multaneously, WSDN should support QoS-aware routing upon des-
ignated network virtualization, thus enabling multiple isolated vir-
tual networks to fulfill respective QoS requirements of users’ ap-
plications. This means WSDN should not only slice network re-
sources for multiple virtual networks so that they can simultane-
ously share the same physical network architecture, but also of- Fig. 3. Overall architecture of NFV.
fer a QoS-aware traffic engineering accordingly [41]. On the other
hand, to serve best-effort traffic which typically does not have
specific QoS requirements (the counterpart of real-time traffic), limitations of the corresponding services. For example, the SGW
WSDN should support throughput-optimal collaborative schedul- and PGW can be co-located with the base station, avoiding long-
ing [42] that maximizes network throughput while reducing delay distance links. Also, NFV can remove bottlenecks in core networks
as much as possible. Throughput-optimal scheduling aims to op- by pooling computing resources. Table 3 compares and briefly
timally and collaboratively distribute network resources of a clus- summarizes these existing NFV solutions.
ter of RRHs among their users based on the statistical properties A summary for the solutions that integrate NFV and WSDN de-
of their traffic flows so that the overall throughput of the SD-BS signs can be found in [33]. Note that from its sophisticated ar-
is maximized. Specifically, each SD-BS can be optimally formed so chitectural design, SoftAir [20] embraces a complete solution for
that fine-grained coordination among a cluster of RRHs is enabled. NFV and perfectly brings the synergy for WSDN and NFV. Specif-
Such promising feature along with the built-in traffic classifier ically, regarding NFV (network cloudification), SoftAir introduces a
on each SD-BS, for the first time, allows performing throughput- fine-grained fronthaul network decomposition that decreases the
optimal collaborative scheduling on a SD-BS. volume of data transferred between RRHs and BBS by implement-
ing partial baseband processing at RRHs and remaining process-
4. Network Function Virtualization (NFV) ing at BBS. Moreover, as each SD-BS has OpenFlow interface im-
plemented, SD-BSs have a uniform interface across different wire-
NFV, a complementary concept to SDN, enables the virtualiza- less standards. These OpenFlow capabilities enable vertical mo-
tion of entire network functions that were tied to hardware be- bility where users can roam across SD-BSs with different wire-
fore to run on cloud infrastructure. Specifically, in conventional ar- less standards. With these, SoftAir also leverages high performance
chitectures, operators purchase and install proprietary devices to controllers and optimizes network management to improve the
deploy each network function, while specialized hardware is usu- scalability and flexibility. In the following, we first briefly overview
ally very expensive but barely configurable. Because of these con- the NFV architecture according to ETSI [44] and then introduce
straints, network operators face problems in low agility, leading several crucial open problems in NFV domain.
to longer product cycles. NFV, emerging as a breakthrough in 5G
cellular systems, decouples physical hardware and underlying net- 4.1. NFV architecture
work functions, and let the network functions run centrally on
generic cloud servers, thus providing advantages in scalability and According to ETSI, the overall architecture of NFV consists of
flexibility. NFV greatly reduces capital expenses (CAPEX) required four key elements: NFV infrastructure (NFVI), virtual network func-
to buy hardware devices and saves operating expenses (OPEX) by tions (VNFs), hypervisors, and NFV management and orchestration
aggregating resources for virtual network functions that run on a (NFV MANO), as shown in Fig. 3 [44,47,48]. Specifically, the main
centralized server pool. component of NFV is VNFs which are software implementations
One use case of NFV is the IoT. Specifically, as the IoT paradigm of network functions, running on a generic cloud infrastructure.
and its widespread applications create an explosive number of de- VNFs are deployed upon the NFVI that includes virtual computa-
vices connected to the network, NFV would go a long way to re- tion, virtual storage, and virtual network resources. These virtual
duce the cost of IoT devices by reducing the functionalities in de- resources, created by hypervisors, bring virtualization over physical
vices and virtualizing those functions. For example, in [43], a sen- hardware resources within the network. In particular, hardware re-
sor function virtualization that enables distributed intelligence in sources might include networking (e.g., switches and RRHs), com-
IoT is proposed through the deployment of modular blocks. More puting (e.g., server), and storage (e.g., data centers) infrastructures.
details of interaction between IoT and NFV will be discussed in Moreover, the NFV MANO framework controls the provisioning of
Section 9. As being another use case of NFV, the usage of pro- VNFs, the configuration of VNFs, and the infrastructure they run
prietary equipments in mobile core networks causes inflexible de- on. MANO can also chain several VNFs to activate an end-to-end
ployments and time consuming, costly network re-planning. In- service. We briefly explain each component [49] as follows.
stead, NFV can make core networks more intelligent, scalable, re- NFV Infrastructure (NFVI): NFVI is the infrastructure platform
silient, and flexible by virtualizing various entities (e.g., MME, PGW, over which the VNFs are deployed. Specifically, NFVI is the col-
SGW, home subscriber server, etc.) and eliminating geographical lection of the physical hardware which include the computing,
24 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

Table 3
Existing NFV comparison [33].

NFV NFVI Hypervisors NFV MANO

SoftAir [20] Fine-grained NFV; solve Network and Wireless/Switch hypervisors Flexible platform for fully or partial centralized
redundant I-Q transmissions architecture (via local control agents)
OpenRoads [27] Simple decoupling via FlowVisor [45] and SNMPVisor Simple extension to WiMAX AP
OpenFlow
OpenRadio [28] Enable multi-core DSP Abstraction layer for PHY and MAC Focus on hardware design without control
architectures functions plane
CloudMAC [29] VAPs for IEEE 802.11 stations Dynamic spectrum use; on-demand AP; VAPs with virtual machines and dumb WTPs
downlink scheduling
Odin [30] LVAPs for IEEE 802.11 stations Hidden terminal mitigation; AP and its No concrete solution for radio heads
function hypervisor
ADRENALINE [31] SDN integrated IT and Network OpenStack [46] Deploy end-to-end VNFs
Orchestrator (SINO)
SoftRAN [32] Rough NFV with big-base Antenna hypervisor No concrete solution
station abstraction

storage and networking hardware and software resources. In gen- MAC layer and baseband processing solutions (e.g., modulation, de-
eral, the NFVI could be geo-distributed over several locations (e.g., modulation, channel coding, source coding, and MIMO precoding)
customer premises and cloud data center) so that the network con- at physical layer. Moreover, various device techniques could also be
nectivity between the locations is also considered as a part of NFVI. adopted to reduce power consumption, such as the migration to
Virtual Network Functions (VNFs): A network function refers energy-efficient hardware, the reduction of CPU clock speeds, the
to the functional component of network infrastructure that pro- shutdown of unused components, and energy-efficient function de-
vides a well-defined functional behavior and external interfaces. sign [49]. Also, reducing the power consumption on data transfer
Some examples of network functions are DHCP servers, firewalls, among virtual machines should be greatly explored.
NAT, gateways, and MME. VNFs are the software implementations Performance and Portability: Performance is a big concern
of network functions that are deployed over virtual resources cre- in virtualization environment, but fortunately the OVS technology
ated over the NFVI. [35] is quite flexible to leverage various offload technologies in
Hypervisors: Hypervisors provide the abstraction of virtual re- hardware-level (e.g., Linux NAPI or Intel DPDK). Also, since NFV
sources over physical hardware for VNFs to run on. That is, hy- virtualizes service components (does not change the original tele-
pervisors setup the virtualization layer that offers virtual machines com service portfolio), OSS/BSS support will be kept as the same.
(e.g., virtual compute and storage) and virtual networks over physi- Specifically, VNFs will be run in a generic industry standard server,
cal computing, storage and networking resources. Hypervisors also and these servers for the NFVI must be manufactured without the
provide a logical slicing of actual network infrastructure into vir- knowledge of function types that would run on in the future. To
tual networks for efficient management. this end, given the unacceptance of quality degradation from NFV
NFV Management and Orchestration (NFV MANO): NFV migration, these generic severs must provide the service compa-
MANO provides the framework for managing and orchestrating all rable to that running on specialized hardware. Moreover, software
infrastructure resources. Specifically, it provides virtual machines, techniques (like multi-threading, independent memory structure to
configures the virtual machines and the physical infrastructure, avoid OS deadlock, and direct access to I/O interfaces) can be ex-
and manages of physical resources for virtual machines. The MANO plored for higher performance [50]. To evaluate the performance
contains three functional blocks: NFV orchestrator, VNF manager, in virtual networks, a phased cloud diagnosis [51] is proposed that
and virtual infrastructure manager. First, the NFV orchestrator han- collects appropriate statistical data. Another possible approach for
dles on-boarding new network services and VNFs, manages global solving the performance degradation is to consider the tradeoff
resources, and validates and authorizes NFVI resource requests. between flexibility and performance by implementing some high-
Second, the VNF manager manages the life-cycle of VNF instances performance functions on hardware while virtualizing other net-
and coordinates the configuration and event reporting between work functions.
NFVI and network management software. Last, the virtual infras- Management and Orchestration: Although NFV brings net-
tructure manager controls and manages the compute, storage, and work flexibility and configurability, it also requires modifications
network resources of NFVI. to the current network management systems and will change
the way networks are deployed and managed. First, a rich set
4.2. Open Problems of programmable interfaces should be developed that supports
complex services for heterogeneous networks [52]. Second, while
While bringing network flexibility, NFV is not yet fully mature most of NFV frameworks only define intra-operator interfaces, the
to be embraced by the industry for 5G systems. There are several inter-operability between operators needs to be investigated. Third,
technical gaps that need to be closed as described in the following. when the management systems are implemented in cloud virtual
Energy-efficient Network Cloudification: One of the major machines, due to dynamic loads, the virtual machine for current
challenges for adopting NFV in 5G systems is to offer network VNF program should be decided on the fly. Forth, as NFV provides
cloudification in an energy-efficient way. Specifically, using hyper- the ability to elastically support network function demands, the
visors and moving functionality to the cloud, NFV should support automation is the key to scale the NFV environment in regard to
virtual functions from physical-, MAC-, to network-layers regard- reliability and availability, stability, resource allocation, and energy
ing energy consumption. In particular, network-layer function vir- efficiency. Last, innovative algorithms to manage the instantiation
tualization decouples the routing function from hardware routers and migration of virtual functions should be developed to improve
and centralizes it at network controller(s) through an open net- the effectiveness of NFV systems.
work interface (e.g., OpenFlow [34]). MAC/PHY-layer function vir- Security and Privacy: VNFs, which NFV runs on third-party
tualization exploits software-defined radio technologies to imple- public cloud, correspond to subscriber services and contain
ment resource (e.g., time, frequency, and code) sharing schemes at
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 25

personal information about the subscriber. Hence, the security of particular, the 60 GHz band, which shows a significant 15 dB/km
the information being processed becomes important [53]. Specif- atmospheric attenuation, is only feasible for short range communi-
ically, as the VNFs might be run over several third-party servers, cations [57]. For other candidate millimeter wave bands (e.g. the
service providers have no direct control of the data, which raises 28 and 39 GHz bands), the effect of atmospheric attenuation is
new concerns on data privacy and security. Moreover, the new de- less dominant yet still stronger than the microwave frequencies,
signed elements, like MANO and NFV, may also bring additional making them only feasible for small range deployments. The field
security vulnerabilities to intrusion detection systems [53]. Last, measurements conducted in urban areas in New York City at 28
the underlying shared NFVI (i.e., computing, storage, and network- GHz showed that the line-of-sight (LOS) transmission provides the
ing) also generates new security threats. best signal strength with a pair of high-gain directional anten-
nas [58]. However, in non-line-of-sight (NLOS) scenario when di-
5. Millimeter wave rect propagation is blocked, additional path loss is observed which
greatly degrades the received signal strength and further limits the
As suggested by the Federal Communications Commission (FCC) propagation distance. From the same field measurements in the
and many research groups from academia and industry, millimeter dense urban environment, it is reported that the average NLOS
wave shows great potential in enabling gigabit per second through- path yields a path loss exponent of 5.76, which is much higher
put with its great amount of available bandwidth. Millimeter wave than the path loss exponent of 2 in free space [59]. The block-
usually refers to the frequency bands at 30–300 GHz, but re- age effect at millimeter wave frequencies is also inevitable, which
searchers also prefer to add the nearby centimeter frequency bands can be caused by common objects, such as buildings, vehicles,
from 24 GHz to 28 GHz in discussion. The unlicensed and light- pedestrians, and even tree branches and foliage [60]. Consequently,
licensed millimeter wave frequency bands provide great amount the coverage range of a millimeter wave base station is dramati-
of available spectrum resources that can support the requirements cally limited to only a few hundred meters, even with high-gain
for high data rate, low latency, and flexible connectivity for a wide (more than 50 dBi at transmitter and receiver combined) anten-
range of users. nas used at both the transmitter and receiver. An outage study
In October 2015, the FCC proposed the rules to release millime- from the measurement campaign in New York City showed that
ter wave spectrum bands for 5G services [54]. Specifically, the fre- the maximum cell radius is 200 meters at 28 and 73 GHz, and the
quency bands that the FCC proposed to authorize for small cells major blockage effect is caused by the dense buildings in urban
deployment are the 28 GHz (27.5–28.35 GHz) and the 39 GHz environments [61].
(38.6–40 GHz) bands. Since wireless local area network (WLAN) at
60 GHz has been deployed as IEEE 802.11ad standard and demon-
strated great success on high data rates in short range indoor en- 5.2. Spatial multiplexing and beamforming
vironments, the FCC proposed to release an adjacent 64–71 GHz
band for similar indoor short distance operations. This additional One solution to compensate the high atmospheric attenuation
bandwidth of 7 GHz will further enhance the data rates for WLAN brought by millimeter wave is to use large antenna arrays at the
services. In addition, the FCC promoted a hybrid licensing scheme base station. Instead of the omnidirectional antennas used by past
which leaves the choice of operation to service providers in the 37 and current generations of wireless communication systems, 5G
GHz band. The hybrid band can help facilitate the development of will deploy a large set of narrow-beam directional antennas that
novel broadband network techniques and also serve as a supple- can pinpoint to multiple users simultaneously. Therefore, large an-
ment to traditional cellular networks [55]. tenna arrays are essential for serving a wide range of users in
In current 4G LTE standards and widely accepted channel mod- highly populated areas.
els, there are several commonly studied propagation scenarios, Among all antenna array techniques, spatial multiplexing and
namely indoor office, indoor shopping mall, rural, dense urban, ur- beamforming are two most popular approaches in research [62].
ban, and highway, etc. Those scenarios determine the arrangement Normally, spatial multiplexing utilizes the multiple antenna ele-
of wireless communication network infrastructures and configura- ments to transmit multiple independent and separately encoded
tion of key channel parameters, including the specific electrical and data streams. While each data stream travels through different
mechanical downtilt angles of the base station sector antenna, the propagation channels, the receivers also equipped with multiple
number of uniform linear arrays to install in each sector antenna, antennas will be able to receive and reconstruct the original data
the height of base stations, and the projected coverage of a base transmitted with better spectral efficiency [63]. Spatial multiplex-
station sector, just to name a few. In the next generation of wire- ing is mostly effective when the propagation paths are rich and
less system, these legacy cell arrangements will continue to ex- can support multiple data streams for transmission, and also when
ist and serve the users. The new communication scenarios should the channel has high signal-to-noise ratio (SNR) which will not af-
include the direct communications among devices at millimeter fect the signal strength when the original signal has to be split into
wave, a millimeter wave multi-user communication scenario, and a multiple data streams [64]. Beamforming, on the other hand, com-
vehicular communication scenario for autonomous vehicles at mil- bines antenna array elements at the base station adaptively and
limeter wave. Fig. 4 depicts the combination of currently supported utilizes the beamforming weights multiplied on each element to
and predicted scenarios that are discussed below in this section. control the directions to which the data streams are transmitted
[65]. In this way, specific users will be directed to receive the data
5.1. Propagation characterization while others will not be interfered. Beamforming performs well
when the channel has limited power or low SNR, as a result of the
Many recent research from both academia and industry can be combination of multiple antenna elements coherently [66,67]. As
found on the topic of millimeter wave communication. One re- for millimeter wave, since the wavelength becomes much shorter,
search focus is the propagation characterization of millimeter wave the physical size of antenna will accordingly be smaller. The hard-
frequencies, which is essential in designing the new wireless prop- ware prototype design for both spatial multiplexing and beam-
agation channel model for 5G network. When the frequency goes forming are carried out in various research groups. In one millime-
up to the range of millimeter wave, the energy absorption caused ter wave beamforming prototype tested by Samsung Electronics in
by atmosphere, rain, and snow becomes increasingly prominent, South Korea at 28 GHz carrier frequency, researchers use a 32 ×
which causes limitation in signal transmission distance [56]. In 32 antenna arrays as the RF transceiver configuration. Their test
26 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

Fig. 4. The vision for 5G wireless communication scenarios. (a) Urban environment with high user density; (b) device-to-device communications; (c) vehicular communica-
tions among autonomous cars and users; (d) dense population in a stadium; and (e) indoor millimeter wave communications.

results demonstrate great potential of millimeter wave beamform-

ing in the small cell deployment scenario [68].

5.3. Terahertz (THz) band for beyond 5G

Besides millimeter wave, there is also increasing attention fo-

cusing on even higher frequencies in the Terahertz (THz) band. The
THz band spans the frequencies from 0.3 THz to 3 THz, with a
broad overlap with millimeter wave frequencies. Although the IEEE
802.15 Terahertz Interest Group was chartered in 2008 to explore
the feasibility of Terahertz band communications, there is yet no
regulation on this extremely high frequency bands, which makes it
a “no man’s land” [69]. However, the THz band offers great amount
of spectrum resources, which show the great potential to support
Fig. 5. The architecture of WSDN-enabled millimeter wave radio access network.
data rates of more than 100 Gbps or even 1 Tbps [70]. Additionally,
with the advances in semiconductor transceiver design at high fre-
quencies, the THz band transceiver architecture also can be back- to maximize the gain for specific user while minimizing the in-
ward compatible with the millimeter wave bands, making the THz terference to nearby RRHs. The channel state information (CSI) es-
band the next frontier in research of wireless communication net- timated by the SD-BS can determine the hosting scheme (i.e. the
works, and creating a paradigm shift from traditional large scale corresponding RRHs) and the transmission mode (i.e. legacy LTE
transceiver design to nanoscale architecture. or millimeter wave) for satisfying QoS. Since the millimeter wave
Current research has covered various topics such as channel propagation links can be constantly changing in channel condi-
modeling, transceiver device design, antenna architecture design, tions such as LOS probability, received signal strength, level cross-
waveform design, network and transport layer design, etc. [69– ing rate, etc., the SD-BS and its associated RRHs will make dynamic
73]. New research areas are being developed, such as Wireless power level adjustments to ensure the system performance. Mil-
NanoSensor networks and the Internet of NanoThings [74,75], and limeter wave backhaul is also very promising in supporting ultra-
the novel concept of Ultra-Massive MIMO by 1024 × 1024 is being high speed wireless backhaul compared to fiber. WSDN architec-
introduced as a means to increase communication distance [76]. ture with millimeter wave backhaul links can solve the deployment
Hence, the THz band is envisioned as one of the key enablers for restricts that fiber faces. The centralized and scalable management
ultra-high-speed short range communications. of the SD-BS will efficiently improve the performance in RAN and
reduces the cost of CAPEX/OPEX. As shown in Fig. 5, the millimeter
5.4. WSDN enabled millimeter wave radio access network wave RAN is enabled by WSDN.
architecture
5.5. Open problems
With the foreseeable deployment of millimeter wave which
means cell size will shrink greatly, the challenge is brought by co- Millimeter wave shows many attractive characteristics on large
ordination among millimeter wave base stations with a dramat- amount of available spectrum resource and acceptable propagation
ically increased density. Hence, the WSDN offers a flexible and behavior in LOS environment, which makes it promising in serving
efficient approach for network resource management of the RAN the 5G technological revolution. The advantage of broad bandwidth
for millimeter wave communications in SoftAir, as illustrated in offers great potential for developing new air interface technologies.
Fig. 2 [20]. The large antenna arrays will also bring high spectral efficiency,
In the coordination of millimeter wave base station, the SD-BS high throughput, and additional channel gain when deployed
can manipulate the adaptive antenna arrays at the RRH in order along with millimeter wave system. However, there are still some
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 27

technical obstacles that must be resolved before moving from the can be reused by indoor access points and outdoor small cell base
prototype to commercial market. The problem of high power con- stations.
sumption of large number of signal processing components is still Dynamic power control: Another open problem of using mil-
an open issue. Also the challenge of channel estimation and pre- limeter wave for communication is how to dynamically control the
diction for adaptive power control also need solutions. In this sub- propagating signal strengths and track the signals in order to adapt
section, we discuss the open problems of millimeter wave commu- to actual channel environments. As mentioned before, at millime-
nications and potential solutions. ter wave frequency the NLOS transmission will cause a big drop in
3-D channel modeling: One open challenge calls for a uni- signal strengths for 15 ∼ 40 dB as compared to LOS, the channel
form 3-D channel model to cover all types of communication sce- condition thus varies significantly with surrounding environments,
narios. So far, there is no widely accepted conclusive model that walking pedestrians, and traffic. In order to deploy millimeter wave
can comprehensively characterize the wide millimeter wave fre- for 5G network, researchers need to dynamically track and con-
quency bands. Research groups have been promoting their devel- trol the power of the signal so that users can experience satisfying
oped channel models in hope for standardization body adoption. high data rate services while base stations have cost-effective en-
The 3-D channel model proposed by a research group in New York ergy consumption.
University is based on measurements conducted in dense urban Cell search: In the 4G LTE systems, the cell search process is
environment, which is considered where the 5G requirement of designed based on omnidirectional transmissions of base stations,
high data rates will mostly need to satisfy [77]. The U.S. National which allows the user equipments (UEs) to establish links to the
Institute of Standards and Technology (NIST) are supporting a col- serving cell. However, in millimeter wave communication, direc-
laborative research effort called “5G mmWave Channel Model Al- tional transmission will be deployed in order to compensate the
liance” to collect insights to systematically characterize mmWave high propagation loss in omnidirectional transmission scenario. But
channels. In Europe, a model specifically design for the 60 GHz the directional transmission scheme causes complexity issues to
frequency bands named “Millimeter-Wave Evolution for Backhaul initial cell search, as both the UEs and BSs will search and lo-
and Access (MiWEBA)” proposed a quasi-deterministic model for a cate suitable transmission space to establish links. One proposed
wide variety of communication scenarios, and took the backhaul directional cell search process provides a solution by letting the
links into model considerations [78]. BS broadcasts synchronization signals at random directions [81].
In general, channel modeling approaches including both statis- The detection process is based on a generalized likelihood ratio
tical and stochastic methods are considered, which also have their test and the results demonstrated that digital beamforming out-
distinct advantages and drawbacks. The statistical models are cre- performs analog beamforming in cell discovery.
ated based on huge amount of data collected from actual field User scheduling and congestion control: Under the circum-
measurements, which is the reflection of the real-world propaga- stances of high path loss and blockage effect in millimeter wave
tion channel, but the data collection process is extremely time- communications, a key solution is to utilize the gain in highly di-
consuming and normally vary a lot with different geographical lo- rectional beamforming to establish a multihop relaying network
cations. The accuracy of the data is also limited by the measure- architecture. Since the multihop architecture provides link diver-
ment equipment as well as the post-processing methods. On the sity and cell range extension, more users can be connected to the
other hand, the stochastic model is based on a step-by-step pro- BSs. However, in the air interface system design, an open problem
cedure following the probability distributions of channel model comes to user scheduling and congestion control scheme. A possi-
parameters given the environment geometry. No exhaustive labor ble solution is to deploy a dynamic duplex scheme to allow max-
is required to complete the channel modeling, but the stochastic imum resource allocated for users, as in contrast of static duplex-
model is not convincing unless validated using either field mea- ing schemes (i.e. TDD and FDD) in current 4G network [82]. How-
surement data or the widely accepted ray-tracing method. For the ever, more solutions are still needed as well as field measurements
5G communication system, a hybrid model, which is the combina- validations.
tion of both statistical and stochastic model, is preferred. The hy- Hardware limitations: One of the open challenges is the high
brid model should consist accuracy, spatial consistency, and envi- cost and energy consumption of RF hardware. Due to the frequency
ronment scalability [11]. In the hybrid channel model, the first step characteristic difference between millimeter wave and legacy mi-
can be implemented by the stochastic model where key channel crowave, the transceiver equipment designed for lower frequencies
parameters can be characterized by probability distributions. Once is no longer suitable for millimeter wave frequencies. As the wave-
the model is completed in various environments, field measure- length decreases, the antenna arrays used at both base station and
ments can be conducted to verify the stochastic model and provide mobile terminals will be replaced accordingly. The good thing is,
modifications in case of discrepancy. with the smaller wavelength at millimeter wave, the physical size
Since in real world environment objects are constantly moving, of antennas will shrink, allowing more antennas to be built in the
the transition between LOS and NLOS will be frequent and unpre- same area as before. However, the downside comes to the equip-
dictable, especially at higher frequencies. In current 3GPP channel ment of RF chains after the antennas. In current wide-deployed
models, the probability of LOS is modeled as a function of sepa- digital transceiver architecture, each antenna is connected to an
ration distance in 2D and the height of the UE in only two en- RF chain, which consists of power amplifiers (PAs), analog-to-
vironments, indoor and outdoor [79]. However, the actual proba- digital converters (ADCs) and digital-to-analog converters (DACs).
bility of LOS is more complicated, and should not be limited in The power consumption of the large number of PAs, ADCs, and
two static environments. Therefore, the millimeter wave channel DACs is huge, making this conventional architecture not feasible
model should also consider the dynamic signal strengths and de- for millimeter wave antenna array architecture. And the manufac-
velop a flexible link transition model between LOS and NLOS. Also, turing cost for large number of antenna elements per base sta-
comparing with current microwave frequencies, millimeter wave tion also prohibits the deployment of traditional architecture. In
demonstrates high penetration loss for concrete walls, making the another word, from an energy efficient and cost-effective perspec-
signal difficult to receive by an indoor receiver from an outdoor tive, the number of RF chains should be less than the number of
base station [80]. However, this can be a benefit since it eliminates antenna elements. A hybrid beamforming architecture is promoted
the interference from nearby cells. Hence in terms of small cell which could take advantage from both analog and digital beam-
system design and base station setup, the same frequency bands forming to achieve higher throughput [83].
28 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

value that still satisfies the operating requirement. If heavy data

traffic occurs in one massive MIMO cell range, all the components
will be active to provide fast data transmission to all users. The
benefit of scalability of massive MIMO is quite attractive to 5G ar-
chitecture researchers and will play a significant role in the 5G net-
work design.

6.1. Time-division duplexing in massive MIMO

As discussed in Section 5.2, massive MIMO is operated on the

principle of spatial multiplexing which requires that the propaga-
tion channel has rich paths. The channel estimation is dependent
on the number of orthogonal pilots sent by the base station to the
user terminal, which is proportional to the number of antenna el-
Fig. 6. Three types of layout of massive MIMO base station antenna array: linear, ements at the base station. Therefore, the frequency-division du-
rectangular, and cylindrical arrays.
plexing (FDD) mode for current LTE systems is not feasible for
massive MIMO channel estimation, since it requires that both the
uplink and downlink channel knowledge should be known to the
Adaptive beam-steering technique: Another one of the im-
base station in order to achieve optimal efficiency. In the case of
portant open challenges in millimeter wave is to support beam-
massive MIMO where several hundreds of antenna elements are
steering where the beams can be dynamically steered to follow
configured at the base station, the downlink channel estimation
user mobility as well as to overcome blockage effects. This adaptive
will become a bulky task for user terminals to estimate all the pi-
beam-steering technique involves designing intelligent prediction
lots from the base station. Hence the channel estimation becomes
algorithms, rapid beam switching methods, as well as overcoming
computationally cumbersome and is not realistic as the user ter-
the hardware constraints for beamforming.
minal has the nature of mobility. The solution is by utilizing the
reciprocity of the uplink and downlink in TDD for channel esti-
6. Massive MIMO mation [85,86]. The user terminals will send pilot waveforms to
base stations for uplink channel estimation, and then the base sta-
In addition to millimeter wave communications, massive MIMO tion can relatively easily perform downlink estimation based on
is proposed as another key enabling solution for 5G networks. the received uplink pilots. The detailed quantitative analysis can
Different from the multiuser MIMO (MU-MIMO) deployed in the be found in [87–89]. Some studies propose a technique called joint
4G system in which only several tens of antenna components are spatial division and multiplexing (JSDM) to exploit the correlation
built on base stations and user terminals, massive MIMO designs of channel vectors and demonstrate the feasibility of FDD systems
hundreds of antennas at base stations to further increase capac- in achieving comparable performance as its counterpart of TDD in
ity and system throughput, as well as to provide more degrees of massive MIMO [90].
freedom [84]. Massive MIMO utilizes spatial multiplexing (previ-
ously discussed in Section 5.2) and time-division duplexing (TDD)
to serve multiple users on the time-frequency resources. Fig. 6 6.2. Full-dimension MIMO
demonstrates three possible types of layout of massive MIMO an-
tenna arrays at base station, namely linear, rectangular, and cylin- One of the scalable implementation of massive MIMO is full-
drical arrays, respectively. The different layouts of massive MIMO dimension MIMO (FD-MIMO), which takes the spatial multiplex-
base stations are designed for different deployment scenarios. For ing in the elevational plane into consideration and utilizes the 3D
example, the linear and rectangular antenna arrays can be installed channel model to achieve better performance [91]. In traditional
on the outer surface of high-rise buildings to serve indoor users spatial channel modeling standards such as 3GPP [92], only sig-
with various heights and outdoor pedestrian users. The rectangu- nals propagating in the azimuth direction are considered, while the
lar antenna array can also serve multiple users using spatial mul- vertical beam pattern and antenna downtilt angles are normally
tiplexing technique. For moving users, the cylindrical arrays can be fixed. This type of antenna array configuration has been proven to
deployed to track the users’ moving routes. be inadequate [93,94]. Recent research realizes the importance of
The benefits of massive MIMO include a huge increase in spec- providing one extra degree of freedom in the elevation domain.
tral efficiency, a reduced latency, and a scalable air interface struc- For example, in some cases where users are distributed at mul-
ture. Since there are many more antenna elements at the base sta- tiple heights in a building (e.g. a stadium or a shopping mall),
tion than those at mobile terminals, the signal pre-coding proce- the FD-MIMO can dynamically steer the vertical beams toward the
dures will be simplified, resulting an enhanced spectral efficiency users. Hence, FD-MIMO shows great potential in massive MIMO
[85,86]. The reduced latency is achieved by the law of large num- application which utilizes the full 3D spatial channel. With the ex-
bers embedded in the nature of massive MIMO which eliminates tra degree of freedom brought by the antenna beams arranged in
the frequency dependence of the channel and the frequency selec- the elevation direction, the system capacity can be increased by
tive fading will not affect the signal strength. Consequently, each 30% [95].
transmitted data stream will no longer suffer from channel distor-
tion and can arrive at the receiver with reduced latency. Due to the 6.3. WSDN-based massive MIMO coordination scheme
large number of base station antenna components, the actual air
interface structure architecture can be designed to have flexibility Since the deployment of large scale antenna arrays in massive
and scalability depending on the real-time data traffic conditions. MIMO will yield huge amount of spatial domain information about
For example, if a cell served by a massive MIMO base station only the channel knowledge, the estimation of channel state informa-
connects to several user terminals, not all base station antenna el- tion requires timely update and quick computations. WSDN pro-
ements will be necessarily active, so the energy consumption per vides an alternative for dynamic spatial domain radio resource pro-
each individual antenna element can be reduced to its minimum cessing [96]. Traditional BS can only collect partial channel state
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 29

information due to the distance limitation, whereas the interfer- cussed architectures are analog, digital, and hybrid arrays. In the
ence from other cells are normally neglected, which can be prob- analog array architecture, all signal processing components are lo-
lematic. By the use of RRH and SD-BS, the WSDN can collect chan- cated at the central processing unit and thus can save space in the
nel knowledge with a widely extended coverage, thus enabling in- base station setup; however, the distance between the antenna and
telligent interference cancellation and spatial domain utilization. signal processing units introduces unneglectable signal loss and
noise, as well as interference. In the digital array architecture, there
6.4. Open problems is a complete RF signal processing chain after each antenna ele-
ment, which reduces the unfavorable noise and interference, but
In this subsection, several open issues in massive MIMO are de- the cost for large number of RF components constrains market ap-
scribed and possible solutions are presented. plication, also the physical size for the signal processing chains is
Near-field vs. far-field channel models: In cellular systems, it insurmountable. Considering the above tradeoffs, the hybrid archi-
is usually assumed that signals, which are considered as plane tecture seems an optimal solution. In the hybrid architecture, the
waves, propagate in the far-field region. This assumption is valid digital baseband processing is done at the central processing unit
for current and legacy generations of wireless systems because of and the RF beamforming only requires a series of phase shifters
the relatively low carrier frequencies. According to the definition of which does not need complicated circuitry [103]. However, the im-
Fraunhofer distance which determines the boundary between the plementation of the large antenna array architecture still holds
near and far fields [97], we know that with higher frequency (i.e. challenges.
smaller wavelength) in millimeter wave and larger dimension of When the antenna elements are placed close to each other,
the antenna, the Fraunhofer distance will increase and thus the there is an inevitable electromagnetic interaction (i.e. mutual cou-
near-field region expands. In that case, the basis of plane wave pling) among them [85]. One solution is to mitigate the effect of
propagation will not be exclusively valid for propagation channel coupling via impedance matching RF circuits with multiple ports
characterization. To overcome this challenge, a distance-dependent [104]. However, as the number of antenna elements increases and
channel model that considers the distinct effects in both near and the spacing shrinks, the influence of mutual coupling will become
far fields should be developed in order to fully understand the prominent and hence new approaches should be considered.
channel characteristics with huge antenna array aperture and small Distributed massive MIMO: The concept of distributed massive
wavelength. MIMO is brought to address the issue of antenna array form factor
Pilot contamination: With the base station requirement of constraints, by utilizing the formation of arbitrarily large number
hundreds of antenna components to be installed to serve multi- of antenna elements opportunistically in both transmission and re-
ple users at the same time, the challenges of synchronization and ception [105]. With the flexible scheme in antenna array form fac-
multiuser interference suppression occur. In small-size MU-MIMO tors, higher transmission and reception directivity can be achieved,
systems, the problem of synchronization is solved by performing as well as the increase in spatial and spectral efficiency. How-
a calibration process through exchanging pilot waveforms between ever, the challenge resides in the synchronization of RF chains in
base station antennas and a so-called “reference” antenna [98,99]. massive MIMO antenna arrays. If digital beamforming is deployed,
This calibration process is very sensitive to the layout of the ref- there should be a synchronization signal sending to each assigned
erence antenna and not feasible for the large scale massive MIMO RF chain to establish the array. The algorithms of maximum ratio
case. In traditional MIMO there exists the problem of pilot contam- combing (MRC) and minimum mean square error (MMSE) beam-
ination, which is caused by the reuse of limited pilot waveforms in forming are being discussed in [106]. The complexity of the algo-
multiple cells in channel estimation. But the effect of pilot contam- rithms and the power consumption for the arbitrary array synchro-
ination is much serious in massive MIMO. Study shows that with nization are still calling for better solutions.
incremental antenna elements without upper bound, the problem Energy consumption and cost: The most energy-consuming
of pilot contamination can lead to severe interference among cells part in a massive MIMO system is the RF signal processing com-
when each cell serves the maximum number of user terminals ponents, which include ADCs, DACs, PAs and some other necessary
[86]. electric components. The manufacture of microwave components
Several solutions are proposed to mitigate the effect of pilot is already matured so the cost will not be a major issue. How-
contamination. One is to decrease the pilot reuse factor, which can ever, if massive MIMO is to be used at millimeter wave frequen-
raise the distance between two cells utilizing the same frequency cies, the design and manufacturing will set higher requirements
band. The technique of fractional frequency reuse (FFR) proposed and thus cost will increase [86]. Testbeds of millimeter wave mas-
to maintain the original pilot reuse factors in cell edges and set sive MIMO system should be verified carefully before mass pro-
an additional pilot waveform to each cell’s center, in order to de- duction, and some key performance indicators (KPIs) (e.g. energy
crease the interference among cells [100]. Another solution is to efficiency) should be taken into evaluation.
utilize the blind pilot decontamination technique to avoid using pi-
lot waveforms in channel estimation [101]. This technique is based 7. Ultra-densification
on the eigenvalue decomposition of channel’s sample covariance
matrix, in the theory of random matrix. The third solution looks The traditional macrocell network architecture has served as
at new precoding algorithms, such as pilot contamination precod- the basic network architecture in the 4G system. Estimations have
ing (PCP) [102]. This algorithm allows multiple base stations with shown that the monthly mobile data traffic around the world will
the same pilot waveform to collaborate constructively and there- exceed 15 exabytes by the year of 2018 [11]. Such tremendous
fore cancel out the interference caused by pilot contamination. growth in bandwidth demand comes from the accelerated devel-
Very large antenna array design: As shown in Fig. 4, various opment of new smartphone, tablets, wearable devices, and other
massive MIMO base station layouts can facilitate diverse user sce- types of new devices that requires network access and desires de-
narios. However, the popular open issue comes to the design of cent connectivity for data roaming and video streaming. With the
such very large antenna arrays. Since the physical space for base unprecedented diversity of mobile devices, applications, and ser-
station equipment is often limited, research has been focused on vices, the 5G cellular network calls for high flexibility at all net-
how to fit hundreds of antenna elements into a limited area with- work layers. However, the traditional high-tower mounted base
out sacrificing system performance and reducing the complexity of stations cannot keep up with this increase. In order to accommo-
the signal processing hardware. Up to now, three commonly dis- date the ever-growing user terminals and offload the congested
30 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

Fig. 7. Ultra-densification as a combination of microcell, picocell, and femtocell.

data flow in 4G base stations, two major solutions favorable to can be improved. Second, it enables a more flexible network archi-
many are spatial densification and spectral aggregation, which can tecture planning. In traditional network planning hexagonal shaped
create more layers of cells and manage the shared spectrum re- cells are considered in almost all scenarios. Numerous results have
sources wisely to increase capacity. To be more specific, spatial proven that in real-world network hexagonal cells may not neces-
densification includes the idea of heterogeneous network (HetNet), sarily serve as the best option [107–110]. For 5G network, the cell
which has been brought to attention in the 4G LTE system. The size and shape should be able to have configuration flexibility in
group of smaller elements for HetNets including picocells, fem- order to achieve both better spectrum and energy efficiency. For
tocells, and distributed antenna elements will serve a significant instance, the concept of moving cells and nomadic nodes will pro-
role in enhancing system performance, as shown in Fig. 7. On the vide small cells with high mobility. Together with D2D commu-
other hand, spectral aggregation takes the challenges of aggregat- nications which will be discussed in later sections, the small cell
ing the non-continuous fragments in bandwidth resource, and uti- BS can be mounted on top of trains, cars, and buses, which allows
lizes spectrum sharing to coordinate the licensed as well as unli- users in close proximity to have network connectivity [11].
censed spectrum in order to achieve higher spectrum usage.

7.2. Spectral aggregation and spectrum sharing

7.1. Spatial densification
In addition to spatial densification where users are associated
A key performance indicator in network performance is the out- to different cells to increase network capacity, the spectral aggre-
age probability. Outage probability evaluates the quality of service gation provides an alternative to condensing the network through
of a cell on the quantitative basis of cumulative distribution func- the means of spectrum resource management. At the early stage
tion of the signal-to-noise-and-interference ratio (SINR). Under the of wireless communications, due to the limitations in signal pro-
premise that high SINR also leads to high data throughput, the cessing techniques and the fact that frequency spectrum resources
users in the center of a cell coverage will tend to have higher data were not envisaged as scarce resources, some spectrum resources
rates than the users at the edge of a cell. And in the multiuser experienced an extravagant deployment. For example, the analog
case, according to the Round-Robin (RR) scheduling algorithm, the TV used to cover a wide spectrum in the VHF to UHF range in the
user with higher SINR will always be assigned to reach system effi- 20th century. However, with the invention and widespread of digi-
ciency. However, it leads to unfairness as the users at the cell edge tal TV which takes much fewer bandwidth for signal transmission,
with lower SINR might not be served at all, causing unbalanced the analog TV gradually resigned globally in late 20 0 0s, leaving
network resource allocation and service degradation for edge users. the white space spectrum underutilized. This precious spectrum
In HetNets, the users originally located at the edge of traditional resource can provide effective broadband access and has shown
microcells will be covered by picocells or femtocells, depending on better performance than current Wi-Fi network. In 2010, the FCC
the cell size and environment, thus the distance between the serv- opened this spectrum white space for public use, and then the
ing base station to the user is shrunk with higher SINR at the re- IEEE standardized this spectrum as IEEE 802.11af for spectrum
ceiver, which in turn decreases the outage probability. sharing [111,112].
The idea of small cell is a realization of spatial densification in One way of integrating the spectrum required by new appli-
network architecture planning, which utilizes low-power nodes to cations into current network layout is through spectrum sharing
provide a cost-efficient alternative to reduce the load factor for tra- techniques with the help of cognitive radio technologies [113,114].
ditional urban microcells. The benefits of ultra-densification over In LTE-A, the approach of licensed assisted access (LAA) is deployed
spatial densification are twofold. First, it allows more capacity in to boost small cell coverage, where the licensed channels are pri-
the network. With multiple layers of cells in one area, a user can mary carriers, serving voice calls in both uplink and downlink.
be assigned to a cell that can provide the best channel through- And the unlicensed spectrum available in the network will serve
put. Hence the data traffic can be dispersed to different micro- UEs when they demand high-throughput data services. The unli-
cells, picocells, and femtocells. In this way the network capacity censed secondary users can obtain the dynamic spectrum resource
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 31

availability and thus make decisions on spectrum usage, without

causing interference or channel collision to licensed primary users.
Additionally, this idea of dynamic spectrum sharing can improve
the scalability of the entire wireless network. In a real-world sce-
nario, when an emergency alert needs to be spread from one cell
to a large area, the current available spectrum resources will be
deployed for that specific purpose, without causing interference to
occupied spectrum or collisions to busy channels.

7.3. WSDN-enabled ultra-densification

The idea of ultra-densification calls for the design of a scalable,

efficient, and flexible radio access network (RAN) for the 5G sys-
tem. The approach of logical separation between the control and
data planes in WSDN is considered promising to realize such re-
quirement. Under the premise of control-data separation architec-
ture in WSDN, there will be two types of base stations, namely
control base stations (CBS) and data base stations (DBS) [115]. The
CBS manages cell coverage and resource allocation, whereas the
DBS executes the signal transmission and reception, within the Fig. 8. The centralized authentication process enabled by WSDN controller in ultra-
control of CBS. In this way, all UEs within the cell range, no matter densified small cells.
busy or idle, are connected to the CBS, but only the busy UEs are
also connected to DBS for data transmission. The separation of the
control and data planes in WSDN can therefore improve the spec- downlink signal carrying data packets, respectively. On the other
tral efficiency and reduce the energy consumption of signaling at hand, the inter-layer interference, in a broad sense, comes from
the base station. all types of signaling and packet streaming from active user ter-
Though appealing in theoretical analysis, the control-data sep- minals and all available cell towers under the macrocell cover-
aration architecture faces a few challenges before full deployment. age [119–121]. Even though the small cells use low-power nodes
The backhaul between the CBS and DBS requires low latency and for smaller coverage area to suppress intra-layer interference, the
high data rate transmission to reflect the UE activity in a timely inter-cell interference generated by the higher power cells would
manner. However this increases the system overhead, which brings still cause service degradation. There are two solutions available to
the question on how to design an efficient signaling mechanism cope with the interference, which are the smart power manage-
between CBS and DBS [116]. ment and inter-cell interference coordination (ICIC). Although in-
Another challenge is the authentication of new UEs admitted to telligent power management research and ICIC solutions are avail-
another small cell upon handoff. Since in current wireless network able for deployment in current LTE network, more advanced and
architecture, the authentication servers are often located far away comprehensive solutions are necessary for network with new pro-
from the local RRH and BS, the round trip time for authentication posed 5G techniques such as massive MIMO and high frequency
request and confirmation is normally up to the scale of hundred systems [120,122,123].
milliseconds, which is unbearable for the 5G requirement on la- User scheduling and mobility management: In ultra-dense
tency [117]. Even with some simplified authentication algorithms networks, other challenges include the user scheduling and hand-
such as mutual authentication [118], the packet overhead is still off among cells. In the current 4G networks, the user will normally
large which makes the authentication process less efficient. be associated to the cell that provides the best connection. How-
Regarding the above open problems, WSDN can be utilized for ever, in case of the HetNet, a single user may find more than one
centralized orchestration, as shown in Fig. 8, which coordinates all cell available for data transmission and reception, on the norm of
handoff requests and confirmation messages, distributes forward- SINR. And the BS will also likely find more than one user equip-
ing instructions to SD-Switches, and assigns RRH to admit new UEs ment that indicates the best channel condition to serve. In this
to its cell range, as described in detail in Section 3.2. In this way, case, both parties in communication link should take more metrics
the solution provided by WSDN-enabled network architecture can into consideration. The power efficiency, which can be expressed
help resolve the latency issue in ultra-densification. as the output signal power divided by the input signal power,
WSDN will also serve as an optimal solution for the chal- can be assessed as another KPI [124]. When the SINR is not the
lenge of inter-cell interference, one of the significant issues exist- sole channel condition estimator, the power efficiency can be in-
ing in HetNets. Since the RRHs are controlled in a centralized man- troduced to the algorithms of cell association. However, when the
ner by the SD-Controllers which can adaptively manage the radio user is in motion, the signal strength may vary and thus the user
power of adjacent cells, the inter-cell interference will be greatly might experience handoff from one cell to another to remain con-
mitigated. nected. In HetNets, users may find more than one cells available to
handoff, therefore an intelligent handoff scheme in HetNet is re-
7.4. Open problems quired to address this problem.
Costs for small cells: At the early stage of small cell’s deploy-
Although many benefits will make ultra-densification an indis- ment in the 4G network, bulky transceiver equipment, complicated
pensable technique for realizing 5G, we still need to tackle the power amplifier and filtering circuits, and heavy cables were be-
challenges before integrating the novel network architecture. hind each small cell base station in the backhaul connections. Both
Intra-/inter-layer interference management and cancellation: the energy consumption and costs set limitation on the potentials
The source of interference can be intra-layer interference and inter- of small cells. An estimation shows that the cost of backhaul net-
layer interference. The intra-layer interference consists of uplink work occupies 30% of the total cost in wireless networks [125]. The
and downlink interference, which originate from the uplink pi- optic fiber is widely deployed in backhaul connection due to its
lot waveforms from user terminals to each small cell and the high bandwidth and robustness compared to the wireless option.
32 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

However, with the 5G projection of large number of small cells in-

stalled in dense urban areas, the fiber infrastructure needs to be
planned efficiently to minimize the cost and also maintain the net-
work resilience. One solution proposed is using WSDN (as describe
in Section 3) to assist a hybrid backhaul architecture, which con-
tains both optic fiber and millimeter wave connections [126]. The
simulated backhaul network was proven to provide cost efficiency
in dynamic resource allocation. However, this promising approach
still needs prototype validation before commercial deployment.
Management and pricing of small cells: Since the main idea
of small cell is to provide higher throughput for a smaller cov-
ered area, then for one macrocell coverage there should be many
more small cells to be deployed, which will induce the manage-
ment issue of small cell installation and maintenance. Network op-
erators are reluctant to increase their budget on OPEX of small
cells, and end-users are prone to choose the less expensive Wi-
Fi service which have similar functionalities, rather than having
the costly small cells. A moderate solution is applying WSDN and
NFV on small cell management, which can be done remotely and
with much lower OPEX [20]. The SDN controller will configure the
Fig. 9. The architecture of mobile cloud computing, in which mobile users and ter-
packet forwarding schemes for a group of small cells in similar minals can offload local data computation using the remote cloud servers.
operating conditions, thus reducing the manual configuration time
and resource.
that is traditionally assumed to be fully accomplished by the mo-
8. Big data and mobile cloud computing bile device. Since more and more sophisticated applications are in-
stalled and running on mobile devices with limited battery life and
Facing higher network throughput, the data storage techniques storage [128], the sheer amount of computation will significantly
will also experience evolution in 5G. Traditional data storage on lo- drain the battery which will be unfavorable to users. Especially
cal devices will no longer be capable to handle the exponentially for those smart devices with small size and weight, such as smart
increasing data cache especially when the user wants to down- watches and bracelets, the demands for high speed data compu-
load huge files or programs, or even stream HD-videos. In recent tation, large amount of local storage, and decent standby time,
years the cloud storage has become popular because of the con- will be some of the most competitive features that attract mar-
venient and on demand service. Users can upload data to cloud ket. Therefore, one benefit of mobile cloud computing is to help
servers through the Internet and therefore save the local storage increase the energy efficiency of the mobile device [129,130].
on their devices. In 5G networks, the cloud storage will become an As shown in Fig. 9, the mobile cloud architecture is fairly
inevitable part, and the mobile cloud computing will become the straightforward and can be divided into three major parts, namely
major method for data computing on a higher level. the cloud server, the mobile access point, and the mobile device.
Additionally, the sheer amount of data generated by mobile de- Similar as the cellular network, the mobile device will connect
vices and networks also intensify the importance of big data an- to the telecommunication network via the base station or access
alytics. Currently, most process of data analytics is performed on point. The telecommunication network is further connected to the
groups of physical machines, using the platform named Apache Internet, which allows the mobile users to access the cloud server
Hadoop (i.e. in-house Hadoop). The platform always requires in- through Internet connection [131].
frastructure to be installed and maintained manually, and also
lacks scalability if the already-installed platform cannot handle 8.2. Mobile edge computing
higher demands, which leads the reconfiguration process to be-
come costly and inefficient. In contrast to the in-house Hadoop, the Another important technique being developed for 5G cloud
big data in cloud offers advantages on cost, efficiency, and scalabil- computing is the mobile edge computing (MEC), which provides
ity. The big data analytics will be even more efficient and robust, the advantages of low latency and high bandwidth at the edge of
through the advantages of software-defined switching and control RAN [132]. The initial idea is proposed by the European Telecom-
brought by WSDN. munications Standards Institute (ESTI), which utilizes the dis-
The enhancements brought by big data and mobile cloud com- tributed computing environment in close proximity to UEs in order
puting to 5G are flexibility and high efficiency in data manage- to reduce network traffic congestion and therefore achieve faster
ment. Since more and more users choose to share files on cloud response time. The applications of MEC can be found in real-time
storage servers, mobile cloud computing offers data management content delivery, big data analytics where time-efficiency is critical.
on the cloud servers. On the contrary to cellular system where the
downlink plays a major role on system performance analysis, mo- 8.3. Open problems
bile cloud computing considers the uplink quality and the backhaul
network. The mobile cloud computing requires a stable connection This subsection discusses some open problems related to big
to the cloud server, which in turn sets high requirements for back- data and mobile cloud computing. The imbalance of network re-
haul data gateway to provide traffic-smart routes [127]. sources is one big challenge for mobile cloud computing, and the
importance of addressing serious cloud security problems should
8.1. Smart data management be emphasized. Since current cloud computing is facing transmis-
sion delays in wide area networks, we argue that some new rout-
One of the promising techniques of mobile cloud computing is ing schemes should be proposed to solve the open problem.
computation offloading, which is a process that allows the remote Limited network resources: The challenges of mobile cloud
server, or cloud, to share part of the computational responsibility computing include the current limited network resources and the
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 33

concern for data security. Since this technique has been available is used to build smarter systems like smart homes, smart cities,
for only a few years, there are many open issues need to be ad- intelligent transport systems, healthcare systems, etc [137]. With
dressed. Even though the network resources should be theoreti- applications in several diverse areas, the advent of IoT will en-
cally equally shared by users and service providers, due to different able a multitude of devices to be connected to the Internet. This
network standards around the world, users from different service paradigm shift from the concept of connected people to connected
providers might encounter difficulties in changing networks when things predicts that the number of devices in network increase to
they travel to other countries. Another critical issue is the data se- 50 billion by 2020 [138]. Because of the sheer volume of IoT de-
curity. With more advanced hacking technologies threatening the vices expected to be connected and the plethora of applications
Internet around the world, more research focus has been put on they could be used in, IoT devices play a significant part in design
network security [127,131]. Current cloud data computing has not of 5G systems and the several parts of the 5G network are being
yet employed robust security defense techniques [133]. This con- designed with the requirements of IoT in mind. Although some of
cern must be addressed before mobile cloud computing becomes the technologies discussed in other sections would enable IoT, we
a norm in 5G network in order to provide a safe environment for present them here separately because of the huge importance of
users to enjoy the benefits of 5G. IoT devices and the role they play in development of 5G systems.
Cloud security: In order to protect the data in cloud storage
from being hacked, the issue of cloud security and privacy should
9.1. Scalability
be fully addressed in the 5G wireless systems. In addition to re-
search on anti-hacking technologies, governments should establish
The widespread application of IoT and the cheap availability of
regulations on cloud security and take appropriate actions against
sensor devices makes it easier and cheaper to deploy many sensors
cyber crimes [134]. For instance, the European Union (EU) has set
to monitor an area or to perform a task. This is expected to cause
up a specific agency, the European Network and Information Secu-
an unprecedented surge in the number of devices connected and
rity Agency (ENISA), to assess the network environment and pro-
the density of devices. This outburst will pose several challenges
vide recommendations. Global initiatives should be encouraged to
in terms of providing reliable coverage to all the devices. As the
protect the cloud security against attackers, as the network has
number of devices increase and the size of the data they transmit
been shared almost by anyone from anywhere.
decreases, simplified access and core network techniques will be
Transmission delays: Another limitation of current mobile
required to accommodate all the devices and reduce the signaling
cloud computing is the delay caused by the data transmission in
and connection setup overheads for each packet. Apart from in-
wide area network (WAN). The transmission delay results from sig-
creasing the number of devices a BS needs to support, increase in
naling and traffic congestion in a highly crowded network archi-
the number of devices in the network also brings the problem of
tecture. In order to tackle this limit, the workload of WAN should
increased number of handoffs that the BS should be able to han-
be separated and released to smaller networks with reduced area
dle. This could be solved by adoption of technologies like SDN that
of coverage. One proposed method is to merge the idea of small
would enable smoother handoffs [16].
cells with cloud computing [128], which utilizes the small cell
base station to gain access to the cloud server, in order to provide
faster computation over the cloud with reduced network latency. 9.2. Intelligent processing and storage
This mobile computation offloading framework is operated on the
joint basis of communication and computing, which brings more Although the sensor devices in IoT collect data, it is processed
promising enabling solutions to the 5G mobile cloud computing and converted to actionable knowledge and intelligence only at
techniques. the backend servers. With the data being collected from IoT de-
Worldwide standardization: Since the Internet is a public vices becoming increasingly complex, cloud servers are often de-
property shared by users worldwide, the mobile cloud computing ployed to collect the data and process it to get insights. Compared
will also be adopted to facilitate boundary-less cooperation among to standard cloud providers, IoT cloud services have different re-
different countries, as predicted in the 5G networks. In the US, quirements in terms of communication protocols, software layers,
the National Institute of Standards and Technology (NIST) has ini- and provisioning models [139]. Although elasticity by providing
tiated the cloud computing program in order to regulate the cloud on-demand computational power is one of the basic features of
computing practices, which can support interoperability, portabil- cloud computing, it should be made scalable to the extent of ac-
ity, and security requirements [135]. The International Telecom- commodating the enormous number of devices IoT would bring in.
munication Union (ITU) also published several recommendations Thus, the service level agreements (SLAs) provided by the cloud
and technical reports in the efforts to standardize and cultivate provider will be critical for time sensitive applications like report-
an ecosystem in the cloud computing [136]. However, the new ing equipment failure, fire, and other emergency reporting. Most
requirements and technologies emerging in the 5G networks will of the existing cloud services come with proprietary interfaces and
eventually drive the evolution of standardization in mobile cloud this will require customization of all IoT devices [140]. This issue is
computing. aggravated when multiple service providers occupy different parts
of processing in the same IoT network.
9. Internet of Things Another approach to ease the computational load at the cloud
servers is to move the computing service towards the edge of the
Internet of Things (IoT) is the key concept that is changing network. Instead of having a centralized cloud computing facility at
many perspectives in 5G cellular networks landscape. IoT is the some data center, the cloud servers and storage are placed near the
network of everyday physical objects, vehicles, appliances, devices, end users, typically in the same building. The physical proximity
buildings, etc. Some examples of the devices that would be part of of the computing server to the nodes decreases the delay due to
the IoT are microwave ovens, washing machines, wearable smart communication. Fog computing also provides location awareness
watches, and health monitors. These devices sense some informa- and geographic distribution features to IoT services [141].
tion and it is passed on to a remote server, mostly through the In- Apart from processing the data, storing of the data generated
ternet. The server can also remotely issue commands to control the by IoT devices will need smart and efficient solutions. These types
device. The information collected at the server is then processed of huge data which cannot be handled by ordinary computing
to gain knowledge about the underlying process. This knowledge hardware are called big data. As discussed in Section 8, although
34 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

there are big data solutions like Apache Hadoop and SciDB com- challenges that are yet to be solved to enable IoT services are dis-
mercially available, they cannot scale enough to meet the needs cussed here.
of IoT [12]. These technologies fall short particularly for applica- Low power consumption: Most of the IoT devices will be stan-
tions where real time actions need to be taken. An interesting ap- dalone sensors equipped with transceivers. These devices are bat-
proach to address this problem is to avoid sending redundant data tery powered and they die off when the battery is drained. Hence,
to the server. Several works on sensor networks have noted that the power consumed by the IoT devices has to be kept minimal,
the data collected by sensors is highly redundant [142]. Reducing in order to extend their lifetime. This makes power consumption
data redundancy could be achieved using some dimensionality re- at device one of the biggest challenges for supporting IoT devices.
duction algorithms like principal component analysis (PCA) or by Another interesting way to increase the lifetime of devices is by
using a smart gateway for the sensor network, which aggregates harvesting several forms of energy available in the environment
the data collected by the network before sending to the remote in which the devices are deployed and using it to power the de-
server [143]. This would reduce the transmission load on the cel- vice. For example, sensors deployed over open fields could use so-
lular network as well as the processing and storage load on the lar power, sensors in industrial automation could generate power
server. from vibrations. These new energy harvesting techniques provide
a sustainable solution to the power consumption problem. The en-
9.3. Network slicing ergy consumption of the device could also be decreased by opti-
mizing the software running on them.
The 5G networks are expected to serve not only mobile phones, Signaling and control overheads: Most of the IoT devices send
but also more diverse types of devices with different system re- data to the server periodically or only when an event occurs and
quirements and functionalities, such as smartwatches, tablets, mas- go to sleep after transmission. The devices are not active and con-
sive IoT devices, mission-critical IoT devices, etc. In order to pro- nected to the network all the time. Moreover, the data sent by the
vide efficient services to each type of devices, the concept of “net- IoT devices is much shorter compared to the packets sent by con-
work slicing” is being discussed, which offers the freedom for net- ventional applications such as video streaming. So the signaling
work operators to slice the RAN into multiple end-to-end virtual- and control overhead of the frames transmitted becomes signifi-
ized network [144]. Each of the sliced networks contains its own cant compared to the size of data in the frame. This makes the
radio access, transport, and core network dedicated to service a overhead in signaling and control signals a major part of the total
single type of device. These dedicated slices provide independent consumed power[145]. This overhead was reduced by using tech-
network sources so there is no interference among other slices. niques such as fast transition to idle as proposed in 3GPP Release
Network slicing offers a great boost for mission-critical IoT ser- 11. Approaches like connection-less communication that fits within
vices, as well as achieves a dynamic network resource allocation the 3GPP standards [146] and the use of WSDN/NFV based solu-
for different traffic scenarios. tions to reduce the signaling and control overheads should be ex-
plored further.
Although the number of IoT devices is going to be mammoth,
9.4. WSDN and NFV based IoT these devices are not connected to the network at all times. The
devices send data to the server once every time period or when
With the WSDN and NFV offering dynamic and scalable net- an event occurs and go back to sleep again. This is a significant
work configuration capabilities, they are seen as the key technolo- transition from the mode of operation of traditional mobile devices
gies that would enable accommodation of 50 billion IoT devices which are connected to the BS at all times. Consequently, the IoT
which are expected to be a key part of 5G cellular network by 2020 device has to synchronize with the BS every time it switches on
∼ 2030. WSDN will be able to intelligently route the huge traffic before it starts transmitting any useful data. Further, the size of
generated by IoT devices through underutilized network resources, data sent each time the device is turned on is small and hence the
providing an increase in performance of the network while spread- power consumed by the synchronization step is significant com-
ing the load across several devices in the network. The level of ab- pared to the power consumed for actual data transmission. So, the
straction provided by WSDN will also enable the IoT applications radio access technique to be used in 5G cellular networks should
to access the data and control the devices without the knowledge be able to support devices with no synchronization or at least sup-
of underlying infrastructure. This makes deployment and manage- port loose synchronization.
ment of IoT devices simpler. The WSDN and NFV based IoT could Low latency: Several applications that use IoT networks are
configure and manage the devices dynamically to optimize re- mission-critical and require very low latency. Applications like
quired performance parameters such as reducing power consump- autonomous cars, drones, eHealth monitoring, industrial process
tion of devices, reducing the packet delay, etc. The centralized con- monitoring are delay-sensitive and require ultra-low latency on the
trol of data flow offered by WSDN/NFV also enables and simplifies order of 1 ms. This very low latency could be achieved by a combi-
data aggregation. nation of several technologies such as new radio access techniques
Another advantage of deploying WSDN and NFV architecture and WSDN.
for the IoT network is the improved security. The security of data Interoperability and standardization Due to the diverse range
transmitted is particularly important for mission-critical applica- of applications, there are various types of IoT devices manufac-
tions. Since WSDN and NFV centralize the network management tured by several vendors and having different capabilities and re-
and control, they will be able to efficiently handle the network se- quirements. An IoT device could range from a simple temperature
curity at different layers. Further, as WSDN and NFV profile each sensor, which monitors temperature and sending it to a server
flow differently, they can handle the traffic from different IoT ser- every hour, to a sensor monitoring critical industrial processes.
vices which have different security profiles. This would ease the Among the IoT devices available in market today, most of them
unnecessary processing of less secure applications. deploy their own standards and interfaces for communicating with
other nodes and servers. This creates incompatibilities when sen-
9.5. Open problems sors from different manufacturers are used simultaneously. Further,
the interfaces at the cloud servers that collect information also
Since IoT is one of the crucial application scenarios of 5G net- need to be standardized to provide interoperability between cloud
works, supporting them is of paramount importance. The open services.
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 35

10. Device-to-device communications

In conventional cellular networks, the majority of communica-

tion occurs between BSs and devices. Even though two users are
in their range of direct communication, the link has to be estab-
lish through the BS, which is not efficient for real-time services re-
quiring high data rates and low latency. In order to enhance spec-
tral efficiency, the idea of direct device communications was pro-
posed in [147] to initially create multi-hop relays among devices
in cellular networks. Since then, several D2D communication use
cases have been widely investigated as part of the LTE-A networks,
in both licensed cellular spectrum and unlicensed bands. Details
about D2D communication classification can be found in a compre-
hensive survey in [148]. The 5G networks are expected to continue
supporting applications that require devices to set up direct com-
munication links with their peers [149]. The applications include
file sharing, gaming, and social networking [150].
The D2D communications will improve QoS of the cellular net-
work as relaying can help increase the data rates of the users at
cell edge. The other scenario where the devices communicate di-
rectly with each other will alleviate the data traffic between nearby Fig. 10. Device-to-device communication scenarios: (a) Device relaying with BS
devices and the BS. This can offload the BS capacity and simultane- controlled link establishment; (b) direct D2D communication with BS controlled
link establishment; (c) direct D2D communication with device controlled link es-
ously provide a better user experience. Meanwhile, D2D communi-
tablishment; and (d) device relaying with device controlled link establishment.
cations bring in three types of gains – proximity gain due to the
reduced distance between transmitter and receiver which allows
high data rates at low power consumption; reuse gain provided by establishment is partially or fully controlled by the BS. Since the
the reuse of spectral resources at other locations given that the BS coordinates the relay link establishment, the device should still
communication range is shrunk; hop gain obtained by using a sin- be in the coverage range of the BS. A scenario where this will be
gle link in D2D while the conventional communication through BS utilized is when the device is within the coverage area of BS but
requires both uplink and downlink communications [151]. the channel quality is poor that higher data rates cannot be sup-
In the application level, D2D communications will unlock sev- ported. But the BS can establish connections with its base data rate
eral innovative applications and services. One important applica- and the device can achieve high data rates through the established
tion is in the area of public safety and security [152]. In scenar- relay link. This type of relaying helps the devices to achieve higher
ios such as accidents in long tunnels, landslides, and earthquakes, QoS and improve the battery life.
where cellular networks are not available or collapsed, D2D com- Direct D2D communication with BS controlled link establishment:
munications can be readily used for rescue and relief operations. The transmission and reception devices are within the transmis-
D2D communications will be greatly helpful in machine-type com- sion range of each other and they communicate with each other
munications, especially vehicle-to-vehicle communications where without the help of BS. However, the BS helps with link estab-
it can be used to get information of accidents and other alerts lishment between the devices. This will enable several novel social
about routes, such as route finding in self-driving cars. For exam- networking applications. This scenario would be advantageous for
ple, the 3GPP Release 12 added D2D communications to proxim- file sharing over short distances as it can avoid the traffic through
ity services (ProSe) in order to enhance public safety [153]. D2D the BS and in turn, offloading the load from BS. It would also pro-
communications can also be used to connect several devices like vide higher data rates as the rate is no longer limited by the con-
smartwatches with smartphones. Although other technologies such nectivity and capacity of the link to the BS.
as Bluetooth also connect devices to mobile phones, a unified pro- Direct D2D communication with device controlled link establish-
tocol would enable lower cost and energy consumption of devices. ment: In this scenario, the source and destination devices are
On the perspective of user mobility, D2D communications offer a within the communication range of each other and they commu-
more reliable environment, where Bluetooth is not designed for nicate directly with each other without the help of BS. It should
high mobility environment. D2D communications will also unlock be emphasized here that the BS does not play any role in this sce-
new types of social networking applications where nearby devices nario. Since the BS is not involved in the setup, this scenario will
can establish direct connections. also enable communications between devices when they are out
of the coverage area. This will come in handy during emergencies
10.1. Types of D2D communication when the device cannot contact the BS.
Device relaying with device controlled link establishment: The
Several works incorporate D2D communications in the architec- last scenario is where the source and destination devices are not
ture of 5G cellular networks. A two-tier 5G cellular network with within the transmission range of each other. Under this circum-
the macrocell and device tiers is proposed in [149]. A device tier is stance, they establish a relay link between each other without the
introduced in addition to the conventional macrocell tier to enable help of BS. Since there is no involvement of BS, the source and
and manage the D2D communications. Depending on the level of destination devices are responsible for the link establishment.
control the BS has on the D2D communications and whether an-
other device is used as a relay, D2D communications can be clas- 10.2. WSDN and NFV for D2D
sified into four types as described below and visually illustrated in
Fig. 10. The versatile and flexible communication framework provided
Device relaying with BS controlled link establishment: A device by D2D can be exploited using WSDN and NFV. The flexibility
with poor coverage communicates with the BS by using the other and configurability that WSDN and NFV provide could be easily
devices to relay its information to the base station. The relay link leveraged to deploy a new paradigm of communication like D2D.
36 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

However, the integration of D2D communication with WSDN and other devices or BS. This operation requires strict resource coor-
NFV concepts is not explored in detail and this leaves many open dination in order to prevent undesirable interference and unex-
problems that should be addressed. One such problem is that pected collision. One solution is that devices intended to transmit
WSDN requires the controller to have a view of the entire network, or receive data can broadcast a busy signal to surrounding devices.
while in many D2D scenarios the decisions are taken locally by the The challenge of limited spectrum resources for autonomous op-
devices. This inherent difference in the requirement of knowledge eration can be addressed by the future utilization of higher fre-
poses a problem in integrating WSDN and NFV. A hierarchical ar- quency bands.
chitecture of mobile cloud formation initiated by the user device Spectrum sharing with cellular systems: Currently the D2D
is proposed in [154]. The mobile clouds proposed are local clusters communication is operated underlaying the cellular networks,
of devices that can communicate with the cloud head. The hierar- which will inevitably cause interference to the existing cellu-
chical architecture enables the distribution of controller function- lar networks. A better spectrum sharing and resource allocation
alities using local and central SDN controllers. The cloud head con- scheme is needed to optimize the cooperation between the two
ducts the formation and management of the devices locally while types of services. A dynamic power control scheme should also be
the central SDN controller supervises the entire network and per- applied on D2D communication devices. The research in [158] pro-
forms resource allocation and routing functionalities. Due to the posed an admission control method on the basis of QoS to realize
strict hierarchical architecture, the scenarios of D2D communica- spectrum resource allocation.
tions that the architecture proposed in [154] will support are lim- Device pairing and link setup: The discovery of device pair
ited. The fact that the WSDN controller has a global view of the and link setup also bring additional challenges to HetNet opera-
network and resource utilization should be exploited intelligently tion [159]. In current LTE systems, there are some network-assisted
to modify the routing and flow tables at the BS or the user device device discovery algorithms discussed. These algorithms either use
to enable D2D communication whenever it is feasible. exhaustive search to locate the pairing device which is computa-
tionally costly (i.e. the Kuhn-Munkres algorithm), or only one side
of the communication is active in pairing selection [160]. The high
10.3. Vehicle-to-vehicle communications
computation load drains device battery fast, making it unfeasible
for long packet transmission or large file sharing. Therefore some
Two important use cases of D2D communications are vehicle-
new device pairing and link setup algorithms are proposed. One of
to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communica-
methods described in [161] includes two steps in pairing strategy.
tions, which can be generalized as the V2X communications. The
The first step is to establish a group of devices that are consid-
vehicles communicating with each other as well as with the in-
ered admissible candidates, which are active and nearby devices,
frastructure will aid development of several technologies like au-
and the second step is the process of resource matching and power
tonomous cars, automated traffic violation monitoring, etc. The
allocation. This method is shown to reduce the computation com-
V2X communications have several distinct challenges due to spe-
plexity while increasing the spectral efficiency.
cific deployed conditions. For example, the speed of the vehicles
Security and privacy: Since the devices are connected directly
would be several tens of kilometers per hour and they should be
and user data is being routed via devices, security and privacy
able to communicate with the vehicles on the opposite direction
must be maintained. This can be accomplished by assigning a set
during the short time when they cross each other [155]. The high
of trustworthy device group in which the devices are authorized to
density of traffic movement also increases the node density of ve-
communicate through. The devices within a trustworthy group can
hicles. A detailed list of QoS requirements for the V2X communi-
set some encryption between them to prevent other users from
cation scenarios is presented in [156].
gaining access to the data being shared.
Currently two major technologies are being considered for V2X
Interference management: In traditional cellular networks, the
communications, namely IEEE 802.11p and the aforementioned LTE
BSs handle the task of interference management among users.
ProSe. Since IEEE 802.11p uses a contention-based medium ac-
Since there is no central controller in D2D scenario, interference
cess technique, its throughput and delay performance degrade at
management will be an issue. The interference caused can be of
high user density. Furthermore, the range of 802.11p is also lim-
two types: with the other D2D communications as well as an
ited, requiring more hops, higher protocol complexity, and longer
ongoing communication with the BS. Approaches like resource
overhead [157]. The LTE ProSe is designed for public safety and
pooling [162], admission control and power allocation [158], re-
consumer applications specifically. Thus, it does not fit the V2X
lay selection and game theoretic techniques like non-cooperative
requirements of high mobility, differentiated QoS, and broadcast
game and bargaining can be used to moderate the interference
communications. There are more open challenges in D2D com-
[163,164].
munications that 5G is expected to address to enhance system
Resource control: Besides the technical challenges given above,
performance.
another practical and administrative problem that should be con-
sidered is control and pricing of D2D services. For example, suffi-
10.4. Open problems cient incentive must be provided for the devices that relay other
device’s data. Otherwise, the relaying device can refuse to do so as
The introduction of D2D communication brings in a lot of open it would be a waste of its resources, such as the device’s battery
problems. The primary challenge is discovering the peer devices and memory.
and identifying the type of services they support, especially in sce- 3-D channel model with spatial consistency: Currently, most
narios where connection is to be established without assistance wireless communication systems are deploying channel models
from the BS. Peer discovery using beacon signals will be time that are based on the assumption of stationary channel, which
and energy consuming. Another challenge is to make decisions on shows advantage on time complexity, reliability, and good estima-
which service to use among direct D2D communication. Analyzing tion to channels with fixed links. However, in the scenario of D2D
the QoS that could be achieved using each of these methods and communications with both terminals in movement, there has been
then deciding which mode to use will again not be energy efficient. an increasing demand on characterizing the rapid time-varying
Resource management for autonomous operation: In the au- nonstationary channel to increase channel modeling accuracy for
tonomous operation mode for D2D communications, devices are the 5G wireless communication systems. In particular, the METIS
operated independently, without the synchronization signal from summarized a number of deficiencies in current channel models,
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 37

in which the spatial inconsistency of current drop-based channel and capacity. A study of European cellular network operators con-
models requires novel modeling approach. cluded that a reduction in energy consumption by 35–60% could be
Some research groups also conducted research and measure- achieved by such sharing of infrastructure between network oper-
ments on the time-varying dual-mobility channel. The method of ators [174].
ray tracing is widely used to characterize the ray propagation in
a dynamic environment [165–168]. However, the granularity of ray 11.2. Harvesting renewable energy resources
tracing results will be affected greatly by the database of the spe-
cific environment and computation capability of workstations. Ad- Another approach to achieve green communication networks
ditionally, the variation of propagation paths and the transition be- is to harvest the renewable energy resources like solar, wind, vi-
tween LOS and NLOS propagation are often neglected. A 3-D time- brations at the BS and use them for its operation, reducing or
varying channel model is proposed for the 5G dual-mobility chan- even eliminating the use of conventional power consumption. A
nel which can address the problem of spatial inconsistency [169]. cognitive radio network that not just utilizes the spectral holes
for transmission, but also minimizes the energy consumption by
11. Green communications opportunistically harvesting energy from ambient sources is pre-
sented in [175]. In addition to harvesting natural sources of en-
The number of devices that would be connected to the network ergy like solar, wind, and vibrations, they also propose to har-
is expected to increase 100 times and the data volume is expected vest synthetic sources of energy like microwave power transfer.
to increase over 10 0 0 times in the next decade. While achieving They provide a comprehensive study of the solar energy that can
these benchmarks is in itself a challenge, we should also meet be harvested and whether a BS operating on solar power can be
these requirements in an affordable and sustainable way. Although made sustainable. They conclude that by storing the surplus en-
the contribution of mobile communications to the CO2 footprint is ergy received in afternoon and used in evenings, continuous self-
less than a percent now, we should try to reduce it further. Op- sustainability can be achieved during periods of abundant sunlight
erators are already facing that the power bills have become a sig- and even in winter, drawing power from the grid can be avoided
nificant part of the operating expenditure. So, lowering the energy for three to six hours a day. Another work [176] incorporated both
consumption and moving towards green communication alterna- solar and wind power to power the base station. They also propose
tives are not only important from an environmental perspective, to use fuel cell based energy sources for deployment in urban ar-
they are also significant from an economic perspective. eas where deploying solar and/or wind powered base stations will
not be feasible. Besides providing a reduction in energy consump-
tion, use of renewable energy resources will also enable setting up
11.1. Network planning and deployment self-sustainable BS in remote locations where power is not avail-
able, thereby improving the coverage.
Energy efficiency of the cellular network can be improved
by adopting several network deployment strategies [170]. These
strategies can be base station cooperation, different topologies of 11.3. User-centric design
cells, and distributed antenna systems. Significant improvements in
the energy efficiency can also be achieved in heterogeneous cellu- The traditional cell based coverage of a geographical area does
lar networks by using small cells. Currently, small cell base stations not provide the required elasticity to accommodate the diverse re-
are placed at locations to enhance the network capacity as well quirements of 5G cellular networks. To overcome this, it is pro-
as keeping the cost of infrastructure and deployment low. There- posed to get rid of BS centric design of cells and move to a user-
fore, by choosing the location of the microcells and relays opti- centric concept of “no more cells” [145]. They propose to retire the
mally within the range of a macro cell, they can significantly of- cell based design to a user-centric design with amorphous cells,
fload the macrocell and produce energy savings while providing a decoupled signaling and data and decoupled downlink and uplink.
better coverage [171]. These optimizations are especially efficient This enables the small cells within the macrocell to be turned off
in areas where extremely high capacity and data rates are needed when they have no traffic. This cannot be done in the traditional
like offices, shopping malls, subway stations, etc., where the user cell based network. The decoupling of uplink and downlink will
density is large. Since most of these places are indoor, indoor ac- enable more efficient resource allocation. This enables a user to
cess points should be deployed so that energy wastage due to wall send uplink data through one cell and get downlink data from an-
penetration can be avoided. other cell when the cells are heavily loaded in downlink and uplink
Another complication in cellular networks is the high variabil- respectively.
ity of network traffic with time due to the patterns in which ev- With the use of SDN that provides separation of control and
eryone tends to access the network at the same time. This causes data plane, this kind of architecture becomes easier to implement.
huge difference between the average and peak hour cellular traf- This could be exploited to provide each user with a single radio re-
fic. Reports indicate that this difference between peak and average source control connection with macro BS and dual data connection
traffic is increasing and the peak rate of traffic is expected to grow with both macro BS and micro BS [177]. The collaboration between
much faster than the rate of growth of average traffic [13]. This macro and micro BS is utilized to minimize control channel over-
makes network operators deploy more base stations to support the head and cell-specific reference signals in order to achieve a pure
peak hour traffic. This causes unnecessary power consumption and data carrier for small cells. This architecture was shown to provide
waste when it is not needed. This could be reduced by systemati- 90% energy efficiency gain while still achieving a throughput im-
cally switching off some of the base stations that are not required provement of more than 17%.
to be operating. Based on the traffic pattern, analytical models can
be developed to identify optimal BS switch off times [172]. It was 11.4. Smaller frame overhead
also observed that the variation of traffic demands among differ-
ent network operators serving the same geographical area is sig- It is also noted in [145] that the traditional cellular networks
nificantly different [173]. Hence, the network infrastructure of sev- are designed for conventional streaming applications. The diverse
eral network operators could be shared among them to dramati- traffic requirements have exposed the inefficiencies of conventional
cally reduce energy consumption while providing better coverage cellular networks and brought in new challenges. There are several
38 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

applications that use small sized persistent bursty traffic like in- 11.6. Open problems
stant messaging services. Other types of applications like MTC and
D2D communications also send small bursts of data [10]. These ap- In the above discussions of green communications, there are
plications generate smaller packets that are sent at regular inter- some tradeoffs and challenges also needed to be resolved. In this
vals. They cause the mobile device to switch between idle and con- subsection, we present some open problems in green communica-
nected states and it may consume much power. The size of these tions and give potential solutions to them.
packets is also small so that this cycle happens very frequently. Power control in green communications: Although there are
Another problem with this type of traffic is that these packets are several green communication solutions proposed for 5G cellular
very small and hence the signaling overhead in header data might networks from the point of energy efficiency, it should be verified
be significant compared to the actual data size. To combat this sce- that achieving energy efficiency does not degrade the performance
nario, a lightweight radio resource connection state without main- of networks in terms of data rate and other requirements. For ex-
tenance overhead for handover and channel status feedback has ample, in the case of multi-tiered network deployments, the user is
been introduced in 3GPP Release 11. There are also many modi- not attached to the BS with maximum power. This causes the BS to
fications proposed to the random access procedures to handle this experience higher interference from other BSs which would affect
type of traffic like implementing predetermined dedicated pream- the received signal quality and hence the data rate that the user
bles or sending the data in the uplink resource allocated for radio could achieve. The feasibility of implementation of some of solu-
resource control requests. Some contention based methods have tions like harvesting solar and wind energy in dense localities and
also been proposed [178] where the devices directly send packets places where sunlight is not available throughout the year should
in a contention-based manner. be studied further.
Energy efficient hardware: Previous and current transceiver
equipment and hardware in wireless communication networks are
designed to achieve good performance in data throughput and
reliability. However, such hardware is normally energy costly. In
11.5. Green metrics 5G systems, operators and equipment manufacturers should weigh
heavier on the energy efficiency when designing and testing net-
In order to evaluate the energy efficiency of wireless commu- work equipment. The research in [182] examined the energy con-
nication systems, a group of green metrics is necessary to be es- sumption in both office equipment and portable devices such as
tablished for the 5G networks. The green metrics are useful in re- laptops and mobile phones. Based on the statistics of global en-
search and development in energy efficient components, standard- ergy consumption, the recommendations are made in perspectives
ization of energy efficient equipment manufacturing, and quanti- of power management, battery life management, and utilization on
fied assessment on system performance. Therefore, based on the energy harvesting to alleviate the issue. A globally adoptable en-
structure of a wireless communication network, the green metrics ergy saving recommendation can be an alternative to encourage
can be measured on three levels, namely, component, equipment, energy saving activities around the world.
and system levels [179]. Energy efficient network architecture: The energy saving net-
According to current signal processing architecture in wireless work architecture will be enabled by multiple technologies, which
communication networks, the electronic components normally are are also discussed in this paper, namely, massive MIMO, ultra-
filters, power amplifiers, A/D converters, and antennas at the RF densification, SDN and NFV, D2D communications, and mobile
side. So on the component level, the green metrics can be charac- cloud computing. The technology of massive MIMO can reach en-
terized as the gain of the RF component, radiated efficiency of an- ergy efficiency by the utilization of hundreds of antenna elements
tennas, or power efficiency in power amplifiers. The measurements for high gains. By separating the control and data planes in BS,
for the green metrics in the above components are straightforward, idle UEs will not waste energy on keeping connection with BS. The
and are normally showed in the component’s specification lists. wireless software-defined network architecture will save configu-
As to the level of equipment, the performance of green met- ration time and energy in traditional hardware. The approach of
rics cannot be easily measured and should be evaluated in differ- enabling direct device connection can further improve energy ef-
ent environments. Although the equipment consists of numerous ficiency in the entire network. Furthermore, a series of reconfig-
electronic components, the green metrics of the equipment are not urable and energy-scalable radio network solutions were surveyed
simple linear additions of those parameters in each component. In in [181], which can also be of important reference to the 5G sys-
the standard operation mode, the equipment should consume less tem design.
amount of energy than in busy operation mode, while in idle op- Battery technology enhancements: The novel battery tech-
eration mode the energy consumption should be the least. In Eu- nologies will also bring revolutionary changes on the next gener-
rope, the standardized metric in this case is the Energy Consump- ation of energy efficient communication networks. Recent electro-
tion Rating (ECR) which can be expressed as the actual energy con- chemical research on novel energy sources has found that sugar
sumption divided by the effective system throughput, as defined by can actually be an excellent energy provider, which offers an order
the ETSI [180]. of magnitude more energy than the same weight lithium-ion bat-
In current wireless networks, green metrics are evaluated in tery in smartphones [183]. The theory behind the sugar battery is
cellular networks, wireless local area networks, satellite systems, based on using enzymes to extract the energy from sugar, which
and ad hoc networks [179]. However, since in the 5G visions, ultra- is similar to human digesting sugar to absorb energy. This proto-
dense networks will be deployed to provide more layers of net- type on sugar battery is promising in the utilization on sensors
works with wider coverage, and new frequency bands will be li- and other small devices which can expand their lifespan without
censed for cellular networks, there should be another novel set adding extra weight. Additionally, unlike the lithium-ion battery
of green metrics for 5G wireless networks. The 5G green metrics which is a limited source and requires professional recycling pro-
should be defined by network layers. Two examples are the signal cedures, the sugar battery is easy to refill and safe to use.
processing energy efficiency in the physical layer and the modula- In addition, some plants have been discovered to be equipped
tion and user association energy consumption in the medium ac- with the capabilities of a solar panel. A group of researchers in
cess control layers. More research efforts should be encouraged to the University of Cambridge found the Photo Microbial Fuel Cells
expedite the standardization of 5G green metrics [181]. (Photo-MFCs), which is a type of moss can transfer solar power
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 39

into electric power [184]. Although the Photo-MFCs are still in the self-interference arising due to the nonlinearities of the analog-to-
early stage of research and cannot harness energy with accept- digital converter and oscillator phase noise. Some works proposed
able efficiency, the trend for the 5G green communications is clear employ a combination of these techniques to achieve superior self-
ahead. interference cancellation.
The full duplex communication can be used in a couple of ways
12. Radio access techniques to increase the capacity of the 5G cellular networks [187]. In the
first scenario, the user equipment sends its data to a BS or poten-
The different requirements explained in Section 2 urge us to re- tially another device in D2D communication while the BS sends its
think about the radio access technique design. The gigabit speed own data to the same UE at the same time and the UE and BS can
demand of 5G networks motivates that the underlying radio access decode the data intended for them while simultaneously transmit-
technique should also be capable of supporting higher data rates. ting its data in the same frequency band. This mode of operation
Since spectrum is a scarce and costly resource, spectral efficiency is called bidirectional transmission. The other scenario is the uni-
is a key factor of the radio access technique that would enable the directional transmission similar to the full duplex relaying [188].
gigabit speeds. Several applications that are envisioned in 5G net- Here, the first node transmits its data to the second node which
work, like tactile Internet, require a very low latency of the order in turn transmits its own data to the third node. Although the full
of 1 ms. This puts a constraint on the perspective of latency of the duplex communications seem to double the spectral efficiency that
radio access technique, so that the lower latency required at higher can be realized, the practical throughput gains that are achieved
layers can be attained. The other applications like IoT have scenar- is about 30–40% [186]. This clearly shows that there is significant
ios where the devices are not connected to the BS at all times. scope for improvement of full duplex technologies.
The power constraints of these devices prevent the devices from Although self-interference cancellation is the primary problem
having full synchronization with the BS. Hence, the multiple ac- in realizing a full duplex communication, it is not the only is-
cess technique to be adopted should be able to support loose or sue that needs to be addressed. The whole of the current network
more preferably no synchronization. As in other wireless systems, stack, was designed with half-duplex communication in mind. This
interference is a concern while adopting a multiple access tech- requires us to rethink and redesign these functionalities for full du-
nique. The technique we adopt should have low out-of-band (OOB) plex communication system. For example, water filling algorithm
emissions so that the guard band between other carriers can be can be used for power allocation to maximize the spectral effi-
minimal. Having a lower interference will require less guard band- ciency of half duplex system but, it is not the optimal scheme
width and ultimately increase the effective spectral efficiency. The for full duplex system as it does not take into account the self-
multiple access technique should also efficiently use the multiple interference. The MAC protocols used in earlier systems like Carrier
antennas that will be present in most of the devices. Hence com- Sense Multiple Access (CSMA) inherently assumes a half duplex
patibility with MIMO technology should also be considered while system and it cannot be directly employed in full duplex system.
choosing a multiple access technique. Finally, the radio access tech- This requires development of novel MAC protocols like the proto-
nique must also be power efficient, so that it does not drain out col proposed in [187]. Similar redesign of higher layer protocols for
the power of low power devices. application in full duplex systems also needs to be explored.
The major innovations in the radio access techniques can be
classified into the introduction of full duplex communications, the 12.2. Multiple access techniques
choice of multiple access technique, modulation, and channel cod-
ing. These innovative improvements on radio access techniques are E-UTRA employs Orthogonal Frequency Division Multiple Access
explained here. (OFDMA) for the downlink channel and Single Carrier - Frequency
Division Multiple Access (SC-FDMA) for the uplink channel to over-
12.1. Full duplex communication come the high peak-to-average power ratio and thereby conserver
power [185]. Although these techniques work well for 4G LTE net-
Full duplex communication systems enable the simultaneous works, they cause inefficiencies in several scenarios foreseen for
transmission and reception of signals in the same frequency bands. 5G networks. This requires us to redesign the multiple-access tech-
This is a significant departure from the traditional half duplex nique for 5G networks. The several multiple access techniques that
communications that use Time Division Duplexing (TDD) or Fre- are currently being considered for 5G systems are explained below
quency Division Duplexing (FDD) in order to achieve duplex com- and their advantages and disadvantages are discussed.
munication. The RAN technology of 4G LTE network, Evolved UMTS
Terrestrial Radio Access (E-UTRA) supports both TDD and FDD 12.2.1. Orthogonal Frequency Division Multiplexing
modes of operation [185]. TDD and FDD modes suffer from per- Orthogonal Frequency Division Multiplexing (OFDM) and Or-
formance and efficiency issues like out-of-date channel state infor- thogonal Frequency Division Multiple Access (OFDMA) have been
mation, inflexible bandwidth allocation, and requirements of guard the popular multi-carrier modulation and multiple access tech-
intervals and frequency bands [186]. Although full duplex systems niques until the 4G LTE [189]. Several advantages of OFDM like
overcome these inefficiencies, it was considered infeasible to im- ease of implementation, robustness against frequency selective fad-
plement a practical full duplex system due to the very high self- ing channels, compatibility with MIMO, simple structure and ef-
interference from the transmitter blinding the intended received ficient implementation using FFT enabled the widespread use of
signal. Recent developments in self-interference cancellation tech- OFDM in several applications like IEEE 802.11 wireless LAN and
niques has enabled the practical realization of full duplex systems. broadband over power line [190].
The self-interference cancellation techniques developed propose Despite all these advantages, OFDM suffers from a lot of disad-
to apply interference suppression at various stages of the re- vantages. Although OFDM is well suited for traditional communica-
ceiver chain [187]. Propagation domain interference suppression tion, it does not support several new scenarios foreseen for 5G net-
tries to avoid the input of the RF amplifier being overwhelmed works. The orthogonality between subcarriers require perfect syn-
by self-interference, analog domain interference cancellation pre- chronization which is difficult to achieve in scenarios like uplink
vents the analog-to-digital converter from being overwhelmed by [191] and in downlink when base stations cooperate to send data
self-interference by applying interference cancellation while digi- [192]. Due to the requirement of cyclic prefix, low latency can-
tal domain interference cancellation attempts to cancel the residue not be achieved in OFDM. OFDM suffers from other well known
40 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

disadvantages like high peak to average power ratio [193], high This makes UFDM more suitable for CoMP transmission where syn-
OOB power emission [194] and pilot contamination problem [195]. chronization between transmitters is hard to achieve. UFMC also
The vast literature available on OFDM has tried to overcome these supports MIMO systems.
shortcomings. For example, [196] proposes an algorithm to reduce In spite of these advantages, UFDM equalization process is very
the sidelobes, [195] tries to reduce the pilot contamination prob- complex. This will be prohibitive for devices with limited compu-
lem. All of them try to overcome only one disadvantage of OFDM tational abilities and are highly power constrained. UFMC is also
and there is no solution that solves all the drawbacks mentioned sensitive to small timing misalignment [210] which makes them
above. unsuitable for applications like MTC and IoT that require loose or
no synchronization. In spite of having several improvements over
12.2.2. Filter Bank Multi-Carrier OFDM and FBMC, UFMC does not satisfy all the requirements to
Filter Bank Multi-Carrier (FBMC) transmits the set of data sym- become the choice for 5G cellular networks.
bols through a bank of modulated filters [197] and it could be
seen as a generalization of OFDM, where OFDM filters the com- 12.2.4. Generalized Frequency Division Multiplexing
plete band while FBMC filters each subcarrier individually [198]. Generalized Frequency Division Multiplexing (GFDM) was ini-
FBMC replaces the sinc-pulse of subcarriers in OFDM with a dif- tially proposed in [211] as a waveform for scenarios where there
ferent shape. This freedom in choosing the filter shape could be is a high degree of spectrum fragmentation like exploiting spectral
exploited and to develop a suitable filter design that satisfies the holes in TV bands. Due to its favorable properties, it is one of the
desired spectral characteristics could be developed. There are two top contenders for the waveform for 5G cellular networks [212].
broad classes of FBMC – staggered multi-tone (SMT) and filtered GFDM is a block based multiplexing technique where each in-
multi-tone (FMT), in which SMT is more popular [199]. dependently modulated block consists of several subcarriers and
FBMC has the advantages of having a very good frequency local- subsymbols. These subcarriers are filtered by the prototype filter
ization achieved by using different prototype filters. It also gets rid along with the filter’s time and frequency domain shifted versions
of the use of cyclic prefix thus increases the spectral efficiency of [212]. This can be viewed as a generalization of OFDM and sin-
the wireless system. There are efficient implementations of FBMC gle carrier frequency domain equalization where OFDM has several
using FFT and IFFT blocks and polyphase filter structures [197,200]. subcarriers and only one symbol is transmitted per subcarrier in
Compared to OFDM, FBMC is robust to frequency offset [201] and a block and single carrier frequency domain equalization has only
does not have strict symbol synchronization [202]. one subcarrier in the whole frequency band and several symbols
FBMC requires the use offset-QAM modulation and does not are transmitted serially. The number of subcarriers and number
support other types of modulation. Similar to OFDM, FBMC also of subsymbols in each subcarrier, along with the prototype filter
suffers from a high peak-to-average power ratio [203]. High spec- structure can be configured to meet the requirements of different
tral efficiency offered by FBMC can be realized only when the applications.
number of symbols transmitted is very high [198]. In practical sce- In GFDM, cyclic prefix is needed to be added only once per
narios where the transmission time interval is of the order of mil- block. Since the block has several subsymbols transmitted serially,
liseconds, the filtering introduces ramp-up and ramp-down times GFDM has higher spectral efficiency compared to OFDM [212]. The
at the end of these intervals and thereby reduces the spectral ef- requirement of only one cyclic prefix per block could also be ex-
ficiency [204]. This will be a major issue in applications like MTC ploited by using a long cyclic prefix which could enable applica-
and IoT where the packet sizes are short. Further, some versions of tions that require loose or no synchronization [213]. The use of
FBMC support only pulse amplitude modulation [197]. This restric- cyclic prefix could also be avoided by multiplying the GFDM block
tion would prohibit the use of higher order modulation techniques with a window function with has smooth ramp-up and ramp-
like 64-QAM which is being used in LTE E-UTRA [185], thereby down at the edge of blocks [212]. GFDM is also compatible with
limiting the data rates supported. FBMC systems are more sensi- MIMO systems and can be combined efficiently with Alamouti
tive to errors in channel measurements and FBMC modulation is space-time codes. Due to the similarity of GFDM waveform to that
less efficient than OFDM systems when there are errors in chan- of OFDM, most of the vast literature developed for synchronization
nel state information (CSI) [205]. These imperfections also cause and timing estimation like [214] can be readily adapted for use in
inter-symbol interference (ISI) and inter-carrier interference (ICI) GFDM systems.
[206] and makes FBMC less compatible with MIMO systems. The The main disadvantage of GFDM system is the computational
advanced wireless communication techniques like spatial multi- complexity. Since GFDM cannot use the FFT/IFFT blocks for mod-
plexing with maximum likelihood detection and Alamouti Space ulation and demodulation like OFDM and FBMC systems, GFDM
Time Block Codes cannot be combined efficiently with FBMC [207]. systems have a huge computational complexity of encoding and
decoding. This might be a forbidding computational requirement
12.2.3. Universal Filtered Multi-Carrier at several IoT devices. GFDM is also more sensitive to carrier fre-
Universal Filtered Multi-Carrier (UFMC) is a generalization of quency offset compared to OFDM [215]. Due to the inherent non-
OFDM and FBMC. While OFDM filters the entire band and FBMC orthogonality, GFDM waveform has higher bit error rate compared
filters each subcarrier individually, UFMC uses filters on a group of to OFDM waveforms [216]. Although, there are interference can-
subcarriers to reduce the out-of-band emissions [208]. UFMC splits cellation schemes proposed [217], the use of successive interfere
the data stream into several sub-streams and a pulse shaping filter cancellation technique will further make it computationally more
is applied to each of those streams [209]. This reduces the interfer- expensive.
ence from adjacent sub-bands. This aggregation of frequency sub-
bands also increases the bandwidth and thus makes subsymbols 12.2.5. Non-Orthogonal Multiple Access
shorter in time, making them more suitable for applications that Distinct from the other multiple access techniques discussed
require short packet transmissions [198]. In addition, UFMC also before, Non-Orthogonal Multiple Access (NOMA) does not provide
does not need a cyclic prefix thus increases the spectral efficiency. an alternative waveform design. Instead, it provides a new dimen-
The filter ramp-up and ramp-down time periods at the start and sion of multiplexing, power domain, which could be used in con-
end of the frame give inherent protection against ISI. UFMC is also junction with other multiple access techniques [218]. NOMA ex-
shown to outperform OFDM under both perfect and imperfect CSI ploits the difference in channel gain between the users and uses it
conditions and in the presence of carrier frequency offset (CFO). to multiplex the data of different users [219].
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 41

Table 4
Comparison of multiple access techniques.

Multiple access technique Advantages Disadvantages

OFDM • Computationally simple • Requires perfect synchronization

• Simple modulation and demodulation • Requires cyclic prefix
• High compatibility with MIMO • High OOB emission
• Well studied • High PAPR
• Pilot contamination
FBMC • Frequency localization • High PAPR
• Efficient implementation • Supports only offset-QAM modulation
• Robust to center frequency offset • Spectrum efficiency achieved only when length of data is large
• Supports loose synchronization • Sensitive to errors in CSI
• Not compatible with MIMO
UFMC • Frequency localization • Complex equalization
• Shorter symbols • Sensitive to timing misalignment
• Soft protection from ISI
• Robust to CSI errors
• MIMO compatibility
GFDM • Spectral efficiency • Computational complexity
• MIMO compatibility • Sensitive to CFO
• ISI protection • High BER
• Loose synchronization
NOMA • New domain of multiplexing • Computational complexity
• Compatible with other multiple access techniques • Requires perfect synchronization
• Improves spectral efficiency

NOMA will be well suited in practical wide area deployments packets without being synchronized with the base station while
where there are several users spread across the coverage area with the mobile phone may be streaming a HD video and it will re-
some users with higher channel gain and some others with poor quire very high data rates and it will be ready to synchronize with
channel gain. This will help pairing the users efficiently so that the the base station perfectly so that it can attain maximum data rate.
NOMA schema could be utilized effectively. To make this pairing, These two scenarios cannot be supported by these multiple ac-
we need only the rough channel state estimate at the transmit- cess techniques simultaneously. This is because all the minimum
ter and not the finer CSI that would be needed in the receiver for resource unit that could be allocated to a user has the same shape
decoding the data. This data could be used to decide whether or in the time-frequency lattice representation. Although LTE E-UTRA
not to use NOMA and if we use, which users should be paired supports this to some level by providing flexible spectrum alloca-
together [220]. tion and supporting variable bandwidths from 1.4 MHz to 20 MHz
NOMA inherently uses successive interference cancellation (SIC) [185]. However, this level of configurability is also very limited and
for decoding and it comes with several drawbacks. The first one we need a multiple access technique where the size and shape of
being the complexity. A user has to decode the other user’s data the time-frequency block allocated to each user can be configured
first and then decode its own data. This doubles the computational individually so that each device attached to the base station can
complexity required at the receiver and this could be prohibitive satisfy their own requirements, without being constrained by the
for low power IoT devices. SIC also comes with the error propaga- other devices in the network.
tion in decoding where the error in decoding the first user’s data A first step in this direction could be the Software Defined Mul-
will cause error in decoding second user’s data as well [221]. SIC tiple Access (SoDeMA) as envisioned in [222]. SoDeMA is said to
will also require the two symbols transmitted at the same time- have the flexibility where all the NOMA and orthogonal multiple
frequency block to be perfectly synchronized and this will be an access schemes coexist in a system and different types of devices
issue in the uplink scenario. in the network can use different part of the spectrum using the
multiple access technique that would be the best fit for them. For
12.3. Open problems example, the IoT device can use GFDM with a long cyclic prefix
while the mobile device streaming video can use some orthogo-
A summary of the advantages and disadvantages of multiple nal multiple access scheme that supports high data rates and the
access techniques discussed in this section are listed in Table 4. schemes to use can be configured adaptively based on the user re-
The several multiple access techniques proposed here provide an quirement and the load on the base station.
improvement over OFDM and other conventional multiple access Computational complexity: The multiple access technique to
techniques. Most of them support the several scenarios that are be adopted should also be computationally simple so that devices
new and are envisioned to be supported by 5G cellular networks. with low computational power can implement them easily. With
But, there is no clear winner that could be chosen conclusively. the 50 billion IoT devices that are expected to be connected to 5G
Although full duplex communication systems look set to be de- network, having a complex decoding requirement will affect the
ployed, there are several aspects of it which need further stud and lifetime of these devices. Thus, low complexity is a necessary re-
research. The open problems in the development of radio access quirement for any candidate multiple access technique.
techniques for 5G network are discussed below. Robustness: The vast majority of the devices in a 5G cellular
Simultaneously accommodating all devices: Although multi- network will be low cost IoT devices. Due to their low cost, they
ple access techniques like UFMC and GFDM have the advantage of cannot afford to have high precision clocks and oscillators. This er-
being adaptable to different scenarios by varying the parameters ror in oscillator may lead to CFO and timing misalignment. Fur-
involved, they do not support those scenarios simultaneously. For ther, going through a synchronization step every time the device is
example, an IoT device and a mobile phone may be in the same connected to the network will create overhead in power consump-
cell range simultaneously. The IoT device will want to send short tion. This makes it necessary for the multiple access technique to
42 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

be robust against the imperfections in timing and frequency mis- support short range for extremely high throughput at millimeter
alignments. wave [226].
Replacing OFDM: Despite of all the problems discussed above,
OFDM remains one of the simplest, efficient and widely deployed
13.2. Europe
multiple access technique. Although there are several alternatives
proposed for replacing OFDM, the industry and the standard bod-
In Europe, the 3rd Generation Partnership Project (3GPP), which
ies would be reluctant for a radical change in the multiple access
is the leading collaborating force in the standardization process of
technique. The multiple access technique adopted for 5G would be
LTE and LTE-A, is focusing on enhancing the current standards to-
an extension or enhancement over the OFDM, at least for the first
wards the 5G era. The 3GPP Release 14 and 15, which are being
releases.
developed and expected to be finalized by 2020, are promising to
Harmonization of radio access techniques: As explained in
provide unprecedented system capacity which is one of the key
[223], the 5G radio access technique will be a protocol harmoniza-
requirements of 5G systems. Release 15 is also expected to be sub-
tion of several multiple access and modulation techniques across
mitted to ITU as the first 5G standard in Europe.
different bands, services and cell types. The important issue in use
Additionally, the 5G Infrastructure Public Private Partnership
of such a technique is the co-existence of several of these tech-
(5GPPP) is another European initiative on collaborations among
niques. Although the user equipment and the radio access tech-
network operators, infrastructure providers, and researchers. The
niques should be as similar as possible across the frequency bands
projects under the collaboration of 5GPPP cover a wide range of 5G
used, some of the techniques like narrowband beamforming for
networks. The Mobile and wireless communications Enablers for
millimeter wave communications are tailored to be used in some
the Twenty-twenty Information Society 5G, also known as METIS
of these bands. This makes the techniques disparate and makes
2020 Project, is a leader in the research for 5G networks [11].
harmonization difficult to achieve. In addition to the harmoniza-
METIS initiated its discussions in late 2012 and covers a wide range
tion of waveform design, the harmonization is also required at
of new concepts and horizontal topics. They proposed a new se-
higher layers. For example, the MAC protocol deployed should be
ries of channel models based on extensive measurements for spe-
independent of the frequency bands or waveforms used at the
cial scenarios in shopping malls, D2D, and V2V communications. In
physical layer, providing the abstraction needed.
particular, the spatial consistency for D2D communications needs
to be accounted for in the channel modeling for 5G, which is a de-
13. Global 5G activities ficiency of current channel models [92,227–229]. The second phase
of the METIS project (METIS-II) was initiated in July 2015 and has a
With the pursuit of unprecedented high data rates and capacity length of 24 months. The METIS-II aims to foster the standardiza-
in wireless communication networks, the research activities for 5G tion of the 5G RAN architecture and functionality design [230]. The
are going on around the world at research institutions, standard 5G millimeter-wave RAN is the focus of the “mmMAGIC” project,
bodies, service operators, and telecommunication companies, with which covers the wide frequency range of 6 ∼ 100 GHz for radio
the focus ranging from new channel models, signal multiplexing access technology design [231].
techniques, large scale antenna arrays, to the core network design. As to the aspect of waveform design which is another fun-
We hereby provide an overview of the global research activities. damental issue in 5G system, a collaborative project in Europe
named as 5th Generation Non-Orthogonal Waveforms for Asyn-
chronous Signaling, in acronym of “5GNOW”, aims at boosting ca-
13.1. The United States
pacity and coverage via novel waveform design and scalable air
interface framework [232]. Moreover, as energy efficiency has be-
In the United States, great research efforts in both academia and
come a global concern when more sophisticated computation and
industry have been dedicated to the innovations in 5G. In order
frequent signaling have caused excessive energy consumption, the
to establish a flexible and self-adaptable network architecture that
“5GrEEn” project dedicates to provide a sustainable solution for en-
can maximize capacity and spectrum efficiency, the development
vironmentally friendly development in the 5G network [170]. An-
of SoftAir, a novel design based on WSDN, has been initiated in
other project issued by 5GPPP is the 5G novel radio multi-service
2015 in the Broadband Wireless Networking (BWN) Laboratory in
adaptive network architecture (5G-NORMA), which aims to develop
Georgia Institute of Technology. The great potential of SoftAir can
an adaptive and efficient 5G mobile network architecture with the
enable a transformation from hardware-based network toward a
capability of dynamic network resource sharing among operators
low-cost, flexible, and programmable software-based network [20].
[233].
At New York University, University of Southern California, and
In order to design the 5G backhaul and frontaul networks,
University of Texas at Austin, groups of researchers are exploring
5GPPP also initiated the “5G-Xhaul” and “5G-Crosshaul” projects,
the great spectrum resources at millimeter wave frequency, which
which are targeted to develop a dynamically reconfigurable back-
will become an indispensable supplement to address today’s con-
haul/fronthaul network with cognitive control plane for small cells
cern on spectrum crunch [58]. Since the deployment of antenna
and C-RANs. Other efforts within 5GPPP include “CogNet” for intel-
arrays has proven to boost capacity for current LTE system, re-
ligent network management, “COHERENT” for coordinated control
searchers in University of Texas at Austin are also proposing the
and spectrum management for 5G HetNets, etc [234].
development of massive MIMO, which is large scale antenna ar-
rays that can serve multiple users at the same time with high data
rates [224]. 13.3. Asia
The AT&T research groups put their major focus on the key
technologies of SDN and NFV to improve cost efficiency and en- In Asia, South Korea, Japan, and China are the leading forces in
hance network performance. They also partnered with Ericsson 5G research activities. Samsung in South Korea announced its suc-
and Intel to initiate 5G trials to implement new network architec- cessfully tested millimeter wave technique for 5G in 2014, with a
ture and prepare for commercial deployment [225]. peak rate of 7.5 Gbps at the carrier frequency of 28 GHz, which is
Another industrial research effort in the US, Qualcomm, has vi- over 30 times faster than the current 4G LTE network [235]. Ko-
sioned that 5G will continue to enhance mobile broadband services rea Telecom aims to provide millimeter wave-based 5G network
at frequency bands currently being utilized (i.e. below 6 GHz), and services for 2018 Winter Olympic Games. In Japan, NTT DoCoMo
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 43

dedicates the 5G research on small cells, which offers higher spec- [10] H. Shariatmadari, R. Ratasuk, S. Iraji, A. Laya, T. Taleb, R. Jantti, A. Ghosh,
tral density for a smaller cell coverage area [25]. They target the Machine-type communications: current status and future perspectives to-
ward 5G systems, IEEE Commun. Mag. 53 (9) (2015) 10–17, doi:10.1109/
2020 Summer Olympic Games to launch their 5G commercial ser- MCOM.2015.7263367.
vices. The four major telecom operators, Korea Telecom, NTT Do- [11] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M. Maternia, O. Que-
CoMo, SK Telecom, and Verizon, are joining forces on the 5G Trial seth, M. Schellmann, H. Schotten, H. Taoka, H. Tullberg, M. Uusitalo, B. Timus,
M. Fallgren, Scenarios for 5G mobile and wireless communications: the vision
Specification Alliance, with the focus on developing a common and of the METIS project, IEEE Commun. Mag. 52 (5) (2014) 26–35, doi:10.1109/
extendable platform for 5G networks. The Chinese government ini- MCOM.2014.6815890.
tiated the IMT-2020 Promotion Group as early as 2013. The re- [12] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, M. Ayyash, Internet of
things: a survey on enabling technologies, protocols, and applications, IEEE
search groups in China are advocating massive MIMO for next gen-
Commun. Surveys Tuts. 17 (4) (2015) 2347–2376, doi:10.1109/COMST.2015.
eration of wireless communications. The Academy of China Mobile 2444095.
is developing a prototype of 128 antenna array elements for perfor- [13] Cisco visual networking index: forecast and methodology, 2014–2019, CISCO
White Paper (2015).
mance analysis [236]. Huawei’s recent demonstration on 5G end-
[14] R. Di Taranto, S. Muppirisetty, R. Raulefs, D. Slock, T. Svensson, H. Wymeersch,
to-end network slicing implementation reveals the efficiency and Location-aware communications for 5G networks: how location information
flexibility that network slicing can achieve to enable multiple net- can improve scalability, latency, and robustness of 5G, IEEE Signal Process.
work applications on a unified physical network architecture [237]. Mag. 31 (6) (2014) 102–112, doi:10.1109/MSP.2014.2332611.
[15] K.-T. Feng, C.-H. Hsu, T.-E. Lu, Velocity-assisted predictive mobility and
location-aware routing protocols for mobile Ad Hoc networks, IEEE Trans.
14. Conclusion Veh. Technol. 57 (1) (2008) 448–464, doi:10.1109/TVT.2007.901897.
[16] X. Duan, X. Wang, Authentication handover and privacy protection in 5G het-
nets using software-defined networking, IEEE Commun. Mag. 53 (4) (2015)
This paper presents an in-depth overview of the ten key en- 28–35, doi:10.1109/MCOM.2015.7081072.
abling technologies which will be landmarks in 5G systems. Specif- [17] A. Ruiz-Martinez, Towards a web payment framework: State-of-the-art and
ically, the requirements of the next generation cellular network challenges, Electron. Commerce Res. Appl. 14 (5) (2015) 345–350. Contem-
porary Research on Payments and Cards in the Global Fintech Revolution.
are first explained and how these ten technologies would aid the http://dx.doi.org/10.1016/j.elerap.2015.08.003.
realization of these requirements is examined. Furthermore, the [18] I.F. Akyildiz, A. Lee, P. Wang, M. Luo, W. Chou, A roadmap for traffic engineer-
strengths and limitations faced by these technologies are investi- ing in sdn-openflow networks, Comput. Netw. J. 71 (2014) 1–30.
[19] S. Jain, A. Kumar, S. Mandal, et al., B4: experience with a globally-deployed
gated in detail, in which the open problems that need to be ex- software defined WAN, in: Proceedings of the ACM SIGCOMM Conference,
plored and resolved before the commercialization of these solu- SIGCOMM’13, August 2013, pp. 3–14.
tions are fully summarized. Finally, the current research activities [20] I.F. Akyildiz, P. Wang, S.C. Lin, SoftAir: a software defined networking archi-
tecture for 5G wireless systems, Comput. Netw. 85 (C) (2015) 1–18.
being undertaken around the world are discussed. Although some [21] X. Jin, L.E. Li, L. Vanbever, J. Rexford, SoftCell: scalable and flexible cellu-
network operators start to promote their “5G services” which are lar core network architecture, in: Proceedings of the Ninth ACM Confer-
actually based on LTE-Advanced technologies, we predict that the ence on Emerging Networking Experiments and Technologies, ACM, 2013,
pp. 163–174.
5G cellular standardization will be completed by the early years
[22] J. Wu, Z. Zhang, Y. Hong, Y. Wen, Cloud radio access network (C-RAN): a
of the next decade and commercial services would be launched by primer, IEEE Netw. 29 (1) (2015) 35–41, doi:10.1109/MNET.2015.7018201.
mid-late 2020s. Although drastic paradigm shifts have never been [23] ARGELA (2016) http://www.argela.com.tr/progran/.
[24] SK Telecom, SK Telecom 5G White Paper: SK Telecom’s View on 5G Vision,
easy and require massive upgrading of network infrastructures, the
Architecture, Technology, and Spectrum, Technical Report, 2014.
need for better services, which motivated such transitions in the [25] N.T.T. DOCOMO, DOCOMO 5G White Paper: 5G Radio Access: Requirements,
past, will continue to occur in the future. Concept and Technologies, Technical Report, 2014.
[26] Project CONTENT FP, 2012--2015. http://cordis.europa.eu/fp7/ict/
future-networks/.
Acknowledgment [27] K.-K. Yap, R. Sherwood, M. Kobayashi, T.-Y. Huang, M. Chan, N. Handigol,
N. McKeown, G. Parulkar, Blueprint for introducing innovation into wireless
The authors would like to thank David M. Gutierrez-Estevez, mobile networks, in: Proceedings of the Second ACM SIGCOMM Workshop on
Virtualized Infrastructure Systems and Architectures, ACM, 2010, pp. 25–32.
Ravikumar Balakrishnan, Elias Chavarria Reyes, Won Yeol Lee, [28] M. Bansal, J. Mehlman, S. Katti, P. Levis, OpenRadio: a programmable wireless
Ozgur Baris Akan, Chong Han, and Wolfgang Gerstacker for their dataplane, in: Proceedings of the First Workshop on Hot Topics in Software
valuable comments and suggestions to improve the quality of the Defined Networks, ACM, 2012, pp. 109–114.
[29] P. Dely, J. Vestin, A. Kassler, N. Bayer, H. Einsiedler, C. Peylo, CloudMAC - an
paper. OpenFlow based architecture for 802.11 MAC layer processing in the cloud,
in: 2012 IEEE Globecom Workshops (GC Wkshps), 2012, pp. 186–191, doi:10.
References 1109/GLOCOMW.2012.6477567.
[30] L. Suresh, J. Schulz-Zander, R. Merz, A. Feldmann, T. Vazao, Towards pro-
[1] I.F. Akyildiz, D.M. Gutierrez-Estevez, R. Balakrishnan, E. Chavarria-Reyes, LTE- grammable enterprise WLANS with Odin, in: Proceedings of the First Work-
Advanced and the evolution to Beyond 4G (B4G) systems, Phys. Commun. 10 shop on Hot Topics in Software Defined Networks, ACM, 2012, pp. 115–120.
(2014) 31–60. http://dx.doi.org/10.1016/j.phycom.2013.11.009. [31] Centre Tecnologic Telecomunicacions Catalunya (CTTC), SDN/NFV Cloud Com-
[2] I.F. Akyildiz, M. Pierobon, S. Balasubramaniam, Y. Koucheryavy, The Internet puting Platform and Core Network for 5G Services, Technical Report.
of Bio-Nano Things, IEEE Commun. Mag. 53 (3) (2015) 32–40, doi:10.1109/ [32] A. Gudipati, D. Perry, L.E. Li, S. Katti, SoftRAN: software defined radio access
MCOM.2015.7060516. network, in: Proceedings of the Second ACM SIGCOMM Workshop on Hot
[3] A. Ghosh, T. Thomas, M. Cudak, R. Ratasuk, P. Moorut, F. Vook, T. Rappaport, Topics in Software Defined Networking, ACM, 2013, pp. 25–30.
G. Maccartney, S. Sun, S. Nie, Millimeter-wave enhanced local area systems: a [33] I.F. Akyildiz, S.-C. Lin, P. Wang, Wireless software-defined networks (W-
high-data-rate approach for future wireless networks, IEEE J. Sel. Areas Com- SDNs) and network function virtualization (NFV) for 5G cellular systems: an
mun. 32 (6) (2014) 1152–1163, doi:10.1109/JSAC.2014.2328111. overview and qualitative evaluation, Comput. Netw. 93, Part 1 (2015) 66–79.
[4] F. Boccardi, R. Heath, A. Lozano, T. Marzetta, P. Popovski, Five disruptive tech- http://dx.doi.org/10.1016/j.comnet.2015.10.013.
nology directions for 5G, IEEE Commun. Mag. 52 (2) (2014) 74–80, doi:10. [34] N. McKeown, T. Anderson, H. Balakrishnan, G. Parulkar, L. Peterson, J. Rexford,
1109/MCOM.2014.6736746. S. Shenker, J. Turner, OpenFlow: enabling innovation in campus networks,
[5] ITU-R, IMT Vision Framework and overall objectives of the future develop- ACM SIGCOMM Comput. Commun. Rev. 38 (2) (2008) 69–74.
ment of IMT for 2020 and beyond, Recommendation ITU-R M.2083-0, Tech- [35] B. Pfaff, J. Pettit, K. Amidon, M. Casado, T. Koponen, S. Shenker, Extending net-
nical Report, International Telecommunication Union, 2015. working into the virtualization layer, ACM Workshop on Hot Topics in Net-
[6] E. Hossain, M. Hasan, 5G cellular: key enabling technologies and research works, 2009.
challenges, IEEE Instrum. Meas. Mag. 18 (3) (2015) 11–21, doi:10.1109/MIM. [36] A. Aissioui, A. Ksentini, A. Gueroui, T. Taleb, Toward elastic distributed
2015.7108393. SDN/NFV controller for 5G mobile cloud management systems, IEEE Access
[7] 3GPP TS 25.913, Requirements for evolved UTRA and evolved UTRAN, 2009. 3 (2015) 2055–2064, doi:10.1109/ACCESS.2015.2489930.
[8] J. Andrews, S. Buzzi, W. Choi, S. Hanly, A. Lozano, A. Soong, J. Zhang, What [37] S.-C. Lin, P. Wang, M. Luo, Control traffic balancing in software defined net-
will 5G Be? IEEE J. Sel. Areas Commun. 32 (6) (2014) 1065–1082, doi:10.1109/ works, Comput. Netw. (2015). http://dx.doi.org/10.1016/j.comnet.2015.08.004.
JSAC.2014.2328098. [38] S.-C. Lin, P. Wang, M. Luo, Jointly optimized qos-aware virtualization and
[9] G. Fettweis, The tactile Internet: applications and challenges, IEEE Veh. Tech- routing in software defined networks, Comput. Netw. 96 (2016) 69–78.
nol. Mag. 9 (1) (2014) 64–70, doi:10.1109/MVT.2013.2295069.
44 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

[39] P. Bosshart, G. Gibb, H.-S. Kim, G. Varghese, N. McKeown, M. Izzard, F. Mu- [68] W. Roh, J.-Y. Seol, J. Park, B. Lee, J. Lee, Y. Kim, J. Cho, K. Cheun, F. Aryan-
jica, M. Horowitz, Forwarding metamorphosis: fast programmable match-ac- far, Millimeter-wave beamforming as an enabling technology for 5G cellular
tion processing in hardware for sdn, in: ACM SIGCOMM Computer Commu- communications: theoretical feasibility and prototype results, IEEE Commun.
nication Review, vol. 43, ACM, 2013, pp. 99–110. Mag. 52 (2) (2014) 106–113, doi:10.1109/MCOM.2014.6736750.
[40] S. Tomovic, M. Pejanovic-Djurisic, I. Radusinovic, SDN based mobile net- [69] I.F. Akyildiz, J.M. Jornet, C. Han, Terahertz band: next frontier for wireless
works: concepts and benefits, Wireless Pers. Commun. 78 (3) (2014) 1629– communications, Phys. Commun. 12 (2014) 16–32.
1644. [70] I. Akyildiz, J. Jornet, C. Han, TeraNets: ultra-broadband communication net-
[41] A.X. Porxas, S.C. Lin, M. Luo, QoS-aware virtualization-enabled routing in soft- works in the terahertz band, IEEE Wireless Commun. 21 (4) (2014) 130–135,
ware defined networks, IEEE ICC’15, 2015. doi:10.1109/MWC.2014.6882305.
[42] S.C. Lin, P. Wang, I.F. Akyildiz, M. Luo, Delay-based maximum power-weight [71] C. Han, A.O. Bicen, I.F. Akyildiz, Multi-ray channel modeling and wideband
scheduling with heavy-tailed traffic (2016). Submitted for publication in characterization for wireless communications in the terahertz band, IEEE
IEEE/ACM Transaction on Networking. Trans. Wireless Commun. 14 (5) (2015) 2402–2412, doi:10.1109/TWC.2014.
[43] F. Van den Abeele, J. Hoebeke, G.K. Teklemariam, I. Moerman, P. Demeester, 2386335.
Sensor function virtualization to support distributed intelligence in the inter- [72] C. Han, A.O. Bicen, I.F. Akyildiz, Multi-wideband waveform design for
net of things, Wireless Pers. Commun. 81 (4) (2015) 1415–1436. distance-adaptive wireless communications in the terahertz band, IEEE Trans.
[44] ETSI GS NFV 003 V1.2.1: Network Functions Virtualisation (NFV); Terminol- Signal Process. 64 (4) (2016) 910–922, doi:10.1109/TSP.2015.2498133.
ogy for Main Concepts in NFV, 2014, (ETSI Industry Specification Group (ISG) [73] O. Momeni, E. Afshari, High power terahertz and millimeter-wave oscillator
NFV). design: a systematic approach, IEEE J. Solid-State Circuits 46 (3) (2011) 583–
[45] R. Sherwood, G. Gibby, K.K. Yapy, G. Appenzellery, M. Casado, N. McKe- 597, doi:10.1109/JSSC.2011.2104553.
owny, G. Parulkar, FlowVisor: A Network Virtualization Layer, Technical Re- [74] I.F. Akyildiz, J.M. Jornet, Electromagnetic wireless nanosensor networks, Nano
port, 2009. Commun. Netw. 1 (1) (2010) 3–19.
[46] OpenStack: The Open Source Cloud Operating System https://www.openstack. [75] I.F. Akyildiz, J.M. Jornet, The internet of nano-things, IEEE Wireless Commun.
org/. 17 (6) (2010) 58–63, doi:10.1109/MWC.2010.5675779.
[47] ETSI GS NFV 002 V1.2.1: Network Functions Virtualisation (NFV); Architec- [76] I.F. Akyildiz, J.M. Jornet, Realizing ultra-massive MIMO communication in the
tural Framework, 2014a, (ETSI Industry Specification Group (ISG) NFV). (0.06-10) terahertz band, Nano Commun. Netw. 3 (4) (2016) 46–54. http://dx.
[48] ETSI GS NFV-MAN 001 V1.1.1: Network Functions Virtualisation (NFV); Man- doi.org/10.1016/j.nancom.2016.02.001.
agement and Orchestration, 2014b, (ETSI Industry Specification Group (ISG) [77] M.K. Samimi, T.S. Rappaport, Statistical channel model with multi-frequency
NFV). and arbitrary antenna beamwidth for millimeter-wave outdoor communica-
[49] R. Mijumbi, J. Serrat, J. Gorricho, N. Bouten, F. De Turck, R. Boutaba, Network tions, in: IEEE Global Telecommunications Conference (GLOBECOM) Work-
function virtualization: state-of-the-art and research challenges, IEEE Com- shop, 2015, pp. 1–7.
mun. Surveys Tuts. 18 (1) (2016) 236–262, doi:10.1109/COMST.2015.2477041. [78] A. Maltsev, A. Pudeyev, I. Karls, I. Bolotin, G. Morozov, W. Keusgen, R. Weiler,
[50] H. Hawilo, A. Shami, M. Mirahmadi, R. Asal, NFV: state of the art, challenges, M. Danchenko, A. Kuznetsov, Quasi-deterministic approach to mmwave chan-
and implementation in next generation mobile networks (vEPC), IEEE Netw. nel modeling in the FP7 MiWEBA project, Wireless Communications and Net-
28 (6) (2014) 18–26, doi:10.1109/MNET.2014.6963800. works, Fraunhofer Heinrich Hertz Institute, Germany, 2014.
[51] M. Ghasemi, T. Benson, J. Rexford, RINC: Real-Time Inference-based Network [79] 3GPP TR 36.873, Study on 3D channel model for LTE, 2014.
Diagnosis in the Cloud, Technical Report, Princeton University, 2015. [80] H. Zhao, R. Mayzus, S. Sun, M. Samimi, J. Schulz, Y. Azar, K. Wang, G. Wong,
[52] R. Mijumbi, J. Serrat, J.-L. Gorricho, S. Latre, M. Charalambides, D. Lopez, Man- F. Gutierrez, T. Rappaport, 28 GHz millimeter wave cellular communication
agement and orchestration challenges in network functions virtualization, measurements for reflection and penetration loss in and around buildings in
IEEE Commun. Mag. 54 (1) (2016) 98–105, doi:10.1109/MCOM.2016.7378433. New York city, in: 2013 IEEE International Conference on Communications
[53] B. Han, V. Gopalakrishnan, L. Ji, S. Lee, Network function virtualization: chal- (ICC), 2013, pp. 5163–5167, doi:10.1109/ICC.2013.6655403.
lenges and opportunities for innovations, IEEE Commun. Mag. 53 (2) (2015) [81] C. Barati, S. Hosseini, S. Rangan, P. Liu, T. Korakis, S. Panwar, T. Rappa-
90–97, doi:10.1109/MCOM.2015.7045396. port, Directional cell discovery in millimeter wave cellular networks, IEEE
[54] F.C.C. 15-138, Notice Of Proposed Rulemaking, Use of Spectrum Bands Above Trans. Wireless Commun. 14 (12) (2015) 6664–6678, doi:10.1109/TWC.2015.
24 Ghz For Mobile Radio Services, 2015. 2457921.
[55] N.O.I. FCC 14-154, In the Matter of Use of Spectrum Bands Above 24 GHz For [82] J. Garcia-Rois, F. Gomez-Cuba, M. Riza Akdeniz, F. Gonzalez-Castano,
Mobile Radio Services, 2014. J. Burguillo-Rial, S. Rangan, B. Lorenzo, On the analysis of scheduling in dy-
[56] Z. Qingling, J. Li, Rain attenuation in millimeter wave ranges, in: 7th Inter- namic duplex multihop mmWave cellular systems, IEEE Trans. Wireless Com-
national Symposium on Antennas, Propagation EM Theory, 2006, pp. 1–4, mun. 14 (11) (2015) 6028–6042, doi:10.1109/TWC.2015.2446983.
doi:10.1109/ISAPE.2006.353538. [83] T. Kim, J. Park, J.-Y. Seol, S. Jeong, J. Cho, W. Roh, Tens of Gbps support with
[57] H. Xu, V. Kukshya, T. Rappaport, Spatial and temporal characteristics of 60 mmWave beamforming systems for next generation communications, in:
GHz indoor channels, IEEE J. Sel. Areas Commun. 20 (3) (2002) 620–630, 2013 IEEE Global Communications Conference (GLOBECOM), 2013, pp. 3685–
doi:10.1109/49.995521. 3690, doi:10.1109/GLOCOM.2013.6831646.
[58] T. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. Wong, J. Schulz, [84] E. Larsson, O. Edfors, F. Tufvesson, T. Marzetta, Massive MIMO for next gen-
M. Samimi, F. Gutierrez, Millimeter wave mobile communications for 5G cel- eration wireless systems, IEEE Commun. Mag. 52 (2) (2014) 186–195, doi:10.
lular: it will work!, IEEE Access 1 (2013) 335–349, doi:10.1109/ACCESS.2013. 1109/MCOM.2014.6736761.
2260813. [85] F. Rusek, D. Persson, B.K. Lau, E. Larsson, T. Marzetta, O. Edfors, F. Tufvesson,
[59] M. Samimi, K. Wang, Y. Azar, G. Wong, R. Mayzus, H. Zhao, J. Schulz, S. Sun, Scaling up MIMO: opportunities and challenges with very large arrays, IEEE
F. Gutierrez, T. Rappaport, 28 GHz angle of arrival and angle of departure Signal Process. Mag. 30 (1) (2013) 40–60, doi:10.1109/MSP.2011.2178495.
analysis for outdoor cellular communications using steerable beam antennas [86] T. Marzetta, Noncooperative cellular wireless with unlimited numbers of base
in New York City, in: 2013 IEEE 77th Vehicular Technology Conference (VTC station antennas, IEEE Trans. Wireless Commun. 9 (11) (2010) 3590–3600,
Spring), 2013, pp. 1–6, doi:10.1109/VTCSpring.2013.6691812. doi:10.1109/TWC.2010.092810.091092.
[60] T. Rappaport, S. Deng, 73 GHz wideband millimeter-wave foliage and ground [87] J. Jose, A. Ashikhmin, P. Whiting, S. Vishwanath, Channel estimation and lin-
reflection measurements and models, in: 2015 IEEE International Confer- ear precoding in multiuser multiple-antenna TDD systems, IEEE Trans. Veh.
ence on Communication Workshop (ICCW), 2015, pp. 1238–1243, doi:10.1109/ Technol. 60 (5) (2011) 2102–2116, doi:10.1109/TVT.2011.2146797.
ICCW.2015.7247347. [88] P. Viswanath, D. Tse, Sum capacity of the vector Gaussian broadcast channel
[61] S. Nie, G. Maccartney, S. Sun, T. Rappaport, 28 GHz and 73 GHz signal outage and uplink-downlink duality, IEEE Trans. Inf. Theory 49 (8) (2003) 1912–1921,
study for millimeter wave cellular and backhaul communications, in: 2014 doi:10.1109/TIT.2003.814483.
IEEE International Conference on Communications (ICC), 2014, pp. 4856– [89] H. Yin, D. Gesbert, M. Filippou, Y. Liu, A coordinated approach to channel es-
4861, doi:10.1109/ICC.2014.6884089. timation in large-scale multiple-antenna systems, IEEE J. Sel. Areas Commun.
[62] S. Sun, T. Rappaport, R. Heath, A. Nix, S. Rangan, MIMO for millimeter-wave 31 (2) (2013) 264–273, doi:10.1109/JSAC.2013.130214.
wireless communications: beamforming, spatial multiplexing, or both? IEEE [90] J. Nam, A. Adhikary, J.-Y. Ahn, G. Caire, Joint spatial division and multiplexing:
Commun. Mag. 52 (12) (2014) 110–121, doi:10.1109/MCOM.2014.6979962. opportunistic beamforming, user grouping and simplified downlink schedul-
[63] Q. Li, G. Li, W. Lee, M. il Lee, D. Mazzarese, B. Clerckx, Z. Li, MIMO techniques ing, IEEE J. Sel. Topics Signal Process. 8 (5) (2014) 876–890, doi:10.1109/JSTSP.
in WiMAX and LTE: a feature overview, IEEE Commun. Mag. 48 (5) (2010) 2014.2313808.
86–92, doi:10.1109/MCOM.2010.5458368. [91] Y.-H. Nam, B.L. Ng, K. Sayana, Y. Li, J. Zhang, Y. Kim, J. Lee, Full-dimension
[64] E. Torkildson, C. Sheldon, U. Madhow, M. Rodwell, Millimeter-wave spatial MIMO (FD-MIMO) for next generation cellular technology, IEEE Commun.
multiplexing in an indoor environment, in: 2009 IEEE GLOBECOM Work- Mag. 51 (6) (2013) 172–179, doi:10.1109/MCOM.2013.6525612.
shops, IEEE, 2009, pp. 1–6. [92] D.S. Baum, J. Hansen, J. Salo, An interim channel model for beyond-3G sys-
[65] A. Molisch, M. Win, J. Winters, Capacity of MIMO systems with antenna tems: extending the 3GPP spatial channel model (SCM), in: 2005 IEEE 61st
selection, in: IEEE International Conference on Communications., 2, 2001, Vehicular Technology Conference, vol. 5, IEEE, 2005, pp. 3132–3136.
pp. 570–574, doi:10.1109/ICC.20 01.9370 04. [93] P. Almers, E. Bonek, A. Burr, N. Czink, M. Debbah, V. Degli-Esposti, H. Hof-
[66] A. Molisch, M. Win, MIMO systems with antenna selection, IEEE Microw. Mag. stetter, P. Kyösti, D. Laurenson, G. Matz, et al., Survey of channel and radio
5 (1) (2004) 46–56, doi:10.1109/MMW.2004.1284943. propagation models for wireless MIMO systems, EURASIP J. Wireless Com-
[67] Y.-S. Choi, A. Molisch, M. Win, J. Winters, Fast algorithms for antenna selec- mun. Netw. 2007 (1) (2007). 56–56
tion in MIMO systems, in: IEEE 58th Vehicular Technology Conference., Vol.
3, 2003, pp. 1733–1737, doi:10.1109/VETECF.2003.1285322.
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 45

[94] X. Lu, A. Tolli, O. Piirainen, M. Juntti, W. Li, Comparison of antenna arrays in [122] S. Geirhofer, P. Gaal, Coordinated multi point transmission in 3GPP LTE het-
a 3-D multiuser multicell network, in: 2011 IEEE International Conference on erogeneous networks, in: 2012 IEEE Globecom Workshops, 2012, pp. 608–612,
Communications (ICC), 2011, pp. 1–6, doi:10.1109/icc.2011.5963126. doi:10.1109/GLOCOMW.2012.6477643.
[95] X. Cheng, B. Yu, L. Yang, J. Zhang, G. Liu, Y. Wu, L. Wan, Communicating in [123] I. Hwang, B. Song, S. Soliman, A holistic view on hyper-dense heterogeneous
the real world: 3D MIMO, IEEE Wireless Commun. 21 (4) (2014) 136–144. and small cell networks, IEEE Commun. Mag. 51 (6) (2013) 20–27, doi:10.
[96] S. Sun, B. Rong, R.Q. Hu, Y. Qian, Spatial domain management and massive 1109/MCOM.2013.6525591.
MIMO coordination in 5G SDN, IEEE Access 3 (2015) 2238–2251, doi:10.1109/ [124] D. Lopez-Perez, M. Ding, H. Claussen, A. Jafari, Towards 1 Gbps/UE in cellular
ACCESS.2015.2498609. systems: understanding ultra-dense small cell deployments, IEEE Commun.
[97] C.A. Balanis, Antenna Theory: Analysis and Design, Wiley-Interscience, 2005. Surveys Tuts. 17 (4) (2015) 2078–2101, doi:10.1109/COMST.2015.2439636.
[98] R. Rogalin, O. Bursalioglu, H. Papadopoulos, G. Caire, A. Molisch, A. Michaloli- [125] BellAir Networks, Cell site backhaul over unlicensed bands, Technical Report,
akos, V. Balan, K. Psounis, Scalable synchronization and reciprocity calibration 2012.
for distributed multiuser MIMO, IEEE Trans. Wireless Commun. 13 (4) (2014) [126] D. Bojic, E. Sasaki, N. Cvijetic, T. Wang, J. Kuno, J. Lessmann, S. Schmid,
1815–1831, doi:10.1109/TWC.2014.030314.130474. H. Ishii, S. Nakamura, Advanced wireless and optical technologies for small-
[99] C. Shepard, H. Yu, N. Anand, E. Li, T. Marzetta, R. Yang, L. Zhong, Argos: prac- cell mobile backhaul with dynamic software-defined management, IEEE Com-
tical many-antenna base stations, in: Proceedings of the 18th Annual Inter- mun. Mag. 51 (9) (2013) 86–93, doi:10.1109/MCOM.2013.6588655.
national Conference on Mobile Computing and Networking, Mobicom, 2012, [127] R. Kaewpuang, D. Niyato, P. Wang, E. Hossain, A framework for cooperative
pp. 53–64. resource management in mobile cloud computing, IEEE J. Sel. Areas Commun.
[100] I. Atzeni, J. Arnau, M. Debbah, Fractional pilot reuse in massive MIMO sys- 31 (12) (2013) 2685–2700, doi:10.1109/JSAC.2013.131209.
tems, in: 2015 IEEE International Conference on Communication Workshop [128] S. Barbarossa, S. Sardellitti, P. Di Lorenzo, Communicating while computing:
(ICCW), 2015, pp. 1030–1035, doi:10.1109/ICCW.2015.7247312. distributed mobile cloud computing over 5G heterogeneous networks, IEEE
[101] R. Muller, L. Cottatellucci, M. Vehkapera, Blind pilot decontamination, IEEE Signal Process. Mag. 31 (6) (2014) 45–55, doi:10.1109/MSP.2014.2334709.
J. Sel. Topics Signal Process. 8 (5) (2014) 773–786, doi:10.1109/JSTSP.2014. [129] N. Fernando, S.W. Loke, W. Rahayu, Mobile cloud computing: a survey, Future
2310053. Gener. Comput. Syst. 29 (1) (2013) 84–106.
[102] A. Ashikhmin, T. Marzetta, Pilot contamination precoding in multi-cell large [130] K. Kumar, J. Liu, Y.-H. Lu, B. Bhargava, A survey of computation offloading for
scale antenna systems, in: 2012 IEEE International Symposium on Informa- mobile systems, Mob. Netw. Appl. 18 (1) (2013) 129–140.
tion Theory Proceedings (ISIT), 2012, pp. 1137–1141, doi:10.1109/ISIT.2012. [131] M. Barbera, S. Kosta, A. Mei, J. Stefa, To offload or not to offload? The band-
6283031. width and energy costs of mobile cloud computing, in: 2013 Proceedings IEEE
[103] D. Ying, F. Vook, T. Thomas, D. Love, Hybrid structure in massive MIMO: INFOCOM, 2013, pp. 1285–1293, doi:10.1109/INFCOM.2013.6566921.
achieving large sum rate with fewer RF chains, in: 2015 IEEE International [132] M. Patel, B. Naughton, C. Chan, N. Sprecher, S. Abeta, A. Neal, et al., Mobile-
Conference on Communications (ICC), 2015, pp. 2344–2349, doi:10.1109/ICC. Edge Computing Introductory Technical White Paper, Mobile-edge Computing
2015.7248675. (MEC) industry initiative (2014).
[104] J. Wallace, M. Jensen, Mutual coupling in MIMO wireless systems: a rigorous [133] S. Subashini, V. Kavitha, A survey on security issues in service delivery mod-
network theory analysis, IEEE Trans. Wireless Commun. 3 (4) (2004) 1317– els of cloud computing, Journal of Network and Computer Applications 34 (1)
1325, doi:10.1109/TWC.2004.830854. (2011) 1–11.
[105] U. Madhow, D.R. Brown, S. Dasgupta, R. Mudumbai, Distributed massive [134] Y. Chen, V. Paxson, R.H. Katz, What is New About Cloud Computing Security?
MIMO: algorithms, architectures and concept systems, in: IEEE Information Technical Report, EECS Department, University of California, Berkeley, 2010.
Theory and Applications Workshop (ITA), IEEE, 2014, pp. 1–7. [135] P. Mell, T. Grance, NIST SP 800-145, The NIST Definition of Cloud Computing,
[106] K. Truong, R. Heath, The viability of distributed antennas for massive MIMO 2011.
systems, in: 2013 Asilomar Conference on Signals, Systems and Computers, [136] C.A. Ardagna, R. Asal, E. Damiani, Q.H. Vu, From security to assurance in the
2013, pp. 1318–1323, doi:10.1109/ACSSC.2013.6810508. cloud: a survey, ACM Comput. Surv. 48 (1) (2015) 2:1–2:50.
[107] J. Andrews, Seven ways that HetNets are a cellular paradigm shift, IEEE Com- [137] O. Vermesan, P. Friess, Internet of Things: Converging Technologies for Smart
mun. Mag. 51 (3) (2013) 136–144, doi:10.1109/MCOM.2013.6476878. Environments and Integrated Ecosystems, River Publishers, 2013.
[108] N. Zhang, N. Cheng, A. Gamage, K. Zhang, J. Mark, X. Shen, Cloud assisted [138] D. Evans, The Internet of Things - How the Next Evolution of the Internet is
HetNets toward 5G wireless networks, IEEE Commun. Mag. 53 (6) (2015) 59– Changing Everything, CISCO White Paper (2011).
65, doi:10.1109/MCOM.2015.7120046. [139] H.-L. Truong, S. Dustdar, Principles for engineering iot cloud systems, IEEE
[109] M. Munoz, C. Rubio, A new model for service and application convergence in Cloud Comput. 2 (2) (2015) 68–76, doi:10.1109/MCC.2015.23.
B3G/4G networks, IEEE Wireless Commun. 11 (5) (2004) 6–12, doi:10.1109/ [140] Integration of cloud computing and internet of things: a survey, Future Gener.
MWC.2004.1351676. Comput. Syst. 56 (2016) 684–700. http://dx.doi.org/10.1016/j.future.2015.09.
[110] A. Ghosh, N. Mangalvedhe, R. Ratasuk, B. Mondal, M. Cudak, E. Visotsky, 021.
T. Thomas, J. Andrews, P. Xia, H. Jo, H. Dhillon, T. Novlan, Heterogeneous cel- [141] F. Bonomi, R. Milito, J. Zhu, S. Addepalli, Fog computing and its role in the
lular networks: from theory to practice, IEEE Commun. Mag. 50 (6) (2012) internet of things, in: Proceedings of the First Edition of the MCC Workshop
54–64, doi:10.1109/MCOM.2012.6211486. on Mobile Cloud Computing, MCC ’12, 2012, pp. 13–16.
[111] W. Webb, On using white space spectrum, IEEE Commun. Mag. 50 (8) (2012) [142] C. Manoj, A.K. Jagannatham, Optimal prediction likelihood tree based source-
145–151, doi:10.1109/MCOM.2012.6257541. channel ml decoder for wireless sensor networks, IEEE Signal Process. Lett.
[112] A.B. Flores, R.E. Guerra, E.W. Knightly, P. Ecclesine, S. Pandey, IEEE 802.11 af: 21 (2) (2014) 135–139, doi:10.1109/LSP.2013.2294794.
a standard for TV white space spectrum sharing, IEEE Commun. Mag. 51 (10) [143] M. Aazam, E.-N. Huh, M. St-Hilaire, C.-H. Lung, I. Lambadaris, Cloud of things:
(2013) 92–100. integration of IoT with cloud computing, in: Robots and Sensor Clouds,
[113] R. Menon, R.M. Buehrer, J.H. Reed, On the impact of dynamic spectrum shar- Springer, 2016, pp. 77–94.
ing techniques on legacy radio systems, IEEE Trans. Wireless Commun. 7 (11) [144] NGMN Alliance, NGMN 5G white paper, Next Generation Mobile Networks,
(2008) 4198–4207. White paper(2015).
[114] Z. Ji, K. Liu, Cognitive radios for dynamic spectrum access-dynamic spec- [145] C.-L. I, C. Rowell, S. Han, Z. Xu, G. Li, Z. Pan, Toward green and soft: a 5G per-
trum sharing: a game theoretical overview, IEEE Commun. Mag. 45 (5) (2007) spective, IEEE Commun. Mag. 52 (2) (2014) 66–73, doi:10.1109/MCOM.2014.
88–94. 6736745.
[115] A. Mohamed, O. Onireti, M. Imran, A. Imran, R. Tafazolli, Control-data sepa- [146] R.P. Jover, I. Murynets, Connection-less communication of iot devices over lte
ration architecture for cellular radio access networks: a survey and outlook, mobile networks, in: Sensing, 12th Annual IEEE International Conference on
IEEE Commun. Surveys Tuts. 18 (1) (2016) 446–465, doi:10.1109/COMST.2015. Communication, and Networking (SECON), 2015, pp. 247–255, doi:10.1109/
2451514. SAHCN.2015.7338323.
[116] T. Zhao, P. Yang, H. Pan, R. Deng, S. Zhou, Z. Niu, Software defined radio im- [147] Y.-D. Lin, Y.-C. Hsu, Multihop cellular: a new architecture for wireless com-
plementation of signaling splitting in hyper-cellular network, in: Proceedings munications, in: Nineteenth Annual Joint Conference of the IEEE Computer
of the Second Workshop on Software Radio Implementation Forum, ACM, and Communications Societies. Proceedings of IEEE INFOCOM, vol. 3, 20 0 0,
2013, pp. 81–84. pp. 1273–1282, doi:10.1109/INFCOM.20 0 0.832516.
[117] D. He, C. Chen, S. Chan, J. Bu, Secure and efficient handover authentication [148] A. Asadi, Q. Wang, V. Mancuso, A survey on device-to-device communication
based on bilinear pairing functions, IEEE Trans. Wireless Commun. 11 (1) in cellular networks, IEEE Commun. Surveys Tuts. 16 (4) (2014) 1801–1819,
(2012) 48–53, doi:10.1109/TWC.2011.110811.111240. doi:10.1109/COMST.2014.2319555.
[118] J. Cao, H. Li, M. Ma, Y. Zhang, C. Lai, A simple and robust handover authenti- [149] M. Tehrani, M. Uysal, H. Yanikomeroglu, Device-to-device communication in
cation between HeNB and eNB in LTE networks, Comput. Netw. 56 (8) (2012) 5G cellular networks: challenges, solutions, and future directions, IEEE Com-
2119–2131, doi:10.1016/j.comnet.2012.02.012. mun. Mag. 52 (5) (2014) 86–92, doi:10.1109/MCOM.2014.6815897.
[119] T. Novlan, R. Ganti, A. Ghosh, J. Andrews, Analytical evaluation of fractional [150] B. Bangerter, S. Talwar, R. Arefi, K. Stewart, Networks and devices for the
frequency reuse for heterogeneous cellular networks, IEEE Trans. Commun. 5G era, IEEE Commun. Mag. 52 (2) (2014) 90–96, doi:10.1109/MCOM.2014.
60 (7) (2012) 2029–2039, doi:10.1109/TCOMM.2012.061112.110477. 6736748.
[120] N. Bhushan, J. Li, D. Malladi, R. Gilmore, D. Brenner, A. Damnjanovic, [151] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklos, Z. Turanyi,
R. Sukhavasi, C. Patel, S. Geirhofer, Network densification: the dominant Design aspects of network assisted device-to-device communications, IEEE
theme for wireless evolution into 5G, IEEE Commun. Mag. 52 (2) (2014) 82– Commun. Mag. 50 (3) (2012) 170–177, doi:10.1109/MCOM.2012.6163598.
89, doi:10.1109/MCOM.2014.6736747. [152] G. Fodor, S. Parkvall, S. Sorrentino, P. Wallentin, Q. Lu, N. Brahmi, Device-to-
[121] A. Damnjanovic, J. Montojo, Y. Wei, T. Ji, T. Luo, M. Vajapeyam, T. Yoo, O. Song, device communications for national security and public safety, IEEE Access 2
D. Malladi, A survey on 3GPP heterogeneous networks, IEEE Wireless Com- (2014) 1510–1520, doi:10.1109/ACCESS.2014.2379938.
mun. 18 (3) (2011) 10–21, doi:10.1109/MWC.2011.5876496.
46 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

[153] X. Lin, J.G. Andrews, A. Ghosh, R. Ratasuk, An overview of 3GPP device-to- [181] C. Han, T. Harrold, S. Armour, I. Krikidis, S. Videv, P.M. Grant, H. Haas,
device proximity services, IEEE Commun. Mag. 52 (4) (2014) 40–48, doi:10. J.S. Thompson, I. Ku, C.-X. Wang, et al., Green radio: radio techniques to en-
1109/MCOM.2014.6807945. able energy-efficient wireless networks, IEEE Commun. Mag. 49 (6) (2011)
[154] M. Usman, A.A. Gebremariam, U. Raza, F. Granelli, A software-defined device- 46–54.
to-device communication architecture for public safety applications in 5G [182] P. Somavat, V. Namboodiri, et al., Energy consumption of personal comput-
networks, IEEE Access 3 (2015) 1649–1654, doi:10.1109/ACCESS.2015.2479855. ing including portable communication devices, J. Green Eng. 1 (4) (2011)
[155] S. Mumtaz, K.M.S. Huq, M.I. Ashraf, J. Rodriguez, V. Monteiro, C. Politis, Cogni- 447–475.
tive vehicular communication for 5G, IEEE Commun. Mag. 53 (7) (2015) 109– [183] K. Moskvitch, Sweet Success for Bio-Battery, 2014. http://www.rsc.org/
117, doi:10.1109/MCOM.2015.7158273. chemistryworld/2014/01.
[156] 5G Automotive Vision (2015) https://5g- ppp.eu/wp- content. [184] K. Simona, Moss-Powered Future, 2014. https://ecopostblog.wordpress.com/
[157] D. Jiang, L. Delgrossi, IEEE 802.11p: towards an international standard for 2014/03/09.
wireless access in vehicular environments, in: IEEE Vehicular Technology [185] 3GPP, Evolved universal terrestrial radio access (E-UTRA) and evolved univer-
Conference, 2008, pp. 2036–2040, doi:10.1109/VETECS.2008.458. sal terrestrial radio access network (E-UTRAN): Overall description Stage 2,
[158] D. Feng, L. Lu, Y. Yuan-Wu, G. Li, G. Feng, S. Li, Device-to-device commu- Release 13 (2016) http://www.3gpp.org/DynaReport/36300.htm.
nications underlaying cellular networks, IEEE Trans. Commun. 61 (8) (2013) [186] N.H. Mahmood, G. Berardinelli, F.M.L. Tavares, P. Mogensen, On the potential
3541–3551, doi:10.1109/TCOMM.2013.071013.120787. of full duplex communication in 5g small cell networks, in: 2015 IEEE 81st
[159] W.H. Chin, Z. Fan, R. Haines, Emerging technologies and research challenges Vehicular Technology Conference (VTC Spring), 2015, pp. 1–5, doi:10.1109/
for 5G wireless networks, IEEE Wireless Commun. 21 (2) (2014) 106–112, VTCSpring.2015.7145975.
doi:10.1109/MWC.2014.6812298. [187] X. Zhang, W. Cheng, H. Zhang, Full-duplex transmission in PHY and MAC lay-
[160] C.-H. Yu, K. Doppler, C. Ribeiro, O. Tirkkonen, Resource sharing optimiza- ers for 5G mobile wireless networks, IEEE Wireless Commun. 22 (5) (2015)
tion for device-to-device communication underlaying cellular networks, IEEE 112–121, doi:10.1109/MWC.2015.7306545.
Trans. Wireless Commun. 10 (8) (2011) 2752–2763, doi:10.1109/TWC.2011. [188] H. Alves, R.D. Souza, M.E. Pellenz, Brief survey on full-duplex relaying and
060811.102120. its applications on 5G, in: 2015 IEEE 20th International Workshop on Com-
[161] L. Wang, H. Wu, Fast pairing of device-to-device link underlay for spectrum puter Aided Modelling and Design of Communication Links and Networks
sharing with cellular users, IEEE Commun. Lett. 18 (10) (2014) 1803–1806, (CAMAD), 2015, pp. 17–21, doi:10.1109/CAMAD.2015.7390473.
doi:10.1109/LCOMM.2014.2351400. [189] T. Hwang, C. Yang, G. Wu, S. Li, G. Li, OFDM and its wireless applications: a
[162] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklos, Z. Turanyi, survey, IEEE Trans. Veh. Technol. 58 (4) (2009) 1673–1694, doi:10.1109/TVT.
Design aspects of network assisted device-to-device communications, IEEE 20 08.20 04555.
Commun. Mag. 50 (3) (2012) 170–177, doi:10.1109/MCOM.2012.6163598. [190] IEEE Standard for Broadband over Power Line Networks: Medium Access Con-
[163] B. Zhou, H. Hu, S.-Q. Huang, H.-H. Chen, Intracluster device-to-device relay trol and Physical Layer Specifications, IEEE Std 1901–2010 (2010) 1–1586,
algorithm with optimal resource utilization, IEEE Trans. Veh. Technol. 62 (5) doi:10.1109/IEEESTD.2010.5678772.
(2013) 2315–2326, doi:10.1109/TVT.2012.2237557. [191] M. Morelli, C.-C. Kuo, M.-O. Pun, Synchronization techniques for orthogonal
[164] E. Hossain, D. Niyato, Z. Han, Dynamic Spectrum Access and Management in frequency division multiple access (OFDMA): a tutorial review, Proc. IEEE 95
Cognitive Radio Networks, 1st Edition, Cambridge University Press, New York, (7) (2007) 1394–1427, doi:10.1109/JPROC.2007.897979.
NY, USA, 2009. [192] P. Banelli, S. Buzzi, G. Colavolpe, A. Modenini, F. Rusek, A. Ugolini, Modulation
[165] S. Jaeckel, L. Raschkowski, K. Borner, L. Thiele, QuaDRiGa: a 3-D multi-cell formats and waveforms for 5G networks: who will be the heir of OFDM?: an
channel model with time evolution for enabling virtual field trials, IEEE Trans. overview of alternative modulation schemes for improved spectral efficiency,
Antennas Propag. 62 (6) (2014) 3242–3256, doi:10.1109/TAP.2014.2310220. IEEE Signal Process. Mag. 31 (6) (2014) 80–93, doi:10.1109/MSP.2014.2337391.
[166] T. Zwick, C. Fischer, W. Wiesbeck, A stochastic multipath channel model in- [193] H. Ochiai, H. Imai, On the distribution of the peak-to-average power ratio in
cluding path directions for indoor environments, IEEE J. Sel. Areas Commun. OFDM signals, IEEE Trans. Commun. 49 (2) (2001) 282–289, doi:10.1109/26.
20 (6) (2002) 1178–1192, doi:10.1109/JSAC.2002.801218. 905885.
[167] C.-C. Chong, C.-M. Tan, D.I. Laurenson, S. McLaughlin, M.A. Beach, A.R. Nix, [194] J. Van De Beek, F. Berggren, Out-of-band power suppression in OFDM, IEEE
A novel wideband dynamic directional indoor channel model based on a Commun. Lett. 12 (9) (2008) 609–611.
Markov process, IEEE Trans. Wireless Commun. 4 (4) (2005) 1539–1552, [195] J. Zhang, B. Zhang, S. Chen, X. Mu, M. El-Hajjar, L. Hanzo, Pilot Contamina-
doi:10.1109/TWC.2005.850341. tion Elimination for Large-Scale Multiple-Antenna Aided OFDM Systems, IEEE
[168] A.O. Kaya, L.J. Greenstein, W. Trappe, Characterizing indoor wireless channels J. Sel. Topics Signal Process. 8 (5) (2014) 759–772, doi:10.1109/JSTSP.2014.
via ray tracing combined with stochastic modeling, IEEE Trans. Wireless Com- 2309936.
mun. 8 (8) (2009) 4165–4175, doi:10.1109/TWC.2009.080785. [196] S. Pagadarai, R. Rajbanshi, A.M. Wyglinski, G.J. Minden, Sidelobe suppres-
[169] S. Nie, C. Han, I.F. Akyildiz, A 3-Dimensional Time-Varying Channel Model for sion for OFDM-based cognitive radios using constellation expansion, in:
5G Wireless Communication Systems, 2016. Manuscript submitted for publi- 2008 IEEE Wireless Communications and Networking Conference, IEEE, 2008,
cation pp. 888–893.
[170] M. Olsson, C. Cavdar, P. Frenger, S. Tombaz, D. Sabella, R. Jantti, 5GrEEn: to- [197] B. Farhang-Boroujeny, OFDM versus filter bank multicarrier, IEEE Signal Pro-
wards green 5G mobile networks, in: 2013 IEEE 9th International Confer- cess. Mag. 28 (3) (2011) 92–112, doi:10.1109/MSP.2011.940267.
ence on Wireless and Mobile Computing, Networking and Communications [198] F. Schaich, T. Wild, Y. Chen, Waveform contenders for 5G - suitability for short
(WiMob), 2013, pp. 212–216, doi:10.1109/WiMOB.2013.6673363. packet and low latency transmissions, in: 2014 IEEE 79th Vehicular Tech-
[171] H.Y. Lateef, M. Dohler, A. Mohammed, M.M. Guizani, C.F. Chiasserini, Towards nology Conference (VTC Spring), 2014, pp. 1–5, doi:10.1109/VTCSpring.2014.
energy-aware 5G heterogeneous networks, in: Energy Management in Wire- 7023145.
less Cellular and Ad-hoc Networks, Springer, 2016, pp. 31–44. [199] F. Schaich, T. Wild, Waveform contenders for 5G–OFDM vs. FBMC vs. UFMC,
[172] M. Marsan, L. Chiaraviglio, D. Ciullo, M. Meo, Multiple daily base station in: 2014 6th International Symposium on Communications, Control and Sig-
switch-offs in cellular networks, in: 2012 Fourth International Conference on nal Processing (ISCCSP), 2014, pp. 457–460, doi:10.1109/ISCCSP.2014.6877912.
Communications and Electronics (ICCE), 2012, pp. 245–250, doi:10.1109/CCE. [200] A. Sahin, I. Guvenc, H. Arslan, A survey on multicarrier communications: pro-
2012.6315906. totype filters, lattice structures, and implementation aspects, IEEE Commun.
[173] M. Aftab Hossain, R. Jantti, C. Cavdar, Energy saving market for mobile op- Surveys Tuts. 16 (3) (2014) 1312–1338, doi:10.1109/SURV.2013.121213.00263.
erators, in: 2015 IEEE International Conference on Communication Workshop [201] T. Fusco, A. Petrella, M. Tanda, Sensitivity of multi-user filter-bank multicar-
(ICCW), 2015, pp. 2856–2861, doi:10.1109/ICCW.2015.7247612. rier systems to synchronization errors, in: 2008 3rd International Sympo-
[174] M. Marsan, M. Meo, Network sharing and its energy benefits: a study of sium on Communications, Control and Signal Processing, 2008, pp. 393–398,
European mobile network operators, in: 2013 IEEE Global Communications doi:10.1109/ISCCSP.2008.4537257.
Conference (GLOBECOM), 2013, pp. 2561–2567, doi:10.1109/GLOCOM.2013. [202] G. Wunder, M. Kasparick, S. ten Brink, F. Schaich, T. Wild, I. Gaspar, E. Ohlmer,
6831460. S. Krone, N. Michailow, A. Navarro, G. Fettweis, D. Ktenas, V. Berg, M. Dry-
[175] S. Raza Zaidi, A. Afzal, M. Hafeez, M. Ghogho, D. McLernon, A. Swami, So- janski, S. Pietrzyk, B. Eged, 5GNOW: challenging the LTE design paradigms
lar energy empowered 5G cognitive metro-cellular networks, IEEE Commun. of orthogonality and synchronicity, in: 2013 IEEE 77th Vehicular Technology
Mag. 53 (7) (2015) 70–77, doi:10.1109/MCOM.2015.7158268. Conference (VTC Spring), 2013, pp. 1–5, doi:10.1109/VTCSpring.2013.6691814.
[176] L.-C. Wang, S. Rangapillai, A survey on green 5G cellular networks, in: 2012 [203] A. Skrzypczak, P. Siohan, J.-P. Javaudin, Analysis of the peak-to-average power
International Conference on Signal Processing and Communications (SPCOM), ratio for OFDM/OQAM, in: IEEE 7th Workshop on Signal Processing Advances
2012, pp. 1–5, doi:10.1109/SPCOM.2012.6290252. in Wireless Communications, 20 06, 20 06, pp. 1–5, doi:10.1109/SPAWC.2006.
[177] X. Zhang, J. Zhang, W. Wang, Y. Zhang, C.-L. I, Z. Pan, G. Li, Y. Chen, Macro- 346413.
assisted data-only carrier for 5G green cellular systems, IEEE Commun. Mag. [204] M. Schellmann, Z. Zhao, H. Lin, P. Siohan, N. Rajatheva, V. Luecken, A. Ishaque,
53 (5) (2015) 223–231, doi:10.1109/MCOM.2015.7105669. FBMC-based air interface for 5G mobile: challenges and proposed solutions,
[178] D. Lin, G. Charbit, I.-K. Fu, Uplink contention based multiple access for 5G in: 2014 9th International Conference on Cognitive Radio Oriented Wire-
cellular IoT, in: 2015 IEEE 82nd Vehicular Technology Conference (VTC Fall), less Networks and Communications (CROWNCOM), 2014, pp. 102–107, doi:10.
2015, pp. 1–5, doi:10.1109/VTCFall.2015.7391184. 4108/icst.crowncom.2014.255708.
[179] T. Chen, H. Kim, Y. Yang, Energy efficiency metrics for green wireless com- [205] M. Payaro, A. Pascual-Iserte, M. Najar, Performance comparison between
munications, in: 2010 International Conference on Wireless Communications FBMC and OFDM in MIMO systems under channel uncertainty, in: European
and Signal Processing (WCSP), IEEE, 2010, pp. 1–6. Wireless Conference, 2010, pp. 1023–1030, doi:10.1109/EW.2010.5483521.
[180] ECR Initiative, Network and Telecom Equipment - Energy and Performance
Assessment, Technical Report, 2010.
 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48 47

[206] I. Estella, A. Pascual-Iserte, M. Payaro, OFDM and FBMC performance compar- [222] L. Dai, B. Wang, Y. Yuan, S. Han, C.-L. I, Z. Wang, Non-orthogonal multiple
ison for multistream MIMO systems, in: Future Network and Mobile Summit, access for 5G: solutions, challenges, opportunities, and future research trends,
2010, pp. 1–8. IEEE Commun. Mag. 53 (9) (2015) 74–81, doi:10.1109/MCOM.2015.7263349.
[207] R. Zakaria, D. Le Ruyet, A novel filter-bank multicarrier scheme to mitigate [223] I. Da Silva, E. Authors, S.E. El Ayoubi, O.M. Boldi, Ö. Bulakci, P. Spapis, M.
the intrinsic interference: application to MIMO systems, IEEE Trans. Wireless Schellmann, H. ERC, J.F. Monserrat, T. Rosowski, et al., 5G RAN architecture
Commun. 11 (3) (2012) 1112–1123, doi:10.1109/TWC.2012.012412.110607. and functional design, METIS II White Paper.
[208] Y. Chen, F. Schaich, T. Wild, Multiple access and waveforms for 5G: IDMA [224] A. Alkhateeb, O. El Ayach, G. Leus, R. Heath, Channel estimation and hybrid
and universal filtered multi-carrier, in: IEEE 79th Vehicular Technology Con- precoding for millimeter wave cellular systems, IEEE J. Sel. Topics Signal Pro-
ference (VTC Spring), 2014, pp. 1–5, doi:10.1109/VTCSpring.2014.7022995. cess. 8 (5) (2014) 831–846, doi:10.1109/JSTSP.2014.2334278.
[209] V. Vakilian, T. Wild, F. Schaich, S. ten Brink, J.-F. Frigon, Universal-filtered [225] AT&T, AT&T Unveils 5G Roadmap Including Trials In 2016. http://about.att.
multi-carrier technique for wireless systems beyond LTE, in: IEEE Globecom com/story/unveils_5g_roadmap.html.
Workshops (GC Wkshps), 2013, pp. 223–228, doi:10.1109/GLOCOMW.2013. [226] Qualcomm, 5G - Vision for the next generation of connectivity, White Paper,
6824990. 2015 https://www.qualcomm.com/invention/technologies/5g.
[210] X. Wang, T. Wild, F. Schaich, Filter optimization for carrier-frequency- and [227] Y.J. Bultitude, T. Rautiainen, IST-4-027756 WINNER II D1. 1.2 V1. 2 WINNER II
timing-offset in universal filtered multi-carrier systems, in: IEEE 81st Ve- Channel Models.
hicular Technology Conference (VTC Spring), 2015, pp. 1–6, doi:10.1109/ [228] J. Medbo, K. Borner, K. Haneda, V. Hovinen, T. Imai, J. Jarvelainen, T. Jamsa,
VTCSpring.2015.7145842. A. Karttunen, K. Kusume, J. Kyrolainen, P. Kyosti, J. Meinila, V. Nurmela,
[211] G. Fettweis, M. Krondorf, S. Bittner, GFDM - generalized frequency division L. Raschkowski, A. Roivainen, J. Ylitalo, Channel modelling for the fifth gen-
multiplexing, in: IEEE 69th Vehicular Technology Conference, 2009, pp. 1–4, eration mobile communications, in: 2014 8th European Conference on An-
doi:10.1109/VETECS.2009.5073571. tennas and Propagation (EuCAP), 2014, pp. 219–223, doi:10.1109/EuCAP.2014.
[212] N. Michailow, M. Matthe, I. Gaspar, A. Caldevilla, L. Mendes, A. Festag, G. Fet- 6901730.
tweis, Generalized frequency division multiplexing for 5th generation cel- [229] The Institute of Electrical and Electronics Engineers (IEEE), IEEE Standard for
lular networks, IEEE Trans. Commun. 62 (9) (2014) 3045–3061, doi:10.1109/ Information technology–Telecommunications and information exchange be-
TCOMM.2014.2345566. tween systems–Local and metropolitan area networks–Specific requirements-
[213] G. Wunder, P. Jung, M. Kasparick, T. Wild, F. Schaich, Y. Chen, S. Brink, I. Gas- Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer
par, N. Michailow, A. Festag, L. Mendes, N. Cassiau, D. Ktenas, M. Dryjanski, (PHY) Specifications Amendment 3: Enhancements for Very High Through-
S. Pietrzyk, B. Eged, P. Vago, F. Wiedmann, 5GNOW: non-orthogonal, asyn- put in the 60 GHz Band, IEEE Std 802.11ad-2012 (Amendment to IEEE Std
chronous waveforms for future mobile applications, IEEE Commun. Mag. 52 802.11-2012, as amended by IEEE Std 802.11ae-2012 and IEEE Std 802.11aa-
(2) (2014) 97–105, doi:10.1109/MCOM.2014.6736749. 2012) (2012) 1–628, doi:10.1109/IEEESTD.2012.6392842.
[214] A. Awoseyila, C. Kasparis, B. Evans, Improved preamble-aided timing es- [230] METIS II Project, Preliminary Views and Initial Considerations on 5G RAN Ar-
timation for OFDM systems, IEEE Commun. Lett. 12 (11) (2008) 825–827, chitecture and Functional Design, 2015 https://metis-ii.5g-ppp.eu/.
doi:10.1109/LCOMM.2008.081054. [231] mmMAGIC Project, mmMAGIC: Millimetre-Wave Based Mobile Radio Ac-
[215] A. Aminjavaheri, A. Farhang, A. RezazadehReyhani, B. Farhang-Boroujeny, Im- cess Network for Fifth Generation Integrated Communications, 2015
pact of Timing and Frequency Offsets on Multicarrier Waveform Candidates https://5g-ppp.eu/mmmagic/.
for 5G, arXiv preprint arXiv:1505.0 080 0(2015). [232] G. Wunder, P. Jung, M. Kasparick, T. Wild, F. Schaich, Y. Chen, S. Brink, I. Gas-
[216] A. Farhang, N. Marchetti, L. Doyle, Low complexity modem design for GFDM, par, N. Michailow, A. Festag, L. Mendes, N. Cassiau, D. Ktenas, M. Dryjanski,
IEEE Trans. Signal Process. PP (99) (2015) 1–1, doi:10.1109/TSP.2015.2502546. S. Pietrzyk, B. Eged, P. Vago, F. Wiedmann, 5GNOW: non-orthogonal, asyn-
[217] R. Datta, G. Fettweis, Z. Kollar, P. Horvath, FBMC and GFDM interference can- chronous waveforms for future mobile applications, IEEE Commun. Mag. 52
cellation schemes for flexible digital radio PHY design, in: 14th Euromicro (2) (2014) 97–105, doi:10.1109/MCOM.2014.6736749.
Conference on Digital System Design (DSD), 2011, pp. 335–339, doi:10.1109/ [233] A. Banchs, M. Breitbach, X. Costa, U. Doetsch, S. Redana, C. Sartori, H. Schot-
DSD.2011.48. ten, A novel radio multiservice adaptive network architecture for 5G net-
[218] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, K. Higuchi, Non- works, in: IEEE 81st Vehicular Technology Conference (VTC Spring), 2015,
orthogonal multiple access (NOMA) for cellular future radio access, in: IEEE pp. 1–5, doi:10.1109/VTCSpring.2015.7145636.
77th Vehicular Technology Conference (VTC Spring), 2013, pp. 1–5, doi:10. [234] 5th Generation Public Private Partnership, mmMAGIC: Millimetre-Wave
1109/VTCSpring.2013.6692652. Based Mobile Radio Access Network for Fifth Generation Integrated Commu-
[219] A. Benjebbour, Y. Saito, Y. Kishiyama, A. Li, A. Harada, T. Nakamura, Concept nications, 2015 http://5g-ppp.eu.
and practical considerations of non-orthogonal multiple access (NOMA) for [235] Samsung Electronics, Samsung Electronics Sets 5G Speed Record at 7.5Gbps,
future radio access, in: 2013 International Symposium on Intelligent Signal Over 30 Times Faster Than 4G LTE, 2014 http://www.samsung.com/global/
Processing and Communications Systems (ISPACS), 2013, pp. 770–774, doi:10. business/networks/insights/news.
1109/ISPACS.2013.6704653. [236] C.-L. I, C. Rowell, S. Han, Z. Xu, G. Li, Z. Pan, Toward green and soft: a 5G per-
[220] M.-R. Hojeij, J. Farah, C. Nour, C. Douillard, Resource allocation in down- spective, IEEE Commun. Mag. 52 (2) (2014) 66–73, doi:10.1109/MCOM.2014.
link non-orthogonal multiple access (NOMA) for future radio access, in: 6736745.
2015 IEEE 81st Vehicular Technology Conference (VTC Spring), 2015, pp. 1– [237] Huawei, 5G: New Air Interface and Radio Access Virtualization, Technical Re-
6, doi:10.1109/VTCSpring.2015.7146056. port, 2015.
[221] A. Benjebbour, A. Li, Y. Saito, Y. Kishiyama, A. Harada, T. Nakamura, System-
level performance of downlink NOMA for future LTE enhancements, in:
2013 IEEE Globecom Workshops (GC Wkshps), 2013, pp. 66–70, doi:10.1109/
GLOCOMW.2013.6824963.
48 I.F. Akyildiz et al. / Computer Networks 106 (2016) 17–48

Ian F. Akyildiz received the B.S., M.S., and Ph.D. degrees in Computer Engineering from the University of Erlangen-Nurnberg, Germany, in 1978,
1981 and 1984, respectively. Currently, he is the Ken Byers Chair Professor in Telecommunications with the School of Electrical and Computer
Engineering, Georgia Institute of Technology (Georgia Tech), Atlanta, GA USA; the Director of the Broadband Wireless Networking (BWN) Laboratory
and the Chair of the Telecommunication Group at Georgia Tech. Since 2013, he is a FiDiPro Professor (Finland Distinguished Professor Program
(FiDiPro) supported by the Academy of Science) with the Department of Electronics and Communications Engineering, at Tampere University of
Technology, Finland, and the founding director of NCC (Nano Communications Center). Since 2008, he is also an honorary professor with the
School of Electrical Engineering at Universitat Politcnica de Catalunya (UPC) in Barcelona, Catalunya, Spain, and the founding director of N3Cat
(NaNoNetworking Center in Catalunya). Since 2011, he is a Consulting Chair Professor at the Department of Information Technology, King Abdulaziz
University (KAU) in Jeddah, Saudi Arabia. He is the Editor-in-Chief of Computer Networks (Elsevier) Journal, and the founding Editor-in-Chief of the
Ad Hoc Networks (Elsevier) Journal, the Physical Communication (Elsevier) Journal and the Nano Communication Networks (Elsevier) Journal. He
is an IEEE Fellow (1996) and an ACM Fellow (1997). He received numerous awards from IEEE and ACM. His h-index is 88 and the total number
of citations is above 69K due to Google scholar as of March 2015. His current research interests are in wireless sensor networks in challenged
environments, 5G Cellular systems, nanonetworks, Terahertz Band, and software defined networks.

Shuai Nie received the B.S. degree in Electrical Engineering from Xidian University in 2012, and the M.S. degree in Electrical Engineering from New
York University in 2014. She is a graduate research assistant in the Broadband Wireless Networking Laboratory (BWN Lab), School of Electrical and
Computer Engineering, Georgia Institute of Technology. Currently, she is working toward the Ph.D. degree in Electrical and Computer Engineering
under the supervision of Prof. Ian F. Akyildiz. Her research interests include the 5G cellular system and telecommunication networks.

Shih-Chun Lin received the B.S. degree in Electrical Engineering and the M.S. degree in Communication Engineering from National Taiwan Uni-
versity in 2008 and 2010, respectively. He is a graduate research assistant in the Broadband Wireless Networking Laboratory (BWN Lab), School of
Electrical and Computer Engineering, Georgia Institute of Technology. Currently, he is working toward the Ph.D. degree in Electrical and Computer
Engineering under the supervision of Prof. Ian F. Akyildiz. His research interests include wireless sensor networks, software defined networking,
large machine-to-machine communication, cognitive radio networks, and statistical scheduling in wireless systems.

Manoj Chandrasekaran received the B.Tech. degree in Electronics and Communications Engineering from Amrita Vishwa Vidyapeetham University,
Coimbatore, India, in 2011 and the M.Tech. degree in Electrical Engineering from Indian Institute of Technology Kanpur, India in 2013. From July
2013 to July 2015, he worked as a Software Development Engineer at Microsoft India R&D Pvt. Ltd. Currently, he is pursuing Ph.D. degree under the
supervision of Prof. Ian F. Akyildiz in the Broadband Wireless Networking Lab of School of Electrical and Computer Engineering, Georgia Institute
of Technology, Atlanta, Georgia. His current research is focused on 5G cellular networks.