Architecture harmonization between cloud radio access network and fog network

SPECIAL SECTION ON EMERGING CLOUD-BASED WIRELESS COMMUNICATIONS AND NETWORKS
Received October 15, 2015, accepted November 19, 2015, date of publication December 17, 2015,
date of current version January 8, 2016.
Digital Object Identifier 10.1109/ACCESS.2015.2509638
Architecture Harmonization Between Cloud
Radio Access Networks and Fog Networks
SHAO-CHOU HUNG1, HSIANG HSU1, SHAO-YU LIEN2,
AND KWANG-CHENG CHEN1, (Fellow, IEEE)
1Graduate Institute of Communication Engineering, National Taiwan University, Taipei 10617, Taiwan
2Department of Electronic Engineering, National Formosa University, Yulin 632, Taiwan
Corresponding author: S.-C. Hung (d02942008@ntu.edu.tw)
This research was supported by the Ministry of Science and Technology, Taiwan ROC, and MediaTek Inc. under the contracts MOST
104-2622-8-002-002 and MOST 104-2221-E-002-082.
ABSTRACT To guarantee the ubiquitous and fully autonomous Internet connections in our daily life, the
new technical challenges of mobile communications lie on the efficient utilization of resource and social
information. To facilitate the innovation of the fifth generation (5G) networks, the cloud radio access
network (RAN) and fog network have been proposed to respond newly emerging traffic demands. The
cloud RAN functions more toward centralized resource management to achieve optimal transmissions. The
fog network takes advantage of social information and edge computing to efficiently alleviate the end-to-
end latency. In this paper, we conduct a comprehensive survey of these two network structures, and then
investigate possible harmonization to integrate both for the diverse needs of 5G mobile communications.
We analytically study the harmonization of cloud RAN and fog network from various points of view,
including the cache of Internet contents, mobility management, and radio access control. The performance
of transition between the cloud RAN and the fog network has been presented and the subsequent switching
strategy has been proposed to ensure engineering flexibility and success.
INDEX TERMS 5G, fog network, cloud radio access network, RAN, heterogeneous network,
edge computing, cloud computing, cache, radio resource management, mobility, mobile communications,
vehicular network.
I. INTRODUCTION
Being deployed for several decades, mobile/cellular
infrastructures successfully provide seamless and reliable
streaming (voice/video) services for billions of mobile users.
From GSM/GPRS, UMTS, to LTE/LTE-A, the transmission
data rates have been enhanced a million-fold. The recent
deployment of the heterogeneous networks (HetNets) [1]–[5]
consisting of macrocells, small cells (femtocells, picocells),
and/or further relay nodes, ubiquitously support basic multi-
media and Internet browsing applications. In many occasions,
it seems satisfactory to primitive human-to-human (H2H)
communication applications using existing network archi-
tectures/technologies. However, to substantially facilitate
human daily activities in addition to basic voice/video and
Internet access services, achieving ‘‘full automation’’ and
‘‘everything-to-everything’’ (X2X) had been regarded as an
ultimate goal not only for future information communication
industry, but also for financial transactions, e-commerce,
social communities, transportation, agriculture, and energy
allocation [6]. ‘‘Full automation’’ implies a significant
enhancement of human being’s sensory and processing capa-
bilities, which embraces unmanned or remotely controlled
vehicles/robots/offices/factories/augmented/virtual reality,
and immerse sensory human interactions of cyber-physical-
social systems. The goal is to employ distributed-autonomous
control to relieve/simplify the network control and evolutive,
by which the resource utilization can be boosted in the
dynamic complex networks, and be re-optimized after the
major environmental changes [7]. On the other hand, X2X
connection implies that diverse entities, including human
and machines are able to form general-sense communities
other than to H2H, such as social networks of human-to-
machine (H2M) and machine-to-machine (M2M) facilitating
the ultimate cyber-physical-social systems [8]–[10]. To name
a few application scenarios include intelligent transportation
systems (ITS) [11], volunteer information networks [12],
Internet of Things (IoT) [13]–[15], smart grids [16], [17] and
much more.
VOLUME 3, 2015
2169-3536
2015 IEEE. Translations and content mining are permitted for academic research only.
Personal use is also permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
3019
www.redpel.com +917620593389

S.-C. Hung et al.: Architecture Harmonization Between Cloud RANs and Fog Networks
To enable these various applications, boosting the
transmission data rates is just one of the diverse require-
ments. The performance in terms of end-to-end transmission
latency [6], energy efficiency, reliability, scalability, cost
efficiency as well as stability shall also be fundamentally
enhanced to enable every aspect of mobile Internet services in
an omni way. As the data traffic from Internet has gradually
been dominating the traffic volume in mobile communication
systems [18], in addition to the improvement of air-interface,
migration to more efficient network architecture is definitely
a must in technology development. Furthermore, a large
portion of current traffic data is user-generated via social
networks such as documents, pictures, videos, messages.
Such data are circulated among the users’ sides according to
their social relationship, which precisely indicates the inter-
play between mobile communication networks and social
network as shown in [19]. More precisely, most of the data
are generated at the edge of networks, but stored and analyzed
in the clouds. Consequently, the current Internet architecture
which partitioning networks into layers will not be able to
support these heterogeneous applications in an affordable
cost [20]. All of these new technology opportunities suggest
the need of evolving a state-of-the-art network architecture
beyond ultra-efficient air-interface. This paper starts with the
introduction of cloud radio access networks in Section II and
fog networks in Section III, with their unique features to deal
with future technology challenges, in spite of quite different
philosophy behind. In Section IV, we propose harmonization
H-CRAN with FogNet in architecture. With emerging cache
technology into our harmonization, we analytically treat the
network optimization as a whole in Section V. In addition to
IoT/X2X and Internet content traffic, in light of automatic
driving, unmanned vehicles, and service robots, we investi-
gate mobility management and subsequent various handover
design, to lay out the framework of this harmonization for 5G
and future mobile communication networks in Section VI.
The overall top-down design paradigm on top of resource
access in this harmonization is presented in Section VII.
Numerical results follow in Section VIII.
II. CLOUD RADIO ACCESS NETWORKS
The great success of mobile communications in past decades
brings billions of user equipments and devices into networks
to demand high bandwidth connections in the air. Following
the breakthrough of multiple-input-multiple-output (MIMO)
technology to approach the Shannon limit in past decades,
the next generation of mobile communications can not solely
rely on the enhancement from air-interface transmission.
Introducing the HetNet architecture to facilitate the concept
of small cells has been shown to further increase system
capacity [3]. In the HetNet, in addition to the Macrocells
formed by the existing eNodeBs, there are heterogeneous
small cell networks (e.g. femto or pico cell) underlay or
overlay the Macrocells. The motivation of such architecture is
to increase the spatial spectrum reuse and increase the whole
network efficiency. For this reason, the use of very dense and
low-power-small-cells with highly spacial spectrum reuse is a
promising way to allow handling such tremendous amount of
devices [21], [22]. However, even though densely deployed
small cells can provide shorter transmission distance and
more efficient spatial reuse, it also introduces additional
inter-cell interference problem and extra management issues.
To compensate such those potential defects, heterogeneous
cloud radio access network (H-CRAN) has been proposed on
top of cloud radio access networks (C-RAN).
The C-RAN architecture can be traced back to the proposal
by IBM [23] and further elaborated in [24]. The concept orig-
inates from the hierarchical network architecture of UMTS,
in which each radio network controller (RNC) coordinates a
number of NodeBs [25]. In UMTS, radio resource manage-
ment (RRM) is conducted by each RNC, while NodeBs only
perform physical signal transmissions/receptions. Although
RRM is subsequently implemented to be performed by
an eNodeB in LTE/LTE-A/EPC, this concept opens the
designs of integrating the radio resource optimization in
individual eNodeBs into a joint optimization. Coordinated
multi-point (CoMP) transmissions/receptions are thus a prac-
tical paradigm of joint resource scheduling/optimization
among multiple eNodeBs [26]–[28] and have been included
in Release 11 of the specifications [29], [30]. The details
of C-RAN were described in [24]. The baseband units
pool (BBUs pool) and the radio head do not have to be
collocated within an eNodeB. Instead, a number of BBUs
and remote radio heads (RRHs) can be separated from an
eNodeB to be massively deployed. Through the fiber-optic
cables to connect an eNodeB and BBUs/RRHs, coverage of
eNodeBs is therefore ubiquitously extended. This C-RAN
architecture has attracted great research interests [31]–[34].
Later, Peng et al. [35] and Lei et al. [36] further revealed
the conceptual realization of H-CRAN, as shown in Fig. 1.
Compared to the C-RAN architecture, with the deployment
of high power nodes (HPNs), the coordination of RRHs
and HPN can be more efficient to alleviate the interference
problem. Through the provided wired/wireless interfaces
(i.e., S1, X2 and Un), not only BBUs/RRHs but also relay
nodes, and HPNs are able to exchange information for joint
resource scheduling/allocation.
FIGURE 1. System architecture of the H-CRAN. RRHs can provide
short-distance communication for UEs to improve transmission rate and
HPN can provide ubiquitous connection to achieve seamless coverage.
3020 VOLUME 3, 2015

Even though it has been analyzed that H-CRAN are more
cost efficient [37], the entire cost structure behind H-CRAN
remains worth further further investigation, due to potential
IoT applications invoking significant growth in terms of the
number of devices and traffic demand in the air. First, the
number of heterogeneous devices (e.g., user equipments,
sensors, vehicles, robots, etc.) is dramatically increasing in
the future, particularly with the growth of IoT applications.
The additional BBU pools are needed to support the high
complexity of resource optimization in H-CRAN. Second, the
traffic amount in the cloud radio increases explosively. More
importantly, most data are generated by the users and is likely
to be propagated via various social media platforms as inter-
action with friends (i.e. local in human networks to highlight
the importance to treat mobile networks and social network
together). With the treelike topology in the H-CRAN, all the
information exchanged among edge users are not suitable
for this characteristic. It will introduce additional amount of
burden on the front-haul and back-haul links, especially in
wireless type [38].
In the H-CRAN, three facts are generally ignored.
(i) Traffic may be exchanged socially and locally. It is
assumed that each packet from each edge entity may be
delivered to another edge entity in the world under the
H-CRAN. However, this assumption may not be generally
practical, as more and more social applications only require
data exchanges in close physical proximity. (ii) Each edge
entity may exchange information with certain edge entities
more frequently than other edge entities, while such entities
of closer interaction can be considered as a social network
in the general sense. Such a social network can be a set of
webs, servers, individuals, or machines/devices. (iii) Down-
link traffic to different edge entities or uplink traffic from
different edge entities may be with strong correlations. For
example, a large number of users may enjoy the same sport
game streaming program simultaneously, and therefore the
downlink traffic to these users is highly correlated. A group
of densely deployed sensors measuring a common physical
quantity may obtain a highly correlated result, and thus the
uplink data to the cloud may also be correlated. It is reported
to adopt in-network computation to significantly enhance
spectrum utilization [39]. As a result, the technical merits of
the H-CRAN simultaneously bring those engineering chal-
lenges to limit the performance of the H-CRAN at the edge
side. This predicament thus motivates the concept of fog
computing and thus fog networking.
III. FOG NETWORK
The Fog-Network (FogNet) was initiated by Cisco to enable
the fog computing technology at the edge of the network [40].
The main characteristics of FogNet include ubiquity, decen-
tralized management and cooperation [41]. FogNets are com-
posed of a large amount of devices connecting to Internet
like it device, wearable devices and self-driving vehicles,
etc. These devices form many ‘‘mini clouds’’ at the edge
of the network and manage themselves in a distributed way.
On the contrast to the cloud computing, the fog computing
facilitates processing/computing capabilities at edge entities,
by which not all information for performance optimization
should be delivered to the cloud. Only the tasks (and corre-
sponding information for optimization) those cannot be well
processed by edge entities are handled by the cloud. For
example, the users in the FogNet can release some of their
own computing/storage capacity to support their neighboring
devices. The users need not download the data from the
core network, instead, they just download the required data
from their neighbors, like the adjacent small cell network, or
other mobile devices through proximity direct links. Without
complex routing in the core network, it can be expected
that the reduction of end-to-end latency is reachable. The
FogNet, therefore, may significantly alleviate the computing
and routing burdens in the cloud-part of networks to achieve
the scalability.
Aryafar et al pioneered the facilitation of fog comput-
ing into a new type of network architecture known as the
FogNet [42]. Under the FogNet, each edge entity having
social messages to be exchanged among other edge entities
does not rely on traffic relay via the cloud, while edge entities
in close physical proximity are able to locally share messages.
This design leads to the concept of socially-aware traffic
managements to significantly decrease the traffic amounts to
be supported by the cloud. Recently, radio access technolo-
gies (RATs) such as device-to-device (D2D) communicat-
ions [43]–[47] and using small cells [48] as smart data/traffic
routing gateways are successful practices of the FogNet.
When a group of edge entities have highly correlated traf-
fic to be delivered to the cloud, each entity does not have
to upload traffic individually. Instead, the common part of
traffic is delivered once by a single edge entity. On the
other hand, when the cloud has highly correlated traffic to
be forwarded to multiple edge entities, the cloud does not
forward traffic individually to each edge entity. Instead, the
cloud only selects one edge entity to forward traffic, then the
selected edge entity autonomously shares traffic with other
edge entities. Consequently, the amounts of traffic supported
by both the fronthaul and backhaul links can be largely
alleviated.
Although the FogNet provides considerable technical
virtues to potentially tackle the issues of complexity, scalabil-
ity, and heavy traffic burdens in the H-CRAN, new challenges
emerge at the same time. First, although traffic can be socially
shared among edge entities, there is no guarantee that all edge
entities needing this traffic are able to successfully receive
this traffic. Therefore, reliability of data delivery turns out to
be the primary concern. Second, the mobility management
and service continuity may not be sufficiently supported
in the FogNet. Third, for the H-CRAN, as the optimized
resource scheduling/allocation is the key requirement, inter-
ference can be well rejected/mitigated. However, due to the
lack of effective resource coordination among edge entities,
interference may drastically impact on the performance of the
FogNet.
VOLUME 3, 2015 3021

IV. IMPACT OF TWO SYSTEMS WITHOUT
HARMONIZATION
These two effective networking strategies of somewhat
opposite approaches can create further challenges in their
co-existence, which primarily include storage management,
load balancing, and interference.
1) STORAGE MANAGEMENT
It is expected that the FogNet can help to reduce the data
loading in the H-CRAN. Ideally, we may expect that the local
information going around in a small set of entities in the social
network are stored in the local FogNet and just those popular
data/media would be uploaded into the cloud-part. However,
without proper exchange of control signaling that takes time
delay, the cloud-part (H-CRAN) and fog-part (FogNet) do not
know whether another system stores such data or not. It may
result in either both sides store or both sides do not. In this
way, the storage resources cannot be utilized optimally and
redundant traffic flows in the network. To solve this problem,
a possible way is to identify the social relationship among the
FogNet and to report this information back to the H-CRAN
to help both appropriately allocate the storage resources and
consequently reduce network traffic by popular data.
2) LOAD BALANCING
Under the coexistence of H-CRAN and FogNet, there are
multiple types of devices ranged from the devices equipped
with a large power transceiver like vehicles to the device
relying only on energy harvesting (EH) such as sensor net-
works. The wireless links include cellular (like UMTS or
LTE), WiFi, mmWave [49], etc. All differ in coverage, band-
width, and capacity. To optimize the performance of the
whole network, an important issue is the load balancing. Load
balancing focuses on how to guarantee that all the network
can optimally operate under their own way. In [50], it has been
pointed out that the most intuitive way like always connect-
ing to the network with the best signal-to-interference-plus-
noise-ratio (SINR) may not be the most efficient scheme.
Therefore, it is critical to design a handover scheme beyond
SINR based for devices to choose between H-CRAN and
FogNet.
3) INTERFERENCE
An indoor device under co-existence of H-CRAN and FogNet
may obtain/send desirable data via D2D links, WiFi, and/or
cellular systems. However, such a scenario may jeopardize
the functions of a local FogNet. For example, the D2D links,
underlaying in the cellular network, may suffer from the
interference from the cellular network. The devices connected
with WiFi may also be interrupted by the cellular network
with Licensed-Assisted Access. It has been expected that
there will be 80% of traffic is generated by the devices in the
indoor environment. If the H-CRAN cannot provide a suitable
way to control the interference to the FogNet, all these data
can be only uploaded to the cloud-part and crash the whole
systems.
TABLE 1. Comparison of H-CRAN and FogNet.
V. HARMONIZATION OF H-CRAN and FogNet
To compare the H-CRAN and the FogNet, the features of
these two architectures are summarized in Table 1, while
each of them is effective to deal with certain new traffic
patterns. All in all, the H-CRAN focus on the global resource
allocation/utilization optimization through a centralized way
and the FogNet facilitates the information exchange and com-
putation at the edge of the network. The design philosophy
of the H-CRAN and FogNet appear opposite each other.
However, instead of arguing preference, we note that different
technical merits in the H-CRAN and in the FogNet may
lead to a complementary harmonization. The necessity of
harmonization includes the following:
1) For the devices of a small form factor like EH in
FogNets, only very limited energy or power is available
for transmission. Without coordination between cloud-
part and fog-part, these small-size devices may suffer
from severe interference and the FogNet may collapse.
2) Wireless fronthual (backhual) [51] is necessary in the
region where the cost of building infrastructure is
large. By offloading the burden from the H-CRAN, the
FogNet can increase the feasibility of wireless fron-
thual (backhual) and therefore decrease the cost of the
network simultaneously.
3) The flexibility of radio resource utilization and the
latency performance can be improved via the FogNet.
The BBU pools can allocate more radio resources to
the hot spots; for example, the traffic storm in sensor
networks due to some urgencies or emergent accidents
like tsunami or earthquake. BBU pools can also broad-
cast the information to smart phones to save the time to
respond.
To enable the harmonization, Peng et al bring the idea
of FogNet into the H-CRAN architecture by taking the
correlation among traffic to/from different edge users into
account [52]. The hybrid architecture FogNet-HCRAN net-
work (abbreviated F-CRAN in the following) are illustrated
in Fig. 2. F-CRAN is composed of cloud-part and fog-part.
In the cloud-part, there exists high power nodes (HPNs) to
cover a wider geographical area and RRHs to provide the
conventional functions in C-RAN. The devices can connect to
Internet through the cloud-part or the fog-part. The fog-parts
are composed of all kinds of devices that can provide ser-
vices to other devices, or furthermore, including cloud-part.
3022 VOLUME 3, 2015

FIGURE 2. Evolution of the architecture of F-CRAN. The whole network is composed of cloud and fog-part., that
is H-CRAN and FogNet. The cloud-part is composed of RRHs and HPN. The aim of cloud-part is to provide the
ubiquitous connection service to all the device. The fog-part is composed of multiple mini clouds coexisting with
the cloud-part. The devices can exchange data locally or connect to the cloud-part through F-APs.
These ‘‘powerful’’ devices form the multiple mini clouds at
the edge of the network. F-APs may play the roles as the coor-
dinators in each of these mini clouds. The F-APs can be inter-
faced into the BBU pool through the fronthual links. F-APs
can provide services from physical layer management, like
basic radio resource management, to the higher layer manage-
ments, like cache memory management. Since large amounts
of data or control signals are processed in the FogNet, the
burden of fronthauls and BBU pools are alleviated. If these
services are removed from the F-APs, the F-APs degenerate
into the RRHs and all the computing load and file exchange
burdens are shifted to BBU pool or even to the core net-
work and cloud, which may significantly increase the latency
resulting in poor quality of user experience. To reduce the
end-to-end latency and take advantages of social relationship
among networks, the technique of cache can be implemented
both in the cloud and fog-part to reduce the burden of the core
network [53], [54].
However, whether these two eagles fuse into one of greater
power or turn into a turkey remains unclear. In the following,
we further investigate sufficient conditions to adequately inte-
grate the H-CRAN and the FogNet. We analytically explore
a unique top-down system design based on the proper alloca-
tion of all kinds of resources in the entire F-CRAN, according
to the need of users or applications, particularly the caching
for social media and virtual reality, and the resource/mobility
management for radio access control of diverse service
requirements. Our results establish the foundation toward a
network architecture of 5G mobile communications.
VI. CACHING IN APPLICATION LAYER
A. CACHING IN WIRELESS NETWORKS
Caching mechanisms are originally a common methodology
to reduce traffic volume and meanwhile access latency in
computer systems like a CPU and database centers [55].
In late 1990s, caching has been implemented in the wired
web application systems, called web cache systems (or HTTP
cache systems), to store copies of heavily accessed docu-
ments in the networks, thereby reducing bandwidth usage,
server load, and improving web retrieving stability, latency
and quality of service (QoS) [56]–[58]. In web caching sys-
tems, a client could store web contents for later reuse, called
a forward position system; moreover, a web server (e.g. a
search engine) may also cache copies of web contents in
content delivery networks (CDNs), called a reverse position
system. These two caching mechanisms operate together in
web caching systems as complements to each other, making
a successful and efficient web content retrieving system. The
reason for the efficacy behind caching mechanisms is that
most of the traffic flows in the Internet are attributed to a
relatively small part of the data or contents in the networks,
a phenomenon which can be traced back to web requests and
proxy traces [59]. In fact, the characteristics of these popular
contents have been verified to follow mathematical forms
as power-law distributions, meaning that the probability of
attaining a certain content ck of rank k ( i.e. the kth popular
content) is proportional to k−ζ , with ζ greater than or equal
to 1. That is Pr {ck} ∝ k−ζ . Power-law distribution explicitly
implies that only a few higher ranks of data occupy most
portion of the traffic volume in the Internet. Identifying the
most popular Internet contents, and caching the contents in
the networks therefore greatly reduce the traffic volume and
latency, since for most Internet users, they need not to acquire
the contents from remote data centers.
In the past decade, the communications industry has
experienced a dramatic variation: mobile Internet traffic grad-
ually dominates wireless networks [18], [60], with distinct
features compared to traditional telephony traffic and short
message service (SMS), which are usually transmitted to
VOLUME 3, 2015 3023

and needed by a single user in the network. People start
to access social media, streaming videos, and other con-
tents that possess traffic characteristics of Internet contents
via mobile devices. This fundamental dissimilitude on traf-
fic characteristics gives designers opportunities to introduce
caching in wireless networks and hence motivates us to
redesign our wireless mobile network together with the core
network to support caching mechanisms. Niesen et al and
Maddah-Ali et al have developed information-theoretic
frameworks to analyze the performance limit of caching
gain by coded caching with respect to cache size available
to mobile users [61], [62], revealing positive outcomes of
caching in wireless networks. Consequently, the engineering
implementation of caching becomes an urgent issue to net-
work designers and researchers.
Even though caching mechanisms have brought
extraordinary success in wired networks and theoretical
foundation in computing has been well established, there
still exist challenges to implement caching mechanisms in
wireless networks, especially the cache in the radio access
network to abate the traffic volume from remote data centers,
with an aim of reducing latency as well. Different caching
mechanisms have been proposed; e.g. in-network caching in
the core network and base stations (BSs) to optimize data
retrieving latency for mobile users [63]; in [64], the authors
discuss caching at BSs and at mobile users as two cases of
caching utilization. However, if caching is considered in a
more general way as a sort of resource in the radio access
network, rather than separate storage capacity at devices
and infrastructures, then the utilization of caching turns into
resource allocation problems in wireless networks [65]. The
resource allocation perspective allows us to ruminate the
harmonization of caching utilization in centralized H-CRANs
(cloud-parts) and distributed FogNets (fog-parts), to result in
a new cache utilization problem in F-CRAN.
B. CACHING UTILIZATION IN F-CRAN
Regarding caching as a general resource than wireless
bandwidth facilitates the design of F-CRAN, which satisfac-
torily addresses three difficulties when implementing caching
in wireless networks. First, considering caching at the infras-
tructures like BS, the backhaul traffic volume would indeed
be greatly reduced, and thus enhances the latency since
the transmission bottleneck on the backhaul is alleviated.
However, the limited wireless cellular bandwidth still acts
as another bottleneck for data acquisition when the number
of mobile users increases [66]. An efficient way to tackle
this problem is to allow D2D communications for direct
content sharing, and hence abate the need of wireless cellu-
lar bandwidth. We expect similar methodologies for virtual
reality (VR) traffic in the future. Introducing D2D commu-
nication in the networks is actually based on caching in the
mobile devices; therefore FogNets are reasonably regarded
as an auxiliary to reduce the burden of the air interface;
making FogNet a promising design for the future realization
of mobile networks. Second, in FogNets, the mobility of
user devices might jeopardize the efficiency of D2D com-
munication and hence the performance of caching due to
the unstable nature of wireless communication. To optimally
utilize device caching is the main challenge. However, with
the aid of H-CRAN, the mobility and interference of devices
could be managed, since H-CRAN integrates all the infor-
mation and is equipped with mighty computational power.
Furthermore, some coded caching protocols could also be
implemented, rendering caching a more operative method-
ology. Third, the information collection and instantaneous
monitoring of the characteristics of the contents becomes
an indispensable part in CDNs. The highly-centralized
H-CRAN provides an opportune solution for this prob-
lem, making caching at infrastructure and traffic monitoring
viable. These three explanations cause H-CRAN and FogNet
perfect complements to each other; the coordination of
H-CRAN and Fog-Network in the F-CRAN is thus totally
different from the early studies about D2D on relaying
purpose [67], [68].
FIGURE 3. Under the F-CRAN architecture, there are three different paths
for data retrieving: (1) retrieve the data via direct connections to other
users (2) retrieve the data from cached copies at BBU pool (3) retrieve the
data from the cloud network (traditional path). FogNet helps to reduce
the burden of cloud-part by the first path.
The complete scenario of the F-CRAN system for
caching as resource utilization is shown in Fig. 3.
As H-CRAN and FofNets are combined via cahcing,
three different main connections for the mobile users
exist: (1) direct links to other devices (D2D communi-
cation) (2) devices to caching in BBUs pool (3) device
to the cloud network (conventional link). To summarize,
in order to integrate the H-CRAN and FogNets via these
three different connections, the overall cache utilization
can be generalized into the optimization of resources
in a network, by incorporating infrastructure caching
(H-CRAN cache) and device caching (D2D communication)
into a new networking design scenario of joint optimization
on the backhaul networking, storage, computing, and radio
resource allocation in the air-interface.
3024 VOLUME 3, 2015

FIGURE 4. Schemes of in-band overlay, in-band underlay, and outband
D2D. In the formulation, we adopt in-band overlay scheme and optimize
the usage of D2D spectrum as caching in devices.
C. GRAPHICAL NETWORK MODEL AND
CHARACTERISTICS OF INTERNET CONTENTS
Let us consider a straightforward scenario with one F-AP and
mobile devices, where the mobile devices forms a FogNet,
as illustrated in Fig. 3. This scenario could be extended to
more than one F-AP under the management of the BBU
pool. Since under the structure of H-CRAN, the cell size
shrinks, leading to smaller mean distances among users, we
adopt a random graph model by assuming that for any two
users under a F-AP, there is a probability p that they can
communicate and share contents with each other. Thus, we
modelled the network topology of direct device communica-
tion by Erdös-Rényi (ER) random graph [69]. This random
graph model offers a performance upper bound for device
caching and D2D communication supported by the FogNets,
since ER model overweights the connectivity of two mobile
devices and ignore possible clustering structures formed by
mobile users. Moreover, ER model also provides mathemati-
cal tractability for performance analysis.
The BBU pool has a given cache size M memory
units (MU), and the storage sizes at the devices are finite.
For an Internet content ck of rank k, its file size is sk, with
k = 1, 2, · · · , K, where K is the number of active Internet
contents. As mentioned in Sec. VI-A, the popularities of
Internet contents being requested in a network during a period
of time can be characterized by power-law distributions [59],
pk = Pr{ck} = HK,ζ k−ζ , with ζ > 0 and HK,ζ being the
normalization factor. The popularity is reasonably assumed
to be constants in the observed period. For the size of the
Internet contents, it is assumed that sk follows Log-Normal
distribution with mean µ and variance σ2 [70]. The powerful
H-CRAN servers and the core network are able to acquire the
information of popularities {pk} and sizes {sk} of the active
Internet contents by traffic monitoring and statistics gather-
ing. For instance, the H-CRAN server can record the type of
data from the logic channels and transport channels and the
uniform resource locator (URL) requested from mobile users.
Here, as suggested in [65], caching utilization among
users in FogNets is actually spectrum utilization of D2D
bandwidth. The spectrum utilization scheme for D2D com-
munication can be categorized into three different types: in-
band overlay, in-band underlay, and out-band (as visualized
in Fig. 4).
1) In-Band Overlay: D2D devices utilize a reserved
fraction of cellular spectrum. Therefore, the devices
need not to perform spectrum sensing.
2) In-Band Underlay: D2D devices and cellular traffic
share the same spectrum. However, D2D devices need
spectrum sensing to control their interference to cellu-
lar network users under a certain threshold.
3) Out-Band: There is a part of unlicensed spectrum avail-
able for the D2D devices and all the D2D links operate
in this band only.
In this paper, we consider the in-band overlay scheme as
a conceptual discussion about spectrum utilization to avoid
complicated spectrum sensing issues, as studied in [71].
Therefore, the total available spectrum in a F-AP cell
is divided a fraction WD2D for D2D communication
and the other fraction WC for original cellular downlink
traffic.
D. FORMULATION OF CACHING UTILIZATION
To optimally utilize caching resource in F-CRAN, we need
to decide which Internet content should be cached in the
H-CRAN and which should be shared in FogNets through
D2D links to minimize the total traffic volume, which
consists of backhaul traffic volume FB and downlink traf-
fic volume FD. Therefore, we introduce two binary vari-
ables with state space {0, 1}. δC
k decides whether the Internet
content ck should be cached at the BBU pool; δD
k decides
whether ck should be obtained using D2D communication in
the FogNets. Therefore, the backhual traffic accounts from
the requests of contents that are neither cached at the BBU
pool, nor can be shared among users. Similarly, the downlink
traffic volume comes from the requests of contents that are
not cached at the BBU pool. We use scalarization method [72]
to optimize the two traffic simultaneously, yielding the
following optimization problem:
minimize
{δC
k ,δD
k }
Ftotal = αFB + βFD (1)
subject to
K
k=1
δC
k sk ≤ M (2)
FD ≤ WC (3)
K
k=1
Npksk − FB − FD ≤ WD2D, (4)
where the constants 0 ≤ α ≤ 1 and β = 1 − α repre-
sent the importance of backhual traffic and downlink traffic
respectively.
The constraints of the optimization problem in (1) should
take the following issues into account. First, the utilization
of caching resource at the BBU pool should not exceed its
storage capacity M, as given in (2). Moreover, the volume of
downlink traffic and D2D communication traffic in FogNet
also should not exceed pre-allocated fraction WC and WD2D,
as respectively described in (3) and (4), where K
k=1 Npksk
is the traffic volume of the active Internet Contents. In other
words, we maximize caching utilization in F-CRAN by min-
imizing the traffic volume between FogNets and BBU pool,
and between BBU pool and the data centers.
VOLUME 3, 2015 3025

FIGURE 5. Optimal caching utilization in F-CRAN for α = β = 0.5. It is
suggested that H-CRAN and the FogNets should cooperatively store the
Internet contents according to their characteristics.
E. OPTIMAL HARMONIZATION OF CACHING
UTILIZATION IN F-CRAN
The optimization problem in (1) is a binary integer
programming problem. To carry out numerical results of opti-
mal caching utilization, we assume 100 mobile users under a
F-AP, and 50 Internet content actively requested by the users.
The popularity of the contents follows power-law distribution
with ζ = 2. The caching capacity at the BBU pool is M = 30
GigaBytes (GB). For the existence of D2D sharing links,
we assume that p = 0.1. We consider the representative
case which the traffic load at backhaul and at downlink is
equally heavy (α = β = 0.5) and µ varies from 1 to 3
with σ2 = 1. We ran the optimization problem for more than
10, 000 times and gathered statistics of the optimal caching
utilization of Internet contents, i.e. to count the normalized
sum of δC
k (number of Internet contents cached by the BBU
pool) and the normalized sum of δD
k (number of Internet
contents shared in the FogNet). The two statistics, shown as
lines with circle markers and lines with cross markers respec-
tively in Fig. 5, offer us the suggestions of optimal caching
utilization regarding F-CRAN. In the figure, it is clear that
for the higher rank Internet contents, caching them at the
BBU pool is suggested to be optimal utilization of the caching
capacity. Nonetheless, for the rest of the Internet contents
and especially with small file sizes, it is suggested that the
requests of them should be directly satisfied in the FogNet.
The essence is that although FogNet indeed help caching and
traffic load releasing, the unstable nature of FogNets should
also be calculated; thus, for Internet contents with small file
sizes, the chance for direct sharing is much higher, resulting in
this optimal caching utilization. The minimized traffic in (1)
under optimal caching utilization with respect to the different
storage size at the BBU pool is shown in Fig. 6. As the
spectrum of D2D communication increases (larger WD2D to
WC ratio), the role of FogNets becomes more important; the
benefit of introducing FogNets into H-CRAN also increases.
The mitigation of traffic loads as well improves the retrieving
latency for Internet contents, making another contribution
from the optimal caching utilization in F-CRAN. The results
of harmonization between H-CRAN and FogNets suggest
FIGURE 6. Minimized total traffic under optimal caching utilization for
different ratios of WD2D and WC . It is clear that with the assistance of
FogNets, the total traffic volume is further alleviated.
that F-CRAN truly provides a more general solution to traf-
fic offloading via caching implementation. Additionally, this
harmonization philosophy of F-CRAN could be easily and
effectively extended to other traffic types in the future than
Internet contents, like VR traffic and popular road traffic
information in vehicular networks.
VII. MOBILITY MANAGEMENT
A. VEHICULAR NETWORK IN FogNet
ITS emerges as an even more important feature in the
5G mobile communication systems than before. Over these
years, ITS has been developed aiming at improving road
traffic safety and at automated driving in the vehicle industry.
Therefore, it is expected that the passengers have the same
demand concerning connectivity performance in the vehicles
as at home and work [73]. On the other hand, the global
market is expected to reach 130 billions by 2019 [74]. These
connected vehicles could provide alternatives to alleviate the
vehicular traffic congestions via intelligent traffic control
and managements [75]. To enable the vehicular communi-
cation, the Federal Communications Commission (FCC) in
the United State has allocated 75MHz centered at 5.9GHz
for ITS system [76]. Therefore, the success of vehicular
network could be an essential part toward the success of the
5G mobile networks due to the ubiquitous deployment of
cellular systems.
The goal of the vehicular networks is to provide
human-safety services which include road safety infor-
mation exchange, emergency alarms, traffic management,
localization and navigation, and even unmanned intelligent
driving [73]. All these applications involve a large amount
of information change and extremely low end-to-end latency
transmissions. To support these services, the early vehicu-
lar network protocol 802.11p [77] has been proposed. The
802.11p mainly focuses on the vehicle-to-vehicle (V2V)
communication through wireless link based on the dedicated
short range communication (DSRC). It is similar to the
D2D links between vehicles. Easy deployment, low cost to
construct and to maintain, and capability to accommodate
the ad-hoc mode V2V scenario, are its technical merits [78].
However, it suffers from intermittent and short-live
3026 VOLUME 3, 2015

connectivity between vehicles. Its scalability, geographic
coverage, insufficient radio access to meet quality of services
(QoS), and lacking overall network architecture to satisfy
ultra-low latency to the cloud or to facilitate appropriate
control functions of physical entities, become the major
concerns. These challenges motivate us to consider other
types of communication network, to achieve the safety and
latency requirements. Therefore, the establishment of the
heterogeneous vehicular networks and corresponding proper
management of the resources in such networks arise as the
new technology challenges regarding vehicular networks [79]
and even networking for service robots. All these suggest new
show-case opportunities for 5G cellular networks.
Nevertheless, the realizations of the HetNets are always a
challenge for wireless network engineers. These challenges
include interference coordination, radio resource allocation,
cooperative radio signal processing and frequent handover
problem. To overcome these obstacles, F-CRAN [52] may
serve as an attractive solution. In the F-CRAN, not only the
HPNs provide ubiquitous connections to the mobile devices,
but also F-APs can provide short distance connections to the
devices at the edge of the networks, by which, the perfor-
mance of the vehicle network can be further improved. For
example, the latency of traffic information can be achieved
by downloading from the nearby F-APs instead of the remote
cloud data center. Furthermore, the vehicles can offload or
exchange the data with the FogNet to reduce the burden of
the cloud network.
B. HANDOVER SCHEME IN F-CRAN
Handover and subsequent mobility management are of
critical importance in the mobile communication networks,
especially in the highly dynamic environment like vehicular
networks. Particularly, with massive deployment of small cell
networks like F-APs, the handovers happen more frequently
and result in a heavy burden on fronthaul and core net-
works [80]. To achieve seamless services for the vehicles with
high mobility, in [81], a survey of F-CRAN architecture is
provided, which discusses how high mobility devices should
be served by macro cell like HPN network and low mobility
should be serviced by small cells like F-APs. With multiple
access networks, traffic flows can be balanced to avoid con-
gestion and performance degradation. Therefore, not only the
switches between different access points in the same network
(horizontal handover) but also between different networks
(vertical handover) are urgently wanted.
The most common approach is designed based on the
straightforward parameter, the received signal
strength (RSS) [82]–[84]. By detecting the RSS through
reference control signals, the mobile devices can access
the best wireless network while entering new cells. More
details can be found in the survey paper [85]. How-
ever, these RSS based handover algorithms might not
be satisfactory under interference, say to cause unneces-
sary handover such as ping-pong effects, which is severer
in the small coverage cell [86]. In addition, these RSS based
algorithms are designed based on the assumption that
the mobile devices can communicate with only one
access point. However, with the facilitation of CoMP, the
F-CRAN architecture can support a mobile device accessing
multiple accessing points. The CoMP technique is presented
to mitigate inter-cell interference and QoS improvement in
highly density network. Through feeding all the information
to the centralized processing server, the interference among
different small cells like F-APs can be mitigated. In [87],
a comprehensive introduction and performance evaluation
are provided, and the evolution of the vehicular networks is
illustrated in Fig. 7. By utilizing CoMP techniques, all mobile
devices can access the F-APs at the fof-part of F-CRAN in
the same frequency without suffering from the interference,
which implies that no need for hard-lined cell definition in the
F-CRAN architecture. Therefore, RSS based approach may
not be the best choice for the F-CRAN architecture to tackle
the handover problem.
FIGURE 7. The evolution of the vehicular network under the F-CRAN
architecture. We use different colors of access points to illustrate
different frequency bands. In the conventional vehicular network, all the
vehicles can connect to one access point with a certain frequency band to
avoid interference. In the F-CRAN architecture, CoMP allows all the
vehicles can connect to multiple ones in the same frequency without
interference. Therefore, the cell concept is alleviated.
Please recall that the small access points are also limited
resources in the wireless networks. Because the performance
gain of the CoMP highly depends on the perfect knowledge
of the channel state information and frequent control signal
exchanges, if all the mobile devices utilize all the small access
points, it may increase the complexity of interference can-
cellation algorithms [81]. On the other hand, for the mobile
devices of urgent operation like having some emergent acci-
dents, it is reasonable to allocate more accessing points for
them to guarantee the emergency information can be trans-
mitted successfully. Therefore, accessing all available small
cell networks may not always be a good strategy throughout
the entire network. A different thinking to design the mobile
networks under such scenario may be needed, If we regard
the access points of the mobile network as another kind
of resource for mobile devices, then the handover can be
VOLUME 3, 2015 3027

generalized to a dynamic resource allocation problem. In this
new perspective, the resources to be allocated are not radio
resource units in the spectrum anymore, but the access points
like the RRHs or F-APs.
Therefore, we stand on the viewpoint of resource alloca-
tion to innovatively reformulate the vertical and horizontal
handover in the F-CRAN architecture. With the realization
of CoMP, the vehicles or mobile machines like robots can
request more resource at the fog-part, like roadside access
points (RAPs) or other small network access points F-APs
in F-CRAN. While it is necessary, the vehicles can ask for
more resource from the cloud-part of the F-CRAN system.
C. TIME DYNAMIC RESOURCE ALLOCATION FOR
HANDOVER SCHEME
As previous description, the handover can be commonly con-
sidered as a detection mechanism. The devices need to make
a decision whether the RSS is over the threshold or not. The
handover problem and resource allocation are usually studied
separately. Nevertheless, the direct impact of the handover
problem of the resource allocation is the mechanism to allo-
cate the resource for the newly arriving devices. It is possible
that the network with the best SINR may not have remaining
resource for the upcoming devices. To solve these two prob-
lems in a unified framework, Stevens-Navarro et al. design
the vertical handover algorithm based on the Markov decision
process (MDP) to dynamically allocate network resource for
the mobile devices [88]. Qin et al. integrates the handover
scheme into the time allocation scheme which simultaneously
reach fairness and optimal link gains [89]. Generally speak-
ing, little attention has been paid in the literatures to jointly
optimize the radio resource allocation, allocation of access
points in mobile network, and handover together. We will
be working on this emerging technology challenge in the
following.
D. STOCHASTIC NETWORK MODEL
We consider the scenario that the mobile machines like vehi-
cles or robots connected to the Internet through the F-RAN
as shown in the Fig. 7. The networks are composed of mobile
devices, F-APs serving locally and HPN covers the whole
network. These mobile devices can access the F-APs or the
HPN through total M channels. To describe the connection
between the vehicles and F-APs, a common approach is to
model the transmission region as a circle. This is the reason
that the we only utilize the wireless link with good long-
term performance, which is dominated by the distance to the
receiver. In this model, the candidate F-APs for vehicles to
handover are only in the circle with the radius R. To find
the general performance, we assume that all the vehicles
and F-APs are uniformly distributed in the infinite flat area.
We also assume that all the vehicles are moving straightly
with velocity v(meters/s) and choosing different direction
randomly.
Under this model, the quality of wireless links depends
on not only channel fading, but also the interference from
other vehicles in the same channels. To connect the quality
of wireless links and the interference, a popular approach is
to describe the quality of wireless links as outage probability
by stochastic geometry [90]. In the stochastic geometry, the
power of the interference from a single source depending on
the distance to the receiver and follows d−α, where α is a
path loss effect coefficient. With this assumption, we can see
that the interference may go to infinity while the distance d
is close to zero, which is not possible in the real world.
Therefore, the stochastic geometry analysis shall supply the
performance lower bound of the outage probability.
E. HANDOVER PROBLEM FORMULATION
We define the queue length U(t) as data in the queue
at tth time slot. To guarantee the system stability, that is,
U limT→∞ 1/T ∞
t=1 U(t) < ∞, the system should
allocate more resource to increase service rate u(t) while it
is necessary. At each time slot t, the service rate u(t) of the
mobile devices is determined by the number of connected
communication links with F-APs and the decision space is
denoted as Dt :
Dt =
{1, . . . , N(t)}, if N(t) = 0
{0}, if N(t) = 0.
(5)
We denote n(t) ∈ Dt as the number of connected F-APs at
the time slot t. Therefore, the number of serviced packets at
each time slot u(t) can be expressed as
u(t) =
n(t)
i=1
1i, (6)
where 1i is the index function of the wireless link correspond-
ing to the ith F-AP. 1i = 1 if the SIR of the ith link larger than
the threshold θ.
From here we can note that the horizontal handover is
the process that the vehicles or mobile devices ask more
wireless link resource from the local F-APs. Once more
resources (F-APs) are allocated to the vehicles, it has faster
service rate or higher bandwidth of transmission. However,
other vehicles may not get the necessary resource to stabi-
lize its own queue. Therefore, a suitable solution is that all
the vehicles minimize its utilization of APs resources while
simultaneously they still can stabilize its own traffic queue.
On the other hand, it may be possible that even though all
the available resources are utilized by vehicles and mobile
users, it is still possible that certain queue(s) cannot be sta-
bilized. For example, such a situation becomes more likely
for the data traffic flow from significantly increasing vehicles
and mobile users in rush hours. A possible alternative to
resolve this dilemma is to execute vertical handover, that
is, the vehicles borrow more resources from cloud-part
(i.e. HPNs of F-CRAN) to offload data traffic. Nevertheless,
the vertical handover is much more complicated than horizon-
tal handover. To avoid the additional burden of the cloud side
and the complexity of vertical handover, the vehicles should
not access the cloud as possible as they can. To describe
3028 VOLUME 3, 2015

the horizontal handover and vertical handover problem
more precisely, we formulate the mathematical problems as
follows.
Horizontal Handover Problem:
min
n(t)∈D(t)
lim
T→∞
1
T
T
t=1
n(t)
subject to lim
T→∞
1
T
T
t=1
E(U(t)) < ∞, (7)
Vertical Handover Problem:
max
n(t)∈D(t)
lim
T→∞
1
T
T
t=1
λp(t)
subject to n = lim
T→∞
1
T
T
t=1
n(t) ≤ Nav
lim
T→∞
1
T
T
t=1
E(U(t)) ≤ ∞, (8)
In (7), we try to minimize the time average utilization
of n(t), the horizontal handover time and simultaneously
keep the data stable, that is, the constraints in (7). In (8),
λp(t) is the data flowing through F-APs networks at time t.
We try to maximize time average of the data flowing through
F-APs networks, which equally means that minimizing the
utilization of vertical handover. The first constraint in (8)
comes from the fact that the number of the utilized F-APs
cannot more than the existing number of F-APs.
F. LYAPUNOV OPTIMIZATION
In general, the most popular way to solve the problem
like (7) and (8) is to formulate by MDP. MDP can help
us to find the best tradeoff between transmission delay and
resource utilization [91]. Even though MDP can approach the
best tradeoff between resource utilization and delay, it may
take lots of time to find the optimal solution, especially in the
scenario that the number of the states are large [92].
In fact, there is always an intuitive way to solve this
type dynamic resource allocation problem, for example, prior
allocating the resource to the queue suffering severe delay.
However, there is always a tradeoff between resource utiliza-
tion and the network performance. Though more resources
suggest better network performance, it is difficult to intu-
itively conclude the tradeoff. Please recall that Lyapunov
optimization [93], originating from Lyapunov drift theory, is
used to develop dynamic control algorithms. It introduces the
drift-plus-penalty theorem concept into the control algorithm.
That is, it gives a cost weighting V to the network utility
and tries to optimize drift-plus-penalty function subject to the
queue stability. For example, we give the cost weighting to
the utilization of n(t) to maximize the service rate with the
horizontal handover. That is,
max
n(t)
E(2U(t)u(t) − Vn(t)). (9)
In the vertical handover, the problem can also be converted
into the similar form as following.
max
n(t)
E(U(t)n(t) − X(t)n(t))
min
λp(t)
E 2U(t)λp(t) − Vλp(t) . (10)
The results of the first equation give us the threshold X(t)/p of
the fully-utilizing APs. The second equation can be arranged
as E((2U(t)−V)λp(t)), therefore, all the arriving data should
be switched through vertical handover if U(t) > V/2 and
through the APs network if U(t) < V/2.
The advantage of Lyapunov optimization is that it provides
the delay upper bound of the system with O(V) and the uti-
lization of resource will reach the optimal utilization within
O(1/V) [93]. On the other hand, Lyapunov optimization also
introduces an interesting concept called virtual queue. The
concept virtual queue can convert the constraint problem, like
in (8), into the stability problems. It makes the Lyapunov opti-
mization be able to solve more general dynamic optimization
problems.
FIGURE 8. Illustration of the autonomous horizontal handover scheme
for the vehicular networks. The vehicles access all the available F-APs
only if the queuing delay is larger than the threshold V /2p, which comes
from the results of (9).
FIGURE 9. Illustration of the autonomous vertical handover scheme for
the vehicular networks. From the figure, we can find that we can set
different threshold via V to adjust the utilization of the vertical handover.
The Fig. 8 and Fig. 9 illustrates the resulting horizon-
tal and vertical handover schemes based on the solutions
of (9) and (10). Fig. 8 illustrates that the vehicles need to
access only one F-AP if the data queue is smaller than the
threshold V/2p. If the data queue exceeds the threshold V/2p,
which means that the queuing delay has been intolerable,
and the vehicles start to access all the available F-APs to
decrease the queuing delay. While considering the vertical
VOLUME 3, 2015 3029

handover into the design, there needs two thresholds in the
final handover scheme, as shown in Fig. 9. The first one
threshold is to determine whether to utilize all the F-APs
or not. The second one is determined by the virtual queue
X(t), which can be interpreted as the indicator whether the
constraint in the original optimization is satisfied or not. After
queue size is larger than the second threshold, the vehicles
just execute the vertical handover.
We consider the environment that the distribution density
of F-AP λap = 5×10−5/m2 corresponding to different vehi-
cle density λv. The velocity of the vehicles is 15m/s which is
about 55km/hr. The mean number of arrival packet in each
time slot t is E(λp) = 200 follows a Poisson distribution.
The length of each time slot is 0.01s. The total iteration are
1000 times, and the length of simulation time T for each
iteration is 25000 time slots.
FIGURE 10. The upper bound of delay of the proposed handover scheme.
If the density of vehicles λv is small enough the to satisfy the stability
condition, the delay can be successfully upper bounded without the help
of vertical handover.
Fig. 10 illustrates the mean delay corresponding to differ-
ent values of V without the help of the vertical handover.
The upper bounds in Fig. 10 is derived from the Lyapunov
optimization and can be converted to mean delay via Little’s
Theorem. In practical operation, we can adjust the value
of V to achieve the different delay requirements. Fig. 11
illustrates the probability of vertical handover corresponding
to different value V. From the figure, we can find that the
probability approaches to 0 while the value of V increases if
the density of the vehicles is small. If the density of the vehi-
cles increases, however, the probability of vertical handover
cannot decreases to 0 with the increment of V. This is the
reason that the density of F-APs is large enough to support all
the data flows from the vehicles. However, when the density
of the vehicles is large, there is no enough F-APs resources
to support the data flows from the vehicles. In such case, the
vertical handover scheme becomes necessary.
VIII. RESOURCE ACCESS CONTROL IN
H-CRAN AND FogNet
With limited spectrum resource, it is inevitable that
FogNet (fog-part) may be underlay or overlay under the
FIGURE 11. The probability of vertical handover of the proposed scheme.
The probability of vertical handover does not decrease as V increasing if
the value of λv is large. It is the reason there is no enough access
resource F-APs and the vertical handover becomes the necessary
alternative ways to service data flows.
H-CRAN (cloud-part). In such case, the successful coexis-
tence of fog-part and cloud-part may rely on the resource
access control. It raises a fundamental question ‘‘is the con-
trolling signals necessary to organize the resource access
between fog and cloud-part?’’. In this section, we show that,
in some situation, the FogNet relying on random access
control may outperform the fully centralized control like
H-CRAN.
A. RESOURCE UTILIZATION SCHEME IN H-CRAN
AND FogNet of F-CRAN
As aforementioned, H-CRAN is a fully centralized radio
access architecture. To reach the radio resource optimization,
H-CRAN can schedule all the radio resources to the users’
needs. On the contrast, FogNet has a distributed architecture
in which all the users may share a common pool of allocated
radio resources. For a long run, we consider the systems
with the centralized resource management or the distributed
scheme such as random access in the form of carrier sense
multiple access (CSMA) separately.
While we consider scheduling in H-CRAN, it needs to
allocate the amount of the necessary radio resources to the
users. In this way, all the users can individually utilize the
radio resources such that there is no interference among users
or statistically under an acceptable interference level. Nev-
ertheless, H-CRAN must allocate more radio resources than
the required amount to against deep fading in the channels.
This scheme is further necessary while the probability of
deep fading channels is large. Even though the performance
of individual user can be improved through allocating more
amount of radio resource than a user’s request, but the per-
formance of the entire network may degrade. Please recall
that there may be no centralized coordination in the FogNet
and all the users compete for the radio resources through
random access like CSMA. Though random access may result
in interference among users, but the fading effects in different
3030 VOLUME 3, 2015

radio resources are varying for different users at different
locations. Such a situation creates an effect similar diversity
communication. From the viewpoint of frequency reuse in
the network, FogNet may enjoy better spectrum efficiency
than the H-CRAN. Therefore, the existence of FogNet in the
H-CRAN invokes a fundamental question: when and how to
efficiently switch between H-CRAN and FogNet for a mobile
UE to get the best performance.
B. H-CRAN MODEL AND PERFORMANCE OF SCHEDULE
We consider the scenario that the UEs adopt a single
frequency band for uplink transmission, and radio resources
are allocated in the basic unit of a resource block (RB).
As each uplink transmission may involve a number of RBs,
these RBs used for one batch of resources in uplink trans-
mission is referred as one batch of resources. A UE requests
for one resource batch for uplink transmission, then a HPN
may allocate exactly one resource batch for this UE through
RRHs. Due to the mobility and distribution of UEs and RRHs
channel conditions among H-CRAN/FogNet are stochastic.
While the deep fading occurs, UEs cannot successfully trans-
mit data with this radio resource. The net throughput con-
tributed from this UE depends on the probability of deep
fading occurrence.
Considering that there are totally M resource blocks in the
frequency domain. We denote the probability of deep fading
as p. To guarantee the successful transmission, the most naive
approach is to allocate the m resource blocks to a UE. In this
way, the throughput of a UE is 1−pm. Obviously, even though
such approach increase the individual performance, but the
total network throughput is severely degraded. To discuss the
throughput of the whole system, we can define the entire
network throughput υ as
υ =
P(Successfully Transmission)
m
=
1 − pm
m
. (11)
It can be interpreted as the probability of successful
transmission per resource block.
C. FogNet MODEL AND PERFORMANCE
OF RANDOM ACCESS
FogNet is a group of UEs and may be without the coordi-
nation of the H-CRAN in the F-CRAN system. To further
enhance the throughput of such system, it prefers that more
users can simultaneously utilize the same resource blocks.
Due to the random location of UEs, different UEs may suffer
from different level of deep fading at the same resource batch.
Therefore, it gives the room for multiple UEs accessing the
same resource blocks.
Among the total M resource blocks, each UE performs
channel estimation at all the M resource blocks and selects
one without deep fading condition. Then these resource
blocks can be fully utilized to enhance the throughput of the
entire network. However, if some of UEs unfortunately select
the same unoccupied resource batch to transmit data, then a
collision occurs and the throughput of UEs degrades, which
is the issue in the FogNet adopting random access.
Grouping is an effective approach to alleviate above
dilemma [94] and later adopted in LTE. While grouping UEs
to form a FogNet, it is desirable to identify a proper size
for a group. A large group can introduce severe competition
for resource blocks, but a small group may result in low
utilization efficiency of the network. It is necessary to find the
number of UEs that can achieve the best system performance
given the limited resource batches and deep fading proba-
bility. Considering that there are totally M resource batches
indexed by m = 1 . . . M to be shared by N UEs. Before trans-
mitting, the UEs will sense all M resource batches without
deep fading and choose one to transmit data. Let
Im,i =
1, ith UE selects the mth resource batch,
0, otherwise,
(12)
be an indication function. Then the probability that ith UE
utilizes mth resource block can be expressed as
q P(Im,i = 1) =
1 − pM
M
. (13)
Then the throughput of the mth resource block is
υ = Nq(1 − q)N−1
. (14)
To find the most suitable number of UEs N∗ in the FogNet,
we need to maximize the equation above.
N∗
= arg max
N
Nq(1 − q)N−1
(15)
We can substitute N∗ into (14) and get the best performance
of the FogNet.
D. SWITCH POINT BETWEEN FogNet AND H-CRAN
The centralized scheduling based approach is regarded as an
effective scheme to enhance the throughput of the network.
However, the performance of the throughput of the entire
network may degrade if the deep fading or other interfer-
ence (from other underlay network) is severe. If we allow
all the UEs compete the resource randomly, a particular
resource block that may be under deep fading for one UE but
has a good channel condition for another UE in a different
location. If this happens, the utilization efficiency of the
resource blocks can be increased without interfering other
UEs. This performance enhancement is boosted by statistical
multiplexing of multiple UEs’ channel access. According to
this argument, it is necessary to find the switch point, which
depends on the value of p, between the H-CRAN and FogNet.
In (11), we can find that the entire system reaches the
largest throughput while m = 1, that is, each UE can be allo-
cated with one resource block. Therefore, the performance of
the H-CRAN can be expressed as 1−p. Then we can compare
the performance with the appropriate grouping number N∗
with the best performance of the FogNet.
The performance of network switching point between cen-
tralized control and random access is illustrated in Fig. 12.
VOLUME 3, 2015 3031

FIGURE 12. The throughput of the FogNet with scheduling based radio
access, a pool of resource shared scheme and the optimal transition
scheme.
We consider the environment with 20MHz bandwidth with
m = 100 RBs in the frequency domain. The length of each
RB is regarded as 4ms. The transmission power of each UE is
20dBm. The scheduling based radio access provides a better
system throughput υ at p = 1 − (1 − 1
m )m−1 is about 0.63,
which confirms our arguments.
IX. CONCLUSION
In recent years, H-CRAN and FogNet were proposed to tackle
the Internet contents, vehicular network and large amount of
devices. The former one focuses on the centralized control to
optimize the whole network and reach the best resource uti-
lization. The latter one takes advantage of the characteristics
of Internet contents to simplify the architecture of the network
in a decentralized way. For the 5G wireless system, the big
question may not to select from these two network archi-
tectures, but the way for devices to properly select between
H-CRAN and FogNet for radio access, which has been over-
looked in the literatures. Appropriate coordination between
these two network architectures to fully utilize the advantages
from each architecture starts with this paper, but definitely
not ends here. There are quite a few works discussing about
H-CRAN and FogNet separately but little attention has been
paid about the coordination of them. Further coordination
between these two network architectures to complement each
other and to fully utilize network and radio resources remains
a subject worth further pursuing. In this research, we illustrate
the conditions and mechanism to switch between H-CRAN
and FogNet from the viewpoint of cache in wireless network,
mobility management and access control, and thus pave a
new avenue to various research directions. Future research
opportunities may include control and signaling between two
network architectures, integration with short-range commu-
nication like millimeter-wave (mmWave), licensed assisted
access (LAA) and energy harvesting (EH) devices into
the H-CRAN. Due to the limited transmission power and
computation ability, these devices may hard to access the
H-CRAN and the only way is toward the FogNet. On the other
hand, to tackle the dynamic environment, sensing as cognitive
radios to coordinate the utilization of radio resources [95]
may be a good candidate to further improve the FogNet
without interfering H-CRAN system. Last but not the least,
this harmonization suggests a good balance to standardize
state-of-the-art mobile communications and more detailed
design and precise analysis remains very much wanted.
REFERENCES
[1] Y. S. Soh, T. Q. S. Quek, M. Kountouris, and H. Shin, ‘‘Energy efficient
heterogeneous cellular networks,’’ IEEE J. Sel. Areas Commun., vol. 31,
no. 5, pp. 840–850, May 2013.
[2] Q. Li, R. Q. Hu, Y. Qian, and G. Wu, ‘‘Intracell cooperation and resource
allocation in a heterogeneous network with relays,’’ IEEE Trans. Veh.
Technol., vol. 62, no. 4, pp. 1770–1784, May 2013.
[3] A. Damnjanovic et al., ‘‘A survey on 3GPP heterogeneous networks,’’
IEEE Wireless Commun., vol. 18, no. 3, pp. 10–21, Jun. 2011.
[4] Y. L. Lee, T. C. Chuah, J. Loo, and A. Vinel, ‘‘Recent advances in radio
resource management for heterogeneous LTE/LTE-A networks,’’ IEEE
Commun. Surveys Tuts., vol. 16, no. 4, pp. 2142–2180, Fourthquarter 2014.
[5] M. Wildemeersch, T. Q. S. Quek, M. Kountouris, A. Rabbachin, and
C. H. Slump, ‘‘Successive interference cancellation in heterogeneous net-
works,’’ IEEE Trans. Commun., vol. 62, no. 12, pp. 4440–4453, Dec. 2014.
[6] S.-Y. Lien, S.-C. Hung, K.-C. Chen, and Y.-C. Liang, ‘‘Ultra-low-latency
ubiquitous connections in heterogeneous cloud radio access networks,’’
[7] J. M. Chapin and V. W. S. Chan, ‘‘Architecture concepts for a future het-
erogeneous, survivable tactical Internet,’’ in Proc. IEEE Military Commun.
Conf., Nov. 2013, pp. 1874–1879.
[8] K.-C. Chen and S.-Y. Lien, ‘‘Machine-to-machine communications: Tech-
nologies and challenges,’’ Ad Hoc Netw., vol. 18, pp. 3–23, Jul. 2014.
[9] S.-Y. Lien, K.-C. Chen, and Y. Lin, ‘‘Toward ubiquitous massive accesses
in 3GPP machine-to-machine communications,’’ IEEE Commun. Mag.,
vol. 49, no. 4, pp. 66–74, Apr. 2011.
[10] A. Rajandekar and B. Sikdar, ‘‘A survey of MAC layer issues and protocols
for machine-to-machine communications,’’ IEEE Internet Things J., vol. 2,
no. 2, pp. 175–186, Apr. 2015.
[11] G. Dimitrakopoulos and P. Demestichas, ‘‘Intelligent transportation sys-
tems,’’ IEEE Veh. Technol. Mag., vol. 5, no. 1, pp. 77–84, Mar. 2010.
[12] C. Chakrabarti and S. Roy, ‘‘Adapting mobility of observers for quick
reputation assignment in a sparse post-disaster communication network,’’
in Proc. Appl. Innov. Mobile Comput. (AIMoC), Feb. 2015, pp. 29–35.
[13] R. Pozza, M. Nati, S. Georgoulas, K. Moessner, and A. Gluhak, ‘‘Neighbor
discovery for opportunistic networking in internet of things scenarios: A
survey,’’ IEEE Access, vol. 3, pp. 1101–1131, Jul. 2015.
[14] C. Perera, A. Zaslavsky, P. Christen, and D. Georgakopoulos, ‘‘Context
aware computing for the Internet of Things: A survey,’’ IEEE Commun.
Surveys Tuts., vol. 16, no. 1, pp. 414–454, Firstquarter 2014.
[15] M. R. Palattella et al., ‘‘Standardized protocol stack for the
Internet of (important) Things,’’ IEEE Commun. Surveys Tuts., vol. 15,
no. 3, pp. 1389–1406, Thirdquareter 2013.
[16] Z. M. Fadlullah, M. M. Fouda, N. Kato, A. Takeuchi, N. Iwasaki, and
Y. Nozaki, ‘‘Toward intelligent machine-to-machine communications in
smart grid,’’ IEEE Commun. Mag., vol. 49, no. 4, pp. 60–65, Apr. 2011.
[17] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi,
‘‘Internet of Things for smart cities,’’ IEEE Internet Things J., vol. 1, no. 1,
pp. 22–32, Feb. 2014.
[18] Erisson, ‘‘Erisson mobility report,’’ Erisson, Stockholm, Sweden, Tech.
Rep., Jun. 2015.
[19] K.-C. Chen, M. Chiang, and H. V. Poor, ‘‘From technological networks to
social networks,’’ IEEE J. Sel. Areas Commun., vol. 31, no. 9, pp. 548–572,
Sep. 2013.
[20] V. Chan et al., ‘‘Future heterogeneous networks,’’ Nat. Sci. Found.,
Arlington, VA, USA, Tech. Rep., 2011.
[21] M. Dohler, R. W. Heath, A. Lozano, C. B. Papadias, and R. B. Valenzuela,
‘‘Is the PHY layer dead?’’ IEEE Commun. Mag., vol. 49, no. 4,
pp. 159–165, Apr. 2011.
[22] N. Bhushan et al., ‘‘Network densification: The dominant theme for wire-
less evolution into 5G,’’ IEEE Commun. Mag., vol. 52, no. 2, pp. 82–89,
Feb. 2014.
3032 VOLUME 3, 2015

[23] Y. Lin, L. Shao, Z. Zhu, Q. Wang, and R. K. Sabhikhi, ‘‘Wireless network
cloud: Architecture and system requirements,’’ IBM J. Res. Develop.,
vol. 54, no. 1, pp. 4:1–4:12, Jan./Feb. 2010.
[24] C. Mobile, C-RAN: The Road Towards Green RAN, China Mobile Res.
Inst., Beijing, China, 2011.
[25] A. Samukic, ‘‘UMTS universal mobile telecommunications system: Devel-
opment of standards for the third generation,’’ IEEE Trans. Veh. Commun.,
vol. 47, no. 4, pp. 1099–1104, Nov. 1998.
[26] Q. Cui et al., ‘‘Evolution of limited-feedback CoMP systems from 4G to
5G: CoMP features and limited-feedback approaches,’’ IEEE Veh. Technol.
Mag., vol. 9, no. 3, pp. 94–103, Sep. 2014.
[27] S. Sun, Q. Gao, Y. Peng, Y. Wang, and L. Song, ‘‘Interference manage-
ment through CoMP in 3GPP LTE-advanced networks,’’ IEEE Wireless
Commun., vol. 20, no. 1, pp. 59–66, Feb. 2013.
[28] O. Onireti, F. Heliot, and M. A. Imran, ‘‘On the energy efficiency-spectral
efficiency trade-off in the uplink of CoMP system,’’ IEEE Trans. Wireless
Commun., vol. 11, no. 2, pp. 556–561, Feb. 2012.
[29] P. Marsch and G. P. Fettweis, Eds., Coordinated Multi-Point in Mobile
Communications: From Theory to Practice. Cambridge, U.K.: Cambridge
Univ. Press, 2011.
[30] J. Lee et al., ‘‘Coordinated multipoint transmission and reception in LTE-
advanced systems,’’ IEEE Commun. Mag., vol. 50, no. 11, pp. 44–50,
Nov. 2012.
[31] M. Peng, S. Yan, and H. V. Poor, ‘‘Ergodic capacity analysis of remote
radio head associations in cloud radio access networks,’’ IEEE Wireless
Commun. Lett., vol. 3, no. 4, pp. 365–368, Aug. 2014.
[32] F. A. Khan, H. He, J. Xue, and T. Ratnarajah, ‘‘Performance analysis
of cloud radio access networks with distributed multiple antenna remote
radio heads,’’ IEEE Trans. Signal Process., vol. 63, no. 18, pp. 4784–4799,
Sep. 2015.
[33] S.-N. Hong and J. Kim, ‘‘Joint coding and stochastic data transmission for
uplink cloud radio access networks,’’ IEEE Commun. Lett., vol. 18, no. 9,
pp. 1619–1622, Sep. 2014.
[34] S.-H. Park, O. Simeone, O. Sahin, and S. Shamai (Shitz), ‘‘Fronthaul
compression for cloud radio access networks: Signal processing advances
inspired by network information theory,’’ IEEE Signal Process. Mag.,
vol. 31, no. 6, pp. 69–79, Nov. 2014.
[35] M. Peng, K. Zhang, J. Jiang, J. Wang, and W. Wang, ‘‘Energy-
efficient resource assignment and power allocation in heterogeneous cloud
radio access networks,’’ IEEE Trans. Veh. Technol., vol. 64, no. 11,
pp. 5275–5287, Nov. 2014.
[36] L. Lei, Z. Zhong, K. Zheng, J. Chen, and H. Meng, ‘‘Challenges on wireless
heterogeneous networks for mobile cloud computing,’’ IEEE Wireless
Commun., vol. 20, no. 3, pp. 34–44, Jun. 2013.
[37] V. Suryaprakash, P. Rost, and G. Fettweis, ‘‘Are heterogeneous cloud-
based radio access networks cost effective?’’ IEEE J. Sel. Areas Commun.,
vol. 33, no. 10, pp. 2239–2251, Oct. 2015.
[38] D. C. Chen, T. Q. S. Quek, and M. Kountouris, ‘‘Wireless backhaul in
small cell networks: Modelling and analysis,’’ in Proc. IEEE Veh. Technol.
Conf. (VTC), May 2014, pp. 1–6.
[39] S.-C. Lin and K.-C. Chen, ‘‘Improving spectrum efficiency via in-network
computations in cognitive radio sensor networks,’’ IEEE Trans. Wireless
Commun., vol. 13, no. 3, pp. 1222–1234, Mar. 2014.
[40] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, ‘‘Fog computing and its
role in the Internet of Things,’’ in Proc. 1st Ed. MCC Workshop Mobile
Cloud Comput. (MCC), 2012, pp. 13–16.
[41] L. M. Vaquero and L. Rodero-Merino, ‘‘Finding your way in the fog:
Towards a comprehensive definition of fog computing,’’ SIGCOMM
Comput. Commun. Rev., vol. 44, no. 5, pp. 27–32, Oct. 2014.
[42] E. Aryafar, A. Keshavarz-Haddad, M. Wang, and M. Chiang, ‘‘RAT
selection games in HetNets,’’ in Proc. IEEE INFOCOM, Apr. 2013,
pp. 998–1006.
[43] G. Fodor, S. Parkvall, S. Sorrentino, P. Wallentin, Q. Lu, and N. Brahmi,
‘‘Device-to-device communications for national security and public
safety,’’ IEEE Access, vol. 2, pp. 1510–1520, Dec. 2014.
[44] Y. Li, D. Jin, J. Yuan, and Z. Han, ‘‘Coalitional games for resource
allocation in the device-to-device uplink underlaying cellular networks,’’
IEEE Trans. Wireless Commun., vol. 13, no. 7, pp. 3965–3977, Jul. 2014.
[45] L. Lei, Z. Zhong, C. Lin, and X. Shen, ‘‘Operator controlled device-
to-device communications in LTE-advanced networks,’’ IEEE Wireless
Commun., vol. 19, no. 3, pp. 96–104, Jun. 2012.
[46] Q. Ye, M. Al-Shalash, C. Caramanis, and J. G. Andrews, ‘‘Distributed
resource allocation in device-to-device enhanced cellular networks,’’ IEEE
Trans. Commun., vol. 63, no. 2, pp. 441–454, Feb. 2015.
[47] F.-M. T. Shao-Yu Lien, C.-C. Chien, and T.-C. Ho, ‘‘3GPP device-to-device
communications for beyond 4G cellular networks,’’ IEEE Commun. Mag.,
2015
[48] J. Oueis, E. C. Strinati, and S. Barbarossa, ‘‘The fog balancing: Load
distribution for small cell cloud computing,’’ in Proc. Veh. Technol. Conf.,
May 2015, pp. 1–6.
[49] J. Qiao, X. Shen, J. Mark, Q. Shen, Y. He, and L. Lei, ‘‘Enabling device-to-
device communications in millimeter-wave 5G cellular networks,’’ IEEE
Commun. Mag., vol. 53, no. 1, pp. 209–215, Jan. 2015.
[50] J. Andrews, S. Singh, Q. Ye, X. Lin, and H. Dhillon, ‘‘An overview of
load balancing in HetNets: Old myths and open problems,’’ IEEE Wireless.
Commun., vol. 21, no. 2, pp. 18–25, Apr. 2014.
[51] J. Zhao, T. Q. S. Quek, and Z. Lei, ‘‘Heterogeneous cellular networks using
wireless backhaul: Fast admission control and large system analysis,’’
IEEE J. Sel. Areas Commun., vol. 33, no. 10, pp. 2128–2143, Oct. 2015.
[52] M. Peng, S. Yan, K. Zhang, and C. Wang. (2015). ‘‘Fog computing
based radio access networks: Issues and challenges.’’ [Online]. Available:
http://arxiv.org/abs/1506.04233
[53] E. Bastug, M. Bennis, and M. Debbah, ‘‘Social and spatial proac-
tive caching for mobile data offloading,’’ in Proc. IEEE Int. Conf.
Commun. (ICC), Jun. 2014, pp. 581–586.
[54] X. Wang, M. Chen, T. Taleb, A. Ksentini, and V. Leung, ‘‘Cache in the air:
Exploiting content caching and delivery techniques for 5G systems,’’ IEEE
Commun. Mag., vol. 52, no. 2, pp. 131–139, Feb. 2014.
[55] J. L. Hennessy and D. A. Patterson, Computer Architecture: A Quantitative
Approach. Amsterdam, The Netherlands: Elsevier, 2011.
[56] A. Balamash and M. Krunz, ‘‘An overview of Web caching replacement
algorithms,’’ IEEE Commun. Surveys Tuts., vol. 6, no. 2, pp. 44–56,
Secondquarter 2004.
[57] J. Wang, ‘‘A survey of Web caching schemes for the Internet,’’ SIGCOMM
Comput. Commun. Rev., vol. 29, no. 5, pp. 36–46, Oct. 1999.
[58] M. Rabinovich and O. Spatscheck, Web Caching and Replication. Boston,
MA, USA: Addison-Wesley, 2002.
[59] L. Breslau, P. Cao, L. Fan, G. Phillips, and S. Shenker, ‘‘Web caching
and Zipf-like distributions: Evidence and implications,’’ in Proc. IEEE
INFOCOM, vol. 1. Mar. 1999, pp. 126–134.
[60] Global Mobile Data Traffic Forecast Update 2014–2019 White Paper,
Cisco Syst., Inc., San Jose, CA, USA, 2015.
[61] U. Niesen, D. Shah, and G. W. Wornell, ‘‘Caching in wireless networks,’’
IEEE Trans. Inf. Theory, vol. 58, no. 10, pp. 6524–6540, Oct. 2012.
[62] M. A. Maddah-Ali and U. Niesen, ‘‘Fundamental limits of caching,’’ IEEE
Trans. Inf. Theory, vol. 60, no. 5, pp. 2856–2867, May 2014.
[63] Z. Ming, M. Xu, and D. Wang, ‘‘InCan: In-network cache assisted
eNodeB caching mechanism in 4G LTE networks,’’ Comput. Netw., vol. 75,
pp. 367–380, Dec. 2014.
[64] E. Bastug, M. Bennis, and M. Debbah, ‘‘Living on the edge: The role of
proactive caching in 5G wireless networks,’’ IEEE Commun. Mag., vol. 52,
no. 8, pp. 82–89, Aug. 2014.
[65] H. Hsu and K.-C. Chen, ‘‘A resource allocation perspective on caching to
achieve low latency,’’ IEEE Commun. Lett., to be published.
[66] H. Chen and Y. Xiao, ‘‘Cache access and replacement for future wireless
Internet,’’ IEEE Commun. Mag., vol. 44, no. 5, pp. 113–123, May 2006.
[67] Y.-D. Lin and Y.-C. Hsu, ‘‘Multihop cellular: A new architecture for
wireless communications,’’ in Proc. 19th Annu. Joint Conf. IEEE Comput.
Commun. Soc. (INFOCOM), vol. 3. Mar. 2000, pp. 1273–1282.
[68] H. Wu, C. Qiao, S. De, and O. Tonguz, ‘‘Integrated cellular and ad hoc
relaying systems: ICAR,’’ IEEE J. Sel. Areas Commun., vol. 19, no. 10,
pp. 2105–2115, Oct. 2001.
[69] M. E. J. Newman. (2002). ‘‘Random graphs as models of networks.’’
[Online]. Available: http://arxiv.org/abs/cond-mat/0202208
[70] P. Sobkowicz, M. Thelwall, K. Buckley, G. Paltoglou, and A. Sobkowicz,
‘‘Lognormal distributions of user post lengths in Internet discussions—
A consequence of the Weber–Fechner law?’’ EPJ Data Sci., vol. 2, no. 1,
p. 2, 2013.
[71] M. G. Khoshkholgh, Y. Zhang, K.-C. Chen, K. G. Shin, and S. Gjessing,
‘‘Connectivity of cognitive device-to-device communications underlying
cellular networks,’’ IEEE J. Sel. Areas Commun., vol. 33, no. 1, pp. 81–99,
Jan. 2015.
[72] D. G. Luenberger and Y. Ye, Introduction to Linear and Nonlinear Pro-
gramming, 3rd ed. Reading, MA, USA: Addison-Wesley, 1973.
[73] 5GPPP. (Oct. 2015). 5G Automotive Vision. [Online]. Available:
https://5g-ppp.eu/wp-content/uploads/2014/02/5G-PPP-White-Paper-
on-Automotive-Vertical-Sectors.pdf
VOLUME 3, 2015 3033

[74] ‘‘Connected car market—Global industry analysis, size, share, growth,
trends and forecast, 2013–2019,’’ Transparency Market Res., New York,
NY, USA, Tech. Rep., 2013.
[75] N. Lu, N. Cheng, N. Zhang, X. Shen, and J. W. Mark, ‘‘Connected
vehicles: Solutions and challenges,’’ IEEE Internet Things J., vol. 1, no. 4,
pp. 289–299, Aug. 2014.
[76] Standard Specification for Telecommunications and Information
Exchange Between Roadside and Vehicle Systems—5 GHz
Band Dedicated Short Range Communications (DSRC) Medium
Access Control (MAC) and Physical Layer (PHY) Specifications,
ASTM Standard E2213-02, 2003.
[77] Amendment 6: Wireless Access in Vehicular Environment,
IEEE Standard 802.11p, Jul. 2010.
[78] G. Araniti, C. Campolo, M. Condoluci, A. Iera, and A. Molinaro, ‘‘LTE
for vehicular networking: A survey,’’ IEEE Commun. Mag., vol. 51, no. 5,
pp. 148–157, May 2013.
[79] K. Zheng, Q. Zheng, P. Chatzimisios, W. Xiang, and Y. Zhou, ‘‘Het-
erogeneous vehicular networking: A survey on architecture, challenges
and solutions,’’ IEEE Commun. Surveys Tuts., vol. PP, no. 99, pp. 1–1,
Fourthquarter 2015.
[80] H. Zhang, C. Jiang, and J. Cheng, ‘‘Cooperative interference mitigation
and handover management for heterogeneous cloud small cell networks,’’
[81] M. Peng, Y. Li, J. Jiang, J. Li, and C. Wang, ‘‘Heterogeneous cloud radio
access networks: A new perspective for enhancing spectral and energy effi-
ciencies,’’ IEEE Wireless Commun., vol. 21, no. 6, pp. 126–135, Dec. 2014.
[82] T. Ali and M. Saquib, ‘‘Performance evaluation of WLAN/cellular media
access for mobile voice users under random mobility models,’’ IEEE Trans.
Wireless Commun., vol. 10, no. 10, pp. 3241–3255, Oct. 2011.
[83] N. W. Sung, N.-T. Pham, T. Huynh, and W.-J. Hwang, ‘‘Predictive associa-
tion control for frequent handover avoidance in femtocell networks,’’ IEEE
Commun. Lett., vol. 17, no. 5, pp. 924–927, May 2013.
[84] J.-M. Moon and D.-H. Cho, ‘‘Efficient handoff algorithm for inbound
mobility in hierarchical macro/femto cell networks,’’ IEEE Commun. Lett.,
vol. 13, no. 10, pp. 755–757, Oct. 2009.
[85] D. Xenakis, N. Passas, L. Merakos, and C. Verikoukis, ‘‘Mobility man-
agement for femtocells in LTE-advanced: Key aspects and survey of han-
dover decision algorithms,’’ IEEE Commun. Surveys Tuts., vol. 16, no. 1,
pp. 64–91, Firstquarter 2014.
[86] H. Kalbkhani, S. Yousefi, and M. G. Shayesteh, ‘‘Adaptive handover
algorithm in heterogeneous femtocellular networks based on received
signal strength and signal-to-interference-plus-noise ratio prediction,’’ IET
Commun., vol. 8, no. 17, pp. 3061–3071, 2014.
[87] R. Irmer et al., ‘‘Coordinated multipoint: Concepts, performance, and
field trial results,’’ IEEE Commun. Mag., vol. 49, no. 2, pp. 102–111,
Feb. 2011.
[88] E. Stevens-Navarro, Y. Lin, and V. W. S. Wong, ‘‘An MDP-based vertical
handoff decision algorithm for heterogeneous wireless networks,’’ IEEE
Trans. Veh. Technol., vol. 57, no. 2, pp. 1243–1254, Mar. 2008.
[89] L. Qin and D. Zhao, ‘‘Channel time allocations and handoff management
for fair throughput in wireless mesh networks,’’ IEEE Trans. Veh. Technol.,
vol. 64, no. 1, pp. 315–326, Jan. 2015.
[90] M. Haenggi, J. G. Andrews, F. Baccelli, O. Dousse, and M. Franceschetti,
‘‘Stochastic geometry and random graphs for the analysis and design
of wireless networks,’’ IEEE J. Sel. Areas Commun., vol. 27, no. 7,
pp. 1029–1046, Sep. 2009.
[91] K. T. Phan, T. Le-Ngoc, M. van der Schaar, and F. Fu, ‘‘Optimal schedul-
ing over time-varying channels with traffic admission control: Structural
results and online learning algorithms,’’ IEEE Trans. Wireless Commun.,
vol. 12, no. 9, pp. 4434–4444, Sep. 2013.
[92] Y. Cui, V. K. N. Lau, R. Wang, H. Huang, and S. Zhang, ‘‘A
survey on delay-aware resource control for wireless systems—Large
deviation theory, stochastic Lyapunov drift, and distributed stochastic
learning,’’ IEEE Trans. Inf. Theory, vol. 58, no. 3, pp. 1677–1701,
Mar. 2012.
[93] L. Georgiadis, M. J. Neely, and L. Tassiulas, ‘‘Resource allocation and
cross-layer control in wireless networks,’’ Found. Trends Netw., vol. 1,
no. 1, pp. 1–144, 2006.
[94] K.-C. Chen, ‘‘Medium access control of wireless LANs for mobile com-
puting,’’ IEEE Netw., vol. 8, no. 5, pp. 50–63, Sep. 1994.
[95] S.-Y. Lien, K.-C. Chen, Y.-C. Liang, and Y. Lin, ‘‘Cognitive radio resource
management for future cellular networks,’’ IEEE Wireless Commun.,
vol. 21, no. 1, pp. 70–79, Feb. 2014.
SHAO-CHOU HUNG received the B.S. and
M.S. degrees in electrical engineering from
National Taiwan University, in 2010 and 2013,
respectively, where he is currently pursuing the
Ph.D. degree with the Graduate Institute of Com-
munication Engineering. His research interests
include 5G network architecture, cognitive radio
networks, and dynamic optimal control in wireless
network.
HSIANG HSU received the B.S. degrees in
electrical engineering and mathematics from
National Taiwan University, in 2014, where
he is currently pursuing the M.S. degree
with the Graduate Institute of Communication
Engineering. His research interests include
resource allocation, optimization, and machine
learning in complex networks.
SHAO-YU LIEN is currently an Assistant
Professor with the Department of Electronic
Engineering, National Formosa University,
Taiwan. Recently, his focuses are particularly on
cyber-physical systems and 5G communication
networks. His research interests include optimiza-
tion techniques for networks and communication
systems. He received a number of prestigious
recognitions, including the IEEE Communications
Society Asia-Pacific Outstanding Paper Award
in 2014, the Scopus Young Researcher Award (issued by Elsevier) in 2014,
the URSI AP-RASC 2013 Young Scientist Award, and the IEEE ICC 2010
Best Paper Award.
KWANG-CHENG CHEN (M’89–SM’94–F’07)
received the B.S. degree from the National Tai-
wan University in 1983, and the M.S. and Ph.D.
degrees from the University of Maryland, College
Park, USA, in 1987 and 1989, all in electrical
engineering. From 1987 to 1998, he worked with
SSE, COMSAT, IBM Thomas J. Watson Research
Center, and National Tsing Hua University, work-
ing on the mobile communications and networks.
Since 1998, he has been with the National Taiwan
University, Taipei, Taiwan. After serving as the Director, Graduate Institute
of Communication Engineering, Communication Research Center, and the
Associate Dean for Academic Affairs, he is currently a Distinguished Pro-
fessor with National Taiwan University and is visiting the Massachusetts
Institute of Technology from 2015 to 2016. He has been actively involving
in the organization of various IEEE conferences as General/TPC Chair/Co-
Chair, and has served in editorships with a few IEEE journals. He also
actively participates in and has contributed essential technology to various
IEEE 802, Bluetooth, and LTE and LTE-A wireless standards. He is an
IEEE Fellow and has received a number of awards, such as the 2011 IEEE
COMSOC WTC Recognition Award, 2014 IEEE Jack Neubauer Memorial
Award, and 2014 IEEE COMSOC AP Outstanding Paper Award. His recent
research interests include wireless communications, network science, and
data science.
3034 VOLUME 3, 2015

Architecture harmonization between cloud radio access network and fog network

Recommended

More Related Content

What's hot (18)

Similar to Architecture harmonization between cloud radio access network and fog network (20)

More from redpel dot com (20)

Recently uploaded (20)

Architecture harmonization between cloud radio access network and fog network