arX

iv:1

709.

0501

7v1

[cs

.IT

] 1

5 Se

p 20

17IEEE ACCESS, VOL. 2016, SI ON RECENT ADVANCES IN SOCIALLY-AWARE MOBILE NETWORKING 1

Recent Advances in Fog Radio Access Networks:

Performance Analysis and Radio Resource

AllocationMugen Peng, Senior Member, IEEE and Kecheng Zhang

Abstract—As a promising paradigm for the fifth generationwireless communication (5G) system, the fog radio access network(F-RAN) has been proposed as an advanced socially-aware mobilenetworking architecture to provide high spectral efficiency (SE)while maintaining high energy efficiency (EE) and low latency.Recent advents are advocated to the performance analysis andradio resource allocation, both of which are fundamental issues tomake F-RANs successfully rollout. This article comprehensivelysummarizes the recent advances of the performance analysis andradio resource allocation in F-RANs. Particularly, the advancededge cache and adaptive model selection schemes are presentedto improve SE and EE under maintaining a low latency level.The radio resource allocation strategies to optimize SE and EEin F-RANs are respectively proposed. A few open issues in termsof the F-RAN based 5G architecture and the social-awarenesstechnique are identified as well.

Index Terms—Fog radio access networks (F-RANs),socially-aware mobile networking, edge cache, 5G, radio

resource allocation

I. INTRODUCTION

As the mobile traffic explosively grows with multitude of

mobile social applications, the socially-aware mobile networks

have emerged as a promising direction to exploit the peoples

behaviors and interactions in the social domain [1]. Socially-

aware mobile network designs can improve shared spectrum

access, cooperative spectrum sensing and device-to-device

(D2D) communications, and have potential to achieve sub-

stantial gains in spectral efficiency (SE), which can meet some

performance requirements of the fifth generation wireless com-

munication (5G) systems [2]. Among all presented socially-

aware mobile networking architectures, the cloud radio access

networks (C-RANs) can be regarded as a promising paradigm

for improving SE and energy efficiency (EE) for 5G [3]. In

C-RANs, the baseband processing is centralized in the base

band unit (BBU) pool and the densely deployed remote radio

heads (RRHs) connected to BBU pool via fronthaul fulfill

the user-centric capability. However, the constrained fronthaul

with limited capacity and long time delay degrades SE and

EE performances. Meanwhile, the full-centralized architecture

brings heavy burdens on the computing capability in the BBU

pool [4].

Taking full advantage of fog computing and C-RANs,

fog radio access networks (F-RANs) have been proposed to

Mugen Peng (e-mail: pmg@bupt.edu.cn) and Kecheng Zhang (e-mail:buptzkc@163.com) are with the Key Laboratory of Universal WirelessCommunications for Ministry of Education, Beijing University of Posts andTelecommunications (BUPT), China.

tackle these aforementioned disadvantages of C-RANs as an

advanced socially-aware mobile networking architecture in 5G

systems [5]. In F-RANs, the RRH and the user equipment

(UE) are capable of local signal processing, cooperative radio

resource management, and distributed storing. Such RRHs and

UEs are named the edge computing access point (EC-AP) and

the edge computing user equipment (EC-UE), respectively.

Since a part of UEs no longer access to the BBU pool,

the heavy burdens on both fronthaul and BBU pool are

alleviated, and the latency is reduced significantly as well.

Therefore, F-RANs can achieve high SE/EE, low latency,

and fantastic reliability for different IoT applications such as

mobile vehicular connectivity, smart home, smart education,

wearable healthcare devices, and industrial automation [5].

Recently, the functionalities and technologies of F-RANs

have been discussed in the 3rd generation partnership project

(3GPP) standard, such as the edge cache and the system

architecture requirements for fulfilling 5G systems [6]. Mean-

while, many works have been done for the IoT in the F-

RAN based 5G systems. The system architecture and key

techniques of F-RANs have been proposed in [5]. Following,

the harmonization of C-RANs and F-RANs from various

points of view has been compared in [7]. The joint compu-

tational and radio resources problems in F-RANs have been

explored in [8]. Furthermore, effective caching strategies for

F-RANs is given by [9], while the content caching and delivery

techniques for 5G systems are discussed in [10] and [11]. All

these publications have illustrated that the cache incorporation

can significantly improve SE, EE and delay performances.

However, the outage probability and ergodic capacity analysis

in F-RANs are still not straightforward, which constrain the

development of F-RANs. Besides, the distributed edge caching

can alleviate the burdens on the fronthaul and BBU pool,

meanwhile decreases latency through shortening the end-to-

end communication distance, which results in the traditional

radio resource strategies unsuitable for F-RANs. Since the

performance analysis and radio resource allocation are two

key issues for fulfilling the requirements of IoT applications

in the F-RAN based 5G systems, this article presents a

framework for these two issues, in which the recent advances

are comprehensively summarized. Meanwhile, the challenges

and open issues for the IoT in F-RANs are discussed as well.

This article is organized as follows. We discuss the edge

cache based performance analysis and radio resource alloca-

tion in Section II and Section III, respectively. Open issues

and challenges are discussed in Section IV. The conclusion is

IEEE ACCESS, VOL. 2016, SI ON RECENT ADVANCES IN SOCIALLY-AWARE MOBILE NETWORKING 2

D2D Mode

Nearest

EC-AP Mode

Global C-RAN Mode

Local distributed

coordination Mode

Centralized Cloud

BBU Pool

Centralized Cloud

Controller

Cloud Content Cache

Cloud Center

Fronthaul

Backhaul

X2/S1

EC-AP

EC-AP

RRH RRH

RRH

HPN

EC-UE

EC-UE

Fig. 1. System architecture of the F-RAN

presented in Section V.

II. CACHE-BASED PERFORMANCE ANALYSIS

As shown in Fig. 1, to fulfill the functions of socially-aware

mobile networking, the hierarchical architecture consists of

cloud center and fog layer in F-RANs. The cloud center is

composed of cloud content cache, cloud BBU pool, and cloud

controller. RRHs connect to the BBU pool via fronthaul, while

the high power node (HPN) connects to the content cloud

through backhaul. EC-AP integrates not only the front radio

frequency (RF), but also the local distributed collaboration

radio signal processing (CRSP), cooperative radio resource

management (CRRM) and caching capabilities. By processing

collaboratively among multiple adjacent EC-APs and even

through the device-to-device (D2D) communication, the over-

load of the fronthaul links can be decreased, and the queuing

and transmitting latency can be alleviated. When all CRSP

and CRRM functions are shifted to the BBU pool, EC-AP

is degenerated to a traditional RRH. EC-UEs denote by UEs

accessing EC-APs and working in the D2D mode [5].

A. Edge caching impacting on the performance of F-RAN

Collaborative strategy to implement caching in infrastruc-

ture and in mobile devices simultaneously is researched in

[11]. With respect to sizes, the utilization is different for

caching in EC-APs and caching in devices, as shown in Fig.

2 (a). For the popular social media (high rank), the best

strategy is to cache it in the EC-AP. Due to the instability

of opportunistic links between devices, the remaining social

media with higher popularity and small size should take

the chance by direct sharing via D2D. Meanwhile, different

caching schemes have significant impacts on the latency, as

shown in Fig. 2 (b). According to different characteristics(i.e.,

popularity and size), the social media contents are distributed

by the infrastructure or by devices to increase the successful

delivery probability, the latency is considerably reduced.

Edge caching significantly improves SE/EE in F-RANs. In

[12], the problem of dynamic content-centric EC-AP clustering

and multicast beamforming with respect to both channel

condition and caching status is studied. Since each UE prefer

to access the EC-APs which cache the content they want, the

0 2 4 6 8 10 12 14 16 18 200

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Rank of Social Media, k

Pro

babi

lity

of C

achi

ng

EC−AP Caching,µ=1EC−AP Caching,µ=0.5EC−AP Caching,µ=0.1Device Caching,µ=1Device Caching,µ=0.5Device Caching,µ=0.1

(a) Optimal strategy of caching utilization in F-RANs[11]

148 IEEE COMMUNICATIONS LETTERS, VOL. 20, NO. 1, JANUARY 2016

3. (a) The minimized flow for different pairs of , η). It is obvious that

large storage capacity in infrastructures and utilization of caching in devices

(larger ) eliminate the traffic load for the RAN. (b) The latency performance.

we unify the caching resources in the RAN, the latency is reduced beyond

ws that the latency performance

of the algorithm and the optimal solution are closed, but with much lower

. It suggests that the algorithm provides suboptimal solutions.

not fully exploited function in present LTE systems. In the

current structure, an entity called PCRF supplies the necessary

information to make policies and charging decisions (Fig. 1).

Two functions exist in the PCRF: the Policy Decision Function

(PDF) and the Charging Rules Function (CRF), which perform

accurate business activities based on the information of user

data from the data center. It actually bridges the gap between

content-aware services (i.e. charging) and the communication

system. Together with PDF and CRF, the traffic flow of social

media can be reasonably monitored, learned, and sorted [14];

that is, the sizes and popularities of the active social media can

be obtained as two vectors and , where is the num-

ber of active social media in the RAN during a period of time.

Therefore, we propose a new decision making entity (Fig. 1)

in the core network for deciding the optimal strategy about

caching. In addition, the optimization problem, suffering from

high computational complexity, can be simply carried out by

an algorithm of low complexity (Algorithm 1) based on the

greedy nature inspected from Fig. 2.

In addition, device connectivity is needed, and can be

obtained by uncomplicated sensing of mobile users. A user

device senses in physical multicast channel (PMCH) and

detects that it has connections, then an estimator for

appears to be)/

, where )/2 is the number

of all possible links, and the estimator is the minimum variance

unbiased estimator (MVUE) for the Bernoulli parameter

With , and , the proposed new entity enables the algo-

rithm and give the suboptimal strategy for resource allocation

Algorithm 1. The Greedy Algorithm

Input: , and

Output: and

1: cache, bandwidth

2: for 1 to do Scan social media

3: if cache then

4: Cache cache cache

5: for 1 to do Scan social media

6: if is not cached in eNB & bandwidth then

7: Share via D2D communication

of caching. The performance of the proposed algorithm approx-

imates that of the optimization in (3), while featuring low

complexity, as shown in Fig. 3(c).

Subsequently, the BS, after receiving the suboptimal strategy

from the on-line algorithm in the core network, could peri-

odically broadcast a signal to the UEs by physical broadcast

channel (PBCH) to tell them the which social media are suit-

able for direct device sharing. Upon receiving the signal, the

UEs will attempt to access the social media by D2D commu-

nication first [13]. If D2D communication fails, the device will

alternatively access the content from the BS. This completes

the implementation of unifying caching resources in the RAN

in present cellular systems.

EFERENCES

[1] Ericsson. (2015). Ericsson Mobility Report [Online]. Available:http://www.ericsson.com/mobility-report

[2] 3GPP. (2015). 3GPP Specification Series [Online]. Available:

[3] I. V. Loumiotis et al., “Dynamic backhaul resource allocation: An evolu-tionary game theoretic approach,” Trans. Commun., vol. 62, no. 2,pp. 691–698, Feb. 2014.

efficiency via in-networkin cognitive radio sensor networks,” IEEE Trans. Wireless

Commun., vol. 13, no. 3, pp. 1222–1234, Mar. 2014.Z. Ming, M. Xu, and D. Wang, “Incan: In-network cache assisted eNodeBcaching mechanism in 4G LTE networks,” , vol. 75,pp. 367–380, 2014.E. Bastug, M. Bennis, and M. Debbah, “Living on the edge: The roleof proactive caching in 5G wireless networks,” IEEE Commun. Mag.

vol. 52, no. 8, pp. 82–89, Aug. 2014.X. Zhuo, Q. Li, G. Cao, Y. Dai, B. Szymanski, and T. L. Porta, “Social-

ve caching in DTNS: A contact duration aware approach,”in Proc. Int. Conf. Mobile Adhoc Sens. Syst. (IEEE MASS), 2011,pp. 92–101.J. Cho, S. Oh, J. Kim, H. H. Lee, and J. Lee, “Neighbor caching in

ad hoc networks,” , vol. 7, no. 11,pp. 525–527, Nov. 2003.

[9] M. Newman, An Introduction. London, U.K.: Oxford Univ.

L. Breslau, P. Cao, L. Fan, G. Phillips, and S. Shenker, “Web cachingand Zipf-like distributions: Evidence and implications,” in Proc. IEEE

, 1999, 126–134.P. Sobkowicz, M. Thelwall, K. Buckley, G. Paltoglou, and A. Sobkowicz,

of user post lengths in internet discussions—Aof the Weber-Fechner law?” EPJ Data Sci., vol. 2, no. 1,

pp. 1–20, 2013.D. G. Luenberger, Introduction to Linear and Nonlinear Programming

y, 1973, vol. 28.W. Sun, E. G. Strom, F. Brannstrom, Y. Sui, and K. C. Sou, “D2D-basedV2V communications with latency and reliability constraints,” in Proc.

IEEE GLOBECOM Workshops, 2014, pp. 1414–1419.[14] E. Bastug, M. Bennis, and M. Debbah, “A transfer learning approach for

(b) Latency performance of caching utilization in F-RANs [11]

Fig. 2. Caching utilization analysis in F-RAN

EC-AP clustering is influenced significantly by the caching

strategies. Threes caching strategies are proposed as baselines:

popularity-aware caching, random caching and probabilistic

caching. For the popularity-aware caching baseline, Each BS

caches the most popular contents until its storage is full. For

random caching, each EC-AP caches contents randomly with

equal probabilities. For the probabilistic caching, each EC-AP

caches one content with the probability of its popularity. Fig. 3

(a) shows the effect of different caching strategies. It is obvious

that the popularity-aware caching baseline performs best be-

cause EC-APs are more likely to cooperate with each other to

achieve high cooperation gain. Random caching is only better

than the no caching strategies because EC-APs can hardly

cooperate with each other. The performance of probabilistic

caching is in between. Meanwhile, both the power assumed

by traditional radio resource and caching are jointly considered

to evaluate the energy efficiency of F-RAN in [13]. The energy

consumption is defined as product of the power of keeping a

content object and the content object size. Particularly, a nested

coalition formation game-based algorithm is proposed to find

the optimal cell association scheme. Besides, a suboptimal

algorithm is designed to reduce the computational complexity.

Fig. 3 (b) illustrates the performance comparison of these

two algorithms. The EE increases as the size of local cluster

cache is enlarged, which indicates that the energy efficiency of

the F-RAN outperforms the C-RAN due to the edge caching

technique.

IEEE ACCESS, VOL. 2016, SI ON RECENT ADVANCES IN SOCIALLY-AWARE MOBILE NETWORKING 3

40 45 50 55 60 65 70 75 800

500

1000

1500

2000

2500

3000

Power Cost (dBm)

Bac

khau

l Cap

acity

(M

bps)

Without CachingRandom CachingProbabilistic CachingPopularity−aware Cachingη = 30

η = 10

η = 1 η = 0.01 η = 10−6

η → + ∞

(a) The impact of different caching techniques on SE[12]

Size of local cluster cache K0 1 2 3 4 5

Ave

rage

effe

ctiv

e ca

paci

ty to

pow

er c

onsu

mpt

ion

ratio

(M

bit/J

oule

)

×10-3

4.5

5

5.5

6

6.5

7

7.5

8

8.5

Algorithm 2 (Based on nested coalition formation game)Algorithm 3 (Suboptimal algorithm)

θlT=0.1, θ

lC=0.3

(marked by square)

θlT=0.1, θ

lC=0.6

(marked by "o")

θlT=0.1, θ

lC=0.9

(marked by "*")

(b) The impact of different caching technique on EE [13]

Fig. 3. Impact of caching technique to SE/EE in F-RANs

B. Mode Selection impacting on the performance of F-RAN

In traditional C-RANs, UEs can only access the RRH to

get the content that they need, which indicates only the global

C-RAN mode is available in C-RANs. However, in F-RANs,

due to the existing of edge caching equipment, UEs can access

to D2D UEs, EC-APs or RRHs according to different channel

and cache conditions to be served. Four basic access mode is

shown in Fig. 1. D2D mode is enabled when the distance

between two adjacent UEs with edge cache is sufficiently

close and the transmitter caches the content that the receiver

needs. EC-AP is triggered when no neighboring D2D UEs can

provide the desired contents while the neighbour EC-AP can.

Furthermore, if more than one EC-AP can provide UEs with

the desired contents, the local distributed coordination mode

is triggered. If the desired contents are just stored in the center

edge, and the desired UE has to access all potential RRHs for

the service, and hence the global C-RAN mode is triggered.

Generally speaking, one UE prefers to being served not via

backhaul or fronthaul, but through accessing the local D2D

UEs or EC-APs which cache the desired content to achieve

high SE/EE and low latency. The major reasons include:

1) the nearer D2D UE or EC-AP brings higher signal to

interference plus noise ratio (SINR), which indicates high

SE and EE; 2) the D2D UE or EC-AP usually consumes

less energy and suffer lower latency because the contents are

delivered to UEs locally without passing through the fronthaul

or backhaul. Compared with the pure C-RAN mode, in which

all contents are achieved by the central cache server, the

adaptive model selection in F-RANs can significantly improve

SE and decrease latency.

Except for the SE and latency, green communication is

another key factor in F-RANs. Ref. [14] compares the energy

consumption of D2D mode and cellular mode to evaluate the

cost performance of different mode selection. It can be shown

in Fig. 4(b) that when the distance between two D2D users is

sufficiently short, the energy of the D2D mode is significant

reduced because the high SINR and low transmit power due to

the short communication distance, and the low circuit power

of the D2D transmitter. In F-RANs, D2D is an effective access

technique to provide UEs with low latency and low EE with

constrained cache capacity, while the global C-RAN mode can

afford all the contents from the centralized cache, which results

in not good SE and EE performance. UEs should select proper

association mode, which is based on the desired content in the

edge cache, the desired SINR, and the energy consumption.

To illustrate the impact of different access modes, in [14],

the authors illustrate the latency performance under different

access mode. Local Video base mode (local VB) means one

UE can communicate with another UE locally to get the

contents, which is the same with the D2D mode in F-RANa.

In Sub video base (subVB) mode, one UE access more than

one neighboring access points to get the contents, which is

the same with the local distributed cooperation mode in F-

RAN. Video base (VB) mode means one UE access the nearest

access point and it is the same with nearest EC-AP mode.

Outside mode means one UE has to connect the BBU pool to

get the contents it wants, which is the global C-RAN mode. It

can be seen from Fig. 4 (a) that the local VB mode achieves

the best performance. The performance of subVB mode and

the VB mode are similar while the subVB performs better

because of the cooperation gain. The latency of outside mode

is biggest because it wastes much time on getting contents

from the Internet and transmitting them through fronthaul.

III. RADIO RESOURCE ALLOCATION

Compared with that in C-RANs, radio resource alloca-

tion in F-RANs is more complex but advanced because the

deployment of equipment with edge caching such as EC-

APs and D2D UEs should be considered. Precoding design,

resource block allocation, user scheduling, and cell association

should be jointly designed to optimize SE, EE and latency

performances. Three kinds of performances are categorized to

illustrate the impact of radio resource allocation in F-RANs.

Spectral Efficiency Optimization: To optimize SE in C-

RANs, one UE can access to a certain subset of neighboring

RRHs. Such subsets of neighboring RRHs are mainly deter-

mined by disjoint clustering scheme or user-centric clustering

scheme [8], in which the cluster formation is based on the

reference signal received power threshold. Disjoint clustering

scheme can mitigate the inter-cell interference but the edge

users may suffer serious interference from neighboring clus-

ter. In user-centric clustering scheme, there exists no edge

users, thus the inter-cluster interference decreases. Besides

the large-scale CRSP, C-RAN executes power allocation and

dynamical scheduling to enhance SE. However, in F-RAN,

IEEE ACCESS, VOL. 2016, SI ON RECENT ADVANCES IN SOCIALLY-AWARE MOBILE NETWORKING 4818 IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 15, NO. 4, JUNE 2013

Fig. 7. Evaluation of SVC resolution schemes. (a) Delay of difference SVC

resolution schemes in the Cloud (b) Overhead of different SVC resolutions

schemes in the Cloud.

TABLE II

ELAYS OF REFETCHING HARING FOR ARIOUS EVELS

hance layers for extremely high scalability, which may practi-

cally bring too much overhead.

C. Prefetching Delays

In ESoV, video segments can be prefetched among VB,

tempVB, and localVBs of the mobile users, based on their

activities in SNSs. we evaluate the required delays for different

levels of prefetching as shown in Table. 2. We here use the

normal resolution con guration of “ ” with 2

second temporal segmentation by default (the same in fol-

lowing tests). We also set the sharing length of “little” as only

the rst 5 seconds of the BL and ELs, that of “parts” as the rst

15 seconds of the BL and ELs, and that of “all” as all BL and

ELs segments.

We can see that prefetching supported by the cloud computing

is signi cantly fast. When prefetching via wireless links, it takes

several seconds. However it is obvious that in most cases [26],

[38] a recipient of the video sharing may not watch immediately

after the original sharing behavior, that is normal users have

signi cant access delay gaps, so this prefetching transmission

delay won’t impact user’s experience at all, but will bring “non-

Fig. 8. Average click-to-play delay for various cases.

buffering” experience in fact when the user clicks to watch at a

later time.

D. Watching Delay

We test how long one user has to wait from the moment that

one clicks the video in the mobile device to the moment that

the rst streaming segment arrives, which is called as “click-to-

play” delay. As shown in Fig. 8, if the video has been cached

in localVB, the video can be displayed nearly immediately with

ignorable delay. When we watch video which is fetched from

the subVC or the VC, it generally takes no more than 1 second

to start. However if the user accesses to AMES-Cloud service

via the cellular link, he will still suffer a bit longer delay (around

1s) due to the larger RTT of transmission via the cellular link.

For the cases to fetch videos which are not in the AMES-

Cloud (but in our server at lab), the delay is a bit higher. This

is mainly due to the fetching delay via the link from our server

at lab to the cloud data center, as well as the encoding delay. In

practical, there are be optimized links in the Internet backbone

among video providers and cloud providers, and even recent

video providers are just using cloud storage and computing ser-

vice. Therefore this delay can be signi cantly reduced in prac-

tice. Also this won’t happen frequently, since most of the pop-

ular videos will be already prepared in the AMES-Cloud.

VIII. CONCLUSION

In this paper, we discussed our proposal of an adaptivemobile

video streaming and sharing framework, called AMES-Cloud,

which ef ciently stores videos in the clouds (VC), and utilizes

cloud computing to construct private agent (subVC) for each

mobile user to try to offer “non-terminating” video streaming

adapting to the uctuation of link quality based on the Scalable

Video Coding technique. Also AMES-Cloud can further seek to

provide “non-buffering” experience of video streaming by back-

ground pushing functions among the VB, subVBs and localVB

of mobile users. We evaluated the AMES-Cloud by prototype

implementation and shows that the cloud computing technique

brings signi cant improvement on the adaptivity of the mobile

streaming.

The focus of this paper is to verify how cloud computing

can improve the transmission adaptability and prefetching for

(a) Average delay of different access mode [14]

(b) Percentage of energy saved in D2D mode com-pared with cellular mode

Fig. 4. Impact of access mode to the performance of F-RAN

cell association is quite different from that in C-RAN because

users prefer to access EC-APs which cache the contents they

interest in. The serving cluster can be determined by two

conditions: 1)The EC-AP caches contents the UE wants; 2)

The reference signal received power from the EC-AP to the

UE is high enough. Once the serving cluster is formulated, user

scheduling and power allocation is executed to optimize the

spectral efficiency of the network. Besides, in F-RAN, spectral

efficiency is no longer an important goal to pursuit because

high spectral efficiency means heavy burden on the fronthaul,

whose capacity may be not high enough to afford too much

data traffic. Thus, the tradeoff between the fronthaul capacity

and the transmit power should be highlighted.

To illustrate the impact of the cell association in F-RANs,

in [15], the ergodic rate for both EC-AP UEs and device-

to-device UEs by taking into account the different nodes

locations, cache sizes as well as user access modes are

analyzed and evaluated. As shown in Fig. 4(a), the numerical

results of the ergodic rate for the local distributed coordination

mode with different EC-AP cache sizes versus cluster radius

threshold Lc are validated. It illustrates how ergodic rate varies

with the number of EC-APs in the cluster when the EC-AP

cache size is fixed. It suggests that more EC-APs serve the

desired UE with the increase of signal and the decline of

interference and leads to the enlargement of cluster SIR and

results in the improvement of ergodic rate of local distributed

coordination mode. Similarly, the larger cache size of EC-APs

Cf suggests that there are more opportunities for the desired

UE to get the contents that it needs, which leads to a higher

45 50 55 60 65 701

1.1

1.2

1.3

1.4

1.5

1.6

1.7

Cluster Radius Threshold Lc (m)

Erg

odic

Rat

e of

Loc

al D

istr

ibut

ed C

oord

inat

ion

Mod

e (b

ps/H

z)

Simulation Cf=200

Simulation Cf=500

Simulation Cf=800

Analytical Cf=200

Analytical Cf=500

Analytical Cf=800

(a) Ergodic rate versus cluster radius [15]

12

Size of local cluster cache K

0 1 2 3 4 5

Avera

ge e

ffective c

apacity (

Mbit/s

/Hz)

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Analytical results of Corollary 4

Monte Carlo results

(Uniform distribution,

marked by "o")

s = 2 (marked by "x")

s = 1 (marked by square)

s = 0.5 (marked by "*")

(a) Effective capacity vs. size of cluster content cache

Size of local cluster cache K

0 1 2 3 4 5

Avera

ge e

ffective c

apacity to a

vera

ge p

ow

er

consum

ption r

atio (

Mbit/J

oule

)

×10-3

2

2.5

3

3.5

4

4.5

5

5.5

6

6.5

Analytical results of (33)

Monte Carlo results

s = 1 (marked by square)

s = 0.5 (marked by "*")

s=0 (Uniform distribution,

marked by "o")

s = 2 (marked by "x")

(b) Average effective capacity to average power consumption ratio vs. size ofcluster content cache

Fig. 4. Performance of a typical C-RAN cluster.

VI. SIMULATION ESULTS

In this section, simulation results are provided to evaluat

the performance of the cluster content caching structure in

C-RANs. In particular, the channel are assumed to be block

fading in each RRU, where = 1 ms and = 1 kHz.

The densities of RRHs and users are given as

10 . The cardinality of the required content set is

, the size of each content object is set as = 1 Mbits,

and the popularity of content objects follows Zipf’s law [31].

The power consumptions of RRHs in the active and the sleep

modes are set as act = 104 W and sle = 56 W, respectively

[29], and the power consumption of cluster content caching

and backhaul transmissions defined in are set as CC = 0 15W and BH = 10 W.

To verify the accuracy of our theoretical analysis, the effec-

tive capacity of a typical user is plotted in Fig. 3, where the

path loss exponent is set as = 4 . As shown in the figure,

the numerical results based on the analytical results given in

Theorem 3 match the Monte Carlo results perfectly, which

shows the validity of our theoretical derivations. Moreover, as

the value of the path loss exponent increases, the effective

capacity of a typical user increases. The reason is that the

receive SINR can be improved as the signal to interference

path loss ratio increases, e.g., ) =

/d . Note that the serving RRH is usually

nearer to the user than the interfering RRHs, e.g.,

and thus is an increasing function of the path loss exponent

, and so is the effective capacity.

The performance of a typical cluster is evaluated in Fig. 4,

where the Zipf exponent is set as = 0 , respectively,

and the QoS exponents are set as = 0 and = 0respectively. As shown in the figure, both the effective capacity

and the energy efficiency increase as the size of cluster cach

is enlarged, since more requests can be responded to locally

with short delay. In particular, compared with the no-caching

structure, e.g., = 0, the average effective capacity and the

average effective capacity to average power consumption ratio

can be improved up to 0.57 Mbit/s/Hz and 0.004 Mbit/Joule

when = 5. Moreover, the performance of cluster content

caching improves faster as increases, since the user interests

converge to fewer popular content objects stored in the cluster

cache.

To further evaluate the performance improvement of clus-

ter content caching, the performance of the proposed RRU

allocation and RRH association algorithms is provided in Fig.

5, where the effective capacity and the effective capacity t

power consumption ratio are plotted, respectively. For the

th RRU, the average effective capacity to power consumption

ratio is defined as (48) on the following page. As shown in

Fig. 5, both the effective capacity and the energy efficiency

increase as increases and decreases. Compared with the

simulation results in Fig. 4, the performance gains of cluster

content caching can be enlarged to 0.95 Mbit/s/Hz and 0.0055

Mbit/Joule when = 5In Fig. 6, the performance comparison of different RRU

allocation and RRH association schemes is provided. Two

previous resource allocation schemes are selected as two

comparable schemes, which are the orthogonal RRU allocatio

scheme and the RRU full-reusing scheme. As shown in Fig.

5(a), Algorithm 2 can always achieve the best average effective

capacity performance. There exists a performance gap between

Algorithm 2 and Algorithm 3, since the utility function of

RRU allocation in Algorithm 3 cannot ensure each content

object makes the best choice. As increases, the interest

conflict among content objects can be alleviated, and the

performance of RRU full-reusing scheme approaches that of

Algorithm 2. Although increasing the density of RRHs can

always improve the effective capacity performance, it is not the

best choice from an energy efficiency perspective. As shown

in Fig. 5(b), the effective capacity to power consumption ratio

increases in the low region, while it keeps decreasing in

the high region. Moreover, the case of multi-casting can

be considered as a special case of our studied scheme, e.g.,

when content objects are served by using orthogonal RRUs,

and the performance of the orthogonal RRU allocation scheme

in Fig.6 can be treated as a lower bound on multi-casting.

VII. CONCLUSION

In this paper, a cluster content caching structure has been

proposed in C-RANs, in which some requested content could

(b) Average EE vs. size of cluster content cache [13]

Fig. 5. Impact of cache based cell association to SE/EE performance ofF-RAN

SE performance.

Energy Efficiency Optimization: Since C-RAN is a fully

centralized network, RRHs only operate as soft relay by

compressing and forwarding the received signals from UEs to

the BBU pool and all of the signal and baseband processing

are performed in the BBU pool. All desired contents that

UEs wants are stored in the BBU pool and delivered to UEs

by fronthaul, which consumes much energy and results in

the low EE. The EE optimization in C-RANs focuses on

maximizing the total overall SE and minimizing the energy

consumed in the BBU pool. However, in F-RANs, the EC-

APs are capable of caching contents and executing signal and

baseband processing. Thus, some UEs can get the preferred

contents from local EC-APs without through fronthaul and

backhaul. With more contents cached in local EC-APs, the EE

performance of F-RAN becomes better. The EE optimization

in F-RANs should highlight the effect of local caching. In [13],

the numerical results of the average effective EE versus the

size of cluster content cache are validate, as shown in Fig. 4(b).

With the larger size of local cluster cache, the EE performance

of F-RANs increases almost 3 times because more UEs can

access EC-APs to locally get the contents cached.

Latency optimization: In C-RANs, all signals and contents

are centralized to the BBU pool from RRHs through fronthaul.

Since RRHs have no capability of signal processing and

IEEE ACCESS, VOL. 2016, SI ON RECENT ADVANCES IN SOCIALLY-AWARE MOBILE NETWORKING 5

0 0.5 1 1.5 2 2.5 3 3.5 43

4

5

6

7

8

9

10

Backhaul delivery delay δ

Avera

ge d

ow

nlo

ad d

ela

y

Exhaustive search, cache size = 1

Exhaustive search, cache size = 2

Proposed algorithm, cache size = 1

Proposed algorithm, cache size = 2

Figure 2. Average download delay versus backhaul delivery delay.

0.5 1.5 2.5 3.51000

1500

2000

2500

3000

3500

4000

4500

5000

Backhaul delivery delay

Avera

ge d

ow

nlo

ad d

ela

y

caching placement strategy

posed caching placement strategy

LCD caching placement strategy

Figure 3. Performance comparison of various caching placement strategies.

part of the segments can be served by the local caches, instea

of having to be fetched from the remote central controller

through the backhaul. The step size and the stopping criterion

are the same as Fig. 2. Fig. 3 demonstrates that our proposed

strategy outperforms the other schemes, and important insights

are revealed. When the backhaul delay is small enough, it

can be regarded as equivalent to the case of infinite caching

capacity. In this case, the best way to save download time

is to maximize the channel diversity for each segment. As a

result, the proposed strategy and the MPC policy coincide at the

point =0. When the backhaul delay increases, the advantage

of caching diversity emerges, and the LCD policy will surpas

the MPC policy. Our proposed strategy and the LCD policy

will converge when backhaul links are suffering from severe

delivery delays. This is because when there is a large , even

a single delivery via backhaul will lead to a huge delay and

thus backhaul transmission should be prevented as much as

possible. Therefore, the strategy that can provide maximum

caching content diversity is favorable.

V. CONCLUSIONS

This paper presented a framework to minimize the aver-

age download delay of wireless caching networks. A caching

placement problem which takes into account physical layer

processing as well as backhaul delays was formulated to

fully exploit the benefit of caching. As the design problem

is an MINLP problem, we relaxed it into a DC optimization

problem and adopted the SCA algorithm to solve it efficiently

Simulation results showed that our strategy can significantly

reduce the average download delay compared to conventional

strategies, and the proposed low-complexity algorithm can

achieve comparable performance to exhaustive search. More

over, we demonstrated that the backhaul propagation delay

will greatly influence the caching placement. Specifically, when

the backhaul delay becomes very small or very large, our

proposed strategy will gradually evolve to the MPC and the

LCD strategy, respectively. In particular, for a practical value

of the backhaul delay, the proposed caching placement serve

as the best strategy. Therefore, it can be concluded that our

work provides a promising model to formulate the download

delay for wireless caching networks, and important insights are

given for determining the optimal caching placement strategy

under different backhaul conditions.

EFERENCES

[1] Cisco Systems Inc., “Cisco visual networking index: Global mobile datatraffic forecast update, 2014-2019,” White Paper, Feb. 2015

[2] N. Golrezaei, A. Molisch, A. Dimakis, and G. Caire, “Femtocachingand device-to-device collaboration: A new architecture for wireless videodistribution,” IEEE Commun. Mag., vol. 51, no. 4, pp. 142–149, Apr.2013.

[3] J. Gu, W. Wang, A. Huang, and H. Shan, “Proactive storage at caching-enable base stations in cellular networks,” in Proc. IEEE Int. Symp. onPersonal Indoor and Mobile Radio Comm. (PIMRC), London, UK, Sept.2013, pp. 1543–1547.

[4] M. Maddah-Ali and U. Niesen, “Fundamental limits of caching,” IEEETrans. Inf. Theory, vol. 60, no. 5, pp. 2856–2867, May 2014.

[5] N. Golrezaei, P. Mansourifard, A. Molisch, and A. Dimakis, “Base-stationassisted device-to-device communications for high-throughput wirelessvideo networks,” IEEE Trans. Wireless Commun., vol. 13, no. 7, pp. 3665–3676, Jul. 2014.

[6] K. Shanmugam, N. Golrezaei, A. Dimakis, A. Molisch, and G. Caire,“Femtocaching: Wireless content delivery through distributed cachinghelpers,” IEEE Trans. Inf. Theory, vol. 59, no. 12, pp. 8402–8413, Dec.2013.

[7] X. Peng, J.-C. Shen, J. Zhang, and K. B. Letaief, “Joint data assignmentand beamforming for backhaul limited caching networks,” in Proc. IEEEInt. Symp. on Personal Indoor and Mobile Radio Comm. (PIMRC)Washington, DC, Sept. 2014.

[8] H. Ahlehagh and S. Dey, “Video-aware scheduling and caching in theradio access network,” IEEE/ACM Trans. Netw., vol. 22, no. 5, pp. 1444–1462, Oct. 2014.

[9] P. Blasco and D. Gunduz, “Learning-based optimization of cache contentin a small cell base station,” in Proc. IEEE Int. Conf. Commun. (ICC)Sydney, Australia, Jun. 2014, pp. 1897–1903.

[10] R. Rejaie, M. Handley, H. Yu, and D. Estrin, “Proxy caching mechanismfor multimedia playback streams in the Internet,” in Proc. Int. WebCaching Workshop, San Diego, CA, Mar. 1999.

[11] A. Ghosh, J. Zhang, J. G. Andrews, and R. Muhamed, Fundamentals ofLTE. Prentice-Hall, 2010.

[12] G. Caire and D. Tuninetti, “The throughput of hybrid-ARQ protocols forthe Gaussian collision channel,” IEEE Trans. Inf. Theory, vol. 47, no. 5,pp. 1971–1988, Jul. 2001.

[13] M. Razaviyayn, M. Hong, and Z.-Q. Luo, “A unified convergence analysisof block successive minimization methods for nonsmooth optimization,”SIAM J. Optim., vol. 23, no. 2, pp. 1126–1153, 2013.

[14] M. Zink, K. Suh, Y. Gu, and J. Kurose, “Characteristics of YouTubenetwork traffic at a campus network-measurements, models, and implica-tions,” Comput. Netw., vol. 53, no. 4, pp. 501 – 514, 2009.

Fig. 6. Average download delay versus backhaul delivery delay [16]

storage, the latency between RRHs and UEs is much smaller

than that between RRHs and the BBU pool. The latency

optimization problem in C-RANs focus on minimizing the

latency of fronthaul, which determines the whole latency of

the network. However, in F-RANs, the latency optimization

problem is more complex. EC-APs can provide contents that

UEs need, which results in the significant decrease of the

latency of fronthaul, while the latency from EC-APs to UEs

can not be ignored. To illustrate the impact of the cache

placement to the latency, [16] formulates an average download

delay minimization problem subject to the caching capacity

constraint of each base station. Fig. 5 shows that the results

given by the efficient radio resource allocation algorithm

almost approaches the optimal value that obtained by the

exhaustive search. Besides, when the cache size becomes en-

larged, the download delay decreases dramatically. Therefore,

it is anticipated that the local caching technique is promising

to improve the delay performance of F-RANs.

IV. OPEN ISSUES AND CHALLENGES

Although the performance analysis and radio resource al-

location for F-RANs have been researched, there are still

many challenges and open issues remaining to be discussed in

the future, including the F-RAN architecture for 5G, and the

social-awareness F-RAN.

A. F-RAN architecture for 5G

It is envisioned that 5G will bring a 1000x increase in terms

of area capacity compared with 4G, achieve a peak rate in

the range of tens of Gbps, support a roundtrip latency of

about 1 ms as well as connections for a trillion of devices,

and guarantee ultra-reliability. Since F-RANs are evolved from

the existing HetNets and C-RANs, it is fully compatible with

the other 5G systems. The advanced 5G techniques, such as

the massive multiple-input multiple-output, cognitive radio,

millimeter wave communications, and non-orthogonal multiple

access, can be used directly in F-RANs. There are some

apparent advantages in F-RANs, including the local caching,

real-time CRSP, and flexible CRRM at the edge devices, the

rapid and affordable scaling that make F-RANs adaptive to

the dynamic traffic and radio environment, and low burdens

on the fronthaul and the BBU pool. Furthermore, the different

IoT applications in 5G, such as mobile vehicular connectivity,

smart city, and industrial automation, should be based on the

F-RAN. It can be anticipated that F-RANs will be qualified

for meeting the high SE, EE, low latency and high reliability

for different kinds of IoTs in 5G.

B. Social-awareness F-RAN

In the traditional D2D scenario, one UE is able to connect to

any other UE via D2D once the channel gain between them is

high enough. However, in practical, UEs are usually carried by

human beings who are actively involved in social interactions

and the social relationships of UEs are quite different from

each others. One UE usually does not communicate with

another unfamiliar UE due to the security. According to the

locations, interests and background, UEs can be divided into

different communities. UEs in the same community tend to

exchange contents with each other while they seldom connect

to UEs in other communities. Compared with the traditional

D2D scenario, this social feature becomes more apparent in F-

RANs as more contents are cached in the edge, and UEs prefer

to be associated with the edge equipment, such as EC-APs and

D2D UEs, other than the centralized BBU pool. Therefore, the

social relationship affects the success probability of a D2D

communication, which contributes a lot to the SE/EE and

latency in F-RANs. Unfortunately, the performance analysis

and radio resource allocation seldom take the social relation-

ship into consideration and how social relationship effects the

performance analysis and radio resource allocation in F-RANs

is still not straightforward. Thus, the social aware F-RAN,

which leverages the community property of social networks

with the communication systems, is worthy to be researched.

V. CONCLUSION

This article has outlined and surveyed the recent advances of

the performance analysis and radio resource allocation for the

socially-aware mobile networking in fog radio access networks

(F-RANs). Compared with the typical cloud radio access net-

works (C-RANs), the performance analysis and radio resource

allocation in F-RANs are more complex but advanced due to

the local edge cache and adaptive model selection. The spectral

efficiency (SE), energy efficiency (EE), and latency of F-RANs

as three major issues have been exploited in this article. In

particular, to show how different caching and model selection

strategies impact on SE, EE and latency, some impressive

numerical results have been summarized. Nevertheless, given

the relative infancy of the field for the F-RAN based 5G

system, there are still quite a number of outstanding problems

that need further investigation. Notably, it is concluded that

greater attention should be focused on the F-RAN architecture

and the social-awareness F-RAN. The presented performance

analysis and radio resource allocation in F-RANs provide

breakthroughs of theories and technologies for the advanced

5G systems, and it is anticipated to transfer these two key

issues of F-RANs that discussed in this article to the standards

organizations.

IEEE ACCESS, VOL. 2016, SI ON RECENT ADVANCES IN SOCIALLY-AWARE MOBILE NETWORKING 6

REFERENCES

[1] P. Costa, C. Mascolo, M. Musolesi, and G. P. Picco, “Socially-awarerouting for publish-subscribe in delay-tolerant mobile ad hoc networks,IEEE Journal on Selected Areas in Communications, vol. 26, no. 5, pp.748 C 760, June 2008

[2] X. Chen, B. Proulx, X. Gong, and J. Zhang, “Exploiting social ties forcooperative D2D Communications: A mobile social networking case,IEEE/ACM Transactions on Networking, vol. 23, no. 5, pp. 1471-1484,Oct. 2015

[3] M. Peng, Y. Sun, X. Li, Z. Mao, and C. Wang, “Recent advances incloud radio access networks: System architectures, key techniques, andopen issues”, IEEE Communications Survey & Tutorial, vol. xx, no. xx,online, Mar. 2016.

[4] Q. Yan and F. R. Yu, “Distributed denial of service attacks in software-defined networking with cloud computing,” IEEE Communications Mag-

azine, vol. 53, no. 4, pp. 52-59, Apr. 2015.[5] M. Peng, S. Yan, K. Zhang, and C. Wang, “Fog-computing-based radio

access networks: issues and challenges”, IEEE Network, vol. 30, no. 4,pp. 46–53, Jul. 2016.

[6] A. Bradai, K. Singh, T. Ahmed, and T. Rasheed, “Cellular softwaredefined networking: A framework,” IEEE Communications Magazine,vol. 53, no. 6, pp. 36 - 43, Jun. 2015.

[7] S. C. Hung, H. Hsu, S. Y. Lien, and K. C. Chen, “Architectureharmonization between cloud radio access networks and fog networks,”IEEE Access, vol. 3, pp. 3019-3034, Dec. 2015.

[8] S. Sardellitti, G. Scutari, and S. Barbarossa, “Joint optimization of radioand computational resources for multicell mobile-edge computing,”IEEE Transactions on Signal and Information Processing over Networks,vol. 1, no. 2, pp. 89-103, Jun. 2015.

[9] S. H. Park, O. Simeone, and S. Shamai, “Joint optimization of cloudand edge processing for fog radio access networks,” to appear in IEEE

Transactions on Information Theory.[10] X. Wang, M. Chen, T. Taleb, A. Ksentini, and V. C. M. Leung, “Cache

in the air: Exploiting content caching and delivery techniques for 5Gsystems,” IEEE Communications Magazine, vol. 52, no. 2, pp. 131-139,Feb. 2014.

[11] H. Hsu and K. Chen, “A resource allocation perspective on caching toachieve low latency,” IEEE Communications Letters, vol. 20, no. 1, pp.145-148, Jan. 2016.

[12] M. Tao, E. Chen, Hao. Zhou, and W. Yu, “Content-centric sparsemulticast beamforming for cache-enabled cloud RAN,” IEEE Transac-

tions on Wireless Communications, vol. PP, No. 99, pp. 1 - 1, DOI:10.1109/TWC.2016.2578922, 2016

[13] Z. Zhao, M. Peng, Z. Ding, W. Wang, and H. V. Poor, “Cluster contentcaching: An energy-efficient approach to improve quality of servicein cloud radio access networks,” IEEE Journal on Selected Areas in

Communications, vol. 34, no. 5, pp. 1207 - 1221, May 2016.[14] X. Wang, M. Chen, T. T. Kwon, L. Yang, and V. C. M. Leung, “AMES-

cloud: A framework of adaptive mobile video streaming and efficientsocial video sharing in the clouds,” IEEE Transactions on Multimedia,vol. 15, no. 4, pp. 811-820, Jun. 2013.

[15] S. Yan, M. Peng, and W. Wang, “User access mode selection in fogcomputing based radio access networks,” ICC 2016, Kuala Lumpur,Malaysia, May 2016, pp. 1-6.

[16] X. Peng, J.C. Shen, J. Zhang, and K. B. Letaief, “Backhaul-awarecaching placement for wireless networks,” in Proc. IEEE Global Com-

mun. Conf. (GLOBECOM), SAN DIEGO, CA, USA, Dec. 2015, pp.1-6.

Mugen Peng (M’05–SM’11) received the B.E.degree in electronics engineering from the Nan-jing University of Posts and Telecommunications,Nanjing, China, in 2000, and the Ph.D. degree incommunication and information systems from theBeijing University of Posts and Telecommunications(BUPT), Beijing, China, in 2005. Afterward, hejoined BUPT, where he has been a Full Professorwith the School of Information and Communica-tion Engineering since 2012. In 2014, he was anAcademic Visiting Fellow with Princeton University,

Princeton, NJ, USA. He leads a Research Group focusing on wirelesstransmission and networking technologies with the Key Laboratory of Uni-versal Wireless Communications (Ministry of Education), BUPT. His mainresearch areas include wireless communication theory, radio signal processing,and convex optimizations, with a particular interests in cooperative com-munication, self-organization networking, heterogeneous networking, cloudcommunication, and internet of things. He has authored/coauthored over 60refereed IEEE journal papers and over 200 conference proceeding papers.

Dr. Peng was a recipient of the 2014 IEEE ComSoc AP OutstandingYoung Researcher Award, and the best paper award in IEEE WCNC 2015,WASA 2015, GameNets 2014, IEEE CIT 2014, ICCTA 2011, IC-BNMT2010, and IET CCWMC 2009. He received the First Grade Award of theTechnological Invention Award in the Ministry of Education of China forthe hierarchical cooperative communication theory and technologies, and theFirst Grade Award of Technological Invention Award from the China Instituteof Communications for the contributions to the self-organizing networkingtechnology in heterogeneous networks. He is on the Editorial/AssociateEditorial Board of the IEEE Communications Magazine, IEEE Access, IET

Communications, International Journal of Antennas and Propagation (IJAP),and China Communications. He has been the guest leading editor for thespecial issues in the IEEE Wireless Communications, IEEE Access, and IET

Communications.

Kecheng Zhang received the B.S. degree intelecommunication engineering from Beijing Uni-versity of Posts and Communications (BUPT),China, in 2012. He is currently pursuing the Ph.D.degree in the Key Laboratory of Universal WirelessCommunications for Ministry of Education at BUPT.His research interests include the radio resource allo-cation optimization and cooperative communicationsin heterogeneous cloud radio access networks.