ieee transactions on vehicular technology, vol. 66, no. 7, july 2017 6569 qoe...

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 7, JULY 2017 6569

QoE-Driven Channel Allocation and HandoffManagement for Seamless Multimedia in

Cognitive 5G Cellular NetworksMd. Jalil Piran, Member, IEEE, Nguyen H. Tran, Member, IEEE, Doug Young Suh, Member, IEEE,

Ju Bin Song, Member, IEEE, Choong Seon Hong, Senior Member, IEEE, and Zhu Han, Fellow, IEEE

Abstract—Cognitive radio (CR) is among the promisingsolutions for overcoming the spectrum scarcity problem in theforthcoming fifth-generation (5G) cellular networks, whereasmobile stations are expected to support multimode operations tomaintain connectivity to various radio access points. However,particularly for multimedia services, because of the time-varyingchannel capacity, the random arrivals of legacy users, and theon-negligible delay caused by spectrum handoff, it is challengingto achieve seamless streaming leading to minimum quality of expe-rience (QoE) degradation. The objective of this paper is to managespectrum handoff delays by allocating channels based on the userQoE expectations, minimizing the latency, providing seamless mul-timedia service, and improving QoE. First, to minimize the handoffdelays, we use channel usage statistical information to compute thechannel quality. Based on this, the cognitive base station maintainsa ranking index of the available channels to facilitate the cognitivemobile stations. Second, to enhance channel utilization, we developa priority-based channel allocation scheme to assign channels to themobile stations based on their QoE requirements. Third, to mini-mize handoff delays, we employ the hidden markov model (HMM)to predict the state of the future time slot. However, due to sensingerrors, the scheme proactively performs spectrum sensing and re-actively acts handoffs. Fourth, we propose a handoff managementtechnique to overcome the interruptions caused by the handoff. Insuch a way that, when a handoff is predicted, we use scalable videocoding to extract the base layer and transmit it during a certaininterval time before handoff occurrence to be shown duringhandoff delays, hence providing seamless service. Our simulation

Manuscript received February 15, 2016; revised July 18, 2016 and Octo-ber 10, 2016; accepted November 7, 2016. Date of publication November 16,2016; date of current version July 14, 2017. This work was supported inpart by the Ministry of Science, ICT and Future Planning, South Korea, inthe ICT R&D Program, 2015; in part by the Basic Science Research Programthrough National Research Foundation of Korea (NRF) funded by the Ministryof Education (NRF-2014R1A2A2A01005900); and in part by the US NSF CPS-1646607, ECCS-1547201, CCF-1456921, CNS-1443917, ECCS-1405121, andNSFC61428101. The review of this paper was coordinated by Dr. B. Canberk.

M. J. Piran is with the Department of Computer Engineering, Sejong Uni-versity, and the Department of Electrical and Radio Engineering, Kyung HeeUniversity, Seoul 02447, South Korea (e-mail: [email protected]).

N. H. Tran and C. S. Hong are with the Department of Computer Engineering,Kyung Hee University, Seoul 02447, South Korea (e-mail: [email protected];[email protected]).

D. Y. Suh and J. B. Song are with the Department of Electrical and Ra-dio Engineering, Kyung Hee University, Seoul 02447, South Korea (e-mail:[email protected]; [email protected]).

Z. Han is with the Department of Electrical and Computer Engineering,University of Houston, Houston, TX 77004, USA (e-mail: [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TVT.2016.2629507

results highlight the performance gain of the proposed frameworkin terms of channel utilization and received video quality.

Index Terms—Cognitive radio (CR) networks, fifth-generation(5G) cellular networks, handoff, multimedia communication, qual-ity of experience (QoE).

I. INTRODUCTION

MOBILE traffic is experiencing its breakneck prolifera-tion and explosive growth by development of diverse

applications, e.g., multimedia services, placing a tremendousstrain on wireless communications capacity. Vodafone revealedthat its customers used 400 petabytes of data on their mo-bile phones during August to October 2015, which is on av-erage twice as much fourth-generation (4G) data as they did onthird-generation (3G) [1]. The popularity of multimedia appli-cations is the main cause of this growth, whereas CISCO pre-dicted that 69.1% of mobile traffic would be occupied by videodown/uploading [2]. According to the mentioned statistics, thenext mobile generation, fifth-generation (5G), will face multi-fold challenges, e.g., uneven distribution and diverse increasingof mobile traffic. Therefore, various stringent requirements aresolicited, such as near zero delay, scalability to support massivenumber of devices, high reliability, spectrum and bandwidthflexibility, network and device energy efficiency, etc. [3]. Spec-trum allocation is considered as one of the most crucial problemsamong the mentioned requirements. It is due to the inefficienttraditional spectrum allocation, in which fixed licensed spec-trum bands within ultrahigh frequency bands that are allocatedto the cellular networks do not have the capability to satisfyusers’ expectation of high-speed wireless access for 5G cellularnetworks. To solve this problem and achieve higher spectrumflexibility in cellular networks, different types of technologiesare under consideration such as carrier aggregation in long-term evolution (LTE-U) and licensed assisted access (LAA)[4], LTE-WiFi aggregation (LWA) [5], MuLTEFire [6], oper-ation on millimeter wave bands [7], and cognitive radio (CR)[8]. According to CR features such as opportunistic and sharedspectrum access, in this paper, we consider CR to be utilized inour framework.

A. Cognitive Radio for 5G Cellular Networks

Due to its unique characteristics such as flexibility, adaptabil-ity, and interoperability, CR is a promising candidate technology

0018-9545 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

6570 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 66, NO. 7, JULY 2017

to address spectrum scarcity issues for 5G cellular networks.Therefore, 5G standardization leaders, such as fifth generationpublic private partnership (5GPPP) in [9], ITU in [10], and IEEEin [11], considered CR as a candidate technology for enhance-ment of spectral efficiency in 5G.

Indeed, there are many essential similarities between 5Gand CR:

1) interworking with different systems and/or networks;2) adaptation based on the access network features (5G case),

respective on the primary system network characteristics(CR case). Remark that the 5G-integrated access net-works and CR primary systems networks have the samemeaning;

3) new and flexible protocols;4) very advanced Physical (PHY) and media access control

(MAC) technologies;5) a very advanced terminal, endowed with the possibility

to scan the environment, with intelligence and decisioncapabilities;

6) resource management (and end-to-end integrated resourcemanagement that should include all the networks involvedin the data transmission process).

Thus, whereas CR was proposed to enhance the spectrumefficiency and 5G is going to imply the interconnection of thewireless networks all around the world, integration of these twotechnologies (cognitive 5G cellular networks) is a promisingsolution to tackle or partly address the next-generation mobilecommunication issues. To enable 5G devices to have cognitivecapability, the mobile devices only need to be equipped withsoftware-defined radio (SDR) to be reconfigured with the othernetwork protocols, e.g., 802.16, 802.11, 802.22, etc. Thus, withthe help of CR technology, 5G devices can discover spectrumopportunities with one or two transceivers and there will be noneed to have additional devices.

B. Problem Statement

In cognitive 5G cellular networks, cognitive-enabled users(CUs) are allowed to occupy the unused licensed/unlicensedchannels that are not used temporally or spatially by licensedusers (LUs) and other CUs. The licensed bands are the fixedspectrum bands that are bought by the operators and on theother hand, the unlicensed bands are the bands that have beenleft free for unlicensed users, e.g., 2.4 GHz ISM, 5 GHz U-NII,and 60 GHz mmWave.

One of the main problems in cognitive 5G cellular networksthat need to be considered is dynamic channel allocation, whichmay seriously degrade service quality and increase transmissionfailure and delay, particularly in multimedia services. Amongdifferent multimedia services, some are considered as more im-portant and critical than the others based on their quality ofexperience (QoE) requirements, delay-sensitive voice over in-ternet protocol (VoIP) service as an instance. Therefore, it isnecessary that better quality channel should be provided to thetraffic classes with higher priority.

Upon successful utilization of a channel by a CU that is notdedicated to CUs, the CU must immediately vacate the primaryspectrum bands whenever any LU reclaims the channel (knownas spectrum mobility). Moreover, spectrum mobility may occur

when a CU moves to another cell or when the quality of thecurrent channel becomes poor. Spectrum mobility presents thespectrum handoff function. During spectrum handoff, the cur-rent communication process should be transferred to anotheravailable channel.

For channel mobility, it takes significant time called as spec-trum handoff delay, which is the time spent to search for anotheravailable channel and radio frequency front-end reconfigurationprocess by CUs. Thus, another severe issue to be considered incognitive 5G cellular networks is the spectrum handoff delay,which particularly in multimedia services gives rise to interrup-tion of service that dramatically reduces QoE.

C. Our Contributions

To overcome with the above-mentioned issues, in this pa-per, we study a database (DB)-assisted and index-based channelallocation and handoff management framework, based on chan-nel usage statistics, hidden markov model (HMM), and scalablevideo coding (SVC). The main contributions of the paper aredescribed briefly as follows.

1) To predict the arrival time of LUs as well as to selectthe best available channel, we propose a DB-assisted andindex-based scheme to collect channel usage informa-tion and rank the available channels in accordance withtheir quality. The cognitive base station (CBS) maintainsa ranking index of the available channels based on theirparameters such as detection accuracy, false alarm prob-ability, LU idle time, and LU arrivals. The channels areranked into different classes based on their QoS param-eters. The use of ranking index decreases interferencewith the LUs significantly, maximizes the CU spectrumoccupancy, minimizes the number of interruptions, andhandoff delay.

2) Whereas all the traffic sources may not have the same QoErequirements, we develop a priority-based channel assign-ment paradigm. The scheme classifies CUs’ traffics basedon their QoE requirements and reserves an appropriateclass of the available channels for each traffic class.

3) We formulate channel state estimation in the temporal do-main in the HMM framework. The transition pattern ofthe LUs is modeled by a Markov chain. We use CUs’experiences and channel history to predict the next timeslot state, e.g., handoff occurrence. To take care of spec-trum sensing errors, as well as prediction errors, our sys-tem performs handoff prediction proactively but channelswitching is performed reactively upon LU detection.

4) We propose a handoff interruption management to pro-vide seamless multimedia service. If the mechanism ofchannel state estimation predicts that a handoff will occurin future slots, we use SVC to encode the video sequenceinto two layers: base layer (BL) and enhancement layer(EL). Due to the lightweight of BL and its higher priority,the server sends only the BL code in a certain intervalbefore handoff occurrence. Then, if arrival of an LU isdetected by the CU, while the CU is searching for a sub-stitute available channel in the index (during the handoffdelay), the receiver presents the prefetched BL code tohide handoff interruption from the user perspective.

JALIL PIRAN et al.: QOE-DRIVEN CHANNEL ALLOCATION AND HANDOFF MANAGEMENT FOR SEAMLESS MULTIMEDIA 6571

5) Our framework can be distinguished from the other relatedproposed scenarios by its unique merits that can be statedin four aspects.

a) According to the fact that multimedia servicesare bandwidth-hungry, and delay- and distortion-sensitive, they need high quality, as well as stableand reliable channels, in order to achieve the targetQoE. However, identifying such kind of channelsneeds a longer sensing time that shortens the trans-mission time and increases delay and distortion. Ourproposed framework with its unique features such asQoE-drivenness and channel quality-awareness in-terestingly outperforms compared to the other pro-posed scenarios.

b) By employing HMM and reactively handoff per-forming, the scheme significantly reduces hand-off delay, limits interference to LUs, and improvesspectral efficiency.

c) While the other related works ignore QoE, thisframework effectively provides seamless multime-dia service and improves QoE, by extracting, trans-mitting, and presenting BL during the handoff delayusing SVC.

d) Finally, the unique feature of the proposed frame-work is to consider all the three important re-quirements that an efficient CR-based scheme re-quires, i.e., channel utilization, QoS provisioning,and QoE-drivenness.

6) We conduct a comprehensive simulation study to evalu-ate the effectiveness of our proposed framework in sev-eral aspects such as the number of interruptions, averagehandoff delay, average number of collisions, the qualityof the reconstructed video, and end-user perception. Inthe cases of the number of interruptions, average handoffdelay, and average number of collisions, we compare theproposed work with reactive and proactive random chan-nel selection, greedy nonpriority channel allocation, andfair proportional channel allocation. The proposed schemereduces the number of interruptions up to 25%, and forhandoff delay and collisions, it outperforms much betterthan the other four methods. Finally, we have observedhow effective the proposed framework improves the qual-ity of the reconstructed video in terms peak signal-to-noiseratio (PSNR) and mean opinion score (MOS).

The rest of this paper is organized as follows. In Section II,we review and discuss the related work. The system modelis presented in Section III. Our proposed QoE-driven channelallocation and handoff management framework is described andformulated in Section IV. Section V validates the derived resultsand analyzes the performance behaviors. Finally, in Section VI,we draw conclusions.

II. LITERATURE REVIEW

In recent years, plenty of research work has been con-ducted to manage the issue of spectrum handoff. Lertsinsrub-tavee et al. [12] investigated the tradeoff between the spectrumhandoff delay and channel busy duration time. For initial andtarget channel selection, a probabilistic approach was intro-duced in [13]. Sheikholeslami et al. provided analytical results

for the switching-enabled policy according to the connection-based spectrum handoff. Wang and Chen [14] studied spec-trum handoff in three aspects: nonspectrum handoff method, thepredetermined channel list handoff, and the spectrum handoffbased on the radio sensing scheme. Then, they evaluated themin terms of link maintenance probability and the effective datarate of SUs’ transmission.

Furthermore, there are some proposed schemes in the liter-ature, which consider two modes: reactive or proactive. In thereactive mode, the spectrum switching by SUs is performed af-ter detection of the LU arrival, while in the proactive mode, SUsforecast the channel status based on the LU activity and vacatethe channel before reappearance of LUs.

In case of reactive, a Markov transition model integrating withthe preemptive resume priority M/G/1 queuing network was in-troduced in [15], in order to calculate handoff delay causedby sensing, handshaking, channel switching, and the waitingtime. Wang and Wang [16] compared the pros and cons ofproactive-sensing and reactive-sensing spectrum handoff tech-niques. Furthermore, the authors proposed a greedy algorithmthat determines suboptimal target available channels. The pro-posed algorithm can automatically switch between the two tech-niques. Although because of the on-demand spectrum sensing,reactive schemes may get an accurate spectrum hole, they areinvolved with longer handoff latency.

On the other hand, in proactive case, a spectrum manage-ment scheme called voluntary spectrum handoff (VSH) [17] isintroduced for cognitive radio networks (CRNs) to minimizemean opinion score secondary user (SU) disruption periodsduring spectrum handoff. The transition probability selectionand reliability-based selection algorithms are employed to esti-mate the spectrum life time, to determine VSH time. Proactivemethods are involved with much smaller latency compared toreactive methods. However, they suffer from two common de-tection errors, i.e., false alarm and miss detection. By predictinga handoff and as the SU leaves the channel in the predicted time,there will be no LU arrival then, which will result in wasting theprecious spectrum resources (false alarm). On the other hand, ifthe scheme does not detect the LU accurately, then it will resultin interference to LUs (miss detection).

Most of the proposed schemes considered reactive handoffor proactive handoff to minimize handoff delay while ignor-ing QoE and channel quality. This may make them unsuitablefor multimedia applications where service interruptions signif-icantly degrade the QoE. To the best of our knowledge, thereis no previous research work concentrating on providing seam-less streaming and improving QoE during handoff delays formultimedia services.

III. SYSTEM MODEL

We adopted a time-slotted cognitive-based cellular systemhavingX primary channels with identical bandwidth. Each pri-mary channel has N time slots and each slot is composed oftwo parts, i.e., sensing (link setup) and data transmission [37].At the beginning of a time slot, the CU performs the sensingfunction via energy-detection (ED) technique to discover thestatus of the channel and sends the collected data to the CBS.The traffic pattern of LUs on the licensed channels is designedas a two-state Markov model, the ON and OFF renewal process


alternating between busy and idle periods. In the queue theory,the LU arrival process follows a Poisson arrival model and ex-ponential service distribution.

Let us suppose that CU arrival also follows a Poisson arrivalmode. The Poisson arrival process is employed to inspect thechannel usage efficiency of the CUs links according to the LUarrival rate and the number of channels. The traffic modeledby the Poisson process with the arrival rate of λ is adopted toa Markov chain model by calculating the success probabilityPsuccess by the chain state estimation.

Although 5G devices owe a licensed spectrum band, with thehelp of CR, they will be able to capture more bandwidth tohave a higher data rate in order to improve the end-user QoE,particularly for multimedia services. Hence, in cognitive 5Gcellular networks, mobile units equipped with CR technologyare considered as CUs, whenever they need to access otherspectrum bands, whether it is licensed or unlicensed. Comparedto normal 5G mobile stations, CUs will be able to adjust theirQoE expectations, according to the network fluctuations, e.g.,dynamic channel state. According to the coverage range of thevarious CBSs in 5G networks and also based on the spectrumtype that they access, our proposed framework is applicable inthe following modes.

1) It can help to extend the coverage to rural and indoorareas to provide broadband internet service by WAN. Thebroadband internet can be provided through a CBS that isconnected to world wide web (WWW). To implement thisscenario, there is a need to deploy one or more fixed CBSsand many available radio channels. The CBS is assumedto adjust its transmission power at or below the thresholdthat has been defined by the local regulator.

2) Another scenario is to use CR as the backhaul, whereasboth CBS and relay have CR functionality. So that bothof them are able to select the same available channel forthe backhaul link dynamically.

3) The next scenario that can be considered is to apply CRto small cells. According to the fact that small cells re-quire low transmission power and more spectra and higherbandwidth, CR can give them the capability to access moreavailable channels. An advantage for this scenario is themitigation of cochannel interference among neighboringsmall cells and reducing the interference between a small-and a macro-cell. This would result in achieving highernetwork throughput as well. This scenario can be consid-ered for both indoors and outdoors.

4) CR can help 5G to access more spectrum for the capacityenhancement. In this scenario, basically the operators usethe licensed bands as the main component carrier, whichfacilitate a stable link to the mobile devices. In cases wherea higher data rate is needed, a CBS discover spectrum isprovided as a supplement.

5) Moreover, although we considered a CR-based frame-work, even by eliminating the CR, our proposed hand-off management technique is applicable to LTE-U, LAA,LWA, and MuLTEFire as well.

A CU is supposed to utilize two transceivers with SDR capa-bilities. One of the transceivers is allocated for transmission andthe other one is for channel sensing with two functionalities:

Fig. 1. Spectrum handoff process.

monitoring the channel usage and collecting channel informa-tion to estimate future channel state [38]. The channel status,busy or idle, is detected at the beginning of each slot by the CUwith ED spectrum sensing techniques [18]. If the status of thechannel in the slot is idle, the CU starts to transmit till the endof the slot.

A CU that occupied a primary channel must leave it uponarrival of incumbents immediately, as recommended in IEEE802.22-WRAN (wireless regional area networks) [19]. In thestandard, it is assumed that the CU needs to vacate the occupiedprimary channels within two seconds from the arrival of LUs.The CU requires to search and move to a new available channelin three cases: when the legacy users reclaim the channel, whenthe quality of underused channel becomes poor, and when CUsmove spatially and the CU transmission coverage may overlapwith an LU that currently uses the same channel.

The process of switching to another available channel poses anon-negligible delay. Such kind of time delays are called as thehandoff delay that is cumulative of the time taken for: channeldiscovering, spectrum sharing, switching to the new availablechannel, and link establishment.

Since the CU has to temporarily break the transmission pro-cess during spectrum handoff to search and switch to anotheravailable channel, handoff may occur several times during atransmission session, particularly in the case of multimediatransmission; the delay caused by handoff processes interruptsthe transmission for several times that prevents providing seam-less transmission and significantly degrades QoE. In order tomanage handoff delays in multimedia services, our scheme fo-cuses on three aspects:

1) obtaining channel statistical information and mitigatinginterference to the LUs;

2) selecting the optimal candidate channels during the hand-off process and minimizing handoff delay;

3) prefetching the light weight BL code to present during thehandoff process to provide seamless service and improveQoE.

Fig. 1 illustrates the operation of a CU in primary channels.We assumed there are X primary channels having a synchro-nized slotted structure. At the beginning, CH#2 is allocated tothe CU by the CBS, because it is assumed as the best availableand most reliable channel in the ranking list in comparison to the


Fig. 2. QoE-driven channel allocation and handoff management scheme.

other available channels and it is matched with the traffic QoErequirements, e.g., CH#1. The CU calculates sensing accuracyand idle duration and forward to the CBS to store into the DB. Inaddition, the CU monitors the other channels and maintains theranking index to choose the backup channel in case of handoff.

At time t0, a spectrum handoff is predicted to occur at t2. Thehandoff delay is predicted to be one slot. The CBS checks theranking index and allocates CH#3 as the substitute channel toresume CU’s transmission over it because it is assumed to bethe first channel in the corresponding class in the index. Then,equal to the predicted handoff delay, only the extracted BL codeis transmitted in advance to the receiver during the second slot tobe viewed during the third slot. At time t2, an LU is detected bythe CU. The sender pauses its transmission, while the receiverstarts to show the prefetched BL code during the third time slot.As the best matched available channel in the ranking index,CH#3 is facilitated to the CU. The CU moves its transmissionto CH#3 and resumes it at t3 without interference to the LUwhich starts its transmission over CH#2 at t2.

IV. PROPOSED QOE-DRIVEN CHANNEL ALLOCATION AND

HANDOFF MANAGEMENT FRAMEWORK

Our objective is to allocate the best available channel and tomanage the unavoidable spectrum handoff by minimizing thehandoff latency and handle the handoff process in such a wayto provide seamless multimedia service as well as improvingQoE. In doing that, our system consists the three phases, eachcomposed of some functionalities, as illustrated in Fig. 2.

1) Phase I: Channel evaluation and allocationa) data collection via ED spectrum sensing technique;b) channel quality estimation based on sensing accu-

racy and channel idle duration;c) channel allocation according to the CU’s QoE re-

quirements and the available channels’ quality.2) Phase II: Channel state estimation

a) parameter estimation using the Baum–Welch algo-rithm (BWA) under the HMM parameters;

b) channel decoding for channel state estimation.3) Phase III: Handoff interruption management

a) extraction of the BL code using SVC and sending itduring a certain interval before handoff occurrence.

The system functionalities related to each phase are explainedand formulated in the next sections.

A. Phase I: Channel Evaluation and Allocation

In this section, we explain and formulate different function-alities of the first phase including data collection, channel qual-ity estimation, and channel allocation. Algorithm 1 illustratesa concise procedure for the channel evaluation procedure. Thefunctionalities related to this phase are described and formulatedin the following sections.

1) Data Collection: Dynamic spectrum access consists ofspectrum sensing, networking, and regulator policy. Spectrumsensing for a fixed spectrum channel can be performed in twodomains: spatial domain, in which the CR-enabled device esti-mates the location of the primary transmitter and based on thatadjusts the transmission power to prevent interference to thetransmitted signals, and temporal domain, in which the CUs tryto discover the time intervals when the primary transmitter isidle.

In this paper, we follow the temporal spectrum sensing anddata collection is performed through channel sensing at the startof each slot. During sensing time, the CU first senses the LUactivity on the target band and decides to transmit if and onlyif there is no active LU. We use the ED technique for channelsensing due to its efficiency and fast noncoherent features thatessentially calculates a running average of the signal power overa window of a given spectrum length.

In our technique, CU compares its received signal with asensing threshold. The sensing threshold is based on the noisefloor that is to specify the transmission state of the primarytransmitter. This technique does not need any prior informationregarding LU’s signal and its performance is poor in the environ-ment with low SNR. However, the sensing function is not always


precise because of the limitation in the observation quantity andnoise.

Normally, two types of errors occur in sensing of LU activi-ties, which are known as false-alarm and miss-detection errors.Miss-detection phenomenon occurs when the user fails to sensethe presence of the LU, which results in interference to theLUs. And on the other hand, the probability of false-alarm oc-curs when the sensing function declares falsely that there is anactive LU, which results in waste of spectrum resources [20].Assuming Rayleigh fading channels [21], miss-detection prob-ability Pmd and false-alarm probability Pfa are, respectively, asfollows using complete and incomplete gamma and generalizedMarcum Q-functions:⎧⎪⎪⎨

⎪⎪⎩

H1 : Presence of Signal if

(∑N

n−1

(Rs [n]

)2)

> Θ

H0 : Absence of Signal if

(∑N

n−1

(Rs [n]

)2)

< Θ(1)

Pfa = P

{( N∑

n−1

(Rs [n]

)2)

> Θ | H0

}

=Γ(m, Θ

2 )Γ(m)

(2)

Pmd = 1 −Q

((Θδ2

− 1 − γ

)√UN

1 + 2γ

)

(3)

Pd = e−Θ2

m−2∑

k=0

1k!

(Θ2

)2+(1 + γ

γ

)m−1

×(

eΘ

2( 1+ γ ) − e−Θ2

m−2∑

k=0

1k!

(Θγ

2(1 + γ)

)k)

(4)

where Rs is the received signal, Θ is the E-D threshold, δ2

is additive white Gaussian noise variance, U is the number of

CUs, γ is SNR,∑N

n−1

(Rs [n]

)2is the output of the integrator,

and the upper incomplete gamma function [22] is defined as theintegral from Γ(a, x) =

∫∞x ta−1e−tdt. The primary signals are

modeled as a two-state Markov chain: H0 and H1. From (2),it can be seen that the probability of false-alarm and SNR areindependent. Thus, when H1 is satisfied, it means that an LU isactive on the channel. The relation between miss-detection andfalse-alarm probabilities [23] is stated as

Pfa = Q

(√Nγ +Q−1(1 − Pmd)

√1 + 2γ

)

. (5)

Based on IEEE 802.22 standard [19], the probabilities ofmiss-detection and false-alarm are assumed to be lower than0.1: 0.01 ≤ Pfa ≤ 0.1 and 0.9 ≤ Pd ≤ 0.99.

2) Channel Quality Estimation: Data collected via spectrumsensing function are stored in a DB at the CBS side. With thehelp of the collected data, the quality of the available channels iscomputed based on sensing accuracy and channel idle duration.Using (2) and (4), the accuracy of spectrum sensing is calculatedbased on false-alarm probability and detection probability:

ω = Pd

(1 − Pfa

)(6)

where 0.81 ≤ ω ≤ 0.99. In addition, the discussed spectrumsensing function enables the CU to identify LU traffic patterns.The CU may obtain various traffic patterns over different chan-nels over the time. The objective of traffic pattern identificationis to predict traffic rate fluctuations based on the evaluationhistory and the CU experiences.

In this paper, we consider deterministic mode of traffic pat-terns, and hence we model the channel usage as an ON–OFF

source alternating model, where ON means busy period and OFF

means idle period. The model is confirmed as a suitable modelbecause it approximates the channel usage pattern at publicsafety bands [24]. The ON and OFF periods are independentidentically distributed. The alternating renewal process can bedesigned in form of a two-state birth-death process:

f(T1) ={νe−νT1 , T1 ≥ 00, T1 < 0

(7)

f(T0) ={υe−υT0 , T0 ≥ 00, T0 < 0

(8)

where ν and υ are the rate parameter of the exponential distri-bution [25] for busy and idle periods with the average 1

ν and1υ , respectively. If we assume T0 and T1 as the duration of OFF

and ON states, and E[T0] and E[T1] as the vectors related tothe mean quantities of the OFF and ON states, the probabilitydensity function (PDF) of busy and idle periods will be fT1(x)and fT0(y), respectively, the time duration till the transition of

the next state isfT 1 (x)fT 0 (y ) , and hence, the secondary opportunity

for the CU can be stated as

CUopp =E[T0]

E[T0] + E[T1]=

1/υ1/υ + 1/ν

. (9)

The maximum-likelihood estimation method is used to obtainthese mean values. To do that, the model uses the CU experi-ences and the collected information stored in the DB. The meanduration of ON and OFF periods is updated according to the sens-ing results gradually. The probabilities that an LU on a specificchannel remains idle while the CU transmission process is going


on and the LU does not arrive while the CU is in sensing modeare calculated based on the scale ζC and shape ζH parametersas follows:

PTt =

(ζC

ζC + ΓT

)ζH

, (10)

PSt =

(ζC

ζC + ΓT

)ζ C

. (11)

Finally, based on the sensing accuracy and mean value of theOFF state, the channel quality is estimated as [26]

QCH =(

1 + logρ ω)E[T0] (12)

where ρ > 1 is to taking the preference of the CU into accountthat is derived using partial derivative of the sensing accuracyand mean duration of the idle state:

∂2QCH

∂ω∂ρ= − ω

E[T0]ρ

(1

ln ρ

)2

< 0 (13)

∂2QCH

∂E[T0]∂ρ= − lnω

(1

ln ρ

)2

> 0 (14)

where ρ ∈ [1.1, 10.0]. As in (12), the quality of the channels iscomputed based on sensing accuracy and idle duration. To sat-isfy some specific QoS parameters, the CU may tend to capturethe channels with higher idle duration rather than higher sensingaccuracy. Therefore, selection of a bigger ρ means that the idletime duration is prior to the sensing accuracy, and vice versa.

3) Channel Allocation: In a CR-based network, the CUs arefacilitated with high bandwidth through heterogeneous wireless

architecture and dynamic channel access mechanisms. How-ever, such kind of networks face the challenges because of thefluctuating nature of the available channels and various require-ments of different applications. Therefore, some special kindof spectrum management schemes are required to address thesechallenges. In this section, we explain our solution to tackle dy-namic channel access issue in cognitive 5G cellular networks,which is abstracted in Algorithm 2.

Generally, the CBS needs to allocate a new channel to a CU inthree cases: when an LU reclaims the channel, when the qualityof the channel becomes poor, and when the CU moves spatiallyout of a cell.

If an LU reclaims the channel, the CU must avoid any harmfulinterference with the LU. This process, through interferencenullification, is done by the listen-before-talk policy in whichthe CU must sense the channel before transmitting any datapacket. Hence, it is assumed that the LU arrival is detectableby the CU within an acceptable time duration and collisionpercentage is negligible.

In any of the mentioned three cases if CUs need to switch toanother channel, the CBS looks at the ranking index and selectsthe best available and the most reliable channel according tothe channel quality and the CU’s QoE requirement. The CBSinforms the CU about the selected channel during the link setup.


Fig. 3. Traffic and channel mapping.

Then, the CU switches to the allocated channel and resumesits transmission process. With more available channels in theranking index, the CU may transmit the packets at a higherdata transmission rate. The CBS periodically updates the DB ofchannel usage statistics. Any new discovered idle channel will beadded to the index. The process of channel sensing is performedindependently on the channels in a mixture of proactive and on-demand modes. Therefore, the CBS is supposed to schedule theconcurrent sensing on the channels to avoid duplicate sensingand duplicate data collection on a specific channel.

In the remaining part of this section, we explain the classifi-cation of the traffic sources based on their requirements.

In the case of multimedia transmission in cognitive 5G cel-lular networks, it is expected that there will be a lot of differ-ent types of bandwidth-hungry and distortion-sensitive trafficsources such as internet protocol television (IPTV), video con-ferencing, video, voice, and image. Each of the traffic sourcesmay require different network QoS parameters. For instance,voice and constant bit rate are sensitive to the delay, while theothers such as data, image, and variable bit rate are not. There-fore, for such kind of delay-sensitive multimedia traffic, ourframework allocates the channel with smaller handoff proba-bility. Both delay and the quality of the received content at thereceiver side are considered as the most important QoE metrics.

Whereas the users in a CR-based network are not supposed touse the allocated channel permanently, dynamic channel accessmay seriously degrade service quality and increase transmis-sion failure and delay. Therefore, to tackle this problem, in theproposed scheme, we allocate the available channels to the CUsbased on their requirements and the quality of the channels.In such a way that the channels with higher sensing accuracyand idle duration are provided with the more delay and failuresensitive traffic sources in order to enhance the QoE.

Traffic mapping to the appropriate channels is done at theCBS for downlink and at the CU for uplink directions. Thechannels allocated to a CU enable it to transmit over uplink andreceive over the downlink [36]. The general scenario is shownin Fig. 3. At the channel scheduler module, the CBS provides anappropriate channel for the traffic coming from a media originserver and heading to a CU. And on the other hand, the CUopts a suitable channel based on the traffic requirements. Asmentioned previously, the ranking index is exchanged betweenthe CBS and CU during link setup.

Upon arrival of a channel request from a CU from a specifictraffic class, the channel scheduler module checks the type of

channel that is required. Then, based on the channel qualitythat is required by the CU, the scheduler checks the rankingindex. If the index is empty, the call will be rejected. Otherwise,it checks the corresponding class in the ranking index. If thecorresponding class is nonempty, the call is accepted. Otherwise,it checks the next class with lower quality. A buffer is establishedfor every admitted request.

The scheduler schedules the accepted calls over the allocatedchannel. To protect the LUs, in the case of capturing licensedbands, upon detection of an LU, the scheduler stops the transmis-sion immediately and allocates another suitable channel fromthe index if available. If there is no channel in the index, theCBS puts the suspended CU into a queue to wait for the nextopportunity.

To compare the efficiency of the proposed channel allocationscheme, we study a greedy nonpriority channel allocation and afair proportional scheme as follows.

a) Greedy nonpriority channel allocation: In this scheme, theCBS does not classify the incoming traffic based on their QoErequirements and just serves them as first-in, first-out (FIFO).Upon arrival of a channel request from a CU, the CBS checkswhether there is any free channel in the index. If so, the CBSallocates the first available channel to the CU. However, toprotect the LUs, the CU senses the channel during the sensingpart of the first slot, if there is no active LU on it, the CU starts itstransmission. Such kind of channel allocation schemes decreasethe channel utilization efficiency.

b) Fair proportional channel allocation: In this scheme, theCBS reserves an equal number of bands and allocates them todifferent priority classes according to the total number of trafficclasses. The channel quality and QoE expectations are not takeninto account, which results in a waste of spectrum resources.

B. Phase II: Channel State Estimation

We use HMM to model the signal characteristics statisticallyregarded as Markov chains that is observable through the mem-oryless channels. To form an HMM, we need to incorporate anprocess that can be observed along with an underlying Markovchain. The observable process is to collect information regard-ing the underlying Markov chain that is said to be hidden. For anunderlying Markov chain, the observed process will be indepen-dent conditionally. The hidden process is in two modes: discreteor continuous finite-state homogeneous Markov chain. Then,the result of the process that can be observed may have finite-alphabet or general-alphabet. Therefore, it is characterized by aPDF or a probability mass function (PMF) appropriately.

The channel state, busy or idle, is hidden due to not beingdirectly observable. Therefore, the results of channel sensingdone by the CU are considered as the observation of the channelstate. Hence, since the variations of the hidden states are basedon LU activities, therefore, it is a hidden Markov process, whilethe collected observation information under a certain hiddenstate is a normal random process.

In this paper, we consider a hidden Markov chainwith Gaussian densities that is represented by a doublystochastic process {(Ht,Ot)}, where t = 0, 1, ... The hiddenstate set is Ht = {h1(t), h2(t), ..., hN (t)} with state space


hi ∈ {0, 1} ∀i ∈ {1, ..., N} indicating the channel is idle andbusy, respectively. {Ht} is a discrete-time finite-state homo-geneous Markov chain that satisfies P (Ht+1 = x | Ht = y) =P (H1 = y | H0 = x)∀t = 1, 2, ... The observation state set forM slots is Ot = {o1(t), o2(t), ..., oM (t)} with state space oj ∈{0, 1} ∀j ∈ {1, ...,M} indicating idle and busy, respectively.The observation set is the local decisions of the CU regardingM-slot LUs’ activities sensing on a specific channel. (Ht,Ot)is the stationary finite state system based on the followings:

1) (Ht,Ot) are jointly stationary, 0 < t ≤ N ;2) P

(Ot+1 = oj+1, Ht+1 = hi+1 | Oj

1 = yj1 , Hi1 = hi1

)=

P(Ot+1 = oj+1, Ht+1 = hi+1 | Ht = hi

);

where Ht and Ot are the state and the output of the station-ary finite state system, respectively. The random variable Ht isconditionally independent given Ot :

p(oK0 | hk0 ) =K∏

k=0

p(ot | ht) (15)

where K is a non-negative integer.At the beginning, the CU can provide the observation set.

The observation set is provided to determine the history of theresults of channel sensing. We assumed that the channel state isfixed during a time slot (not changed). It means that the statusof one slot can be either idle (flag = 0) or busy (flag = 1). Theflag is used to specify the number of OFF and ON slots of a bandto conclude the number of LU arrivals.

If we consider S(t) = {s1(t), s2(t), ..., sN (t)} as the set ofchannel states, the status of CHn at time t is sn (t) based onsome corresponding state transitions probabilities, where thestate space is S : {0, 1}. Then, on (t) is the corresponding resultof spectrum sensing function. We describe our model as anHMM [27] by its parameters Λ = (A,B, π), where we have thefollowing.

1) A = [aij ]N×N ∀1 ≤ i, j ≤ N is the state transition matrixthat defines transitioning probability from one state toanother or to the same state.

2) B = [bj (k)]N×M is the output symbol probability matrixthat computes the probability of providing various outputsymbols while being in a specific state.

3) π = {P (s1 = hi)} is the initial state probability vector.The parameters, i.e., the probabilities of state transition, ob-

servation symbol emission, and initial state distributions, arecalculated, respectively, as follows:

aij =P(hn (t) = sj | hn (t− 1) = si

),

N∑

j=1

aij =1 (16)

bj (k) = P(ot = sj | sn (t) = sj

),

N∑

k=1

bj (k) = 1 (17)

πj = P(sn (t) = sj

),

N∑

j=1

πj = 1 (18)

where 0 ≤ aij , bj (k) ≤ 1; i, j ∈ N, k ∈M, πj ≥ 0, π ={π1, π2, ..., πN }, and S = {s1, s2, ..., sN } denotes N observa-tion symbols. To estimate the status of the next slot, we needto specify the model parameter of each channel according to

the observation set. Therefore, HMM predictor is employedto predict the state of oM+1 based on the experienced Mobservations.

To do that, HMM is supposed to produce the observationsequence having maximum-likelihood probability. Therefore,the parameters are adapted by maximizing the probability ofP (O|Λ). Whereas we are supposed to have a DB to store CUexperiences, we exploit the collected information to facilitateHMM. The data collected by the CU are used to compute theprobabilities of occupancy of the channels using the maximum-likelihood approach. To do that, we first calculate the joint dis-tribution P (o; s) of the sensing data and estimated occupancy.Then, the joint distributions for all possible channels busy se-quences are calculated, and finally the distribution that resultsin the maximum probability and related channel busy state se-quence is obtained according to the estimation of the real chan-nel busy status:

P((o1(t), o2(t), ..., oM (t)

)(s1(t), s2(t), ..., sN (t)

))=

[

P(O1 = o1(t)

)P(S1 = s1(t)

)]

×[

P(O2 = o2(t)

)

P(S2 = s2(t)

)]

× · · · ×[

P(OM = oM (t)

)P

×(SN = sN (t)

)]

. (19)

First, to calculate the parameters and train the HMM model forthe future channel state prediction, the observation sequence isused as the training sequence. The observation set regarding thechannel state needs to be specified in order to obtain the historyof spectrum sensing results. To do this, we used BWA [28] thatis derived form of the expectation-maximization algorithm [29]to estimate the HMM parameters. Using BWA, the HMM modelparameters Λ = (A,B, π) are defined in the following format:

π = (π0, π1) (20)

A =

[a00 a01

a10 a11

]

(21)

B =

[b00 b01

b10 b11

]

. (22)

Based on a model Λ and an observation set O, the obser-vation evaluation problem P (O|Λ) is resolved by employinga forward–backward procedure using forward and backwardvariables:

αt(i) = P (Ot, sn (t) = i | Λ) (23)

βt(i) = P ({ot+1, ot+2, ..., oT }|sm = i,Λ). (24)


The recursive relation for both of them is calculated as

αt+1(i) = bj (t+ 1)N∑

i=1

αt(i)aij ∀j ∈ N ; 1 ≤ t ≤ T − 1

(25)

βt(i) =N∑

i=1

βt+1(j)aijb(ot+1) ∀j ∈ N ; 1 ≤ t ≤ T − 1

(26)

where α1(i) = πj bj (o1), 1 ≤ j ≤ N , and βT (i) = 1, 1 ≤ i ≤N , Then, the model parameters are estimated as

aij =∑T −1

t=1 αt(i)aijbj (ot+1)βt+1(j)∑T −1

t=1

∑Nj=1 αt(i)aijbj (ot+1)βt+1(j)

(27)

bj (k) =

∑Tt=1,ot =k


∑T −1t=1


(28)

πj =N∑

j=1

αt(i)aijbj (ot+1)βt+1(j)P (O | Λ)

. (29)

Using (25) and (26), P (O|Λ) can be obtained as

P (Ot | Λ) =N∑

i=1

αt(i)βt(i). (30)

Then, we calculate the joint probability of sensing resultsfollowed by either busy or idle slot oM+1:

P (Ot, 1 | Λ) =N∑

i=1

(N∑

j=1

αM (j)aij

)

· bi(oM+1 = 1) (31)

P (Ot, 0 | Λ) =N∑

i=1

(N∑

j=1

αM (j)aij

)

· bi(oM+1 = 0). (32)

By having the estimated parameters and decoded channelstate, we predict spectrum handoff occurrence for the comingtime slot (M + 1) according to

{P (Ot, 1 | Λ) ≥ P (Ot, 0 | Λ) =⇒ handoff

P (Ot, 1 | Λ) < P (Ot, 0 | Λ) =⇒ no handoff.(33)

To estimate the status of two conductive slots, suppose that thetransition rate of idle to busy is r01 = P (Xt+1 = 1 | Xt = 0)and the transition rate of busy to idle is r10 = P (Xt+1 = 0 |Xt = 1). Then, the probability of idle and busy based on thetransition rate is, respectively, as follows:

P (0) =r10

r01 + r10(34)

P (1) =r01

r01 + r10. (35)

And, using the Chapman–Kolmogorov equation [23], theprobability of the two consecutive states based on the

transition rate is calculated by

P(si+ 1 = 0 | si = 0

)=

r10

r01 + r10

+r01

r01 + r10e−(r10 + r01)tslot , (36)

P(si+ 1 = 0 | si = 1

)=

r10

r01 + r10

− r10

r01 + r10e−(r10 + r01)tslot (37)

where tslot is time slot duration. Equation (36) calculates the idleprobability of two consecutive idle slots whereas (37) computesthe probability that slot n is idle and the next slot n+ 1 isbusy. Then, using (10), (11), and (23)–(28), the probability ofavailability of at least one channel for CU is formulated as

Pavail = 1 −M∏

m=1[

1 −(

N∑

i=1

( N∑

j=1

αM (j)aij

)

· bj (oM+1 = 0

)

· P

(

PTt ≥ ϕ

)]

(38)

where ϕ is the collision threshold, and the failure probability,which is the probability that the CU tries to capture a channeland fails to transmit, is

Pfail =M∏

m=1[

1 −(

N∑

i=1

( N∑

j=1

αM (j)aij

)

· bj (oM+1 = 0

)

· P

(

PTt ≥ ϕ

)]

.

(39)

Using (38) and (39), the probability of successful transmis-sion of CU is equal thereby to

Psuccess =(

1 −M∏

m=1

[

1 −(

N∑

i=1

( N∑

j=1

αM (j)aij

)

· bj (oM+1 = 0

)

· P

(

PTt ≥ ϕ

)])( M∏

m=1

[

1 −(

N∑

i=1

( N∑

j=1

αM (j)aij

)

· bj (oM+1 = 0

)

· P

(

PTt ≥ ϕ

)])

. (40)

Finally, by having the initial arrival time of the LU that issensed by the CU and stored in the database, the maximumnumber of spectrum opportunities [30] captured by the CU isdetermined as follows:

CUOFF =

[

− 1tslot

(

tinit +logϕ

λ

)]

(41)


where λ is the LU arrival rate, and tinit is the initial arrival timeof the LU.

C. Phase III: Handoff Interruption Management

In the previous section, we discussed handoff prediction sce-nario in detail and we stated that to manage handoff interrup-tions when a handoff is predicted, the server extracts and sendsonly the lightweight BL code. In this section, we present ourproposed handoff management scheme. And subsequently, wediscuss the evaluation metrics for the received video qualityassessment.

As stated previously, in cognitive 5G networks, the CUs needto switch to another available channels when an LU reclaimsthe channel, and/or when the quality of the current channelbecomes poor, and/or even when the CU moves to another cell.The spectrum mobility results in handoff delay, which makesservice interruption in multimedia services. In order to overcomesuch kind of challenges, in this section, we present our handoffinterruption management scheme.

We use SVC to encode the video sequence in a scalable mode.To do that, we use a temporal-SNR scalability mode to encodethe video sequence into one BL (L0) and one EL (L1) with thesame spatial resolution, but with different frame rate and quality.The quality is determined by quantization parameter (QP), thehigher the QP, the lower the PSNR and bitrate [32]. We set higherQP and lower frame rate to the BL to generate lightweight BL.The reason is to overcome the interruptions caused by spectrumhandoff and thereby provide seamless multimedia service. TheEL bitstream is to enhance the quality of the received content.However, decoding of the EL depends on the successful de-

coding of BL. Moreover, the EL must be transmitted, received,and decoded entirely, otherwise it does not improve the contentquality at all.

In our scheme, when a handoff is predicted, equal to themean duration of handoff delay, only the lightweight BL codeis sent in advance. Then, at the time of LU arrival, while theCU is looking at the ranking index and trying to select thebest available channel, the receiver views only the prefetchedBL code. The handoff interruption management procedure isabstracted in Algorithm 3.

In the proposed network, the CBS is assumed to allocate thechannels to the CUs based on both the quality of the channelsand the requirements of the traffic sources. In order to overcomewith the issue of signaling overhead, in addition to the bitraterequired by the traffic source, an approximate amount of bitrateis considered for the extra BL bitrate, which may be required tobe sent additionally in advance in order to manage the handoffdelay. One of the salient advantages of the proposed frame-work is that the CBS is aware of both channel quality and CUs’requirements. Thereby, our system enhances both channel uti-lization and service quality. For the cases that the CBS predictsthat maybe the current channel does not have the capacity tocarry the extra BL code and there is the probability of overhead,for the current slot also transmits only BL code due to the higherpriority of BL code. In the case that the CBS predicts that thecurrent channel does not have the capacity to carry the extra BLcode and may overhead happens, according to the higher proba-bility of BL code, only BL code will be sent for the both currentslot and channel switching (handoff delay) slot. So, there willnot be service interruption at the receiver side.

Although, comparing to the other related proposed schemes,the distinctive superiority of the proposed framework can be


seen in this algorithm. However, in multimedia services, thedegree of delight (seamless service) or annoyance (interruptedservice) of the end users is the main efficiency determinativefactor, which is ignored by the other proposed spectrum handoffschemes, our framework interestingly is able to improve the enduser satisfaction by providing seamless multimedia services.

In the remaining part of this section, we explain the evaluationmetrics for the reconstructed video quality.

To evaluate the efficiency of video transmission schemes,traditionally the delivered video quality was measured in termsof PSNR or distortion as a metric scale of QoS. For a videosequence, PSNR or distortion is defined as the average of thecorresponding assessments over all of the frames. However,visual masking phenomenon is not considered in the PSNRevaluation mode. Therefore, we consider PSNR as an objectiveevaluation metric and MOS as a subjective evaluation mode. Wedescribe and formulate both PSNR and MOS as follows:

1) PSNR: Modeling of PSNR or distortion is done in dif-ferent ways such as: modeling of distortion as a continuousfunction of the content rate [31] and discreet values based onthe number of received layers [33], and modeling PSNR as alinear/piece-wise linear nondecreasing utility function [34]. Wemodel PSNR as a linear function of the bit rate:

ψorg = θ · (rtot − rL0) + ψL0 = θrL1 + ψL0 (42)

where rtot is the total bitrate that is the sum of BL bitrate (rL0)and EL bitrate (rL1), and the R−D model parameter (θ) ischosen according to the spatial–temporal characteristics of thevideo and codec. If we consider rL0 as ratio of packet loss inBL, then PSNR of the reconstructed video at the receive sidewill be

ψrec = θ(rL0 − rL0rL1) + ψL0. (43)

2) MOS: In CR-based networks, the condition of the chan-nels varies over time. Therefore, CUs may experience packetloss because of the low quality of the captured channels orchannel switching as we discussed in the previous sections.Thus, based on frame rate rf , transmission rate rt , packet er-ror rate re , modulation η, and coding scheme σ, the MOS iscalculated as

Υ =a1 + a2rf + a3(ln rt)

1 + a4re + a5(re)2(44)

where re = 11+eη (SINR−σ ) , and the coefficients a1 − a5 are de-

rived as in [35]. Generally, packet delivery fails due to spectrumhandoff and poor channel quality.

V. PERFORMANCE EVALUATION

In this section, the effectiveness of our proposed frameworkis evaluated through a simulation study. According to our objec-tives, i.e., 1) handoff delay minimization, 2) seamless transmis-sion, and 3) QoE improvement, we assess the effectiveness ofour framework in terms of 1) handoff latency, 2) average num-ber of interruptions, and 3) the quality of reconstructed video atthe receiver side, in terms of PSNR in dB and MOS.

TABLE ISIMULATION PARAMETERS

Parameter Value

ϕ 0.2x 5δ 2 −87 dBPd [0.9, 0.99]P fa [0.01, 0.1]rL 0 1 Mbpstslot 100 ms

A

[0.2 0.80.3 0.7

]

TABLE IICHANNEL RANKING PARAMETERS

CH#1 CH#2 CH#3 CH#4 CH#5

P fa 0.091 0.092 0.089 0.091 0.091Pd 0.921 0.942 0.971 0.991 0.901ν 0.114 0.378 0.367 0.075 0.170υ 0.674 0.515 0.515 0.575 0.749λ 0.632 0.532 0.436 0.321 0.632

A. Simulation Setup

Evaluation parameters used to assess the performance ofour framework are listed in Table I. Moreover, sensing period,switching delay, and transmission duration are supposed to be10, 10, and 90 ms, respectively. CU arrival is assumed to fol-low a Poisson arrival mode. We set 10 and 20 HMM statesand assumed that the states of channels follow an exponentialdistribution. The CU estimates the parameters through the dis-cussed techniques in the previous sections and then based onthe estimated metrics the quality of the channels is predicted.The parameters are spectrum sensing accuracy, mean durationof ON and OFF states, and LU arrival rate.

B. Simulation Results and Discussions

We categorize the results in three parts according to our ob-jectives. In each section, we compare our result with the otherrelated schemes as well.

1) Channel Evaluation: In our framework, we consideredchannel ranking index to facilitate the best and most availablechannels to the CU based on its requirements to decrease thehandoff delay and the number of interruptions. For the ranking,we consider various random parameters that are listed in Table IIfor five primary channels as instances. However, we conductsimulation on an example scenario with 20 primary channels.The obtained results can be readily generalized to any numberof primary channels. Based on the parameters, the CU estimatesthe channel quality and ranks in a descending order.

Fig. 4 shows the result of channel ranking. From the figure, itis clear to see that the channel rank changes with ρ in such a waythat channel sensing accuracy is the main metric with smaller ρ,and with a bigger ρ, the channel idle duration is in priority. Forexample, because CH#4 has the highest sensing accuracy whenρ is low, it is the best available channel. While on increasing ρ,


Fig. 4. Channel ranking based on the channel quality estimation.

Fig. 5. Number of handoff vs. LU arrival rate.

CH#5 demonstrates to have higher idle duration time, the rankof this channel increases as well. This kind of ranking procedurefits well for multimedia services due to the intelligent channelallocation that greatly satisfies the user requirements. Moreover,it reduces the number of spectrum handoff as well, since one ofthe reason for handoff was poor channel quality. Such kind ofchannel allocation prevents allocating the poor quality channelto high bandwidth demand applications.

Fig. 5 compares the number of interruptions of our proposedmethod with random channel selection in two proactive and re-active modes, greedy nonpriority, and fair proportional channelallocation algorithms. The proposed method decreases the num-ber of interruptions compared with random selection-proactivemethod, because in the random selection-proactive method, theCU needs to leave the channel before LU arrival where thearrival of LU is predicted based on the channel statistical infor-mation. However, due to common channel sensing errors (e.g.,false-alarm) in which the CU predicts a wrong arrival of LUs, theCU vacates the band where there is no claim from any LU. This

Fig. 6. Average handoff delay.

results into wasting of precious spectrum resources. However,although our method predicts proactively, it performs handoffaction reactively, which resolves the issue of false-alarm. Inaddition, our proposed framework performs better than the ran-dom selection-reactive mode, because it selects already evalu-ated and stable channels, while random selection blindly selectsthe available channels. It results into selecting even poor qualitychannels, and hence the CU needs to perform handoff againwhen the quality of the selected channel becomes unacceptable.

Moreover, comparing with greedy nonpriority channel alloca-tion, although this scheme allocates already evaluated channelsto the CU, the channel allocation is done in an FIFO mode.Thus, because of nonconsideration of the traffic requirementsof the traffic, it is possible that after some time the channelcapacity does not satisfy the requirements and the CU needsto change the channel, hence, increasing the number of inter-ruptions. Finally, compared with the fair proportional channelallocation scheme, because of static channel reservation in thisscheme for each priority class, the quality of channels and CUrequirements are not taken into account. However, our schemeperforms intelligently, where with the help of the ranking indexand the DB, the best available and the most stable channels arefacilitated to the CU based on its requirements and the numberof interruptions decreases significantly. As a result that is shownin Fig. 5, the proposed framework, comparing the average num-ber of interruptions for the other schemes, reduces the numberof interruptions up to 25%.

Fig. 6 compares the spectrum handoff delay of our frameworkwith the random selection scenario in reactive and proactivemethods, greedy nonpriority, and fair proportional schemes. Asdiscussed for the last simulation result, the other schemes areinvolved in a higher number of interruptions and thereby longerhandoff delay. Our scheme has the lowest latency due to theassistance of the ranking index. The reason is that the CU doesnot need to search and compare the quality of available channelat the time of LU arrival. The channel searching and qualityevaluation are already done and in the shortest time possible,hence the CBS allocates the best available channel to the CU in


Fig. 7. Average number of collisions.

minimum duration of time and the CU switches to it in order tocontinue its transmission.

The average number of collisions for the proposed schemecompared with the other schemes is presented in Fig. 7. Collisionmay occur in two cases: 1) when the CU cannot detect the arrivalof LUs and does not leave the channel upon arrival of LUs (miss-detection) and 2) when a channel is detected as free, but it isoccupied by an LU (false alarm).

As shown in the figure, the error probability for randomselection-proactive is more than random selection-reactive, be-cause in addition to the two cases, proactive method is involvedwith the prediction error, as well. It means that if the predic-tion output is not accurate, the CU does not even try to sensethe arrival of the LU. The reactive method is involved withthe two cases, but the involvement is less than the proactivemethod. However, our proposed scheme works better, becauseit employs two techniques for LU detection, namely, predictionand detection. And for the channel selection, it selects only theevaluated channels in the ranking index. In greedy and fair pro-portional schemes since the handoff occurrence is higher thanthe proposed scheme, the chance of collision is higher as well.

2) Evaluation of Video Quality: Finally, the effectiveness ofour proposed framework in terms of quality of the reconstructedvideo is evaluated.

We used Soccer standard SVC test video sequences in704 × 576 pixels resolution (600 frames). JSVM 9.19.7 ref-erence software was used to encode the video sequences with agroup of pictures (GOP) length of 30 frames at 30 frames persecond and the rate of 200 kbps. The video sequence is encodedinto one BL and one EL. We consider a delay deadline for eachtraffic class based on their requirements. For instance, 0.25,0.5 s, no delay deadline for conversational video class, stream-ing video class, and interactive class, respectively. Whereas theconversational video class, video conferencing for example, hasthe lowest delay deadline, it has the highest priority. An instancefor interactive traffic class is nonreal-time data application withno delay restriction.

The end user would decode the received ELs if a significantamount of parity packet of the layer were received otherwise it isdiscarded. Using JSVM reference software, the received data aredecoded and PSNR is measured for the reconstructed video tobe compared with the original video PSNR. In the normal caseswhen the CU is transmitting over a captured channel, both BLand EL can be received by the receiver. Thus, the EL is decodedto improve the quality of the video. However, when a handoffoccurs, normally there should be a service interruption. But inour scheme, before handoff occurrence, the receiver receivesBL code in advance for the estimated handoff delay duration.Hence, during handoff delay the receiver shows the prefetchedBL code. Despite the lower quality, the end user does not realizethe handoff interruption.

Fig. 8 shows the impact of the estimated accuracy of channelsensing and the idle duration on the quality of the reconstructedsequence. Clearly, the quality of the reconstructed video is cor-related with both of the metrics, i.e., sensing accuracy and idleduration. It can be concluded from the figures that the highestvideo quality is obtained from the channels that can facilitatemore opportunities to the CU, and the CU can detect the chan-nels with less sensing false alarm or miss detection. We adjustedthe preference parameter [1.2, 1.8, 2.7, 5]. Increasing ρ meansthat the idle time duration is preferred by the CU, and therefore,sensing accuracy has less impact on the video quality.

Fig. 9 compares MOS vs. LU arrival rate for differenttraffic classes based on the proposed scheme, random chan-nel selection, greedy nonpriority channel allocation, and fairproportional schemes. For conciseness, we present the sim-ulation results for three traffic classes, i.e., conversationalvideo class, streaming video class, and interactive classes tosee the performance of the proposed scheme. And for theother schemes, we just show their performance for streamingclass. The proposed handoff interruption management schemeachieves obvious improvement in average MOS compared withthe other schemes. The conversational video class has lowerMOS compared with the streaming class because of its largerdelay deadline, while the interactive class with no delay deadlinehas minimal changes in its MOS.

The performance of the proposed scheme compared to theother schemes is shown in streaming traffic class. When therate of LU arrival increases, the number of channel switch-ing increases as well. It happens because of several reasonsas explained in the previous sections as well: sensing errors,prediction errors, channel allocation regardless of channel qual-ity, and CU requirements. Anyhow, by increasing the channelswitching, the probability of packet loss increases as well, whichresults in deterioration of the end user satisfaction. In this case,the behavior of the proposed handoff management frameworkis interesting. As shown in Figs. 5 and 6, the other schemes areinvolved with higher number of interruptions and longer hand-off delay. In those schemes, when an LU reclaims the channel,the CU has to stop transmission, find a new available channel,and resume its transmission over it. This process causes interrup-tions, and hence decreases QoE. However, our scheme performsintelligently. By increasing the number of LU arrival, althoughthe number of spectrum handoff increases, at least the end usercan see the video sequence (BL only) even with a lower quality.


Fig. 8. PSNR of the reconstructed video based on channel quality. (a) ρ = 1.2. (b) ρ = 1.8. (c) ρ = 2.7. (d) ρ = 5.

Fig. 9. MOS of different traffic classes vs. the arrival rate of LUs.

Thereby, the MOS of the proposed scheme is higher than theother schemes.

VI. APPLICATION SCENARIOS AND REQUIREMENTS

The proposed framework is a promising solution to be ap-plied in many distinct application areas that are envisioned for5G where current wireless networks will struggle to deliver.Specifically, CR will facilitate various versions of radio tech-nology to share the same spectrum band in a more efficient way.The followings are just some instances of the target applicationsthat do not only need higher data rate but need an improvedQoE:

1) Internet of media things (IoMT), which is the collection ofinterfaces, protocols, and associated media-related infor-mation representations that enable advanced services andapplications based on human-to-machine and machine tomachine (M2M) interaction in physical and virtual envi-ronment. Media things refers to the things with at leastone of audio and/or visual sensing and/or actuating capa-bilities.

2) Mobile health care applications, such as smartphone-based applications for the monitoring and treatment oflong-term medical conditions from mental health prob-lems to diabetes. Even more sophisticated application can

be envisioned such as far-distance robotic surgery usinghaptic technology.

3) Virtual reality (VR) and augmented reality (AR) appli-cations, which is for military application as an instance(superimposing a digital view on a physical view). Suchapplications need large volumes of data for the end-userdevices, e.g., headsets and displays with very low delayand high reliability.

4) Many other applications such as smart cities, moving net-works, industrial telecontrol, e-transportation, etc.

New applications and key design principles of the systemlead to many stringent requirements that are needed to meet thecoming mobile broadband system. Some of the non-negligiblerequirements are

1) The system should be prepared to service a tremendousnumber mobile users.

2) The radio latency should be lower than one millisecondin order to achieve fast procedure response time and highdata rates with low cost.

3) New evolution of battery technology is expected to in-crease better life time.

4) Very low cost of devices needs to be ensured.5) A peak data rates of 10 Gbps is expected and even more

data rate for cell-edges.

VII. CONCLUSION

In this paper, we addressed the issue of efficient channel al-location based on QoE requirements as well as service interrup-tions caused by spectrum handoff, specifically for multimediaapplications. The proposed framework efficiently evaluates thequality of available channels based on sensing accuracy, idleduration, and the arrival rate of LUs. In our priority-based chan-nel allocation scenario, according to the quality of the availablechannels, the CBS maintains a channel ranking index to pro-vide the CUs the most reliable channel based on the qualityof the available channels and QoE requirements of the var-ious traffic classes, hence, improving channel utilization anddecreasing the number of interruptions. Then, we developed ascheme by employing HMM to estimate the status of the futuretime slot, handoff occurrence for instance. Handoff predictionsignificantly reduces interference to the LUs and enhance spec-tral efficiency as well. After that, by taking into account theunavoidable spectrum handoff and non-negligible delay caused


by handoff, we proposed a handoff interruption managementscheme. Our framework interestingly is able to provide seam-less multimedia service over cognitive 5G cellular networks.

REFERENCES

[1] L. Lin, “Unlicensed spectrum will improve customer 4G mobile experi-ence,” Vodafone Webpage, Nov. 2015. [Online]. Available: http://www.vodafone.com/

[2] Cisco VNI Forecast, “CISCO visual networking index: Global mobiledata traffic forecast update 2015–2020,” CISCO Visual Networking Index,Feb. 2016.

[3] D. Liu et al., “User association in 5G networks: A survey and an out-look,” IEEE Commun. Surveys Tuts., vol. 18, no. 2, pp. 1018–1044,Jan. 2016.

[4] N. Rupasinghe and I. Guvenc, “Licensed-assisted access for WiFi-LTEcoexistence in the unlicensed spectrum,” in Proc. IEEE Globecom Work-shops, Dec. 2014, pp. 894–899.

[5] D. Marcin and M. Szydelko, “A unified traffic steering framework for LTEradio access network coordination,” IEEE Commun. Mag., vol. 54, no. 7,pp. 84–92, Jul. 2016.

[6] B. Fernando, S. K. Ray, and J. A. Gutierrez, “Understanding the currentoperation and future roles of wireless networks: Co-existence, competitionand co-operation in the unlicensed spectrum bands,” IEEE J. Select. AreasCommun., vol. 34, no. 11, pp. 2829–2837, Nov. 2016.

[7] W. Roh et al., “Millimeter-wave beamforming as an enabling technol-ogy for 5G cellular communications: Theoretical feasibility and pro-totype results,” IEEE Commun. Mag., vol. 52, no. 2, pp. 106–113,Feb. 2014.

[8] X. Hong, J. Wang, C. Wang, and J. Shi, “Cognitive radio in 5G: A per-spective on energy-spectral efficiency trade-off,” IEEE Commun. Mag.,vol. 52, no. 7, pp. 46–53, Jul. 2014.

[9] 5G Vision, “The 5G infrastructure public private partnership: The nextgeneration of communication networks and services.” Feb. 2015. [Online].Available: https://5g-ppp.eu/

[10] C. Langtry, “ITU-R activities on 5G,” in Proc. IEEE World Forum InternetThings, Dec. 2016.

[11] W. Jiang and H. Cao, “SIG on cognitive radio for 5G.” Dec. 2015. [Online].Available: http://cn.committees.comsoc.org/

[12] A. Lertsinsrubtavee, N. Malouch, and S. Fdida, “Spectrum handoffstrategy using cumulative probability in cognitive radio networks,” inProc. Int. Congr. Ultra-Modern Telecommun. Control Syst., Oct. 2011,pp. 1–7.

[13] F. Sheikholeslami, M. Nasiri-Kenari, and F. Ashtiani, “Optimal proba-bilistic initial and target channel selection for spectrum handoff in cog-nitive radio networks,” IEEE Trans. Wireless Commun., vol. 14, no. 1,pp. 570–548, Jan. 2015.

[14] L. C. Wang and A. Chen, “On the performance of spectrum handoff forlink maintenance in cognitive radio,” in Proc. IEEE 3rd Int. Symp. WirelessPervasive Comput., May 2008, pp. 670–674.

[15] C. W. Wang and L. C. Wang, “Analysis of reactive spectrum handoffin cognitive radio networks,” IEEE J. Select. Areas Commun., vol. 30,no. 10, pp. 2016–2028, Nov. 2012.

[16] L. C. Wang and C. W. Wang, “Spectrum handoff for cognitive radionetworks: Reactive-sensing or proactive-sensing,” in Proc. 27th IEEE Int.Perform. Comput. Commun. Conf., Dec. 2008, pp. 343–348.

[17] S. Yoon and E. Ekici, “Voluntary spectrum handoff: A novel approachto spectrum management in CRNs,” in Proc. IEEE Int. Conf. Commun.,May 2010, pp. 1–5.

[18] W. Zhang and C. K. Yeo, “Sequential sensing based spectrum handoffin cognitive radio networks with multiple users,” Comput. Netw., vol. 58,pp. 87–98, Jan. 2014.

[19] Wireless Regional Area Network (WRAN) Specific Requirements Part 22:Cognitive Wireless RAN Medium Access Control (MAC) and PhysicalLayer (PHY) Specification: Policies and Procedures for Operation in theTV Bands, IEEE Std. 802.22, Jul. 2011.

[20] M. J. Piran, M. E. Ahmed, A. Ali, J. B. Song, and D. Y. Suh, “Channelallocation based on content characteristics for video transmission in time-domain-based multichannel cognitive radio networks,” Mobile Inf. Syst.,vol. 2015, 2015, Art. ID. 934715.

[21] M. Jin, Q. Guo, J. Xi, Y. Li, Y. Yu, and D. Huang, “Spectrum sensingusing weighted covariance matrix in Rayleigh fading channels,” IEEETrans. Veh. Technol., vol. 64, no. 11, pp. 5137–5148, Nov. 2015.

[22] S. Atapattu, C. Tellambura, and H. Jiang, “Relay based cooperative spec-trum sensing in cognitive radio networks,” in Proc. IEEE Global Telecom-mun. Conf., Dec. 2009, pp. 1–5.

[23] D. J. Lee and W. Y. Yeo, “Channel availability analysis of spectrumhandoff in cognitive radio networks,” IEEE Commun. Lett., vol. 19, no. 3,pp. 435–438, Jan. 2015.

[24] B. Vujicic, N. Cackov, S. VujicCic, and L. Trajkovic, “Modelingand characterization of traffic in public safety wireless networks,” inProc. Int. Symp. Perform. Eval. Comput. Telecommun. Syst., Jul. 2005,pp. 214–223.

[25] G. Yuan, R. C. Grammenos, and Y. Yang, “Performance analysis of selec-tive opportunistic spectrum access with traffic prediction,” IEEE Trans.Veh. Technol., vol. 59, no. 4. pp. 1949–1959, Dec. 2009.

[26] X. Xing, T. Jing, Y. Huo, H. Li, and X. Cheng, “Channel quality predictionbased on Bayesian inference in cognitive radio networks,” in Proc. IEEEINFOCOM, Apr. 2013, pp. 1465–1473.

[27] S. R. Eddy, “Hidden Markov models,” Current Opinion Struct. Biol.,vol. 6, no. 3, pp. 361–365, Jun. 1996.

[28] L. R. Welch, “Hidden Markov models and the Baum–Welch algorithm,”IEEE Inf. Theory Soc. Newslett., vol. 53, no. 4, pp. 10–13, Dec. 2003.

[29] A. P. Dempster, N. M. Laid, and D. B. Rubin, “Maximum likelihoodfrom incomplete data via the EM algorithm,” J. Roy. Statist. Soc., vol. 39,no. 1, pp. 1–38, Apr. 1977.

[30] J. Lee and H. Park, “Channel prediction-based channel allocation schemefor multichannel cognitive radio networks,” J. Commun. Netw., vol. 16,no. 2, pp. 209–216, Apr. 2014.

[31] H. Schwarz, D. Marpe, and T. Wiegand, “Overview of the scalable videocoding extension of the H. 264/AVC standard,” IEEE Trans. Circuits Syst.Video Technol., vol. 17, no. 9, pp. 1103–1120, Sep. 2007.

[32] S. Ullah, K. Thar, and C. S. Hong, “Management of scalable video stream-ing in information centric networking,” Multimedia Tools Appl., vol. 2016,pp. 1–28, 2016.

[33] T. Schierl, S. Johansen, A. Perkis, and T. Wiegand, “Rateless scalablevideo coding for overlay multisource streaming in MANETs,” J. VisualCommun. Image Represent., vol. 19, no. 8, pp. 500–507, Dec. 2008.

[34] M. Schaar, S. Krishnamachari, S. Choi, and X. Xu, “Adaptive cross-layerprotection strategies for robust scalable video transmission over 802.11WLANs,” IEEE J. Select. Areas Commun., vol. 21, no. 10, pp. 1752–1763,Dec. 2003.

[35] A. Khan, L. Sun, E. Jammeh, and E. Ifeachor, “Quality of experience-driven adaptation scheme for video applications over wireless networks,”IET Commun., vol. 4, pp. 1337–1347, Jul. 2010.

[36] L. Chen, W. Chen, B. Wang, X. Zhang, H. Chen, and D. Yang, “System-level simulation methodology and platform for mobile cellular systems,”IEEE Commun. Mag., vol. 49, no. 7, pp. 148–155, Jul. 2011.

[37] C. Wang, H. Chen, Q. Yin, A. Feng, and A. F. Molisch, “Multi-user two-way relay networks with distributed beamforming,” IEEE Trans. WirelessCommun., vol. 10. no. 10, pp. 3460–3471, Aug. 2011.

[38] Y. Li, Q. Yin, L. Sun, H. Chen, and H. Wang, “A channel quality metricin opportunistic selection with outdated CSI over Nakagami-m fadingchannels,” IEEE Trans. Veh. Technol., vol. 61, no. 3, pp. 1427–1432,Jan. 2012.

Md. Jalil Piran (S’10–M’16) received the M.Sc. de-gree in computer science and engineering from Jawa-harlal Nehru Technological University, India, and thePh.D. degree in electronics and radio engineeringfrom Kyung Hee University, Seoul, South Korea, in2011 and 2016, respectively. He is currently a Post-doctoral Researcher with the Department of Com-puter Engineering, Kyung Hee University.

He was a Research Assistant with Media Labora-tory, Kyung Hee University, as well as the Commu-nication Research Center, International Institute of

Information Technology, India. His research interests include mobile and wire-less communication, cognitive radio networks, and multimedia communication.

Dr. Piran has been a delegate from South Korea for the ISO/IEC MPEG(Moving Picture Experts Group) Forum since 2013.


Nguyen H. Tran (S’10–M’11) received the B.S. de-gree from Hochiminh City University of Technol-ogy, Vietnam, and the Ph.D degree in electrical andcomputer engineering, from Kyung Hee University(KHU),in 2005 and 2011, respectively.

Since 2012, he has been an Assistant Profes-sor with the Department of Computer Science andEngineering, KHU. His research interest is apply-ing analytic techniques of optimization, game theory,and stochastic modelling to cutting-edge applicationssuch as cloud and mobile-edge computing, datacen-

ters, heterogeneous wireless networks, and big data for networks.Dr. Tran received the best KHU thesis award in engineering in 2011 and sev-

eral best paper awards, including IEEE ICC 2016, APNOMS 2016, and IEEEICCS 2016. He is the Editor of IEEE TRANSACTIONS ON GREEN COMMUNICA-TIONS AND NETWORKING.

Doug Young Suh (S’89–M’90) received the B.Sc.degree in nuclear engineering from the Department ofNuclear Engineering, Seoul University, Seoul, SouthKorea, and the M.Sc. in electrical engineering andPh.D. degrees in electrical engineering from the De-partment of Electrical Engineering, Georgia Instituteof Technology, Atlanta, GA, USA, in 1980, 1986, and1990, respectively.

In September 1990, he joined the Korea Academyof Industry and Technology and conducted researchon high-definition television until 1992. Since Febru-

ary 1992, he has been a Professor with the College of Electronics and Informa-tion Engineering, Kyung Hee University, Seoul. His research interests includenetworked video and video compression.

Dr. Suh has been a Korean delegate for the ISO/IEC Moving Picture ExpertsGroup since 1996.

Ju Bin Song (S’99–M’02) received the B.Sc. andM.Sc. degrees in electronics engineering from theDepartment of Electronics, Kyung Hee University,Seoul, South Korea, in 1987 and 1989, respectively,and the Ph.D. degree in electronics and electricalengineering from the Department of Electronic andElectrical Engineering, University College London(UCL), London, U.K., in 2001.

From 1992 to 1997, he was a Senior Researcherwith the Electronics and Telecommunications Re-search Institute, Daejeon, South Korea. He was a

Research Fellow with the Department of Electronic and Electrical Engineer-ing, UCL, in 2001. From 2002 to 2003, he was an Assistant Professor with theSchool of Information and Computer Engineering, Hanbat National University,Daejeon. From 2009 to 2010, he was a Visiting Professor with the Depart-ment of Electrical Engineering and Computer Science, University of Houston,Houston, TX, USA. He has been a Professor and the Department Head withthe Department of Electronics and Radio Engineering, Kyung Hee University,since 2003. He has been the Vice Dean with the School of Electronics andInformation, Kyung Hee University, since 2015. His current research interestsinclude resource allocation in communication systems and networks, coopera-tive communications, game theory, optimization, cognitive radio networks, andsmart grid.

Dr. Song served as a technical program member for international confer-ences and is the Workshop Chair for the Seventh EAI International Conferenceon Game Theory for Networks for 2017. He received the Kyung Hee UniversityBest Teaching Award in 2004 and 2012.

Choong Seon Hong (S’95–M’97–SM’11) receivedthe B.S. and M.S. degrees in electronic engineeringfrom Kyung Hee University, Seoul, South Korea, in1983 and 1985, respectively, and the Ph.D. degree ininformation and computer science from Keio Univer-sity, Minato, Japan, in 1997.

In 1988, he joined Korea Telecom (KT), wherehe worked on broadband networks as a member ofTechnical Staff. In September 1993, he joined KeioUniversity. He worked for the TelecommunicationsNetwork Laboratory, KT, as a senior member of Tech-

nical Staff and was the Director of the Networking Research Team until August1999. Since September 1999, he has been a Professor with the Departmentof Computer Science and Engineering, Kyung Hee University. His researchinterests include future internet, ad hoc networks, network management, andnetwork security.

Dr. Hong is a member of ACM, IEICE, IPSJ, KIISE, KICS, KIPS, andOSIA. He has served as the General Chair, a Technical Program CommitteeChair/Member, or an Organizing Committee Member for international confer-ences such as NOMS, IM, APNOMS, E2EMON, CCNC, ADSN, ICPP, DIM,WISA, BcN, TINA, SAINT, and ICOIN. In addition, he is currently an AssociateEditor of the IEEE TRANSACTIONS ON NETWORK AND SERVICE MANAGEMENT,the International Journal of Network Management, and the Journal of Com-munications and Networks and is an Associate Technical Editor of the IEEECommunications Magazine.

Zhu Han (S’01–M’04–SM’09–F’14) received theB.S. degree in electronic engineering from TsinghuaUniversity, Beijing, China, in 1997, and the M.S. andPh.D. degrees in electrical and computer engineeringfrom the University of Maryland, College Park, MD,USA, in 1999 and 2003, respectively.

From 2000 to 2002, he was an R&D Engineer ofJDSU, Germantown, Maryland. From 2003 to 2006,he was a Research Associate with the University ofMaryland. From 2006 to 2008, he was an AssistantProfessor with Boise State University, Idaho. Cur-

rently, he is a Professor in the Electrical and Computer Engineering Departmentas well as in the Computer Science Department, University of Houston, Hous-ton, TX, USA. His research interests include wireless resource allocation andmanagement, wireless communications and networking, game theory, big dataanalysis, security, and smart grid.

Dr. Han received an National Science Foundation Career Award in 2010,the Fred W. Ellersick Prize of the IEEE Communication Society in 2011, theEURASIP Best Paper Award for the Journal on Advances in Signal Processingin 2015, IEEE Leonard G. Abraham Prize in the field of Communications Sys-tems (best paper award in IEEE JSAC) in 2016, and several best paper awardsin IEEE conferences. He is currently an IEEE Communications Society Distin-guished Lecturer.

ieee transactions on vehicular technology, vol. 66, no. 7, july 2017 6569 qoe...

Documents