The explosive growth of video-related services, which are bandwidth-hungry and delay-sensitive, and the need for

quality of experience (QoE) provisioning for users post a great challenge for the depleting radio spectrum. Cognitive radio (CR), as a powerful technology for sharing spectrum, is a promising solution to the above problem. A CR accesses spec-trum in an opportunistic manner. The basic idea is to allow CR users (CUs) to access the licensed spectrum, with limited interference to primary users (PUs). Spectrum sharing can be achieved over long timescales (e.g., in the form of spectrum leasing) or over much shorter timescales (e.g., as in dynamic spectrum access).

Video streaming over CR networks (CRNs) has attracted great interest in recent years, and considerable advances have been made. As illustrated in Fig. 1, when there is a shortage of available spectrum for CUs, they can take advantage of the underutilized licensed and unlicensed bands to enable various video services, such as online gaming, video conferencing, and video-on-demand streaming. Even if a CU owns a licensed spectrum, the user can still employ CR to obtain more band-width in order to support a higher data rate to improve the QoE of the video session. Compared to users in traditional networks, CUs can adjust their QoE requirements according to the dynamic channel conditions in CRNs, or they can adjust their transmission parameters or strategies, such as power, modulation and coding scheme, and relay selection, to adapt to varying network conditions. Since spectrum allocation is not

fixed in CRNs, it is possible to optimize spectrum allocation whenever channel conditions change in order to achieve a global optimal QoE for multiple CUs.

In this article, we first introduce the preliminaries on video streaming over CRNs (VCRN), including spectrum sensing and QoE modeling. We then examine the technical challenges and review existing solutions in QoE-driven VCRN. Next, the unique features and challenges of QoE optimization in CRNs are examined. Finally, we discuss open problems and conclude this article.

Preliminaries for VCRNBefore planning on how to stream videos over a CRN, there are two major tasks to be accomplished: identifying the avail-able channels for video transmission and defining the QoE metric to be incorporated into the transmission strategy and for performance evaluation.

Spectrum SensingUsually a CR implements the cognitive loop as illustrated in Fig. 2. Through spectrum sensing, the CR is aware of the radio environment. The decision making unit, usually in the form of a cognitive engine, then decides to reconfigure the software defined radio as the radio environment changes to maintain the QoE of CUs and avoid interfering with PUs.

Spectrum sensing is critical in QoE-driven VCRN in that it identifies possible transmission opportunities. A reliable spec-trum sensing result should not only discover the available idle spectrum for CUs, but should also not misidentify the unavail-able spectrum as available spectrum for CU access in order to avoid interference to PUs. In addition, spectrum sensing is neatly coupled with the other aspects in QoE-driven VCRN.

Trade-off between Reliability and Delay: One of the most important factors in determining QoE is the data rate. Video streaming is a bandwidth-hungry service, so a high data rate may be needed to achieve the target QoE. CUs have to iden-tify a sufficient amount of idle channels with spectrum sens-ing. However, trying to identify more idle channels usually requires a longer spectrum sensing time, which shortens the

20 IEEE Network • November/December 2015

AbstractVideo streaming, as a bandwidth-hungry and delay-sensitive application, can be supported with the Cognitive Radio (CR) technology, which is effective in exploiting the underutilized allocated spectrum. In this article, we examine the challenges of quality of experience (QoE) driven video streaming over CR Networks (CRN) and review existing solutions proposed in the literature. We also discuss the unique fea-tures in the optimization of QoE-driven video streaming over CRNs. This article is concluded with a discussion of several open problems that call for substantial future research efforts.

A Survey of QoE-Driven Video Streaming over Cognitive Radio Networks

Zhifeng He, Shiwen Mao, and Tao Jiang

0890-8044/15/$25.00 © 2015 IEEE

Zhifeng He and Shiwen Mao are with Auburn University. Tao Jiang is with Huazhong University of Science and Technology. This work is supported in part by the US National Science Founda-tion (NSF) under Grant CNS-1247955 and the Wireless Engineering Research and Education Center (WEREC) at Auburn University, and in part by the Major State Basic Research Development Program of China (973 Program) with Grant number 2013CB329006, the National Science Foundation of China with Grant 61428104, and Joint Special-ized Research Fund for the Doctoral Program of Higher Education (SRFDP) and Research Grants Council Earmarked Research Grants (RGC ERG) with Grant 20130142140002.

video transmission time and causes a larger trans-mission delay. Therefore, it is necessary to achieve a trade-off between spectrum sensing time and video transmission time. This feature distinguish-es VCRN from other applications that are not as bandwidth-hungry, such as voice, text messaging, and data file transmission.

Interaction with Other Factors: QoE is also dependent on other factors such as bit error rate and delay, which are affected by the chosen video modulation and coding scheme, channel conditions, and so on. For example, if more time is spent on spectrum sensing to find idle channels with better conditions, we may choose a more aggressive modulation and coding scheme to increase the transmission rate. However, as discussed previously, greater delay is going to occur as more time is spent on spectrum sensing, and therefore, QoE may be compromised. Therefore, it is important to jointly optimize multi-ple design factors, such as spectrum sensing, mod-ulation and coding scheme selection, and relay selection, for QoE provisioning.

QoE ModelingQoE is a metric to evaluate the viewer’s satisfac-tion with perceived video quality. An adequate QoE model should accurately reflect the performance of the system and the experience of the users, thus providing useful guidelines for configuring video transmission parameters [1]. QoE mod-eling is especially important for VCRN, where a CR has great flexibility to adjust its operating parameters to adapt to the conditions of the CRN.

QoE Modeling in Traditional Networks: In subjective tests, human viewers score the quality of video they have viewed, and the scores are processed into either the mean opinion score (MOS) or degradation MOS (DMOS) metric. MOS is the average score of a tested video, and DMOS is obtained by averaging the arithmetic difference between the scores of the source video and the processed video. It has been found that subjective tests suffer from drawbacks such as limited asses-sors, limited distortion types, and high cost.

In the objective quality model, a metric is calculated as a function of QoS parameters and external factors. It is often modeled as a function of network parameters (e.g., packet loss rate, transmission delay, and jitter), coding parameters (e.g., bit rate and frame rate), and some external factors such as video content type, viewer demography, and device type [1, 2]. Common objective quality models include moving picture quality metric (MPQM), perceptual video quality measure (PVQM), and visual signal-to-noise ratio (VSNR), among others [1].

QoE Modeling in CRNs: Some QoE models, such as MOS and DMOS, have been adopted in the literature on video streaming over CRNs, and the adopted models in CRNs have no difference from those developed for traditional networks. In addition to consideration of dynamic channel availability in CRNs, the QoE of a video session should also include consideration of the delay in recovering the lost video data caused by channel outages, which determines the amount of time needed to deliver the video sequence. Possible solutions to the problem of uncertainty of channel availability may include, but are not limited to, backup channels, spectrum handoff, spectrum sharing, and scalable video coding (SVC). However, different CUs may have different capabilities to

support such techniques, so it is important to consider what CUs can afford when optimizing QoE-driven video streaming over CRNs.

One feature that distinguishes QoE-driven VCRN from that of traditional networks is the possible interference caused to PUs when CUs are trying to utilize idle licensed spectrum. Therefore, controlling the interference with PUs should also be integrated into the QoE metric in some suitable form for VCRNs.

Unique Features of QoE-Driven VCRNFrom the above discussions, it can be seen that there are some unique features that distinguish QoE-driven VCRN from that of traditional wireless networks. We briefly discuss these fea-tures in this section. • Spectrum sensing: In regard to QoE-driven VCRN, one

needs to consider the interaction of spectrum sensing and other design factors such as channel allocation, and modula-tion and coding scheme, which bring about a new dimension of uncertainty and complexity.

• Spectrum handoff: In CRNs, CUs need to perform spectrum

IEEE Network • November/December 2015 21

Figure 1. Illustration of video streaming over CRNs.

Yout

ube

Video

gamm

ing

Video chat

Licensed band II

Licensed band I

Primary user

Primary user

Unlicensed band

Primary networks Cognitive radio network

CR user

CR user

CR user

Internet

Spectrum broker Other

cognitive radio

networks

Figure 2. The cognitive loop.

Cognitive engine:decision making for resource allocation

Spectrum sensing Software defined radio

Radio environment

handoff in order to switch to other available channels to resume transmission as soon as possible when a PU reap-pears on a channel. The authors in [3] propose a learn-ing-based QoE-driven spectrum handoff scheme for VCRN.

• Interference control: VCRN needs to control the interfer-ence to PUs when CUs are trying to access the licensed spectrum. Power adaptation is an effective approach to con-straining the interference [4].

• Admission control: In CRNs, admission control would be useful for limiting the number of active video sessions such that each of them can have a satisfactory QoE [5]. This is a more challenging problem than that in traditional net-works, since the channel capacity is highly dynamic due to PU transmissions. Alternatively, the video sessions should be prioritized as in [3, 6], so a lower-priority session may be blocked to ensure higher-priority sessions have access to the channels.

• Machine learning: As shown in Fig. 2, a CR should learn the radio environment and then make adaptation decisions to tune the radio to the changing environment. This provides a perfect setting for the application of machine learning techniques. In addition, there have been some recent works on machine-learning-based QoE modeling [1]. A potential topic for future research is applying machine learning tech-niques in QoE-driven VCRN.

Technical Challenges and Existing SolutionsChannel Allocation and RoutingVideo streaming a is bandwidth-hungry and delay-sensitive service that should be carried by channels in good condition and routes with low delay. The available channels found by CUs may be significantly different in frequency, channel gain, and available time for transmission, which define the channel conditions. In a practical wireless network, CUs may have different channel gains with respect to the same channel. Routes with different numbers of hops may cause different delays. Therefore, effective channel selection and routing algorithms should be designed for VCRN. The authors in [7] study the problem of streaming multi-user videos over the downlink of a CRN, aiming to maximize the overall QoE of all CUs. Channel allocation to different types of videos is based on the fact that different types of videos may have different parameter coefficients in the same QoE model. For example, videos of low motion (e.g., news) have a higher QoE than videos of high motion (e.g., football) [2]. This fact is similar to the spatial diversity of the spectrum resource and thus can be exploited in the optimization of channel alloca-tion in CRNs.

In most cases, frequency reuse among different sessions may be implemented to support multi-user video stream-ing. Channel assignment among different sessions should be carefully designed so that interference among the video sessions are kept at a tolerable level and frequency reuse can be maximized. This problem is addressed in [8]. In this article, the authors assume that multiple video sessions are streamed sequentially via multiple hops over the CRN, and frequency is reused to support multiple sessions concur-rently. For a newly initiated video session, given the routing and channel assignment strategy of the existing sessions, the authors calculate which routing and channel assign-ment strategy for the new session can maximize the fre-quency reuse index while limiting the interference caused to other sessions. Since the bandwidth consumption for each sequentially arriving video session is minimized one at a time, the network can support more video sessions with a fixed amount of radio resources.

Video Coding and Adaptive Modulation Coding

The authors in [9] study the problem of streaming fine gran-ularity scalability (FGS) videos in CRNs. FGS video is espe-cially suitable for CRN since an FGS video stream is divided into a base layer and an enhancement layer. The base layer contains the most important information of the video stream, and the reconstructed video sequence can be guaranteed to have the quality of the basic level even if only the base layer is received. The enhancement layer contains less important information, but receiving the enhancement layer improves the quality. If we truncate the enhancement layer at an arbi-trary bit location, the remaining bits are still decodable, and thus the decoded video quality of a CU will be commensurate with its channel quality. Therefore, the authors propose to further divide the enhancement layer into multiple sublayers of different priorities, and a sublayer with lower priority will be transmitted only when the sublayers of higher priority are all acknowledged.

To deal with the dynamic transmission environment in the CRN, the authors of [10] propose using adaptive modulation and coding (AMC) in video streaming. The basic idea is that when channel conditions are good, AMC schemes with larger constellation sizes and higher channel coding rates are pre-ferred. The goal is to minimize the expected video distortion while keeping the packet loss at an acceptable level, consid-ering that channel availability may quickly change over time. Fountain code is used in [11] to combat packet loss in CRNs. With fountain code, the source generates an endless supply of packets using K input packets. As long as at least any N K packets are received, the original K packets can be recovered with an error probability. Thus, N affects the overhead and delay.

The authors have derived the optimal N to maximize the spectrum efficiency of a video session with the maximum tol-erable delay.

Admission ControlDue to the time-varying channel availability in CRNs, it is possible that there is no sufficient amount of idle spectrum to support all requested video sessions. Therefore, it is important to incorporate admission control. This can be achieved by a comparison of the estimated capacity that can be harvested with the capacity needed to satisfy the QoE requirements of all the CUs. If there are no sufficient spectrum opportunities, new video session requests will be blocked. Alternative priority can be introduced to treat the video sessions differently.

Considering the fact that different CUs may have different QoE requirements, the authors in [3] propose a priority-based queuing model in which CUs with higher priority are allowed to stream their video sessions over lower-priority CUs. When the network resource is constrained, higher-priority CUs have more channel access opportunities, and lower-priority CUs are blocked. A similar approach is adopted in [6] where the authors notice that in resource allocation to multiple video sessions, some types of videos may be more delay-sensitive than others. In the proposed scheme, the CU requiring a more delay-sensitive video session is assigned a higher priority, and vice versa. When the number of available channels is not enough to support all the CUs, the available channels are first assigned to higher-priority CUs.

Buffer Storage ManagementDue to the dynamic channel status in CRNs, the receiver may not be able to receive video frames consistently from the air. However, the video player would keep on consuming the received frames in the resource buffer (RB). The RB will soon

IEEE Network • November/December 201522

be depleted (i.e., buffer underflow) if the channel is bad or unavailable for an extended period of time. Such buffer under-flow seriously degrades the QoE perceived by the CUs.

Some solutions have been proposed to solve this problem.In [12], the authors propose lowering the playout speed of the video sequence to a certain extent to decrease the probability of playout outage, when the RB becomes hungry. This is based on the observation that varying the video playout rate by 25 to 50 percent could be unnoticeable and therefore acceptable. Moreover, since the perceptual quality of videos with drastic motion degrades more than that of videos with little motion when the playout rate is decreased, the authors also propose to adjust the playout rate based on the motion intensity of the videos.

Alternatively, the authors of [13] solve this problem by allo-cating channels according to the buffer storage of the CUs. CUs may have different channel data rates and different buf-fer storage sizes, which result in different abilities to tolerate network dynamics. When a CU with more buffer storage is competing with a CU with less buffer storage to access a chan-nel, we may prefer to allocate the channel to the one with less buffer storage.

Interference ManagementIn CRNs, frequency reuse may be applied among multiple CUs since sometimes there may not be enough channels to support all CUs. For a video packet to be correctly decod-ed by the receiver, it is required that the signal-to-interfer-ence-plus-noise ratio (SINR) of the packet at the receiver be greater than a threshold, which is dependent on the signal power, noise power, and interference caused by the transmis-sion of other video sessions.

Power Adaptation — Power adaptation can also be considered as resource management. One way to ensure a desired SINR at the receiver is to control the transmit power of the target session and that of other video sessions in order to guarantee the strength of the desired signal and limit the interference to an acceptable level.

When there are concurrent video sessions in a CRN, it is necessary to control the power of each transmission in order to constrain the mutual interference, and meanwhile achieve the required SINR for each transmission. The authors of [14]

propose to use relay between the source and the destination so that a better transmission route is built with smaller path loss, and the transmit powers of the source and relay are optimized so that the target SINR can be achieved.

In addition to leveraging the performance of a CRN, power adaptation also plays an import-ant role in keeping the interference to PUs at an acceptable level. In [4], the authors assume that CUs and PUs can operate on the same channel as long as the interference imposed on the PUs by CUs is lower than a threshold. The allocat-ed power is optimized by maximizing the overall quality of all CUs under interference constraints.

Interference Alignment — Another way to miti-gate the interference from other transmissions is interference alignment. The idea is that multiple senders jointly encode signals to multiple receiv-ers according to the channel conditions so that interfering signals will be cancelled out at each receiver, and each receiver is then able to decode its desired information.

QoE Optimization in VCRNCross-Layer OptimizationWhile the above-mentioned techniques seem to serve differ-ent purposes and work on different layers, it is worth noting that they may affect or even conflict with each other. For instance, channel assignment determines the channel condi-tion of the data links, based on which an appropriate mod-ulation and coding scheme should be applied. The content type of the video sequence and the buffer storage at the CR receiver can also affect the resource management strategy [6, 12]. From the existing QoE models, we can see that fac-tors at different layers contribute to the user experience [1]. Therefore, factors from different layers should be jointly considered in designing the CR video streaming strategy, as shown in Fig. 3.

Most of the literature on VCRN is focused on QoE-driv-en cross-layer optimization. In [11], two design objectives, the transmission success probability and spectrum efficien-cy, are modeled as functions of channel condition param-eters and channel coding overhead. The best combination of channel and coding schemes is selected to achieve a trade-off between the two objectives. A different approach is developed in [4] by considering power allocation, where the authors jointly optimize the transmit power and chan-nel allocation for different video sessions for CR receivers so that the weighted sum quality of all CUs is maximized while keeping the possible interference to PUs below a threshold.

Generally speaking, the more factors taken into consid-eration, the better the performance that can be achieved. However, this also results in a more complex problem that often appears to be a mixed integer nonlinear non-convex optimization problem, which is NP-hard in general. Therefore, heuristic algorithms are often applied to solve the cross-layer optimization problems.

Trade-off among Multiple ObjectivesIn QoE-driven VCRN, we usually deal with multiple objec-

tives. For example, we may want to maximize the QoE of CUs, while minimizing the interference to PUs. Generally these objectives are like the different sides of a coin and cannot be achieved at the same time; a compromise has to be made.

IEEE Network • November/December 2015 23

Figure 3. Illustration of the cross-layer optimization framework.

Application layer

Transportation layer

Network layer

Link layer

Physical Layer

Adaptive video playout, video content type

Error detection & Correction,

ARQ

Routing information

Channel conditions, spectrum

access

Modulation, coding, power

control

Design constraints

(delay bound, power budget,

interference limit)

Design

Parameters of different layers

Cross-layer optimization (QoE models, algorithms)

Optimization

Output (strategy)

Input

CU Performance and PU Performance Trade-off — In overlay CRNs, where CUs access the licensed channels in an opportu-nistic manner, spectrum sensing time and the reliability of the sensing result can affect the utility of CUs as well as that of PUs. Adjusting the spectrum sensing parameters may improve the utility of CUs but may also lead to degradation of the util-ity of PUs, and vice versa. How to balance the utility of CUs and that of PUs by tuning the spectrum sensing parameters is studied in [3, 7].

In underlay CRNs, where CUs simultaneously transmit on the same licensed channel as PUs, CUs control their transmit powers so that the interference to PUs is kept at an acceptable level, as proposed in [4].

CU Performance and Delay Trade-off — To combat packet losses in wireless networks, especially in CRNs where the channels are not always available over time, one should encode the video sequences using a strong coding scheme. However, apply-ing strong coding schemes usually means high redundancy as more video data packets are generated and transmitted to the receiver. It may take a longer period of time for the receiver to receive and decode the video sequence.

The quantization step size in the video coder can also affect the distortion of reconstructed video and the delay of recon-structing the video, as pointed out in [10]. The quantization step size can be optimized to achieve a trade-off between dis-tortion and delay. The basic idea is that a larger quantiza-tion step leads to a higher data compression rate but has less information contained per packet. Therefore, the receiver may recover the video sequence with greater distortion but less delay.

Key Factors in QoE ModelingGenerally speaking, optimization of QoE-driven video stream-ing is more challenging than that of QoS-driven video stream-ing, since QoE not only depends on objective parameters such as data rate, packet loss rate, and delay, but is also affected by other factors such as video service type, viewer demogra-phy, video length, and display device, as specified in [1]. Such non-network-related factors may not have a direct impact on video QoS but do influence users’ viewing experience. In the optimization of QoS-driven video streaming, one usually only needs to consider the objective parameters mentioned above. Therefore, optimization of QoE-driven video streaming involves more design factors, which leads to a more challeng-ing problem.

Open Problems and Future DirectionsQoE-driven VCRN is a relatively new research area that has not been widely studied until recently. Although promising techniques and algorithms are proposed, there are still many research problems that have not been solved satisfactorily or even considered at this time. Here we briefly discuss some open problems that are worth investigating. • CU energy consumption: In order to promptly identify

returning PUs on a licensed channel, CUs need to perform spectrum sensing periodically, or even continuously. Fur-thermore, encoding and decoding video also require a lot of computation and consume considerable power. These two fundamental tasks will seriously shorten the battery lifetime of mobile CUs. There is not much work done on this aspect of VCRN. Possible solutions should involve spectrum sens-ing optimization and energy-efficient video codec design.

• CU feedback: During a video session, network parameters, such as channel conditions and receiver buffer storage, may change over time. Due to time-changing channel conditions

in CRNs, CUs may not be able to find an available chan-nel on which to send feedback promptly. The problem is more challenging when there are no common control chan-nels as in some CR networks. This makes it difficult for the sender to adjust transmissions and thus downgrades the CR video performance. Effective feedback mechanisms for CUs should be investigated.

• Admission control for SVC videos: Current admission con-trol works exploit the different QoE requirements of video sessions by prioritizing them. With the widely used SVC, a video sequence is divided into multiple layers, where the base layer provides the basic quality and an enhancement layer further strengthens the video quality if all the lower layers are received. An admission control scheme may first admit or reject video session requirements according to the capacity needed for guaranteeing the basic QoE of the vid-eos. It then treats the admitted videos differently based on their varying QoE requirements and characteristics, and the various layers of an admitted video session differently based on their significance in enhancing the quality of the received video.

• QoE modeling for VCRN: As discussed earlier, the exist-ing QoE models are mostly developed for video streaming over traditional wireless networks. As CRNs are different in many important aspects, it is necessary to design QoE mod-els based on the unique features of CRNs to fully explore their potential.

• Security and privacy: Security is also an important issue when ensuring the QoE of CUs. Attacks on a CU’s privacy will bring about information leakage, and attacks on the CU’s resource (e.g., licensed channel) will bring about serious degradation of the experience of CUs [15]. New research is needed to protect a CU’s privacy and video data from being tampered with or intercepted, and to protect the licensed channels for both PUs and SUs.

• Emerging new applications: As it is an emerging application, a 3D video contains far more information than the tradi-tional 2D video. Greater bandwidth is needed to support 3D video streaming, which makes CR technology especially suitable due to the capability it provides to CUs to unlock the potential of the under-used radio resource.

ConclusionsVCRN is a relatively new research area with many technical challenges to be addressed. In the preliminaries for VCRN, spectrum sensing result reliability should be guaranteed, and adequate QoE models should be developed. Efficient algo-rithms for resource, buffer, and interference management, cross-layer optimization, and trade-off among multiple objec-tives need to be developed in order to ensure satisfactory QoE performance. We have therefore concluded this article by a discussion of several open research problems.

References[1] Y. Chen, K. Wu, and Q. Zhang, “From QoS to QoE: A Tutorial on Video

Quality Assessment,” IEEE Commun. Surveys & Tutorials, no. 99, Oct. 2014, pp. 1–41.

[2] A. Khan, L. Sun, and E. Ifeachor, “QoE Prediction Model and Its Appli-cation in Video Quality Adaptation over UMTS Networks,” IEEE Trans. Multimedia, vol. 14, no. 2, 2012, pp. 431–42.

[3] Y. Wu et al., “A Learning-based QoE-Driven Spectrum Handoff Scheme for Multimedia Transmissions over Cognitive Radio Networks,” IEEE JSAC, vol. 32, no. 11, 2014, pp. 2134–48.

[4] H. Saki, A. Shojaeifard, and M. Shikh-Bahaei, “Cross-Layer Resource Allocation for Video Streaming over OFDMA Cognitive Radio Networks with Imperfect Cross-Link CSI,” Proc. ICNC 2014, Honolulu, HI, Feb. 2014, pp. 98–104.

[5] F. Wang, J. Huang, and Y. Zhao, “Delay Sensitive Communications over Cognitive Radio Networks,” IEEE Trans. Wireless Commun., vol. 11, no. 4, Apr. 2012, pp.1402–11.

IEEE Network • November/December 201524

[6] T. Jiang, H. Wang, and A. V. Vasilakos, “QoE-Driven Channel Allocation Schemes for Multimedia Transmission of Priority-Based Secondary Users over Cognitive Radio Networks,” IEEE JSAC, vol. 30, no. 7, Aug. 2012, pp. 1215–24.

[7] Z. He, S. Mao, and S. Kompella, “QoE Driven Video Streaming in Cog-nitive Radio Networks: The Case of Single Channel Access,” Proc. IEEE GLOBECOM 2014, Austin, TX, Dec. 2014, pp. 1388–93.

[8] Y. Ding and L. Xiao, “Video On-Demand Streaming in Cognitive Wireless Mesh Networks,” IEEE Trans. Mobile Computing, vol. 12, no. 3, Mar. 2013, pp. 412–23.

[9] D. Hu et al., “Scalable Video Multicast in Cognitive Radio Networks,” IEEE JSAC, vol. 28, no. 3, Apr. 2010, pp. 334–44.

[10] H. Luo, S. Ci, and D. Wu, “A Cross-Layer Design for the Performance Improvement of Real-Time Video Transmission of Secondary Users over Cognitive Radio Networks,” IEEE Trans. Circuits and Systems for Video Tech., vol. 21, no. 8, Aug. 2011, pp. 1040–48.

[11] H. Kushwaha et al., “Reliable Multimedia Transmission Over Cognitive Radio Networks Using Fountain Codes,” Proc. IEEE, vol. 96, no. 1, Jan. 2008, pp. 155–65.

[12] R. Yao et al., “Perceptual Experience Oriented Transmission Scheduling for Scalable Video Streaming over Cognitive Radio Networks,” Proc. IEEE GLOBECOM 2013, Atlanta, GA, Dec. 2013, pp. 1681–86.

[13] S. Li, T. H. Luan, and X. Shen, “Channel Allocation for Smooth Video Delivery Over Cognitive Radio Networks,” Proc. IEEE GLOBECOM 2010, Miami, FL, Dec. 2010, pp. 1–5.

[14] Z. Guan et al., “On the Effect of Cooperative Relaying on the Perfor-mance of Video Streaming Applications in Cognitive Radio Networks,” Proc. IEEE ICC 2011, Kyoto, Japan, June 2011, pp. 1–6.

[15] Z. Gao et al., “Location Privacy in Database-Driven Cognitive Radio Net-works: Attacks and Countermeasures,” Proc. IEEE INFOCOM 2013, Turin, Italy, Apr. 2013, pp. 2751–59.

BiographiesZhifeng he [S’15] received his M.S. degree in micro electronics and solid state electronics from Beijing University of Posts and Telecommunications, China, and his B.S. degree in electronics information science and technol-

ogy from Shandong University of Technology, Zibo, China, in 2012 and 2009, respectively. Since 2012, he has been pursuing his Ph.D. degree in the Department of Electrical and Computer Engineering, Auburn University, Alabama. His current research interests include cognitive radio, mmWave communications and networking, and multimedia communications and optimization.

Shiwen Mao [S’99, M’04, SM’09] received his Ph.D. in electrical and com-puter engineering from Polytechnic University, Brooklyn, New York, in 2004. Currently, he is the Samuel Ginn Endowed Professor in the Department of Electrical and Computer Engineering, Auburn University. His research interests include wireless networks and multimedia communications. He is a Distin-guished Lecturer of the IEEE Vehicular Technology Society. He is on the Edi-torial Boards of IEEE Transactions on Multimedia, the IEEE Internet of Things Journal, IEEE Communications Surveys & Tutorials, and IEEE Multimedia, among others. He is the Vice Chair--Letters of the IEEE ComSoc Multimedia Communications Technical Committee. He received the 2013 IEEE ComSoc MMTC Outstanding Leadership Award and the NSF CAREER Award in 2010. He is a co-recipient of the IEEE WCNC 2015 Best Paper Award, the IEEE ICC 2013 Best Paper Award, and the 2004 IEEE Communications Society Leonard G. Abraham Prize in the Field of Communications Systems.

Tao Jiang [M’06, SM’10] is currently a chaired professor in the School of Electronics Information and Communications, Huazhong University of Science and Technology, Wuhan, China. He received a Ph.D. in information and communication engineering from Huazhong University of Science and Tech-nology in April 2004. From August 2004 to December 2007, he worked in academia, including positions at Brunel University and the University of Mich-igan-Dearborn. He has authored or co-authored over 200 technical papers in major journals and conferences, and eight books/chapters in the areas of com-munications and networks. He served as Symposium Co-Chair for IEEE GLO-BECOM 2013 and IEEE WCNC 2013. He has served or serving as Associate Editor of IEEE Transactions on Signal Processing, IEEE Communications Surveys & Tutorials, IEEE Transactions on Vehicular Technology, and the IEEE Internet of Things Journal, among others. He is a recipient of the NSFC Distinguished Young Scholars Award in China.

IEEE Network • November/December 2015 25