Fei Wang, Zesong Fei , Jing Wang

School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China.

* The corresponding author, email：feizesong@bit.edu.cn

I. INTRODUCTION

Mobile video are popular in everyday life.The traffic of mobile video accounted for 55 percent of total mobile data traffic in 2015.What’s more, as a forecast from [1], threefourths of the world’s mobile data traffic will be video by 2020. Hypertext transfer protocol(HTTP) adaptive streaming (HAS), as an outstanding technology in mobile video transmission. Most video content providers like You-Tube and Netflix are using this method to deliver their video. This proposes a great challenge about HAS mobile video resource management.

In traditional wireless resource management methods in video aspect, researchers focus on the station side to schedule the radio access resource among different users [2]–[8].The quality of experience (QoE) is proposed as the overall quality of an application or service, as perceived subjectively by the end-user[9]. The QoE enhancement is also a target of many 5G technologies [10]. With QoE becoming more and more important in video service[11], the client side plays a key role in the wireless resource management, many resource management methods based on QoE are proposed.

The author in [12] uses the video QoE model with packet error rate, sending bit rate and frame rate parameters for optimizing network resource utilization for video over wireless network. But this model is not clear with the parameters of both application and network layers, because the packet error rate in the network layer can impact the frame rate in the application layer. What’s more, the author doesn’t concern the real buffer status or playing situation of end user. The work in [13]highlights the optimization of jointly considering QoE model and playout time. The author takes the QoE model as an incomplete information. The QoE model is taken as the unknown information in the paper and the author gives the bound of MOS loss. But the analysis of the real video service is not given in the paper. A machine learning approach is proposed in [14]. They use SSIM as MOS metric and the rate scaling factor. In unsupervised phase,the Restricted Boltzmann Machine (RBM)is used and the linear classi fier is used in the supervised phase. But the author also doesn’t concern the real video transmission and end user’s playing situation. In paper [15], authors present the resource allocation approach based on QoE in downlink OFDMA systems with the target of minimizing transmit power. The results show the better power efficiency with QoE requirements.

In this paper, the authors propose BDRM as a new schedule approach to enhance HAS QoE in LTE network.

Although many resource management methods based on QoE have been proposed as we analyze, all the existing researches only take the QoE as a optimizing metric in concept and don’t concern about what end users’ buffer and real playing situation are.What’s more, these methods do not focus on video transmission in HAS. The video server in HAS encodes the original video into different bit rates and cuts the whole video into many segments with same duration. The multi-segment and multi-rate features make the video playing process in a different way.When video-on-demand or live event is under HAS transmission, client probes the current bandwidth and requests the sequential video segments with a proper bit rate. Different from the traditional real time message protocol(RTMP) stream’s push-based style, the HAS is pull-based from the client side. It provides the dynamic video rate choice under different network environment. Microsoft, Apple Inc.and Adobe System Inc. propose the Microsoft Smooth Streaming (MSS) [16] [17], HTTP Live Streaming (HLS) [18], HTTP Dynamic Streaming (HDS) [19] as typical private protocol of HAS separately. Because the multirate and multi-segment features and pull-based scheme of HAS, a new resource management method is needed. In this paper, we propose a buffer-driven resource management (BDRM)method to enhance HAS QoE in LTE network.The resource in this paper is referred to as resource block (RB) in each schedule slot. The RBs are scheduled and dispatched among the users in each cell. The contributions of our work are presented as follows.

Firstly, we propose the HAS QoE influencing factors from the technical view in real playing situation. We focus on the multi-segment and multi-rate feature of HAS in mobile network to analyze the QoE factors of HAS.Most of paper [20]–[23] are concerned on the transmission technology of HAS, but not on the HAS’s feature. Because the application layer is the final layer facing the end users.It reflects the human visual system’s feeling from the end users [24]–[27]. The initial delay,rebuffering and quality level are proposed as the three parameters from application layer deciding the HAS QoE for the speci fic video clip and equipment.

Secondly, we propose the buffer as the drive of whole schedule and the proposed BDRM approach schedules the wireless resource referring to both client side and station side. Because HAS video playing is a space-separate and time-consecutive process for different users, the BDRM method decomposes the HAS QoE maximization problem into client and base station sides separately to solve it in mobile HAS video playing scene.The traditional method only takes the base station side without considering the user’s buffer. At client side, a buffer probe and rate request algorithm is proposed. Different from the traditional HAS scheme probes only average download rate of the previous video segment, our quality decision algorithm contains download rate, playback buffer and segment duration. It shows a more swift to guarantee the less rebuffering events. What’s more, after the client shows first request of quality,we use greedy algorithm to give the resource allocation in each schedule to make a better quality level with updated rate request. The drive of buffer and twice rate request schemes make the BDRM take advantage of HAS’s multi-segment and multi-rate features.

What’s more, to make the results more sufficient, we implement the work in multi-cell and multi-user situation in LTE network. We take the throughput, initial delay, rebuffering,quality level, MOS and MOS fairness from the view of objective and subjective assessment metrics. In order to make the subjective work use less time and cost, we propose the weighted similar MOS search method to use less real MOS samples from end users to evaluate more playing scenes.

The results show that the BDRM method decreases rebuffering events and increases the mean MOS compared with proportional fair(PF), Max C/I and traditional HAS schedule(THS) methods. It enhances the end user’s subjective feeling and guarantees the throughput and fairness. The BDRM can be used in the complex and inconstant network for HAS transmission and provide users good QoE.This paper shows a further step of the wireless resource management of HAS to enhance the QoE in the LTE network.

The remainder of the paper is organized as follows. Section II describes the key factors influencing HAS QoE. The BDRM scheme and details are explained in Section III. The simulation of multi-cell multi-user HAS video playing scenes in LTE network and weighted similar MOS search method are introduced in Section IV. In Section V, we compare the PF, Max C/I, THS and BDRM in objective and subjective assessment metrics. The results show that the BDRM has less rebuffering events. The statistical data of MOS also validate the BDRM’s excellent performance.Finally, in Section VI we conclude this paper and gives the future work.

II. MULTI-PARAMETERS QOE PROBLEM IN APPLICATION LAYER

Although many factors can affect the QoE of end user [24], the subjective and objective factors or technical and nontechnical factors.These factors contain the watch environment,psychological, sociological aspects, pricing policy, terminals, video format, etc. For different speci fic research objectives, we can focus on the speci fic factor and set other factors in the fixed mode. For example, if researcher wants to analyze the different terminals’ impact on the video QoE, the other factors like video content, video rate, etc. should be same when using different terminals. For a speci fic video in mobile HAS playing scene in a certain terminal, the application layer parameters directly in fluence the end user’s feeling. In our paper, we take the initial delay, rebuffering and quality level as the three factors that users can feel directly from the video. In this section, we analyze these three factors’ effect on HAS QoE from the view of end users.

2.1 Initial delay

When users touch the play button on the video player in Internet, it shows initial delay before video playing. The initial delay is influenced by many factors like the video bit rate, the video encoder, the buffer size of video player.Usually, an initial delay constant is set in the video player in order to make enough time for the video downloading to avoid the interruption in the following play. The initial delay can be taken as buffer size of video player, while it is always in the second unit not in the byte unit. It can be calculated as follow

whereIrepresents the initial delay in second or

In the HAS speci fic technology manual [18],the iOS HLS player requests video bit rate in subminimal level. This is a balance to make the end user have less initial delay even the network is not good enough and guarantee the end user have a relative good enough video rate.

2.2 Rebuffering

The rebuffering is the stalling of the HAS video playing when the buffer size of player is empty. The buffer time is the constant function of download time and play time. It’s simply shown as follow

whereB(t)is the buffer time,Bis the full buffer time setting in the video player.is the downloaded video clip duration after the initial delay and theis the real play time of progress bar. Specially in HAS video play situation, each time a video bit rate request from the end user. The buffer can be updated like this

All the existing analyses about the relationship between rebuffering and video QoE is only for HTTP progressive download and RTP streaming not for HAS [28] [29]. Because of the multi-segment and multi-rate feature, the HAS can be analyzed in each segment. We allocate the rebuffering event in each segment.

2.3 Quality level

2.4 Space decomposition of HAS QoE problem

From the view of the userm, the QoE of video play related to the initial delay, the total buffer times and the quality level is written as follow

The average QoE performance of all theMusers is de fined as

The aim of the buffer-driven resource management can be expressed as

This target is the general target of the whole HAS video process. This problem is about three application layer parameters, but it has the relationship with both end equipment and station. Different from the traditional video resource management methods in [12]–[14],[20]–[23], all the key processes are in the base station side. What’s more, the schedule process of the HAS is the continuous time process, not a point in time. Because video playing is an space-separate and time-consecutive process for different users, the BDRM method decomposes the HAS QoE maximization problem into client and base station sides separately with space decomposition solve it in multicell and multi-user video playing scene in LTE network. In next section, we show the BDRM in detail.

III. BUFFER-DRIVEN DYNAMIC RESOURCE MANAGEMENT

To specify and solve the problem (8), we must specify the process of theI(t),N(t)andL(t)in the HAS transmission in mobile network. The BDRM method is the method concerning both equipment and base station side. We can find that theBin (1) and (3) is the key factor in fluencing the schedule. It plays as a drive of the BDRM’s two-side process. A buffer probe and rate request model is proposed and implemented in mobile users side. It ensures the mobile equipment to make a higher video rate request.The station processes the dynamic resource management to find scheduled users who have lessN(t)and higherL(t).

3.1 System architecture

In Fig. 1, the whole system architecture is presented. It contains both base station side and end user side. The feature of this system is that a scheduling node is added in the LTE platform. Traditional HAS client post the rate request to base station and the HAS server delivers the corresponding bit rate segment to the end user. In our system, the buffer level of the end user is an important factor for rate requesting. The scheduling node receives all user’s rate request. But this rate request is not directly posted to the HAS server. It calculates all the RB resource and reassigns the rate of each user. The update bit rate is posted to HAS server and it’s the final video bit rate of each segment. Our proposed buffer probe and rate request algorithm and dynamic resource management are introduced in next two parts.

3.2 Buff er probe and rate request algorithm

In the user side, each mobile equipment proposes their own requirement of HAS video rate. Each user don’t know the whole wireless network environment, they only can request the video bit rate from the schedule scheme in the video player. In the traditional schedule scheme of the HAS video, the end user only concern about the bandwidth to decide the next video segment’s bit rate. But in our buffer probe and rate request model, we take the video download rate, video

playback buffer and the segment duration into account. The average video download rate is calculate by

whereDthe data size of downloaded video more intelligent, we classify two cases of the rate request scene.

Fig. 1 Buffer-driven resource management system architecture for HAS in mobile network

request model. In Fig. 2, we observe that the proposed scheme selects the bit rate more moderate and flat. Even the bandwidth becomes the lowest, the end user can also watch the higher bit-rate video. In Fig. 3, the change of bit rate based on the traditional scheme has a constant delay of bandwidth, because the traditional scheme selects the bit rate just based on the download rate of the previous schedule time. The bit rate with proposed scheme changes differently when we consider both bandwidth and buffer. It can be seen that the traditional scheme induces two rebuffering event, while our proposed scheme avoid these rebuffering and make the user watch the video in a fluent way.

3.3 Dynamic resource management

In the base station side, the station receives the HAS end user’s different rate request. The aim of the station side is to maximize the quality level in the limited resource as

From (4) and (14), the target is got as follow

Fig. 2 The request video bit rate of traditional scheme and proposed scheme

Fig. 3 The buffer time of traditional scheme and proposed scheme

IV. SIMULATION IN TESTBED

4.1 Simulation setup

In our test work, we design the speci fication of the framework like Fig. 1. It’s a video-on-demand schedule system in LTE network with multi-cell and multi-user settings. The video source is from the open source Sintel [38].The video codec is H.264 and the audio codec is AAC. In order to make our test cover different scenes, we take six typical clips from the whole movie. Fig. 4 shows six clips containing three scenes. Fig. 4 (a)(b) are still scenes without moving of roles. Fig. 4 (c)(d) are the slow moving scenes and the hero just slowly walks in video. Figs. 4 (e)(f) are the fast moving scenes： one is the hero running in the market and the other is fighting with dragon. We take all these clips also considering the brightness of pictures. Because the HAS is based on HTTP protocol in application layer, the video server can use the traditional HTTP server like Apache [39]. In order to decrease the initial buffering time and guarantee a relative good bit rate [18], the first two segments are requested the subminimum-value 210kbps. The m3u8 table is a list file containing the different video bit rate and the mapping between bandwidth and bit rate. Table I shows a typical of the clip2. The bandwidth is a little larger than the bit rate as the audio rates are contained.The schedule node and LTE platform dispatch the wireless resource and update the video bit rate request to pull them from HAS server.The network parameters are shown in Table II.

4.2 Weighted similar MOS search

Because videos in HAS scheduled by different methods are ultimately watched by end users in their smart phone, the MOS from end users is the most pertinent indicator to assess the MOS gain of resource management methods.In order to assess the different resource management results in the view of the real MOS from volunteers, we launch a subjective test with typical video transmission scenes over LTE network. As a universal structure of testbed targeting at MOS evaluation, four phases are shown in Fig. 5. After the subjective test,the database collects all the MOS and the objective parameters.

The ideal situation of doing subjective test is transmitting all the video playback scenes in three different resource management methods with double-blind test [27],but different network conditions and different resource management methods produce too many different video play scenes. This makes the work to be too complex. It will cost much time and human resources.

In order to simplify the subjective testing work and make the test results similar to the accurate MOS value by end users, we proposed the similar MOS search method. In Fig.6, we choose 90 video scenes typically in different network fluctuation conditions from our simulation. The normal network is stable and the change times of segment video rate mostly distribute between 1 to 5 times in the lab environment [40]. The experimental design in absolute category rating (ACR) method. The ACR method speci fies that after each presentation the subjects are asked to evaluate the quality of the sequence [27]. 90 volunteers are invited to mark 90 scenes in random and we make sure that each scene is marked with 6 different subjective MOS.

?

Fig. 4. The clips of different scenes from the Sintel movie: (a)(b) are still scenes, (c)(d) are slow moving scenes, (e)(f) are fast moving scenes

Fig. 5 The universal architecture of QoE evaluation

Table I The m3u8 file of clip2

Table II Simulation parameters

What’s more, in order to evaluate the fairness of different method, we proposes the MOS fairness metric. It is de fined as

The biggerFis, the more fair case is. The maximum value ofFis 1 in which all the users achieve the same MOS and it’s the greatest fairness case.

V. RESULTS

In this section, we show the performance of the proposed BDRM method. The comparisons are displayed between PF, MAX C/I,THS and BDRM. The assessment indicators are mean cell throughput, initial delay, rebuffering and MOS from objective to subjective scope. The results show that our proposed BDRM method can decrease the rebuffering times, improve the end user MOS apparently and guarantee a high fairness.

5.1 Performance with objective indicator

Fig. 8 shows that the rebuffering percent of whole play of video with BDRM method is only 1.96%. It’s much less than the 74.31%with Max C/I and 11.1% with PF. THS with its bandwidth probe scheme can decrease its rebuffering percent to 7.01%, but it is also higher than the BDRM. We can note that the Max C/I has the highest mean average cell throughput, but the majority resource is only limited in the small group and larger group has the lower bandwidth to play the lowest bit rate video. This makes that most of the video play under Max C/I are stall and causes rebuffering events frequently. The initial delay of THS and BDRM are the same because the initial video bit rate are set to be the same value in the simulation as recommendation in HLS developer guide [18]. While the first segment rate of PF and BDRM is decided by the initial resource dispatch from the first schedule and the initial delay is calculated by mean of all users in different cell. From the initial delay and rebuffering percentage of PF and BDRM,we can conclude that although the longer initial delay can increase the buffer time to avoid rebuffering, but the good scheme from the view of both end user and base station can decrease rebuffering more apparently. From above analyses in objective scope, it is obvious that the proposed BDRM method has the lowest rebuffering percent of different methods and have a throughput increase than the PF and THS. But these indicators can’t re flect the comprehensive performance, we will show the subjective metric in next subsection.

Fig. 6 The video rate change times (a) and distribution of different scenes (b)

?

5.2 Performance with subjective indicator

In Fig. 9, the three dimensions of MOS are presented. The mean MOS is calculated by all the users in different cell and it re flects all user’s subjective feeling to the whole video play. Some of the MOS are not marked directly from the user, but gotten by using weighted similar MOS search method. The MOS standard deviation (STD) and MOS fairness are two indicators re flecting the MOS fluctuation and distribution. Normally, the smaller MOS STD is, the higher MOS fairness is. Fig. 9 presents that the BDRM has the mean MOS 3.94. It gets 50% gain than the Max C/I mean MOS 2.15 and 15% gain than the PF mean MOS 3.42. On the other hand, the BDRM also perform better in fairness. The MOS fairness of BDRM is 0.989 and is a little higher than the 0.968 and 0.972 of HAS and PF method.The Max C/I MOS fairness value is 0.895 and this show the resource are not dispatch in equal. Meanwhile, the BDRM has the least MOS STD value of all methods.

Fig. 10 shows the cumulative distribution function (CDF) of mean MOS for different methods. As expected, the 50% of all users have the mean MOS above 4 with RDRM method. The top 50% of mean MOSs with PF and THS are 3.6 and 3.5. The Max C/I CDF cure shows that although about 10% of all mean MOS is 4.4, 60% is under 2.1. The distribution of Max C/I MOS induces the lower mean MOS.

Fig. 7 The mean cell throughput of different methods

Fig. 8. The initial delay and rebuffering percentage of different methods

Fig. 9 The mean MOS, MOS STD and MOS fairness of different methods

We can see that the BDRM can improve mean MOS and guarantee the fairness of all users. The buffer probe and rate request from end users guarantee the best rate of their own and the schedule node in base station side allocates the resource to make the fairness.The updated video bit rate is scheduled by the quality level. The drive of buffer and twice rate request schemes make BDRM take full advantage of HAS’s multi-segment and multirate features. The BDRM’s two-side comprehensive schedule features guarantee the MOS gain and fairness.

From the objective and subjective assessment metrics of different methods, the BDRM shows the best performance. The BDRM can make the video play with less rebuffering event and higher end user’s MOS.

Fig. 10 The MOS CDF of different methods

VI. CONCLUSIONS AND FUTURE WORK

In this paper, we propose BDRM as a new schedule approach to enhance HAS QoE in LTE network. Unlike the traditional resource schedule methods only focusing on base station side, we take both station and client sides into account to fit the multi-segment and multi-rate feature of HAS. The buffer from end user is taken as the drive of whole schedule from client to station. The proposed HAS QoE in fluencing factors are composed of initial delay, rebuffering and quality level. The BDRM method decomposes the HAS QoE maximization problem into client and base station sides separately. The problem is solved in multi-cell and multiuser video playing scene in LTE network. In client side, the buffer probe and rate request algorithm is implemented by each user separately to guarantee the less rebuffering event and decide which HAS segment rate level to be requested. While, in the base station side, the schedule of wireless resource is made to maximize the quality level of all access clients and update the requested rate pulled from HAS server. The drive of buffer and twice rate request schemes make BDRM take full advantage of HAS’s multi-segment and multi-rate features. As the simulation results, compared with PF, Max C/I and THS methods, the proposed BDRM method decreases rebuffering percentage, increases the mean MOS of all users to 3.9 from 3.42 with PF and 2.15 with Max C/I and guarantees a high fairness with 0.98 from objective and subjective assessment metrics view. Except HAS, we will do future investigation on the burgeoning video services like virtual reality (VR) and augmented reality(AR) and research these application in next generation mobile network.

ACKNOWLEDGEMENTS

This work was supported by the 863 project(Grant No. 2014AA01A701) and Beijing Natural Science Foundation (Grant No. 4152047).

[1] Cisco Systems Inc, “Cisco visual networking index: Global mobile data traffic forecast update 2015-2020 white paper,” Feb. 2016.[Online]. Available: http://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/mobile-white-paper-c11-520862.html.

[2] Q. Zhang, W. Zhu and Y.-Q. Zhang, “Channel-adaptive resource allocation for scalable video transmission over 3G wireless network,”IEEE Trans. Circuits Syst. Video Technol., vol. 14,no. 8, pp. 1049—1063, Aug. 2004.

[3] H. Jiang and W.-H. Zhuang, “Resource allocation with service differentiation for wireless video transmission,”IEEE Trans. Wireless Commun.,vol. 5, no. 6, pp. 1456—1468, Jun. 2006.

[4] Z. He and D. Wu, “Resource allocation and performance analysis of wireless video sensors,”IEEE Trans. Circuits Syst. Video Technol., vol. 16,no. 5, pp. 590—599, May 2006.

[5] P. Pahalawatta, R. Berry, T. Pappas and A.Katsaggelos, “Content-aware resource allocation and packet scheduling for video transmission over wireless networks,”IEEE J. Sel. Areas Commun., vol. 25, no. 4, pp. 749—759, May 2007.

[6] H. Zhu “Radio resource allocation for OFDMA systems in high speed environments,”IEEE J.Sel. Areas Commun., vol. 30, no. 4, pp. 748—759,May 2012.

[7] S. Gnanavel, S. Ramakrishan and N. M. Kumar,“A systemic resource allocation for multi-user video transmission over wireless network,” inProc. Int. Conf. Radar, Commun. Comput. (ICCR),Tiruvannamalai, Dec. 2012, pp. 282—288.

[8] Z.-S. Fei, C.-W. Xing and N. Li, “QoE-driven resource allocation for mobile IP services in wireless network,”Sci. China Inf. Sci., vol. 58, no. 1,pp.1—10, Jan. 2015.

[9] ITU-T Recommendation P.10/G.100, “Vocabulary for performance and quality of service,” Jul.2006.

[10] S. Chen, F. Qin, B. Hu, X. Li and Z. Chen, “User-centric ultra-dense networks for 5G: challenges, methodologies, and directions,”IEEE Wireless Commun., vol. 23, no. 2, pp. 78-85, Apr.2016.

[11] S. Chen and J. Zhao, “The requirements, challenges, and technologies for 5G of terrestrial mobile telecommunication,”IEEE Commun.Mag., vol. 52, no. 5, pp. 36-43, May 2014.

[12] 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, no. 11, pp.1337—1347, Jul. 2010.

[13] L. Zhou, Z. Yang. Y. Wen, H Wang and M. Guizani, “Resource allocation with incomplete information for QoE-Driven multimedia communications,”IEEE Trans. Wireless Commun., vol. 12, no.8, pp. 3733—3745, Aug. 2013.

[14] A. Testolin et al., “A machine learning approach to QoE-based video admission control and resource allocation in wireless systems,” inProc.13th Annu. Mediterranean Ad Hoc Net. Workshop (MED-HOC-NET), Piran, Jun. 2014, pp.31—38.

[15] S. E. Ghoreishi and A. H. Aghvami “Power-eき-cient QoE-aware video adaptation and resource allocation for delay-constrained streaming over downlink OFDMA,”IEEE Commun. Lett., vol. 20,no. 3, pp. 574—577, Mar. 2016.

[16] A. Zambelli, “Smooth streaming technical overview,” Mar 2009. [Online].Available:http://www.iis.net/learn/media/on-demand-smoothstreaming/smooth-streaming-technical-overview.

[17] A. Zambelli, “A history of media streaming and the future of connected TV,” The Guardian, Jul. 2013. [Online]. Available: http://www.theguardian.com/media-network/media-networkblog/2013/mar/01/history-streaming-future-connected-tv.

[18] Apple Inc., “HTTP live streaming overview,” Feb.2016. [Online]. Available: https://developer.apple.com/streaming.

[19] Adobe Systems Inc., “HTTP dynamic streaming,”May 2015. [Online].Available:http://www.adobe.com/products/hds-dynamic-streaming.html.

[20] J. M. Jeong and J. D. Kim, “Effective bandwidth measurement for dynamic adaptive streaming over HTTP,” inProc. Int. Conf. on Infor. Netw.(ICOIN), Cambodia, Jan. 2015, pp. 375—378.

[21] W. Pu, Z.-X. Z and C.-W. Chen, “Video adaptation proxy for wireless dynamic adaptive streaming over HTTP,” inProc. 19th Int. Packet Video Workshop (PV), Munich, May 2012, pp.65—70.

[22] S. Lee, K. Youn and K. Chung, “Adaptive video quality control scheme to improve QoE of MPEG DASH,” inProc. IEEE Int. Conf. Consum.Electron. (ICCE), Las Vegas, Jan. 2015, pp. 126—127.

[23] S. Garcia, J. Cabrera and N. Carcia, “Quality-optimization algorithm based on stochastic dynamic programming for MPEG DASH video streaming,” inProc. IEEE Int. Conf. Consum. Electron. (ICCE), Las Vegas, Jan. 2014, pp. 574—575.

[24] P. Brooks and B. Hestnes, “User measures of quality of experience: why being objective and quantitative is important,”IEEE Netw., vol. 24,no. 2, pp. 8—13, Apr. 2010.

[25] ITU-T Recommendation P.800.1, “Mean opinion score (MOS) terminology,” Jul. 2006. [Online].Available: http://www.itu.int/rec/T-REC-P.800.1-201607-I.

[26] ITU-T Recommendation P.800.2, “Mean opinion score interpretation and reporting,” May 2013. [Online]. Available: http://www.itu.int/rec/T-REC-P.800.2-201607-I.

[27] ITU-T Recommendation P.910, “Subjective video quality assessment methods for multimedia applications,” Apr. 2008. [Online].Available:http://www.itu.int/rec/T-REC-P.910-200804-I

[28] R. K. P. Mok, E. W. W. Chan, and R. K. C. Agar,“Measuring the quality of experience of HTTP video streaming,” inProc. 12th IFIP/IEEE Inter.Symp. Integr. Netw. Manag. and Workshops (IM),Dublin, May 2011, pp. 485—492.

[29] T. Hossfeld et al., “Initial delay vs. interruptions:Between the devil and the deep blue sea,” inProc. 4th Int. Workshop on Quality of Multimedia Experience (QoMEX), Yarra Valley, Jul. 2012,pp. 1—6.

[30] A. Ismail and S. BAulent, “Statistical evaluation of image quality ¨ measures,”J. of Electron.Imag., vol. 11, no. 2, pp. 206—223, Apr. 2002.

[31] Z. Wang, L. Lu. A. C. Bovik, “Image quality assessment: from error visibility to structural similarity,”Signal Process.: Image Commun., vol. 19,no. 2, pp. 121—132, Feb. 2004.

[32] M. H. Finson and S. A. Wojf, “A new standardized method for objectively measuring video quality,”IEEE Trans. Broadcast., vol. 50, no. 1, pp.312—322, Sep. 2004.

[33] ITU-T J.247 “Objective perceptual multimedia video quality measurement in the presence of a full reference,” Aug. 2008.

[34] O. Sugimotom, R. Kawada, M. Wada and S. Matsumoto “Objective measurement scheme for perceived picture quality degradation caused by MPEG encoding without any reference pictures,”Proc Spie, vol. 4310, pp. 13—18, Jan. 2001.

[35] M. Fiedler, T. Hossfeld and P. Tran-Gia, “A generic quantitative relationship between quality of experience and quality of service,”IEEE Netw.,vol. 24, no. 2, pp. 36—41, Mar. 2010.

[36] P. Reichl, S. Egger, R. Schatz and A. D’Alconzo,“The logarithmic nature of QoE and the role of the Weber-Fechner Law in QoE assessment,” inProc. IEEE Int. Conf. Commun. (ICC), Cape Town,May 2010, pp. 1—5.

[37] P. Reichl, B. Tuffin and R. Schatz, “Logarithmic laws in service quality perception: where microeconomics meets psychophysics and quality of experience,”Telecommun. Syst., vol. 52, no. 2,pp. 587—600, Feb. 2013.

[38] Blender Foundation, “Watch Sintel online,” Sept.2015. [Online]. Available: https://durian.blender.org/download.

[39] Apache Software Foundation, “What is the apache software fundation,” Sept. 2015. [Online]. Available: http://www.apache.org/foundation/howit-works.html.

[40] T.-T. Yuan et al.,” A QoE-based cell range expansion scheme in heterogeneous cellular networks,”Sci. China Inf. Sci., vol. 58, no. 8, pp.1-11,Jun. 2015.