QoE-Oriented Two-Stage Resource Allocation in Femtocell Networks Peilong Li, Ying Wang, Weidong Zhang, Yan Huang
Key Laboratory of Universal Wireless Communications, Ministry of Education Wireless Technology Innovation Institutes, Beijing University of Posts and Telecommunications P.O. Box 92, BUPT, Beijing, China Email:
[email protected] Abstract—In this paper, we study the radio resource allocation for femtocell networks in orthogonal frequency division multiple access (OFDMA) systems from the aspect of users’ quality of experience (QoE). Three types of services are considered including audio stream, data stream and video stream. With the metric being mean opinion score (MOS), a two-tier network is modeled and optimization problem is formulated for maximizing the system performance with lower consumed power. Given the Weber-Fechner Law, we investigate the logarithmic nature of QoE. Afterward, a two-stage iterative algorithm is proposed to resolve such optimization problem. Simulation results show that the overall MOS achieved is significantly higher than the traditional algorithm with a lower power consumption.
D
E ployment
I. I NTRODUCTION
of femtocellls based on orthogonal frequency division multiplexing (OFDM) brings enormous benefits to both users and operators. For instance, due to the mitigation of the interference, better signal qualities can be achieved, resulting in a enhancement of system capacity. However, the femtocell networks are deployed by customers without proper network planning, so their interference environment tends to be much complicated. Moreover, the existing macrocells may also cause great interference in the two-tier networks, (which are composed of macro-tier and femto-tier). Therefore, proper resource management is necessary for femtocells to maintain reasonable communication quality. Many studies [2][3] have been conducted to allocate radio resource in femtocells aiming to improve throughput or reduce consumed power. However, with the prevalence of various mobile devices, wireless users may require different types of services such as downloading files and watching videos. Thus, conventional metrics like date rate and spectrum efficiency cannot reflect the users’ satisfaction directly. On the other hand, users’ quality of experience (QoE) is drawing more and more attentions, which evaluates the performance of communication systems in terms of users’ subjective opinion. Therefore, the previous resource management approaches with regard to conventional metrics possibly further improve the transmission performance for the users with high QoE. As a result, no significant performance improvement can be achieved on the overall QoE. In order to address such issue, it is preferable to utilize the knowledge of QoE to exploit radio resource. Some works has been carried out on this topic. A QoEaware power allocation algorithm is presented in [4], aiming
to maximize the overall QoE in a network. However, little attention has been paid to the multi-application scene. [5] proposed a QoE-driven cross-layer optimization algorithm, which considers both system efficiency and the user quality fairness. But only the video delivery is concerned in this algorithm which may fail to handle the situation of multiservice. Sacchi in [7] addresses a balanced strategy to jointly maximize the QoE based on game theory. But its QoE-evaluating model has not paid enough attention to the characters of different services, which play a key role in QoE assessment. And to our best knowledge, few works take energy efficiency into consideration when designing the QoE-oriented algorithm. In this paper, we study the QoE-oriented radio resource allocation problem for femtocells in OFDMA systems. Every femtocell is assumed to provide multiple types of services to different users. With the mean opinion score(MOS) being the metric to capture user satisfaction, a two-stage iterative algorithm is proposed to improve the overall QoE as well as reducing the consumed power. The rest of this paper is organized as follows. Section II gives the system model, the QoE metric and the problem formulation. Section III presents the algorithm to resolve such problem. Section IV gives the simulation results and the corresponding analysis, followed by the final conclusion in Section V. II. S YSTEM M ODEL A. Network Model
We consider the downlink of a multiuser multi-service twotier network based on Long Term Evaluation (LTE). Assuming femtocells are randomly distributed in one macrocell, that active user equipments(UEs), and the − femtocell has each running one specific kind of service. The frequency band resource is reused in all cells, which can be divided into blocks(RBs) on each slot. To illustrate the model more clearly, several sets are defined as follows: ∙ Femtocells:F = { 1 }; 2 ∙ UEs in :U = { 1 }; 2 ∙ RBs:R = { 1 } ; 2 ∙ MOS of UEs in : } M = { 1 2
978-1-4799-4449-1/14/$31.00 ©2014 IEEE
Power consumed by UEs in 𝐹𝑚 : P 𝑚 = {𝑃1𝑚 , 𝑃2𝑚 . . . 𝑃𝑁𝑚𝑢𝑚 } For 𝑅𝑘𝑚 of femtocell 𝐹𝑚 , the transmit power is 𝑃𝑘𝑚 , and the channel gain is 𝐺𝑚 𝑘 , the Signal to Interference Plus Noise Ratio (SINR) 𝛾𝑘𝑚 on 𝑅𝑘𝑚 is : ∙
𝛾𝑘𝑚 =
𝑃 𝑚 ⋅ 𝐺𝑚 𝑃𝑘𝑚 ⋅ 𝐺𝑚 𝑘 𝑘 ∑𝑘 = ′ , ′ 𝜔𝑘 + 𝜎 𝑃𝑘 ⋅ 𝐺𝑘 + 𝑛∕=𝑚 𝑃𝑘𝑛 ⋅ 𝐺𝑛𝑘 + 𝜎
(1)
where 𝑃𝑘′ and𝐺′𝑘 represent the macrocell’s transmit power and channel gain respectively. 𝜎 stands for the thermal noise power. According to the Shannon theorem, the achievable channel capacity 𝐶𝑘𝑚 on 𝑅𝑘𝑚 is: 𝐶𝑘𝑚 = 𝐵𝑤 ⋅ 𝑙𝑜𝑔(1 + 𝛾𝑘𝑚 ).
(2)
Denote the set of RBs occupied by 𝑈𝑙𝑚 as R𝑙𝑚 = {𝑅𝑙1 , 𝑅𝑙2 . . . 𝑅𝑙𝑁 }. The throughput 𝑇𝑙𝑚 of 𝑈𝑙𝑚 is the sum of channel capacity of these RBs, which can be calculated by: 𝑇𝑙𝑚 =
𝑙𝑁 ∑
𝐶𝑘𝑚 ,
(3)
𝑘=𝑙1
And the power 𝑃𝑙𝑚 consumed by 𝑈𝑙𝑚 is: 𝑃𝑙𝑚
=
𝑙𝑁 ∑
𝑃𝑘𝑚 .
(4)
𝑘=𝑙1
B. QoE Metric According to the different characters, three typical application are deployed in this paper including audio stream, data stream and video stream. Each user is assumed to be in the service of single application. The utility function of QoE is expressed in terms of MOS estimated for each stream. MOS is represented by a real number ranging from 1 to 4.5. The relationship between MOS and user satisfaction are illustrated in Fig.1. Details of estimating MOS for three applications are
Fig. 1: Relationship between MOS and user satisfication[14]
described below. 1) Audio Stream: Different methods have been proposed to evaluate MOS of audio stream before, such as Perceptual Evaluation of Speech Quality (PESQ) and Pseudo-subjective Quality Assessment (PSQA). This paper adopts a PESQ-based MOS estimation method proposed in [8], which is a function of the transmission rate 𝑅 and the packet error probability (PEP ). Fig.2 shows experimental curves for MOS estimation as a function of PEP for different voice codecs. Accounting for
Fig. 2: PESQ-based MOS vs. packet error probability(PEP)for different transmission rate[8]
the high data rate of LTE, a reasonable assumption is that all the audio stream are coded by G.711. Thus the MOS function of PEP can be formulated as: 1 ) + 4.3. (5) 𝑀 𝑂𝑆 = 0.5 ∗ 𝑙𝑜𝑔( 1 + 60 ∗ 𝑃 𝐸𝑃 2) Data Stream: Data Stream service refers to surfing webs or downloading files on the internet. Download bandwidth, corresponding to user’s throughput here, directly influences the waiting time until the download is completed.Hence, experiments in [9] shows that MOS increases logarithmically with increasing throughput. In addition, users’ tolerance for waiting time is influenced by the size of the file. This paper adopts the MOS function modeled in [9] as follows: 0.775 (6) 𝑀 𝑂𝑆 = √ 𝑙𝑛(𝑇 ) + 1.268, 𝑠 where, 𝑠 stands for the file size with unit of MB and 𝑇 is the throughput of a user. 3) Video Stream: Video stream service in wireless communication requires high data rate, high continuity and realtiming. MOS for video stream is complicated to evaluate because users’ QoE is related to not only transmission rate of the stream but also the content of the video. [10] points out that, compared with images’ Mean Square Error(MSE) or Peak Signal Noise ratio(PSNR), Video Structural Similarity Index (VSSIM) can evaluate users’ QoE more reasonable, since human eyes are more sensitive to the distortion of images’ structure than that of images’ pixels. According to the data from Video Quality Expert Group(VQEQ)[13], MOS can be mapped to VSSIM as follows: ⎧ 𝑉 𝑆𝑆𝐼𝑀 < 0.7 ⎨ 1 12.5 ⋅ 𝑉 𝑆𝑆𝐼𝑀 − 7.75 0.7 ≤ 𝑉 𝑆𝑆𝐼𝑀 ≤ 0.98 𝑀 𝑂𝑆 = ⎩ 4.5 𝑉 𝑆𝑆𝐼𝑀 > 0.98. (7) An example of a video utility curve for five different video sequences is shown in Fig.3. From the figure we can observe that videos with more dynamic scenes are more demanding for data rate than static videos. Fig.3: As[5][6] MOS function can be formulated as follows: 𝑅 (8) 𝑀 𝑂𝑆 = 𝑘𝑖 ⋅ 𝑙𝑜𝑔𝛼𝑖 ( ) + 4.5, 𝑣𝑖
III. ALGORITHM DESIGN The problem formulation indicates a multi-objective optimization problem. A two-stage iteration algorithm namely MMRP(Maximize MOS and Reduce Power) is proposed, in which we uses two stages to maximize the total MOS of UEs and minimize the total transmit power respectively. These two stages are repeated iteratively until a terminate condition is satisfied. The algorithm is illustrated in Fig.4. The terminate condition is formulated as Eq.9. 𝑚
∣1 − (
Fig. 3: MOS vs. data rate for different videos[5]
in which the group of factor {𝑘𝑖 , 𝛼𝑖 , 𝑣𝑖 } for the 𝑖 − 𝑡ℎ video represents the impact of content in different videos. 𝑘𝑖 and 𝛼𝑖 are factors related to the temporal(movement) and spatial(brightness, blurriness) features of each video. 𝑣𝑖 refers to the required rate determined by the codec. 𝑣𝑖 is randomly generated and 𝑘𝑖 , 𝛼𝑖 is given in Table I.
video1 0.8 2
video2 0.6 5
video3 0.4 10
video4 0.2 15
𝑚
𝑝𝑀 𝑂𝑆𝑖𝑚 /
𝑁𝑢 ∑
𝑖=1
𝑙𝑀 𝑂𝑆𝑖𝑚 )∣ < 𝛾,
(9)
𝑖=1
where, 𝑝𝑀 𝑂𝑆𝑖𝑚 is 𝑈𝑖𝑚 ’s MOS before a certain iteration of MMRP while 𝑙𝑀 𝑂𝑆𝑖𝑚 stands for that after the iteration correspondingly. 𝛾 is a threshold which can be set according to the accuracy requirement. The flow chart of this algorithm is shown in Fig.4 NO
TABLE I: factors corresponding to different videos VIDEO 𝑘𝑖 𝛼𝑖
𝑁𝑢 ∑
Begin
Stage2: Decrease usersÿ power
Stage1: Maximize the sum of MOS
Meet the end condition?
End
YES
Fig. 4: Flow Chart of the two-stage iteration algorithm
Details of the two stages are described as follows. C. Problem Formulation
A. Stage I:Maximizie the Sum of MOS
In the network architecture, each femtocell manages its own resource independently. Thus, no information is needed to exchange through the backhaul of femtocells. Taking the both the MOS and power consumption into consideration, the primary objective is maximizing the sum of MOS and the slave one is to reduce the overall consumed power. Based on the network model and QoE metric described above, for a certain femtocell, taking 𝐹𝑚 as an example, the problem can be formulated as follows: 𝑚
𝑚𝑎𝑥 :
𝑁𝑢 ∑
𝑀 𝑂𝑆𝑖𝑚
𝑖=1 𝑚
𝑚𝑖𝑛 :
𝑁𝑢 ∑
𝑃𝑖𝑚 ,
𝑖=1
𝑠.𝑡.
⎧ ⎨
∑𝑁𝑢𝑚
𝑚 𝑖=1 𝑃𝑖 ≤ 𝑃𝑡𝑜𝑡𝑎𝑙 ∀𝑖 ∈ {1, 2 . . . 𝑁𝑢𝑚 }, 0 ≤ 𝑃𝑖 ≤ 𝑃𝑚𝑎𝑥 ⎩ ∀𝑗 ∈ {1, 2 . . . 𝑁 }, ∑𝑁𝑢𝑚 𝜒 ≤ 1 𝑅 𝑘=1 𝑗,𝑘
where 𝑃𝑡𝑜𝑡𝑎𝑙 is the total power of femtocell base station. And 𝑃𝑚𝑎𝑥 is the maximum power allowed for each UE.𝜒𝑗,𝑘 ∈ {0, 1} is a decision binary variable that is equal to 1 if 𝑅𝑗 is occupied by 𝑈𝑘𝑚 , 0 otherwise.
In this stage, the optimization objective: 𝑚
𝑚𝑎𝑥 :
𝑁𝑢 ∑
𝑀 𝑂𝑆 𝑚 𝑖 ,
𝑖=1
retrogrades to a constrain: 𝑚
𝑁𝑢 ∑
𝑃𝑖𝑚 = 𝑃𝑡𝑜𝑡𝑎𝑙 .
(10)
𝑖=1
[9] reveals the logarithmic nature of QoE according to the Weber-Fechner Law. Accordingly, MOS of a specific user is a logarithmic function of the throughput. In the meantime, throughput is also a logarithmic function of transmit power. Therefore, MOS will exhibit a strong diminishing marginal benefit in terms of power when RBs is settled. In other words, a user’s MOS , if already being high, will have a limited increase with a certain amount improvement of power, while a low MOS will have a stronger enhancement with the same amount of power improvement. So we present a new approach described in Algorithm Stage I. B. Stage II:Decrease Users’ Power The Weber-Fechner Law shows that the operation of the human sensory system can be traced back to the percipience of so called ”just noticeable differences” [11] [12]. Therefore, we assume that users can tolerate a small decrease in MOS. Taking 𝑀 𝑂𝑆𝑜 as the current MOS of a user and 𝜆(0 < 𝜆 < 1)
Algorithm Stage I:Maximize the Sum of MOS /* Reallocate Resource */ forall the 𝑈𝑖𝑚 ∈ U 𝑚 do if 𝑀 𝑂𝑆𝑖𝑚 == 4.5 then Free the redundant RBs and power of 𝑈𝑖𝑚 untill 𝑀 𝑂𝑆𝑖𝑚 just equals 4.5; else 𝑚 Put 𝑈𝑖𝑚 into the set of not satisfied users U𝑛𝑠 ; end end 𝑚 do forall the 𝑈𝑗𝑚 ∈ U𝑛𝑠 while Available RBs or power is not used up do Allocate RBs or power to 𝑈𝑗𝑚 until 𝑀 𝑂𝑆𝑗𝑚 just equals 4.5; end end /* Exchange power */ Set the step size of changing power to be Δ𝑃 ; repeat forall the 𝑈𝑖𝑚 ∈ U 𝑚 do Calculate Δ𝑀𝑖+ = 𝑀 𝑂𝑆𝑖𝑚 (𝑃𝑖𝑚 + Δ𝑃 ) − 𝑀 𝑂𝑆𝑖𝑚 (𝑃𝑖𝑚 ); Calculate Δ𝑀𝑖− = 𝑀 𝑂𝑆𝑖𝑚 (𝑃𝑖𝑚 ) − 𝑀 𝑂𝑆𝑖𝑚 (𝑃𝑖𝑚 − Δ𝑃 ); end Search among U 𝑚 to find: Δ𝑀 + = 𝑚𝑎𝑥𝑀𝑘+ ; Δ𝑀 − = 𝑚𝑖𝑛𝑀𝑙− ; and the corresponding 𝑈𝑘𝑚 and 𝑈𝑙𝑚 ; if Δ𝑀 + > Δ𝑀 − then Set 𝑃𝑘𝑚 = 𝑃𝑘𝑚 + Δ𝑃 ; Set 𝑃𝑙𝑚 = 𝑃𝑙𝑚 − Δ𝑃 ; end until Δ𝑀 + ≤ Δ𝑀 − ;
𝑚
where 𝑀 𝑂𝑆 ′ 𝑖 is the final MOS of 𝑈𝑖𝑚 after this stage. And the corresponding algorithm is presented in Algorithm Stage II described in the following: Algorithm Stage II:Reduce Users’ Power forall the 𝑈𝑖𝑚 ∈ U 𝑚 do 𝑚 Set 𝑃 ′ 𝑖 = 𝑃𝑖𝑚 ; 𝑚 Set 𝑀 𝑂𝑆 ′ 𝑖 = 𝑀 𝑂𝑆𝑖𝑚 ; 𝑚 while 𝑀 𝑂𝑆 ′ 𝑖 ≥ 𝜆 ⋅ 𝑀 𝑂𝑆𝑖𝑚 do 𝑚 ′ Set Δ𝑃 = 𝛿 ⋅ 𝑃 ′ 𝑖 ; 𝑚 Calculate 𝑇 𝑒𝑚𝑝𝑀 = 𝑀 𝑂𝑆𝑖𝑚 (𝑃 ′ 𝑖 − Δ𝑃 ′ ); 𝑚 if 𝑇 𝑒𝑚𝑝𝑀 ≥ 𝜆 ⋅ 𝑀 𝑂𝑆𝑖 then 𝑚 if Δ𝑃 ′ /(𝑀 𝑂𝑆 ′ 𝑖 − 𝑇 𝑒𝑚𝑝𝑀 ) ≥ 𝜇 then ′𝑚 Set 𝑃 𝑖 = 𝑇 𝑒𝑚𝑝𝑃 − Δ𝑃 ′ ; 𝑚 Set 𝑀 𝑂𝑆 ′ 𝑖 = 𝑇 𝑒𝑚𝑝𝑀 ; end end end end
IV. S IMULATION AND D ISCUSSION The scenario employed for this experimental evaluation is a Macrocell with a number of Femtocells uniformly distributed in the coverage of the Macrocell. UEs are uniformly distributed in each cell. The channel model is described in section II and the corresponding parameters are as follows: TABLE II: Basic Parameters PARAMETER Bandwidth Femtocell number
VALUE 10MHz 50
PARAMETER Carrier Frequency RB number
VALUE 2.0GHz 50
TABLE III: Parameters of Macrocell
as the ”tolerate-factor”. The acceptable change from 𝑀 𝑂𝑆𝑜 is defined as 𝑀 𝑂𝑆𝑐 = 𝜆 ⋅ 𝑀 𝑂𝑆𝑜 . On the other hand, accounting for the strong diminishing marginal benefit of MOS, great amount of power may be saved with a little decrease of MOS when current MOS is high. Denote Δ𝑃 ′ as the decreasing step size of power and 𝜇 as the ”cost-effective threshold”. If Δ𝑃 ′ /(𝑀 𝑂𝑆(𝑃 ) − 𝑀 𝑂𝑆(𝑃 − Δ𝑃 ′ )) ≥ 𝜇, then the decrease of MOS is cost-effective. Given the above definition, the optimization problem for stage II is defined as follows: 𝑚
𝑚𝑖𝑛 :
𝑁𝑢 ∑
𝑃𝑖𝑚 ,
𝑖=1
⎧ ⎨
∑𝑁𝑢𝑚
𝑃𝑖𝑚 ≤ 𝑃𝑡𝑜𝑡𝑎𝑙 ∀𝑖 ∈ {1, 2 . . . 𝑁𝑢𝑚 }, 0 ≤ 𝑃𝑖 ≤ 𝑃𝑚𝑎𝑥 𝑠.𝑡. ∑𝑁𝑢𝑚 ∀𝑗 ∈ {1, 2 . . . 𝑁 }, 𝑅 𝑘=1 𝜒𝑗,𝑘 ≤ 1 ⎩ 𝑚 ∀𝑖 ∈ {1, 2 . . . 𝑁𝑢𝑚 }, 𝑀 𝑂𝑆 ′ 𝑖 ≥ 𝜆 ⋅ 𝑀 𝑂𝑆𝑖𝑚 , 𝑖=1
PARAMETER Coverage Shadowing standard deviation Penetration Loss UE Antenna gain Total BS TX power(Ptotal) Minimum distance between UE and cell UE number
VALUE 1732m 8 dB 20 dB 0 dBi 46 dBm 35 m 50
TABLE IV: Parameters of Femtocell PARAMETER Room size Exterior wall penetration loss Noise figure femto BS Femto BS Min/Max TX Power max UE number Minimum distance between UE and cell
VALUE 12m x 12m 20 dB 8 dB 0/10 dBm 10 20 cm
In order to demonstrate the effectiveness of our algorithm, a baseline algorithm is implemented in the simulation. In the baseline algorithm, RBs are allocated to UEs with Round-robin scheduling, and power is equally distributed on every RB.
TABLE V: Parameters of Experiment PARAMETER 𝑃𝑠 𝛽 𝛾1
VALUE 0.01 mw 0.1 0.99
PARAMETER 𝛿 𝜆 𝛾2
VALUE 0.01 0.99 0.99
MMRP schedule Baseline schedule
0.8
Cumulative Distribution Function
V. C ONCLUSION In this paper, we have discussed a novel conception that radio resource can be allocated according to QoE in femtocells where quantities of multimedia services may take place. After modeling the femtocell network and QoE metric, a two-stage iteration algorithm MMRP has been promoted according to the logarithmic nature of QoE. Simulation results shows that the MMRP is an efficient resource management algorithm that optimizes users’ QoE as well as reduces the consumed power. Future work should address a more accurate method to evaluate QoE metric.
1 0.9
such as proportional fair(PF) algorithm. Notwithstanding its limitation, this study does suggest that MMRP is beneficial in the aspect of enhancing users’ QoE.
0.7 0.6 0.5 0.4 0.3 0.2
ACKNOWLEDGEMENT
0.1 0
1
1.5
2
2.5 3 Mean Opinion Score
3.5
4
4.5
Fig. 5: CDF curve of MOS
The experiment parameters are listed in TABLE V: Fig.5 shows the Cumulative Distribution Function(CDF) of all the users’ MOS during the simulation. It shows that the proportion of high-MOS users under the MMRP scheduling is higher than that under the baseline scheduling. Moreover, according to the statistic data, the average MOS under MMRP is 4.41 while that under baseline is 3.90. That is a promotion of 13.1%. It is reasonable to conclude that MMRP scheduling can improve users’ QoE effectively. Fig.6 illustrates each HeNB’s consumed power. 11 MMRP schedule Baseline schedule 10
Power/mW
9
8
7
6
5
0
5
10
15
20
25 30 HeNB Index
35
40
45
50
Fig. 6: Power consumption
It is obvious that almost every HeNB’s power is reduced by MMRP scheduling. From the statistic data we know that the average power consumed under the baseline scheduling is 9.56W while that under MMRP scheduling is 8.08W, which is a reduction of 15.5%. Therefore, MMRP scheduling can reduce HeNB’s consumed power observably. It should be noted that this simulation has compared only the baseline algorithm with MMRP. The results can not be taken as evidence that MMRP performs better in terms of both improving MOS and reducing power than other algorithms
This paper is supported by key project (2013ZX03001025002),National Natural Science Foundation of China (Project 61121001) and PCSIRT (No.IRT1049). R EFERENCES [1] Mobile Broadband Access at Home, Informa Telecoms & Media, August 2008. [2] Mhiri F. Sethom K., Bouallegue R., Pujolle G.. AdaC: Adaptive Coverage Coordination Scheme in Femtocell Networks. Wireless and Mobile Networking Conference (WMNC), 2011 4th Joint IFIP. [3] Ladanyi, A.,Lopez-Perez, D.,Juttner, A.,Xiaoli Chu,Jie Zhang. Distributed resource allocation for femtocell interference coordination via power minimization. GLOBECOM Workshops (GC Wkshps), 2011 IEEE. 2011:744-749. [4] Lili Xie, Chunjing Hu , Wenjun Wu, Zhenning Shi,QoE-Aware Power Allocation Algorithm in Multiuser OFDM Systems, 2011 Seventh International Conference on Mobile Ad-hoc and Sensor Networks (MSN). [5] Srisakul Thakolsri, Serdar Cokbulan, Dan Jurca, Zoran Despotovic, Wolfgang Kellerer,QoE-driven cross-layer optimization in wireless networks addressing system efficiency and utility fairness,2011 IEEE GLOBECOM Workshops (GC Wkshps). [6] Mohammed Shehada, Srisakul Thakolsri, Zoran Despotovic, Wolfgang Kellerer,QoE-based Cross-Layer Optimization for Video Delivery in Long Term Evolution Mobile Networks,2011 14th International Symposium on Wireless Personal Multimedia Communications (WPMC). [7] Claudio Sacchi,Fabrizio Granelli,Christian Schlegel,A QoE-Oriented Strategy for OFDMA Radio Resource Allocation Based on Min-MOS Maximizatio,IEEE COMMUNICATIONS LETTERS, VOL. 15, NO. 5, MAY 2011 [8] S. Khan, S. Duhovnikov, E. Steinbach, M. Sgroi, W. Kellerer. Applicationdriven Cross-layer Optimization for Mobile Multimedia Communication using a Common Application Layer Quality Metric. IWCMC06, July 3C6, 2006, Vancouver, Brithis Columbia, Canada [9] Reichl, P., Egger, S., Schatz, R. The Logarithmic Nature of QoE and the Role of the Weber-Fechner Law in QoE Assessment. 2010 IEEE International Conference on Communications (ICC). [10] Shehada M., Thakolsri S., Despotovic, Z.. QoE-based Cross-Layer Optimization for video delivery in Long Term Evolution mobile networks. 2011 14th International Symposium on Wireless Personal Multimedia Communications (WPMC). [11] M. Varela, valuation PseudoCsubjective de la Qualit dun Flux Multimdia, PhD Thesis, University of Rennes 1, France, 2007. [12] E. H. Weber, De Pulsu, Resorptione, Auditu Et Tactu. Annotationes Anatomicae Et Physiologicae, Koehler, Leipzig 1834. [13] International Telecommunication Union(ITU). Final Report From The Video Quality Experts Group On The Validation of Objective Models of Video Quality Assessment. COM 9-80-E. June 2000. [14] ITU-T G.107, The E-model, a computational model for use in transmission planning.
