Customizable Algorithm for Data Scheduling in Long Term Evolution Networks Jahyr Gonçalves Neto, Fabbryccio A. C. M. Cardoso, Paulo A. Valente Ferreira* and Max H. M. Costa Department of Communications, Department of Systems and Energy* School of Electrical and Computer Engineering University of Campinas - Unicamp Campinas, SP, Brazil janeto, [email protected], [email protected], [email protected] Abstract - Long Term Evolution (LTE), a standard developed by the 3rd Generation Partnership Project (3GPP), is currently the 4G mainstream technology, providing technology convergence for this generation of cellular communication systems. LTE networks promise data transmission rates of 100 Mbps in the downlink channel. For this high throughput it needs to optimize t;he use of network resources. LTE uses scheduling techniques for radio resource allocation based on communication channel conditions. This paper presents a heuristic customizable algorithm for multi-user data scheduling in the LTE downlink channel that can prioritize selected types of services. The proposed scheduling scheme is a modified version of the proportional fairness scheduler that uses Analytic Hierarchy Process (AHP) techniques. This allows to aggregate performance criteria to improve resource allocation based on general traffic conditions and a set of predefined service priorities. The approach for this customization is to assign weights to certain parameters used in the resource allocation process, thus modifying the relative importance of these parameters and creating a hierarchy of criteria. Criteria for different types of data are added through comparison matrices. A comparison matrix quantifies the judgment of the decision maker with respect to the relative importance of different criteria. Thus the scheduler can be made flexible and adaptive to different traffic scenarios. Examples are given where video traffic is prioritized. Keywords: Scheduling; Multicriteria decision making; Long Term Evolution; Scheduler; Analytic Hierarchy Process, AHP

I. INTRODUCTION Even without achieving all the requirements of ITU (International Telecommunication Union) to be considered a fourth-generation cellular network, LTE (Long Term Evolution) [1] is popularly called a 4G [2] technology. LTE is a mobile IP technology carefully designed to provide much lower latency and much higher spectral efficiency than previous technologies of third generation. LTE wireless access technology is based on Orthogonal Frequency Division Multiplexing Access (OFDMA) [3], which enables the achievement of significant improvements in the allocation of radio resources and link adaptation. Furthermore, with a simple network architecture of low cost, LTE not only provides a faster connection, but also a variety

of new applications previously available only via broadband wired networks. An example of such applications is navigation on websites with video content embedded in Web pages. More than a tendency, this type of content is causing a much higher data traffic in wired broadband access networks, and it is inevitable that the same will occur in mobile cellular networks [2]. Foreseeing this growing demand for data rate, mobile Internet and interactive services, eg, VoIP, video streaming and online gaming, LTE was included in the itinerary of 3GPP (3rd Generation Partnership Project) to ensure the competitiveness of 3G operators and enable an upgrade path for current networks. The original goal of bit rate for the system was set at 100 Mbps downlink and 50 Mbps in the uplink, which can be achieved by the system when operating with a single antenna (Single Input Single Output - SISO) and bandwidth of 20 MHz. For narrower bandwidths, the corresponding bit rates can be calculated from the appropriate scaling of the 20 MHz bit rate. However, the requirements for LTE system operation are different from other wireless communication systems. Thus, the adaptation of some techniques for use in LTE system requires further investigation [4]. In addition, there is no specific standardization for the scheduler and the implementation is up to the network service provider [5]. This paper proposes a heuristic adaptive scheduling scheme that can be customized by the operator according to the demand for network traffic services. The scope of work is the downlink of LTE networks, where data services can be specially treated to avoid network congestion. The development is based on a customization of the Proportional Fairness algorithm using the Analytic Hierarchy Process (AHP) tool [6]. The AHP is a multicriteria decision making method that makes it possible to add performance criteria to algorithm parameters, such as user channel quality indicator (CQI), historical data received by the user, or data rate achievable by users. These variables are used in the calculations for allocating radio resources to users in a comparison of different

scheduling algorithms. The main idea of the AHP method is to select relative weights that describe the relative importance of each criterion, classifying the decision alternatives. This is done using pairwise comparison matrices, which allow to modify the resource allocation among a set of alternatives The proposed scheme is designed to perform as well as proportional fairness when the multicriteria decision making is not fulfilled, that is, in a scenario where the parameters of the comparison matrices are all equal to one. Some related work can be highlighted as follows. In [7] a study for implementation and computation of resource allocation for mixed traffic data is presented. In [8] some scenarios are proposed to evaluate the network performance. In [9] a cell is isolated for LTE network performance verification. In [10] the capacity of multiuser scheduling systems in different scenarios is analyzed. In [11] the AHP method is used in load balancing to avoid unicast and multicast mixed services congestion in LTE networks. The AHP technique is used to calculate the weights of the required radio resources. In [12] the authors propose a scheduling algorithm for LTE networks that takes into account the channel quality and the type of service for radio resources allocation. From an initial classification of real-time and non real-time traffic, audio and video users are added to a list of prioritization. The algorithm assigns allocation blocks to users considering the order of the list.

II. LTE SYSTEM An OFDMA scheduler allocates pairs of radio resources blocks (1ms and 12 subcarriers) to different users attempting to achieve the highest possible data flow in the cell [13,15]. These resource blocks pairs in LTE are the smallest units of radio resource allocation. These blocks can be independently modulated by bit rate flows from different users. Several scheduling algorithms found in the literature, have been employed in OFDMA systems. The best known are Best CQI or Maximum Carrierto-Interference Ratio (C/I) [7], Round-Robin [14] and Proportional Fairness (PF) [15]. The Best CQI technique classifies users according to the instantaneous CQI, providing more radio resources for those with better channel quality. This technique maximizes the total flow but limits the capacity of users in the system, since only users with good communication conditions will receive radio resources. The advantage is the easy implementation. The Round-Robin technique makes the fairest scheduler. The user who has been waiting for the most time will be serviced first. However, this technique does not take into account the channel quality and due to system link adaptation, it can allocate a sizeable resource share to a bad channel indicator user. This can significantly degrade the overall data throughput.

Proportional Fairness [16] is the most commonly used technique [17]. It seeks to attend the tradeoff between maximizing the overall data throughput and enabling a minimum level of service for all users. This is usually accomplished by adding some form of prioritization inversely proportional to the anticipated radio resources directed to a given user. The customizable algorithm proposed in this paper utilizes the Proportional Fairness algorithm as a basis, adopting the AHP multicriteria decision making method to set different weights for different criteria, thereby adjusting the preferences in a customizable manner.

III. PROPOSED SCHEDULING METHOD Decision problems involving a finite number of alternatives arise frequently. Examples include choosing an university for study, selecting an employee for a particular job, acquiring a home or a car, predicting the outcome of an election, or reducing costs of department to meet new budget constraints. The analytic hierarchy process (AHP) is a prominent tool for dealing with decisions under certainty, where subjective judgment is quantified logically and then used as a basis for a decision. The AHP technique is designed for situations in which ideas, feelings and emotions that affect the decision process are quantified to provide a numerical scale to determine priorities of alternatives [6]. The general structure of the AHP may include multiple hierarchies of different criteria weights. The general structure of the AHP may include multiple hierarchies of criteria. A corresponding matrix for each hierarchy quantify the judgment of the decision maker with regard to the relative importance of different criteria A) Algorithm Description The proposed scheduling scheme is a modified version of the Proportional Fairness scheduler that uses the AHP technique. The details of the development and operation of the Proportional Fairness algorithm are discussed in [14, 20]. The steps for our implementation of AHP method in a customizable algorithm scheduler are as follows: 1. Definition of the criteria for the first hierarchy. Each hierarchy can have n criteria. The chosen ones are: achievable data rate (R) and CQI (C). These criteria were selected by their importance in the resource allocation process; 2. Generation of a comparison matrix A, n x n, pairwise, which quantifies the judgment of the decision maker with regard to the relative importance of different criteria. According to [6], pairwise comparison is made in such a way that the criterion in row i (i = 1,2, ..., n) is classified in relation to all other criteria. Denoting the element (i, j) of the matrix A as aij, AHP method uses a discrete scale from 1 to 9 where aij = 1 means that i and j have equal importance For example, aij = 5 indicates that i

is significantly more important than j, and aij = 9 indicates that it is far more important than j. Other values between 1 and 9 are interpreted accordingly. For consistency, if aij = k then aij = 1 / k. In addition, all diagonal elements aij of A must equal 1. 3. Normalization of the matrix A. Each element of the matrix A is divided by the sum of the corresponding column. It generates a new normalized matrix N. The definition of the relative weights of each criterion is obtained by the mean of each row of the normalized matrix N; 4. Definition of alternatives and the relative importance between them; 5. Creation of new matrices, each within its respective criterion. The matrices sizes depend on the amount of alternatives within each criterion; 6. Normalization of AR and Ac matrices in new matrices NR and Nc, repeating the procedure in step 3; 7. Definition of the alternatives weights within each hierarchy criteria by calculating the average of each row of the normalized matrices NR and Nc; 8. Calculate the compound weight of each component. B) Case Study The decision problem of the proposed algorithm is to prioritize selected types of services for efficiently allocating radio resources. For this particular case study, video users are selected to be prioritized through the use of tailor made criteria comparison matrices. To demonstrate the procedure, a simple example involving the allocation of weights for two criteria and three users as alternatives is shown. The pairwise comparison matrix for two criteria, namely user achievable data rate and CQI, was chosen to establish the first hierarchy of criteria. The chosen weights compose the comparison matrix A, given by

1 1/ 5 A=  5 1 

(1)

The symbols R and C are used to denote the respective criteria. The first row and first column of the matrix A refer to criterion R (achievable rate), and the second row and column refer to criterion C (CQI). The fact that the value a21 = 5 means, in this case, that CQI is much more important than the achievable rate. This assignment implies that a12 = 1/5. The relative weights of R and C can be determined through the normalization in the new matrix N, given by

 0.17 0.17  N =   0.83 0.83 

(2)

The process requires dividing the elements of each column by the sum of the elements of the same column. Thus, to calculate matrix N, we divide the elements of the first column by (5 + 1 = 6) and the element of the second column by (1 + 1/5 = 1.2). Therefore, the desired relative weights for R and C criteria, wR and wC respectively, are calculated as the average of the corresponding row. The calculations result in 0,17 and 0,83 respectively, which indicate that for this case, R’s importance is proportional to 17%, whereas C is 83% important, that is, ( wR , wC , ) = (0.17;0.83) . The relative weights for three user alternatives (1, 2, 3) are determined within each criterion C and R using two comparison matrices AC (3) and AR (4), whose elements are based on the judgment of the relative importance of file transfer (F), audio (D) and video (V) traffic. The first row and column of matrices AC and AR refer to service F. The second row and column refer to the audio service (D), and the third row and column refer to video (V).

 1 1 / 4 1/ 7    AC =  4 1 1/ 3  7 3 1  

(3)

 1 1/ 2 1/ 3    AR =  2 1 1/ 2  3 2 1  

(4)

Repeating the procedure of normalization of matrix N, normalized matrices NC (5) and NR (6) are obtained, given by

 0.0833 0.0588 0.0968    N C =  0.333 0.2352 0.2258   0.5833 0.70588 0.6775   

(5)

 0.1666 0.1428 0.1818    N R =  0.333 0.2857 0.2727   0.5 0.5714 0.5455  

(6)

The means of the rows of matrix NC were calculated as follows. The average of the first row, corresponding to F traffic weight, is wCF = 0.0796. In the second row, the D traffic weight is wCD = 0.2646. Finally, in the third row, corresponding to V traffic weight, we have wCV = 0.6556. Repeating the procedure for matrix NR, we obtain ( wRF , wRD , wRV ) = (0.1638;0.2972;0.5389) . To demonstrate the decision process, Fig. 1 presents the weights of criteria and alternatives.

 1 1/ 4 1/ 7   0.0796   0.2395       AC wmc =  4 1 1/ 3   0.2646  =  0.8019  (10) 7 3 1   0.6556   2.0075   Thus we compute

nmax = 0.2395 + 0.8019 + 2.0075 = 3.0489

(11)

Considering n = 3 users, it is possible to calculate the consistency index (CI) of AC and the random consistency index (RI) of AC as Fig. 1: Structure of the decision problem

The first hierarchy criterion represents the importance of the relative weights of CQI (C) and achievable rate (R) in the user selection process for resource allocation. The alternatives are the users with weights assigned to the traffic type related to hierarchy criterion 1. The procedure consists of multiplying the weight of criteria C (wC) by the weight of user traffic i ∈ (1, 2,3) and then multiply the weight of criteria R (wR) by the weight of user traffic i ∈ (1, 2,3) .The determination of the compound weight is given by the sum of these products. For comparison between algorithms, consider that in the Proportional Fairness scheduler, R(k) represents the achievable data rate for user i in transmission interval k, which depends on the current conditions of the user communication channel. Ti(k) represents the accumulated historical data by user i in transmission interval k [21]. The user selection process in this scheduler is based on the proportion

P (i ) =

Ri( k ) Ti ( k )

CI =

nmax − n 3.0489 − 3 = = 0.0244 n −1 3 −1

(12)

RI =

1.98(n − 2) 1.98(3 − 2) = = 0.66 n 3

(13)

Finally we calculate the consistency ratio as

CR =

In the Customizable Scheduler, different users impose an additional hierarchical level in the AHP tree. Thus the weights calculated in the present work modify P, increasing or decreasing the resource allocation rate for different data services. C) Consistency Level In this case study, the Ac (3) and AR (4) matrices do not have identical columns. Thus, the calculation of the consistency ratio for each is required.

Repeating the procedure for matrix AR given in (4), we represent the average vector by wmR , and calculate

AR wmR

 1 1/ 2 1/ 3  0.1638   0.4921       =  2 1 1/ 2  0.2972  =  0.8943  3 2    1    0.5389   1.6248 

wmR = ( wRF , wRD , wRV ) = (0.1638; 0.2972; 0.5389) (9) Denoting the mean vector in (18) by

wmc we have

(15)

This gives

nmax = 0.4921 + 0.8943 + 1.6248 = 3.0112

(16)

Considering n = 3, we calculate the consistency index and the random consistency index of AR

CI =

nmax − n 3.0112 − 3 = = 0.056 n −1 3 −1

(17)

RI =

1.98(n − 2) 1.98(3 − 2) = = 0.66 n 3

(18)

Consider the row mean vector of the normalized matrices in (5) and (6), given below.

wmc = ( wCF , wCD , wCV ) = (0.0796; 0.2646; 0.6556) (8)

(14)

As CR < 0.1, the consistency level of AC is considered acceptable.

(7)

The user with the highest value of P(i), given in Eq. 7 is scheduled at each time.

CI 0.0244 = = 0.0370 RI 0.66

Then we calculate the consistency ratio of AR as

CR =

CI 0.056 = = 0.0085 RI 0.66

(19)

IV. SIMULATION RESULTS For the considered scenario implementation and testing we used the LTE system level simulator [18] from Vienna University of Technology. This system level simulator allows us to evaluate the performance of the AHP-PF scheduler algorithm in the context of LTE networks. The simulation is performed by defining a region of interest (ROI) where the eNodeBs and the mobile terminals are positioned. An important parameter to be set up in the simulator is the simulation time. This parameter is configured in terms of transmission time intervals (TTI), an LTE parameter set to 1 millisecond, corresponding to the time of one subframe in the LTE signal frame structure. The network design consists of a grid in which seven eNodeBs are placed with one central eNodeB and six adjacent ones. In the simulated scenario we have considered 3 users connected to one sector of the central eNodeB, randomly distributed in the Region of Interest (ROI). Table II summarizes the system configuration used for the simulation. The module utilized is able to schedule user traffics with mixed data types. We considered an example which combined voice, file transfer (ftp) and video traffic sources. The customizable scheduler differentiates between the various types of data traffic. Thus, in addition to the use of channel quality and other performance criteria, it can prioritize access to network resources for selected types of data.

The scheduler was customized using the criteria of CQI and achievable throughput as illustrated in Fig. 1. We use resource block allocation rate and data throughput as figures of merit to evaluate the customizable scheduler performance as compared to the proportional fairness approach. Fig. 2 presents the allocated RBs curves for mixed traffic using the proposed customizable scheduler. The blue curve represents video traffic, the green curve represents file transfer traffic and the red curve represents the audio traffic.

Assigned RBs - Customizable Scheduler (CS) 25 Video User FTP User Audio User 20

Number of RBs

Again, as CR < 0.1, the consistency level of AR is considered acceptable

15

10

5

0

0

10

20

30

40

50 Time (ms)

60

70

80

90

100

Fig 2. Allocated RBs curves using the proposed Customizable Scheduler. .

Fig 3 shows the allocated RBs curves for mixed traffic using the Proportional Fairness Scheduler. The curve colors are the same as in Fig 2.

Assigned RBs - Proportional Fairness Scheduler (PF) 25

TABLE II summarizes the parameters used in the simulation.

Video User FTP User Audio User 20

Parameters Frequency Bandwidth SISO Transmission Simulation length (TTI) Slot Duration Distance between eNodeBs eNodeB Transmission Power Network Layout Number of Users User Equipment Speed Number of RBs Bandwidth of RBs Number of Subcarriers Delay CQI report Number of subcarriers per RB Subcarrier Spacing

Values 2,14 GHz 5MHz 1x1 1000 0,5ms 500m 20 W 7 eNodes / 3 sectors 3 5 Km/h 25 180 kHz 300 3 ms 12 15kHz

Assigned RBs

TABLE II - Simulation Parameters 15

10

5

0

0

10

20

30

40

50 Time (ms)

60

70

80

90

100

Fig 3. Allocated RBs curves using the Proportional Fairness scheduler.

In the simulations, the fact that there were no data contributing users in the neighboring cells did not seem to be a determining factor for the improved data throughput. According to [19], the power control employed in LTE networks is aimed at improving system capacity, while generating the minimum possible interference from users in adjacent cells.

Fig. 4 shows a comparison between the RBs allocation curves for the video user using the customizable scheduler and the proportional fairness scheduler. As we can note, the customizable scheduler technique provides better RB allocation for video traffic source, once it was prioritized by the weight choices in the comparison matrix. The difference between the two curves in Fig. 4 may be increased or decreased depending on the elements of the AHP comparison matrix. The adjustment of this matrix must be made by the management of the network in order to achieve the desirable levels of prioritization among the users. In a closed loop scenario, the comparison matrix may be continually adjusted to optimize given measures of performance. The AHP provides a knob for priority control among users and optimization of various criteria. The customizable scheduler achieves the same performance as the proportional fairness scheduler when all comparison matrix elements are equal to 1.

25 PF CS

Assigned RBs

20

15

10

5

0

10

20

30

40

50 Time (ms)

60

70

80

90

V. CONCLUSIONS This paper presents a heuristic, adaptable multi-user scheduling algorithm, customizable to prioritize selected traffic sources in the LTE downlink channel. The proposed scheduler algorithm is based on proportional fairness, and uses the Analytic Hierarchy Process (AHP) technique to incorporate performance criteria and to adjust resource allocation according to user traffic. The use of decision making techniques to designate performance criteria appears as an interesting solution for differential treatment of multiple traffic sources. The flexibility of the present technique is provided by the choice of different performance criteria or parameters in the AHP pairwise comparison matrices. It is possible to improve resource allocation for selected users in applications that demand special treatment such as distance learning, live streaming, video on demand, online gaming, and others.

Assigned RBs Video User - CS vs PF

0

memory and scenario complexity. Very long lengths may exceed memory capacity, affecting simulation results.

100

Future work is planned to take into account new scenarios, for example, with high speed vehicles. Furthermore, emphasis should be given to the analysis of the most adequate selection of AHP comparison matrices to achieve desired levels of user prioritization and rate performance. Additionally, it is important to consider the overhead imposed by these and other scheduling techniques [20].

Fig 4. Comparation between Customizable Scheduler (CS) and Proportional Fairness (PF) for the video user.

The overall cell performance is presented in Fig. 5. As we can see in the graph, the customizable scheduler obtained better throughput performance than the proportional fairness scheduler.

ACKNOWLEDGEMENT The authors would like to thank CNPq, CAPES and UNIFEV for their support. REFERENCES [1]

Total Cell Throughput AC vs PF 18

[2]

16

Throughput (Mbps)

14

[3]

12

[4]

10 PF CS

8 6 4

[5]

2 0

0

10

20

30

40

50 Time(ms)

60

70

80

90

100

Fig 5. Comparation between Customizable Scheduler (CS) and Proportional Fairness (PF) for total cell throughput.

In our simulations, the traffic generation length of the selected channel is based on a tradeoff between computer

[6] [7]

3 GPP TS 36.300, 2010, “3rd GENERATION PARTNERSHIP PROJECT”, 3GPP TS 36.300: Overall description – Stage 2. Mar. 2010. V8.12.0. MCQUEEN, D., “3GPP LTE – The Momentum Behind LTE Adoption”, IEEE Communications Magazine. February, 2009. ZOU, W. Y.; WU, Y., “COFDM: An Overview”, IEEE Tans. On Broadcasting. Vol. 41, nº 1. March, 1995. XUE, M.; SANDRASEGARAN, K.; RAMLI, H. A.; LIN, C. “Performance Analysis of Two Packet Scheduling Algorithms in Downlink 3GPP LTE System”, 2010 IEEE 24th International Conference on Advanced Information Networking and Applications Workshop. ALI, S.; ZEESHAN, M., “A Delay-Scheduler Coupled Game Theoretic Resource Allocation Scheme for LTE Networks”, Frontiers of Information Technology (FIT), 2011 - Page(s): 14 – 19. TAHA , H. A. “Operations Research: An Introduction”, 8th Edition. Pearson Publisher, 2008. BALINT, C.; BUDURA, G.; MARZA, E., “Scheduling Techniques Evaluation in LTE Systems with Mixed Data Traffic”, 9th International Symposium on Electronics and Telecommunications (ISETC), Timisoara, Romania, Nov. 10-11, 2010.

[8]

BASUKALA, R; RAMLI, H. A. M.; SANDRASEGARAN, K.; CHEN, L., “Impact of CQI Feedback Rate/Delay on Scheduling Video Streaming Services in Lte Downlink”, 12th IEEE International Conference on Communication Technology (ICCT), Nanjing, China, Nov. 11-14, 2010. [9] SIMONSSON, A.; HAGERMAN, B.; CHRISTOFFERSSON, J.; KLOCKAR, L.; KOUTSIMANIS, C.; COSIMINI, P.R., “LTE Downlink Inter-Cell Interference Assessment in an Existing GSM Metropolitan Deployment”, 72nd IEEE Vehicular Technology Conference Fall (VTC 2010-Fall), Ottowa, Canada, Sep. 6-9, 2010. [10] KAMPEAS, J., COHEN, A., GUREWITZ, O., “Capacity of Distributed Opportunistic Scheduling in Heterogeneous Networks”. 50th Annual Allerton Conference on Communication, Control, and Computing (Allerton), Monticello, Illinois, Oct. 1-5, 2012. [11] WANG, M., Feng, C. e Zhang, T., “Analytic hierarchy process in load balancing for the multicast and unicast mixed services in LTE”, Wireless Communications and Networking Conference (WCNC), 2012 IEEE. April 2012. Shanghai. Pag 2735-2740. [12] AKYILDIZ, H. A., Akkuzu, B., Hökelek, I. e Çırpan, H., “A LTE downlink scheduler with reconfigurable traffic prioritization”, IEEE International Black Sea Conference on Communications and Networking (BlackSeaCom), Odesa, Ukraine, May 27-30, 2014, pp. 5862. [13] LUO, H., “Quality-driven cross-layer optimized video delivery over LTE”, IEEE Communications Magazine, Vol. 48, nº 2, pp. 102-109, 2010. [14] RAMLI, H. A. M.; BASUKALA, R; SANDRASEGARAN, K.; PATACHAIANAND, R., “Performance of Well Known Packet Scheduling Algorithms in the Downlink 3GPP LTE System”, IEEE 9th Malaysia International Conference on Communications (MICC), Kuala Lumper, Malaysia, Dec. 15-17, 2009. [15] SADIQ, B.; MADAN, R.; SAMPATH, A., “Downlink Scheduling for Multi-class Traffic in LTE”, EURASIP Journal on Wireless Communications and Networking - 3GPP LTE and LTE Advanced. Vol 9, November 2009. [16] KIM, I.; HAN, Y.; CHUNG, H. K., “An Efficient Synchronization Signal Structure for OFDM Based Cellular Systems”, IEEE Trans. On Wireless Communications, Vol. 9, January 2010. [17] JUNGNICKEL, V. et al., “Interference-aware scheduling in the multiuser MIMO-OFDM downlink”, IEEE Communications Magazine, Vo. 47, nº. 6, pp. 56-66, 2009. [18] [Online] “LTE Simulator”. Available at: http:/www.nt.tuwien.ac.at/ ltesimulator/. [19] DAHLMAN, E.; PARKVALL, S.; SKOLD, J. BEMING, P., “3G Evolution - HSPA and LTE for Mobile Broadband”, 1st ed. Academic Press, 2007. [20] KOLEHMAINEN, N., “Channel Quality Indication Report Schemes for UTRAN Long Term Evolution Downlink”, IEEE Vehicular Technology Conference, pp, 2522-2526, Singapur, May 11-14, 2008. [21] VUKADINOVIC, V.; KARLSSON, G., “Video Streaming Performance Under Proportional Fair Scheduling”, IEEE Journal on Selected Areas in Communication. Vol 28, pp, 399-408, April, 2010.