IEEE ICC 2015 - Communications QoS, Reliability and Modeling Symposium

QoS and Energy Efficient Resource Allocation

in Downlink OFDMA systems
Dionysia Triantafyllopoulou, and Klaus Moessner
Institute for Communication Systems (ICS)
University of Surrey
Guildford, Surrey, GU2 7XH, United Kingdom
email: {d.triantafyllopoulou, k.moessner}@surrey.ac.uk

algorithms is compared, with M-LWDF achieving higher

AbstractIn this paper we present and evaluate the system throughput and fairness.
performance of a resource allocation algorithm to enhance the In [8], a two-level scheduling algorithm is described. In the
Quality of Service (QoS) provision and energy efficiency of upper level, the amount of data of each source in order to
downlink Orthogonal Frequency Division Multiple Access
satisfy its delay constraint is calculated. At the lower level the
(OFDMA) systems. The proposed algorithm performs resource
allocation using information on the downlink packet delay, the PF scheduler is used. A weighted round-robin resource
average delay and data rate of past allocations, as well as the allocation algorithm is proposed in [9] to improve system
downlink users buffer status in order to minimize packet throughput, guarantee application layer QoS, in terms of video
segmentation. Based on simulation results, the proposed distortion, and ensure fairness, taking into consideration the
algorithm achieves significant performance improvement in channel quality, the packet delay constraints and the average
terms of packet timeout rate, goodput, fairness, and average
data rate of each user. In [10], a resource allocation algorithm
delay. Moreover, the effect of poor QoS provision on energy
efficiency is demonstrated through the evaluation of the that operates in three phases is presented. In the first phase, the
performance in terms of energy consumption per successfully resource blocks are allocated to the User Equipment (UE)
received bit. devices based on their Channel Quality Information (CQI).
Then, the packet delays are predicted. Finally, the
Index Terms Energy efficiency, Orthogonal Frequency transmission order is rearranged and the packets that cannot
Division Multiple Access (OFDMA), Quality of Service (QoS),
meet their delay requirements are discarded.
resource allocation.
Very useful conclusions regarding the trade-off between the
I. INTRODUCTION energy efficiency (EE), i.e., the ratio of throughput over the
total power consumption expressed in bits per Joule, and QoS
ne of the major challenges in future mobile
O communication networks is the need for increased
capacity, Quality of Service (QoS) provision and energy
in OFDMA networks are drawn in the recent bibliography.
Specifically, the relation between energy efficiency and
spectral efficiency, i.e., the ratio of throughput over the
efficiency. Among the most prominent approaches to achieve bandwidth expressed in bits per second per Hertz, in a single-
these goals is the design and employment of efficient resource cell OFDMA network is shown to be a quasi-concave function
allocation schemes. To this end, significant research progress [11]. In [12], the authors propose a method for energy efficient
has been made in the area of downlink resource allocation in resource allocation and an algorithm for trading bandwidth for
Long Term Evolution (LTE), and Orthogonal Frequency energy efficiency during low load periods. A scheduling
Division Multiple Access (OFDMA) networks in general [1]. technique that, combined with the use of an energy efficient
n [2]-[10], emphasis is given on resource allocation scheduler, allows users whose energy consumption is
schemes that aim at enhanced QoS provision. Specifically, in dominated by control channel overhead to reduce their overall
[2], the authors propose and compare the performance of two energy expenditure by reducing the number of resource blocks
scheduling algorithms, i.e., a maximum-rate scheduler, which allocated to them is presented in [13]. In [14], an optimization
prioritizes the users based on the highest supported bit rate problem for the energy efficiency maximization subject to
according to their channel quality, and a proportional-fair (PF) users minimum transmission rate requirements and maximum
scheduler, which improves fairness among users. Two of the transmission power constraints is described.
most notable resource allocation rules, namely the exponential According to the related literature, downlink resource
(EXP) rule and the log rule are described in [3] and [4], allocation is usually performed per resource block, i.e., in each
respectively. In [5], the performance of the well-known subframe a resource block is allocated to the user who
maximum-rate, round robin, PF, EXP/PF [6], and maximum- maximizes a specific metric. The overall performance of the
largest weighted delay first (M-LWDF) [7] resource allocation system highly depends on the parameters that comprise this
metric. However, this approach does not take into

978-1-4673-6432-4/15/$31.00 2015 Crown 5967

 IEEE ICC 2015 - Communications QoS, Reliability and Modeling Symposium

consideration the effect of packet segmentation on the overall II. SYSTEM MODEL
system performance. Specifically, in case the allocated The system model consists of a single LTE macro cell and a
resource blocks are not enough to transmit a users packet as a number of UE devices, randomly deployed in the macro cell
whole, the need for packet segmentation introduces overhead, coverage area. For the remainder of this document the terms
due to the introduction of separate protocol headers in each user and UE are used interchangeably. Each user has an active
packet segment. Furthermore, in the case of real-time real-time video connection on the downlink and the eNodeB is
applications, where Automatic Repeat reQuest (ARQ) responsible to allocate the available resources in a fair, QoS
techniques are not employed, the loss of one packet segment and energy efficient manner, employing the proposed resource
results in the need to discard all the already received segments allocation algorithm. Table II summarizes the parameters used
of the same packet and the waste of the respective resources for the formulation and performance evaluation of the
used for their transmission. Moreover, as shown in Table I, the proposed algorithm.
proposals that focus on QoS provision to real time applications In the time domain, downlink LTE transmissions are
do not consider energy efficiency [2]-[10], while the ones that organized into radio frames, each of which consists of two

subframes of length
each. Each subframe consists of two

mainly focus on energy efficiency do not explicitly support the half-frames. A half-frame consists of five equally sized

equally sized slots. Each slot consists of 

strict constraints of realistic real-time applications in terms of

symbols, including cyclic prefix. The exact value of 

packet delay [12]-[14]. OFDM

Motivated by the above, in this paper we propose a QoS-
oriented and energy efficient resource allocation algorithm for depends on the cyclic prefix length, which is configured by the

transmitted signals in each slot consists of  

downlink LTE systems. Resource allocation is performed higher layers. The resource grid describing the downlink

subcarriers and  OFDM symbols. The smallest physical

taking into consideration the packet delays in the downlink

direction, the average delay and data rate of allocations in the

past, as well as downlink buffer status in terms of packet resource in LTE is a resource element, consisting of one
segmentation. The main contributions of this paper with subcarrier during one OFDM symbol. Resource elements are

consists of 
respect to the reviewed literature are summarized as follows: grouped into resource blocks, where each resource block
1) Consideration of the effect of packet segmentation,
domain and one slot consisting of 

consecutive subcarriers in the frequency
performed at the Radio Link Control (RLC) layer, on the
OFDM symbols in
resource allocation. The proposed algorithm i) aims at the time domain [15]. A scheduling block consists of two
allocating enough resources in each subframe to each user
equal to
, and is the minimum amount of resources that can
consecutive resource blocks, spanning a subframe of length
in order to transmit their packets as a whole and ii)
prioritizes users whose pending packets are already
be allocated to a user in a subframe.
segmented in order to transmit packet segments as soon as
possible and avoid the discarding of already received
III. THE PROPOSED ALGORITHM
As a first step, the set of active users  is sorted in
packet segments due to expiration.

descending order of  . This is a metric that aims to

2) Demonstration of the effect of QoS on energy efficiency
through the evaluation of the system performance in terms
of the total energy consumption per successfully received provide higher resource allocation priority to users with
bit. increased waiting time with respect to the delay threshold,
This paper is organized as follows. Section II introduces the high average delay and low average data rate of their

transmitted packets. To this end,   is defined as

system model. Section III describes in detail the proposed allocations in the past, as well as segments of already
downlink resource allocation algorithm, whose performance is
  =  exp

evaluated through simulations in section IV. Finally, section V

(1)
contains conclusions and discusses on plans for future work. ,

power spent for already transmitted packet segments,  is a

In order to avoid wasting of resources and transmission

TABLE I.
QoS AND ENERGY CONSUMPTION PARAMETERS CONSIDERED IN DOWNLINK RESOURCE ALLOCATION
Reference Traffic model QoS parameters Energy efficiency parameters
PF-Multiuser [2] Not explicitly specified Average bit rate maximization, fairness -
Mutually independent ergodic Markov
EXP-Rule [3] Throughput -
chains with countable state spaces
Log Rule [4] i.i.d. Bernoulli process, 1kb packet size Average packet delay, 99th percentile delay -
EXP/PF, M-LWDF [5] 128 kb/s video streaming service System throughput, packet loss rate, fairness -
Packet loss rate minimization, average
H.264 video, voice over Internet Protocol
[8] goodput maximization, fairness, peak signal to -
(VoIP), infinite buffer (best effort)
noise ratio (PSNR)
[9] H.264 video PSNR -
Invalid packet rate, goodput, packet average
[10] H.264 video -
delay, packet dropping rates
TCoM [12], [13] Constant size buffer model Data rate RF energy consumption gain
Not explicitly specified, minimum rate
QA-ERS [14] QoS satisfaction index Energy efficiency
requirement per user 128 kb/s

5968
 IEEE ICC 2015 - Communications QoS, Reliability and Modeling Symposium

TABLE II.
DEFINITION OF SYSTEM MODEL PARAMETERS
Parameter Definition
Downlink scheduling metric of user
Segmentation bias
Downlink queuing delay of user (s)
, Queuing delay threshold of user (s)

Average downlink delay of user (s)
Average downlink rate of user (b/s)
Average delay and rate calculation factor
Instantaneous downlink rate of user (b/s)
Length of HoL data packet of user (b)
Modulation of user (b/symbol)
Coding rate of user
Number of required scheduling blocks of user with
,
Modulation and Coding Scheme (MCS) ,
Number of data carrying resource elements in a

downlink scheduling block
Set of available scheduling blocks for user with
, ,
MCS ,
Number of bits that can be allocated to user with
,,
MCS ,
, SNR of user on scheduling block
, Transmission power of user on scheduling block
Acceptable packet loss rate of user
Total number of resource blocks per slot
Number of OFDM symbols per downlink resource

block
Number of subcarriers per resource block

, Number of resource blocks per scheduling block
Subframe duration (s)
Set of users
Set of available scheduling blocks
Set of allocated scheduling blocks to user
Fig. 1. Flowchart of the proposed algorithm
parameter that gives resource allocation priority to users
scheduling blocks in order to transmit its HoL packet as a
whose head-of-line (HoL) packet is segmented. The aim of
  ,  , depends on the HoL packet size  , the
whole. The required amount of scheduling blocks,
this is to transmit segments of the same packet as closely as

discarding of the already successfully received segments.  is modulation  and coding rate  , and the number of data
possible to each other in order to avoid their expiration and the
carrying resource elements per downlink scheduling block 
 , HoL packet of user  is segmented
formulated as follows:

 =  ,
as follows:
1  , else   ,   =  .
(2)
where 0  1.

(5)

  is the delay of the HoL packet of user , while ,

user  is determined in (6), shown at the bottom of the page,

This number is calculated for all MCSs. Then, the MCS of

where , ,  is the number of available scheduling blocks

is the delay threshold, beyond which the packet is no longer

  and   are the average delay and data rate,
 of user  with MCS  ,  . A scheduling block  is
considered usable and is discarded by the users buffer.

respectively, experienced by user  in the past, and are considered available to user  with MCS  ,   if its Signal-
to-Noise Ratio (SNR) , exceeds a specific threshold  .
  =   + 1 
   1 and
calculated using a weighted moving average formula:

 
Therefore, in case there are enough scheduling blocks to
  =   1
+  
(3)
accommodate the users HoL packet as a whole, the MCS
where   is the instantaneous downlink data rate of user 
(4)
1,
selected is the one that results in the need for the minimum
and 0  1. The incorporation of    and   in

number of scheduling blocks. Otherwise, the MCS is the one
  allows the prioritization of users that were served with
allocated to this user, i.e., , , = , , 
 log  .
that results in the maximum number of bits that can be
high average delay and low average data rate in the past, thus

in each subframe is shown in Fig. 1. For each user  , in

increasing the fairness of the proposed solution. A flowchart of the proposed resource allocation performed

descending order of  , the proposed downlink resource

In order to avoid packet segmentation, in each subframe the
proposed algorithm aims at allocating to each user enough

arg min   ,  !,  ,  :   ,   , , 

 ,   =  "
arg max , , !,  ,  :   ,   > , , 
,
(6)
,

5969
 IEEE ICC 2015 - Communications QoS, Reliability and Modeling Symposium

allocation algorithm performs the following steps, if there are TABLE III.
PERFORMANCE EVALUATION PARAMETERS
scheduling blocks available for allocation, i.e., set is non-
Parameter Value
empty: Physical layer parameters Channel bandwidth: 10MHz,
1) Firstly, the users MCS is determined, based on (5) and Subframe duration ( ): 1ms,
(6). Number of RBs ( ): 50

users HoL packet as a whole, i.e.,   ,  

Resource block format
2) If there are enough scheduling blocks to accommodate the Number of subcarriers per RB ( ):
12, Number of symbols per RB

, , , the set ,, is ordered in descending

( ): 7, Subcarrier spacing: 15kHz

order of SNR , , and the   ,   scheduling

Reference Signal transmissions 4 Reference Signals per RB
Downlink control region size 2 OFDM symbols per subframe
TDD configuration Configuration 1, DL:UL 3:2

necessarily contiguous, are the ones that comprise # , i.e.,

blocks of this set with the highest SNR, which are not Modulation and Coding
QPSK 1/2, 16-QAM 1/2, 64-QAM 3/4

the set of all scheduling blocks allocated to user  in this

Schemes
Inter-eNodeB distance 500m
Path loss model 128.1+37.6log , : distance from
subframe. the eNodeB (km)
3) Otherwise, if the available scheduling blocks of set Transmitter antenna gain 18dBi

whole, i.e.,   ,   > , , , all of them are

, , are not enough to accommodate the packet as a Receiver antenna gain 0dBi
Cable loss 2dB

allocated to user  and the packet needs to be segmented.

Receiver Noise Floor -95dBm

4) When the resource allocation for user  is finalized, the

Interference margin 3dB

user is removed from  and all its allocated scheduling

Control channel overhead 1dB
Shadowing Log normal, =8dB

blocks, i.e., belonging to # , are removed from the set

Fading Rayleigh
Maximum transmission power 20W

If and  , the resource allocation algorithm

of available scheduling blocks. Maximum tolerable delay
20ms
(, )
RLC mode Unacknowledged mode (UM)
proceeds to the next user, otherwise the resource allocation for Traffic model H264 video traffic QCIF 176x144
this subframe is complete and the algorithm terminates. Protocol header sizes RTP/UDP/IP with ROCH
Compression: 3 bytes, PDCP: 2bytes,
RLC: 3 bytes, MAC: 2 bytes, CRC:
IV. PERFORMANCE EVALUATION 3bytes
To evaluate the performance of the proposed resource Segmentation bias ( ) 0.9
Average delay and rate
allocation algorithm, a simulation model was built in calculation factor ()
0.2
MATLAB. The performance of the system employing the
proposed algorithm is compared to three legacy systems that [17], with a rate of 30 frames per second (fps). The created
employ the PF, EE, and M-LWDF algorithms, respectively. video traffic frames are provided to the resource allocator.
Specifically, according to the PF algorithm, a scheduling The channel model simulates the physical layer channel
block is allocated to the user who maximizes the ratio of conditions by providing path loss, shadowing, and short-term

 =   . The EE algorithm allocates a

fading. It produces bit errors randomly for each connection,
,
instantaneous data rate to the average data rate, i.e.,
based on the allocated scheduling blocks and the MCS per
scheduling block to the user who maximizes the ratio of user. Perfect channel knowledge is assumed for the purposes
of Adaptive Modulation and Coding (AMC). The link budget
i.e., ,  = % '&, (, where &, is the transmission
instantaneous data rate to the transmission power required,
parameters are summarized in Table III [18].
power of user  on scheduling block . However, it has to be
The resource allocator is the entity that is responsible for
allocating the downlink resources to the different UE devices
noted that no power control is performed in downlink LTE following either the proposed algorithm, or the legacy
systems, therefore all scheduling blocks have the same approaches.
transmission power level. Therefore, in this case the EE The simulation scenario considers an increasing number of
algorithm is similar to the maximum-rate algorithm. Finally, users, each one with one downlink video connection. The
user who maximizes the product ,  =
the M-LWDF algorithm allocates a scheduling block to the systems performance is evaluated in terms of packet timeout


% log ) , (    , where ) is the
rate, delay, goodput, fairness, and energy efficiency of

acceptable packet loss rate of user .

successfully received bits. All simulation model parameters
are summarized in Table III. In order to achieve statistical
The simulation environment consists of a single LTE cell accuracy, 50 simulation runs were executed.
and a variable number of UE devices within the cells Fig. 2 depicts the packet timeout rate, i.e., the number of
coverage area. The individual subsystems of the simulation packets that are discarded due to expiration in the unit of time,
model employed are as follows: with respect to an increasing number of users. In the case of
The traffic generator uses the Joint Scalable Video Model the PF and EE systems, the packet timeout rate follows a sharp
(JSVM) reference software [16] in order to generate variable- increase with the increase of the number of users due to the
length video traffic frames for each UE, starting at a random fact that the increased congestion results in excessive packet
instance within the first 33 ms of a simulation run. The video delays and packet expirations that cannot be avoided, since
sequence used is the well-known Highway video sequence delay is not considered in their resource allocation process. In
the case of the M-LWDF algorithm, the packet delay is taken

5970
 IEEE ICC 2015 - Communications QoS, Reliability and Modeling Symposium

0.9 18
PF PF
0.8 EE 16 EE
M-LWDF M-LWDF
Proposed Proposed
0.7 14

0.6 12

Delay (ms)
0.5 10
Rate

0.4 8

0.3 6

0.2 4

0.1 2

0 0
5 10 15 20 25 5 10 15 20 25
Number of users Number of users
Fig. 2. Average packet timeout rate versus the number of users. Fig. 3. Average delay versus the number of users

into consideration, however, the packets are not prioritized the increase of the number of users, as a result of the
based on their segmentation status, therefore the excessive increasing congestion, which leads to excessive packet delays
delay of at least one packet segment results in the expiration and timeouts. However, all the legacy systems experience a
and discarding of the whole packet. Therefore, the system rapid deterioration of the goodput with the increase of the
employing the proposed algorithm significantly outperforms number of users. On the contrary, the system employing the
the three legacy systems in terms of packet timeout rate. This proposed algorithm achieves a significantly improved
is a result of the prioritization of users based on their packet goodput, even in the cases of increased number of users.
delays with respect to their delay threshold as well as their In order to highlight the interdependency of the energy
segmentation status, therefore significantly reducing the efficiency and QoS provision, Fig. 6 depicts the systems
performance in terms of energy efficiency of successfully
packet expirations.
received bits. This is defined as the amount of data
Fig. 3 depicts the average packet delay with respect to an
successfully concatenated at the receivers RLC layer (in Mb)
increasing number of users. In the PF and EE systems, the
for a given amount of transmission energy (in J) and
average packet delay increases very quickly with the increase
represents the average energy consumption per successfully
of the number of users. This is a result of the fact that these
received bit. As it can be seen, in the proposed system the
algorithms do not take into consideration the packet delay in
energy efficiency of received bits is more than 6-times
the resource allocation process. As already described above,
improved compared to that of the legacy systems. This is a
even though the M-LWDF algorithm takes into consideration
result of the fact that, due to packet segmentation performed at
the packet delay it does not prioritize the packets based on
the RLC layer, a packet segment loss may be unrecoverable at
their segmentation status, therefore the increased delay of at
the receiving side, therefore leading to the waste of already
least one packet segment results in the delayed packet
received packet segments, whose transmission consumed
reassembly at the receiving side. Therefore, the system
energy. This could be partly mitigated by efficient ARQ
employing the proposed algorithm, which prioritizes users
schemes. However, these are not appropriate for real-time
with segmented packets and increased delay with respect to
applications, since the required retransmissions induce
their delay threshold, significantly outperforms the three
additional delays that may result in a packet having expired
legacy systems in terms of average packet delay.

evaluated using the Jain Index of Fairness, i.e., *+ =

Fig. 4 depicts the fairness of the four systems. Fairness is 1


|.|
, where
 is the

throughput of user  [19]. As expected, the PF and M-LWDF
0.9

0.8
systems that take into consideration the average data rate of
0.7
past allocations achieve higher fairness, compared to the EE
Fairness

system, which only considers the users instantaneous data 0.6

rates in the resource allocation. The system that employs the
0.5
proposed algorithm achieves improved fairness compared to
all the legacy systems. This is a result of the fact that the 0.4

packet delay    in the user prioritization, favoring users

proposed algorithm also takes into consideration the average 0.3
PF
EE
that have experienced high average delay in past allocations. 0.2
M-LWDF
Fig. 5 depicts the average goodput, i.e., the number of 0.1
Proposed
5 10 15 20 25
useful bits that reach the application layer in the unit of time. Number of users
As it can be seen, the goodput follows a declining course with
Fig. 4. Fairness versus the number of users.

5971
 IEEE ICC 2015 - Communications QoS, Reliability and Modeling Symposium

22 0.07

20
0.06
18

Energy Efficiency (Mb/J)

16 0.05
Goodput (kb/s)

14
0.04 PF
EE
12
M-LWDF
0.03 Proposed
10

8 0.02

6 PF
EE 0.01
4 M-LWDF
Proposed
2 0
5 10 15 20 25 5 10 15 20 25
Number of users Number of users
Fig. 5. Average goodput versus the number of users. Fig. 6. Energy efficiency of successfully received bits versus the number of
users.
before being reassembled at the receiving side. This result [4] B. Sadiq, S.J. Baek, and G. De Veciana, Delay-optimal opportunistic
highlights the effect that enhanced QoS provision has on scheduling and approximations: the log rule, IEEE/ACM Transactions
energy efficiency, since the lower packet loss rate and the on Networking, vol.19, no.2, pp.405-418, April 2011.
prioritization of segmented packets of the proposed system [5] H.A.M. Ramli, R. Basukala, K. Sandrasegaran, and R. Patachaianand,
Performance of well known packet scheduling algorithms in the
results in lower waste of already transmitted packet segments, downlink 3GPP LTE system, in IEEE 9th Malaysia International
and a larger amount of packets successfully being reassembled Conference on Communications (MICC), Kuala Lumpur, Malaysia,
by the receiver RLC layer. pp.815-820, Dec. 2009.
[6] J.-H. Rhee, J.M. Holtzman, and D.-K. Kim, Performance analysis of
the adaptive EXP/PF channel scheduler in an AMC/TDM system, IEEE
V. CONCLUSION Communications Letters, vol.8, no.8, pp.497-499, Aug. 2004.
In this paper we introduced a downlink resource allocation [7] M. Andrews, K. Kumaran, K. Ramanan, A. Stolyar, P. Whiting, and R.
Vijayakumar, Providing quality of service over a shared wireless link,
algorithm for OFDMA systems, which focuses on QoS IEEE Communications Magazine, vol.39, no.2, pp.150-154, Feb 2001.
provision in real-time applications and energy efficiency. The [8] G. Piro, L.A. Grieco, G. Boggia, R. Fortuna, and P. Camarda, Two-
proposed algorithm prioritizes users based on their estimated Level Downlink Scheduling for Real-Time Multimedia Services in LTE
packet delay, the average delay and data rate of past Networks, IEEE Transactions on Multimedia, vol.13, no.5, pp.1052-
1065, Oct. 2011.
allocations, as well as their buffer status with regards to packet [9] H. Luo, S. Ci, D. Wu, J. Wu, and H. Tang, Quality-driven cross-layer
segmentation. Simulation results highlight the considerable optimized video delivery over LTE, IEEE Communications Magazine,
performance improvement achieved by the proposed vol.48, no.2, pp.102-109, Feb. 2010.
algorithm compared to the PF, EE, and M-LWDF algorithms [10] W.K. Lai, and C.-L. Tang, QoS-aware downlink packet scheduling for
LTE networks, Elsevier Computer Networks, vol.57, no.9, pp.1689-
in terms of packet timeout rate, goodput, fairness, and average 1698, May 2013.
delay. In order to emphasize on the negative effect of poor [11] C. Xiong, G.Y. Li, S. Zhang, Y. Chen, and S. Xu, Energy- and
QoS provision on energy efficiency, the system was also Spectral-Efficiency Tradeoff in Downlink OFDMA Networks, IEEE
evaluated in terms of energy consumption per successfully Transactions on Wireless Communications, vol.10, no.11, pp.3874-
3886, Nov. 2011.
received bit. Our plans for future work include the extension [12] S. Videv, and H. Haas, Energy-Efficient Scheduling and Bandwidth-
of the proposed downlink resource allocation to a multicell Energy Efficiency Trade-Off with Low Load, in IEEE International
scenario, and the enhancement of its functionality with Conference on Communications (ICC), Kyoto, Japan, pp.1-5, June 2011.
interference coordination and avoidance features. [13] S. Videv, H. Haas, J.S. Thompson, and P.M. Grant, Energy efficient
resource allocation in wireless systems with control channel overhead,
in IEEE Wireless Communications and Networking Conference
ACKNOWLEDGMENT Workshops (WCNCW), Paris, France, pp.64-68, April 2012.
[14] X. Xiao, X. Tao, and J. Lu, QoS-Aware Energy-Efficient Radio
We would like to acknowledge the support of the University Resource Scheduling in Multi-User OFDMA Systems, IEEE
of Surrey 5GIC (http://www.surrey.ac.uk/5gic) members for Communications Letters, vol.17, no.1, pp.75-78, Jan. 2013.
this work. [15] 3GPP TS 36.211 V12.3.0 (2014-09): 3rd Generation Partnership Project;
Technical Specification Group Radio Access Network; Evolved
Universal Terrestrial Radio Access (E-UTRA); Physical channels and
REFERENCES modulation (Release 12).
[1] F. Capozzi, G. Piro, L.A. Grieco, G. Boggia, and P. Camarda, [16] Joint Scalable Video Model (JSVM) reference software,
Downlink Packet Scheduling in LTE Cellular Networks: Key Design http://www.hhi.fraunhofer.de/de/kompetenzfelder/image-
Issues and a Survey, IEEE Communications Surveys & Tutorials, processing/research-groups/image-video-coding/svc-extension-of-
vol.15, no.2, pp.678-700, Second Quarter 2013. h264avc/jsvm-reference-software.html
[2] R. Kwan, C. Leung, and J. Zhang, Proportional Fair Multiuser [17] YUV Video Sequences, http://trace.eas.asu.edu/yuv/
Scheduling in LTE, IEEE Signal Processing Letters, vol.16, no.6, [18] H. Holma, and A. Toskala, LTE for UMTS: Evolution to LTE-Advanced,
pp.461-464, June 2009. Chichester, U.K., John Wiley & Sons Ltd, 2011.
[3] S. Shakkottai, and A.L. Stolyar, Scheduling for multiple flows sharing [19] R. Jain, D. Chiu, and W. Hawe, A Quantitative Measure of Fairness
a time-varying channel: The exponential rule, Translations of the and Discrimination for Resource Allocation in Shared Computer
American Mathematical Society-Series 2, vol.207, pp.185-202, 2002. Systems, DEC Research Report TR-301, Sept. 1984.

5972