Content uploaded by Abdulaziz M. Ghaleb
Author content
All content in this area was uploaded by Abdulaziz M. Ghaleb on Nov 24, 2014
Content may be subject to copyright.
On the Extension of Traditional Resource Allocation
Algorithms in LTE-A to joint UL-DL Scheduling
with FDD Carrier Aggregation
Abdulziz M. Ghaleb∗, Elias Yaacoub∗
∗Qatar Mobility Innovations Center, Doha, Qatar
Email: {aghaleb, eliasy}@qmic.com
Ayad Atiyah Abdulkaޠ
†College of Engineering, Tikrit University, Iraq
Email: al.ayad@yahoo.com
Abstract—Throughput and spectral efficiency maximization
are two of the most challenging issues to be addressed by current
and future cellular networks. To achieve these goals however,
radio resources need to be utilized in a coordinated manner
to provide better network performance and ensure adequate
user quality of service (QoS). The radio resources are controlled
by LTE base station (eNodeB) and can be scheduled either
independently or jointly by considering both uplink (UL) and
downlink (DL). This paper addresses the joint UL-DL scheduling
in a carrier aggregation (CA) supportive LTE-advanced networks.
This work provides good insights of the implementation of the
traditional scheduling algorithms for joint UL-DL scheduling for
LTE-A utilizing the functionality of CA. We present preliminary
results from a MATLAB-based IS-Wireless system level simulator
that confirm the importance of the fair sharing of the resources
between the UL and DL channels while maintaining a good
QoS. Furthermore, simulation results show that the resource
utilization and spectral efficiency can be significantly improved by
dynamically changing the bandwidth portion between the UL and
DL. Finally, we also show that there is a significant improvement
in both UL and DL data rates using the proposed CA and joint
scheduling schemes.
Keywords—traditional scheduling algorithms; carrier aggrega-
tion; joint scheduling; resource allocation; LTE-A
I. INTRODUCTION
Scheduling is a key functionality of the radio protocol
stack, and it is performed by the MAC scheduler in the
eNodeB. The scheduler is responsible for allocating the radio
resources in both directions (DL and UL) while considering
the QoS requirements for all active radio flows (bearers). This
is achieved by allocating the available radio resource blocks
(RBs) to specific User Equipments (UEs) within the sector
or the cell for the transmission and reception of variable-size
Transport Blocks (TB). The scheduler runs every subframe or
Time Transmission Interval (TTI) and allocates the RBs to the
UEs in the DL and UL. A single TB may be allocated to a
UE per TTI [1]. The scheduling algorithms have a significant
impact on the performance of the individual eNodeB and
overall LTE network.
CA is an added enhancement in the LTE-A system by
which carriers or bands are aggregated together and users
can be scheduled on continuous or non-continuous component
carriers. The feature was primarily added to LTE in order to
provide substantial improvements to meet the IMT-A require-
ments for 4th Generation Networks (4G)[2].
The need for joint UL-DL scheduling is justified by the fact
that the evolutions of 4G wireless communication paved the
way to the advent of two-way or bidirectional communication.
This forces the traffic to be both symmetric (in bidirectional
application such as, online gaming, BitTorrent and video
conferencing) and asymmetric (web traffic) [3-5]. This type
of resource allocation has not been investigated enough in the
literature. There is no concrete framework defined for joint
UL-DL scheduling in LTE/LTE-A with carrier aggregation.
The existing work on LTE-/LTE-A scheduling is mostly
limited to UL or DL, separately. However, there are some
works on the joint UL-DL scheduling in other wireless
technologies. For example, the authors of [5] presented an
opportunistic joint UL-DL scheduling scheme for Wireless
Local Area Networks (WLANs) and focused on the unfair-
ness suffered by WLAN UL [6]. However, there exist some
differences in the access mechanisms between the WLANs and
LTE. The authors of [7] provided a resource allocation model
for the joint LTE/LTE-A system in DL only and considered CA
and the backward compatibility with conventional LTE users.
However, the authors focused on the fairness and neglected the
QoS provisioning which is a key element in the 4G network.
Furthermore, UL-DL joint scheduling was not addressed. A
lot of work has been done in joint cooperative scheduling
between the cells, either in UL or DL but not joint UL-
DL. The authors of [8] introduced cooperative scheduling for
jointly performing resource allocation for various UEs attached
to different cooperating eNodeBs for dynamic interference
coordination in the 3GPP LTE UL. Similar work has been
presented in [9] to address the joint scheduling and resource
allocation in UL orthogonal frequency division multiplexing
(OFDM) networks using the existing gradient-based schedul-
ing framework. The joint resource allocation in orthogonal
frequency division multiple access (OFDMA) systems was
addressed in [4, 10-13].
In this paper, we present a comparison of the imple-
mentation of the traditional resource allocation algorithms,
namely Round Robin (RR), Maximum signal-to-interference-
plus-noise ratio (MaxSINR) and Proportional Fair (PF), for the
QoS-aware joint UL-DL scheduling for FDD LTE-A utilizing
the functionality of CA. The framework presented enforces
the fair sharing of the resources between the UL and DL
channels while maintaining good QoS. The performance of
the implementation of the three traditional resource allocation
methods is compared and analyzed.
The rest of the paper is organized as follows. Section
II provides a detailed description of the joint UL-DL model
used in the simulation and gives and overview of the use of
the traditional scheduling methods in our joint schemes. The
simulation setup and results for the performance analysis of the
implementation of traditional resource allocation algorithms on
the proposed method are presented in Section III. Finally, the
conclusions are drawn in Section IV.
II. MO DE L DESCRIPTION
For FDD LTE air interface, the resource allocation for each
direction implies allocating portion of the bandwidth, which is
interpreted as a number of RBs. In this paper, we developed a
joint UL-DL algorithm, based on a utility function that decides
whether to give more resources (bandwidth) to the DL or the
UL based on the QoS Class Identifier (QCI) matrix and load
matrix, fairness and sum rate maximization. Then, each user
will be assigned a number of RBs in each direction based on
its streams QCIs, load, and SINR.
CA is utilized by adding a third carrier beside the UL and
DL carriers to be shared by both channels. Let BW be the total
bandwidth (in RBs) and BWU L,B WDL and BWsh are the
bandwidths for the UL, DL and shared bandwidth, respectively.
Each UE has a number of traffic flows (streams). The QoS of
each flow is defined by the QCI which is considered as a packet
filter. Let Nindicates the number of UEs, UE1to U EN. The
traffic flow ifor a UE j,j∈ {1,2, ..., N}, is characterized by
a QCI indicated as qcij
iand a load lj
i,i∈ {1,2, ..., 9}. The
load matrix and QCI matrix for the UL (LUL and QCI U L)
and DL (LDL and QCID L) are defined in (1), (2), (3) and (4).
The weights for the UL and DL from the shared bandwidth
are indicated as WUL and WDL and calculated using (5) and
(6) where (•) is the dot product operation.
LUL =
l1
1,UL · ·· ln
1,UL
.
.
.....
.
.
l1
9,UL ... ln
9,UL
(1)
QCI U L =
qci1
1,UL · ·· qcin
1,UL
.
.
.....
.
.
qci1
9,UL ... qcin
9,UL
(2)
LDL =
l1
1,DL ··· ln
1,DL
.
.
.....
.
.
l1
9,DL ... ln
9,DL
(3)
QCI DL =
qci1
1,DL ··· qcin
1,DL
.
.
.....
.
.
qci1
9,DL ... qcin
9,DL
(4)
WUL =XQC I U L •√LUL =
N
X
j=1
9
X
i=1
qcij
i,UL ×lj
i,UL (5)
WDL =XQCID L •√LDL =
N
X
j=1
9
X
i=1
qcij
i,DL ×lj
i,DL (6)
The traffic load for the jth UE in the UL lj
i,UL with the
QCI qcij
i,UL is defined as the requested traffic by the jth UE,
in Bytes, over the total similar requested traffic by all UEs
multiplied by the number of UEs requesting for this type of
traffic. Similarly, The traffic load for the jth UE in the DL
lj
i,DL with qcij
i,DL is defined as the amount received traffic
by the eNodeB to be transmitted to the jth UE, in Bytes,
over the total received traffic with same QCI multiplied by the
number of UEs having same traffic. The bandwidth portions
for the UL and DL (from BWsh ) are calculated as (7) and
(8), respectively.
BW U L
sh =WUL
WUL +WD L ×BWsh (7)
BW DL
sh =WDL
WUL +WD L ×BWsh (8)
A. Round Robin
This scheduling method performs allocation that satisfies
fairness in the amount of resources given to each user which is
not fair in terms of user throughput. It is based on the idea of
being fair in the long term by assigning equal number of RBs
to all active users without taking into consideration the channel
conditions, Channel Quality indicator (CQI) feedback. Hence,
this technique does not satisfy the user’s QoS requirements
and lack resource optimization [14]. The joint RR scheduling
method takes into considerations the BW DL
sh and BW U L
sh and
assigns the RBs for each direction accordingly. The scheduler
performs RR scheduling among the users in each direction. It
starts from the highest priority user then continues to the next
priority and examines all queues from there too.
B. Maximum SINR
The ordinary MaxSINR scheduling algorithm assigns the
RBs to the users based on their channel quality taken form the
CQI feedback to the eNodeB; user with higher CQI level will
have the priority to use the channel resources. This algorithm
targets the optimal cell throughput by assigning the RB to the
user with highest CQI. Hence, it enhances the data rate for
this user and the overall cell data rate at the expense of the
fairness between the users [14]. Similar to the joint RR, joint
Max SINR will first divide the resources between the UL and
DL based on BW DL
sh and BW U L
sh and then assigns the RBs
for each user in each direction according to its mechanisms.
C. Proportional Fair
Although there are many versions of PF algorithms, joint
PF scheduling algorithm was introduced in [15] and adopted
in this work. The RBs are assigned to the UEs with the
best relative channel quality i.e. a combination of CQI, load
and desired level of fairness in order to satisfy a balance
between maximizing the cell throughput and fairness achieving
a minimum required QoS by users. The weight for each UE (in
UL and DL) is calculated using (9) and (10), and the RBs are
allocated for the users according to (11) and (12),respectively.
wUE j
i,UL
=ρj
UL ×qcij
i,UL ×qlj
i,UL (9)
wUE j
i,DL
=ρj
DL ×qcij
i,DL ×qlj
i,DL(10)
RBsj
i,UL =f loor
wUE j
i,UL ×B W UL
sh
PN
j=1 P9
i=1 wUE j
i,UL
(11)
RBsj
i,DL =f loor
wUE j
i,DL ×BW D L
sh
PN
j=1 P9
i=1 wUE j
i,DL
(12)
The flow of the algorithms used in our for joint scheduling
is shown in Algorithm 1. Choosing the number of TTI (NTTI),
which determines how adaptive the algorithm is, depends on
the resource fluctuations and the nature of the traffic, whether
Constant Bit Rate or Bursty traffic. Simple learning algorithms
can be added to determine the optimal value of NTTI for each
case.
Algorithm 1 Joint UL-DL algorithm
Require: Number of users, N
QoS Matrices QCI U L and QCI DL
Load Matrices LUL and LDL
1: Check the No. of attached UEs, N.
2: Read the measurements from UEs
3: Read the requirement for the UEs (LUL ,LDL ,QCIU L , and
QCI DL )
4: Compare BW U L and BW DL with the users requirements
5: if (BW U L and BW DL ) are enough then
6: Perform Traditional scheduling
7: else
8: Calculate (WUL and WDL )
9: Calculate BW U L
sh and BW DL
sh
10: if Scheduler == MaxSINR then
11: Perform MaxSINR for BW U L
sh and BW DL
sh
12: else if Scheduler == RR then
13: Perform RR for BW U L
sh and BW DL
sh
14: else if Scheduler == PF then
15: Read the users’ weights
16: Perform PF for BW U L
sh and BW DL
sh based on (11) and
(12)
17: end if
18: end if
19: Check again each NTTI
20: Go back to Line 1
III. RESULTS AND DISCUSSION
A. Simulation Setup
One-cell LTE-A network was considered for the simulation.
Three UEs are connected to the eNodeB through its LTE air
interface in the FDD mode. The Matlab-based IS-Wireless
system level simulator was used for the core simulation
(building UEs, eNodeB and modelling the channel, pathloss
and so on). We integrated the the joint version of the traditional
scheduling algorithms to the simulator, and developed a MAT-
LAB graphical user interface. The three users are simulated
with single streams. Table I lists the default parameters unless
stated otherwise. The scheduling is performed using a 3 MHz
bandwidth for each of the UL, DL, and the additional carrier
used for CA. Three UEs are placed with a single eNodeB.
With a channel bandwidth of 3 MHz, there will be 15 usable
RBs as shown in Table II [16]. These RBs will be shared by
the three UEs in the downlink, uplink and shared portion of
the bandwidth. The simulation setup is shown in Fig. 1.
Fig. 1. Simulation scenario
B. Results
The performance of the joint UL-DL scheduling is pre-
sented by simulation. As aforementioned, three UEs are simu-
lated with a single eNodeB where the goal is to display the im-
pact of several parameters on the overall system performance.
The performance is compared with independent (non-joint)
scheduling with and without CA.
TABLE I. DE FAULT PARA MET ER S
Parameter Settings
Bearer Type Vary ( QCI 1- QCI 9)
Scheduling request error Disabled
Scheduling Mode Link adaptation
Scheduling Algorithm Proportional Fair
PDCP Compression Disabled
Path loss Model Free space
Multipath Model 3GPP model
Environment Type Urban
eNodeB height 20 m
UEs Height 1.5 m
Carrier Frequency 2600 MHz
Transmission Mode SISO
Cyclic Prefix Normal (7 symbols per slot)
Duplexing Mode FDD
BLER 10−1
ACK-to-NACK error Disabled
NACK-to-ACK error Disabled
Antenna types Omnidirectional
eNodeB Max. Tx power 40 Watt/46dBm
UE Max Tx power 200mWatt/23dBm
Modulation and coding scheme Adaptive modulation & coding
Channel Bandwidth (MHz) 3 DL, 3 UP, 3 Joint
TABLE II. LTE BA NDW ID TH AN D NO.OF RBS
Bandwidth (MHz) 1.4 3 5 10 15 20
Usable RBs 6 15 25 50 75 100
LTE comprises traffic classes with different QoS attributes
to define the traffic characteristics for services. Bearers are set
between UE and core network, where each bearer is associated
with different QoS attributes, such as QCI and Guaranteed
Bit Rate (GBR). Generally, bearers are classified into GBR
bearers Non-GBR bearers. The bearer will be admitted or pre-
empted according to allocation and retention priority. Once
the bearer is successfully established, the network will be
able to prioritize the service according to the QCI type.
The characteristics for the nine standardized QCI with their
respective parameters are shown in Table III [17]. The data rate
in the DL is generally higher than the UL in LTE, according to
[18]. It is affected, in both directions, by the Modulation and
coding scheme (MCS) order and number of users. However,
for our simulation MCS is adaptive based on the SINR. Hence,
we do not predict the maximum data rate for the cell. Instead,
the performance is evaluated in terms of the improvement
for a given configuration. The results were obtained with the
tabulated setting for the UEs’ traffic and QoS requirements in
Table IV.
TABLE III. LTE STANDA RD IZE D QCIS
QCI Resource
Type Priority
Packet
Delay
Budget
Packet
Loss
Rate
Example
Application
1 GBR 2 100 ms 102VOIP
2 GBR 4 150 ms 103Video Call
3 GBR 5 300 ms 106Streaming
4 GBR 3 50 ms 103Real-Time Gaming
5 Non-GBR 1 100 ms 106IMS Signaling
6 Non-GBR 7 100 ms 103Interactive Gaming
7 Non-GBR 6 300 ms 106TCP Applications
8 Non-GBR 8 300 ms Browsing, Email
9 Non-GBR 9 300 ms Download, etc.
TABLE IV. SETTING FOR UE S IN TH E SI MUL ATIO N
UL/DL UL DL
UE UE 1 UE 2 UE 3 UE 1 UE 2 UE 3
QCI 118151
L0.5 0.5 0.5 0.2 0.4 0.3
The averaged data rates obtained by each user in the DL
and UL using the RR normal and joint scheduling are presented
in Fig. 2. From the data presented, we can observe that the
joint method offers an improvement in the data rate for each
user due to the added radio resources. The joint scheduling
adds some fairness between the UL and DL (in terms of data
rate). However, the users in each direction still have the same
ratio of the resource as that of the normal scheduling which
does not satisfy the QoS and load Requirements by each user.
For example, UE2should have the highest data rate in the DL
based on its requirements which is not the case.
Fig. 2. UL and DL average UEs data rates (RR)
Similarly, the averaged data rates obtained by each user
in the DL and UL using the MaxSINR normal and joint
scheduling are presented in Fig. 3. It is clear that this method
lacks the fairness among the users both in the UL and DL
which is due to the unequal radio resources distribution among
the users. Again, the joint method adds some fairness between
the UL and DL by dividing the joint spectrum according to (7)
and (8). However, the data rates of each user using the joint
scheduling will depend on the SINR of each user in the joint
spectrum as well. From Fig. 3, we can observe that UE2has
less data rate than UE1and UE3in the DL. In the UL, UE3
has higher data rate than UE1and UE2.
Fig. 3. UL and DL average UEs data rates (MaxSINR)
The joint PF method serves the best since it considers
the load of the network and QoS requirements. The averaged
data rates obtained by each user in the DL and UL using the
PF normal and joint scheduling are presented in Fig. 4. The
scheduler gives UE2the highest data rate in the DL and gives
UE3less data rate than that for UE1and UE2in the UL. This
satisfies the requirements of each UE as shown in Table IV.
Fig. 4. UL and DL average UEs data rates (PF)
The cell data rates in the UL and DL are presented in Table
V for the three different scheduling methods, the traditional
and their joint counterparts. The high data rates for the RR
method are justified by the allocation methodology followed
by the simulator since all the resources are allocated in the
UL which is not the case for the other two. This is expected
to be different somehow when many users are competing for
the same resources in a single TTI, since less RBs will be left
unoccupied.
The cell spectral efficiency (for the UL and DL together) is
presented in Fig. 5. We compare the spectral efficiency for the
system with standard scheduling mechanism, scheduling with
CA and joint scheduling with CA.The scheduling with CA
only was performed by aggregating two 3MHz bands in both
TABLE V. CEL L DATA RATE
Scheduling UL DL UL % DL %
Standard RR 3.08 2.63
Joint RR 5.12 5.66 66% 115%
Standard MaxSINR 2.56 3.63
Joint MaxSINR 5 6.57 95% 88%
Standard PF 1.22 4.05
Joint PF 3.6 6.57 195% 62%
UL and DL and using standard scheduling in both directions.
The spectral efficiency was not presented for UL and DL
separately since it cannot be extracted easily for the joint
schemes. The joint scheduling increases the spectral efficiency
using all methods. On the other hand, CA does not have any
impact on the spectral efficiency since it increased the data
rate by simply adding additional bandwidth.
Fig. 5. Spectral efficiency
IV. CONCLUSION
In this paper, the implementation of the traditional schedul-
ing algorithms for joint UL-DL scheduling for LTE-A network
has been addressed. Three resource allocation algorithms,
namely RR, MaxSINR and PF have been discussed and
compared for the QoS-aware joint UL-DL scheduling for FDD
LTE-A utilizing the functionality of CA. The averaged data
rates obtained by each user in the DL and UL using the
normal and joint scheduling are presented for the mentioned
algorithms. Also, the system performance in terms of spectral
efficiency with standard scheduling mechanism, scheduling
with CA and joint scheduling with CA have been compared
and analyzed. Simulation results show that the performance of
joint scheduling is better than that of the normal scheduling
for all cases. Moreover, the framework presented enforces
the resources’ fair sharing between the UL and DL channels
while maintaining good QoS. Finally, findings also show that
the proposed scheme can greatly improve the throughput and
spectral efficiency by dynamically adjusting the bandwidth
portion between the UL and DL.
ACK NOW LE DG ME NT
This work was made possible by NPRP grant 4-347-2-127
from the Qatar National Research Fund (a member of The
Qatar Foundation). The statements made herein are solely the
responsibility of the authors.
REFERENCES
[1] N. Abu-Ali, A. Taha, M. Salah, and H. Hassanein, “Uplink Scheduling
in LTE and LTE-Advanced: Tutorial, Survey and Evaluation Frame-
work,”IEEE Communications Surveys & Tutorials, vol. PP, pp. 1-27,
2013.
[2] R. Ratasuk, D. Tolli, and A. Ghosh, “Carrier Aggregation in LTE-
Advanced,” in IEEE 71st Vehicular Technology Conference (VTC 2010-
Spring), pp. 1-5, 2010.
[3] W. Saad, Z. Dawy, and S. Sharafeddine, “A utility-based algorithm for
joint uplink/downlink scheduling in wireless cellular networks,” Journal
of Network and Computer Applications, vol. 35, pp. 348-356, 2012.
[4] N. Zorba, E. Yaacoub, A. M. El-Hajj, and Z. Dawy, “A modified joint
uplink-downlink opportunistic scheduling for Quality of Service guaran-
tees,” in IEEE International Conference on Communications Workshops
(ICC), pp. 1000-1004, 2013.
[5] J. Yoo, H. Luo, and C.-k. Kim, “Joint uplink/downlink opportunistic
scheduling for Wi-Fi WLANs,” Computer Communications, vol. 31, pp.
3372-3383, 2008.
[6] S. N. and and V. Sharma, “A Joint Uplink/Downlink Opportunistic
Scheduling Scheme for Infrastructure WLANs,” CoRR, Octorber, 2013.
[7] L.-x. Lin, Y.-a. Liu, F. Liu, G. Xie, K.-m. Liu, and X.-y. Ge, “Resource
scheduling in downlink LTE-advanced system with carrier aggregation,”
The Journal of China Universities of Posts and Telecommunications, vol.
19, pp. 44-123, 2012.
[8] P. Frank, A. Muller, H. Droste, and J. Speidel, “Cooperative interference-
aware joint scheduling for the 3GPP LTE uplink,” in IEEE 21st Interna-
tional Symposium on Personal Indoor and Mobile Radio Communica-
tions (PIMRC), pp. 2216-2221, 2010.
[9] H. Jianwei, V. G. Subramanian, R. Agrawal, and R. Berry, “Joint
scheduling and resource allocation in uplink OFDM systems for broad-
band wireless access networks,” IEEE Journal on Selected Areas in
Communications, vol. 27, pp. 226-234, 2009.
[10] K. Sungyeon and L. Jang-Won, “Joint Resource Allocation for Uplink
and Downlink in Wireless Networks: A Case Study with User-Level
Utility Functions,” in IEEE 69th Vehicular Technology Conference (VTC
Spring 2009), pp. 1-5, 2009.
[11] A. M. El-Hajj and Z. Dawy, “On optimized joint uplink/downlink
resource allocation in OFDMA networks,” in IEEE Symposium on
Computers and Communications (ISCC), pp. 248-253, 2011.
[12] S. Jaewoo, J. Hyun-Cheol, and A. Donggun, “”Joint Proportional Fair
Scheduling for Uplink and Downlink in Wireless Networks,” in IEEE
73rd Vehicular Technology Conference (VTC Spring 2011), pp. 1-4,
2011.
[13] A. M. El-Hajj and Z. Dawy, “Joint dynamic switching point configuration
and resource allocation in TDD-OFDMA networks: optimal formulation
and suboptimal solution,” Transactions on Emerging Telecommunications
Technologies, DOI: 10.1002/ett.2782, 2014.
[14] F. Capozzi, G. Piro, L.A. Grieco, G. Boggia, P. Camarda, “Downlink
Packet Scheduling in LTE Cellular Networks: Key Design Issues and
a Survey,” IEEE Communications Surveys and Tutorials, vol.15, no.2,
pp.678,700, 2013.
[15] A. M. Ghaleb, E. Yaacoub, A. Abdulkafi, “QoS-Aware Joint Uplink-
Downlink Scheduling in FDD LTE-Advanced with Carrier Aggregation,”
International Symposium on Wireless Communication Systems, 2014.
[16] 3GPP TS 36.213 V11.4.0, “3GPP TSG RAN Evolved Universal Terres-
trial Radio Access (E-UTRA) Physical layer procedures,” Release 11,
2013.
[17] 3GPP TS 23.203 V12.4.0, “Technical Specification Group Services and
System Aspects; Policy and charging control architecture,” March 2014.
[18] A. M. Ghaleb, D. Chieng, A. Ting, A. Abdulkafi, K.-C. Lim, and H.-S.
Lim, “Throughput performance insights of LTE Release 8: Malaysia’s
perspective,” in 9th International Wireless Communications and Mobile
Computing Conference (IWCMC), pp. 258-263, 2013.
