Content uploaded by Zeinab Shmeiss
Author content
All content in this area was uploaded by Zeinab Shmeiss on Jun 29, 2018
Content may be subject to copyright.
Downlink Scheduling in LTE: Challenges,
Improvement, and Analysis
Mohamad Omar Kayali, Zeinab Shmeiss, Haidar Safa, and Wassim El-Hajj
Department of Computer Science
American University of Beirut
Beirut, Lebanon
Email: {mmk77, zhs07}@mail.aub.edu, {hs33, we07}@aub.edu.lb
Abstract—Long Term Evolution (LTE) was developed by 3GPP
to cope with the increasing demand for better Quality of service
(QoS) and the emergence of bandwidth-consuming multimedia
applications. Today’s data transmission networks face extreme
challenges in providing high data rate and low latency. Scheduling
paradigms such as Round Robin, Best Channel Quality Condition
and Proportional Fair are commonly adopted in current LTE
downlink scheduling algorithms, but they are far from optimal
for satisfying latency requirements. In this paper, we first survey
the state of the art downlink scheduling algorithms in LTE and
identify their main challenges. We then formulate the LTE downlink
scheduling problem as an optimization problem in order to meet
the flow deadlines, then incorporate the formulation within the
surveyed scheduling algorithms, to produce better performance.
We consider strict deadlines for different types of packets with
the goal of maximizing resource distribution. Additionally, in our
formulation the buffer state for each user is taken into consideration
in order to minimize the packet loss. We evaluate the proposed
formulation using LTE-Sim and study its positive impact on the
existing LTE downlink scheduling algorithms; the performance in
terms of QoS, packet loss and fairness is improved throughout all
evaluations.
Index Terms—LTE, QoS, Downlink Scheduling, Integer Linear
Programming, Buffer State.
I. INTRODUCTION
Long term evolution (LTE) offers revolutionary performance
when compared to its predecessors, GSM/EDGE and UMT-
S/HSPA network technologies. LTE is part of the evolved
packet system (EPS) which was developed by the third gen-
eration partnership project (3GPP). EPS incorporated several
technologies such as orthogonal frequency division multiplex
(OFDM) and multiple inputs multiple outputs (MIMO). It is an
all-IP network, consisting of the evolved UTRAN (eUTRAN)
and the evolved packet core (EPC), as shown in Figure
1. The eUTRAN contains only one simplified entity, the
evolved NodeB (eNB) that is responsible for radio resource
management, making the core of the network flatter and less
complex, thus allowing for reduced latencies. The eUTRAN
is the access point of the user equipment (UE) to the network.
The EPC is comprised of several components such as the
packet data network gateway (P-GW), the serving gateway (S-
GW), the mobility management entity (MME), and the home
subscriber server (HSS). LTE specification was published and
first deployed in 2009 [1]. It supports high data transmission
rate of more than 100 Mb/s and operates on the downlink and
the uplink air interfaces. Downlink is the transmission of data
from the base station to the mobile station. As opposed to
downlink, uplink is the transmission of data from the mobile
station to the base station.
Fig. 1: Evolved Packet System Overview
OFDM has been adopted as the downlink transmission
scheme for the 3GPP LTE. It involves frequency division mul-
tiplexing (FDM) and multi-carrier communication. It divides
the available bandwidth into many sub-carriers and allows
multiple users to access the system simultaneously. Guard
bands are used between sub-carriers to avoid interference.
In multi-carrier FDM, the data of the user can be split into
multiple sub-streams and transmit them in parallel to make the
data rate higher. Orthogonality allows sub-carriers to overlap
and save bandwidth. Thus, achieving higher data rate.
A scheduler is a key element in the eNB which assigns
the shared physical resources to different users. There are
several downlink scheduling algorithms such as the Round
Robin (RR) Scheduling, Best Channel Quality Indication (Best
CQI) Scheduling, and Proportional Fair (PF) Scheduling. In
RR, terminals are assigned one after another without taking
any factor into consideration. Although this method results in
poor performance, fairness is guaranteed since all terminals
are equally scheduled. Best CQI scheduling assigns resource
blocks to the user with the best radio link conditions. On
the other hand, Proportional Fair (PF) scheduling tries to
maximize total throughput while providing all users at least a
minimal level of service. Thus, it balances between through-
put and fairness among all the UEs. Because the scheduler
aims to maximize the system performance, the design of the
scheduling algorithms has became a major issue. However, the
sharp growth of QoS applications makes the aim much more
challenging. It is well known that best-effort applications that
require non-real time traffic do not call for strict requirements
on packet delay, whereas real time services are delay sensitive
and should be transmitted as soon as possible. The 3GPP
specifications did not define scheduling algorithms that support
real-time QoS applications [2].
In this paper, we formulate the LTE downlink scheduling
as an optimization problem where the objective is to optimize
parameters such as the resource distribution, data rate, packet
delay, and even buffer overflow. Our formulation is then
incorporated in the existing state of the art LTE downlink
scheduling algorithms leading to enhanced performance on all
fronts. The remaining of this paper is organized as follows. In
Section 2, we present some basic concepts and related work. In
Section 3, we mathematically formulate the scheduling prob-
lem in LTE using integer linear programming and incorporate
the formulation in the surveyed algorithms. In Section 4, we
evaluate the performance of the proposed solution. Conclusion
is drawn in Section 5.
II. BAS IC C ON CE PT S AN D RE LATE D WO RK
In a base station (BS), the scheduler is a key element that
assigns the shared physical resources to different users in
the cell. LTE downlink physical resource is represented as a
time-frequency resource grid consisting of multiple Resource
Blocks (RB). A resource block (RB) is the smallest unit of
resources that can be allocated to a user. A resource block is
180 kHz wide in frequency and has a duration of 0.5 msec (one
slot). Figure 2 shows the resource blocks which are divided
into multiple Resource Elements (RE).
Fig. 2: Resource Blocks
Many state-of-the-art downlink scheduling algorithms were
proposed in the literature. The most relevant ones for our work
are:
1) Proportional Fair (PF): It aims to balance between
maximizing the bit rate and fairness. The PF schedul-
ing effectively reduces variations in user bit rates. It
was shown that as long as the user average signal-to-
interference-plus-noise ratio (SINR) are fairly uniform the
PF scheduler provides an ultimate fairness performance
with a moderate loss in throughput [3].
2) Maximum-Largest Weighted Delay First (M-LWDF):
It uses LWDF for bounded delay and PF for fairness. M-
LWDF scheduler outperforms other downlink scheduling
algorithms by granting higher system throughput and
guaranteeing fairness [4]. Thus, M-LWDF is considered
one of the best candidates for real-time packet scheduling.
3) EXP/PF: It uses exponential rule for bounded delay
and PF for fairness. It was demonstrated that M-LWDF
scheduler outperforms EXP/PF algorithm for lower loads
while the computationally complex EXP/PF scheduler
performs better for higher loads in downlink LTE system
supporting multimedia services [5].
4) EXP-rule: It uses exponential rule for bounded delay and
PF for channel awareness. [6] proposes the EXP rule to
provide QoS guarantees for real-time packet scheduling.
5) LOG-rule: It uses logarithm rule for bounded delay and
PF for fairness. The LOG rule scheduler was designed
to balance in QoS metrics taking the mean delay and
robustness into consideration [7]. Moreover, it allocates
resources to users in the same way as EXP rule to
maximize throughput.
6) DP-VT-MLWDF: It is a scheduler that maximize the
QoS performance of real-time traffic while sacrificing
an acceptable performance of nonreal-time traffic by
effectively using the delay priority function [8].
Every scheduler uses different strategy for assigning re-
sources. Table I shows the parameters used in the process of
resource allocation.
TABLE I: Parameters Used by Schedulers
PF M-LWDF EXP/PF [EXP/LOG]-rule
SINR x x x x
Throughput x x x x
HoL PD x x x
Target Delay x x x
Target PLR x x
Queue Length
Buffer State
The LTE downlink scheduling problem has been addressed
by many researchers in the field. [9] addresses the state-
of-the-art algorithms that tackle the problem of scheduling
QoS applications. However, these algorithms fail to meet
all QoS requirements mentioned in the literature [10]. Some
work such as [8] proposes a variation to some state-of-the-art
algorithms in order to increase QoS performance by taking
into consideration the delay sensitivity. Alternatively, some
proposes solutions that tackle the problem of scheduling by
monitoring the buffer state of end users [11]. However, they
didn’t take into account the QoS of real time and non-real
time applications.
The downlink scheduling problem was formulated as an
optimization problem in approaches [12], [13], [14], [15].
However, [13], [15] didn’t take into consideration the buffer
state or major QoS parameters. Moreover, [12], [14] took into
consideration the QoS but only for specific class of real time
applications such as VoIP or Video Stream.
III. PROP OS ED WORK
Our proposed approach takes into consideration all QoS
parameters including the buffer state of end users and formu-
lates the scheduling problem as a linear optimization problem.
Our main goal is to optimize the usage of available network
resources in order to provide better QoS in LTE networks.
This is done by selecting the optimal parameters and the
appropriate scheduling algorithm to obtain the best QoS.
Indeed, our solution aims to maximize QoS performance while
maintaining fairness among all packets. The value of the
parameters used in the scheduling process can be determined
based on the following factors [12]:
1) Quality of Service (QoS): data that has the lowest Quality
Class Identifier (QCI) value will have the highest metric.
2) Channel Quality Indicator (CQI): based on the CQI value,
the highest throughput will have the highest metric.
3) The transmission queues state: the longest queue will
have the highest metric.
4) Resource Allocation History: based on the performance
history, the lowest previous throughput will have the
highest metric.
5) Buffer State: based on the buffer state at the UE, the
buffer with the highest available space will have the
highest metric.
It is worth mentioning that overwhelming the user with
packets will result in buffer overflow which leads to packet
loss. In our scheduling process the buffer status is considered
while assigning resource blocks to different users which is
done to avoid packet loss. UEs provide the eNodeB scheduler
with its buffer state information such as the terminal buffer
length Lbuff and the amount of current received packets
Ncurr . The variable Ncurr is initialized to 0 when the data
traffic process is established. The increment step of variable
Ncurr is 1. The Ncurr is flushed to 0 when it reaches buffer
length Lbuff [11]. The buffer of user iis buffiwhich is
described as follows:
Buf fi=Lbuf f −Ncurr
Lbuff
(1)
In order to decrease the packet dropping probability due
to buffer overflow, our approach schedules more resources to
users with higher spare buffer space.
We next discuss the proposed problem formalization. Sup-
pose we have M available RBs and U users. We define the
variable matrix x[M, U] such that:
xi,j (t) = (1if ith RB is allocated to the jth user at time t
0otherwise
(2)
where the sum of all xi,j is less than or equal to 1 for any
particular RB implying that only one user can employ the ith
resource block. We also define the matrix F[M, U ](t)where
Fi,j represents the behavior of the jth user on the ith RB at
time taccording to each of the selected algorithms (Figure 3).
Fig. 3: Fi,j Matrix
We will formulate many state-of-the-art scheduling algo-
rithms using our optimization solution by adding to them all
the QoS parameters described in Table II. In the proportional
fairness scheduling algorithm [3], Fi,j (t)is given as:
Fi,j (t) = ri,j (t) = log (1 + SIN Ri,j(t))
Ri(t−1) (3)
where SINR is the signal to noise ratio reported by the UE
and Ri(t−1) is the past average throughput achieved by user
jat time t - 1.
In the M-LWDF scheduling algorithm [12], Fi,j (t)is given
as:
Fi,j (t) = αj∗DHOLj∗ri,j (t)for real time applications
ri,j (t)otherwise
(4)
where
αj=log δj
τj
with δjis the acceptable packet loss rate, τjis the delay
threshold, and DHOLjis the head of line delay for the jth
user.
In EXP/PF scheduling algorithm [12], Fi,j (t)is given as:
Fi,j (t) = exp(αj∗DHO Lj−χ
1+
√χ)∗ri,j (t)for real time applications
ri,j (t)otherwise
(5)
where
χ=1
Nrt
Nrt
X
k=1
αk∗DHOLk
and Nrt is the number of active downlink real-time flows.
In delay priority scheduler (DP-VT-MLWDF)[8], Fi,j(t)is
given as:
Fi,j (t) = (αj∗Qj(t)
τj−DHO Lj
∗ri,j (t)for real time applications
ri,j (t)otherwise
(6)
with Qj(t)being the length of token queue for the jth user.
Furthermore, the end-user buffer (Buf fj) status which is
reported by the UE can be easily expressed through an opti-
mization constraint where we force the scheduler to assign RB
to the user such that the eNodeB does not transfer data more
than the UE can handle. When we do so we are preventing the
UE buffer from overflow. Thus, minimizing the ratio of packet
loss making the system perform better on the long term.
The formalization of the optimization approach is expressed
as follows:
maximize
X
M
X
i=1
U
X
j=1
xi,j Fi,j (t)
subject to C1 :
U
X
j=1
xi,j ≤1∀i= 1, ..., M
C2 : xi,j ∈0,1
C3 :
M
X
i=1
xi,j ri,j (t)≤rj∀j= 1, ..., U
C4 :
M
X
i=1
xi,j ri,j (t)≤Buf fj∀j= 1, ..., U
(7)
Constraint C1 ensures that every RB is allocated to at most
one user. C2 forces the value of the variable xi,j to be 0 or
1. C3 implies that the total data rate assigned to the users
should not exceed their demands, and also should not lead to
an overflow on the user side buffer which is expressed in con-
straint C4. The optimization problem presented is a 0-1 integer
linear programming in which the variables are restricted to be
integers. This type of problems is considered NP-hard where
the unknowns are binary. Thus, relaxation must be done on
constraint C2 by replacing it with the following constraints:
0≤xi,j and xi,j ≤1. After this relaxation, the problem is
transformed to linear programming optimization which can be
easily solved by the two-phase simplex algorithm. However,
in order to preserve the binary property of the unknowns the
outcome of the simplex algorithm is rounded to 0 or 1.
IV. EVALUATION AND PRELIMINARY RES ULTS
In this section, we present the performance of some of
the state-of-the-art algorithms (PF, MLWDF, EXP/PF, DP-
VT-MLWDF) with and without the optimization while taking
the buffer state into consideration. LTE-Sim simulator was
used for simulations. The experiments were performed on a
uniformly distributed real-time and nonreal-time applications
on the available users. To ensure accuracy, each run was con-
ducted 5 times, with different number of users and the average
was taken. Table III summarizes the system parameters used
in the simulations.
TABLE III: Simulation Parameters
Parameters Values
Bandwidth 10MHz
Number of RBs 50
Frame Structure FDD
Cell Radius 1 kmn
No. of Users 5:5:20
UE speed 3 km/h
In our evaluation, we measured the average packet loss
ratio and the average total throughput for the video stream
applications and nonreal-time applications. Figure 4 shows
the packet loss ratio for video stream applications. It can
be noticed that as the number of users increase the packet
loss ratio for PF, EXP/PF, and DP-VT-MLWDF increases in
both (A) and (B). However, the PLR in MLWDF scheduler
is not affected by the number of users. Moreover, Figure 5
presents a comparison of the PLR between both optimization
and non-optimization approach for 20 users. We noticed that
the PLR for MLWDF, EXP/PF, and DP-VT-MLWDF is less
when using the optimization approach. On the other hand, the
PLR for PF scheduler is the same for both approaches. This
is because the PF scheduler doesn’t take the packet priority
into consideration while assigning resources. Consequently,
adding the formulation we proposed to the existing scheduling
algorithms, either improved on the ”average packet loss ratio”
metric or did not worsen it.
(a) Without Optimization
(b) With Optimization
Fig. 4: PLR of Video Stream Applications.
TABLE II: Standardized QoS Class Identifiers for LTE
QCI Resource Type Priority Packet Delay Budget [ms] Packet Loss Rate Example services
1 GBR 2 100 10−2Conversational voice
2 GBR 4 150 10−3Conversational video (live streaming)
3 GBR 5 300 10−6Non-Conversational video (buffered streaming)
4 GBR 3 50 10−3Real time gaming
5 non-GBR 1 100 10−6IMS signaling
6 non-GBR 7 100 10−3Voice, video (live streaming), interactive gaming
7 non-GBR 6 300 10−6Video (buffered streaming)
8 non-GBR 8 300 10−6TCP based (e.g., WWW, e-mail)
9 non-GBR 9 300 10−6TCP based (e.g., WWW, e-mail)
Fig. 5: PLR of Video Stream Applications for 20 Users
Figure 6 presents the throughput for video stream appli-
cations with and without optimization. As presented in the
figure 6-(a), in all cases, the throughput tends to increase as
the number of users increases. However, and as shown in figure
6-(b) and figure 7, when the suggested optimization is incor-
porated, the optimized approach outperforms the traditional
approaches, in terms of throughput, for all selected algorithms.
Hence, the optimization techniques force the scheduler to
select the best choice with the best throughput.
(a) Without Optimization
(b) With Optimization
Fig. 6: Throughput of Video Stream Applications.
Fig. 7: Throughput of Video Stream Applications for 20 Users
Figure 8 shows the packet loss ratio of non-real time appli-
cations. The PLR tends to remain within the same range when
the number of users exceeds 10 users. Moreover, the results
tend to be slightly less when using the non-optimized ap-
proach. This is because the optimized approach gives priority
for QoS packets while sacrificing an acceptable performance
for non-real time traffic.
(a) Without Optimization
(b) With Optimization
Fig. 8: PLR of Non-Real Time Applications.
Figure 9 shows the throughput when using non-real time
applications. The results show that when the optimized for-
mulation is incorporated in the scheduling algorithms (PF,
MLWDF, and EXP/PF), the algorithms perform better than
the regular non-optimized approach.
(a) Without Optimization
(b) With Optimization
Fig. 9: Throughput of Non-Real Time Applications
V. CONCLUSION
In this paper, we have addressed the problem of down-
link scheduling for QoS packet flow in LTE networks. The
state-of-the-art scheduling algorithms have been formulated
using integer linear programming, and solved. The formulation
considers in addition to the regular QoS parameters, the UE
buffer state parameter as an enhancement. The effects of
the proposed approach have been studied and evaluated to
demonstrate that it is suitable to provide better services for
QoS and best-effort applications. Obtained simulation results
confirm the effectiveness of the proposed approach. They show
the performance of the original state-of-the-art scheduling al-
gorithms and how incorporating our optimization formulation
have impacted positively their performance while measuring
parameters such as throughput and the packet loss rate.
REFERENCES
[1] C. Cox. An introduction to LTE: LTE, LTE-advanced, SAE and 4G
mobile communications, 2012.
[2] Bin Liu, Hui Tian, and Lingling Xu. An efficient downlink packet
scheduling algorithm for real time traffics in lte systems. In 2013 IEEE
10th Consumer Communications and Networking Conference (CCNC),
pages 364–369, Jan 2013.
[3] R. Kwan, C. Leung, and J. Zhang. Proportional fair multiuser scheduling
in lte. IEEE Signal Processing Letters, 16(6):461–464, June 2009.
[4] H. A. M. Ramli, R. Basukala, K. Sandrasegaran, and R. Patacha-
ianand. Performance of well known packet scheduling algorithms in the
downlink 3gpp lte system. In 2009 IEEE 9th Malaysia International
Conference on Communications (MICC), pages 815–820, Dec 2009.
[5] R. Basukala, H. A. M. Ramli, and K. Sandrasegaran. Performance
analysis of exp/pf and m-lwdf in downlink 3gpp lte system. In 2009
First Asian Himalayas International Conference on Internet, pages 1–5,
Nov 2009.
[6] Sanjay Shakkottai and Alexander L Stolyar. Scheduling for multiple
flows sharing a time-varying channel: The exponential rule. Translations
of the American Mathematical Society-Series 2, 207:185–202, 2002.
[7] Bilal Sadiq, Seung Jun Baek, and Gustavo De Veciana. Delay-optimal
opportunistic scheduling and approximations: The log rule. IEEE/ACM
Transactions on Networking (TON), 19(2):405–418, 2011.
[8] Yuan-Ping Li, Bin-Jie Hu, Hui Zhu, Zong-Heng Wei, and Wei Gao.
A delay priority scheduling algorithm for downlink real-time traffic in
lte networks. In Information Technology, Networking, Electronic and
Automation Control Conference, IEEE, pages 706–709. IEEE, 2016.
[9] F. Capozzi, G. Piro, L. A. Grieco, G. Boggia, and P. Camarda. Downlink
packet scheduling in lte cellular networks: Key design issues and
a survey. IEEE Communications Surveys Tutorials, 15(2):678–700,
Second 2013.
[10] Giuseppe Piro, Luigi Alfredo Grieco, Gennaro Boggia, Rossella Fortuna,
and Pietro Camarda. Two-level downlink scheduling for real-time
multimedia services in lte networks. Trans. Multi., 13(5):1052–1065,
October 2011.
[11] Yan Lin and Guangxin Yue. Channel-adapted and buffer-aware packet
scheduling in lte wireless communication system. Proc. Int. Conf.
Wireless Communications, Networking and Mobile Computing, WiCOM,
Dalian, China, 2008.
[12] Tarik Ghalut, Hadi Larijani, and Ali Shahrabi. Qoe-aware optimization
of video stream downlink scheduling over {LTE}networks using
{RNNs}and genetic algorithm. Procedia Computer Science, 94:232 –
239, 2016. The 11th International Conference on Future Networks and
Communications (FNC 2016) / The 13th International Conference on
Mobile Systems and Pervasive Computing (MobiSPC 2016) / Affiliated
Workshops.
[13] Raymond Kwan, Cyril Leung, and Zhang Jie. Proportional fair multiuser
scheduling in lte. IEEE Signal Processing Letters, 16(6):461–464, 2009.
[14] Duy-Huy Nguyen, Hang Nguyen, and Eric Renault. We-mqs-voip
priority: An enhanced lte downlink scheduler for voice services with
the integration of voip priority mode. International Journal of Advanced
Computer Science and Applications IJACSA, 7(7):560–567, 2016.
[15] Yunzhi Qian, Canjun Ren, Suwen Tang, and Ming Chen. Multi-service
qos guaranteed based downlink cross-layer resource block allocation
algorithm in lte systems. In Wireless Communications & Signal
Processing, 2009. WCSP 2009. International Conference on, pages 1–4.
IEEE, 2009.
