A downlink scheduler supporting real time services in LTE cellular networks

A downlink scheduler supporting real time services
in LTE cellular networks
Emmanouil Skondras1, Angelos Michalas2, Aggeliki Sgora1, Dimitrios D. Vergados1
1
Department of Informatics, University of Piraeus, Piraeus, Greece, Email: {skondras, asgora, vergados}@unipi.gr
2
Department of Informatics Engineering, Technological Education Institute of Western Macedonia,
Kastoria, Greece, Email: amichalas@kastoria.teiwm.gr
Abstract—The wide spread of real-time services in wireless net-
works demands scheduling mechanisms supporting strict Quality
of Service (QoS) requirements. Nevertheless, the specifications
of the LTE standard for mobile connectivity defined by the
3rd Generation Partnership Project (3GPP) does not impose
any specific scheduler for the proper allocation of resources to
services. Therefore, several LTE schedulers have been proposed
in the literature meeting the QoS requirements of modern
services. In this paper a QoS aware scheduler for the LTE
downlink is proposed namely the FLS-Advanced (FLSA) aiming
at prioritizing real-time traffic. The proposed scheduler has
been built on three distinct levels assigning the available radio
resources to services according to their requirements. Based on
simulation results, the FLSA outperforms in terms of packet
loss ratio, attainable throughput and fairness the performance of
existing schedulers including PF, MLWDF, EXP/PF, FLS, EXP
RULE and LOG RULE.
I. INTRODUCTION
Nowadays, modern services with increased QoS require-
ments have been developed. Currently, the LTE network tech-
nology uses an all-IP, purely packet switched architecture to
provide high communication speeds and satisfy the constraints
of recent services. The first release (rel. 8) provides data
rates up to 100 Mbps for downlink and up to 50Mbps for
uplink. LTE-Advanced (rel. 11) significantly enhances the rel.
8 specifications and improves the network capacity to provide
data rates up to 1Gbps for downlink and up to 500Mbps for
uplink, respectively.
The scheduling is implemented in the MAC layer and the
resource allocation is performed from the eNodeB. Informa-
tion is transmitted in frames of 10ms length and every frame is
split into 10 sub-frames of 1 Time Transmission Interval (TTI)
length each. The minimum resource which can be allocated
for transmission is called Resource Block (RB). The scheduler
assigns Resource Blocks (RBs) to users in each TTI and the
number of available RBs per TTI depends on the system
bandwidth. Flexible bandwidths of 1.4 and up to 20 MHz are
provided and totally 110 RBs are available for allocation in
the case of 20MHz bandwidth.
Scheduling information is exchanged between the eNodeB
and user equipment (UE) using signaling messages transmitted
inside the first part of each downlink sub-frame, which is
called control region. Downlink control signaling is performed
using different channel types. The UE is informed about
the size of the control region using the Physical Control
Format Indicator Channel (PCFICH). In particular, signaling
information about the downlink scheduling assignments and
uplink scheduling grants is transmitted using the Physical
Downlink Control Channel (PDCCH) and Physical Uplink
Control Channel (PUCCH) channels, respectively.
This paper focuses on the downlink packet scheduling of
LTE. A three level scheduler, the FLS-Advanced (FLSA), is
proposed, aiming at QoS aware resource allocation in order
to satisfy the requirements of strict real times services. The
remainder of the paper is organized as follows. In section 2,
existing scheduling strategies for LTE along with related re-
search literature are revisited. Section 3 describes the proposed
scheduler. Furthermore, section 4 presents the performance
evaluation of the FLSA in comparison with existing scheduling
models and section 5 concludes our work.
II. SCHEDULING STRATEGIES FOR LTE
Several downlink packet schedulers have been proposed
in the current literature. They can be classified into two
groups namely non-QoS and QoS aware, respectively. A non-
QoS aware scheduler does not take into account parameters
that affect the service quality, while a QoS aware distributes
resources considering the specific constraints of each service.
This section makes a brief overview of available scheduling
strategies for LTE systems.
A. Non-QoS aware schedulers
In [1] the non-QoS aware schedulers Maximum Throughput
(MT), Proportional Fair (PF) and Throughput to Average
(TTA) are evaluated in terms of throughput and fairness index.
The MT scheduler aims at the maximization of the overall
system throughput and allocates resources using the metric
calculated by formula 1, where di
k(t) represents the available
throughput in the kth
RB of the ith
TTI. Correspondingly,
the PF aims at fair distribution of network resources to users.
Its metric is estimated using formula 2, where ¯Ri
(t − 1)
denotes the past average throughput. Finally, the TTA metric is
calculated using formula 3 where di
(t) represents the available
throughput in the ith
TTI. Simulation results in [1], [2] showed
that the MT achieves higher throughput, the TTA has better

performance in terms of fairness index while the PF provides
both satisfactory throughput and fairness.
mMT
i,k = di
k(t) (1)
mP F
i,k =
di
k(t)
¯Ri(t − 1)
(2)
mT T A
i,k =
di
k(t)
di(t)
(3)
The Blind Equal Throughput (BET) non-QoS aware sched-
uler described in [3] distributes equally the throughput to
the users using formula 4. The authors of [4] compare the
MT, the PF and the BET schedulers using TCP traffic in
a vehicular environment. Numerical results showed that the
MT and the PF schedulers achieve similar throughputs, while
their performance is better than the one achieved by the BET
algorithm.
mBET
i,k =
1
¯Ri(t − 1)
(4)
Moreover, the authors of [5] evaluate the performance
of video streaming using the MT, the PF and the Round
Robin (RR) which allocates resources sequentially to users.
Simulation results showed that the PF algorithm achieves
higher Mean Opinion Score (MOS) and throughput for a large
number of users, while the MT performs better in terms of
Freezing Delay Ratio (FDR).
B. QoS aware schedulers
In [6] a QoS aware downlink scheduler is proposed which
considers requirements of Guaranteed Bit Rate (GBR) and non
- Guaranteed Bit Rate (non-GBR) bearers as well as channel
conditions for resource allocation. Scheduling is performed
in two phases. At the time domain (TD) phase a list of
GBR and a list of non-GBR users requiring transmission
are created and their priorities are defined. At the frequency
domain (FD) phase the allocation of users to RBs is performed.
Evaluation results showed that the suggested scheduler can
handle effectively VoIP, Video, HTTP and FTP traffic.
Furthermore, the authors of [7] propose a scheduling al-
gorithm in wireless networks which uses a utility function to
perform resource allocation for both QoS aware and BE traffic.
Numerical results showed that the proposed solution satisfies
users with different QoS requirements while it provides fair-
ness to BE users.
The Modified Largest Weighted Delay First (M-LWDF)
and the Exponential/PF (EXP/PF) QoS aware schedulers are
described in [8] and [9], respectively. Both algorithms extend
the PF metric by taking into consideration network factors
such as delays and packet losses that affect the service quality.
Specifically, the M-LWDF metric is estimated using formula
5, while the EXP/PF metric is calculated by formula 6.
The DHOL,i parameter represents the head of line delay.
Additionally, the αi value is determined by formula 7, where
δi is the target packet loss ratio and τi is the delay constraint.
Finally, the X value is calculated by formula 8, where Nrt is
the number of active real time flows. Numerical results showed
that the M-LWDF scheduler performs better as the network
load is low, while the EXP/PF algorithm gives better results
as the load increases.
mM−LW DF
i,k = ai · DHOL,i · mP F
i,k (5)
m
EXP/P F
i,k = exp(
ai · DHOL,i − X
1 +
√
X
) · mP F
i,k (6)
αi = −
logδi
τi
(7)
X =
1
Nrt
·
Nrt
i=1
ai · DHOL,i (8)
The LOG RULE and the EXP RULE schedulers presented
in [10] and [11] take into account the head of line packet
delay and the channel quality reported by UEs, to support
delay sensitive flows. Especially, the LOG RULE metric is
estimated using formula 9, where Γi
k is the spectral efficiency
for the ith
user on the kth
subchannel. Correspondingly, the
EXP RULE metric is estimated using formula 10, where bi
and c are configurable parameters. Simulation results showed
that both LOG RULE and EXP RULE could guarantee delay
constraints by configuring scheduler parameters according to
users target delays.
mLOGRULE
i,k = bi · log(c + ai · DHOL,i) · Γi
k (9)
mEXP RULE
i,k = bi · exp(
ai · DHOL,i
c + (1/Nrt) j DHOL,j
) · Γi
k
(10)
The FLS scheduler described in [12] implements a two
level QoS aware strategy. The upper level uses formula 11
to estimate the ui(k) quota of data that the ith
real time flow
must transmit at the kth
frame to succeed its QoS constraints.
In this formula, qi(k) represents the queue length in the kth
frame, Mi the number of coefficients used and ci(n) the nth
coefficient value. Coefficients are used in order to guarantee
the required delay constraints for real time flows. The number
Mi of coefficients is estimated using formula 12, where τi
represents the target delay. Additionally, the coefficient value
ci(n) is determined by formula 13. Accordingly, the lower
level uses the PF metric to allocate network resources to
real time flows for transmitting their quota of data. Whereas,
the remaining resources are allocated to best effort flows.
Simulation results showed that the proposed model improves
performance of real time flows compared to the EXP RULE
and LOG RULE schedulers.

Best-effort
queue
Real-time queue jReal-time queue i
YES
ui(t)
Link
Adaptation
CQI
Middle Level Scheduler (M-LWDF)
Schedule RBs in TTI for ui(t)
Upper Level Scheduler (FLSA)
Compute data to be transmitted ui(t)
Free RBs?
Lower Level Scheduler (M-LWDF)
Schedule RBs in TTI for:
1. qi-ui(t) data for real time flows
2. Best effort flows
Fig. 1. The three-level scheduler
ui(k) = qi(k) +
Mi
n=2
[qi · (k − n + 1)
− qi · (k − n + 2) − ui · (k − n + 1)] · ci(n)
(11)
τi = (Mi + 1) · Tf (12)
ci(n) =
⎧
⎪⎨
⎪⎩
0 where n = 0
1 where n = 1
ci(n − 1)/2 where n ≥ 2
(13)
The authors of [2] classify LTE downlink schedulers into
five categories namely the channel-unaware, the channel-
aware/QOS-unaware, the channel-aware/QoS-aware, the semi-
persistent for VoIP support and finally the energy aware
scheduling algorithms. Simulations that performed for the
channel-aware/QoS-aware category showed that the FLS
scheduler outperforms the M-LWDF, the EXP/PF, the EXP
RULE and the LOG RULE in terms of packet losses and
packet delays as the number of video flows increases.
III. THE PROPOSED SCHEDULER
This section describes the proposed FLSA scheduler which
aims at the optimization of real time flows in the LTE down-
link. It has been built on three distinct levels as presented in
figure 1. The three levels cooperate each other for dynamically
assigning radio resources to users in each TTI. Real time flows
receive higher priority than the best effort ones because of their
strict service constraints that must be fulfilled.
A. The Upper Level of the Scheduler
The upper level of FLSA uses formula 11 of FLS to estimate
the quota ui(k) of data that the ith
real time flow should
transmit in each kth
TTI, to succeed its QoS constraints. In
other words, ui(k) quota is estimated in each kth
TTI of a
frame, whereas in FLS it is estimated once at the beginning of
each kth
frame. Performance improvement has been observed
due to the fact that in FLS, when a real time flow transmits
its ui(k) quota of data, it loses the opportunity to continue
the transmission until the beginning of the next frame. By
recalculating the formula 11 in each TTI (instead of estimating
it only at the beginning of each frame), the FLSA provides
more resources to real time flows that have remaining data for
transmission.
B. The Middle Level of the Scheduler
In every TTI, the middle level uses the M-LWDF scheduler
to allocate resource blocks (RBs) to real time flows for
transmitting their ui(t) quota of data obtained from the upper
level. Parameters such as signal to interference plus noise ratio
(SINR), throughput, head of line delay, target delay, target
packet loss ratio and queue length, are considered according
to formula 5. Moreover, the use of the QoS aware M-LWDF
scheduler realizes improved resource distribution among the
real time flows in comparison with the FLS scheduler which
at the second level uses the non-QoS aware PF algorithm.
C. The Lower Level of the Scheduler
The third level has been added to allocate the remaining
RBs of each TTI to both real time and best effort flows using
the M-LWDF algorithm, in a way similar to [13]. Specifically,
RBs are allocated to real time flows, for transmitting their
qi − ui(t) data where qi denotes the queue length for the flow
i, as well as to best effort flows, considering the fact that
the later have no specific service constraints. In comparison
with the FLS scheduler, which allocates the remaining RBs
only to best effort flows using the PF scheduler, this level
of FLSA further improves the resource distribution in a QoS
aware manner.
IV. PERFORMANCE EVALUATION
The performance of the FLSA was evaluated against the
PF, M-LWDF, EXP/PF, FLS, EXP-RULE and LOG-RULE
schedulers using the open source simulator LTE-Sim [14].
Especially for the EXP-RULE metric the used parameter set is
ai ∈ [5/(0.99 · τi), 10/(0.99 · τi)], bi = 1/E[Γi
] and c = 1 as
proposed in [11] for best performance. Table I summarizes the
factors considered by each scheduler for resource allocation,
demonstrating that the FLSA is the most complete strategy.
In the simulations an LTE network topology composed by
19 cells is considered. Each cell contains 1 eNB with 43
dBm transmission power and 10 MHz downlink bandwidth
providing 50 RBs available for allocation per TTI. System sub-
frequencies have been distributed to cells as presented in figure
2, to avoid interference between neighbor cells. Full bandwidth
periodic Channel Quality Indication (CQI) reporting scheme

TABLE I
THE PARAMETERS CONSIDERED IN EACH SCHEDULER
Scheduler SINR Throughput
HOL
Delay
Max.
Delay
Max.
PLR
Queue
Length
PF x x
M-LWDF x x x x x
EXP/PF x x x x x
FLS x x x x
FLSA x x x x x x
EXP
RULE
x x x x
LOG
RULE
x x x x
Fig. 2. The simulated topology
is applied and each user equipment (UE) reports its downlink
Signal to Interference plus Noise Ratio (SINR) to eNB, in
every TTI. The eNB quantizes the reported SINR value and
calculates the CQI as described in [14], Then, it uses the
CQI to guarantee a maximum block error rate (BLER) less
than 10% regardless of the scheduling strategy applied. A
number of users, between the range [10, 40], is moving
inside the borders of each eNB according to the random way-
point mobility model. Each user receives two real time flows,
including an H264 video with bitrate equal to 440 kbps and
a voice over IP (VoIP) using the G.729 codec. Furthermore,
one best effort flow is added as background traffic. Table II
summarizes the simulation parameters.
In general, QoS aware schedulers increase the packet loss
ratio (PLR) for maintaining the required target delay. This
strategy is based on the assumption that real time services
such as VoIP and Video have no advantages of receiving
expired packets. Thus, since the delay constraint is satisfied,
the algorithms are evaluated in terms of PLR, so as to have
a comprehensive view about the performance improvements.
Figures 3 and 4 illustrate the impact of the target delay
parameter in the PLR for VoIP and video flows, respectively,
TABLE II
THE SIMULATION PARAMETERS
Parameter Value
Simulation time 100 seconds
Downlink bandwidth 10 MHz
Modulation QPSK, QAM-16 and QAM-64
Number of cells 19
Cell radius 0.5 km
Number of users up to 40 users per cell
Users mobility Random way point
Traffic models
Real time traffics:
H264 video at 440 kbps,
VoIP using G.729 codec
Best effort traffic: Web
50 75 100 125 150
0
1
2
3
4
5
6
7
8
x 10
−3
Target delay(ms)
VoIPPacketLossRatio
PF
MLWDF
LOG
FLSA
FLS
EXPRULE
EXPPF
Fig. 3. VoIP packet loss ratio using different target delays
for the case of having 20 users per cell. While the target
delay increases from 50ms to 150ms, the PLR decreases.
Additionally it may be observed that FLSA compared with
FLS exhibits lower PLR while their difference ranges up to
7%.
In figures 5 and 6, the PLR for VOIP and video flows
is presented while the number of cell users varies from 10
to 40. The considered target delays are set to 100ms and
150ms for VoIP and video flows respectively, as determined
by the LTE QoS class specifications for these service types.
As shown, FLSA results in a lower PLR than the rest of the
algorithms. Specifically, FLSA shows a marginal decrease of
its PLR compared to FLS for VoIP flows. Also, its PLR is
10% lower than that of FLS for video flows.
The analysis of the throughput provides an important insight
on the performance of the FLSA in comparison with the other
schedulers. Accordingly, as presented in figures 7 and 8 the
FLSA outperforms the rest of the schedulers, independently
of the number of users for VoIP and video flows, respectively.
This is expected due to the recalculation of formula 11 in each
TTI by the upper level of FLSA as well as the inclusion of
the lower level to the scheduler. More specifically, the FLSA

50 75 100 125 150
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Target delay(ms)
VideoPacketLossRatio
PF
MLWDF
LOG
FLSA
FLS
EXPRULE
EXPPF
Fig. 4. Video packet loss ratio using different target delays
10 20 30 40
0
0.01
0.02
0.03
0.04
0.05
0.06
Users
VoIPPacketLossRatio
PF
MLWDF
LOG
FLSA
FLS
EXPRULE
EXPPF
Fig. 5. VoIP packet loss ratio
10 20 30 40
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Users
VideoPacketLossRatio
PF
MLWDF
LOG
FLSA
FLS
EXPRULE
EXPPF
Fig. 6. Video packet loss ratio
10 20 30 40
0
0.5
1
1.5
2
2.5
3
3.5
x 10
5
Users
VoIPThroughput(bps)
PF
MLWDF
LOG
FLSA
FLS
EXPRULE
EXPPF
Fig. 7. VoIP throughput
10 20 30 40
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
x 10
6
Users
VideoThroughput(bps)
PF
MLWDF
LOG
FLSA
FLS
EXPRULE
EXPPF
Fig. 8. Video throughput
succeeds higher throughputs than the rest of the algorithms
providing rates of up to 330kbps for VoIP and up to 4.7 Mbps
for video services.
The proposed scheduler is also evaluated in terms of Jain
fairness index, which is estimated using formula 14 where n
is the number of the service flows and xi is the throughput of
the ith
flow.
Jain Fairness =
(
n
i=1 xi)2
n ·
n
i=1 x2
i
(14)
Flows with the same service constraints must receive similar
QoS to avoid the situation of having satisfied users against
dissatisfied ones of the same service type. The maximum value
of fairness is 1 while the more a scheduler accomplishes a
value close to 1, the more the resource allocation is fair.
As presented in figure 9 the fairness for VoIP flows is very
close to 1. Additionally, figure 10 demonstrates that the FLSA

50 75 100 125
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Users
VoIPFairnessIndex
PF
MLWDF
LOG
FLSA
FLS
EXPRULE
EXPPF
Fig. 9. VoIP fairness index
10 20 30 40
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Users)
VideoFairnessIndex
PF
MLWDF
LOG
FLSA
FLS
EXPRULE
EXPPF
Fig. 10. Video fairness index
scheduler improves the fairness for the video flows compared
with the rest of the schedulers.
V. CONCLUSION
In this work a QoS aware scheduler called FLSA is pro-
posed improving the resource allocation for real time services
in the LTE downlink. FLSA has been built on three distinct
levels which cooperate each other to allocate the network
resources to users. Specifically, in every TTI the upper level
estimates the quota of data that each real time flow should
transmit to succeed its QoS constraints. Then, the middle
level uses the M-LWDF algorithm to allocate RBs to real
time flows in each TTI for transmitting their quota of data
obtained by the upper level. Thereafter, if there are free RBs
available in the current TTI, the lower level allocates them
both to real time and to best effort flows, using the M-LWDF
algorithm. The performance of FLSA for real time flows is
compared against other scheduling algorithms in terms of PLR,
attainable throughput and fairness. Simulation results showed
that the FLSA scheduler outperforms the rest of the schemes
by achieving better resource allocation for real time services.
ACKNOWLEDGEMENTS
The publication of this paper has been partly supported by
the University of Piraeus Research Center (UPRC).
REFERENCES
[1] D. Zhou, N. Baldo, and M. Miozzo, “Implementation and Validation of
LTE Downlink Schedulers for ns-3,” in Simulation Tools and Techniques
(SimuTools ’13), 6th International ICST Conference on, 2013, pp. 211–
218.
[2] F. Capozzi, G. Piro, L. A. Grieco, G. Boggia, and P. Camarda, “Down-
link packet scheduling in LTE cellular networks: Key design issues and
a survey,” Communications Surveys and Tutorials, IEEE, vol. 99, 2012.
[3] A. Pokhariyal, K. I. Pedersen, G. Monghal, I. Z. Kovacs, C. Rosa, T. E.
Kolding, and P. E. Mogensen, “HARQ aware frequency domain packet
scheduler with different degrees of fairness for the UTRAN long term
evolution,” in Vehicular Technology Conference, 2007. VTC2007-Spring.
IEEE 65th. IEEE, 2007, pp. 2761–2765.
[4] D. Zhou, W. Song, N. Baldo, and M. Miozzo, “Evaluation of TCP per-
formance with LTE downlink schedulers in a vehicular environment,” in
Wireless Communications and Mobile Computing Conference (IWCMC),
2013 9th International. IEEE, 2013, pp. 1064–1069.
[5] H. A. M. Ramli, K. Sandrasegaran, R. Basukala, R. Patachaianand, and
T. S. Afrin, “Video Streaming Performance Under Well-Known Packet
Scheduling Algorithms,” International Journal of Wireless & Mobile
Networks (IJWMN)l, vol. 3, pp. 25–38, 2011.
[6] Y. Zaki, T. Weerawardane, C. Gorg, and A. Timm-Giel, “Multi-QoS-
aware fair scheduling for LTE,” in Vehicular technology conference (VTC
spring), 2011 IEEE 73rd. IEEE, 2011, pp. 1–5.
[7] L. Chen, B. Wang, X. Chen, X. Zhang, and D. Yang, “Utility-based
resource allocation for mixed traffic in wireless networks,” in Computer
Communications Workshops (INFOCOM WKSHPS), 2011 IEEE Con-
ference on. IEEE, 2011, pp. 91–96.
[8] M. Andrews, K. Kumaran, K. Ramanan, A. Stolyar, P. Whiting, and
R. Vijayakumar, “Providing quality of service over a shared wireless
link,” Communications Magazine, IEEE, vol. 39, no. 2, pp. 150–154,
2001.
[9] K. S. R. Basukala, H. Mohd Ramli, “Performance analysis of EXP/PF
and M-LWDF in downlink 3GPP LTE system,” in AH-ICI 2009. 1st
Asian Himalayas International Conference on Internet. IEEE, 2009,
pp. 1–5.
[10] B. Sadiq, R. Madan, and A. Sampath, “Downlink scheduling for mul-
ticlass traffic in LTE,” EURASIP Journal on Wireless Communications
and Networking, vol. 2009, no. 14, 2009.
[11] S. Bilal, M. Ritesh, and S. Ashwin, “Downlink scheduling for multiclass
traffic in LTE,” EURASIP Journal on Wireless Communications and
Networking, vol. 2009, 2009.
[12] G. Piro, L. A. Grieco, G. Boggia, R. Fortuna, and P. Camarda, “Two-
level downlink scheduling for real-time multimedia services in LTE
networks,” Multimedia, IEEE Transactions on, vol. 13, no. 5, pp. 1052–
1065, 2011.
[13] H. A. M. Ramli, R. Basukala, K. Sandrasegaran, and R. Patachaianand,
“Performance of well known packet scheduling algorithms in the down-
link 3GPP LTE system,” in Communications (MICC), 2009 IEEE 9th
Malaysia International Conference on. IEEE, 2009, pp. 815–820.
[14] G. Piro, L. A. Grieco, G. Boggia, F. Capozzi, and P. Camarda, “Sim-
ulating LTE cellular systems: An open-source framework,” Vehicular
Technology, IEEE Transactions on, vol. 60, no. 2, pp. 498–513, 2011.

A downlink scheduler supporting real time services in LTE cellular networks

More Related Content

What's hot (19)

Similar to A downlink scheduler supporting real time services in LTE cellular networks (20)

More from University of Piraeus (20)

Recently uploaded (20)

A downlink scheduler supporting real time services in LTE cellular networks