Content uploaded by Amr Abdelbari
Author content
All content in this area was uploaded by Amr Abdelbari on Dec 27, 2020
Content may be subject to copyright.
Received: 14 February 2019 Revised: 28 August 2019 Accepted: 1 October 2019
DOI: 10.1002/dac.4229
RESEARCH ARTICLE
Throughput enhanced scheduling (TES) scheme for
ultra-dense networks
Huseyin Haci Amr Abdelbari
Department of Electrical and Electronics,
Near East University, Mersin 10, Turkey
Correspondence
Amr Abdelbari, Department of Electrical
and Electronics, Near East University,
Nicosia, TRNC, Mersin 10, Turkey.
Email: amr.abdelbari@neu.edu.tr
Summary
This paper proposes two novel packet scheduling schemes, called as through-
put enhanced scheduling (TES) and TES plus (TES+), for future ultra-dense
networks. These schemes introduce two novel parameters to the scheduling
decision making and reformulate the parameters used by the state-of-the-art
schemes. The aim is to have a more balanced weight distribution between delay
and throughput-related parameters at scheduling decisions. Also include a new
telecommunications related parameter into scheduling decision making that
has not been studied by popular schedulers. The performance of novel schemes
is compared with well-known schemes—proportional fairness (PF), exponen-
tial/proportional fairness (EXP/PF), and M-LWDF. For performance evaluation,
five performance metrics—average spectral efficiency and delay, quality of ser-
vice (QoS) violation ratio, jitter, and Jain's fairness index—are investigated. The
simulation results show that proposed schemes can outperform all the compared
scheduling schemes.
KEYWORDS
5G networks, EXP/PF, M-LWDF, packet scheduling, real-time traffic, ultra-dense networks
1INTRODUCTION
Today, long-term evolution (LTE) systems serve a wide range of applications including real time (RT) (eg, video streaming,
VoIP, and online gaming) and nonreal-time (nRT) applications (eg, web browsing and uploading files). RT applica-
tions require a delay sensitive and high throughput channel to satisfy QoS requirements.1nRT applications are served
with respect to a best-effort manner.2The fifth generation (5G) wireless communication networks will serve a number
of novel technologies that require new ways of communications like Internet of Things (IoT), cloud servers, and
Multiple-Input-Multiple-Output (MIMO).3,4 Accordingly, it is envisioned that there will be more stringent QoS require-
ments and demand for higher throughput to satisfy the needs for these ultra-dense mobile networks.5Ultra-dense
networks stand for the networks, where the number of users can be much more than the available radio resources, such
as number of subcarriers or resource blocks in LTE. A case of ultra-dense networks considered in this paper is that the
number of users in a cell can be up to four times more than the available resource blocks. Further in the 5G networks,
it is expected that demand for RT services will be four times over compared with nRT services.6Thus, the next genera-
tion of wireless communication networks is expected to give higher priority to the RT QoS requirements (eg, delay and
throughput), and many researches focus on developing new techniques that satisfy the higher rate of demand on a very
strict QoS requirements.7-9
In the wireless communications, any packet has to go through a number of stages before it is transmitted. A very impor-
tant stage that significantly affects the system performance is the resource allocation (RA) stage. In this stage, adaptive
Int J Commun Syst. 2019;e4229. wileyonlinelibrary.com/journal/dac © 2019 John Wiley & Sons, Ltd. 1of14
https://doi.org/10.1002/dac.4229
2of14 HACI AND ABDELBARI
modulation and coding is applied with respect to the channel state information (CSI) of users that will be served by the
system. The process of choosing a subgroup of users to be served is called the user scheduling stage. As the chosen group
of users directly affects the RA stage, it also has a significant effect on the overall system and user performance.10-12
The proportional fairness (PF) scheduling scheme used in 3G networks has achieved a high throughput on the long
term and kept a high level of fairness among users. However, in 4G networks, a large demand for RT traffic exists, and
PF cannot guarantee the delay QoS requirements for these RT traffic.13 To address this drawback of PF, modified largest
weighted delay first (M-LWDF) and exponential/proportional fairness (EXP/PF) schemes are proposed.14 These popular
techniques can help to address QoS requirements by sacrificing some of the system throughput. In 5G and beyond, there
is a need for much greater system throughput compared with 4G (1000 x increase)4; thus, new scheduling techniques that
can significantly improve the system throughput and address the QoS and fairness requirements are needed.15
Scheduling users depends on multiple factors such as the instantaneous and average channel state, the delay of
head-of-line (HOL) packets, the delay threshold requirements, and the type of service demanded.16,17 The three latter
factors set the QoS requirements for the scheduler to meet, and the two former factors are used by the scheduler to maxi-
mize the system throughput while satisfying the QoS requirements.9The scheduling scheme needs to consider the delay
threshold requirements with up-most importance. In RT streams, packets that arrive after their delay threshold being
expired will not be used and be discarded.2In nRT streams, the packets can be queued for a longer time period compared
with RT since they can tolerate longer delay. Another important aspect for the scheduler is to make sure that the users
are being served equally in order to preserve the fairness. Other aspects that affect the user scheduling include the sta-
tus of queued packets (control and/or management packets) and the probability of arrived packets to have errors because
of the variance in the channel, ie, interference from other cells, fading, and shadowing.11 All of the aforementioned
aspects of wireless communications affect the scheduling scheme decisions as well as the system performance. There are
many researches that focus on studying a subgroup of these aspects to improve the system performance.18,19 However,
the scheduling approaches taken by these researches do not include a crucial telecommunications related parameter
called as jitter. Jitter is the variance in the delay of head-of-line packets and in high amounts can cause interruptions at
delay-sensitive multimedia playback and degrade the quality of experience for users. Another drawback of state-of-the-art
schemes is being too conservative in the scheduling decisions by having much of the weight on delay-related parameters.
Thus, these schemes may not achieve the performance requirements for 5G and beyond network. This paper proposes
novel scheduling schemes, TES and TES+, that incorporate jitter and signal to interference plus noise ratio (SINR) at the
scheduling decision making and behave much more opportunistic by reformulating the scheduling parameters used by
popular schemes.
The paper is organized as follows. In Section 2, related works are discussed. The system model is provided in Section 3,
where the network architecture and scheduler model are introduced. Section 4 presents the proposed scheduling algo-
rithms and related analysis. In Section 5, a comprehensive performance investigation between the proposed schemes, PF,
M-LWDF, and EXP/PF is provided.
2RELATED WORKS
There are a number of popular researches that have developed user scheduling schemes for high speed wireless net-
works. The common goal is to maximize the system throughput while achieving user QoS requirements and fairness
among users. In the previous studies,18,20-27 the performance of the well-known schemes are compared, and the advan-
tage of each of them has been discussed. Ghariani and Jouaber22 studied the energy efficiency for popular LTE scheduling
schemes and showed that M-LWDF is more energy efficient than EXP/PF and PF, especially at ultra-dense wireless net-
works. Chayon et al23 extend the EXP/PF scheme to improve the performance of cell-edge users in high-speed wireless
networks. The proposed scheme considers a threshold value based on user congestion in the regions (inner/cell-edge),
besides the typical parameters of EXP/PF to prioritize users. The simulation results show approximately 30% increase
in cell-edge users throughput. Liu et al24 propose a particle swarm optimization algorithm based on EXP/PF to combat
with co-channel interference for cell-edge users. The scheme is applied in downlink coordinated multipoint MIMO trans-
missions. Significant improvement in total system capacity and outage probability has been shown by the simulations.
Ramli and Isa25 extend the M-LWDF scheme to exploit the multicarrier transmissions (carrier aggregation) in Long Term
Evolution Advanced (LTE-A) networks. The proposed scheme manages the component carriers, modulation, and coding
used on each carrier to improve QoS satisfaction and retransmission performance. The simulation results show 16.7%
improvement in QoS satisfaction for real-time multimedia users. Liqiang et al26 propose a novel scheduling scheme for
HACI AND ABDELBARI 3of14
5G communications based on PF, M-LWDF, and sparse code multiple access (SCMA). The paper shows that SCMA-based
systems outperform orthogonal frequency division multiplexing (OFDMA)-based systems. Also, when M-LWDF scheme
is employed with SCMA, the system can provide low packet loss ratio, high fairness index, and time delay sensitivity to PF
scheme employed with SCMA. Yang and Zu27 modified the traditional M-LWDF scheme to adapt to cognitive radio net-
works. The authors introduced a classification (primary/secondary user)-based parameter to the traditional metric and
significantly improved the packet dropping rate and system block rate.
Nguyen et al28,29 propose a new scheduling scheme called as E-MQS scheduler. The scheduling decisions are based on
user perception parameters as well as channel and QoS aware parameters. The user perception parameters are obtained
from a computational model called as E-model.29 Since 5G communication networks are expected to be very dense net-
works, employing scheduling schemes that introduces high computation burden per user is not practical. Frame Level
Scheduler (FLS) is another interesting scheduling scheme30 that has a two-level scheduling architecture. At the upper
layer, the scheduler employs a closed control loop method, and the lower layer employs a PF rule. The simulation results
show that, at dense networks, the FLS scheduler leads to significantly longer delay time for users compared with EXP/PF
and M-LWDF scheduler.29 Thus, it is not practical to employ FLS scheme at future ultra-dense networks. Nguyen and
Rao,31 Chen,32 and Wang and Cai33 developed optimal solutions for wireless multiuser scheduling problem. The solutions
are based on linear programming and machine learning tools. The authors applied various techniques, such as statistical
based classification to reduce the computational complexity of the proposed solutions. However, these researches are still
computationally not easy to solve to be practical in ultra-dense networks.
Despite the interesting research mentioned above, a scheduling scheme that can be applied in practice at future
ultra-dense cellular networks should be computationally easy to perform and require minimal information about users.
PF, M-LWDF, EXP/PF, and their derivatives schemes with these properties still have a room for improvement. This paper
aims to design a novel scheme that can be applied in practice and achieve better performance than these schemes by
improving their design. At the rest of this section, analysis of PF, M-LWDF, and EXP/PF is driven for the convenience of
the reader to better understand the novelty of the proposed scheme.
2.1 Proportional fairness
The well-known PF scheme allocates a RB depending on the knowledge of users' instantaneous channel state and the
average data rate over a sliding window. This is to maximize the throughput achieved by the users in long term and achieve
the fairness between users.10 The PF scheduling metric for the ith user is calculated by
wi=ri
Ri(t),(1)
where riis the instantaneous data rate of the ith user, and Ri(t)is the average data rate calculated over sliding window by
Ri(t+1)=(1−1
tc
)∗Ri(t)+(1
tc
)∗ri,(2)
where tcis the window size.18
The main drawback of PF is that it does not consider delay and related QoS requirements of RT traffic. This may cause
unacceptable QoS violation for RT traffic in dense networks. Thus, PF cannot be employed in 5G and beyond networks.
Although impractical, this paper chooses to include the PF scheme in the performance comparison as the benchmark
scheme, especially for the spectrum efficiency performance. This is because PF scheme has more degrees of freedom
compared with M-LWDF and EXP/PF schemes to achieve higher spectrum efficiency.
2.2 Modified largest weighted delay first
The M-LWDF scheme schedules users based on the parameters of PF scheme plus the delay of HOL packets and the toler-
ance for the HOL packet's delay to exceed the delay threshold.17 The latter parameter is also denoted as the probability of
delay to exceed a given threshold.29 M-LWDF addresses the problem of PF and can provide high system throughput while
achieving fairness among users.34 However, in this paper, it is shown that this scheme can have problems in future dense
networks, and its performance can be further improved. The M-LWDF scheduling metric for the ith user is calculated by
wi=𝛼i∗DOLi∗ri
Ri(t),(3)
4of14 HACI AND ABDELBARI
where DOLirepresents the time difference between current time and the time HOL packet of the ith user is created,35
and 𝛼iis calculated by
𝛼i=log(𝛿i)
Ti
,(4)
where 𝛿iis the probability of DOLito exceed the delay threshold Ti. The latter parameter lets the users with a low delay
tolerance to have a higher scheduling priority than users with higher delay tolerance.
2.3 Exponential/proportional fairness
The EXP/PF is a scheduling scheme that is evolved from the PF, which provides RT streams with higher priority over nRT
streams.36 EXP/PF uses the ordinary PF metric for nRT traffic, and for RT traffic, the metric is very similar to M-LWDF;
except an exponential function is applied to delay-related parameters. This provides much more weight to delay-related
parameters at scheduling decision making.37 The EXP/PF metric for the ith user is calculated by
wi=exp(𝛼i∗DOLi−X
1+√X
)∗ ri
Ri(t),(5)
where X is given by
X=1
NRT
∗
NRT
∑
i=1
𝛼i∗DOLi,(6)
where NRT is the number of RT streams in the present scheduling time.
3SYSTEM MODEL
This paper considers a wireless cellular network, which includes one base station (BS) and Nmobile users (MUs) within
a cell as shown in Figure 1.38 It is assumed that orthogonal frequency division multiplexing (OFDM) is used in the system
to divide a frequency selective wide-band channel into a number of flat-fading, narrow-band subchannels.
The BS serves all communication processes both from the BS to MUs (Downlink) and from the MUs to the BS (Uplink).
Additionally, it successfully monitors MUs' instantaneous channel state on each resource block (RB). An RB in an
OFDM-based system is a frequency-time domain allocation of radio resources. In 4G systems, RB consists of 12 subcar-
riers spaced by 15 kHz (total bandwidth of 180 kHz) and the time duration of 7 OFDM symbols (total time duration of
1ms).
21 The BS receives MU data requests via the uplink channels and determines type, size, and other relevant informa-
tion of the requested data, then queues the requests in a form of packets to be transmitted to the MUs. Packet scheduler
unit in the BS manages which the RBs will be allocated to the ith user ,1≤i≤N,asshowninFigure2.Inordertosim-
ulate the future very dense networks model, it is considered that there may be up to four times more MUs in a cell then
available RBs. For example, considering a cellular network with 50 RBs, the number of MUs in the cell can be up to 200.
A wireless channel is assumed between the BS and each MU that is independent and identically distributed with
Rayleigh fading. Rayleigh fading distribution is a common method used to theoretically describe the wireless multipath
fading channel in an urban environment with a random movement of Nusers.39
FIGURE 1 An example of a wireless network architecture
HACI AND ABDELBARI 5of14
FIGURE 2 A packet scheduler model
4THE PROPOSED SCHEDULING ALGORITHMS
A scheduling scheme that is designed to handle a mixture of RT and nRT traffic should have two fundamental parts—the
delay-related parameters part, denoted by 𝜌ifor the ith user and the throughput related parameters part, denoted as 𝛾ifor
the ith user, in this paper. Considering the state-of-the-art schemes EXP/PF and M-LWDF,
𝜌i={𝛼i∗DOLi,𝑓or M-LWDF,
exp (𝛼i∗DOLi−X
1+√X),𝑓or EXP/PF,(7)
and 𝛾i=ri
Ri(t)for M-LWDF and EXP/PF. Then a common expression to calculate the scheduling metric for the ith user in
the system can be rewritten from Equations (3) and (5) as
wi=𝜌i∗𝛾i.(8)
Further, on a given subcarrier, the i*th user that wins the scheduling contention is given by
i∗=arg max
i∈Ω {wi},(9)
where arg max denotes the argument of maximum, and Ωis the set of users with data to be transmitted. For i*th user, the
weight of 𝜌i∗on the scheduling decision is given by
Z𝜌i∗=𝜌i∗
𝜌i∗+𝛾i∗
,(10)
and the weight of 𝛾i∗on the scheduling decision is given by
Z𝛾i∗=𝛾i∗
𝛾i∗+𝜌i∗
.(11)
A performance limiting factor of EXP/PF and M-LWDF is not to provide enough significance on Z𝛾i∗at scheduling
decisions. Figures 3 and 4 show the ratio of Z𝜌i∗and Z𝛾i∗on scheduling decisions at 1000 scheduling events for EXP/PF
and M-LWDF, respectively. It can be seen from the figures that for both scheduling schemes, Z𝜌i∗>Z𝛾i∗most of the time.
This makes the scheduling decisions too conservative with respect to delay and thus significantly limits the achievable
system throughput. It is crucial to have a good balance between Z𝜌i∗and Z𝛾i∗on scheduling decisions to let the scheduler
be more opportunistic in scheduling users with better channel conditions. This may improve the overall system and user
6of14 HACI AND ABDELBARI
FIGURE 3 Weight on scheduling decision for
exponential/proportional fairness (EXP/PF) versus scheduling event
FIGURE 4 Weight on scheduling decision for modified largest
weighted delay first (M-LWDF) versus scheduling event
throughput as well as the users' delay performance. This paper proposed TES and TES+ schemes to achieve a better
balance between Z𝜌i∗and Z𝛾i∗by reformulating 𝜌iand introducing two new terms to 𝜌i∗and 𝛾i∗of the state-of-the-art
schemes. The aim is to increase the significance of Z𝛾i∗on scheduling decisions to a degree that high throughput users
can be frequently scheduled while the delay (QoS) threshold violations of all users still be minimized, also to study the
effect of a crucial telecommunications-related parameter in scheduling decisions.
Because of various QoS requirements of RT and nRT traffics, a two-part indexing system is used at TES and TES+
schemes to calculate index values for RT and nRT traffics via different expressions. Accordingly, the expression used for
a corresponding traffic type fits better to its requirements. The expression proposed for TES scheme to calculate the ith
user's scheduling metric is given by wi=𝜌i∗𝛾i,where
𝜌i={exp (−Ti−DOLi
Ti),𝑓or RT traffic,
𝛼i∗DOLi,𝑓or nRT traffic,
(12)
and
𝛾i=log(b∗SINRi)∗ ri
Ri(t),𝑓or both RT and nRT traffic,(13)
where SINRiis the signal to interference plus noise ratio of the ith user. In Equation (13), log{.}represents the logarithm
function, and bis a constant. The first term in Equation (13) is introduced to 𝛾ito increase Z𝛾i∗directly proportional to
HACI AND ABDELBARI 7of14
the SINR of the ith user. A user with higher SINR has a higher chance to get scheduled than a user with a worse channel
condition. The constant band log{.}function is applied to SINR in order to normalize the effect of SINR on Z𝛾i∗(ie, for
SINR not to have too much effect on scheduling decision). The second term in Equation (13) is included in order to achieve
fairness among users and maximize the average user throughput in the long term.11 In formulation for Equation (12),
an exponential function is applied to the term for RT traffic since the delay parameters play a much significant role in
satisfying the QoS requirements. And a linear function is employed at the term for nRT traffic since the delay is relatively
less important.
Further, by the TES+ scheme, this paper proposes to include jitter as a novel parameter to scheduling decision making.
Jitter is the variation in the head-of-line packet delay. It is denoted by 𝜎2
DOLiin this paper and given by
𝜎2
DOLi=E[(DOLi)2]−(E[DOLi])2,(14)
where E[(DOLi)] is the expected value or mean of DOLigiven by
E[DOLi]= 1
S
S
∑
s=1
DOLs,i,(15)
where S represents the number of packets successfully transmitted by the ith user. In other words, the packets that are
dropped because of exceeding delay threshold are not counted at S. High amounts of jitter may cause out-of-order or loss
of multimedia data, ie, interrupted video frames, and voice packets, and poor quality of experience to users. Thus, it is a
crucial telecommunications that has not been taken into account by the popular schedulers.
At TES+, Equation (8) is used to calculate the scheduling index for the ith user, where jitter is incorporated into the 𝜌i
part of Equation (12), and the new expression is given by
𝜌i={exp(𝜎2
DOLi−Ti−DOLi
Ti
),𝑓or RT traffic.
𝛼i∗DOLi,𝑓or nRT traffic.
(16)
And the 𝛾ipart of the expression of TES+ is the same as in Equation (13).
Figure 5 shows the ratio of Z𝜌i∗and Z𝛾i∗on scheduling decisions at 1000 scheduling events for TES and TES+. A single
figure is given to show the performance of both schemes since their performance is similar to each other. It can be seen
from the figure that Z𝛾i∗>Z𝜌i∗. This gives more chance to the scheduler to make opportunistic decisions and may signif-
icantly improve the system performance. A comprehensive performance investigation between EXP/PF, M-LWDF, and
proposed methods is provided in Section 5 of this paper.
FIGURE 5 Weight on scheduling decision for TES and TES+
versus scheduling event
8of14 HACI AND ABDELBARI
5SIMULATION AND RESULTS
The simulation is done using MATLAB (version R2018a) program. Five performance metrics—average spectrum effi-
ciency, delay QoS violation, average delay, jitter, and Jain's fairness index—are investigated to compare proposed schemes
with PF, M-LWDF, and EXP/PF scheduling schemes.
5.1 Performance metrics
Average spectrum efficiency performance metric measures the total throughput achieved by scheduled users during the
course of simulation divided by the total number of subcarriers multiplied with the number of scheduling events.38
The delay QoS violation performance metric is the ratio of dropped RT packets to the total number of RT packets to be
transmitted.10 This metric shows how effective the scheduling scheme is to satisfy the delay constrains on the RT traf-
fic, which in practice requires a very low violation value. The average delay performance metric is a measure of all the
delay times experienced by scheduled RT streams averaged over the total number of RT packets served. Jitter performance
metric measures the deviation from the periodicity of the incoming data packets. This metric shows how successful the
scheduling scheme is to provide stable, timely flow of data. The well-known Jain's fairness index is used to measure the
degree of fairness experienced among users by employing the proposed scheduling scheme as well as the PF, M-LWDF,
and EXP/PF. Jain's index measures the distribution of RBs over the active users. A high degree of fairness is achieved if
the performance perceived by users is well balanced (ie, similar to each other).40
5.2 Simulation results
In the simulations, 1000 scheduling events are considered, and the total bandwidth is 10 MHz, where each subcarrier is
considered to have 15-kHz band. The power spectral density of additive white Gaussian noise (AWGN) is assumed to be
unity (normalized). The delay threshold of the RT streams is assumed to be 10 ms, which is envisioned to be in the 5G and
beyond networks.5The movement of users is set to be random within 700 m radius, which is suitable for urban microcell
environment, and the traffic streams are selected between RT and nRT randomly for all users within the service area with
50:50, 70:30, and 80:20 ratio for RT to nRT traffic. These ratios are chosen according to the recent study.6By Cisco,6it is
expected that in the future, the Internet traffic will be dominated by RT (especially videos traffic) traffic up to four times
compared with nRT traffic. It is considered that there are 50 RBs available at the BS, and the number of users is changed
from 20 to 200. For the convenience of the readers, all the parameters used in the simulations, and their setting values are
summarized by Table 1.
Figure 6A shows the system throughput expressed in spectral efficiency. It is shown that the proposed schemes
(TES and TES+) achieve significantly higher throughput than PF, EXP/PF, and M-LWDF schemes, where the perfor-
mance of popular M-LWDF and EXP/PF follow similar trend. PF scheme achieves higher throughput than EXP/PF
and M-LWDF since it just considers maximizing the long-term throughput of users and does not consider delay con-
strains. EXP/PF has a better throughput compared with M-LWDF when the number of users is lower than 50 users.
With the increase in the number of users, M-LWDF performs better than EXP/PF since it prioritizes users because
of delay constrains using a logarithmic function while EXP/PF uses an exponential function. TES and TES+ schemes
outperform all other considered schemes because of the first term in Equation (13) that provides additional weight to
throughput-related parameters at decision making. Thus, TES scheme has a more balanced trade-off between through-
put maximization and QoS satisfaction. Moreover, TES+ scheme achieves higher throughput than TES scheme for all the
TABLE 1 Simulation parameters Parameters Values
Number of simulation events 1000
System bandwidth 10 MHz
Sub-carrier bandwidth 15 kHz
Cell radius 700 m
Number of RBs 50
Number of UEs 20-200
RT:nRT ratio 50:50, 70:30, 80:20
TTI Duration 1 msec
RT delay threshold 10 msec
Number of OFDM symbols per slot 7
HACI AND ABDELBARI 9of14
FIGURE 6 Performances for 80:20 RT:nRT ratio. A, Spectrum Efficiency, B, Average Delay, C, Delay QoS Violation, D, Jain's Fairness
Index, and E, jitter for ultra-dense network where RT and nRT ratio is (80:20)
considered number of users. This shows that incorporating the jitter term into decision making, by Equation (16), let the
scheduler be more opportunistic. Thus, TES and TES+ schemes are promising scheduling schemes for future ultra-dense
networks.
In Figure 6B, the average delay performance for RT streams is shown. The highest delay is experienced with PF as it
does not consider delay-related parameters at its decisions. Although M-LWDF and EXP/PF have a lower average delay
compared with PF, their performance is significantly worse compared with TES and TES+. This is because M-LWDF
and EXP/PF sacrifice from throughput while they try to meet delay requirements in a conservative way (ie, too much
weight on delay parameters at decision making). Accordingly, from the queuing theory, the reduced throughput results
into higher average delay. Further, TES+ achieves considerably better performance than TES, because of the additional
sensitivity to the variance of packet delay in Equation (16).
10 of 14 HACI AND ABDELBARI
Figure 6C shows the delay QoS violation ratio, which is the ratio of packets lost because of exceeding an RT delay
threshold to that of total number of transmitted RT packets. PF has one of the highest violation, and the ratio increases
rapidly when the number of users increases to more than 40 (ie, not even at a very dense network). The performance of
M-LWDF and EXP/PF shows similar trend. Their performance is better than PF's up to 120 users (ie, moderately dense
network) but becomes similar to PF's when the network gets densely populated. This is due to low spectral efficiency
achieved by these schemes. Above 120 users, the effect of high non-RT traffic and delay sensitivity in EXP/PF cost it more
packet loss than M-LWDF and PF schemes. It is shown by Figure 6C that TES provide the minimum delay violation ratio
compared with other scheduling algorithms because of the enhancements in both throughput and delay performances.
Furthermore, TES+ outperforms TES especially at dense network scenarios, which confirms one more time that the jitter
term has a significant effect on the scheduling decisions.
In Figure 6D, the fairness experienced among users is shown by employing various schemes. M-LWDF and EXP/PF
are the schemes with lowest performance because they are the most conservative schemes to schedule RT traffic over
nRT, which is not so fair to nRT users especially when the number of RT users increases. PF scheme provides better
performance than M-LWDF and EXP/PF since it does not consider delay-related parameters at its decisions (ie, prioritize
RT over nRT traffic). Therefore, it can be more fair among RT and nRT traffic scheduling decisions. TES and TES+ achieve
the highest fairness compared with other schemes even with a very dense network. This is due to having a more balanced
weight distribution between throughput and delay-related parameters at scheduling decisions. All algorithms experience
a degradation in fairness index as the number of users increases because the RBs become scarce, and scheduling schemes
need to make more stringent decisions to meet RT traffic QoS requirements.
Figure 6E shows the jitter performance. It can be seen from the figure that EXP/PF and M-LWDF have the two worse
performances since these are the schemes that achieve the lowest spectral efficiency, see Figure 6A. PF, TES, and TES+
have similar performances when the network is lightly loaded (ie, up to 80 users). However, as the network starts to get
densely populated, TES+ outperforms all the other schemes since it is the only scheme that incorporates jitter into its
decisions.
FIGURE 7 Performances for different RT:nRT traffics ratios A, Delay QoS Violation, B, Average Delay, C, Jain's Fairness Index, and D, Jitter
HACI AND ABDELBARI 11 of 14
FIGURE 8 Performances of scheduling schemes for RT video communications (video conferencing) A, Packet Loss Ratio, B, Average
Delay, and C, Jitter
Figure 7A-D compares the performances of TES, TES+, and M-LWDF in a densely populated network with various ratio
of RT to nRT traffic, where the RT traffic will take up to 70% of the total traffic demand.6The simulation results include two
RT to nRT traffic ratios, where 50% RT - 50% nRT (1:1), and 70% RT - 30% nRT (2.3:1) traffic are considered. M-LWDF is
chosen as the scheme to have further performance comparison with TES and TES+, since it is the state-of-the-art scheme
that has the third best performance (especially average delay and delay QoS violation ratio metrics) after the proposed
schemes.
Figure 7A compares the delay QoS violation ratio for M-LWDF, TES, and TES+ schemes when 50:50 and 70:30 RT to
nRT traffic is considered. It can be seen that violation ratio increases for all schemes with the increase of the RT to nRT
traffic ratio. However, the performances for TES and TES+ even at 70:30 is better than the performance of M-LWDF at
50:50. This shows that the proposed schemes can deliver a superior performance to the popular schemes even at a stringer
scenario. One shall notice that future networks will be populated mainly by RT traffic users, so working well in scenarios
with high number of RT users should be a condition for future scheduling schemes.
Figure 7B shows the average delay performance experienced by RT traffic when M-LWDF, TES, and TES+ schemes are
employed at scheduling. It can be seen that for all schemes, the average delay increases as the number of users and the
RT to nRT ratio increase. For all the considered scenarios, TES+ scheme significantly outperforms M-LWDF and TES
especially when the network becomes densely populated.
Figure 7C shows the Jain's fairness index performance perceived by users when M-LWDF, TES, and TES+ schemes
are employed at scheduling for different RT to nRT traffic ratios. The performance of all the considered schemes is not
affected much with the change of the RT:nRT ratio. This shows that the fairness of the schemes is related strongly to the
number of users and weakly to the ratio of RT to nRT users.
Finally, Figure 7D shows the jitter performance for the M-LWDF, TES, and TES+. It can be seen that TES and TES+
schemes significantly outperform the M-LWDF scheme. Thus, by having a more balanced weight distribution between
throughput and delay-related parameters, the scheduler can have more freedom and schedule users more periodically to
meet the requirements. This considerably improves the jitter performance.
12 of 14 HACI AND ABDELBARI
Figure 8A-C show the performance of the considered scheduling schemes for a real application that is expected to dom-
inate the traffic in 5G and beyond networks.6This application is video conferencing (VC). VC is an application that can be
classified under RT video communications with strict QoS requirements. VC requires an average throughput of 1 Mbps,
latency less than 150 ms, packet loss ratio no more than 1%, and jitter less than 30 ms.41 Presenting the schedulers' perfor-
mance for a real application is important to give a notion to the reader on how well the considered schemes can perform
on practice. The performance metrics have been chosen with respect to QoS requirements for interactive multimedia
applications.41,42
Figure 8A shows the packet loss ratio performance versus system utilization. To provide a high QoS VC application
shall encounter no more than 1% packet loss ratio. It can be seen from the figure that PF, M-LWDF, and EXP/PF schemes
can only support up to 75 users with such requirement, where TES and TES+ schemes can support up to 110 and 125
users, respectively. This means TES and TES+ can serve about 47% and 67% more users while achieving a high QoS.
In Figure 8B, delay performance is shown, which constitutes a part of the latency QoS requirement. Latency has propa-
gation, transmission, and queuing delay components. The scheduling delay is under the queuing component. In order to
satisfy the 150-ms latency requirement, scheduling delay should be no more than 10 ms.5Itcanbeseenfromthefigurethat
only TES and TES+ schemes can stay under this value. Finally, Figure 8C presents the jitter performance. It can be seen
that TES and TES+ can significantly outperform the popular schemes, especially when the network is densely populated.
By incorporating the jitter term at its formulation, TES+ scheme provides the best performance, and the performance gap
between TES and TES+ increases as the number of users increase.
6CONCLUSION
Scheduling schemes take a crucial part in determining the system performance and user QoS satisfaction achieved by
telecommunication networks. Thus, there is a need for much research to develop schemes that can meet demands of
5G and beyond networks. This paper proposed two novel schemes, called as TES and TES+, that introduce a two terms
to the scheduling decision making and reformulates the parameters used by well-known schemes. The aim was to have
a more balanced weight distribution between throughput and delay-related parameters at scheduling decisions, also to
study the effect of a crucial telecommunications-related parameter, jitter, on scheduling performance. For this reason, the
performance of both TES and TES+ schemes is shown individually for the five popular performance metrics. The readers
may well investigate and understand the effect of jitter on scheduling performance by comparing the curves of TES and
TES+ for each metric. The performance results show that TES and TES+ can achieve higher average system throughput
and fairness and lower average delay, delay QoS violation, and jitter compared with popular schemes PF, M-LWDF, and
EXP/PF. For future research, TES and TES+ scheme can be improved to incorporate intercell interference information
into their decision making for multicell environment.
ORCID
Huseyin Haci https://orcid.org/0000-0002-0720-479X
Amr Abdelbari https://orcid.org/0000-0002-5861-7762
REFERENCES
1. Ng TS, Chuah TC, Tan YF. QoS-based radio network dimensioning for LTE networks with heavy real-time traffic. Int J Commun Syst.
2014;27(11):2455-2480. https://onlinelibrary.wiley.com/doi/abs/10.1002/dac.2447
2. Basukala R, Ramli HAM, Sandrasegaran K. Performance analysis of EXP/PF and M-LWDF in downlink 3GPP LTE system. In: 2009 First
Asian Himalayas International Conference on Internet; 2009; Kathmandu, Nepal:1-5.
3. Benchaabene Y, Boujnah N, Zarai F. 5G cellular: Survey on some challenging techniques. In: 2017 18th International Conference on
Parallel and Distributed Computing, Applications and Technologies (PDCAT); 2017; Taipei, Taiwan:348-353.
4. Ge X, Tu S, Mao G, Wang C, Han T. 5G ultra-dense cellular networks. IEEE Wirel Commun. 2016;23(1):72-79.
5. Sanchez MI, Asadi A, Draxler M, et al. Tackling the increased density of 5G networks: The crowd approach. In: 2015 IEEE 81st Vehicular
Technology Conference (VTC Spring); 2015; Glasgow, UK:1-5.
6. Cisco. Cisco visual networking index: Global mobile data traffic forecast update, 2016-2021. White Paper; 2017.
7. Jang H-C, Huang C-W, Yeh F-K. Design a bandwidth allocation framework for SDN based smart home. In: 2016 IEEE 7th Annual
Information Technology, Electronics and Mobile Communication Conference (IEMCON); 2016; Vancouver, BC, Canada:1-6.
8. Kayali MO, Shmeiss Z, Safa H, El-Hajj W. Downlink scheduling in LTE: Challenges, improvement, and analysis. In: 2017 13th
International Wireless Communications and Mobile Computing Conference (IWCMC); 2017; Valencia, Spain:323-328.
HACI AND ABDELBARI 13 of 14
9. Hendaoui S, Zangar N, Tabbane S. PDAS: PLR-delay aware scheduler for real time application in LTE-A. In: 2017 Sixth International
Conference on Communications and Networking (ComNet); 2017; Hammamet, Tunisia:1-7.
10. Haci H, Zhu H, Wang J. Novel scheduling for a mixture of real-time and non-real-time traffic. In: 2012 IEEE Global Communications
Conference (GLOBECOM); 2012; Anaheim, CA, USA:4647-4652.
11. Cao Y, Li VOK. Scheduling algorithms in broadband wireless networks. Proc IEEE. 2001;89(1):76-87.
12. Mohammed H, Radcliffe Pj. Next generation wireless mobile system efficient, fair, class based packet scheduling algorithm. Wireless Pers
Commun. 2013;72(4):1969-1991. https://doi.org/10.1007/s11277-013-1116-x
13. Mohammed H, Radcliffe PJ. A packet scheduling scheme for 4G wireless access systems aiming to maximize revenue for the telecom
carriers. Telecommun Syst. 2014;57(4):347-366. https://doi.org/10.1007/s11235-013-9840-9
14. Priya LR, Soundar KR. Lte: An enhanced hybrid domain downlink scheduling. Cogn Syst Res. 2018;52:550-555. http://www.sciencedirect.
com/science/article/pii/S1389041718302717
15. Taboada I, Aalto S, Lassila P, LiberalF. Delay- and energy-aware load balancing in ultra-dense heterogeneous 5G networks. Trans Emerging
Telecommun Tech. 2017;28(9):e3170. https://onlinelibrary.wiley.com/doi/abs/10.1002/ett.3170
16. Sa
glam Mi, Kavas A, Kartal M. A new long-term evolution downlink delay based lean scheduler. IEEJ T Electr Electr Eng. 2018;0(0).
https://onlinelibrary.wiley.com/doi/abs/10.1002/tee.22860
17. Wang Y-C, Jhong D-R. Efficient allocation of LTE downlink spectral resource to improve fairness and throughput. Int J Commun Syst.
2017;30(14):e3287. https://onlinelibrary.wiley.com/doi/abs/10.1002/dac.3287
18. Capozzi F, Piro G, Grieco LA, Boggia G, Camarda P. Downlink packet scheduling in LTE cellular networks: Key design issues and a survey.
IEEE Commun Surv Tut. 2013;15(2):678-700.
19. Subramanian R, Sandrasegaran K, Kong X. Performance comparison of packet scheduling algorithms in LTE-A HetNets. In: 2016 22nd
Asia-Pacific Conference on Communications (APCC); 2016; Yogyakarta, Indonesia:185-190.
20. Benchaabene Y, Boujnah N, Zarai F. Performance comparison of packet scheduling algorithms for voice over ip in LTE cellular network.
In: 2016 4th International Conference on Control Engineering Information Technology (CEIT); 2016; Hammamet, Tunisia:1-7.
21. Nasralla MM, Martini MG. A downlink scheduling approach for balancing QoS in LTE wireless networks. In: 2013 IEEE 24th Annual
International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC); 2013; London, UK:1571-1575.
22. Ghariani T, Jouaber B. Energy consumption evaluation for LTE scheduling algorithms. In: 2015 International Symposium on Networks,
Computers and Communications (ISNCC); 2015; Hammamet, Tunisia:1-5.
23. Chayon HR, Dimyati K, Ramiah H. An efficient packet scheduling algorithm to improve the performance of cell-edge user in LTE network.
In: 2017 IEEE 13th Malaysia International Conference on Communications (MICC); 2017; Johor Bahru, Malaysia:130-135.
24. Liu J, Li X, Ji H, Tang Y. Traffic-pairing scheme based on particle swarm optimization in downlink comp-mu-mimo system. In: 2014 IEEE
Wireless Communications and Networking Conference (WCNC); 2014; Istanbul, Turkey:1779-1784.
25. Ramli HAM, Isa FNM. Improving real-time multimedia scheduling in practical mobile cellular channels. In: 2016 6th International
Conference on Intelligent and Advanced Systems (ICIAS); 2016; Kuala Lumpur, Malaysia:1-6.
26. Liqiang W, Xinyue L, Shan G, Xingling H. Resource allocation strategy in SCMA based multiple access system. In: 2017 First International
Conference on Electronics Instrumentation Information Systems (EIIS); 2017; Harbin, China:1-6.
27. Yang W, Zu Y. The research of QoS guarantee mechanism of the secondary users in cognitive radio networks. China Commun.
2016;13(10):146-152.
28. Nguyen D, Nguyen H, Renault É. Performance evaluation of E-MQS scheduler with mobility in LTE heterogeneous network. In: 2017
IEEE International Conference on Communications (ICC); 2017; Paris, France:1-6.
29. Nguyen DH, Nguyen H, Renault E. E-MQS - a new downlink scheduler for real-time flows in LTE network. In: 2016 IEEE 84th Vehicular
Technology Conference (vtc-fall); 2016; Montreal, QC, Canada:1-5.
30. Ang EM, Wee KK, Pang YH, Lau SH. Two-level scheduling framework with frame level scheduling and exponential rule in wireless
network. In: 2014 International Conference on Information Science Applications (ICISA); 2014; Seoul, South Korea:1-4.
31. Nguyen P, Rao B. Optimal fair opportunistic scheduling for wireless systems via classification framework. IEEE Trans on Cogn Commun
Netw. 2015;1(2):185-199.
32. Chen T, Shan H, Wang X. Packet scheduling for on-demand data services to high-speed trains over wireless links. In: 2013 IEEE Global
Communications Conference (GLOBECOM); 2013; Atlanta, GA, USA:4507-4512.
33. Wang X, Cai L. Limiting properties of overloaded multiuser wireless systems with throughput-optimal scheduling. IEEE Trans Commun.
2014;62(10):3517-3527.
34. Jang H-C, Hu C-P. Fairness-based adaptive QoS scheduling for LTE. In: 2013 International Conference on ICT Convergence (ICTC); 2013;
Jeju, South Korea:626-631.
35. Kuokkwee W, Ee Mae A, Wee K, Han Pang Y, Phang K. A performance analysis on packet scheduling schemes based on an exponential
rule for real-time traffic in LTE. EURASIP J Wirel Commun Netw. 2015;2015:201.
36. Biernacki A, Tutschku K. Comparative performance study of LTE downlink schedulers. Wireless Pers Commun. 2014;74(2):585-599.
https://doi.org/10.1007/s11277-013-1308-4
37. Lai WK, Tang C-L. QoS-aware downlink packet scheduling for LTE networks. Comp Netw. 2013;57(7):1689-1698.
38. Ali S, Zeeshan M. A utility based resource allocation scheme with delay scheduler for LTE service-class support. In: 2012 IEEE Wireless
Communications and Networking Conference (WCNC); 2012; Shanghai, China:1450-1455.
39. Rappaport TS. Wireless Communications: Principles and Practice. 2nd ed., Prentice Hall Communications Engineering and Emerging
Technologies Series: Upper Saddle River, N.J: Prentice Hall; 2002.
14 of 14 HACI AND ABDELBARI
40. SHI H, Prasad RV, Onur E, Niemegeers IGMM. Fairness in wireless networks:issues, measures and challenges. IEEE Commun Surv Tut.
2014;16(1):5-24.
41. Tesfamicael AD, Liu V, Foo E, Caelli B. Modeling for a QoS-efficient real-time video conferencing system in the cloud. In: 2018 26th
International Conference on Software, Telecommunications and Computer Networks (SOFTCOM); 2018; Split, Croatia:1-6.
42. Al-Shammari BKJ, Al-Aboody N, Al-Raweshidy HS. IoT traffic management and integration in the QoS supported network. IEEE Internet
Things J. 2018;5(1):352-370.
Howtocitethisarticle: Haci H, Abdelbari A. Throughput enhanced scheduling (TES) scheme for ultra-dense
networks. Int J Commun Syst. 2019;e4229. https://doi.org/10.1002/dac.4229
A preview of this full-text is provided by Wiley.
Content available from International Journal of Communication Systems
This content is subject to copyright. Terms and conditions apply.