ISSN: 2449 – 0539

BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29

A Composite Packet Scheduling

Algorithm for LTE Downlink
O.J. Ifere1, L.M. Bello2 and D.S. Shu’aibu3
1,2,3
Department of Electrical Engineering, Faculty of Engineering
Bayero University Kano

ABSTRACT
This paper proposes a new scheduling algorithm for Long Term Evolution (LTE) downlink systems called
Composite Scheduler (CMP) for offering simultaneously Real Time (RT) and Non-Real Time (NRT) applications
resources. CMP algorithm dynamically controls channel resources in a manner which reduces packet drop in an
overloaded system while still guaranteeing delay bounds and high throughput of real-time services a problem
inherent in the Earliest Deadline First (EDF) scheduler.TheCMP is obtained by modifying the EDF scheduling
algorithm using natural logarithm, number of RT users and LTE Quality of Service (QoS) Class Identity (QCI)
priority index which reduces the domino effect experienced by the scheduler during overload. Simulation was
done using LTE-Sim with varying number of users ranging from 5 – 60 in a cell radius of 2km and each user
receiving one video, one Voice over Internet Protocol (VoIP), one infinite buffer and one Internet Protocol (IP)
Multimedia Subsystem (IMS) flow simultaneously. The proposed algorithm is evaluated against Proportional
Fair (PF), EDF and Modified Largest Weighted Delay First (MLWDF) scheduling algorithms for real-time
services. The result shows great increase in spectral efficiency by about 10%, 21% decrease in packet loss, an
improved fairness of 12%, lowers delays by 75ms and 10 times increase in throughput for RT services when
compared to normal EDF scheduler.

1.0 INTRODUCTION
In wireless communication systems, packet effort schedulers, the QoS requirement of a
schedulers are employed to manage flow of user (e.g. delay) is not considered whereas
traffic from user queues by assigning such requirements are considered and are of
shared resource to users at a given time. utmost importance for QoS schedulers. The
Packet scheduler determines the order in goal of a good resource scheduler is to
which packets are executed. The process of achieve the required QoS and provide an
assigning users’ packets to appropriate optimal balance of spectral efficiency and
shared resource to achieve some fairness of the system (Furht & Ahson,
performance guarantee is called packet 2011).
scheduling(Furht & Ahson, 2011). Several Long Term Evolution (LTE) is considered
packet scheduling algorithms have been the successor to Universal Mobile
employed in cellular networks to handle Telecommunication System (UMTS) for
resource allocation. Advancement in wireless mobile communicationaccording
technology has seen numerous schedulers to Third Generation Partnership Project
been developed for wireless cellular (3GPP) in its Release 8 and later Release
networks. Schedulers maybe grouped into 9(Iwamura, Umesh, & Hapsari, 2009). It
best-effort or Quality of Service (QoS) aims at tackling a number of problems
schedulers(Furht & Ahson, 2011). In best including: increased number of data

Also available online at https://www.bayerojet.com 15

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
consumption, higher data rates demanded Kakishima, & Yasukawa, 2012).Resource
by Real Time (RT) applications, improved allocation through packet scheduling plays
mobility, high delay experience etc. a key role in enabling QoS guarantees.
(Iwamura et al., 2009). LTE implements an In this paper a new packet scheduling
all Internet Protocol(IP) flat network algorithm called Composite Scheduler
architecture which is an improvement to (CMP) is proposed for both RT and Non-
UMTS core network with optimization in RT (NRT) applications in a mixed traffic
packet switching traffic, mobility and scenario.
QoS(Kurose & Ross, 2008). LTE utilizes The remaining part of thepaper is organized
Orthogonal Frequency Division Multiple as follows. Section 2 provides a background
Access (OFDMA) for its downlink of packet scheduling and similar literatures
communication and employs several studied. The proposed CMP scheduling
technologies to achieve its goal: Resource algorithm for downlink LTE is presented in
allocation, Channel Quality Indicator Section 3. Section 4 discusses the simulated
(CQI), Adaptive Modulation and Coding parameters used in the study and section 5
(AMC), Multiple Input Multiple Output discusses the simulation scenario. Results
(MIMO), Hybrid Automatic Repeat of the study is discussed in section 6 and
Request (HARQ)(Kawamura, Kishiyama, conclusion is made in section 7.

2.0 BACKGROUND
Packet schedulers are employed in cellular
networks to prioritize/sort user packets.
There are two main types of packet
schedulers: Uplink packet schedulers which
are employed when User Equipment (UE)
sends data to eNodeB (eNB) and Downlink
packet schedulers which are employed
when the eNB sends data to the UE. Packet
scheduling in LTE involves two key
aspects: first the ordering/prioritization of Fig. 1. Packet Scheduler Model
packets which is done in the form of The UE to be scheduled and its
assigning users’ metrics based on their corresponding number of RBs is selected
importance and secondly the allocation of by the packet scheduler based on channel
RBs to users for transmission. Fig. 1 condition and QoS requirements. The
illustrates the procedure followed when information of channel quality of each UE
carrying out resource allocation by the is available to the packet scheduler through
eNB. CQI which is available to the eNB via
reports of UEs through Physical Uplink
Control Channel (PUCCH). This value is
then used to choose the appropriate
Modulation and Coding Scheme (MCS) for
the UE. The eNB informs the UE on the
selected MCS and allocated number of RBs

Also available online at https://www.bayerojet.com 16

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
through the Physical Downlink Control Multiplexing (OFDM) which is a multi-
Channel (PDCCH) and data is transferred carrier modulation scheme (Srikanth S,
from the eNB through the Physical Kumaran V, & Manikandan C, 2006). In
Downlink Shared Channel (PDSCH). This OFDMA, the shared spectrum is divided
process is repeated for every TTI both in time and frequency domain (OFDM
(Radhakrishnan, Neduncheliyan, & symbols and sub-carriers respectively) and
Thyagharajan, 2016). assigned to multiple users (Srikanth S et al.,
The downlink communication between 2006). Fig. 2a shows the LTE Frequency
eNB and UE is carried out in a channel Division Duplex (FDD) (Type 1) frame
which employs OFDMA. OFDMA is one structure and Fig. 2b illustrates how a frame
of the multiple access schemes derived is divided in time and frequency domain.
from Orthogonal Division Frequency

Fig. 2a. LTE FDD Frame Structure(Techplayon, 2017)

Fig. 2b. LTE Frame Resource Component(Techplayon, 2017)

At the time domain, the available spectrum OFDM symbols (Normal Cyclic Prefix) or
is divided into frames. Each frame consist 6 OFDM symbol (Extended Cyclic Prefix).
of 10 sub-frames or Transmission Time The available spectrum bandwidth is
Interval (TTI) of 1ms each and each sub- divided into sub-channels of 180 kHz in the
frames consist of 2 time slots (Afroz, frequency domain. A sub-channel consists
Heidery, Shehab, Sandrasegaran, & of 12 succeeding and equally spaced sub-
Shompa, 2015). Each time slot consists of 7 carrier of 15 kHz. Users are allocated a

Also available online at https://www.bayerojet.com 17

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
portion of this time-frequency resource especially with increasing congestion
known as Resource Blocks (RB) for (Afroz, Sandrasegaran, & Ghosal, 2014;
transmission. The RB is a time-frequency Nwawelu et al., 2017). Another scheduler is
radio resource spanning a time slot (0.5ms) the Earliest Deadline First (EDF) scheduler
in time domain and a sub-channel (180 studied in(Stankovic, Spuri, Ramamritham,
kHz) in the frequency domain. However, & Buttazzo, 1998). It is a pre-emptive,
the smallest allocable resource for dynamic scheduler that seeks to reduce the
transmission is a pair of Physical Resource delay experienced by the overall system. In
Blocks (PRB) since the TTI in LTE last for this scheduler, each users’ packet is queued
1ms (2 time slots) (Iacobucci, 2013). Also, based on their delay and users with higher
LTE supports several channel bandwidths delay constrained services are given higher
for transmission (i.e. 1.4 MHz to 20 MHz) priority. User’s metrics are computed as
and this dictates the number of RBs presented in (2):
available for allocation (i.e. 6 – 100)
(Anritsu, 2015). The smallest unit however 1
𝐸𝐷𝐹 = 𝐷−𝐻𝑂𝐿 2
is called Resource Element (RE) and it is
the building block of RB. It represents a
single sub-carrier during an OFDM symbol whereD is the delay target of user flow, and
interval. HOL is the packet Head of Line delay. A
major drawback to this scheme is the huge
Over the years, several packet scheduling packet dropped experienced during
algorithms have been proposed to handle congestion which greatly affects the system
the resource sharing/prioritization in performance. This effect known as the
mobile networks. In (Jalali, Padovani, & domino effect is studied in (Stankovic et al.,
Pankaj, 2000) a Proportional Fair (PF) 1998)
algorithm was proposed. It is a scheduling
algorithm aiming to balance a user’s In (Saleh & Dong, 2010) the authors
throughput and fairness. Using CQI and compared the traditional First Come First
AMC, the scheduler is able to obtain the Serve (FCFS) scheduler with the EDF to
users achievable/maximum throughput and evaluate their performance in terms of miss
average throughput over a period of time ratio, delay and average size of buffer. The
which is then used to calculate the users’ supremacy of the EDF was ascertained
metric. The user’s metric is thus computed under the studied conditions, however both
as shown in (1) (Jalali et al., 2000): schedulers are unsuitable for LTE due to
lack of prioritization of RT flow (i.e. FCFS)
𝑟𝑖(𝑡) and high packet drop as earlier mentioned (1)
𝑃𝐹 = 1
𝑅𝑖(𝑡)
(i.e. EDF).The Modified Largest Weighted
where 𝑟𝑖(𝑡)is the user’s achievable
Delay First (MLWDF) scheduling
throughput and 𝑅𝑖(𝑡)= is the average
algorithm proposed in(Andrews et al.,
throughput over a given period.
2001) was developed to handle and provide
higher priority to RT applications in a view
The PF scheduler though able to offer high
of providing QoS for these flows. It
level of fairness to all users, fails to meet
supports users of multiple services with
the deadline requirement of RT users
different QoS requirements. To achieve

Also available online at https://www.bayerojet.com 18

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
this, it considers the users maximum and resource allocation. This scheme was able
average throughput over a period like in PF to handle a low loaded system but failed and
together with the HOL of each packet, the led to starvation, low throughput and high
maximum allowable time of a users’ packet delays in NRT services as the network
and the probability of QoS requirement grew. The second approach known as
violation. In this scheme a user metric is Delay-Fairing (DF) Approach employed
calculated in (3): adding an exponential factor to the PF
algorithm based on the users’ deadline for
RT applications and an unchanged PF
𝑟𝑖(𝑡) (3)
𝑀𝐿𝑊𝐷𝐹 = 𝑎𝑖(𝑡) 𝑤𝑖(𝑡) 𝑅𝑖(𝑡) 3 algorithm for NRT users which enabled the
scheduler meet users’ deadlines but again
led to starvation in a highly loaded scenario.
where 𝑟𝑖(𝑡) and 𝑅𝑖(𝑡) are the same as in PF In (Dardouri & Bouallegue, 2014) a
algorithm. 𝑤𝑖(𝑡)represents the weight of comparative analysis of PF, MLWF and
the user which is equivalent to HOL packet Exponential/PF scheduling under low
in the user queue or length of user queue. network load (5-20 users) was done. A
𝑎𝑖(𝑡)reflects the strictness of QoS mixed traffic scenario of BE, video and
guarantee.During system congestion, the VoIP flow in a pedestrian and vehicular
algorithm prioritizes RT flow leading to environment was used and metrics such as
better performance of such flows but throughput, fairness index, delay, packet
degrading performance for NRT flows loss ratio (PLR) and spectral efficiency
through a phenomenon known as were considered. The MLWDF was seen to
“Starvation”(Nwawelu et al., 2017) which provide better performance in video flows
led to unfairness in the system. In (Xian, while Exponential/PF and PF performed
Tian, Xu, & Yang, 2011) the authors seek better in VoIP flows. It was also observed
to improve the fairness problem of that by changing from a slower speed
MLWDF. To achieve this, they provided a pedestrian environment to a higher speed
theoretical analysis of MLWDF fairness vehicular environment users’ performance
which showed that its fairness was affected drop as expected due to increase in
by channel conditions, packet arrival multipath losses in the network. (Basukala,
process and the ratio of QoS requirement of Ramli, & Sandrasegaran, 2009) carried out
users’ queues. Based on their analysis an a performance analysis on EXP/PF and
Enhanced MLWDF (EM-MLWDF) MLWDF algorithms in a multimedia
algorithm was proposed whose fairness was environment with RT video streaming
independent of this factors and thus services and NRT web browsing services.
improved the fairness of users but more Evaluation was done via throughput:
complex.(Müller, Schwarz, & Rupp, 2013) average RT and NRT throughput, packet
investigated the ability to provide QoS loss for RT services and fairness for NRT
guarantee for RT services based on the PF services. Result should that MLWDF
algorithm. In view of this, two scheduling performed better with fewer users (< 85)
procedures were proposed: a Two-Layer providing PLR of less than 1% and average
(TL) scheme in which all the RT services RT throughput of 100kbps. Once users
are first served before NRT services with increased, MLWDF performance dropped
both layers employing the PF algorithm for as its means of prioritizing users via HOL

Also available online at https://www.bayerojet.com 19

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
alone was not sufficient and RT services throughput over a period, channel
deadline got missed. Conversely, EXP/PF conditions etc.). In addition, the review also
performed better at higher users (>125 highlighted their drawback e.g. EDF high
users) providing higher throughput and packet loss during congestion, FCFS and
lower PLR at this range. PF lack of prioritization mechanism for RT
The different works been reviewed flows and MLWDF starvation problem. In
provided an understanding to the several light of this a new algorithm is proposed to
possible parameters employed in mitigate these weaknesses.
scheduling (e.g. delay, HOL, users’ average

3.0 COMPOSITE SCHEDULER

The proposed algorithm employs a wherea is a positive constant used in
composition of deadlines, retention priority adjusting the strictness to QoS of the
number from LTE QCI (Sheta, Zaki, & system. a is computed as in (5):
Keshk, 2013), and number of real-time
users to compute the users’ metric. During a = 𝑙𝑜𝑔10 (𝑁𝑅𝑇 ) 5
congestion, the earliest deadline first
scheduling algorithmsis known to behave where 𝑁𝑅𝑇 is the number of RT
unpredictable leading to a high possibility users.Normally as users increases in EDF
of its packet been dropped as it is unable to scheduler, more packets are lost due to the
meet all users’ deadline. This phenomenon inability of the scheduler to meet their
known as the domino effect (Stankovic et deadlines. By reducing theeffects of
al., 1998) is caused when the system deadline-ness by means highlighted above
experiences transient overload leading to fewer packets are lost and thus higher
packet loss.To reduce this effect and throughput is been achieved.
thereby improve the performance of the Also, as the number of users increases and
EDF algorithm, the dependence on the deadline of RT flow becomes less
deadliness alone as means of prioritization efficient due to the effect of (4), a Priority
need to be reduced. coefficient (P) is employed to increase the
To achieve this, the composite scheduler priority of these flows using LTE QCI
aka CMP employs a natural log on the EDF priority index as shown in (6), thereby
algorithm and also divides this by the reducing packet loss both with small and
logarithm to base 10 of the number of RT large users.
users as in (4) which reduces the metric
dependence on user deadlines as the (
1
)
network grows.The choice of the log 𝑃=𝑎 𝑝𝑟𝑖𝑜𝑟𝑖𝑡𝑦 6
function is to allow for gradual and smooth
degradation and increase of a user metric.
where𝑃 is the Priority coefficient
andpriority is the LTE QCI priority
𝑙𝑜𝑔2 (𝐸𝐷𝐹)
4 indexnumber. (4)
𝑎

Also available online at https://www.bayerojet.com 20

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
To achievethe metric of each user, (4) and
(6) are combinedto compute (7): where 𝑟𝑖(𝑡)and 𝑅𝑖(𝑡)are as in PF algorithm
and EDF is the Earliest Deadline First
𝑙𝑜𝑔2 (𝐸𝐷𝐹) (
1
) 𝑟𝑖(𝑡) scheduling
𝐶𝑀𝑃 = x 𝑎 𝑝𝑟𝑖𝑜𝑟𝑖𝑡𝑦 x 7
𝑎 𝑅𝑖(𝑡) (7)

algorithm.

4.0 SIMULATION PARAMETERS

The simulation parameters used are summarized in the Table 1.

Table 1: Simulation Parameters

Simulation Parameters Values
Frame Structure Frequency Division Duplex (FDD)
Cell Radius 2 km
Bandwidth 5 MHz
Maximum Delay 100 ms
Simulation Duration 120 sec
User speed 3 kmph
Number of Users 5 to 60
Traffic Types Video: H264(242kbps), VoIP:
G.729(8.4kbps), Infinite Buffer,
IMS(8.4Kbps)

5.0 SIMULATION SCENARIO

To evaluate the proposed scheduling Video, one Infinite Buffer and one IP
algorithm, we compare its performance Multimedia System (IMS) packet
with algorithms earlier mentioned (PF, simultaneously. An IMS application is
EDF and MLWDF). The performance implemented by simply reusing the VoIP
comparison is carried out via LTE-Sim traffic model present in LTE-Sim.
(Piro, Grieco, Boggia, Capozzi, & Performance metrics used are average
Camarda, 2011) in a 5MHz single cell with throughput, Jain’s Fairness Index
interference scenario which provides a (Huaizhou Shi, Prasad, Onur, &
proper representation of the LTE model. Niemegeers, 2014), delay and packet loss
The scenario is configured to compose of ratio (PLR). For delay comparisons the PF
users ranging from 5 to 60 moving scheduler is omitted for visualization due to
randomly in a random walk mobility model its extremely poor performance which
at a speed of 3 kmph and a cell radius of 2 hindered better evaluation of other
km. Each user receives one VoIP, one algorithms.

Also available online at https://www.bayerojet.com 21

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29

6.0 SIMULATION RESULT AND DISCUSSION

The downlink throughput result is shown in considerable performance increase with a
Fig.3. Throughput which represents the 20.7%, 35.23% and 8.07% at60 userswhen
number of successful packets transmitted compared to the EDF, PF and MLWDF
shows the CMP to perform better than other schedulers respectively. In Fig. 3b the video
schemes in all flow types. CMP better performance is shown withCMP providing
performance is as a result of its ability to higher performance with only MLWDF
meet deadlines quickly and reduce packet coming close when users are greater than
loss at the same time. MLWDF is next with 40. At 15 users, CMP performed better with
its ability to prioritize RT while also 1761%, 38.36% and 351% against PF,
reducing packet loss via balancing users MLWDF and EDF respectively. Table 2
queue using HOL. This technique is inferior shows the simulated result of video
to the CMP which is able to schedule throughput. In Fig. 3c, IMS performance is
packets faster via employing EDF while shownwith CMP outperforming all
still reducing packet drop. EDF and PF both schedulers especially above 25 users. A
perform poorly. The EDF increasing packet 46.73%, 17.25% and 31.39% improvement
loss due to its inability to handle channel is observed at 60 users against PF, MLWDF
variation especially during congestion and EDF schedulers respectively. Table 3
leading to high packet loss and the PF lack shows the throughput performance
of prioritization mechanism hinders their improvement of CMP to other schedulers as
performances. Fig. 3a illustratesVoIP users increased from 15 to 60.
performance.CMP is shown to provide a

Fig.3.The Throughput of (a) VoIP (b) Video and (c) IMS

Also available online at https://www.bayerojet.com 22

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
Table 2: Video Throughput Simulated Result
Users PF MLWDF CMP EDF
5 256,767.84 650,006.76 849,343.66 289,176.87
10 219,158.68 747,716.80 1,026,387.55 396,332.92
15 49,447.11 665,099.72 920,261.26 203,926.75
20 23,417.50 510,186.94 677,597.39 70,744.31
25 22,032.99 448,517.57 559,582.23 26,724.31
30 27,097.28 420,172.10 497,360.90 25,929.64
35 26,559.83 376,342.45 432,882.87 23,938.65
40 25,676.64 330,711.38 364,643.57 22,567.82
45 29,836.41 313,180.86 345,271.42 22,142.84
50 26,964.36 269,014.54 295,848.24 19,635.28
55 28,172.26 257,727.21 282,920.71 19,840.67
60 30,982.03 250,249.41 276,144.70 18,159.31

Table 3:Throughput performance comparison with respect to CMP for IMS, Video and VoIP.

IMS VIDEO VoIP

Users PF MLWDF EDF PF MLWDF EDF PF MLWDF EDF

15 9.51% -1.17% 49.11% 1761.10% 38.36% 351.27% 8.94% 0.65% 49.02%
20 31.59% 1.59% 68.70% 2793.55% 32.81% 857.81% 30.09% 2.90% 83.98%
25 38.19% 6.74% 67.82% 2439.75% 24.76% 1993.91% 34.14% 4.25% 58.77%
30 41.89% 11.30% 61.17% 1735.46% 18.37% 1818.12% 35.36% 6.84% 53.87%
35 41.92% 14.01% 56.14% 1529.84% 15.02% 1708.30% 34.49% 5.38% 53.83%
40 47.09% 17.06% 67.97% 1320.14% 10.26% 1515.77% 36.01% 4.94% 52.27%
45 50.02% 15.76% 49.27% 1057.22% 10.25% 1459.29% 38.64% 7.63% 33.24%
50 48.95% 18.03% 44.92% 997.18% 9.97% 1406.72% 34.47% 6.34% 25.15%
55 48.11% 19.27% 44.08% 904.25% 9.78% 1325.96% 33.02% 8.28% 26.04%
60 46.73% 17.25% 31.39% 791.31% 10.35% 1420.68% 35.23% 8.07% 20.70%

The PLR is shownin Fig 4 which is a ratio to the PF and MLWDF a performance
of packet loss to packet sent. CMP is again increase of 90.86% and 43.18% at 20 users
seen to be superior especially when and 28.59% and 10.49% at 60 users
compared to the EDF scheduler. The reason respectively is observed. In Fig 4b the PLR
for its superiority is as discussed earlier. It of video traffic is evaluated, the superiority
should be noted that the PF lack of is again shown by CMP providing 39.35%,
prioritization mechanism earlier mentioned 12.43% and 37.13% improvement at 15
results in RT packets not scheduled on time users and 5.54%, 0.46% and 6.08% at 60
leading to high packet loss. In Fig. 4a for users to PF, MLWDF and EDF
VoIP, evaluation showed a performance respectively. For IMS as in Fig. 4c, CMP
increase of 97.64% at 5 users and 22.11% provides the lowest loss especially when
at 60 users to EDF. Also, when compared users are greater than 15 with a 91.65%,

Also available online at https://www.bayerojet.com 23

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
71.50% and 94.10% performance gain at 25 respectively. Table 4 shows the PLR
users and a 39.52%, 23.85% and 33.62% at performance improvement of CMP to other
60 users against PF, MLWDF and EDF schedulers as users increased from 10 to 60
.

Fig.4.PLR of (a) VoIP (b) Video and (c) IMS

Table 4:PLR performance comparison with respect to CMP for IMS, Video and VoIP.
IMS VIDEO VoIP

Users PF MLWDF EDF PF MLWDF EDF PF MLWDF EDF

10 43.66% 25.61% 97.73% 55.03% 23.94% 51.12% 22.94% 16.67% 97.53%
15 89.26% 24.01% 96.50% 39.35% 12.43% 37.13% 88.43% 25.89% 96.30%
20 92.69% 55.77% 95.71% 25.52% 6.02% 25.51% 90.86% 43.18% 95.17%
25 91.65% 71.50% 94.10% 18.84% 3.31% 19.51% 84.57% 45.73% 88.90%
30 84.56% 66.75% 87.61% 14.51% 1.97% 15.28% 70.79% 36.26% 76.41%
35 76.06% 56.12% 78.86% 11.84% 1.34% 12.54% 60.24% 26.76% 66.63%
40 65.50% 43.50% 70.02% 9.46% 0.75% 10.09% 49.61% 16.11% 55.87%
45 57.59% 36.16% 57.93% 8.12% 0.70% 8.78% 42.52% 12.86% 39.56%
50 50.14% 31.29% 48.61% 6.82% 0.57% 7.38% 36.62% 12.65% 30.52%
55 44.78% 28.43% 43.92% 6.14% 0.50% 6.67% 32.00% 12.53% 27.24%
60 39.52% 23.85% 33.62% 5.54% 0.46% 6.08% 28.59% 10.49% 22.11%

as it also adopts the EDF though its

In Fig 5 the delay performance is shown. deadline consideration is not as strict as that
Performance shows the EDF to outperform of the EDF. The MLWDF which balances
all other schemes followed by CMP, users based on HOL performs better than
MLWDF and PF. This performance was the PF which has no deadline consideration
expected as the EDF schedules users based at all leading to very high delays especially
on their deadline alone thereby providing with increasing users. In Fig. 5aforVoIP
the lowest delay possible. CMP flows next flows, CMP performed better than PF and

Also available online at https://www.bayerojet.com 24

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
MLWDF with a 64.97% and 76.20% at 20 flows which require more time for
users and 98.65% and 7.16% at 60 users transmission. The delay for IMS flow is
respectively. Fig. 5b shows the delay in shown in Fig. 5c. CMP outperformed both
video flow. CMP provides lower delays as PF and MLWDF with 91.11%, 82.52% at
compared to MLWDF when users are less 15 users and 98.82%, 19.21% at 60 users
than 35 but slowly degrades above. Its respectively. Table 5 shows the Delay
inability to provide similar delay as in VoIP performance improvement of CMP to other
is due to the larger packet size of video schedulers as users increased from 5 to 60

.
Fig.5.Delay of (a) VoIP (b) Video and (c) IMS

Table 5:Delay performance comparison with respect to CMP for IMS, Video and VoIP.

IMS Delay Video Delay VoIP Delay

Users PF MLWDF PF MLWDF PF MLWDF

5 51.82% 45.78% 95.13% 37.14% 53.81% 50.26%
10 66.20% 75.34% 98.83% 15.85% 64.97% 76.20%
15 91.11% 82.52% 99.59% 8.12% 90.68% 81.73%
20 98.15% 81.39% 99.74% 3.65% 96.79% 72.02%
25 98.19% 70.85% 99.76% 1.30% 97.31% 54.59%
30 98.18% 56.16% 99.77% 0.06% 98.07% 47.13%
35 98.54% 51.39% 99.78% -0.73% 98.31% 45.02%
40 98.57% 45.02% 99.79% -1.59% 98.33% 35.45%
45 98.67% 38.39% 99.80% -1.96% 98.32% 24.44%
50 98.67% 30.96% 99.81% -2.00% 98.41% 16.07%
55 98.81% 24.90% 99.81% -1.77% 98.57% 10.35%
60 98.82% 19.21% 99.80% -1.21% 98.65% 7.16%

Fig. 6 shows the fairness index from the resource allocated to a user/flow-type
results. Fairness determines whether users defines its fairness in wireless
or applications are receiving a fair share of networks(Huaizhou Shi et al., 2014). The
system resources(Jain, R. K., Chiu, D. M. result showed CMP to largely outperform
W., & Hawe, 1984) and the amount of other schemes especially as users increases

Also available online at https://www.bayerojet.com 25

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
with an exception to video as seen in Fig. respectively.In Fig.6c, the IMS fairness
6b. The large packet size of video hinders index is evaluated. It was observed that the
CMP from scheduling it on time thereby CMP best improvement against the PF and
reducing the number of users served. EDF algorithm was at 35 users providing
Nonetheless it still performs slightly better 37.52% and 63.33% gain respectively and
than other schemes. In Fig.6a for VoIP, at 50 users for MLWDF providing 12.74%
CMP is seen to outperform other schemes gain. Table 6 shows the Fairness Index
in fairness when users are above 25 with a performance improvement of CMP to other
28.12%, 5.04% and 21.44% increase schedulers as users increased from 10 to 60.
against the PF, MLWDF and EDF

Fig.6. Fairness Index of (a) VoIP (b) Video and (c) IMS

Table 6:Fairness Index performance comparison with respect to CMP for IMS, Video and
VoIP.
IMS Video VoIP

Users PF MLWDF EDF PF MLWDF EDF PF MLWDF EDF

10 0.57% 0.03% 21.77% 44.32% 15.57% 51.94% -0.51% -0.66% 41.30%
15 5.39% 0.36% 34.92% 61.69% 8.52% 12.67% 4.70% -0.46% 38.34%
20 15.60% 1.03% 56.39% 20.70% 1.25% 18.59% 15.45% -0.35% 54.41%
25 24.85% 0.52% 45.36% 33.84% 8.09% 12.58% 28.28% 1.45% 62.33%
30 38.65% 7.67% 39.99% 49.41% 3.68% 9.38% 28.51% 3.10% 42.23%
35 37.52% 11.98% 63.33% 17.52% 3.92% -9.15% 31.79% 7.35% 54.49%
40 36.98% 11.62% 52.52% 12.82% 4.38% -11.24% 27.93% 4.65% 45.18%
45 32.22% 11.46% 40.76% 9.88% 3.84% -9.16% 28.12% 5.04% 21.44%
50 33.52% 12.74% 52.68% 23.31% 2.26% 10.34% 26.64% 4.92% 27.66%
55 30.31% 11.42% 42.03% 29.90% 1.21% 19.41% 23.49% 6.15% 29.18%
60 30.36% 9.73% 28.76% 36.36% 5.27% 28.91% 21.77% 5.07% 28.48%

The spectral efficiency which measures to provide the highest spectral efficiency
how efficient the spectrum or bandwidth is due to its reduction in packet loss and fast
utilized is shown in Fig.7. CMP can be seen scheduling as earlier discussed. It is

Also available online at https://www.bayerojet.com 26

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
followed by MLWDF, EDF and lastly PF affects their spectrum utilization.Table 7
scheduler. The large packet loss associated shows the spectral efficiency performance
with EDF (during congestion) and PF (due improvement of CMP to other schedulers as
to its inability to handle RT flow) largely users increased from 5 to 60.
Table 7:Spectral Efficiency performance comparison with respect to CMP.
Users PF MLWDF EDF
5 43.85% 6.96% -9.23%
10 73.43% 7.85% -3.50%
15 97.60% 7.32% 8.64%
20 86.04% 4.73% 6.94%
25 72.35% 2.52% 18.71%
30 62.21% 1.55% 23.09%
35 54.77% 0.89% 22.68%
40 49.71% 0.32% 18.47%
45 47.98% 1.14% 19.90%
50 45.38% 1.41% 15.78%
55 45.12% 2.29% 13.45%
60 48.96% 2.04% 9.86%

Fig.7. Spectral Efficiency

7.0 CONCLUSION
This paper has studied the scheduling of performance. Simulated results show CMP
packets in LTE downlink systems. A novel to provide better throughput, PLR
scheduling algorithm named CMP have andfairness in all studied flow types. Also,
been introduced to improve overall cell it provides the highest cell spectral
throughput (spectral efficiency) of the efficiency (approx. 0.24 at 10 users and
system. The CMP scheduler is a 0.19 at 60 users) when compared to other
modification of the EDF scheduling schedulers. As far as delay and fairness
algorithm with the aim of reducing its high index were concerned, the scheduler faired
packet loss thereby improving positively offering delay of approx. 0.07sec

Also available online at https://www.bayerojet.com 27

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
for video flows which had higher data and EDF schedulers.Future work would
packets and fairness indices of approx. consider more QoS metrics for LTE system
0.054 when the system was loaded with 60 and comparison with other schedulers such
users. The performance of CMP show it is as Exponential Proportional Fair,
more suitable for RT flows in a mixed Logarithm Rule, Earliest Based Deadline
traffic scenario compared to PF, MLWDF etc.

REFERENCES

Afroz, F., Barua, S., & Sandrasegaran, K. Comparative Study of Scheduling

(2014). PERFORMANCE Algorithms for LTE Networks. 8(3),
ANALYSIS OF FLS , EXP , LOG 467–472.
AND M-LWDF PACKET Furht, B., & Ahson, S. (2011).
SCHEDULING ALGORITHMS IN HSDPA/HSUPA handbook.
DOWNLINK 3GPP LTE SYSTEM. Huaizhou Shi, Prasad, R. V., Onur, E., &
International Journal of Wireless & Niemegeers, I. G. M. M. (2014).
Mobile Networks (IJWMN), 6(5), 77– Fairness in Wireless Networks:Issues,
91. Measures and Challenges. IEEE
Afroz, F., Heidery, R., Shehab, M., Communications Surveys & Tutorials,
Sandrasegaran, K., & Shompa, S. S. 16(1), 5–24.
(2015). COMPARATIVE Iacobucci, M. S. (2013). Reconfigurable
ANALYSIS OF DOWNLINK radio systems : network architectures
PACKET SCHEDULING and standards.
ALGORITHMS IN 3GPP LTE Iwamura, M., Umesh, A., & Hapsari, W. A.
NETWORKS. International Journal (2009). Further Enhancements of LTE
of Wireless & Mobile Networks LTE Release 9. NTT DOCOMO
(IJWMN), 7(3), 1–21. Technical Journal, 2(1).
Afroz, F., Sandrasegaran, K., & Ghosal, P. Jain, R. K., Chiu, D. M. W., & Hawe, W. R.
(2014). Performance Analysis of PF, (1984). A quantitative measure of
M-LWDF and EXP/PF Packet fairness and discrimination. Eastern
Scheduling Algorithms in 3GPP LTE Research Laboratory, Digital
Downlink. Australasian Equipment Corporation, pp. 2–7.
Telecommunication Networks and Jalali, A., Padovani, R., & Pankaj, R.
Applications Conference (ATNAC), (2000). Data throughput of CDMA-
87–92. HDR a high efficiency-high data rate
Andrews, M., Kumaran, K., Ramanan, K., personal communication wireless
Stolyar, A., Whiting, P., & system. VTC2000-Spring. 2000 IEEE
Vijayakumar, R. (2001). Providing 51st Vehicular Technology
quality of service over a shared Conference Proceedings (Cat.
wireless link. IEEE Communications No.00CH37026), 3, 1854–1858.
Magazine, 39(2), 150–153. IEEE.
Anritsu. (2015). LTE Resource Guide. Kawamura, T., Kishiyama, Y., Kakishima,
Basukala, R., Ramli, H. A. M., & Y., & Yasukawa, S. (2012). LTE-
Sandrasegaran, K. (2009). Advanced—Evolution of LTE—
Performance Analysis of EXP/PF and Radio Transmission Experiments.
M-LWDF in Downlink 3GPP LTE NTT DOCOMO Technical Journal,
System. AH-ICI 2009, 1–5. 14(2).
Dardouri, S., & Bouallegue, R. (2014). Kurose, J. F., & Ross, K. W. (2008).

Also available online at https://www.bayerojet.com 28

 ISSN: 2449 – 0539
BAYERO JOURNAL OF ENGINEERING AND TECHNOLOGY (BJET) VOL.15 NO.1, JANUARY, 2020
pp15-29
Computer networking : A Top-down Sheta, N., Zaki, F. W., & Keshk, S. (2013).
Approach. Packet scheduling in LTE mobile
Müller, M. K., Schwarz, S., & Rupp, M. network. International Journal of
(2013). QoS Investigation of Scientific & Engineering Research,
Proportional Fair Scheduling in LTE 4(6).
Networks. IEEE. Srikanth S, Kumaran V, & Manikandan C.
Nwawelu, U. ., Anil, C. ., & Ahaneku, M. . (2006). Orthogonal Frequency
(2017). COMPARATIVE Division Multiple Access : Is it the
ANALYSIS OF THE Multiple Access System of the Future?
PERFORMANCE OF RESOURCE AU-KBC Research Center, Anna
ALLOCATION ALGORITHMS IN University, India.
LONG TERM EVOLUTION Stankovic, J. A., Spuri, M., Ramamritham,
NETWORKS. Nigerian Journal of K., & Buttazzo, G. C. (1998).
Technology (NIJOTECH), 36(1), 163– Deadline Scheduling for Real-Time
171. Systems : EDF and Related
Piro, G., Grieco, L. A., Boggia, G., Algorithms. Springer US.
Capozzi, F., & Camarda, P. (2011). Sulthana, S. F., & Nakkeeran, R. (2014).
Simulating LTE Cellular Systems : an Performance Evaluation of Downlink
Open Source Framework. IEEE Packet Scheduling Algorithms in
Transactions on Vehicular LTE. Proceeding of International
Technology, 60(2), 498–513. Conference on Advances in
Radhakrishnan, S., Neduncheliyan, S., & Communication Networks and
Thyagharajan, K. K. (2016). A Review Computing, CNC, 38–44.
of Downlink Packet Scheduling Techplayon. (2017). LTE Frame Structure.
Algorithms for Real Time Traffic in Retrieved September 12, 2018, from
LTE-Advanced Networks. Indian http://www.techplayon.com/wp-
Journal of Science and Technology, content/uploads/2018/04/LTE-Frame-
9(4). 1.png
Saleh, M., & Dong, L. (2010). Comparing Xian, Y. J., Tian, F. C., Xu, C. B., & Yang,
FCFS & EDF scheduling algorithms Y. (2011). Analysis of M-LWDF
for real-time packet switching fairness and an enhanced M-LWDF
networks. 2010 International packet scheduling mechanism.
Conference on Networking, Sensing Journal of China Universities of Posts
and Control, ICNSC 2010, 698–703.
and Telecommunications, 18(4), 82–88.

Also available online at https://www.bayerojet.com 29