See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/308417643

Delay and Jitter Analysis in LTE Networks

Conference Paper · October 2016

DOI: 10.1109/WINCOM.2016.7777202

CITATIONS READS

2 156

2 authors, including:

Hamza Dahmouni
Institut National des Postes et Télécommunications
23 PUBLICATIONS 141 CITATIONS

SEE PROFILE

All content following this page was uploaded by Hamza Dahmouni on 17 October 2017.

The user has requested enhancement of the downloaded file.

 Delay and Jitter Analysis in LTE Networks
Nabil MESBAHI* Hamza DAHMOUNI
Institut National des Postes et Teleommunications Institut National des Postes et Teleommunications
INPT, 2, av. Allal El Fassi, Madinat Al Irfane, INPT, 2, av. Allal El Fassi, Madinat Al Irfane,
10000, Rabat, Morocco 10000, Rabat, Morocco
Email: [email protected] Email: [email protected]

Abstract—The emergence of new multimedia applications and interpreted as the requirements of IP transport network which
real-time services have necessitated networks with guaranteed provides the underlying path.
quality of service (QoS). For mobile networks, LTE is expected
to provide low-latency and capacity improvement for interactive Obviously, QoS requirements increase during the period
and real-time applications which have critical delay and jitter of congestion, the most sensitive applications at this period
requirements. Indeed, the problem of measuring the end-to-end are interactive real-time services such as VoLTE and stream-
network performance over LTE networks is more difficult and
largely open. The topic of this paper is to highlight the behavior
ing in which the restrictive performance indicators are the
of end-to-end (e2e) delay and jitter under increasing load, in e2e delay and jitter [2]. In fact, even if it is noticed that
particular, we focus on the impact of the LTE Backhaul and LTE radio access reduces the latency [3] compared to the
Backbone networks on these metrics. Furthermore, we address previous cellular technologies, in reality it does not employ
some QoS requirements that should be taken into account for fixed delay because of fast retransmissions used to repair
LTE-EPC planning and design. The main finding is that the QoS
requirements depend heavily on the network configuration and
erroneous transmissions [4]. In this case, LTE radio access
geographical size, and we have concluded that a proper QoS network transmissions can introduce jitter which has a negative
treatment have to be done at each network element in order to impact on the quality of real-time applications more than
guarantee a successful e2e QoS. the delay. Moreover, Backhaul network is becoming even
Keywords– LTE, QCI, Traffic routing, Backhaul, e2e Delay, more complex by the expanding adoption of full IP-based
Jitter, QoS, Performance evaluation. technologies and the exponential growth of traffic demand [5].
I. I NTRODUCTION Operators and service providers have less direct control over
the Backhaul, and find it more difficult to gain visibility into
Today providing quality of service (QoS) in a sustainable it because several components do not understand QCI values.
manner is widely recognized. The support of a variety of Consequently, LTE IP network must scale smoothly to support
applications and services with different QoS requirements has increasing traffic volume and complex heterogeneous services
become the main challenge in the design of mobile network especially delay and jitter sensitive applications, which require
operators. This has been the fundamental premise behind the real-time QoS-based optimization.
development of Long Term Evolution (LTE) technology which
is designed to support higher traffic volumes, and driving the In order to meet such targets, it is important to understand
shift in Backhaul architectures to an all IP model. the impact of each LTE component on the delay and jitter
LTE is an end-to-end IP network which provides IP con- which are considered as the most restrictive indicators for
nections from the terminal to the core network. QoS implies real time applications. Our objective is to investigate the
services to be differentiated based on the QoS Class Identifier performance of real-time services by examining the behavior
(QCI) which determines the level of priority of each service of the e2e delay and jitter for different load levels. The main
class and specifies the maximum one way allowed values result is that there is a complex relationship between jitter and
in terms of delay, jitter and packet loss [1]. Moreover, the delay which behave in different manner. Also, we show that
QCI identifiers aim to determine the user plane treatment delay and jitter depend heavily on the network configuration,
for IP packets transported on a bearer and its corresponding geographical size and how much delay is consumed by the
type: GBR (dedicated) and non-GBR (default). However, the network components. We have concluded that operators must
maximum values allowed for delay, jitter and packet loss manage traffic more actively to meet requirement in terms of
associated with each QCI type are applicable, as specified delay and jitter for real-time applications.
in 3GPP TS 23.20, only between mobile components from The rest of the paper is organized as follows: In Section
mobile infrastructure perspective. In fact, the application of II we present a detailed overview on related work. Section
these values do not go down into IP Backhaul and back- III contains a description of our analytical model of jitter.
bone networks that each application may traverse for e2e In Section IV, we present a simulation based-study on the
communication because, routers or switches, however, do not behavior of delay and jitter under increasing load and the
understand 3GPP QCI values. These values are not directly impact of LTE network elements on the e2e QoS of real time
978-1-5090-3837-4/16/$31.00
c 2016 IEEE services. Finally, section V draws some conclusions.
 II. R ELATED WORK this purpose, we proposed in [16] an analytical expression
A. Delay and jitter in IP networks for the e2e delay and jitter incurred by Poisson traffic as
a function of the traffic load, bandwidth and latency. This
Numerous studies have been presented in the literature
model is used in [17] as a metric to solve the optimal routing
addressing various QoS aspects in IP networks. Few of them
scheme for traffics sensitive to delay or jitter. Results show that
have been devoted to the study of delay and jitter. In this
the optimal routing for jitter-constrained and delay-constrained
context, an analytical model of end-to-end delay and jitter
flows is different and depends heavily on which one of the two
of constant bit rate (CBR) traffic produced by a single node
parameters is considered. This issue is evaluated by proposing
was proposed in [6]. The model was extended for the multi-
in [18] an algorithm for the optimization of traffic flows in
node case and the accuracy was assessed by simulation. The
MPLS networks. This algorithm aims to provide the optimal
paper [7] proposes a control scheme which aims to reduce the
routing scheme for a traffic sensitive to both delay and jitter.
buffer space in order to provide bounds on jitter for real-time
channels in packet switched networks. Authors of [8] propose III. A NALYTICAL J ITTER F ORMULATION FOR P OISSON
a scheduling algorithm which aims to guarantee low delay TRAFFIC
and jitter. The proposed model was assessed by simulation
In this section, we introduce the delay and jitter model
Pn de-
and results show that the end-to-end delay and jitter produced
veloped in our previous work [16]. We denote by λ = i=1 λi
along a path is significantly reduced and bounded by a number
the total arrival rate and µ the service time. We then have
of Inter-Departure Time.
ρ = µλ and η = λ − µ. The jitter of a tagged flow k produced
B. Delay and jitter evaluation in LTE networks by a single node can be approximated by:
LTE is an emerging technology in which the Quality of
  
1 −η η −η

Service (QoS) provisioning is crucial for providing a range Jk ≈ 1−e λk

+e λk
(1)
η λk
of IP-based services [9]. Several studies have been conducted
in order to evaluate multi-media applications over LTE net- For the multi-node case, we consider that the tagged flow k
works [10], [11]. To this end, [12] analyzes the impact of goes through N tandem nodes. We add a superscript n to all
channel quality indicator (CQI) on the performance of a the parameters of the previous section in order to indicate the
scheduling algorithm serving video streaming. Results show node.
(N )
that a minimum rate requirement of aperiodic CQI reports is Let Jk be the average packet jitter through N tandem
(n)
recommended to meet the QoS requirement. Moreover, authors nodes and Tj be the delay of packet j through node n. The
of [13] show that prioritization of the VolP service has not a path jitter of the tagged flow k is given by
large impact on the quality of other services due to small " N #
(N )
X (n) (n)
VoIP packet sizes but allows more efficient radio resource Jk = E | Tj+1 − Tj | (2)
utilization. n=1
LTE architecture is designed to support high data traffic
From Eq.2 we can see that:
and a guaranteed QoS to e2e IP based service. However
N
the increasing volume of traffic leads to an increasing of its (N )
X (n)
complexity. In this case, the traffic complexity and delay and Jk ≤ Jk (3)
jitter sensitive services can force the Backhaul network to n=1

become the performance bottleneck in LTE networks [14]. At low load, the queues become independent. In this case, the
The authors of [5] analyze the impact of Backhaul sub-frame absolute value of the sum in Eq.2 can be approximated by the
misalignment within the framework of LTE-Advanced and sum of the absolute values [16].
they show that if directional antennas are available for Back- N
(N ) (n)
X
hauling, misalignment does not degrade system performance. Jk = Jk (4)
Also, [15] proposes a model for power consumption where n=1
the Backhaul is taken into account. The paper shows that the
However, at high load the queues become strongly coupled
impact of Backhaul on the total power consumption can be
and can not be approximated by the sum of absolute values.
significant.
We have shown in our previous work [16] that:
C. Our previous work on jitter evaluation N
(N ) (n)
X
Network traffic prediction and characterization has become Jk = K (n) Jk (5)
an increasingly task due to its importance in network perfor- n=1
mance and scheduling. An accurate modeling of traffic must
K (n) represents the auto-correlation function given by [16]
take into account several factors related to the network struc-
(n)
ture and traffic parameters. Poisson is among the models which 1 λk (R(n) )2
K (n) = .[(L(n) 2
) + ] (6)
is widely used for the characterization of real IP traffic because (L(n) )2 + (R(n) )2 (n)
λk + µ/(R(n) )2
it can provide helpful insights to compare and understand
ρ(n) ρ(n)
the behavior of some network performance parameters. For Where L(n) = 1−ρ(n)
and (R(n) )2 = (1−ρ(n) )2
 UE 0.7
UE

0.6

eNode−b eGW 0.5

Server

Delay and jitter (s)

0.4
Fig. 1: NS2 - LTE network architecture
0.3
Jitter
Delay
0.2
IV. D ELAY AND JITTER EVALUATION IN LTE NETWORKS
A. Delay and jitter evaluation under increasing load
0.1
QoS evaluation of real time services under increasing load
is a real scenario in an all-IP network architecture designed to 0
support high data traffic and a wide range of applications. The 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
aim for any real time application is to utilize the low delay and Traffic load
jitter features available within LTE network in order to ensure
Fig. 2: Delay and jitter as a function of load
an improvement over the previous standards available on the
2G and 3G networks. However, the increasing of traffic volume
in these networks affects negatively the service delivered to the
user. LTE behavior is usually variable due to several random be positioned to support the introduction of LTE and the
events including busy hours and high profile events. overall network evolution in order to avoid becoming the
In this section, we investigate the behavior of delay and jitter performance bottleneck.
under increasing load. The traffic is modeled as a Poisson pro-
cess whose inter-arrival and service time follow an exponential B. Impact of access, Backhaul and core networks
distribution. The simulation is carried out by simplifying the LTE is an all IP based traffic network. All the interfaces
LTE network architecture instead of considering a complex between different LTE network elements are IP transport
mix of component and technologies. To this end, we consider links. However, data packets can be transported and tunneled
the architecture presented in Fig.1 which consists of a finite over other links and protocols including IP over ATM and
number of IP components and LTE interfaces and links. IP over MPLS. But, from an LTE viewpoint, we are always
It is clear that the increasing of load affects negatively the talking about IP links. Currently, LTE is becoming the
network performance especially the delay, packet loss and global network technology choice. The optimization of these
throughput. However, this behavior is no longer true for the networks is an essential element in order to make the LTE
jitter. In fact, there is a complex relationship between jitter and network operable in any environment. In fact, the Backhaul is
delay which behave differently under increasing load. Results a critical part of LTE network because it can be built on a mix
of Fig.2, show that the jitter decreases as the load increases and of wired and wireless network technologies and equipments
delay increases under increasing load. This issue is explained which transport data and signaling traffic. Moreover, and
in [16] as follows : “at low load, the queue becomes nearly from IP transport network viewpoint, QoS is achieved by
empty and arriving packets do not have to wait. In this case, queuing and scheduling which perform the prioritization of
the delay depends on the inter-arrival process and the time service classes in order to guarantee a certain required QoS
in the queue. However, at high load, the queue becomes large level. However, even if QoS, in LTE IP transport network, is
and all arriving packets have to wait in the queue. In this managed through a Diff-Serv model, each node can function
case, the inter-arrival process does not contribute much to the independently because routers or switches do not understand
sojourn time. Consequently, the variation in the delay is due 3GPP QCI values. This raises the following questions.
to the service time only, and it is not unreasonable to expect
that the jitter will be smaller in this case”. What is the effect of LTE IP transport network on e2e
The integration of all applications over a simplified and delay and jitter? and how can the operators manage their
common architecture have some challenges. In fact, the in- network in order to implement an efficient QoS policy?
crease of traffic volume leads to an increasing of its complexity
through which operators have to monitor and manage the Once we have answered this question by investigating the
traffic flows according to the type of applications. Moreover, behavior of path delay and jitter. Fig. 3 and 4 show the delay
operators have not only to tackle with a staggering increase in and jitter budget calculation at each LTE network element.
traffic volumes, but also they have to optimize their Backhaul We observe that the average packet delay and jitter incurred
and Backbone. It is important to ensure that Backhaul should in the Down-link air interface represent both nearly 30% of
 3
0.12 End−to−end
End−To−End Server to eGW
2.5
Server to eGW 0.1
eGW to eNode−b
eGW to eNode−b eNode−b to UE
2 eNode−b to UE
0.08

Jitter (s)
Delay (s)

1.5
0.06

1
0.04

0.5 0.02

0 0
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9
Traffic load Traffic load

Fig. 3: Delay incurred at each LTE network component Fig. 4: Jitter incurred at each LTE network component

Tagged flow
(Poisson)
the average e2e delay and jitter. This result is in harmony
1 2 5
with the VoIP service in which the access radio network must
guarantee a maximum one-way delay of 50 ms [19] while the Background flow
lower recommended value for e2e delay is 150ms. However, 30% of the link capacity

the nature of the wireless medium is unreliable and can be Fig. 5: Scenario 1, network of 5 links with background traffic
subject to errors due to the unpredictability of the environment
and propagation conditions. The performance focus for the
deployment of an efficient access wireless network is a com-
plex task because operators have less direct control over the edges of the second network architecture is done trough six
radio access network and find it more difficult to gain visibility (6) links. All the links have the same capacity and only the
into it. Consequently, a comprehension of the backbone and first network model of Fig. 5 is loaded at 30% of the capacity
Backhaul networks comportment and interdependent behavior of their links. Our goal is to show the difference between the
towards the delay and jitter is a key step in understanding how delay and jitter produced by a routing through a minimum
to optimize or even predict the LTE networks performance number of loaded links toward a high number of unloaded
properties. links.
We have shown in the previous section that the behavior We have plotted the delay and the jitter as a function of load
of delay and jitter is different under increasing load. The on Fig. 7. Results show that the difference between the delays
jitter decreases as the load increases and delay increases under incurred in the first and second scenarios is not important. The
increasing load. Moreover, Fig. 4 shows that jitter increases behavior of the e2e delay is not complex because the packet
as the number of nodes increases and that this effect is more delays through a queue are strongly correlated with each other
important at low load. Our results are obtained by considering and equal to the sum of delays incurred at each node. If an
a simple network architecture but in reality, the Backhaul and application is sensitive only to delay, this effect can be reduced
Backbone networks consist of a mix of components which by multiplexing the data packet into several flows. However,
lead to an increasing of the number of links trough which in reality, there is a dependence between the service time and
data packets can traverse before arriving at the radio access the arrival time at the following router, which results a strong
network. These networks may vary according to operators and correlation between queues in tandem. This correlation have
depending on EPC locations and geographical size. a negative impact on the e2e jitter which is less than the sum

C. Impact of Network Size

Tagged flow
In order to understand the impact of EPC size, we consider (Poisson)
1 2 6
the two IP network architectures shown in Fig. 5 and 6
representing the Backbone and Backhaul networks. The source Fig. 6: Scenario 2, network of 6 links and without background
and the destination of the first network model (Fig.5) are traffic
connected through five (5) links while the connection of the
 related to the concavity of the jitter and the convexity of delay
0.6 as a function of flow. We have then exterminated the impact
of Backbone and Backhaul networks on the performance of
real time applications. The main finding is that the QoS
0.5
requirements depend heavily on the network configuration and
geographical size. We have concluded that operators have to
0.4 create a mapping policy between 3GPP QCI and IP QoS
Delay (s)

values in order to define properly the optimal routing scheme

Delay of scenario 1 according to the QoS requirements of applications.
0.3
Delay of scenario 2
R EFERENCES
Jitter of scenario 1
0.2 Jitter of scenario 2 [1] 3GPP, “Quality of service (QoS) concept and architecture,” TS 23.107.
[2] J. Puttonen, H.-H. Puupponen, K. Aho, T. Henttonen, and M. Moisio,
“Impact of control channel limitations on the LTE VoIP capacity,”
0.1 International Conference on Networks, Menuires, France, 2010.
[3] 3GPP, “Digital cellular telecommunications system (Phase 2+), Uni-
versal Mobile Telecommunications System (UMTS), LTE, services and
service capabilities,” TS 22.105.
0
0 0.1 0.2 0.3 0.4 0.5 [4] ——, “Universal mobile telecommunications system (UMTS); LTE; IP
Traffic load Multimedia Subsystem (IMS); Multimedia telephony; Media handling
and interaction,” TS 26.114.
[5] O. Bulakci, A. S. Nedelcu, A. B. Saleh, S. Redana, and J. Hmlinen, “Im-
Fig. 7: Delay and jitter incurred by the scenarios 1 and 2 pact of backhaul subframe misalignment on uplink system performance
of LTE-advanced relay networks,” Vehicular Technology Conference
(VTC Fall), 2012.
[6] O. Brun, C. Bockstal, and J.-M. Garcia, “Analytical approximation of
of link jitters as shown in Eq. 3 and Fig. 4. We can see, in the jitter incurred by CBR traffics in IP networks,” Telecommunication
Systems, vol. 33, 2006.
Fig. 7, that the jitter incurred by the first scenario is much [7] D. Verma and D. F. H. Zhang, “Delay jitter control for real-time
lower than the jitter at the second scenario and this becomes communication in a packet switching network,” IEEE TRICOMM91,
more important when the traffic rate is low. The e2e jitter pp. 35–43, April 1991.
[8] T. Szymansky, “A low jitter guaranteed rate scheduling algorithm for
increases as the number of nodes increases and decreases packet switched IP routers,” IEE Transactions on Communications,
under increasing load. Consequently, the QoS requirements 2009.
of real time applications depend also on the routing scheme [9] P. Mogensen, W. Na, and I. Z. Kovacs, “LTE capacity compared to
the shannon bound,” IEEE 65th Vehicular Technology Conference -
which must define the optimal path which take into account VTC2007-Spring, pp. 1234–1238, 2007.
the complex behavior between delay and jitter. [10] T. Aziz, M. Masum, M. Babu, S. Rahman, and J. Nordberg, “Mobility
We can conclude that the delay and especially jitter, depend Impact on the End-to-End Delay Performance for VoIP over LTE,”
International Conference on Communication Technology and System
heavily on the network configuration, geographical size and Design, vol. 30, pp. 491–498, 2012.
may vary depending on the routing scheme and how delay [11] H. H. Soliman, H. M. El-Bakry, and M. Reda, “Real-time transmission
budget is consumed by LTE network elements in reality. Any of video streaming over computer networks,” 11th WSEAS Int. Conf.
Electronics, Hardware, Wireless and Optical communications (EHAC
impact on the performance of a network component can have ’12), no. 51-62, 2012.
an impact on the overall network performance. [12] R. Basukala, H. A. M. Ramli, K. Sandrasegaran, and L. Chen, “Impact
A simple answer of the previous questions is that a proper of CQI feedback rate/delay on scheduling video streaming services in
L TE downlink,” Communication Technology (ICCT), 2010 12th IEEE
QoS treatment have to be done at each network element International Conference, pp. 1349 – 1352, 2010.
in order to guarantee a successful e2e QoS for real time [13] I. Siomina and S. Wanstedt, “The impact of QoS support on the end
applications. In fact, a definition of a QoS policy on every node user satisfaction in LTE networks with mixed traffic,” International
Symposium on Personal and Mobile Radio Communications, 2008.
is the key element to guarantee the QoS requirement for each [14] O. Bulakci, A. B. Saleh, S. Redana, B. Raaf, and J. Hamalainen, “Flex-
defined class. For this purpose, it is desirable that operators ible backhaul resource sharing and uplink power control optimization in
perform a mapping between 3GPP QCI and IP QoS values LTE-advanced relay networks,” Vehicular Technology Conference, 2011.
[15] S. Tombaz, P. Monti, K. Wang, A. Vastberg, M. Forzati, and J. Zan-
(DSCP for native IP or EXP for MPLS) so that every network der, “Impact of backhauling power consumption on the deployment
element can know the QoS requirements of each application of heterogeneous mobile networks,” IEEE Global Telecommunications
in order to define the optimal routing scheme. Conference (GLOBECOM), 2011.
[16] H. Dahmouni, A. Girard, and B. Sanso, “An analytical model for jitter
in ip networks,” Annals of Telecommunications, published on-line by
V. C ONCLUSION Springer, 2011.
In this paper, we have introduced the jitter model developed [17] H. Dahmouni, A. Girard, M. Ouzineb, and B. Sanso, “The impact of
jitter on traffic flow optimization in communication networks,” IEEE
in our previous work which aims to produce a simple formula Transactions on Network and Service Management, pp. 279–292, 2012.
for the e2e delay-jitter for a network of M/M/1 queues. [18] N. Mesbahi and H. Dahmouni, “An efficient algorithm for traffic flow
Through a simulation based-study, we have investigated the optimization in mpls networks,” International Conference on Protocol
Engineering (ICPE) and International Conference on New Technologies
behavior of delay and jitter in LTE network under increasing of Distributed Systems (NTDS), pp. 1 – 6, 2015.
load. The main result is that, under increasing load, the be- [19] M. Poikselk, H. Holma, J. Hongisto, J. Kallio, and A. Toskala, “Voice
havior of delay and jitter in LTE networks is different. This is over LTE (VoLTE),” Wiley, 2012.

View publication stats