04 Voice Over LTE

Voice over LTE
PRASANNA GURURAJ RAGHAVENDRARAO

Master of Science Thesis
Wireless and Mobile Communications Group Department of Telecommunications Faculty of Electrical Engineering, Mathematics and Computer Science Delft University of Technology
II
Prasanna Gururaj Raghavendrarao
III
Voice over LTE

For the degree of Master of Science in Wireless and Mobile Communications Group (WMC) at Department of Telecommunications at Delft University of Technology
Prasanna Gururaj Raghavendrarao 29.6.2012
Faculty of Electrical Engineering, Mathematics and Computer Science Delft University of Technology Delft, The Netherlands
IV
Delft University of Technology Department of Telecommunications
The undersigned hereby certify that they have read and recommend to the Faculty of Electrical Engineering, Mathematics and Computer Science for acceptance a thesis entitled Voice over LTE by Prasanna Gururaj Raghavendrarao in partial fulllment of the requirements for the degree of Master of Science.
Dated: 29.6.2012 Supervisors: dr.ir. Jos Adema (KPN)
ir. Gerard Fossung (KPN)
dr. R.R. Venkatesha Prasad Readers: dr.ir.Jos Weber
dr.ir. Bert Jan Kooij
Abstract
Long Term Evolution (LTE) is the latest high speed mobile broadband technology that is gaining widespread attention due to its high data rates and improved Quality of Service (QoS). Initially, LTE was seen as a technology for supporting high speed data, but there is a growing interest in the industry to support voice over LTE. The support of voice over LTE has lot of challenges owing to the fact that both voice and data trac are to be carried over the same radio and core networks. The optimum usage of resources in the radio network is of high importance as there is a growing need to improve the capacity at reduced cost. The transport network is another key area that needs to be carefully planned according to the capacity of the radio network. Dierentiation and scheduling of resources in the transport network plays a key role in guaranteeing good end to end performance for both voice and data services. In this thesis, the impact of dierentiation and scheduling of resources in the transport network on the end to end performance of voice over LTE is investigated. The results indicate that without proper prioritization and scheduling of resources in the transport network , the performance of voice is severely aected when the transport network is congested with data trac. To overcome this scenario, we prioritize voice over data trac and analyse its performance for dierent transport network scheduling algorithms. From the results, it is clear that with proper classication and scheduling of resources in the transport network, signicant increase in voice capacity is observed. On the other hand, by totally prioritizing voice, performance of the data trac is aected to a large extent. Hence, to achieve a balance, voice users are classied into dierent priority levels and the performance of voice and data in this scenario is investigated. The analysis for all these scenarios are based on simulations using OPNET simulation tool.
ii
Table of Contents
Acknowledgements 1 Introduction 1-1 Solutions for Supporting Voice over LTE . . . . . . . . . . . . . . . . . . 1-1-1 1-1-2 1-2 1-3 1-4 1-5 Circuit Switch(CS) fallback . . . . . . . . . . . . . . . . . . . . . Voice over LTE via IP Multimedia Subsystem (VoLTE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xi 1 1 1 2 3 4 5 5 7 7 7 9 11 12 13 13 14 14 15 15 15 15
Motivation for the Thesis Problem Denition . . . . Related Work . . . . . . Organization of the Thesis
2 Background 2-1 Introduction . . . . . . . . . . . . 2-2 LTE Network Architecture . . . . . 2-3 QoS Architecture in LTE . . . . . . 2-4 IMS Network Architecture . . . . . 2-5 Dierentiated Services Architecture 2-6 Scheduling Strategies . . . . . . . 2-6-1 Strict Priority Scheduling . 2-6-2 2-6-3
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
Weighted Round Robin (WRR) Scheduling . . . . . . . . . . . . . Weighted Fair Scheduling . . . . . . . . . . . . . . . . . . . . . .
3 Simulation Model 3-1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2 Introduction to OPNET Modeller . . . . . . . . . . . . . . . . . . . . . . 3-2-1 Overview of LTE Model in OPNET . . . . . . . . . . . . . . . . .
iv
Table of Contents
3-2-2 Issues in LTE model . . . . . . . . . 3-3 Changes in LTE model . . . . . . . . . . . . 3-3-1 LTE S1 process model In E-Node B . 3-3-2 LTE S1 NAS Process model in EPC . 3-3-3 GTP Process model in E-Node B and 3-4 Simulation Environment . . . . . . . . . . . 3-4-1 Mobile Node . . . . . . . . . . . . . 3-4-2 E-Node B . . . . . . . . . . . . . . 3-4-3 IMS Model . . . . . . . . . . . . . . 3-4-4 Application Conguration . . . . . .
. . . . . . . . . . . . EPC . . . . . . . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
18 18 18 20 20 22 22 23 25 26 27 27 27 28 28 29 31 31 31 35 36 37 37 37 40 42 43 43 43 48
4 Results 4-1 Introduction . . . . . . . . . . . . . . . . . . 4-2 QoS parameters for Voice . . . . . . . . . . . 4-3 Scenario 1 . . . . . . . . . . . . . . . . . . . 4-3-1 Description of the Scenario . . . . . . 4-3-2 Analysis of results . . . . . . . . . . . 4-4 Scenario 2 . . . . . . . . . . . . . . . . . . . 4-4-1 Description of the Scenario . . . . . . 4-4-2 Analysis of results . . . . . . . . . . . 4-4-3 Impact on FTP trac . . . . . . . . . 4-4-4 Summary of the Results for Scenario 2 4-5 Scenario 3 . . . . . . . . . . . . . . . . . . . 4-5-1 Description of the Scenario . . . . . . 4-5-2 Analysis of results . . . . . . . . . . . 4-5-3 Impact on FTP trac . . . . . . . . . 4-5-4 Summary of results for Scenario 3 . . . 4-6 Scenario 4 . . . . . . . . . . . . . . . . . . . 4-6-1 Description of the scenario . . . . . . . 4-6-2 Analysis of Results . . . . . . . . . . . 4-7 Comparison of Scenarios . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
5 Technical Details of Voice over LTE via IMS based solution - An Operator Perspective 5-1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2 VoLTE Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3 Options for integrating LTE with existing CS/PS networks . . . . . . . . . 5-3-1 5-3-2 Independent PS based solution . . . . . . . . . . . . . . . . . . . Enhanced Single Radio Voice call continuity (SRVCC) / IMS Centralized Services (ICS) . . . . . . . . . . . . . . . . . . . . . . . .
49 49 49 51 51 53
Table of Contents
6 Conclusion 6-1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Glossary List of Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55 55 56 59 59 60
vi
Table of Contents
List of Figures
1-1 Circuit Switch Fallback [2] . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2 VoLTE [2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-1 LTE Network Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2 QoS Architecture in LTE [8] . . . . . . . . . . . . . . . . . . . . . . . . . 2-3 IMS Network Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1 LTE Network Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2 Data Flow in LTE Network [13] . . . . . . . . . . . . . . . . . . . . . . . 3-3 3-4 3-5 3-6 3-7 3-8 3-9 3-10 3-11 3-12 4-1 4-2 4-3 4-4 4-5 4-6 Protocol Architecture . . . . . . . . . . . . . . . . . . . GTP Encapsulated Packet . . . . . . . . . . . . . . . . . E-Node B Node Model . . . . . . . . . . . . . . . . . . lte _ s1 Process model in E-Node B . . . . . . . . . . . Node Model in EPC . . . . . . . . . . . . . . . . . . . . lte _ s1 _ nas Process model in EPC . . . . . . . . . . . GTP Process Model . . . . . . . . . . . . . . . . . . . . LTE Simulation Network . . . . . . . . . . . . . . . . . . Mobile Conguration . . . . . . . . . . . . . . . . . . . . IMS Proxy session control function conguration attribute Voice Packet End to End Delay vs No. of VoIP Packet Delay Variation Vs No. of VoIP users . End to End Delay Vs No. of VoIP Users . . . S1 Delay Vs No. of VoIP Users . . . . . . . . PDV vs No. of VoIP users . . . . . . . . . . . Packet Loss Rate vs No. of VoIP users . . . . users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 3 8 9 11 16 16 17 17 19 19 20 21 22 23 24 25 29 30 32 32 33 34
viii
List of Figures
4-7 4-8 4-9 4-10 4-11 4-12 4-13 4-14 4-15 4-16 4-17 4-18 4-19 4-20 4-21 4-22
Mean Opinion Score vs No. of VoIP users . . . . . . . FTP Transfer time vs No. of VoIP users . . . . . . . FTP Throughput vs No. of VoIP users . . . . . . . . End to End Delay vs No. of VoIP users . . . . . . . . S1 delay vs No. of VoIP users . . . . . . . . . . . . . Packet delay variation vs No. of VoIP users . . . . . . Packet Loss Rate vs No. of VoIP users . . . . . . . . Mean Opinion Score vs No. of VoIP users . . . . . . . Mean FTP transfer time vs No. of VoIP users . . . . FTP Throughput vs No. of VoIP users . . . . . . . . Packet end to end delay for high priority VoIP users . Packet end to end delay for normal priority VoIP users Packet delay variation . . . . . . . . . . . . . . . . . Packet Loss Rate . . . . . . . . . . . . . . . . . . . . Mean Opinion Score . . . . . . . . . . . . . . . . . . FTP Transfer time . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
34 35 36 38 38 39 40 40 41 42 44 44 45 46 46 47 50 52 53
5-1 VoLTE Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2 LTE-3G Integrated architecture . . . . . . . . . . . . . . . . . . . . . . . 5-3 SRVCC/ICS Architecture . . . . . . . . . . . . . . . . . . . . . . . . . .
List of Tables
2-1 EPS QOS Bearer Denitions [9] . . . . . . . . . . . . . . . . . . . . . . . 2-2 Assured Forwarding Drop Precedence Classication . . . . . . . . . . . . . 3-1 EPS Bearer to DSCP Mapping . . . . . . . . . . . . . . . . . . . . . . . 3-2 E-Node B Conguration Parameters . . . . . . . . . . . . . . . . . . . . . 3-3 VoIP Conguration Parameters . . . . . . . . . . . . . . . . . . . . . . . 4-1 MOS satisfaction level . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-2 Scenario Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-3 Number of satised VoIP users . . . . . . . . . . . . . . . . . . . . . . . 5-1 VoLTE Relevant Interfaces and Protocols . . . . . . . . . . . . . . . . . .
10 13 23 24 26 28 43 48 50
List of Tables
Acknowledgements
First, I would like to thanks my supervisors at KPN Dr.ir. Jos Adema and ir. Gerrard Fossung for their valuable inputs and suggestions during the writing of this thesis. I would also like to thank Perry Jackson for giving me an opportunity to do this thesis at KPN and the Mobile Innovation Voice Team for their kind support and encouragement during the course of this thesis. Next, I would like to thank my supervisor Dr. R.R. Venkatesha Prasad for his invaluable support and constant encouragement during the writing of this thesis. Last but not the least, I would like to thank my family and friends for their continuous love and support.
Delft 29.6.2012
xii
Acknowledgements
Chapter 1 Introduction
Long Term Evolution (LTE) is a fourth generation technology which is standardized in the Release 8 specications by the 3GPP. It is capable of providing high data rates (100 Mbps in downlink and 50 Mbps in uplink) as well as support high speed mobility. It has a completely packet switched core network architecture unlike its predecessor UMTS which is capable of supporting both the Circuit Switched (CS) as well as Packet Switched (PS) core networks.
1-1
Solutions for Supporting Voice over LTE
The absence of CS domain in the LTE network has led the industry and standardization bodies like the 3GPP to propose various solutions to support voice in the LTE network. The two most important among them widely being considered for deployment are as follows: 1-1-1 Circuit Switch(CS) fallback
The Circuit Switch fallback solution dened in [1], provides a convenient way in reusing the existing GSM/UMTS network to support voice in LTE network. This solution is standardized in [1] and provides the operators with exibility to roll out LTE as a data only overlay network and use the existing CS network for supporting voice functionality. The network architecture of CS fallback is shown in the Figure 1.1. The user performs a combined registration with both the LTE as well as GSM/UMTS network during the initial registration procedure. This combined registration is facilitated by the Mobility Management Entity(MME) in the LTE network which performs the registration in the 2G/3G network on behalf of the user. During the initiation of
Master of Science Thesis Prasanna Gururaj Raghavendrarao
Introduction
Figure 1-1: Circuit Switch Fallback [2]
the voice call by the user, the MME redirects the request towards the MSC server in the CS domain. On successful reservation of the resources in the CS domain for the call, the MSC server shall respond to the MME on the status of the request. The MME then instructs the E-Node B to request the user to perform a handover to the GSM/UMTS network. The ongoing data session for the user in the LTE network is suspended if the destination network is a GSM network. If the destination network is an UMTS network, then a separate handover of the existing data bearers from LTE to UMTS network takes place after registration by the user in the UMTS network. This solution has several disadvantages like increase in call set up time due to the handover procedure and disruption of data transmission throughout the duration of the voice call when the user falls back to a GSM network. This solution can be used during the initial roll out when LTE is more used for high speed data and voice is completely handled by legacy circuit switched networks. Hence CS fallback is being seen only as a temporary solution during the initial roll out of the LTE network. 1-1-2 Voice over LTE via IP Multimedia Subsystem (VoLTE)
In this solution, voice functionality is provided by the IP Multimedia Subsystem (IMS). IMS is a core network architecture that is integrated on top of the LTE network as shown in Figure 1.2. The IMS network is mainly used to provide all the basic services for voice
Prasanna Gururaj Raghavendrarao Master of Science Thesis
1-2 Motivation for the Thesis
that are provided by the existing CS networks. In addition, it also provides enhanced multimedia services like video conference, real time gaming etc. The main advantage of
Figure 1-2: VoLTE [2]
using an IMS based solution is that it completely utilizes the LTE architecture rather than relying on the existing CS networks for supporting voice feature. The IMS network is also capable of integrating with the legacy 2G/3G networks and thus can support voice call continuity even when the subscriber moves out of LTE coverage. Hence, the subscriber can experience the same services even when roaming into legacy networks. This solution is being projected as the long term solution as it is capable of providing enhanced services to the LTE network and also supports integration with the existing 2G/3G networks.
1-2
Motivation for the Thesis
The VoLTE solution mentioned in the above section is widely being considered for deployment by operators across the world as it provides simultaneous support of both voice and data in the LTE network. In VoLTE, voice is carried in the LTE network as Voice over IP (VoIP) packets. Hence the VoLTE architecture is signicantly dierent from the 2G/3G networks which have distinct CS capabilities for voice. IP based networks are mainly designed for best eort services which do not provide any strict guarantees on the quality of service demands of the various services that are oered to the users. In legacy networks like GSM/UMTS, IP based networks were mainly used for carrying data services like FTP, HTTP etc. However, with the growth of mobile broadband technologies like High Speed Packet Access (HSPA) and LTE, there is a growing need for carrying both voice and data in the same IP based network. Such an architecture could lead to signicant reduction in the costs for operation and maintenance of the networks. It also enables the operators to introduce new IP based services like Rich Communication Suite (RCS), that can provide the users with improved quality of experience at reduced costs. Hence the VoLTE solution should provide the users with a better quality of experience at reduced costs than the existing CS networks.
Introduction
In VoIP based networks, the user perceived Quality of Experience (QoE) depends on various QoS parameters like delay, jitter, latency, packet loss etc. In addition, since both the data and voice are carried over the same PS network in LTE, there needs to be proper classication among them for scheduling of network resources in the radio and core network domains. During congestion periods, scheduling algorithms used in both the radio and core networks for allocation of resources play a critical role in meeting the stringent delay and packet loss requirements of VoIP service as well as the packet loss requirements of the data service. The capacity of LTE radio network is very high and it can provide a peak cell throughput of around 300 Mbps in the downlink in the 4x4 MIMO conguration. This places a direct challenge on the transport network with respect to the scheduling of the resources. The motivation of this thesis is to study the eects of congestion in the transport network and to analyse its impact on the performance of voice in LTE network.
1-3
Problem Denition
In mobile broadband networks like LTE, the high performance of the radio network can be realized with proper scheduling of resources for dierent types of services. But proper scheduling of resources in the radio network alone is not sucient to guarantee a good end to end performance. During periods of high congestion, packet losses might occur in the transport network which can reduce the overall performance of the service that is oered to the user. Hence, the transport network between the radio and core networks is another area which needs proper dimensioning and scheduling of resources for various types of services. The transport network is not aware of the QoS architecture of LTE. This implies that the various bearers that are used to classify the services in LTE domain needs to be mapped to IP based QoS techniques.
The Dierentiated services architecture (Diserv) which is commonly used in IP based networks is used to classify the various types of services in the LTE transport network. The Diserv architecture needs to be integrated with the LTE QoS architecture to guarantee good end to end performance. The scheduling of resources in the transport network is another area which needs proper attention as the choice of scheduling algorithms is pivotal for optimum usage of resources. There are various scheduling strategies like Weighted Fair scheduling, Strict Priority scheduling and Weighted Round Robin scheduling that are used to schedule the packets based on the priority of each type of service. For real time trac like VoIP, the role of the classication and scheduling strategies is of paramount importance as they play a crucial role in guaranteeing the end to end quality of service to the users. During periods of congestion, real time services like VoIP can be severely impacted if there is a marginal increase in the end to end delay between VoIP packets or there is a packet loss in the transport network. The aim of this thesis is to study the various transport network scheduling strategies and to analyse their impacts on VoIP trac during congestion periods. The analysis is done based on simulations using OPNET simulation tool.
1-4 Related Work
1-4
Related Work
The transport of voice over LTE has a lot of challenges with respect to QoS as mentioned earlier. In the literature, there are a number of studies which are focussed on the optimum scheduling of resources for supporting VoIP service. In [3], Siomina et.al. have analysed the impact of prioritizing VoIP over other services in the radio network. The performance of prioritized VoIP is compared with Best Eort VoIP and the advantage in terms of increase in capacity is explained. In [4], Zaki, et.al. have studied the impact of dynamic packet scheduling on the performance of VoIP in LTE. Puttonen, J [5] and Yasir Zaki [6], have studied the impacts of MAC scheduling algorithm for dierent types of services. Most of these studies are focussed on the scheduling of resources in the LTE radio network. To the best of my knowledge there are very few studies that have been done on analysing the impact of scheduling in the LTE transport network. The most relevant study in this aspect is done in [7] in which Li, et.al. have studied the impact of dimensioning in the transport network. In this study, analytical models have been proposed for dimensioning the transport network for real time and non real time services and the proposed models are veried by simulations.
1-5
Organization of the Thesis
The thesis is organized as follows. In chapter 2, the background information related to the network architecture of LTE and IMS networks is introduced followed by a brief explanation on the QOS concepts in LTE. The chapter also gives an overview on the Diserv architecture that will be used in the transport network for classication of various services like voice, FTP etc. The chapter concludes with the explanation on the various scheduling strategies that will be used in the transport network. In chapter 3, the details of the OPNET modeller are presented. The limitations of the LTE model in OPNET and the changes that were done on the various process models are explained. The conguration details of the various nodes in the LTE network and the nal OPNET simulation environment that will be used for performing the analysis is presented at the end of this chapter. In chapter 4, the results of the simulation are presented. The chapter begins with a brief introduction of the various metrics that were used to perform the analysis followed by the evaluation of dierent congestion scenarios. In chapter 5, the technical impacts on deploying VoLTE solution is presented. This chapter begins with an introduction on the VoLTE architecture followed by various scenarios that are being considered for integration of VoLTE with the existing 2G/3G networks. The idea behind this study is to get an industry perspective on the technical impacts of VoLTE solution.
Introduction
In chapter 6, the main results of the thesis are summarized and topics for further research have been proposed.
Chapter 2 Background
2-1
Introduction
This chapter begins with an overview on the LTE network architecture which explains the functions of the various elements present in the LTE network. The QoS concept in LTE is presented in section 2.3. The QoS concept in LTE is based on bearers that uniquely dene the type of treatment for the packet ows between the mobile and the gateway nodes in the network. Hence this section provides the necessary information required for a better understanding of the QoS concept in LTE. The voice over LTE solution also requires an IMS core network that performs the necessary signalling and media related functions for providing voice services. The IMS core network architecture is presented in Section 2.4 to provide an overview on the functions of the key elements in IMS domain. In Section 2.5, the Dierentiated Services architecture is explained which will be used for packet classication in the transport network. In Section 2.6, the scheduling strategies that will be used to analyse the performance of VoIP are explained.
2-2
LTE Network Architecture
The LTE network architecture is shown in the Figure 2.1. The network architecture called the Evolved Packet System (EPS) has a at IP based architecture and is divided into the Evolved Universal Terrestrial Radio Access Network E-UTRAN and Evolved Packet Core (EPC). The overall architecture consists of ve elements which are explained as follows. E-UTRAN The radio network called the E-UTRAN comprises of the E-Node Bs that are interconnected to each other over the X2 interface and connected to the core network elements
Background
Figure 2-1: LTE Network Architecture
over the S1 interface. The E-Node Bs are responsible in scheduling and allocation of the radio resources for the users in the LTE network. The E-Node B terminates the control plane signalling messages as well as the user plane data with the EPC over the S1 interface.
EPC The EPC is the core network comprising of four elements which are Mobility Management Entity (MME), Serving gateway, Packet Data Network (PDN) gateway, Proxy and Charging Rules Function (PCRF) and Home Subscriber Server (HSS). MME MME is the most important element in the EPC as it terminates the control plane signalling from the user. Some of the functions performed by MME include authentication, mobility management, security and retrieval of subscription information from the HSS. Serving gateway Serving gateway is responsible for forwarding the user plane packets from the mobile towards the PDN Gateway. It is also responsible for tunnelling the user plane IP packets using the GPRS tunnelling protocol (GTP) when the user moves across dierent E Node Bs and serves as a mobility anchor for the user plane packets in the LTE network. PDN Gateway Packet data network gateway is the end node in the LTE network. It acts as an edge router and routes the user plane IP packets from the mobile nodes to other networks like Internet, IMS etc. It is also responsible for allocation of IP address to the user. PCRF PCRF is responsible for enforcing various operator policies on the network like guarPrasanna Gururaj Raghavendrarao Master of Science Thesis
2-3 QoS Architecture in LTE
anteed QoS, maximum bit rate provisioned for a user etc. It communicates with the PDN-gateway in enforcing these policies for various users in the LTE network. HSS HSS is the master database containing all the subscription information of the user along with the subscription for various services that are oered by the operator. It also comprises of the authentication centre which stores all the keys required for ensuring the encryption and integrity of the data in the network.
2-3
QoS Architecture in LTE
In LTE, the QoS is provided by means of a bearer which uniquely identies the packet ow between the user and the PDN-GW and is responsible for the priority that is given to a packet ow across the LTE network. Bearers are established after the successful authentication and registration of the user in the LTE network. The LTE bearer architecture is shown in the Figure 2.2.
Figure 2-2: QoS Architecture in LTE [8]
Each bearer is associated with a Trac Flow Template (TFT) which is used to differentiate the types of packets that ow through it. The TFT does this classication based on one of the following parameters: Port numbers ToS/DSCP Values Source/Destination address Protocol (TCP/UDP)
10
Background
QCI
Resource Type
Priority
Packet Delay Budget
1 2
GBR GBR
2 4
100 150
Packet Error Loss Rate 102 103
Services
3 4
GBR GBR
3 5
50 300
103 106
5 6 7 8 9
Non-GBR Non-GBR Non-GBR Non-GBR Non-GBR
1 6 7 8 9
100 300 100 300 300
106 106 103 106 106
Voice. Voice Conversation (Real Time Streaming). Real Time Gaming. Non Conversational Video (buered video). IMS Signalling. Video (Buered Streaming). Interactive Gaming. Video (Buered Streaming). Video (Buered Streaming).
Table 2-1: EPS QOS Bearer Denitions [9]
The bearers are classied as two types namely the default and dedicated bearers. Default bearers are established during the allocation of IP address to the user by the PDN Gateway. Default bearers provides the basic IP connectivity to the LTE network and does not provide any guaranteed QoS for the packets that are transmitted across this bearer. Dedicated bearers are used for specic services like voice, video streaming etc and are established based on the subscription prole of the user.
The bearers are also classied as Guaranteed Bit Rate (GBR) and Non Guaranteed Bit Rate (N-GBR). As the name indicates the GBR bearers provide guaranteed QoS to the packets that ows through this bearer and is less likely to be aected during heavy congestion at the network. On the other hand, the N-GBR bearers are used for services that do not have strict QoS constraints. As shown in the Figure 2.2, each bearer in LTE is characterized by a QoS Class Identier (QCI), Allocation and Retention Priority (ARP), packet delay budget and maximum bit rates. The QCI uniquely identies the type of bearer that is provisioned for the user at the radio and core networks. It is used in determining the type of treatment a packet ow experiences at each of the nodes in the LTE network. The ARP is used to decide whether a bearer can be admitted and is also used to release the bearers based on priority levels when the network is congested. The Table 2.1 [9] summarizes the values for each type of bearer.
2-4 IMS Network Architecture
11
2-4
IMS Network Architecture
IP Multimedia Subsystem is a core network architecture standardized by the 3GPP [10] to provide multimedia services like voice, streaming services like video on demand etc over an IP backbone independent of the underlying access network through which the user registers with it . The most important service provided by an IMS network is the Multimedia Telephony Service (MMTel) which is the basic voice over IP service but with guaranteed QoS. IMS is also capable of interworking the circuit switched 2G/3G network with packet switched networks like LTE. Hence IMS based voice is envisioned as the ultimate target solution for supporting voice in advanced next generation networks like LTE. The Figure 2.3 presents the key elements of an IMS network. The main elements of IMS core network are the Proxy Call Session Control Function(P-CSCF), Serving Call Session Control Function (S-CSCF), Interrogating Call Session Control Function (I-CSCF), Breakout Gateway Control Function (BGCF), Media Gateway Control Function (MGCF) and Media Resource Function (MRF). The main functions of these entities are explained as follows:
Figure 2-3: IMS Network Architecture
P-CSCF : P-CSCF is a SIP proxy server in the IMS domain which is the rst point of contact for the user within the IMS domain. All the requests of the user to the elements in the IMS domain as well as to the application servers are routed through the P-CSCF. In addition, the P-CSCF performs functions like subscriber
12
Background
authentication and establishment of security association with the mobile. It may also authorize QoS resources for the voice bearer by way of a policy decision function. S-CSCF : S-CSCF is the main element in the IMS domain which performs important functions like subscriber registration, authorization for using specic application servers, DNS lookup to retrieve the address of the destination etc. It downloads the user proles from the HSS for performing authorization of the subscriber. I-CSCF : I-CSCF is a SIP server that acts as a last point of contact in the IMS domain i.e. it is at the edge of the IMS domain and all requests from other IMS domains as well as requests from remote application servers are routed through the I-CSCF. During initial registration, the I-CSCF queries the HSS to assign a S-CSCF for the specic user. BGCF : BGCF performs breakout to other domains when routing of the request based on ENUM lookup is failed at the S-CSCF. It is mainly used when the destination user is a PSTN user and the call needs to be transferred to the CS domain. MGCF : MGCF is used to translate SIP signalling into ISUP signalling for communication towards PSTN and other CS networks. It also controls the media gateway which translates the RTP into CS media stream. MRF : MRF is used in transcoding between dierent codecs and provides media related functions like mixing of media streams and playing tones etc. The MRF is subdivided into Media resource function controller (MRFC) and Media resource function processor (MRFP) which perform the media translation activities in the control and user plane respectively.
2-5
Dierentiated Services Architecture
The Dierentiated services (Diserv) architecture dened in [11] is used by the E-Node B and the PDN gateway to map the QCI to a DSCP value in uplink and downlink respectively. This mapping at the E-Node B and the PDN gateway allows for the classication of the packets in the underlying transport network. The Diserv architecture consists of various Per Hop Behaviours (PHB) that are used to identify and classify the packets and apply appropriate QoS treatment at the transport network. The PHB classes are broadly classied into three classes namely Expedited Forwarding(EF), Assured Forwarding(AF) and Best Eort(BE). The EF class has the highest priority and is generally used for delay critical services like signalling, voice etc. The AF class consists of several sub classes with dierent levels of drop precedences as shown in the Table 2.2. The drop precedence enables the
2-6 Scheduling Strategies
13
Drop Precedence Level 1 Level 2 Level 3
AF 4X AF41 (DSCP 34) AF42 (DSCP 36) AF43 (DSCP 38)
Af 3X AF31 (DSCP 32) AF32 (DSCP 30) AF33 (DSCP 28)
Table 2-2: Assured Forwarding Drop Precedence Classication
operator to provide various levels of QoS for dierent types of services. The Best Eort class is the default PHB and has the least priority among the three classes. The AF class and BE class employ the Weighted Random Early Detection technique to detect congestion of queues based on pre dened thresholds. When the number of packets in the queue exceeds a minimum threshold, the WRED technique starts dropping of packets based on the weight assigned to each queue. If the link is heavily congested and the number of packets in the queue exceeds the maximum threshold then all incoming packets to the queue are dropped.
2-6
Scheduling Strategies
The transport network consists of a scheduler which assigns the available network bandwidth based on certain priorities and weights. The scheduler uses the classication of the packets based on DSCP to form these strategies for allocation of resources in the transport network. In this work, three dierent scheduling strategies are evaluated and their performance is compared for delivering high quality voice service in LTE network. The following section gives an overview on the dierent scheduling strategies that are implemented in the transport network.
2-6-1
Strict Priority Scheduling
In this type of scheduling, the packets are grouped into four levels of priority namely low, normal, medium and high. Packets which are very sensitive to delay like voice are given a high priority and services like streaming which have tolerable delay budgets are given medium priority. TCP based services like HTTP and FTP are mapped to normal and low priorities respectively as they have less constraint on the delay budgets. The scheduler always processes the high priority packets before servicing packets in other queues. This scheduling is especially useful for services like VoIP which have stringent delay requirements. The major drawback of this scheduling is, when the network is congested with high priority trac like voice, the low priority data trac will completely devoid of resources and hence the overall throughput of the network is reduced.
14
Background
2-6-2
Weighted Round Robin (WRR) Scheduling
This type of scheduling is based on the classical round robin scheduling where dierent types of services are served in a round robin manner. The only addition in WRR is the presence of weight which determines the number of packets that are removed from the queue. The packets are grouped into various queues and each queue is assigned a weight. Based on the weight, the scheduler calculates the bandwidth for each queue and corresponding to this bandwidth, number of packets in the queue are removed at a time before moving to the next queue. Hence WRR does packet by packet scheduling in a round robin manner. 2-6-3 Weighted Fair Scheduling
In Weighted Fair scheduling dened in [12], the packets are grouped into various queues and each queue is assigned a weight which determines the fraction of the total bandwidth available to the queue. In our case, there are dierent PHBs such as EF, AF and BE are assigned weights based on the priority of the trac. The bandwidth for each queue is based on the weights and is expresses as BWk = Wk BW W (2-1)
The Weighted Fair scheduling assigns the bandwidth for each service based on the weight assigned to each queue and not based on the number of packets. Hence when various types of trac like VoIP, FTP, HTTP are owing in the network, the bandwidth for each service is proportional to its weight and independent of the size of the packet in the queue. The main dierence between Weighted Round Robin and Weighted Fair is that the former does packet by packet scheduling in each turn whereas the latter does bit by bit scheduling. Weighted Fair hence has an advantage in the fact that it is aware of the true size of the packets in each queue while performing scheduling whereas Weighted Round Robin is not aware of the same.
Chapter 3 Simulation Model
3-1
Introduction
This chapter begins with an explanation on the details of the OPNET simulation environment used for modelling the LTE network. In section 3.3, the issues with the LTE model in OPNET are presented briey and the modications that were performed to resolve these issues are highlighted. The conguration parameters of the LTE network and the simulation settings for the VoIP and FTP process models are provided in the subsequent sections.
3-2
Introduction to OPNET Modeller
The OPNET simulation environment [13] is a discrete event simulation tool that is used in analysing the performance of various networks like LTE, WiMAX, Wi-Fi and Zigbee. The models library in OPNET consists of a large number of models supporting variety of protocols like TCP, UDP, SIP and is capable of simulating applications like voice, video, FTP etc. In this thesis, the LTE model in OPNET is used along with application models like voice and FTP. The details of the simulation environment are presented in the following sections. 3-2-1 Overview of LTE Model in OPNET
The OPNET modeller has a hierarchical environment consisting of the network model, node model and process model. All the three models need to be congured to perform the simulation. The LTE network model in OPNET is shown in Figure 3.1. The model consists of mobile nodes , an E-Node B and an EPC. The LTE core network consisting of the MME, serving gateway and PDN-gateway is modelled by a single device represented as the EPC in the Figure 3.1. The LTE attribute denition node is used to
16
Simulation Model
Figure 3-1: LTE Network Model
dene various conguration parameters like DL and UL frequencies, bandwidth and the various bearers that will be congured on the mobile nodes. The LTE model implements most of the features that are standardized by the 3GPP. However, it has some limitations in the establishment of bearers and hence signicant changes are required in the model to perform our analysis. The Figure 3.2 gives the data ow in LTE network.
Figure 3-2: Data Flow in LTE Network [13]
It is seen that for each bearer in the radio network, there is a corresponding S1 bearer in the transport network. This S1 bearer uses the GPRS Tunnelling Protocol (GTP).
3-2 Introduction to OPNET Modeller
17
Hence for each bearer that is established between an user and EPC, there is a separate GTP tunnel established for control plane signalling as well as user plane data. The signalling GTP tunnel is used for transmitting all the signalling information related to the establishment of the bearer. The data GTP tunnel is used in forwarding all the user plane IP packets from the user to EPC and vice versa. The Figure 3.3 gives the complete protocol architecture across various nodes in the LTE network. The GTP-U layer in the E-Node B represents the tunnels that are created between the E-Node B and EPC. In the uplink when the E-Node B receives IP packets from the mobile, the GTP layer encapsulates the received IP packet and copies the contents of the inner IP header to the outer IP header. The same process is repeated in the downlink direction when the EPC node receives an IP packet from outside the LTE network. The encapsulated GTP packet structure is shown in the Figure 3.4.
Figure 3-3: Protocol Architecture
Figure 3-4: GTP Encapsulated Packet
18
Simulation Model
3-2-2
Issues in LTE model
In the LTE model present in OPNET, the process of GTP encapsulation is not implemented in accordance to the QoS type of the bearer . The packets entering the E-Node B in the uplink are encapsulated into an IP packet without any classication based on the type of bearer (DSCP mapped to BE by default). The same issue is there in the downlink when the packets entering EPC are encapsulated without proper classication. Due to this problem, when there are dierent types of services like voice, FTP, HTTP, video streaming etc., there is no proper classication of packets at the IP level in the transport network. As explained in the problem denition in Chapter 1, the intermediate nodes in the transport network between E-Node B and EPC are not aware of the classication based on bearers. The type of scheduling strategies used in the transport network also has no meaning, if all the packets are classied with same priority. So, it is very important for the E-Node B and EPC to perform packet level classication by mapping the bearer type to DSCP. Hence changes are required in the process models in the E-Node B and EPC. The following sections illustrate the changes that were performed in the E-Node B and EPC to achieve this objective.
3-3
Changes in LTE model
As mentioned in the previous section, there is no packet level classication among the bearers in the transport network which needs to be implemented. This section presents an overview on the changes that were done in the E-Node B and EPC nodes in the OPNET. 3-3-1 LTE S1 process model In E-Node B
The node model for the E-Node B is shown in Figure 3.5. The node model gives an overview on the various layers of the 3GPP LTE stack that are implemented in E-Node B. In this node model, there are two processes lte_ s1 and gtp (highlighted in Figure 3.5) that are to be modied. The lte _ s1 process model at the E-Node B is shown in the Figure 3.6. This process is run every time when a dedicated bearer is created in the LTE network. The s1 _ msg _ rcvd represents the state in which the E-Node B has received a new bearer request message from the mobile. The state runs a dedicated bearer setup function and commands the GTP layer in the E-Node B to create a tunnel for this dedicated bearer towards the EPC in the uplink direction. So, the change that needs to be performed in this process is to map the QoS type of the tunnel created to the type of the bearer that is received in the request. This is done as follows: Each bearer congured in the mobile node is mapped to a bearer ID which uniquely identies the bearer in the network.
3-3 Changes in LTE model
19
Figure 3-5: E-Node B Node Model
Figure 3-6: lte _ s1 Process model in E-Node B
The bearer is also congured with a TFT as explained in Section 2.3 which is used to perform the mapping between the bearer level QoS and DSCP value in the IP header. So there is an indirect mapping between the bearer ID and DSCP value. By using this mapping, the functions in the process model are changed such that during the creation of GTP tunnel between the E-Node B and EPC, the DSCP value corresponding to the bearer ID is also taken into consideration.
20
Simulation Model
3-3-2
LTE S1 NAS Process model in EPC
The node model of the EPC is shown in Figure 3.7. Similar to the E-Node B node model, the EPC also has the lte _ s1 _ nas and gtp process models that are responsible for bearer creation and tunnel creation respectively. Hence these two processes are to be changed to overcome the issues explained in the previous section. The Figure 3.8 shows
Figure 3-7: Node Model in EPC
the lte _ s1 _ nas process model. This process model is used to setup the S1 bearer that carries the data between the E-Node B and EPC for the mobile in the downlink direction. In the Figure 3.8, the state s1 _ msg _ rcvd represents the state in the EPC corresponding to the one explained in the previous subsection at the E-Node B. This state acts as a trigger towards the state nas _ msg _ rcvd which indeed actually contains the functions responsible for setting up the S1 bearer towards the E-Node B. The mapping procedure used in the previous section for E-Node B is again followed here in the EPC. 3-3-3 GTP Process model in E-Node B and EPC
The GTP process model that runs in both the E-Node B and EPC nodes is shown in Figure 3.9. The GTP-U block performs the encapsulation of the user plane IP packet received at the E-Node B and delivers it to the UDP layer for transport towards the EPC. The GTP-U block consists of four states which are idle, tunnel search, gtpencap and to UDP. When a packet arrives at the E-Node B, the process goes from the idle
3-3 Changes in LTE model
21
Figure 3-8: lte _ s1 _ nas Process model in EPC
state to the tunnel search state. If the tunnel corresponding to the bearer is found, the process goes to the gtpencap state, where the packet is encapsulated and sent to the UDP module. Else, the process goes to the tunnel management state, where the tunnel creation function is executed before the encapsulation of packet is performed. The same process is repeated at EPC in the downlink direction when a packet arrives from outside the LTE network. Hence the gtpencap state is where the actual process of mapping between the bearer ID and the DSCP takes place and the contents of the inner IP header are copied to the outer IP header. The various functions that were used to perform this mapping in the GTP layer were modied to overcome the limitations that were mentioned in Section 3.1.2.
22
Simulation Model
Figure 3-9: GTP Process Model
3-4
Simulation Environment
This section lists all the conguration details of the various nodes that were used in the analysis. The network topology used for performing this simulation is shown in the Figure 3.10. In the network topology, there are two cells represented by E-Node B 1 and E-Node B 2 connected to the EPC via an Edge router. Each cell consists of 30 LTE users. The EPC node is connected to an IMS network via an edge router. The Ethernet links between the E Node B and the EPC are 5 Mbps. All other links in the core network are of 10 Mbps capacity. There are two FTP servers connected to the EPC node which are used by the mobile nodes for establishing FTP sessions in the network. The conguration details of each of the nodes are explained below. 3-4-1 Mobile Node
The LTE mobile nodes are congured to run VoIP and FTP services. Each mobile node is congured to run one type of application at a time. The Figure 3.11 shows the important conguration details of the mobile nodes in the network. The EPS bearer conguration attribute denes four bearers namely Platinum, Gold, Silver and Bronze.
3-4 Simulation Environment
23
Figure 3-10: LTE Simulation Network
Each of the bearer is assigned to a TFT packet lter which in our case is the DSCP value. The Table 3.1 shows the mapping between the bearer type and DSCP value. This mapping is used by the mobiles to identify the type of bearer for dierent types of
Bearer Type Platinum Gold Silver Bronze DSCP EF AF 11 AF 43 BE
Table 3-1: EPS Bearer to DSCP Mapping
services like voice, FTP, etc. The mobility feature in the mobile nodes is set to disabled as we assume that all the mobiles are stationary in the area around the E-Node B. 3-4-2 E-Node B
The E-Node B in the network is congured with 3 MHz bandwidth. The total capacity of each cell is limited to 10 Mbps. The channel between the mobile nodes and E-Node B is congured to be an error free channel as the primary objective of this analysis is to investigate the impact of congestion in the core network. Hence various physical layer eects like multipath and interference eects are not modelled in these simulations.
24
Simulation Model
Figure 3-11: Mobile Conguration
The MAC scheduler implemented in the OPNET E-Node B module uses a priority scheduling among the guaranteed and non guaranteed bit rate bearers which implies the guaranteed bit rate bearers are always allocated radio resources ahead of the non guaranteed bit rate bearers. To avoid the scenario of packets getting dropped due to non availability of resources in the radio network, the peak usage of each cell is limited to 50 percent of the total capacity. The summary of the conguration parameters of the E-Node B is listed in the Table 3.2.
Parameter Name Bandwidth UL Frequency DL Frequency Channel Characteristics No. of Transmit/Receive Antennas
Value 3MHz 1920 MHz 2110 MHz Error Free 2
Table 3-2: E-Node B Conguration Parameters
3-4 Simulation Environment
25
3-4-3
IMS Model
The IMS model used in this simulation environment is used from the contributed models section available in [13]. It consists of proxy, serving and interrogating call session
Figure 3-12: IMS Proxy session control function conguration attribute
control functions (P/I/S-CSCF) which are used in signalling procedures for the VoIP calls between the dierent users in the network. The IMS signalling ow in the LTE network requires the highest priority as it is the rst procedure that is invoked towards the establishment of the VoIP call between the users. Hence all the IMS signalling packets are marked with the highest priority in both the radio and core networks. The Figure 3.12 gives the conguration attribute of the P-CSCF in the IMS network.The domain name and area congured in these servers are also congured in the mobiles and using these attributes, each mobile registers with the IMS network. The three call session control functions are used to route the signalling between two VoIP users before the establishment of the media path. The SIP signalling procedure dened in [1] is followed for establishment of the VoIP calls between the users in the network. The IMS model is used in our simulation only to emulate the real world scenario as the main focus of our study is on the user plane voice bearer and not on the control plane signalling data.
26
Simulation Model
3-4-4
Application Conguration
We use two trac models namely voice and FTP for performing this analysis. The details of the trac models are explained below.
VoIP model
We use the commonly used G.711 voice codec for all the simulations. The codec has a bit rate of 64 Kbps with 20 milliseconds frame size and 1 frame per packet. Hence, there are 50 packets that are transmitted per second. The RTP/UDP/IP layers add headers to each packet and hence the overall bandwidth is around 90 kbps. In our simulations, silence suppression is used and is modelled as an exponential distribution with talk spurt length of 1.2 seconds(mean) and silence length of 0.8 seconds(mean). A summary of the conguration details is given in Table 3.3.
Parameter Name Codec Frame Size Voice Activity Factor Silence Suppression Value G.711 (64 Kbps) 20 ms 0.6 Enabled
Table 3-3: VoIP Conguration Parameters
FTP model
The FTP server is congured to send a le of size 1 MB upon request by each mobile. The inter repetition time between requests is 30 seconds. There is a separate TCP connection established for each request between the mobile and the server.
Chapter 4 Results
4-1
Introduction
In this chapter, the performance of the various scheduling scheduling strategies that were explained in Chapter 2 are investigated. The chapter begins with an introduction on the various QoS parameters that were used for performing the analysis. In all the simulations, only voice and FTP services are used. The details of the trac models for voice and FTP are as explained in Chapter 3.
4-2
QoS parameters for Voice
The following are the parameters that were used to determine the QoS of the VoIP call in the LTE network. Packet End to End delay : This parameter gives the total voice packet delay i.e. the mouth to ear delay between the users. In all simulations, the mean end to end delay is shown for the all the users in the network. Packet Delay Variation (PDV) : This parameter gives the variance in the end to end delay among all the packets received at the user. The mean of this PDV is shown for all the users in the network. S1 Delay : S1 delay is the one way delay between the E-Node B and the EPC. This parameter gives the mean time taken for a packet to traverse between the E-Node B and EPC. The S1 delay is measured at the E-Node B. Packet Loss Rate (PLR): The packet loss rate gives the number of voice packets that are lost in the network due to congestion. The packet loss rate is measure at the EPC node, since the congestion is in the core network. The mean of the PLR for all the users in the network is shown for all the simulations.
28
Results
In addition to the above QoS parameters, the Mean Opinion Score (MOS) is also presented for all the simulations. The MOS is a measure of the Quality of Experience for the VoIP users in the network. The E-Model dened in [14] is used to calculate the MOS based on the R-factor. The R-factor called the rating factor is used to measure the quality of the VoIP call based on various parameters like packet end end delay, packet loss etc. The R-factor is expressed as follows [14] R = 94.2 Id Ie (4-1)
where Id is the impairments caused due to the mouth to ear delay and Ie is the impairment caused due to packet losses in the network. The R-factor is mapped to a MOS score using the following mapping dened in [14]: M OS = 1 + 0.035R + 7 106 R(R 60)(100 R), 0 R 100 M OS = 1, R 0 M OS = 4.5, R > 100 (4-2) (4-3) (4-4)
The OPNET software uses the above model to calculate R factor and is mapped to the MOS using the above equation. The MOS score is mapped to the level of satisfaction of the users based on Table 4.1 [14].
MOS score 4.3 - 5 4 - 4.3 3.6 - 4 3.1 - 3.6 2.6 - 3.1 Less than 2.6 Quality of VoIP call experienced by the user Very much satised Satised Many users satised Many users dissatised Nearly all users dissatised Not recommended
Table 4-1: MOS satisfaction level
In all the simulations, the average of the MOS for all VoIP users in the network is presented.
4-3
Scenario 1
This scenario is used to illustrate the signicance of QoS in the transport network by mapping both VoIP and FTP users with the same priority in the transport network. 4-3-1 Description of the Scenario
Case 1: In this case, only voice trac is generated in the network. The number of voice users in the network is periodically increased from 20 to 100. There are totally 25 LTE mobiles running VoIP application in each cell and each user is
4-3 Scenario 1
29
congured to establish multiple VoIP sessions simultaneously. The VoIP session is carried over a Guaranteed Bit Rate (GBR) bearer (QCI 1 in Table 2.1) and is mapped to default best eort (DSCP-BE) QoS in the transport network. Case 2: In this case, both voice and FTP trac are generated in the network. The number of voice users are same as the previous case and there are totally 10 FTP users (5 in each cell). VoIP users are mapped to GBR bearer as in case 1 and FTP users are mapped to Non GBR bearer (QCI 9 in Table 2.1) in the downlink. Both the voice and FTP are mapped to the same best eort QoS in the transport network. So, the packets entering the nodes EPC, Edge Router 1 and E-Node B 1 & 2 are served with First In First Out Scheduling (FIFO). Hence, this case analyses the performance of best eort VoIP service when the network is congested with data service.
4-3-2
Analysis of results
The Figures 4.1 and 4.2 show the mean end to end delay and mean packet delay variation for VoIP users in the network. For case 1, the delay is constant at 80 ms whereas when there is an ongoing FTP session in case 2, there is a signicant increase in the end to end delay for VoIP users. In case 1, since there are only VoIP users present in
Figure 4-1: Voice Packet End to End Delay vs No. of VoIP users
the network, the total trac in the link between the Edge Router 1 and the EPC is still within the total bandwidth even when the number of VoIP users is large. At 100 VoIP
30
Results
Figure 4-2: Packet Delay Variation Vs No. of VoIP users
users the peak trac that can be expected is maximum at 9 Mbps. This value is arrived by assuming that all the users are sending voice packets at the same time and the bandwidth required for a single voice call is 90 Kbps after adding the RTP /UDP/ IP headers to the actual voice payload. The peak trac will never be reached as each user has an exponential distribution on the talk spurts and silence periods. So, the mean end to end delay for all VoIP users in the network remains constant. This also explains the Figure 4.2, which shows no variation in delay among the packets received at the mobile. In case 2, since there is no priority among voice and FTP, the smaller VoIP packets are getting queued in the core network and the edge router till the larger FTP packets are processed in each node. This causes a larger variation among the packets received at the mobile as shown in Figure 4.2. The minimum mean delay is 150 ms, when the number of VoIP users is 20 and is much higher than the acceptable limit of 100 ms. From the scenario, it is evident that to achieve an acceptable QoS for VoIP in LTE, there needs to be proper classication in the transport network.
4-4 Scenario 2
31
4-4
Scenario 2
In Scenario 2, the voice users are accorded the highest priority and the FTP users are mapped to the lowest priority. Each VoIP user is assigned to a Platinum bearer (QCI 1 in Table 2.1) and mapped to the highest EF QoS class in the transport network. The FTP users are assigned to a Bronze bearer (QCI 9 in Table 2.1) and mapped to the BE QoS class in the transport network. 4-4-1 Description of the Scenario
This scenario evaluates the performance of the scheduling algorithms explained in Chapter 2. The scenario is subdivided into three cases as follows: Case 1: In this case, the Weighted Fair scheduling algorithm explained in Chapter 2 is analysed. The high priority VoIP trac is assigned a weight of 7 and the FTP trac is assigned a weight of 3. Hence, the total bandwidth assigned for voice users is 7 Mbps and the total bandwidth for FTP users is 3 Mbps. Case 2: In this case, the Strict Priority scheduling algorithm explained in Chapter 2 is analysed. In terms of priority as explained earlier the VoIP users are mapped to high priority and FTP users are mapped to low priority. Case 3: In this case, the Weighted Round Robin scheduling algorithm explained in Chapter 2 is analysed. The weights for the voice and FTP services are same as those assigned in Case 1. 4-4-2 Analysis of results
This section presents an analysis on the various parameters that were explained in Section 4.1.
Packet end to end Delay and PDV
In Figure 4.3, the mean end to end delay for the voice packets is shown. In case 1 and case 3, the Weighted Fair and Weighted Round Robin scheduling algorithms a denite bandwidth is assigned to the VoIP users. Hence till this bandwidth limit is reached, the mean end to end packet delay shown in Figure 4.3 is constant at 80 ms. When the number of VoIP users is 80, the mean end to end delay is around 100 ms for both the cases which is still within the acceptable limit. The increase in the end to end delay is attributed to the fact that the peak bandwidth for 80 VoIP users is around 7.2 Mbps. This bandwidth is higher than the provisioned bandwidth for VoIP users based on the weight which is calculated to be around 7 Mbps. The delay falls within the 100 ms threshold as the peak bandwidth for voice will not be reached due to the exponential distribution of the talk spurts between the users.
32
Results
Figure 4-3: End to End Delay Vs No. of VoIP Users
Figure 4-4: S1 Delay Vs No. of VoIP Users
The Figure 4.4 shows the mean one way S1 delay i.e. the time taken for the voice packets to reach the E-Node B from the EPC.
4-4 Scenario 2
33
Figure 4-5: PDV vs No. of VoIP users
When the number of VoIP users is 80, the S1 delay shows a substantial increase which explains the overall increase in the end to end delay at the mobile node. When the number of VoIP users reaches 100, there is a signicant increase in the bandwidth demand of the VoIP users, leading to more waiting time in the queues at the core network as shown in Figure 4.4. The mean end to end delay at this point is around 120 ms which is beyond the tolerable limit. The PDV shown in Figure 4.5, follows a similar pattern like the mean end to end delay with the variation increasing to 0.3 ms. The PDV is small compared to the No QoS case explained in Scenario 1. This is because of of the separate bandwidth provisioning for VoIP and FTP users in these two scheduling algorithms. The mean end to end delay and delay variation for Strict Priority scheduling is also shown in Figures 4.3 and 4.5. The delay remains constant at 80 ms and packet delay variation is negligible. As explained in the section 2.6, the Strict Priority scheduling always performs better for VoIP users which are assigned a higher priority compared to the FTP users.
PLR and MOS
In Figures 4.6 and 4.7, the PLR for the VoIP users in the network and the corresponding MOS is shown. From the Figure 4.6, it is seen that for case 1 and case 3, the packet loss rate is almost negligible till number of VoIP users is equal to 80 when the packet loss rate reached the threshold of 2 percent. Beyond this, with an increase in the number of
34
Results
Figure 4-6: Packet Loss Rate vs No. of VoIP users
Figure 4-7: Mean Opinion Score vs No. of VoIP users
VoIP calls, the packet loss rate also increases signicantly. The value beyond this point is of no signicance, as more than 2 percent drop in the number of packets implies that the VoIP calls are dropped. The high packet loss rate beyond this point attributes to the sharp decrease in the value of the Mean Opinion Score shown in Figure 4.7. In case 1, the minimum value of MOS is 2.8 as shown in Figure 4.7 whereas in case 3, the minimum MOS value is 3. In both cases, the lower value of MOS implies that all the users are dissatised and hence beyond 80 users, there is no possibility of having
4-4 Scenario 2
35
more number of VoIP users for case 1 and case 3. In case 2, the PLR is negligible as shown in Figure 4.6 and hence there is no impact on the MOS values for case 2. 4-4-3 Impact on FTP trac
The increase in the number of VoIP users will have a direct impact on the le transfer time for the FTP users in the network. The Figure 4.8 shows the FTP transfer time for the three cases. The transfer time is increased by almost twice in case 1 and case
Figure 4-8: FTP Transfer time vs No. of VoIP users
3 when the number of VoIP users reaches 100. The increase is mainly due to the the congestion in the link between the EPC and the Edge router, thereby leading to more queuing of packets in the EPC. The transfer time for case 2 shows a large increase when the number of VoIP users in the network increases beyond 80. This leads to an undesirable situation where there are more TCP re transmissions in the network leading to increased congestion. The Figure 4.9, shows the total FTP trac received by all the users in the network. It is seen from the Figure 4.9, that there is signicant drop in the FTP throughput when the number of VoIP users increase in the network for all the three cases. For case 1 this drop in throughput is around 30 percent whereas for case 3 the drop in throughput is around 50 percent. Weighted Fair scheduling performs better than Weighted Round Robin due to the fact that it performs bit by bit scheduling and hence is aware of the actual packet size of FTP packets before scheduling of resources. So, there is less delay for servicing large FTP packets in Weighted Fair compared to Weighted Round Robin which explains the behaviour in Figure 4.9. For case 3 there is almost a 75 percent drop in throughput when the number of VoIP users in the network
36
Results
Figure 4-9: FTP Throughput vs No. of VoIP users
increases from 80 to 100. Hence the overall QoS for FTP trac is severely degraded due to the increase in the number of voice users for case 2 when compared to case 1 and case 3. 4-4-4 Summary of the Results for Scenario 2
In this scenario, the performance of VoIP when mapped to Platinum bearer is analysed for three scheduling algorithms. Within the acceptable QoS thresholds (100 ms end to end delay and 2 percent packet loss), the number of VoIP users for case 1 and case 3 is 80 Users. The case 2 has a higher capacity of 100 users within the acceptable limits but it comes at a cost as Strict Priority scheduling totally starves the resources for low priority trac when the network is congested with high priority trac. Hence, the FTP transfer time shows more than 50 percent increase compared to case 1 and 3 which is not acceptable in environments with mixed trac.
4-5 Scenario 3
37
4-5
Scenario 3
In Scenario 3, each VoIP user is assigned to a Gold bearer (QCI 7) in both directions and mapped to the AF 11 QoS class in the transport network. The FTP users are assigned to the Bronze bearer as in Scenario 2 and mapped to the BE QoS class in the transport network. 4-5-1 Description of the Scenario
This scenario evaluates the performance of VoIP when assigned to a normal priority in the radio and core networks. The scenario is subdivided into three cases as earlier described in Scenario 2. Case 1: In this case, the Weighted Fair algorithm is analysed as in Scenario 2. The VoIP trac is assigned a weight of 5 and the FTP trac is assigned a weight of 3 in the transport network. Hence, the total bandwidth assigned for voice users is around 6 Mbps and the total bandwidth for FTP users is around 4 Mbps. Case 2: In this case, VoIP users are assigned Normal priority and FTP users are assigned low priority. Case 3: The weights for the Weighted Round Robin algorithm are same as those in case 1. 4-5-2 Analysis of results
Packet end to end delay and PDV
In Figure 4.10, the mean end to end delay for the VoIP users are shown. The mean end to end delay from the Figure 4.10 for case 1 and case 3 shows a signicant increase when the number of VoIP users is beyond 60. At 70 users, the delay crosses the threshold of 100 ms. This behaviour is due to the fact that when the number of VoIP users crosses 70, the maximum peak bandwidth for voice reaches around 6.3 Mbps which is slightly more than the provisioned bandwidth for voice which is 6 Mbps. The increase in the end to end delay is marginal until the number of voice users reaches 80 when the peak bandwidth is around 7.2 Mbps. At this point, more number of VoIP packets are buered in the interface of EPC node thereby increasing the end to end delay. The Figure 4.11 exactly proves this point as the mean one way delay between the EPC and E- Node B increases signicantly when the number of VoIP users crosses 70. The case 2 in this scenario follows the same behaviour as in scenario 1. This is mainly due to the fact that merely changing the priority to normal does not aect the VoIP quality as still VoIP packets are served rst by the scheduler before the FTP users are served. The PDV is shown in Figure 4.12 for all the three cases and as explained for scenario 1, the packet delay variation has no signicant impact when the number of VoIP users are increased.
38
Results
Figure 4-10: End to End Delay vs No. of VoIP users
Figure 4-11: S1 delay vs No. of VoIP users
4-5 Scenario 3
39
Figure 4-12: Packet delay variation vs No. of VoIP users
PLR and MOS
The Figure 4.13 shows the PLR for this scenario. The PLR for case 1 and case 3 exceeds the threshold value of 2 percent when the number of VoIP users is around 65 users. Beyond this point, there is a signicant increase in the packet loss rate which implies that beyond this point the calls will get dropped. In case 2, there is no packet loss as it follows the same behaviour as explained in scenario 2. The Figure 4.14 shows the average MOS for all the users in the network. It is seen that the quality of experience degrades for case 1 more than for case 3 when the number of users is increased beyond 60.
40
Results
Figure 4-13: Packet Loss Rate vs No. of VoIP users
Figure 4-14: Mean Opinion Score vs No. of VoIP users
4-5-3
Impact on FTP trac
In this scenario, the FTP trac is not impacted much when the number of VoIP users is increased. This can be observed in the Figure 4.15 which shows the mean le transfer time of all the FTP users. When the number of VoIP users are increased beyond 80, the
4-5 Scenario 3
41
mean FTP transfer time for case 1 almost remains constant and for case 3 it increases marginally by 3 percent. In comparison with scenario 2, we observe that the mean FTP transfer time is almost reduced by 50 percent when the number of VoIP users is 100 for scenario 3. This is largely due to the fact that a higher percentage of bandwidth is assigned to the FTP users in this scenario.
Figure 4-15: Mean FTP transfer time vs No. of VoIP users
The throughput for the FTP trac in downlink is shown in Figure 4.16. In this scenario, there is no change in the FTP throughput when the number of VoIP users is increased from 80 to 100 for case 1 whereas there is a slight decrease in case 3. This behaviour is due to the same concept explained in Scenario 2. For Strict Priority scheduling, there is a signicant drop as in scenario 2 and hence there is a severe degradation in FTP throughput when compared to other two scheduling strategies.
42
Results
Figure 4-16: FTP Throughput vs No. of VoIP users
4-5-4
Summary of results for Scenario 3
In this scenario, the performance of VoIP when mapped to Gold bearer is analysed for the three dierent scheduling algorithms. Within the acceptable QoS thresholds (100 ms end to end delay and 2 percent packet loss), the number of VoIP users for case 1 and case 3 is 65 Users whereas for case 2 it remains the same as in previous scenario at 100 users. The main reason behind the drop in the capacity of VoIP users is due to the fact that the VoIP users are mapped to AF bearer in the transport network which has a lower bandwidth limit compared to the previous scenario where the EF class had an higher bandwidth limit. The FTP throughput achieved in this scenario is much better compared to the previous scenario when the network is congested with VoIP users for case 1 and case 3.
4-6 Scenario 4
43
4-6
Scenario 4
In this scenario, the VoIP users are mapped into both Platinum and Gold bearers i.e. the VoIP users are split into two groups mapped into high priority Platinum bearer and normal priority Gold bearer. The description of the scenario is as follows. 4-6-1 Description of the scenario
In the transport network the high priority VoIP users are mapped to EF class and normal priority VoIP users are mapped to AF 11 class. The FTP users are mapped into Bronze bearer (QCI 9). The scheduling strategy is used such that for high priority VoIP users, Strict Priority scheduling is used. The remaining available bandwidth is shared between the normal priority VoIP users and FTP users using the Weighted Fair scheduling algorithm. The scenario is divided into four cases according to the number of high priority and normal priority VoIP users. In all the cases, the weights for normal priority VoIP users and FTP users are set to 6 and 3 respectively. The Table 4.1 shown below gives the details of each of the four cases.
Application Type VoIP (High Priority) VoIP (Normal Priority) Case 1 30 40 Case 2 20 60 Case 3 30 60 Case 4 20 80
Table 4-2: Scenario Description
4-6-2
Analysis of Results
Packet end to end delay and PDV
The Figures 4.17 and 4.18 shows the packet end to end delay for the premium and normal VoIP user. In Figure 4.17, we see that there is no signicant change in the end to end delay for the four cases. This is mainly due to the fact that for premium users, we use strict priority scheduling and hence they are always served rst even during the time of congestion. There is also no signicant PDV due to the same reason and hence PDV is not plotted for the premium VoIP users. In Figure 4.18, the packet end to end delay for the normal VoIP user is shown. The end to end delay for case 1 is around 80 ms. The peak bandwidth for normal users when all of them send packet simultaneously is around 3.6 Mbps which is less than the provisioned bandwidth of 4.8 Mbps. Hence there is very less waiting time in the core network for the VoIP packets which explains the less packet end to end delay. Due to the same reason, there is no variation in packet delay as seen in Figure 4.19. In case 2 and case 3, the end to end delay is around 100 ms. This is a signicant change compared to the case 1 but still the value is within the acceptable limits of end to end
44
Results
Figure 4-17: Packet end to end delay for high priority VoIP users
Figure 4-18: Packet end to end delay for normal priority VoIP users
delay. For both the cases, the peak bandwidth is around 5.4 Mbps which is higher than the provisioned bandwidth. Hence the packers are buered in the EPC node, which leads to an increase in the end to end delay. For case 4, the delay is 140 ms which
4-6 Scenario 4
45
is beyond the acceptable value. The peak bandwidth in this case is around 7.2 Mbps which is well beyond the provisioned bandwidth. This leads to a congestion in the EPC node leading to more waiting times. The values of PDV for cases 2 and 3 are around 0.5 ms whereas for case 4 the PDV is around 2 ms as shown in Figure 4.19.
Figure 4-19: Packet delay variation
PLR and MOS
The packet loss rate for the premium VoIP users is null since they are served using strict priority. As there is no packet loss in this case and the delay is within the limits, the value of MOS is greater then 4.3 which implies there is very high quality of experience for premium users. The PLR and MOS for premium VoIP users follow the same pattern as in Figures 4.6 and 4.7. The Figure 4.21 shows the PLR for normal VoIP users. There is no packet loss for case 1 as the congestion scenario is not yet reached and the bandwidth is within the limits. For case 2 and case 3, there is a packet loss of around 1.5 percent which is still within the acceptable value of 2 percent. For case 4, there is a large packet loss of around 10 percent which implies that the calls are dropped. The Figure 4.22 shows the MOS for the normal VoIP users. It is clear from the Figure 4.22 that the MOS for cases 2 and 3 are less compared to case 1 but still within the acceptable value.
46
Results
Figure 4-20: Packet Loss Rate
Figure 4-21: Mean Opinion Score
Impact on FTP trac
The Figure 4.23 shows the FTP transfer time for all the four cases. It is seen that for case 1 the transfer time is 24 seconds and for case 2 the transfer time is around 27 seconds. The total number of VoIP users in the network is 70 users (premium + normal) for case 1. In Figure 4.15, at the same point, it is seen that the transfer time
4-6 Scenario 4
47
is around 26 seconds in Scenario 3 . Similarly the total number of VoIP users in case 2 is 80 and at the same point the transfer time seen in Figure 4.15 is 30 seconds. Hence in these two cases for the same number of VoIP users, there is a marginal decrease in the FTP transfer time compared to scenario 3. In Figure 4.23, the FTP transfer time for case 3 and case 4 are 50 and 65 respectively. In comparison with Figure 4.15, the transfer time is signicantly higher for both the cases. This is due to the fact that in case 3 in Figure 4.23, there are more number of VoIP users (premium + normal = 90) which are within the acceptable limits of QoS compared to 4.15 and hence the VoIP capacity is increased at the expense of the FTP throughput. In case 4, the number of normal VoIP users are very high which leads to more congestion in the core network and hence the throughput for both the VoIP and FTP users are signicantly aected leading to a poor performance for both the services.
Figure 4-22: FTP Transfer time
48
Results
4-7
Comparison of Scenarios
In this section, a comparison of the scenarios 2, 3 and 4 is done for a better understanding of the results. The table 4.2 gives the number of satised VoIP users in each scenario which are within the acceptable limits of 100 ms delay and 2 percent packet loss. The corresponding FTP transfer time at this point is also shown in Table 4.2. The number of voice users in Scenario 4 shown in the Table 4.3 is the total number of voice users (premium+normal).
Scenario Name Scenario 2 Scenario 3 Scenario 4 case 1 80 65 70 No. of VoIP users case 2 case 3 case 4 100 80 X 100 65 X 80 90 20 Corresponding FTP Transfer time case 1 case 2 case 3 case 4 38 130 40 X 24 130 24 X 24 26 50 65
Table 4-3: Number of satised VoIP users
From the table 4.2, it is seen that the Scenario 4 has a better capacity in terms of number of VoIP users compared to the other two scenarios. This is explained as follows. In Scenario 2, the emphasis is more on increasing the VoIP capacity at the cost of increase in transfer time for FTP users when there is congestion in the network. Though there no delay guarantees for FTP users in this scenario, there should not be total degradation of throughput for FTP as in case 2. Hence in Scenario 2, the maximum number of VoIP users that can be supported with acceptable QoS limits for voice is 80. Comparing this with Scenario 4, there are 90 VoIP users that can be supported within the acceptable QoS limits. Hence there is about 10 percent increase in the VoIP capacity at a cost of increase transfer time in Scenario 4 when compared to Scenario 2. The Scenario 3 has strict bounds on the QoS of data trac i.e. the mean FTP transfer time is not increased by more than 20 percent when there is congestion in the network. Hence in Scenario 3, the maximum number of VoIP users that can be supported while ensuring that the increase in delay for data trac is within the bounds is 65. Comparing the Scenario 3 with Scenario 4, we see that for the same criteria i.e increase in delay nor more than 20 percent the number of VoIP users that can be supported is 80. Hence there is a 20 percent increase in the VoIP capacity in Scenario 4 when compared to Scenario 3. Hence by grouping of VoIP users into dierent levels of priority an increase in capacity is achieved when compared to mapping them to a single specic service class.
Chapter 5 Technical Details of Voice over LTE via IMS based solution - An Operator Perspective
5-1
Introduction
This chapter gives an overview on the various technical impacts of VoLTE solution on the existing CS and PS networks. The VoLTE solution will introduce the voice functionality in the LTE network using the new IMS framework which is widely being accepted as the long term solution for supporting voice in LTE network. IMS based voice is widely seen as the better solution in the current scenario capable of delivering voice in the LTE network. Hence, operators worldwide or considering the deployment of an IMS based solution. This chapter explains the technical details of VoLTE solution from an industry perspective.
5-2
VoLTE Architecture
The Figure 5.1 shows the important elements in the VoLTE architecture. The architecture shows a scenario where the LTE network is deployed as a separate PS network. The IMS network is deployed as an overlay to the LTE network and it provides the basic call origination/termination functionalities as well as value added services like Presence, Instant messaging etc. The user after obtaining an IP address from the LTE network performs a registration operation with the IMS network which enables the users to get access to the basic services like voice and also other value added services based on subscription. The Table 5.1 gives the relevant protocols and interfaces for VoLTE solution [15].
50
Technical Details of Voice over LTE via IMS based solution - An Operator Perspective
Figure 5-1: VoLTE Architecture
Nodes MME HSS PCRF P-CSCF I/S-CSCF HSS I/S-CSCF AS P-CSCF I/S-CSCF
Interfaces S6a Rx Cx ISC Mw
Protocols Diameter Diameter Diameter SIP SIP
Table 5-1: VoLTE Relevant Interfaces and Protocols
In the above architecture, LTE is deployed as a standalone network and there is no integration with the 2G/3G networks. During the initial roll out of LTE, the coverage will be minimum and hence there should be some way of integrating the LTE network with the 2G/3G network. This integration is quite challenging owing to the fact that LTE has a completely packet switched architecture. As stated earlier, the voice in LTE is carried as VOIP packets. When the user is roaming outside of LTE coverage i.e. in 2G/3G domains the voice call needs to be switched from VoIP based to legacy TDM
5-3 Options for integrating LTE with existing CS/PS networks
51
based call. Hence there are a few possible solutions for integrating LTE and 2G/3G networks which are explained in the following sections.
5-3
Options for integrating LTE with existing CS/PS networks
During initial LTE deployments, the coverage is going to be limited and hence it is required to integrate the LTE network with the existing 2G/3G network. When loss of LTE coverage is detected, the user should be able to attach to 2G/3G network. If there is an ongoing voice call in the LTE network, then a handover needs to be performed to the 2G/3G network without interruption of voice call. There are two architectures for integration of LTE with 2G/3G networks which are seen as a possible approaches for achieving voice call continuity between LTE and 2G/3G networks. They are as follows: Independent PS based solution. Enhanced Single Radio Voice call continuity(SRVCC) / IMS Centralized Services(ICS). 5-3-1 Independent PS based solution
In PS based solution, voice over IMS is implemented in both the LTE and 3G networks. During loss of LTE coverage, a PS handover is performed towards the 3G network thereby providing seamless mobility between LTE and 3G networks. Thus both the voice and data sessions that are active in the LTE network are simultaneously transferred to the 3G network, there by preventing loss of voice/data during the loss of LTE coverage. The Figure 5.2 shows the architecture of PS based solution. The handover procedure is dened in [16]. The overview of the procedure is briey summarized below: The user is initially attached to the LTE network and a voice call is established via IMS in the LTE network. When loss of LTE coverage is detected, the E-Node B in the LTE network initiates an handover towards the MME which then forwards the same to the SGSN. The MME also separates the voice bearers from the non voice bearers and performs a mapping between the LTE bearer and 3G PDP context. The target SGSN reserves the necessary resources in the 3G network and also creates a session request towards the serving gateway. The SGSN reverts back to the MME on successful completion of the reservation procedure. The MME in the LTE network then performs the handover execution procedure by sending handover command towards the E-Node B The E-Node B then sends a handover command to the UE containing the radio access network parameters of the target 3G network.
52
Figure 5-2: LTE-3G Integrated architecture
The UE can continue the voice session after successful completion of the handover procedure.
Advantages and Limitations
The major advantage is, it is seen as the simplest solution for integrating LTE with 3G network as it involves minimum changes in terms of network architecture. The existing PS network for 3G can be reused easily without any major upgrades. 3G network has PS based capabilities and hence handover of voice from LTE to 3G can be accomplished easily via IMS without signicant interruption.
The independent PS based solution cannot be taken as a target solution as it requires complete coverage of 3G network. Since both voice and data are carried in the 3G PS network, higher bandwidth is required. Hence the advanced release of UMTS which is HSPA+ is needed to support high data rates for carrying both data and voice in the network simultaneously. The existing CS network is not reused in this scenario which can be a major factor in the future when the legacy networks like 2G become obsolete.
5-3 Options for integrating LTE with existing CS/PS networks
53
5-3-2
Enhanced Single Radio Voice call continuity (SRVCC) / IMS Centralized Services (ICS)
The integration of the LTE network with the 2G/3G network based on Enhanced Single Radio Voice Call Continuity (SR-VCC)/IMS Centralized Services architecture is shown in Figure 5.4
Figure 5-3: SRVCC/ICS Architecture
Enhanced SRVCC
In the Enhanced SR-VCC based approach dened in [16], the call control of the LTE network lies within IMS network. The Service Control and Centralization Application server (SCC AS) in the IMS network is the responsible element for anchoring the call in IMS. The SIP signalling messages from the user attached to the LTE network and the destination user is relayed via the SCC AS. The mobile is also assigned a Session Transfer Number for SRVCC (STN-SR) by the SCC AS during the initial registration and is used during handover of the call from the LTE to 2G/3G network. The handover procedure dened in [17] for enhanced SRVCC is briey explained as follows: The E-Node B initiates a handover procedure towards the MME when a loss of LTE coverage is detected based on the measurement reports from the UE. The MME splits the voice and data bearers and initiates a handover procedure towards the Enhanced MSC server in the CS domain.
54
The Enhanced MSC server in the CS domain is responsible for reservation of bearers in the CS domain. This is done by forwarding the handover request message to the target MSC server to which the LTE user will be registered in the CS domain. In addition, the enhanced MSC server initiates the transfer of the call in the IMS domain, by using the STN-SR. The SCC AS in the IMS domain executes the session transfer procedure in IMS domain and the media bearer is switched towards the CS domain. After a successful completion of the access transfer procedure in the CS domain, the Enhanced MSC server indicates the successful completion of the procedure to the MME. The MME sends a handover command to the UE via the E-Node B and the UE attaches to the CS domain by following the CS domain attach procedure and the call ow is switched to the MSC Server/Media Gateway in the CS domain.
IMS Centralized Services
IMS Centralized services (ICS) dened in [17] is an extension of the Enhanced - SRVCC and leads towards complete integration of 2G/3G networks with LTE network. The ICS architecture is same as shown in Figure 5.4. The dierence lies in the fact that in ICS based approach calls that are originating in 2G/3G network i.e. calls from legacy mobiles are also anchored at IMS network. Since, the call control for both the CS (2G/3G) and PS (LTE) domains is within the IMS, seamless handover of users between the 2G/3G and LTE networks can be facilitated easily.
Advantages and Limitations
The main advantage of using the SRVCC/ICS based approach is it facilitates the integration of the legacy CS networks with the LTE network. During the initial roll out of LTE when the coverage is minimum, the SRVCC based approach enables easy roaming between LTE and 2G/3G domains. It also enables the users to experience the same services independent of the access network to which the user is connected. Lastly, IMS Centralized Services approach is the way towards the future when the legacy networks like 2G/3G become obsolete and a common infrastructure would save lot of costs in operating these networks. The drawback of SRVCC/ICS based approach is, it requires signicant upgrades in the existing CS networks. In the CS domain, elements like MSC server are to be upgraded which involve signicant costs. The other major limitation is the long handover time when a user moves from LTE to 2G domain which causes a signicant disruption when there is an ongoing call in the LTE network.
Chapter 6 Conclusion
6-1
Conclusion
In this thesis, the performance of voice over LTE is analysed when the transport network is congested with data trac. The analysis was carried out using OPNET simulation tool. The LTE model in OPNET had signicant limitations in the classication of bearers in the transport network. This led to a situation where there was no prioritization of the bearers in the transport network. Hence to being with, various functions in the process models of the E-Node B and EPC were modied to achieve proper classication of bearers in the transport network. The modication was performed such that each bearer in the LTE network will be mapped to a specic DSCP in the IP header. This enabled us to do classication of IP packets for dierent services like voice and FTP in the transport network. The importance of classication of voice and data trac in the transport network is realized from the results of Section 4.2. Without proper classication, we see that there is a 50 percent increase in the packet end to end delay for voice even when there is no congestion in the transport network. In Sections 4.3 and 4.4, the role of scheduling algorithms on the performance of voice and data was analysed. We see that the capacity for voice users is higher when there is absolute priority for voice in the transport network. But there is a drawback of using Strict Priority for voice, as there is a signicant degradation in the performance of data trac at times of congestion. The Weighted Fair and Weighted Round Robin algorithms were used to overcome this drawback. A comparative analysis was carried out to understand the performance of voice when these scheduling algorithms are implemented in the transport network. We see that the capacity of voice users in the network is reduced when the voice bearers are mapped to a AF service class in the transport network. This is mainly due to the
56
Conclusion
reduced bandwidth allocated for voice users in the AF class than in the EF class. There is a trade-o between the performance of voice and data trac, depending on the type of classication in the transport network. Finally, we present a dierent approach where voice users are mapped into two priority levels and mapped to both EF and AF classes in the transport network. With this approach, we see that we can add more voice users in the network within the acceptable QoS levels than mapping voice into a single EF or AF service class in the transport network. At times of high congestion, there is a signicant reduction in the performance of voice users belonging to the AF service but in a more controlled manner. Such an approach enables the operators to oer dierent levels of service quality for voice users. It also enables the operators to drop calls belonging to normal service class when there is heavy congestion the core network.
6-2
Future Work
In our thesis, we have used only the VoIP and FTP trac models to analyse the performance of VoIP in the network. LTE supports very high data rates and hence services like Video streaming, Interactive gaming can also be used by the mobile users. In the current LTE specications, there are dierent bearers that have dened for each service as seen in Table 2.1. But the mapping of these bearers to IP based QoS is also important for classication in the transport network. In the future, when there are no CS based networks like GSM and voice is carried entirely over PS based networks like LTE, there will be a signicant impact on the delivery of voice when multiple services are present in the network. In such scenarios, when there is multiple level of classication for dierent types of services, each type of service has dierent QoS requirements and mapping them to IP based QoS in the transport network needs to be done carefully. This will be an interesting area to investigate as the role of scheduling algorithms in the transport network become much more important owing to the fact that over provisioning of bandwidth for one type of service has a direct impact on the performance of other service. The usage of admission control is another area that needs to be investigated. Most of the studies that have been done in this area are focussed on the radio network i.e. the use of admission control is studied when there is congestion in the radio network. But when admission control takes into account the availability of resources in both the radio and core networks, ecient link usage in the transport network can be achieved without increasing the link capacity in the transport network.
Bibliography
[1] 3GPP Technical Specication 23.272, "Circuit Switched (CS) fallback in Evolved Packet System (EPS)", Stage 2 (Release 10); http://www.3gpp.org, 2011. [2] Alcatel-Lucent Strategic White Paper, "Options for Providing Voice over LTE and Their Impact on the GSM/UMTS Network"; www.alcatel-lucent.com, August 2009. [3] Siomina, I.; Wanstedt, S.; , "The impact of QoS support on the end user satisfaction in LTE networks with mixed trac," IEEE 19th International Symposium on Personal, Indoor and Mobile Radio Communications, pp.1-5, 15-18 Sept. 2008. [4] Zaki, Y.; Weerawardane, T.; Gorg, C.; Timm-Giel, A., "Multi-QoS-Aware Fair Scheduling for LTE," IEEE 73rd Vehicular Technology Conference (VTC Spring) vol., no., pp.1-5, 15-18 May 2011. [5] Puttonen, J.; Henttonen, T.; Kolehmainen, N.; Aschan, K.; Moisio, M.; Kela, P.; , "Voice-Over-IP Performance in UTRA Long Term Evolution Downlink," IEEE Vehicular Technology Conference, vol., no., pp.2502-2506, 11-14 May 2008. [6] Yasir Zaki, Nokila Zahariev, Thushara Weerawardane, Carmelita Grg and Andreas Timm-Giel, "Optimized Service Aware LTE MAC Scheduler: Design, Implementation and Performance Evaluation", OPNET workshop, Washington, D.C., August 29September 1, 2011. [7] Li, X.; Toseef, U.; Weerawardane, T.; Bigos, W.; Dulas, D.; Goerg, C.; Timm-Giel, A.; Klug, A.; , "Dimensioning of the LTE S1 interface," Third Joint IFIP Wireless and Mobile Networking Conference (WMNC), vol., no., pp.1-6, 13-15 Oct. 2010. [8] Ekstrom, H.; , "QoS control in the 3GPP evolved packet system," IEEE Communications Magazine , vol.47, no.2, pp.76-83, February 2009. [9] 3GPP Technical Specication 23.203, "Policy and charging control architecture (Release 11)", www.3gpp.org, 2012
58
Bibliography
[10] 3GPP Technical Specication 23.228, "IP Multimedia Subsystem (IMS); Stage 2 (Release 11) http://www.3gpp.org, 2012. [11] S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, W. Weiss, "An architecture for Dierentiated Services", "Request for Comments 2475, Internet Engineering Task Force", December 1998. [12] A. Demers, S. Keshav, and S. Shenker "Analysis and simulation of a fair queueing algorithm", In Symposium proceedings on Communications architectures and protocols ", ACM, New York, NY, USA, 1989. [13] OPNET Modeller, www.opnet.com accessed on December 2011. [14] ITU-T Recommendation G.107, "The E-Model, a computational model for use in transmission planning", 2011. [15] 3GPP Technical Specication 23.401, "General Packet Radio Service (GPRS) enhancements for Evolved Universal Terrestrial Radio Access Network (E-UTRAN) access " Stage 2 (Release 10), http://www.3gpp.org, 2011. [16] 3GPP Technical Specication 23.216, " Enhanced Single Radio Voice Call Continuity (SRVCC),Stage 2(Release 11) " http://www.3gpp.org;, 2012. [17] 3GPP Technical Specication 23.292, " IMS Centralized Services Stage 2(Release 11) " http://www.3gpp.org, 2012.
Glossary
List of Acronyms
3GPP ARP BGCF CS Diserv
Third Generation Partnership Project Allocation and Retention Priority Breakout Gateway Control Function Circuit Switched Dierentiated Services (Evolved Universal Terrestrial Radio Access Network)
E-UTRAN EPC FIFO GTP HSS HSPA I-CSCF ICS IMS LTE MOS MSC MIMO MGCF
Evolved Packet Core First In First Out GPRS Tunnelling Protocol Home Subscriber Server High Speed Packet Access Interrogating Call Session Control Function IMS Centralized Services IP Multimedia Subsystem Long Term Evolution Mean Opinion Score Mobile Switching Centre Multiple Input Multiple Output Media Gateway Control Function
60
Glossary
MME MRF PDN PCRF P-CSCF PDV PS PLR QCI QoE RCS RTP S-CSCF SCC AS TFT VoLTE WRR
Mobility Management Entity Media Resource Function Packet Data Network Proxy and Charging Rules Function Proxy Call Session Control Function Packet Delay Variation Packet Switched Packet Loss Rate QoS Class Identier Quality of Experience Rich Communication Suite Real Time Protocol Serving Call Session Control Function Service Control and Centralization Application Server Trac Flow Template Voice over LTE via IP Multimedia Subsystem Weighted Round Robin

04 Voice Over LTE

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

04 Voice Over LTE

Uploaded by

Copyright:

Available Formats

Voice over LTE

PRASANNA GURURAJ RAGHAVENDRARAO

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Voice over LTE

Prasanna Gururaj Raghavendrarao 29.6.2012

Master of Science Thesis

Prasanna Gururaj Raghavendrarao

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Delft University of Technology Department of Telecommunications

Dated: 29.6.2012 Supervisors: dr.ir. Jos Adema (KPN)

ir. Gerard Fossung (KPN)

dr. R.R. Venkatesha Prasad Readers: dr.ir.Jos Weber

dr.ir. Bert Jan Kooij

Master of Science Thesis

Prasanna Gururaj Raghavendrarao

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Weighted Round Robin (WRR) Scheduling . . . . . . . . . . . . . Weighted Fair Scheduling . . . . . . . . . . . . . . . . . . . . . .

Prasanna Gururaj Raghavendrarao

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Master of Science Thesis

Prasanna Gururaj Raghavendrarao

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Master of Science Thesis

Prasanna Gururaj Raghavendrarao

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Master of Science Thesis

Prasanna Gururaj Raghavendrarao

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Prasanna Gururaj Raghavendrarao

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

Solutions for Supporting Voice over LTE

Figure 1-1: Circuit Switch Fallback [2]

1-2 Motivation for the Thesis

Figure 1-2: VoLTE [2]

Motivation for the Thesis

1-4 Related Work

Organization of the Thesis

Prasanna Gururaj Raghavendrarao

Master of Science Thesis

LTE Network Architecture

Figure 2-1: LTE Network Architecture

2-3 QoS Architecture in LTE

QoS Architecture in LTE

Figure 2-2: QoS Architecture in LTE [8]

Packet Delay Budget

Packet Error Loss Rate 102 103

Non-GBR Non-GBR Non-GBR Non-GBR Non-GBR

100 300 100 300 300

106 106 103 106 106

Table 2-1: EPS QOS Bearer Denitions [9]

2-4 IMS Network Architecture

IMS Network Architecture

Figure 2-3: IMS Network Architecture