VoIP Jitter in 3GPP Long Term Evolution Networks

Transcription

1 INSTITUTO TECNOLÓGICO Y DE ESTUDIOS SUPERIORES DE MONTERREY CAMPUS MONTERREY PROGRAMA DE GRADUADOS EN MECATRÓNICA Y TECNOLOGÍAS DE INFORMACIÓN VoIP Jitter in 3GPP Long Term Evolution Networks by Christian Alberto Rodríguez García Thesis Presented as a partial fulfillment of the requirements for the degree of Master of Science in Electronic Engineering Major in Telecommunications Monterrey, N.L. December 2009

2 Instituto Tecnológico y de Estudios Superiores de Monterrey Campus Monterrey División de Mecatrónica y Tecnologías de Información Programa de Graduados The members of the thesis committee hereby approve the thesis of Christian Alberto Rodríguez García, B.S. as a partial fulfillment of the requirements for the degree of Master of Science in: Electronic Engineering Major in Telecommunications Thesis Committee: David Muñoz Rodríguez, Ph.D. Thesis Advisor César Vargas Rosales, Ph.D. Synodal Gabriel Campuzano Treviño, Ph.D. Synodal Joaquín Acevedo Mascarúa, Ph.D. Director of the Graduate Program December 2009

3 To my family, Cristina García González, Juan José Rodríguez Uc and Karla Brisol Rodríguez García

4 ACKNOWLEDGMENTS This work is devoted with affection to my parents, María Cristina and Juan José, for their unconditional support along my life. Without your love, guidance, and comprehension, I had never made it - thanks will never suffice. To my dear sister, Karla Brisol, for being more than my best friend. To all and every one of my family members, especially to my grandfathers, Trinidad González Anaya, Lucino García Ochoa, María Elena Catzín and Eladio Rodríguez for being an inspiration in my life. To my family in heaven that always encouraged me. I would like to express my gratitude to my thesis advisor David Muñoz Rodríguez, Ph.D. for his professional advice. Because, without his guidance this thesis would not have been possible. I also want to thank to my professor, and friend, Alejandro Aragón Zavala, Ph.D. for instilling in me the passion for telecommunications. Finally but not the last, I want to thank God for giving me strength, patience, and the wonderful opportunity to be alive. Christian Alberto Rodríguez García December 2009 V

5 VoIP Jitter in 3GPP Long Term Evolution Networks Christian Alberto Rodríguez García, B.S. INSTITUTO TECNOLÓGICO Y DE ESTUDIOS SUPERIORES DE MONTERREY, 2009 Thesis advisor: David Muñoz Rodríguez, Ph.D. 3GPP LTE is the next step towards 4G mobile communications with performance comparable to wire-line networks. Careful planning and design must be carried out to assure a successful deployment for both, users and network operators. LTE must be able to adapt to a variety of traffic such as data, voice, and video. Services currently provided through circuit-switched systems are expected to have a similar equivalent in LTE, an all-ip based network. Voice is the most widespread service and represents the main revenue for network operators. Users expect at least the same Quality of Service provided by CS networks while operators look for an increase in capacity and reduction of costs. Such objectives can be reached through Voice-over-IP (VoIP). This service has the following characteristics: Bursty low bitrate traffic, strict packet delay-based QoS, and a high number of simultaneous users. Furthermore, VoIP is highly sensible to jitter. Jitter is a common issue in packet-switched networks, where packets arrive at random times at the receiver. For voice services, it implies disruptions in speech intelligibility and poor QoS. This master thesis studies the jitter phenomenon for the LTE downlink, where bottlenecks arise naturally due to user queues, when it is operated under VoIP traffic. Specifically the impact on jitter due to network congestion, retransmissions and different modulation and coding schemes, for diverse radio channel conditions, are analyzed. Abstract

6 CONTENTS Acknowledgments... V List of Figures... X List Of Tables... XII Chapter 1 Introduction Problem Description Objective Justification Contribution Thesis Organization... 4 Chapter 2 3GPP Long Term Evolution The standardization process Design Targets Architecture LTE Physical Layer Bandwidths, frequency bands and duplexing OFDM Physical Resources Physical signals Cell-Specific Downlink Reference Signals Synchronization Signals Downlink L1/L2 control signaling Link Adaptation Modulation and Coding Scheme Scheduler VII

7 2.8.1 Channel-status reports Hybrid-ARQ Chapter 3 Voice-over-IP VoIP Codecs Quality Criteria VoIP traffic model Generating VoIP traffic VoIP Traffic Simulator Chapter 4 Introduction to Jitter The Jitter Concept LTE Jitter Sources Scheduler Buffer HARQ retransmissions Radio Link Control Functions Mobility Other Jitter Sources Jitter management Jitter buffer Scheduler strategies Chapter 5 VoIP Jitter in LTE Simulation Scenario Simulation Description LTE Physical layer LTE MAC Protocol Simulation results SNR = 5 db SNR = 8 db SNR = 12 db Chapter 6 Conclusions and Future Work General Conclusions Future Work VIII

8 Appendix A Multi-carrier transmission A.1 The Muti-carrier Concept A.1.1 Channel Capacity A.1.2 Wider bandwidths A.2 Multi-carrier transmission A.3 OFDM as a multi-carrier transmission A.3.1 OFDM implementation using IDFT/DFT A.3.2 Cyclic-Prefix A.3.3 OFDM Subcarrier Spacing A.3.4 Number of subcarriers Vita IX

9 LIST OF FIGURES Figure 2.1: LTE Architecture... 8 Figure 2.2: DL LTE protocols... 9 Figure 2.3: FDD and TDD Figure 2.4: OFDM concepts Figure 2.5 OFDM and OFDMA Figure 2.6: Time-domain structure Figure 2.7: Time slot Figure 2.8: Resource Block for normal-cp Figure 2.9: Reference Symbols in a subframe Figure 2.10: Synchronization Signals Figure 2.11: L1/L2 control region Figure 2.12: LTE rate control Figure 2.13: Constellation diagrams Figure 2.14: Scheduling units Figure 2.15: Channel dependent scheduling Figure 2.16: Soft combining Figure 3.1: VoIP Codecs: AMR-NB and AMR-WB Figure 3.2: VoIP packets and SID packets Figure 3.3: Voice Quality (Source: ITU) Figure 3.4: VoIP Traffic Model Figure 3.5: Inverse Discrete Transform Figure 4.1: Jittered packets Figure 4.2: Resource scheduler Figure 4.3: HARQ retransmission for the DL Figure 4.4: Jitter buffer Figure 5.1: Simulation Scenario Figure 5.2: LTE downlink PHY structure Figure 5.3: Scheduler flow diagram Figure 5.4: L1/L2 Control Region, RS, and Data Region Figure 5.5: BLER curves obtained from SISO AWGN simulations for all 15 CQI values. From CQI 1 (leftmost) to CQI 15 (rightmost) X

10 Figure 5.6: SNR-CQI mapping Figure 5.7: Jitter behavior (SNR = 5 db) Figure 5.8: Jitter behavior (SNR = 8 db) Figure 5.9: Jitter behavior (SNR = 12 db) Figure 5.10: Jitter cell profile (SNR = 5 db) Figure 5.11: Jitter cell profile (SNR = 8 db) Figure 5.12: Jitter cell profile (SNR = 12 db) Figure A.1: Operation regions Figure A.2: Subcarrier spacing Figure A.3: a) FDM, b) OFDM Figure A.4: OFDM modulation and demodulation Figure A.5: Digital implementation of OFDM Figure A.6: There is no intra-cell interference for OFDM Figure A.7: Corruption due to time dispersion Figure A.8: Cyclic-prefix insertion XI

11 LIST OF TABLES Table 2.1: LTE Frequency bands Table 2.2: LTE Resource configuration Table 3.1: VoIP traffic model parameters Table 3.2: VoIP Traffic Simulation Results Table 5.1: Simulation parameters Table 5.2: CQI and MCS recommendations Table 5.3: Cell jitter; SNR = 5 db Table 5.4: Cell jitter; SNR = 8 db Table 5.5: Cell jitter; SNR = 12 db XII

12 Chapter 1 INTRODUCTION With more than 2 billion users around the world, there is no doubt that 2G and 3G UMTS cellular technologies are a complete success adopted by most countries and mobile network operators [1]. The first release, published in 1999, considered a circuit-switched (CS) data network, establishing a dedicated channel between transmitter and receiver. Later on, the standard considered a packetswitched (PS) cellular network known as HSPA, but still supporting CS services. The latest release of the UMTS wireless technology is the so-called 3GPP Long Term Evolution (LTE), an all-ip network. Initiated in 2004 by the 3rd Generation Partnership Project (3GPP), the Long Term Evolution (LTE) project focused on enhancing the Universal Terrestrial Radio Access Network (UTRAN) and optimizing 3GPP s radio access architecture. Targets were to have peak data rates of 100 Mbps in the downlink and 50 Mbps in the uplink. Orthogonal Frequency Division Multiple Access (OFDMA) and Single- Carrier Frequency Division Multiple Access (SC-FDMA) were selected as the multiple access technologies for the DL and UL respectively. The defined data modulation schemes are QPSK, 16QAM, and 64QAM 1 for both DL and UL. Furthermore, Multiple-Input Multiple-Output (MIMO) antenna technology is also supported, increasing capacity. LTE is extremely flexible, using a number of defined channel bandwidths between 1.4 and 20 MHz (contrasted with UTRA s fixed 5 MHz channels). To suit as many frequency band allocation arrangements as possible, both paired (FDD) and unpaired (TDD) band operation is supported. LTE can co-exist with earlier 3GPP radio technologies, even in adjacent channels, and calls can be handed over to and from all 3GPP s previous radio access technologies. 1 Optional for the uplink 1

13 Chapter 1. Introduction 2 The LTE architecture has been greatly simplified compared to past 3GPP's technologies, turning the hierarchical structure into a flat structure. All the user functionality is centralized in a single entity, the so-called evolved-nodeb. This design has several advantages: reduces the Round-Trip delay Time (RTT), scheduling decision are made faster (1 ms), and coordination among entities is improved. LTE is an all-ip network; in other words, only packet-switched services are supported. While data traffic and its corresponding revenue are increasing, the voice service still makes the majority of operators income. Therefore, LTE is designed to support not only data services efficiently, but also good quality voice service with high efficiency. As LTE radio only supports packet services, the voice service will also be Voice over IP (VoIP), not CS voice [2]. The use of VoIP instead of CS voice represents savings for operators, since the CS related part of the network will not be needed anymore. It is expected that VoIP can bring better capacity than CS voice due to more efficient utilization of resources. 1.1 PROBLEM DESCRIPTION Voice-over-IP represents savings for users and network operators. However, supporting VoIP in packet-switched mobile networks faces certain challenges due to its strict delay requirements and jitter sensibility. Jitter is the variation of delay, where packets arrive at random times at the receiver. In other words, the kth packet is expected to arrive at a time but it is received at, where is jitter. When jitter is constant, it can be filtered out or compensated in a deterministic way. However, it often exhibits a random behavior [3]. Jitter results in speech intelligibility disruptions [4]; hence the end-toend jitter has to be small enough not to be noticeable. 3GPP Long Term Evolution, an all-ip based network, is not exempt from jitter. Hence, research about this phenomenon is necessary to assure the feasibility of VoIP services over LTE.

14 Chapter 1. Introduction OBJECTIVE In order to determine the feasibility of VoIP services over the LTE mobile networks, the purpose of this thesis is to analyze the jitter phenomenon. Particularly the impact on jitter caused by network congestion, retransmissions, and the modulation and coding scheme, for different radio channel conditions is studied. VoIP traffic, physical layer and MAC layer simulations are developed. 1.3 JUSTIFICATION In the packet-switched LTE network, services must be provided in a fast, efficient and reliable way, including services substituting their CS counterparts. Voice services will be offered in the form of Voice-over-IP. Since voice is the most widespread service, special care must be taken to assure a successful deployment of future LTE networks. In the literature exists a variety of studies about VoIP over LTE [5] [6] [7]. Nevertheless, they mainly focus on capacity, coverage, or scheduling issues. However, there are not researches identifying the jitter phenomenon and its behavior. This thesis pretends to research jitter under diverse channel conditions, and its impact on the VoIP QoS. 1.4 CONTRIBUTION 3GPP LTE is the next step towards 4G mobile communications with performance comparable to wire-line networks. Careful planning and design must be carried out to assure a successful deployment for both, users and network operators. LTE must be able to adapt to a variety of traffic such as data, voice, and video. Services currently provided through circuit-switched systems are expected to have a similar equivalent in LTE, an all-ip based network. Voice is the most widespread service and represents the main revenue for network operators. Users expect at least the same Quality of Service provided by CS networks, while operators look for an increase in capacity and reduction of costs. Such objectives can be reached through Voice-over-IP (VoIP). This service has the following characteristics: Bursty low bitrate traffic, strict packet delay-based

15 Chapter 1. Introduction 4 QoS, and a high number of simultaneous users. Furthermore, VoIP is highly sensible to jitter. Jitter is a common issue in packet-switched networks, where packets arrive at random times at the receiver. For voice services, it implies disruptions in speech intelligibility and poor QoS. This master thesis studies the jitter phenomenon for the LTE downlink, where bottlenecks arise naturally due to user queues, when it is operated under VoIP traffic. Specifically the impact on jitter due to network congestion, retransmissions, and different modulation and coding schemes, for diverse radio channel conditions, are analyzed. 1.5 THESIS ORGANIZATION The thesis structure is described now. An overview of LTE is presented in Chapter 2. The discussion begins exposing the architecture and design targets established by 3GPP. Then, the main technologies necessary to support the outstanding key features of LTE are introduced. Focus is made on OFDM, link adaptation, scheduling, and the HARQ retransmission scheme; key elements in the performance of VoIP. The VoIP concept is analyzed in Chapter 3. First, the AMR voice codec used in LTE is described. Then, the quality criterion for VoIP services is presented. Further discussion focus on the VoIP traffic model and its implementation. Chapter 4 provides a description of the jitter phenomenon under LTE. In Chapter 5, the performance of LTE under VoIP traffic is tested through simulations for different channel conditions. It will be shown that LTE, as an all-ip network, will be able to offer VoIP services successfully as long as the number of users in the cell can be estimated correctly. Final conclusions are presented in Chapter 6. Further research under the same line of study is also proposed. Finally, Appendix A offers an explanation of the multi-carrier and OFDM concepts.

16 Chapter 2 3GPP LONG TERM EVOLUTION 3GPP Long Term Evolution is the next step towards 4G mobile communications. Higher user data rates, increased capacity, and reduced delay/jitter, are some of the driving forces behind the evolution of the Universal Terrestrial Radio Access Network (UTRAN). This Chapter provides the background necessary to comprehend LTE. Both, the architecture and air interface are presented. 2.1 THE STANDARDIZATION PROCESS LTE consists of a series of standards and specifications defined by the 3rd Generation Partnership Project (3GPP). A clear understanding of the standardization process shows, for example, why certain air interfaces were chosen as part of the standard instead of any other alternative. The process is described now: Standardization starts with the requirement phase, where the standardization body decides what should be achieved with the standard. In the architecturephase, the main architecture is decided, i.e., how to meet the requirements. The interfaces and technologies are proposed. For the detailed specification phase, the parameters for the architecture are detailed. Finally, in the testing and verification phase, the interfaces are proved to work as expected. This is an iterative process since any phase can directly affect the others. 5

17 Chapter 2. 3GPP Long Term Evolution DESIGN TARGETS Initiated in 2004, the Long Term Evolution project focused on enhancing the Universal Terrestrial Radio Access Network (UTRAN) and optimizing 3GPP's radio access architecture. The design targets were [5]: 1. Support scalable bandwidths a) 1.25, 2.5, 5.0, 10.0 and 20 MHz Peak data rate that scales with system bandwidth. a) DL (2 Ch. MIMO) peak rate of 100 Mbps in 20 MHz channel. b) UL (1 Ch. TX) peak rate of 50 Mbps in 20 MHz channel. 3. Supported antenna configurations. a) DL: 4x2, 2x2, 1x2, 1x1. b) UL: 1x2, 1x1. 4. Spectrum efficiency. a) DL: 5 bit/s/hz (3 to 4 x HSPA Rel. 6) b) UL: 2.5 bit/s/hz (2 to 3 x HSPA Rel. 6) 5. User throughput. 6. Latency 7. Mobility a) DL average: bit/s/hz (3 to 4 x HSPA Rel. 6 ) b) UL average: bit/s/hz (2 to 3 x HSPA Rel. 6) c) DL cell edge 2 : bit/s/hz (2 to 3 x HSPA Rel. 6) d) UL cell edge : bit/s/hz (2 to 3 x HSPA Rel. 6 ) a) Control-plane < ms: Delay generated for transiting from a non-active state to an active-state, where the terminal is able to send/receive data. There are two measures. The first one corresponds to the transition from a camped state, where the user terminal is unknown to the RAN (100 ms). The other measure is the transition from a dormant state, where the user terminal is known by the RAN but radio resources have been not assigned (50 ms). b) User-plane < 5 ms: The user-plane latency requirement is expressed as the time it takes to transmit a small IP packet from the terminal to the RAN edge node or vice versa in an unloaded network. 1 2 Final specifications consider 1.4, 3, 5, 10, 15 and 20 MHz. 5 th percentile - 95% of the users have better performance

18 Chapter 2. 3GPP Long Term Evolution 7 a) Optimized for low speeds (<15 Km/h) b) High performance at speeds up to 120 Km/h 8. Coverage c) Maintain link at speeds up to 350 Km/h (500 Km/h for certain frequencies) a) Full performance up to 5 Km b) Slight degradation at 5 Km - 30 Km c) Operation up to 100 Km is not precluded by the standard Additionally, the related VoIP service requirements are: 1. The E-UTRA should efficiently support various types of service. These must include currently available services like web-browsing, FTP, video-streaming or VoIP, and more advanced services (e.g. real-time video or push-to-talk) in the packet-switched domain. 2. VoIP should be supported with at least as good radio backhaul efficiency and latency as voice over UMTS circuit-switched (CS) networks. 3. Voice and other real-time services supported in the CS domain in Release 6 shall be supported by E-UTRAN via the packet switched domain with at least equal quality as supported by UTRAN (e.g. in terms of guaranteed bit rate) - over the whole speed range. 2.3 ARCHITECTURE The LTE architecture has been greatly simplified compared to past 3GPP's technologies. An all-ip flat architecture has been adopted to support the outstanding design targets. The main entities and interfaces are shown in Figure 2.1. A lot of functionalities, which in past 3GPP's architectures were placed in different entities, have been centralized in the enodeb (base station). A new interface called X2 connects the enodebs, enabling direct communication between them. The E-UTRAN 3 is connected to the Evolved Packet Core (EPC) through the S1 interface which connects the enodebs to the Mobility Management Entities (MME) and the Serving Gateway (S-GW or SAE Gateway) through a many to many relationship. 3 E-UTRAN is the official standard's name for LTE, the entire radio network.

19 Chapter 2. 3GPP Long Term Evolution 8 enb: Enhanced Node B, or base MME/SAE Gateway MME/SAE Gateway station UE: User Equipment EPC: Evolved Packet Core o MME: Mobility Management Entity (Control Plane) enb enb E-UTRAN o SAE: System Architecture Evolved (User Plane) E-UTRAN: Evolved Universal Terrestrial Radio Access Network enb UE Figure 2.1: LTE Architecture The radio protocol architecture of E-UTRAN is specified for the control-plane and user-plane. The control plane performs the radio resource control (RRC). The user-plane is divided in protocols with the following functions (see Figure 2.2): Packet Data Convergence Protocol (PDCP) performs IP header compression to reduce the number of bits over the air interface. It also handles the ciphering/deciphering and integrity functions. Radio Link Control (RLC) is responsible for segmentation/concatenation, RLC retransmission handling, and in-sequence delivery to higher layers. Medium Access Control (MAC) handles the HARQ retransmissions, scheduling for DL and UL, link adaptation, etc. Physical Layer (PHY) is responsible for coding/decoding, modulation/demodulation, multi-antenna mapping, and other typical physical layer functions.

20 Chapter 2. 3GPP Long Term Evolution 9 IP packet IP packet User#i User #j SAE bearers PDCP Header compression PDCP Header compression Ciphering Deciphering MAC Payload selection Priority handling, payload selection Retransmission control MAC scheduler RLC Segmentation, ARQ MAC multiplexing Hybrid ARQ Radio bearers Logical channels RLC MAC Reassembly, ARQ MAC demultiplexing Hybrid-ARQ Redundancy version PHY Coding Transport channel PHY Decoding Modulation scheme Antenna and resource assignment Modulation Antenna and resource mapping Demodulation Antenna and resource demapping enodeb Mobile terminal (UE) Figure 2.2: DL LTE protocols 2.4 LTE PHYSICAL LAYER BANDWIDTHS, FREQUENCY BANDS AND DUPLEXING LTE can be operated in different bandwidth sizes. The main reason for this is that the amount of spectrum available depends on the frequency band and the particular operator's situation. Originally it was stated in [6] as a list of LTE spectrum allocations from 1.25 to 20 MHz, although final specifications consider only 1.4, 3, 5, 10, 15 and 20 MHz. Pair and unpair spectrum, i.e. FDD and TDD modes, are supported. Frequency Division Duplex (FDD) entails that downlink and uplink take place in

21 Chapter 2. 3GPP Long Term Evolution 10 different, sufficiently separated, frequency bands. Time Division Duplex (TDD) implies that downlink and uplink transmission take place in different nonoverlapping slots. Figure 2.3 shows this concept. Figure 2.3: FDD and TDD Table 2.1: LTE Frequency bands Band UL Range (MHz) DL Range (MHz) Mode Main Region (s) FDD Europe, Asia FDD Americas (Asia) FDD Europe, Asia (Americas) FDD Americas FDD Americas FDD Japan FDD Europe, Asia FDD Europe, Asia FDD Japan FDD Americas FDD Japan FDD Americas FDD Americas FDD Americas FDD TDD Europe, Asia (not Japan) TDD Europe Asia TDD TDD TDD TDD Europe TDD China TDD Europe, Asia

22 Chapter 2. 3GPP Long Term Evolution 11 LTE can be deployed in current cellular frequency bands (IMT-2000) and new frequency allocations as they become available. For instance, the 700 MHz frequency band previously used for analog television in the United States, it is now considered a potential band for LTE operation. The identified frequency bands by 3GPP are shown in Table 2.1 [6] OFDM Orthogonal Frequency Division Multiplexing has been chosen as the downlink transmission scheme fo The subcarrier spacing is Δf = 15 KHz 4. Likewise the OFDM symbol duration is Tu=1/Δf = 66.7 µs. Both co of an OFDM signal is given by the relationship BW = Nc. Δf, where Nc is thenumberofsubcarri Δf = 1/T U Pulse shape T u = 1/Δf a) Subcarrier spacing b) OFDM symbol duration Figure 2.4: OFDM concepts OFDM can be implemented digitally through the IDFT at the transmitter, and the DFT at the receiver. The FFT size N FFT should be preferably selected asn FFT =2 n for some integer, so OFDM can be performed by means of the efficient radix-2 IFFT/FFT. The sampling rate is given as f s = Δf N FFT = 15,000 N FFT, thus the FFT size must be chosen such that the sampling theorem is satisfied. For example, if LTE is operated in a 20 MHz bandwidth, then NFFT = 2048 and the resulting sampling rate would be MHz. Commonly, specifications express time units relative to the smallest time unit in LTE,T s, corresponding to the sample period of a 20 MHz OFDM signal with a FFT size of That ists = 1/f = 1/ (15 OFDM can be made completely resistant to multi-path delay spread. This is possible because the long symbols used for OFDM can be separated by a guard interval known as the cyclic prefix (CP), where the CP is a copy of the end of the KHz is also considered for multi-cell broadcast messages

23 Chapter 2. 3GPP Long Term Evolution 12 OFDM symbol inserted at the beginning. The CP has been chosen to be slightly longer than the longest expected delay spread in the radio channel. LTE defines two cyclic prefix lengths, the normal CP and the extended CP. Normal CP is the expected operation mode for LTE, and its size has been set at ~4.7 us, enabling the system to cope with path delay variations up to about 1.4 Km. Since the length of an OFDM symbol is ~66.7 us, about a reduction of 6.6% in the effective data rates is experienced. Extended CP is designed to provide robustness against multipath effect in larger cells, and for use with multi-cell broadcast messages. It provides protection for up to 10 Km delay spread with a capacity loss of 20%. Physical resources are organized in both, time and frequency domains. However, traditional OFDM systems split the system bandwidth into diverse frequency bands and assign them to the users indefinitely (similar to FDM). Hence radio resources may not be fully exploited. OFDM can be used as a robust Multiple Access scheme, the so-called Orthogonal Frequency-Division Multiple Access (OFDMA) which incorporates elements of TDMA. OFDMA allows subsets of the subcarriers to be allocated dynamically among the different users on the channel, for every unit of time. The result is a more robust system with increased capacity. This is due to the trunking efficiency of multiplexing low rate users and the ability to schedule users by time and frequency, which provides resistance to multi-path fading. A comparison between OFDM and OFDMA is shown in Figure 2.5. Subcarriers Subcarriers Used 1 Symbols (Tine) Symbols (Tine) User 2 User 3 OFDM OFDMA Figure 2.5 OFDM and OFDMA A drawback of OFDM/OFDMA is that parallel transmission of multiple subcarriers leads to larger variations in the instantaneous transmission power. Thus, multi-carrier transmission will have a negative impact on the transmitter power-amplifier efficiency, implying increased transmitter power consumption and increased power-amplifier cost. This is specially an issue at the mobile terminal. Single-Carrier Frequency Division Multiple Access (SC-FDMA) was selected as the multiple access scheme for the uplink because it shares multi-carrier characteristics, while decreasing the Peak-to-Average Power Ratio (PAPR). A detailed description of SC-FDMA can be found in [1] [8].

24 Chapter 2. 3GPP Long Term Evolution PHYSICAL RESOURCES LTE radio resources are organized as a bi-dimensional time-frequencygrid 5. The largest unit of time in LTE is the 10 ms radio frame, which is further subdivided into ten 1 ms subframes, each of which is split into two 0.5 ms slots (Figure 2.6). Each slot comprises 7 OFDM symbols for normal cyclic prefix operation, and 6 for the extended cyclic prefix case (see Figure 2.7) One radio frame, T f= xTs;= 10 ms One slot, Tslot = 15360xTs; = 0.5 ms #0 #1 #2 #3 #18 #19 One subframe Figure 2.6: Time-domain structure The basic physical resource unit is composed by a subcarrier during an OFDM symbol, the so-called resource element(re). Theoretically the scheduler could assign resources in a per-re basis, increasing flexibility. Albeit, an overwhelming amount of overhead would be required to handle every single resource element, causing a reduction in power efficiency and user data rates. Normal CP Δf=15kHz 5.2 µs 160 samples LTE slot: 0.5 ms samples (Assumed Sampling Frequency f s = MHz) 4.7 µs 144 samples Extended CP Δf = 15kHz Special OFDM symbol: 71.9 µs 2208 samples 66.7 µs 2048 samples 66.7 µs 2048 samples 16.7 µs 512 samples OFDM symbol: 83.3 µs 2560 samples OFDM symbol: 71.3 µs 2192 samples Figure 2.7: Time slot 5 Spatial-domain is also considered for MIMO communications (1 grid per antenna)

25 Chapter 2. 3GPP Long Term Evolution 14 LTE defines a resource block(rb) as 12 consecutive subcarriers (180 KHz) during one time slot (0.5 ms). One slot consists of 7 or 6 OFDM symbols for normal-cp and extended-cp respectively 6. Figure 2.8 illustrates a RB for normal- CP. One resource element QPSK. 2bits 16QAM, 4bits 64QAM, 6bits \f = 15 khz One resource block (12x7 = 84 resource elements) 12 sub-carriers, 180 khz Figure 2.8: Resource Block for normal-cp The number of RBs depends on the total system bandwidth, as shown in Table 2.2. Note that a frequency guard band is considered at the end of the LTE spectrum to avoid out-of-band emissions. For example, for a 5 MHz system bandwidth there are 25 RBs occupying a transmission bandwidth of 4.5 MHz. Table 2.2: LTE Resource configuration System BW 1.4 MHz 3 MHz 5 MHz 10 MHz 15 MHz 20 MHz Subframe-duration 0.5 ms Subcarrier-spacing 15 KHz Sampling frequency 1.92 MHz 3.84 MHz 7.68 MHz MHz MHz MHz FFT size Number of occupied subcarriers Number of RBs Transmission BW 1.08 MHz 2.7 MHz 4.5 MHz 9 MHz 13.5 MHz 18 MHz (Efficiency) (77%) (90%) (90%) (90%) (90%) (90%) Number of OFDM symbols 7/6 per subframe (Normal/Extended) CP length Normal (4.7/9) x 6 (4.7/18) x 6 (4.7/36) x 6 (4.7/72) x 6 (4.7/108) x 6 (4.7/144) x 6 (us/samples) (5.2/10) x 1 (5.2/20) x 1 (5.2/40) x 1 (5.2/80) x 1 (5.2/120) x 1 (5.2/160) x1 Extended (16.7/32) (16.7/64) (16.7/128) (16.7/256) (16.7/384) 16.7/512 6 Hence, a RB is formed by 84 or 72 resource elements.

26 Chapter 2. 3GPP Long Term Evolution 15 A resource block pairis formed by two consecutive time-domain RBs. That is, 12 consecutive subcarriers (180 KHz) along 1 subframe (1 ms). A resource block pair is the minimum resource unit for scheduling purposes. The reason to define a RB in the first place is because certain control signals are mapped in particular RBs. 2.6 PHYSICAL SIGNALS CELL-SPECIFIC DOWNLINK REFERENCE SIGNALS To carry out coherent demodulation of different downlink physical channels, a mobile terminal needs estimates of the downlink channel. More specifically, in case of OFDM transmission, the terminal needs an estimate of the complex channel of each subcarrier. One way to enable channel estimation in case of OFDM transmission is to insert known reference symbols into the OFDM timefrequency grid, the so-called Cell-specific downlink reference signals 7. These reference signals are transmitted in every downlink subframe, and span the entire downlink cell bandwidth [8]. LTE defines four reference symbols per resource block, separated in time and frequency as shown in Figure 2.9. To estimate the channel over the entire time-frequency grid as well as reducing the noise in the channel estimates, the mobile terminal should carry out interpolation/ averaging over multiple reference symbols. Figure 2.9: Reference Symbols in a subframe 7 Additionally, there also exist the UE-specific reference signals (to be used for an explicit UE) and MBSFN reference signals (for multi-cell broadcast).

27 Chapter 2. 3GPP Long Term Evolution SYNCHRONIZATION SIGNALS To assist the cell search, two special signals are transmitted on the LTE downlink, the Primary Synchronization Signal (PSS) and the Secondary Synchronization Signal (SSS). In case of FDD, the PSS is transmitted within the last symbol of the first slot of subframes 0 and 5, while the SSS is transmitted within the second last symbol of the same slot (i.e., just prior to the PSS). In the frequency domain they are transmitted on 62 subcarriers within 72 reserved subcarriers around DC subcarrier. 10 ms radio frame subframe #0 #1 #2 #3 #4 #5 #6 #7 #8 # subcarrier or 6 resource blocks 1.08 Mhz Systembandwidth Secondary Synchronization Signal OFDM symbol Primary Synchronization Signa Other resource allocation of variable bandwidth Figure 2.10: Synchronization Signals DOWNLINK L1/L2 CONTROL SIGNALING To support the transmission of downlink and uplink transmissions, there is a need for downlink control signaling. This control signaling is often referred to as downlink L1/L2 control signaling, indicating that the corresponding information partly originates from the physical layer (Layer 1) and partly from the MAC layer (Layer 2). The downlink control signaling corresponds to three physical channels: Physical Control Format Indicator Channel (PCFICH): Informs the terminal about the size of the control region (1, 2, or 3 OFDM symbols). There is one PCFICH in each cell. Physical Downlink Control Channel (PDCCH): It is used to signal downlink scheduling assignments and uplink scheduling grants. Each PDCCH carries signaling for a single terminal (or a group of terminals).

28 Chapter 2. 3GPP Long Term Evolution 17 Physical Hybrid-ARQ Indicator Channel (PHICH): It is used to signal hybrid- ARQ ACKs in response to uplink transmissions. There are multiple PHICHs in each cell. The downlink L1/L2 control signaling is transmitted within the first part of each subframe. Thus each subframe is divided into a control region followed by a data region. The control region occupies 1, 2, or 3 OFDM symbols (up to 4 in case of a 1.4 MHz bandwidth). The size of the control region can be dynamically varied on a per-subframe basis to adjust to the instantaneous traffic situation. In case of a small number of users being scheduled in a subframe, the required amount of control signaling is small and a larger part of the subframe can be used for data transmission. One subframe Control region (1-3 OFDM symbols) Control signaling Reference symbols Figure 2.11: L1/L2 control region 2.7 LINK ADAPTATION Link adaptation deals with how to set the transmission parameters of a radio link to handle variations of the radio-link quality. Unlike the early versions of UMTS, which used closed-loop power control to support CS services with a roughly constant data rate, link adaptation in LTE adjusts the transmitted information data rate dynamically (Figure 2.12). The radio-link data rate is controlled by adjusting the modulation scheme and/or the channel coding rate. In case of advantageous radio-link conditions a higher-order modulation, for example 16QAM or 64QAM, together with a high code rate is appropriate. Similarly, in case of poor radio-link conditions, QPSK and low-rate coding is used. For this reason, link adaptation by means of rate control is sometimes also referred to as Adaptive Modulation and Coding (AMC) [8].

29 Chapter 2. 3GPP Long Term Evolution 18 Figure 2.12: LTE rate control A key issue in the development of LTE was if the RBs allocated to a user in a subframe should use the same Modulation and Coding Scheme (MCS), or whether the MCS should be frequency-dependent within each subframe. It was shown that the throughput gains for a frequency-dependent MCS does not justify the overhead required handling the RBs. Consequently, all the RBs assigned to a user within a subframe uses the same MCS, but it can change between subframes MODULATION AND CODING SCHEME Modulation Digital modulation allows for higher data rates in a fixed bandwidth. According to the modulation scheme, one or more bits can be carried per modulation symbol. LTE defines the QPSK, 16QAM and 64QAM modulation schemes which can carry 2, 4 and 6 bits respectively, for both downlink and downlink. The constellation diagrams for these modulation schemes are shown in Figure QPSK 2 bits/symbol 16QAM 4 bits/symbol 64QWI 6 bits/symbol Figure 2.13: Constellation diagrams

30 Chapter 2. 3GPP Long Term Evolution 19 Coding Rate The channel coding scheme chosen for user data was turbo coding. Turbo codes have the benefits of their near-shannon limit performance outweighing the associated costs of memory and processing requirements. A nominal rate-1/3 Turbo Code is used in LTE. Additional coding rates are obtained by puncturing/repetitions. 2.8 SCHEDULER The scheduler determines at a large extend the overall system performance, especially in highly loaded networks. The scheduler controls, for each instant of time, to which users the shared resources should be assigned. It also determines the data rate to be used for each link. LTE has access to both, time and frequency domains. Scheduling decisions are made every 1 ms (TTI) with a granularity of 180 KHz in the frequency domain, as illustared in Figure This is commonly refered as a RB-pair, the scheduling unit in LTE. Gains in system capacity can be achieved if the channel conditions are taken into account in the scheduling decisions. This is known as channel-dependent scheduling (Figure 2.15). Figure 2.14: Scheduling units Channel-dependent scheduling allows for full flexibility in terms of the resources used and can handle large variations in the amount of data to transmit at the cost of the scheduling decision being sent on the control channel of each subframe for both, time and frequency domains. However, some services, most notably VoIP, are characterized by regularly occurring transmission of relatively small payloads. In order to avoid control channel limitations for VoIP traffic in LTE, the concept of semi-persistent scheduling was adopted.

31 Chapter 2. 3GPP Long Term Evolution 20 Figure 2.15: Channel dependent scheduling The semi-persistent resource allocation method adopted in LTE is talk spurt based persistent allocation, and in DL direction the method works as follows. At the beginning of a talk spurt, a persistent resource allocation is done for the user and this dedicated time and frequency resource is used to transmit initial transmissions of VoIP packets. At the end of the talk spurt, persistent resource allocation is released. Thus, the released resource can be allocated to some other VoIP user [10]. Usually, only time-domain decisions are allowed CHANNEL-STATUS REPORTS An important part of the support for downlink scheduling is channel-status reports provided by terminals to the network, reports on which the base station can base its scheduling decisions. Although referred to as channel-status reports, what a terminal delivers to the network are not explicit reports of the downlink channel status. Rather, what the terminal delivers are recommendations on what transmission configuration and related parameters the network should use if/when transmitting to the terminal on the downlink shared channel. The terminal has typically based these recommendations on estimates of the instantaneous downlink channel conditions, thus the term channel-status report. The most important channel-status report is the so-called Channel Quality Indicator 8 [6]: The CQI provides the enodeb information about the link adaptation parameters the UE can support at the time (taking into account the transmission mode, the UE receiver type, number of antennas and interference situation 8 Rank Indicator (RI) and Pre-coding Matrix Indicator reports are used for MIMO schemes.

32 Chapter 2. 3GPP Long Term Evolution 21 experienced at the given time) [2]. This report is represented by a CQI index which indicates the modulation scheme and coding rate that should, preferably, be used for the downlink transmission such that the BLER does not exceed 10%. Details about the CQI will be given in Chapter HYBRID-ARQ Transmissions over wireless channels are subject to errors, for example due to variations in the received signal quality. LTE implements error detection and correction through HARQ, which makes use of the following techniques: Forward Error Correction(FEC): The basic principle is to introduce redundancy in the transmitted signal. This is done by adding parity bits computed from the information bits. Thus, the number of bits transmitted over the channel is larger than the number of original information bits. Automatic Repeat Request(ARQ): The receiver uses an error-detecting code, typically a Cyclic Redundancy Code (CRC), to determine if the packet is in error or not. If the packet is error-free, a positive acknowledgment (ACK) is sent to the transmitter. On the other hand, if an error is detected a retransmission is requested through a negative acknowledgment (NACK). The LTE HARQ protocol uses multiple parallel stop-and-wait process (see [9] for details about this protocol). The number of hybrid ARQ processes directly affects the delay budget in the UE and the enodeb. The smaller the number of hybrid ARQ processes the better from a round-trip time perspective but also the tighter the implementation requirements. Taking transmission, reception, and processing delays into account, it can be calculated that the retransmission of the packet is possible 8 ms after the previous transmission. Thus, the number of parallel HARQ processes is fixed to 8. LTE implements HARQ slightly different for the DL and UL. For Downlink, an adaptive asynchronous HARQ retransmission scheme is considered. Adaptive means that retransmission can take place using different MCS and distinct RBs. Asynchronous refers to the fact that retransmission can happen any time after the 8 ms retransmission delay.

33 Chapter 2. 3GPP Long Term Evolution 22 For Uplink, non-adaptive synchronous HARQ is considered. Non-adaptive indicates that the retransmission must be completed using the same resource allocation and MCS as the original transmission. It requires less overhead than the HARQ for DL. Since it is synchronous, a retransmission occurs exactly 8 ms after the previous transmission. LTE supports soft combining. In traditional ARQ schemes, the erroneous data packets are discarded and a retransmission is requested. However, these packets still contains portions of useful data that could be used for future retransmissions. There are two types of soft combing: Chase Combining: Retransmissions use exactly the same coding scheme as the original transmission, as shown in Figure 2.16 a). The receiver uses maximum-ratio combining to increase the received for each retransmission. Incremental Redundancy: Every time a retransmission occurs, the coding rate is adapted to increase redundancy. Thus, additional to the received Eb/N0 gain, there is also a coding gain (Figure 2.16). a) Chase Combining b) IR Combining Figure 2.16: Soft combining

34 Chapter 3 VOICE-OVER-IP VoIP (Voice-over-IP) is simply the transmission of voice traffic over IP-based networks. The Internet Protocol (IP) was originally designed for data networking. The success of IP in becoming a world standard for data networking has led to its adaption to voice networking. 3.1 VOIP CODECS GSM networks started with the Full rate (FR) speech codec and evolved to Enhanced Full Rate (EFR). The Adaptive Multi-Rate (AMR) codec was added to 3GPP Release 98 for GSM to enable codec rate adaptation to the radio conditions. AMR data rates range from 4.75 Kbps to 12.2 Kbps. The highest AMR rate is equal to the EFR. AMR uses a sampling rate of 8 KHz, which provides Hz audio bandwidth. The AMR-Wideband (AMR-WB) codec was added to 3GPP Release 5. AMR-WB uses a sampling rate of 16 khz, which provides Hz audio bandwidth and substantially better voice quality and mean opinion score (MOS). As the sampling rate of AMR-WB is double the sampling rate of AMR, AMR is often referred to as AMR-NB (narrowband). AMR-WB data rates range from 6.6 Kbps to Kbps. The typical rate is Kbps, which is similar to the normal AMR of 12.2 Kbps. AMR-WB offers clearly better voice quality than AMR-NB with the same data rate and can be called wideband audio with narrowband radio transmission. The bandwidths for the AMR-NB and AMR-WB codecs are illustrated in Figure 3.1.a, while a comparison of these codec with audio bandwidth is exemplified in Figure 3.1.b. The smallest bit rates, 1.8 and 1.75 Kbps are used for the transmission of Silence Insertion Descriptor Frames (SID) [2]. 23

35 Chapter 3. Voice-over-IP 24 1 [unian ear Hz Wideband AMR Hz Narrowband AMR Hz a) Codec bandwidth b) Audio bandwidth Figure 3.1: VoIP Codecs: AMR-NB and AMR-WB LTE specifications suggest that for VoIP performance tests, the AMR 12.2 codec should be used with the parameters described in Table 3.1. It provides a similar reference point for comparisons with actual cellular systems [8]. The resulting capacity of the AMR-NB12.2 Kbps would also be approximately valid for AMR-WB Kbps. Table 3.1: VoIP traffic model parameters Parameter Voice Codec Encoder frame Value RTP AMR 12.2, Source Rate 12.2 Kbps 20 ms Voice activity factor (VAF) 50% (a=b=0.01) SID payload SID Packet every 160 ms during silence 15 bytes (5 Bytes + Header) Protocol overhead with compressed header Total voice payload on air interface 10bits+padding (RTP-pre-header), 4 bytes (RTP/UDP/IP), 2 Bytes (RLC/Security), 16 bits (CRC) 40 Bytes There are two types of VoIP frames for the AMR 12.2 codec: Voice frames and SID frames (see Figure 3.2):

36 Chapter 3. Voice-over-IP 25 Voice Frames. At a voice source rate of 12.2 Kbps, a voice frame generated every 20 ms consists of 244 bits. The total protocol overhead per voice frame includes 10-bits of RTP pre-header and 2-bits padding resulting in a total of 236 bits (32 bytes). Furthermore, a compressed RTP/UDP/IP header consisting of 4 bytes is attached to the packet making the total size of 36 bytes. With 2 bytes of Layer 2 overhead consisting of RLC and security header and 2 bytes CRC, the total VoIP payload size transmitted over the air interface becomes 320 bits (40 bytes) every 20 ms [7]. SID Frames. These frames contain comfort noise, and connection information. SID frames are delivered during silent states every 160 ms (8 voice frames) consisting of 5 bytes of information and 10 bytes of header. For the AMR 12.2 codec, 120 bits must be carried every 160 ms. VoIP packets on the active period SID packets on the silent period 20 ms Headers Payload 160 ms Figure 3.2: VoIP packets and SID packets 3.2 QUALITY CRITERIA Considering the nature of radio communication it is not practical to aim for 100% reception of all the VoIP packets in time. Instead, certain degree of missing packets can be tolerated without notably affecting the QoS perceived by the users. For voice services, usually 1% is tolerated [7]. The voice quality perception degrades as the end-to-end delay increases as depicted in Figure 3.3 [10]. LTE assumes a delay below 200 ms for mobile-tomobile communication. Under this assumption, the delay budget available for radio interface is considered as 50 ms (from enodeb to UE).

37 Chapter 3. Voice-over-IP E-Model Rating R Mouth-to-Ear-Delay/ms Users Very Satisfied Users Satisfied Some Users Dissatisfied Many Users Dissatisfied Nearly All Users Dissatisfied Figure 3.3: Voice Quality (Source: ITU) The system capacity for VoIP is defined as the number of users supported in the cell when more than 95% of the users are satisfied. A VoIP user is satisfied if 98% of its packets experience a delay of less than 50 ms [8]. 3.3 VOIP TRAFFIC MODEL Consider the two-state voice activity model shown in Figure 3.4. The probability of transitioning from state 0 (silence or in active state) to state 1 (talking or active state) is a while the probability of staying in state 0 is (1-a). On the other hand, the probability of transitioning from state 1 to state 0 is denoted b while the probability of staying in state 1 is (1-b). The updates are made at the speech encoder frame rate R= 1/T, where T is the encoder frame duration (20ms). a (1 - a) Silence (State 0) Talking (State 1) (1-b) Figure 3.4: VoIP Traffic Model

38 Chapter 3. Voice-over-IP 27 respectively: The probabilities of being in state 0 and state 1 denoted as and P O = b a + b P 1 = a a + b state 1: The Voice Activity Factor is the probability of being in taking state, that is, VAF = P1 = a a + b The mean silence duration and mean talking duration in terms of number of voice frames 1 can be written as: E [TS] = 1 a E [TS] = 1 b The probabilities that silence duration or a talking duration is n voice frames long are given by: PTs = a( 1-a) n - 1, n = 1,2,... (3.1) (PTs =b(1-b) n-1,n =1,2,... ) (3.2) Since the states transitions from state 1 to state 0 and vice versa are independent, the mean E[T A E] between active state entries is given simply by the sum of the mean time in each state, That is: 1 A voice frame duration is 20 ms.

39 Chapter 3. Voice-over-IP 28 E[TAE] = E[TS] +E[T T] = 1 a + 1 b given by Accordingly, the mean rate of arrival R AE of transitions into the active state is RA E = 1 / E[TA E] 3.4 GENERATING VOIP TRAFFIC The VoIP model traffic can serve as a guide on the number of resource allocation requests. Likewise, it can also be used to generate VoIP traffic as will be described in this section. The number of packets in silence and talking states must be determined to simulate VoIP traffic. The discrete inverse transform (DIT) would be used for such purpose [12]. Consider the probability mass functions in Equation 3.1 and Equation 3.2 describing the probability of a user staying in talking and silence duration respectively. For any probability mass function the following condition must hold: P(N = n i)=p i, i = 1,...,Σpi = 1 i To generate N, the discrete random variable representing the number packets in any state, generate a uniform random number u and set: N = n i if p p i - 1 +p i That is, N = n 1, u p 1 n 2, p 1 <u p 1 +p 2 n 3, Pi + P2 < u pi + p 2 + p3 n t p Pi- 1 <u<p 1 + -p i _ 1 + p i

40 Chapter 3. Voice-over-IP 29 Because is uniform distributed on (0, 1), it follows that for (0<a<b<1): P(a< u b)=b-a Consequently, P i-1 Σ pj < u Σ pj j=1 i j=1 = Pi which proves that N has the desired probability mass function. Graphically, it represents a mapping between the CDF and the number of packets. For example, consider Equation 3.1 for the talking state case. The CDF is easily obtained and plotted in Figure 3.5. Since u is a uniform number between 0 and 1, it represents the probability that the duration in talking state is n packets. Find the value n that produces u. Finally set N=n. Figure 3.5: Inverse Discrete Transform 3.5 VOIP TRAFFIC SIMULATOR The VoIP traffic simulator was developed applying the VoIP traffic model concepts; its respective parameters were initialized according to the 3GPP recommendations in Table 3.1 [8]. Figure 3.4 depicts the theoretical and simulated

41 Chapter 3. Voice-over-IP 30 probabilities that a talking subframe is n frames long. Since the voice activity factor is 50%, the silence probability mass function will be exactly the same. The obtained results of the simulation are displayed in Table 2.1. Simulation and theoretical results do agree, proving the validity of the results Figure 3.3: VoIP Traffic Model Table 3.2: VoIP Traffic Simulation Results Parameter Voice frame periodicity SID frame periodicity Value 20 ms 160 ms Voice Activity Factor (VAF) 0.5 Probability of staying in state 1 ( 1 -b ) 0.99 Probability of staying in state 0 ( 1 - a) 0.99 Probability of transitioning from state 1 to 0 (b) 0.01 Probability of transitioning from state 0 to 1 ( a) 0.01 Mean talking duration E[T T] Mean silence duration E[T S] Mean successive transitions into the 1 state E[T A E] 100 voice packets (2 sec) 100 voice packets (2 sec) 200 voice packets (4 sec) Mean rate of arrivals into the active state( R A E) 0.25 talk-spurts/second

42 Chapter 4 INTRODUCTION TO JITTER In wire-line systems, channels are typically clean and end-to-end transmissions are almost error-free, requiring no retransmissions. However, a wireless channel could be unfavorable, resulting in bit errors and corrupted packets. Packets may have to be retransmitted multiple times to ensure successful reception, and the number of retransmissions depends on the dynamic radio channel conditions. This could introduce significant delay variations. Furthermore, unlike the circuit channels which have a dedicated fixed bandwidth for continuous transmission, packet transmissions are typically bursty and share a common channel that allows multiplexing for efficient channel utilization. This operation results in loading-dependent delay (jitter). 4.1 THE JITTER CONCEPT In a packet-switched network, such as LTE, data is sent by the transmitter as a continuous stream of packets spaced evenly apart. However, due to network congestion, retransmissions, etc., this steady stream could vary over time. Consider Figure 4.1 showing the jitter concept. A transmitter sends packets sequentially and periodically every kt seconds, where k = 0, 1, 2,... is the number of frame. The kth packet is expected to arrive at the receiver at a fixed time t k = kt + t p, where t p is the propagation time. However, as the kth packet travels along the PS network it suffers delay variations t n, called jitter. Then, the packet would be received at a time t k = kt +t p + t. Jitter is defined as the variation in delay that the receiver experiences, or alternatively, a variation in the delivery rate. Jitter can be defined by using the known arrival intervals (20 ms for VoIP), and subtracting the consecutive delays of packets that were not lost. When jitter is a constant is can be filtered out or 31

43 Chapter 4. Introduction tojitter 32 compensated in a deterministic way. However, often exhibits a random behavior [3]. Transmitter T 2T 3T 4T Packet 1 Packet 2 Packet 3 Packet 4 Time Receiver Packet 1 Packet 2 Packet 3 Packet 4 tp tp tp tp T1 = 0 T2 T3 T4 tp - Propagation time T - Packet periodicity Ti - Jitter Figure 4.1: Jittered packets Jitter is a source of speech intelligibility disruptions [4]; the end-to-end jitter has to be small enough not to be noticeable. Delay and jitter are not the same concept. However, as will be explained, there is a trade-off between jitter and delay, and that is the reason why commonly both terms are used. 4.2 LTE JITTER SOURCES SCHEDULER BUFFER The generic function of a resource scheduler is to schedule data to a set of UEs on a shared set of physical resources. In general, scheduler algorithms can make use of two types of measurement information, channel-state information and traffic measurements (volume and priority). The algorithm used by the resource scheduler is closely related with the adaptive and modulation scheme and the retransmission protocol (Hybrid-ARQ).

44 Chapter 4. Introduction to Jitter 33 As network load increases, the physical resources will become scarce and users will be placed in the scheduler buffer. The queues dynamics, which impact the throughput, delay and jitter characteristics of the link seen by the application, depend heavily on network congestion and the MCS (packet sizes). These concepts are shown in Figure 5.2. Figure 4.2: Resource scheduler Another jitter source in the scheduler is due to packet fragmentation. When a packet cannot be sent in one resource scheduling unit, it will be segmented, until the packet has been completely transmitted HARQ RETRANSMISSIONS Due to unfavorable instantaneous channel conditions, packets could arrive corrupted at the receiver. Consequently a retransmission would be requested. Consider Figure 4.3, showing a single HARQ process for the downlink direction. At the enodeb, a packet is sent in the subframe n and received after a propagation time t p, in the subframe n of the receiver. Then the UE will attempt to decode the received signals during a time tue, possible after soft combining. In the subframe n + 4 of the receiver, an ACK/NACK is sent by the uplink channel to the enodeb. The enodeb, processes this information during time tenb and retransmits the packet at subframe n + 8. Thus, a retransmission occurs at least 8 ms after the previous transmission. For VoIP frames up to 6 retransmissions could be possible for a delay bound of 50 ms.

45 Chapter 4. Introduction to Jitter 34 Figure 4.3: HARQ retransmission for the DL RADIO LINK CONTROL FUNCTIONS In LTE, retransmissions of missing or erroneous data units are handled primarily by the hybrid-arq mechanism in the MAC layer, complemented by the retransmission functionality of the RLC protocol. The reasons for having a two-level retransmission structure can be found in the trade-off between fast and reliable feedback of the status reports. A feedback error rate of around 1% results common for hybrid-arq processes. Such an error rate is in many cases far too high; high data rates with TCP may require virtually error-free delivery of packets to the TCP protocol layer [8]. The RLC protocol can be operated in three modes to adapt to the type of transmission: Transparent Mode (TM) bypasses the RLC functions. No retransmissions, no segmentation/reassembly, and no in-sequence delivery take place. This configuration is used for broadcast channels where the information should reach multiple users. Unacknowledged Mode (UM) supports segmentation/reassembly and insequence delivery, but not retransmissions. This mode is used when errorfree delivery is not required, for example VoIP. Acknowledged Mode (AM) is the main operation mode for TCP/IP data transmission. Segmentation/reassembly, in-sequence delivery and retransmission of erroneous data are also supported. VoIP services are operated in unacknowledged mode, where certain packet lost is tolerated. That is, jitter produced by the RLC retransmission scheme is not

46 Chapter 4. Introduction to Jitter 35 an issue. However, in-sequence delivery of packets is still a requirement which will introduce a fixed delay, the so-called jitter buffer size. This will be clarified in the next section MOBILITY As a mobile terminal moves through the network, the propagation time will change. Furthermore, the link adaptation algorithm will adapt the transmission parameters, e.g. the modulation and coding scheme, to the radio channel conditions. This will result in fluctuating data rates. Mobility also implies handovers among sectors or cells. The handover process will introduce unexpected variations in delay, due to the unpredictable coordination time between enodebs OTHER JITTER SOURCES There exist other jitter sources which are not considered for this thesis since they are negligible for network-level simulations, or because they are beyond the scope of this research. Some examples include: External Networks: If packets come from another packet-switched network such as Internet, they could already be jitter. Routing and buffers are typically the reasons behind this undesired impairment. Hardware: Because electronic circuits are not completely synchronized, small variation in the encoding times, processing speeds, etc., are present. 4.3 JITTER MANAGEMENT Although a jitter-free packet-switched network is unfeasible, the jitter phenomenon can be contained. Even real-time services can tolerate certain jitter as long as it is below an established delay bound. Some techniques to deal with jitter are presented in this section JITTER BUFFER Since packets arrive at their destination at random times due to jitter, the user may perceive anomalies in the stream, experienced as static, strange noise effects, garbled words or even missed words or syllables. In Figure 4.4 a jitter

47 Chapter 4. Introduction to Jitter 36 buffer is depicted, which is a common method used at the receiver side to counteract variations in the delay. Basically, incoming packets are stored for a predefined period of time and then they are played-out at the expected rate (often constant). In other words, the receiver holds the first packet in a buffer for a while before sending it to the voice decoder. The amount of time a packet is hold is known as jitter buffer size. Figure 4.4: Jitter buffer If the jitter buffer size is too short, packets will still experience jitter. On the other hand, if it is too long the delay will cause packet lost, and degradation for sensitive-delay applications such as interactive applications and real-time-services such as VoIP. Because of the jitter buffer, there is a trade-off between delay and jitter LTE assumes an end-to-end delay below 200 ms. Under this assumption, the delay budget available for radio interface is 50 ms (from enodeb to UE). There is discussion about the ideal buffer size; even adaptive jitter buffers which change the size dynamically. For the discussion of this thesis, the buffer size will be assumed as 50 ms. Hence, if the jitter caused by the LTE air interface is less than 50 ms, then the jitter buffer will be exchanged for delay SCHEDULER STRATEGIES The LTE Scheduler can optimize over several metrics. However, a critical factor which must always be present is the queue dynamics. The proposed LTE scheduler used for this research takes decisions based on reducing delay - consequently jitter. Chapter 5 will give details about the proposed scheduler.

48 Chapter 5 VOIP JITTER IN LTE Previous Chapters provided to the reader the knowledge necessary to comprehend the jitter phenomenon in LTE networks, especially for VoIP services. From now on, VoIP jitter is quantified by means of physical layer and MAC layer simulations. 5.1 SIMULATION SCENARIO Consider the simulation scenario shown in Figure 5.1 for the LTE downlink. It is assumed that VoIP packets and SID packets arrive to the enodeb with no jitter. According to the traffic load and the instantaneous radio channel conditions, packets could be scheduled immediately or placed in the scheduler buffer until resources become available. Let us consider the first case. Once a packet has been selected to be transmitted, the appropriated modulation and coding scheme is determined by the base station. To help on the scheduler decisions, every UE sends periodical reports about the radio channel conditions (1 ms for this simulation). Feedback information is sent in the form of CQIs, indicating the maximum MCS that can be supported by the UE, such that the Block Error Rate does not exceed 10%. Once the scheduler has determined the resource allocation and MCS, the enodeb notifies to the UE through the Physical Downlink Control Channel (PDCCH). Then the packet is sent through the Physical Downlink Shared- Channel (PDSCH), while storing a copy in the HARQ buffer in case a retransmission is requested. The UE will try to decode the information sent by the enodeb. According to the result, the mobile terminal could send an ACK if the packet was received successfully or an NACK if the erroneous packet could not be recovered by the FEC. A HARQ retransmission takes at least 8 ms for the DL. Note that a 37

49 Chapter 5.VoIP Jitter in LTE 38 retransmission could also be placed in the scheduler queue in case of high traffic conditions. enodeb Channel UEs Scheduler Buffer UE1 RBs, MCS CQI, ACK/NACK VoIP Traffic Generator UE1 UE2 UEn RBs, MCS CQI, ACK/NACK UE2 Hybrid-ARQ Buffer RBs, MCS CQI, ACK/NACK UEn Figure 5.1: Simulation Scenario The jitter sources under this scenario are: 1. Scheduler buffer: Due to network congestion, some packets will have to wait in the queue. Besides, packets could also be placed in the scheduler buffer if the destination is experiencing poor channel conditions. 2. Packet fragmentation: VoIP traffic is characterized by low bitrates, what implies small packets. However, in poor radio conditions a low order, more robust MCS would be chosen, so a single RB-pair could even not be enough to send a complete VoIP packet. The packet would be fragmented, requiring more than 1 TTI to be completely transmitted. The VoIP packet cannot be used by the voice decoder until it is complete.

50 Chapter 5.VoIPJitter in LTE Retransmission requests. Due to the wireless channels impairments, packets could arrive corrupted to the receiver. In this case a retransmission would be requested. 8 HARQ processes are used in LTE, implying a retransmission delay of at least 8 subframes. The Jitter T n kexperienced by the kth packet, of the nth UE, is defined as the time difference between the moment the packet was successfully received by the UE, t (n) UEk, and the instant when it arrived to the enodeb, t (n) (n) enbk. That is, T k = t (n) UEk - t(n) enbk. The time transmission interval is not considered for jitter calculations since it is a fixed delay. The propagation delay is neglected. Of course, the nth user will receive K packets, comprising the user jitter profile. In addition, the cell jitter profile can be obtained by taking the overall behavior of the N users in the cell. Cell jitter profiles would be analyzed in further sections. 5.2 SIMULATION DESCRIPTION The simulation can be split into 3 sections. VoIP traffic (see Chapter 3.5, page 29), PHY layer, and MAC protocol (Scheduler) LTE PHYSICAL LAYER Developing a whole new simulator for the LTE air interface would be an overwhelming task and beyond the scope of this thesis. However, results as real as possible are desired. The Institute of Communications and Radio Frequency Engineering of the Vienna University of Technology has developed the LTE linklevel simulator for the Matlab platform, and provided under academic use. It has quickly gain popularity because of its features and reliability [9]. The structures of the transmitter and receiver are shown in Figure 5.2 for the downlink. Nevertheless, neither traffic models nor the required algorithms to study the jitter phenomenon had been implemented, i.e., a continuous data stream was supposed. The original simulator was enhanced, by the author of this thesis, to support a variety of traffic models - VoIP in this case. The PHY layer simulation parameters used in the simulations are described in Table 5.1.

51 Chapter 5.VoIP Jitter in LTE 40 Figure 5.2: LTE downlink PHY structure Table 5.1: Simulation parameters Parameter Value Scenario/direction 1 Cell / Downlink SNR 5, 8, 12 Channel Bandwidth 5 MHz Number of RBs 25 Subcarrier Spacing 15 KHz Cyclic Prefix Normal Symbols per subframe 14 L1/L2 control region 3 OFDM symbols Reference symbols per subframe 4 Duplexing FDD HARQ mode Adaptive asynchronous HARQ soft combining Off HARQ processes 8 HARQ max retransmissions Infinite CQI delay 1 ms CQI resolution 1 RB Possible MCSs MCS = CQI Antenna configuration SISO Channel model Typical Urban

52 Chapter 5.VoIP Jitter in LTE 41 Some comments about the simulation parameters: The SNRs are a sample of a worst, mean, and best case scenario. Although a semi-persistent scheduler will be used, 3 OFDM symbols are considered for the simulation. The interest is to analyze the VoIP performance under strict restrictions. Soft combining is not used to analyze a baseline system. The maximum number of retransmission was set to infinite, so that packet losses did not impact in the jitter behavior. The MCS is adjusted according to the CQI recommendations. This will be clarified in further sections LTE MAC PROTOCOL The scheduler plays a key role in the performance of LTE networks. However, the scheduling and rate-adaptation algorithms are vendor specific. In this master thesis a fair user, delay-optimized resource allocation algorithm is proposed. It is fair in the sense that every user has the same probability to be chosen. It is delay optimized, signifying that decisions are taken based in time. Note, that the term delay is used in this context. The reason for this is that if the maximum jitter experienced by a packet is covered by the jitter buffer size, it will be traded-off for delay. Thus, the goal is to keep the maximum number of packets with minimum jitter. The scheduler flow diagram is illustrated in Figure 5.3. Instructions are performed sequentially. Every section of the scheduler is described in the following subsections. Data and Control Regions The total number of physical resources depends on the bandwidth assigned to the LTE system. For VoIP performance evaluations a 5 MHz channel bandwidth is suggested, and results are scaled accordingly for other bandwidths [12]. This corresponds to 25 RB-pairs. For each subframe, the scheduler must determine how many of these radio resources will be employed for user data and how many for signaling and control purposes. The number of data subcarriers varies between subframes. However, an approximation can be obtained as follows:

53 Chapter 5.VoIP Jitter in LTE 42 1 ms subframe size, 5 MHz BW Each subframe comprises of 14 OFDM symbols 1 and 300 REs per symbol First 3 OFDM symbols comprises of control + pilot = 900 REs (worst case) 5th, 8th and 12th symbol comprises additional pilots; 50 x 3 = 150 REs PSS and SSS 2 take = 132 REs around the DC subcarrier; subframes 0, 5 Data subcarriers (RE) in DL: 300 x = 3,018 (worst case) Number of overhead RE s: 1182 REs (worst case) Figure 5.4 illustrates the L1/L2 control region, reference symbols, and the data region. Synchronization signals are omitted for clarity. 1 OFDM symbols 2 5th symbol contains 2 RS, hence 72 2 x 6 =60

54 Chapter 5.VoIP Jitter in LTE 43 Start The Scheduler makes decisions every 1 ms Feedback ACK/NACK Input Packets received correctly? Yes Free HARQ Buffer No Save Scheduler Buffer Retransmission staus Read Scheduler Buffer Any packet needs retransmission? Yes # retransmissions < Max. retransmissions? Yes Free HARQ Buffer Free Scheduler Buffer No Priority 1 Read Scheduler Buffer Read Any other packet in the scheduler buffer? Yes Time scheduler buffer > time-bound? Yes Free HARQ Buffer Free Scheduler Buffer No Priority 2 Incoming packets (VoIP Traffic) Input New packets? (VoIP Traffic) Yes Priority 3 Read Priority list RBs available? No Save Scheduler Buffer Yes Feedback CQI Input CQI > 1? No Yes Choose MCS AMC Save HARQ Buffer End Figure 5.3: Scheduler flow diagram

55 Chapter 5.VoIP Jitter in LTE 44 OFDM symbol RB = 1 R R 12 subcarrier 180 KHz R R R R R R 1 Slot 0.5 ms 1 Slot 0.5 ms RB = 2 R R 12 subcarrier 180 KHz R R 25 RBs 300 subcarriers 4.5 MHz R R R R L1/L2 Control Region Data Region... RB = 25 R R 12 subcarrier 180 KHz R R R R R R 1 Subframe TTI = 1 ms Figure 5.4: L1/L2 Control Region, RS, and Data Region

56 Chapter 5.VoIPJitter in LTE 45 Link Adaptation Link adaptation is a functionality of the LTE scheduler. The MCS determines both the modulation alphabet and the Effective Code Rate (ECR) of the channel encoder. To help on the scheduler decisions, the UEs sends recommendations about the maximum supported MCS that ensures abler 10-1, the so-called Channel Quality Indicator. The possible MCS suggested by a mobile terminal are described in Table 5.2. Table 5.2. CQI and MCS recommendations CQI Index Modulation Coding Rate Efficiency (Bits per resource element) 0 No transmission QPSK 78/ QPSK 120/ QPSK 193/ QPSK 308/ QPSK 449/ QPSK 602/ QAM 378/ QAM 490/ QAM 616/ QAM 466/ QAM 567/ QAM 666/ QAM 772/ QAM 873/ QAM 948/ To obtain the BLER for the MCS corresponding to each CQI value, AWGN simulations were performed. Figure 5.5 shows the BLER results of CQIs 1-15 without using HARQ. Each curve is spaced approximately 2 db from each other. The SNR-to-CQI mapping required to achieve this goal can thus be obtained by plotting the 10% BLER values of the curves in Figure 5.5 over SNR, like it is shown

57 CQI BLER Chapter 5.VoIP Jitter in LTE 46 in Figure 5.6. Using the obtained line, an effective SNR can be mapped to a CQI value [14] BLER, 1.4MHz, SISO AWGN, 5000 subframes SNR [db] Figure 5.5: BLER curves obtained from SISO AWGN simulations for all 15 CQI values. From CQI 1 (leftmost) to CQI 15 (rightmost) 15 SNR-CQI mapping model SNR [db] Figure 5.6: SNR-CQI mapping For VoIP traffic the used BLER target in DL is in the order 10% for the first transmission [10]. Since CQI provides approximately such BLER, the MCS can be adjusted according to the CQI suggestions.

58 Chapter 5.VoIPJitter in LTE 47 Priority System Before actual radio resource assignment takes place, a priority system orders the packets according to the following criteria 1. Retransmission requests. Since retransmissions take at least 8 ms to be completed, it gets the higher priority. 2. Buffer time. This weight is obtained as the sum of the time in the queue experienced by all the packets of the same user. 3. New packets. Incoming VoIP/SID packets get the lower priority. Note that the priority system applies for packets not for users. That is, all the users are treated equally, but packets experiencing more delay jitter get higher priority. It is important to mention that, if certain UE is experiencing bad instantaneous channel conditions (CQI < 2), it will not be scheduled until the next subframe even if it has the higher priority. Instead, the next user in the priority list will be chosen. Resource assignment It will be assumed a semi-persistent scheduling implementation. That is, users will always transmit in the same frequency allocation, but scheduling decisions can be taken in the time domain. The number of users in each simulation has been chosen as a multiple of 25, since there 25 RBs in 5 MHZ. Thus for 25, 50 and 75 UEs there would be 1, 2, and 3 users per RB respectively. 5.3 SIMULATION RESULTS Simulations are computationally intensive because of the complexity of physical and MAC layers simulations. Coding, decoding, modulation, channel estimation, link adaptation, scheduling, and other functions are performed continuously for each user, packet and fragment. The tests were run for 12, 000 subframes (12 seconds), for different SNR scenarios. Before analyzing the results, it is important to describe some concepts. The cell outage probability is defined as. P{outage} = P(T > 50)

59 Chapter 5.VoIPJitter in LTE 48 Where T is the jitter experienced by all the packets in the cell. Likewise, LTE specifies the capacity and satisfaction criteria as [14]. "System capacity is defined as the number of users supported in the cell when more than 95% of the users are satisfied. A VoIP user is satisfied if 98% of its packets experience a delay of less than 50 ms" Note that satisfaction is a stricter criterion than cell outage probability, since it is based in the number of satisfied users and not only in the total jitter experienced in the cell SNR = 5 db Consider the simulation results in Table 5.3 and Figure 5.7, describing the jitter behavior for a SNR = 5 db. It can be concluded the following. The satisfaction quality criterion is completely accomplished for 25 and 50 users, although 0.22 % and 0.34% of the packets are above the desired bound (50 ms). However, when the number of mobile terminals increases to 75, the satisfaction criterion is not strictly achieved (93.3% < 95%). Nevertheless, jitter is a minor issue in this case, since the mean of the delay is 12.2 and the deviation is only This indicates that jitter could be tolerated for the 0.61% of the packets. Now, let us study the case for 100 users. At a glance it could seem from Figure 5.7 as if jitter were under acceptable limits since the mean is 18.9 ms, and the standard deviation is 27.6 ms. However, the satisfaction percent drops dramatically to 66%; indeed, the outage probability is 5.8%. This is not acceptable. If users keep increasing up to 125 and 150, the means are and 171.4, clearly above the 50 ms target. Finally, for 175 UEs the satisfaction percent is only 3.4, and the outage probability is 69%. Now, let us analyze the jitter cell profile 3 shown in Figure Consider the case for 25 users. It can be observed, that jitter is concentrated inside the 50 ms bound, so 100% capacity is achieved. Most of the packets (10.5%) suffer a jitter of 2 ms due to packet fragmentation (the source is clear, since packets does not have to wait because there is not contention). However, as expected, the retransmission scheme will introduce jitter. Recall that a packet sent in subframe n, takes n + 8 subframes to be retransmitted. Thus, erroneous packets generate "replicas" with a 3 Jitter cell profiles show the probability that a packet is jittered by Tms milliseconds

60 Jitter (ms) Chapter 5.VoIP Jitter in LTE 49 periodicity of 8 ms. As the number of users keeps increasing, the jitter cell profile spreads, leading to the subsequent satisfaction drop. Table 5.3: Cell jitter; SNR = 5 db Users Mean Deviation Mode Max P outage Satisfaction x x x x x x x Jitter behavior Mean Dispersion Users Figure 5.7: Jitter behavior (SNR = 5 db)

61 Chapter 5.VoIP Jitter in LTE SNR = 8 db Now, let us study the case for the average case (SNR = 8 db). Consider Table 5.4 and Figure 5.8. The jitter behavior seems steady up to 100 users, with a satisfaction criterion of 100%. Note that under the strict capacity criterion, capacity has increased from 50 to 100 compared to the previous case (SNR = 5 db). Although the satisfaction is 88% for 125 UEs, the outage probability is only 1%. At 150 users, 81.3% satisfaction is achieved with an outage probability of 3.69%. Operation for 175 and 200 users is not suggested since the outage probability is 22.6 % and 45.1 % respectively. Table 5.4: Cell jitter; SNR = 8 db Users Mean Deviation Mode Max P outage Satisfaction x x x x x x x x It can be noted from Figure 5.8 that network congestion has a small impact on jitter up to 125 users. In this zone, jitter is mainly due to HARQ retransmission (see the jitter profiles in Figure 5.8). Then, jitter will increase exponentially with the number of users. Jitter cell profiles for this scenario are illustrated in Figure Under no contention (25 users), 24% of the packets are jittered by 1 ms; an excellent performance measure. Nevertheless, network congestion will cause that the jitter cell profile spreads, as the previous case, leading to the subsequent reduction in satisfaction.

62 Jitter (ms) Chapter 5.VoIP Jitter in LTE Jitter behavior Mean Dispersion Users Figure 5.8: Jitter behavior (SNR = 8 db) SNR = 12 db As can be noted from Table 5.5 and Figure 5.9, the capacity for this case is 225 users. Network congestion has small impact up to this case. Then, when the number of UEs in cell reaches 250, the satisfaction criterion drops to 78.4% and the cell outage probability is 2.8%. From the jitter cell profile depicted in Figure 5.12, it is easy to see that from 25 to 100 users the packets are concentrated around 0 and 1 ms. Due to retransmissions, with a periodicity of 8 ms, replicas appears along the jitter profile. As the number of users increases, then packets will be placed in the scheduler buffer. This will lead to the spreading of the jitter cell profile, and the subsequent reduction in the satisfaction quality criterion.

63 Jitter (ms) Chapter 5.VoIP Jitter in LTE 52 Table 5.5: Cell jitter; SNR = 12 db Users Mean Deviation Mode Max P outage Satisfaction x x x x x x x x x x x Jitter behavior Mean Dispersion Users Figure 5.9: Jitter behavior (SNR = 12 db)

64 Probability Probability Probability Probability Chapter 5.VoIP Jitter in LTE 0.12 SNR = 5 db 0.12 SNR = 5 db Jitter (ms) Jitter (ms) a) 25 UEs b) 50 UEs 0.12 SNR = 5 db 0.12 SNR = 5 db Jitter (ms) Jitter (ms) c) 75 UEs d) 100 UEs Figure 5.10: Jitter cell profile (SNR = 5 db) 53

65 Probability Probability Probability Probability Chapter 5.VoIP Jitter in LTE 0.25 SNR = 8 db 0.25 SNR = 8 db Jitter (ms) Jitter (ms) a) 25 UEs b) 50 UEs 0.25 SNR = 8 db 0.25 SNR = 8 db Jitter (ms) Jitter (ms) c) 75 UEs d) 100 UEs 54

66 Probability Probability Chapter 5.VoIP Jitter in LTE 0.25 SNR = 8 db 0.25 SNR = 8 db Jitter (ms) Jitter (ms) e) 125 UEs f) 150 UEs Figure 5.11: Jitter cell profile (SNR = 8 db) 55

67 Probability Probability Probability Probability Chapter 5.VoIP Jitter in LTE 0.35 SNR = 12 db 0.35 SNR = 12 db Jitter (ms) Jitter (ms) a) 25 UEs b) 50 UEs 0.35 SNR = 12 db 0.35 SNR = 12 db Jitter (ms) Jitter (ms) c) 75 UEs d) 100 UEs 56

68 Probability Probability Probability Probability Chapter 5.VoIP Jitter in LTE 0.35 SNR = 12 db 0.35 SNR = 12 db Jitter (ms) Jitter (ms) e) 125 UEs f) 150 UEs 0.35 SNR = 12 db 0.35 SNR = 12 db Jitter (ms) Jitter (ms) g) 175 UEs h) 200 UEs 57

69 Probability Probability Probability Probability Chapter 5.VoIP Jitter in LTE 0.35 SNR = 12 db 0.35 SNR = 12 db Jitter (ms) Jitter (ms) i) 225 UEs j) 250 UEs 0.06 SNR = 12 db 0.06 SNR = 12 db Jitter (ms) Jitter (ms) k) 275 UEs l) 300 UEs Figure 5.12: Jitter cell profile (SNR = 12 db) 58

70 Chapter 6 CONCLUSIONS AND FUTURE WORK 6.1 GENERAL CONCLUSIONS 3GPP Long Term Evolution is the next step towards 4G mobile networks, promising higher data rates, increased capacity, and reduced delay. LTE is a flat, all-ip network, where voice services will be provided through Voice-over-IP (VoIP). VoIP services are highly sensitive to jitter, the delay variation, since it causes speech intelligibility disruptions. The main jitter sources in LTE were identified. Network congestion, packet fragmentation and HARQ retransmissions influence the arrival variability of voice packets. Furthermore, the radio channel conditions dictate the performance of all of these components as demonstrated in the VoIP traffic, PHY and MAC layer simulations. The scheduler plays a key role in the queue dynamics; therefore jitter. Scheduler strategies should consider that jitter does not exceed the quality bound, (50 ms for LTE), such that the jitter buffer can handle effectively this phenomenon. Indeed, for VoIP service is crucial to assure a minimum data rate (according to the requirements of the VoIP codec) at any time. Three different scenarios were analyzed corresponding to diverse channel conditions. In every scenario the MCS was selected to achieve a 10% BLER, as suggested by the CQI reports sent by the UE. In this way a reference point between each scenario was established. In poor radio conditions (SNR = 5 db), low-order modulation and coding schemes induced packet fragmentation. This phenomenon increased network congestion since several TTIs were required to send a single packet. Likewise, every fragment could be retransmitted several times provoking the exponential 59

71 Chapter 6. Conclusions and Future Work 60 growth of jitter. According to the 3GPP VoIP satisfaction criterion, up to 50 users could be supported in this case. For average (SNR = 8 db) and good (SNR = 12 db) channel conditions, packet fragmentation was reduced since higher order MCSs were used. The main jitter source for few users in cell was due HARQ retransmissions. However, as the number of mobile terminals increased, packets began to be placed in the scheduler buffer, leading to network congestion. The number of supported users was determined as 100 and 225 respectively. Comparing the results of the different scenarios the following conclusions can be established. Network congestion leads to the spreading of the jitter cell profiles (see Figure 5.10, Figure 5.11 and Figure 5.12), while HARQ retransmissions generate replicas of such widening with a periodicity of 8 ms. Indeed packet fragmentation, inducing a jitter granularity of 1 ms, is preferable over HARQ retransmissions that take at least 8 ms to be accomplished. Even more, it can be inferred that conservative MCS could improve the jitter behavior even in the worst case scenario. It comes from the fact that, although the number of fragments would increase, along the jitter spreading, the BLER and the number of HARQ retransmissions would be decreased. Then, the jitter replicas would be diminished. Of course, this implies that the jitter caused by the added fragments does not surpass the delay bound. Under the strict settings considered in this thesis (maximum overhead, simple HARQ retransmissions, time-based scheduling), LTE prove to be able to handle jitter. However, aspects such as network congestion have to be considered in the deployment stage. For instance, the cell size determines the number of users, and radio channel conditions; key component in jitter dynamics. 6.2 FUTURE WORK The following ideas can suggested for further research: Analyze jitter under conservative MCSs. Study the case for HARQ soft combing. Consider the effects of dynamic scheduling.

72 Appendix A MULTI-CARRIER TRANSMISSION In modern communications systems intensive multimedia applications such as video streaming, videoconferencing, voice, etc., are being demanded by users. The necessity for higher data rates in a limited spectrum has developed a series of transmission schemes which improve the spectrum efficiency, increment capacity, and reduce the corruption caused by radio-channel time dispersion. One of such schemes is the so-called, Orthogonal Frequency Division Multiplexing (OFDM), a multi-carrier system. A.1 THE MUTI-CARRIER CONCEPT A.1.1 CHANNEL CAPACITY Shannon provided the theoretical tools to determinate the maximum rate C, known as channel capacity, where the radio link is impaired only by additive white Gaussian noise. This relationship is given by: C = BW 1og 2 (l + S N ) BW is the available bandwidth, S is the signal power and N denotes the power noise. It is clear that channel capacity depends on the signal-to-noise ratio and the bandwidth available. Suppose a communication system with rate R. The signal power can be expressed as S = E b R where E b is the energy per information bit, and N = N 0 BW is the power spectral density in W/Hz. Since the actual information rate cannot exceed the capacity of the system: R C = B W lo g 2 ( l + S N ) = BW log 2 ( 1 + Eb-R N0-BW ) 61

73 Appendix A. Multi-carrier transmission 62 Let γ = R/BW be defined as the bandwidth utilization of the channel. γ log 2 (1 + γ Eb N0 ) It is convenient to express this inequality as a lower bound on the required received energy per information bit, normalized to the noise power density, necessary for a given data rate. E b E b N 0 N 0 min { } 2γ -1 = γ The lower bound is plotted in Figure A.1. Two regions can be identified from this plot: a power-limited region and a bandwidth limited region. Figure A.1: Operation regions Power-limited region. When the bandwidth utilization is less than 1, the channel is underutilized; an increment in the bandwidth of the communication system will not improve significantly the performance of the channel. The actual constrain is the power of transmission. Bandwidth-limited region. If the bandwidth is not increased, a very high power is required to achieve higher data rates.

74 Appendix A. Multi-carrier transmission 63 Working on a bandwidth-limited region is highly inefficient since un proportional signal-to-noise ratios are required to achieve high data rates. The system should be operated in the power-limited region. This is the reason why wider bandwidths are necessary for the next generation of wireless communication networks such as 3GPP Long Term Evolution. LTE can handle bandwidths up to 20 MHz. A.1.2 WIDER BANDWIDTHS Larger bandwidths are necessary to provide high data rates in a power efficiently way. However, transmitting a single wideband signal has some constraints. Allocating continuous spectrum is complicated since it is a scarce and expensive resource. The use of wider bandwidths increases the complexity of transmitters and receivers. For example, the amplifiers must work on a wider linear area. Furthermore, the time dispersion problem associated with wider-band transmission arises naturally. Time dispersion occurs when the transmitted signal propagates to the receiver via multiple paths with different delays. In a frequency domain, a time-dispersive channel corresponds to a non-constant channel frequency response. This radio-channel frequency selectivity will corrupt the frequency domain spectrum of the transmitted signal and lead to higher error rates for a given signal-to-noise/interference ratios. Every radio channel is subject to frequency selectivity, at least at some extend. However, the extent to which the frequency selectivity impacts the radio communication depends of the bandwidth of the transmitted signal with, in general, larger impact for wide-band transmissions. The amount of radio-channel frequency selectivity also depends on the environment with typically less frequency selectivity (less time dispersion) in case of small cells and in environments with few obstructions and potential reflectors such as rural environments. As a mobile terminal is moving through the environment, the detailed structured of the multi-path propagation, and thus also the detailed structure of the channel frequency response, may vary rapidly in time. The rate of the variations in the channel frequency response is related to the channel Doppler spread.

75 Appendix A. Multi-carrier transmission 64 A.2 MULTI-CARRIER TRANSMISSION In muti-carrier transmissions instead of transmitting a single wideband signal, multiple narrowband signals, often referred as subcarriers, are frequency multiplexed and transmitted over the same radio-link. The impact in terms of signal corruption due to radio-channel frequency selectivity depends on the bandwidth of each subcarrier. Consider a wideband system with a total transmission bandwidth BW and a data rate R. The coherence time of the channel is assumed to be B c < B, so the signal experiences selective fading. The basic principle behind multi-carrier transmission is to split the system into N subsystems placed in parallel. Every substream will have a bandwidth B N = BW / N and a data rate R N = R/N. If is chosen sufficiently large then B N << B c and every subcarrier will experience approximately flat fading. In the time domain this implies that the symbol time of every subcarrier T N = 1/ B N is much bigger than the delay spread in the channel σ τ = 1 /Bc so little ISI corruption is experienced. A drawback of conventional multi-carrier transmission is that the spectrum of each subcarrier cannot overlap to avoid distortion. Typically frequency guard bands are added between subcarriers, resulting in low frequency spectrum utilization. A second disadvantage is that multi-carrier systems usually have big peak-toaverage power ratios (PAPR), which is an aspect to consider in mobile terminals. A.3 OFDM AS A MULTI-CARRIER TRANSMISSION Orthogonal Frequency Division Multiplexing is a kind of multi-carrier transmission. The main differences between OFDM and a straightforward multicarrier transmission scheme are. The use of relative large number of narrowband subcarriers. As an example, WCDMA multi-carrier evolution consists of four subcarriers in a bandwidth of 20 MHz, each with a bandwidth of 5 MHz. In contrast, OFDM could consist of hundreds of subcarriers transmitted at the same time over the same radiochannel. Tight frequency-domain packing with a subcarrier spacing Δf = 1/T u where T u is the per-carrier modulation-symbol time (Figure A.2). Basically, the

76 Appendix A. Multi-carrier transmission 65 spectrum overlaps without causing interference due to the orthogonally property between subcarriers as shown in Figure A.3. Figure A.2: Subcarrier spacing Figure A.3: a) FDM, b) OFDM An OFDM signal is produced through a bank of Nc modulators, where each modulator corresponds to an OFDM subcarrier, like the one shown in Figure A.4.a. Therefore, the OFDM signal x(t), during the time mt u t<(m+1)t U, can be expressed as: Nc-1 k=0 a) Bank of Modulators b) Bank of correlators Figure A.4: OFDM modulation and demodulation