Table of Contents Cover Title page List of Contributors Foreword Acknowledgments List of Abbreviations 1 Introduction 1.1 To the reader 1.2 Content 1.3 Scope Reference 2 LTE Backhaul 2.1 Introduction 2.2 LTE Backhaul Planes 2.3 Radio Features of LTE and LTE-A 2.4 Requirements for LTE Backhaul (SLAs) 2.5 Transport Services 2.6 Planning Problems 2.7 LTE Backhaul Technologies 2.8 Small Cell Backhaul 2.9 Future Radio Features Affecting Backhaul 2.10 Related Standards and Industry Forums 2.11 Operator Example References 3 Economic Modeling and Strategic Input for LTE Backhaul 3.1 Introduction 3.2 Strategic Input for Planning 3.3 Quantifying benefits 3.4 Quantifying costs 3.5 Case router 3.6 Wireless Backhaul Case Study References Further Reading 4 Dimensioning Aspects and Analytical Models of LTE MBH Networks 4.1 Introduction 4.2 Dimensioning Paradigm 4.3 Applications and QoE: Considerations 4.4 Dimensioning Requirements 4.5 Traffic Models 4.6 Network models 4.7 Dimensioning References 5 Planning and Optimizing Mobile Backhaul for LTE 5.1 Introduction 5.2 Backhaul Network Deployment Scenarios 5.3 Network Topology and Transport Media 5.4 Availability and Resiliency Schemes 5.5 QoS Planning 5.6 Link Bandwidth Dimensioning 5.7 Dimensioning Other Traffic Types 5.8 Base Station Site Solutions 5.9 Security Solutions 5.10 IP Planning 5.11 Synchronization Planning 5.12 Self-Organizing Networks (SON) and Management System Connectivity 5.13 LTE Backhaul Optimization References 6 Design Examples 6.1 Introduction 6.2 Scenario #1: Microwave 6.3 Scenario #2: Leased Line Reference 7 Network Management 7.1 Introduction 7.2 NMS Architecture 7.3 Fault Management 7.4 Performance Management 7.5 Configuration Management (CM) 7.6 Optimization 7.7 Self-Organizing Network (SON) 7.8 O&M Protocols 7.9 Planning of Network Management System References 8 Summary Index End User License Agreement List of Tables Chapter 02 Table 2.1 Delay and loss requirements for different planes (upper bound) Table 2.2 Synchronization requirements for LTE mobile standards (see Chapter 6 of Metsälä and Salmelin, 2012) Table 2.3 LTE feature-dependent phase synchronization requirements Table 2.4 DSL standards suitable for LTE backhaul Table 2.5 Categorization of non-ideal backhaul in 3GPP Release 12 work (3GPP, 2014) Table 2.6 Categorization of ideal backhaul in 3GPP Release 12 work (3GPP, 2014) Chapter 03 Table 3.1 Estimates of backhaul capex and opex as percentage of operator revenue Table 3.2 Opex types relevant to the LTE backhaul Table 3.3 Annual cash flows of Example 3.7 Table 3.4 Cumulative net cash flows of Example 3.7 Table 3.5 Net present value calculation Table 3.6 Related cost items for Alternative A (leased lines) and Alternative B (wireless) Table 3.7 Cash flows (thousands) Table 3.8 NPV of Example 3.8 (thousands) for Alternative B (compared against Alternative A) Chapter 05 Table 5.1 Availability and corresponding downtime per year Table 5.2 Availability calculations for different hardware configurations Table 5.3 Unavailability for different types of links is calculated from the outdoor and indoor units Table 5.4 Unavailability comparison for different types of links a Table 5.5 Availability comparison for different types of links Table 5.6 Protection latency times from different layers Table 5.7 A simple example of traffic type to service class mapping, with four classes Table 5.8 An example of traffic type to service class mapping, with eight classes Table 5.9 Overheads for S1 user plane transport Table 5.10 Typical cost of selected S1 signaling procedures for DL and UL, including SCTP/IP/Eth/L1 overhead. The UL/DL cost is calculated over 60 minutes Table 5.11 Example control plane traffic model Table 5.12 Single-server queue types with example applications Table 5.13 Summary of dimensioning approaches Table 5.14 Planning parameters for IPsec Table 5.15 Planning parameters for PKI Table 5.16 Example for QoS setting for user, control, synchronization and management planes Chapter 06 Table 6.1 Input information for both scenarios Table 6.2 Example eNB IP address allocation Table 6.3 Summary of IP addresses in the design example Table 6.4 Summary of availability calculations for eNB in the example topology Table 6.5 Mapping of traffic types to classes Table 6.6 Summary of provider service offering of the candidate service providers List of Illustrations Chapter 02 Figure 2.1 Network element symbols Figure 2.2 Definition of network areas Figure 2.3 LTE related e2e planes of backhaul network Figure 2.4 LTE U-plane protocol stack Figure 2.5 LTE C-plane protocol stack Figure 2.6 Frequency aligned signals Figure 2.7 Phase and frequency aligned signals Figure 2.8 Time, phase and frequency aligned signals Figure 2.9 Management plane protocol stack Figure 2.10 Principle of carrier aggregation Figure 2.11 Carrier aggregation improves average cell throughput both in UL and DL (Nokia Networks, 2014) Figure 2.12 Principle of MIMO Figure 2.13 MIMO peak data rates with different antenna configurations (Nokia Networks, 2014) Figure 2.14 Principle of CoMP (intra-eNB) Figure 2.15 Principle of relay nodes Figure 2.16 Examples of achievable cell throughput gains in some relay node scenarios (Nokia Networks, 2014) Figure 2.17 Principle of time domain enhanced inter-cell interference coordination Figure 2.18 User plane peak capacities of LTE base station for different frequency bands Figure 2.19 Average capacity of LTE base station for different frequency bands Figure 2.20 X2 handover signaling Figure 2.21 Security architecture Figure 2.22 Topology example Figure 2.23 FTP throughput for different shaping configurations Figure 2.24 Simplified network picture Figure 2.25 Unsuccessful inter-frequency handover, real time per day for the base stations shown in Figure 2.24 Figure 2.26 Unsuccessful inter-system handover, real time per day for the base stations shown in Figure 2.24 Figure 2.27 Different radio KPIs per day for another base station which changed from in sync to out of sync Figure 2.28 Chain of L2 and L3 VPNs in case of microwave backhaul Figure 2.29 L3 VPN in case of optical backhaul Figure 2.30 High-level planning process Figure 2.31 Physical media used in the backhaul for 2013 and 2018 (Infonetics Research, 2014) Figure 2.32 Radio data rate as a function of modulation with different radio channel bandwidths Figure 2.33 Scheduler coordination alternatives (eCoMP) Figure 2.34 Principle of dual connectivity Figure 2.35 Alternative architectures of dual connectivity Figure 2.36 Impact of dynamic ABS muting pattern adjustment on system performance Figure 2.37 Operator example combining different leased line providers and own equipment Chapter 03 Figure 3.1 2015 Estimate for network capex by network component. Figure 3.2 Telco’s opex. Figure 3.3 Comparison of alternatives Figure 3.4 Router investment Figure 3.5 Annual net cash flows of Example 3.7 Figure 3.6 Example 3.7 net cash flow and cumulative net cash flow Figure 3.7 Alternative A: Leased line access (shaded area considered) Figure 3.8 Alternative B: Wireless access (shaded area considered) Figure 3.9 Cumulative net cash flow (thousands) Chapter 04 Figure 4.1 The connectivity and interfaces of an LTE eNB Figure 4.2 The dimensioning process Figure 4.3 Simulation results showing the measured throughput of TCP CUBIC connections not in timeout over a 10 Mbps transport link serving a first in, first out (FIFO) buffer with maximum delay thresholds set to 50, 100 and 200 ms. The buffer was managed by RED with minimum threshold set to 50% of the maximum delay and maximum drop probability set to 10%. The number of simultaneous TCP connections was increased from 30 to 200. The available per connection nominal bandwidth is the link capacity divided by the number of TCP connections Figure 4.4 Simulation results showing the ratio of the measured throughput and the available (nominal) bandwidth of TCP CUBIC connections not in timeout. The simulation setup was the same as the one shown in Figure 4.3. As the load is increased by starting more and more connections, an increasing number of connections enter in timeout, thus the available bandwidth is taken by those connections that are not in timeout Figure 4.5 Simulation results showing the ratio of the TCP CUBIC in exponential back- off. The simulation setup is the same as the one in Figure 4.3. As the load is increased by starting more and more connections, increased numbers of connections enter timeout Figure 4.6 The anatomy of an HTTP object download Figure 4.7 The anatomy of a Web page download Figure 4.8 Simulation evaluation of the simple capacity calculation. The calculated capacity is shown in the function of the target download time (the quality requirement). The results show that at reasonable target download time values the overtime of the Web page download versus the target download time is less than 20% Figure 4.9 Web page download rate as a function of time. The same Web page was downloaded over an LTE system with a point-to-point transport link connecting the S- GW and the eNB. The link capacity was calculated with the simple capacity calculation based on the target Web page download time set to 5 s (a) and 20 s (b) Figure 4.10 The cumulative distribution function of the YouTube media rate in function of the itag Figure 4.11 The anatomy of a YouTube download Figure 4.12 Dimensioning requirements Figure 4.13 Relationship between the modeling levels Figure 4.14 Example of the daily load profile Figure 4.15 The source of multiplexing gain due to geographic diversity Figure 4.16 Identifying the busy hour: the busy hour is the time of the day when the highest amount of transport resources are needed. Due to the complexity of the traffic mix, the identification of the busy hour is difficult as the required amount of transport resources largely depends on the traffic mix Figure 4.17 Busy hour load and session level models. The session level models capture the dynamics of the session arrival process and allow the calculation of the number of simultaneous sessions during the busy hour Figure 4.18 The typical tail distribution of the number of simultaneous sessions. In this example, the average number of simultaneous sessions is 20 and the quantiles Q are p shown for p = 0.9, p = 0.99, and p = 0.999 Figure 4.19 Microscopic level behavior of the On–Off traffic source Figure 4.20 The number of sessions in the ‘On’ state changes dynamically Figure 4.21 Typical dependence of the required and effective bandwidth on the number of simultaneous sessions Figure 4.22 Packet level behavior within an On period Figure 4.23 Link capacity and the arrival process: if identical streaming sources with deterministic TTI are aggregated, the required link capacity will be a function of the aggregated traffic process. In this example, the packets are always served before a new packet arrives; therefore, the required link capacity is not increasing with the number of sessions Figure 4.24 Link capacity and the arrival process: if identical streaming sources with deterministic TTI are aggregated, the required link capacity will be a function of the aggregated traffic process. In this example, the packets are not served before a new packet arrives; therefore, the link capacity is increasing with the number of sessions. Dimensioning should guarantee that the delay of the second, third, etc. packet is below the maximum allowed time Figure 4.25 Link capacity and the arrival process: if identical streaming sources with deterministic TTI are aggregated, the required link capacity will be a function of the aggregated traffic process. If there is no strict delay requirement, dimensioning should guarantee that the buffer content is served during one TTI. The required link capacity in this case is linearly increasing with the number of connections Figure 4.26 Packet and burst level queues in the PDF of the buffer occupancy Y Figure 4.27 The TCP source is probing the system for the available bandwidth. As at the connection setup it has no knowledge on the system conditions, data transfer is started at a low rate, which is increased exponentially (slow-start). During this period, the source is not using the system efficiently. In a similar way, a slow-start is executed after a timeout or a longer idle period, during which the resources are not fully utilized Figure 4.28 When the requested object (file, software update, etc.) is large enough, the TCP connection spends the majority of its lifetime as an elastic connection Figure 4.29 The TCP connections established during a Web browsing session are not acting as elastic sources as they spend significant time within the slow-start phase Figure 4.30 The difference between the real and ideal elastic sources Figure 4.31 Users may establish TCP connections through several S-GW to various content servers, that is the TCP connections served by an eNB can have diverse routes Figure 4.32 Queuing model of a transport node. The figure assumes that the outgoing link is the bottleneck link; therefore, the node is modeled by its output scheduler Figure 4.33 An example of delay and waiting packets in a node Figure 4.34 Example of a transport network topology with uplink and downlink traffic Figure 4.35 Queueing network model of the reference transport topology. The topology is converted into two independent queuing networks with directed links corresponding to the uplink and downlink traffic considered not correlated Figure 4.36 Example of delay requirements derived from the expectations of different connections sharing the same link Figure 4.37 End-to-end protection over two domains Chapter 05 Figure 5.1 Detailed planning is a continuous process Figure 5.2 MPLS pseudowire Figure 5.3 MPLS L2 VPN supports multipoint-to-multipoint connectivity Figure 5.4 L3 VPN Figure 5.5 Network deployment example using IP access Figure 5.6 Network deployment example using Ethernet in the access Figure 5.7 Simple network topologies Figure 5.8 Ring availability from an eNB point of view presented as an equivalent parallel system Figure 5.9 HSRP/VRRP principle Figure 5.10 BFD-triggered gateway protection Figure 5.11 Synchronization protection using IEEE1588v2 timing over packet Figure 5.12 A simple QoS model for an access layer transport node. Other scheduler types could also be used Figure 5.13 WRR/WFQ scheduler with four queues Figure 5.14 Strict priority queuing scheduler with four queues Figure 5.15 The operation of traffic shaping. Transport node #2 has a buffer size of four packets Figure 5.16 Traffic shaping and policing for the leased line provider use case. Service provider polices packets entering its network Figure 5.17 An example of distribution of data traffic over 24 hr for three sites and at the network level Figure 5.18 Example load distribution factor per site versus site daily traffic volume. Each dot is one site