A comprehensive resource containing the operating principles and key insights of LTE networks performance optimization
1,377 119 34MB
English Pages 848 [845] Year 2017
Table of contents :
fmatter
ch1
ch2
ch3
ch4
ch5
ch6
ch7
ch8
ch9
ch10
ch11
ch12
ch13
ch14
ch15
ch16
ch17
ch18
app
gloss
refs
index
ins
LTE Optimization Engineering Handbook
LTE Optimization Engineering Handbook Xincheng Zhang
China Mobile Group Design Institute Co., Ltd. Beijing, China
This edition first published 2018 © 2018 John Wiley & Sons Singapore Pte. Ltd All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from this title is available at http://www.wiley.com/go/ permissions. The right of Xincheng Zhang to be identified as the author of this work has been asserted in accordance with law. Registered Offices John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA John Wiley & Sons Singapore Pte. Ltd, 1 Fusionopolis Walk, #07-01 Solaris South Tower, Singapore 138628 Editorial Office 1 Fusionopolis Walk, #07-01 Solaris South Tower, Singapore 138628 For details of our global editorial offices, customer services, and more information about Wiley products visit us at www.wiley.com. Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some content that appears in standard print versions of this book may not be available in other formats. Limit of Liability/Disclaimer of Warranty While the publisher and authors have used their best efforts in preparing this work, they make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives, written sales materials or promotional statements for this work. The fact that an organization, website, or product is referred to in this work as a citation and/or potential source of further information does not mean that the publisher and authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for your situation. You should consult with a specialist where appropriate. Further, readers should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages. Library of Congress Cataloging‐in‐Publication Data Names: Zhang, Xincheng, 1970– author. Title: LTE optimization engineering handbook / Xincheng Zhang, China Mobile Group Design Institute Co., Beijing, China. Description: Hoboken, NJ, USA : Wiley, [2017] | Includes bibliographical references and index. | Identifiers: LCCN 2017019394 (print) | LCCN 2017022857 (ebook) | ISBN 9781119159001 (pdf ) | ISBN 9781119158998 (epub) | ISBN 9781119158974 (cloth) Subjects: LCSH: Long-Term Evolution (Telecommunications)–Handbooks, manuals, etc. | Wireless communication systems–Handbooks, manuals, etc. | Computer network protocols–Handbooks, manuals, etc. Classification: LCC TK5103.48325 (ebook) | LCC TK5103.48325 .Z4325 2017 (print) | DDC 621.3845/6–dc23 LC record available at https://lccn.loc.gov/2017019394 Cover Design: Wiley Cover Images: (Yin Yang) © alengo/Gettyimages; (Feng shui compass) © Liuhsihsiang/Gettyimages Set in 10/12pt Warnock by SPi Global, Pondicherry, India 10 9 8 7 6 5 4 3 2 1
v
Contents About the Author xvi Preface xvii Part 1 1
LTE Basics and Optimization Overview 1
LTE Basement
3
1.1 LTE Principle 3 1.1.1 LTE Architecture 6 1.1.2 LTE Network Interfaces 7 1.2 LTE Services 11 1.2.1 Circuit‐Switched Fallback 12 1.2.2 Voice over LTE 13 1.2.3 IMS Centralized Services 16 1.2.4 Over the Top Solutions 16 1.2.5 SMS Alternatives over LTE 17 1.2.6 Converged Communication 19 1.3 LTE Key Technology Overview 19 1.3.1 Orthogonal Frequency Division Multiplexing 20 1.3.2 MIMO 21 1.3.3 Radio Resource Management 22 2
LTE Optimization Principle and Method
24
2.1 LTE Wireless Optimization Overview 24 2.1.1 Why LTE Wireless Optimization 24 2.1.2 Characters of LTE Optimization 24 2.1.3 LTE Joint Optimization with 2G/3G 25 2.1.4 Optimization Target 25 2.2 LTE Optimization Procedure 26 2.2.1 Optimization Procedure Overview 26 2.2.2 Collection of Mass Nerwork Measurement Data 28 2.2.3 Measurement Report Data Analysis 30 2.2.4 Signaling Data Analysis 31 2.2.5 UE Positioning 32 2.2.5.1 Timing Advance 33 2.2.5.2 Location Accuracy Evaluation 35
vi
Contents
2.2.5.3 Location Support 36 2.2.5.4 3D Geolocation 37 2.2.6 Key Performance Indicators Optimization 42 2.2.7 Technology Evolution of Optimization 43 2.3 LTE Optimization Key Point 44 2.3.1 RF Optimization 44 2.3.1.1 RSRP/RSSI/SINR/CINR 44 2.3.1.2 External Interference 48 2.3.2 CQI versus RSRP and SINR 51 2.3.2.1 CQI Adjustment 51 2.3.2.2 SINR Versus Load 54 2.3.2.3 SINR Versus MCS 56 2.3.3 Channel Power Configuration 58 2.3.3.1 RE Power 58 2.3.3.2 CRS Power Boosting 64 2.3.3.3 Power Allocation Optimization 66 2.3.4 Link Adaption 67 2.3.5 Adaptive Modulation and Coding 69 2.3.6 Scheduler 70 2.3.6.1 Downlink Scheduler 72 2.3.6.2 Uplink Scheduler 74 2.3.7 Radio Frame 75 2.3.8 System Information and Timers 76 2.3.8.1 System Information 76 2.3.8.2 Timers 81 2.3.9 Random Access 83 2.3.10 Radio Admission Control 85 2.3.11 Paging Control 86 2.3.11.1 Paging 86 2.3.11.2 Paging Capacity 92 2.3.11.3 Paging Message Size 95 2.3.11.4 Smart Paging 95 2.3.11.5 Priority Paging 96 2.3.12 MIMO and Beamforming 97 2.3.12.1 Basic Multi‐Antenna Techniques 100 2.3.12.2 2D‐Beamforming 101 2.3.12.3 2D MIMO and Parameters 104 2.3.12.4 Massive‐MIMO 105 2.3.13 Power Control 107 2.3.13.1 PUSCH/PUCCH Power Control 107 2.3.13.2 PRACH Power Control 109 2.3.14 Antenna Adjustment 111 2.3.14.1 Antenna Position 112 2.3.14.2 Remote Electrical Tilt 113 2.3.14.3 Antenna Azimuths and Tilts Optimization 117 2.3.14.4 VSWR Troubleshooting 118 2.3.15 Main Key Performance Indicators 120
Contents
Part 2 3
Main Principles of LTE Optimization 123
Coverage Optimization
125
3.1 Traffic Channel Coverage 125 3.1.1 Parameters of Coverage 126 3.1.2 Weak Coverage 128 3.1.2.1 DL Coverage Hole 128 3.1.2.2 UL Weak Coverage 128 3.1.2.3 UL and DL Imbalance 129 3.1.3 Overlapping Coverage 129 3.1.4 Overshooting 130 Tx1/Tx2 RSRP Imbalance 132 3.1.5 3.1.6 Extended Coverage 132 3.1.7 Cell Border Adjustment 135 3.1.8 Vertical Coverage 137 3.1.9 Parameters Impacting Coverage 138 3.2 Control Channel Coverage 138 4
Capacity Optimization
140
4.1 RS SINR 140 4.2 PDCCH Capacity 141 4.3 PUCCH Capacity 144 4.3.1 Factors Affecting PUCCH Capacity 145 4.3.2 PUCCH Dimensioning Example 151 4.4 Number of Scheduled UEs 152 4.5 Spectral Efficiency 153 4.6 DL Data Rate Optimization 154 4.6.1 Limitation Factor 156 4.6.2 Model of DL Data Throughput 157 4.6.3 UDP/TCP Protocol 158 4.6.4 MIMO 161 4.6.4.1 DL MIMO 161 4.6.4.2 4Tx/4Rx Performance 163 4.6.4.3 Transmission Mode Switch 163 4.6.4.4 UL MU‐MIMO 164 4.6.5 DL PRB Allocation and Utilization Mechanism 165 4.6.6 DL BLER 167 4.6.7 Impact of UE Velocity 169 4.6.8 Single User Throughput Optimization 170 4.6.8.1 Radio Analysis – Assignable Bits 171 4.6.8.2 Radio Analysis – CFI and Scheduling 171 4.6.8.3 Radio Analysis – HARQ 171 4.6.9 Avarage Cell Throughput Optimization 172 4.6.10 Cell Edge Throughput Optimization 172 4.6.11 Some Issues of DL Throughput 173 4.6.11.1 Antenna Diversity not Balanced 173 4.6.11.2 DL Grant is not Enough 173 4.6.11.3 Unstable Rate 175
vii
viii
Contents
4.7 UL Data Rate Optimization 175 4.7.1 Model of UL Data Throughput 176 4.7.2 UL SINR and PUSCH Data Rate 176 4.7.3 PRB Stretching and Throughput 179 4.7.4 Single User Throughput Optimization 180 4.7.4.1 Radio Analysis – Available PRBs 181 4.7.4.2 Radio Analysis—Link Adaptation 181 4.7.4.3 Radio Analysis – PDCCH 182 4.7.5 Cell Avarage and Cell‐edge Throughput Optimization 182 4.7.6 Some Issues of UL Throughput 183 4.8 Parameters Impacting Throughput 185 5
Internal Interference Optimization
188
5.1 Interference Concept 188 5.2 DL Interference 190 5.2.1 DL Interference Ratio 191 5.2.2 Balance Between SINR and RSRP 192 5.3 UL Interference 192 5.3.1 UL Interference Detection 194 5.3.2 Generation of UL Interference 196 5.3.2.1 Cell Loading Versus Inter‐Cell Interference 196 5.3.2.2 Unreasonable UL Network Structure 197 5.3.2.3 Cross slot interference 199 5.3.3 PUSCH Tx Power Analysis 200 5.3.4 UL Effect of P0 and α 202 5.3.5 PRACH Power Control 204 5.3.6 SRS Power Control 206 5.3.7 Interference Rejection Combinin 209 5.4 Inter‐Cell Interference Coordination 210 5.5 UL IoT Control 210 5.5.1 UL Interference Issues and Possible Solutions 210 5.5.2 UL IoT Control Mechanism 210 5.5.3 PUSCH UL_SINR Target Calculation 212 5.5.4 UL Interference Criteria 213 6
Drop Call Optimization
216
6.1 Drop Call Mechanism 216 6.1.1 Radio Link Failure Detection by UE 217 6.1.2 RadioLink Failure Detection by eNB 220 6.1.2.1 Link Monitors in eNB 220 6.1.2.2 Time Alignment Mechanism 221 6.1.2.3 Maximum RLC Retransmissions Exceeded 224 6.1.3 RadioLink Failure Optimization and Recovery 225 6.2 Reasons of Call Drop and Optimization 227 6.2.1 Reasons of E‐RAB Drop 227 6.2.2 S1 Release 230 6.2.3 Retainability Optimization 233 6.3 RRC Connection Reestablishment 233 6.4 RRC Connection Supervision 239
Contents
7
7.1 7.2 7.3 7.4 7.5 7.6 8
Latency Optimization
244
Mobility Optimization
254
User Plane Latency 244 Control Plane Latency 247 Random Access Latency Optimization 247 Attach Latency Optimization 248 Paging Latency Optimization 250 Parameters Impacting Latency 250
8.1 Mobility Management 255 8.1.1 RRC Connection Management 256 Measurement and Handover Events 256 8.1.2 8.1.3 Handover Procedure 260 8.1.3.1 X2 Handover 261 8.1.3.2 S1 Handover 267 8.1.3.3 Key point of X2/S1 Handover 267 8.2 Mobility Parameter 269 8.2.1 Attach and Dettach 272 8.2.2 UE Measurement Criterion in Idle Mode and Cell Selection 273 8.2.3 Cell Priority 276 8.3 Intra‐LTE Cell Reselection 276 8.3.1 Cell Reselection Procedure 278 8.3.2 Inter‐Frequency Cell Reselection 279 8.3.3 Cell Reselection Parameters 282 8.3.4 Inter‐Frequency Reselection Optimization 283 8.4 Intra‐LTE Handover Optimization 285 8.4.1 A3 and A5 Handover 285 8.4.2 Data Forwarding 290 8.4.3 Intra‐Frequency Handover Optimization 291 8.4.4 Inter‐Frequency Handover Optimization 292 8.4.5 Timers for Handover Failures 296 8.5 Neighbor Cell Optimization 297 8.5.1 Intra‐LTE Neighbor Cell Optimization 297 8.5.1.1 Neighbor Relations Table 297 8.5.1.2 ANR 298 8.5.2 Suitable Neighbors for Load Balancing 299 8.6 Measurement Gap 299 8.6.1 Measurement Gap Pattern 299 8.6.2 Measurement Gap Versus Period of CQI Report and DRX 304 8.6.3 Impact of Throughput on Measurement Gap 304 8.7 Indoor and Outdoor Mobility 305 8.8 Inter‐RAT Mobility 306 8.8.1 Inter‐RAT Mobility Architecture and Key Technology 307 8.8.2 LTE to G/U Strategy 309 8.8.3 Reselection Optimization 314 8.8.3.1 LTE to UTRAN 315 8.8.3.2 UTRAN to LTE 319 8.8.4 Redirection Optimization 320 8.8.4.1 LTE to UTRAN 320
ix
x
Contents
8.8.4.2 UTRAN to LTE 322 8.8.5 PS Handover Optimization 322 8.8.5.1 LTE to UTRAN 322 8.8.5.2 UTRAN to LTE 324 8.8.6 Reselection and Redirection Latency 325 8.8.7 Optimization Case Study 326 8.9 Handover Interruption Time Optimization 326 8.9.1 Control Plane and User Plane Latency 329 8.9.2 Inter‐RAT Mobility Latency 332 8.10 Handover Failure and Improvement 332 8.11 Mobility Robustness Optimization 335 8.12 Carrier Aggregation Mobility Optimization 341 8.13 FDD‐TDD Inter‐mode Mobility Optimization 345 8.14 Load Balance 346 8.14.1 Inter‐Frequency Load Balance 346 8.14.2 Inter‐RAT Load Balance 348 8.14.3 Load Based Idle Mode Mobility 349 8.15 High‐Speed Mobile Optimization 351 8.15.1 High‐Speed Mobile Feature 353 8.15.2 Speed‐Dependent Cell Reselection 354 8.15.3 PRACH Issues 356 8.15.4 Solution for Air to Ground 358 9
Traffic Model of Smartphone and Optimization
360
9.1 Traffic Model of Smartphone 360 9.1.1 QoS Mechanism 362 9.1.2 Rate Shaping and Traffic Management 366 9.1.3 Traffic Model 371 9.2 Smartphone‐Based Optimization 372 9.3 High‐Traffic Scenario Optimization 372 9.3.1 Resource Configuration 374 9.3.2 Capacity Monitoring 375 9.3.3 Special Features and Parameters for High Traffic 377 9.3.4 UL Noise Rise 379 9.3.5 Offload Solution and Parameter Settings 379 Part III 10
Voice Optimization of LTE 383
Circuit Switched Fallback Optimization
385
10.1 Voice Evolution 385 10.2 CSFB Network Architecture and Configuration 386 10.2.1 CSFB Architecture 386 10.2.2 Combined Register 387 10.2.3 CSFB Call Procedure 392 10.2.3.1 Fallback Options 392 10.2.3.2 RRC Release with Redirection 393 10.2.3.3 CSFB Call Procedure 395 10.2.4 Mismatch Between TA and LA 397
Contents
10.3 CSFB Performance Optimization 402 10.3.1 CSFB Optimization 402 10.3.1.1 Main Issues of CSFB 402 10.3.1.2 CSFB Optimization Method 403 10.3.2 CSFB Main KPI 407 10.3.3 Fallback RAT Frequency Configuration Optimization 409 10.3.4 Call Setup Time Latency Optimization 411 10.3.4.1 ESR to Redirection Optimization 416 10.3.4.2 Twice Paging 416 10.3.5 Data Interruption Time 418 10.3.6 Return to LTE After Call Complete 419 10.4 Short Message Over CSFB 422 10.5 Case Study of CSFB Optimization 423 10.5.1 Combined TA/LA Updating Issue 423 10.5.2 MTRF Issue 425 10.5.3 Track Area Update Reject After CSFB 425 10.5.3.1 No EPS Bearer Context Issue 428 10.5.3.2 Implicitly Detach Issue 428 10.5.3.3 MS Identity Issue 428 10.5.4 Pseudo Base Station 428 11
VoLTE Optimization
434
11.1 VoLTE Architecture and Protocol Stack 435 11.1.1 VoLTE Architecture 435 11.1.2 VoLTE Protocol Stack 435 11.1.3 VoLTE Technical Summary 438 11.1.4 VoLTE Capability in UE 439 11.2 VoIP/Video QoS and Features 442 11.2.1 VoIP/Video QoS 442 11.2.2 Voice Codec 444 11.2.3 Video Codec 446 11.2.4 Radio Bearer for VoLTE 449 11.2.5 RLC UM 454 11.2.6 Call Procedure 457 11.2.6.1 LTE Attach and IMS Register 458 11.2.6.2 E2E IMS Flow 458 11.2.6.3 Video Phone Session Handling 462 11.2.7 Multiple Bearers Setup and Release 466 11.2.8 VoLTE Call On‐Hold/Call Waiting 467 11.2.9 Differentiated Paging Priority 468 11.2.10 Robust Header Compression 470 11.2.10.1 RoHC Feature 470 11.2.10.2 Gain by RoHC 470 11.2.11 Inter‐eNB Uplink CoMP for VoLTE 475 11.3 Semi‐Persistent Scheduling and Other Scheduling Methods 477 11.3.1 SPS Scheduling 477 11.3.2 SPS Link Adaptation 478 11.3.3 Delay Based Scheduling 481 11.3.4 Pre‐scheduling 482
xi
xii
Contents
11.4 PRB and MCS Selection Mechanism 484 11.4.1 Optimized Segmentation 484 11.4.2 PRB and MCS Selection 485 11.5 VoLTE Capacity 486 11.5.1 Control Channel for VoLTE 487 11.5.2 Performance of Mixed VoIP and Data 488 11.6 VoLTE Coverage 491 11.6.1 VoIP Payload and RoHC 492 11.6.2 RLC Segmentation 492 11.6.3 TTI Bundling 498 11.6.4 TTI Bundling Optimization 502 11.6.5 Coverage Gain with RLC Segmentation and TTI Bundling 507 11.6.6 MCS/TBS/PRB Selection 509 11.6.7 Link Budget 510 11.7 VoLTE Delay 513 11.7.1 Call Setup Delay 516 11.7.1.1 Call Setup Time 516 11.7.1.2 Reasons for Long Call Setup Time 516 11.7.2 Conversation Start Delay 519 11.7.3 RTP Delay 521 11.7.4 Handover Delay and Optimization 525 11.8 Intra‐LTE Handover and eSRVCC 527 11.8.1 Intra‐Frequency Handover 527 11.8.2 Inter‐Frequency Handover 528 11.8.3 Single Radio Voice Call Continuity Procedure 529 11.8.4 SRVCC Parameters Optimization 539 11.8.4.1 Handover Parameters 539 11.8.4.2 SRVCC–Related Timer 539 11.8.5 aSRVCC and bSRVCC 543 11.8.6 SRVCC Failure 543 11.8.7 Reducing SRVCC Voice Gap and eSRVCC 545 11.8.7.1 Voice Interruption Time during SRVCC 545 11.8.7.2 eSRVCC 549 11.8.8 Fast Return to LTE 552 11.8.9 Roaming Behavior According to Network Capabilities 555 11.9 Network Quality and Subjective Speech Quality 555 11.9.1 Bearer Latency 558 11.9.2 MoS 561 11.9.2.1 Voice Quality 561 11.9.2.2 Video Quality 570 11.9.3 Jitter 571 11.9.4 Packet Loss 572 11.9.5 One Way Audio 575 11.9.6 PDCP Discard Timer Operation 576 11.10 Optimization 577 11.10.1 Distribution of Main Indicators of Field Test 580 11.10.2 Compression Ratio and GBR Throughput 584 11.10.3 RB Utilization 584 11.10.4 BLER Issue 587
Contents
11.10.5 Quality Due to Handover 589 11.10.6 eSRVCC Handover Issues 589 11.10.7 Packet Loss 592 11.10.7.1 Packet Loss due to Poor RF 592 11.10.7.2 Packet Loss due to Massive users 592 11.10.7.3 Packet Loss Due to Insufficient UL grant 592 11.10.7.4 Packet Loss due to Handover 601 11.10.7.5 Packet Loss Due to Network Issue 601 11.10.8 Call Setup Issues 601 11.10.8.1 Missed Pages 602 11.10.8.2 IMS Issues 604 11.10.8.3 Dedicated Bearer Setup Issues 609 11.10.8.4 CSFB Call Issues 612 11.10.8.5 aSRVCC Failure 612 11.10.8.6 RF Issues 612 11.10.8.7 Frequent TFT Updates 617 11.10.8.8 Encryption Issue 618 11.10.9 Call Drop 619 11.10.9.1 Call Drop 619 11.10.9.2 Radio Link Failure 622 11.10.9.3 RTP‐RTCP Timeout 624 11.10.9.4 RLC/PDCP SN Length Mismatch 626 11.10.9.5 IMS Session Drop 626 11.10.9.6 eNB/MME Initiated Drop 632 11.10.10 Packet Aggregation Level 632 11.10.11 VoIP Padding 633 11.10.12 VoIP Ralated Parameters 635 11.10.13 Video‐Related Optimization 635 11.10.13.1 Video Bit Rate and Frame Rate 637 11.10.13.2 Video MoS and Audio/Video Sync 637 11.10.14 IMS Ralated Timer 637 11.11 UE Battery Consumption Optimization for VoLTE 638 11.11.1 Connected Mode DRX Parameter 643 11.11.2 DRX Optimization 644 11.11.2.1 State Estimation 644 11.11.2.2 DRX Optimization and Parameters 644 11.11.2.3 KPI Impacts with DRX 648 11.11.3 Scheduling Request Periodicity and Disabling of Aperiodic CQI 652 11.12 Comparation with VoLTE and OTT 654 11.12.1 OTT VoIP User Experience 654 11.12.2 OTT VoIP Codec 657 11.12.3 Signaling Load of OTT VoIP 658 Part IV 12
Advanced Optimization of LTE 663
PRACH Optimization
665
12.1 Overview 665 12.2 PRACH Configuration Index 669 12.3 RACH Root Sequence 673
xiii
xiv
Contents
12.4 PRACH Cyclic Shift 674 12.4.1 PRACH Cyclic Shift Optimization 674 12.4.2 Rrestricted Set 679 12.5 Prach Frequency Offset 682 12.6 Preamble Collision Probability 683 12.7 Preamble Power 684 12.8 Random Access Issues 687 12.9 RACH Message Optimization 689 12.10 Accessibility Optimization 692 12.10.1 Reasons for Poor Accessibility 692 12.10.2 Accessibility 693 12.10.3 Accessibility Analysis Tree 695 12.10.4 Call and Data Session Setup Optimization 697 12.10.5 RACH Estimation for Different Traffic Profile 698 13
Physical Cell ID Optimization
702
Tracking Areas Optimization
711
13.1 Overview 702 13.2 PCI Optimization Methodology 703 13.2.1 PCI Group Optimization 705 13.2.2 PCI Code Reuse Distance 705 13.2.3 Mod3/30 Discrepancy Analysis 708 13.2.4 Collision and Confusion 708 13.3 PCI Optimization 709 14
14.1 TA Optimization 712 14.1.1 TA Update Procedure 713 14.1.2 TA Optimization and TAU Failure 715 14.2 TA List Optimization 716 14.3 TAU Reject Analysis and Optimization 719 15
Uplink Signal Optimization
721
15.1 Uplink Reference Signal Optimization 721 15.1.1 Coding Scheme of UL RS 722 15.1.2 Correlation of UL Sequence Group 723 15.1.2.1 UL Sequence Group Hopping 725 15.1.2.2 UL Sequence Hopping 726 15.1.2.3 UL Cyclic Shift Hopping 726 15.1.3 UL Sequence Group Optimization 727 15.2 Uplink Sounding Signal Optimization 729 15.2.1 SRS Characters 730 15.2.2 Wideband SRS Coverage 736 15.2.3 Dynamic SRS Adjustment Scheme 736 15.2.4 SRS Selection Dimension and Confliction 737 15.2.5 SRS Conflict and Optimization 739 16
HetNet Optimization
741
16.1 UE Geolocation and Identification of Traffic Hot Spots 741 16.2 Wave Propagation Characteristics for HetNet 745
Contents
16.3 New Features in HetNet 746 16.4 Combined Cell Optimization 747 16.5 Cell Range Expansion Offset 748 16.6 HetNet Cell Reselection and Handover Optimization 751 17
QoE Evaluation and Optimization Strategy
17.1 QoE Modeling 753 17.2 Data Collecting and Processing 756 17.3 QoE‐Based Traffic Evaluation 757 17.3.1 Online Video QoE 757 17.3.1.1 Video Quality Monitoring Methods 761 17.3.1.2 RATE Adaptive Video Codecs 763 17.3.1.3 Streaming KPI and QoE 764 17.3.1.4 Video Optimization 766 17.3.2 Voice QoE 769 17.3.3 Data Service QoE 770 17.3.3.1 Web browsing 770 17.3.3.2 Online Gaming 774 17.4 QoE Based Optimization 776 18
Signaling‐Based Optimization
780
18.1 S1‐AP Signaling 780 18.1.1 NAS signaling 782 18.1.2 Inactivity Supervision 783 18.1.3 UE signaling Management 785 18.2 Signaling radio bearers 786 18.3 Signaling Storm 788 18.4 Signaling Troubleshooting Method 788 18.4.1 Attach Failure 788 18.4.2 Service Request Failure 796 18.4.3 S1/X2‐Based Handover 796 18.4.4 eSRVCC Failure 798 18.4.5 CSFB Failure 800 Appendix 802 Glossary of Acronyms 820 References 823 Index 825
752
xv
xvi
About the Author Xincheng Zhang graduated from the Beijing University of Posts and Telecommunications in 1992. He has worked in mobile communication for 25 years as a technical expert with a solid understanding of wireless communication technologies. Starting out in the early days of GSM rollouts, he has many years of planning and optimization experience in 2G, 3G, 4G, and 5G networks, working in operator and vendor environments. He is working as a senior wireless network specialist in the fields of antenna arrays, analog/digital signal processing, radio resource management, and propagation modeling, and so on. He has participated in many large‐scale wireless communication system designs and optimization for a variety of cellular systems using various radio access technologies, including GSM, CDMA, UMTS, and LTE.
xvii
Preface Mobile communication has become ubiquitous and mobile Internet traffic is continuously growing due to the technology that provides broadband data rates (3G, LTE) and the growing number of mobile dongles and mobile devices like tablets or smartphones that enable the usage of a tremendous number of internet applications through the mobile access. Mobility, broadband, and new device technology have changed the way people connect and communicate. Smartphones have changed the characteristics of the control and user plane, leading to a huge impact on RAN and e2e network capacity, end‐user experience, and perception of the network, which has changed with the advent of new devices and applications. Subscribers want the same internet experience that they have at home, anytime, anywhere, so the long‐term network is under strain and optimization is needed. Many of the new services aim to enhance the experience of a phone conversation by allowing sharing of content other than speech. The quality of all these services needs to be monitored to ensure that users experience a high‐quality service. Low‐bit cost is an essential requirement in a scenario where high volumes of data are being transmitted over the mobile network. To achieve the proposed goals, a very flexible network that aggregates various radio access technologies is created. This network should provide high bandwidth, from 50‐100 Mbps for high mobility users, to 1Gbps for low mobility users, technologies that permit fast handoffs, which is necessary in a QoS framework that enables fair and efficient medium sharing among users. The core of this network should be based on internet protocol version 6—IPv6, the probable convergence platform of future services. The other key factor to the success of the network is that the terminals must be able to provide wireless services anytime, everywhere, and must adapt seamlessly to multiple wireless networks, each with different protocols and technologies. Subscriber loyalty has shifted to devices and applications; quality of experience becomes the fundamental service provider’s differentiation. In this background, the 3GPP long term evolution (LTE) is created and adopted all over the world. High‐speed, high‐capacity data standard for mobile devices is on its way to becoming a globally deployed standard for the fourth generation of mobile networks (4G) supported by all major players in the industry. LTE builds on EUTRAN, a new generation radio access network, and the evolved packet core (EPC), which provides flexible spectrum usage and bandwidths, high data rate, low latency, and optimized resource usage. As LTE has been used as a mobile broadband service, we need to understand the effects of the LTE terminals providing services and how to optimize the network. Actually, for operators, the challenge is not only to optimize 2G, 3G, and 4G but also how to balance the use of those systems, including WiFi. The entire service delivery chain needs to be optimized and the optimization aims to improve network efficiency and the mobile broadband service quality.
xviii
Preface
It is known that LTE doesn’t have basic voice and SMS support. To mitigate this, 3GPP roposes a fallback to circuit‐switched (CS) network for voice and SMS. Although voice has p loosened its weight in the overall user bill with the rise of more and more data services, voice is the dominant source of revenue for operators and is expected to remain so for the foreseeable future. On one hand, 3GPP defines the concept of CS fallback for the EPC, which forces the UE to fall back to the GERAN/UTRAN network where the CS procedures are carried out. On the other hand, voice over LTE will become a mainstream mobile voice technology. VoLTE ecosystem is building up fast due to its strong end‐to‐end VoLTE solution portfolio including the LTE radio, EPC, mobile softswitch, IMS, and its extensive delivery capabilities of complex end‐to‐ end projects. It is the world’s most innovative voice solution for LTE‐based networks and big VoLTE growth is expected since wide‐scale commercial VoLTE started in Korea during 2012. The operators will want to have the best possible observability for this new voice service with fast call setup, low latency, and high speech quality. Actually, VoLTE will be among the most critical and complex technologies mobile operators will ever deploy as VoLTE testing is quite complex due to inherent intricacy of the technology covering the IMS/EPC core, radio network, and UE/IMS client. They expect to be able to monitor how their customers experience accessibility, retainability, as well as the quality of the voice service. Obviously, much of the observability is already in place, but there are reasons to believe that there are missing parts. Under this background, to meet customers’ requirements for high‐quality networks, LTE trial networks must be optimized during and after project implementation. The basis and the main inputs that allowed the creation of this handbook were based on optimization experience, whereas the scope of this book is to provide network engineers with a set of processes and tasks to guide them through the troubleshooting and optimization. For a network optimization engineer, he/she needs to know how good the quality of mobile broadband applications is, and how the network capabilities impact the performance, and how to identify the most critical network KPIs that impact customer experiences. This book is divided into four parts. The first is called “LTE Basics and Optimization Overview,” and proceeds with an introduction to general principles of data transfer of LTE. This chapter is dedicated to the reader who is not acquainted with this area. The second part, titled “Main Principles of LTE Optimization,” and the third part, “Voice Optimization of LTE,” makes up the core of the book, since it describes coverage, capacity, interference, mobility optimization, and includes two chapters that provide step‐by‐step optimization of CFSB and VoLTE. The fourth part “Advanced Optimization of LTE” takes a more applied perspective in PRACH, PCI, TA, QoE, Hetnet, and signaling optimization. Thanks to the many people in China who shared their views acquired from years of experience and valuable insights in wireless optimization, the Optimization Handbook covers the basics of optimization rules, solutions, and methods. It is evident that this book does not cover many other important areas of optimization of LTE networks. Nonetheless, I sincerely hope that readers will find the information presented to be interesting and useful to inspire you to go and do optimization with a renewed vigor in order to help you build a better LTE network. January 1, 2017
Xincheng Zhang
1
Part 1
LTE Basics and Optimization Overview
3
1 LTE Basement Mobile networks are rapidly transforming—traffic growth, bit rate increases for the user, increased bit rates per radio site, new delivery schemes (e.g., mobile TV, quadruple play, IMS), and a multiplicity of RANs (2G, 3G, HSPA, WiMAX, LTE)—are the main drivers of the mobile network evolution. The growth in mobile traffic is mainly driven by devices (e.g., smartphone and tablet) and applications (e.g., mainly web browsing and video streaming). To cope with the increasing demand, mobile networks have based their evolution on increasingly IP‐centric solutions. This evolution relies primarily on the introduction of IP transport, and secondly, on a redesign of the core nodes to take advantage of the IP backbones. The first commercial LTE network was opened by Teliasonera in Sweden in December 2009, and marks the new era of high‐speed mobile communications. The incredible growth of LTE network launches boomed between 2012 and 2016 worldwide. It is expected that more than 500 operators in nearly 150 countries will soon be running a commercial LTE network. Mobile data traffic has grown rapidly during the last few years, driven by the new smartphones, large displays, higher data rates, and higher number of mobile broadband subscribers. It is expected that the mobile broadband (MBB) subscriber numbers will double by 2020, reaching over 7 billion subscribers, that MBB data traffic will grow fourfold by 2020, reaching over 19 petabytes/ month. The internet traffic, MBB subscriber, and relative mobile data growth is illustrated in Figure 1.1.
1.1 LTE Principle To provide a fully mature, real‐time–enabled, and MBB network, structural changes are needed in the network. In 2005, the 3GPP LTE project was created to improve the Universal Mobile Telecommunications System (UMTS) mobile phone standard to cope with future requirements, which resulted in the newly evolved Release 8 (Rel 8) of the UMTS standard. The goals include improving efficiency, lowering costs, improving services, making use of new spectrum opportunities, and better integration with other open standards. Long‐term evolution (LTE) is selected as the next generation broadband wireless technology for 3GPP and 3GPP2. The LTE standard supports both FDD (frequency division duplex), where the uplink and downlink channel are separated in frequency, and TDD (time division duplex), where uplink and downlink share the same frequency channel but are separated in time. After Rel 8, Rel 9 was a relatively small update on top of Rel 8, and Rel 10 provided a major step in terms of data rates and capacity with carrier aggregation, higher‐order Multi‐Input‐Multi‐Output (MIMO) up to eight antennas in downlink and four antennas in uplink. The support for heterogeneous network (HetNet) was included in Rel 10, also known as LTE‐Advanced (Figure 1.2). LTE Optimization Engineering Handbook, First Edition. Xincheng Zhang. © 2018 John Wiley & Sons Singapore Pte. Ltd. Published 2018 by John Wiley & Sons Singapore Pte. Ltd.
Bandwidth demand
140,000.0 Video Conferencing
Video Streaming
Audio Streaming
120,000.0
P2P
100,000.0
Video Streaming / TV VoIP
80,000.0
Ecommerce
Web / Internet
Web Surfing
60,000.0
Rich text Email
high
40,000.0 VoIP (VoLTE)
low
Text Email
low
high
20,000.0 0.0
Delay demand
2011 2012 2013 2014 2015 2016 2017 2018 2019
Internet traffic on LTE
Mobile traffic type (Source: ABI Research) Mobile Data Traffic
8,000 7,000
18 000
6,000
16 000
Europe
LAT
APAC total
MEA
NAM
14 000
5,000
12 000
4,000
10 000 8 000
3,000
6 000 2,000
4 000 2 000
1,000
MBB subscriber growth
Figure 1.1 The internet traffic, MBB subscriber, and relative mobile data growth.
MBB data traffic
20
19
20
20
17
18
20
16
20
15
20
14
20
13
20
20
11
12
20
10
20
09
20
08
20
07
20
06
20
05
20
2020
2019
2018
2017
2016
2015
2014
2013
2012
2011
2009
2010
2008
2007
2006
0
2005
0
20
MBB Subscriber in Million
20 000
LTE Basement Phase 2+ (Release 97)
Release 99
Release 6
Release 8
GPRS 171.2 kbit/s
UMTS 2 Mbit/s
HSUPA 5.76 Mbit/s
LTE +300 Mbit/s
Release 9/10 LTE Advanced
GSM 9.6 kbit/s
EDGE 473.6 kbit/s
HSDPA 14.4 Mbit/s
Phase 1
Release 99
Release 5
HSPA+ 28.8 Mbit/s 42 Mbit/s Release 7/8
Figure 1.2 3GPP standard evolution. 5G 10 Gbps Peak Average
LTE-A
1 Gbps 5G in 2020 (Ave. ~1Gbps Peak ~5Gbps)
LTE 100 Mbps HSDPA, HDR
Cat. 11 (Ave. ~240Mbps, Peak ~600Mbps)
10 Mbps
1 Mbps
WCDMA, CDMA2000
Cat. 9 (Ave. ~180Mbps, Peak ~450Mbps) Cat. 6 (Ave. ~120Mbps, Peak 300Mbps) Cat. 4 (Ave. ~24Mbps, Peak 150Mbps) Cat. 3 (Ave. ~12Mbps, Peak 100Mbps)
100 Kbps
HSDPA (Ave. ~2Mbps, Peak 14Mbps) 2000
2005
2010
2015
2020
Figure 1.3 Downlink data rate evolution.
Among the design targets for the first release of the LTE standard are a downlink bit rate of 100 Mbit/s and a bit rate of 50 Mbit/s for the uplink with a 20‐MHz spectrum allocation. Smaller spectrum allocation will of course lead to lower bit rates and the general bit rate can be expressed as 5 bits/s/Hz for the downlink and 2.5 bits/s/Hz for the uplink. Rel 10 (LTE‐Advanced), was completed in June 2011 and the first commercial carrier aggregation network started in June 2013 (Figure 1.3). LTE provides global mobility with a wide range of services that includes voice, data, and video in a mobile environment with lower deployment cost. The main benefits of LTE include (Figure 1.4): ●●
Wide spectrum and bandwidth range, increased spectral efficiency and support for higher user data rates
5
LTE Optimization Engineering Handbook
100% CDF
6
“Average”Tput ~0.12bps/Hz 50% “Cell Edge”Tput ~0.06bps/Hz (95% coverage) 5% cell edge
cell centre Tput
Figure 1.4 Throughput of a user, 10 users evenly distributed in cell.
●●
●● ●● ●●
Reduced packet latency and rich multimedia user experience, excellent performance for outstanding quality of experience Improved system capacity and coverage as well as variable bandwidth operation Cost effective with a flat IP architecture and lower deployment cost Smooth interaction with legacy networks
LTE air interface uses orthogonal frequency division multiple access (OFDMA) for downlink transmission to achieve high peak data rates in high spectrum bandwidth. LTE uses single carrier frequency division multiple access (SC‐FDMA) for uplink transmission, a technology that provides advantages in power efficiency. LTE supports both FDD and TDD modes, with FDD, DL, and UL transmissions performed simultaneously in two different frequency bands, with TDD, DL, and UL transmissions performed at different time intervals within the same frequency band. LTE supports advanced adaptive MIMO, balance average/peak throughput, and coverage/cell‐edge bit rate. Compared to 3G, significant reduction in delay over air interface can be supported in LTE, and it is suitable for real‐time applications, for example, VoIP, PoC, gaming, and so on. Spectrum is a finite resource and FDD and TDD system will support the future demand, which are shown in Figure 1.5. TDD spectrum can provide 100‐150MHz of additional bandwidth per operator, TD‐LTE spectrum with large bandwidth will be a key to operators future network strategy and one of the way to address capacity growth. 1.1.1 LTE Architecture
LTE is predominantly associated with the radio access network (RAN). The eNodeB (eNB) is the component within the LTE RAN network. LTE RAN provides the physical radio link between the user equipment (UE) and the evolved packet core network. The system architecture evolution (SAE) specifications defines a new core network, which is termed as evolved packet core (EPC) including all internet protocol (IP) networking architectures (Figure 1.6). Evolved NodeB (eNB): Provides the LTE air interface to the UEs, the eNB terminates the user plane (PDCP/RLC/MAC/L1) and control plane (RRC) protocols. Among other things, it performs radio resource management and intra‐LTE mobility for the evolved access system. At the S1 interface toward the EPC, the eNB terminates the control plane (S1AP) and the user plane (GTP‐U).
LTE Basement B42 (3,5GHz)
250 TDD FDD 200
200
150
B43 (3,7GHz)
BW (MHz)
B40 (2,3GHz) B3 (1,8GHz)
100 B28 (700MHz)
100
90 75 60
50 B5 (850MHz)
35 25
B8 (900MHz)
B2 (1,9GHz) B39 20 (1,9GHz)
200
70 60 B1 B10 (2,1GHz) (1,7/2,1GHz)
50
B7 (2,6GHz) B38 (2,6GHz)
0 0
200
400
600
800
1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400 3600 3800 4000 Frequency (MHz)
Figure 1.5 Spectrum of LTE.
Mobility Management Entity (MME): A control plane node responsible for idle mode UE tracking and paging procedures. The Non‐Access Stratum (NAS) signaling terminates at the MME. Its main function is to manage mobility, UE identities, and security parameters. The MME is involved in the EPS bearer activation, modification, deactivation process, and is also responsible for choosing the SGW for a UE at the initial attach and at time of intra‐ LTE handover involving core network node relocation. PDN GW selection is also performed by the MME. It is responsible for authenticating the user by interacting with the home subscription server (HSS). Serving Gateway (SGW): This node routes and forwards the IP packets, while also acting as the mobility anchor for the user plane flow during inter‐eNB handovers and other 3GPP technologies (2G/3G systems using S4). For idle state UEs, the SGW terminates the DL data path and triggers paging when DL data arrives for the UE. Packet Data Network Gateway (PDN GW): Provides connectivity to the UE to external packet data networks by being the point of exit and entry of traffic for the UEs. The PDN GW performs among other policy enforcement, packet filtering for each user and IP address allocation. Policy and Charging Rules Function (PCRF): The PCRF supports policy control decisions and flow based charging control functionalities. Policy control is the process whereby the PCRF indicates to the PCEF (in PDN GW) how to control the EPS bearer. A policy in this context is the information that is going to be installed in the PCEF to allow the enforcement of the required services. Home Subscription Server (HSS): The HSS is the master database that contains LTE user information and hosts the database of the LTE users. 1.1.2 LTE Network Interfaces
LTE network can be considered of two main components: RAN and EPC. RAN includes the LTE radio protocol stack (RRC, PDCP, RLC, MAC, PHY). These entities reside entirely within the UE and the eNB nodes. EPC includes core network interfaces, protocols, and entities. These entities and protocols reside within the SGW, PGW, and MME nodes, and partially within the eNB nodes.
7
HSS • Subscription Profiles • Security information • MME (IP) address for UE S6a
UE • IMEI (equipment) • IMSI (SIM card) • Temporary GUTI • User Plane IP
MME • Mobility Management • Session Management • Security Management • Selects SGW based on TA • Selects PGW based on APN S11
S1-MME
PCRF:Policy & Charging Rules Function
DNS • TA to SGW IP query • APN to PGW query
PGW: Packet Data Network Gateway HSS:Home Subscriber Server EPC:Evolved packet Core PCRF • QoS rules • Charging rules
Gx
Rx
X2 S1u eNB • Radio control and resource management • Inter eNB communication via X2
Figure 1.6 Nodes and functions in LTE.
S5/S8 SGW • Data forwarding • Data buffering
SGi
PDN (i.e. IMS or internet)
PGW • Gateway between the internal EPC network and external PDNs • User IP address allocation • User plane QoS enforcement
SGW:Serving Gateway UE:User Equipment EUTRAN:Evolved UTRAN eNodeB:Enhanced Node B VLR:Visitor Location Register MSC:Mobile Switching Centre MME:Mobility Management Entity LTE Uu: LTE UTRAN UE Interface
LTE Basement
Uu: Uu is the air interface connecting the eNB with the UEs. The protocols used for the control plane are RRC on top of PDCP, RLC, MAC, and L1. The protocols used for the user plane are PDCP, RLC, MAC, and L1. LTE air interface supports high data rates. LTE uses OFDMA for downlink transmission to achieve high peak data rates in high spectrum bandwidth. LTE uses SC‐FDMA for uplink transmission, a technology that provides advantages in power efficiency. S1: The interface S1 is used to connect the MME/S‐GW and the eNB. The S1 is used for both the control plane and the user plane. The control plane part is referred to as S1‐ MME and the user plane S1‐U. The protocol used on S1‐MME is S1‐AP on the radio network layer. The transport network layer is based on IP transport, comprising SCTP on top of IP. The protocol used on S1‐U is based on IP transport with GTP‐U and UDP on top. The X2 interface is a new type of interface between the eNBs introduced by the LTE to perform the following functions: handover, load management, CoMP, and so on. X2‐UP protocol tunnels end‐user packets between the eNBs. The tunneling function supports are identification of packets with the tunnels and packet loss management. X2‐UP uses GTP‐U over UDP/ IP as the transport layer protocol similar to S1‐UP protocol. X2‐CP has SCTP as the transport layer protocol is similar to the S1‐CP protocol. The load management function allows exchange of overload and traffic load information between eNBs, which helps eNBs handle traffic load effectively. The handover function enables one eNB to hand over the UE to another eNB. A handover operation requires transfer of information necessary to maintain the services at the new eNB. It also requires establishment and release of tunnels between source and target eNB to allow data forwarding and informs the already prepared target eNB for handover cancellations. NAS is a control plane protocol that terminates in both the UE and the MME. It is transparently carried over the Uu and S1 interface. S6a: S6a interface enables transfer of subscription and authentication data between the MME and HSS for authenticating/authorizing user access to the EUTRAN. The S6a interface is involved in the following call flows, initial attach, tracking area update, service request, detach, HSS user profile management, and HSS‐initiated QoS modification, and so on. S11: Reference point between MME and SGW. This is a control plane interface for negotiating bearer plane resources with the SGW. The above‐mentioned LTE network interfaces are shown in Figure 1.7. IP connection between a UE and a PDN is called PDN connection or EPS session. Each PDN connection is represented by an IP address of the UE and a PDN ID (APN). As shown in Figure 1.8, there are two different layers of IP networking. The first one is the end‐to‐ end layer, which provides end‐to‐end connectivity to the users. This layers involves the UEs, the PGW, and the remote host, but does not involve the eNB. The second layer of IP networking is the EPC local area network, which involves all eNBs and the SGW/PGW node. The end‐to‐end IP communications is tunneled over the local EPC IP network using GTP/UDP/IP. Moreover, in LTE, IDs are used to identify a different UE, mobile equipment, and network element to make the EPS data session and bearer establishment, which can refer to Annex “LTE identifiers” for reference; the summary of IDs is shown in Table 1.1.
9
SGi Interface Comunicates CPG with external networks.
Diameter SCTP IP L2
S6a Interface AAA interface between MME and HSS that enables user access to the EPS
L1
IMS/External IP networks
HSS
IP L2
S11 Interface Control plane for creating, modifying and deleting EPS bearers. MME
IP L2
S1-MME Interface Reference point for control plane protocol between E-UTRAN and MME
TCP IP L2 L1
L1
S10 S1-AP
Gx
L2
L1
SCTP
Rx Interface Transport policy control, charging and QoS control.
IP S10 Interface AAA interface between MME and HSS that enables user access to the EPS
L1 Diameter
SGi
UDP
GTPv2-C UDP
PCRF
GTP-C
S6a
Rx
IP L2
S11
Gx Interface Provides transfer of policy and charging Rules from PCRF to PDN Gw.
PDN GW S5/S8 Serv GW
S5/S8 Interface Control and user plane tunneling between Serving GW and PDN GW
S1-U
S1-MME
L1
Diameter TCP IP L2 L1 GTP-C/GTP-U UDP IP
X2-AP
GTP-U
SCTP
UDP IP L2 L1
Figure 1.7 LTE network interfaces.
X2 Interface Connects neigboring eNBs
eNB X2
S1-U Interface Reference point for user plane protocol between E-UTRAN and MME
GTP-U
L2
UDP
L1
IP L2 L1
LTE Basement UE Control plane
end-to-end layer Uu
UE IP address
App IP PDCP
User plane
eNB
RRC Signaling
MME
SGW
PGW
S1 S11 GTP-C S11 GTP-C Signaling S1 Bearer
DRB
eNB
S5 Bearer SGW
S5/S8
PGW
SGi
IP GTPu UDP IP
APN
GTPu UDP IP
End to End service PDCP
GTPu UDP IP
EPS Bearer (ID) E-RAB (ID)
RLC
RLC
MAC
MAC
L2
L2
L2
PHY
PHY
L1
L1
L1
P D N
S1-u
Figure 1.8 LTE‐EPC control and data plane protocol stack. Table 1.1 Classification of LTE identification. Classification
LTE identification
UE ID
IMSI, GUTI, S‐TMSI, IP address, C‐RNTI, UE S1AP ID, UE X2AP ID
Mobile equipment ID
IMEI
Network element ID
GUMMEI, MMEI, Global eNB ID, eNB ID, ECGI, ECI, P‐GW ID
Location ID
TAI, TAC
Session/bearer ID
PDN ID (APN), EPS bearer ID, E‐RAB ID, DRB ID, LBI, TEID
1.2 LTE Services LTE is an all packet‐switched technology. The telephony service on LTE is a packet‐switched mobile broadband service relying on specific support in LTE radio and EPC, which is needed to meet the expectations of telephony. On the other hand, the handling of voice traffic on LTE handsets is evolving as the mobile industry infrastructure evolves toward higher, eventually ubiquitous, and finally, LTE availability. Central to the enablement of LTE smartphones is to meet today’s very high expectation for the mobile user experience and to evolve the entire communications experience by augmenting voice with richer media services. Voice solutions of LTE include VoLTE/SRVCC, RCS, OTT, CSFB, SVLTE, and so on. LTE radio and EPC architecture does not have a circuit‐switched (CS) domain available to handle voice calls as being done in 2G/3G. The voice traffic in the LTE network is handled through different procedures. The first one, which is mainly used, still remains on the circuit switch network (e.g., 2G or 3G) by maintaining either parallel connection and registration on these network or by switching to them whenever a voice call is initiated or terminated. The second one, which is when the voice call stands over LTE, the voice service is named VoLTE or VoIMS when the IP multi‐media system (IMS) service function is included. Video in LTE is one of the most importanr services. The demand for video content continues to grow among data services. Web video traffic growth has accelerated, as the number of internet‐ enabled devices has increased and more people depend on the mobile internet. Recently, a group of key operators, infrastructure, and device vendors announced a joint effort to facilitate the evolution of mobile communication toward RCS (rich communication suite). The core feature set of RCS includes the following services: enhanced phonebook, with service capabilities and presence enhanced contacts information; enhanced messaging, which
11
12
LTE Optimization Engineering Handbook
enables a large variety of messaging options including chat and messaging history, and enriched call, which enables multimedia content sharing during a voice call. It is believed that RCS is a promising evolution in LTE, many operators have announced to support the RCS. 1.2.1 Circuit‐Switched Fallback
The basic principle of circuit‐switched fallback (CSFB) is that once originating or receiving a CS voice call by the UE connected over LTE, it will move to either GSM or UMTS network (fallback) where the call proceeds. One major requirement for the realization of CSFB is the overlay of LTE with GSM, UMTS, or both. It is the quickest implementation both at terminal and at network sides and is mandatory for international roaming scenarios. With CSFB, UE will attach to the network through LTE, MME will ask MSC to update UE location in its database, when the UE is operating in LTE (data connection) mode and when a call comes in, the LTE network pages the device. The device responds with a special service request message to the network, and the network signals the device to move to 2/3G to accept the incoming call. Similarly for outgoing calls, the same special service request is used to move the device to 2/3G to place the outgoing call. CSFB for operator means very little investment since only few modifications are required in the network, additional interface (SGs) between MME and MSC is required shown in Figure 1.9. With basic CSFB implementation, the additional delay to set up the voice call is less than 1.3s to 3G or about 2.8s to 2G, which is acceptable from an end‐user perspective. This delay is significantly reduced with the activation of PS handovers when falling back to 3G and of RRC release with 3GPP Rel 9 redirections to 2G/3G. The CSFB option offers complete services and feature transparency by enabling mobile service providers to leverage their existing GSM/ UMTS network for the delivery of CS services, including prepaid and postpaid billing. SGs interface is used to carry signaling to move the access network carrying the voice traffic from 2/3G to the LTE and from LTE to 2/3G. This interface maintains a connection between the MSC/VLR and the MME and its main role is to handle signaling and voice by SGsAP application. Gn‐C interface is the interface connecting the MME to the SGSN in the pre‐Rel 8, it is replaced by the S3 for Rel 8 or later. This interface is required when a CSFB call is established to initial the signaling with SGSN. In case CSFB with PS handover the data established over the LTE will be carried over 2/3G network, the interface Gn‐C or S3 is used to establish the signaling sessions with the SGSN to forward pending data over the LTE toward the 2/3G packet core. To forward the data from the PGW, an additional interface named Gn‐U is required between the SGSN and the PGW in pre‐Rel 8 and the S4 interface between the SGSN and the SGW in Rel 8 or later. Uu
UTRAN
Iu-ps
SGSN Gs
Gb
UE
Um
Iu-cs
GERAN
MSC/ VLR
A Gn SGs S1-MME
MME S11
LTE Uu
E-UTRAN
S1-U
Figure 1.9 Standard architecture for CSFB.
SGW
S5/S8
PGW
LTE Basement
Data
LTE (eNode B)
Voice
2G/3G Base Station
Figure 1.10 Dual radio handsets.
CSFB is a single radio solution of handset, in order to make or receive calls, the UE must change its radio access technology from LTE to a 2G/3G technology, and uses network signaling to determine when to switch from the PS network to the CS network. The shortcoming is that someone on a voice call will not be able to use the LTE network for browsing or chatting, and so on. Except CSFB, dual‐radio handsets (SVLTE) shown in Figure 1.10 support simultaneous voice and data— voice provided through legacy 2G or 3G network and data services provided by LTE. Dual‐radio solutions use two always‐on radios (and supporting chipsets), one for packet‐switched LTE data and one for circuit‐switched telephony, and as a data fallback where LTE is not available. The dual radio has the benefit in which simultaneous CS voice and LTE data is available; the drawback is the complexity from the device point of view, since more radio components are required increasing the cost, size, and power consumption. Dual‐radio solutions also force the need for double subscriber registration leading to split legacy and LTE records in the subscriber data managers. As a matter of fact, lack of dual‐radio eco‐system for 3GPP markets and the top six main chipset vendors are addressing the 3GPP market with singe‐radio terminal and CSFB, while the top chipset vendors for 3GPP2 markets are supporting dual‐radio solution for the 3GPP2 market. The above considerations have lead to a clear split in the market for early LTE support of voice services with mobile networks based on 3GPP technologies adopting CSFB, while 3GPP2 markets have adopted a dual‐radio solution for early LTE deployments. CSFB addresses the requirements of the first phase of the evolution of mobile voice services, which commercially launched in several regions around the world in 2011. CSFB has become the predominant global solution for voice and SMS inter‐operability in early LTE handsets, primarily due to inherent cost, size, and battery life advantages of single‐radio solutions on the device side. CSFB is the solution to the reality of mixed networks today and throughout the transition to ubiquitous all‐LTE networks in the future phases of LTE voice evolution. 1.2.2 Voice over LTE
After CSFB, LTE voice evolution introduces native VoIP on LTE (VoLTE) along with enhanced IP multimedia services such as video telephony, HD voice and rich communication suite (RCS) additions like instant messaging, video share, and enhanced/shared phonebooks. The voiceover LTE solution (VoLTE) is defined in the GSMA1 Permanent Reference Document (PRD) IR.92,2 based on the adopted one‐voice profile (v 1.1.0) from the One Voice Industry Initiative. Video‐related additions are described in GSMA IR.94. 1 At the 2010 GSMA mobile world congress, GSMA announced that they were supporting the one voice solution to provide voice over LTE. After that, industry aligned 3GPP based e2e solution for GSM equivalent voice services over LTE. 2 The VoLTE IR.92 is from October 2010 put in maintenance mode and only corrections of issues that may cause frequent and serious misoperation will be introduced.
13
14
LTE Optimization Engineering Handbook
VoLTE specifies the minimum requirements to be fulfilled by network operators and terminal vendors in order to provide a high‐quality and interoperable voice over LTE service. The VoLTE solution is scalable and rapidly deployed, offering rich multimedia and voice services, seamless voice continuity across access networks, and the re‐use of existing network investments including business and operational support assets. In terms of the operators, the deployment of VoLTE means that it is opened to the mobile wideband speech path of evolution. Also, VoLTE can offer a competitive advantage by providing a superior voice service quality with HD voice and video, shortening setup times for the calls and guaranteeing bit rate, and offering simultaneous LTE data together with the voice call. Finally, a richer end‐user experience; to be able to provide end users the benefit of real‐time communications can be another VoLTE attraction. Better multimedia, video‐conferencing, or video chat while still maintaining a voice call, are all possible revenue opportunities of VoLTE. Introducing VoLTE on a standard‐based IMS provides the service provider with a true converged network where services are available regardless of the access type network. Blending services with an IMS service architecture enables an operator to cost‐effectively build integrated service bundles. VoLTE can evolve voice services into rich multimedia offerings, including HD voice, video calling, and other multimedia services (i.e., start a voice session, add and drop media such as video, and add callers, presence) available anywhere on any device, combining mobility with service continuity. VoLTE is an advancement from today’s voice and video telephony to full‐fledged multimedia communication to utilize the full potential of LTE and to improve customer experience. The IP‐based call is always anchored in IMS core network to carry and establish a voice call over an LTE network. Now, in both 3GPP and 3GPP2 markets, there is a clear consensus to adopt the IMS‐based VoLTE solution for the LTE deployments. Two transport modes are also used on the network and determines the quality of the voice call over an IP network. The VoIP’s best effort, mainly over the internet and based on some widely deployed applications, such as Skype, Google talk, and MSN, uses this mode with no guarantee of the quality. Other technology such as LTE propose to carry the VoIP with the guarantee of the quality of this call over the end‐to‐end network. For VoLTE, the installed solution aims at being partially compliant with GSMA PRD IR.92.3 One voice was an effort to use already‐defined standards to specify a mandatory set of functionality for devices, the LTE access network, the evolved packet core network, and the IP multimedia subsystem in order to define a voice and SMS over LTE solution using an IMS architecture. Some VoLTE handsets are already commercial including the features such as emergency call, location based services, and so on. In case VoLTE through IMS is the mode used, two connections are required with the LTE network—Rx interface between the P‐CSCF and the PCRF and the Gx interface between the PCRF and the PGW for dynamic PCC rules. The Gm interface is a virtual interface established between the SIP application on the end user and the P‐CSCF function of the IMS network where it is connected (Figure 1.11). Along with VoLTE introduction, 3GPP also standardized Single Radio Voice Call Continuity (SRVCC) in Rel 8 specifications to provide seamless continuity when an UE handovers from LTE coverage (E‐UTRAN) to UMTS/GSM coverage (UTRAN/GERAN). With SRVCC, which is depicted in Figure 1.12, the calls are anchored in IMS network while UE is capable of transmitting/receiving on only one of those access networks at a given time. SRVCC protocol evolution have different types according to the function. There are bSRVCC (before alerting
3 Complementary scenarios are also beign defined in the VoLTE profile extension (IR.93) to cope with the cases where LTE coverage needs to be complemented with existing WCDMA/GSM CS coverage.
LTE Basement A
GERAN Gb
S4 Gn
SGSN IuPS IuCS UTRAN
S3/Gn
HSS
ISUP
Sv
S6d
Gm
MSC
IMS Gi
SGW
Gm
Rx Sv
S6a EUTRAN
S5
MME
S1-MME
S11
PGW
Gx SGi
PCRF
S1-U
Figure 1.11 Standard architecture for VoLTE.
SRVCC
VoLTE
CS Core
SRVCC
SRVCC
CS
Legacy RAN
SRVCC
IMS SRVCC
SRVCC
Evolved Packet Core
LTE RAN
SRVCC
CSFB Semi-Persistent Scheduling TTI Bundling Common IMS SRVCC RCS
CSFB Fast Return after CSFB Emergecy call on VoLTE Emergecy call w/SRVCC
Rel-8
Rel-9
eSRVCC aSRVCC
Rel-10
SRVCC function
rSRVCC vSRVCC
Rel-11
Figure 1.12 SRVCC and evolution.
SRVCC), aSRVCC (alerting phase SRVCC), vSRVCC (video SRVCC), and vSRVCC (reverse SRVCC, HO 3G/2G → LTE). Up to now, VoLTE launches are taking place in Korea, the United States (AT&T, T‐Mobile, Verizon), Russia (MTS), and Asia (NTT Docomo, SingTel, M1, Starhub, HKT). T‐Mobile U.S. launched VoLTE in Seattle on May 22, 2014. AT&T launched in three markets on May 23, 2014 with “crystal clear conversations.” SingTel launched on May 31, 2014 in Singapore using 4G Clear Voice. In 2015 and 2016, more and more countries launched VoLTE, like China, Canada, France, and Denmark.
15
16
LTE Optimization Engineering Handbook
1.2.3 IMS Centralized Services
In a CS network telephony, services are provided by the MSC (based on the subscription data in the HLR). In IMS telephony, services are provided by the telephony application server. Multiple service engines introduce synchronization problems and differences in user experience. IMS centralized services avoids these problems by assuming that only one service engine will be used. IMS plays an essential role in IMS centralized services. The UE performs SIP (session initiation protocol) registration with the IMS network. IMS‐AKA (IMS‐authentication and key agreement) procedures are followed for authentication. Integrity protection, whereby integrity of SIP signaling messages is ensured, is mandatory. The use of ISIM (IP multimedia services identity module) or USIM (UMTS subscriber identity module) is required during the IMS authentication. SIP signaling messages are ASCII text messages and could thus be quite large. Hence, signaling compression is mandatory to reduce the bandwidth requirements, especially for over‐the‐air transmission. IMS centralized services (ICS) enable the use of the IMS telephony service engine for originating and terminating services regardless if a UE is connected via a LTE PS access network or connected via a GSM/WCDMA CS access network. For terminating calls, ICS determines the access network currently in use by a UE to deliver the call via the correct access network. ICS requires an IMS service centralization and continuity application server. 1.2.4 Over the Top Solutions
At the same time, there are already a number of applications providing over the top (OTT) voice service on smartphones, which can be used over Wi‐Fi connection but also over cellular networks. OTT application is completely transparent to network and also out of operators’ control. OTT services are those provided without special consideration at the network level (i.e., no special treatment with respect to QoS). Examples of these types of services are YouTube, Vimeo, and DailyMotion, which are very popular today. Skype and GoogleTalk have nearly a billion registered users worldwide. Apple has sold countless iPhones and iPads, many of which are capable of FaceTime video calling. These services are provided directly by content providers (and usually over content delivery networks), generally without any arrangement with the network providers sitting between the content and its consumers. Nowadays, some OTT solutions, such as Skype and FaceTime, often come preinstalled on smartphones, and as these devices become much more widespread, the adoption of OTT solutions for video‐calling services will also increase. LTE supports high bandwidth, low latency, always online, all IP and other characteristics, it is convenient for the development of OTT. OTT application providers have delivered very popular voice, video, messaging, and location services that are shifting consumers’ attention and usage. In addition, while OTT players currently generate revenue using the operator’s network for service delivery, the operator itself doesn’t gain any associated increase in revenues. The Figure 1.13 shows MoS performance based on data from the South Korean market’s most OTT‐friendly operator. In the future, the proportion of OTT voice may be more and more high, especially in the area of long distance calls, as these solutions are familiar to subscribers and have driven user expectations. However, a fully satisfactory user experience cannot be provided by OTT solutions, as there are no QoS measures in place, no handover mechanism to the circuit‐switched network, no widespread interoperability of services between different OTT services and devices, and no guaranteed emergency support or security measures. Consequently, the adoption of OTT clients is directly dependent on mobile broadband coverage and the willingness of subscribers to use a service that lacks quality, security, and flexibility. For example, with VoLTE, using
LTE Basement Range : – 105.000000 AND RSRP < = –95.000000 > – 95.000000 AND RSRP < = –80.000000 > – 80.000000 AND RSRP < = 0.000000
Figure 2.8 Optimization method by advanced geolocation algorithms.
both indoors and outdoors, three dimensions in horizontal and vertical directions by using MRs from the user equipment, wherever they are, even within buildings. 2.2.5.1 Timing Advance
Different UEs in the cell may have different position, and therefore, different propagation delay, thus this may affect uplink synchronization. eNB’s timing will be phase‐synced to GPS within 100ns to support timing advance (TA) to achieve the tight phase‐sync. TA characteristics can be assumed that uplink arrives 0.3us too late compared to downlink. This error needs to be minimized with a correction of the UL timing. Due to the minimum step length of 0.52us
33
34
LTE Optimization Engineering Handbook
Table 2.2 Positioning methods. Method
Handset impact Accuracy
Availability
Cell ID
No impact
From 200 m to 5 Km
200 m in urban to 5 Km in rural area
CID/TA
No impact
From 100 m to >1Km
Mainly use CGI/TA (GSM and LTE) or RTT (3G) and UERxTxDiff (LTE)
E‐CGI (GSM)
No impact
150‐400 m
Any environment
AECID
No impact
From 50 m in urban to 500 m in rural area
Any environment, but it requires high penetration of A‐GPS in order to reduce maintenance cost
A‐GPS
HW & SW
10‐30 meter
Doesn’t perform well indoor.
OTDOA
SW
50‐150 meter
Only LTE, did not work well in 2/3G
U‐TDOA
No Impact
50‐100 meter
Cost is high for deploying LMUs.
3D RSS fingerprinting No Impact
10‐30 meter
Required 3D maps & network configuration
Hybrid Positioning
between accuracies of the involved methods
No Impact
0,52us
The mobile is somewhere in the yellow arc. 0,26us 0,26us
A
DL
n*TA
Ideal position
pos(X,Y,Z) B
UL without TA UL useing TA
Speed of light: 300*106 m/s s = v*t; 300*106*0.52*10–6 = 156 m; 300*106 *0.26*10–6 = 78 m Two types of TA: TA (type 1) = (eNB Rx – Tx) + (UE Rx – Tx) Continuous TA TA (type 2) = eNB Rx – Tx PRACH TA Figure 2.9 Timing advance.
(TA granularity is 16* Ts = 16/(15000*2048) = 0.52 µs) the UE is told to send the next frame 0.52us earlier. After the correction is applied, the frames will arrive 0.22us to early. This is treated as good since the error has decreased. The least measurable error is 0.26us (Figure 2.9). UE distance and UE velocity relative to eNB can be calculated by TA. UE distance can be calculated based on the median TA value in a series of TA reports. UE velocity relative to eNB can be calculated the difference between the first and last TA values in a series of TA reports, and convert the TA difference to distance difference relative to eNB and then UE velocity is based on the duration for the reports.
LTE Optimization and Principle and Method
In urban areas, UE is usually connected to eNB under a no‐line‐of‐site condition (reflections on building and terrain). This effect results in an over‐estimation of the distance between UE and eNB impacting on the positioning. 2.2.5.2 Location Accuracy Evaluation
Location determination algorithms are the backbone of operator‐offered location‐based services. There are legacy techniques listed below to determine an estimate of the location of a wireless user. Other techniques are possible, but are typically less applicable and require much more complicated analysis and computation. Cell‐ID method The cell‐ID method provides the identification of the serving sector for an active wireless subscriber as an estimate of their location. The cell‐ID can be obtained from signaling information that is used to set up or maintain a wireless call. This technique has limited accuracy since it can only identify the location of a user to the area covered by a cell, which can be quite large. Normally cell‐ID and other UE/EUTRAN measurements can be used to calculate UE’s position, it can be got other informations include the coordinates of the radio antenna, number of sectors in an eNB, beam orientations, and transmitting powers. The cell‐ID method can provide estimated accuracy is 100 m 67% and 300 m 95%. TA+AoA method UE distance can be calculated based on the median TA value, AoA (angle‐of‐arrival, the angle (usually azimuth) from which a signal arrives relative to a reference angle (geographical north) of an antenna array) can be used to determine the azimuth of the UE. Assisted GPS (A‐GPS) Assisted GPS makes use of the GPS capabilities of a UE and the availability of a GPS satellite network to provide very accurate location information for an individual user. Assisted GPS is a refinement of GPS where the network stores information that can be provided to the UE to help the UE speed up the process of acquiring the required number of GPS satellites to allow for GPS geolocation. However, not all UEs are equipped with GPS capabilities today. Furthermore, GPS features tend to use a significant amount of power, causing wireless subscribers to sometimes turn off the GPS functions on their GPS‐enabled phones. Also, GPS requires line‐of‐sight access to multiple GPS satellites, which means that it does not work well indoors. An assisted GPS method can provide an estimated accuracy of 50 m 67% and 100 m 95%. OTDOA OTDOA positioning method relies on the UE measuring the Reference Signal Time Difference (RSTD) on Positioning Reference Signals (PRS) sent from the reference cell and a number of neighboring cells as it requires that the UE can detect and measure on at least two neighbors; the two neighbors must also have “decent” geometry relative to the UE (e.g., not in line as seen from the UE). The OTDOA method can provide estimated accuracy is 100 m 67% and 300 m 95%. Triangulation Triangulation is a mathematical technique where the location of a point in a three‐dimensional environment can be determined by knowing the distance between that point and at least three other fixed points in the network. The more accurate the information about these distances,
35
36
LTE Optimization Engineering Handbook
Figure 2.10 Choose signals from three sites and estimate the distance from UE to sites. –9
9d
Bm
1d
Bm
–8
–85 dBm –98
dBm
–96
dBm
Site1 Site2
Site3
the more accurately the calculated position of the point of interest can be obtained. Location estimates of lesser accuracy can be obtained by knowing the distance to two or even one other point in the network (Figure 2.10). In wireless networks, obtaining the distances from the point of the wireless user to the other fixed points is generally accomplished by measuring the time it takes to transmit information from the point of interest to the other fixed points and using that time to estimate the distance based on the propagation speed of the information. Adjustments can be made to the estimate of the distance by the known delays in the propagation. In urban environments, the triangulation method can provide estimated accuracies of 150 meters most of the time. Several tests were made by combined TA+AoA and triangulation method and in the field at different locations in a live network. Some analysis of the test results are provided in Figure 2.11. It was begun with individual tests to see if there is a systematic error in the coordinate detection produced by the tool. Then it will have an assessment of all the tests to see what accuracy can be achieved with an increase in samples. The blue dots are the measured coordinates relative to the mobile location at (0, 0). It is seen that the over all offset are unevenly scattered around (0, 0). On average, the offset in X is 14.9 meters and in Y is −22.5 meters. The standard deviation is 94.4 meters (Figure 2.11). 2.2.5.3 Location Support
The control plane of location support provides positioning information of a user in the EPS system. The positioning methods supported by this feature are assisted GPS (A‐GPS), observed time difference of arrival (OTDOA), and enhanced cell ID, and they generate UE positioning data used to position a LTE UE with high accuracy in the network. In A‐GPS, assistance data are retrieved by the network from GPS signal being sent by the GPS satellite constellation.
LTE Optimization and Principle and Method
Figure 2.11 Scatter plot of the coordinate offset in meters from all the tests.
GHData relative to FieldData
150 100
Distance in meters
50 0 –50 –100 –150 –200 –250 –100
–50
0
50
100
150
200
250
300
Distance in meters
The UE performs the GPS measurements with the help of assistance data provided by the network. This position can then be used by emergency services or commercial location services. Positioning data between positioning nodes and the UE is transferred on the control/ user plane. Positioning tasks to be performed by eNB include: ●● ●● ●● ●● ●● ●● ●● ●● ●●
Configuration of positioning sub‐frames (for OTDOA) Broadcast of position reference symbols Inform UE about positioning sub‐frames, by broadcasting or by dedicated signaling Measure TA (for ECID) Request UE measurements (for ECID) Measure GNSS timing of cell frames Forward measurements to eSMLC Terminate LPPa protocol Forward LPP containers (assistance data and UE measurements)
Following RAN events/measurements are located on the map as a function of actual statistically distributed usage (user id unknown) (Figure 2.12). 2.2.5.4 3D Geolocation
Drive test and 2D geolocation provide a network view at street level. Traditional optimization doesn’t consider what is happening in buildings. So, how do we identify problems in the buildings, low coverage areas, traffic hot spots, and so on? Imagine you have a multi‐use high‐rise building with 20 floors. Is the majority of the mobile traffic coming from the restaurant on the first floor, or the office on the tenth floor? 3D geolocation can answer this question and help plan the most cost‐efficient and effective solution. As most of the traffic is indoor, network performance in‐building is very important, because accurate indoor positioning of UEs can have a evaluation of indoor cell loading and efficiency by analyzing how much traffic should belong to an indoor cell but is absorbed by surrounding outdoor cells. A 3D method geolocates network events not just on a horizontal two dimensions (2D) but also at different floors in high‐rise buildings, that is, on a vertical z‐axis (3D). A 3D vision of real traffic with outdoor versus indoor traffic and height with massive geolocation brings
37
38
LTE Optimization Engineering Handbook
Value value < –110 –110< = value > choise S‐TMSI >>> S‐TMSI >> choise IMSI >>> IMSI > CN‐Domain
Provided by the S1AP: paging message.
Paging‐v890‐IEs > lateNonCriticalExtension > Paging‐v‐920‐IEs >> cmas‐Indication‐r9 >>nonCriticalExtension
Omitted
88
LTE Optimization Engineering Handbook UE
eNB
EPC
UE
eNB
NAS:Paging (S_TMSI)
EPC NAS:Paging (IMSI)
NAS:Paging (S_TMSI)
NAS:Paging (IMSI)
Same as the procedure of service request initiated by UE
Same as the procedure of attach request initiated by UE
Figure 2.58 Paging procedure. Table 2.28 DRX/DTX parameter. Parameter
Range
Default
Default paging cycle
32rf (0), 64rf (1), 128rf (2), 256rf (3)
64rf (1)
Inactivity timer
10…65535 s, step 1 s
10 s
Paing nB
1/32T, 1/16T, 1/8T,1/4T,1/2T, 1T, 2T, 4T; T is the defaultPagingCycle.
1T (2)
aging information from the network. Paging DRX cycle defines time between two POs of the p same UE. The RRC layer tries to send the paging message during a period specified by the parameter pagingDiscardTimer, after which the paging message is discarded. It is recommended that the pagingDiscardTimer should be set equal to or smaller than T3413. To guarantee at least one retransmission attempt by the RRC layer, the pagingDiscardTimer must be set to a larger value than the defaultPagingCycle. If the MME does not receive the service request within T3413 seconds, it resends the S1‐AP paging message according to the paging profile. The DRX/ DTX parameters are present in Table 2.28. The default paging cycle defines the cell specific paging DRX cycle duration (periodicity of the paging). It also determines the maximum paging DRX duration applicable in the cell. Value rf32 corresponds to 32 radio frames, rf64 corresponds to 64 radio frames, and so on. One radio frame is 10ms. Current recommended value is 64rf, since it improves the paging time. Higher parameter values can save battery in idle mode as listening to paging in less frequently but it also means that call setup is getting longer due to longer average paging time. pagingNb defines the number of possible paging occasions per radio frame, that is, the density of paging occasions. It is used to calculate the number and position of paging occasions (PO) and paging frames (PF). By increasing pagingNb the number of paging occasions per second is increased. 3GPP defined there are maximum 16 users per paging occasion. There is one paging record per UE in each paging occasion so it is possible to page a maximum of 16 UEs. Therefore, increasing pagingNb means it is possible to page more UEs, and increasing the paging capacity (Figure 2.59). SFN0 nB = ½T PO
nB = T nB = 2T
SFN1
SFN2
SFN3
Figure 2.59 pagingNb.
LTE Optimization and Principle and Method Possibility to page this terminal
UE receiver circuitry
UE receiver circuitry
switched off
switched off
Subframe DRX cycle
Paging DRX Cycle Paging DRX Cycle Paging DRX Cycle Paging DRX Cycle
PF
PF
PF
PF
PF
PF
#0
#4
#5
#9
PO
PO
PO
PO
PDCCH
PDSCH
T_Frequency P-RNTI
T_Format
Paging message
Figure 2.60 Paging frame and occasion.
The positions that paging messages are transmitted on the Uu interface are fixed, which are indicated by the paging frames (PFs) and paging occasion (PO) subframes. One PF is one radio frame, which may contains one or multiple POs. All attached UEs are distributed on all paging frames within one paging DRX cycle (based on IMSI). One PO is a subframe where the P‐RNTI is contained. The PO is transmitted over the PDCCH, the P‐RNTI value is fixed, that is, FFFE. UEs read paging messages over PDSCH according to the P‐RNTI. To receive paging messages from E‐UTRAN, UEs in idle mode monitor the PDCCH channel for P‐RNTI value used to indicate paging. If the UE detects a group identity used for paging (the P‐RNTI) when it wakes up, it will process the corresponding downlink paging message transmitted on the PCH (Figure 2.60). LTE frames numbering has two components, one at frame level, that is, system frame number (SFN) and second one at subframe level, that is, subframe number. So the UE has to know both SFN and subframe number to locate exact position of its page. The SFN of a paging frame (PF) is derived from the following formula:
PF
SFN modT
T div N x UE _ IDmod N
The subframe number i_s of a PO is derived from the following formula:
i_s
UE _ ID/N mod Ns
T is DRX cycle of the UE. UE can get the T from two different sources, one from the system information (SIB2) and the one from upper layer (NAS). If upper layer send the value, the UE
89
90
LTE Optimization Engineering Handbook
Table 2.29 Example of paging offset and sub‐frame, TDD (all UL/DL configurations). Ns
PO when i_s=0
PO when i_s=1
PO when i_s=2
PO when i_s=3
1
9
N/A
N/A
N/A
2
4
9
N/A
N/A
4
0
4
5
9
compares the value signaled from the NAS and the value of DefaultPagingCycle and uses the smaller value between them, otherwise UE has to use the value from SIB2. N = min(T, nB), which means the smaller one among T and nB. nB can be any one of 4T, 2T, T/2, T/4, T/8, T/16, T/32, which comes from SIB2 (IE nB). UE_ID is IMSI mod 1024, calculated by MME and UE. If the paging is triggered by the MME, the UE_ID value is the UE identity index value contained in the paging message on the S1 interface without having to signal the IMSI across the S1 interface. It is a 10 bit string value. Thus, UE are divided in “N” groups according to their UE identity, all UE with equal values for “UE_ID mod N” share the same paging frames. Network send paging with S‐TMSI, but if something (e.g., network failure) happens during registration and it fails to allocate TMSI to the UE, network would send paging with IMSI. If UE get the paging with IMSI, it should tear down all existing bearers and delete TAI, TAI List, KSIASMI and get into EMM‐DEREGISTERED status. And then redo “attach request.” Ns = max(1, nB/T), which means that Ns is the larger value between 1 and NB/T Table 2.29 gives an example of paging offset and sub‐frame under TDD (all UL/DL configurations) system. Example: Example: T = 64 nB = 2T = 128 UE_ID = IMSI mod(1024)
N = min (T,nB) = 64 Ns = max (1,nB/T) = 2 UE_ID = IMSI mod 1024 = e.g 0 PF : SFN mod T= (T div N)*(UE_ID mod N) = 0, 64, 128…
Although 3GPP allows up to four paging occasions per radio frame. Assume only supports up to 1 paging occasion per radio frame (pagingNb = oneT), that translates into 100 paging occasions per second. Since each paging message may contain up to 16 paging records the maximum air interface capacity is 1600 paging records per second per cell, which is more than what eNB can provide in the control plane, that is, it has been observed a CPU load of roughly 4% for 30 S1 pagings per second. As a consequence, 500 S1 paging messages per second would occupy 2/3 of CPU processing power for paging alone. Based on the above, the default pagingNb= quarterT (equivalent to 400 possible paging messages per second) is the maximum recommended value. UE may also be paged by the network when there is data addressed to that particular UE. UE returns to EMM_ACTIVE/RRC_connected mode as soon as packet arrival is detected. The delay depends on the paging DRX cycle, time to acquire UL synchronization, and time to set up the RRC connection with the eNB. MME Initiated paging fow is shown in Figure 2.61.
LTE Optimization and Principle and Method UE
eNB
MME
SGW
PGW
k data
Downlin Page Page Response Time
Page RACH Preamble RACH Response RRC Connection Request RRCConnection Setup S1-AP Initia l UE Messa ge Attach Requ est eNB UE S1 AP ID IE
Total Delay
RRCConnectionSetup Complete Attach Request
tup req
ntext se
Initial co
Security Mode Command Security Mode Complete RRCConn Reconfiguration Req RRC conn Reconfiguration Complete
Initial Cont
ext Setup
UL DATA
Response
DL DATA
Update Be arer Request arer Modify Be Response
Figure 2.61 MME initiated paging flow
Here are three paging examples as following. Example 1 > - Paging with s‐TMSI PCCH-Message ::= SEQUENCE +-message ::= CHOICE [c1] +-c1 ::= CHOICE [paging] +-paging ::= SEQUENCE [1000] +-pagingRecordList ::= SEQUENCE OF SIZE(1..maxPageRec[16]) [1] OPTIONAL:Exist | +-PagingRecord ::= SEQUENCE | +-ue-Identity ::= CHOICE [s-TMSI] | | +-s-TMSI ::= SEQUENCE | | +-mmec ::= BIT STRING SIZE(8) [00000001] | | +-m-TMSI ::= BIT STRING SIZE(32) [000000000000000000000 | 00000000001] | +-cn-Domain ::= ENUMERATED [ps] +-systemInfoModification ::= ENUMERATED OPTIONAL:Omit +-etws-Indication ::= ENUMERATED OPTIONAL:Omit +-nonCriticalExtension ::= SEQUENCE OPTIONAL:Omit Example 2 > - Paging with IMSI PCCH-Message ::= SEQUENCE +-message ::= CHOICE [c1] +-c1 ::= CHOICE [paging] +-paging ::= SEQUENCE [1000]
91
92
LTE Optimization Engineering Handbook
+-pagingRecordList ::= SEQUENCE OF SIZE(1..maxPageRec[16]) [1] OPTIONAL:Exist | +-PagingRecord ::= SEQUENCE | +-ue-Identity ::= CHOICE [imsi] | | +-imsi ::= SEQUENCE OF SIZE(6..21) [15] | | +-IMSI-Digit ::= INTEGER (0..9) [0] | | +-IMSI-Digit ::= INTEGER (0..9) [0] ...... | | +-IMSI-Digit ::= INTEGER (0..9) [9] | +-cn-Domain ::= ENUMERATED [ps] +-systemInfoModification ::= ENUMERATED OPTIONAL:Omit +-etws-Indication ::= ENUMERATED OPTIONAL:Omit +-nonCriticalExtension ::= SEQUENCE OPTIONAL:Omit Note: IMSI is used as paging identity when reattach from UE is required. It indicates that error occurred. Example 3 > - Paging for system Info Modification RRC_LTE:PCCH-Message PCCH-Message ::= SEQUENCE +-message ::= CHOICE [c1] +-c1 ::= CHOICE [paging] +-paging ::= SEQUENCE [0100] +-pagingRecordList ::= SEQUENCE OF OPTIONAL:Omit +-systemInfoModification ::= ENUMERATED [true] OPTIONAL:Exist +-etws-Indication ::= ENUMERATED OPTIONAL:Omit +-nonCriticalExtension ::= SEQUENCE OPTIONAL:Omit 2.3.11.2 Paging Capacity
The higher amount of paging frames the more PDCCH and PDSCH resources may be used for paging. More aggregated paging records in one RRC paging message that is sent in one PO causes less PDSCH occupation comparing to higher number of RRC paging messages with less paging records inside. When average paging load is low, one can reduce value of pagingNb, thus more UEs under one PO will be aggregated and in average more page records per each RRC paging message is expected. What is more, less POs will decrease PDCCH load that less P‐RNTIs will be used to sent the same amount of paging records. On the other hand more paging records per one PO will cause higher blocking probability, it should be balanced. The the number of EPS paging attempts, received/discarded S1AP Paging messages by eNB can be got from PM counter listed in Table 2.30. Paging DRX cycle does not change the maximum paging capacity because the number of resources that can be used for paging is not changed. Paging DRX cycle reduces blocking probability at the cost of call setup time. Table 2.30 The number of EPS paging attempts, received/discarded S1AP Paging messages. EPS paging attempts
This counter counts the the number of paging attempts of initial and repeated.
PageS1Received
This counter counts the number of received S1AP Paging messages in the RBS.
PageS1Discarded
The number of S1AP Paging messages that are discarded and not routed to any cell
LTE Optimization and Principle and Method defaultPagingCycle = rf32
Paging Occasion (per Paging Frame)
Paging Frame
nB = 4T
=> 0
16
32
nB = 2T 0
16
32
nB = T
4
9
0
4
9
0
4
9
0
4
9
0
4
9
0
4
9
0
4
9
0
4
9
=> 0
16
32
nB = (½)T 0
16
32
0
16
32
nB = (¼)T nB = (1 8)T
0
=>
=> =>
0
16
32
=>
The value of nB determines the rate at which paging frames occur from the cell perspective, i.e. there is an impact upon paging capacity.
=>
nB = (1 16)T 0
16
32
nB = (1 32)T
=> 0
16
32
Figure 2.62 Paging frame and paging occasion.
PagingNB is used to calculate the number of POs within one paging DRX duration, which in turn is used to calculate the PO. PagingNb changes the maximum paging capacity because it increases number of resources that can be used for paging, it reduces blocking probability and increases maximum number of pagings. Paging frames are distributed on all radio frames according to pagingNb parameter value. PagingNb has impact on number of POs per considered time frame, higher value of pagingNb more POs per time period. Paging frame and paging occasion example are shown in Figure 2.62. If eNB KPI indicates more paging discards, it can be considered to increase the maximum number of paging records parameter or to allow more number of paging occasions, that is, increase nB (Table 2.31). Paging capacity has impact on call setup time, reachability of UE, and tracking area size. Excessive paging blocking value leads to delay or failure of the paging procedure. Here are four type of paging DRX cycle and pagingNb settings usually configured in a live network. It can be seen that higher value of pagingNb (½) will result to reduced number of UEs assigned to one PO (20UEs/PO or 10UEs/PO) and longer paging DRX cycle will result more POs within one cycle reducing number of UEs assigned to one PO (Figure 2.63). Assumption1: 320 UEs/cell, pagingNb = ½, 32/64 paging DRX cycle Assumption2: 320 UEs/cell, pagingNb = ¼, 32/64 paging DRX cycle Paging blocking probability is another factot that impacts on eNB paging capacity as it limits maximum number of pagings that can be handled by one eNB. Pagings arrival is poisson distribution Table 2.31 Paged UEs per second per cell. nB
T/32
T/16
paging occasions per radio frame
1/32
1/16
Ns
T/8
T/4
1/8
T/2
1/4
T
1/2
2T
1
4T
2
4
1
1
1
1
1
1
2
4
Number of paging record per radio frame
16
16
16
16
16
16
32
64
Max No. of paged UEs per second per cell
50
100
200
400
800
1600
3200
6400
93
94
LTE Optimization Engineering Handbook Paging frame (10 ms)
50 PO/s = max 800 pagings/s 32rf Paging DRX Cycle (16POs)
32rf Paging DRX Cycle
1 second
32rf Paging DRX Cycle
32rf Paging DRX Cycle
320/16 = 20UEs/PO
50 PO/s = max 800 pagings/s
320/32 = 10UEs/PO
1 second
64rf Paging DRX Cycle (32POs)
64rf Paging DRX Cycle
25 PO/s = max 400 pagings/s 32rf Paging DRX Cycle (16POs)
32rf Paging DRX Cycle
1 second
32rf Paging DRX Cycle
32rf Paging DRX Cycle
320/8 = 40UEs/PO
25 PO/s = max 400 pagings/s
320/16 = 20UEs/PO
1 second
64rf Paging DRX Cycle (32POs)
64rf Paging DRX Cycle
Figure 2.63 Four type of paging DRX cycle and pagingNb.
Table 2.32 PDSCH resources occupation for paging. N_PRB I_TBS
1
2
3
0
16
32
1
24
2
32
3 4
4
5
6
7
8
9
10
56
88
120
152
176
208
224
256
56
88
144
176
208
224
256
328
344
72
144
176
208
256
296
328
376
424
40
104
176
208
256
328
392
440
504
568
56
120
208
256
328
408
488
552
632
696
in a live network and blocking probability determines average number of records per one RRC paging message. For the calculation of consumption of PDSCH resources, assume 16 pagings records in one PO, the length of S‐TMSI is 32bits, the length of MME record is 8bits. For each PO, the maximum transmit payload are 16*40+1+1=642 bits, it will occupy 10 PRBs PDSCH resources shown in Table 2.32. It can be seen that when PDSCH resources is 10 PRBs and MCS level is 4, the TB size is 696 bits accordingly. Actually, in a live network, it is normally not to use higher MCS level for the paging channel. Now let’s take an example of how to calculate eNB’s paging capacity. It is simply assumed the traffic model (three cells/eNB) of a network is listed below (variety of applications and smartphone types cause that the real network behavior might be significantly different): ●● ●● ●● ●● ●● ●● ●●
Users (RRC connected + RRC idle) per eNB – 450 MTC/MOC for VoIP – 50%/50% MTC/MOC for PS/background – 30%/70% VoIP holding time – 90 sec; PS data mean holding time – 312 sec. PS data sessions per user per busy hour – 1; VoIP sessions per user per busy hour – 1. Background traffic per every user InactivityTimer: 10 sec. So, in this traffic model, eNB’s paging capacity is 4.3 pagings/s/eNB.
LTE Optimization and Principle and Method
2.3.11.3 Paging Message Size
The physical layer will add a 24 bit CRC to the transport block and then complete channel coding. The default channel coding value of 0.12 means that high quantities of redundancy are added before transmitting the paging message across the air‐interface. The value of 0.12 and QPSK are always used on the PDSCH when transferring a paging message that helps to make paging more reliable. Assuming a resource block includes 132 resource elements per resource block pair within a subframe. Table 2.33 illustrates the requirement for a large number of resource blocks when the paging load is very high. 2.3.11.4 Smart Paging
LTE allows the operator to configure a first page is to be distributed, to the area of a single eNB, TA or the whole TA list. The selected area in which the page is sent can be decided based on different criteria: ●●
●●
●●
UE‐related criteria, which is used to send a page to a single eNB or a TA for known stationary UEs such as electricity meters UE last reported location criteria, which is used to send a page to a single eNB or a TA when a the location of the UE is unknown Service‐related criteria by APN, QCI, and ARP. These criteria are used for time critical services to guarantee that the page is sent directly to all eNBs in TA list
In general, paging strategy is step‐by‐step paging in a live network. When the MME wants to page the UE, first paging the last visit of the eNB, if the UE can not be paged, then last eNB and its neighboring eNBs will be paged, if the UE still can not be paged, then eNBs in last TA will Table 2.33 PDSCH paging load. Number of paging records
1
Paging message size (bits)
56
Transport block size (bits)
56
# Bits after channel coding
667
# RE
334
# RBs
3
2
104
120
1200
600
5
3
144
144
1400
700
6
4
192
208
1934
967
8
5
232
256
2334
1167
9
6
280
280
2534
1267
10
7
320
328
2934
1467
12
8
384
392
3467
1734
14
9
408
488
4267
2134
17
10
456
488
4267
2134
17
11
496
552
4800
2400
19
12
544
552
4800
2400
19
13
584
600
5200
2600
20
14
632
632
5467
2734
21
15
672
696
6000
3000
23
16
720
776
6667
3334
26
95
96
LTE Optimization Engineering Handbook Page strategy
IMEI TAC
User level
IMSI
MSISDN
The basic paging stragegy is: Trigger 1
Trigger 2 Service level
APN
ARP
Voice
QCI
SMS
last eNB -> last eNB and its neighboring eNBs -> eNBs in last TA -> eNBs in last TA List (3GPP standard)
Figure 2.64 Paging strategy.
be paged, and finally eNBs in last TA List will be paged. Paging the last visit of the eNB, UE paging success ratio can reach to 70% to 90%, and reduce paging signaling in large quantities. At the same time, different levels or different characteristics of users can be set to different paging attributes. With different attributes of paging, it can be greatly reduced the paging signaling load (amount of paging S1 paging messages decrease 50%). Intelligent intelligent paging strategy can be set different attributes, according to different levels or different characteristics of users based on APN, QoS, IMSI, IMEI, and so on. The paging strategy can be further refined and optimized, for example, accurate paging scheme can only be applied for slow moving UEs, for high mobility users, paging over tracking area, and gradually expand the range of paging shall be applied, thus will greatly reduced the paging signaling load (Figure 2.64). 2.3.11.5 Priority Paging
In a system with mixed paging priorities, higher priority paging can preempt existing lower priority paging in the queue, the paging success rate will be higher for higher priority of the paging message compared to lower or no priority. Assuming there are non‐prioritized paging messages and prioritized paging messages per second, the prioritized paging messages shall be equally distributed between priority 1 and 8. All paging messages shall be sent equally distributed in time based on priority. S1AP paging message listed in Table 2.34, which shows a paging payload. 3GPP allows up to 16 S1 pages per modem‐to‐UE page message. In case when more than 16 pagings are considered Table 2.34 S1AP paging message IE and semantics description. IE/Group name
Source
Semantics description
Comments
UE identity index value
MME
IMSI mod 1024
–
UE paging identity
MME
S‐TMSI or IMSI*
–
Paging DRX
MME
Paging DRX cycle
CN domain
MME
CN Domain – PS or CS
When more than 16 pagings are sent in one PO, they are put in RRC paging message according to descending paging priority value
List of TAIs
At least one TAI shall be present; up to 16
> TAI list item >> TAI
MME
>>> PLMN identity
MME
>>> TAC
MME
Paging priority
MME
Used in control plane overload cases
LTE Optimization and Principle and Method
Token rate–r, short interval (10 ms) or continues
excess flow P1
excess flow Tokens?
P2
Accepted pagings
P3
Tokens?
Excess paging
Accepted pagings
Tokens?
prioLevel1 Excess paging
prioLevel2
Paging
Accepted pagings
prioLevel3
Excess paging
Figure 2.65 Priority paging.
to be sent in one PO, they are put in RRC paging message according to descending paging priority value. Figure 2.65 gives a description of Token based priority paging. 2.3.12 MIMO and Beamforming
From 3GPP roadmap of multi‐antenna point of view, the evolution of LTE releases in 3GPP brings significant new capability in the domain of multi‐antenna operation. Advanced multi‐ antenna solutions are key components to achieve the LTE network requirements for high peak data rates, extended coverage, and high capacity. Nowadays, market is driven by advanced antennas and more complex passive antennas. Beamforming and spatial multiplexing illustrated in Table 2.35 have been widely deployed in the live network, while diversity and beamforming aim to improve received signal power of a single information stream, spatial multiplexing aims to share the signal power between multiple parallel streams. Smart antenna is a multiple antenna elements system, which combined with signal processing to dynamically select or form the “optimum” beam pattern for each user. Smart antennas usually categorized as switched beam and adaptive array, and there are four types of smart antennas as shown in Figure 2.66: uniform linear array (a), circular array (b), two‐dimensional grid array (c) and three‐dimensional grid array (d). Now in the LTE live network, usually classical antenna and active antenna are deployed. A classical antenna consists of subelements, two antenna ports per column or smart antenna that has at least eight antenna port. Smart antenna can shape a beam by weighting of subelements so called beamforming (BF). The beam shape could be fixed vertically or adaptable horizontally Table 2.35 Beamforming and spatial multiplexing. 2D Beamforming
Digital beamforming with integrated radios and antennas
3D Beamforming
3D beamforming is integrated with radios and antennas, support user specific beamforming. Require phased‐array technology, including analoque, digital and hybrid beamforming technologies
MIMO and massive MIMO
More antenna elements required for capacity increase. However higher frequency => smaller antennas
97
LTE Optimization Engineering Handbook
∆x
y
(a)
x
∆Φ
∆z y x
∆z
∆y
y
∆x
(b)
x (c)
(d)
∆x
y
z x
Figure 2.66 Four types of smart antennas based on element arrangement.
one column
Subelement weights
Figure 2.67 Classical antenna.
ments
w1
subele
antenna port
PA
w4
PA
w5
w8
w1
PA PA
antenna ports
98
PA
w4
PA
w5
PA PA PA
w8
PA
Figure 2.68 Flexible active antennas.
to any direction and shape by changing amplitude and phase. An active antenna, composed of a multiple set of low power active transceivers modules, which are connected to eNB through CPRI. PAs are tightly integrated inside antenna, one PA per subelement. Active antenna offers beamforming more flexibility to tune tilt angles without mechanical actions, more flexibility to change the antenna vertical beam (Figure 2.67 and Figure 2.68).
LTE Optimization and Principle and Method
For beamforming, two precoding methods were evaluated using a full buffer traffic model in the 3GPP 3D UMi scenario: one is SVD precoding using the uplink long term spatial channel correlation matrix, the other is codebook based precoding using grid‐of‐beams, where the precoder matrix is chosen based on the uplink long term spatial channel correlation matrix. For spatial multiplexing, the eNB is able to transmit and receive multiple parallel information streams within the same spectrum which requires multiple antennas at both ends of the radio link and the maximum number of parallel streams is equal to the minimum of the number of transmit and receive antennas, therefore increasing the spectral efficiency. Most MIMO implementation nowadays is X‐pol (one is 45 degree, the other is ‐45 degree) MIMO and not spatial MIMO. The evolution of DL transmission modes on top of Rel 8 was the enhanced transmit diversity, beamforming and spatial multiplexing, Rel 9 adds dual layer beamforming. Rel 10 extends the dual layer mode of TM8 to TM9 with up to eight layers. Rel 11 adds TM10 with up to eight layers and provide support for DL beamforming on dedicated control channel and optimized DL CoMP operation, and in Rel‐10 UL SU‐MIMO with up to four‐layer, UL reference signal enhancements for improved UL MU‐MIMO (Rel 10) and UL CoMP (Rel 11) will be supported. 3GPP transmission modes (downlink) are present in Table 2.36. It is worth to mention that 3D beamforming is planning introduced in 3GPP Rel 13, it is not only a new beamforming feature but a whole new product and site solution, instead of 8 pipes with fairly high power per pipe, the idea is to use 64 pipes with low power per pipe. Either beamforming or spatial multiplexing, the goal of optimizing a MIMO system is to achieve the highest throughput and connectivity possible in a given environment by leveraging the multipath potential of the environment.
Table 2.36 3GPP transmission modes (downlink).
3GPP
Rel‐8
Transmission scheme tx mode of PDSCH
Antenna Port Feedback
Beam‐ Spatial Spatial forming multiplexing diversity
1
Single‐antenna port
CRS
CQI
No
No
No
2
Transmit diversity
CRS
CQI
No
No
Yes
3
Open‐loop spatial multiplexing
CRS
CQI, RI
No
1‐4 layer
Yes
4
Closed‐loop spatial multiplexing
CRS
CQI, PMI, RI Yes
1‐4 layer
No
5
Multi‐user MIMO
CRS
CQI, PMI, RI Yes
Yes
No
6
Closed‐loop Rank=1 precoding
CRS
CQI, PMI, RI Yes
No
No
7
Beamforming single‐ antenna port; port 5
DM RS
CQI
No
No
Rel‐9
8
Dual layer beamforming
DM RS
CQI, PMI, RI Yes
1‐2 layer
No
Rel‐10
9
Closed‐loop spatial multiplexing
DM RS
CQI, PMI, RI Yes
1‐8 layer
No
Rel‐11
10
TM9 with DL CoMP and E‐PDCCH
DM RS
CQI, PMI, RI Yes
1‐8 layer
No
Yes
99
100
LTE Optimization Engineering Handbook
2.3.12.1 Basic Multi‐Antenna Techniques
For LTE 10 transmission modes, only four MIMO technology programs are focused on in this part: beamforming, transmit diversity, SU‐MIMO, and MU‐MIMO. With at least eight antennas at the transmitter and only single antenna at the receiver (referred to as MISO) it is possible to obtain beamforming (BF). With this method the transmission signal is steered in a beneficial direction toward the UE, as the UE moves throughout the cell, the switched beam system detects the signal strength and continually switches the beam as necessary. This is accomplished by adjusting the phase and amplitude of the different antenna elements by multiplying the signal with complex weights. Each element of the array will be fed with the same signal and this enables “smart” antennas to modify their radiation pattern dynamically to alter the direction and shape even a beamforming antenna enables the entire eNB signal to be directed to the single user. It would be possible to dip the beam by causing element B to lag in phase behind element A, thus the line describing points where the radiation arrives in phase is no longer horizontal but instead dips toward point Z as shown in Figure 2.69. BF support to tune shapes horizontally and vertically according to typical UE positions and traffic needs, and even horizontal/vertical sectorization. Transmit diversity aims to increase the robustness of data transmission. When the same data is transmitted redundantly over more than one transmit antenna, this is called TX diversity. This increases the SNR. Space‐time codes are used to generate a redundant signal. Transmit diversity mode includes two transmitting antenna’s SFBC (space‐frequency block code) and four transmitting antenna’s SFBC+FSTD. In LTE, transmit diversity is used as a fallback option for some transmission modes, such as when spatial multiplexing cannot be used. Control channels, such as PBCH and PDCCH, are also transmitted using transmit diversity. MIMO includes single‐user mode SU‐MIMO and multi‐user mode MU‐MIMO. For SU‐ MIMO, data is divided into separate streams for one UE, which are then transmitted simultaneously over the same air interface resources under spatially uncorrelated channels (Figure 2.70). MU‐MIMO, the program will be the same time‐frequency resources through space division with sharp beams, assigned to different users under spatially correlated channels (Figure 2.71). Beamforming and MU‐MIMO prefer correlated antennas, precoding weights are chosen to maximize signal toward a given UE and are related to the instantaneous channel gains and phases of the signal paths through the MIMO channel. With correlated antenna ports as shown in Figure 2.71, the signal paths through the MIMO channel will all have the same instantaneous Y
Phase shif
A Feed
Phase shifters
θ
X
B
Z
Figure 2.69 Cell shaping and sectorization.
e
P
S-
y la
De
Figure 2.70 SU‐MIMO (left) and transmit diversity (right).
LTE Optimization and Principle and Method
Figure 2.71 Correlated (left) and decorrelated antennas (right).
v
v
W
W
ϕ t
t Ideal weights have same gain on all ports only phase changes
Ideal weights differ in gain and phase across ports
fading, so the ideal weights differ only in their phase slope. With decorrelated ports, the fade depth varies across the multiple signal paths through the MIMO channel. Optimal weights would ideally bias power away from the deeply faded paths and are therefore complex, with different gain and phase. It is worth to say beamforming is implicit in multi‐layer SU‐MIMO and MU‐MIMO in LTE, besides improving a UE’s signal strength or quality. 2.3.12.2 2D‐Beamforming
Beamforming (BF) algorithm can make wave to any direction and shape by changing amplitude and phase. BF principle is based on channel reciprocity, for TDD system, uplink and downlink has similar channel response and similar covariance matrices.6
HDL HUL
R DL R UL R HH H
Normally, HUL is estimated by eNB based on sounding reference signal (SRS transmits eriodically), which is used to determine the precoding, for example, grid of beams or p Eigenvalue‐based beamforming, evaluating the required periodicity and calculate DL weights. There are many possible ways of choosing the beamforming vector. Two different algorithms for estimating the precoding are usually used, the simpler grid of beams (GoB) algorithm, and the more complex eigenvalue based beamforming (EBB). The GoB‐based method calculates the DL beamforming weights based on the angle of arrival information. The EBB‐based algorithm calculates the DL beamforming weights based on the SVD decomposition of the spatial channel matrix. GoB based method is effective/efficient in small angle spread (AS) while considered less effective in large AS cases. EBB based methods can work in large AS cases. For FDD system, although uplink and downlink channels on separate frequencies experience independent fading, they should have similar spatial characteristics. Using UL spatial correlation matrix to derive the DL precoder matrix would result in a 5% to 10% loss in average user throughput and a 4% to 50% loss in cell‐edge user throughput compared to using closed‐loop feedback to derive the precoder matrix. In a TDD system with Na receiver and transmitter antennas, let the transmitted signal from a user with a single transmit antenna at subcarrier k of Nsc subcarriers be denoted X(k). Then, the received signal can be written as Y (k ) H (k ) X (k ) D(k ), where Y (k ) [Y1 (k ),Y2 (k ),,YN a (k )]T is the received signal at the eNB, H (k ) [ H1 (k ), H 2 (k ),, H N a (k )]T is the uplink spatial channel vector, and D(k ) [ D1 (k ), D2 (k ),, DN a (k )]T is the interference and noise at the eNB receiver antennas. In the receiver, an estimate H est (k ) of the channel vector H (k ) is calculated from the user’s reference signals. The precoding vector w(k ) [w1 (k ), w2 (k ),, w N a (k )]T is then
6 If the delay is within the channel coherence time, it will approximately hold.
101
102
LTE Optimization Engineering Handbook
CW1
u4 u3 u2 u1
Short term weight w1
Long term weight w1 1
5
w5
w2
w2 w6
6
2
CW1
w12
λ/2
w21 3
7
w4 4
8
w3 w7
Short term weight w11
w22 CW2
w8
u4 u3 u2 u1
Long term weight w1 1
5
w5 2
6
3
7
w4 4
8
w2 w6 w3 w7
λ/2
w8
Figure 2.72 Single‐stream (left) and dual‐stream (right) beamforming.
c hosen using this estimated channel vector H est k . It can be seen that the gain from the beamforming is related to the factor g H (k )* w(k ) . While 3GPP Rel 8 of the LTE specification defines transmission mode (TM) 7 for beamforming with one layer, Rel 9 defines transmission mode (TM) 8, to support dual‐stream beamforming, in which two different and independently coded data streams are separately transmitted from two logical antenna ports, combines beamforming with 2x2 MIMO spatial multiplexing capabilities. This feature implemented for TDD only, can result in improvements in coverage and capacity. Either of the two data streams is generated by four or eight antennas in beamforming transmission mode. The two data streams both form directional beams toward the target UE, which increases SINR. Dual‐stream beamforming incorporates both spatial multiplexing and beamforming during downlink transmission. This helps provide spatial multiplexing gains, diversity gains, and array gains (Figure 2.72). TM8 is a mode where data is transmitted over two spatial layers as two independent beamformed streams. It has the similar covariance matrices as one layer beamforming: RDL ≈ RUL(R = HHH) H Hermitian operator matrix transposition conjugate complex
In general, TM8 is expected to have higher throughput at cell edge, slightly better or similar throughput as TM3 in medium points, lower throughput in good points. Assuming the channel vectors at PRB i estimated from SRS from two polarization antenna elements groups are described by hi,1 and hi,2 respectively, then the instantaneous spatial channel covariance at current sub‐frame n is computed by averaging over two polarization antenna elements groups over all the used PRBs.
R inst n
i
h i ,1 h iH,1
h i ,2 h iH,2
The long‐term BF covariance matrices (weight) averaging over time by a recursive filter of first order, if long term BF covariance matrices is used, it can get better performance in channel models with low AS (dominant eigenvector).
Rave n
Rave n 1
1
Rinst n
where α is forgetting factor, which is inverse proportional to the settling time of filter. As stated above, the performance of beamforming strongly depends on the instantaneous channel information. If the channel condition of an UE varies frequently, then measurement results or reported results on the channel condition do not exactly reflect the current channel
LTE Optimization and Principle and Method
Table 2.37 TM8 Port7/8 beamforming gain imbalance. No
RSRP
SINR
CQI0
CQI1
MCS CW0
MCS CW1
1
–56.95
23.01
13.78
14.66
25.17
25.19
2 3 4 5 6 7 8
–81.07 –75.57 –99.23 –94.81 –103.51 –91.28 –95.44
14.46 14.70 9.01 10.88 10.00 5.44 0.23
8.82 11.75 6.62 7.23 5.52 4.81 3.65
5.64 6.25 7.62 9.73 8.75 6.74 4.50
19.12 20.69 12.78 14.04 11.58 6.99 4.61
10.21 11.65 16.63 19.30 17.72 9.90 7.37
TM8 Port7/8 BF gain imbalance
condition in time, thus the dual‐stream beamforming’s performance will decrease. It is also worth to mention that often TM8 Port7/8 beamforming gain imbalance had been observed in the field test as shown in Table 2.37. The key in beamforming is to generate the weighting vectors. There are two classes of DL beamforming schemes, cell‐specific DL broadcasting beamforming for cell‐specific channel and UE‐adaptive DL beamforming for PDSCH. There are different algorithms for calculating the optimum UE‐adaptive DL beamforming weightings. For example, it is possible to determine from the direction of the received uplink signal (DoA or AoA) if the angular spread is small, or from the uplink sounding reference signals channel estimation to calculate the beamforming weightings. Beamforming weight calculation by channel estimation is described below (Figure 2.73): ●● ●●
●●
Frequency (PRB)
●●
step 1: collect snapshots of instantaneous H (data matrix) step 2: determine covariance matrix: R = HHH, the format is M x M = 8 x 8, which depends on UE, subcarrier and TTI step 3: averaging of R → Rav, over frequency (over 4 most recently sounded PRBs) and over time (over past SRS receptions (IIR or FIR approach)) step 4: eigenvalue decomposition (EVD): Rav = V Λ VH, Λ is diagonal matrix with eigenvalues, decreasing absolute value => 1. eigenvalue is dominant, V and VH are orthogonal eigenmatrices (M x M)
SRS hopping SRS periodicity
D
S
U
D
D
t
1 HH k Hk Rav,f = 4 k ∈ {f with SRS}
Figure 2.73 Averaging of R.
D t+5
S
U
D
D
D
S
U
t + 10
D
D
time (TTI)
(1−α)Rav,t(t−5) + αRav,f(t) IIR Rav,t(t) = ,
1
t
Rav,f(k)
Lk = t–5(L+1)
FIR
103
104
LTE Optimization Engineering Handbook ●●
●● ●●
step 5: determine dominant eigenvector e1, corresponds to max eigenvalue => e1 = v1 = first column of VH, e1 is the optimum BF vector w (maximizes “ergodic capacity” C, “averaging” over all possible channels H, cuboid!) C = maxE{log 2 det(I w H R w} w/ w 1
step 6: use dominant eigenvector for DL BF: w = e1 implementation: step 4 and 5 merged by very fast “power method” => e1 directly
2.3.12.3 2D MIMO and Parameters
This part is foucused on spatial multiplexing of 2D MIMO. Spatial multiplexing allows a radio link composed of M transmit and N receive antennas (MxN) to exchange up to N independent data streams (codewords). Number of codewords is decided based on transmission mode, RI sent from UE, and so on. For UEs with low SINR beamforming should be used and for high SINR UEs spatial multiplexing should be used (Figure 2.74). For a spatial multiplexing system the corresponding generalization of the classic Shannon formula reads as:
C
log 2 1 SNR1
log 2 1 SNR2
log 2 1 SNRk
where SNRk = Sk/N now denotes the SNR of the kth information stream, k = min(n,m). Each MIMO mode can be switched to transmit diversity mode (TM2), and TM3 through TM9. Before pushing to MIMO parameters, a few basic concepts should be reviewed. Codeword (CW) is a transport block that has been processed by the physical layer in terms of CRC addition, channel coding, and rate matching. When two codewords are transferred, they do not need to be of equal size. CQI reporting, link adaptation, and HARQ run independently for each codeword. LTE supports simultaneous transmission resources in the same block by two relatively independent codeword, which is by spatial multiplexing (SM) technology to achieve. Each set of data sent through the antennas in a spatial multiplexing operation is called a layer. Layer mapping is needed for MIMO that maps the modulated symbols belonging to either one or two codewords onto a number of “layers” where the number of layers are less than or equal to the number of antenna ports, and a channel matrix rank is corresponding. PMI, the signal is “pre‐coded” (i.e., multiplied with a precoding matrix) at eNB side before transmission, optimum precoding matrix is selected from predefined “codebook” known at eNB and UE side. Rank, equivalent to the total number of layer. Rank indicator (RI), the number of layers that can be supported under the current channel conditions and modulation scheme. RI indicates M Tx
precoding Modulation + coding
Select # code words
Modulation + coding
RI
Layer mapping
CQI
Figure 2.74 Spatial multiplexing procedure.
V
N Rx H
Demod UH Demod
PMI
H = UΣVH
LTE Optimization and Principle and Method
Table 2.38 Legacy 2D MIMO parameters. Parameter
Range
Default
CQI threshold for fallback to closed Loop MIMO 1 CW Mode
0…16, step 0.1
11
CQI threshold for activation of closed loop MIMO 2 CW Mode
0…16, step 0.1
13
Rank threshold for fallback to closed loop MIMO 1 CW Mode
1…2, step 0.05
1.4
Rank threshold for activation of closed loop MIMO 2 CW Mode
1…2, step 0.05
1.6
CQI threshold for fallback to MIMO diversity
0…16, step 0.1
9
CQI threshold for Activation of open loop MIMO SM
0…16, step 0.1
11
Rank threshold for fallback to MIMO diversity
1…2, step 0.05
1.4
Rank threshold for Activation of open loop MIMO SM
1…2, step 0.05
1.6
the number of freedom degrees measured by the UE, which represents the maximum capacity of the Tx/Rx channel in terms of independent streams. Antenna port, it is not equal to the number of antennas, but rather a different channel estimation reference signal pattern. For ports 0 to 3, corresponding to RS transmission pattern of the multi‐antenna; for Port 4, corresponding to the PMCH, MBSFN case of RS; for port 5, corresponding to the UE Special RS. MIMO thoughput gains depend on above factors. In reality, maximizing rich scattering conditions within a cell, configuring the eNB to properly match MIMO parameters settings to real‐world conditions, and ensuring that UEs can take full advantage of the multipath conditions. Selection of the correct SU‐MIMO mode depends on factors such as mobility, CQI, and channel correlation (Rank). MIMO optimization process requires accurate measurement of these multipath conditions in order to achieve the best performance for a given environment while avoiding the time and expense of guesswork. Finally, the legacy 2D MIMO parameters are listed in Table 2.38. In a live network, SU‐MIMO optimization is the primary focus of operators attempting to maximize throughput gains. eNB make MIMO decision mainly based on UE report RI that decided by RS SINR and radio environment (the lower channel correlation the better). For low SINR, the two codewords is not easy to distinguish even with the lower channel correlation, if SINR > 12dB, the two codewords is easier to use MIMO, sometimes if the channel correlation is high, SM mode doesn’t increase throughput even with high SINR. From the test data, it is found rank1 CQI reflects the DL SINR fairly well, rank2 CQI have a heavy dependency over channel correlations. Higher correlated channel yield much smaller chances of rank2 reports, and the values are much lower. From Figure 2.75, it can be seen that the proportion of scheduled dual streams and the number of RI=2 are approximate linear relationship, i.e. the more probability of the reported RI=2, more dual streams will be scheduled. In addition, CQI and the dual streams scheduled are also approximately proportional relationship in a live network. 2.3.12.4 Massive‐MIMO
Many operators are quite interested in high‐rise building coverage, but legacy 2D beamforming only allowed controlling the beam pointing in azimuth direction. Massive MIMO techniques is introduced that exploit both the azimuth and elevation dimensions that are characterized by focusing the transmit power radiated to a user in the cellular system based on digital beamforming methods such that the peak of a resulting beam can be dynamically controlled in azimuth as well as in elevation direction, so massive MIMO can significantly improve user data
105
LTE Optimization Engineering Handbook 30 SM Throughput (Mbps)
TxDiv Throughput (Mbps)
25 For low SNRs, TX Diversity gives slightly better throughput than SM, since the power from both eNB TX antenna is aggregated to decode one code word. This is in line with theoretical simulations
20 15 10
Although the two cases are static TX diversity and SM, switching point could be around this area for dynamic adaptation of MIMO
5 0
–4
–2
0
2
4
6
8
10 12 14 16 18 20 22 24 26 SINR (dB)
2 Average Reported Rank
106
1.9
No correlation
1.8
Med correlation
1.7
High correlation
1.6 1.5 1.4 1.3 1.2 1.1 1 –10
0
10
20
30
DL SINR (dB) RI = 2 reported ratio RI = 2 scheduled ratio
100.00% 80.00% 60.00% 40.00% 20.00% 0.00% 100.00%
RI = 2 scheduled ratio
90.00% 80.00% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00% 5
7
9
Figure 2.75 MIMO ratio versus RI and CQI.
11
13
CQI
LTE Optimization and Principle and Method
rate in a high‐rise scenario. Massive MIMO deployed high‐gain adaptive beamforming and high‐order spatial‐multiplexing with a large‐scale array. The antenna type to be used is typically a planar antenna array consisting of multiple cross‐polarized antenna elements arranged in a rectangular two‐dimensional grid. From 3GPP perspective, DMRS and/or CSI‐RS should support more than eight antenna ports, for example, 32 or 64 or even more antennas/ports, beamforming main lobe sharpens as number of antenna elements increases. Massive MIMO, many subelements are controllable from baseband (Figure 2.76). Massive MIMO needs acquiring channel state information at the eNB for downlink MU‐ MIMO, also needs efficient signaling for multiplexing large numbers of UEs. TDD is often viewed as an easier problem than FDD as it leverages DL/UL reciprocity; for FDD, a codebook feedback solution may have difficulty that large number of antenna elements require large codebook size, which will result extremely high UE codebook search complexity. The UE needs receive CSI‐RS and computes azimuth and elevation PMI. User‐specific beams now formed in elevation domain to provide optimal throughput at all levels of the building. 2.3.13 Power Control
Power control and proper power configuration will reduce inter‐cell interference and power consumption. This leads to higher cell capacity and the control of the maximum data rate for UE at cell edge and limit the interference that cell‐edge users create to the neigboring cells. In addition, it helps to prolong the battery life of the UE. In LTE, fractional power control (FPC) is introduced to allow a more flexible trade‐off between spectral efficiency and cell‐edge rates. 2.3.13.1 PUSCH/PUCCH Power Control
The basic power control of PUSCH aspects are open‐loop power control with slow aperiodic closed loop correction factor, it is based on fractional pathloss power control. With fractional pathloss power control, it will be easier to set power control parameters that will enable higher UL peak rate in the cell without sacrificing cell‐edge performance. It enables a trade‐off between maximized UL cell edge bitrates versus improving the overall UL cell capacity. The formula of PUSCH power control is:
PPUSCH i
min PMAX , 10.log10 M PUSCH i
P0 _ PUSCH
. PL
TF
TF i
f i
where PMAX is the maximum allowed power that depends on the UE power class. MPUSCH(i) is the bandwidth factor, expressed in number of resource blocks taken from the resource allocation valid for uplink subframe i. P0_PUSCH is a parameter obtained as a sum of a cell‐specific nominal component p0NominalPUSCH signaled from higher layers and a UE‐specific component p0UePUSCH. P0_PUSCH is the basic starting point of open loop power control. α is partial pathloss compensation factor, a cell‐specific parameter signaled from higher layers in order to support fractional power control. α=1 corresponds to classic UL power control, that is, full pathloss compensation. PL is the downlink path loss estimate calculated in the UE. ΔTF(TF(i)) denotes the power offset depending on PUSCH transport format TF(i). f(i) is PUSCH close loop power control adjustment derived from TPC command in subframe i‐4. Both accumulated and non‐accumulated power control rules are used. Compared with PUSCH’s fractional pathloss power control, PUCCH only uses a complete pathloss power control mechanism. The PUCCH power control procedure is used to guarantee the required error rate; it aims at achieving a target SIR the value of which guarantees the
107
Physical Antenna Connectors
Virtual Antennas
Antenna Ports
DMRS or Data
Physical Antenna Sub -elements
Wprec
Wvirt
Wfeed
CSI-RS
Reference - or data signals
Precoder
Antenna Array
Constrained Virtualization
Unrestricted Virtualization (per RE) st
21 floor
User specific beams Broadened elevation sector beam.
250
13th floor
RX
TX
63 m
Site
8th floor
RX
6 RX
Average sum rate (Mbps/cell)
Highrise
200 150 100 50
Matched filter Zero-forcing Interference free
23 m nd
2 80 m
floor
0
0
100
200
300
Number of antennas
Figure 2.76 Massive‐MIMO principle.
400
500
LTE Optimization and Principle and Method
required error rate. Higher settings of this parameter will improve PUCCH reception, but will also drive higher UE TX power leading to interference to neighboring cells, and vice versa. PPUCCH(i) = min{ PCMAX, P0_nominal_PUCCH+Pathloss+h (nCQI,nHARQ)+∆F_PUCCH(F)+g(i) (UE specific parameter) Unlike PUSCH, the PUCCH power control is based on SINR instead of power spectral density (PSD) as there is no link adaptation on PUCCH; fixed MCS is used over PUCCH. The UE‐received PUCCH SINR will be compared with the target PUCCH SINR and TPC values will be generated based on the difference between the two values. The target PUCCH SINR default value will be estimated based on the minimum required by the received SINR in order to achieve a certain erasure or BER performance. eNB sets a semi‐static nominal power P0 for all UEs in the cells first (P0_PUSCH for PUSCH and P0_PUCCH for PUCCH) and broadcast it to all UEs by SIB2 (UplinkPowerControlCommon: p0NominalPUSCH, p0NominalPUCCH). P0_PUSCH values range is (−126 to 24) dBm. Each UE has a UE‐specific nominal offset power (P0_UE_PUSCH for PUSCH and P0_UE_PUCCH for PUCCH), which is sent to UE by dedicated RRC signal (UplinkPowerControlDedicated: P0_UE_PUSCH, P0_UE_PUCCH). It is worth to note that P0_PUSCH is different for semi‐persistent grant and dynamic scheduled grant (SPS‐ConfigUL: p0NominalPUSCH‐Persistent). Table 2.39 gives the main parameters of power control. Figure 2.77 shows two set of p0NorminalPusch (−106dBm versus −96dBm) comparation. When p0NorminalPusch is set to −106dBm, uplink throughput reduces by approximately 2Mbps in good RF conditions as lower uplink received power target leads to lower UL_SINR and further leading to lower MCS assignment by link adaptation. With −106dBm, uplink received power target is lower by 10dB. Hence, this leads to a much lower uplink Tx_power. 2.3.13.2 PRACH Power Control
Open‐loop power control is applied for initial transmission of RACH. The transmit power is determined taking into account the total UL interference level and the required SINR operating point, which can be determined at the UE as: PRACH _ msg1
min PCMAX , PL P0 _ PREAMBLE
PREAMBLE
N PREAMBLE 1
RAMP _ UP
Table 2.39 The related parameters of power control. Parameter
Description
Range
Default
P0UEPucch
Power offset for UE PUCCH TX power calculation
−8…7 dB, step 1 dB
0 dB
P0UEPusch
Power offset for UE PUSCH TX power calculation
−8…7 dB, step 1 dB
0 dB
P0NomPucch
Nominal power for UE PUCCH TX power calculation
−127…‐96 dB, step 1 dB −96 dB
P0NomPusch
Nominal power for UE PUSCH TX power calculation
−127…‐96 dB, step 1 dB −100 dB
srsPwrOffset
Power Offset For SRS Transmission Power Calculation
0…15, step 1
α
α, Indicates the compensation factor for path loss.
α0 (0), α0.4 (1), α0.5 (2), α 1 (7) α 0.6 (3), α 0.7 (4), α 0.8 (5), α 0.9 (6), α 1 (7)
7
109
LTE Optimization Engineering Handbook Throughput 25
–96 dBm
Mbps
20
–106 dBm
15 10 5 0 Good
Medium
Poor
UL Tx Power 25 20 dBm
110
–96 dBm
15 10
–106 dBm
5 0
Medium
Good
–5
Poor
Figure 2.77 Two set of p0NorminalPusch (–106dBm versus –96dBm) comparation.
P0‐PREAMBLE is the preamble received power set point determined at the eNB. This parameter is calculated from the target SINR operating point, and the UL interference‐plus‐noise (IN) power in the PRACH resource. P0 _ PREAMBLE
SINRT arg et
IN
M arg in
ΔPREAMBLE is the power offset value dependent on PRACH preamble format, which is given by prach‐ConfigIndex. The preamble format–based power offset values are presented in Table 2.40. According to the estimated received power of RACH preamble, the eNB is able to know the SNR condition of the UE initialized the random access. Thus it will assign a certain power to the UE so that it can send message 3 with reasonable power to enable it to receive the message 3 correctly so that in most cases UE will not have to restart a new random access due to the failure of message 3 transmission. Table 2.40 ΔPREAMBLE value. Preamble format
DELTA_PREAMBLE value
0
0 dB
1
0 dB
2
−3 dB
3
−3 dB
4
8 dB
LTE Optimization and Principle and Method Tx power to meet the target Rx power
UE
eNB RACH preamble
x
x
x
RACH preamble
+∆Ramp
RACH preamble
+N∆Ramp
RAR y y
y
RRC Connection Req
+N∆Ramp + δmsg2
RRC Connection Req
+N∆Ramp + δmsg2
PO_Pre PO_Pre
PO_Pre
(δmsg2)
RRC Connection Req
+N∆Ramp + δmsg2
Target Rx power
PO_Pre + ∆msg3 PO_Pre + ∆msg3
PO_Pre + ∆msg3
RRC Connection Setup
Closed loop power control with accumulation
RRC Connection Setup complete (+NAS setup) NAS Setup + Authentication
P0_nominal_PUSCH
RRC Connection Reconfiguration
Figure 2.78 Power control algorithm.
MSG 3 Tx Power 10 * log 10 Mpusch Last Preamble Power deltaPremableMsg 3 PC _ msg 2
PC_msg2 is determined in the range of −6 to 8 dB (eight values) based on the preamble detection performance. So, the power control algorithm during the whole UE RACH procedure is shown in Figure 2.78. 2.3.14 Antenna Adjustment
Besides antenna azimuths and tilting, antenna placement also has big impact on other cell interference in real environment. A poor site design can have a significant impact upon the performance of a potentially good site. Site design involves identifying an appropriate location for each antenna and the eNB cabinet. When there is a requirement to achieve a specific isolation from another radio system then that isolation is easier to achieve if the antennas are separated vertically rather than horizontally. The most important requirement is that antennas should be mounted such that their main beams are not obstructed. This should include both the horizontal and vertical half power beamwdiths, that is, the beamwidths at which the antenna gain has decreased by 3 dB. In the case of a roof‐top site, obstructions could be other antennas or cabins located on the same or a neighboring roof. In the case of mast or pole
111
112
LTE Optimization Engineering Handbook
Main beam is not obstructed
Poor Position
Good Position
Figure 2.79 Examples of poor and good roof‐top antenna locations.
mounted antennas, obstructions could be trees or nearby buildings. Figure 2.79 illustrates examples of poor and good roof‐top antenna positions. 2.3.14.1 Antenna Position
In the example of the poor antanna installation position, the antenna is located behind and slightly higher than some existing antennas. In this case the main beam of the antenna is obstructed and its performance is likely to be deteriorated. In the example of the good position, the antenna is located in front of and slightly lower than some existing antennas. In this case the main beam of the antenna is not obstructed, although care should be taken that the rear lobe of the antenna does not cause interference to the other radio systems. There is also a requirement to ensure that the edge of the roof‐top does not cause shadowing of the antenna. If an antenna is positioned on the edge of a roof‐top then it is unlikely to incur any shadowing from the roof‐ top itself. However, as the antenna position is moved away from the edge then the antenna is more likely to incur shadowing. Antennas that are located away from the edge should be mounted with an increased height. A general rule is that if you can walk in front of the antenna then it should be mounted 3 m above the roof‐top. Figure 2.80 illustrates the principle of shadowing from a roof‐top and suggests a range of heights that could be used to avoid shadowing. In most cases an antenna would be mounted less than 10 m from the edge of the building and its suggested height would be obtained by dividing the distance to the edge by two. Wherever possible, antenna mountings should allow the height and azimuth of the antenna to be adjusted. In the case of antennas mounted on walls then the azimuth should be configured to ensure that the horizontal beamwidth of the antenna is not compromised. In general, a 15‐degree safety margin should be added to each side of the half‐power horizontal beamwidth and then a check made to ensure that the composite beamwidth is free from obstruction. Figure 2.81 illustrates the principle of avoiding shadowing from the walls upon which antennas are mounted.
General rule Clearance angle
h d
d < 10 m
h > d/2
10 < d < 20 m
h > d/3
d > 20 m
h > d/4
Figure 2.80 Principle of avoiding shadowing from a roof‐top.
LTE Optimization and Principle and Method
Figure 2.81 Principle of avoiding shadowing from walls.
Direction of main beam
Poor Position
15° safety margin
15° safety margin Direction of main beam
Half power beam width
Good Position
When there are other antennas on the same mast, the same roof‐top or the same wall, then the isolation from those antennas should be maximized without compromising the position of the LTE antennas. The isolation requirement will depend upon the systems to which the antennas belong. The isolation requirement can be translated into a physical separation using curves that plot the measured isolation as a function of physical separation. These curves depend upon the gain patterns of the antennas being used and whether or not the antennas are cross‐polar. As an example, the LTE system requires 40 dB of isolation from the UMTS system. If the antennas have a vertical separation then there should be at least 0.2 m between the base of one antenna and the top of the other antenna. If the antennas have a horizontal physical separation and a horizontal beamwidth of 65° then there should be at least 0.5 m between them. Whenever possible, a vertical separation should be combined with a horizontal separation to increase the achieved isolation. In cases where these separations cannot be achieved, then the isolation requirement should be solved in other ways. Alternatively, the isolation requirement can be achieved using a diplexor and allowing the two radio systems to share the same feeders. A diplexor typically offers 40 dB of isolation. Radio systems may also share the same antennas. In general, this has the drawback of restricting both radio systems to using the same antenna downtilts; that is, downtilts cannot be configured separately for each system. Antennas that have remotely controllable tilts are generally more expensive, but tilt changes can be made with relative ease. 2.3.14.2 Remote Electrical Tilt
Antenna tilting is a very powerful method to control network capacity and performance optimization. With the tilt, it directs irradiation further down (or higher), concentrating the energy in the new desired direction. The remote electrical tilt (RET) function enables the operator to control and optimize the coverage area by modifying the inclination of installed antennas, without the need for climbing masts. The RET provides electrical tilt for tuning and optimizing the network by adjusting the vertical lobe‐angle (adjusting the phase‐shifter on the antenna) of the antenna. The RET unit can be mounted on any antenna with tilting capability, regardless of height, gain, or band. The RET unit communicates over an interface by the open specifications defined
113
114
LTE Optimization Engineering Handbook Triple Band
Dual Band
Single Band
Manual Adjustment Knobs
Electrical Tilt Indicators
iRET Tilt indicator - reading a value of 4° (+/– 1°)
RetSubunit
RET AuPort
RfBranch
TMA
RfBranch
Rx
TmaSubunit
RfPort
Tx/Rx & DC
RBS
Figure 2.82 Remote electrical tilt.
by the antenna interface standards group (AISG) to ensure basic interoperability of antennas and to control infrastructure (Figure 2.82). Antenna tilt is defined as the angle of the main beam of the antenna below the horizontal plane. Positive and negative angles are also referred to as downtilt and up‐tilt respectively. Antenna downtilt can be adjusted mechanically or electrically. Electrical tilt is realized by wire
LTE Optimization and Principle and Method
+
Antenna Axis
Mechanical downtilt, the pattern in the front goes down, and behind goes up.
Mechanical Tilt
Mechanical Tilt Electrical Tilt
Total Tilt
Figure 2.83 Mechanical downtilt, electrical tilt, and total downtilt.
feed phase shifting. This phase shifting is realized by modification of feed cable length in antenna factory. In electrical downtilt, phases of antenna elements are adjusted so that desired tilt angle is achieved by tilting main, side, and back lobes uniformly contrary to mechanical downtilt. The total tilt is the inclination of the maximum of the antenna’s main beam with respect to the horizontal plane. The performances of vertical sectorization depend on antenna tilt, vertical beam width, and front‐to‐back ratio, and so on (Figure 2.83). Downtilt calculation: Downtilt=arctan(h/D)+ (Beamwidth/2). The optimal amount of tilt is a trade‐off between coverage and interference reduction, it depends on the real‐time traffic situation with a varying degree of user clustering or hotspots. In a realistic network, the traffic characteristics are dynamically changing and the optimum tilt to the current traffic conditions, which referred to as automatic tilt control. It is possible to pre‐define tilt settings for different times of the day (such as rush hour, midday, evening, night) and different times of the week based on historic data (Figure 2.84). The antenna effects are combined as a sum of antenna gain, horizontal pattern, and elevation pattern. The sum of horizontal and vertical patterns is limited for a common front‐to‐back attenuation Am and SLAv. The antenna gain of horizontal and vertical radiation patterns are shown below:
A
min 12
A
min 12
3 dB
3 dB
, Am , Am
25dB
, SLAv , SLAv
20dB A
,
min{
AH
AV
, Am
Assume that the antenna vertical (3dB) beamwidth is 120 and antenna height is 30m. The relation of downtilt and Dmin, Dmax is shown in Table 2.41. The algorithm for tilt control is based on relative load between different cells covering the same area/cluster. First, find Max load and Min load in the cluster, then if the Max load minus Min load is greater than a predefined margin, find downtilt for sector with Max load and do up‐tilt for the sector with Min load. The load for cell number m is defined as: Lm = Im/N + Im, Where Im is the total interference experienced and N is thermal noise.
115
116
LTE Optimization Engineering Handbook
0°
Downtilt Angle θ Side Lobe
3dB Beamwidth
Main Lobe
Rapidly Decreasing Signal Strength Region
h Usable Signal Area
Shadow Area
Dmax
Vertical Beamwidth
Antenna Height
Dmin (m) Dmax (m)
Figure 2.84 Downtilt calculation.
Table 2.41 The relation of downtilt and Dmin, Dmax. Downtilt (°)
Dmin (m)
Dmax (m)
Downtilt (°)
Dmin (m)
Dmax (m)
0
285
infinite
10
105
429
1
244
infinite
11
98
343
2
213
infinite
12
92
285
3
189
infinite
13
87
244
4
170
infinite
14
82
213
5
154
infinite
15
78
189
6
141
infinite
16
74
170
7
130
1719
17
71
154
8
120
859
18
67
141
9
112
572
19
64
130
LTE Optimization and Principle and Method
2.3.14.3 Antenna Azimuths and Tilts Optimization
The antenna system plays an important role in mobile communications. Antenna height, tilts, and azimuths (to certain extent) are strong primary RF shaping factors. The performance of the entire network is affected by improper type, location, or configured parameters of the antenna system. Figure 2.85 gives an example of how reduced antenna height improved application coverage. The antenna tilt is mainly affected by the coverage radius of the cell and the average SINR value in the coverage area, and the optimization of the tilt must take into account the balance between RF coverage and SINR. Figure 2.86 is the driving test results of the impact of the tilt on the SINR value. Usually in the tilt adjustment will encounter the following three typical scenarios: uplink coverage limited, overshooting, and coverage holes, and so on. Uplink coverage limited: When a cell is in a continuous coverage zone, due to the large tilt, worse propagation conditions or big penetration loss, and so on, the terminals may encounter rdaio link failure, although RSRP is still higher, it is not able to make the handover. This situation may be uplink coverage limited, which can be considered to reduce the tilt to improve coverage. Tilt
35% cells with >40 m, >15°
45% cells with 10 Mbps
High antenna (>40 m) gives 35% calls CQI >10; 40% user throughput >10 Mbps
30 20
cells potentially serving high rise Indoor Traffic
10 0 0
20
40
60
80
Figure 2.85 Antenna height versus tilt. 20 18 16
Average SINR
14 12 10 8 6 4 2 0
0
2
4
6
8
10
12
Antenna Downtilt, degrees
Figure 2.86 Average SINR versus tilt.
14
16
Height (m)
117
LTE Optimization Engineering Handbook
Overshooting: When a cell is in a continuous coverage zone, due to the small tilt, good propagation conditions, and so on, overshooting will happen. When a cell has coverage beyond the intended coverage area, it leads to interference problems especially if the signal strength of the overshooting cell is high. This issue can be determined by observing the handover area. Usually the solution is to change the antenna configuration of the overshooting cell, for example, tilting down the antenna, redirecting the antenna orientation, or reducing the antenna height. Coverage holes: The serving RSRP within this area is below the minimum required signal level (qRxLevMin) to set up and maintain LTE service. There are two causes for coverage holes: lack of signal power and high interference. The coverage holes may cause handover failure or session drop. At this time it should adjust tilts of the cells to make up the hole based on the location of coverage hole and the surrounding RF conditions. 2.3.14.4 VSWR Troubleshooting
VSWR stands for voltage standing wave ratio, it is a measure of how efficiently RF power is transmitted from a power source, through a transmission line, into a antenna. VSWR is the ratio of the peak amplitude to the minimum amplitude of a standing wave, calculated by the ratio of the highest voltage to the lowest voltage along the transmission line. This ratio is a function of the reflection coefficient, which in simple terms is just a measurement of power reflected from the antenna (Figure 2.87).
Back to Transmitter
Voltage Amplitude
Reverse Power
Vmax
Transmission Line
Vmin
Forward Power
To Antenna
2 1.5 Voltage (normalized)
118
1 0.5 0 –0.5 –1 –1.5 –2
0
5
10
Forward wave Composite wave
Figure 2.87 VSWR concept.
15
20
25
Reflected wave Detected standing wave
30
LTE Optimization and Principle and Method
VSWR
V max / V min
Return loss (RL) is the loss of power in the signal returned/reflected by a discontinuity in a transmission line. It is usually expressed as a ratio in decibels (dB).
RL dB
10 log10
Pi Pr
Where RL is the return loss in dB, Pi is the incident/forward power and Pr is the reflected/ reverse power. The relation between VSWR and RL is present as below:
Example: by substituting VSWR = 1.5 and RL/20 = X, it can get: 1 10 x / 1 10 x 1.5 x 2.5 0.5 10 x 1.5 10 1 RL 20 * 0.6989 13.979 ~14
1.5
10 x 1 / 10 x 1 10 x 1 5 10 x X log10 5 0.6989,
A high value enables detection of a low reflected power (high return loss). A low value requires a high reflected power (low return loss) to generate an alarm. The normal VSWR range is 1 to1.5. If current VSWR is more than a specified threshold, then eNB will generate relevant alarm. The VSWR antenna supervision enables supervision of the feeder cables. Measurements are made on the reflected radio power, making it possible to detect breaks or loose connections in the cables connected to the radio unit. An alarm is raised if return loss is below the configured VSWR sensitivity value. It is possible to activate VSWR antenna supervision on all RF ports supporting measurements of reflected radio power in the radio unit. Supervision can be made on RF ports used for downlink transmitter branch where power is transmitted. Possible causes of VSWR issues include improperly installed antenna, damaged or defective antenna hardware, damaged feeder cable or jumper cable (i.e., excessive bend radius), dirty or loose cable connection, improperly sealed/weatherproofed cable connection, water in the antenna of antenna feeder, and snow on the antenna, and so on.
119
120
LTE Optimization Engineering Handbook
2.3.15 Main Key Performance Indicators
It is necessary for the LTE to provide (periodically or on demand) a range of key performance indicators (KPI). The aim of KPI monitoring is in order to provide global information of network behaviour, detect the different problems, analyzing and correlating them with the rest of statistics in order to fix problems encountered. These KPIs ensure adequate analysis of LTE performance in commercial operation environments. The operator‐associated target performance guidelines and the resultant measured KPIs give a means to measure actual performance against target performance. The KPIs will provide the ability monitor network performance at the user plane, control plane, and network level ensuring a full understanding on the overall LTE and eNB performance. These KPIs will enable the monitoring and subsequent troubleshooting of network performance, and will support performance analysis at the user plane, control plane, and network level. Take TDD_LTE, a 2300 MHz operation, or a 20 MHz bandwidth system, for example, LTE RAN KPIs are proposed in Table 2.42. Table 2.42 TDD_LTE RAN KPIs. Drive test based
Recommended KPI
Drive test OSS based based
Session Setup Success >= 99.3 % Rate (%)
>= 98.5 %
DL RLC Throughput (Mbps)
>= 23.2
‐
E‐RAB establishment successrate (%)
>= 99.15 %
UL RLC Throughput (Mbps)
>= 9.2
‐
Minutes Per >=55min/drop >= 45 min/drop DL PDCP Abnormal Release (%) Throughput (Mbps)
>= 20
‐
Uplink User Throughput (Mbps)
>= 12
>= 15
UL PDCP Throughput (Mbps)
>= 8
‐
Downlink User Throughput (Mbps)
>= 30
>= 35
DL Application Throughput (Mbps)
>= 21
>= 24.5
Uplink Peak User Throughput (Mbps)
>= 30
>= 33
Latency Downlink (ms) = 65
>= 70
Packet Loss Rate ‐ Uplink
= 99.2 %
Radio received interference power PUCCH
= 99.1 %
‐
Voice Telephony ‐ Jitter
‐
Added ERAB establishment success Rate
>= 99.4 %
‐
Voice Telephony(VT) ‐ = 92.0 %
VT ‐ Speech Quality
‐
>= 4.1 [MOS]
DL PDSCH Throughput (Mbps)
>= 30
>= 35
VT ‐ Session Setup Success Rate
>= 99.0%
>= 98.7 %
UL PDSCH Throughput (Mbps)
>= 12
>= 15
VT ‐ Session Setup Time
SM 40000 30000 16QAM => 64QAM 20000
QPSK => 16QAM
10000 0
–6
–4
–2
–1
1
2
4
6
7
9
11
13
15
16
19
26
28
31
32
DL C(I+N) - dB
Figure 4.17 DL throughput versus SINR and switching SINRs based on simulations.
161
LTE Optimization Engineering Handbook
code words are sent at the same time. Although the two cases are static TX diversity and SM, the switching point could be around SNIR = 10 to 13 dB for dynamic adaptation of MIMO mode (open loop or closed loop MIMO) as shown in Figure 4.18. Optimum switching point is suggested to be at around SNIR = 18 dB for UEs at medium speed of 5 to 30 km/h. Antenna correlation and antenna power imbalance are the key factors affect the MIMO per formance. Both of these factors increase the SNR difference between streams. In order to mini mize antenna correlation needed to consider practical deployment, power imbalance between antenna branches were discussed in Chapter 3. Figure 4.19 shows a measurement example with a real UE and antenna correlation artificially altered with a fading simulation, which are the terms of different degrees of correlation. In order to switch from SM to TxDiv, it is required for the CQI‐ or RI‐averaged values to be below certain thresholds (RiThreshold2 and CqiThreshold2) defined for these two metrics. In order to switch from TxDiv to SM it is required for the CQI and RI averaged values to be above 25 Throughput (Mbps)
SM Throughput (Mbps)
TxDiv Throughput (Mbps)
20
SM Throughput (Mbps)
15 TxDiv Throughput (Mbps)
10 5
SINR (dB)
0 –4
–2
0
2
4
6
8
10
12
14
16
18
20
22
24
26
Figure 4.18 DL throughput versus SINR.
MIMO Usage Probability
100% 80% 60% 40% 20%
27
25
23
21
19
17
15
11
13
9
7
5
3
1
–1
–3
–5
0% –7
162
SNR CQI
SM
SM
CqiThreshold1 CqiThreshold2 Time RI
Filtered cqi, ri
RiThreshold1 RiThreshold2 Time
Figure 4.19 MIMO mode versus measured SNR in unloaded network.
28
30
32
34
Capacity Optimization
a certain thresholds (RiThreshold1 and CqiThreshold1) defined for these two metrics. The recommended values for these parameters involved in the open loop dynamic MIMO are: ●●
RiThreshold1 = 1.8, RiThreshold2 = 1.4, CqiThreshold1 = 11, CqiThreshold12 = 9
The thresholds defined for the switching are based on the CQI/RI, allow for hysteresis. Depending on the difference between these up and down thresholds the switching between the modes will be faster or slower. 4.6.4.2 4Tx/4Rx Performance
It is known that four‐antenna cell application has been widely used in networks, which can be used to extend the cell coverage and increase user data rates. The main motivation for 4TX/4RX is enhanced coverage and enhanced cell‐edge data rates. Compare with 2TX/2RX and 4TX/4RX, the average improvement in the user data rate was +50% in UL and +25% in DL. The data volumes increased even more because the cell coverage area also increased. The typical antenna solution for 4TX/4RX is XX‐polarized antenna in a single radome. The number of antennas is not increased compared to 2TX/2RX but the antenna just gets slightly wider, which is shown in Figure 4.20. Another antenna option would be two separate X‐polar ized antennas, which would be more difficult for the site solution. The simulations and the field measurements indicate that XX‐polarized antenna gives better performance. 4Rx can improve UL coverage for all channels, the cell average bit rate and primarily the cell‐edge bitrate. With 4Rx, interference rejection combining (IRC) generally improves the per formance when 4‐way Rx diversity is used. Figure 4.21 shows that due to 4RX, cell‐edge is extended with MIMO 4x2 versus 2x2, and cell‐edge UL throughput gets a better performance due to the antenna spatial diversity and up to 3 dB of coverage improvement. With the IRC feature activated, the system dynamically switches between the combining methods IRC and maximum ratio combining (MRC) depending on interference situation. 4.6.4.3 Transmission Mode Switch
3GPP Rel 9 defines multiple transmission mode, which is controlled by RRC; mode switching/ selection is higher layer configured and on a slow basis (several seconds) in order to ensure optimum UE DL throughput and performance. The transmission mode is selected based on wideband SINR and UE speed. The CQI report is firstly mapped to SINR, then sum over streams (in case of rank 2) and finally average over frequency. The beamforming weighting factor is based on periodic SRS (10 ms), SRS Rx_SINR impacts to beamforming weighting factor is considered. Beamforming performs good enough for high speed even with 350 km/h coverage due to intra‐polarization beamforming. This still can give higher gain than broadcasting and inter‐cell interference of beamforming is more
Figure 4.20 Antenna options for 4TX4RX.
Single XX-pol antenna
Two X-pol antennas
163
LTE Optimization Engineering Handbook DL Tput (Mbps): 2x20 W
DL Tput (Mbps): 4x10 W
DL Tput
DL Tput (Mbps): 4x20 W 60 50 40 30 20 10 0
Figure 4.21 DL and UL throughput versus pathloss.
DL Tput (Mbps): 4*20 W
DL Tput (Mbps): 2*20 W
DL Tput (Mbps): 4*10 W
–75 –80 –85 –90 –95 –100 –105 –110 –115 –120 –125 Path Loss (dB) UL Tput (Mbps): 2RX
25 UL Tput
164
UL Tput (Mbps): 4RX
DL Tput (Mbps): 4RX
20 15
DL Tput (Mbps): 2RX
10 5 0
–75 –80 –85 –90 –95 –100 –105 –110 –115 –120 –125 Path Loss (dB)
robust to FSS performance. But TM3 provides higher peak rate due to more robust to SRS channel estimation error and SRS capacity problem. TM3/TM8 inter‐mode switch is based on the DL SINR estimation in eNB. TM3 is used in good RF condition to get better DL throughput. TM8 is used in poor and middle RF condition to get better throughput by utilizing beamforming gain. But in rapid fading channel, TM8 may have worse performance than TM3 as shown in Figure 4.22. From Figure 4.22, it can be seen that TM3/8 switch has lower throughput than TM3 and TM8 below SINR 23 dB and high DL BLER in high SINR environment in TM8. 4.6.4.4 UL MU‐MIMO
Contrary to DL, 3GPP Rel 8 and 9 do not standardize single user MIMO in UL, UE is able to transmit only one stream of data, instead multi‐user MIMO (or virtual MIMO, V‐MIMO) is supported, which allows pairs of UEs with appropriate radio conditions to be scheduled on the same time and frequency radio resources. Usually with SINR threshold and DoA threshold, UEs can be filtered to be paired. The pairing criterions are based on radio conditions of individual UEs and potential pairs (othogonality), pairing candidates are chosen from among the UEs that are to be scheduled in the same TTI. Advanced eNB receiver is needed to be able to separate data streams from MU‐MIMO UEs. Othogonality between the UEs got from the SRS soundings is obtained from the channel coefficients of UE pairs. Orthogonality is a wideband value. Calculation of the channel vec tor is done in the same way as long term beamforming vector (hi, hi are also wideband values). Oij
1
* hUEi hUEj
hUEi
hUEj
2
Only the UE pairs with wideband orthogonality metric above the threshold can be considered for pairing. UE pairs with low orthogonality will interfere one another too much, as MU‐MIMO receiver implementation does not cancel the interfering UE.
Capacity Optimization
Figure 4.22 DL throughput versus transmission mode.
DL PHY throughput (kb/s) 80000 70000 60000 50000 40000 30000 20000 10000 0
–17 –14 –11 –8 –5 –2
1
4
7
TM38
10 13 16 19 22 25 28
TM3
TM8
DL BLER (%) 20 Fast fading
15 10 5 0 –17 –14 –11 –8 –5 –2
1
4
7
10 13 16 19 22 25 28
In single‐user mode preliminary MCS is calculated based on the wideband SINR measure ment provided by physical layer. UEs that are scheduled in MU‐MIMO mode will be assigned with MCS lower than in single user mode. Assuming SINRSIMO(i) and SINRSIMO(j) as the UL wideband SINR of UE i and UE j before pairing respectively, the UL wideband SINR of UE i after pairing SINR MU‐MIMO(i) and SINR MU‐MIMO(j) are:
SINRMU
MIMO
i
SINRMU
MIMO
j
1 Oij * SINRSIMO j 1 SINRSIMO j 1 Oij * SINRSIMO i 1 SINRSIMO i
* SINRSIMO i ;
* SINRSIMO j ;
The scheduling metric relative SINR for the whole UE pair is calculated by averaging over two relative SINR of UEs in the UE pair. It notes that when both of the MU‐MIMO multiplexed TB shall be re‐transmitted, they shall not be multiplexed instead of with non‐overlapped RBs allocation. 4.6.5 DL PRB Allocation and Utilization Mechanism
In order to reduce the inter‐cell interference in low load scenarios, an enhancement of the resource utilization optimized strategy in live network is introduced. The start index of resource
165
LTE Optimization Engineering Handbook
allocation is based on ICIC feature for decreasing the PRB collisions between cells as shown in Figure 4.23. In this way, cell‐edge users (cell‐edge/non‐edge classification is done using the event A3) are prioritized for allocation close to the start index. Usually, an operator starts the PRB allocation at the start index with “cell‐edge VoIP” UEs, continuing with “cell‐edge best effort” UEs, and finally continuing with “non‐edge” UEs. After change PRB start position offset, the expected gain in non peak hour is obvious, but not in busy hour. From Figure 4.24, it can be seen that in a non‐peak hour, a 194 kbps improvement (11:00‐16:00) for DL user throughput that there is no changes in a BH. In live network, trying to avoid high PRB utilization is necessary. There are very limited means of reducing of PRB utilization. However, some of these methods can be tried on case‐by‐ case basis. ●● ●● ●● ●●
Traffic offload to less utilized neighboring cells Reduce control channel resources (Before that check PDCCH utilization) Add bandwidth as bandwidth increase would increase number of PRBs Reduce inactivity timer value so that inactive user can be released early.
PRB Collision
0 0
Parameter
Current
dlConfigurableFrequencyStart
0
0 0
0
0
0
Starting point for allocations Configured
0
0
0
0
0
Proposal S1 = 0, S2 = 34, S3 = 67
7000
400
6000
350
5000
300 250
4000
200
3000
150
2000
100
1000
50
No Change
RRC User
Active User DL
0
3/22 11:00 3/22 12:00 3/22 13:00 3/22 14:00 3/22 15:00 3/22 16:00 3/22 17:00 3/22 18:00 3/22 19:00 3/22 20:00 3/22 21:00 3/22 22:00 3/22 23:00
3/15 11:00 3/15 12:00 3/15 13:00 3/15 14:00 3/15 15:00 3/15 16:00 3/15 17:00 3/15 18:00 3/15 19:00 3/15 20:00 3/15 21:00 3/15 22:00 3/15 23:00
0
DL User Throughput
DL Cell Throughput
Figure 4.24 Throughput gain after PRB start position offset changed.
Throughput (kbps)
450
67
34 0
34
67
0
Figure 4.23 PRB start position offset.
User
166
67
34
Change parameter-PRB start position offset
Capacity Optimization UL average PRB utilization
100 90 80
DL average PRB utilization
70 60 50 40 30 20 10 0 0
200
400
600
800
1000
1200
1400
1600
Figure 4.25 Average PRB utilization.
It is worth noting that UL PRB utilization grows very rapidly compared to the DL because there is no MIMO or 64QAM in UL, therefore, the UL efficiency is much less compared to the DL. From Figure 4.25 it can be seen that for 400Kbps throughput, ~20% PRB utilization in DL and ~90% PRB utilization in UL. 4.6.6 DL BLER
There is a strong correlation between high BLER and low throughput, so it is clear that the issue is purely in RF, not an issue for any backhaul/transport problems. In most cases, ratio of retransmission is the major impact factor for throughput, BLER (block error rate) is needed to check in case of problem. BLER loop convergence algorithm provides the link adaptation in order to meet the target BLER. DL and UL link adaption are targeting to sustain certain BLER for the first transmission. iBLER (initial BLER) can be expressed as:
iBLER
f TBS , coding rate , modulation, SINR
BLER target denotes the initial BLER target that is used to determine MCS together with CQI. BLER target is defined by parameters dlTargetBler and ulTargetBler. Default value for both parameters is 10%. BLER target could be changed and optimized for different radio envi ronments. If the BLER is higher than 10%, the channel condition is poor and will result in low throughput. BLER is a metric that eNB controls within a configurable target (DL initial BLER target) to optimize the RF operation. Too‐high BLER performance beyond the target often indicates some aspect of RF problems. In live network, DL iBLER is below 10% when the average CQI is above 4. DL UE throughput is decreasing as DL iBLER increases, DL iBLER is decreasing as average CQI increases (i.e., RF improves), which is shown in Figure 4.26. The expected range of iBLER is between 5% and 15%. If the performance is significantly (e.g., more than double) poorer, it may be usually caused by several reasons: ●●
UE RF estimate accuracy issue with CQI report – for example, CQI reports are much inflated comparing with UE’s actual RF condition
167
LTE Optimization Engineering Handbook 80
70
Average of DL iBLER DL iBLER distribution
60
DL iBLER (%)
50
40
30
20
10
0 0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
CQI Index 16.00
80
14.63 14.00
DL UE Throughput (Mbps)
168
70
Average UE DL throughput UE DL throughput distribution
12.00
60
10.54 10.00
50 8.70 7.51
8.00
40 6.21
6.00
30
5.01 4.16 3.50
4.00
3.05
20 2.60
2.18
2.00
1.78
1.42
10 1.01
0.72 0.58 0.34 0.18 0.00 0.12 0.00
0.00 0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
0
100
DL iBLER (%)
Figure 4.26 DL iBLER versus CQI and DL UE throughput versus DL iBLER. ●●
●●
UE’s tx power or RF condition is poor such that the ACK/NAK it sends over UL is too weak for eNB to detect UE’s RF is very poor such that it misses frequently the DL grants forcing eNB to retransmit the packets. High BLER caused by inter‐cell interference
In this scenario, it have really good CINR value and really good RSRP value, but inter‐cell interference where in the reference signal (RS) from an adjacent sector interferes with the traf fic resource blocks in the serving sector can be seen. The signature to look for is high RSRP, high CINR, and very high BLER, greater than 10%. As a result of the link adaptation deployed
Capacity Optimization
Figure 4.27 BLER target issue example.
in LTE, the eNB would start reducing the MCS (modulation and coding scheme) given to the UE. DL residual BLER One of the possible issues that could lead to the decrease of throughput is related with the excess of residual BLER. In order to evaluate this possible issue, it is advised to check the DL residual BLER distribution that it is based on measurement of number of MAC PDUs that are not acknowledged by the UE after the maximum allowed number of transmissions. The DL residual MAC BLER is distributed by different intervals as it follows: ●● ●● ●● ●● ●●
DL BLER low (from 0% to 1%) DL BLER low‐medium (from 1% to 2%) DL BLER high‐low (from 2% to 5%) DL BLER high‐medium (from 5% to 10%) DL BLER high (above 10%)
With the intervals above, means that it must have the samples as much as possible on the low BLER classification. An example of incorrect setting of BLER target is shown in Figure 4.27. In the site A base station, the cell download rate is too low (20Mbps), MCS is also low, but RSRP and DL SINR is pretty good, and 16QAM ratio is high, while BLER is low. By preliminary analysis, the phenomenon is not caused by interference. It is found that DL BLER target value is configured at 1%, so that scheduler will decrease MCS to guarantee BLER target. For FTP service, it does not require such a high BLER requirements, and this will lead to not be able to use higher‐order MCS, which leads to low download rate. After modified DL BLER target from 1% to 10%, DL data rate can achieve above 35Mbps. 4.6.7 Impact of UE Velocity
The DL throughput is impacted by UE velocity. The network performance influence by UE velocity relative to eNB is shown in Figure 4.28, negative velocity means that UE is moving closer toward eNBs, and positive velocity means thet UE is moving farther away from eNBs. The UE velocity group X stands for group [X, X+10], UE mobility may help or harm RF condition (i.e., CQI), the UE experiences and thus DL UE throughput.
169
LTE Optimization Engineering Handbook DL UE Throughput, CQI, DL iBLER v.s UE Velocity Reative to eNB 600000
16.00 547349 14.00
500000 11.40
12.00 10.59
11.97 11.11 10.20
10.00 8.43
10.91
10.50
10.59
10.72
10.67
10.57
10.61
9.95
9.74
400000
8.80 300000
8.00
6.00 200000
Sample Count
DL UE Throughput(Mbps), CQI, DL iBLER (%)
170
4.00 100000 2.00 254
739
1268
2614
4317
8690
–80
–70
–60
–50
–40
–30
21891
29756
–20
–10
46431 12438
5890
3816
2207
1438
479
20
30
40
50
60
70
0
0.00 0
10
UE Velocity group (MPH) Average of UE_DL_Thp(Mbps)
Average of Avg_CQI_Reported(Overall)
Average of DL iBLER(%)
Count of UE_DL_Thp(Mbps)
Figure 4.28 Impact of UE velocity (relative to eNB).
From Figure 4.28, it can be seen that as UE velocity increases, average UE DL throughput and CQI decreases a bit, while DL iBLER increases. 4.6.8 Single User Throughput Optimization
This part is focused on the radio analysis to improve the DL single user throughput. Single‐user throughput optimization needs UE trace logs to spot the problem. Here we pre sent an example of trace parses. This example is a modified output of DL, which summarizes a number of traces onto one line, like transmission mode, MCS (modulation and coding scheme), PRB occupied, TBS (transport block size), assignable bits, HARQ, CQI/RI, DL BLER, and so on. Note how RI=2 is reported, then transmission changes from TxDiversity to MIMO. HARQ ACK/NACK refers to the transmission four subframes earlier.
sfn | sf |mode |dlModul | mcs1 |mcs2 280 | 4 |TxDi | 64QAM | 16 | 0 280 | 5 | | | | 280 | 6 |TxDi | 64QAM | 18 | 0 280 | 7 |TxDi | 64QAM | 18 | 0 280 | 8 |MIMO | 16QAM | 13 | 13 280 | 9 |MIMO | 16QAM | 13 | 13 281 | 0 |MIMO | 16QAM | 12 | 12 281 | 1 |MIMO | 16QAM | 13 | 13 281 | 2 |MIMO | 16QAM | 13 | 13 281 | 3 |MIMO | 16QAM | 13 | 13 281 | 4 |MIMO | 16QAM | 13 | 13 281 | 5 |TxDi | 16QAM | 30 | 0 281 | 6 |MIMO | 16QAM | 30 | 30 281 | 7 |MIMO | 16QAM | 30 | 30
|prb |Ndf |Tbs1 |Tbs2 |AssBits | Harq | dlBler | cqi |ri | 25 | Y |7736 | 0 |8771784 | | | 11 | 2 | | | | | |A | 0% | | | 25 | Y |7992 | 0 |8764088 | A | 0% | | | 25 | Y |7992 | 0 |8756144 | A | 0% | | | 25 |Y Y |5736 | 5736 |8748192 | A | 0% | | | 25 |Y Y |5736 | 5736 |8736760 | | | | | 25 |Y Y |4968 | 4968 |8737384 | A | 0% | | | 25 |Y Y |5736 | 5736 |8763568 | N | 0% | | | 25 |Y Y |5736 | 5736 |8776208 | N N | 2% | | | 25 |Y Y |5736 | 5736 |8800856 | N N | 4% | | | 25 |Y Y |5736 | 5736 |8825504 | N N | 6% | | | 25 | N |7992 | 0 |8862160 | N N | 8% | | | 25 |N N |5736 | 5736 |8862200 | N N | 10% | | | 25 |N N |5736 | 5736 |8862200 | A A | 10% | |
| | | | | | | | | | | | | | |
Capacity Optimization
4.6.8.1 Radio Analysis – Assignable Bits
If the UE is sending with high CQI (in the range 10 to 15) and RI=2 but throughput is still very low, then the next check should be assignable bits. Assignable bits means the amount of data in the DL buffer available for the scheduler to schedule for this UE. A classic symptom of low assignable bits is that the UE is scheduled with a high MCS but a low number of PRBs. This scheduler strategy maybe attempts to send with the highest possible MCS and least number of PRBs so that leftover PRBs could be assigned to another UE, or, another symptom is that the UE is not scheduled every TTI (and nothing else is available to schedule). Possible causes for low assignable bits include data received from core network is not enough to fill the RLC buffers in eNB and RLC status messages are not being received fast enough and RLC buffers are full. RLC status messages are sent between eNB and UE to inform about lost RLC packets. It needs to check that non‐TCP based traffic is not being sent with too large packet size. For iperf‐based traffic, the recommended MTU size is 1360 bytes (default is 1470). For RLC status messages not being received fast enough, it needs to check that RLC discards which will trigger TCP congestion control and lower throughput (discards on UDP traffic will not affect throughput). 4.6.8.2 Radio Analysis – CFI and Scheduling
SIBs require PDCCH resources, typically SIBs consume four or eight CCEs of PDCCH resources. If a UE is in good SINR conditions, the scheduler may allocate only one CCE for that UE. In that case, because of limited positions in PDCCH, it is quite likely that a PDCCH colli sion occurs especially in low system bandwidths. If a UE is in bad SINR conditions, the scheduler may allocate a large number of CCEs for that UE (two or four or eight CCEs) depending on the configured CFI there may only be common search space available or it may still collide with other PDCCH users when other DL users are scheduled. 4.6.8.3 Radio Analysis – HARQ
Each transport block transmission is represented as a HARQ process. Each HARQ process data is held in memory until NDI is toggled (NDI – new data indicator2 (physical layer bit tog gled for new data, i.e., new data is to be sent). This allows fast retransmission of erroneously received data. The scheduler’s representation of an HARQ process is as follows: feedback sta tus (ACK, NAK, DTX, Pending), TBS, MCS, and RV (redundancy version, HARQ has 4 redun dancy versions, rv0, rv2, rv3, rv1). Increasing the default number of transmissions means that RLC parameters also need to be modified and will require larger RLC buffers. The parameter tPollRetransmit also impacts the DL throughput; as the RLC send data by AM transmit mode, receiver will feedback ACK/ NACK to transmitter, and if time is out of the value of tPollRetransmit, the transmitter will retransmit related PDU, so the value of tPoll is important and the recommended value is tPollRetransmit=40 ms. If tPollRetransmit set too low, it will result to premature initiated retransmission rather than received ACK/NACK, this is due to a time out in advance. This will affect the normal transmis sion and reduce data rate. If tPollRetransmit set too high, it will lead to delay in the launch, can not quickly complete the normal transmission, resulting in a decline in data rate. Table 4.7 gives the field test of the responding UE PDCP throughput, RB Num/slot and percentage of lower than 2Mbps with different tPollRetransmit settings of signaling and data. 2 Do not confuse with newDataFlag, which is scheduler internal flag where 1 means new data and 0 means retransmission.
171
172
LTE Optimization Engineering Handbook
Table 4.7 UE throughput with different tPollRetransmit settings.
Test (ms)
RSRP
SINR
PDCP throughput
RB Num/slot
percentage of lower than 2Mbps
signaling 80, data 500
−79.2
18.5
33.5
94.811
1.14%
signaling 160, data 160
−78.5
19.3
29.4
74.122
3.97%
signaling 80, data 80
−78.3
18.5
27.0
86.974
8.00%
signaling 40, data 40
−79.5
18.3
31.9
93.672
1.71%
signaling 20, data 20
−78.4
19.3
30.6
93.577
0.54%
In case of rank 2 spatial multiplexing there are 16 HARQ process per UE instead of 8, but there are two processes that share the same ID; scheduler sees them as separate processes that are coupled to each other. 4.6.9 Avarage Cell Throughput Optimization
Average cell throughput is the average number of successfully transmitted data bits in one second per DL bandwidth of all active users in one data frame. In a live network, if the RSRP/SINR does not show any special issues, with normal maximum values and distribution according to the specific location, coverage of the cells and distribution of traffic in the cells, that means trying to increase RSRP close to the site and avoiding serving users far from the node. This fact will lead to an increase in the average throughput of the cells as well. The parameter of crsGain is another factor impacted throughput. The configuration of crsGain to values higher than 0 will increase the area of coverage, increasing the power allocated to reference signal, but decreasing power in other resources, therefore, the average throughput in the cell can be affected. Due to this fact, RS boost must be used only in specific cases, and it is not a recommended configuration to be implemented in the whole network. The other impact factors include bandwidth, control format indicator (CFI) format, UE category, MIMO transmission mode, and loading. For TDD LTE system, UL‐DL configuration and special subframe configuration also impact the average cell throughput. The DL through put calculation flow is shown in Figure 4.29. The target average DL throughput obtained for the LTE sectors of the capacity layer (high‐ frequency band) and coverage layer (low‐frequency band) is different. Unlike the coverage layer, the objective of capacity layer is to provide a throughput as high as possible. The mean frequency efficiency can be achieved to 2 bits per hertz in the capacity layer. 4.6.10 Cell Edge Throughput Optimization
Reasonable DL coverage threshold should also reflect the throughput requirements of DL and UL. In most cases LTE UL is always limited coverage due to the mobile terminal is power lim ited. The problem is how much Tx_power will be limited. It depends on the DL power and DL/ UL rate requirements. A typical case is that volume of DL traffic and data rate are always higher than the UL. Therefore, RF DL/UL imbalance is normal, and that will match the data rates. If the imbalance is very high, which means that the DL‐received power is either too high (may interfere with), or too low (because the DL power is insufficient, can not fully play the UL power of the potential). A very important fact is that the UL performance and coverage do not
Capacity Optimization BW
UL/DL config
Transmission mode
Determine PRB number of DL subframe MCS index
PRB number
PRB number
CFI
Determine total RE of DL sbframe Transmission mode Deduct common channel/signal overhead
Determine TB size UE category Physical bits number
Determine the physical bits number
N TBS valid? Y throughput
Figure 4.29 DL throughput calculation flow.
depend entirely on the RSRP and RSRQ. Therefore, higher DL transmit power may improve the DL cell average and cell‐edge data rate, but it is not able to improve the UL coverage. Generally speaking, the diversity of MIMO can improve the cell‐edge throughput peak. Therefore, in order to improve the UL coverage performance, through higher levels of diversity (using 4RX diversity), they can keep a low noise in LNA application. On the other hand, LTE uses universal frequency reuse (N=1) without soft hand off. Consequently, high levels of interference and low SINR can be expected near the cell‐edge. Traffic channel performance at the cell edge can also be enhanced via ICIC feature. 4.6.11 Some Issues of DL Throughput
Throughput issues can be anywhere in the network. A series of questions are needed to ask first: How many eNBs are affected? How many subscribers/UEs are affected? Was the test done using UDP or TCP? 4.6.11.1 Antenna Diversity not Balanced
During field testing, it is found RSRP and RSRQ display good (SINR is around 27 dB, RSRP is around ‐80dBm) but DL throughput can not achieve 35Mbps, which is shown in Figure 4.30. It is found that the imbalance of the two antenna diversity reception is the cause of the prob lem. RSRP of antenna0 and antenna1 display huge power gap, RANK2 SINR is low shown in Figure 4.30. Antenna 0’s RSRP is around ‐76dBm, antenna 1 is around ‐95dBm; the gap is more than 20 dB. Play back all the test data, and it can be seen that the antenna1’s RSRP continued at a low level. It is suspected that the antenna interface issues or problems with the antenna. After adjusting the eNB transmit antenna (antenna TX power gap, check RRU) and test again, the test antenna is found to be received normally, and throughput increased by about 40Mbps. 4.6.11.2 DL Grant is not Enough
From Figure 4.31, it can be seen that under good SINR (17.5 dB), DL throughput was only able to reach to around 1Mbps. The reason is too low number of DL grant allocation, the average is around 200 for a period of time.
173
174
LTE Optimization Engineering Handbook
Figure 4.30 Antenna diversity not balanced.
Figure 4.31 Low DL grant allocation.
The reasons of low number of DL grants are generally include: UE is in DRX state, the DL is not scheduled or rarely scheduled, PDCCH false alarms or missed detection, and lack of RB resources, and so on. In this case, DRX was off and PDCCH misdetection ratio is nor mal, it is found finally that there were other users occupied the PDSCH resources after OMC counter‐investigation. When locking to a single‐user test, the DL Grant can reach to 600, and DL throughput resumed normal.
Capacity Optimization
4.6.11.3 Unstable Rate
In this case, the cell radio coverage was good, RSRP was about −75dBm, the average SINR was about 25 dB; in addition, BLER and double MCS are normal, but the DL data rate is often below 10Mbps or no rate. The issue is DL user data rate was unstable, sometimes reduced to 10Mbps, even decreased to 0Mbps. The problematic cell coverage seemed to be well, and MCS was normal, but the PDCP data rate was not qualified. The investigation work was following: first, check if there is any equip ment (including UE, FTP Sever, etc.) alarm. Next, check the radio parameters configuration, like RRU power, bandwidth, time slot allocation, and so on, were right or not. If the rate was still so unstable after changed, someone still needs to check the cable between RRU and antenna and whether the line sequence was correct or not. Sometimes, mapping of the RRU port and antenna mouth (line sequence) does not meet the requirements during installation and causes an unstable data rate.
4.7 UL Data Rate Optimization UL throughput optimization is not a trivial task in since there are different features that affect the UL throughput. UL scheduler assignments will decide how many PRBs are allocated to each UE, UL adaptive modulation and coding will decide the MCS to be used by each UE every time it is granted UL resources, and adaptive transmission bandwidth will reduces the number of PRBs assigned to the UE in the UL based on the UEs power headroom reports in order to favor retainability of the call. The general troubleshooting strategy is described in the following along with different factors responsible for poor UL throughput. ●● ●● ●● ●● ●● ●●
High BLER (bad coverage) UL interference (high RSSI) Low power headroom Scheduling algorithm Low demand Other (VSWR, backhaul capacity) Analysis flow for UL throughput investigation:
●● ●● ●●
●● ●●
Alarm and parameter/feature check: make baseline audit for parameter and feature. RSSI: High UL RSSI would impact the UL throughput. Percentage of 16 QAM samples: Low usage of 16 QAM modulations scheme in UL would impact the UL throughput. PUCCH and PUSCH SINR: Poor UL_SINR conditions would impact UL throughput. Power limited UE: High number of power limited UE indicates poor UL coverage.
For UL, the above mentioned is the fundamental areas of analysis for UL. It will begin this sec tion with an overview of UL scheduling and link adaptation. BSR is the mechanism the UE uses to inform the eNB about the amount of data waiting in its RLC buffers. PHR is the mechanism the UE uses to inform the eNB about remaining power at the transmitter (or power limitations). The num ber of PRBs available for UL scheduling has some 3GPP specified limitations, which is different from DL. This means that, for example, the maximum number of PRBs for a single UE able to be scheduled in 5 MHz is 20 and not 23 (with two reserved for PUCCH). The areas of analysis for UL: ●● ●●
UL scheduling strategy. BSR (buffer status report), Values ranges from 0 up to >15000 bytes using 64 index values. for example, index 0 for BSR=0, index 1 for 0 < BSR 10 to < = 20
>20 to < = 30
cdf
>30 to < = 40
>40
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
Duration (ms) pdf 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
cdf
>30 to >50 to >70 to >90 to >110 >130 >150 >170 < = 50 < = 70 < = 90 < = 110 to to to < = 130 < = 150 < = 170
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
Duration (ms)
Figure 7.7 Contention‐free (left) and contention‐based (right) RACH access latency.
250
LTE Optimization Engineering Handbook UE NAS
eNB
LTE Uu
S1 - MME
MME
AS Attach Request RRC Connection Request Random Access Request / Response (MSG1 / MSG2)
UE Identification / Contention Resolution (MSG3 / MSG4)
RRC Connection Setup / RRC Connection Setup Complete
ueCapabilityEnquiry / ueCapabilityInformation
NAS Authentication Request / NAS Authentication Response
The attach time is the time between the sending of the Attach request NAS message from the UE (inside the RRC connection setup complete message) and the receipt at the MME of the Attach complete message from the UE
NAS Security Mode Command / NAS Security Mode Complete
securityModeCommand / securityModeComplete
RRC Connection Reconfiguration / Reconfiguration Complete
Attach Accept Attach Complete
Figure 7.8 Attach procedure.
After RRC connection, eNB forwards attach request to MME initiating the set up of the S1 control plane. This triggers authentication and key agreement procedure involving the HSS and the AUC. Security keys and vectors are generated and passed on to the MME. When UE performs the initial attach procedure, MME need to request for the UE’s IMSI to reconfirm UE’s identity and authentication. Then, MME initiates the establishment of the default date radio bearer as part of the default EPS bearer to finish attach procedure (Figure 7.8). Field result shows that under either stationary(good RF) or mobility, the average attach duration is around 220 to 240 ms, the minimum can even reach to 160 ms.
7.5 Paging Latency Optimization The time duration for paging was measure from paging message to RRC reconfiguration complete message. The latency is determinated by the paramtere DRX paging cycle and RF conditions. Paging procedure and latency analysis is shown in Figure 7.9. The average latency increase with degraded SNR due to fading conditions, the range of variation is usually several milliseconds. Prescheduled ping latency outperforms non‐prescheduled ones as expected, lab result also shows ping latency versus increased path loss in Figure 7.10.
7.6 Parameters Impacting Latency Latency is just one of the performance index. Relevant KPIs should be considered and balanced. Control plane latency involves the network attachment operation while user plane latency only considers the latency of packets while UE is in connected state. Some mechanisms
Latency Optimization
UE
MME
eNB Paging
Paging Random Access Procedure NAS: Service Request
S1 - AP: INITIAL UE MESSAGE(FFS) + NAS Service Request + eNB UE signalling connection ID S1 - AP: INITIAL CONTEXT SETUP REQUEST
RRC: Radio Bearer Setup (NAS Message)
RRC: Radio Bearer Setup Complete
+ (NAS message) + MME UE signalling connection ID + Security Context + UE Capacity Information (FFS) +Bearer Setup (Serving SAE - GWTEID, QoS profile) S1 - AP: INITIAL CONTEXT SETUP COMPLETE
+ eNB UE signalling connection ID + Bearer Setup Confirm (eNB TEID)
UE
eNB
MME Paging
69 ms
Delay for RACH Scheduling Period
Processing delay in UE
Paging RACH Premble
TA+ Scheduling Grant RRC Connection Request RRC Connection Setup
76 ms
Processing delay in eNB
168 ms Processing delay in eNB
Processing delay in UE 10, RRC Connection Setup Complete
Figure 7.9 Paging procedure and latency analysis.
including scheduling request periodicity, prescheduling, DRX and PRACH parameters, and so on, will impact network latency. Below setting can shorten the latency on air‐interface: ●●
●●
●●
Change SR periodicity from 10ms to 5ms (drawback: increase UE power consumption and network interference) Enable prescheduling (drawback: increase UE power consumption and network interference; waste of channel resources; decrease UE throughput when large amount of UEs in the network) Disable DRX (drawback: increase UE power consumption)
PRACH parameters impacting control plane latency (attachment operation) and user plane latency are given in Table 7.3. The inactivity timer will ensure that the UE is active even for busy traffic. If no data is send/ received, the UE goes to sleep and wakes up during the next DRX cycle. Figure 7.11 shows the effect of DRX on latency. The result from ping responses suffers the same effect, because they need to wait for the maximum long DRX cycle.
251
LTE Optimization Engineering Handbook Userplane Latency_32Bytes Non–prescheduled vs. pre–scheduled
35
Latency (ms)
30
32.4
31.4
30
25 20 15
17.2
15.6
15.4
10 5 0
A
B
A
high SINR A
B
A
medium SINR Non Pre-scheduled
B
B
low SINR Pre-scheduled
Ping latency vs pathloss, AWGN
120
100
Very poor coverage causing higher BLER
80 Ping latency (ms)
252
60
40
20
0 –50
–45
–40
–35
–30
–25
–20
–15
–10
–5
0
pathloss relative to maximum pathloss (dB)
Figure 7.10 Paging latency for different SNR locations.
Table 7.3 Parameters impacting control plane latency. Name
Recommended
Name
Recommended
preambleInitialReceived TargetPower
94dBm
macContentionResolutionTimer
Sf64
preambleTransMax
n8
maxHARQmsg3Tx
4
preambleTransmitPowerStepSize dB4 aUGtriggerDelayforRACHmsg4
5
maximumNumberOfDLTransmisions 4 RACHMessage4
Latency Optimization eNB VoIP Packets
VoIP Packet 1
VoIP Packet 2
VoIP Packet 1
Scheduled Packets
Transmission Delay
PRBs Matrix
UE DRX State Talk-Spurt detected
No available PRBs
On-Duration
DRX cycle
On-Duration
Three types of DRX parameters in live network Support both long and short - DRX drx-Config DRX-Config : setup onDuration Timer : psf2 drx-Inactivity Timer : psf100 drx-Retransmission Timer : psf2 longDRX-CycleStartOffset : sf40 sf40 : 1 shortDRX-Cycle : sf20 dxShortCycle Timer : 1
Figure 7.11 DL UE scheduling with DRX.
Only support long - DRX drx-Config DRX-Config : setup onDuration Timer : psf60 drx-Inactivity Timer : psf80 drx-Retransmission Timer : psf8 longDRX-CycleStartOffset : sf320 sf320 : 281 DRX not supported drx-Config DRX-Config : release
253
254
8 Mobility Optimization Mobility is the key procedure for ensuring that users can move freely within a network. The LTE mobility can be divided into “intra‐LTE mobility” and “inter‐LTE mobility” (inter‐working with 2G/3G and CDMA2000). It can be further divided into RRC_connected and RRC_idle mode mobility. Inter‐radio access technology handover (IRAT‐handover) is one type of “inter‐LTE mobility” in RRC_connected mode while cell reselection is referred to as RRC_idle mode mobility. The idle mode tasks can be divided into four processes: PLMN selection, cell selection and reselection, location registration, and manual CSG ID selection. The UE performs a PLMN selection when switched on, and for example, when a new PLMN is found at return from lack of coverage. Cell selection aims to find one suitable cell to camp on, which refers to UE in idle mode by monitoring the signal quality of neighbor cell and the current serving cell to choose the best cell service signal process. Cell reselection aims to camp on the best cell according to the evaluation criteria, which is performed continuously by the UE in RRC_idle. RRC_connected mobility (LTE handover) consists of four distinct phases: measurement configuration, measurement reporting, handover evaluation, and handover execution. Handover process in LTE is hard handover; it means that it has to break the wireless connection first and re‐establish the connection after it handover to a new cell, thus it will impact the user experience in network. It is generally assumed that downlink measurements, done by the UE, are used for the handover decision. Measurements performed by the source eNB may also assist the decision. The reporting can be done on an event triggered and/or periodic basis. In event‐triggered reporting, the UE measures, evaluates, and reports standardized events, which are used for triggering network actions, for example, handover. The events and parameters are controlled by the network and used by the UE, for example, offsets, hysteresis, averaging time, thresholds, and so on. The UE sends a report only when the network is interested in the information, that is, only when event criteria are fulfilled. As an example to fully understand the scope of each event and how the different handover are triggered, Figure 8.1 is provided. The UE will not perform measurements of neighbor cells when the serving cell RSRP is above Threshold1 (Th1). This saves UE battery life and avoids unnecessary handovers in good radio conditions. The eNB will command the UE to perform measurement gaps when one of the Th2 events is reported in a measurement report. Measurement gaps will be cancelled if the serving cell RSRP rises above Th2a. As a last resort, if no suitable intra‐frequency or inter‐frequency/ IRAT cells are found by the UE and the serving cell RSRP falls below Th4, an RRC connection release with re‐direction can be triggered by the eNB. Note that if Th2 values are set to a high value, this will increase the measurement gaps. As DRX cannot be used in parallel with measurement gaps and measurement gaps have higher LTE Optimization Engineering Handbook, First Edition. Xincheng Zhang. © 2018 John Wiley & Sons Singapore Pte. Ltd. Published 2018 by John Wiley & Sons Singapore Pte. Ltd.
Mobility Optimization
No measurements except serving cell
Intra–LTE + Inter Freq measurements
Intra–LTE measurements
Intra–LTE + Inter Freq + WCDMA + GSM measurements RRC Intra–LTE + Inter Connection Freq + WCDMA Release and measurements re-direction
RSRP (dBm) Th1
Th2a
Th2_InterFreq
Th2_WCDMA
Th2_GSM
Th4
Figure 8.1 Event thresholds configuration for different handover types.
priority than DRX the benefits of DRX (i.e., increasing battery time) wouldn’t be observed when Th2 values are set too high. Mobility for UEs is very important to ensure PS service continuity and reduce the interruption, latency, and PS drops. Tuning the handover thresholds/timers is needed to reduce unnecessary handover attempts toward overshooting cells or due to shadowing.
8.1 Mobility Management Mobility management refers to the process of establishing, maintaining, and releasing physical channel between E‐UTRAN and UE, which is shown in Figure 8.2. In the system of E‐UTRAN, according to the connection state of RRC, the mobility management is divided into two categories: connected state and idle state. The purpose of handover is to ensure that a UE in RRC_connected mode is served continuously when it moves. Mobility including intra‐frequency handover, inter‐frequency handover,
S1 based WCDMA/ GSM/ CS
CS Fallback
• Packet Mobility • Packet Handover
LTE
SRVCC
LTE
X2 based
WCDMA/GSM
Figure 8.2 Mobility management.
S2a based • Packet Mobility
CDMA 2000
• Packet Handover
S101/103 based
• Packet Mobility
S2a based
• Packet Mobility • Packet Handover
S3/S4 based
Gn based
• Packet Mobility • Packet Handover
• Packet Handover
255
256
LTE Optimization Engineering Handbook Mobility types From LTE LTE LTE LTE LTE
Mandatory SIB
Optional SIB
SIB1-SIB2-SIB3 SIB1-SIB2-SIB3 SIB5 SIB1-SIB2-SIB3 SIB6 SIB1-SIB2-SIB3 SIB7 SIB1-SIB2-SIB3 SIB8
SIB4
LTE-Intra-Freq
SIB4
LTE-Inter-Freq
To
WCDMA GERAN HRPD
Figure 8.3 SIB for handover.
inter‐RAT handover and handover between TDD and FDD LTE. From the mobility point of view, the mobility can be classified into intra‐LTE handover, coverage trigger session continuity, inter‐frequency load balancing, service‐triggered mobility, and subscriber‐triggered mobility. The parameters of mobility can be derived from SIB message as shown in Figure 8.3. Figure 8.4 shows the different states of the UE/MS in GSM, 3G, and LTE, also shown the transitions of these states while the UE/MS is in RRC_idle and RRC_connected mode. Actually, IRAT mobility from LTE is only supported through session continuity or release with redirect. When an IRAT mobility event is triggered while in LTE network, the eNB shall intiate a UE release with redirect information to reselect to 3G, 2G, or CDMA network. 8.1.1 RRC Connection Management
RRC connection management involves RRC connection establishment, RRC connection reconfiguration, RRC connection re‐establishment, and RRC connection release. ●●
●●
●●
●●
RRC connection establishment: This procedure is performed to establish an RRC connection. RRC connection establishment involves signaling radio bearer 1 (SRB1) establishment. The procedure is also used to transmit the initial NAS dedicated information or messages from the UE to the E‐UTRAN. RRC connection reconfiguration: This procedure is performed to modify an RRC connection, for example, to establish, modify, or release radio bearers, to perform handovers, and to configure or modify measurements. As a part of the procedure, NAS dedicated information may be transmitted from the E‐UTRAN to the UE. RRC connection re‐establishment: This procedure is performed to re‐establish an RRC connection after a handover failure or radio link failure. RRC connection re‐establishment involves the restoration of SRB1 operation and the re‐activation of security. A UE in RRC_ connected mode, for which security has been activated, may initiate the procedure in order to continue the RRC connection. The connection re‐establishment will succeed only if the cell has a valid UE context. RRC connection release: This procedure is performed to release an RRC connection. RRC connection release involves the release of the established radio bearers and the release of all radio resources.
8.1.2 Measurement and Handover Events
The UE performs radio measurements of its surrounding radio environment. The eNB controls the UE measurement in idle mode with the broadcast system information blocks and in
1xRTTCS Active
E-UTRA RRC_CONNECTED
Handover
Handover
HRPD Active
GSM_Connected CELL_DCH
Handover
E-UTRA RRC_CONNECTED
Handover GPRS Packet transfer mode
CELL_FACH Connection establishment/release
CELL_PCH URA_PCH
CCO with NACC Reselection
Connection establishment/release
1xRTT Dormant
1x
Reselection
E-UTRA RRC_IDLE
Reselection
LTE
Figure 8.4 LTE Inter RAT mobility procedures.
HRPD Idle
UTRA_Idle
Reselection
CCO, Reselection
Connection establishment/release
E-UTRA RRC_IDLE
Connection establishment/release
Reselection
GSM_Idle/GPRS Packet_Idle
CCO, Reselection
EVDO
3G
LTE
GSM
258
LTE Optimization Engineering Handbook Measurement object
Object ID
Object Measurement Report ID ID ID
Report Report config. ID ID
LTE carrier frequency 1
1
1
1
1
1
Event A1
LTE carrier frequency 2
2
2
2
2
2
Event A3
UMTS carrier frequency 1
3
3
3
3
3
Event B2
UMTS carrier frequency 2
4
4
4
3
4
Event B2
5
5
5
5
GERAN set of carrier frequencies
Figure 8.5 Measurement configuration.
connected mode with a RRC measurement control message. Two parameters are used to trigger measurements of the neighboring cells, idle mode uses sIntrasearch, and connected mode uses sMeasure. When UE is in connected mode, a UE measurement consists of two parts: measurement object and report configuration. This pair is referenced by a measurement ID. In connected mode, when the eNB provides the UE with a measurement configuration, it includes the following parameters (Figure 8.5): ●●
●●
●●
●●
●●
Measurement objects: The objects on which describes the radio technology and frequency to measure on. Reporting configurations: A list of reporting configurations where each reporting configuration consists of reporting criterion and reporting format. Reporting criteria triggers the UE to send a measurement report (e.g., RSRP). Reporting format quantities that the UE includes in the measurement report (e.g., RSRP and RSRQ). Measurement identities: A list of measurement identities where each measurement identity links one measurement object with one reporting configuration. Quantity configurations: One quantity configuration is configured for intra‐frequency measurements, and one per RAT type. Measurement gaps: Periods that the UE may use to perform measurements, that is, no (UL, DL) transmissions are scheduled.
Measurement procedures distinguish the following types of cells: the serving cell, cells listed within the measurement objects, and cells that are not listed within the measurement objects but are detected by the UE on the carrier frequencies indicated by the measurement objects. When the measurements triggering conditions are met, the UE initiates the measurement reporting procedure and sends a measurement report message to the eNB. The message includes the information related to the event previously configured by the eNB. From the measurement report sent by the UE, the eNB may take a handover decision. Up to now, LTE measurement items by eNB and UE are listed below: ●● ●●
UE measurements: CQI, RSRP, RSRQ, and inter‐RAT measurements for handover. eNB measurements: DL RS Tx power, received interference power, thermal noise power, TA, average RSSI, average SINR, UL CSI, detected PRACH preambles, transport channel BLER, and so on.
Mobility Optimization
RSRP
RSRP
A3 trigger
A5 trigger
RSRP
A2 trigger
A5 thold, y‘
A1 threshold A3 offset A2 threshold
A1 trigger
RSRP serving move direction
RSRP neigh
RSRP serving
A5 thold, x‘ RSRP neigh
move direction
RSRP serving move direction
Figure 8.6 Handover events.
3GPP defines in TS36.331 several handover triggers in LTE. The handover events A1, A2, A3, A4, A5, B1, and B2 currently supported by eNB are listed below (Figure 8.6): ●● ●● ●● ●● ●●
●●
●●
●●
Event A0: Periodical reporting Event A1: Serving becomes better than threshold Event A2: Serving becomes worse than threshold Event A3: Neighbor becomes offset better than serving Event A4: Neighbor cell becomes better than a threshold value. Event A4 is mainly used to configure UE to perform measurements on inter‐frequency carriers for offloading purpose Event A5: Serving becomes worse than threshold1 and neighbor becomes better than threshold2 Event B1: Inter‐RAT neighbor cell becomes better than a threshold value. Event B1 is use for CS fallback to UTRAN and GERAN Event B2: Serving becomes worse than threshold1 and inter RAT neighbor becomes better than threshold2
The parameters of hysteresis of intra_LTE mobility are concluded in Table 8.1 and Table 8.2. Measurement command message is included in RRC connection reconfiguration message. Measurement configuration is depicted in Figure 8.8.
Table 8.1 Use of hysteresis for various mobility events – intra‐frequency mobility. Mobility event
Type
Event triggered condition
Event cancelled condition
Start/stop of measurements of intra‐frequency cells
A1/A2
RSRPserv RSRPserv + a3Offset + hysA3Offset
RSRPneigh threshold3 OR RSRPneigh + hysThreshold3 a3OffsetRsrqInterFreq + hysA3OffsetRsrqInterFreq
RSRQneigh – RSRQserv < a3OffsetRsrqInterFreq – hysA3OffsetRsrqInterFreq
Coverage handover (inter‐frequency
A5
RSRPserv + hysThreshold3InterFreq < threshold3InterFreq AND RSRPneigh – hysThreshold3InterFreq > threshold3aInterFreq
RSRPserv – hysThreshold3InterFreq > threshold3InterFreq OR RSRPneigh + hysThreshold3InterFreq < threshold3aInterFreq
Stop of measurements A1 of inter‐frequency cells
RSRPserv > threshold2a + hysThreshold2a
Note: Event‐A2—serving cell becomes worse than a threshold—can be triggered by different thresholds. Event‐A2 can be used to trigger, for example, RRC connection release with redirection, for inter‐frequency measurements, or inter‐RAT measurements when UE encountering bad coverage can measure on several inter frequency and/or IRAT target frequencies. Mobility control at cell edge needs to define and prioritize the preferred frequencies for measurements and mobility actions and possibility to perform blind handover or release with redirect to that RAT at critical threshold. In LTE network design, there is a search zone introduced shown in Figure 8.7, at search threshold, start inter‐frequency and IRAT measurements.
Search zone
Search threshold
A2 -> Search activity
Search for other Freqs/RATs. If found, do HO or RwR. Based on A3/A5 + B2 or Blind RwR/HO
Figure 8.7 Search zone.
8.1.3 Handover Procedure
Intra‐LTE handover have two versions: S1‐based and X2‐based. The X2 handover is normally used for inter‐eNB handover and is preferred because of its packet‐forwarding feature to avoid data loss. If the X2 interface is not available or the source eNB is configured to use S1 based handover, then S1 handover is used. From the UE perspective, the two procedures are identical. In fact, the UE doesn’t even know which handover procedure is executed. The successful handover was determined by verifying the following message sequence as shown in Figure 8.9: ●● ●● ●●
Measurement report sent from UE to source cell. RRC connection reconfiguration message including MobilityControlInfo sent by a source cell to UE, and RRC connection reconfiguration complete sent from UE received by a target cell.
Mobility Optimization
8.1.3.1 X2 Handover
Based on UE measurement and RRM information, the source eNB decides that a handover to the target eNB is necessary. The source eNB sends a handover request message over the X2 interface to the target eNB. The message contains necessary information to prepare the handover at the target side. The target eNB allocates resources for the target cell and the UE is allocated a new C‐RNTI for identification in the target cell. The target eNB sends a handover request acknowledge message to the source eNB, which in turn sends handover command over the air interface to the UE, including necessary information (e.g., the new C‐RNTI) so that the UE can perform the handover. From statistics in a live network, about 30% of drop calls was due to X2 is failure (Figure 8.10). During the X2 handover procedure, the eNB performs (Figure 8.11): ●● ●●
●● ●● ●● ●● ●●
An X2AP handover preparation procedure An X2AP SN status transfer procedure if the PDCP SN status preservation applies for at least one of the (RLC‐AM) radio bearers handed over An RRC connection reconfiguration procedure for mobility within EUTRAN A X2 user‐plane data forwarding An X2AP UE context release procedure An S1AP path switch request procedure A UE context deletion in the source eNB (with associated resources)
The source eNB sends PDCP sequence number (SN) information to the target eNB in an SN status transfer message. This information is necessary to avoid missing or duplicating PDCP packets when the uplink and downlink user data paths are switched from the source eNB to the target eNB. Also, the source eNB now forwards the received downlink user data packets to the target eNB instead of sending them to the UE. The downlink user data packets are buffered in the target eNB until the handover is completed. As soon as the handover command (RRC connection reconfiguration) message is received, the UE buffers the uplink user data until the handover has been completed, detaches from the source cell, and synchronizes with the target cell using the non‐contention based random access procedure. Next, the UE sends a handover confirmation (RRC connection reconfigura tion complete) message to the target eNB to indicate that the handover procedure is completed as far as the UE is concerned. Now the UE can start sending the buffered uplink user data and the target eNB can forward the downlink user data to the UE. The uplink user data is sent via the target eNB directly to the serving gateway, since the uplink tunnel endpoint identifier (TEID1) in the S‐GW was conveyed to the target eNB already. The target eNB sends path switch message to MME to inform that the UE has changed cells, including its downlink tunnel endpoint identifier (TEID). MME forwards it to the S‐GW so that the S‐GW can send the downlink user data directly to the target eNB. Before the S‐GW can release any user plane resources toward the source eNB, it sends one or more “end marker” packets to the source eNB as an indication that the downlink data path has been switched. It should be noted that these packets do not contain any user data, and are transparently forwarded by the source eNB to the target eNB to help it decide when the last forwarded packet was received. After receiving an acknowledgment message, the target eNB informs the source eNB about the success of the handover. As a final step, the source eNB releases all air interface and control plane resources associated with the UE context, and the handover is completed. Figure 8.12 is an example from eNB trace, which filter out the UE related message.
1 Different S1 bearers are identified by their tunnel endpoint identifier (TEID), which is allocated by the endpoints (eNB and S-GW) of the GTP tunnel.
261
Measurement identity
Reference number in measurement report
Measurement objects
Reporting configurations Quantity configuration
A carrier frequency A list of neighboring cell offsets IRAT neighboring cells (no neighbor list for intra-LTE mobility) Reporting criteria: periodical or event-triggered Reporting format: quantities (e.g. number of cells to report) e.g. RSRP or RSRQ One quantity for intra freq, one for inter and one for each RAT type
Figure 8.8 Measurement configuration.
LTE Optimization Engineering Handbook
UE
eNB 1. RRC Connection Reconfiguration (Measurement Control)
Legend L3 signalling
Packet data
L1/L2 signalling UL allocation User data 2. Measurement Reports
THOoverall 4. Admission Control and Resource Allocation DL allocation
Handover Preparation
3. HO Decision
Handover Execution
5. RRC Connection Reconfiguration
Detach from old cell and synchronise to new cell 6. Synchronisation 7. UL allocation + TA for UE 8. RRC Connection Reconfiguration Complete
9. Release Resources Packet data
Handover Completion
264
Figure 8.9 Intra‐eNB handover.
Handover procedure works as expected, data is forwarded over X2, path is switched to target cell and average user plane interruption time was about 200ms but varied between 100ms and 500ms depending on the loading in the network. The source eNB issues X2 handover request message to the target eNB by passing necessary info to prepare the handover and gets the X2 handover request acknowledge message back. The detail content in the two messages is listed in Figure 8.13.
Mobility Optimization Source eNB
UE
Target eNB
Serving GW
MME
0. Area Restriction Provided 1. RRC Connection Reconfiguration (Measurement Configuration)
Packet data
Packet data UL allocation
Legend L3 signalling
2. Measurement Reports
Handover Preparation
L1/L2 signalling
3. HO Decision
Tx2ACLOCpre
User data
4. Handover Request 5. Admission Control and Resource Allocation
Tx2ACLOCoverall
Tx2ACLOCarea
6. Handover Request Acknowledge
DL allocation 7. RRC Connection Reconfiguration
Deliver buffered and in transit packets to target eNB
Handover Execution
Detach from old cell
8. SN Status Transfer Data Forwarding Buffer Packets from Source eNB 6. Synchronisation 10. UL allocation + TA for UE 11. RRC Connection Reconfiguration Complete
T 12a. RRC Connection Reconfiguration (Measurement Configuration) x2ACLOCarea/path/assignment Tx2ACLOCcomp
13. Path Switch Request
12b. RRC Connection Reconfiguration Complete
14. User Plane Update Request
Packet data 17. Path Switch Request Acknowledge TDATAtwdD2
TDATAtwdsd
16. User Plane Update Response
18. UE Context Release
Release S1, X2 signalling and radio resources. Continue forwarding packets
Release X2 Signalling Connection
Data Forwarding End Marker 19. Release Remaining Resources
Packet data
Figure 8.10 X2 handover procedure.
Figure 8.11 X2AP handover preparation.
Begin sending S1-U data Release Other X2 Resources
15. Switch DL Path
Packet data
Handover Completion
End Marker
265
266
LTE Optimization Engineering Handbook
Figure 8.12 X2 handover procedure from test tool. X2 HO Request Message HO Preparation Info Serving cell configuration -MIB/SIBs
UE radio configuration and capability -existing radio bearers (SRBs/DRBs)
UE security capabilites, Last visited cell lists, ... ERAB ID, ERAB level QoS parameters, Target cell ID Global MME indentity …… X2 Handover Request Acknowledge eNB-UE-X2AP-ID Admitted E-RAB list GTP TEID Target eNB to Source eNB Transparent Container A new C-RNTI Target eNB security algorithm identifiers A dedicated RACH preamble (optional) ……
Figure 8.13 X2 handover request message and handover request acknowledge message.
It is worth to pay attention to the X2 round‐trip delay, if it is large and the UE is moving at high speed, the UE could lose contact with the source eNB due to increasingly poor radio conditions resulting in failure to send handover commands to the UE and possible call drop. Different propagation conditions cause handover performance to be more or less sensitive to
Mobility Optimization
this delay. For the most delay‐sensitive applications, a one‐way delay over X2 of up to 50 to 70 ms is probably not noticeable, for example, for VoIP. 8.1.3.2 S1 Handover
Source eNB decides to initiate S1‐based handover to target eNB based on UE measurement report when there is no X2 connectivity to the target eNB, or by an error indication from the target eNB after an unsuccessful X2‐based handover. The measurement procedures are performed in the same way as for X2 handover. When S1 handover is executed, the source eNB initiates the handover preparation by sending the handover required massage to the serving MME. The source MME sends a forward relocation request to the target MME over the S10 interface. During the S1 handover procedure, the eNB performs: ●● ●●
●● ●● ●● ●●
An S1AP handover preparation procedure An S1AP SN status transfer procedure if the PDCP SN status preservation applies for at least one of the (RLC‐AM) radio bearers handed over An RRC connection reconfiguration procedure for mobility within EUTRAN A X2 user‐plane or S1 user‐plane data forwarding An S1AP handover notify procedure A UE context deletion in the source eNB
The eNB general procedure for inter‐eNB S1 handover is shown in Figure 8.14. The target MME sends a handover request to the target eNB. The target eNB may admit or reject this request with its admission control function. If it admits the request, the handover request is acknowledged. Then the forward relocation is acknowledged with a response and now the source MME sends the handover command to the source eNB, which forwards it to the UE. Now the UE can perform the random access procedure in the target cell and the handover confirm is sent to the target eNB. Handover notify is sent to target MME in order to indicate to the target MME that the handover has succeeded. The context in the old eNB is released. From an end‐to‐end view, S1 handover can be used to achieve the following: ●● ●● ●● ●●
S1 handover with MME and SGW relocation S1 handover with MME relocation (no SGW relocation) S1 handover with SGW relocation (no MME relocation) S1 handover without any EPC node relocation
8.1.3.3 Key point of X2/S1 Handover
Intra LTE handover phases can be divided into handover preparation phase, handover execution phase, and handover completion phase. The handover preparation phase is the phase when the source RAN request the target RAN to prepare to accommodate the UE. The target RAN performs admission control and reserves cell resources required for the UE and prepares the UE access and sends the needed access information in a container sent transparent through the source RAN to the UE. The handover execution phase starts with the source RAN command the UE to make a handover. The source RAN performs packet forwarding to target RAN during the execution phase. The phase ends when the UE access the target RAN and confirm the new connection in the target cell. In the handover completion phase the user data path to the SAE gateway is switched over to the target RAN. The handover complete phase starts with the target RAN notice the MME and SAE gateway that the handover has occurred. The MME/SAE gateway switches the bearer from source RAN to target RAN. Bearer between SAE gateway and the RAN is set up per eNB
267
UE RRC Connected
Source eNB
Target eNB
Source S-GW
Target S-GW
Source MME
Target MME
1. RRC Connection Reconfiguration (Measurement Conf) 2. RRC Measurement Report (Event A3) 3. HO Decision 4. S1 Handover Required (Source to target Transparent Container)
5. S10 Forward Relocation Request 6. S11 Create Session Req/Res 7. S1 Handover Request
8. Admission Control 9. S1 Handover Request Acknowledge 10. S10 Forward Relocation Response 11. S11 Create Bearer Req/Res Up Forwarding 12. S1 Handover Commond 13. RRC Connection Reconfiguration Handover Commond Regenerate Security Keys
14. MAC: Random Access Preamble 15. MAC: Random Access Response (UL Allocation+TA) 16. RRC Connection Reconfiguration Complete (Handover Confirm)
17. S1 Handover Notify
18. Data Transfer in Target 19. S10 Forward Relocation
RRC Connected
Figure 8.14 Inter eNB S1 handover with MME relocation.
20. S1 UE Context Release (Cause: Successful Handover)
Complete/ACK
Mobility Optimization
and not per cell. The target eNB informs the source eNB that the handover is complete. The source eNB finishes the packet forwarding (if used) and releases the UE resources. If the UE is moving to a cell in the same eNB then performs no switching of the bearers. Handover preparation by source eNB and target eNB are shown in Figure 8.15. Handover user plane interrupt time UL or DL interrupt time can be calculated from the instant that UE receives handover command to disconnect user plane to the old cell to the instant that user plane is re‐established. Interrupt time is different depending on a) DL or UL b) DL data forward option. With data forwarding, DL interrupt time is packet delay time, without data forwarding, DL interrupt time is packet loss time. The example in Figure 8.16 showed that user plane S1 handover interruption time is 48 ms. S1 handover failure analysis The symptoms in Figure 8.17 will usually lead to unsuccessful S1 handover, which will bring in low throughput and higher call drop rate in a live network. In Figure 8.17, MME does not reply to S1 handover request till tS1relocprep expired; in this case, we tried to prolong tS1relocprep, and will expect more core network rejection failures. S1 rejection with “unknown‐target ID,” it is suspected that the interworking between source MME and target was unsuccessful due to core network configuration fault or mapping fault. IF S1 rejection is with “ho‐failure‐in‐target‐EPC‐eNB‐or‐target‐system,” in this case the failure cause is quite clear: core network configuration needs to be checked to indicate the fault.
8.2 Mobility Parameter When the UE is powered on, before RRC, it says that the UE is in idle mode. In order to save battery power, UEs enter in idle mode where they are connected to EPC only (not to eUTRAN), that is, context of UE is deleted from eNB and is maintained at MME and SGW only. eNB releases all dedicated resources when UE enters idle mode. In this mode, UE shall only wake up at fixed intervals of time to check whether it has any incoming notifications for data, also UE can wake up at any time when it has any data to send. This mode is called RRC_idle and state is called EMM_registered state. EMM‐registered has two substates: idle (ECM‐idle + RRC‐ idle), and connected (ECM‐connected + RRC‐connected). The two EPS mobility management (EMM) states including EMM‐deregistered and EMM‐registered state, which describe the mobility management states that result from the mobility management procedures, for example, attach, detach, and tracking area update procedures. The two EPS connection management (ECM) states including ECM‐idle and ECM‐connected state, which describes the signaling connectivity between the UE and the EPC. Table 8.3 depicts RRC idle and RRC connected state. Idle mode mobility management is UE chooses a suitable cell to camp according to the system broadcast message sent by the eNB, in order to improve the success rate of UE access and quality of service. Connected mode mobility management is when UE moves in the connected state; the network offers the continuity of the service by handover for UE. UE moves from idle to connected by sending an initial NAS message. The initial NAS message in 3GPP Rel 8 can be attach request, detach request, tracking area update request, service request, or extended service request. When UE goes from RRC_connected to R RC_ idle, the bearers marked X are taken down that is shown in Figure 8.18.
269
X2 Handover
Intra-RBS Handover Intra-RBS Handover
X2 Handover
Receive X2 Handover Request message from Source RBS
S1 Handover
Get UE Information Get Cell Information
License exceeded? -number of connected users
X2 Handover of Intra-RBS?
Intra-RBS Handover
X2 Handover
S1 Handover: Place in Source To-Target Transparent S1AP container with target cell details
Yes
Handover Allowed? -UE Cap OK? -Resources Avail?
No
X2 Handover of Intra-RBS?
Send S1 Handover Required message
No
Yes
Yes Place in S1 Handover Required Message
Send X2 Handover Request message
Extract Source-ToTarget S1AP Container
No No
Place in X2 Handove Handover Request Message
Intra-RBS Handover
Receive S1 Handover Request message to MME
Get Handover Preparation Information -UE/Cell Information
Place UE/Cell Information in Handover Preparation Information RRC Container
Yes
S1 Handover
Send S1 Handover Failure to MME Send X2 Handover Preparation Failure to Source RBS
Build RRC Reconfiguration Message
X2 Handover Place in Handover Command of Intra-RBS? RRC Message
Yes
No
Create Target-To-Source eNB container Send S1 Handover Request Ack to MME Send X2 Handover Request Ack to Source RBS
Handover preparation by source eNB
Figure 8.15 Handover preparation.
Handover preparation by target eNB
Figure 8.16 User plane S1 handover interruption time.
tS1relocprep expired
unknown-target ID
Figure 8.17 Example of S1 handover failure. Table 8.3 RRC idle and RRC connected state. UE
RRC connected
RRC idle
Network
›Listens to the PDCCH for its assign cell RNTI
›Knows the UE on a cell level
›Has E‐RABs established
›Has a UE context in core nodes and an eNB.
›May send/receive data on the shared channels
›Controls mobility based on UE measurement reports.
›Listens to the PD‐CCH for a paging RNTI
›Knows the UE to within a tracking area.
›Performs random access and connection establishment procedure when it is paged or needs to send/receive data.
›Has a UE context only in core nodes.
›Controls mobility based on system information
›Needs to page the UE to send/ receive data.
Note: An EPS bearer carries traffic between the UE and PGW using an enhanced radio access bearer (E‐RAB) between the UE and SGW. An E‐RAB uniquely identifies the concatenation of an S1 bearer and the corresponding data radio bearer. When an E‐RAB exists, there is a one‐to‐one mapping between this E‐RAB and an EPS bearer of the NAS. The E‐RAB is can be thought of a PDP context of previous 3GPP releases.
272
LTE Optimization Engineering Handbook NAS Signaling Connection RRC Connection
S1 Connection
S11 Tunnel
S5 Tunnel
Signaling Radio Bearer (SRB) E-UTRAN Radio Access Bearer (E-RAB) Data Radio Bearer (DRB)
S1 Bearer
S5 Bearer
EPS Bearer
Figure 8.18 The bearers changed as RRC_connected to RRC_idle.
8.2.1 Attach and Dettach
UE needs to register with the network to receive services by the attach messages when a UE is first connecting to the network after being switched on. The UE enters the EMM‐registered state by a successful registration with an attach procedure. The RRC connection is established, with the NAS message attach request, transferred to the network as part of the procedure. The attach request message is provided to the MME in the “initial UE message.” The UE is identified either by the GUTI or the IMSI (in the attach request message). After registration, a context is established for the UE in the MME, and a default bearer is established between the UE and the PDN GW, thus enabling always‐on IP connectivity to the UE. The HSS provides the subscribed QoS (APN‐AMBR, default EPS bearer QCI) for the default APN to the MME. The MME selects and derives the addresses of the SGW and PGW and creates the default EPS bearer in the SGW by sending the message create session request. The MME provides the IMSI, APN name, subscribed QoS, PDN GW S5/S8 address, and the PDN type to the SGW. The default EPS bearer is established using the initial radio bearer establishment procedure. The TEID and the IP address for the UL of the user plane received from the SGW are used. The NAS message attach accept including activate default EPS bearer context request is transferred to the UE during the default radio bearer establishment. If the UE was identified by the GUTI in the attach request message the MME will re‐allocate the GUTI, that is, the new GUTI will be provided in the attach accept message. Finally, the NAS message “attach complete” (including the session management message activate default EPS bearer context accept) is transferred from the UE (Figure 8.19). The MME updates the SGW with the TEID and IP address for the DL of the user plane, received from the eNB by sending the message modify bearer request. There are four tunnels that are set up when a UE attaches to the network, two S1 tunnels, one S5 tunnel and one S11 tunnel. If a user has dedicated bearers each bearer will be a GTP tunnel. First the S11 and S5 tunnel (S5, S1, and S11 tunnels are GTPv2 tunnels) is set up, next is the S1 tunnel from the SGW to the eNB, and finally the S1 tunnel from the UE to the eNB. There is a tunnel setup between the PGW and the PDN, this tunnel is a GTPv1 tunnel. GTPv1 is unsecure and the data in the tunnel can be seen using a protocol analyzer. Data within a GTPv2 tunnel is secure and cannot be seen. Detach is the process of turning a UE off and disconnecting from the network, which is used to remove bearers and clear states in the network. The network or the UE explicitly requests detach and signal with each other. There are three types of detach: UE‐iInitiated detach procedure, MME‐initiated detach procedure, and HSS‐initiated detach procedure, which is shown in Figure 8.20.
Mobility Optimization UE
eNB
MME
RRC: RRC Connection setup complete S1AP: Initial UE message NAS: Attach Request NAS: PDN connectivity request NAS: Attach Request NAS: PDN connectivity request
SGW
PGW
HSS
the first message sent to the MME to establish a connection
Authentication and NAS security procedure S6a: Update Location request
S6A (MME to HSS) diameter signalling
S6a: Updated Location answer
S11: Create bearer request
S11: Create bearer response
S5: Create bearer request S5: Create bearer response
S1AP: Initial Context setup request RRC: RRCConnectionReconfiguration
NAS: Attach accept NAS: Activate Default EPS bearer request
NAS: Attach accept NAS: Activate Default EPS bearer request RRC: RRC ConReconfigComplete
S1AP: Initial Context setup response
RRC: UL information Transfer S1AP: Uplink NAS Transport
Default bearer establishment
The HSS accesses the database and responds with user information to the MME
This message from the MME is a three in one message
NAS: Attach complete NAS: Activate Default EPS bearer accept NAS: Attach complete NAS: Activate Default EPS bearer accept
Figure 8.19 Attach procedure with initial EPS bearer establishment (3GPP TS23.401).
The detach procedure allows the UE to inform the network that it does not want to access the EPS any longer or the network to inform the UE that it does not have access to the EPS any longer. The UE sends the NAS message “detach request” to the MME, using the UL NAS signaling transfer procedure. If the reason for the detach was not “power off,” the MME sends the NAS message “detach accept” to the UE, and releases the EPS bearers in the SGW by sending the message delete session request. The S5/S8 bearer is released using the S5/S8 bearer release procedure. 8.2.2 UE Measurement Criterion in Idle Mode and Cell Selection
When the UE is switched on, it performs the PLMN and cell selection procedure, which involves sweeping UE supported frequency bands, and searching the carrier frequencies with highest power for cells in a non‐forbidden PLMN. The selection of PLMN can be automatic or manual, and pre‐stored frequency and RAT information can be used to accelerate the procedure. The cell selection criteria is defined in 3GPP standards 36.3042 UE procedures in idle mode to determine which LTE cell to camp on. It is based on the measured reference signal received power (RSRP) level in the cell. Cell selection parameter is shown in Table 8.4. The cell selection criterion is fulfilled if:
Srxlev 0,Srxlev Qrxlevmeas
Qrxlevmin Qrxlevminoffset
Pcompensation
2 The cell camping criteria in Rel 9 is based on both level and quality while cell reselection to a cell with different absolute priority is based on either level or quality.
273
UE
eNodeB
MME
Serving GW
PDN GW
UE
PCRF
1. Uplink NAS Signalling Transfer NAS Message =“Detach Request”
eNodeB
MME
Serving GW
PDN GW
PCRF
1. Downlink NAS Signalling Transfer NAS Message =“Detach Request” 2. Delete Session Request
2. Downlink NAS Signalling Transfer NAS Message =“Detach Accept”
3. S5/S8 PDN Connectivity Release
3. Delete Session Request
4. Delete Session Response
5. Uplink NAS Signalling Transfer NAS Message =“Detach Accept”
4. S5/S8 PDN Connectivity Release 5. Delete Session Response
6. MME-initiated Connection Release
UE
6. MME-initiated Connection Release
eNodeB
MME
Serving GW
PDN GW
PCRF
HSS
1. Cancel-Location-Request IMSI, Cancellation Type = SUBSCRIPTION_WITHDRAWAL 2. Cancel-Location-Answer
3. MME-initiated Detach from ECM-CONNECTED state
Figure 8.20 UE‐initiated, MME‐initiated and HSS‐initiated detach from ECM‐connected state.
Mobility Optimization
Table 8.4 Cell selection parameter. function
Srxlev
Cell selection RX level value calculated by the UE (dB)
Qrxlevmeas
Measured cell RX level value (RSRP)
Qrxlevmin
Indicates the minimum required receive level used in intra‐frequency E‐UTRA cell reselection, which contains information relevant when evaluating if a UE is allowed to access a cell and defines the scheduling of other system information, default: −128dBm
Qrxlevminoffset
This parameter defines an offset to be applied in cell selection criteria by the UE when it is engaged in a periodic search for a higher‐priority PLMN. (dB). Increasing the value of this parameter would determine an earlier reselection of the target neighboring cell. Decreasing the value would determine a later reselection of the target neighboring cell. If you do not use inter‐PLMN mobility, this parameter is inhibited.
Pcompensation
Pcompensation is a compensation factor to penalize the low power mobiles. Max(PMax – UE Maximum Output Power, 0) (dB)
PMax
Indicates the maximum TX power of the UE in the cell. If this parameter is not included in the SIB3, the UE uses the maximum TX power
Inter Freq Target Cell
SourceCell CDMA Target Cell
CellSelectionReselectionConf:: qrxlevmin in SIB1 CellReselectionConfLte:: qrxlevmin in SIB3 CellReselectionConfLte:: qrxlevmin in SIB5 CellReselectionConfUtraFdd or Tdd:: qrxlevmin in SIB6 CellReselectionConfGERAN:: qrxlevmin in SIB7 Snonservingcell = –FLOOR (–20 x log10Ec/Io) in SIB8 (No qrxlevmin)
Intra Freq Target Cell
WCDMA Target Cell
GERAN Target Cell
Figure 8.21 Qrxlevmin configuration.
Qrxlevmin configures the cell minimum required RSRP level used by the UE in cell reselection as shown in Figure 8.21. Once the cell selection is accomplished, UE dominated reselection is about to happen, the following rules are used by the UE: ●●
●●
●●
Intra‐frequency criterion: If Srxlev ≤ sIntraSearch, the UE performs intra‐frequency measurements, if Srxlev > sIntraSearch, the UE does not perform these measurements (Figure 8.22) Inter‐frequency and/or inter‐RAT criterion: If Srxlev ≤ SNonIntraSearch, the UE performs inter ‐frequency and/or inter‐RAT measurements, if Srxlev > SNonIntraSearch, the UE does not perform these measurements sIntraSearch: Specifies the threshold (RSRP) for intra‐frequency measurements, i.e. how bad must the serving cell before the UE starts to measure on neighboring cells
Figure 8.23 gives a 4G measurements example. Measurement starts when Srxlev Rs) within a time interval (TreselectionEUTRA), as shown in Figure 8.25.
R _ s Qmeas , s Qhyst ;
R _ n Qmeas ,n Qoffset
where Qoffset is: qOffsetCellEUtran: Cell individual offset in the intra‐frequency and equal priority inter‐frequency cell ranking criteria. qOffsetFreq: Frequency specific offset in the equal priority inter‐frequency cell ranking criteria. Different cell reselection parameters settings will lead to different network performance. If sintrasearch sets to low, this means neighbor reselections are happening very late and can result in paging/RACH/RRC setup failures. This has been proposed that it should to be changed to 62, which means neighbors for reselection are always monitored. A drawback for this can be increased battery consumption for UEs. Figure 8.26 presents three sets of parameters configuration and their performance. The recommended setting can reduce the number of reselections with only moderate decrease in RF performance. In LTE, it is possible to use different settings for the cell reselection parameters tReselectionEutra and qHyst based on the estimated speed of the UE. The UE estimates its speed as the number of cell reselections per evaluation period. The speed is divided into normal, medium, and high
Mobility Optimization S (dB) 60 Signal quality on neighbor Cell B gets better than on serving Cell A but ...
50 40
Rs = Qmeas,s + QHyst Rn = Qmeas,n – Qoffset
Cell B
30
Sintrasearch
20
Srxlev,s (Cell A) Srxlev,n (Cell B)
10 0
Cell A 1 2
–10
3 4 5
6 7 8 9 10 11 12 13 14 15 16
Parameter values considered: • Qrxlevmin = –110 dBm
–20
• QHyst, Qoffset, Pcomp assumed 0
Time (sec)
... measurement on any neighbor will start when
Cell B selected after Rn > Rs satisfied for TReselectionEUTRAN
SservingCell < Sintresearch
• Rs = Qrxlev,s; Rn = Qrxlev,n • TReselectionEUTRAN: 3 seconds
Figure 8.25 Intra‐frequency cell reselection.
Parameter/Metric Qrxlevmin
Simulated Setting 1 –120 dBm
Original Simulated Setting 2
–120 dBm
Recommended Simulated Setting 3
–120 dBm
Sintrasearch
62 dB
62 dB
62 dB
QHyst
2 dB
4 dB
8 dB
0
0
0
TreselectionEUTRAN
1 sec
2 sec
4 sec
DRX Cycle
1.28 s
1.28 s
1.28 s
Qoffsets,n
Number of reselections
96
42
24
Percentage of time OOS
0%
0%
0%
Average RSRP (Camped Cell) Average RSRQ (Camped Cell)
–96.74 dBm –8.8 dB
–96.84 dBm
–97.82 dBm
–8.9 dB
–9.7 dBm
Figure 8.26 Different cell reselection parameters settings.
speed. At higher speeds, the tReselectionEutra timer and the qHyst values can be adjusted to trigger the cell reselection quicker and vice versa. 8.3.2 Inter‐Frequency Cell Reselection
LTE cell reselection uses priority‐based levels. Parameter sNonIntraSearch decides when UE starts searching and measuring all priority cells. Priority based cell reselection is configured so that LTE frequencies have either higher or lower priority than 2/3G serving cell priority.
279
280
LTE Optimization Engineering Handbook
In normal case, LTE frequency is configured to have higher priority, to help push capable UEs toward LTE. Cell reselection priority is the main criteria that determines UEs will camp on which layer. UE will perform call establishment on the layer it is camped on. Idle mode load distribution between layers helps reduce the need for the connected mode load balancing. Load‐ based adaptation of cell reselection thresholds can push cell‐edge UEs down to lower‐priority layer. Inter‐frequency cell reselection measurement trigger depends on the absolute priorities of the serving and non‐serving layers. For each neighboring frequency/layer, it is necessary to define the corresponding priority. UEs close to cell center will camp on high priority although RSRP is higher on the other frequency, UEs close to cell edge will camp according to priority. The idle mode distribution is possible to be adjusted with the thresholds. Inter‐frequency and inter‐RAT cell reselection are decided by higher‐priority cell reselection threshold ThreshXHigh and lower‐priority cell reselection threshold ThreshXLow. ThreshXLow: Threshold for the Srxlev value of the target cell for cell reselection toward a lower‐priority inter‐frequency or inter‐RAT frequency. UE will perform cell reselection toward a lower‐priority inter‐RAT when the Srxlev value of the serving cell is below threshServingLow and the Srxlev value of the target cell is above threshXLow for period of time. ThreshXHigh: Threshold for the Srxlev value of the target cell for cell reselection toward a higher‐priority inter‐frequency or inter‐RAT frequency. The larger of the two values, it is more difficult to select to the corresponding cell. Table 8.7 is for reference. Priority based inter‐frequency/inter‐RAT cell reselection: 1) Low priority to high priority transition is shown in Figure 8.27: Periodic search for higher‐ priority layer Qrxlevmeasneighbour qRxLevMinInterF interFrqThrH 2) High‐priority to low‐priority transition is shown in Figure 8.28: Search for a lower priority if the UE received level is below a threshold Qrxlevmeass
qRxLevMinIntraF thresholdSrvLow qRxLevMinOffsett P _ compensation AND Qrxlevmeasn qRxLevMinInterF interFreqThrL P _ compensation
Table 8.7 ThreshXHigh and ThreshXLow. Parameter
Recommend
GERANNFREQGROUP
ThreshXHigh
−11dBm
ThreshXLow
−11dBm
CDMA2000HRPDBANDCLASS
ThreshXHigh
−18dBm
ThreshXLow
−18dBm
UTRANFDDNFREQ
ThreshXHigh
0dB
ThreshXLow
0dB
UTRANTDDNFREQ
ThreshXHigh
0dB
ThreshXLow
0dB
E‐UTRANINTERNFREQ
ThreshXHigh
0dB
ThreshXLow
0dB
Mobility Optimization LTE Freq-A PRIO = 2 (highest) LTE Freq-B Serving cell PRIO = 1
PRIO = 1
Srxlev
Low prio Treselection
Threshx,high PRIO = 2
Cell Re-selection to Freq-A Search for higher prioritized layers at reguler intervals
Figure 8.27 Low‐priority to high‐priority transition. LTE Freq-B Serving cell PRIO = 2 LTE Freq-A PRIO = 1 (lowest)
Srxlev
Low prio
Threshx,low Treselection Threshserving,low
Search for lower/same priority layers starts
Cell Re-selection to Freq-A
tReselectionEutra/Utra
Higher Priority Cell
Srxlev A
D
Lower Priority Cell
Lower Priority Cell
(QrRxlevmin + sNonIntra Search) C
Threshx,low
B
Higher Priority Cell
Threshserving,low Srxlev = Measured RSRP + Qrxlevmin Srxlev > 0 is a must for suitable cell
Figure 8.28 High‐priority to low‐priority transition.
A: Start to measure inter‐frequency (RAT) target cell (Srxlev ≤ SNonIntraSearch) B: Srxlev ≤ Threshservinglow, can now attempt to connect to lower‐priority cell C: The lower‐priority target cell Srxlev > ThreshXLow is met, tReselectionEutra/Utra timer starts D: The lower‐priority target cell has been above threshold for tReselectionEutra/Utra, the UE will reselect to lower cell. 3) Equal priority cell transition is shown in Figure 8.29.
281
282
LTE Optimization Engineering Handbook Start to measure inter-frequency target cell (Srxlev≤ SNonIntraSearch)
Srxlev Serving Cell
tReslectionEutra A
C
(QrRxlevmin + sNonIntraSearch)
Target Cell
B
Target cell is above QHyst for time tReselectionEutra, UE will reselect to target cell.
Target cell is stronger than serving cell by QHyst.
QHyst
Figure 8.29 Equal priority transition.
8.3.3 Cell Reselection Parameters
The Qhyst is a hysteresis value of the serving cell used by the UE for ranking criteria in cell reselection, preventing too frequent reselection back and forth between cells of nearly equal rank. When a neighboring cell is ranked as better than the serving cell (that is, Rn > Rs) during a time interval tReselectionEutra (default value is 2s), the UE performs a cell reselection to the better‐ranked cell. It is worth to note that Qhyst should theoretically be set to a value at least equal to the handover hysteresis as cell reselection in idle mode is not as important as handover in connected mode. Qoffset is an offset in the cell ranking criterion of neighbor LTE cells, pertains to system information block 4 (SIB4) or neighbor cell in measurement configuration in RRC connected mode. It consists of a cell individual part and a frequency specific part. The frequency specific part applies to equal priority inter‐frequency cells only. If cell re‐selection is between the same priority, then,
For intra freq, Rankneighbor For inter freq,Rankneighbor
RSRPneighbor RSRPneighbor
qOffsetCell qOffCell qOffFrq
Increasing the value of Qoffset would determine a later reselection of the target neighboring cell (larger target cell list). Decreasing the value of Qoffset would determine an earlier reselection of the target neighboring cell (smaller target cell list) (Figure 8.30). One example of cell re‐selection to the same priority LTE cell between Micro and Macro (applicable if Micro is deployed with the same priority as Macro cell) in a live network, the re‐selection parameters can be set as: F1/F1: If there is a need to promote Micro, one must tune qHyst (direction Micro ‐ > Macro) or qOffsetCell (Macro ‐ > Micro). F1/F2: If there is a need to promote Micro, one must tune qHyst (direction Micro ‐ > Macro) or qOffFrq (Macro ‐ > Micro). qOffsetCell could be used to further differentiation between cells within Micro layer. The recommended re‐selection parameters are shown in Table 8.8.
Mobility Optimization
SI:
RS :U Em ll r es ele easu res cti on RS pa ram RP ete rs
ce
s
ea
: RS
m UE
Cell reselect ion? tReselectionE utra Rn > Rs?
UE performs cell reselection autonomously based on measurements Serving cell
P
SR
sR ure
Rs = QmeasS + QHyst Rn = QmeasN – Qoffset
Neighboring cell
RSRP sIntraSearch sNonIntraSearch qHyst(s)
Qmeas(n)
R(n) R(s)
qoffset(s)
Qmeas(s) tReselectionEutra time
Figure 8.30 Cell reselection parameters. Table 8.8 Re‐selection parameters. Parameter
Description
Recommend
q‐Hyst
This parameter configures the hysteresis value of the serving cell used by the UE for ranking criteria in cell reselection
2dB ~4 dB
Treselection
This parameter specifies the value of the cell reselection UE timer in the cell on the indicated EUTRAN frequency
1s ~2s
qOffsetCell
cell‐specific offset for reselection ‐ > to differentiate between particular cells within intra‐freq layer
0
qOffFrq
frequency‐specific offset for reselection ‐ > to differentiate between frequency layers
0
8.3.4 Inter‐Frequency Reselection Optimization
Idle mode inter‐frequency reselection from lower‐ to higher‐priority cell and from higher‐ to lower‐priority cell procedure are shown in Figure 8.31. Parameter sNonIntraSearch dictates when UE starts searching and measuring all priority cells. Higher‐priority carriers and RATs are searched even if UE is in good RF condition, lower
283
UE camp on serving cell N Exist inter-freq neighbor with higher reselection priority?
Y
Measurement performed by UE
t-reselectionEUTRA expires -> cell reselection to target cell
Y
StargetCell >threshXHigh?
N Y
SystemInfo SrxLev > SnonIntraSearch?
N
Reselection priority of Inter-fre neighor is lower than serving fre?
Measurement performed by UE
Y
SrxLev < threshServingL ow & Stargetcell > threshXLow?
N
N N
Figure 8.31 Idle mode inter‐frequency reselection strategy.
Fulfill S criteria & Rn > Rs?
Y
Y
Mobility Optimization
thresholds are set in a high‐priority carrier to keep the UEs in the carrier, higher thresholds are set in a lower‐priority LTE carrier to encourage the UE to re‐select higher‐priority carriers or another RAT. For reselection from higher priority toward lower‐priority inter‐frequency/IRAT cell, it should reduce ping‐pong by appropriately setting tReselectionRAT (in SIB5 or 6), threshServ ingLow (in SIB3), and threshXLow (in SIB5 or 6). For reselection from lower priority toward higher‐priority inter‐frequency/IRAT cell, it should reduce ping‐pong by appropriately setting tReselectionRAT (in SIB5 or 6) and threshXHigh (in SIB5). An example of the parameters from SIB 3 and 5 messages are shown in Figure 8.32: intra‐frequency (1900–1900) and inter‐frequency (1900–700), and intra‐frequency (700–700) and inter‐ frequency (700–1900). Key idle mode parameter settings for 1900 and 700 MHz band cells can be referred to Table 8.9. As shown in Table 8.9, a UE reselects from a B2 cell to a B17 cell if the serving B2 cell RSRP is −120 dBm. A UE reselects from a B17 cell to a B2 cell if the neighbor B2 cell RSRP is > −96 dBm.
8.4 Intra‐LTE Handover Optimization Intra‐LTE handover feature manages the UE in connected mode and allows for seamless mobility from one LTE cell to another. In contrast to idle mode, connected mode mobility is entirely managed by the LTE RAN based on configured and received measurement reports from the UE. Only hard handover is supported in LTE. Good handover performance will ensure the UE throughput and experience. By modifying the handover parameters can avoid or reduce the too early, too late and ping‐pong handovers, so as to improve the system performance. In a live network, either event (A3, A4, or A5) can be used in the LTE system with the intra and inter‐frequency handover decision. 8.4.1 A3 and A5 Handover
Event A3 means neighbor becomes offset better than serving cell, event A3 is used for c onnected mode handover. The formular for Event A3 triggered is shown below:
RSRP at serving cell a3Offset
RSRP at neighbour cell
The formular for Event A5 triggered is shown below:
RSRP at serving cell a5Threshold 1 AND RSRP at target a5Threshold 2
Offset, hysteresis, and timetotrigger values play major role in Intra handover. So if it considers offseta3 = 3dB, hysteresisa3 = 1dB, and timetotriggera3 = 640msec then it could say that neighboring sector has to be 4dB higher than serving cell for 640msec to make the UE to generate event a3 (Figure 8.33). The formular of event A3 entering condition: Mn Ofn Ocn Hys Ms Ofs Ocs Off The formular of event A5 entering condition: Ms hysteresis thresholdEutraRsrp and Mn offsetFreq hysteresis threshold 2 EutraRsrp ●● ●●
●●
Mn = measurement result of the neighboring cell [dBm] Ofn = MeasObjectEUTRA::offsetFreq, Ofn is the frequency specific offset of the frequency of the neighbor cell [dB] Ocn = cellIndividualOffset for neighboring cell [dB], can be used to fine tune the handover hysteresis on a cell‐to‐cell basis, which make the handover from cell1 to cell2 more difficult by decreasing the cellIndividualOffset of the cell2
285
Figure 8.32 SIB 3 and 5 messages.
Mobility Optimization
Table 8.9 Idle mode parameter settings for high‐priority cell and low‐priority cell. LTE Cell Selection & Reselection Parameters B2(1.9GHz)
cellReselectionPriority
B17(700MHz)
5
4
−106
−124
sNonintraSearch
6
4
threshServingLow
6
4
qRxLevMin
threshXHigh(B2 Neighbor)
‐
threshXLow(B17 Neighbor)
10
4
‐
threshXLow(UMTS)
30
30
threshXHigh(Femto)
14
14
RSRP
RSRP Neighbour Cell
RSRP
a5threshold2 Serving Cell RSRP
a3offset Serving Cell RSRP
a5Threshold1
RSRP Neighbour Cell time
time a3TimeToTrigger a3ReportInterval Meas. Report
a5TimeToTrigger a5Report Interval Meas. Report Meas. Report
Meas. Report
Equal power borders
Figure 8.33 Better cell handover concept for A3 and A5. ●● ●● ●●
●● ●●
Hys = reportConfigEUTRA::hysteresis [dB] Ms = measurement result of the serving cell [dBm] Ofs = MeasObjectEUTRA::offsetFreq, Ofs is the frequency specific offset of the serving frequency [dB] Ocs = cellIndividualOffset for serving cell [dB] Off = eventA3Offset, Off is the offset parameter for this event [dB]
Ocn, Ofn, Ocs and Ofs brings means for optimization of intra‐LTE handovers and/or realization of particular mobility management strategy. It introduces cell‐ and frequency‐specific offsets, which allow for differentiation of handover trigger criteria toward particular frequency layer and/or particular neighbor cells. These mobility offsets can be applied to intra‐ and inter‐frequency A3 and A5 events. Especially, the cellIndOffNeigh parameter may be of special interest in case macro and micro layers run with the same frequency carrier, while offsetFreqInter and offsetFreqIntra may be useful for micro‐macro inter‐frequency scenario. A3 event entry condition (intra‐frequency)
RSRPneigh
cellIndOffNeigh
RSRPserv
cellIndOffServ a3Offset hysA3Offset
A3 event entry condition (inter‐frequency) RSRPneigh cellIndOffNeigh offsetFreqInter RSRPserv a3OffsetRsrpInterFreq hysA3OffsetRsrpInterFreq
cellIndOffServ offsetFreqIntra
287
LTE Optimization Engineering Handbook
For handover procedure, two types of RRC connection reconfiguration message need to be investigated as shown in Figure 8.34. The first RRC connection reconfiguration message is sent by source RBS to UE indicating all information for target cell like RACH parameters, root sequence, reference power, PUSCH/PUCCH nominal values, CQI reporting interval, and handover type. If target cell doesn’t have resources, it won’t see this message and handover preparation failure is observed. The second RRC connection reconfiguration message after
EPC 10 no . A to w a ll be ta fte ar rg r er et su s eN cc are B es re sfu le l tr as an ed sit ion
288
MME
SGW 9.
Pa
th
8.
Pa
th
itc
Sw
h
h
Re
qu
es
t
es
tA
ck
no
wl
ed
Target eNB
Source eNB
4. 1. A3 (F RR or C M wa Co ea rd nn su in e re g ct m ta ion en rg R tR et e ep ce co or ll i nfi t nf gu ro ra m tio at n io n to UE )
Re
qu
itc
3. Resource available confirmation along with target cell specific information that UE may need to Sync
Sw
B N te e ge RA eNB r et pl Ta B et m to s- C arg o C T nc es n Sy oc ith io at to pr d w r u g ig in CH ce nf try A yn co E h R ts s e U g nR ge 5. rou io th UE ct e n 6. on C C R R 7.
Figure 8.34 Two types of RRC connection reconfiguration message.
ge
Mobility Optimization
Table 8.10 Intra LTE ‐ A3,A5 parameters. Parameter
Description
Range
Defaults
a3Offset
The offset value for eventA3
−30…30dB, step 0.5 dB
6(3dB)
hysteresisA3
Parameter for entering/leaving measurement report triggering condition.
1dB
a3TimeToTrigger The period of time that must be met for a5TimeToTrigger the UE to trigger a measurement report for Event A3 (A5), it depends much on the speed of the UE and the coverage scenarios.
0ms (0), 40ms(1), 64ms(2), 80ms (3), 100ms (4), 128ms(5), 160ms(6), 256ms (7), 320ms (8), 480ms (9), 512ms (10),.…, 5120ms (15)
320ms
a3ReportAmount Number of reports when periodical a5ReportAmount reporting is used. 0 means that reports are sent as long as the event is fulfilled.
1r (0), 2r (1), 4r (2), 8r (3), 16r (4), 32r (5), 64r (6), infinity (7)
Infinity (7)
a3ReportInterval The interval for event triggered a5ReportInterval periodical reporting
240ms 120ms(0), 240ms(1), 480ms(2), 640ms(3), 1024ms(4), 2048ms(5), 5120ms(6), 1.024s (7), 1min(8), … 60min(12)
cellIndividual OffsetEUtran
Offset value specific to the neighbor cell relationship. This parameter can be applied individually to each neighbor cell with load management purposes.
low value will delay the HO, and the higher the value allocated to a neighbor cell, the “more attractive” it will be.
‐
a5Threshold1
RSRP threshold1 used for triggering the EUTRA measurement report for Event A5
0…97 dB, step 1 dB −140 + X
‐
a5Threshold2
RSRP threshold2 used for triggering the EUTRA measurement report for Event A5
0…97 dB, step 1 dB
‐
Note: increasing TimeToTrigger and hysteresis might reduce ping pong effect and unnecessary handover but at the same time increasing too much might lead to handover drop.
handover is sent by target (new serving) cell to UE indicating all information for itself like carrier frequency, A3/A2 thresholds and Smeasure parameter values. Normally, there are three ways of optimizing handovers in LTE, via the modification of the parameters a3offset and hysteresisa3, by changing the parameter timetotriggereventa3 and via the modification of the parameter filtercoefficient for event a3. The configuration of A3 should be different in different areas, considering the factor of overlapping, distance of sites etc. For A3 parameters setting shown in Table 8.10, it is recommended to follow the optimization rules below. ●●
●●
a3offset should always be larger than hysteresisa3 if you want UE to handover to cells with an RSRP at least equal to the RSRP value of its serving cell, and ensuring a3offset > hysteresisa3 avoids ping‐pongs. The higher the value of a3offset + hysteresisa3 the more difficult for calls do handover to other cells. This is very useful where it has coverage holes (not a one to one deployment scenario on top of 3G cells). The smaller the value of a3offset + hysteresisa3 the faster release the calls to neighboring cells. This is useful in those scenarios where a large number of LTE cells exist in a given geographical area.
289
290
LTE Optimization Engineering Handbook
8.4.2 Data Forwarding
This feature of data forwarding works for DL data only and is used during intra‐LTE handover. User‐plane tunnels can be established between the source eNB and the target eNB (or S‐SGW and T‐SGW) during handover preparation, it helps to decreases the packet loss rate, improve end‐user experience in case of time sensitive applications due to data loss or TCP slow‐start characteristics. As soon as the source eNB receives the handover request acknowledge, or as soon as the transmission of the handover command is initiated in the downlink, data forwarding is initiated if necessary. The packet forwarding function shall forward all buffered and incoming PDCP SDUs from the source to the target RAN, that is, IP packets before ROHC and encryption, so ROHC context is not needed to be forwarded. Packet forwarding is only performed in downlink (Figure 8.35 and 8.36). During handover preparation phase, source eNB attempts to allocate forwarding transport bearers toward the target eNB. If this is successful, a forwarding user plane tunnel is established for each E‐RAB between the source and target eNB. During handover execution phase, all buffered and incoming data is forwarded from the source to target eNB. Once handover is complete, the target eNB sends a path switch message to the MME. The MME then sends a user plane update request message to the SGW, which switches the user‐plane path from source to target eNB. The source eNB continues to forward packets while packets are being received from the SGW and the source eNB buffer contains data. When the END MARKER packet is Figure 8.35 Data forwarding. S1 HO Packet Forwarding S-SGW
T-SGW
IP S-eNB
T-eNB X2 HO Packet Forwarding
UE
S-eNB
MME Handover Required
Handover Command
Handover Command eNB status transfer
T-eNB Handover Request
Handover Request ACK MME status transfer
Handover confirm Data transfer Handover notify
Figure 8.36 S1 data forwarding procedure.
Mobility Optimization
received in the target eNB or the forwarding ordering timer expires, the target eNB discards any further forwarding data packets. Forwarding tunnels are then released. 8.4.3 Intra‐Frequency Handover Optimization
In LTE, the earliest a handover can be set up after security mode completed during an attach or idle to active setup. Handover optimization can be done by parameters adjustment, although each of the key handover parameters (hysteresis, eventA3offset, filtercoefficientRSRP/RSRQ, timeToTrigger) have a unique purpose, adjusting them may have many common effects which are illustrated in Table 8.11. Additional handover‐related parameters includes cellIndividualOffset (allows altering effective handover hysteresis in a certain direction for a given serving cell ‐ > neighboring cell pair), maxReportCells (max number of cells reported by UE on the MR), reportingAmount/report ingInterval and s‐Measure (UE will look for neighboring cells only if the serving cell RSRP falls below this threshold). Handover optimization needs to decrease drop call during and after handover procedure. The usual phenomenon is UE cannot successfully access the PRACH in the target cell or UE drops soon after sending RRC connection reconfig complete in target cell. ●●
●●
●● ●●
●●
Symptom: T304 expires. In this case timer T304 expires and UE makes RRC connection re‐ establishment with cause “handoverFailure.” (T304 is started after receiving handover command in source cell and stopped in target cell after successful PRACH procedure) Symptom: UE log indicates RRC connection reconfiguration complete (“PRACH msg3”) is transmitted but no further DL messages from target cell are seen and call drops Solution: increase T304 timer to 1000ms or longer, parameter t304IntraLte Solution: increase the max number of UL HARQ retransmissions, parameter harqMaxTrUl. (Parameter harqMaxMsg3 is for contention‐based RA with non‐dedicated preamble) Solution: increase initial preamble target receive power, parameter ulpcIniPrePwr (Value of −90dBm decreased handover drops considerably in a network)
Handover optimization also needs to decrease drop call before handover procedure. The usual phenomenon is handover is triggered too late, UE drops before handover command, sometimes ping‐pong handovers increase if A3 used for bad RF. ●●
Possible fix: If using A3 handover reduce a3offset and time to trigger. In some cases also filterCoefficientRSRP may need to be reduced.
Table 8.11 Handover (HO) parameters optimization. Smaller ●●
●● ●● ●●
●●
Faster HO trigger – can be useful in fast‐rising cell situations Increased HO ping‐ponging More frequent throughput interruptions Increased likelihood for UE served by a stronger cell. Hence, can improve HO success rate / throughput / BLER perf but offset by impact from more frequent HO interruptions / HO to momentarily strong cells More suitable for fast moving UE environments, and higher RF loading on the network
Larger ●● ●● ●● ●●
●●
Slower HO rate Reduced HO ping‐ponging Fewer throughput interruptions Increased likelihood for UE staying on a weaker cell. Reduced HO rate can impact HO success rate / throughput / BLER, but offset by fewer interruptions / HO to more stable cells. More suitable for predominantly stationary usage / slow moving UE environments usage / lighter RF loading
291
292
LTE Optimization Engineering Handbook –70 1
106 211 316 421 526 631 736 841 946 1051 1156 1261 1366 1471 1576 1681 1786 1891
–80
–90
–100
–110 RSRP_instant
–120
RSRP_FC(K = 4) RSRP_FC(K = 11)
–130
Figure 8.37 filterCoefficientRSRP simulation analysis. ●●
One should consider using A5 triggered coverage‐based handover to quickly trigger handover in bad RF. This way, there is no need to reduce A3 trigger time, and less ping‐pong results. In this handover strategy, A3 is the handover for normal usage and A5 is for the bad‐RF quick reaction handover.
If handover is triggered too early, the target cell SINR can be too weak when handover occurs. If handover is triggered too late, the source cell SINR can be too low. This can result in an abnormal release before handover. Besides, optimization of hysteresis and timeToTrigger should be performed in conjunction with optimization of parameter filterCoefficientRSRP. Low value of this parameter might allow ping‐pong handover operation. High value of this parameter might delay the handover and possible lead to lost connection to the serving cell. filterCoefficientRSRP simulation analysis is shown in Figure 8.37. It is recommended filterCoefficientRSRP value = “fc8.” Finding the optimum pair of (filter CoefficientRSRP, hysteresis, and timeToTrigger) should be considered to one of the following {(fc6,2,80), (fc8,3,40), (fc8,4,20), (fc5,1,100)}, in current cell and neighbor cell. It was proven that the time to trigger below 100 ms have the same impact on performance as UE measures RSRP every 100ms. 8.4.4 Inter‐Frequency Handover Optimization
The eNB must include consideration of inter‐frequency neighbors in all measurement, handover, and redirection algorithms. There are two inter‐frequency mobility preparation methods: blind‐based handover and measurement‐based handover. This includes support of measurement objects (MeasObjectEUTRA) in the RRC connection reconfiguration message that include eutra‐carrierInfo and measurement bandwidth values differing from those of the current cell. Gaps must be supported for inter‐frequency measurements to allow the UE sufficient time to retune its radio the frequency being measured. Also the eNB must support the
Mobility Optimization
IdleModeMobilityControlInfo (interFreqPriorityList option) in the RRC connection release message, which is used to direct UE camping behavior (Figure 8.38). For inter‐frequency handover, the system supports bad coverage method based on RSRP and RSRQ with below triggers: based on bad coverage, based on interference, based on load, based on service, and based on subscription (SPID3 information from MME). The example of inter‐ frequency handover in Figure 8.39 shows that when UE in layer F1, mobility control is based on coverage, when UE in layer F2, mobility control is based on interference (load). Usually two types of mobility triggers are used to move UEs between layers for inter‐frequency handover control strategy, coverage‐based mobility, and load‐based mobility, used in when losing serving cell coverage and when serving cell load reaches a configurable threshold. The use case is for a cell on a low frequency with coverage relation to a cell at higher frequency. In such case it is possible to handover UEs to the higher frequency layer when the UE is close enough to eNB. For layer F1(higher frequency), when event A2 (RSRP x3%, moved all to active.
Active LB
UEs to be off-loaded due to started A4 measurement.
Figure 8.43 Load‐balancing strategy. Table 8.16 Gap pattern configurations supported by the UE.
Gap pattern Id
Measurement gap length (MGL, ms)
Measurement gap repetition period (MGRP, ms)
Minimum available time for inter‐ frequency and inter‐RAT measurements during 480ms period (Tinter1, ms)
0
6
40
60
1
6
80
30
LTE measurement gaps are specified by 3GPP to have 6ms4 gap (see Table 8.16). The periodicity of these gaps is specified in periods of 10ms (1 frame). Larger periodicity of the gaps leads to better UL/DL throughput performance at the cost of inter‐frequency/IRAT monitoring, which could lead to longer required times to perform the handover. By default, Gap 1 is assigned for ANR while Gap 0 is assigned for mobility as shown in Table 8.17. After measurement gap is configured, the UE searches for cells in the target LTE frequency, which will take several measurement gaps since all 504 PCIs need to be scanned. If there are several LTE frequency layers, the search and measurement time is multiplied accordingly since the UE can only tune its RF receiver to one frequency during a measurement gap. The measurement procedure is similar if the target frequency is UTRA. However, in that case, the number of gaps required for cell search depends on the number of neighbor cells, scrambling codes that are not provided in the neighbor list need not be measured by the UE. For GSM target frequency, the UE is required to decode the BSIC on the 8 strongest measured BCCH frequencies. 4 6ms is enough to be able to search for primary and secondary synchronization channels of inter-frequency cells, even for the worst case frame timing where the gap starts at the beginning of the second or seventh subframe of the measured cell. This is because PSS/SSS is transmitted at 5ms interval at the end of the first and sixth subframes.
301
302
LTE Optimization Engineering Handbook
Table 8.17 Gap 1 and Gap 0. Gap pattern
Feature
Event Type
Gap 1
IF ANR
Inter‐freq A3/A5
Gap 1
IRAT ANR
ReportStrongestForSON
Gap 0
InterFrequencySessionContinuity
A2/Inter‐freq A3/A5
InterFrequencyLTEHandover Gap 0
WcdmaSessionContinuity
B2
Gap 0
SRVCCtoUTRAN
B2
Gap 0
InterFrequencyLoadBalancing
A4
Gap 0
InterRatOffloadToUtran
B1
Table 8.18 Maximum allowed time to identify a detectable cell with measurement gaps, per layer.
Max time to identify a detectable E‐UTRA FDD inter‐frequency cell
Max time to identify a detectable UTRA FDD cell
Max time to decode BSIC of a GSM cell (no other layers measured)
Long DRX cycle length
40ms gap
80ms gap
40ms gap
80ms gap
40ms gap
80ms gap
no C‐DRX
3.84 sec
7.68 sec
2.4 sec
4.8 sec
2.16 sec
5.28 sec
40ms 64ms
2.56 sec
30 sec
80ms
3.2 sec
‐
128ms
3.2 sec
‐
256ms
5.12 sec
320ms
6.4 sec
9.6 sec
3.2 sec
5.12 sec
‐
6.4 sec
6.4 sec
‐
If a cell is detectable and eligible to trigger a measurement report, the maximum time for UE to send the measurement report is called the cell identification time. The maximum allowed identification time depends on the C‐DRX long cycle and the target system, which are listed in Table 8.18. Actually, the UE vendor is free to implement faster search and measurement than required by 3GPP, but network parameters will in most cases need to be optimized for the worst performing fraction of UE models in the network. For this reason, C‐DRX can be disabled for the duration of urgent inter‐frequency measurements to reduce the probability of call drop due to measurement delay. If RRC connection containing inter‐frequency/IRAT ANR comes first, then GP1 (reconfiguration message default value) will be used. If Inter‐frequency/IRAT mobility measurement for handover comes later, then measurement gap will be updated as GP0 (default value). If inter‐frequency/ IRAT mobility and ANR are in the same RRC connection reconfiguration message, GP0 will be used, which is shown in Figure 8.44. An illustrated subframe scheduling in Figure 8.45 has been to show the measurement gap, all DL transmissions granted after measurement start‐4 will not be ACK/NACKed, due to the DL
Mobility Optimization
Figure 8.44 Gap parameters from RRC connection reconfiguration message. No DL scheduling
Frame Subframe
Measurement gap
N–1
Not immediately transmit after gap
N
N+1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7
PDCCH/PDSCH 0 1 2 3 4 5 PUSCH/PUCCH 0 1 2 3 4 5 6 7 8 9
8 9 0 1 2 3 4 5 6 7 8 9
6 7 8 9 0 1 2 3 4 5 6 7 8 9 7 8 9 0 1 2 3 4 5 6 7 8 9
PUSCH/PUCCH is transmitted.
Figure 8.45 Scheduling strategy during measurement gap.
scheduling will stop at measurement start‐4. PUSCH/PUCCH can still be scheduled in the four subframes before measurement gap to send CQI/PMI/RI report, and so on, in order to keep monitoring the radio. Measurement gaps are initiated by eNB with RRC connection reconfiguration message when Event A2 is triggered. The eNB deactivates measurement gaps once a suitable target cell has been reported by the UE or an event A1 is triggered by the UE. When measurement gaps are activated, the eNB informs the UE of the gap pattern. Note that DRX and measurement gaps are not used in parallel. Measurement gaps have higher priority than DRX. The gap pattern is configured by parameter measurementGapsPattern. Note, however, that when Gap is activated on CSFB trigger, the period is hardcoded to 40 ms regardless of the setting of parameter measurementGapsPattern. The starting position of the measurement gap in a given gap period is defined by the UE‐specific Measurement Gap Offset (MGO). It is determined so that the UE performance is degraded as little as possible by measurement gaps. This is achieved by choosing the MGO so as to minimize the collision of measurement gaps with the other periodic transmissions listed below: CQI/ PMI/RI reporting and SRS transmissions, ACK/NACK transmissions on PHICH and PUCCH (except the last ACK/NACK, that is, when the max number of retransmissions has been reached), QCI1 transmissions and SIBs transmissions and so on.
303
304
LTE Optimization Engineering Handbook
CQI Report
0
Measurement gap
40
DRX on duration
CQI Report
80 CQI Measurement (8 ms before CQI reporting)
CQI Report
120
Figure 8.46 CQI report alligned with DRX cycle and measurement gap.
Note that the MGOs of the different UEs are also chosen so that their different measurement gaps are distributed over time so that during the measurement gap of some UE, a sufficient number of other UEs can transmit (receive) and the total cell throughput performance will not degraded. The MGO is in the set {0, 1,…, 39} in gap pattern 0 and in the set {0, 1, …, 79} in gap pattern 1. 8.6.2 Measurement Gap Versus Period of CQI Report and DRX
Measurement gap needs to consider periodic CQI report and DRX period to not miss CQI reports from the UE sent while the UE is active (on duration). When a measurement gap is started, it needs to determine where the measurement gap should start by presenting a MGO for the UE. The UEs CQI periodic reporting is also configured with an interval and an offset in relation to subframe starts. Measurement gap has to be aligned (equal to or multiples of CQI and DRX cycle) with CQI report configuration to not lose the CQI reports from the UE. Currently DRX is aligned with CQI so that CQI reports are sent while the UE is active (on duration). Measurement gap will be placed before the UEs CQI measurement, to avoid a collision with CQI measurement and reporting and to allow for extended DRX active right after the DRX on period. Figure 8.46 is an illustration of measurement gap period, CQI measurement, CQI report, and DRX cycle. Here it assumes all the cycles are 40ms. 8.6.3 Impact of Throughput on Measurement Gap
During a measurement gap the UE cannot be scheduled in uplink or downlink, and therefore, measurement gaps induce throughput loss for the UE, there is no loss of cell throughput, however. From lab trial results, measurement gap with GP0 (40ms) has 25% throughput decrease compared with baseline. Measurement gap with GP1 (80ms) shows 12.8% increase compared with GP0. So up to 15% and 25% DL throughput reductions have been observed for a UE, which is being scheduled with 80 ms and 40 ms measurement gaps respectively. Reducing the amount of time a connected mode UE is required to make inter‐frequency or inter‐RAT measurements can reduce this throughput impact. From inter‐frequency (2.6G@20M 1.8G@10M) handover parameter optimization field test, Method1 is basic, Method2 is aggressive, Method3 is Method2‐based plus quick Event A5 handover. Three sets of parameters tested are listed below: ●● ●●
●●
Method1: the gap is triggered at −100 dBm, A3 = 3dB Method2: the gap is triggered at 1.8G to −110 dBm, A3 (2.6G →1.8G) =8dB, A3 (1.8G → 2.6G) = −12 dB Method3: method2 based settings plus A5 (2.6G → 1.8G) = −116dBm
Mobility Optimization Empirical CDF
1 0.9
basic
0.8 aggressive
0.7
cdf
0.6 basic aggressive aggressive + quick a5
0.5 0.4 aggressive + quick a5
0.3 0.2 0.1 0
0
10
20
30 40 50 60 MAC DL throughput, Mbps
70
80
90
Figure 8.47 Different handover parameters impact DL throughput.
From Figure 8.47 it can be seen method2 settings result in UE staying too long (no 2.6G layer in bad coverage). For method3 settings, A5 trigger used to push UE quickly to 1.8G in bad coverage. In this case, Method2 and Method3 are recommended.
8.7 Indoor and Outdoor Mobility Indoor hotspots are assumed to be covered by indoor distributed system as well as small cells for higher efficiency. Indoor wave propagation losses are frequency dependent. Indoor and outdoor mobility optimization evolves priority‐based cell reselection, A2 + A5, A2 + A4, A2 + A3‐based handover, and load balance and indoor leakage optimization items. In case of indoor deployment on a dedicated carrier, the indoor leakage outside the building will be controlled as much as possible. To ensure service continuity between outdoor macro‐ cell network and indoor network, a good overlap must be guaranteed to execute a handover indoor/outdoor. The indoor to outdoor cell overlap is calculated from the scanner measurement. It consists in calculating the RSRP delta between Top1 RSRP of indoor cell and RSRP of outdoor macro cell when Top1 RSRP > = RSRP target + indoor outdoor overlapping margin. Usually, indoor cell will deploy on a dedicated carrier, indoor cells except those covering the first floor must have no F1 outdoor macro cells neighbors. The F1 outdoor macro cell must have as neighbors the indoor cell covering the first floor. Cells covering the building exits located essentially at the first floor, which is called a transition cell (Figure 8.48). For cell reselection parameters setting, it is similar as outdoor cell (with same or different frequency/priority) reselection (Table 8.19). For handover parameters setting, there are usually three configurations as shown in Table 8.20. ●●
●● ●●
configuration 1: A2 + A5 for both direction between indoor and outdoor, which can keep the UE indoors as long as possible, and meanwhile, any signal leaks can be controlled effectively configuration 2: A2 + A5 is used for indoor to outdoor, A2 + A4is used for outdoor to indoor configuration 3: A2 + A3 for both direction between indoor and outdoor
305
306
LTE Optimization Engineering Handbook
F2 3rd
F2
floor
Indoor pico cell
F2
F2
F2
F2
Outdoor macro cell
2nd floor
F1
1st floor
Figure 8.48 Transition cell.
Table 8.19 Reselection parameters setting between indoor and outdoor (example).
1 2 3
Outdoor cell
Indoor cell
Outdoor cell priority = 7, indoor cell = 5
Outdoor cell priority = 5, indoor cell = 7
RSRP_out −98dBm
RSRP_out > −100 dBm, RSRP_in 4dB
RSRP_out‐RSRP_in > 4dB
Outdoor cell priority = 5, indoor cell = 7
Outdoor cell priority = 5, indoor cell = 7
RSRP_in > −98dBm
RSRP_out > −100 dBm, RSRP_in −102dBm, RSRP_in −98dBm
RSRP_out > −102dBm, RSRP_in 3dB
RSRP_out‐RSRP_in > 3dB
8.8 Inter‐RAT Mobility The part is about the strategy of mobility in RRC idle mode (i.e., inter‐RAT cell reseletction) and in RRC connected mode (redirection, PS HO, CSFB or CCO) between LTE and other RAT (Utran or Geran). When a LTE cluster or site is driven, it is desired that all the time UE are connected to LTE instead of IRAT to other technology. However, this sometimes is not possible, as LTE coverage becomes weaker in some areas it causes UE to switch to IRAT or to a better‐serving UMTS or GSM network. Cell reselection is used for inter‐RAT (IRAT) mobility in idle mode. The UE receives the information from the eNB for IRAT and inter‐frequency cell reselection through system information messages (SIB3 and SIB6) and dedicated signaling that determine when it is appropriate
Mobility Optimization
RRC-CONNECTED
RRC-CONNECTED
Bad coverage detection triggers Release with Redirect: Redirect Information Frequency
RRC-IDLE
Handover Command Move to reserved resources
CellReselection according to redirect information (GSM, WCDMA, LTE IF) “RRC-CONNECTED”
“RRC-CONNECTED”
Figure 8.49 Release with redirect and handover.
to begin measuring other RATs and frequencies along with specifics of how the measurements should be triggered, prioritized, and ranked prior to the UE performing IRAT cell reselection. The S criterion is again used to select the good cells for cell reselection. In RRC connected mode, there are in principle two ways of inter‐working between LTE and other RATs triggered based on event A2, and optionally, also by event B2. The inter‐working can be performed by a prepared handover (network controlled) where the UE does not leave the connected state (handover) or by a cell re‐selection/release and redirect (UE controlled or network assisted) where UE via idle state performs network assisted cell reselection (Figure 8.49). 8.8.1 Inter‐RAT Mobility Architecture and Key Technology
In Figure 8.50, an inter‐working network between LTE and GERAN/UTRAN is illustrated. The network shall ensure that the user has ongoing service performance by providing inter‐RAT handover functionalities. The UTRAN/GERAN network must be coverage overlapped with the eUTRAN network, and the neighboring frequency and neighboring cell relation shall be configured in the RAN nodes. LTE IRAT capabilities can be divided into two major areas: CS and PS mobility, PS mobility includes cell reselection, cell redirection, NACC, and PS handover mode, which is shown in Figure 8.51. Table 8.21 presents an example of mobility strategy between LTE and UTRAN PS domain. LTE will be broadcast as the highest‐priority RAT in all three technologies, followed by 3G, and then 2G. Reselection will only be made to a lower‐priority RAT when the destination GERAN MS
Um
Abis
BTS
A
BSC
-cs
Iu
G
UE
Uu
b
NodeB
Iub
RNC
Iu-ps
UTRAN
MSC/VLR Gs GnGp SGSN Gn
UE
LTE-Uu
E-UTRAN FDD X2
UE
LTE-Uu
E-UTRAN TDD
S1-MME S1 -U ME
-M
S1
E-UTRAN
Figure 8.50 IRAT interworking topology.
S1-U
MME
D SGs
Gr
HSS/HLR
S6a Gn
PCRF Gx
Rx
S11 S-GW
S5
P-GW/GGSN
Gi/SGi
PDN
307
308
LTE Optimization Engineering Handbook IRAT MOBILITY
Packet service (PS) mobility (including VOIP)
Cell Reselection
Redirection
Circuit Service (CS) Mobility
NACC (Network assisted cell Change)
PS Handover
Figure 8.51 IRAT strategy.
Table 8.21 Mobility Between LTE and UTRAN PS domain.
LTE => UMTS
UMTS=> LTE
UE State
Strategy
Description
idle
Cell reselection
The sib6 message is about the 3G neighbor cells. The system send the sib6 message to the UE side. And according to the sib6 and the LTE signal, the UE decides whether it is necessary to reselect to the 3G.
connected
R8 redirection
eNB sends RRC connection release message with ARFCNof UMTS. UE will access UMTS with the frequency.
R9 redirection
eNB sends RRC connection release message with ARFCNof UMTS and ralated SI information of the cell. UE will access UMTS with the frequency. SI is acquired by RIM procedure.
PS HO
When PS HO is triggered, eNB send mobility from EUTRA command message for UE handover to UTRAN. This feature is an end‐to‐end option that requires new functionalities in the UE, the access and the core network.
Idle/cell_PCH/ URA_PCH
Cell reselection
According to SIB19 or UTRAN mobility information ‐ > Dedicated priority information, E‐UTRA frequency, priority and cell reselection threshold, UE reselection from UTRA to LTE.
RRC connected
redirection
UTRA send UE E‐UTRA frequency by RRC connection release (redirection info) message for UE accesses to LTE.
cell_DCH
PS HO
Inter‐RAT PS HO feature can provide lossless handover between eUTRAN and GERAN/UTRAN, but the procedure is rather complicated.
c overage is expected to be better. A non‐zero setting of threshXLow for both 3G and 2G targets will be used to achieve this, ensuring that reselection does not occur when LTE and the destination RAT have poor coverage. A UE will reselect to a lower‐priority RAT when both the following criterion are met: RSRP
RSCP / RSSI
qRxLevMin EUtran
threshServingLow
qRxLevMin Utran/GERAN
threshXLow
Mobility Optimization
Redirection is a mechanisms to force a UE to switch from RRC‐connected to RRC‐idle mode and reselect to inter‐frequency/inter‐RAT neighbor cell for PS session continuation and/or CS call setup (CS Fallback). Redirection is triggered by either RSRP based event A2 or RSRQ‐ based event A2. Having both RSRP and RSRQ criterion, UE can be redirected to other LTE frequency/RAT layer to avoid performance degradation not only due to poor coverage but also due to extensive interference. Redirection target layer is selected by eNB according to parameter settings and UE capabilities. Selection of target for redirection is done without triggering any target layer measurements, that is, blind selection. NACC: The serving cell controls the redirection using the UE measurement data. UE receives system info from target cell before release from the serving cell. Note that NACC is only applicable to GERAN and is an alternative for GERAN that does not support PS handover (NACC is not applicable to UTRAN to eUTRAN mobility, but to GERAN to UTRAN) PS Handover: In addition to cell control used for redirection, target core communicates with the source core to establish the UE context. Unsent data in the serving RAN will be forwarded to the target UTRAN. This mobility procedure enables the establishment of network resources in the target UTRAN system prior to commanding the UE to move to the target cell. The overview of IRAT classification and mobility actions is shown in Figure 8.52. 8.8.2 LTE to G/U Strategy
The UE shall only perform cell reselection evaluation for E‐UTRAN frequencies and inter‐RAT frequencies that are given in system information and for which the UE has a priority provided. Through the configuration of the frequency priority, networks can easily lead terminal reselect to the high‐priority cell to camp, so as to achieve the balance network loading and improve resource utilization, make the UE signal quality better, and so on (Figure 8.53). In idle mode, priority‐based cell reselection is applied, the order of the priorities is LTE > UMTS > GSM. UE will camp on LTE with priority first, then UMTS, GSM finally. In connected mode, UE will move to UMTS (GSM) by redirection/handover in LTE poor coverage. When the user returns to the LTE overlay area in the UMTS (GSM) network, the redirection/handover to LTE can be applied. Figure 8.54 shows an example of LTE3G interworking in both idle/connected mode. To realize LTE3G interworking in idle mode, the wireless network needs to software upgrade the eNB like 3G neighbor configuration and reselection parameter, RNC needs to configure 4G neighbor configuration and reselection parameter, the core network needs to upgrade MME, MSC (should support SGs interface), and so on. There are four phases in LTE‐ > UMTS and UMTS‐ > LTE cell reselection: start measurement, measurement, decision, and execution. According to SIB3, SIB6, UE proceeds LTE‐ > UMTS reselection based on the priority. For UE in UMTS idle mode state or Cell_PCH/URA_PCH state, according to UMTS SIB19, which contains priority information (for E‐UTRA frequencies and UTRA cells) and LTE serving cell and neighboring cells (Figure 8.55). For LTE to GSM reselection, all descriptions of UTRAN above apply except UTRAN is replaced by GERAN and SIB 6 is replaced by SIB7. SIB7 contains GERAN frequency and neighboring cells for cell reselection mainly. GSM to LTE cell reselection is managed in “blind search” mode including PS cell reselection in NC0 mode (mobile controlled cell reselection). This means that 2G/LTE reselection algorithm is implemented at the MS side and performed autonomously. Inter‐RAT neighboring information and parameters SI2Quater message is broadcasted, all the parameters for the existing algorithm and the priority algorithm are sent. Then it is up to the MS to decide which algorithm to use in idle mode depending on its capabilities.
309
Release with Redirect to LTE (Blind)
Fast return (Blind) Reselect to LTE (higher priority)
Reselect to LTE (higher priority) HPRIOTHR
sNonlntraSearch Reselect to WCDMA (lower priority)
threshServingLow a1a2SearchThresholdRsrp b2ThresholdRsrp utranB1ThresholdRscp
PS HO to WCDMA
qRxLevMin a2CriticalThresholdRsrp
Coverage Triggered Session Continuity (Blind)
LTE Acceptable Performance
Decreasing signal strength
threshHigh
qRxLevMin (Eutra)
CSFB to GSM (Blind)
GSM
LTE
WCDMA
CSFB to WCDMA (Blind)
Reselect to GSM (lower priority) QRXLEVMINE
Coverage Triggered Session Continuity (Blind)
threshXLow(Utran) qRxLevMin
b2Threshold2RscpUtra qRxLevMin(Utran) threshXLow(Geran) qRxLevMin(Geran)
Figure 8.52 IRAT classification and mobility actions.
ACCMIN
Mobility Optimization 1 2 3 4 5 6
LTE 5
Cell Reselection from low to high priority Cell Reselection from high to low priority 3
Coverage based HO/Redirection
2 1 UMTS
4
6
Service based HO/Redirection CSFB Fast Return
1
GSM
5
6 2
3
Figure 8.53 Inter‐RAT frequencies priority. 1- Event A2 is sent by the UE after 640ms of poor coverage detection 2- Blind RWR to 3G Cell
1- UE Idle mode on LTE Network Starts Measuring on UtranFreq: 2- UE reselects 3G Cell after 2s
–122
–120
1- UE Idle on 3G Network starts Measuring on EUtranFreq: 2- UE reselects LTE cell after 2s
–118 RSRP (LTE)
–112
CONNECTED MODE LTE to 3G Session Continuity
IDLE MODE LTE to 3G Reselection
IDLE MODE 3G to LTE Reselection
RSCP (3G)
Figure 8.54 LTE3G interworking.
For 3G to LTE cell reselection, UE starts measuring IRAT cells in search of a better cell to camp on when the following criteria is met:
Srxlev
Snonintrasearch OR cellReSelPrio inter RAT
cellReSelPrrio serving
When the measured RSRP level will be lower than qrxLevMin + sNonintrSearch (i.e., −130 + 20 = −110 dB) the terminal will start IRAT measurements as shown in Figure 8.56. The
311
UE camps on LTE (idle mode)
UE camps on 3G (idle mode, or PCH state)
N Y
Start measurement
measurement
Y
LTE_priority > UMTS_priority
LTE(SqualServingCell)≤ 2(QRxLevMin + SNonIntraSearch)
N
Y
Start measure UMTS
decision
Y
execution
Start L to U reselection
Start measurement
Start measure UMTS
N
UMTS(SrxlevnonServingCell)> 2*(QRxLevMin + ThreshXHigh) for TReseIUtran
N
measurement
UMTS_priority > LTE_priority
UMTS(SqualServingCell)≤ Qqualmin
N
Y
Start measure LTE
Start measure LTE
N
N
LTE(SqualServingCell)≤ 2(QRxLevMin + ThrshServLow) and UMTS(SrxlevnonServingCell)> 2*(QRxLevMin+ThreshXLow) for TReseIUtran
+ Sprioritysearch2
N LTE(SrxlevnonServingCell)> 2*(Eqrxlevmin + ThdToHigh) for Treselection
decision
LTE(SrxlevnonServingCell)> 2*(Eqrxlevmin + ThdToLow) and UMTS (SqualServingCell)≤ Qqualmin for Treselection
Y
Y
Figure 8.55 LTE‐ > UMTS and UMTS‐ > LTE cell reselection.
execution
Start U to L reselection
Y
Mobility Optimization qrxLevMin = –130 dBm
IRAT measurements will start at –110 dBm
sNonIntrSearch = 20
RSRP decreases
–110
–130
Figure 8.56 Example of thresholds to start IRAT measurements.
Table 8.22 Example parameter values for IRAT reselection in idle mode. Inter‐frequency and inter‐ sNonIntrsearch RAT measurements threshold
20 dB
Defines the threshold (in dB) for inter‐RAT and inter‐frequency measurements.
Minimum required RX level in cell
qrxLevMin
−130 dBm Specifies the Mini required RX RSRP level
UTRA minimum required received level
qrxlevminUTRA
−115 dBm Minimum required Rx level in the cell.
Cell reselection priority
cellReSelPrio
7
Absolute priority of the LTE carrier frequency
UTRA carrier frequency absolute priority
uCelResPrio
3
Absolute priority of the UTRA carrier frequency
UTRA cell reselection timer
tResUtra
2s
UTRA cell reselection timer
Threshold serving low
threshSrvLow
14 dB
Threshold for the serving frequency used in reselection evaluation toward lower‐ priority LTE frequency or RAT
UTRA inter‐frequency threshold low
Threshx,low
8dB
Threshold used in reselection toward the frequency X priority from a higher‐priority frequency
condition regarding priorities is not fulfilled in this case, that is, LTE network has higher priority than 3G but it is enough one of the conditions is fulfilled to start measurements. 3G network can have higher or lower absolute priority than the LTE. This will depend on the operator’s strategy. If the operator wants to keep the terminals in LTE as long as possible it should give lower priority to 3G. Once the IRAT measurements are started, the cell reselection to the best cell on 3G layer will happen if for a time period equal to tResUtra (2 sec in the example):
SServingCell threshSrvLow and SnonServingCell , x Threshx , low
Based on example values from Table 8.22, when the RSRP level of the serving cell will be lower than −130 + 14 = −116dBm and the 3G level is greater than −115 + 8 = −107 dBm.
313
314
LTE Optimization Engineering Handbook
8.8.3 Reselection Optimization
Cell reselection information is related about the available inter‐RAT frequencies, the cell reselection priorities, and the threshold values for cell reselection is provided to the UE through the system information in the cell. Usually, LTE cell priority is 7 and 3G priority is 4. Snonintrasearch drive cell detection frequency and lower‐priority cell detection for UE. If the Srxlev value of the serving cell falls below the threshServingLow value, the UE attempts to reselect a cell on an inter‐RAT frequency with cell reselection priority lower than the frequency where the UE is camping. Cell reselection occurs if the UE finds a cell with an Srxlev value greater than the threshXLow value for that frequency (Figure 8.57). For 3G, the LTE frequencies and the parameters for priority based cell reselection are sent on the broadcast channel in SIB19. For GSM, each GSM cell broadcasts information via the s ystem information message SI2quater, includes neighboring cells (UMTS and LTE), thresholds for IRAT and priority between GSM, UMTS and LTE cells (Figure 8.58). 3G to 4G and 4G to 3G signaling are shown in Figure 8.59.
Signal strength Reselection Low Priority Signal to UMTS
Snonintrasearch Threshservinglow Threshxlow
LTE High Priority Qrxlevmin Starts measuring IF or IRAT freq Condition1: Srxlev (source) < ThreshServingLow
Signal strength
Low Priority Signal
TreselectionUTRA
&
Condition2: Srxlev (target) > ThreshXLow
Reselection to LTE High Priority Signal
ThreshxHigh
TreselectionRAT Condition: Srxlev (target) > ThreshXHigh
Figure 8.57 IRAT measurement and reselection.
Time
Time
Mobility Optimization
IDLE
CONNECTED RSRP–110]
LTE RSRP>–116 RSRP >–116
RSRP–112 or EcNo>–14]
WCDMA EcNo>–12
Fast Return
LTE
RSRP–109] RSRP–102
RWR
WCDMA
[RSCP S-nonintrasearch?
UE performs measurements of IRAT frequency cells of equal or lower CRP.
NO
CRP = Cell reselection priority
Figure 8.60 IRAT cell reselection.
Table 8.23 IRAT cell reselection parameter. Parameters
Value
Specification
SnonIntraSearch
31
Threshold when iRAT measurement are required for cell reselection.
ThreshServingLow
24
Threshold that the signal strength of the serving cell must be below for cell reselection toward iRAT cell
ThreshX,low
0
Threshold that the signal strength of the target iRAT cell must be above
According to Table 8.25, UE will start searching for a lower‐priority 3G frequency when:
RSRP
qRxLevMin sNonIntraSearch
128 16
112dBm
The UE will reselect to 3G when the below is fulfilled for a time tReselectionUtra: RSRP
qRxLevMin EUtranCellFDD
RSCP
threshServingLow
qRxLevMin UtranFreqRelation
threshXLow
128 8 115 4
120d dBm and , 111dBm
In addition, inter‐RAT cell reselection to a even lower‐priority GSM frequency is performed by the UE in LTE when the LTE and 3G criteria listed below are fulfilled during tReselection seconds: ●● ●● ●●
Serving LTE cell RSRP below qRxLevMin [EUtranCellFDD] + threshServingLow Target UMTS cell RSCP is NOT above qRxLevMin [UtranFreqRelation] + threshXLow Target GSM cell RxLev is above qRxLevMin [GeranFreqGroupRelation] + threshXLow
Using the current parameter values, the UE will reselect to GSM cell when LTE RSRP is below −120 dBm AND UMTS RSCP is NOT above −111 dBm AND GSM Rxlev is above −98 (−100 + 2) dBm for 2 seconds at least.
317
3G reselection Reselection when cell>–90 dBm, threshXLow = 30
Femto reselection Reselection when cell>–106 dBm, threshXHigh = 14 threshXhigh:11 is IE value, real configure value is 22. TDL RSRP>–98dBm
Figure 8.61 SIB6 and SIB19.
Mobility Optimization
Table 8.24 LTE to UTRAN reselection parameters. Parameter
Setting
qRxLevMin
−124 dBm
threshServingLow
4 dB
qRxLevMinOffset
2
threshXLow
6
sintrasearch
46 dB
sNonIntraSearch
8 dB 2
tReselectionUtra b2Threshold1Rsrp
−120
a2ThresholdRsrpPrim
−118
Table 8.25 Reselection parameters in SIB messages. Impact
SIB1 SIB3
Parameter
Measure
Decision
Recommended
Qrxlevmin (LTE)
√
√
−128
PEMAX (LTE)
√
√
23
Snonintrasearch (LTE)
√
Threshservinglow (LTE)
√
√
cellReselectionPriority (LTE) SIB6
5 ~ 7
Threshxlow (to 3G decision threshold) Qrxlevmin (3G) P‐MaxUTRA (3G) t‐Reselection‐UTRA (3G)
2
cellReselectionPriority (3G)
LTE RSRP
tReselectionUtra
qRxLevMin + sNonIntraSearch qRxLevMin (EUtranCellFDD) + threshServingLow
qRxLevMin
2 ~ 3
LTE to WCDMA cell reselection
WCDMA RSCP
qRxLevMin (UtranFreqRelation) + threshXLow qRxLevMin (UtranFreqRelation)
Figure 8.62 LTE to 3G reselection (example).
8.8.3.2 UTRAN to LTE
UMTS to LTE cell reselection to a higher‐priority LTE frequency is performed by the UE in UMTS if the measured LTE cell RSRP is greater than qRxLevMin + threshHigh during tReselection seconds. The measurement in UMTS for LTE neighbor cells will be enabled continuous. Idle
319
320
LTE Optimization Engineering Handbook LTE RSRP
Reselect from WCDMA to LTE threshHigh sNonIntra Search
Hysteresis region Stable on either WCDMA or on LTE
threshServingLow qRxLevMin Neither is suitable
Reselect from LTE to WCDMA
Fall off LTE and camp on WCDMA qRxLevMin
threshXLow
UMTS RSCP
Figure 8.63 IRAT reselection strategy.
mode cell reselection evaluation is performed every DRX cycle. The UE will reselect to LTE when the below is fulfilled for a time tReselectionUtra:
RSRP
threshHigh qRxLevMin EUtranFreqRelation
10
128
11 18dBM
The related parameters (threshHigh, qRxLevMin etc.) are included in SIB19 in UMTS. The whole IRAT reselection strategy is shown in Figure 8.63. It illustrates the hysteresis and reselection regions for transition between LTE and UMTS in idle mode. 8.8.4 Redirection Optimization
Cell redirection is a cell reselection when UE is in RRC connected state. It enables a multi‐RAT UE to be quickly redirected toward a target LTE (2/3G) cell while a UE is camping in a 2/3G (LTE) cell in RRC connected state (and involved in a data transfer). The redirection will make sure the retainability of the PS service and good end‐user experience. 8.8.4.1 LTE to UTRAN
The LTE serving cell controls the redirection using UE RF measurement data. Two options are supported: blind redirection and measured redirection. Blind redirection from LTE to UTRA is used if UE does not support event B2 measurement (measurement purpose = “Mobility‐Inter‐RAT‐ to‐UTRA”), or event B2 measurement report is not available when mobility from LTE to UTRA has to be triggered. Compared with a blind redirection without UMTS radio measurements, the redirection with measurement improves the end‐user QoE (quality of experience) by redirecting the UE from an LTE cell to an UMTS overlay in a timely fashion. Measurement‐based redirection from LTE to UMTS is used if PS handover is not supported by the UE or is not activated in eNB. eNB may broadcast six UMTS frequencies in redirection message. The blind redirection is triggered when no PS handover is implemented or activated nor supported between those two systems and when the UE is not supporting inter‐system measurement gap capability. The blind redirection is based on serving cell measurement (Event A2 measurement) only rather than measured redirection, which is based on serving and target measurement (Event A2 and Event B2 measurements). The measured redirection is the same as the previous one with the enhancement on the measurement gap functionality. This procedure is triggered when no PS handover is implemented or activated nor supported and with UE supporting inter‐system measurement gaps on radio frames. The enhancement is mainly related to radio side, which improves UMTS redirection time duration and efficiency. Measurement gaps periods are set by RRC signaling reconfiguration procedure toward UE by fixing, length, periodicity and offset information in
Mobility Optimization
Sometimes try changing A2 parameter to trigger IRAT earlier and reduce drops during redirect RSRP Serving Cell RSRP threshold4
threshold4
–110 dBm
reporting condition met A2 condition met after Time To Trigger
a2TimeToTriggerRedirect time Connected Mode
Triggered A2
Other case
Decision
iRAT measurement
Blind Redirection Start A1 and B2 A1 reported
1
3
A5B2 timer expires
2 B2 reported
Decision
Other case
Initiate Redirection
Figure 8.64 RRC connection release with redirect procedure.
radio subframes. The need for measurement gaps by the UE is specified in the UE capabilities. This scenario is using event A2 and B2 type measurement to indicate LTE serving cell degradation and redirection decision (Figure 8.64). Event B2 based redirection parameters are shown in Table 8.26. The example of redirection procedure and related parameters is shown in Figure 8.65.
321
322
LTE Optimization Engineering Handbook
Table 8.26 Redirection parameters. Parameters
Values
UEMeasurementsActive
true
a1ThresholdRsrpPrim
−90
Comments
Decides whether iRAT measurement should be started RSRP threshold value for the primary event A1 measurement
a2ThresholdRsrpPrim
−102
RSRP threshold value for the primary event A2 measurement
b2Threshold1Rsrp
−104
RSRP threshold value of the serving cell for the Event B2 measurement
b2Threshold2RscpUtra
−119
RSCP threshold value for the Event B2 measurement
a5B2MobilityTimer
3000
Specifies the time the UE are allowed to perform Event A5 and B2 measurements
8.8.4.2 UTRAN to LTE
When UE was in connected mode in UMTS, the UE would receive the event 3A‐ or 3C‐related information. And then the UE would start the measurements. UE reported the 3A or 3C, and the RNC will send the RRC release information, which include the target frequency and with redirection information. There are eight frequency sent by RNC in RRC redirection information. And when UE received the command, the redirection will be executed after the TimeToTriger timer reached. Table 8.27 is the threshold of the 3A and 3C event. Figure 8.66 shows the procedure of UTRAN to LTE redirection. When the measurement threshold was satisfied, UE send measurement report to RNC. The redirection decision made by RNC and if the redirection will be executed, then RRC connection release will be sent to UE with target frequency information, finally UE sends the service request to MME and finishes tracking area update procedure. Table 8.28 gives an example of the threshold for trigger of event 3A. If you want to make the redirection from UMTS to LTE easily, thres1 can be set a little higher (−50 dBm), thres2 can be set a little lower (−110 dBm). These two parameters need to be set consist with reselection threshold, for example, the threshold for UTRAN to LTE reselection trigger EqrxlevMinRsrp can be set to −112dBm in this case shown in Table 8.28. 8.8.5 PS Handover Optimization
The redirection is based on the RRC release information and found the target frequency and cell to access again. While PS handover flow will include the measurement and decision and action, and it can based on coverage and capacity and RRM will be discussed between serving and target eNB, and then prepare handover and UE finished handover with commands. In theory inter‐RAT PS handover feature can provide lossless handover between eUTRAN and GERAN/UTRAN, but the procedure is rather complicated. 8.8.5.1 LTE to UTRAN
This part introduces the UE measurement–based packet switched (PS) handover procedure to move UE from LTE to UTRAN. eNB will trigger the PS handover when UE is leaving LTE coverage area and moving into UTRAN coverage area, and the UE measurement report indicates that the LTE radio condition becomes worse than a threshold and the UTRA radio condition becomes better than a threshold. Comparing with redirection, PS handover from LTE to UTRAN has the advantage of allocating the resources in UTRAN prior to the execution of PS handover. Besides, PS handover has the capability of data forwarding from source LTE to target UTRAN. It thus reduces the service interruption time and ensures better performance to packet loss–sensitive services.
The source eNB will give a command to the UE to reselect a cell in the target access network via the RRC connection release.
Figure 8.65 Field test of redirection procedure and related parameters.
324
LTE Optimization Engineering Handbook
Table 8.27 3A and 3C. Events
Event description
In
Leave
3A
RSCPScell thres2
QUsed TUsed H 3 a /2 and MOtherRAT CIOOtherRAT TOtherRAT H 3 a /2
QUsed TUsed H 3 a /2 or MOtherRAT CIOOtherRAT TOtherRAT H 3 a /2
3C
RSRPNcell > thres
MOtherRAT CIOOtherRAT TOtherRAT H 3c /2
MOtherRAT CIOOtherRAT TOtherRAT H 3c /2
UE Trigger phase
Node B
RNC
eNB
RRC_PH_CH_RECFG RRC_PH_CH_RECFG_CMP
Measurement phase
RRC_MEAS_CTRL RRC_MEAS_RPRT RRC_RRC_CONN_REL
Decision phase Execution phase
RRC_RRC_CONN_REL_CMP TrackingAreaUpdateRequest
Figure 8.66 UTRAN to LTE redirection.
If PS handover is to be performed, eNB will select the best UTRA cell as the PS handover target cell. eNB will send a handover required message to the MME and start timer TS1relocprep. If the reservation of resources in the target UTRAN cell is successfully completed, MME will send a handover command message to eNB. eNB will then stops the timer TS1relocprep and enter the handover execution phase. If timer TS1relocprep expired, or eNB receives a S1 handover preparation failure message, handover preparation fails. Once a handover command message is received from MME, eNB will stop timer TS1relocprep and start timer TS1relocoverall, eNB will send a MobilityFromEutraCommand to the UE with purpose set to “handover” and targetRAT‐Type set to “ultra.” If UE context release command is received from MME, PS handover is successful. eNB will send a UE context release complete to MME. eNB will stop timer TS1relocoverall and release UE context and associated resources. If timer TS1relocoverall expires, eNB considers the UE to have lost radio coverage and will trigger the release of all UE associated resources by sending an UE context release request to MME and release all UE associated resources in eNB. 8.8.5.2 UTRAN to LTE
For UTRAN to E‐UTRAN handover procedures, the UTRAN has two main tasks: processing the handover decision and initiating the handover procedure by sending RANAP relocation required message with relevant content to source SGSN. Prior to the handover preparation phase, UTRAN has to manage UE LTE capacity, monitor the radio condition change and LTE neighboring cells, and eventually activate CM measurements. In the handover execution phase, UTRAN will process CN message RANAP relocation command, command the UE to handover to the target eNB via the message RRC handover from UTRAN command, which contains the transparent container with radio aspects parameters set up by target eNB during preparation
Mobility Optimization
Table 8.28 Example of the threshold for trigger of event 3A. NLTE RSRP > thresholdOtherSystem + Hysteresis/2 ‐ CIO and SUMTS RSCP RRC connection request/ channel request(s)
RRC connection request— > TAU accept/RAU accep(s)
Delay(s)
TD‐L‐ > TD‐S
500
1.488
2.203
3.691
TD‐S‐ > TD‐L
500
0.114
1.06
1.209
Table 8.30 Parameters comparison of idle‐mode reselection and connected‐mode redirection. Idle mode
Connected mode
Description
Values Idle/connected
Qrxlevmin
N/A
Absolute minimum RF value
−122/−122 dBm
Treselection
Time To TriggerA2Prim
Time to Trigger
2000/640 ms
ThreshServingLow
a2ThresholdRsrpPrim
Minimum RSRP value to trigger
−116 dBm/−116 dBm
qHyst
hysteresisA2P rim
Hysteresis
0 dBm/1dBm
325
326
LTE Optimization Engineering Handbook
8.8.7 Optimization Case Study
One of the things that needs to be monitored in LTE mobility domain is related with the success rate of the handover procedures on source cell. In order to evaluate this it is advised to check success rate indicators, which represents the percentage of the successfully performed handovers. It has possibilities to monitor the different types of LTE‐LTE handover: intra‐eNB, inter‐ eNB, intra‐frequency (X2), inter‐eNB intra‐frequency (S1), inter‐eNB inter‐frequency (X2), and inter‐eNB inter‐frequency (S1). Case 1: ping‐pong reselection In order to prevent the ping‐pong reselection, the reselection threshold of LTE/UMTS/GSM should be considered. In order to prevent the LTE reselection to GSM network quickly after reselection to the UMTS network, 3G signal threshold of LTE to UMTS network needs to be higher than that of UMTS to GSM. In order to prevent ping‐pong reselection between the LTE and UMTS network, the 4G LTE network signal threshold of UMTS to LTE network should be higher than LTE reselection to UMTS threshold (Figure 8.67). Case 2: UE redirection to LTE failure due to EPLMN misconfigured The function of UMTS to LTE redirection is enabled, it is found that after RNC proceeded redirection to LTE, the UE still stayed on UMTS. As shown in Figure 8.68, after UE reported event 3C, RRC connection release was transmitted. LTE frequency was carried in the release message, but the UE was back to UMTS after LTE SIB reading (Figure 8.68). Considering the particularity of the live network, the PLMN of UMTS, LTE of the two operators were inconsistent. Therefore, 3G/4G are required to configure the equivalent PLMN, the inspection found that CS domain of LTE and UMTS were configured the IRAT system equivalent PLMN, but PS domain of UMTS side was not configured the equivalent PLMN of IRAT system, it can be seen by the results of TAU/LAU/RAU attach procedure. After configuring the PS domain equivalent PLMN, UE was redirected to the LTE to initiate the TAU process, as shown in Figure 8.69. For IRAT operation, when UE redirects the PLMN from one system to another one, if the UE cannot access the target system after reading the SIB message, it needs to inspect whether the PLMN is consistent and equivalent PLMN is misconfigured by the core network.
8.9 Handover Interruption Time Optimization After a handover command is triggered, the UE disconnects from the serving eNB before setting up a connection with the target eNB and stops receiving data. This is the point in time where data interruption starts. For long handover interruption time, interference and missing neighbors need to be analyzed. The period of time where the UE can not exchange data is referred to as handover interruption time. It includes the time to execute radio access network procedures, the time for UL and DL radio resource control signaling, and the time taken to notify and execute the data path switching. Handover preparation time is the time duration the eNB takes to prepare the handover. Handover execution time is the time duration RRC signaling is interrupted during handover. Handover preparation time and handover execution time can be according to: Handover preparation time = T (RRC connection reconfiguration) − T (measurement report) Handover execution time = T (RRC connection reconfiguration complete) − T (RRC connection reconfiguration) Table 8.31 gives the most popular handover lantency/interruption KPIs.
UE RAT Type
UE RAT Type
WCDMA
LTE
WCDMA
RAT type
RAT type
LTE
03 07 11 15 19 23 27 31 35 39 43 47 51 55 59 03 07 11 15 19 23 27 31 35 39 43 47 51 55 59 0: 0: 0: 0: 0: 0: 0: 0: 0: 0: 0: 0: 0: 0: 0: 1: 1: 1: 1: 1: 1: 1: 1: 1: 1: 1: 1: 1: 1: 1: :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 :2 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13 13
Time
Figure 8.67 Ping‐pong reselection optimization before and after.
48 52 56 00 4 8 12 16 20 24 28 32 36 40 44 48 52 56 00 04 08 2 6 20 24 28 32 36 40 44 7: 7: 7: 8: :0 :0 8: 8: 8: 8: 8: 8: 8: 8: 8: 8: 8: 8: 9: 9: 9: :1 :1 9: 9: 9: 9: 9: 9: 9: :4 :4 :4 :4 :48 :48 :4 :4 :4 :4 :4 :4 :4 :4 :4 :4 :4 :4 :4 :4 :4 :49 :49 :4 :4 :4 :4 :4 :4 :4 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16
Time
328
LTE Optimization Engineering Handbook
Figure 8.68 UE redirection to LTE failure.
Figure 8.69 UE redirection to LTE success after PLMN configured.
Table 8.31 KPIs focused on the volume. KPI Name
HO interruption delay for VoIP HO interruption delay for best effort service LTE TO UMTS PS HO – interruption time and latency Voice call setup time for CSFB to UTRAN using Redirection/PSHO Data interruption time due to LTE to UTRAN CSFB via Redirection Voice call setup preparation time for CSFB to 1xRTT by idle Dual‐Rx UE LTE to CDMA HO interruption time type HO (data) Enhanced non‐optimized LTE‐to‐HRPD handover preparation latency Data interruption time for LTE to GERAN PS mobility using CCO with NACC LTE‐to‐GERAN handover via Redirection interruption time Voice call setup time for CSFB to GERAN using CCO with NACC/Redirection Voice call interruption time for SRVCC HO to UTRAN
Mobility Optimization
Figure 8.70 Handover interruption time estimation.
Take S1 handover, for example, the handover interruption time is estimated in Figure 8.70. For CBRA, the interruption time is estimated to approx 62 ms in DL and 32 ms in UL. Transport delay in both directions is not included in the figures. For CFRA, eNB knows that it is a handover access from the expected UE when receiving random access preamble. This means that the eNB can start transfer of DL data earlier which will reduce the interruption time by approx 10 ms to approx 52 ms. 8.9.1 Control Plane and User Plane Latency
Handover control plane interruption time is defined as the duration between the handover command at source eNB and the RRC reconfiguration complete at target eNB. In average the time from measurement report to RRC connection reconfiguration took 40 to 50ms. From RRC connection reconfiguration message to RRC connection reconfiguration complete, it took the UE 30 to 40ms (Figure 8.71). User plane (UP) interruption time has been noticed to be too high in field test, especially with FTP, handover procedure works as specified by 3GPP and user plane interruption times are varying roughly even between 100ms to 350ms most of the time. After disconnecting from the serving eNB, the UE typically waits for the next random access opportunity to execute a random access procedure to acquire service with the target eNB and eventually be able to resume any data exchange. An example of X2 handover UP latency test that uses the different eNBs to identify last/ first package is given in Figure 8.72. It shows before handover, UE1 (IP address 10.63.0.2) received the last package at T1 = 15.850252s, and in the time UE1 stays at eNB1 (IP address 10.100.100.10) (as two S1 links are mapped to a same mirror port, so there are two messages with same source and destination). At T2 = 15.899244s, the first package is received by UE1 after handover to eNB2 (IP address 10.100.100.2). So handover UP latency = 15.899244−15. 850252 = 49 ms. Handover UP is also can be computed the UP handover latency based on the signaling and RLC PDU, shown in Figure 8.73.
329
330
LTE Optimization Engineering Handbook UE
Target eNB
Source eNB Measurement Report
40–50 ms RRC Connection Reconfiguration
Handover Command
30–40 ms RRC Connection Reconfiguration Complete RRC Connection Reconfiguration RRC Connection Reconfiguration Complete
Source eNB
UE
Target eNB
Measurement report Last RLC PDU
T304 starts
RRC Connection Reconfig
Minimize PRACH access latency
Random Access Procedure
T304 stops
UP interruption time
RRC Connection Reconfig Complete 1st RLC PDU
Figure 8.71 Control plane (left) and user plane (right) latency.
Figure 8.72 An example of X2 handover UP latency test.
After optimization with properly tuned parameters, less than 50 to 60ms FTP layer interruption time can be achieved in good RF conditions. For more handover interruption reduction improvement that aims to shorten the LTE S1 and X2 handover interruption time by triggering the path switch (for X2 handover) and the handover notify (for S1 handover) messages to MME earlier when the target eNB receives the RA message 3 from the UE, which is needed to be earlier than the reception of RRC connection reconfiguration
Mobility Optimization
Figure 8.73 UP handover latency based on the signaling and RLC PDU.
UE
Source eNB
Target ANR
MME
SGW
RRC:MEASUREMENT REPORT HO Decision X2AP: HANDOVER REQUEST Resource allocation X2AP: HANDOVER REQUEST ACKNOWLEDGE RRC: RCC CONNECTION RECONFIGURATION
Switch to Target RAN
X2AP: SN STATUS TRANSFER
MAC: CBRA RACH PREAMBLE MAC: CBRA RACH RESPONSE RRC CONNECTION REQ REQ/MSG3 / MSG3
S1: PATH SWITCH REQUEST
SETUP/MSG4 RRC CONNECTION SETUP / MSG4 S1: PATH SWITCH REQUEST RRC: RRC RRC: RRCCONNECTION CONNECTION RECONFIGURATION RECONFIGURATION COMPLETE COMPLETE
DL data
S11: UP update req
Path Switch
Figure 8.74 Reduction on the S1 and X2 handover Interruption time.
complete message from the UE but with the same confidence on the UE presence in the target cell. Figure 8.74 shows for the X2 handover case the point when RA message 3 is received in the target eNB and where the path switch request will be triggered. The procedures (in circle) that follow this point will therefore trigger the DL data path switch ~12 ms earlier.
331
332
LTE Optimization Engineering Handbook
The system shall be configured to use RA message 3 instead of RRC connection reconfigura tion request for initiating intra eNB switch. During incoming intra‐eNB handover execution and upon the reception of RA message 3, the RAN use this message as an earlier trigger to confirm internal dispatcher of the UE presence in the target cell and initiate the release of resources at the source cell. During incoming X2 handover execution and upon the reception of RA message 3, the RAN use this as an earlier trigger to send the S1‐AP path switch message to MME. During incoming S1 handover execution and upon the reception of RA message 3, the RAN use this message as an earlier trigger to send the S1‐AP handover notify message to MME. 8.9.2 Inter‐RAT Mobility Latency
For inter‐RAT mobility, UL trigger can be possible to set per QCI in the same way as bad coverage DL trigger based on GINR. GINR can be calculated as:
GINR
psdRX
psdTX
N
I
where psdRX (Received signal PSD), psdTX is the estimated UL Tx PSD based on power headroom reports from the UE; N + I is the received noise and interference averaged over the bandwidth. Measurement‐based and non‐measurement based “InterRAT cell reselection” and “Release with redirection” from the LTE network to both UMTS and GSM are supported as shown in Figure 8.75. For inter‐RAT mobility latency, timing is an indication only as it is very terminal dependent. Cell search time may be reduced if inter‐RAT measurement done before release with redirect. The UE will take much time for cell search. The UE first performs PLMN selection and finds its home PLMN where the MCC and MNC of the PLMN identity match the MCC and MNC of the USIM’s IMSI. The UE may optionally use information it has previously stored if it is available in order to reduce the time needed to select a PLMN.
8.10 Handover Failure and Improvement Mobility procedure can be divided into handover preparation and handover execution phases. Handover preparation is the phase in which the target cell assigns the necessary radio resources for taking over the connection and sending back a handover command message containing the new radio parameters to the source cell. Handover execution phase starts when the previously received handover command message is sent to the UE and successfully finished after the UE has arrived at the target cell. In most of the cases, handover failure could be due to poor radio conditions or badly tuned handover parameters. For example, with too little overlap between cells, handover may fail, with too much cell overlap, higher interference occurs and cell‐edge throughput can be reduced. So a balance must be achieved by adjusting overlap margins and cell sizes. This can be achieved with parameters and physical changes. Downlink mobility issues include RLC failure on SRB1 that UE doesn’t receive RRC reconfiguration (handover command) or PDCCH decoding error (UE miss detect the PDCCH order). Uplink mobility issues include random access failure in the target cell or target cell didn’t receive RRC reconfiguration complete message from UE. If statistic analysis indicates the main reason for low handover success rate in the network was due to downlink, it could be either due to the DL RLC transmission for RRC handover
Re d
ire
ct
Mobility Optimization
UL trigger
Handover
nd
Bli Triggered DL trigger (from UE)
Non
ct
re
-bli
nd
A5, B2, B1
i ed
R
Handover
12 s
5–6 s Cell search
4–5 s 2–4 s
300 ms
PS handover PS handover
Cell search
Cell search
SIB reading
SIB reading
Bearer setup LAU/RAU
Bearer setup LAU/RAU
Bearer setup LAU/RAU
Release with Redirect with NACC (SI)
Release with redirect
Radio link failure
IRAT to WCDMA 15+ s
9–11 s Cell search
Cell search
Cell search
SIB reading
SIB reading
Bearer setup LAU/RAU
Bearer setup LAU/RAU
Bearer setup LAU/RAU
8–9 s
300 ms
PS handover
IRAT to GSM
Figure 8.75 Inter‐RAT mobility latency.
command reaches max transmit attempts or due to PDCCH decoding issue. Downlink should be focused to improve the handover success rate by improving RLC robustness, improving PDCCH robustness, and reducing downlink interference by means of RF tuning, and so on. For mobility troubleshooting, handover preparation failures and handover execution failures should be identified when monitoring LTE. In Figure 8.76, the example is about handover preparation failure. The source eNB initiates the procedure by sending the handover request message to the target eNB. When the source
333
334
LTE Optimization Engineering Handbook E-RAB QoS information Input Target eNB
Resources are granted by target eNB?
HO preparation failure
N
Y Default value for Trelocprep is: 5s
Configure required resources
Reserve a C RNTI for UE
Reserve a RACH preamble (optional)
source eNB
target eNB HANDOVER REQUEST
HANDOVER PREPARATION FAILURE
Figure 8.76 Example of handover preparation failure due to Trelocprep expiry.
eNB sends the handover request message, it shall start the timer Trelocprep. Upon reception of the handover request acknowledge message the source eNB shall stop the timer Trelocprep, start the timer TX2relocoverall and terminate the handover preparation procedure. If there is no response from the target eNB to the handover request message before timer Trelocprep expires in the source eNB, the source eNB should cancel the handover preparation procedure toward the target eNB by initiating the handover cancel procedure with the appropriate value for the cause IE. Here is another example of handover preparation failure as shown in Figure 8.77. If the target eNB does not admit at least one non‐GBR E‐RAB, or a failure occurs during the handover preparation, the target eNB shall send the handover preparation failure message to the source eNB with the cause IE. If the target eNB receives a handover request message containing RRC context IE that does not include required information as specified in TS 36.331, the target eNB shall send the handover preparation failure message to the source eNB. Figure 8.78 presents the possible causes of handover failures. Handover preparation may fail because the target eNB cannot provide the necessary resources for the handover during congestion or cannot interpret the contents of the handover request message. If the source eNB does not receive a response to its X2AP handover request message from the target eNB, the source eNB will send the handover cancel message. Handover execution failure that may be caused by incorrect parameter settings in the target cell (e.g., PCI collision in the target cell). From the possible causes for handover degradation, it can be concluded according to the mobility optimization steps as shown in Figure 8.79. To better understand what is happening with the handovers use a map to plot the locations of the target and each source cell and also draw a xx km (nearly triple inter‐site distance) ring around the target to identify the source cells with the highest number of failures and add pointers that show the sector’s azimuth. You must first check all KPIs for the site, for example, handover oscillation level, noise floor, interference, high or low traffic, PCI conflicts exist or not, handover preparation success rate, and execution success rate.
Mobility Optimization
Figure 8.77 Example of handover preparation failure due to no resource granted.
The meaning of the different handover cause value is described in Table 8.32.
8.11 Mobility Robustness Optimization In live network, it can be found that in certain scenarios measurement report messages count is significantly higher than number of handovers which indicates inability to execute handover quickly. In such scenarios, UE receives ACKs for PHICH and RLC for the measurement report but does not receive RRC reconfiguration message. Boxes in Figure 8.80 indicate the range of RSRP/RSRQ values where handovers took place easily.
335
336
LTE Optimization Engineering Handbook HO preparation failures
HO execution failures
Possible Causes
Other reasons for poor mobility
Possible Causes
1. Resources not granted by target eNB 2. Incorrect parameter settings in target 3. Congestion in target cell 4. Target cannot interpret the contents in handover request message 5. License issue 6. Target cell down 7. Improper MME configuration 8. Insufficient SR and CQI resources for target
1. Incorrect parameter settings in target (PCI collision in target cell) 2. Hardware faults 3. Unreasonable weight of handover attempt to target cells (If handover is triggered too late, the source cell SINR can be too low. This can result in an abnormal release before handover.) 4. Swapped sectors 5. Overshooting 6. Unreasonable neighbor relation
Possible Causes 1. Poor radio conditions (interference) 2. Badly tuned handover parameters -Handover hysteresis and time-totrigger settings are required to prevent excessive ping-pong handovers. -Such behavior increases signaling, risk of failure, and decreases throughput
Figure 8.78 Possible causes of handover failures.
Low HO Succ
HoPre fail
HoExe fail
Target Cell congest?
N X2 HO? N Check X2 definition
Scell->Multi-tcell
Y Expansion, distributary, control coverage
S1 HO? Y Check core configuration, cut over...
Single neighbor
Multi Scell->Tcell
Scell alarm?
Tcell alarm? Y
Y Alarm processing
Alarm processing
N
Overlapping?
Check coverage, power
Y Control coverage
N Tcell interference?
N
Checking interference
Parameter, earfcn/pci, external defining
Figure 8.79 Mobility optimization steps.
The next step will further analysis of handovers for triggering event per LTE cluster, cell or user. It needs to find that the number of A3‐, A5‐, B2‐based handovers in a live network. Mobility optimization feature enabled in RAN targets the following intra‐LTE mobility issues: ●●
●●
●●
Connection failure is due to too early handover: The UE has a connection failure during a handover procedure or soon (1s) in the same cell and if the UE re‐establishes in a different cell, this usually happened just before handover, it was observed that the UE SINR was poor and this caused the UE to drag the call in poor radio environment rather than handover faster to a neighbour cell with better RSRP Connection failure is due to handover to wrong cell: The UE has a connection failure during a handover procedure or soon ( Rs; Relation between (Qmeas,n – Qmeas,s) and Qoffset is Qmeas,n – Qmeas,s > Qhyst + Q offse,. which leads to Qoffset , Qmeas,n – Qmeas,s ; Qoffset , Qmeas,n – Qmeas,s .
8.12 Carrier Aggregation Mobility Optimization Carrier aggregation (CA) is used to increase the bandwidth by combining up to five carriers intra‐band or inter‐band, and thereby increase the peak bitrates. CA UE throughput is much higher than signal carrier, due to additional carrier, frequency selection and scheduling gain and high category UE gain. The UE have one Primary cell (Pcell), this will also be the cell where the UE is connected to, from a EPC perspective. The cell that is not the Pcell will be the Secondary cell (Scell). Mobility is based on Pcell coverage, there is no changes in cell selection/reselection for carrier aggregation. CA configured eNBs are compatible with non‐CA eNBs (for handover, etc.), and that non‐CA UE can co‐exist on a cell simultaneously with CA configured UE. UE uses Pcell to monitor system information, maintain RRC connection, monitor RLF, random access, and so on, all security input and NAS mobility information is communicated
341
342
LTE Optimization Engineering Handbook PCC
SCC
SCC
PCC
SCC
SCC
PCell
SCell
SCell
PCell
SCell
SCell
PUCCH & PUSCH
PUSCH Only
PDSCH & PDCCH
PDSCH & Optional PDCCH
Figure 8.84 PCC and SCC.
RRC reconfiguration, add Scell IE.
Not CA configured
RRC reconfiguration, remove Scell IE.
Activation MAC control element
CA configured
Deactivation MAC control element
Active Scell triggered based on buffer status
CA activated
Deactive Scell triggered based on buffer status or poor Scell CQI
Figure 8.85 Dynamic Scell selection.
using Pcell. UL/DL carrier corresponding to Pcell is primary component carrier (PCC) and to Scell is secondary component carrier (SCC). The Scell operates on a secondary frequency and can be configured once the RRC connection is established. Separate UL PC is required for different CC due to difference in propagation condition in non‐contiguous CA and for different interference conditions in UL CC in contiguous CA (Figure 8.84). The similar Pcell and Scell coverage is the precondition to provide CA capability for the UEs. Before the second carrier cell goes on air, single‐site verification shall be conducted carefully to make sure there are no defects, and second carrier RF optimization tuning shall be done aligning to the first carrier coverage. Similar antenna model, height, downtilt and azimuth is preferred for multi‐carriers. There are always coverage mismatching due to the separate mounted antennas and different bands, which will degraded CA throughput due to poor Scell performance. The mitigation is to activate the feature— dynamic Scell selection as shown in Figure 8.85. It is worth to note that CA configuration/deconfiguration is performed on RRC level, CA activation/deactivation is performed on MAC layer. At attach, reestablishment and incoming handover, the eNB will check CA license, CA neighbor cell configuration (i.e., SCell candidate) and UE capability. Figure 8.86 shows SCell selection is performed by RRC connection reconfiguration in attach procedure. SCell activation/deactivation is performed by MAC control element, which can be shown through RF conditions, resources scheduling, and DL/UL throughput, and so on, in driving test. In CA activated state, the UE is ready to receive data transfer on the SCell (DL assignments sent on SCell, HARQ ACKs are sent to eNB via PCell) and report the SCell’s CSI to eNB via PCell. SCell activate/deactivation can be triggered based on “need” or “coverage.” It notes that eNB will not deconfigure but deactivate the SCell when UE goes out of SCell coverage. PCell always changes due to handover, in the new PCell, old SCell is removed and new SCell is configured. All of this is done in the same RRC reconfiguration message as the handover itself. For CA mobility, a new measurement event A6 is introduced for CA, an intra‐frequency neighbor becomes offset better than SCell for which neighbor cells on an SCC are compared to SCell of that SCC. A6 works in the same way as A3, it reports the strongest cell that matches the configured frequency. If service triggered mobility is used to change the inter‐frequency handover thresholds, CA will continue to be possible with the new combination of PCell and SCell (Figure 8.87).
Mobility Optimization
Figure 8.86 SCell’s configuration and activation and average DL throughput (example).
343
LTE Optimization Engineering Handbook
Low CQI of Scell trigger Scell deactivation Scell deactivation observed by DL MAC transport block
Average DL Throughput (Mbps)
344
12 11 10 9 8 7 6 5 4 3 2 1 0
11.5
6.4 5.3
Pcell (CA UE)
4.3
Scell (CA UE)
Pcell and Scell (CA UE - Total)
Serving Cell (Non CA UE)
Figure 8.86 (Continued)
Pkt version = 7 RRC Release Number.Major.minor = 10.7.1 Radio Bearer ID = 1, Physical Cell ID = 68 Freq = 5230 SysFrameNum = N/A, SubFrameNum = 0 PDU Number = DL_DCCH Message, Msg Length = 93 SIB Mask in SI = 0x00 Interpreted PDU:
Serving Cell Info: PCI = 68
value DL–DCCH–Message :: = { 2. message C1: rrcConnectionReconfiguration: { rrc–TransactionIdentifier 0, criticalExtensions c1 : rrcConnectionReconfiguration–r8 mobilityControlInfo { targetPhysCellId 79, carrierBandwidth { dl–Bandwidth n50 handovertype intraLTE : { sCellToReleaseList–r10 { 1 }, sCellToAddModList–r10 { { sCellIndex–r10 1, cellIdentification–r10 { physCellId–r10 79, },dl–CarrierFreq–r10 2100
Figure 8.87 PCell and SCell handover.
Handover Command
Handover type: Intra Frequency PCC: PCI = 79
Pcell always changes due to handover
In the new Pcell, old Scell is removed and new Scell is configured. RRC reconfig message becomes bigger. SCC configuration during handover SCC: PCI = 79
Mobility Optimization
As CA is performed at MAC layer, PDCP and RLC are not aware whether a certain packet will be transmitted in the PCell or in the SCell. All counters for MAC and physical layer are registered on the cell where the data is transmitted. CA expected to increase PDCP throughput in the network since a PDCP packet may be sent using multiple cells in parallel. A CA user using SCell resources, may therefore increase observed PDPC throughput in the user’s PCell but potentially decrease observed PDPC throughput in the SCell. CA‐related KPIs include accessibility, retainability, DL throughput, the number of configured CA UEs, percentage of CA scheduled, percentage of CA traffic, PDCCH utilization, and so on.
8.13 FDD‐TDD Inter‐mode Mobility Optimization LTE supports mobility of TDD/FDD dual band UE during RRC‐connected mode (intra LTE handover) and in idle mode (cell reselection), while this UE is moving between LTE TDD coverage area is adjacent to or overlay with LTE FDD area. FDD‐TDD carrier aggregation combines excellent FDD coverage with large TDD capacity that the way of primary cell on FDD Scell on TDD to boost downlink capacity is many operators’ choice that own spectrums in both LTE modes. Mobility between LTE FDD and TDD will be of increasing importance for operators that have spectrum for both LTE modes, allowing operators to seamlessly offer mobile broadband services on FDD and TDD spectrum, increasing capacity and improving end‐user experience. FDD TDD mobility is very similar to inter‐frequency mobility, including load management. UE Capability of PS handover between FDD and TDD is indicated by FGI (bit 30) (Figure 8.88). Cell selection is priority (LTE FDD high priority) and threshold controlled, when close to antenna the UEs will select high band cell, at cell edge the UEs will select low band cell. Table 8.33 gives the cell reselection parameters in FDD‐TDD deployment areas. For FDD‐TDD connected mode mobility strategy, the eNB takes handover decision based on radio criterion or for load balancing or offload reason. The way to manage handover preference toward a same or different frame structure, that is, TDD‐ > TDD versus TDD‐ > FDD. X2 handover, which is shown in Figure 8.89 is recommended within MME pool area for TDD to FDD handover or FDD to TDD handover (TDD and FDD eNB may be mixed in the same pool area). FDD‐TDD carrier aggregation introduced in 3GPP R12 will be based on existing LTE CA mechanisms. TDD cells are typically deployed on higher bands with reduced coverage. The short duration of uplink in many TDD deployments (3DL:1UL) decreases further the coverage, eight receive antennas can improve TDD cell coverage but there is still significant gap to coverage of FDD cells. FDD‐TDD CA allows to keep PCell on FDD with good uplink coverage and only use the TDD downlink for boosting downlink peak rate. From the link budget comparison
D
S
U
D
D
D
S
U
D
D
Secondary cell TDD
D
D
D
D
D
D
D
D
D
D
Primary cell FDD DL
aggregation
aggregation U
U
U
U
U
U
Figure 8.88 Example of FDD‐TDD carrier aggregation.
aggregation U
U
U
U
Primary cell FDD UL
345
346
LTE Optimization Engineering Handbook
Table 8.33 FDD‐TDD cell reselection parameters. Idle in FDD Start measurement CRP RSRP
To FDD
6
Idle in TDD
Cell reselection threshold
Start measurement Cell reselection CRP RSRP threshold
−76 (sIntraSearch:54) N > S + 4 dB
6
Always
FDD > −112
To TDD
5
−114
TDD > −112, FDD S + 4 dB
To 3G
4
−114
3G > −103, FDD −103, TDD carrierFreq
FDD
MME
GW
>PUCCHConfiDedicated
Measurement Report A5
>> tddAckNackFeedback Mode bundling 1 Handover Request
>SIB1 Info >>TDD Config
2 Handover Request Acknowledge 3 RRC Connection Reconfiguration
Transparent Container dl-CarrierFreq
4 SN Status Transfer
>antennaPortsCou nt an2 E-RABs Subject to transfer
5
PDCP Receive Status
RRC Connection Reconfiguration Complete
UL Count Value 6
DL Count Value Path Switch Request
Old eNB X2AP ID New eNB X2AP ID
UE Context Release
Modify Bearer Request
Path Switch Request Ack
Figure 8.89 X2‐based handover procedure between TDD and FDD.
as shown in Table 8.34, it can be seen that UL cell‐edge throughput is around 300 kbps, and there are around 4dB difference, which result in much more TDD sites required.
8.14 Load Balance 8.14.1 Inter‐Frequency Load Balance
As LTE networks are deployed, operators will have multiple combinations of technologies and carriers to serve their subscribers equipped with multi‐mode devices. Operators want to make the best possible use of their spectrum while ensuring the best possible QoS to their subscribers. One way to achieve this is through ensuring proper balancing of the load across its technologies/ carriers.
Table 8.34 Link budget comparison.
Scenario
2600MHz 20MHz TDD
Uplink link budget
Data
UE output power
23
Resource blocks (RBs)
18
Feeder loss Power per RB Thermal noise
10.5 3 0.301
Gains (antenna UL + DL) Jumper loss Body loss Penetration loss Margins (LNF) Max. pathloss unloaded
900MHz 10MHz FDD
850MHz 10MHz FDD
VoIP
Data
VoIP
Data
VoIP
Data
VoIP
Data
VoIP
Data
VoIP
2
7
2
8
2
8
2
7
2
7
2
20
14.4
20
14
20
14.1
20
14.3
20
14.4
20
0.012
0.301
0.012
0.33
0.012
0.325
0.012
0.309
0.012
0.305
0.012
−4
−10.8
−4
−10.8
−4
−10.8
−4
−10.8
−4
−10.8
−4
−10.8
−122.4
−129.2
−122.4
−129.2
−122.4
−129.2
−122.4
−129.2
−122.4
−129.2
−122.4
−129.2
3
0
3
0
3
0
3
0
3
0
3
20 0.2 0 18 6.7 128
Utilization
2%
Interference margin
0
Max. pathloss
1800MHz 10MHz FDD
−174
User bitrate RBS sensitivity
2100MHz 10MHz FDD
0
RBS noise figure SINR
2600MHz 10MHz FDD
128
141.3 0% 13.3 128
132
141.3
131.6
141.3
131.7
141.3
131.9
141.3
131.9
141.3
2%
0%
2%
0%
2%
0%
2%
0%
2%
0%
0
9.4
0
9.8
0
9.7
0
9.5
0
132
132
131.6
131.6
131.7
131.7
131.9
131.9
131.9
Range
0.32
0.41
0.49
0.57
1.12
1.19
ISD
0.48
0.62
0.74
0.86
1.68
1.79
9.4 131.9
348
LTE Optimization Engineering Handbook
Inter‐frequency load balancing aims to move connected UEs to another frequency within LTE established by load relations when or before the serving cell becomes loaded. Inter‐frequency load balancing is used to distribute the load across several carriers, traffic will be managed across the two LTE carriers. Load balancing can occur across cells either within the same eNB or in a neighbor eNB, over X2. It is worth to note that load relations can be automatically and dynamically configured of the optimal inter‐frequency. Cell loading information are provided to L3 from L2 in order to calculate DL GBR, DL non‐GBR, and PDCCH load. There are three steps for load balancing as following. Step1: A4 threshold starting the measurement for offload. Estimated DL/UL PRB usage per cell is updated by new admitted requests and periodic modem PRB usage reporting. In addition to the parameters above, the measurement object and Report ConfigID for the inter‐frequency load balancing using A4 threshold mechanism need to be defined in the eNB. For the load information exchange between eNB is over the X2 interface. Load balancing is triggered when current PRB usage exceeds a configurable percentage of total PRBs available in the cell. Step2: UE selection, which means a number of UEs are selected as candidates for load balancing. UE candidates for offloading must support inter‐frequency handover, UE candidates start with lowest priority for offloading. Step3: Execution of offloading to target carrier. MME selects unloaded target carrier for UE measurements, based on X2 resource status reporting information. For equal bandwidth carriers the handover margin can be decided and tuned based on load balancing, for example, for 800M and 1800M layers with 10MHz bandwidth, a3 offsets need to be set asymmetrically, to prevent skewed load. If the carriers have different bandwidths, the throughput difference between layers should be taken into account as well (Figure 8.90). Load balancing can also be triggered when load in source cell (SC) is above the pre‐configured load threshold, and the load difference is larger than the pre‐configured load difference threshold as shown in Figure 8.91. An load balance magnitude to be determined for each target cell (TC), suitable UEs need to be moved (selection/reselection) in order to meet the determined magnitudes. 8.14.2 Inter‐RAT Load Balance
The feature of inter‐RAT load balance is applied for reducing the risk of UE trapped in over‐ utilized LTE cells unable to provide acceptable performance and reducing the risk of depleted 3G cells when the ratio of 3G only UE declines. The load balancing features measure the traffic
Establish load relations
Exchange load info.
Load status
Select candidates UEs
Step1
Step2
Load balance action (handover)
New load status
Step3
Figure 8.90 Load balance procedure.
If load_different> load_different_thresh, load balancing triggered!
load_SC load_TC2
Source cell
Target cell2
Figure 8.91 Load balance strategy (example).
load_different load_TC1
Target cell1
Mobility Optimization
F1
LTE cells
F1
Double arrows are IFLB relations; single arrows are inter-RAT relations;
F2 Frequency
F3
3G cells
Distance
Figure 8.92 IRAT load balance strategy (example).
load in each LTE cell with different frequency and 3G cell. LTE cells with inter‐frequency load balance (IFLB) relations exchange traffic load information and also need to monitor own traffic load versus an 3G offload threshold. Figure 8.92 depicts the IRAT load balance strategy. Inter‐frequency load balance attempts to distribute traffic evenly between overlaid LTE inter‐frequency cells that cover the same area. Inter‐RAT offload attempts to offload LTE traffic above an offload threshold to 3G cells that cover the same area. The trigger for the report of measurement results is a B1 event (neighbor becomes better than threshold) for IRAT load balance. The B1 event represents acceptable coverage for offload to the 3G target cell. Reactive load control is triggered when congestion condition is detected during call/bearer admission. Purpose of reactive load control is to move some UEs to less loaded inter‐freq/ inter‐RAT carriers or release some UE/bearers to relieve the congestion condition. Congestion conditions include number of calls exceed limits, number of modem contexts exceed, number of data bearers exceed limits, and number of PRB consumptions exceed limits. 8.14.3 Load Based Idle Mode Mobility
Load balancing aims to move connected UEs to another frequency within LTE when or before the serving cell becomes loaded. Cell load is determined by the total UL/DL PRBs consumption. The idle mode load distribution between layers helps reduce the need for the connected mode load balancing. The idle mode distribution is possible to adjust with the thresholds and by that the load distribution. Basic load management should primarily be done through steering of UEs between frequency layers in idle mode. Good idle mode distribution can be achieved by setting a higher cell reselection priority to the higher frequency band, cell‐center UEs will camp on the higher band and cell‐edge UEs will camp on the lower band, distribution can be controlled through parameter threshServingLow. For equal priority based cell reselection, it needs to increase qOffsetCellEUtran for reselection to non‐congested neighbor cells, or decrease qOffsetCellEUtran for reselection to high capacity cell. Priority‐based cell reselection aims to ecourage easier cell reselection in idle mode. If inter‐ frequency carrier/inter‐RAT frequency is with cell reselection priority higher than the serving frequency, then decrease threshXHigh for that frequency relation. If inter‐frequency carrier/ inter‐RAT frequency is with cell reselection priority lower than the serving frequency, then decrease threshXLow for that frequency relation (Figure 8.93).
349
350
LTE Optimization Engineering Handbook RSRP Higher priority Lower priority
Lower priority
Higher priority RSRP threshold + UE measurement uncertainty UEs camping on blue cell
UEs camping on orange cell
Cell radius
Figure 8.93 Load‐based idle mode mobility.
1 Srxlev > 0
UE starts Inter freq meas (and iRAT) for equal priority cell reselection
RSRP Serving Cell
4
Eutran cell is reselected with other freq
RSRP equal priority eUTRAN cell
tReselectionEUTRAN or tReselectionRAT
3
Qrxlevmin(SIB1) +Qrxlevminoffset(SIB1) +Pcompensation(SIB1) +sNonIntraSearch(SIB3)
Automatically adapt SIB3 cell reselection parameters based on serving cell load.
Rn > Rs
qOffsetCell
LTE equal priority cell - criteria not met
2
Qrxlevmin(SIB5) +Qrxlevminoffset(SIB1) +Pcompensation(SIB5)
Traditional “S” criteria
UE starts ranking cells qHyst
time qrxlevminoffset = 0 if same PLMN
Figure 8.94 Load‐based adaptation of cell reselection thresholds.
Load‐based adaptation of cell reselection thresholds can push cell‐edge UEs down to lower‐ priority layer, as shown in Figure 8.94. This is able to provide the capability to adjust cell selection settings according to cell load, thus helping to balance traffic and load among cells. Load information exchange over X2 interface, allowing the serving eNB to gather load information from the neighboring cells through the X2 interface. Automatically adapt SIB3 cell reselection parameters based on serving cell load, which used to favor idle UEs cell reselection to less loaded neighbour cells, when serving cell is heavily loaded. This feature can be used to impact intra‐frequency, inter‐frequency and/ or IRAT reselection. Here is an example for F1 (high priority) and F2 (low priority) inter‐frequency reselection. Operator can specify the value to use at different load levels for the following parameters: q‐ Hyst, threshServingLow, threshXLow, threshXHigh, s‐nonIntraSearch, s‐nonIntraSearchP, and s‐nonIntraSearchQ, and so on, which are shown in Table 8.35. According to the above parameters settings, downlink/uplink average and edge throughput, the ratios of UE distribution in F1 and F2 band are shown in Table 8.36.
Mobility Optimization
Table 8.35 Parameters settings for different load levels. Reselection
IFHO
F2‐ > F1
F1‐ > F2
F2‐ > F1
F1‐ > F2
F1‐band, high priority Load ratio: (F1:F2 = 6:4)
threshXHigh = −92
sNonintraSearch = −90, threshServingLow = −93, threshXLow = −125,
a2 = −90, a2 = −83, a5−1 = −80, A5−1 = −93, a5−2 = −92 A5−2 = −125
F1‐band, high priority Load ratio: (F1:F2 = 95:5)
threshXHigh = −102
a2 = −100, sNonintraSearch = −100, a2 = −68, threshServingLow = −103, a5−1 = −65, A5−1 = −103, a5−2 = −102 A5−2 = −125 threshXLow = −125,
F1‐band, high priority Load ratio: (F2:F1 = 99:1)
threshXHigh = −105
a2 = −103, sNonintraSearch = −103, a2 = −65, threshServingLow = −106, a5−1 = −62, A5−1 = −106, a5−2 = −105 A5−2 = −125 threshXLow = −125,
F2‐band, high priority Load ratio: (F2:F1 = 4:6)
sNonintraSearch = −86, threshXHigh = −87 threshServingLow = −88, threshXLow = −102,
When F1, F2 with equal priority
qHyst = 1, qOffsetFreq = 3,
a2 = −83, a2 = −85, a5−1 = −88, a5−1 = −86, a5−2 = −102 a5−2 = −87 a2 = −85, a2 = −83, a5−1 = −86, a5−1 = −88, a5−2 = −88 a5−2 = −86
qHyst = 1, qOffsetFreq = 3,
Table 8.36 Performance of different parameters settings. DL/UL avg THP (Mbps)
DL/UL edge THP (Mbps)
HO
IFHO
Ratio in F (idle)
Ratio in D (idle)
F1‐Hi prio −1 (60%; 40%)
18.9/9.3
3.9/0.05
176
61
42%
58%
F1‐Hi prio −2 (95%; 5%)
18.6/13.9
4.6/4.9
118
7
11%
89%
F1‐Hi prio −3 (99%; 1%)
18.0/14.6
4.7/7.1
106
1
4%
96%
F2‐Hi prio −1 (60%; 40%)
22.2/8.8
4.4/3.5
203
101
57%
43%
F1, F2 with equal priority
21.7/6.6
5.2/0.05
107
20
90%
10%
8.15 High‐Speed Mobile Optimization The higher the velocity that the UE experiences, the more severe the effect of fast fading that the system suffers. Therefore, it is more difficult to achieve the same performance in a high‐ speed scenario as in a normal speed one. 3GPP has defined high‐speed and performance limits in case of up to 350km/h UE speed. Fast‐moving UEs cause Doppler shifts (frequency offsets) in the received uplink signal as shown in Figure 8.95. Since the UE synchronizes to a Doppler‐ shifted signal in downlink, and in uplink it will be roughly doubled, the Doppler shift is proportional to the velocity of the UE and to the carrier frequency. In this case the eNB uplink signal receiver suffers from a huge Doppler shift, which causes severe performance degradations.
351
352
LTE Optimization Engineering Handbook
It is needed to estimate the Doppler spread of the channel, i.e. rate at which the amplitude and phase of the received signal changes. Doppler spread occurs when there are many signal paths with different Doppler shifts, which are added constructively or destructively at the receiver antennas.
UE travelling with speed v
Maximum Doppler Shift θ
Ds/2
Doppler shift + ∆f Minimum Doppler Shift Doppler shift – ∆f
Dmin
eNodeB A Maximum Doppler Shift
Doppler shift + ∆f Ds/2
eNodeB B
Perfect frequency sync
Frequency offset
Railway track
Figure 8.95 Doppler shift.
Frequency offsets rotate the received signal constellation. If it is too high, the constellation has rotated more than 360 degress and the receiver can’t resolve which frequency offset it is. Scenarios with Doppler shift are mainly line‐of‐sight scenarios where the mobile has a velocity component into the direction of the eNB. Typical scenarios refers to high speed trains where the eNB is mounted near to the tracks or highways where the cars are driving fast and the eNB is located quite near to the highway. The formular of maximum Doppler shifts (fd) and Doppler shift fs(t) are: f cu
fd
f cu ,
f cd
v c
2.57 109 2.69 109
350 / 3.6 1705 Hz, 3.108
fs t
f d cos
f cd
t
where are UL and DL highest carrier frequencies. v is the UE travelling speed, c is the speed of light, and θ(t) is the signal arrival angle shown in Figure 8.96. The formular is that the UE already sees a shifted signal in DL and does frequency synchronization on that and then transmitting in UL the eNB sees roughly twice the shift. Besides, when a UE approaches and passes by the cell antennas will cause a quick change in the frequency offset, that is when a UE moves first toward the eNB and then moves away from it causing positive and negative Doppler shifts respectively. It is difficult to track and compensate for large and fast varying frequency offsets of the UE in this scenario. The effect of the frequency offset is that the subcarriers are no longer orthogonal since the side lobes of the interfering subcarriers no longer has a zero crossing at the main lobe of the Reference symbols Subframe (1 ms) 1 2∆T
≈
1000 Hz on PUSCH 1750 Hz on PUCCH
PUSCH symbols:
∆T = 0.5 ms
Figure 8.96 Frequency offset compensation for PUSCH/PUCCH.
Mobility Optimization
desired subcarrier. High‐frequency offsets cannot be correctly estimated with the frequency offset estimators on PUSCH and PUCCH in LTE. Time difference between reference symbols, ΔT, gives aliasing at frequency larger than 1/2ΔT, which is shown in Figure 8.96. For PUCCH the frequency offset will affect the orthogonality of the orthogonal cover used for format 1/1a/1b (the orthogonality is used to distinguish different users signal using the same cyclic shift). When the frequency offset is severe, these users will disturb with each other. When the frequency offset is very high, users using different orthogonal cover may be impossible to separate, that is, the orthogonal cover will look the same at the receiver. To be able to maintain a connection at high Doppler, the eNB must be sure that the UE transmits the periodic CQI at every occasion. DRX cycle must be set according to CQI period, to ensure the eNB receives CQI periodically. 8.15.1 High‐Speed Mobile Feature
In order to estimate and compensate for high frequency offsets, the high‐speed UE feature is introduced in a live network. To mitigate problems in high‐speed deployments eNB shall support a Doppler shift estimation for each UE based on the received signal in different uplink physical channels (PRACH, PUSCH, and PUCCH) and signals (SRS). Those estimates shall be provided to RRM. Some features deal only with enhancing link level performance for high speed users by intoduction of eNB UL receiver improvments (e.g., Doppler shift estimation). For example, antanna tuning‐tilt for high‐speed scenario is calculated below:
The tilt of antenna á atan H /D * 360 / 2 *
b/2 e _ tilt
where H = antenna height – rail height, D = coverage range, b = beam width. For high‐speed mobile UE, the mobility characterization will suffer the issues such as frequent handover, ping‐pong handover, high handover failure rate and drop rate, group handover, and signaling congestion. It is suitable to use frequency diversity mode rather than frequency‐selective scheduling, or transmit diversity rather than spatial multiplexing for a UE at a high speed. Distinguish high‐speed UE via speed estimation by layer1 or layer 3 method, for L1 method, Doppler frequency estimation is used, that will cost 100 ms for estimation. For L3 method, according to v = d/t, UE determines mobility state based on the number of cell changes which occur within a defined period. Judge UE moving direction and set different priorities to forward cell and backward cell as shown in Figure 8.97, which is benefit in accelerating handover and alleviating ping‐pong, for example, CIO = 3 dB for forward cell, CIO = 0 dB for backward cell as below configuration. ●● ●●
If UE is moving from west to east:1‐ > 2, 2‐ > 3, 3‐ > 4, CIO = 3; 2‐ > 1, 3‐ > 2, 4‐ > 3, CIO = −3 If UE is moving from east to west:1‐ > 2, 2‐ > 3, 3‐ > 4, CIO = −3; 2‐ > 1, 3‐ > 2, 4‐ > 3, CIO = 3
Besides, railway network plan is usually adpot the “Z” plan shown in Figure 8.97, that can overcome multipath fading. Combined cell shown in Figure 8.98 is another important feature in high‐speed scenario, which configures multiple sector carriers (RRUs) to belong to same cell. All sector carriers are considered as one logical cell, with same PCI, same CRS, and system information as macro cell to fit to light load. Drive test results showing throughput gain when using combined cell compared to separate cells. It is still worth noting that downlink power allocations in terms of reference signal boosting can cause unnecessary large coverage areas and handover execution problems, so typically RS boosting is avoided for the high‐speed scenarios, and uplink power settings with full pathloss compensation while maximizing the signaling robustness is also suggested.
353
354
LTE Optimization Engineering Handbook
Cell 4
Cell 3
Cell 2
Cell 1
East
West Cell 3
Cell 2
Cell 1 earlier
Cell 4
earlier
earlier
East
West Cell 3
Cell 2
Cell 1 earlier
earlier
Cell 4 earlier
Cell B
Cell A
Cell D
Cell C
Figure 8.97 UE movement and eNB site planning.
8.15.2 Speed‐Dependent Cell Reselection
Cell reselection performance optimization is as similar as handover. The main idea is that speed up the cell reselection in order to have the UEs always under the best cell, ping‐pong cell re‐selections need to be avoided by, for example, speed‐dependent cell re‐selection feature. The speed‐dependent scaling of cell reselection criteria is used to influence the cell reselection criteria for fast‐moving UE. It helps the UE to respond more quickly to cell changes when moving at high speed. A UE may enter three different mobility states: normal mobility, medium
Mobility Optimization Separate Vs Combined cell DL Throughput
70 60
Throughput [Mbps]
Combined Cell 50 40 30 20 Separate Cells 10 0
0
5
10
15 Time (s)
20
25
30
Separate Vs Combined cell DL Throughput Combined Cell
Throughput [Mbps]
20
Separate Cells
10
0
0
5
10
15 Time (s)
20
25
30
Figure 8.98 Combined cell.
mobility, and high mobility as shown in Figure 8.99. The medium and high states are specified by the related parameter nCellChangeMedium and nCellChangeHigh resepectively defining number of cell reselections within sliding time window tEvalution [240sec] that determines the UE shall enter mobility states medium or high. High‐mobility state criteria is detected if number of cell reselections during time period tEvaluation exceeds nCellChangeHigh. Medium mobility state is detected if number of cell reselections (n) during time period tEvaluation exceeds nCellChangeMedium and does not exceed nCellChangeHigh. Consecutive cell reselections between two cells are not taken into account: the UE does not count consecutive reselections between the same two cells into mobility state detection criteria if same cell is reselected just after one other reselection.
355
356
LTE Optimization Engineering Handbook NCRItEvaluation
nCellChange High
nCellChange Medium
tHystNormal
Enter Enter “Medium” “High”
Enter “Medium”
Enter “Normal”
normal
time
tEvaluation
medium
nCellChgHigh > n > nCellChgMed high
n > nCellChgHigh
Figure 8.99 Mobility states determination of a UE.
8.15.3 PRACH Issues
High‐frequency offset impact on PRACH also causes problems. Random access preambles are generated from same root sequence using cyclic shifts, on the receiver, the detector correlates the received signal with the predefined preambles. Frequency offsets will cause correlation peaks for other cyclic shifts of the same sequence and too high‐frequency offsets causes the original peak to be entirely replaced by another peak, thus may cause false detection as shown in Figure 8.100. Fortunately, high‐speed detection can be realized through this PRACH characteristics. If the weighted power sum caused by Doppler shift and spread is greater than a given threshold, the UE is high speed. Impact of high frequency offset
f = 0 Hz f = 625 Hz
Main peak
Additional peak f = 1250 Hz
Figure 8.100 Impact on PRACH.
Mobility Optimization UE experiences higher frequency offset due to Doppler shift which causes false and aliased peaks.
Main peak cyclic shift y Main peak cyclic shift x
Detection interval x
the additional peaks in a matched filter are handled by extended detection interval
Extended detection interval x
Detection interval y
Additional peak cyclic shift x
Cyclic shift y removed in restricted set
Figure 8.101 Unrestricted set and restricted set.
Table 8.37 Supported cell ranges depending on restricted set of cyclic shift length (delay spread = 5.2us). Ncs Config index
Ncs, cyclic shift length
0
15
length of single cyclic shift, μs
9.1
km
1.4
Ncs Config index
8
Ncs, cyclic shift length
68
1
18
12
1.8
9
82
2
22
15.8
2.4
10
100
length of single cyclic shift, μs
59.6
km
9.0
73
11.0
90.2
13.5
3
26
19.6
2.9
11
128
116.9
17.5
4
32
25.3
3.8
12
158
145.5
21.8
5
38
31
4.7
13
202
187.4
28.1
6
46
38.7
5.8
14
237
220.8
33.1
7
55
47.2
7.1
The high‐speed UE feature applies restricted set of cyclic shifts on PRACH (i.e., the RRC parameter highSpeedFlag is set to true), which restricts the number of cyclic shifts that can be used for each sequence to make room for higher‐frequency offsets and allows for extended detection intervals including additional peaks in matched filter output to ensure a properly working PRACH detector at high Doppler shifts. The restricted set of cyclic shifts choose a set of cyclic shifts that allows for secondary peaks, the probability of false detections of the preambles is reduced. In this case, three detection intervals are needed, one for the main peak and one for each of the secondary peaks, compared to the unrestricted set where only one detection interval is needed. This means that at least three times more RACH root sequences are required for a given number of preambles. This increases the number of root sequences needed for PRACH in a cell, but improves the PRACH detector performance and makes it possible to estimate the frequency offset on PRACH. In general, the restricted set can reduce in main peak of matched filter output that does not cause missed detection and additional peaks in matched filter output of receiver will not cause false detection (Figure 8.101). The restricted set preambles are generated by masking some of the cyclic shift positions in order to retain acceptable false alarm rate while maintaining high detection performance for very‐high‐speed UEs. After frequency offset estimation from preamble detector, the result will be sent to PUSCH receiver for compensation. The supported cell ranges for restricted set are shown in Table 8.37.
357
LTE Optimization Engineering Handbook (RACH root sequence) Range: 0...837 Default: 0
(PRACH configuration index) Range: 3...53 Default: 4
Example for SIB2 extract: prach-Config { rootSequenceIndex 264, prach-ConfigInfo { prach-ConfigIndex 3, highSpeedFlag TRUE, zeroCorrelationZoneConfig 9, prach-FreqOffset 4 } },
(PRACH high speed flag) Range: 0 (false), 1 (true) Default: 0
(PRACH cyclic shift) Range: 0...15 Default: 1 (PRACH frequency offset) Range:: 0...94 Default: –
Figure 8.102 Example for SIB2.
In a live network, UL and DL Doppler shift might not a big problem if eNB have “High‐speed UE” feature, but Doppler spread will degrade UL and DL performance. Cell merge will outperform non‐cell merge because energy overlay gain over bi‐direction Doppler loss. The parameter highSpeedFlag can be got from SIB2 and RRC connection reconfiguration message. Figure 8.102 gives an example of high‐speed PRACH parameters in SIB2. 8.15.4 Solution for Air to Ground
Analyze how is LTE standard affected by Doppler shift in case where UE is moving with speed that airplane can reach (relative to the ground), maximum cell size that can be achieved (if currently supported 100km is not enough) and link budget for that solution, interference with services in neighboring frequency bands, and required output power, and so on.
190
Pathloss according to the empirical propagation model (EPM-73) fc = 1 GHz
hRBS = 100 m hAirplane = 10000 m
180
fc = 2 GHz fc = 5.8 GHz
170
Path loss (dB)
358
160
fc = 5.8 GHz
150
fc = 2 GHz fc = 1 GHz
140 130 120 110 100
0
50
100
150 200 Distance (km)
Figure 8.103 The cell range of air to ground coverage.
250
300
Mobility Optimization
Cell size is around 150 to 250 km, which will bring large round trip times (1 ms for 150 km cell radius), and impact on PRACH detector, scheduling, time advance commands, signaling, and so on. For transmission gap in TDD, there needs to be larger than maximum round trip time. High Doppler shift, Downlink (from base station to airplane):
fd
f cdl
v c
1 kHz for f cdl
fd
1 GHz
Uplink:
fd
f cul
f cdl
v c
fd
2 kHz for f cul
f cdl
1 GHz
Path loss and range High altitude for airplanes is the line‐of‐sight propagation and long reflection region. The cell range depends on carrier frequency as shown in Figure 8.103, for 140 dB path loss: 127 km @ 1 GHz, 59 km @ 2 GHz, 22 km @ 5.8 GHz.
359
360
9 Traffic Model of Smartphone and Optimization 9.1 Traffic Model of Smartphone Network traffic is mainly driven by smartphone applications, today’s smartphone applications mostly generate relatively small data bursts. It is important to understand the traffic characteristics for the network optimization purposes. Some years ago traffic was, if not homogeneous, at least a bit easier to predict. For many years, for example, voice has been an important service with very predictable behavior. Today, with the advent of mobile broadband there is really no such service, that is, a service with very predictable behavior that dominates data transfer. Instead, many use mobile broadband all the time to be able to remain connected to their favorite social network. This connection is often done using apps that may be designed without consideration of cellular network characteristics. The nature of app traffic is unpredictable. Also, no two users probably have the same user patterns (update frequency depends on how many friends you have, what they do, OS, etc.). Smartphone’s trigger huge number of small packet in LTE network, always‐on applications require keep‐alive message, frequent transitions between LTE RRC states and causing signaling increase. Numerous applications require that an always‐on mobile‐broadband experience is seamlessly delivered and presented to the end user. Furthermore, many applications may be designed without specific consideration of the characteristics of cellular networks, and consequently may exhibit traffic profiles not well suited to those connections. When attempting to provide such always‐on connectivity at the RAN level, trade‐offs are often encountered between UE power consumption, user experience, data transfer latency, network efficiency, and control plane signaling overhead. Furthermore, the optimum trade‐off point may vary according to application characteristics, or their activity or status. Current trends indicate that the above issues will only increase in significance over the coming years. It is imperative, therefore, that the ability of LTE to efficiently handle and manage such traffic is continually improved. On the other hand, applications characterization drives understanding of demand fundamentals, for example, web page size, packet size, busy hour attempts, avarage call throughput, signaling, and bearer traffic generated by application (Figure 9.1). This control/user plane activity daily profile shows significant differences for different terminal types. Normally, 30% to 60% of all smartphone and M2M terminals are practically “always on,” having user plane activity during all hours of the days. This is due to periodical reporting activity for most M2M terminals and regular background activity for most smartphone terminals. PC terminals are usually not “always on,” but their usage is mostly bound to actual user interaction, for example, web browsing or social networking. Therefore, user plane activity for PC terminals differs significantly from that of smartphones and M2M. LTE Optimization Engineering Handbook, First Edition. Xincheng Zhang. © 2018 John Wiley & Sons Singapore Pte. Ltd. Published 2018 by John Wiley & Sons Singapore Pte. Ltd.
Traffic Model of Smartphone and Optimization Session Count
Session Count CDF
Total Load CDF
100%
Session Count
80% 60% 40% 20%
1K
10K
0%
100K 1M 10M Session Payload (bytes)
100 90
Control plane
80
Smartphone-centric network
70
Mixed network
60 50 40 30 20
PC-centric network
10 0
0
10
20
30
50 60 40 User plane
70
80
90
100
Avarage call throughput Video Streaming
Interactive Video
Web Browsing
Audio Streaming Email Gaming MMS
SMS
Busy hour call attempts
Figure 9.1 Model reflects variability of mobile application usage.
In such traffic model scenario, the maximum number of active users depends on the capacity licenses purchased by an operator. In the case, the maximum number of users is starting to limit user traffic inactivity timer that has mentioned before can be tuned to release inactive users faster. The parameter dictates the time each UE stays connected without downloading or
361
362
LTE Optimization Engineering Handbook Connected Users
90
70
80
60
70
50
60 50
40
40
30
30
20
20
10
10 0 21-01-14 0:00
0 22-01-14 0:00
23-01-14 0:00
Users_connected
24-01-14 0:00
tInactivityTimer
RRC Attempts
7000
70
6000
60
5000
50
4000
40
3000
30
2000
20
1000
10
0 21-01-14 0:00
0 22-01-14 0:00
23-01-14 0:00
LTE_02_RrcSetup_Att
24-01-14 0:00 tInactivityTimer
Figure 9.2 When inactivity timer decreased, users reduced, RRC attempts increased.
uploading data before it is shifted to idle mode. Reducing the value from 61 to 10 seconds can increase signaling load due to increased number of RRC setup attempts but can also reduce the drop rate as less number of users are connected unnecessarily with a chance of dropping the connection as shown in Figure 9.2. This can also reduce congestion in the system due to limitation of connected user’s license in the network. By decreasing inactivity timer, users can be moved earlier from active state to idle state. However, this will have impact on user experience, as there are more frequent session establishments in the case of new data transmission. By changing inactivity timer, an operator can compromised between fast‐release system resources due to inactivity and impact on user experience due delays caused by frequent session establishments. In addition, quality of service (QoS) mechanism controls the service performance, reliability and usability of a mobile service, because different bearer traffic requires different QoS. The following part is about QoS mechanism. 9.1.1 QoS Mechanism
The policy control routing function (PCRF) deploys a set of operator‐created business rules that dictate the dservice QoS strategy of the evolved packet system, which is shown in Figure 9.3.
Traffic Model of Smartphone and Optimization
External IP Networks
IP connectivity
PGW
Gx
IP connectivity
S5/S8
PCRF
SGW QoS parameters: S11
QCI/ GBR/ MBR
IP connectivity S1
MME
eNodeB
IMS Layer MMTEL App
SIP
P-CSCF Rx
Access Layer
PCRF Gx
3GPP Bearer Protocols
PGW
To Request / Grant Access QoS for handling the session media To Request / Get feedback on Loss of Access Qos except handling for Emergency / priority Calls
Figure 9.3 LTE QoS mechanism, Rx and Gx interface.
The PCRF communicates with the PGW data and push appropriate policy rules across the Gx interface. These QoS requirements can be passed by the PGW to the SGW over the S5/S8 interface. The SGW communicates these requirements to MME over the S11 interface. Finally the MME communicates the QoS requirements to the eNB over the S1‐MME interface. The QoS requirements of the EPS bearer are implemented in the radio network with the UL/DL scheduler and transport network with IP differentiated services code point (IP DSCP) and Ethernet priority bits (Pbits). In LTE network the service QoS is implemented between UE and P‐GW. The QoS framework for UL/DL basically consists of the following building blocks, a radio network, a transport network, the core network, configuration management system, the scheduler for UL/DL data inside the eNB, and the function to control the whole QoS handling. LTE QoS architecture is defined in TS 23.402 and 23.203 standards, UE negotiates its capabilities and expresses its QoS requirements during a SIP session setup or session modification procedure. QoS generally refers to the following parameters, QoS class identifier (QCI). VoLTE requires to support
363
364
LTE Optimization Engineering Handbook UL Service Data Flows
EPS Bearer QoS
DL Service Data Flows
Service Requests
- GBR: QCI, ARP, GBR, MBR
Packet Filter
- Non-GBR: QCI, ARP PCRF Radio Bearer L1/L2 configuration L1/L2 configuration UE-AMBR UE
UE-AMBR eNB
APN-AMBR PDN GW
Gx
Figure 9.4 SDF.
dedicated service data flows (SDFs1) assigned with desirable QCI for VoIP SIP signaling and RTP packet. In LTE, services are mapped to SDF as shown in Figure 9.4, several SDFs can be treated as an SDF aggregate, all SDFs in an SDF aggregate must have the same QCI/ARP. SDF aggregates mapped to UL/DL EPS bearers which uniquely defines EPS bearers. Traffic flow template (TFT) defines rules so that UE and network knows which IP packet should be sent on particular dedicated bearer. An UL TFT in the UE binds an SDF (or SDF aggregate) to an EPS bearer in the uplink direction, a DL TFT in the SGW binds an SDF (or SDF aggregate) to an EPS bearer in the downlink direction. Every dedicated EPS bearer is associated with a TFT to classify the packet. It usually has rules on the basis of IP packet destination/source or protocol used. The TFT can allow all packages or filter on specific IP address and/or TCP/UDP port, which is shown in Figure 9.5. UE negotiates its capabilities and expresses its QoS requirements during a SIP session setup or session modification procedure, like media type and bit rate, direction of traffic, packet size, packet transport frequency, usage of RTP payload for media, types, and bandwidth adaptation etc. The QoS concept as used in LTE networks is class‐based, where each bearer type is assigned one QoS class identifier (QCI) by the network to ensure proper QoS for bearer traffic in LTE networks. QCI is used to determine packet forwarding treatment, it can be used to mark packets with DSCP. 3GPP has standardized 9 QCI values and mapping to resource type (GBR, non‐GBR), priority, packet delay budget and packet error loss rate (Table 9.1). Besides QCI, allocation and retention priority (ARP) is used to decide whether bearer establishment or modification request can be accepted in case of resource limitations, ARP can also be used to decide which bearer(s) to drop during resource limitations, and has no impact on packet forwarding treatment. The ARP allocated to both the default bearer and the IMS signaling bearer should have a high value this minimize the risk of deactivation of these two bearers that should be always established until the UE disconnects or detaches from the networks. For video telephony, voice bearer has a higher ARP, and video to another bearer has a lower ARP. In a congestion situation, the eNB can then drop the video bearer without affecting the voice bearer. This would improve service continuity. Linked EPS bearer ID (L‐EBI): each dedicated bearer is always linked to one of default bearers. L‐EBI tells dedicated bearer which default bearer it is attached to. (L‐EBI = EBI) The above description is shown in Figure 9.6. Currently a QoS mapping do exist today in both 3G and LTE but is not aligned between the two different radio access technologies. 3GPP TS23.401 has defined mapping between standardized Rel 8 QCIs and pre‐Rel 8 QoS parameter values. 1 SDF: An aggregate set of packet flows associated with an application service. In case of the VoIP service, the SIP based IMS signaling and the RTP based voice media flows together build the service data flow.
UE
PGW
TFT
Bearer #1
TFT
TFT
Bearer #2
TFT
TFT
Bearer #3
TFT
Figure 9.5 TFT. Table 9.1 Standardized QCIs. Packet delay QCI Resource type Priority budget
Packet error loss rate
Example Services
1
2
100 ms
10−2
Conversational voice
2
4
150 ms
10
Conversational video (Live streaming)
3
5
300 ms
GBR
4
−3
10−6 10
−3
3
50 ms
1
100 ms
10−6
6
7
100 ms
10
7
6
300 ms
8
8
9
9
5
Non GBR
−3
10−6
Non‐conversational video (Buffered streaming) Real time gaming IMS signaling Voice, video (live streaming), interactive gaming Video (Buffered streaming) TCP‐based (e.g., www, e‐mail, chat, ftp, p2p file sharing, progressive video, etc.)
Notes: 1 Packet delay budget (PDB), one-way between UE and gateway. 2 Packet loss rate (PLR) is only air loss counted. 3 VoIP is a GBR service, which means that VoIP users are prioritized against MBB users in the same network. 4 Default bearer is set up with QCI 9 (for non-privileged users) or QCI 8 (for premium users). 5 Guaranteed bit rate (GBR), the minimum guaranteed bit rate per EPS bearer. Specified independently for UL and DL. 6 Maximum bit rate (MBR), the maximum guaranteed bit rate per EPS bearer. Specified independently for UL and DL. 7 APN-aggregate maximum bit rate (A-AMBR), limits the aggregate bit rate that can be expected to be provided across all non-GBR bearers and across all PDN connections of the APN. Specified independently for UL and DL. 8 UE-AMBR limits the aggregate bit rate that can be expected to be provided across all non-GBR bearers among APNs of a UE. Specified independently for UL and DL
366
LTE Optimization Engineering Handbook QoS of EPS bearers
Allocation and retention priority (ARP)
Qos class identifier (QCI.0–255)
Priority
Resource type
Packet loss rate
Packet delay budget
Non-GBR
Default Bearer QCI 5– 9 A-AMBR UE-AMBR APN IP address ARP
Priority level (0–15) (Per bearer)
Pre-emption capability (shall not trigger preemption, may trigger pre-emption) (Per bearer)
GBR
Dedicated Bearer
ARP used for call admission control
Dedicated Bearer
QCI 5–9 A-AMBR UE-AMBR TFT ARP L-EBI
Pre-emption vulnerability (not pre-emptable, pre-emptable) (Per bearer)
QCI 1–4 AMBR GBR TFT ARP L-EBI
Figure 9.6 QoS profile. MME
eNB
UE
INITIAL CONTEXT SETUP REQUEST (E-RAB Level Qos parameters > QCI, ARP) QoS Mapping RNL / TNL Scheduler: Priority / LCG TNL: DSCP rrc Connection Reconfiguration Includes pdcp-SN-Size, UM/AM, snFieldLength, t-Reordering, logicalChannelConfig > priority, and logicalChannelGroup
LCG mapping example
LCID:1
LCID:2
LCID:x
LCID:y
LCID:z
LCID:w
LCID:q
SRB1
SRB2
QC11
QC15
QC17
QC18
QC19
LCG0
LCG1
LCG2
LCG3
rrcConnectionReconfigurationComplete INITIAL CONTEXT SETUP RESPONSE
Figure 9.7 Uplink QOS mapping.
For uplink, the scheduling of the traffics is important to set correct priority per QCI, and also allocate the bearers into logical channel groups (LCG) in such way that the intended scheduling strategy will be used in the UL. LCG mapping in eNB and associated signaling are depicted in Figure 9.7. 9.1.2 Rate Shaping and Traffic Management
Wireless networks bandwidth changes constantly and it becomes important to support traffic shaping or bit rate adaptation to guarantee high QoE. Traffic shaping provides a mean to control the volume of traffic being sent into a network in a specified period (bandwidth throttling), or the maximum rate at which the traffic is sent (rate limiting). This control can be accomplished in many ways and for many reasons; however, traffic shaping is always achieved by delaying packets. Rate shaping in EPS is described in 3GPP TS 23.107 as the so‐called token bucket algorithm. The rate shaping function can be seen as a virtual bottleneck that throttles
Traffic Model of Smartphone and Optimization
Token rate, r
Bucket size, b Token bucket
Token, representing the allowed data volume
Incoming packets
R
Buffered before sent to next link Incoming E-RAB Requests Admission Control
Max Threshold Min Threshold
Incoming Rate (Rin)
Active Queue Management Queue length > Max: Always drop Min < Queue length < Max: Some drops Queue length > Min: Never drop
AMBR/MBR
Rate Shaping
Rate Shaping according to a token bucket, based on AMBR or MBR (ΣRx ≤ AMBR or R ≤ MBR) Rate after packet shaping (Rshaped)
Traffic rate
Scheduling
Figure 9.8 Rate shaping.
the output rate of that buffer to a certain peak burst rate and to a certain average bit rate. The tokens are consumed by the data packets, if there are at least as many tokens in the bucket, the packet is transmitted. This is the nechanism that shaping the aggregate traffic for a user to the aggregate maximum bit‐rate (AMBR). First is admission control for E‐RAB requests. For non‐GBR type E‐RABs admission control is simplified since packets are forwarded on a best‐effort basis. However, GBR‐type E‐RABs requires more elaborate admission control. Request is rejected if the request exceeds available resources. Admission control may consider many different types of resources (shared channel, control channel, transport network resources, memory, etc). After that, active queue management (AQM) is working, which can be viewed as packet‐ level congestion control. It can, for example, drop packets to indicate a congested state to the source at an early stage. GBR queues handled by AQM are not congested, and the scheduler has time enough to serve the queues. Then is data rate shaping concept which function is delaying packets of the data flow so they conform to MBR or UE‐AMBR, using a token bucket as shown in Figure 9.8.
367
368
LTE Optimization Engineering Handbook
Traffic Class Signalling Indicator Traffic Handling Priority
Quality of Service Class Identifiers
Transfer Delay SDU Error Delay GBR
GBR
MBR / AMBR
MBR
Allocation Retencion Priority
Allocation Retencion Priority ..
... Pre-Rel8 QoS per bearer
AMBR
Rel8 QoS per bearer
Rel8 QoS per UE
Set by PDN-GW per bearer
Set by HSS per UE
QCI (Quality of service class identifier, per bearer); GBR (Guaranteed bit rate, per bearer); MBR (Maximum bit rate, per bearer); ARP (Allocation and retention policy, per bearer); AMBR (Aggregated maximum bit rate, for all bearers for a UE)
Figure 9.9 QoS profile.
Rate shaping can be seen as an artificial resource limit, resulting in a longer queue if maximum bit rate is violated. The token rate is set to the DL‐AMBR value for DL and to the sum of UL‐ AMBR and all GBR values for UL. The bucket size is selected so that sufficiently large packets are allowed in order to not reduce the efficiency/capacity of the air interface. Standardized 3GPP QoS parameters are realized by using QCI and ARP in LTE and EPC as well as using DiffServ code points (DSCP) in LTE, EPC, and IMS. ARP decides whether new bearer modification or establishment request should be accepted due to resource limitations considering the current resource situation. For GBR bearers the QoS Profile additionally includes a guaranteed bit rate (GBR) and a maximum bit rate (MBR). Further, all Non‐GBR bearers of a UE are limited by an aggregated maximum bit rate, UE AMBR,2 also part of the QoS profile (Figure 9.9). APN‐AMBR shared by all non‐GBR bearers with the same APN, downlink bandwidth management is done in PDN GW and uplink bandwidth management in UE, UE‐AMBR shared by all non‐GBR bearers of the UE. Bearers and the associated establish/modify signaling procedures are required to reserve resources (processing + transmission capacity) before SDFs mapped to that bearer can be accepted into the network. On the backhaul nodes, the network may encounter traffic drops, and poor voice and video quality due to lack of prioritizing and congestion management in the backhaul nodes. To allow traffic separation in transport network, gateways and LTE RAN translate from bearer‐level QoS (QCI) to transport level QoS (DSCP) (Figure 9.10). From the core network the main important parameter for QoS are received, such as the QCI, DSCP, and ARP values. These values are used for QoS translation to find the proper scheduling attributes such as the scheduling strategy, mapping of QCIs to logical channel groups (LCG) for uplink scheduling based on the data as configured by the OSS configuration management system in a so‐called QCI table. DSCPs are used for layer 3 routing and prioritization of different traffic types but also used for mapping to P bit (Priority Bits, the values of which are used for prioritization by transport equipment handling the Ethernet layer) and queues for layer 2 2 The UE AMBR is used in eNB to limit the bit rate of a UE and it is defined by the subscription. The core network uses APN AMBR, which is the aggregated maximum bit rate allowed across all non-GBR bearers associated with the same APN. APN AMBR is not visible in RAN.
Traffic Model of Smartphone and Optimization (PCRF)
QCI
3GPP QoS Profile
Initialize / Upgrade / Downgrade 3GPP QoS Profile 3GPP QoS Profile
DSCP
DSCP
DSCP
QCI
UL/DL Ethernet Radio pBits (UL) Scheduler
Ethernet pBits
Ethernet pBits
Ethernet Ethernet pBits (DL) pBits (UL)
Ethernet pBits
eNodeB
Transport
Transport
Packet Core
Transport
QoS Scheduling
Figure 9.10 PCRF controls End‐2‐End QoS including radio and transport.
EPC/LTE QoS Profile
QCI
Mapping function IP datagram
ARP
Takes place in RBS and AGW
(Transport) IP header DSCP Mapping function
Ethernet frame
MBR/GBR
Ethernet header p-bits
Data
Takes place in devices on edge between L3 and L2 network DSCP
Data
Figure 9.11 QCI mapping at DSCP and P/bit level.
traffic handling on the access transport network, which enabling the transport network to prioritize between different data flows over the S1 interface. DSCP indicates what forwarding treatment, or per‐hop behavior (PHB), shall be applied at each node or router along the path. PHBs can be implemented by employing a range of queue service and/or queue management disciplines on a node’s or router’s output interface queue. The PDN GW and SGW use the QCI information to set the correct DSCP values on the signaling traffic. The DSCP attribute of the QciProfilePredefined defines the mapping between QCI and DSCP. Operator controls QoS by defining QCI’s and related characteristics, mapping services to QoS profiles, packet filters and bearers, mapping QCIs to DSCP (IP layer) and mapping DSCP’s to priority bits (ethernet layer). This corresponds to mapping from RAN to transport and IP network (Figure 9.11). Packet forwarding treatment in transport network is selected based on the DSCP carried in the tunnel header (IP). The recommended QoS mapping is shown in Table 9.2. DSCP is mapped with QCI to realize QoS in LTE network, an example is shown in Figure 9.12. The eNB scheduler is an essential QoS enabler. DL and UL are treated separately. In DL, the priority is simply determined per QCI, traffic can be separated per QCI and a different treatment
369
LTE Optimization Engineering Handbook
Table 9.2 Recommended QoS mapping. Traffic Type
DSCP
Network Synch
LU
7
Routing, network control
CS6
6
QCI5–IMS Signaling (Non‐GBR)
CS5
6
40
S1AP/X2AP‐Inter‐node Signaling
CS3
6
24
QCI1–GBR Conversational Voice
EF
5
46
QCI3–GBR Real Time Gaming
AF41
5
34
QCI2–GBR Conversational Video (Live Streaming)
AF42
5
36
QCI4‐GBR Non‐Conversational Video (Buffered Streaming)
AF43
5
38
QCI6‐Non‐GBR TCP Specific Services
AF31
4
26
OAM Access and Bulk Data
CS2
4
16
P‐bits
DSCP code
4 CoS
3 CoS
54
1
1
48
2
QCI7‐Non‐GBR Conversational/Speech & Live Streaming
AF11
2
10
QCI8‐Non‐GBR TCP’Premium bearer’
AF12
2
12
QCI9‐Non‐GBR TCP Default Bearer
AF13
2
14
Guaranteed Bit Rate
QCI 1
Voice
3
2
4
3
transport COS1(EF)
2 Unused
3 4 5
Non Guaranteed Bit Rate
370
6 7 8 9
IMS Signaling Video High Priority Data Regular Priority Data Low Priority Data
Control(CS4)
IMS Core
COS2(CS3) COS3(CS2) Default(BE) Default(BE)
Figure 9.12 DSCP mapped with QCI.
per QCI can be applied. In UL, it is determined per LCG, and based on the currently established dominant bearer of the LCG to ensure that traffic types are separated in the uplink, traffic separation must be ensured between the different data flows within the UE. The QCI of that dominant bearer can be different from UE to UE. The mapping of a radio bearer (or logical channel) to a logical channel group is done at radio bearer setup time by the eNB based on the corresponding QoS attributes of the radio bearers such as QCI. QCIs can be mapped to different LCGs, then buffer status reporting from the UE
Traffic Model of Smartphone and Optimization
QoS Aware
DRBn
DRBn
DRB1
DRB1
DRBo
SRB1/2 is mapped to LCG 0
Ue Scheduler
Strict Priority/WFQ per sector
eNB uplink
UE uplink
QoS Aware
QoS Aware
eNB downlink
DRBo
Strict Priority/WFQ per Ue
Strict Priority/WFQ per sector
Figure 9.13 Schematic overview of downlink and uplink scheduling.
is done for traffic types separately (UE reports an aggregate buffer status for the combination of radio bearers in a logical channel group. The eNB knows the radio bearers contained in the group and their priorities). The UE prioritizes internally between logical channels according to logical channel priorities (derived from QCI). The parameter controlling the mapping of LCG to QCIs is logicalChannelGroupRef. In the uplink, the UE has four LCGs, which groups together traffic of different logical channels. So a LCG consists of one or more QCIs. For this reason it is important to correctly configure the LCGs. This limitation is applied by 3GPP to keep the buffer status report (BSR) format small in the uplink MAC layer, the buffer status report is sent per LCG and not per logical channel. It also reduces the number of queues required to be supported in the UE side for uplink traffic separation (Figure 9.13). 9.1.3 Traffic Model
As we know, smartphones make it easy for people to surf the web and watch online videos, leading to much higher bandwidth use; tablet and notebook devices will send data even higher. iPad‐like devices will chew even more bandwidth than the smartphone because of its larger screen, which is driving bearer and signaling traffic growth. Video, web browsing, and audio streaming will dominate bearer traffic on wireless web, especially user generated content (video, photos, data backup) will stress the uplink. Detailed traffic model incorporating average user behaviors and key application characteristics. For each application, specific characteristics and session attempt rates were estimated. When UE has attached with the LTE network, there are both idle users and connected users. From traffic model perspective, the UE states are connected (UE that is RRC_connected or ECM_connected) and idle (UE that is RRC_idle, ECM_idle). The model for packet data service session consists of one or more subsessions,3 which are the sequences of user activities throughout a particular application. The field based characteristics of each application are defined in order to estimate user plane model, which is shown in Figure 9.14. Traffic profile is mainly defined by: ●● ●●
Duration of simulation – define in seconds length of simulation Number of subscribers – number of users with given traffic profile
3 If the dormancy timer is larger than the off-time, then a call may consist of several subsessions. If the dormancy timer is smaller than the off-time, then one call is composed of only one subsession.
371
372
LTE Optimization Engineering Handbook Web browsing BHCA
volume Avg per subscriber metric
Gaming BHCA
volume
BHCA
p
DL throughput
q
...
r Outcome
Streaming Video BHCA
duration Handset
xx %
Device→Appl mapping
PC
xx % Smartphone xx %
on
ti ca pli Ap Mix
...
Figure 9.14 Traffic model estimated.
●● ●●
User type of creation – constant or poisson Non‐real‐time services parameters: number of packet calls, call volume, reading time, number of datagrams in packet call, size of datagram, time between datagrams, and so on.
Take an example of inter‐packet arrival time comparison between Youtube and FTP. The mean inter‐packet time of FTP is 0.96 ms, Youtube is 15.9 ms. The maximum inter‐packet time of FTP is 0.89 seconds, Youtube is 1.29 seconds. Around 96.6% of Youtube ethernet packets are 1414 bytes, this corresponds to raw video packet size of 1356 bytes, while the mean FTP packet size is 1402 bytes.
9.2 Smartphone‐Based Optimization Smartphone experience includes response time, always on friendly, and battery life, and so on, which cause challenges to existing LTE networks. The key to good smartphone experience is that UE should maintain active RRC connection; the number of RRC connected users supported by eNB is an important metric for estimating whether eNB is smartphone‐enabled. When UE maintains active RRC connection for longer times, UE battery life and increased handover signaling load needs further consideration. Always‐on applications need to send and receive small packets frequently to keep IP connectivity open—it is called “heart beat” or “keep alive.” Typical frequency is once per minute, or once every few minutes. The amount of data is very small, 80%
END Y
PUCCH,SRS,PDCCH Utilization>75%
Y
Optimize inactivity timer OptimizeConfiguration (e.g.Optimize RB#, sounding period)
N
2. Add cells
Figure 9.18 Capacity monitoring method.
If the active user license utilization is greater than 80%, it needs to expand the license capacity or add a baseband board. If the active user license utilization is greater than 80%, there are a great number of online users and most have no data transmission. In this case, adjust the user inactivity timer base on the operator’s strategies to adjust the number of users in the RRC_ connected state. If the physical resource block (PRB) utilization is greater than 80% and the CPU usage also reaches 80%, it needs to expand the board capacity. If the PRB utilization is greater than 80% and the CPU usage does not reach 80%, perform flow control on cell resources. If PUCCH or SRS resource utilization is greater than 70%, turn on the PUCCH adaption switch or add more cells. Also it is needed to make sure that the eNB can automatically adjusts the number of PDCCH symbols based on the CCE load. If the total number of preambles exceed 75 to 100 times per second in busy hours, adjust the PRACH backoff algorithm or even start the flow control mechanism on the radio interface to smooth the RACH load peak and decrease the average RACH delay. At the same time, considering capacity expansion is required.
Traffic Model of Smartphone and Optimization
9.3.3 Special Features and Parameters for High Traffic
In case of extreme high capacity, it needs to take all actions to avoid overflow in the system as degrading performance for the end‐users. This special features for high traffic involves many different aspects such as dimensioning, parameter optimization and software features. If number of admitted users becomes higher than the system can handle, this will lead to that the users in the system gets little throughput. The high cell load causes each user already in the system to generate even more load. When number of users increases, the DL data payload drops after reaching a certain level of users. For high‐traffic sectors, the feature of call admission threshold can be changed to more than 90% on all these sectors to allow for more users to connect to LTE, thereby reducing call failure rate. Overload control is the key to ensure a good flow of users in the system. It is needed to control the arrival rate of new users. By just letting all users to access the network will in the end cause the network to stall with consequences. In overload situations it will always be better to serve the users in the network before allowing new users to enter. At high capacity it needs also to ensure that voice users can always access, packet domain users can afford to wait in such congested situations. The features for high traffic are shown in Figure 9.19. The dimensioning perspective is also important. Keep control of interference and maximize the carrier capacity, keep track of your network bottlenecks, and expand capacity in time. Try to be one step ahead when it comes to expand capacity in the radio network. GBR admission control ‐ To ensure that the QoS is maintained for GBR bearers in the system. A headroom is created to allow for fluctuating radio conditions and mobility, also be used to control the ratio between GBR and non‐GBR traffic. Priority paging ‐ At the eNB, the load control and overload protection mechanisms determine if any paging messages need to be discarded. If discards are necessary due to high paging load or high CPU processing load, the paging messages with highest priority are handled first. Load based access barring ‐ Uses barring information from SIB2 broadcast messages to inform the UE of the congestion level in the system. To release inactive UE at high load handover ‐ Automatically alleviates PUCCH resource exhaust at high load, to improve accessibility in high‐load networks, increase efficiency of
› Dimensioning – Cater for traffic increase – Monitor capacity bottle necks
› Control Arrival rate
– Apply admission control – GBR Admission control
– MIMO / CA /UL COMP / 4way RX-div
– Priority Paging – Load based uplink time alignment timer adjustment
– IRAT Offload from LTE
› Reduce overhead
› Maximizing carrier Efficiency
Overload Control
– CA based IFLB – Idle mode settings – Admission triggered IFLB – Service/Priority triggered IF HO
› Maximize network capabilities
– PDCCH Link adaptation – PDCCH Power Boost
› Control interference – ICIC / IRC – Frequency selective scheduling – Parameter optimization
› Block new access requests – Dynamic Load Control – Release of inactive UE at handover – Load based Access Class Barring
Figure 9.19 Features for high traffic.
377
378
LTE Optimization Engineering Handbook Inactivity Timer (10 s) TAT
UE SETUP
DL TRAFFIC
TAC
OUT-OF-SYNC
PUCCH RESOURCES USED
ALLOC PUCCH
UE RELEASE
NO UL RESOURCES USED EXCEPT RANDOM ACCESS
RELEASE PUCCH
Figure 9.20 Two timers setting of high load cell.
PUCCH scheduling request resource usage. There are two parameters needed to be considered. One is inactivity timer value to 4 sec for high load eNB’s to reduce load on high capacity sites. The other is the time alignment timer (TAT, a timer used for supervision and control of uplink synchronization, range, 0, 500, 750, 1280, 1920, 2560, 5120, 10240[ms], 0 means UEs always in sync), which is needed to set to an operator controlled value for new UEs, it means that these UEs will go out of sync after TAT has expired. The feature has a larger gain in handover intense areas as reducing inter‐cell interference, as fewer UEs are connected at cell border (Figure 9.20). The time alignment timer that will (re‐)start every time that a time alignment command is received. If the time alignment timer expires, the UE concludes that prior to any uplink transmission the random access procedure must be used to obtain uplink sync and all PUCCH resources and any assigned SRS resources will be released. PDCCH link adaptation (LA) could be used in high traffic scene. If no LA is used, 8 CCEs are always required to reach the cell edge. Actually, many users are also likely to require a less robust channel coding on PDCCH than the maximum 8 CCEs. PDCCH link adaptation can increase efficiency by dynamically assigning CCEs based on link quality. So PDCCH LA margin is introduced in LTE, which is added to PDSCH for PDCCH link adaptation, curent value is fixed 10 dB in all radio conditions. As many of the UEs don’t require 8 CCEs, reduction in the PDCCH link adaptation offset reduces the number of medium and good radio condition UEs, which are allocated 8 CCEs. It is recommended to reduce the fixed 10 dB margin to 3 dB. RLC parameters also impact performance in a high‐load cell. The transmitting side of the RLC connection can request a status message through a poll request. Upon expiry of this timer, the transmitting side will consider any PDUs, which have not been positively acknowledged for retransmission. Poll retransmit timer is triggered when the sender doesn’t get an acknowledgement back from the receiver about an earlier poll request regarding reception of successful PDUs. Default value is 50 ms for RLC (DRB) and 45 ms for SRB, it can be changed to 200 ms will give the receiver more time to acknowledge receipt of PDUs, thus lowering the need for retransmission from the sender. The parameter Max_Retx_Threshold determines the number of times a packet is retransmitted at the RLC layer. The objective here is to reduce the number of DL RLC retransmissions (default value is 32, recommend changing is to 16) for high load cell, that is, congested sites where resources can be freed up by reducing this number. Still, UE timers and eNB timers need to be modified in a high‐load cell to improve call failure rate. For example, increase T300 from default 400 ms to 1000 ms, changing T302 value from 4 s
Traffic Model of Smartphone and Optimization
to 8 s, increase T301 from default value of 400 ms to 1500 ms, changing T304 value from 1 to 2 s, and changing T311 from 5 s to 3 s, and so on. eNB timers are vendor specific, need to be aligned with UE timers. 9.3.4 UL Noise Rise
The UL noise rise is the key limiting resource in a scenario with many users. It is the amount of users active in RRC_connected state that cost resources. Each user consumes part of the UL noise rise even though no or very small amount of data is being transmitted. The consequence is that cell shrinks, uplink coverage decreases, and as a result, calls are dropped, accessibility suffers, throughput is starved. UL noise rise can be observed by: ●● ●●
●●
Distribution of the SINR values calculated for PUCCH/PUSCH The measured noise and interference power on PUSCH, according to 36.214 of the SINR values calculated for PUSCH The measured noise and interference power on PUCCH, according to 36.214
For a high‐load cell, α factor (i.e., pathloss compensation factor for PUSCH) should be set to default 0.8 to ensure that cell‐edge UEs won’t transmit at higher power (increasing interference), which in turn forces other UEs to transmit with a higher power. p0NominalPucch and p0NominalPusch can be used to adjust power control target, it will trade off between coverage and capacity. Lowing this should help to improve UL noise rise interference. Current setting of the two parameters are as following (Figure 9.21): ●● ●●
p0NominalPucch = −116 dBm, higher value can cause high UL interference in high load condition p0NominalPusch = −96 dBm (should be reduced to default −103dBm to reduce UL interference peaks at high load)
9.3.5 Offload Solution and Parameter Settings
This part investigates parameter settings to determine if more optimization can be achieved and if some of the settings can be applied market wide to improve high‐capacity cell performance. High PDCCH CCEs utilization In general, PDCCH CCEs utilization above 75% indicates high PDCCH CCEs utilization. High PDCCH CCEs utilization can be caused by large number of active UEs, and/or poor coverage. RRC connection setup failure/incoming handover preparation failure under a high‐ capacity condition usually has high PDCCH utilization behind it. High PDCCH CCEs utilization indicates the SRB traffic is using the majority of the scheduling opportunities and leaves very little opportunities for DRB traffic under high‐load conditions. The eNB is limited on CCEs during low throughput. Scheduling request and/or CQI resource congestion A UE is allowed to connect to a cell if there are free scheduling request and CQI resources. When a UE locates at high‐traffic cell, scheduling request, and/or CQI resource congestion can be caused by large number of users. Even sometimes it was accompanied with retainability degradation, which indicated a drop caused by MME side. The operator needs to configure the number of PUCCH resources for a scheduling request and CQI to control the trade off between the number supported users and the uplink peak throughput. Solution of offloading traffic
379
LTE Optimization Engineering Handbook RSRP vs UE Tx Power/PUSCH Throughput
30
25000
25 A
UE Tx Power(dBm)
15
15000
10 10000
5 C
0
UE Tx Power(kbps)
20000
20 B
5000
–5
D
–10 2
–6
–6
4 –6 6 –6 8 –7 0 –7 2 –7 4 –7 6 –7 8 –8 0 –8 2 –8 4 –8 6 –8 8 –9 0 –9 2 –9 4 –9 6 –9 –1 8 0 –1 0 0 –1 2 0 –1 4 0 –1 6 0 –1 8 1 –1 0 1 –1 2 1 –1 4 1 –1 6 1 –1 8 2 –1 0 2 –1 2 2 –1 4 26
0 RSRP (dBm) C Total UE Tx Power/RB (pZeroNominalPusch = –96dBm)
D Total UE Tx Power/RB (pZeroNominalPusch = –103dBm)
A PUSCH Throughput (pZeroNominalPusch = –96dBm)
B PUSCH Throughput (pZeroNominalPusch = –103dBm)
–95 P0 = –95 P0 = –99 P0 = –103 P0 = –107 P0 = –111 P0 = –115 P0 = –119
–100 –105 IRB,UL [dBm]
380
–110 –115 –120 –125 –130 110
115
120
125
130
135
140
145
150
Lsa,cellrange [dB]
Figure 9.21 p0 optimization in a high‐load cell.
Besides adding carrier/site to balance the high load, the solution of offloading traffic is as efficient way that is summarized below: ●●
By mobility allignment: It can be done by encouraging easier cell reselection in idle mode, aims to increase qOffsetCellEUtran for reselection to non‐congested neighbor cells for equal cell reselection priority, decreases qOffsetCellEUtran for reselection to high‐capacity cell, for priority‐based cell reselection, adjusts parameter threshXHigh and threshXLow, and, encouraging early handover in connected mode, aims to increase cellIndividualOffsetEUtran for a certain EUtranRelation to push some traffic to neighboring non‐congested cells.
Traffic Model of Smartphone and Optimization ●●
●●
By decrease cell size: RF adjustment to decrease cell coverage, downtilt to reduce coverage footprint on the high‐capacity sector, while may require to up tilt the neighboring non‐congested sectors. Also it can increase cell size defining parameters: Qrxlevmin or, changed Delta PSD for cell specific reference signal relative to the reference from 3 dB to 0 dB to reduce cell size under low inter‐site distance. For further, it can decreases p0NominalPusch and p0NominalPucch in a high‐load condition. By alleviate PDCCH scheduling congestion: Reduce SRB traffic aims to reduce the number of UL/DL RLC retransmissions for SRB and DRB, or, to extends the time for new poll if no RLC status report is received for SRB and DRB.
381
383
Part 3
Voice Optimization of LTE
385
10 Circuit Switched Fallback Optimization Circuit switched fallback (CSFB) addresses the requirements of the first phase of the evolution of mobile voice services, which began on a commercial scale in 2011. CSFB is the solution to the reality of mixed networks today and throughout the transition to ubiquitous all‐LTE networks in the future phases of LTE voice evolution. On the other hand, CSFB is the first step in enabling mainstream LTE handsets with the cost, size, and battery life advantages of single‐ radio solutions to LTE data in combination with 2/3G voice as well as making the initial LTE investments smaller. When the user’s device is paged via LTE with an incoming call, or when the user initiates an outgoing call, the device switches from LTE to 2/3G. This is a circuit switched (CS) fallback function in EPS, which enables the provisioning of the CS services when the UE is served by E‐UTRAN without the need for IP multimedia subsystem (IMS) support. A CS fallback– enabled terminal connected to LTE can use 2/3G to connect to the CS domain in areas where LTE coverage is overlapped with 2/3G coverage, and the call remains in the CS domain until it is completed. Compared with VoIP and standby phones, CSFB is easier to realize, that means we don’t need to do much to upgrade the network, and CSFB can handle various kinds of phones. CSFB may be used as a generic telephony fallback method securing functionality for incoming roamers as well.
10.1 Voice Evolution This voice communication evolution can be characterized into three major phases. In the first phase, currently underway, all voice traffic is handled by legacy CS networks, while data traffic is handled by LTE packet‐switched (PS) networks, 2/3G network used primarily for voice service of legacy terminals. Usually, the operator may deploy LTE with 3GPP CSFB solution across LTE coverage area. In CSFB, whenever an LTE handset generates or receives a voice call it is automatically transferred to the 2/3G networks. Once the call is finished the device reverts to LTE. It relies on interrupting the LTE connection when the terminal is forced to move to the 2/3G network. This might be a big problem, depending on the application that is being used prior to the voice call. The second phase in LTE voice evolution introduces native VoIP on LTE (VoLTE) along with enhanced IP multimedia services such as video telephony, this solution relies on SRVCC (single radio voice call continuity) or PSHO at edge of coverage. Compare to CSFB, VoLTE avoids 4G data service interruptions and preserves the LTE data experience during speech communications. The third phase converges the enhanced capacity and services of all‐IP networks (voice and video over IP and RCS, such as instant messaging, video share, and enhanced/shared phonebooks) for continuous coverage across the broader range of network LTE Optimization Engineering Handbook, First Edition. Xincheng Zhang. © 2018 John Wiley & Sons Singapore Pte. Ltd. Published 2018 by John Wiley & Sons Singapore Pte. Ltd.
386
LTE Optimization Engineering Handbook 2/3G coverage
LTE
CSFB
LTE
LTE LTE
SRVCC
LTE LTE
Legacy CS voice
CSFB LTE
CS
CS
LTE E LTE LTE LTE E LTE LTE LTE LTE E LTE LTE
LTE LTE
CS voice service via CSFB
IMS Voice service over LTE and CS access LTE
Legacy phone
CSFB phone
CS attach
MSS
Legacy architecture
VoLTE phone
CSFB CS attach
MSS
CSFB Architecture
Mainly continuous LTE coverage
Areas with continuous LTE coverage
Spotty LTE coverage
SRVCC
SRVCC AKS
IMS
CSFB CS attach
MSS
SRVCC / ICS Architecture
LTE E LTE E LTE LTE LTE LTE E LTE LTE E LTE LTE LTE LTE LTE LTE LTE E LTE LTE E LTE 2G - 3G LTE LTE E LTE LTE LTE LTE LTE E LTE LTE LTE LTE LTE LTE LTE LTE LTE E LTE LTE E LTE LTE
IMS Voice service over LTE and HSPA
VoLTE phone
LTE HSPA
IMS
Mobile telephony evolution target
Figure 10.1 Voice evolution and network evolution.
access methods, including LTE, 3G, and WiFi, with interoperability across operators and legacy telephony domains (Figure 10.1). Migration from CSFB to VoLTE is typically done in several steps starting from CSFB, then combination of CSFB and VoLTE, and finally, full VoLTE where all calls are fully utilizing VoLTE. SRVCC enables handover from VoLTE to CS speech when the UE is running out of LTE coverage. Also reverse SRVCC enables handover from CS speech back to IMS‐based VoIP. The handling of voice traffic on LTE handsets is evolving, as the mobile industry infrastructure evolves toward higher, and eventually ubiquitous LTE availability. When subscribers are attached to LTE access, the UE informs the network what it supports, either data centric, or voice centric. If the UE is voice centric, it shall report what it supports, CS voice only (CSFB), IMS PS voice only (voLTE), CS voice preferred (IMS PS voice as secondary), or IMS PS voice preferred (CS voice as secondary). CS fallback is triggered to overlapping CS domain (2/3G) whenever voice service is requested, and resumed LTE access for PS services after call completion.
10.2 CSFB Network Architecture and Configuration 10.2.1 CSFB Architecture
The key point from CSFB view is the SGs interface between MSC and MME that SGs interface connects the MSC/VLR and the serving GPRS support node and mobility management entity (SGSN‐MME). This interface is used for registration in the MSC/VLR of the UE by performing combined procedures, to page the UE on behalf of the MSC/VLR and to convey CS‐related services. CSFB‐evolved node functions require a software upgrade and are listed in Figure 10.2. Following functions shown in Table 10.1 need to be enabled in MSC‐BSC (RNC) and SGSN‐ MME for CSFB. RIM functionality supports NACC/system information distribution for CSFB from LTE to 2/3G. RIM is recommended to pre‐populate LTE eNB with 2/3G SIB data. RIM procedures setup associations between external UTRAN or GERAN cells and the BSS that hosts that cell and receive NACC information for the cell. This process occurs automatically, with RNC/BSC updating eNB of changes to subscribed cell relations. It is required to enable feature “Release with SI” toward 2/3G, which speeds up call setup time toward 2/3G.
Circuit Switched Fallback Optimization 1. RIM (O) 2. DMCR (O) 3. Iu-flex/MSC Pool support (O) 4. PS Handover (O) 1. RIM (O) 2. Fast return to LTE (O) 3. A-flex/MSC Pool support (O) 4. PS Handover (O) 5. Dual Transfer Mode (O)
UTRAN
1. RIM (O) 2. PS Handover (O) SGSN
IuPS
Gs
Gb
Uu GERAN
S3
IuCS A
MSC-S
Um
1. LTE to CS Fallback (M) 2. SMS over SGs (M) 3. MSC Pool (O) 4. MT Roaming Forward (O) 5. Dual Transfer Mode in GSM (O)
SGs UE
E-UTRAN
LTE-Uu
1. CS Fallback (M) 2. DMCR (O) 3. Dual Transfer Mode (O)
MME
S1
1. CS Fallback to GSM and WCDMA (M) 2. RIM (O)
1. CS Fallback (M) 2. SMS over SGs (M) 3. RIM (O) 4. MSC Pool support (O)
M: mandate, O: optional
Figure 10.2 CSFB architecture and functions.
Table 10.1 MSC‐BSC (RNC) and SGSN‐MME functions. MSC‐BSC
SGSN‐MME
Feature provide and LKF installation on APG SGs feature activation Check SGs interface parameters SGs setup IS part SGs interface setup and activation
SMSoSGs and CSFB requirements: RAN information management (RIM) feature activation RIM support enables SI transfer between RATs SMS over SGs and CSFB to GSM and WCDMA feature activation IP‐based interface configuration SCTP‐based interface configuration TA‐LA to MSC‐S BC mapping SGsAP configuration
The support for RIM procedures show in Figure 10.3 that the feature makes it possible for an eNB to receive information for a specific GERAN/UTRAN cell or to create and maintain relationships with GERAN/UTRAN cells. Such relationships are referred to as RIM associations. Once a RIM association is established in the eNB by the operator, the eNB receives updates from the BSC/RNC when SI for the GERAN/UTRAN cell in that RIM association changes. When moving from CSFB region to non‐CSFB region, “data centric” LTE data modems should stay on LTE, “voice centric” LTE smartphones should re‐select to 2/3G (Figure 10.4). CSFB‐configured region is based on tracking area; this CSFB feature can be enabled/disabled per the tracking area. When moving from CSFB region to non‐CSFB region, the tracking area update is triggered, but tracking area update accept contains a tracking area update but does not provide location area information. It is worth to note that UE ID in 2/3G and LTE is different, as shown in Figure 10.5. 10.2.2 Combined Register
A dual‐mode UE can attach both to the LTE network, and to 2/3G networks before voice calls via CSFB can be initiated. The UE with CSFB ability not only can access to EURTAN from EPS, but also can access to CS domain from GERAN/UTRAN.
387
388
LTE Optimization Engineering Handbook
eNodeB
MME
SGSN
BSC
1. eNB Direct Information Transfer 2. RAN Information Relay 3. RIM PDU 4. RIM PDU 5. RAN Information Relay 6. MME Direct Information Transfer
1. Establish RIM Association through the CN. SGSN
MME 2. Report SI changes for Utran cell belonging to RIM association, to the eNB over the lu.
RNC
E-NodeB
UtranCell = N
RIM Association
ExternalUtranCell = N
eNB has the latest system information for the target Utrancell.
Figure 10.3 RIM procedure and association. Cs-domain Registration: FAIL Voice centric LTE smartphones should re-select to 2G/3G in non-CSFB
LTE TDD
LTE TDD
TD- SCDMA
TD- SCDMA
GSM
GSM
CS-domain Registration: OK
CS-domain Registration: OK
Figure 10.4 CSFB region (left) and non‐CSFB region (right).
M-TMSI (Subscriber confidentiality)
SGSN
MME
P-TMSI EPC
RNTI (Scheduling)
MSC
eNB
TMSI IP Address (Packet Forwarding)
Figure 10.5 UE ID in 2/3G and LTE.
S-GW P-GW
Circuit Switched Fallback Optimization
SGsAP
SGsAP
SCTP IP
SCTP IP
L2
L2
L1
a new IE, mobile class mark
5. MSC performs location update. BSC/ RNC
MSC/ VLR
2G/3G Circuit Core
G-MSC
L1 MSC Server/VLR
4 and 6, Location Update SGs Request/Response
1. Attach request (Type = IMSI/EPS Attach, Additional Info = ”SMS Only or Both”) MME
7. Attach Complete
2. MME performs authentication & bearer establishment. 3. MME derives VLR number for LA update.
MME handles the combined PS and CS Attach. For the CS attach it maps TA -> LA and makes a location update over SGs
Figure 10.6 Combined register. Table 10.2 SGs procedures and messages. SGs Procedure
Used SGsAP messages
Location update
Location update accept, Location update reject, TMSI reallocation complete
EPS detach
EPS detach indication, EPS detach ACK.
IMSI detach
IMSI detach indication, IMSI detach ACK.
Paging
Paging request, Service request, Paging reject and UE unreachable
Alert
Alert request, Alert ACK, Alert reject,
Activity indication
UE activity indication
Reset
Reset indication and reset ACK
SMS
Uplink unitdata, Downlink unitdata and release request
The SGs‐interface connects the SAE domain with the classic CS domain with MSCs and 2/3G radio network. The SAE domain consists of the EPS with the MME node and the LTE radio network. The SAE domain is completely packet‐based so the only way to provide CS services to SAE is via the SGs interface. The SGs interface use a SCTP (stream control transmission protocol) connection between the MME and the MSC/VLR. It is needed to note that all new interfaces in the SAE architecture begins with an “S.” The SGs‐interface is an “expansion” of the existing Gs interface between the MSC and SGSN. The SGs‐related procedures and messages are shown in Figure 10.6 and Table 10.2. In CSFB deployment, LTE (TA, tracking area)‐2/3G (LA, location area), and LA‐MSC should be mapped correctly in MME. The TA and LA are mapped by connecting them to the same geographical area in the SGSN‐MME. By this mapping, MME can determine the corresponding MSC, and initiate MSC location–combined update request for HSS/HLR. The combined procedures enable a UE supporting both CS and PS services, to connect to both types of services through the EPS network. The UE requests the MME for combined procedures in an attach request or tracking area update (TAU) request message. The MME establishes initial registration of UE, and maintains UE location information updated with the MSC/VLR. The SGSN‐MME sends information regarding the UE’s location to the MSC/VLR. The MSC/ VLR does not have any information about tracking areas, so the SGSN‐MME has to send location information based on the location area in which the UE is located by TA‐LA mapping.
389
390
LTE Optimization Engineering Handbook
Figure 10.7 “Combined EPS/IMSI attach” in Attach request message.
When a UE send Attach request message to MME, the MME can get a CSFB indicator. The attach type is “combined EPS/IMSI attach” in Attach request message, which is shown in Figure 10.7. In attach request message, the following information can be got: EPS attach type, CSFB capability, and UE’s usage setting. Then the combined (TA/LA) update will be done; it will register with the CS domain in a proper MSC. A mapping between LTE TAs and 2/3G LAs that enables both SGSN‐MME and MSC/VLR to keep track of UE’s location. Whenever the UE moves and changes TA within the LTE network, its position is kept updated in the MSC/VLR according to the TA/LA mapping configured. Then MME will send Attach accept message to UE with the TMSI and LAI, which is assigned by MSC. MME configures the TA‐LA and LA‐MSC mapping relation, determines the MSC, and then MME will do combined registration with MSC, that is, they will trigger MSC register in HLR. After the MME derives the LAI from the TAI of the current LTE cell, then MME sends the “location update request” to the target MSC/VLR. If EMM causes “network failure,” it is found in the Attach accept message, and it means the combined attach was “successful” and is for EPS services only (Figure 10.8). MSC sends the UE a TMSI reconfiguration command by location update accept message, after the TMSI reconfiguration, UE send TMSI reollocation complete message to the core network, indicates that the TMSI reconfiguration has been completed. Once the location update is accepted by MSC, it will create/update the SGs association and mark the subscriber as IMSI and EPS attached. The MSC will then perform the location update procedures toward the HLR in the same way as it is done for an A‐ or Iu‐interface. In attach accept message, the following information TAC, M‐TMSI and LAC can be got. By receiving this message, the UE shall start T3411 and T3402 (refer to Annex). If either of these two timers expires, UE will re‐trigger combined TA/LA updating with IMSI attach. If the UE is not successful for the non‐EPS part LA updating, UE will try five times each time, since after five times, T3402 will start, wait for another 12 mins, and then another five times will initiate. When UE performs detach procedure, it can either be received as IMSI detach, EPS detach, or combined EPS/IMSI detach. The detach procedure can be initiated by UE, MME, or HSS. Depending on detach type, the UE is marked as EPS detached or IMSI detached or both. This procedure is initiated by EPC (MME) by sending an EPS detach indication or IMSI detach indication to the MSC. The MSC removes the SGs association to the EPC (MME).
Figure 10.8 Attach request and attach accept message.
392
LTE Optimization Engineering Handbook
10.2.3 CSFB Call Procedure
LTE CSFB call procedure can be divided into five parts: LTE attach request, CSFB process, mobile originating call, CS radio bearer session, alerting, and PS radio bearer data transfer. 10.2.3.1 Fallback Options
After CSFB has been deployed, eNB supports one of the following methods to fallback to the 2/3G network to make a CS voice call. Depending on feature activation, output from RRM handover algorithm and UE capabilities, the eNB decides which type of CS fallback will be used—RRC release with redirect, PS handover, or (in case of 2G target RAT) inter‐RAT cell change order, and so on (Figure 10.9). ●●
●●
●●
●●
Redirect based CSFB to 2/3G – CSFB via redirect allows for a service‐based redirection from LTE to UTRAN or to GERAN during the call setup. The operator can configure priorities for the target frequency bands per cell. eNB selects the highest priority layer supported by UE. UE does not have any pre‐information about the target cell. It is needed to configure the carrier frequency of the target RAT by RRC connection release message, then, UE access to target 2/3G cell (can be based on field measurements or blind). The baseline method is a blind redirect to target RAT. With blind redirect time, the transition from LTE to 2/3G for a non‐stationary UE is around 2 seconds. Redirect based CSFB to 2/3G with SIB (System information block) – Compare to the upper one, the UE does not have to re‐read the whole system information from air interface. RAN information management (RIM) procedure may be deployed for retrieving system information from 2/3G, and it is possible to provide information about the target cell prior to releasing from LTE. The system information provided together with the RRC connection release message will reduce the overall call setup time. Redirect based CSFB to 3G with deferred SIB11/12 reading (deferred measurement control reading (DMCR)) – With this feature UE defers reading system information blocks type 11, 11bis and 12, which may be distributed to several segments depending on the size of the neighbor list in the target 3G cell before RACH access. The gain in setup time depends on SIB contents and scheduling, but several hundred milliseconds are practical by deferring reading SIB the CSFB call setup delay is reduced. Field tests indicate that DMCR reduces the SIB acquisition time from 2 seconds to about 1 second. The information whether the 3G cell supports DMCR is broadcast in SIB3, which the UE will have to acquire before deciding if SIB11/12 reading can be skipped. PSHO based CSFB to 3G—eNB starts upon the CSFB trigger IRAT measurements to identify suitable target cells. eNB performs a PS handover once a suitable UTRAN target cell has been found based on measurements during the handover preparation phase, using handover instead of release with redirect will give a shorter interruption in the ongoing PS data session and target cell SI provided to UE during the handover procedure. Resources are reserved beforehand in target radio and call setup delay is reduced and reliability of call setup is increased. The feature of a PSHO–based CSFB needs a 3G network to be upgraded to the right software.
RIM start a CSFB Call
without RIM
Figure 10.9 RIM.
R9 RRC redirect with 3G frequency PSC,SI
R8 RRC redirect with 3G frequency
measure 3G cells which will fall back and synchronize
measure 3G cells which will fall back and synchronize
camping and call set up
read SI for target cells
camping and call set up
Circuit Switched Fallback Optimization
10.2.3.2 RRC Release with Redirection
In a live network, the fallback method of RRC release with redirection is widely used so that it can be simply deployed. The available variations for CSFB to 2/3G RRC release with redirection are the following: ●● ●● ●●
Release with redirection—Basic Release with redirection—SIB skipping Enhanced with redirection—SI tunneling (3GPP Rel 9)
The call flows for all three variants are similar and the distinction lies in how the SIB messages are handled. Reading of SIBs in the target cell once the UE has moved to the fallback RAN is a large contributor to the delay budget. The first variant reads all SIBs and thus causes the longest delay. The second reads only the mandatory SIBs—1, 3, 5, and 7—and the remaining SIBs are provided once the UE is connected to the target cell. The third variant reads no SIBs at all and thus has the shortest delay. All SIB information is tunneled via the core from source to target RAN. In Rel‐8, the CSFB release with redirection, at a voice call origination attempt or when receiving a page for CS voice (via SGs interface), the fallback RAT’s frequencies is transmitted by RRC connection release message. The UE changes RAT and starts accessing the new RAT attaching to the indicated target frequency, then transmits a CM service request to the BSC/RNC. Once the CS service request is accepted, the terminal will start the CS call. Rel‐8 CSFB, the release with the redirection procedure is shown in Figure 10.10. For Rel 8 redirection, eNB only carries the frequencies of the target cell, and the UE needs to select a frequency and read the SIB. For Rel 9 redirection (SI tunneling) shown in Figure 10.11, the system information can be regularly updated via RIM signaling and cached at the source eNB, it avoids the delay of configuring and performing target RAT measurements. The main risk is if the provided information is misaligned or outdated, in which case the UE is forced to make the call setup as if there were no system information available. This feature will decrease the amount of time required to setup a CS call since it will not be necessary for the UE to download all SIB containers.
UE initiated a CFSB call
UE
eNB
MME
Optionally: DL NAS-PDU Transfer „CS SERVICE NOTIFICATION“ (in case of MTC) CSFB release with redirection
RRC: ULInformationTransfer (EXTENDED SERVICE REQUEST)
S1AP: UL NAS TRANSPORT S1AP: UE CONTEXT MODIFICATION REQUEST (CS Fallback Indicator)
UE measures the target frequency and sync to the new RAT
Selection of redirect target S1AP: UE CONTEXT MODIFICATION RESPONSE RRC: RRCConnectionRelease L2ACK
(With Redirect Info)
UE read target RAT’s SIB
S1AP: UE CONTEXT RELEASE REQUEST
S1AP: UE CONTEXT RELEASE COMMAND
UE leaves cell
UE access the new RAT and make a call
Figure 10.10 Rel 8 redirection procedure.
S1AP: UE CONTEXT RELEASE COMPLETE
Delete Bearer
CSFB is triggered in the eNB through initial context setup request and UE context modification request The MME sends an S1-AP UE context modification request with a CSFB indicator to the eNB
393
394
LTE Optimization Engineering Handbook 1. eNB registers information (id and signaling strength/quality) about neighboring cells from measurement reports from the UE’s.
2. eNB requests system information (SI) for the overlapping 2/3G cells via the packet core network. 3. SI request/response relayed by the MME MM E
S1
eNB LTE
S3
6. eNB provides SI to UE at CSFB.
BS NodeB
2 / 3G 7. UE do not need to read SI in 2/3G => Faster call setup!
A-bis Iub
BSC RNC
Gb IuPs
SGSN
- 4. SI request/response Gb IuPS relayed by the SGSN
5. BSC/RNC provides SI based on regular intervals (requests) or when ever the configuration is changed.
Figure 10.11 RAN information management (RIM).
Optional enhancements for 3G, the method of RRC release with redirection (R8) + DMCR (deferred measurement control reading function), is also supported shown in Figure 10.12. UE awaits reading SIB11 and 12 with instructions from the target cell (SIB3 support deferred measurement control reading configuration) and processes other time‐consuming SIBs (related to measurement control).
1. UE falls back to 3G
S1
eNB
MME
LTE SGs
2. UE reads SIB3 = DMCR on/off
BS
Iub
RNC
3G
3a. DMCR on = UE awaits with reading SIB 11, 11bis and SIB 12 => Faster call setup! SIB 11, 11bis and 12 are read later during the call
Figure 10.12 Deferred measurement control.
3b. DMCR off = UE reads SIB 11, 11bis and 12 at their regular broadcast intervals (every 1.4 s). Call setup continues after SIB’s are read.
Iu
MSC
Circuit Switched Fallback Optimization CSFB WITH LOAD SHARING Calls redirected to frequencles In Round Robin method
F1 F2 F3 ARE OF SAME PRIORITY
LTE
F2
F3
Calls
F1
WCDMA GSM
F1
F2
F3
Figure 10.13 Fallback frequency selection.
CS fallback will always direct the UE to whichever frequency relation has the highest priority. If the same priority is configured for multiple frequencies, the eNB will redirect the UE based on a simple round robin algorithm for the load sharing, which is shown in Figure 10.13. 10.2.3.3 CSFB Call Procedure
The UE sends an NAS extended service request message to the MME requesting a CSFB service when the UE requests the CS service or as a response to a paging message with a CSFB flag. CSFB is triggered in the eNB through initial context setup request and UE context modification request message sent by the MME. For mobile originating (MO) voice calls, the UE sends Extended service request message indicating mobile originating voice call from a CSFB terminal. The eNB triggers release with redirect, redirecting the UE to the CS domain to continue with the call setup. The eNB makes the decision to what neighbor cell the UE should be released to and sends an RRC_release_ with_redirect message to the UE and a S1‐AP UE context release request to the MME. The UE changes RAT and starts accessing the new RAT attaching to the indicated target frequency. Finally, the UE transmits a CS service request to the BSC/RNC or 2/3G network. Once the CS service request is accepted, the UE will start the CS call. The PS services are moved to target the network if possible. MO CSFB call procedure is shown in Figure 10.14. The call flows for mobile (MT) terminating call setup is presented in Figure 10.15. For terminated CS call in LTE, the MSC pages the UE over LTE via the SGs interface, the UE changes radio mode to 2/3G and receives the call. When MSC/VLR pages users, MSC needs to configure the paging parameters of SGs interface and 2/3G network, including paging times, paging interval, TMSI/IMSI paging, whether to continue A/Iu interface paging or SGs and A/Iu interface simultaneously paging after SGs paging failure, and so on. Before MSC/VLR pages the UE, it will check the UE’s SGs association state. If the SGs association is null, paging UE in 2/3G network in accordance with its storage location information, if SGs association is associated, then paging UE by SGs interface, and starts the SGs interface paging timer. SGs paging message include IMSI, LAI and TMSI etc. When MME received paging message from MSC through SGs, it needs to check whether the UE is attached or detached, the SGs interface is associated or not, and the LA/TA matches or not, and so on. If the UE in the LTE network is detached, MME will response paging reject (IMSI detached). If the UE is in RRC connection state, MME directly sends the CS service notification (IMSI or TMSI) to the UE. If the UE is idle, MME sends paging (S‐TMSI) message to
395
eNB
UE
BSS/RNS
MME
MSC
SGW/PGW
SGSN
Extended Service Request S1 AP-Request message with CSFB indicator S1 AP-Response message Optional Measurement Report Solicitation RRC connection release , S1 AP-: S1 UE Context Release Request S1 UE Context Release UE changes RAT then LA Update or Combined RA/LA Update or RA Update or LAU and RAU Suspend Suspend Request / Response 8Update bearer(s) CM Service Request
A/Iu-cs message (with CM Service Request)
Service Reject Location Area Update CS MO call
Routing Area Update or Combined RA/LA Update
Figure 10.14 MO CSFB call procedure.
If the MSC is changed
Circuit Switched Fallback Optimization
The UE sends a CM service request to the BSC/RNC. Once the request is accepted, the UE starts the CS session
Before call proceeding, only MO UE need to analyze. After that, the core network starts paging MT UE, MO and MT UE need to combind analyze.
Figure 10.15 MO and MT CSFB call procedure.
UE under its TA/TA list for CS and SMS service. When UE received the paging message, MME will send paging response (service request message) through SGS interface to MSC. Finally, the UE informs the MME that it needs to set up the CS call and it needs the CSFB, via the NAS Extended service request message. The UE will be ordered to release the LTE connection it will be redirected to the CS domain. 10.2.4 Mismatch Between TA and LA
For CSFB and session continuity, it needs to have a proper TAC‐LAC mapping where the UE on 4G will be redirected to the underlying 2/3G site as MME needs to know to which MSC the UE should be connected to. Operators who have existing 2G or 3G networks should plan their TA boundaries to coincide with their routing area (RA) or location area (LA) boundaries. LA and RA boundaries used for the 2G and 3G systems are likely to be relatively mature and may have already been optimized in terms of their locations. Existing 2/3G counter data can be used to estimate the quantity of paging, which is likely to be experienced across a specific TA. There might be specific requiremets to coordinate TA code (TAC) planning with LA code (LAC) planning in case of mobility during CSFB or in case of cell reselection to 2/3G. When UE makes a combined EPS/IMSI attach or combined EPS/IMSI, by TA update request message, it receives a LAC from MME inside of the attach/TAU accept message. In case of CSFB call, LAC is needed when MSC is paging a UE over SGs interface as MME maps LAC received from MSC to TAC and sends the
397
398
LTE Optimization Engineering Handbook
paging to eNBs over S1AP. Therefore, a mapping table between LA identity and TA identity must be configured at the MME. With CSFB enabled, in order to minimize CS call setup delay, it is necessary for the MME to inform the MSC in which LA the UE is located. Delays will be introduced if MSC is informed with the incorrect LA. There could be two scenarios: ●● ●●
If the incorrect and actual LA are controlled by the same MSC ⇒ minor delay If the incorrect and actual LA are controlled by different MSC ⇒ more additional delay Thus, the general rule is TA should be correctly mapping to LA, as shown in Figure 10.16.
TAL1
TAL3
TAL2
TAC2
TAC6 TAC4
TAC1
TAC10 TAC7
TAC5
TAC3
TAC12 TAC8
TAC9 TAC11
LAC1
LAC2
LAC3
LA Update
TA Update
Mapping table MME uses TA when roaming in LTE Coverage!
LA1
MSC/VLR#1
Figure 10.16 TA and LA mapping.
MSC/VLR MSC/VLR#1 MSC/VLR#2 MSC/VLR#2
Identifying LA and MSC/VLR from TA
MME
TA1
LA LA1 LA2 LA2
TA TA1 TA2 TA3
TA2
TA3 LA2
MSC/VLR#2
Circuit Switched Fallback Optimization
2G/3G RAN
MSC Page
Page Response/ Call Setup
MME
AGW
LTE
PSTN
Location Update
PDN
Same geographical area
Figure 10.17 Matched LA/TA.
UE
eNB
BSC/ RNC
MME
MSS
Location Update
Time to transition to target RAT
Extended service request Transition to target RAT Location Update Request
Authentication, identity request, TMSI reallocation Location Update Accept
Call setup
CM Service Request CS call setup
Figure 10.18 Unmatched LA/TA.
The TA and LA mapping relationship of MME are pre‐configured. TA should totally match the UE’s true LA in ideal situation. UE need to be registered in geographically “correct” LA while in LTE so that it can respond to a page to the MSC serving the 2/3G cells in the area where the UE currently is located. One LA could contain several TAs, but one TA could only belong to one single LA. Any matched LA/TA, UE will establish a call quickly after falling back to 2/3G and send CM service request or Paging response message. Otherwise, there will be extra LAU procedure. If the MSC changed, the call may failed without LA update procedure (Figure 10.17 and 10.18). Unmatched LA/TA, the target RAT LAC differs from the one UE obtained in LTE, if the two different LAs are controlled by the same MSC, it will cause the UE execute LAU procedure after fallback to 2/3G network. Usually, it will bring 2 more seconds of delay. Before the CM service request or Paging response message on 2/3G, LAU is necessary, and IMEI checking and TMSI reallocation may be done at this point, and UE should also mark Follow‐On‐Request
399
400
LTE Optimization Engineering Handbook
This TA Streches over two areas serviced by two different MSCs (or MSC pools)
MSC1 LA
TA1 = LA1 LA6
LA23 LA22 A TA list is not completely located within an LA, causing LA is different after UE fall back (MSC pool is not changed)
TA2 LA2
LA3
TA3 LA5
TA4 = LA4 TA5 = LA5
2G/3G LA
LTE TA
Figure 10.19 Unmatched LA/TA areas served by same/different MSC pools.
(MO) to make NAS connectivity and successful CM_service request or CSMT1 (MT). If the new LA is served by a different MSS that is not the one registered in combined attachment and caused the MT call to fail, because the update location message is received from a MSC, which is not aware of any previous paging request, the 3GPP feature of MTRR/MTRF is needed to avoid call drop. An example of the areas of unmatched LA/TA is shown in Figure 10.19, it notes that with MSC pool solution each MSC within the pool can control all LAs belonging to the pool; therefore, in this case the critical areas are limited to the border between the pool areas. The example shown in Figure 10.20 gives an explanation of how the UE knows if it belongs to a different LAC. 3G SIB1 broadcasts common information to all UEs in the cell related to cell access parameters and information related to scheduling of other SIBs. The NAS IE inside SIB1 defines the LAC to which the cell belongs: “cn‐CommonGSM‐MAP‐NAS‐SysInfo” = 7922 Hex (which means 31010). At this point the UE already knows it is camping in a cell belonging to a different LAC. Actually, in a live network, a 100% correct mapping of LTE and 2/3G cells (TA‐LA mapping) is nearly impossible to maintain, especially due to cell breathing and varying indoor coverage, and so on. To address the setup delays and setup failure risks in MSC border areas, as well as eliminate the LAU delay time, a few features for overcoming the MSC border issue: ●●
MSC pool: MSC pool architecture, also known as Iu/A‐Flex, conforms to the 3GPP Rel 5 specifications for connection of RAN nodes to multiple core network nodes. With MSC pool architecture, all MSC servers within a pooled area serve all cells in the pool, eliminating MSC borders and the time delay of inter‐MSC LAUs within the pool. When the UE falls back across the pool and result the call failure problem, there are three solutions: delete the excess neighbor cell, delete the excess frequency points, and replane the 2/3G MSC pool border.
1 CSMT: A flag in location update request message used in CS fallback for MT call for shortening the call setup time. When the MSC receives a LAU request, it shall check for pending terminating CS calls and, if the “CSMT” flag is set, maintain the CS signaling connection after the LAU procedure for pending terminating CS calls. The UE includes the “CSMT” flag in the location update request message that informs the new MSS to delay releasing the NAS signaling connection in order to wait for the incoming call setup from the gateway MSS.
Figure 10.20 How the UE knows if it belongs to different LAC.
402
LTE Optimization Engineering Handbook ●●
●●
Mobile terminating roaming retry (MTRR, TS 23.018), allows that the MSC/VLR where the UE is registered routes the call further to the MSC where the UE has fallen back to. The probability is higher that this use case occurs if no MSC pool is deployed (at MSC borders) or when MSC pool is deployed it can occur as MSC pool borders. The feature of MTRR needs upgrade the HLR, GMSC, and MSC. Additional delay has to be added for an inter‐ MSC location update case (2‐10 seconds). Mobile terminating roaming forward (MTRF, TS 23.018), which is subscriber registration in the new MSC and call forwarded from old to new MSC, needs additional call setup delay (~5 s) and additional load on HLR2 and MSC. MTRF need MSC software upgraded. MTRF is a newer version of the MTRR standard and it solves the MSC border issue by forwarding calls directly from the old MSC to the new MSC in case a fallback is done over an MSC border. MTRF has the advantage over MTRR of not needing inter‐operator agreements and not re‐routing calls back to the GMSC for a second HLR interrogation. This makes MTRF more reliable and easier to deploy. Field tests show that MTRF activation will enable the failure rates decreased from around 9% to less than 3% in the critical areas. The additional call setup time due to MTRF for a CSFB call is around 1 sec.
10.3 CSFB Performance Optimization The main CSFB‐specific optimization tasks to be discussed in this section are minimization of call setup time, improved call setup success rate, and accelerating the return to LTE after the call is completed. CSFB mobile originated (MO) calls and CSFB mobile terminated (MT) calls are expected to have different performances. MO calls are quite easy in that mobility issues can be handled by the UE autonomously; apart from the lack of coverage, the UE is normally able to find a suitable target cell and establish the call, network is minimally involved in this process. MT calls are more tricky that CN mobility is critical as the call is already routed to a specific MSC, and the incoming call is pending in the MSC. 10.3.1 CSFB Optimization 10.3.1.1 Main Issues of CSFB
When originated a CSFB call, the main issues lead to unsuccessful calls are listed below: During paging procedure: the wrong software version of the network, frequency locking of the terminal, terminal software problems, 4G weak coverage, user’s implicit detach, when the user is during fallback, TAU and other procedure conflicts happened, and so on. During fallback procedure: the CSFB function of eNB is not enable, 2/3G frequency configured is incorrect in eNB, weak coverage of 4G or target network, fallback to different MSC POOL, and Pseudo base station case. Call setup procedure: the CSFB function of 2/3G BSC/RNC is not enable, incorrect setting of core network parameter, authentication and handove issues caused by 2/3G weak coverage, the terminal setting blacklist, 2/3G network congestion, 2/3G channel assignment failure, and so on. CSFB call setup time needs special care of CSFB optimization. The main contributor to the additional call setup time is from reading system information (SI) in the new cell.
2 An MTRF call requires some 380% additional load on the HLR and 65% additional load on the VMSC per call since HLR registrations are normally not done on a call basis.
Circuit Switched Fallback Optimization UE can not be paged during TAU procedure Paging
LTE coverage issue UE can not be paged during reselection between RAT
RF optimization and TACLAC match optimization
fallback failure Fallback and access to target cell
CS call establish failure-MSC Pool issue CS call establish failure-non optimal frequency of target RAT
4G-2/3G neighbour optimization
CS call establish failure--Pseudo base station issue
Return back to LTE
Reselection, fast return failure
Figure 10.21 Main issues of CSFB.
Return to LTE procedure: When the call is released, shorter time from 2/3G return to LTE is expected, so the reselection failure, reselection with redirect to LTE failure and fast return failure may occur. In summary, the overview of CSFB‐specific optimization tasks is shown in Figure 10.21. 10.3.1.2 CSFB Optimization Method
CSFB solution can provide optimal reliability for originating call setup, that is, the same reliability that the target system would offer in a given location. Terminating call setup reliability is more challenging and can be affected by TAC/LAC mismatch and/or excessive delay. For CSFB deployment, the LTE mandate feature includes eNB CSFB function, LTE to 2/3G session continuity, redirect (RIM) function, and so on. 2/3G mandate feature includes RIM function, fast return function etc. Before RF optimization, 2/3G neighbor and frequency allocation, FreqRelation, and TAC to LAC mapping should be checked. For CSFB optimization, the four steps including combined attach, paging, RRC release procedure, and access to 2/3G should be detail analyzed to spot the CSFB failure problem. The CSFB optimization KPIs include success rate, access, fallback and return latency, and so on, as shown in Figure 10.22. Table 10.3 gives a simple description of the main CSFB failures, issue‐related nodes, and solutions. Abnormal call procedure Low CSFB originated call setup success rate analysis procedure is shown in Figure 10.23. According to low fallback success ratio and call setup reject, there are different analysis steps. The causes of CSFB call connection failure include: ●● ●● ●●
CSFB fallback failure: CSFB fallback failure will directly impact CSFB call connection 2G secure and authentication procedure failure The added 2G frequency have TCH or SDCCH congestion
403
404
LTE Optimization Engineering Handbook Reasons Extended service request RRC connection release
RRC connection request
TAU or handover happened
RRC connection request (3G)
RRC connection request (3G)
Fall back with RIM
Call confirmed
Multi RRC connection request, interference
Latency Call confirmed
Setup
Setup
Access failure
Access failure
Procedure conflict
Service setup failure
LAU
Alerting
MultiRAB assignment, physical channel reconfig
Interference, sync, poor coverage, wrong parameter setting or other faults TAU or handover happened during extended service request
Call drop
Pilot pollution
Network structure, pilot power setting, antenna height/tilt etc.
Call drop
Interference, poor coverage, power setting etc.
Interference, congestion, wrong parameter setting or other faults
Figure 10.22 CSFB optimization method. Table 10.3 The main CSFB failures and solution. Failure reason
Issue location
Solution
Failure before fallback
SGs paging failure
RF, user
Paging optimization
Paging success, but fallback failure
RF
Optimize the issued cell
Failure after fallback
Assignment failure
RF
Check MSC and target cell response assignment
Call drop
RF
Optimize target cell
Recovery on timer expiry
RF, user
Check RF conditions and MSC‐related timers
Authentication failure after fall back to 2/3G
Low CSFB originated call setup success rate CSFB originated call setup reject
low
CSFB originated call fall back success ratio
· UE authentication issue leads to implicit detach · Analyz through EMM cause
UE not received RRC release with redirection message,need to check RF coverage and interference
high
UE search other RAT cell and try to access
The number of CSFB originated fall back + the number of CSFB originated call setup reject < The number of CSFB originated request
2/3G issues
The feature of CSFB in eNB is not enabled, there is no 2/3G cell frequency in RRC release message
LTE misconfigured 2/3G neighbour cell frequency UE search other RAT cell and try to access
eNB did not transmit RRC connection release message, the CSFB feature is not enable in eNB
The number of UE context modified failure
· UE context modified failure is due to frequently handover · Analyze through failure cause
The number of UE context setup failure
Analyze through failure cause
Figure 10.23 The analysis procedure of abnormal call setup.
Circuit Switched Fallback Optimization
Abnormal combined attach Un‐optimized TAC‐LAC mapping will lead to undesired location update between MSC pools and delays in call setup, in worst cases it will lead to call failure. Check combined attach planning by MME signaling trace, analyze the number of S1_release and LAU message, optimization ECI to CI mapping, and find if there is misconfigured LAC to MSC mapping record in MME, check the type of TAU is “combined TA/LA updating with IMSI attach” or not, and analyze the EMM cause value. There are four main type of abnormal combined attach in a live network, CSFB not p referred, SMS only, EPS only, and low combined attach success rate. According to these four scenarios, core network and RF issues should be spot as shown in Figure 10.24. Abnormal combined attach optimization procedure is shown in Figure 10.25. Paging failure CS paging procedure in case of CSFB is based on three steps: paging is transmitted by core network (CN), UE is reached by paging message on LTE cell, and redirection to target system, and paging answer from the last cell. The causes of CSFB paging failure include: ●●
●●
If there is a paging through S‐Gs interface while UE is returning to LTE: While the call ended and UE return to LTE, if TAU procedures haven’t completed, the S‐Gs interface paging will fail in some MME product vendor. LTE network implicit detach UE: If LTE network is implicit detach UE, the second call would fail. There are many possibilities that LTE implicit detach UE, such as defect in equipment
Combined attach access
CSFB not preferred
SMS only
check CSFB feature for CN
EPS only
low success rate
check EMM Cause TA-LA mapping
1. poor RF 2. check EMM and Extend EMM Cause
Figure 10.24 Abnormal combined attach.
Attach success ratio low low
low
Attach complete ratio
Attach complete message is lost, check RF coverage
Attach accomplished ratio
high
The number of EPS only
high
The number of attach reject message is big. The reason can be found by EMM cause and Extended EMM cause
The number of CSFB not preferred
The reason can be found by EMM cause, if EMM cause is MSC temporarily not reachable, check TA/LA mapping in MME
Figure 10.25 Abnormal combined attach optimization.
The number of SMS only
Core network should support CSFB
405
406
LTE Optimization Engineering Handbook Check if paging is transmitted by CN N
1. check terminal UE is on the TAU procedure 2. check terminal UE is on IRAT(2/3/4G) procedure 3. check S1 is abnormal or not
Y 1. check TAC is wrong or not a. frequently TAU before b. check if last TAU/LAU is abnormal 2. check if RF is poor or not (UE unreachable) 3. check if the number of PDCCH CCE is enough and check if the MCS of paging message is suitable 3. check TMSI is valid or not, check TMSI allocation is success or not 4. check UE issues (paging no response) 5. Paging reject a. IMSI detached for EPS services b. IMSI unknown c. IMSI detached for non-EPS services d. rejected by user
Figure 10.26 Paging failure troubleshooting procedure.
●●
and the QoS cannot be modified dynamically (check if the Dynamic QoS Modification feature is enabled). Poor wireless environment: Poor wireless environment or high interference will lead to a UE cannot receive or resolve paging message.
If paging is not transmitted by CN, it is needed to check that UE is doing data service request or TAU, or UE is interworking between 2/3/4G, these cases maybe caused by CN issues, or there is S1 interface alarm exist. If paging has transmitted by CN, a major cause leading to paging failure on LTE side is the “not camping on a suitable cell,” and it is usually caused by: too many recurrent IRAT reselections including fast return from 2/3G, poor RF, and so on. The troubleshooting procedure is shown in Figure 10.26. Receive RRC connection release issue The normal RRC connection release message includes 2/3G frequency and target cell information and BCCH. In some cases, when UE sends Extended service request, UE starts T3417ext timer, if no RRC connection release is sent from eNB, at the expiration of this timer (10 seconds) UE reselects to UMTS or GSM. eNB logs can be analyzed to find out why eNB does not send RRC connection release with redirect. The causes of CSFB call fallback failure include: ●●
●●
●● ●●
Poor wireless environment: Poor wireless environment or high interference will lead RRC connection failure or cannot receive RRC connection release message. CSFB to 2/3G feature is not enable: When the CSFB feature not be enabled in eNB, eNB will still release RRC connection release message with no frequency. In this situation, the UE will fallback to 3G. CSFBPrio (2/3G) is not set correctly. eNBs with unreasonable 2G frequency: When an eNB added too many 2G frequency (more than 15), there is bad quality, high interference, or across MSC pool 2G frequency will all lead to failure in fallback procedure.
For the abnormal case, it is needed to check eNB CSFB feature is enable or not, check the cause of service reject and inital context setup failure. RRC connection release issues troubleshooting procedure is shown in Figure 10.27.
Circuit Switched Fallback Optimization
Receive RRC connection release Y
N
1. check RRC connection release contains CSFB frequency or not 2. check if there is a handover command before RRC connection release 3. check if there is a TAU procedure going on
1. check CSFB feature enable 2. check CSFB 2/3G frequency configured or not 3. check if there is implicit detach issue
Figure 10.27 RRC connection release issues troubleshooting procedure.
Camp on 2/3G Cell Y End
N 1. check 2/3G cell RF (not the best frequency) 2. MSC pool issue 3. check if there is a pesedo BTS
Figure 10.28 Camp on 2/3G cell issues troubleshooting procedure.
Camp on 2/3G cell After UE received RRC connection release message, the next step is camping on the specific 2/3G cell. During the procedure, some issues mostly found in a live network. It needs to check RF, MSC pool, and pesedo BTS issues, as shown in Figure 10.28. If the UE is successful camping on 2/3G cell, it is still needed to analyze the signaling in 2/3G cell, that is, CM service request/RRC connection setup complete to call proceeding, call proceeding to radio bearer setup, radio bearer setup to alerting, and so on. After setup message is transmitted by the MO UE, the network will page the MT UE, and proceed with the extended service request procedure of MT UE (Figure 10.29). Finally, it needs to focus on the return back to LTE issues after the call is completed from RRC connection release in 2/3G to tracking area update accept in LTE. Tracking area update accept message includes TAU result, T3412, GUTI, TAI, and matched LAC information. Usually long fast return (FR) latency and FR failure are due to movement of the UE or frequency planning, or ECI planning is unreasonable, and so on. The main causes of return to LTE failure are listed below. ●●
●● ●●
Poor wireless environment in hook area: Only when the RSRP satisfied Qrexlevmin (120dBm‐ 124dBm) can the UE access a LTE network. Thus, if the 4G wireless environment is poor, UE cannot access LTE network and stay in 2G. LTE network implicit detach UE Wrong setting of cell reselection priority: If the cell reselection priority of LTE network hasn’t been set to 7, UE will fail in returning to LTE.
10.3.2 CSFB Main KPI
The CS fallback network capability is realized by using the SGs interface mechanism between the MSC server and the MME. The SGs reference point between the MME and MSC server is used for the mobility management and paging procedures between the EPS and CS domain, and it is based on the Gs interface procedures specified in TS 23.060. The SGs reference point is also used for the delivery of both mobile originating and mobile terminating SMS.
407
408
LTE Optimization Engineering Handbook
Secuity key, service type, and UE capability etc.
Bearer capability, coding scheme and called ID etc.
CM Service Request Routing Area Update Request securityModeCommand securityModeComplete Setup
MO Call proceeding
securityModeCommand securityModeComplete Identity Request Identity Response
radio bearer setup radio bearer setup complete
Call Proceeding
MT Paging/CS service notification Extended service request rrc connection reconfiguration complete rrc connection release rrc connection setup complete Call confirmed
radio bearer setup radio bearer setup complete Alerting
If apply early assignment, MO RAB setup will not wait for MT Call
Alerting
Figure 10.29 Signaling in 2/3G cell.
CSFB E2E optimization, it is mainly focused on CSFB paging success rate, CSFB call fallback success rate, CSFB call connection success rate, CSFB return back to LTE success rate, and the related latency KPI. CSFB service KPI optimization needs to be carried out with both the core network and wireless side to promote CSFB service quality. The key interfaces involved in CSFB service such as SGs, S1‐MME, and Mc, and so on, and the key related signaling flows need to be analyzed, to find the key process and the performance of each interface in detail, mining the hidden troubles of the interfaces in the network. The CSFB main KPIs from signaling messages are shown in Figure 10.30. CSFB service processes is complex in that it involves the PS and CS domain. A single indicator of OMC can not reflect the problems. It is necessary that CSFB signaling analysis tools require data collection, correlation, and analysis of signaling processes across different domains of the interface, including the CSFB MO/MT call setup success rate, return to LTE success rate, and the related delay analysis through the 2/3/4G network. The multi‐interface signaling message eNodeB
BSC
MME
Paging Request/CS Service Notification Extended Service Request
MSC Paging Request
1
2
UE Context Modification Request
1
CSFB paging success rate
2
CSFB call fall back success rate
3
CSFB call connection success rate
4
CSFB return to LTE success rate
Service Request
UE Context Modification Response UE Context Release Request UE Context Release Command UE Context Release Complete
Fall back to 2G LU Request / Paging Response Alerting
3 4
Voice Clear Command
5
Clear Complete TAU Request
6
TAU Accept
Figure 10.30 CSFB main KPIs from signaling message.
Circuit Switched Fallback Optimization eNB
MME
BSC
MSC
(1) Extended Service Request (2) InitialContext Setup Respose (3) UEContext Release Complete
(4) CM Service Request (5) Alerting (6) Release Complete
(7) Tracking Area Update Request
eNB
MME
(2) Paging/CS Service notification (3) Extended Service Request (5) InitialContext Setup Respose (6) UEContext Release Complete
MSC
BSC
(1) SGSAP Paging Request (4) SGSAP Service Request
(4) Paging (7) Paging Response (8) Alerting
(10) Tracking Area Update Request
(9) Release Complete
Figure 10.31 MO and MT KPIs related signaling.
by real‐time correlation processing, analysis can realize the end‐to‐end optimization, quickly locate the CSFB terminal and network equipment issues, and improve CSFB service quality and customer perception. MO (mobile originated) and MT (mobile terminated) KPIs are list below, the related signalings are shown in Figure 10.31. ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●●
MO CSFB success rate = A_Alerting/S1_Extended_service_request MT CSFB success rate = A_Alerting/SGs_paging request MO fallback success rate = A_CR(RRC connection release)/S1_Extended_service_request MT fallback success rate = A_PAGING_RESP/SGS_Paging request MO call setup delay = A_Alerting‐‐‐‐‐S1_Extend_Service_request MTpaging success rate = SGSAP Service request/SGS_Paging request MT call setup delay = A_Alerting‐‐‐‐‐SGs_Paging MO fallback delay = A_CR‐‐‐‐‐S1_Extended_service_request MT fallback delay = A_Paging_response‐‐‐‐‐SGs_paging request Return time to LTE = Tracking area update complete‐rrc connection release Return to LTE success rate = Tracking area update complete/MO CSFB call release.
10.3.3 Fallback RAT Frequency Configuration Optimization
In standard CSFB procedure, there is no UE measurement during the fallback, which means CSFB RRC release with redirection is a blind release. UE is allowed to pick any cell on the indicated frequency, or may even try other frequencies/RATs if no cell can be found on the target frequency. The selected RAT frequency is based on UE capability (band/RAT is supported) and prioirty (csFallbackPrio). CSFB frequency priorities defined with the csFallbackPrio parameter for each available frequency relation. Based on these two criterias, when several possible target frequencies have the same priority, the eNB applies a round robin scheme for the selection. It
409
410
LTE Optimization Engineering Handbook
Figure 10.32 Wrong GERAN relation frequency configured.
should add all the 2/3G frequency into one group so that it can be found all the 2/3G frequency in RRC release message. If 2/3G neighbor frequencies won’t be defined, CSFB call service request would be rejected. Besides, it is needed to define both WCDMA and GSM as candidates, by defining both UtranFreqRelation and GeranFreqGroupRelation. By setting parameter csFallbackPrio for UtranFreqRelation to a higher value would set higher CSFB priority for corresponding UMTS frequency. Take GSM for example, GERAN starting ARFCN is based on the configuration in eNB that know how many ARFCNs are defined in OMC database. To speed up CSFB to GSM, it will be smart to geographically configure ARFCNs based on the collected 2G ARFCN and neighboring 2G ARFCNs. Usually the operator can find that the problem area is the target GSM cells’ frequencies are not set in LTE cell, which causes this call back to a bad GSM cell, may lead to poor call quality and finally causes the connect failure. An example is shown in Figure 10.32. Example 1 Assume CSFB GERAN relation frequency is cell A, cell A and cell B is co‐BCCH, B is located in mountain, and when UE fallback to cell B, the rx level is −93 to −99dBm, which result to UE lost. In this case, frequency point of the MSC pool boundary needs to be optimized, to avoid the same frequency configured across the MSC pool. Meanwhile it needs to control the 4G and 2G coverage, to avoid cross coverage. eNB can not configure the GERAN f requency point not belonging the pool. Through the pool boundary, the configured 2G frequency points need to be verified, if the 2G frequency points’ signal Rx lever of the other pools were too strong, it is proposed the BTS and antenna parameters need to be adjusted to control coverage. For GSM, SDCCH establishment success rate impacts CSFB call setup success rate. SDCCH assignment success depends on GSM network quality and the degree of interference. In the CSFB process, as long as the SDCCH assignment fails, then CSFB call setup will fail. Therefore, the fallback GSM cell can be priority selected from the high SDCCH setup success rate cells.
Circuit Switched Fallback Optimization
Figure 10.33 GSM neighbor across the pool.
Example 2 Before call setup in GSM, MT UE camped on the 4G cell belongs to TAC 16640, MSC pool2. After fallback to GSM, the GSM cell belongs to LAC 16793, MSC pool1. During the procedure, UE did not send paging response, and the call failed. After deleting the GSM neighbor across the pool, the issue was solved. Example 2 is shown in Figure 10.33. 10.3.4 Call Setup Time Latency Optimization
CSFB call setup time latency (MO/MT and idle/connected) is calculated as below. For originating call setup the start trigger is when the call is placed, which triggers the Extended service request from the UE to MME. The end trigger is when the UE has moved to 2/3G RAN and received the CS ringing from MSC on the target cell to setup the call on 2/3G. Thus, the call setup delay for originating CSFB call is:
Call Setup LatencyCSFB MO ms
TCS Alert TExtended Service Requestt
In the terminating call setup case, the start trigger is when the UE is paged by the MSC via the message CS service notification and sends Extended service request to MME. The end trigger is when the UE has moved to 2/3G and received the CS ringing from MSC on the target cell to setup the call on 2/3G. Thus, the call setup delay for terminating call setup is:
Call Setup LatencyCSFB MT ms
TCS Alert TExtended Service Requestt
411
412
LTE Optimization Engineering Handbook
Delay is the most critical aspect as CSFB call setup time might be prolonged due to the additional time required to measure the target RAN frequency. It can negatively impact the user perception. MO/MT CSFB call setup time latency is shown in Figure 10.34. In case of terminating calls excessive delay can bring to failures due to timers in the core network. In the best case, the additional delay introduced by CSFB can vary from 0.7 to 2.5 seconds, but it can reach 6−10 seconds or more in case of failures. If originating and terminating UE are both CSFB calls, the additional delay introduced will be much longer. The most critical issue affecting the delay is the redirection to a wrong 2G or 3G frequency, that is, the target RAT/ frequency is not suitable to allow the radio access (Figure 10.35). Let’s take an example of LTE to UMTS, to explain CSFB delay (one side) main contributor and statistics: ●● ●● ●●
LTE release delay, time for signaling on LTE: ~100 ms for idle UEs UMTS cell access delay, time to find and select a UMTS cell: 0.4 sec SIB reading, time to read relevant system information from UMTS cell: 0.2 sec
CSFB call setup time latency Start: NAS extended service request; End: NAS alerting.
Figure 10.34 CSFB call setup time latency.
LTE release delay (fallback to 2/3G)
time to find and select a 2/3G cell and read relevant system information from 2/3G
LTE release delay (fallback to 2/3G)
time to find and select a 2/3G cell and read relevant system information from 2/3G
LTE release delay (fallback to 2/3G)
time to find and select a 2/3G cell and read relevant system information from 2/3G the additional delay introduced by CSFB
Figure 10.35 CSFB typical delay.
2/3G call
LAU procedure in MSC pool
2/3G call
LAU in different MSC pool +MTRF (terminate UE)
2/3G call
Circuit Switched Fallback Optimization ●● ●●
Call setup delay, time to establish call from RRC connection request to alerting: 6.7−8.5 sec If originating and terminating UE are both CSFB call, the additional delay introduced will be 1.2−1.8 sec under RIM procedure.
As mentioned earlier, CSFB will prolong the call setup time. Longer call setup time can result in bad end‐user experience and would be perceived as outages. If we compare a call in the classic GSM network with a CSBF call it will be seen that there are some extra interfaces and procedures that are used. The main contributor to the additional call setup time is from reading system information (SI) in the new cell. CSFB to 2G (GSM) today takes ~1.0−1.5 sec longer than CSFB to 3G in both MO and MT. The reason is slower reading of SIB information in 2G compared to 3G. As you can see the call flows in Figure 10.36, CSFB causes this delay because of the additional interfaces and procedures involved. While in a normal 2G‐2G call it has two nodes and the UE in the CSFB cases it has four nodes involved. Therefore, in 3G, it is typical to cause delays of 1.5 sec while in 2G it increases to 2.5 seconds. As we have discussed in Section 10.2.3.1, there are two features on top of CSFB to improve the call setup delay—one is RIM, the other is DMCR. As mentioned earlier, there are four basic methods for directing the UE to the target system in CSFB call setup, these methods can be either blind or measurement‐based. Call setup time latency of these methods to fallback to 2/3G network is listed in Table 10.4. For redirect, the time to access 3G is typically around one second; obviously this assumes optimized neighbor list and target RAT with sufficient radio quality. Call setup time with PS handover can be optimized by shortening the time to trigger to first measurement report and by optimization of target RAT neighbor lists. In case of a CSFB without PSHO, the data might take 5 to 10 sec to resume, while for CSFB with PSHO it is done almost directly. MSC MSC
2/3G
UE
MME
LTE
Page
UE
Page
Page
Page
Page Page response
2/3G
Extended service request Release Release Cell change
Call setup
Read SI Page response
RIM: –1.2 s improvement for 3G –2.0 s improvement for 2G DMCR: 0.7s improvement for 3G
Call setup
Figure 10.36 Comparison of classic CS call setup in 2/3G (left) and CSFB call setup (right). Table 10.4 Call setup time latency of the four methods.
Mobility method between LTE and 3G
3G/LTE in same CSFB MSS
Overlay CSFB MSS
1) Redirect based CSFB to 3G
+1.0 − 1.5 sec
+2.0 − 3.0 sec
2) PSHO based CSFB to 3G, (delta to 1)
−0.4 sec
−0.4 sec
3) Redirect based CSFB to 3G with SIB, (delta to 1)
−0.3 sec
−0.3 sec
4) Redirect based CSFB to 3G, deferred SIB 11/12 reading (delta to 1)
−0.4 sec
−0.4 sec
413
414
LTE Optimization Engineering Handbook
CSFB with measurement‐based redirect takes some 440 ms longer compared to CSFB without measurements, timing limit for the number of 2/3G cells and carriers must be considered during parameters setting and network planning. The example delay between Extended service request to RRC connection release with redirect info to 3G is listed in Table 10.5. For GSM neighbor cell frequency planning, it is recommended to configure the overlapping cells’ BCCH (3*7 = 21) frequencies as the fallback frequencies. The measured call setup time can also be improved (for incoming calls) with DRX paging cycle optimization and other parameters. By adjusting, that is, parameter DRXPagingCycle from 1280 ms to 640 m, call setup time can decrease 400 ms, decrease T3413PagingTimer (interval between two paging in S1 interface) also can decrease call setup time. In a live network, the paging parameters should be analyzed and fine‐tuned according to the real traffic. Figure 10.37 gives the parameters related with CSFB call setup latency. For CSFB and session continuity, it needs to have TAC to LAC mapping where the UE on 4G cell will be redirected to the underlying 2/3G site. Every TAC can be mapped to one and only one LAC. Accordingly, one TAC can be defined in two TALs but shall be fully included within one LAC to have 1 TAC to LAC mapping. If TAC and LAC does not match in co‐site LTE and 2/3G cells, it will lead to even longer call setup times due to an additional LAU procedure. At the border of MSC pool, it will even cause connect failure, or increase a 1‐second time delay. Figure 10.38 gives an example of CSFB call setup latency under TAC and LAC matches and not matches. Table 10.5 Call setup time latency of different number of 3G cells and carriers. Modus
average time
redirect without 3G measurement
110 ms
redirect with 3G measurement 1 neighbor 3G cell @ 1 UTRAN carrier
556 ms
redirect with 3G measurement 6 neighbor 3G cells @ 1 UTRAN carrier
707 ms
redirect with 3G measurement 12 neighbor 3G cells @ 1 UTRAN carrier
760 ms
redirect with 3G measurement 6 + 6 neighbor 3G cells @ 2 UTRAN carriers
853 ms
UE
eNBs in TAI List
MME
eNB
maxNoOfPaging Records: 16 nB: T (1 PO in 1 PF) DRX : 1.28~0.64s
N3413 : 4 S1 Paging
RRC Paging RRC Paging
MSC
SGs-Paging-Request T3413: 3~5s
pagingDiscardTimer: 3s
S1 Paging S1 Paging
1.28 Sec defaultPagingCycle maxNoOfpagingRecords 16 ¼T nB PagingDiscardTimer 3 Sec tInactivityTimer 5–10 Sec S1_T3413-PagingTimer 3 Sec
S1 Paging SGs-Paging-Request Service Request CSFB To 2/3G
RNC Paging Response or LU
Figure 10.37 CSFB call setup latency–related parameters.
Figure 10.38 CSFB call setup latency under TAC and LAC matches (left) and not matches (right).
416
LTE Optimization Engineering Handbook
Table 10.6 Typical CSFB to GSM call setup latency distribution (second).
Extended service request‐‐ > Alerting
Extended service request ‐‐ > RRC connection release
RRC connection release ‐‐ > CM service request
11.4
0.066
1.941
9.363
1.223
5.1
0.115
1.488
3.545
0.448
5.3
0.11
1.94
3.235
0.474
6.8
1.004
2.39
3.406
0.465
4.6
0.082
1.261
3.257
0.47
14.6
0.202
1.403
12.996
0.47
5.7
0.204
1.83
3.68
0.537
5.5
0.114
2.093
3.346
0.454
6.6
1.509
1.597
3.469
0.444
6.1
0.956
1.822
3.272
0.484
5.6
0.218
1.929
3.491
0.455
5.2
0.521
1.475
3.209
0.63
CM service request‐‐ > Alerting
channel release‐‐> TAU Accept
10.3.4.1 ESR to Redirection Optimization
For CSFB, UE initiates the call by sending Extended service request (ESR) to MME. The time of transition to target RAT can be measured from Extended service request to RRC connection release. Once the UE is in target RAT, it initiates location update procedure if the unmatched LAC is met or even MSS is changed; this will increase the location update time, moreover, IMEI checking and TMSI reallocation may be done at this point. The duration of the procedure can be measured from RRC connection release to CM Service request. After that, the UE is able to start the actual CS call setup and the duration of this procedure can be measured from CM Service Request to Alerting. After the CS call is finished, the UE should return back to LTE and the duration of this procedure can be measured from channel release to TAU accept. The duration between CM Service request to alerting, is basically the same for ordinary CS call in the target system. Comparing to other network, the delay in 4G (ESR to RRC redirection) is a little bit longer in Table 10.6. Test result showing the variation of ESR to RRC redirection is still great; from procedures comparison between idle and connect mode, it can be seen that CSFB from idle mode took much longer time than connected mode due to its signaling exchange to MME which occupies the longest time (Figure 10.39). For CSFB latency in 4G (ESR ‐ > RRC connection release), as call from idle mode takes much longer time than connection mode, one solution to reduce the 4G CSFB latency is to increase the inactivetimer setting to keep the UE in RRC connection. 10.3.4.2 Twice Paging
MSC server paging parameter is 9 s + 6 s, namely if the first paging failure (MSC server does not receive the paging response message), MSC server will send the second paging after 9 s, this is LTE twice paging strategy. Combining MO and MT call signaling to analysis as shown in Figure 10.40, it can be found left signaling reflect that MO’s setup is sent on 10:41:48.789; however, TC fallback time is 10:41:58.189, that means MT UE call prepare to fallback to GSM after 10s waiting. Obviously,
ESM rrcConnectionSetup Procedure
0.066s
0.004s
Connected mode
0.017s
UEContextModification Procedure
0.000s
initialContextSetup Procedure
(SecurityMode Procedure) 0.001s rrcConnectionRelease
Idle mode
0.168s
(0.027s)
0.001s rrcConnectionRelease
Figure 10.39 ESR to RRC redirection from connected mode and idle mode. UE
MME
x x x x
MSC
1st SGs paging request
S1/RRC page S1/RRC page
3s
S1/RRC page
3s
S1/RRC page
3s 3s
Incoming call
Ts5 = 9s
2nd SGs paging request UE Unreachable Voice mail service
Figure 10.40 LTE twice paging strategy and an example from field test.
418
LTE Optimization Engineering Handbook
MT UE did not receive the first paging message due to poor coverage of LTE, and this cause the total call connection period lasts over 18 seconds. 10.3.5 Data Interruption Time
If a user is in an active PS data session (e.g., streaming media) when a voice call is initiated, the inter‐RAT transition and routing area update will interrupt the data transfer. The interruption time will depend on the mobility mechanism, as summarized in Figure 10.41 for an example of UMTS. Using handover‐based CSFB, the data stream interruption time of 0.3 seconds is unlikely to be noticeable. The user will experience much higher (5 seconds) data stream interruption Redirection
Handover Rel-8/ Rel 9 Handover
Rel 9
Rel-8
S1 Tunnel Skip SIBs
Basic
0.3
RRC Release
0.2
0.2
0.2
Acquisition on UTRAN
0.2
0.2
0.2
0.4
2.0
Read MIB & SIBs Camp on Cell
0.1
0.1
0.1
Connection Setup
0.3
0.3
0.3
Optional RAU Procedure
4.0
4.0
4.0
4.8
5.1
6.8
Total Data Interruption Time
0.3
Data rate (kbps) ~5–10s w/o PSHO 3) Resume
LTE
LTE WCDMA GSM (DTM) Time (s) CSFB call 3) Suspend 2) Resume
Figure 10.41 Data interruption time.
4) Resume
Circuit Switched Fallback Optimization
in the redirection‐based Rel 9 SI tunneling and Rel 8 Skip SIBs methods, which may be mitigated in practice by the fact that user attention will already be diverted to initiating an outgoing call or receiving an incoming call. 10.3.6 Return to LTE After Call Complete
After the CSFB call is finished, the UE should return back to LTE if LTE network is a preferred access and coverage exists. This is possible by either UE‐controlled normal inter‐RAT cell reselection or by redirect/handover based on a vendor‐proprietary trigger or by fast return mechanism. Return to LTE after call completion procedure is shown in Figure 10.42. Especially, the 2/3G feature release with redirect to LTE functions as well as fast return, to prevent a UE from being stuck in 2/3G after a CS session. It is not a coverage triggered release with redirect, and is triggered based on 2/3G channel switching state changes, it is a blind mobility mechanism. The latency of return back to LTE measurement can be calculated as: Start: UMTS/GSM RR channel release message; End: LTE NAS Tracking area update accept message. Release with redirect to LTE is an IRAT mobility function that provides basic mobility for an LTE‐capable UE from 2/3G to LTE. Redirection to LTE is controlled by the network and can be initiated at several trigger points. A redirection to LTE can be initiated from the active states CELL_DCH, CELL_FACH, or URA_PCH. When the LTE coverage is good, UE redirected quickly to LTE (~2 to 3 s). With bad or non‐existing LTE coverage, UE will search for 10s for LTE frequencies included in the Release with redirect message; if not successful, UE shall search all LTE frequencies, which most likely will return to UTRAN more than 15 s (Figure 10.43). HLR
MSC
UE
RNC Rrc connection release Rrc connection release complete
MME
eNB RRC connection establishment Tracking Area Update Authentication Request Authentication Response Security Mode Command Security Mode Complete
Location update Tracking Area Accept Tracking Area Complete After the RRC connection release, a new combined attach process is need
EMM Info UE Context Release Command UE Context Release Complete RRC Connection Release
Figure 10.42 Return to LTE procedure.
419
The info of redirect to LTE in the RRC release message was sent, meaning the feature “Release with redirect to LTE” can be achieved. The LTE-CSFB-UE can return to LTE as soon as CS call ending.
Figure 10.43 Release with redirect to LTE.
Circuit Switched Fallback Optimization
For the feature of fast return (FR), there are two methods. One is network side FR, and when the call is ended, the 2/3G network releases the user channels and distributes LTE frequencies, and without waiting for changing to an idle state, the UE will access LTE cells according to the indicated LTE frequencies. The other one is UE independent FR, and when the call ends, based on the cell information before fallback, the UE return 4G independently (need UE support). Field tests show that the minimum fast return to LTE latency is 0.5 s (Figure 10.44). Without the feature of fast return, the normal procedure is that the user stays in 2/3G after the CS call has been released and perform normal idle mode mobility. For idle mode cell reselection, when the call ended, the user stays in 2/3G cells, it can reselect to a high priority LTE cells by reading the 2/3G system information, which contains LTE neighbor cell information. User may stay in 2/3G for at least 10−15 s, as shown in Figure 10.45, which will have impact to data performance. Fast return to LTE after call release is introduced to decrease the packet data transfer outage time in order to ensure that 4G‐capable UEs will spend as much time as possible under LTE coverage. With that feature, multi‐RAT mobiles can be requested to select an LTE cell directly after the speech call release, hence preventing the mobile to first select a 2/3G cell and then re‐select an LTE cell. As soon as the CS speech call is over, the UE is redirected to the LTE layer by methods list below: Fast return is done by including the LTE EARFCN in the IE “Cell selection indicator after release of all TCH and SDCCH” of the channel release message. UE will skip the LA and RA update procedures in 2/3G, and the user can quickly resume data services in LTE. When the network side enable fast return function, it can be seen carrying a 4G macro base stations and ventricular frequency division sites in Release 2/3G talk after the end of the message, so that the UE can be quickly re‐select back to the 4G network. Fast return to LTE at CS call completion needs about 1 second, which is a faster procedure than 2/3G to LTE cell reselection (15–20s). Longer return time to LTE can affect end‐user experience. Longer or delayed return time to LTE can exploit UE behavior and would delay return to LTE through chain reactions in certain circumstances. TAU rejections or PS session deactivation can also result in delays in return time to LTE. 2G
4G Measure LTE neighbor cells
CS call release
Sync to target LTE cell and read SIB
TAU
Figure 10.44 Fast return.
4.62s 2.87s
2G Release
Read 2G SIB, Location area update
9.73s 1.75s Routing area update
8.53s
1.2s
Measure LTE cell
Return to LTE
UE out of service
Figure 10.45 CSFB to 2/3G, cell reselection back to LTE, and typical latency.
UE out of service
421
422
LTE Optimization Engineering Handbook
10.4 Short Message Over CSFB Short message in LTE is the same service as in GSM and WCDMA. The solution re‐uses the SGs interface between MSC and MME defined to CSFB to carry SMS messages encapsulated in the NAS signaling channel between MME and terminals. In the SGs interface, NAS procedures are used to translate circuit based SMS messages from the legacy network into the equivalent packet–based SMS messages for the LTE network and vice versa, which does not require “fallback” to 2/3G network to send or receive SMS. Thus, the “SMS over SGs” method can be deployed without deploying CSFB. The solution is ideal for initial LTE services considered to be too complex for these initial terminals and so the NAS signaling based solution developed for CSFB has been extended to offer a low‐cost solution for SMS to data‐only devices. ”SMS over SGs” method can be used in an LTE data only network to provide SMS coverage for LTE UEs. At least one MSC in the legacy network must be equipped, through a software upgrade, to use the SGs interface. If CSFB is set up between the legacy network and the LTE network, then all the MSCs in the legacy network are already set up with the SGs interface. So it does not need to do any additional work to support SMS, if CSFB is already set up. UE needs SMS service but CSFB does not indicate this specific condition with the “SMS‐ only” indication in the EPS/IMSI attach request and combined TA/LA update procedures. This allows an operator to deploy the SGs for SMS delivery over LTE only (without CSFB support), as shown in Figure 10.46. In addition, this allows the MME to use a dedicated algorithm for the selection of the MSC that supports those UEs. In a live network, it is possible that only certain MSCs in the network (one in minimum) is configured to support SGs when the network only supports SMS for SGs operation. However, such a minimal configuration can cause inter‐MSC location updates to be performed at every movement into/out of LTE coverage. SMS is delivered over signaling channel and no data path needs to be established. Once the UE is attached to both MSC and EUTRAN and if MSC wants to deliver a SMS to UE, it will simply send a downlink unit data to MME with SMS content. MME will dump this message in NAS message and send it to UE and in the same way if UE
1. Subscriber registers in MSC by CS signaling over Uu, S1and SGs-interface (attach/Location update).
Uu
3. SMS page over SGs, S1, Uu eNB
S1
MME
LTE SGs
2. Incoming SMS to subscriber in LTE
4. UE responds and receives CS-SMS while roaming in LTE MSC-S
Figure 10.46 SMS over SGs Interface.
SMS-SC
Circuit Switched Fallback Optimization
LTE RAN
UE UE
LTE RAN
EPC
CS Core
EPC
CS Core
SMS
1. Service Request
1. Paging
2. Uplink SMS Delivery
3. Downlink SMS Delivery report
SMS Delivery report
2. Service Request 3. SMS Delivery / Downlink data 4. SMS Delivery report / Uplink data Delivery report
Figure 10.47 Overview of mobile originating and terminating SMS.
wants to send a SMS it will dump the message in NAS message and send it to MME. Mobile originating and terminating SMS procedure is shown in Figure 10.47.
10.5 Case Study of CSFB Optimization CSFB optimization e2e view The common reasons of CSFB failures include pseudo GSM base station, GSM TCH/SDCCH channel congestion, not received paging messages by the called terminal, 4G coverage issues like poor RSRP as well as with good RSRP but poor SINR, 4G cell overlapping, and overshooting, and so on, neighbor cell issues, PCI mod 3 issues, handover issues, TAC‐LAC mapping issues, and other issues. 10.5.1 Combined TA/LA Updating Issue
It is a combined TA/LA updating when CSFB UE is doing TA updating. It required not only tracking area updating success, but also location area updating success. For CSFB optimization, when MT failure occurred, it should be firstly checked whether LA/TA is mapping correctly. In this case as shown in Figure 10.48, only track area updating success in TAU accept but no TA/LA updated, location area updating is failure and cause is MSC temporarily not reachable. Then UE will send track area update request every 10s until reach tracking area updating attempt counter. Finally, the UE starts to select UTRAN radio access technology. The other example is misconfiguration of TAC/LAC that results CSFB UE can not camp on LTE network. When the phone is powered on, 4G signal flashes quickly, then access to 3G immediately. From the signaling shown in Figure 10.49, it is found that when the UE finishes the attach procedure, detach request follows up. UE attach request message and core network gives the attach result, which are shown in Figure 10.50. As shown in Figure 10.49, it is found that after obtained the subscriber information from HSS, the UE sends session request to SGW without combined TA/LA update. After investigation, it is found the reason is that there is no configuration of TAC‐LAC‐MSC mapping in the MME.
423
Figure 10.48 Combined TA/LA updating issue.
Circuit Switched Fallback Optimization
UE capture the subscriber information from HSS
UE sends session request to SGW
Figure 10.49 UE signaling after power on.
10.5.2 MTRF Issue
Without MTRF feature, MSC pool boarder issue will cause call setup failure. When the UE attached it is associated with an MSC based on the TA/LA mapping, if at the borders of two cells, a CFFB is triggered and the UE selects a 2/3G cell belonging to an MSC (pool) other than what the subscriber is registered with, and in the this case normally the CS call should fail. In order to avoid call setup failures, the MTRF features was implemented and with MTRF the call and subscriber is transferred from the registered MSC to the MSC, which was selected during the CSFB call. MTRF is a supplementary feature to CSFB for boarder cases where the LTE cells overlap with 2/3G cells belonging to different MSC, but it can be used in 2/3G only networks to decrease the call setup failure rates. Usually, there are two scenarios for the MTRF: one on borders where mapping of TA/LA might not be correct and another for physical movement of UE between two MSCs. In a field test shown in Figure 10.51, it is found the call failure in 3G LA 54057(MSC2). After the terminated UE location area update procedure in 54507(MSC2), UE can not receive the setup message, and stay in connected state, after 10 secs receive the signaling connection release message from the network, the RF conditions is really good during the whole procedure. This is the typical issue related with LA update. After signaling analysis, it can be positioned as MTRF issue between MSC1 and MSC2, resulting in the failure of the called. When the UE fallback to different MSC, it will not be able to acquire the information of the initial attached MSC. The cause of the case is the possibility that TAs and LAs are not 100% overlapped at the border between MSCs coverage. For this kind of failure, it needs to make sure the LTE subscriber is physical located in the identified critical area, and the LTE subscriber to be called is registered in old MSC, and further verified that the LTE subscriber sends a location update request with CSMT flag toward the new MSC. 10.5.3 Track Area Update Reject After CSFB
When the CSFB call is completed, tracking area update reject usually happened after release with redirect to LTE. It would delay the reestablishment time for PS service.
425
UE request : Combined EPS/IMSI attach
Figure 10.50 Attach request and result.
Core network assign: EPS only
Figure 10.51 Signaling procedure.
428
LTE Optimization Engineering Handbook
10.5.3.1 No EPS Bearer Context Issue
UE needs to initiate track area update request after UE finishes call session and return to LTE network, but network replies track area update reject sometimes, the cause value (40) is “No EPS bearer context,” as shown in Figure 10.52. Compare rejecting “Track area update request” with accept “Track area update request,” as shown in Figure 10.53, it shows if there are not any EBI (EPS bearer identity) active specially EBI5 inactive. But accept “Track area update request” with EBI5, which is the default EPS bearer. The UE shall delete the list of equivalent PLMNs and deactivate all the EPS bearer contexts locally, if any, and shall enter the state EMM‐DEREGISTERED.NORMAL‐SERVICE. The UE shall perform a new attach procedure. For some vendor, it had been found if eNB doesn’t enable dynamic QoS modification feature, UE will still use 2Mbps UL/DL bandwidth or be detached directly when it returns LTE form 2G with data traffic. So it is needed to enable the feature so that the different networks can negotiate the user’s QoS dynamically. 10.5.3.2 Implicitly Detach Issue
For implicitly detached, the network detaches the UE, without notifying the UE. MME doesn’t inform any other node about it, so when new TAU is coming, it’ll be rejected, and new attach will be performed by UE. This is typically the case when the network presumes that it is not able to communicate with the UE, for example, due to radio conditions. The UE shall delete the list of equivalent PLMNs and delete any mapped EPS security context or partial native EPS security context, and shall enter the state EMM‐DEREGISTERED.NORMAL‐SERVICE. If the rejected request was not for initiating a PDN connection for emergency bearer services, the UE shall perform a new attach procedure. An example shown in Figure 10.54 that as checked UE previous behavior in UMTS network, it is found that UE deactivated PDP context before UE entered LTE network. Further to check the trace signaling in MME, it is found that before UE sends TAU to come back LTE after CSFB, MME receives the “delete bearer request” from SGW and responses the request, so MME deletes the bearer but cannot inform UE, which was in 2/3G, also, that is why the UE TAU fails due to implicitly detach. When UE re‐enters LTE network, UE has no EPS context, and its state in EPC is EMM‐DEREGISTERED, which will cause UE implicitly detach. EPC will reject any behaviors of UE before UE performed a new attach, so “Track area update request” was rejected by EPC (Figure 10.55). UE fallback to UMTS network, when UE has no PS service, PDP context will be deactivated. If UE has PS service, PDP context will be hold on. “Track area update reject” is a normal phenomenon when PDP is deactivated by UE. 10.5.3.3 MS Identity Issue
From field tests, sometimes the reason for TAU rejection is “MS identity cannot be derived by the network” cause #9 as shown in Figure 10.56. These cases usually happened for communication problem between SGSN and MME. 10.5.4 Pseudo Base Station
The LAU procedure failed after UE fallback to illegal 2/3G base station and call failed. One example shows that the UE starts the CSFB call in the LTE cell and fallback to GSM cell (ARFCN/ BSIC: 67/52,RxLev: −54dBm, LAC: 14555, Cell ID: 4113) in Figure 10.57. Then the LAU request is rejected by the network and call failure. After verification, the GSM cell (LAC:14555, Cell ID:4113) does not exist in the GSM network, which is an illegal GSM base station cell. The solution of these kinds of issues is trying to get rid of the illegal base station or delete the illegal 2G base station frequency point relation in the LTE cell.
Network starts the connection release, and at the same time, UE sends an attach request
Figure 10.52 Track area update reject.
Figure 10.53 Two‐track area update request message.
Figure 10.54 Implicitly detach.
Figure 10.55 TAU reject caused by implicitly detached.
Figure 10.56 MS identity issue.
Figure 10.57 Pseudo base station.
434
11 VoLTE Optimization Voice over LTE (VoLTE1) enriches LTE network with more voice‐related capability with common EUTRAN and EPC infrastructure. Common packet‐based mobility is also applicable for both LTE data and VoLTE services. In this book, “VoLTE” is used as the collective notation for both voice and conversational video over LTE. VoLTE voice call is based on GSMA PRD IR.92 and VoLTE video call is based on GSMA PRD IR.94. Introducing VoLTE on IMS provides the service provider with a true converged network where services are available regardless of the access type network. VoLTE provides a first line telephony service with high voice quality and short call setup. Voice and video are using QoS bearers with guaranteed bit rate to secure the service characteristic. Voice over LTE allows very fast call establishment (~1 sec) versus CSFB toward 3G (~5 sec) and even more in case of CSFB toward the GSM (~8 sec). VoLTE avoids 4G data service interruptions and preserves the LTE data experience during speech communications while the throughput of concurrent data session is typically reduced in case of CSFB to 3G and even suspended in case of CSFB to 2G. Table 11.1 depicts the differences between CSFB and VoLTE UE call procedure. The LTE standardization work has established that voice will predominantly be supported by an all‐IP network centralized on the IP multimedia subsystem (IMS). IMS architecture provides integrated voice, data, and multi‐media services interworking between different access networks. IMS‐based VoLTE puts the IMS in the center of the voice core network, managing the connectivity between subscribers and the implementation of policy control. The voice service is then managed by a specially designed VoIP application server. IMS‐based VoLTE is standardized by 3GPP and it’s considered to be the target network infrastructure from long‐ term perspective. Usually initial LTE coverage is non‐contiguous, when UE is out of LTE coverage, single radio voice call continuity (SRVCC,2 voice call handover to CS in GSM or UTRAN) is used to keep the voice call continuity. VoLTE phone can work in a variety of modes, attached to the different networks (2G, 3G and LTE), even also attached to two networks by dual standby terminal and therefore when UE originate/terminate the call, terminal/network needs to select which the network to be accessed. This process is called domain selection according to the network registration information. Terminating access domain selection (T‐ADS) realizes the function of domain selection, T‐ADS is processed by application server to determine if the call is for PS or CS domains. 1 In this part, VoLTE also called non‐native VoLTE, which is SIP client‐based. Third‐party applications can register to IMS system and establish VoLTE call. 2 OTT call might drop in this case.
LTE Optimization Engineering Handbook, First Edition. Xincheng Zhang. © 2018 John Wiley & Sons Singapore Pte. Ltd. Published 2018 by John Wiley & Sons Singapore Pte. Ltd.
VoLTE Optimization
Table 11.1 Difference between CSFB and VoLTE UE call procedure. CSFB UE ‐ LTE access selected
VoLTE UE ‐ LTE access selected
Detect available network Attach to the EPC and CS network over LTE Setup Internet APN and do some browsing Paging/call preparation between UE and MSC over LTE Place a call/receive a call
Detect available network Attach to the LTE network Setup IMS APN and find P‐CSCF(s) Register in IMS Place a call/receive a call (keeping current LTE access)
VoLTE signaling and payload packets are supported by VoIP‐specific protocol stacks. E2E connections are managed using SIP with IMS, it requires specific QoS for voice bearers and SIP signaling to achieve performance for a satisfactory user experience and requires specific feature enhancement to achieve capacity and performance.
11.1 VoLTE Architecture and Protocol Stack 11.1.1 VoLTE Architecture
Voice service on EUTRA is available when IMS is installed in the core network. The VoLTE network architecture consists of E‐UTRAN, LTE core, PDN, and IMS. It interworks with 3G, which consists of UTRAN, UMTS core, and a circuit switched (CS) network. The MME provides functions that allow LTE and 3G to interwork. Voice service is based on VoIP session, which is controlled by a companion SIP session. Both devices need to be registered on IMS for VoLTE to VoLTE calls. The solution is characterized by a SIP proxy, VoIP application server (AS), and media gateway (MGW) co‐located with the MSC. It uses pre‐configured policy rules on the PDN‐ GW for binding VoIP sessions to EPS bearer and for QoS provisioning. IMS core network is also needed to be with centralized CSCF, HSS, and VoIP AS, using dynamic policy provisioning with co‐located PCRF (Figure 11.1). The P‐CSCF is the first point of contact in IMS system for the UE for mobile access networks. The P‐CSCF forwards the SIP messages received from the UE to an I‐CSCF, E‐CSCF or S‐CSCF (and vice versa). 11.1.2 VoLTE Protocol Stack
All new interfaces of VoLTE and the new network protocols are represented in Table 11.2. Protocol stack for VoLTE audio packet, extracted from official document IR.92 – “GSMA PRD IMS Profile for Voice and SMS.” Session initiation protocol (SIP) is a popular protocol used to create, modify, and terminate multimedia sessions, essentially negotiating a media session between two users. SIP is not a transport protocol and does not actually deliver media, leaving that task to RTP/RTCP. Session description protocol (SDP) negotiates the multimedia characteristics of the session between sender and receiver (codecs, addresses, ports, formats, bandwidth require for the session). IMS multimedia uses real‐time transport protocol (RTP) over UDP, RTP was originally defined in 1996 then redefined in RFC 3550 in 2003. RTP added a sequence number in order to
435
436
LTE Optimization Engineering Handbook
MSS
MGW
GERAN/UTRAN
Call control signalling User plane traffic Other control signalling
SGs
D
TAS HSS
Sv
Cx S6a
Rx PCRF
MME LTE-Uu
S1-MME
Sh I/S-CSCF BGCF
Ma, ISC Mi
MGCF
Mw P-CSCF Mn
Gx
S11
Mb eNB
S1-U
S-GW
P-GW
MGW
Figure 11.1 Nodes for VoLTE.
Table 11.2 Interfaces of VoLTE and protocols. Interfaces
Components LTE
Protocol
Sv
MSC server – MME(SGSN)
GTP‐Cv2
I2* Mj/Mg/Mx
MSS‐Server – IMS‐I‐CSCF
SIP
Cx
HSS(NSN)‐IMS Core
Diameter
Mw
P‐CSCF –Core IMS
SIP
Gm
P‐CSCF –UE
SIP
Mb
ATGW‐ MSC server
RTP
Rx
AF‐PCRF
Diameter
Gx
PCEF –PCRF
Diameter
identify the lost packets. Together with a new timestamp field it allows the receiver to play the packets in the correct order. Other new fields are SSRC (synchronization source, all the packets have the same SSRC identifier indicating that they are from the same source) and CCRC (contribution source) allow the tracking of one or multiple (in case of a conference) sources for the packet. RTP is used in conjunction with the real time transport control protocol (RTCP). While RTP carries the media streams (audio or video), RTCP monitor transmission statistics and quality of service information. RTCP uses a separate flow from RTP. It is transported over UDP as well, and its purpose is to collect statistics on a given media connection including packet loss, jitter, round trip delay, and monitor the quality of the data transmission. RTCP provides feedback on the transmission and reception quality of data carried by RTP periodically. Each RTCP packet always contains either a sender or receiver report, sender report includes an absolute timestamp,
VoLTE Optimization IMS Signaling
LTE Signaling
SIP*/ SigComp/ IPSec
VOIP Traffic
RTP
RTCP
UDP
TCP
NAS
IP
RRC ROHC
PDCP AM
RLC
AM
UM
MAC PHY Gm Uu
S-GW
S1/u
UE
S1/S8
PDN-GW
SGI
P-CSCF
eNodeB
SDPSIP
SDPSIP UDP/IP
UDP/IP Relay
Relay
GTPv1-U
GTPv1-U
GTPv1-U
GTPv1-U
UDP/IP
UDP/IP
UDP/IP L2
UDP/IP
MAC
UDP/IP L2
L2
UDP/IP L2
L2
L2
L1
L1
L1
L1
L1
L1
L1
PDPC
PDPC
RLC
RLC
MAC L1
Mb S-GW
S1/u
Uu UE
S1/S8
PDN-GW
SGI
BGF
eNodeB
AMR RTP/RTCP
RTP/RTCP UDP/IP
UDP/IP Relay
Relay
GTPv1-U
PDPC
PDPC
RLC
RLC
UDP/IP
UDP/IP
UDP/IP
UDP/IP
MAC
MAC
L2
L2
L2
L1
L1
L1
L1
L1
L2 L1
GTPv1-U
GTPv1-U
GTPv1-U
UDP/IP
UDP/IP
L2
L2
L1
L1
Figure 11.2 VoLTE‐related protocol stack.
to enable synchronization with different streams, receiver report includes number of received packets, to enable QoS determination. For VoLTE, RTCP is not sent during active media transfer but is sent when the call is placed on hold. IMS signaling is carried by SIP messages, which are carried over UDP or TCP, and compared to TCP, UDP has less overhead and no transport layer retransmission (to avoid increase in end‐to‐end delay) (Figure 11.2). The response codes are consistent with, and extend, HTTP 1.1 response codes. There are six classes SIP messages defined: 1xx provisional, 2xx successful, 3xx redirection, 4xx request failure, 5xx server internal error, and 6xx global failure, which are shown in Figure 11.3 and Table 11.3.
437
438
LTE Optimization Engineering Handbook
SIP MESSAGES
REQUESTS/METHODS
REGISTER INVITE ACK BYE CANCEL OPTIONS
RESPONSES
PROVISIONAL
FINAL
RFC 3261 100–199 100 Trying 180 Ringing, and 183 Session progress
MESSAGE SUBSCRIBE PUBLISH NOTIFY INFO UPDATE REFER PRACK
> 199 2xx Success 3xx Redirect 4xx Client Mistake 5xx Server Failure 6xx Global Failure
OWN RFCs
A
Request
B
Provisional response Provisional response Final response
Figure 11.3 SIP messages.
11.1.3 VoLTE Technical Summary
VoLTE delivers voice with high QoS (QCI 1, GBR EPS bearer) in transport and radio scheduling, but admission and congestion control may be efficiently used to preserve the voice quality experience in presence of heavy traffic loads, so non‐GBR EPS bearers (QCI = 5) are used for
VoLTE Optimization
Table 11.3 SIP messages code. 400 Bad Request 401 Unauthorized 402 Payment Required 403 Forbidden 404 Not Found 405 Method Not Allowed 406 Not Acceptable 407 Proxy Authentication Required 200 OK 408 Request Timeout 202 Delivered 409 Conflict 300 Multiple Choices 410 Gone 411 Length Required 301 Moved 413 Request Entity Permanently Too Large 302 Moved 414 Request‐URI Too Long Temporarily 415 Unsupported 305 Use Proxy Media Type 380 Alternative 416 Unsupported Service URI Scheme 100 Trying 180 Ringing 181 Call Is Being Forwarded 182 Queued 183 Session Progress
420 Bad Extension 421 Extension Required 423 Registration Too Brief 480 Temporarily Unavailable 481 Call/Transaction does not exist 482 Loop Detected 483 Too Many Hops 484 Address Incomplete 485 Ambiguous 486 Busy Here 487 Request Terminated 488 Not Acceptable Here 491 Request Pending 493 Undecipherable
500 Server Internal Error 501 Not Implemented 502 Bad Gateway 503 Service Unavailable 504 Gateway Time‐out 505 Version Not Supported 513 Message Too Large 600 Busy Everywhere 603 Decline 604 Does Not Exist Anywhere 606 Not Acceptable
SIP and XCAP. In a live network, voice domain selection is IMS PS voice preferred and CS voice is secondary. VoLTE can provide base telephony service and supplementary services and supplementary service management using Ut with XCAP procedures. The main part of VoLTE need focused is described as the following: VoLTE bearer management, includes PDN connection for IMS APN,3 signaling bearer setup, P‐CSCF discovery, home‐routed PDN connection/APN for Ut, handling of loss of PDN connection, signaling, and GBR bearer. IMS feature part, includes ISIM based authentication (USIM fallback), IPSec protection of signaling, both Tel‐URI and SIP URI, SigComp, GBA (recommended) or http digest authentication for Ut, early dialogues and media, and IMS emergency. IMS media, includes AMR narrow‐band and wide‐band codec and payload format, RTP profile/data transport, RTCP usage, and Jitter buffer management. SMS, includes SMS over IP (IMS) and SMSoSGs. Wireless feature as shown in Table 11.4, includes CDRX, semi‐persist scheduling, TTI‐bundling, forward handover with context fetch, and so on. 11.1.4 VoLTE Capability in UE
The UE indicates its E‐UTRA capabilities in the UE E‐UTRA capability information element at connection setup (RRC UE capability information message). The FGI (feature group indicator) indicates the functionalities supported by the UE. FGI 3, 7, and 27 will indicate the VoLTE capability of UE, and more information can be found in 3GPP TS 36.331 V9.16.0 (2013‐09), which is shown in Figure 11.4 and Table 11.5. 3 In the VoLTE work in GSMA, access point name (APN) used for IMS services has been defined in GSMA IR.88. The IMS APN shall be added to the list of APN names that are used in the LTE/EPC network, this means that IMS APN is added to the subscriber EPS profile in HSS and policy controller.
439
Table 11.4 VoLTE main wireless feature. Features
Description
Benefits
CDRX
Connected mode short DRX allows UE to go to sleep between frames
Better talk time
HD vocoder
Core network and device support for AMR‐WB
HD voice with high quality
Semi‐persist scheduling(SPS)
Efficient scheduling for VoIP‐type traffic
System efficiency and capacity
TTI‐bundling
Bundle 4 TTI together in UL
Improve VoIP link budget
E‐911 support w/positioning
Emergency call with positioning over UP or CP, using E‐CID with TA,OTDOA with PRS and A‐GPS
LBS application
Voice call continuity(VCC)
Voice call continuity to 3G/2G CS and PS domain
Needed if LTE coverage is not ubiquitous
Forward handover with context fetch
Improve handover reliability
Robust mobility
Bit
Definition
Note
3
. 5 bit RLC UM SN . 7 bit PDCP SN
VoLTE Bit 7 = 1
4
Short DRX cycle
Bit 5 = 1
5
Long DRX cycle, DRX command MAC control element
7
RLC UM
VoLTE
9
EUTRA RRC_CONNECTED to GERAN GSM_Dedicated handover
SRVCC Bit 23 = 1
11
EUTRA RRC_CONNECTED to CDMA2000 1xRTT CS Active handover
SRVCC Bit 24 = 1
20
If bit number 7 is set to ‘0’: -SRB1 and SRB2 for DCCH + 8x AM DRB If bit number 7 is set to ‘1’: -SRB1 and SRB2 for DCCH + 8x AM DRB -SRB1 and SRB2 for DCCH + 5x AM DRB + 3x UM DRB
27
. EUTRA RRC_CONNECTED to UTRA FDD or UTRA TDD CELL_DCH CS handover, if the UE supports either only UTRAN FDD or only UTRAN TDD . EUTRA RRC_CONNECTED to UTRA FDD CELL_DCH CS handover, if the UE supports both UTRAN FDD and UTRAN TDD
28
TTI bundling
29
Semi Persistent Scheduling
-Regardless of what bit number 7 and bit number 20 is set to, UE shall support at least SRB1 and SRB2 for DCCH + 4x AM DRB -Regardless of what bit number 20 is set to, if bit number 7 is set to ‘1’, UE shall support at least SRB1 and SRB2 for DCCH + 4x AM DRB + 1x UM DRB
SRVCC Bit 13 = 1
RRC featureGroupIndicators= ‘11111110 00001101 11011000 10000000’ The indexing starts from index 1, which is the leftmost bit in the field
Figure 11.4 FGI bits – extract of VoLTE‐related bits.
Figure 11.4 (Continued)
442
LTE Optimization Engineering Handbook
Table 11.5 Feature group indicator (bit number 3, 7, and 27). 3
5bit RLC UM SN; 7bit PDCP SN
can only be set to 1 if the UE has set bit number 7 to 1.
Yes, if UE supports VoLTE
7
RLC UM
can only be set to 0 if the UE does not support VoLTE
Yes, if UE supports VoLTE
‐ EUTRA RRC_CONNECTED to UTRA FDD/TDD CELL_DCH CS handover, if the UE supports either only UTRAN FDD or only UTRAN TDD ‐ EUTRA RRC_CONNECTED to UTRA FDD CELL_DCH CS handover, if the UE supports both UTRAN FDD and UTRAN TDD
related to SRVCC ‐ can only be set to 1 if the UE has set bit number 8 to 1 and supports SR‐VCC from EUTRA defined in TS 24.008
Yes for FDD, if UE supports VoLTE and UTRA FDD
27
11.2 VoIP/Video QoS and Features 11.2.1 VoIP/Video QoS
The introduction of VoIP poses a number of challenges over LTE access network. To make VoIP attractive for commercial deployment, the capacity will also be desired to be either comparable to or more than that of legacy circuit system. To provide the quality of CS voice without excessive degrading the system capacity, end‐to‐end quality of service (QoS) support in the wireless and wireline packet network infrastructure is essential. Voice is sensitive to data loss, but robust coding and well‐functioning error concealment units makes voice more tolerant to data loss than the other media types. Voice communication has very stringent end‐to‐end delay requirements and voice is thus very sensitive to delay and jitter. Video telephony is very sensitive to data loss, video telephony requirements on frame errors are almost a factor of 10 less than for voice. Video telephony has also very stringent end‐to‐end delay requirements and is thus very sensitive to delay and jitter. SIP signaling, XCAP4 signaling and other signaling must be error‐free for the transaction to be successful. TCP and SIP retransmissions are the mechanisms that guarantee that the data transfer becomes error free even though the EPS bearer introduces packet losses. SIP signaling is time critical (to allow fast call setup, etc.) and thus have a rank sensitive for delay/jitter, whereas XCAP signaling has less‐stringent delay/jitter requirements. Policy and charging control (PCC) enables QoS supervision and control for the media parts of a SIP session. Policy and charging rule function (PCRF) supports 3GPP standardized PCC procedures and makes policy and charging decisions based on input from user subscription information, services information, and so on. PCRF creates policy rules based on session data and push appropriate policy rules (bandwidth, QoS, traffic flow) to P‐GW, and P‐GW interprets the rules and takes actions to establish required EPS bearers for VoLTE. PCC rules contain service data flow (SDF) description and charging and QoS properties to be applied for the flow identified by the SDF through the Rx and Gx interface. When a VoLTE call is to be setup, a dedicated bearer for voice with QCI1 will be setup, initiated from the P‐CSCF over Rx to PCRF, and then PCRF check the policy control and request
4 XML Configuration Access Protocol (XCAP) is the protocol that is used by the UE to configure various parameters for supplementary services such as call hold and call wait.
VoLTE Optimization
1a) SIP INVITE (on SIP bearer) Applic ation
1b) SIP 18x/200 (on SIP bearer) P-CSCF/ IMS AGw
8*) Optional: Bearer established notification LTE radio unit
6a) Session Management Request incl. TFT
2) Application/ Service Info (start/end session)
EPS Bearer QoS
Rx
7) RRC Reconf.
eNodeB
5) E-RAB Setup Request
MME
EPS Bearer QoS
S&P GW
4) Create Bearer Request EPS Bearer QoS
6b) Session Management Response
3) Policy and Charging Rules Provision
PCRF
EPS Bearer QoS
P-CSCF
INVTE sip Content-Type: application/sdp Content-Length: 540 v=0 o = sip:+16309798028 s = session c = IN IP4 135.185.13.175 b = CT:1000 t=0 0 m = audio 51306 RTP/AVP 4 a = rtpmap:4 AMR-WB/8000 m = video 7834 RTP/AVP 34 35 a = rtpmap:35 H264/90000 a = rtpmap:34 GVA/90000 a = sendrcv
m = audio RTP/RTCP, AMR-WB,BW* m = video RTP/RTCP, H264/GVA,BW*
PCRF
Audio RTP/RTCP: QCI 1, ARP, GBR Video RTP/RTCP: QCI 6/7, ARP, GBR
IMS PGW QCI 5 - SIP SIGNALLING QCI 1 – AMR-WB QCI 6/7 H.264/GVA
PGW Packet Filter(s) 1 Packet Filter(s) 1 Packet Filter(s) 1
TFT1
Control Sig
IMS (P-CSCF)
TFT2
Audio
MGW or UE
TFT3
Video
UE
8
8
8
Figure 11.5 QoS negotiation principles at session setup.
the dedicated bearer to the PGW through the Gx interface, and all the way through the network to the UE (Figure 11.5). In LTE, eNB only knows about the service is the QCI, service can be dependent triggered by QCI, for example, scheduler, coverage thresholds (individual bad coverage settings per QCI, for example, QCI1 = −80dBm, QCI9 = −85dBm), RLF settings per QCI, DRX settings per QCI, redirected carrier based on QCI, and handover trigger based on QCI, and so on. Operator also marks DiffServ DSCP (diffserv code point) properly based in the QCI value of the used bearer, which is related to transport QoS. For VoLTE, SIP control signaling and RTP audio packet is assigned with the higher priority of LTE QCI (5 and 1) and DSCP marking over transport network. Transport network QoS is provided by mapping QCI to DSCP for the uplink over S1 and in the downlink for packet forwarding over X2. eNB marks DSCP priority on S1‐U, S1‐MME and X2 interfaces by mapping from LTE QCI, SGW, and PGW marks DSCP priority on S5/S8 and SGi interface by mapping from LTE QCI.
443
444
LTE Optimization Engineering Handbook
Example: QCI 5 and 1 to DSCP mapping SIP
Speech
11.2.2 Voice Codec
For VoLTE UE to VoLTE UE call, voice codec is negotiated by the UEs during the SIP session opening. They have to select a common type. Possible codecs for voice include EVCR‐A, EVCR‐B, AMR, AMR‐WB5 (AMR‐wideband) and even EVS.6 For instance, AMR 12.2, which RTP payload size is 32 bytes, AMR‐WB 12.65 RTP payload size is 33 bytes, AMR WB 23.85 RTP payload size is 60 bytes, and SID RTP payload size is 7 bytes. The UE must be able to operate with any subset of the 8 modes for AMR and any subset of the 9 modes for AMR‐WB. Tandem free operation (TFO) and transcoder free operation (TrFO) must be supported in the IMS core network for CS interworking. VoLTE performance specifications, including video, is listed below. Video quality in terms of the objective estimate is from PEVQ (ITU‐T J.247). Rule of thumb and need to optimize and fine‐tune per operator basis is relevant as a service KPI requirement (Figure 11.6): ●●
●●
●●
Speech latency (end‐to‐end delay) requirement is from standard ITU‐T G.114 and 3GPP TS 22.105, no more than 150 ms is preferred, at maximum 400 ms. Packet loss, E2E conversational voice packet loss rate is defined as 1% in TS23.203 for VoLTE audio packet transported between UE and PGW, at maximum 3% FER), video telephone at maximum 1% FER, and data service requires 0% FER. Radio performance is linked to speech quality (MoS) through: M-to-E
300 ms
Users very satisfied
ec
d Co
225 ms
64
ps
.2 12
Users satisfied
ps
kb
Some users dis-satisfied
kb
FER 1%
Many users dis-satisfied
5%
Figure 11.6 Radio performance is linked to codec, FER, and mouth‐to‐ear delay. 5 AMR‐WB specification: 3GPP TS 26.171 and GSMA IR.36. 6 EVS, codec for enhanced voice services, the EVS standard is the first 3GPP codec to deliver speech and audio in super‐wideband full‐HD voice quality, bringing mobile audio on par with the audio experienced through today‘s digital media services. The algorithm delivers speech and audio up to 20 kHz audio bandwidth, outperforming the audio quality of today’s mobile phone calls.
VoLTE Optimization
●●
●● ●● ●●
●●
Codec, AMR 12.2 kbps and above provides good speech quality, Frame erasure rate (FER), less than 1% provides good speech quality, Mouth‐to‐ear delay, less than 225 ms provides good speech quality. The packet delay budget (PDB) requires no more than 80 ms for voice, and no more than 130 ms for video. PDB is measured from entering PDCP in eNB to leaving PDCP in the UE and vice versa, which includes 20 ms budget for interactions between PCRF, PGW, and eNB (20 ms (PCRF‐eNB) and 80 ms (air interface). The Number of HARQ retransmissions should be considered that cannot exceed PDB 80 ms in live network. Call setup time is 3 to 6 sec. Video path delay (camera‐to‐display delay) should less than 400 ms. Video frame loss: as well as video frame rate > = 25 Hz and video frame loss = 350 kbps
Video path delay (VPD) Residual packet loss ratio Session Setup time ROHC Scheduler TTI bundling
QCI1
QCI1
QCI2
Prio. > QCI2
QCI5 scheduled with non‐GBR, low delay, very low packet loss rate QoS
QCI1 scheduled with GBR/MBR DL/UL = 40 kbps, low delay, moderate packet loss rate, uses RLC‐UM and usually only 2 HARQ transmissions QCI2 scheduled with minimum rate proportional fair 400Kbps QCI 7/8/9 scheduled with minimum rate proportional fair 0 kbps
The default EPS bearer keeps the UE connected to the network. When there is no user traffic and thus the UE state changes to idle, E‐RAB is deactivated and only the S5 bearer stays on. However, as soon as new user traffic arrives, E‐RAB is re‐established, allowing the traffic to be delivered between the UE and the P‐GW. For uplink, different RB mapping of the logical channels to the UL logical channel groups, QoS for default bearer is provided by HSS, QoS for dedicated bearer is provided by PCRF, shown in Table 11.9. LTE supports up to two VoIP bearers per UE. For UE with two VoIP bearers established and semi‐persistent scheduling (SPS) method is configured, SPS transport block size (TBS) shall be the sum or the maximum of the payload size of the two VoIP bearers. The downlink semi‐ persistent scheduler shall multiplex MAC SDU of two VoIP bearers into one MAC PDU in each SPS period. In the uplink direction, eNB MAC layer shall be able to demultiplex two MAC SDUs from one MAC PDU and deliver each MAC SDU to corresponding RLC entity for each VoIP bearer. The following voice RB metrics are mainly used to measure the quality of service for VoIP. One is bandwidth, the other is E2E delay. The bandwidth depends on the type of voice codec. Generally, the required bandwidth for a single call, one direction, is 12.2 kbps for AMR‐NB codec or 23.85 kbps for AMR‐WB codec. Typically, these codec delivers one voice packet each 20 ms. For AMR‐NB, each packet is sent in one ethernet frame. With every packet of size 256 bits, headers of additional protocol layers are added. These headers include RTP + UDP + IP headers and PDCP/RLC/MAC headers with preamble of sizes 32 + 8 + 8 + 8, respectively. Therefore, a total of 312 bits needs to be transmitted 50 times per second in one direction. The guaranteed bit rate (GBR) for VoLTE is determined by the formula below: VAD BW
1 VAD
VoTE _ IP _ PayloadSize Overhead Voice _ SamplingRate Silence _ IP _ PayloadSize Overhead Silence _ SamplingRate
8bit / byte
453
454
LTE Optimization Engineering Handbook
Where: ●● ●● ●● ●● ●● ●● ●●
●●
VAD is the voice activity factor from 0% to 100% Voice_SamplingRate is the frame transmission frequency per standards: 20 ms Silence_SamplingRate is the silent indicator: 160 ms VoLTE_IP_PayloadSize (32 bytes) + Overhead (40 bytes) for AMR 12.2 kbps = 72 bytes Silence_IP_PayloadSize (5 Bytes) + Overhead (40 bytes) for AMR 12.2 kbps = 45 bytes Number of packets per second: 1000/20 = 50 The GBR for a VoLTE call with 100% activity factor using only AMR 12.2 kbps codec mode should consume a bandwidth of 28.8 kbps BW = [1 × (72)/20 ms + (1 – 1) × (45)/160 ms] × 8 bit/byte = 28.8 kbit/s
From Figure 11.12, we can see that the average rate of QCI 1 is 11.8 kbps, QCI 5 is 10.5 kbps, the defaulkt QCI 9 is only 0.04 kbps in a field test. End‐to‐end latency including bearer delay (RTP packets) and signaling delay (SIP signaling). Table 11.10 shows the typical voice delay budget. According to ITU G.114, less than 200 ms of mouth‐to‐ear audio packet delay is needed to make the user very satisfied and no more than 280 ms is needed to make the user satisfied. 11.2.5 RLC UM
An RLC entity can be configured to operate in TM, UM, or AM mode, which are shown in Table 11.11. There are RLC services and RLC functions. The RLC functions that are performed by the RLC entities are concatenation, padding, data transfer, error correction, in‐sequence delivery, duplicate detection, flow control, RLC re‐establishment, protocol error detection, and recovery. The RLC UM (unacknowledged) mode provides a unidirectional data transfer service without sending any feedback to the transmitting entity, therefore, there are no re‐transmissions of packets. UM RLC is mainly utilized by delay‐sensitive and error‐tolerant real‐time applications, especially VoIP, and other delayed sensitive streaming services. RLC UM mode can be configured per bearer type, eNB will choose RLC mode to use based on the bearer setup procedure and QoS requirements. RLC UM is useful for services that tolerate a higher packet loss rate but require lower latency, unlike RLC AM with a OTT ARQ reordering mechanism by RLC ARQ. RLC UM is “in‐ascending‐order delivery,” which unlike RLC AM with “in‐sequence order delivery,” same but gaps allowed (Figure 11.13). The RLC UM machine starts re‐ordering when detecting a gap, same as RLC AM. But when the timer expires, UM does not, like AM, do its own retransmission but rather trusts lower layers to have provided enough time for re‐ordering. RLC UM ignores any gaps that might still occur and continues to deliver in ascending order to higher layers. If any missing RLC SDUs arrive later they will be discarded since they arrive outside the re‐ordering window. The RLC mode is configured for each QCI by eNB. If the source eNB has a bearer with QCI configured with the RLC UM feature and the target eNB has mapped the same QCI to RLC AM, the bearer is rejected. Data forwarding at intra‐LTE (X2) handover for RLC (UM) Data forwarding is executed in user plane (UP) tunnels, established between the source eNB and the target eNB during the handover preparation. Data forwarding must be supported for X2 handover for both UM and AM bearers. Also X2 handovers including data forwarding will work for VoIP only and for the combination of VoIP and data simultaneously. There is one tunnel established for DL data forwarding per each E‐RAB for which data forwarding is applied (Figure 11.14).
35
30
25
20
15
10
5
0
100
200
300
PDCPThrput Qci1_UL(kbps)
400
500
600
PDCPThrput Qci5_UL(kbps)
700
0 1000
PDCPThrput Qci9_UL(kbps)
100
Bit rate (Kbit/s)
Number of bits [Kbit]
15
SS
RE
_
_IN
N SIO
ES
_S
3 _18
OG PR
SIP
10
50
0
20
_ SIP
ING ING _OK R _ 0 180 P_20 IP_ SI
K _O S
SIP message Average rate 5
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
Time
0
Figure 11.12 Data rate of RB in field test. Table 11.10 Voice delay budget. Delay component
Range(ms)
Comments
UE delay (UL/DL)
31‐44
Air interface one‐way delay(UL/DL)
4 ~ 53/1 ~ 37
For 6 HARQ transmission using dynamic scheduling as the worse case
eNB delay
2 ~ 4
eNB processing delay for packet L1/L2 processing
SGW
0.1 ~ 0.5
Packet forwarding
PGW
0.1 ~ 0.5
S1‐U
2 ~ 15
S5
2 ~ 15
IP network
18 ~ 41
Propagation delay is mainly proportional to distance(5us/km) Assuming 2 ms processing and queuing and 2000 miles OC3
456
LTE Optimization Engineering Handbook
Table 11.11 Three modes of RLC. Transparent Mode
Unacknowledged Mode
Acknowledged Mode
No segmentation and reassembly of RLC SDUs
Segmentation and reassembly of RLC SDUs
Segmentation and reassembly of RLC SDUs
No RLC headers are added
RLC headers are added
RLC headers are added
No delivery guarantees
No delivery guarantees
Reliable in sequence delivery service
Suitable for carrying voice
Suitable for carrying streaming traffic
Suitable for carrying TCP traffic
Transmitter Transmission buffer
New RLC AMD PDU-s
Receiver
Link
MUX
DEMUX
Insequence RLC AMD PDU-s
Re-ordering
“3” was lost
(MAC) Backpressure 1
2
3
4
1
2
4 Time
Figure 11.13 RLC UM insequence order delivery.
UE
Source eNB
Target eNB
MME
SGW
Handover request Handover Request ack RRC conn reconf RRC conn reconf Cmp
SN Status Transfer Path switch request
Modify bearer request
GTP-U Modify bearer response UE context release
Path switch request ack
Figure 11.14 Signaling flow (data forwarding at X2 handover for RLC_UM).
VoLTE Optimization
RLC UM will be applied to VoLTE (QCI 1 or QCI 2). The RLC in UM feature provides the means to deliver VoIP without delays due to retransmission and may be configured to reduce the control signaling overhead of RLC and PCPC layers. Another advantage is a faster delivery but with a higher risk of packet loss. The QCI = 2 characteristics defined in 23.203 (PELR ≤ 10−3, PDB ≤ 150 ms) can be realized with RLC UM, but also with RLC AM, as RLC AM minimizes packet loss at the expense of an increased delay. However, using RLC AM may give rise to delay spikes, which are caused by the RLC AM protocol while it waits for a retransmission. RLC AM will give rise to quite a few RLC status reports as the overhead is insignificant since video packets are relatively larger. The RLC status reports will just get a “free ride” with the data packets of the reverse direction of the bi‐ directional video flow. The operator can improve the network performance by optimizing RLC parameters, including the maximum number of ARQ retransmission, the length of poll retransmit timer, and especially SN lengths of RLC PDU and PDCP PDU, which are described below. It is recommended to configure the VOIP media bearer (RLC UM) to reduce the control signaling overhead of RLC and PDCP layers by shortening the default sequence number length from 12 bits to 5 bits for RLC and to 7 for PDCP, as specified in 3GPP TS36.322 and TS 36.323,16 respectively. Further information can be found in the reference RLC in unacknowledged mode. The relevant eNB parameters are: ●● ●●
rlcSNLength, 5 bits (with rlcMode = UM) pdcpSNLength, 7 bits (with rlcMode = UM)
For RLC UM bearers, both RLC and PDCP SN are reset at handover and eNB starts sending UM packets with PDCP SN starting from 0 in the target. Actually, improper configuration for pdcpSNLength and rlcSNLength would also cause negative effects. Setting them to abnormally small values will lead to wraparound on the sequence numbers and hence increase the drop rate if RLC/PDCP SN length for QCI 1 has a mismatch in source and target cell.17 If they are set differently in serving cell and target cell, it would result in handover failures or drops after handover. 11.2.6 Call Procedure
VoLTE services high level steps include detect available network, attach to the LTE network, setup IMS APN and find P‐CSCF, register in IMS, and place/receive a call. From the attach procedure, UE may include capabilities such as UE SRVCC capability in the “Attach request” message. In the “Attach accept” message as well as in the “Tracking area update accept” message, MME provides information to the UE on the network capabilities, for example, emergency service support indicator, location service support indicator, and IMS voice over PS session supported indicator. When the PDN connection for the IMS APN is established in the setup IMS APN and find P‐CSCF procedure, the list of P‐CSCF addresses are provided by the PGW to the UE in the “PDN connectivity accept” message and the IMS APN default bearer QCI 5 is established for the IMS signaling. After register in IMS, MO/MT call triggers UE to do service request to move UE to connected mode and then it can send and receive user plane traffic (SIP invite) (Figure 11.15). 16 3GPP TS36.322: Radio link control (RLC) protocol specification. 3GPP TS36.323: Packet data convergence protocol (PDCP) specification 17 In a live network, In some cases it was observed that PDCP SN length was same in both the source and target cell but the HO failed.
457
458
LTE Optimization Engineering Handbook MME
eNB
S/PGW
HSS
PCRF
CSCF
AS
Radio setup Registration
EPS Registration, Default bearer setup and VoLTE support discovery IMS registration and user authentication IMS VolP session setup
Call establishment
EPS dedicated bearer setup Ongoing voice call
Call release
IMS VolP session and dedicated bearer release
Figure 11.15 High level e2e VoLTE call flow.
11.2.6.1 LTE Attach and IMS Register
As part of attach procedures, VoLTE‐related capabilities and relevant network information (voice and emergency support, and P‐CSCF address) are provided between UE and network. If the IMS APN is not default APN, the UE will initiate a separate PDN connection after attach to establish the IMS PDN. In the attach request message the UE signals if it supports SRVCC to GERAN/UTRAN, or if the UE is performing a combined attach procedure. If it supports SRVCC, it indicates its supported speech codecs for CS speech calls. In the attach accept message the UE is informed whether the network supports IMS VoIP, emergency calls, and EPS/CS location services (Figure 11.16). The UE need to register (authenticate) toward the IMS before any services can be utilized. The registration may take place any time after the UE has attached to LTE and acquired an IP address to P‐CSCF. Once the LTE attach is performed, the terminal start registration at the IMS (S‐CSCF) to be able to receive IMS services. Prior to registering with IMS the UE must establish another PDN connection to the IMS APN by a SIP registration request. VoLTE mandates the use of IMS AKA (i.e., UICC based authentication based on same authentication mechanism as for the access). As a consequence of the authentication, IPsec is established for SIP signaling security between UE and P‐CSCF (Figure 11.17). UE needs to perform re‐registration procedure before registration timer expires. For VoLTE, the application servers also need to know about registration status through a third‐party register. The registration may take place any time after the UE has attached to LTE and acquired an IP address to P‐CSCF. If the user profile includes a service trigger for register, the S‐CSCF sends a third‐party register message to the application server indicated in the trigger, to inform the application server about the change in registration status. The IMS registration latency is: IMS registration latency [ms] = T200 OK − TREGISTER where TREGISTER is the timestamp for SIP register (step 3) being sent from the UE (start trigger for the registration procedure), T200 OK is the timestamp for 200 OK (Figure 11.18). 11.2.6.2 E2E IMS Flow
After register to the IMS domain with a SIP register message the UE can initiate a VoLTE session, using SIP invite. The SIP invite is used to find the called party, and negotiate the media to be used. Figure 11.19 shows that the originating (MO) call triggers UE to do a service request
VoLTE Optimization UE Attach
MME
eNB NAS:Attach Request
HSS
EPG
SAPC
P-CSCF
Diameter:AIR Diameter:AIA
NAS:Authentication Request NAS:Authentication Response NAS:Security Mode Command NAS:Security Mode Complete
GTPv2:Create Session Request
Diameter:CCR
EBI:5, QCI:9(Default Bearer) GTPv2:Create Session Response
Diameter:CCA
S1AP:Initial Context Setup EBI:5 , QCI:9(Default Bearer) EBI:5, QCI:9(Default Bearer) NAS:Attach Accept, activate default EPS bearer request GTPv2:Create Bearer Request EBI:5, QCI:9(Default Bearer) Linked EBI:5, QCI:5(IMS SIP Signaling) S1AP:UE capability info indication S1AP:Initial Context Setup Response NAS:Attach Complete, activate default EPS bearer accept GTPv2:Modify Bearer Request GTPv2:Modify Bearer Response Default Bearer(EBI:5, QCI:9) is established S1AP:E-RAB Setup Request EBI:6, QCI:5(IMS SIP Signaling) NAS:activate dedicated EPS bearer context request EBI:6, QCI:5(IMS SIP Signaling) S1AP:E-RAB Setup Response NAS:activate dedicated EPS bearer context accept GTPv2:Create Bearer Response
Dedicated Bearer(EBI:6, QCI:5) is established
EBI:6
IMS SIP REGISTER Procedure via Dedicated Bearer for QCI:5
Figure 11.16 Attach procedure for VoLTE user.
LTE Attach
HSS
IMS registration (SIP)
(3) Location registration (Download APN for VoLTE) (1) Power-on
LTE/EPC
MME/SGW
(7) Completion of Attach VoLTE terminal
(5) Terminal IP address and P-CSCF allocation IMS
(4) Setting up of bearer for VoLTE
(2) Attach request eNodeB
(9) a. Registration and authentication (9) c. Service information download
(P-CSCF address)
PGW (6) Bearer response (P-CSCF address)
(8) IMS registration request (9)b. Authentication (10) Completion of IMS registration
Figure 11.17 LTE attach and IMS register.
SIP bearer
P-CSCF
S-CSCF
AS
459
460
LTE Optimization Engineering Handbook UE
Step 1 and 2 is the LTE attach procedure and UE acquires IP address from PDN gateway.
LTE RAN
UE power on
EPC
IMS/MTAS
1: Attach procedure, P-CSCF discovery
2: UE requested PDN connectivity
401 unauthorized response occurs when UE has not yet been assigned a serving CSCF.
3: REGISTER 401 Unauthorized 4: REGISTER 200 OK
UE ready to use
5: SUBSCRIBE
Step 5 and 6 are optional IMS services for subscribing to registration state.
200 OK 6: NOTIFY 200 OK
Figure 11.18 IMS registration latency. UE
eNB Service Request
MME
SGW
PGW
HSS
Service Request Authentication (if required)
Authentication (if required)
Initial Context Setup
Radio Bearer Establishment
Initial Context Setup Complete Modify Bearer Request Modify Bearer Response
PGW SIP INVITE FROM IMS
PCRF
DOWNLINK DATA [SIP INVITE]
Modify Bearer Request
IP-CAN Session Modification
Modify Bearer Response
SGW
MME
eNB
UE
Downlink data Notification DDN ACK Paging Paging
Service Request Procedures Stop Paging DOWNLINK DATA [SIP INVITE]
Figure 11.19 MO and MT call service request.
VoLTE Optimization
UE
eNB
User initate a call Reestablishment of bearers if UE is in idle mode
EPC
IMS
Service Request Procedure SIP INVITE
SIP 100 Trying
SDP answer with resource reservation. If needed, establish resources for early media (ring tones or announcements)
SIP183 Session Progress Dedicated EPS Bearer Establishment
Rx/Gx: MMTel Session Est.
SIP PRACK
Only DL media enabled (for early media)
SIP 200 OK
Confirmation of resources available
SIP UPDATE SIP 200 OK
If early media used, ring tone is from network, otherwise generated locally in UE Call picked up and media flowing in both directions.
SIP 180 Ringing SIP 200 OK Update of flow status of affected PCC rules (if required)
Rx/Gx: Flow/gate update
PCC update to open gates for bidirectional media
SIP ACK
Figure 11.20 Originating call flow. IMS
EPC
eNB
SIP INVITE Paging Service Request Procedure SIP INVITE SIP 100 Trying
SDP answer with resource reservation
Wake up UE if in idle mode Reestablishment of bearers if UE is in idle mode
SIP 183 Session Progress Rx/Gx: MMTel Session Est.
Dedicated EPS Bearer Establishment SIP PRACK SIP 200 OK
Confirmation of resources available PCC update to open gates for bi-directional media
UE
SIP UPDATE SIP 200 OK
Indication of incoming call
SIP 180 Ringing SIP 200 OK Rx/Gx: Flow update
Update of flow status of affected PCC rules SIP ACK
User answers the call
Figure 11.21 Terminating call flow.
and move the UE to connected mode and then it can send and receive SIP invite. For terminated (MT) call, the incoming SIP invite results in that a UE in idle mode is paged and as response the UE initiates service request to establish user plane connectivity and the SIP invite can be forwarded to the UE. SIP session establishment start from UE transmits SIP invite to UE received SIP 200 OK message, including dedicated radio bearer setup and session establishment procedure, and so on. The E2E IMS signaling flow is described below and shown in Figure 11.20 and Figure 11.21: ●●
●●
100 Trying: IMS core sends 100 trying message to originating UE once the message is forwarded to the destination. 183 session progress: this message is basically indication that the call is being processed. It’s typically used either to prevent the call from being timed out, as we await the destination user answering the call. It can also be used to support what’s called early media, that is, the
461
462
LTE Optimization Engineering Handbook
●●
●●
●● ●●
ability to send media (in‐band ringing or network announcements) to the calling party before the call is completed. Destination UE indicates codec choice in SIP183 session progress. SIP183 message also triggers network‐initiated dedicated EPS bearer setup at source and destination UE. PRACK (provisional response acknowledgement): PRACK improves network reliability by adding an acknowledgment system to the provisional responses (1xx). PRACK is sent in response to provisional response. Source UE sends a SIP PRACK message indicating codec selection and preconditions extension, “Precondition” increases establishment time by 0.5 s in RRC connected state but reduces probability of call establishment failure in live network. Destination UE responds with a SIP 200 OK acknowledging SIP PRACK. Network‐initiated EPS bearer setup and transfer of SIP PRACK and SIP200 messages occurs concurrently. 180 Ringing: Destination UE locally alerts the user and sends across a ring back tone to the source UE through SIP180 ringing. When the user responds to the alert, destination UE sends SIP200 OK response to original SIP Invite. 200 OK: The request was successful. Source UE responds with an SIP ACK message. Media streams are established.
An example of originating call and terminating call flow in alive network is shown in Figure 11.22. Source UE IMS client forms a SIP invite message, which includes QoS preconditions18 and order of preference for audio codecs. SIP invite message transmitted as part of IPSec packet to EUTRAN. For QoS preconditions, resources reserved before users receives ring tone, when user answers, the users are guaranteed to be able to talk. However, the reservation of the resources prior alerting the user can increase the call setup time. Two different set of traffic flow templates (TFTs) for audio and video is provided to UE when both reservations are successful. For without QoS preconditions, UE are alerted and can answer the call prior resource reservation, there is no guarantee that the user will be able to get any resources for the call (Figure 11.23). When UE A and UE B are belong to different home network, the E2E call procedure is shown in Figure 11.24. Figure 11.25 shows the SIP invite message details. It is identified by a call identifier, local tag, and a remote tag. A call‐ID contains a globally unique identifier for this call, host name, or IP address. The combination of the To tag, From tag, and call‐ID completely defines a peer‐to‐ peer SIP relationship between two users and is referred to as a dialog. Figure 11.23 shows a SDP message carried as message body in the SIP message, describing the session parameters. Default EPS bearer is used to send the initial SIP invite that contains the SDP information. SDP carries the requested media, bandwidth, and source transport address (audio). 11.2.6.3 Video Phone Session Handling
A VoLTE and video‐calling device performs the same network attach, IMS domain authentication and registration procedures as a VoLTE device, plus adding video capability information. At video call setup the client signals in the SDP included in the SIP invite message that two media streams shall be setup by IMS in the EPS. These two media streams shall be full
18 UE supports SIP preconditions per GSMA IR.92 recommendations. The recommendation is to support preconditions for the reason to avoid “ghost ringing” in case resource allocation fails after 180 rings. QoS precondition also prevents that voice/video traffic by mistake is sent on the IMS signaling bearer (which could cause disturbance on other traffic).
MO Call Orig
NW/IMS
MT
RACH + RRC Conn. Setup SIP: Invite (SDP Offer)
Paging + RRC Conn. Setup
SIP: 100 Trying SIP: Invite (SDP Offer) SIP: 100 Trying SIP: 183 Session Progress RRC Reconfig
RRC Reconfig
SIP: 183 Session Progress
Inform User
SIP: PRACK
SIP: PRACK
SIP: 200 OK (PRACK)
SIP: 200 OK (PRACK)
SIP: UPDATE
SIP: UPDATE
SIP: 200 OK (UPDATE)
SIP: 200 OK (UPDATE) SIP: 180 Ringing
SIP: 180 Ringing
SIP: 200 OK (SDP Answer)
SIP: 200 OK (SDP Answer) SIP: ACK
SIP: ACK E2E RTP Audio
Figure 11.22 Call procedure with “precondition.”
Alert User User Answer
UE A
eNB
EPC
IMS
QCI = 5 Call initiation
Resource reservation Note: Time between resource reservation until QCI = 1 bearer is up is undefined.
Ringing Answer
RTP
RTP
QCI = 1 UE A
eNB
EPC
IMS
Call initiation Ringing
QCI = 5
Answer (SDP)
RTP Resource reservation Note: Time between resource reservation until QCI = 1 bearer is up is undefined.
QCI = 1 RTP
RTP
Figure 11.23 With QoS preconditions and without QoS preconditions. UEA ’s home network UEA
P-CSCF INVITE
INVITE
UEB ’s home network
S-CSCF
100 Trying
AS
I-CSCF
HSS
S-CSCF
AS
P-CSCF
UEB
INVITE
100 Trying
INVITE 100 Trying
LIR LIA INVITE
INVITE 100 Trying
INVITE
INVITE 100 Trying
183
183 183 183
183 PRACK
183
183
183
PRACK
PRACK PRACK
PRACK
PRACK
PRACK 200 (OK) 200 (OK) 200 (OK)
200 (OK)
200 (OK)
200 (OK)
Dedicated bearer (QC11) established
200 (OK)
Dedicated bearer (QC11) established UPDATE
UPDATE
UPDATE UPDATE
UPDATE
UPDATE
UPDATE 200 (OK)
200 (OK) 200 (OK)
200 (OK) 200 (OK)
200 (OK)
180 (Ringing)
200 (OK) 180 (Ringing)
180 (Ringing)
180 (Ringing)
200 (OK) ACK
180 (Ringing)
200 (OK)
180 (Ringing) 200 (OK)
180 (Ringing)
200 (OK)
200 (OK)
200 (OK)
User Answers
200 (OK)
200 (OK) ACK
ACK ACK
ACK ACK
Figure 11.24 Call procedure when UE A and UE B are belong to different home network.
Ringing
180 (Ringing)
ACK
VoLTE Optimization
Media negotiation UE IP address IP address used for media stream Time session created and how long intended to last Payload type: 104 Codec: AMR-WB SR: 16000 Hz Codec mode: 0, 1, 2 Mode-change-cap: capability to restrict the mode change Max-red: elapses between the first transmission of a frame and any redundant transmission
Required bandwidth for RTP traffic is 30 kbps No required bandwidth for RTCP traffic (RS=0, RR=0)
WB DTMF NB DTMF
Precondition: resource reservation maxptime: maximum limit of 12 speech frames (240ms) per RTP packet ptime: one speech frame (20ms) encapsulated in each RTP packet
Figure 11.25 Session description protocol.
UE
PGW
PCRF
IMS
Originating network
Terminating network
IMS
PCRF
PGW
UE
Ongoing active voice call
User adds video
SIP Re-INVITE (voice/video)
SIP Re-INVITE (voice/video)
SIP Re-INVITE (voice/video) SIP 200 OK P-CSCF updates with new video
SIP 200 OK
AAR AAA
P-CSCF updates with new video
IMS interacts with PCC to create a new video bearer for the ongoing call.
Dedicated EPS Bearer establishment (video component)
AAR AAA Dedicated EPS Bearer establishment (video component)
SIP 200 OK Ongoing active call with voice and video
Figure 11.26 Adding video to ongoing voice call.
duplex and synchronized for lip sync. The P‐CSCF, based on the negotiated SDP, creates one Rx session toward the EPC per call leg and uses the Rx interface toward the PCRF to request two dedicated EPS bearer one for voice and one for video on top of the already existing default EPS bearer used for SIP. The EPS is responsible to set up the two dedicated EPS bearers using the network‐initiated bearer setup procedures in sequence. Adding video to ongoing voice call procedure is described in Figure 11.26.
465
466
LTE Optimization Engineering Handbook
11.2.7 Multiple Bearers Setup and Release
LTE permits UE to establish up to eight simultaneous data radio bearers (DRB) and each radio bearer can have a unique QoS profile. Prior to introducing VoLTE, a LTE user typically set up one default bearer for data on QCI {7,8,9}. A default bearer is bearer able to carry all kinds of traffic (no filter) without QoS and it is first established for a new PDN connection and remains established throughout the lifetime of the PDN connection. It is typically created during the attach procedure. The UE will have one IP address used for all data services. All the traffic for the user is sent across one DRB and there is no QoS‐based traffic separation. VoLTE introduces a second default bearer. A VoLTE user in the eNB will always have at least two default bearers configured—one for IMS signaling, one for internet connection. Voice/video traffic shall be separated from data traffic for the UE therefore dedicated bearers will be set up on demand for the voice and video components to enable this traffic separation. QCI1 and QCI2 are dedicated bearers associated to the IMS APN default bearer (QCI5). Dedicated bearers apply packet filters or traffic flow templates (TFTs) on top of the default bearer. The TFTs are installed in the UE to make sure that media traffic is forwarded on the respective dedicated bearers. TFTs consist of IP address, port and protocols to specify which traffic shall run upon the dedicated bearer. Two separate steps are considered before the UE is successfully registered in both LTE and the IMS domain. The first part will be the combined PS and CS (for SMS over sGs) attach procedure including the activation of the default internet APN. The second part is the setup of the IMS default bearer and the IMS registration. After completing these two parts successfully the UE is registered to the network and IMS domain. If both the UE and the network are VoLTE capable, the user is now able to start a VoLTE service. Once the radio bearers are established for the UE (now in RRC_connected mode) at the eNB for the default IMS APN (QCI 5) and default internet APN (QCI 7–9) the UE will only change back to RRC_Idle mode in case there is no activity more for the duration of tInactivityTimer on both default non‐guaranteed bit rate bearers. For VoLTE, voice media should always be mapped to a separate EPS bearer as it has GBR running the RLC unacknowledged mode protocol. Voice service can tolerate error rates on the order of 1%, while benefiting from reduced delays. SIP signaling should also be protected against congestion by being mapped to a separate EPS bearer running RLC acknowledged mode protocol. So, the biggest challenge in the LTE RAN within VoLTE is to handle the relation between signaling (SIP) and media (RTP), the SIP signaling is carried on the QCI 5 bearer, which has highest priority to accommodate, for example, fast call setup. For example, in a live network, 2.6 billion eRAB setups out of which 16 million are VoLTE, that is, only 0.6% of all eRAB attempts are VoLTE (QCI1). In conclusion, an IMS UE will have two bearers allocated all the time, the default bearer to internet APN is QCI = 9 (non‐GBR RLC‐AM), the default bearer to IMS APN is QCI = 5 (non‐ GBR RLC‐AM). During a VoLTE call a third bearer will be activated (QCI = 1), voice bearer QCI1 would be configured during voice call initiation (SIP invite). QCI1 is a dedicated bearer to the IMS APN (GBR RLC‐UM), dedicated bearers are created for QoS differentiation purposes (Figure 11.27). When the UE changes state from EPS connection management (ECM)‐connected to ECM‐ idle all radio resources are released but information about the PDN connection remains stored in the packet core. All radio resources are restored again when a network or UE‐initiated service request is received. QCI 8(9), default bearer setup is always the first step for any service when UE attaches to EPC and remains active as long as UE is attached to EPC (always on). IMS default bearer (QCI5), the
VoLTE Optimization MO
NW/IMS
MT
RACH + RRC Conn. Setup (DRB3 + DRB4) SIP: Invite Paging + RRC Conn. Setup (DRB3 + DRB4)
QCI 9 and QCI 5 are established during the attach procedure
SIP: 100 Trying SIP: Invite SIP: 100 Trying SIP: 180 Ringing SIP: 180 Ringing SIP: 200 OK RRC Reconfig (DRB5)
RRC Reconfig (DRB5) SIP: 200 OK SIP: ACK
EPS DRB RLC ID ID 5 3 AM 6 4 AM 7 5 UM
SIP: ACK
QCI
Use
9 5 1
BE SIP RTP
E2E RTP Audio
Connection Established
Default bearer established (QCI9 & QCI5)
Call established (SIP)
Dedicated bearer established (QCI1)
time RRC_IDLE
RRC_CONNECTED
Figure 11.27 Multiple bearers setup.
first dedicated radio bearer, is set up for SIP control signaling right after the default bearer setup during the initial attach. QCI5 will be established with the second PDN connection with IMS APN. This will avoid SIP control signaling setup delay when VoLTE call is invoked. QCI 1, MO UE requests a voice call through SIP invite. MT UE CQI1 bearer is only established after MT UE responds with SIP OK to original SIP invite. The above QCI profiles re shown in Table 11.12. Once a VoLTE has been established, the user might want to end the VoLTE call (video phone). It is expected that the user will have two separate buttons on its terminal in order to end the video only or to directly end both voice and video call according to IR.94 from GSM association. Figure 11.28 gives an illustration of VoLTE call end and bearer release procedure. After VoLTE call end, the UE still has both default bearers active, which might be used for additional SIP signaling or uplink and downlink data transmission. In the case the UE (still in RRC_connected mode) does not send or receive anything any more on both the default bearer’s timer, tInactivityTimer, is started, which is currently set to 60s in more live network. Once the timer, tInactivityTimer, expires the UE will move from RRC_connected mode to RRC_idle mode and the radio bearers are released for the non‐GBR default bearers. 11.2.8 VoLTE Call On‐Hold/Call Waiting
For VoLTE call on‐hold/call waiting, it has been decided to modify the bandwidth only in the “call waiting” call case since it will need to double the bandwidth. Only at release of the second call will trigger another bearer modification. For the “call on‐hold” call case, no bearer
467
468
LTE Optimization Engineering Handbook
Table 11.12 QCI configuration. QCI 1
QCI 2
QCI 5
QCI 8/9
Priority
1
4
2
8/9
Scheduling strategy
Delay based
Proportional fair with Min rate
Resource fair
Resource fair or proportional fair
Logical channel group
1
2
1
3
DRX Profile
1
2
0
0
DRX Priority
99
100
1
1
RLC mode
UM
UM
AM
AM
AQM mode
GBR (2)
GBR (2)
OFF (0)
Non‐GBR (1)
pdb
80
150
100
300
pdbOffset
50
50
0
0
pdcpSNLength
7
12
12
12
rlcSNLength
5
10
10
7
ROHC
TRUE
FALSE
FALSE
FALSE
Switch from RRC_Connected to RRC_Idle state
QCI = 2, Video QCI = 1, Speech
QCI = 1, Speech
QCI = 8, internet
QCI = 8, internet
QCI = 8, internet
QCI = 5, IMS signaling
QCI = 5, IMS signaling
QCI = 5, IMS signaling
VoLTE + ViLTE DRBs IMS + internet default bearers
ViLTE DRB released
VoLTE DRB User inactivity tInactivity Timer expires released on default bearers on default bearers
Figure 11.28 VoLTE call end and bearer release.
modification is triggered since the bandwidth will not be set to “0” but only the packet filters will be adapted, which will not trigger an E‐RAB modification (Figure 11.29). 11.2.9 Differentiated Paging Priority
The problem occurs if an operator would need to use a more aggressive paging for VoLTE calls than for VoLTE SMS. Both services are handled by the same IMS APN. If the IMS APN is used for additional IMS services than VoLTE, such as RCS, then additional non–call‐related signaling can be expected on the IMS APN. The same UE will need different paging profiles depending on the service. If the SGW receives a downlink packet while the UE is IDLE, a data downlink notification is sent to MME requesting paging of the UE. It is possible to configure number of tries for the last visited eNB, TA, and TAI list. Profile 1: MME sends paging messages to all eNB in the TAI list held by MME Profile 2: MME starts by paging eNBs in last visited tracking area, if no success the TAI list is paged Profile 3: MME starts by paging last visited eNB, if no success the MME pages eNBs in last visited TA, if still no success all eNBs in the TAI list are paged.
VoLTE Optimization
eNodeB
UE
MME
1) E-RAB Modify Request List of RABs and corresponding QoS profile to be modified as well as an optional NAS message per RAB Admission Control (if features are enabled) 1) Admission GRANTED 2) Admission BLOCKED conditional *) E-RAB Modify Response Inside this message “E-RAB Failed to Modify List” Cause Value: Radio resources not available 2) RRC Reconfiguration Message Includes the optional NAS message “Radio Modify Setup” 3) RRC Reconfiguration Complete Message 4) E-RAB Modify Response Contains a list of all successfully modified RABs and possibly a list of all RABs that failed to be modified
Figure 11.29 Call flow for the bearer modification due to “call waiting.”
The timer for paging messages is configurable in 2 to 15 s. The default value is 3 seconds. For Profile 1, successful paging will take less than 3 s. The number of messages needed is equal to the number of eNBs in the TAI list. For Profile 3, the time for paging a UE that has not moved is less than 3 s. The number of messages needed is ~1. If the UE has moved since last connected, the time for successful paging of the three profiles is less than 3 s, less than 6 s, and less than 9 s depending on how many paging messages are needed. So differentiated paging using separate bearers or APN for VoLTE calls and other IMS services, this way paging profile 1 could be used for VoLTE calls and paging profile 3 be used for other IMS services. In this case that MME will need additional information for deciding which paging profile to be used. One proposal is to provide a service class IE in the data downlink notification message so that MME can decide which paging profile to use. When SGW receives a DL packet and the UE is idle, the DL packet needs to be marked with the service class to be
469
470
LTE Optimization Engineering Handbook
used for paging differentiation. SGW will use this information when sending the data downlink notification. A proposal is to have the P‐CSCF mark service class using the DSCP field (IPv4 and IPv6) or alternatively the IPv6 flow label field (not present in IPv4 packets). 11.2.10 Robust Header Compression 11.2.10.1 RoHC Feature
IPv4 and IPv6 protocol headers are large relative to voice payload. These headers are transmitted frequently for every voice packet, which has negative performance impacts on over‐the‐air capacity and cell‐edge performance. Header compression is necessary in order to provide voice services on the packet switched (PS) domain with similar packing efficiency associated with the circuit switched (CS) domain. For VoLTE, the IP/UDP/RTP header added to a voice packet for VoIP services is significant (e.g., AMR 12.2 kbps with 32 bytes payload is encapsulated by 40/60 bytes of overhead for IPv4/IPv6). During talk spurt period, ROHC (IETF RFC3095, robust header compression) is expected to compress IP/UDP/RTP header to about 4/6 bytes in average for IPv4/IPv6. Figure 11.30 shows the VoIP frame size of MAC transport block when RoHC is used. For all these codecs, the RTP header part (IP + UDP + RTP) is 40 bytes in IPv4 and 60 bytes in IPv6, and the compressed packet length is between 1 and 42 for IPv4, 1 and 62 for IPv6, which obtaining a reduction up to 50% and more in the best case that shown in Table 11.13. Dimensioning aspect by examples below are described. Example 1: AMR‐WB 12.65 RTP payload is 33bytes, Step 1: [33(AMR‐WB 12.65) +3(RoHC) + 7(PDCP, RLC, MAC, BSR, PHR header)]*8 = 344 bits. Step 2: The desired transport block size (TBS) of 344 is compared with 3GPP TS 36.213 TBS table. The result is 344. The total number of information bits including CRC = 344 + 24 = 368 bits, the effective Layer 1 bit rate: 368/20 = 18.4 kbps. Example 2: AMR 12.2 RTP payload is 32bytes, the effective layer 1 bits with and without RoHC are shown in Table 11.14. Basically the UE could support up to nine different RoHC profiles.19 If it would support any RoHC profile it will always need to support also the uncompressed mode variant. At minimum, UE and network must support “RTP/UDP/IP” profile (0x0001) to compress RTP packets and “UDP/IP” profile (0x0002) to compress RTCP packets, the format of the header of the RoHC packet is different from profile 0x0001. The UE and network must support these profiles for both IPv4 and IPv6. Below an example of a UE, which reports in the RRC UE EUTRA capability message that RoHC profiles it supports. Multiple “packet streams” will be created per bearer that can use different profiles and different modes of operation, each packet stream is identified by context ID (CID). RoHC profiles can be got from S1AP initial context setup message, which is shown in Figure 11.31. 11.2.10.2 Gain by RoHC
IETF RoHC is the only protocol selected by 3GPP to support IP packets header compression. The RoHC algorithm establishes a common context at the compressor and decompressor by transmitting full header and then gradually transition to higher level of compression. In RoHC layer, the compressed IP packet streams flow from compressor to the decompressor inside a RoHC channel, which multiplexes different IP packet streams that are compressed differently using different profiles. In the opposite direction, the RoHC channel carries the compressed IP 19
GSMA PRD IR.92 “IMS Profile for Voice and SMS” 4.0, March 2011.
VoLTE Optimization VoIP frame
Application Layer RTP H
VoIP frame
UDP H
RTP H
VoIP frame
UDP H
RTP H
VoIP frame
RTP Layer
UDP Layer IP H
IP Layer
40 bytes (IPv4) or 60 bytes (IPv6) IP H
PDCP Layer
UDP H PDCP H
RTP H RoHC H
VoIP frame VoIP frame
4 bytes RLC & MAC Layers
MAC H
RLC H
PDCP H
RoHC H
VoIP frame
Protocol
Payload
AMR wide-band 23.85 kbps
61 octets
AMR wide-band 12.65 kbps
33 octets
AMR wide-band 8.85 kbps
24 octets
AMR wide-band 6.6 kbps
18 octets
AMR 12.2 kbps
32 octets
AMR 7.95 kbps
22 octets
AMR 5.9 kbps
16 octets
AMR 4.75 kbps
14 octets
EVRC 8.8 kbps
22 octets
SID1
7 octets
Protocol
Header
PDCP
1 octet
RLC UM header
1 octet
MAC
1 octet (basic header)
BSR
2 octets (header in MAC if present)
PHR
2 octets
RTP/UDP/IP header
1–60 octets
Figure 11.30 VoIP frame size with RoHC.
471
472
LTE Optimization Engineering Handbook
Table 11.13 Robust header compression gains.
Protocol
Total hdr size (bytes)
Min. compressed hdr size (bytes)
Compress gain (%)
IP4/TCP
40
4
90
IP4/UDP
28
1
96.4
IP4/UDP/RTP
40
1
97.5
IP6/TCP
60
4
93.3
IP6/UDP
48
3
93.75
IP6/UDP/RTP
60
3
95
Table 11.14 The effective layer 1 bits of AMR 12.2 payload. AMR NB 12.2 w/RoHC
AMR NB 12.2 w/o RoHC
bits
256
256
RTP/UDP/IP header
bits
32
320
PDCP header
bits
8
16
RLC header
bits
8
16
MAC header
bits
8
8
Total size
bits
312
616
PHY TBS
bits
328
616
Protocol overhead
Source payload
packet streams, as well as the possible RoHC feedback packets for the associated RoHC channel. Figure 11.32 shows the RoHC channels and contexts. Each context corresponds to an individual compression profile, which has different compression algorithms. A context is identified with a context ID (CID). At RoHC layer, the compressed IP packet streams flow from compressor to the decompressor inside a RoHC channel, which multiplexes different IP packet streams that are compressed differently using different profiles. In the opposite direction, the RoHC channel carries the compressed IP packet streams, as well as the possible RoHC feedback packets for the associated RoHC channel. ROHC has three modes of operation, unidirectional (U‐mode), bidirectional‐optimistic (O‐mode), and bidirectional‐reliable (R‐mode). U‐Mode packets are only sent in one direction, from compressor to decompressor, but will not be used as the preferred mode of operation for VoLTE as it is mainly for unidirectional traffic. This mode, therefore, makes RoHC usable over links where a return path from decompressor to compressor is unavailable or undesirable. O‐ Mode (bidirectional optimistic) is similar to the U‐mode, except that a feedback channel is used to send error recovery requests and (optionally) acknowledgments of significant context updates from the decompressor to compressor. The O‐mode aims to maximize compression efficiency and sparse usage of the feedback channel. R‐Mode (bidirectional reliable) differs in many ways from the previous two. The most important differences are a more intensive usage of the feedback channel and a stricter logic at both the compressor and the decompressor that
• 0x0000 ROHC uncompressed (RFC 4995) • 0x0001 ROHC RTP (RFC 3095, RFC 4815), this profile compresses RTP headers efficiently. Such headers are common in a VoIP call or in a video stream; • 0x0002 ROHC UDP (RFC 3095, RFC 4815), this profile will be used to compress the control signaling of the VoIP call, i.e. the RTCP packets • 0x0003 ROHC ESP (RFC 3095, RFC 4815) • 0x0004 ROHC IP (RFC 3843, RFC 4815) • 0x0006 ROHC TCP (RFC 4996) • 0x0101 ROHCv2 RTP (RFC 5225) • 0x0102 ROHCv2 UDP (RFC 5225) • 0x0103 ROHCv2 ESP (RFC 5225) • 0x0104 ROHCv2 IP (RFC 5225) QCI 1 bearer Pkt stream 1 (RTP/UDP/IP): CID #X, profile 0x0001 Pkt stream 2 (UDP/IP): CID #Y, profile 0x0002
Figure 11.31 Possible RoHC profiles.
LTE Optimization Engineering Handbook RoHC channel 1 (DTCH UE< = eNB) CID 0 - IP Stream CID 1 - RTP/UDP/IP Stream
Decompressor
Compressor Feedback Packets
CID 2 - TCP/IP Stream Feedback Packets
474
CID 3 - ESP/IP Stream
CID 0 - IP Stream CID 1 - RTP/UDP/IP Stream
Compressor
Decompressor
CID 2 - TCP/IP Stream CID 3 - ESP/IP Stream UE RoHC
eNB RoHC RoHC channel 2 (DTCH UE = > eNB)
IP
UDP
RTP
Data
IP
RTP
Data
In-order delivery and duplicate detection
Sequence Numbering Compress
UDP
ROHC Context
ROHC Context
Decompress
Ciphering
Deciphering
Add PDCP Header
Remove header
CH
Data
Figure 11.32 RoHC compression architecture.
prevents loss of context synchronization between compressor and decompressor except for very high residual bit error rates. The optimal RoHC operation mode depends on feedback abilities, error probabilities and distributions, effects of header size variation, and so on. RoHC always starts in U‐mode and transitions to another mode based on the decompressor feedback (Figure 11.33). RoHC can provide coverage and capacity utilisation improvements to the network. Figure 11.34 shows the RoHC header size distribution in a live network, 98% of RoHC headers size is smaller or equal to 40 bits. The most important factor for high VoIP capacity is RoHC functionality, especially in uplink. With RoHC functionality, coverage will be improved and inter‐site distance can be increased. For link budget purposes RoHC header size is assumed to be 40 bits. Compare to without RoHC, 3 dB gain due to less bandwidth needed per user (lower MCS used). In conclude, RoHC
VoLTE Optimization
Unidirectional Mode
FO
SO
O mode
R mode
bidirectional
bidirectional
Error feedback
No
Some
Intensive
Efficiency
Low
High
High
)
)
FB (U
FB (U )
U mode unidirectional
Packet direction
FB (R
FB (O )
IR
FB (O) Optimistic Mode
Reliable Mode
FB: feedback channel
FB (R) IR
FO
RoHC feedback packets carry the header compression control information: ACK: decompression success, NACK: decompression failure, STATIC-NACK: static context, invalid/ not established, D_MODE: indicating the desired, compression mode.
SO
IR
FO
SO
Figure 11.33 RoHC modes.
Figure 11.34 RoHC header size distribution.
100%
PDF
80% 60% 40% 20% 0%
504
120
24
32
40
56
40
RoHC header size [bits] 100%
CDF
95% 90% 85% 80% 75%
24
32
40
56
120
504
RoHC header size [bits]
will be enabled for voice (QCI 1) to improve coverage and system capacity. In case there is less data to transmit, UL interference is reduced and thereby increases the over‐the‐air capacity for best effort users and the cell range specifically for the uplink. 11.2.11 Inter‐eNB Uplink CoMP for VoLTE
LTE has frequency reuse of 1. That means a lot of interference on cell edges. In effect, cell‐edge UEs are received with similar power by serving a neighbor cell, but to the neighbor cell, this is interference. Uplink coordinated multi‐point reception (UL CoMP) is a feature that combines antenna signals from multiple cells of the same carrier frequency in order to improve uplink
475
476
LTE Optimization Engineering Handbook 8. RLC Data 7. Decoded VoIP data 3a. UL Comp Req. C-RNTI, PCI, MCS, PRB, ...
X2
(3b.) Start/Stop
4.
–20ms
2. Start Relaxed UL CoMP?
Decoding As usual
Cooperating eNB
Serving eNB
Decoding Only send data if CRC = ok
6. VoIP packet 5. UE Grant
1. Measurement report A3 Entering/or leaving
Figure 11.35 Inter‐eNB uplink CoMP.
throughput by increasing the received signal power and taking this interference and turn it to the useful signal, while the neighbor cell can do it too. The objective of UL CoMP is to improve UL SINR by combining antenna signals from multiple sector carriers belonging to different cells. The benefit is largest for UE that are in the border between two sectors, either two macro sectors, or between a macro sector and a small cell. UL CoMP for VoLTE can realize soft handover for VoLTE bearer, but X2 delay limits the gains. When UL CoMP is used, the expected gains such as increased coverage for VoIP, improved VoIP satisfaction, and reduced call drop rate could be achieved (Figure 11.35). Since the VoIP capacity is limited by the worst users (cell‐edge users), CoMP may be even more beneficial for VoIP than for full buffer traffic. Uplink CoMP can be used if digital processing is centralized in a common point in the network, connecting with remote radio units via fiber. These solutions are not feasible with uncoordinated small cells connected to the macro network via the S1/X2 interfaces. Both macro diversity and interference suppression can be achieved depending on the type of signal combining. Further, a spatial division multiplexing (SDM) scheme can be applied, where multiple users are scheduled at the same time in different nodes, using the same uplink resources. Up to three cells can be included in so called CoMP set ‐ > serving cells can have up to two neighbor cells in each CoMP set. For each UL transmission in the serving cell, a linear SINR average over 4 RX antennas is calculated: SINR _ S _ lin 0.25 * SINR _ S _ lin 1
SINR _ S _ lin 2
SINR _ S _ lin 3
SINR _ S _ lin 4
VoLTE Optimization
For each UL transmission in the neighbor cells from the serving cell CoMP set, a linear SINR average over 4 RX antennas is calculated: SINR _ N 1 _ lin 0.25 * SINR _ N 1 _ lin 1 SINR _ N 1 _ lin 3
SINR _ N 1 _ lin 4
SINR _ N 2 _ lin 0.25 * SINR _ N 2 _ lin 1
SINR _ N 2 _ lin 3
SINR _ N 1 _ lin 2 SINR _ N 2 _ lin 2
SINR _ N 2 _ lin 4
11.3 Semi‐Persistent Scheduling and Other Scheduling Methods PDCCH becomes a bottleneck when VoIP capacity is growing and adoption of semi‐persistent scheduling (SPS) allows a reduction of the PDCCH consumption, which in turns allows an increase in the number of data users accessing the cell and improves the cell throughput for non‐VoIP applications. Without SPS, the 100 VoIP bearers would use the entire PDCCH capacity, thus user perception of the VoIP quality is very bad (delays, frames drop) and also cell throughput is heavily impacted as regular traffic could not be granted at all. It notes that TTI bundling and SPS can’t be enabled simultaneously for TDD as defined by 3GPP. 11.3.1 SPS Scheduling
Persistent allocation of time and frequency resources can be used for each initial HARQ to each VoIP flow, retransmissions are dynamically scheduled in frequency is also fixed in time due to synchronous HARQ. At the admission accords to a predefine frequency‐time pattern, resource allocation can be with a fixed MCS and number of PRBs, for each initial Tx and subsequent re‐Tx of a VoIP frame, a scheduler grant is sent, the unused semi‐static resource can also be used by the dynamic scheduler. SPS scheduling mechanism is a combination of persistent scheduling for initial transmissions and dynamic for retransmissions. Compared to a dynamic scheduler, SPS minimizes control overhead and low DL grant usage, since grants are required only for retransmissions, but semi‐ persistently scheduled VoIP users do not have the benefit of channel‐sensitive scheduling. Frequency selective dynamic scheduling may provide some gain over frequency hopping for VoIP users at low dopplers and MCS and PRB allocation are based on control channel feedback, buffer occupancy and UE power headroom. Usually, we use dynamic scheduling for select users at the cell edge at low to medium speeds who need the extra performance boost to meet VoIP quality targets, use semi‐persistent scheduling for the remaining users. SPS gives the ability to support non VoIP traffic along with high VoIP load. Table 11.15 describes the comparation of dynamic scheduler and semi‐persistent scheduler. This part of optimization aims to enable the network achieved the target performance by how to use the two‐way scheduling strategy. SPS parameters are included in the “RadioResourceConfigDedicated” IE of the RRC connection reconfiguration message. An example is described below: Example RRC : rrcConnectionReconfiguration ....... sps-Config { semiPersistSchedC-RNTI ‘00...010’B,
477
478
LTE Optimization Engineering Handbook
sps-ConfigUL setup : { semiPersistSchedIntervalUL sf640, implicitReleaseAfter e2 ....... }
SPS-Config parameter SPS C-RNTI (UE ID used to schedule an SPS unicast transmission, activation, reactivation and retransmission) SPS DL Config SPS interval DL No. of configured HARQ processes for SPS SPS UL Config SPS interval UL Implicit release after (No. of empty transmissions (padding PDU) on the UL before the UE implicitly releases the UL SPS grant)
Value 003D – FFF3
10 – 640 ms 1–8 10 – 640 ms 2,3,4,8
Table 11.15 Dynamic scheduler versus semi‐persistent scheduler. Dynamic scheduler
Semi‐persistent scheduler
VoIP priority over data
Adaptive QoS weighting algorithms
Fixed absolute priority of SPS VoIP over data
PDCCH grant signaling
High: At every initial Tx in DL/UL Also at HARQ Retx on DL
Low: At SPS activation on DL/UL; At SPS release and HARQ Retx on OL; At explicit UL SPS release.
Link adaptation
MCS/PRB determined per packet (every 20 ms) based on user’s RF and BLER
MCS/PRB determined per talk spurt (every several seconds)
UL Power Control
Slow power control (200 ms)based on FPC, fixed target SINR based on pathloss
Fast power control (~20 ms) with adaptive SINR target based on BLER performance
RLC segmentation
Yes
No
11.3.2 SPS Link Adaptation
Different from the dynamic scheduler is the MCS/PRB/modulation assignment, which will vary depending on the air link condition, SPS link adaptation is enhanced to minimize the padding, thereby improving initial BLER. This is done through a combined selection of the number of PRB and MCS (based upon SINR and TBS input to each), instead of a sequential approach, selecting first the MCS then the number of PRB. In the case of the SPS allocation, the MCS is decided at the start of the talk spurt and used throughout the talk spurt. After SPS activation, the same PRBs are used every 20 ms for the initial transmission of the VoIP frame. For example, uplink MCS/PRB/modulation is fixed at 2PRBs, MCS = 10 when SPS is activated. UL SPS also
VoLTE Optimization
adds an outer loop power control (since there is no rate control) to avoid packet segmentation issue, which makes it even more efficient with scheduling grants compared to dynamic scheduler. DL and UL SPS activation are enabled by RRC configuration, including SPS‐C‐RNTI, UL transmission pattern interval, implicit release after count, and so on. When a SPS bearer is established, the eNB sends a RRC connection reconfiguration message with the new SPS bearer added. To activate the SPS, the DL/UL scheduler sends an SPS grant in PDCCH, addressed to the SPS‐C‐RNTI assigned to the UE. If the UE correctly decodes the grant and correctly receives the first SPS transport block (first SPS MAC packet), it sends an ACK. When the eNB receives the ACK for the SPS packet transmission, the SPS activation is considered successful, and vice versa, the reserved PDCCH resources are freed (as no longer needed). The minimum radio link quality conditions for SPS activation includes the UE is in speech active state, the normalized power headroom, the wideband SINR, no measurement gap actived for the UE and no SPS activation prohibit timer is running. DL and UL SPS activation is shown in Figure 11.36 and Figure 11.37. To deactivate the SPS, the DL/UL scheduler will send another SPS grant, if the UE correctly decodes the grant, it sends an ACK. When the eNB receives the ACK, the SPS deactivation is considered successful. DL always uses explicit release procedure by sending special form of DCI1A to the UE in case of: ●● ●● ●● ●●
●● ●● ●●
VoIP inactivity (more than 20 ms‐periods have elapsed with the SPS buffer empty) Measurement gap activation procedure SPS activation failure Potential collision with PDSCH carrying common channels (e.g., paging, D‐BCH) or positioning reference signals or eMBMS After inactivity timer of SPS release expires BLER is higher than threshold for SPS SPS grant parameter value (i.e., MCS or number of PRBs) changed due to GBR modification
Figure 11.36 DL SPS activation.
479
Figure 11.37 UL SPS activation.
VoLTE Optimization VoIP SID RLC SDU detected Grants are sent until implicit release
2 MAC PDUs with 0 MAC SDUs detected
drxShortCycle = 40 ms
4 ms
4 ms
2 MAC PDUs with 0 MAC SDUs sent on sps
(SID) BSR + Data
SR
PDU with 0 MAC SD U
PDU with 0 MAC SD U
BSR + Data
Grant
4 ms
grant
spsAct
Grant
grant
spsAct
(SID)
eNB
UE
Implicit Release
160 ms
x
4
ms
ms
Implicit Release
Figure 11.38 SPS implicit released (example). ●● ●●
Last VoIP bearer deletion SINR falls below the minimum SINR For SPS
In such above cases, SPS is released by an implicit release. An example of implicit release is shown in Figure 11.38 that the release will occur after two consecutive MAC PDUs without any MAC SDUs. When an SPS release is initiated, all transmissions (intial and retransmissions) are managed by the dynamic scheduler. The SPS release is considered successful if an ACK is received. 11.3.3 Delay Based Scheduling
On top of the QoS aware scheduler, the idea is that VoIP packet is not scheduled immediately but that several packets are packed in the same transport block. eNB shall check whether there is a need to allocate resource to the UE at every UL scheduled TTI. A delay‐based scheduler will prioritize users whose packets are getting close to their PDB (packet delay budget). The amount of bundling is determined by the setting of bundling time, which, in turn, is influenced by the operator via the parameter PDB (Figure 11.39). The goal of delay‐based scheduling is to ensure that all VoIP packets of UEs in good and poor channel conditions are received correctly within assumed delay target, that is, delay–based scheduling is a mechanism that allows to match the packet aggregation level to the delays on radio interface. Figure 11.39 shows that under relatively fair channel conditions, aggregation can be increased, while in good channel conditions, more packets can be aggregated. The scheduler evaluates the actual scheduling delay of voice packet and compares it with a delay target to decide UE priority, if measured delay is bigger than delay target, UE priority is increased, otherwise UE priority is decreased. The idea is that VoIP packet is not scheduled immediately but several packets are packed in the same transport block. So depending on the configured threshold, one can control on how many VoIP packets are bundled at one transport block by knowing the VoIP packet size of 328 bits. However, there is also a requirement for delay, so that buffering is not possible in all cases. Delay‐based scheduling provides the ability to observability for real‐time service in terms of the degree to which packets meet their packet delay. The benefit of DBS (delay based scheduler) is improved capacity. It provides more opportunities to schedule best effort traffic. Uplink packet sizes can be predicted so as to allow uplink grants requests to be minimized, thereby reducing control channel usage. VoIP packets for UEs in good radio conditions can be bundled without compromising VoIP quality. When VoIP packets are bundled, resources are freed that can be used for scheduling of MBB services. The main key areas where this scheduler contributes to are voice performance, smartphone efficiency, and network efficiency.
481
482
LTE Optimization Engineering Handbook
w P1
Scheduler internal parameters
P2 P3 PX PDB1
2 el
v
i at
eg
r gg
on
Le
80
v
n
tio
a
eg
r gg
A
3 el
Le
60
A
40
Delay Term
No
io at
g
re
PDB3
100
n
g Ag
PDB2
20 80
70
60
50
40
30
20
10
0
–10
remaining delay budget [ms]
Figure 11.39 Packet delay budget and its configurable attributes.
The combined DBS and SABE (service aware buffer estimation) scheduler is prefered in a live network that will automatically keep track of whether the user is in TALK or SID (silence indicator) state in order to control the packet‐bundling periodicity and grant estimation size. The combinded DBS and SABE scheduler adapts to handle different codec rates, RoHC on/off, IPv4 or IPv6, and so on. The delay added by combined DBS and SABE is expected to be less than PDB + pdbOffset; therefore, these two functions should work together without a problem. It’s worth to note that DBS and SABE are not beneficial or efficient for video. It is not possible to recognize a video packet in the system in the same way as a VoIP packet is recognized as it is not possible to determine the age of a video packet. 11.3.4 Pre‐scheduling
For LTE uplink, the scheduler is located in the eNB and the buffers are in the UE. Therefore, the uplink scheduler has no direct knowledge of the amount of data available in the UE. The UE can inform the uplink scheduler that it has data by sending a scheduling request (SR) on PUCCH or PUSCH. The terminal can also send a buffer status report (BSR) together with data on PUSCH indicating the size of every priority queue in the UE. There will always be a delay between the time when the data arrives in the buffer and the time when the uplink scheduler is informed. VoIP is a very delay‐sensitive service so minimizing the delay of the uplink will increase the VoIP capacity. Pre‐scheduling is a method to minimize UL delay by means of blindly giving PUSCH grants to a UE in advance, without receiving buffer status information (SR, BSR). Pre‐scheduling
VoLTE Optimization PING
UE
5 10 15 20 PING
5
SR
1
Grant PING
1 4 1
PING PING
PING
UE 0 5 10 PING
eNB
SR encod. and alignment 3
Figure 11.40 Total round trip delay(23 ms) versus total round trip delay with pre‐scheduling (13 ms).
eNB processing 3GPP specified delay
1 3 Core network delay 2 1 1
Time [ms]
PING
Delay Breakdown
Server
eNB
0
Delay Breakdown
Server
Grant PING
PING
PING PING
4 1
1 1
3GPP specified delay 1 3 Core network delay 2
15 20 Time [ms]
means the eNB periodicity grants the UE even if UE reported BSR = 0 (UL buffer empty) or the UE has not sent an SR. Pre‐scheduling improves UE response time and thereby reduce latency, but with the cost of higher UE battery power consumption and also occupy the resources of the network without guarantee that these resources will be used for transmission, since the grants are giving blindly to the UE. The UE will respond to the UL grants even if it has no data to send. In this case it will send padding data (the BSR is called “Padding BSR”) and thus UE will not go DRX and UL out‐of‐sync state. All in all, the UE will less often be in a battery saving state when the UE has no data to send when pre‐scheduling is enabled. Pre‐scheduling can be automatically deactivated (timer controlled) to reduce the impact on UE power consumption. As mentioned in Chapter 2, the most common benchmark for observability of the IP packet latency in the network latency is the ping test. It is used to test the reachability of the host on IP network and to measure the round trip delay in the network. The theoretical analysis, which has been confirmed by the measurements in the network, shows that when prescheduling was used in uplink transmission, the latency performance was about 6 ms better than the case with ordinary scheduling request (see Figure 11.40). One example is shown in Figure 11.41, one advantage is paging time can be reduced by 30% to 50% by pre‐scheduling, and another one is the corresponding time inprovement on a complete web download is 3% to 10%. Pre‐scheduling targets the cases where the traffic load in the network is low or moderate. Actually, prescheduling grants are sent only if there are free resources and only in good radio condition. In other hand, counters and drive tests results show that uplink is heavily impacted by the prescheduling, due to increase of UL interference. In particular, the selected transport format and the uplink BLER is much worse when prescheduling is active, power consumption while transmitting doesn’t change too much.
483
484
LTE Optimization Engineering Handbook No Prescheduling, m = 27 ms, std = 3 ms 1 ms Prescheduling, m = 14 ms, std = 1ms 5ms Prescheduling, m = 16 ms, std = 2 ms 10ms Prescheduling, m = 18 ms, std = 3 ms
0
5
10
20
25 [ms]
30
15
35
No Prescheduling, mean DL time = 6325 ms 1 ms Prescheduling, mean DL time = 5690 ms 5 ms Prescheduling, mean DL time = 6065 ms 10 ms Prescheduling, mean DL time = 6139 ms
0
1
2
3
4
5
6
[s]
Figure 11.41 The example of pre‐scheduling.
11.4 PRB and MCS Selection Mechanism Varying transport formats (modulation, coding scheme, number of RBs and TBS) result in different SINR requirements. VoLTE data rate may vary depending on codec rates. Coverage for VoLTE users depends on the cell‐edge SINR in relation to chosen TBS. Enhance the UL coverage of VoLTE in bad radio conditions, the mechanism of PRB and MCS selection that the UL scheduler uses to control the minimum number of PRBs and the minimum MCS value assigned to scheduled PRBs. PRB override and MCS override usually working together only for VoIP calls. The scope is to be able to achieve the VoIP codec rate (e.g., 12.65 kbps AMR‐WB) and thus obtain a good call quality even in bad radio conditions. 11.4.1 Optimized Segmentation
For a UE in bad radio conditions, the UL scheduler is going to start segmentation of packets in order to use a more robust MCS for the transmission of each individual packet. Segmentation of the VoIP packet introduces delay to transmission of the packet, thus limiting the maximum useful bit rate achievable on the radio link. To ensure the reasonable reliable VoLTE services at poor UL RF condition, two types of constraints need to be addressed to residual block error rate (rBLER) and the packet transfer delay. When the uplink scheduler struggles to maintain its target initial block error rate (iBLER) performance when the total amount of power used for the transmission cannot be increased. The typical behavior of the uplink scheduler is to start segmenting the packets to allow the use
VoLTE Optimization
of more robust MCS for the transmission of each individual packet. The number of payload bit per transmission is consequently reduced, thereby increasing the amount of power per transmitted bit, which in turns leads to an improved iBLER performance. The packet segmentation introduces delay to the transmission of the packet and inherently introduces a bottleneck to the maximum useful bit rate that can be achieved on the radio link (since the number of bit per TTI is restricted). Beyond a critical level of segmentation (i.e., below a minimum payload size per TTI) the radio link throughput becomes lower than the VoLTE codec rate and the overall packet delay start to build up. Under various fading channel conditions, optimized segmentation yields 3‐4 dB improvement in terms of maximum acceptable uplink path loss (MAPL) for an AMR 12.2 kbps codec managed with uplink dynamic scheduler, comparing with the case no optimized segmentation is enforced. 11.4.2 PRB and MCS Selection
If initial BLER is too low, meaning MCS/PRB selection is too conservative, then PRBs used for initial SPS packet transmission will be high even if the retransmission will be low. If initial BLER is too high, meaning MCS/PRB selection is too aggressive, then the retransmission of the SPS packet may be high and also, when users move to a worse RF condition, SPS may not work. In this part, we try to find the right MCS/PRB table to make the initial BLER to be around 10% for SPS packets. There is a trade‐off between the initial BLER for SPS and the total PRB usage. SPS PRB and MCS selection procedure is shown in Figure 11.42. Example The 12.65 kbps AMR‐WB codec delivers a 32‐byte speech frame every 20 ms, consequently the total payload to be transported over four segments is: 3byte RoHC header + 1byte PDCP header + 6 bytes for 1st segment header size + 3 x (4 bytes for subsequent header size) + 32 bytes useful payload = 176 bit header + 256 bits useful payload = 432 bits → Consequently the minimum transport block size per segment is 432/4 = 108 bits, and the four prefered TBS can be selected in Table 11.16. In a live network, as described in the example above, the following MCS rules should therefore be used for enforcing a maximum of four segments per speech frames as different number of scheduled RBs will affect uplink coverage. In Table 11.15, if the PUSCH grant size is 1 PRB,
Figure 11.42 SPS PRB and MCS selection.
485
486
LTE Optimization Engineering Handbook
Table 11.16 Transport block size. IMCS TB size (in bits)
NPRB
0
1
2
3
4
5
6
7
8
9
10/11
1
16
24
32
40
56
72
328
104
120
136
144
2
32
56
72
104
120
144
176
224
256
296
328
3
56
88
144
176
208
224
256
328
392
456
504
4
88
144
176
208
256
328
392
472
536
616
680
5
120
176
208
256
328
424
504
584
680
776
872
the minimum MCS = 8, if the PUSCH grant size is 2 PRB, the minimum MCS = 4, if the PUSCH grant size is 3 PRB, the minimum MCS = 2, and if the PUSCH grant size is 4 PRB, the minimum MCS = 1.
11.5 VoLTE Capacity As specified by 3GPP, the requirement for VoIP capacity is at least 200 users for each cell in the 5 MHz bandwidth. Basically, three things can limit the VoIP capacity in an LTE system: control channel capacity, the number of resource blocks, or UE power. In other words, VoLTE capacity mainly depends on scheduling capacity and PUSCH capacity. More symbols allocated for the PDCCH per TTI gives better capacity, lower voice codec bit rates permits more users, and higher allowed radio interface delay also permits more users. From network design point of view, smaller cell size permits more users. Some features in VoLTE can also help increasing capacity. RoHC is deployed in VoLTE, which increases the VoIP capacity in cases where PDSCH or PUSCH capacity is the limiting factor. RLC UM increases the VoIP capacity in cases where the number of transmissions per TTI is limited. DBS and SABE scheduling can also increase the capacity for VoIP users. The VoLTE capacity is defined according to the number of QCI 1 bearers managed by the cell. Based on the actual bitrate per VoIP call listed in Table 11.17, we can calculate the number of simultaneous VoIP calls. VoIP system capacity is defined as the number of users in the cell when more than 95% of the users are satisfied. VoIP capacity is defined as the min of the two capacities. Outage criteria: less than 5% of VoIP users in the system are in outage; Delay criteria: less than 5% of VoIP users have its 98th percentile delay greater than 50 ms. User outage criteria includes: ●●
●●
●●
Packet error rate (PER): the ratio between num of packets in errors after max num of HARQ Tx plus num of packets discarded at UE and num of total packet transmitted by the UE. VoIP outage: a UE is in VoIP outage if its PER exceeds 2% Call dopping criteria: A VoIP call is dropped if more than a certain percentage (10%) of the VoIP packets is lost in an observation window of 10s (sliding window of talk time). Lost VoIP packet criteria: A packet (or any segment) is discarded at eNB if not sent within the delay budget or a packet (or any segment) is discarded at UE if received beyond the delay budget (it can happen in case of retransmission) or a packet (or any segment) is lost due to HARQ failure.
VoLTE Optimization
Table 11.17 Source codec bit‐rates for the AMR codec (from 3GPP TS26.071) and AMR‐WB codec (from 3GPP TS26.171). Codec mode
Source codec bit‐rate
Codec mode
Source codec bit‐rate
AMR_12.20
12.20 kbit/s (GSM EFR)
AMR‐WB_23.85
23.85 kbit/s
AMR_10.20
10.20 kbit/s
AMR‐WB_23.05
23.05 kbit/s
AMR_7.95
7.95 kbit/s
AMR‐WB_19.85
19.85 kbit/s
AMR_7.40
7.4 kbit/s (IS‐641)
AMR‐WB_18.25
18.25 kbit/s
AMR_6.70
6.70 kbit/s (PDC‐EFR)
AMR‐WB_15.85
15.85 kbit/s
AMR_5.90
5.90 kbit/s
AMR‐WB_14.25
14.25 kbit/s
AMR_5.15
5.15 kbit/s
AMR‐WB_12.65
12.65 kbit/s
AMR_4.75
4.75 kbit/s
AMR‐WB_8.85
8.85 kbit/s
AMR_SID
1.80 kbit/s
AMR‐WB_6.60
6.6 kbit/s
AMR‐WB_SID
1.75 kbit/s
11.5.1 Control Channel for VoLTE
With a baseline dynamic VoIP packet scheduling approach, where each packet is transmitted by layer1 control signaling, multi‐UE channel sharing is an important aspect when optimizing air interface capacity for VoLTE traffic. PDCCH being transmitted to each scheduled terminal, its overhead can be significant and limiting factor since VoIP consists of many small packets. This can result in insufficient control channel resources for scheduling all physical resource blocks while wasting part of PDSCH capacity. Highest achievable PDCCH capacity [users per cell] is given by: NVoIP ,max
N grants SE per TTI consumption 2
N grants
N CCE N CCE ,ave
Factor 2 in the formular above appears because required Ngrants is sum of requirements for UL and DL. NCCE is the number of allocated CCEs in the cell which depends on bandwidth. NCCE,ave is the average number of CCEs per PDCCH which depends on cell size and load. In a live network, DL scheduling entity (SE) consumption estimation can be got from:
DLSEperTTI ,VoLTE
VAF 40
1 VAF * 1 % Retrans. 160
In the live network, segmentation and TTI‐Bundling will occur on the UL due to UE UL power limitation, average UL SE considering mixed radio condition (neither segmentation nor TTI‐Bundling, segmentation, TTI Bundling) is given below. ULSEperTTI ,VoLTE
VAF 1 VAF 40 160 VAF 1 VAF %Seg . * 2 * 40 160 VAF 1 VAF * 1 % Retrans. %TTI Bunding * 4 * 40 160
1 %Seg . %TTI Bunding *
487
488
LTE Optimization Engineering Handbook
Table 11.18 SINR versus CCE. Aggregation level
1
2
4
8
Max number of CCEs
444
222
111
55
Swith point (dB)
6.3
2.4
0.5
−2
According to the network Ng = 1 configuration, PDCCH resources are aggregated in groups of 1, 2, 4, and 8 CCEs, and different aggregation level needs different RS SINR that is shown in Table 11.18. According to the real user distribution in a live network, we can estimate the number of VoIP users in a cell in a simple way,
444 * 3/10 222 * 2/10 111 * 2/10 55 * 3/10 216
In a live network, VoLTE packets consists of two parts: activity (Talk) and inactivity (Silence). Standard voice is packetized in 20‐ms intervals. With DBS it will be 40 ms, for the calculations 20 ms is used, for the inactivity part, SID frames arrive in 160 ms interval. In this case, the VoIP capacity in a cell can still be higher. For UL, the L1/L2 control signaling is carried by PUCCH, CQI information indicating the channel quality estimated by the terminal, and UL scheduling requests indicating that the terminal needs UL resources for PUSCH transmissions, as the number of served UE increases, the CQI overhead can increase significantly, which may become prohibitively large with VoLTE traffic while a large number of UEs must be supported. To maintain the UL control overhead at a reasonable level, CQI feedback should be reduced both in time and in the frequency domain. A reduction of CQI overhead can be achieved with slightly reduced traffic capacity by means of wideband CQI with loss of frequency scheduling gain. 11.5.2 Performance of Mixed VoIP and Data
VoLTE call quality is dependant on the LTE network handling thousands of concurrent SIP sessions per second. Due to the fact that VoIP is a guaranteed bit rate (GBR) service it will be served (scheduled during time domain scheduling phase) before all non‐GBR services, dedicated QoS for voice in LTE will impact data applications running on the same limited bandwidth resources. The average data throughput for a user with a VoIP bearer established is decreased comparing to the case when only the data bearer is established. Non‐GBR UE throughput decreases with an increase of the number of VoIP UEs in the cell, due to PDCCH block and the low priority, as compared with VoIP UE. VoIP UE throughput is stable regardless of the number of VoIP UEs in the cell. UL non‐GBR UE throughput is almost zero when the number of VoIP UEs is larger than 130, because non‐GBR traffic has no chance to be scheduled (Figure 11.43). From Figure 11.44, total cell throughput decreases with higher number of VoIP users in the cell. VoIP traffic increases linearly, has higher priority thus is not blocked. Due to lower priority non‐GBR traffic has much less scheduling occasions; therefore, throughput generated by non‐ GBR UEs decreases when more VoIP UEs are in the network. Using SPS for VoIP alleviates the demand for scheduling grants on PDCCH, allowing higher data throughput. UL SPS adds outer loop power control (since there is no rate control) to avoid packet segmentation issue, which makes it even more efficient with scheduling grants c ompared to DS.
VoLTE Optimization Parameter Layout
Setting 3GPP Macro Case 1 according to TR 25.814
ISD
500 m
Link
DL + UL
Duplex mode
TDD DL/UL:TDD Conifg 1 Special subframe: Config 5 according to 3GPP TR 25.814 3GPP_TR25814_3sector_70deg_14dBi
TDD DL/UL Config Antenna type Transmission modes
TM1
Operational band (MHz)
2 GHz
Bandwidth (MHz)
10M Hz eNB 0.8W per PRB
Output power indoor penetration loss
UE 23dBm
Receive Diversity
2 RX Std. dev = 8 dB; corr. distance = 50 m
Slow fading Fast fading
20 dB
TU3
UL mean # of scheduled UEs Per TTI 10 8 non-GBR VoIP
6 4 2 Ideal Real
Ideal Real
Ideal Real
Ideal Real
Ideal Real
Ideal Real
Ideal Real
0
10
50
100
130
140
150
175
Number of VoIP UEs Per Cell
Figure 11.43 10 MHz TDD cell UL mean # of schedeled UEs per TTI.
Overall, there is a degradation in data throughput with DS compared to SPS, and the degradation increases with the number of voip users increases. In Figure 11.45, we can see that 100 voice calls, the degradation is around 34%; 50 voice calls, the degradation is about 13%; 20 voice calls, no degradation is found. Both UL and DL show that SPS scheduler improves the residual data throughput when the number of voip users is large, which causes grant shortage, that is, for 50 VoIP users, the UL and DL gain of SPS versus DS is 13% and 5%, respectively. For 100 VoIP users, the UL and DL gain of SPS versus DS is 34% and 13%, respectively. When the number of VoIP users is small, or the grant is not the limiting factor, SPS scheduler performs similar with DS scheduler. In a live network, voice outage in a cell is sensitive to interference over thermal (IoT). Higher IoT level is due to inreasonable network structure, more users, higher SINR targets, and higher bandwidth occupancy with SPS compared to DS. Figure 11.46 gives the relation between IoT and number of UEs.
489
UE Throughput (kbps)
UE Throughput (kbps)
LTE Optimization Engineering Handbook Mean UL non-GBR UE TP
250 200 150 100 50 0
10
50
100 120 130 140 150 Number of VoIP UEs Per Cell
175
Mean UL VoIP UE TP
8 7 6 5 4 3
10
50
100
120
130
140
150
Number of VoIP UEs Per Cell
DL mean TP
Cell Throughput (kbps)
7000
Total TP
6000
no-GBR TP VoIP TP
5000 4000 3000 2000 1000 0
10
50
100 130 140 150 Number of VoIP UEs Per Cell
175
UL mean TP
2500
Total TP Cell Throughput (kbps)
490
Non-GBR TP
2000 1500 1000 500 0
10
50
100
130
140
Number of VoIP UEs Per Cell
Figure 11.44 Performance of mixed VoIP and data.
150
175
VoLTE Optimization
Residual data throughput (Mbps)
16.00 14.00
DS
12.00
SPS
10.00 8.00 6.00 4.00 2.00 0.00
0
50
100
150
200
Number of voice calls
Figure 11.45 Benefits of SPS for VoIP on uplink (10 MHz carrier, TDD, multicell). Figure 11.46 IoT versus number of UEs.
8
IoT dB
6 4 2 0
40
50
60
70
80
90
100
110
UE Number 14 12
IoT (dB)
10 8 6 4 DS, AL = 4
2
SPS
0 0
20
50
100
Number of Voice Calls
11.6 VoLTE Coverage As the industry moves toward deployments of VoLTE, one of the most important performance aspects is to be able to provide similar coverage for VoLTE as end users have come to expect from 2/3G voice. However, VoLTE introduces several coverage challenges that were not present in 3G circuit switched voice. For instance, VoLTE deployments have targeted use of high‐ definition voice codes such as AMR 12.65 kbps for improved voice quality, while many 2/3G voice network designs have been based on lower rate narrow band voice even AMR 5.9 kbps, which can provide larger coverage since fewer bits need to be sent. Further, VoLTE has
491
492
LTE Optimization Engineering Handbook
a dditional IP, PDCP, RLC, and MAC header overhead compared to circuit switched based voice. This extra IP overhead for VoLTE compared to 2/3G CS voice again means fewer bits for coding, reducing receiver sensitivity. Fortunately, there are aspects of LTE that help make up for the shortcomings that will be discussed in this part. In order to get good VoIP coverage, segmentation and TTI bundling is used. With segmentation, VoIP packets are split into smaller parts in order to distribute transmission over several TTIs. Since the UE output power is limited, it is important to achieve continuous transmission to maximize the transmitted energy. TTI bundling is used to minimize overhead, by autonomous retransmission over four consecutive TTIs. This reduces the need for segmentation and also control signaling needed to schedule each transmission. Also, minimum bandwidth allocation could be adopted to enhance the UL coverage. The techniques discussed in this part will achieve different compromises in terms of experienced BLER, protocol overhead, required scheduling grants, efficient MCS, bandwidth utilization, and packet delay. 11.6.1 VoIP Payload and RoHC
VoIP coverage mainly determined by VoIP throughput requirement. VoIP throughput is defined by the voice codec and overhead introduced by lower layers. The VoLTE voice codec, L2/L1 SDU bits is shown in Table 11.19. Different voice code rate have different SINR requirement, reducing AMR codec rate from 12.2 kbps to 5.9 kbps brings additional ~1 dB gain in DL and ~1.2 dB gain in UL. Table 11.20 gives the SINR requirement of three types voice code rate. Use of RoHC can considerably compress the header size in voice packets and reduce the user plane traffic over the air interface. The high compression ratio obtained not only increases network capacity but also provides coverage improvements for VoLTE compared to other VoIP alternatives. The test result of bandwidth consumption for AMR and AMR‐WB (Figure 11.47). Table 11.21 and Table 11.22 gives the UL and DL VoIP bitrate requirement with and without RoHC. Here is an example of field test, under two UEs VoIP call with mobility slow speed (30 km/h EVA), the inter‐arrival time of originator/terminator VoIP calls, and average VoIP throughput are shown in Figure 11.48, Figure 11.49, and Figure 11.50. 1) Originator is under the condition of inter‐arrival VoIP packets are 20 ms, inter‐arrival SID packets are 160 ms; first DTXed packets inter‐arrival is 60 ms. We can get the average VoIP inter‐arrival time is 30 ms, the maximum inter‐arrival time is 167 ms, the minimum inter‐ arrival time is 0 ms. 2) From terminator, impact of the network jitter can be noticed, and we can get the average VoIP inter‐arrival time is 30 ms, the maximum inter‐arrival time is 180 ms, the minimum inter‐arrival time is 0 ms. 3) From the throughput pattern and statistics, we can observe the ‘DTX on’ effect, SID frames will be sent if there is no speech activity. We can get the average throughput is 24 kbps, the maximum throughput is 38 kbps, the minimum throughput is 4 kbps. 11.6.2 RLC Segmentation
When a VoLTE UE moves toward the cell edge, SINR received by eNB starts to decrease due to UE power limitation. To maintain the initial block error rate (iBLER), UL dynamic scheduler automatically starts segmenting the VoIP packets from PDCP into multiple smaller packets. This is called a packet segmentation algorithm, which is used as an extension to link adaptation for uplink cell edge coverage improvement.
Table 11.19 VoIP payload and L2/L1 throughput.
Voice Codec
Codec Source Rate [kbps]
Frame Size [bit]
Padding [bit]
PDCP Header [bit]
L2 SDU [bit]
RLC header [bit]
MAC header [bit]
CRC [bit]
L1 SDU [[bit]
AMR‐NB 1.8
1.8
36
4
40
80
8
8
24
120
AMR‐NB 4.75
4.75
95
1
40
136
8
8
24
176
AMR‐NB 5.15
5.15
103
1
40
144
8
8
24
184
AMR‐NB 5.9
5.9
118
2
40
160
8
8
24
200
AMR‐NB 6.7
6.7
134
2
40
176
8
8
24
216
AMR‐NB 7.4
7.4
148
4
40
192
8
8
24
232
AMR‐NB 7.95
7.95
159
1
40
200
8
8
24
240
AMR‐NB 10.2
10.2
204
4
40
248
8
8
24
288
AMR‐NB 12.2
12.2
244
4
40
288
8
8
24
328
AMR‐WB 1.75
1.75
35
5
40
80
8
8
24
120
AMR‐WB 6.6
6.6
132
4
40
176
8
8
24
216
AMR‐WB 8.85
8.85
177
7
40
224
8
8
24
264
AMR‐WB 12.65
12.65
253
3
40
296
8
8
24
336
AMR‐WB 14.25
14.25
285
3
40
328
8
8
24
368
AMR‐WB 15.85
15.85
317
3
40
360
8
8
24
400
AMR‐WB 18.25
18.25
365
3
40
408
8
8
24
448
AMR‐WB 19.85
19.85
397
3
40
440
8
8
24
480
AMR‐WB 23.85
23.85
477
3
40
520
8
8
24
560
Table 11.20 SINR requirement of different voice code rate. Codec
AMR‐NB 12.2
Bits/frame
DL
UL
AMR‐NB 7.95
328
AMR‐NB 5.9
240
#PRBs
MCS index
10% BLER (1st Tx)
2% BLER (4th Tx)
1
18
[dB] 11,87
[dB] 1,46
2
11
5,54
3
7
6
3
MCS index
200
10% BLER (1st Tx)
2% BLER (4th Tx)
[dB]
[dB]
MCS index
10% BLER (1st Tx)
2% BLER (4th Tx)
[dB]
[dB]
15
9,65
0,27
13
7,87
–0,81
–2,03
8
3,73
–3,31
7
3,35
–3,61
2,72
–4,09
6
2,23
–4,17
4
0,76
–5,36
–0,10
–5,63
2
–0,92
–6,27
1
–1,55
–6,72
10% BLER (1st Tx)
2% BLER (4th Tx)
MCS index
10% BLER (1st Tx)
2% BLER (4th Tx)
10% BLER (1st Tx)
2% BLER (4th Tx)
[dB]
[dB]
[dB]
[dB]
[dB]
[dB]
#PRBs
MCS index
1
17
11,43
1,95
2
10
5,36
3
7
6
3
MCS index
15
9,37
1,28
13
7,95
0,41
–1,97
8
3,83
–2,75
7
3,11
–3,07
2,54
–3,62
6
1,35
–4,25
4
0,78
–4,73
–0,33
–5,58
2
–1,37
–6,43
1
–1,89
–6,79
VoLTE Optimization 60.0
150%
40.0
100%
20.0
50%
0.0
Normal Consumption
Consumption With Consumption With RoHC RoHC & VAD Kbps
0%
% Reduction
Figure 11.47 RoHC versus VoIP throughput. Table 11.21 VoLTE bitrate estimation – downlink. AMR12.2
AMR WB12.65
AMR WB23.85
No RoHC
RoHC
No RoHC
RoHC
No RoHC
RoHC
Payload and headers
616
320
624
328
848
552
Number of segments
1
1
1
1
1
1
TBS per segment
616
328
648
328
872
568
CRC added
640
352
672
352
896
592
Bitrate[kbps]
30.8
16.4
32.4
16.4
43.6
28.4
SID packet size (bit)
56
56
56
56
56
56
Number of SID packet/sec
10
10
10
10
10
10
SID bitrate[kbps]
0.56
0.56
0.56
0.56
0.56
0.56
Activity factor
60%
60%
60%
60%
60%
60%
Overall bitrate[kbps]
18.7
10.1
19.7
10.1
26.4
17.3
Packet segmentation algorithm is not an event‐triggered mechanism, it is done automatically and only for UEs with poor radio channel. In worsening radio conditions scheduler performs packet segmentation on layer 2 in order to use more robust MCS and transmits the packet over multiple TTIs. Since RLC/MAC overhead is transmitted more than once, more resources are consumed to transmit the same amount of user data, more resources on PDCCH are utilized and also on PHICH due to transmission of ACKs/NACKs for HARQ purposes. Also the basic dynamic scheduler has the problem that it may excessively segment voice packets trying to maintain 10% iBLER as the path loss increases, leading to build‐up of voice frame segments in the queue leading to excess delay and this becomes the factor which limits the link budget (Figure 11.51). RLC segmentation reduces the payload bit per transmission and increases the amount of power per transmitted bit. It allows to use more robust MCS. Consequently, iBLER is reduced. Each segment occupies a separate HARQ process, the RLC layer reassembles the packet when all segments are successfully received. Each segment needs an additional grant, more resources on PDCCH are utilized, which increases the burden on the PDCCH. Link adaptation in dynamic scheduler (DS) reduces MCS level to a point where eventually a VoIP packet is segmented into many separate smaller MAC SDUs, and improves the link budget.
495
Table 11.22 VoLTE bitrate estimation – uplink. AMR12.2
AMR WB12.65
AMR WB23.85
Segmentation
TTI Bundling
Segmentation
TTI Bundling
Segmentation
TTI Bundling
No RoHC
RoHC
No RoHC
RoHC
No RoHC
RoHC
No RoHC
RoHC
No RoHC
RoHC
No RoHC
RoHC
Payload with RTP/UDP/IP/PDCP
584
288
584
288
592
296
592
296
816
520
816
520
Payload and all headers
616
320
616
320
624
328
624
328
848
552
848
552
4
4
4
4
4
4
4
4
4
4
4
4
Size per sub‐packet
162
88
616
320
164
90
624
328
220
146
848
552
TBS per segment
176
104
632
328
176
104
632
328
224
176
904
552
CRC added
200
128
656
352
200
128
656
352
248
200
928
576
180.8
110.4
Number of segments
Bitrate[kbps]
35.2
20.8
SID packet size (bit)
56
56
Number of SID packet/s
10
10
SID bitrate[kbps]
0.56
0.56
126.4
65.6
35.2
20.8
56
56
56
56
10
10
10
10
0.56
0.56
0.56
0.56
126.4
65.6
44.8
35.2
56
56
56
56
56
56
10
10
10
10
10
10
0.56
0.56
0.56
0.56
0.56
0.56
Activity factor
60%
60%
60%
60%
60%
60%
60%
60%
60%
60%
60%
60%
Overall bitrate[kbps]
21.3
12.7
76.1
39.6
21.3
12.7
76.1
39.6
27.1
21.3
108.7
66.5
VoLTE Optimization 180
Inter-Arrivel [ms]
160 140 120 100 80 60 40 20 :04 :17 :24 :30 :37 :43 :00 :55 :02 5.6 :01 :02 :02 :34 :41 :46 :52 :59 :05 :10 :17 :23 :30 :37 :42 :48 :55 2.1 5.7 :12 :18 :25 :33 :38 :45 :52 :58 :5.4 :09 :16 :23 :29 :34
0
180 160 140 120 100 80 60 40 20 0
14.8 17.8 14.8 10.9 17.7 13.2 19.6 15.9 12.2 16.6 13.3 10.4 17.0 14.3 11.2 16.1 12.7 19.2 15.4 10.5 17.5 13.5 10.4 17.2 12.3 18.9 15.8 12.2 16.7 12.9 18.8 15.9 13.2 18.8 15.5 12.4 18.3 15.5 10.0 16.7 13.2 19.7 14.7
Inter-Arival [ms]
Figure 11.48 Inter‐arrival time of originator VoIP calls from field test.
Figure 11.49 Inter‐arrival time of terminator VoIP calls from field test.
Throughput IP [kbits/sec]
40 35 30 25 20 15 10 5 0
Figure 11.50 VoIP throughput distribution.
Each segment is wrapped with RLC/MAC header and CRC checksum and is transmitted in a separate transport block causing additional overhead. VoIP packet size with RLC segmentation and overhead analysis are presented in Table 11.23. One example of segmentation of AMR‐NB 12.2 kbps is shown in Figure 11.52. Due to smaller segment sizes, RLC segmentation can help the link budget that the RLC segmentation allows higher total energy accumulation within the delay budget, and more robust MCS can be used increasing coverage but in cost of capacity degradation.
497
498
LTE Optimization Engineering Handbook To fit 3GPP defined TBS of 328
41 Bytes (328b) MAC RLC PDCP ROHC (1Byte) (1Byte) (1Byte) (4Byte)
AMR12.65 (33 Byte)
Pad (1Byte)
CRC (3Byte)
RLC payload is segmented
MAC RLC (1Byte) (1Byte)
Segment 1 (N1 Bytes)
CRC (3Byte)
MAC RLC (1Byte) (1Byte)
Segment 2 (N1 Bytes)
CRC (3Byte)
MAC RLC (1Byte) (1Byte)
Segment 3 (N1 Bytes)
CRC (3Byte)
Figure 11.51 RLC segmentation.
Segmentation is also one of the methods for reducing the number of retransmissions. RLC SDUs are segmented at the RLC layer and the resulting segments transmitted in subsequent TTIs. When a RLC SDU is divided into many segments, each transport block has its own RLC/ MAC header. The transport blocks are transmitted in consecutive TTIs using different HARQ processes. The size of each RLC/MAC header is at least 24 bits but can be more. In addition, layer 1 adds 24 bits of CRC to each transport block. This means that segmentation increases the number of bits used for MAC and RLC headers as well as for CRC. Sending packet segments in consecutive TTIs allows more energy per voice packet to be aggregated, increasing coverage. Packet segmentation can provide up to ~5 dB coverage gain (by sacrificing capacity). Real gains will depend on used codec, link adaptation settings and number of segmentations. In a live network, the typical VoLTE RLC PDU distribution can be shown in Figure 11.53. Table 11.24 gives the coverage gain of AMR12.2 VoIP codec of regular transmission and segmentation. Since the maximum number of supported VoIP users per cell depends mostly on PDCCH resources, it significantly drops with the increase of segmentation order. Each segment (and HARQ retransmission) will require a new PDCCH allocation and a new PHICH in DL (for transmission of ACKs/NACKs per segment). The higher the achieved coverage gain, the greater is the capacity loss. Table 11.25 gives the PDCCH resources occupation with regular transmission and different segmentations. 11.6.3 TTI Bundling
TTI Bundling is intended particularly for addressing LTE uplink link budget issues and balance the uplink coverage with the downlink coverage footprint. The main reason for the imbalance of the UL/DL link budgets is that UE maximum transmit power is 200 mW, as opposed to tens of watts for the downlink. UE cannot use enough energy during one TTI in order to send successfully a VoIP packet. The uplink, however, is believed to be coverage limited, and therefore, TTI bundling can be very beneficial when the UE is close to the cell edge specifically for VoLTE services. The eNB activates TTI bundling if the SINR for the UE drops below a configurable threshold when BLER increases and link adaptation has no more options for MCS/PRB reduction. When TTI bundling is activated,TTI bundling uses four automatic retransmissions in four consecutive uplink TTI with a common ACK/NACK for HARQ and that means the receiver waits over the total transmission period (four TTIs) before sending feedback. The four consecutive uplink TTIs are called TTI bundle, and these four transmissions are non‐adaptive with identical MCS/RB location but different redundancy versions. Different redundancy versions are employed to achieve incremental redundancy soft combining gain. Only one uplink grant and one HARQ feedback channel is transmitted for a bundle. HARQ retransmission
Table 11.23 VoIP packet size with RLC segmentation and overhead analysis. no segment Segment Size
2 segments Segment Size
L1 SDU
Voice Codec
[bits]
[bits]
Segment overhead
AMR‐NB 1.8
120
120
0%
AMR‐NB 4.75
176
176
0%
AMR‐NB 5.15
184
184
0%
AMR‐NB 5.9
200
200
AMR‐NB 6.7
216
AMR‐NB 7.4
4 segments Segment Size
L1 SDU [bits]
Segment overhead
80
160
33%
112
224
23%
112
224
0%
120
216
0%
232
232
AMR‐NB 7.95
240
AMR‐NB 10.2
8 segments
L1 SDU
Segment Size
L1 SDU
[bits]
[bits]
Segment overhead
[bits]
Segment overhead
64
256
100%
56
448
233%
80
320
68%
64
512
159%
22%
80
320
65%
64
512
152%
240
20%
80
320
60%
64
512
140%
128
256
19%
88
352
56%
64
512
130%
0%
136
272
17%
88
352
52%
64
512
121%
240
0%
144
288
17%
96
384
50%
72
576
117%
288
288
0%
168
336
14%
104
416
42%
72
576
97%
AMR‐NB 12.2
328
328
0%
184
368
12%
112
448
37%
80
640
85%
AMR‐WB 1.75
120
120
0%
80
160
33%
64
256
100%
56
448
233%
AMR‐WB 6.6
216
216
0%
128
256
19%
88
352
56%
64
512
130%
AMR‐WB 8.85
264
264
0%
152
304
15%
96
384
45%
72
576
106%
AMR‐WB 12.65
336
336
0%
192
384
12%
120
480
36%
80
640
83%
AMR‐WB 14.25
368
368
0%
208
416
11%
128
512
33%
88
704
76%
AMR‐WB 15.85
400
400
0%
224
448
10%
136
544
30%
88
704
70%
AMR‐WB 18.25
448
448
0%
248
496
9%
144
576
27%
96
768
63%
AMR‐WB 19.85
480
480
0%
264
528
8%
152
608
25%
96
768
58%
AMR‐WB 23.85
560
560
0%
304
608
7%
176
704
21%
112
896
50%
[bits]
[bits]
500
LTE Optimization Engineering Handbook
When a RLC SDU is divided into many segments, each transport block has its own RLC/MAC header.
No fragmentation, ROHC, same as TTI-B case (OH 23%)
MAC (1B)
2 fragments (OH 32%) 3 fragments (OH 37%) 4 fragments (OH 43%)
PDCP ROHC (1B) (4B)
RLC (1B)
AMR12.65 (34B)
CRC (3B)
41Bytes (328b) MAC (1B)
RLC (1B)
MAC (1B)
22Bytes (176b) RLC (13 B) (1B)
(20 B)
CRC (3B) CRC (3B)
15Bytes (120b) MAC (1B)
RLC (1B)
(10 B)
CRC (3B)
(5 B)
CRC (3B)
12Bytes (96b) MAC (1B)
8 fragments (OH 58%)
RLC (1B)
7Bytes (56b)
Number of Segments
Segment Size [bits]
Maximum Allowable Path Loss* [dB]
1 (no segmentation)
328
161.13
2
184
162.51
4
112
164.22
8
80
165.16
Figure 11.52 Example: AMR‐NB 12.2 (uplink).
40 35 30 PDCP PDU is segmented into small RLC PDUs due to poor RSRP, size from 6 byte to 40 byte
25 20 15 10 5 0
Figure 11.53 RLC PDU distribution.
of a TTI bundle is also transmitted as a bundle, occurs 16 TTIs after the previous transmission in order to be synchronized with normal (non‐bundled) HARQ retransmissions (8 TTIs). TTI bundling effectively utilizes more time resources by bundling a single VoIP packet into four consecutive TTIs. UL coverage is improved by allocating more energy per packet avoiding the excessive overhead from packet segmentation. Usually TTI bundling can provide gains up to 2 to 4 dB in uplink coverage. From Figure 11.54, we can see that TTI bundling needs reduced segmentation with less RLC and MAC overhead compared to normal operation, and also less control signaling, that is, HARQ feedback and PDCCH grants sent to UE. Figure 11.55 shows the measurements of RSRP, PUSCH BLER, and POLQA MoS without and with TTI bundling. The HARQ retransmissions ratio increases quickly after certain point in pathloss has been experienced. Once target BLER of ~10% cannot be maintained, MoS decreases quickly. Without TTI bundling, the VoLTE coverage RSRP threhold is around −114 dBm, with TTI Bundling is −117 dBm. Figure 11.56 shows the measurements of UL_SINR without and with TTI bundling. UL_SINR is more than 3 dB better when UE is TTI‐B on than off. The activation and de‐activation of TTI Bundling in the eNB is dynamic and is based on the channel quality. To prevent continuous (ping‐pong) activation and de‐activation of TTI bundling, hysteresis are available to avoid this behavior and unnecessary processor load increase in the eNB. TTI bundling is configured (is activated/deactivated) per UE via RRC control messaging as shown below.
VoLTE Optimization
Table 11.24 Coverage gain of AMR12.2 VoIP codec of regular transmission and segmentation. regular(1tx)
regular(4tx)
2 segments
3 segments
From PDCP
bits
304
304
304
304
No of segments
#
1
1
2
3
L2 overhead (RLC + MAC)
bits
16
16
16
16
Segment size
bits
320
320
168
120
No of HARQ transmissions
#
Modulation and coding scheme
1
4
4
4
MCS3
MCS3
MCS2
MCS0
No of PRBs
#
6
6
4
5
Actual transport block size
bits
328
328
176
120
Energy gain (frequency domain)
dB
% of VoIP per TTI
0
0
1.76
0.79
93%
93%
86%
84%
TX power per VoIP per TTI
dBm
22.67
22.67
22.36
22.27
Total energy per packet
dBm
22.67
22.67
25.37
27.04
Total energy per packet (HARQ)
dBm
22.67
28.69
31.39
33.06
Energy gain (time domain)
dB
0
6.02
8.72
10.39
Required SINR (10%/1Tr)
dB
−1.3
−1.3
−1.1
−3.6
Total delta compared to 10% BLER dimensioning
dB
0
6.0
10.3
13.5
Table 11.25 PDCCH resources occupation with regular transmission and segmentation. regular(1tx) regular(4tx) 2 segments 3 segments
PRB used per VoIP packet (cell edge)
6
24
32
60
PDCCH used per VoIP packet (cell edge)
1
4
8
12
PDCCH capacity used per VoIP packet (cell edge)
1.2%
4.8%
9.7%
14.5%
PDCCH capacity used per VoIP packet (non cell edge) 0.36%
0.36%
0.36%
0.36%
No. of supported VoIP users (5% users in cell edge)
247
171
121
93
No. of supported VoIP users (10% users in cell edge)
224
123
77
56
No. of supported VoIP users (40% users in cell edge)
142
46
24
17
value DL-DCCH-Message ::= { message c1 : rrcConnectionReconfiguration : { rrc-TransactionIdentifier 3, criticalExtensions c1 : rrcConnectionReconfiguration-r8 : { mac-MainConfig explicitValue : {ul-SCH-Config { maxHARQ-Tx n16, periodicBSR-Timer sf5, retxBSR-Timer sf320, ttiBundling TRUE },
501
502
LTE Optimization Engineering Handbook
Segmentation –i.e. one RLC SDU using 4 RLC PDUs -Four PDCCH allocations -Four HARQ feedbacks
Overhead
RLC SDU
RLC PDU TB
RLC PDU TB
RLC PDU
RLC SDU
RLC PDU
TB
RLC PDU
TB
Over head
One RLC SDU transmitted in one RLC PDU -One PDCCH allocation -One HARQ feedback
TB
time
time
Figure 11.54 Segmentation and TTI bundling.
When TTI bundling mode20 is activated, PUSCH single transport block is transmitted over four consecutive TTIs but with different redundancy version, only one UL grant is given for the transmission of the whole bundle, retransmission of a TTI bundle, which is also transmitted as a TTI bundle, occurs 16 TTIs after previous (re)transmission. The procedure of FDD and TDD UL transmission with TTI bundling are shown in the Figure 11.57 and Figure 11.58. TTI bundling gain is introduced by the energy collected from additional transport blocks received during the assumed service delay budget. Higher air link latency budget of VoIP services (typically 50 ms) allows for up to seven packet retransmissions (initial transmission + six retransmissions). From Figure 11.59, we can see that TTI bundling gain can be calculated according to: TTI bundling gain 10 * log 12 / 7
2.34 dB
In conclusion, in those poor radio conditions, UL resources over multiple consecutive TTIs can be assigned with a single grant, which decreases the signaling overhead. While TTI bundling is a very useful feature to increase coverage, it comes at the impact of capacity. VoIP capacity will be reduced when large number of VoIP users have TTI bundling enabled, since four consecutive subframes are used by these users. In a live network, we should limit the use of TTI bundling only when it have reached the limits of MCS/PRB override. TTI bundling activation requires a higher layer RRC reconfiguration to enable, and only a single specific format (MCS = 6, 1 PRB) can be used to take advantage of the 4 ms subframe bundling, thus will limit scheduler’s link adaptation flexibility. It should be known that for FDD LTE, HARQ round trip doubled from 8 ms to 16 ms, with TTI bundling, VoLTE data rate is (328 + 24 bits)/(4 HARQ Tx * 4 ms TTI) = 22 kbps. 11.6.4 TTI Bundling Optimization
Since TTI bundling will take four subframes of uplink, which impact the capacity greatly, for TTI bundling optimization, it’s reasonable and necessary to restrict the TTI bundling user number, based on the configured SINR and TTI bundling number. TTI bundling increases the probability of successful transport block decoding by eNB. The UE eligibility for TTI bundling is evaluated at context creation (i.e., call setup, handover, call re‐establishment) and updated during normal call operation (bearer creation or deletion or maybe bearer modification). In a situation when BLER increases and link adaptation has no
20
Feature group indicator processing required to determine if UE supports TTI bundling.
VoLTE coverage (above POLQA2.5) increased by 2.5dB if harqMaxTrUITti Bundling is 24. TTI Bundling OFF –100
TTI Bundling ON (retrans 24times)
+2.5dB gain
RSRP[dBm] Poly. (RSRP[dBm])
–108
RSRP : –114.5dBm
–112
RSRP : –117dBm
–112 –116 –120
–124
–124
TTIB ON
POLQA MOS Poly. (POLQA MOS)
4
3
3
POLQA MOS
POLQA MOS
5
4
Poly. (POLQA MOS)
POLQA : 2.5
2
2
1
POLQA : 2.5
1
0
0 PUSCH BLER[%]
100
PUSCH BLER[%]
100
PUSCH BLER[%] Poly. (PUSCH BLER[%])
80
PUSCH BLER[%]
80
60
60
40
40
Poly. (PUSCH BLER[%])
20
20
Figure 11.55 With TTI Bundling the RSRP level goes down to −117 dm, 2.5 dB gain achieved.
59:02.4
57:36.0
56:09.6
54:43.2
53:16.8
51:50.4
50:24.0
48:57.6
47:31.2
51:21.6
49:55.2
48:28.8
47:02.4
45:36.0
44:09.6
42:43.2
0 41:16.8
0
Poly. (RSRP[dBm])
–108
–116
POLQA MOS
RSRP[dBm]
–104
–120
5
RSRP[dBm]
–100
RSRP[dBm]
–104
LTE Optimization Engineering Handbook
Tti-Feature-ON TTI-Feature-OFF Poly. (Tti-Feature-ON) Poly. (TTI-Feature-OFF)
UL HARQ NACK Rate
1
2
3
4
5
RSRP
6
7
8
9
10 11 12 13 14 15 UL_SINR
PUSCH BLER
POLQA MOS
–96 –98 –100 –102 –104 –106 –108 –110 –112 –114 –116 –118 –120 –122 –124
80 70 60 50 40 30 20 10 0
PUSCH BLER [%], POLQA MOS * 10
–3 –2 –1 0
RSRP [dBm]
504
Figure 11.56 UL_SINR is more than 3 dB better when UE is TTI‐B on. TTI#
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
UL grant on 4 redundancy versions of the same packet PDCCH Tx on PUSCH RV0 RV2 RV3 RV1 Decoding ACK/NACK on PHICH
RV0 RV2 RV3 RV1
N
N Retransmission of bundle
1st bundle transmission HARQ RTT: 16 TTIs
Figure 11.57 FDD UL transmission with TTI bundling.
more options for MCS/PRB reduction while radio conditions for handover are not fulfilled, TTI bundling can be triggered to keep the voice call quality (in terms of delay and packet loss rate) before UE will either change the cell or RF conditions becomes better (Figure 11.60). The eNB periodically monitors every 100 ms the eligible UEs for TTI bundling activation/deactivation. Once the criteria for entering TTI bundling mode are fulfilled, eNB triggers intra‐cell handover procedure by sending RRC connection reconfiguration message toward the UE (Figure 11.61).
VoLTE Optimization TDD Configuration 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 D S U U D D S U U D D S U U D D S U U D D S U U D D S U U D D S U U D D S U U D D S U U D D S U U D D S U U D D S U U D R R V V 0 2
6
6
R R V V 3 1
8
R R V V 0 2
NACK
R R V V 3 1
NACK
UL grant
R R V V 0 2
R R V V 3 1
8
6
7 ms
Figure 11.58 TDD UL/DL configuration 1 with TTI bundling.
N
PHICH
Classical Decoding transmission 8ms RTT PUSCH 328 TTI 1 2 (7 TBSs)
3
4
5
N
6
7
N
N
PHICH
TTI Bundling Decoding 16ms RTT PUSCH 328 328 328 328 RV0 RV1 RV2 RV3 (12 TBSs) TTI 1 2 3 4 5
6
7
N
N
N
328 328 328 328 328 328 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
8
N
N
328 328 328 328 328 328 328 328 RV0 RV1 RV2 RV3 RV0 RV1 RV2 RV3 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
Figure 11.59 TTI bundling gain.
Channel Quality
TTI-B activation
LTE frequency2 layer
HO conditions fulfilled
Decision to switch UE to TTI Bundling LTE frequency1 layer Time
TTI Bundling area
Figure 11.60 TTI bundling triggered.
UE
eNB
eNB decides to configure UE with TTI bundling mode based on UL channel quality RRC: RRC Connection Reconfiguration (includes MobilityControllnfo)
DCCH/UL-SCH/PUSCH (SRB1)
PHY: CBRA Random Access Preamble √RACH/PRACH
MAC: CBRA Random Access Response √DL-SCH/PDSCH
MAC: Random Access Msg3 DCCH/UL-SCH/PUSCH (SRB1)
UL Grant PDCCH RRC: RRC Connection Reconfiguration Complete (sent in TTI Bundling mode) DCCH/UL-SCH/PUSCH (SRB1)
Figure 11.61 Dynamic switching to/from TTI bundling.
505
LTE Optimization Engineering Handbook
TTI bundling activation ccriterias include load criteria and poor RF performance conditions. For load criteria, current total GBR PRB utilization in the cell should be less than TTI bundling activation load threshold, and number of users in the cell already in TTI bundling configuration should be less than the maximum number of TTI bundling users. The consideration to disable the feature might be in heavily loaded eNBs due to high VoIP traffic load at cell edges. If the majority of this VoIP traffic experiences poor channel conditions it may trigger an excessive number of UEs using TTI bundling. This might have impact on the end‐user experience and performance. For poor RF performance criteria, the estimated achievable UL_SINR with 1PRB grant should be less than uplink link budget threshold for a period time, and the measured BLER at TTI bundling measurement points (HARQ Tx point) should bigger than TTI bundling activation BLER threshold within the IIR filter window. With TTI bundling ON, the PRB allocation size for the UE is restricted to NPRB d1(segmentation transition point for d = d1). From uplink link budget it should be possible to identify SINR threshold for transition to segmentation. When UE moves close to cell border, as the SINR value falls below a pre‐defined threshold where distance d > d2 (TTI bundling transition point for d = d2), TTI bundling will occur (Figure 11.64). In a live network, the fraction of VoIP users in different regions depend on the cell size and frequency, larger cells will have more users in segmentation and TTI bundling, higher frequency also leads to more users in segmentation and TTI bundling. Figure 11.64 gives a field test result of TTI bundling comparing with packet fragmentation (Figure 11.65). Although coverage gain can be expected with RLC segmentation and TTI bundling, the voice frame delay for UEs at the coverage limit for min MCS/PRB–based RLC segmentation and TTI bundling are increased. From test in a live network, the average increasement of ninety‐eighth RTP delay %‐tile is 30 ms and 50 ms for RLC segmentation and TTI bundling respectively. The results here are based on single user. In a loaded scenario, grant limitations may increase the delay for RLC segmentation. d1/dmax
dmax
Region with no segmentation and TTI bundling
d2
Region with segmentation only
Prx rb [dBm/RB]
–100
Region with TTI bundling d1
0%
20%
d2/dmax
d/dmax [%] 40%
60%
80%
100%
–105 –110 –115 –120 –125
dmax = cell range d1= point for transition to segmentation d2 = point for transition to TTI bundling
Figure 11.64 Cell regions for segmentation and TTI bundling transition point. 250 PDCCH cost of cell edge users
508
with TTI bundling without TTI bundling
4 segments no TTI bundling
200
150
2 segments no TTI bundling 4 segments TTI bundling
100
50
no segmentation no TTI bundling no segmentation TTI bundling
0 0.00
2.00
4.00
2 segments TTI bundling
6.00 Coverage Gain (dB)
8.00
Figure 11.65 TTI‐bundling comparing with packet fragmentation in a live network.
10.00
12.00
VoLTE Optimization
11.6.6 MCS/TBS/PRB Selection
The basic dynamic scheduler has the problem that it may excessively segment voice packets trying to maintain 10% iBLER as the path loss increases, leading to a build‐up of voice frame segments in the queue leading to excess delay and this becomes the factor that limits the link budget. Some amount segmentation is good for increasing subframe utilization, but too much segmentation can lead to excessive queuing delay since since each segments of a VoIP packet needs separate grant and uses a separate HARQ process, and the PDCCH consumption is higher. The semi‐persistent scheduler does not do any voice frame segmentation, instead the packet size is held constant and hence the iBLER will rise as the path loss increases, eventually leading to post‐HARQ errors, and this becomes the factor which limits the link budget. Based on RLC segmentation and TTI bundling, the usual TBS sizes for VoIP are suggested in Table 11.28 under dynamic and SPS scheduler. The PRB allocation size is derived from the smallest size that fits the initial TBS, rounded up to the RBG boundary. The TBS is obtained by adding the PDCP/RLC/MAC overhead. TBS VoIP codec VoIP frame size 4 bytes RoHC header
MAC , RLC overhead
Table 11.28 TBS index table for VoIP. NPRB ITBS
1
2
3
4
5
6
7
8
9
10
0
16
32
56
88
120
152
176
208
224
256
1
24
56
88
144
176
208
224
256
328
344
2
32
72
144
176
208
256
296
328
376
424
3
40
104
176
208
256
328
392
440
504
568
4
56
120
208
256
328
408
488
552
632
696
5
72
144
224
328
424
504
600
680
776
872
6
328
176
256
392
504
600
712
808
936
1032
7
104
224
328
472
584
712
840
968
1096
1224
8
120
256
392
536
680
808
968
1096
1256
1384
9
136
296
456
616
776
936
1096
1256
1416
1544
10
144
328
504
680
872
1032
1224
1384
1544
1736
11
176
376
584
776
1000
1192
1384
1608
1800
2024
12
208
440
680
904
1128
1352
1608
1800
2024
2280
13
224
488
744
1000
1256
1544
1800
2024
2280
2536
14
256
552
840
1128
1416
1736
1992
2280
2600
2856
15
280
600
904
1224
1544
1800
2152
2472
2728
3112
16
328
632
968
1288
1608
1928
2280
2600
2984
3240
17
336
696
1064
1416
1800
2152
2536
2856
3240
3624
18
376
776
1160
1544
1992
2344
2792
3112
3624
4008
19
408
840
1288
1736
2152
2600
2984
3496
3880
4264
20
440
904
1384
1864
2344
2792
3240
3752
4136
4584
509
510
LTE Optimization Engineering Handbook
The MCS is derived by comparing The UE reported SINR to the thresholds configured by MCS transition table. It is worth to note that the derived MCS is limited to the first 15 MCSs that are supported for SPS. Next step, it needs mapping to MCS/#PRBs with regards to the required TBS, each combination of MCS/#PRBs is correlated with SINR value required for proper signal reception and decoding by a receiver. Allocating more PRBs brings required SINR improvement due to the fact that lower MCS can be used, but UE power will be constrainted in most cases limit UL #PRBs. For optimization, MCS/#PRBs is one of the items which need to consider. Table 11.29 gives the required TBS, combination of MCS/#PRBs and SINR requirement in VoIP packet transmission. For VoLTE coverage, how to configure the upper bound the number of RLC segmentation that a voice packet can be broken into by limiting the smallest MCS and number of PRBs that can be selected by the scheduler, is one of the important work. After enforcing the minimum number of PRBs, MCS override approach would look like this: min MCS C Different values of (A, B, C) have been investigated via simulations to find the optimum combination in terms of maximum achievable uplink path loss for an AMR 12.65 kbps codec, which is shown in Figure 11.66. In Figure 11.66, the simulation result show optimized segmentation (three segments with 2 PRB grant → min MCS = 4, 3 PRB grant → min MCS = 2, 4 PRB grant → min MCS = 1) will yields 2.5 to 3 dB gain in terms of maximum acceptable uplink path loss for an AMR 12.65 kbps codec. When TTI bundling is also activated in the eNB, optimized segmentation may deferred the time when TTI bundling is needed to be configured.
2 PRB grant
min MCS A , 3 PRB grant
min MCS B, 4 PRB grant
11.6.7 Link Budget
AMR‐WB operates like AMR with nine different bit rates. The lowest bit rate providing excellent speech quality in a clean environment is 12.65 kbit/s. Higher bit rates are useful in background noise conditions and for music. Also lower bit rates of 6.60 and 8.85 kbit/s can provide reasonable quality especially if compared to narrow band codecs. The different voice transport formats (modulation, coding scheme, number of resource blocks, and transport block size) have different SINR requirements. When TTI bundling is active/disable, the related SINR requirement can be shown in Table 11.30. HARQ retransmissions will further lead to lower SINR requirements resulting in higher coverage (in cost of lowered capacity), as shown in Table 11.31. Based above assumption, calculating a link budget for VoIP is very similar to the method used for a data service. The UL link budget, maximum acceptable uplink path loss(Lpmax) is given by the following equation: Lpmax PUE , RB SeNB BIUL BLNF LBL LBPL Ga LJ where PUE,RB is the UE output power per RB [dBm], SeNB is the eNB sensitivity [dBm], BIUL is the interference margin (“noise rise”) [dB], BLNF is the log‐normal fading margin [dB], LBL is the body loss [dB], LBPL is the building penetration loss [dB], Ga is the antenna gain [dB], LJ is the jumper loss [dB]
Table 11.29 SINR requirement in VoIP packet transmission. 1
NPRB TBS
2 req.
3 req.
6 req.
9 req.
12 req.
15 req.
… req.
MCS Index
Mod. Order
TBS
SINR
TBS
SINR
0
2
0
16
–0,5
32
–2,2
56
–2,7
152
–2,7
224
–3,2
328
–3,1
392
–3,3
…
1
2
1
24
0,1
56
–1,2
88
–1,7
208
–1,9
328
–2,1
424
–2,3
520
–2,5
…
Index
TBS
SINR
TBS
SINR
TBS
SINR
TBS
SINR
TBS
SINR
…
2
2
2
32
0,6
72
–0,6
144
–0,4
256
–1,4
376
–1,7
520
–1,8
648
–1,8
…
3
2
3
40
1,0
104
0,4
176
0,2
328
–0,3
504
–0,6
680
–0,7
872
–0,6
…
4
2
4
56
1,7
120
0,8
208
0,8
408
0,3
632
0,2
840
0,1
1064
0,1
…
5
2
5
72
2,3
144
1,3
224
0,9
504
1,0
776
0,8
1032
0,7
1320
0,7
…
6
2
6
328
#N/A
176
2,0
256
1,4
600
1,6
936
1,6
1224
1,4
1544
1,4
…
7
2
7
104
3,7
224
3,1
328
2,5
712
2,5
1096
2,5
1480
2,5
1800
2,3
…
8
2
8
120
4,4
256
3,8
392
3,5
808
3,3
1256
3,3
1672
3,2
2088
3,2
…
9
2
9
136
5,0
296
4,6
456
4,4
936
4,3
1416
4,2
1864
4,0
2344
4,0
…
10
2
10
144
5,4
328
5,4
504
5,1
1032
4,8
1544
4,7
2088
4,7
2664
4,8
…
11
4
10
144
6,2
328
5,9
504
5,7
1032
5,6
1544
5,4
2088
5,4
2664
5,5
…
12
4
11
176
7,0
376
6,5
584
6,3
1192
6,3
1800
6,1
2408
6,1
2984
5,8
…
13
4
12
208
8,0
440
7,5
680
7,4
1352
7,1
2024
6,9
2728
6,9
3368
6,7
…
14
4
13
224
8,5
488
8,3
744
8,0
1544
8,1
2280
7,7
3112
7,8
3880
7,7
…
15
4
14
256
9,4
552
9,2
840
8,9
1736
8,9
2600
8,7
3496
8,7
4264
8,5
…
16
4
15
280
10,1
600
9,9
904
9,5
1800
9,3
2728
9,2
3624
9,1
4584
9,2
…
17
4
16
328
11,4
632
10,4
968
10,1
1928
9,7
2984
10,0
3880
9,6
4968
9,8
…
18
4
17
336
11,6
696
11,3
1064
11,1
2152
10,9
3240
10,8
4392
10,9
5352
10,6
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
512
LTE Optimization Engineering Handbook
130.5 130 129.5 129 128.5 128 127.5 127
(10,7,5)
(6,3,2)
(4,2,1)
(3,2,1)
1 segment
< 2 segments
< 3 segments
< 4 segments
Figure 11.66 Different values of MCS override with the minimum number of PRBs allocation. Table 11.30 SINR requirements for voice codec (TTI bundling is active/disable). TBS
MCS TBS Modulation Veh‐A, 3 km/hr Veh‐A, 50 km/hr Veh‐A, 120 km/hr Ped‐A, 3 km/h
AMR 12.2, 328
6
6
QPSK
–5.6/1.4
–4.8/2.7
–3.1/3.2
–5.6/0.4
AMR 4.5, 144
10
10
QPSK
–8.0/–2.9
–7.4/–1.9
–5.8/–1.3
–7.9/–3.7
AMR 5.9, 224
14
13
16QAM
–1.5/1.7
–1.1/2.9
0.73/3.4
–1.5/0.6
AMR 7.95, 256 15
14
16QAM
–0.9/3.0
–0.06/3.8
1.5/4.4
–0.9/1.2
AMR 12.2, 328 17
16
16QAM
–0.5/0.9
0.3/1.5
1.9/2.0
–0.5/–1.1
Table 11.31 HARQ gain. Number of transmissions (2% BLER target)
Required UL SINR (MCS7/3 PRBs) [dB]
HARQ gain [dB]
1 (initial tx., 10% BLER)
2.54
–
2
0.19
2.35
3
–2.05
4.59
4
–3.62
6.16
For VoLTE coverage, downlink is not the limitation, especially for voice service. Voice with enhancements such as TTI bundling or RLC segmentation could outperform 2G voice, but TTI bundling and RLC segmentation are done at the cost of the number simultaneous users. Voice coverage can be also improved by reducing AMR codec rate. Table 11.32 gives the VoLTE link budget under TDD 20 MHz system bandwidth, DL:UL(2:2), ETU3 channel model, and 50% cell loading. In addition, it is known that IMS signaling coverage is not limited for VoLTE. The SIP message size has about 1000byte, according to its QoS requirements, the SIP message should be transmitted to IMS core in 80 ms. Assuming the edge of the average data rate is 100 kbps, the SIP coverage radius can be estimated by above link budget method, and the conclusion is IMS signaling coverage is not limited by using dynamic scheduling. For uplink control channel link budget estimation, the following comparision of VoLTE PUSCH and CCH is described. PUSCH, wideband SRS, RACH message 1 (including Format 0, 2, 3), and message 3 (including 56 bits and 104 bits length) are compared under dense urban, urban, suburban and rural environment. Assuming the maximum of HARQ of message 3 is 3 and 5 accordingly, we can see in Table 11.33: Wideband SRS is not uplink coverage limitation in VoLTE coverage. Comparing to the VoLTE coverage with TTI bundling, message 3 (104 bits) with Max. HARQ = 3 is uplink coverage
VoLTE Optimization
Table 11.32 VoLTE link budget (TDD 20 MHz). Uplink
AMR12.2 k
AMR23.85 k
Antenna configuration
1Tx*8Rx
1Tx*8Rx
1Tx*8Rx
1Tx*8Rx
Modulation
QPSK
QPSK
QPSK
QPSK
Code rate
0.16
User data rate(kbps) # of Occupied RBs
0.2
PUSCH
0.31
0.31
104
164
256
512
3
3
8
16
UE EIRP MS power (dBm)
23
23
23
23
MS EIRP (dBm)
18.2
18.2
14.0
11.0
16.5
16.5
16.5
16.5
eNB sensitivity BS antenna gain (dBi) BS diversity gain (dB)
0
0
0
0
Transmission line loss (dB)
0.50
0.50
0.50
0.50
Thermal noise (kT) (dBm/Hz)
–174.0
–174.0
–174.0
–174.0
15
15
15
15
BS noise power (dB)
–117.9
–117.9
–117.9
–117.9
BS noise figure (dB)
3.5
3.5
3.5
3.5
–5.3
–4.3 –7.8
–7.8
Subcarrier separation (KHz)
SNR for MCS Level ‐ 1%BLER (dB) SNR for MCS Level ‐ 10%BLER(dB) BS sensitivity ‐ composite Shadow fading margin (dB) Interference margin (dB) Penetration loss (dB)
–123.2
–122.2
–125.7
–125.7
8.3
8.3
8.3
8.3
4.0 20
4.0 20
4.0 20
4.0 20
MS body loss (dB)
3
3
0
0
Terminal loss (dB)
6
6
6
6
Total system margin (dB) Outdoor maximum allowable path loss
41.3
41.3
38.3
38.3
136.6
135.6
137.9
134.9
limitation (assuming Max. HARQ = 3 = 5, message 3 is not the coverage bottleneck with c urrent configuration). TTI bundling can provide about 4 dB improvement in VoLTE link budget compared to no TTI bundling.
11.7 VoLTE Delay Delay is crucial for VoIP, therefore delay target must be controlled by the scheduler. It is done by a special delay term DT(t) added to criterion calculation in time domain scheduling (this step in the list of UEs that will be scheduled is now created). VoIP UEs are scheduled before non‐GBR UEs and within the pool of UEs with GBR bearers, and the selection of users for scheduling in given TTI is based on the following formula:
C t
PF t * DT t
513
514
LTE Optimization Engineering Handbook
Table 11.33 Comparision of VoLTE PUSCH and UL control channel. Coverage comparison
RACH
Msg1
Msg3 (HARQ = 3)
DU
U
SU
RU
Format0
1.4 km
2.3 km
4.5 km
16.8 km
Format2
1.7 km
2.7 km
5.2 km
19.9 km
Format3
1.7 km
2.7 km
5.3 km
20.0 km
56 bits
1.1 km
1.7 km
3.6 km
13.6 km Limitation
104 bits
1.0 km
1.5 km
3.2 km
12.2 km
Msg3 (HARQ = 5)
56 bits
1.3 km
2.0 km
4.3 km
16.4 km
104 bits
1.1 km
1.8 km
3.9 km
14.7 km
WB‐SRS
WB‐SRS
48 RBs
1.7 km
2.8 km
6.4 km
24.3 km
VoLTE WB‐AMR 12.65
PUSCH
SPS and No TTI‐B
0.7 km
1.1 km
2.5 km
7.7 km
Dynamic scheduling
0.9 km
1.4 km
3.0 km
10.2 km
max 4 segments
1.0 km
1.6 km
3.6 km
13.4 km
TTI‐B
1.2 km
1.9 km
4.2 km
14.3 km
where PF(t) is proportional fair term, DT(t) is delay term, the delay term is based on VoIP packet delay, that is, the longer the packet stays in the buffer the higher DT(t) value is. It is worth a reminder that in Chapter 11.2.2, the packet delay budget (PDB) requires no more than 80 ms for voice, and no more than 130 ms for video. PDB is measured from entering PDCP in eNB to leaving PDCP in the UE and vice versa. PDB will be increased as more UEs are in a cell due to PF(t) and DT(t) (see Figure 11.67), the 98%‐ile per‐UE VoIP packet delay; however, At 170 calls, > 95% of the calls have delay less than 50 ms, but at 180 calls, the delay is larger than 50 ms for ~40% of the calls
95% 0.9 0.85 0.8 0.75 0.7 0.65 0.6 0.55
170 VoIP call's, delay target 50 ms 180 VoIP call's, delay target 50 ms 50 ms target
50%
0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05
0% 40
200 VoIP call's, delay target 80 ms 80 ms target
210 VoIP call's, delay target 80 ms
At 200 calls > 97% of calls have delay less than 80 ms & at 210 calls, the delay is larger than 80 ms for ~5% of the calls 80 ms
50 ms
50
60
Figure 11.67 PDB versus No. of UEs.
70
80
98 percentile delay (sec)
VoLTE Optimization
capacity with a 50‐ms delay target is ~170 VoIP calls/sector and capacity with 80 ms delay target is ~210 VoIP calls/sector, thus we can see that from ~30 ms longer VoIP delay, 1.23 × (210/170) capacity improvement will be achieved. An example shown in Figure 11.68, it is possible to aggregate VoIP packets into larger transport block and send them all at once, for example, only one UL grant is needed to send three aggregated VoIP packets, and thus will cause long packet delay. For a typical VoLTE to VoLTE call flow, the key focused MO (mobile originated) metrics includes ring back delay, call setup time, and conversation delay, and so on, and the key MT (mobile terminated) metrics includes paging delay and media delay, and so on, as shown in Figure 11.69.
VoIP frame
20ms
SR VoIP frame
20ms
1. With packet aggregation these packets are delayed. They will be send when proper aggregation level is achieved
UL Grant
UL Transm.
SR
UL Grant
UL Transm.
UL Grant
Scheduling Request
VoIP frame
UL Transm.
4. TB in this UL transmission contains 3 VoIP packets.
2. Without aggregation packet should be send here but scheduler doesn’t provide UL grant
3. 3 packets are aggregated. Once the SR is send after 3rd packet arrival, UL grant is provided.
Figure 11.68 Packet aggregation causes longer delay.
MO UE
EUTRAN/IMS
MT UE
SIP: INVITE (SDP Offer): UE_TO_NETWORK RACH + RRC Connection procedure + Default: Bearer Setup SIP: 100 TRYING
Ring back delay
Paging delay
Paging + RACH + RRC Connection procedure + Default: Bearer Setup SIP: INVITE (SDP Offer): NETWORK_TO_UE SIP: 100 TRYING
SIP: 180 Ringing SIP: 180 Ringing SIP: 180 Ringing SIP: PRACK SIP: PRACK
Alert User
Call setup time
SIP: 200 OK (PRACK) SIP: 200 OK (PRACK)
Answer SIP: 200 OK (INVITE)
Dedicated Bearer Setup
Dedicated Bearer Setup
Media delay
SIP: 200 OK (Invite) SIP: ACK: UE_TO_NETWORK SIP: ACK: NETWORK_TO_UE
Conversation delay Disconnect delay
Voice Call In-Progres (RTP Media)
SIP: BYE: (UE_TO_NETWORK)
Figure 11.69 Typical VoLTE delay.
SIP: BYE: (NETWORK_TO_UE) SIP: 200 OK (BYE) SIP: 200 OK (BYE)
515
516
LTE Optimization Engineering Handbook
11.7.1 Call Setup Delay 11.7.1.1 Call Setup Time
Call setup time is the time from the VoLTE user initiates the call until the user receives notification about the called party being alerted. This call type, VoLTE to VoLTE, and call setup time is measured entirely from the UE, which originates the call. The typical call setup time is 3 to 5 seconds. For originating UE, SIP invite to a 180‐ring message is around 4.7 s. For terminated UE, paging message to 200 OK messages is 390 ms, as shown in Figure 11.70. Note: Network logs are required to check delays between nodes to spot precondition and early media delay. Sometimes the high value of operator’s call setup time is caused by the additional functions: “precondition, early media.” Precondition needs additional ~600 ms delay, while early media needs additional ~700 ms delay. The call setup time are specified for preconditions are used or early media is used. Preconditions ensure resources are allocated prior to call setup completion and early media is network ring back tone enabled as announcements prior to call start. In a live network, engineers usually modify call setup time calculation formula with appropriate end trigger depending on call case. The following list states the possible combinations of all call cases and corresponding end triggers. Note that start trigger is always SIP Invite for all cases. ●● ●● ●● ●●
No preconditions and no early media – end trigger is 180 rings Preconditions and no early media – end trigger is 180 rings No preconditions and early media – end trigger is 200 OK (PRACK) Preconditions and early media – end trigger is 200 OK (UPDATE)
Assuming media negotiation should be done during call setup between two UEs to use the same types of voice media, but both initial negotiation bandwidth was inconsistent. At that time, SBC open the proactive transcoding function to ensure that the subsequent media negotiations have enough bandwidth to increase the bandwidth of the SDP. After the completion of the subsequent media negotiations, which are then sent back to the original bandwidth results (see Figure 11.71): first reserved 49 kbps bandwidth and subsequent modified the reserved bandwidth to 38 kbps. Media negotiation is necessary during call setup, but it increased the interface of Rx, Gx, and EPS unnecessary overhead. The call setup time is very much impacted by the activity state in the two UEs involved in the test call. The activity state referred to is the ECM state (EPS connectivity management). The ECM state has two states: ECM‐idle and ECM‐connected. The call setup time 95th percentile target is: idle UE 3
2
Speech path delay
Drive test