WILEY Backhauling Fronthauling for Future Wireless Systems User Manual

Download

BACKHAULING/ FRONTHAULING FOR FUTURE WIRELESS SYSTEMS

www.ebook3000.com

BACKHAULING/ FRONTHAULING FOR FUTURE WIRELESS SYSTEMS

Edited by

Kazi Mohammed Saidul Huq and Jonathan Rodriguez

Instituto de Telecomunicações, Aveiro, Portugal

www.ebook3000.com

Registered Office

John Wiley & Sons, Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, United Kingdom

For details of our global editorial offices, for customer services and for information about how to apply for permission to reuse the copyright material in this book please see our website at www.wiley.com.

The right of the author to be identified as the author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988.

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 the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher.

Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books.

Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners. The publisher is not associated with any product or vendor mentioned in this book.

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. It is sold on the understanding that the publisher is not engaged in rendering professional services and neither the publisher nor the author shall be liable for damages arising herefrom. If professional advice or other expert assistance is required, the services of a competent professional should be sought

Library of Congress Cataloging‐in‐Publication Data

Names: Huq, Kazi Mohammed Saidul, editor. | Rodriguez, Jonathan, editor. Title: Backhauling/fronthauling for future wireless systems / edited by Kazi Mohammed Saidul Huq, Jonathan Rodriguez. Description: Chichester, UK ; Hoboken, NJ : John Wiley & Sons, 2017. | Includes bibliographical references and index. Identifiers: LCCN 2016026831 (print) | LCCN 2016042959 (ebook) | ISBN 9781119170341 (cloth) | ISBN 9781119170358 (pdf) | ISBN 9781119170365 (epub) Subjects: LCSH: Wireless communication systems. Classification: LCC TK5103.2 .B33 2017 (print) | LCC TK5103.2 (ebook) | DDC 384.5–dc23 LC record available at https://lccn.loc.gov/2016026831

A catalogue record for this book is available from the British Library.

Cover image: Gettyimages/Petrovich9

Set in 11/13pt Times by SPi Global, Pondicherry, India

10 9 8 7 6 5 4 3 2 1

www.ebook3000.com

Contents

List of Contributors ix Preface xi Acknowledgements xiii

1 Introduction: The Communication Haul Challenge 1

Kazi Mohammed Saidul Huq and Jonathan Rodriguez

1.1 Introduction 1 References 7

2 A C‐RAN Approach for5G Applications 9

Kazi Mohammed Saidul Huq, Shahid Mumtaz and Jonathan Rodriguez

2.1 Introduction 9

2.2 From Wired toWireless Backhaul/Fronthaul Technologies 11

2.3 Architecture forCoordinated Systems According toBaseline 3GPP 12

2.4 Reference Architecture forC‐RAN 15

2.4.1 System Architecture forFronthaul‐based C‐RAN 15

2.4.2 Cloud Resource Optimizer 16

2.5 Potential Applications forC‐RAN‐based Mobile Systems 20

2.5.1 Virtualization ofD2D Services 20

2.5.2 Numerical Analysis 21

2.6 Conclusion 24

References 27

3 Backhauling 5G Small Cells withMassive‐MIMO‐Enabled

mmWave Communication 29

Ummy Habiba, Hina Tabassum and Ekram Hossain

3.1 Introduction 29

3.2 Existing Wireless Backhauling Solutions for5G Small Cells 31

3.3 Fundamentals ofmmWave andMassive MIMO Technologies 32

3.3.1 MmWave Communication 32

3.3.2 MU‐MIMO withLarge Antenna Arrays 33

www.ebook3000.com

vi Contents

3.4 MmWave Backhauling: State oftheArt andResearch Issues 34

3.4.1 LOS mmWave Backhauling 35

3.4.2 NLOS mmWave Backhauling 36

3.4.3 Research Challenges forBackhauling in5G Networks 37

3.5 Case Study: Massive‐MIMO‐based mmWave BackhaulingSystem 40

3.5.1 System Model 41

3.5.2 Maximizing User Rate 44

3.5.3 Matching Theory forUser Association 45

3.5.4 Numerical Results 48

3.6 Conclusion 51 Acknowledgement 51 References 51

4 Fronthaul foraFlexible Centralization inCloud Radio Access Networks 55

Jens Bartelt,

Dirk Wübben, Peter Rost, Johannes Lessmann andGerhardFettweis

4.1 Introduction 55

4.2 Radio Access Network Architecture 57

4.3 Functional Split Options 58

4.4 Requirements ofFlexible Functional Splits 60

4.4.1 Split A 61

4.4.2 Split B 62

4.4.3 Split C 63

4.4.4 Split D 64

4.4.5 Summary andExamples 64

4.5 Statistical Multiplexing inaFlexibly Centralized Network 67

4.5.1 Distribution ofFH Data Rate per Base Station 67

4.5.2 Outage Rate 68

4.5.3 Statistical Multiplexing onAggregation Links 69

4.6 Convergence ofFronthaul andBackhaul Technologies 73

4.6.1 Physical Layer Technologies 73

4.6.2 Data/MAC Layer Technologies 75

4.6.3 Network Layer Technologies 77

4.6.4 Control andManagement Plane 78

4.7 Enablers ofaFlexible Functional Split 78

4.8 Summary 80 Acknowledgement 82 References 82

5 Analysis andOptimization forHeterogeneous Backhaul Technologies 85

Gongzheng Zhang, Tony Q. S. Quek, Marios Kountouris, Aiping Huang andHangguan Shan

5.1 Introduction 85

5.2 Backhaul Model 88

5.2.1 Network Model 88

www.ebook3000.com

Contents vii

5.2.2 Delay Model 89

5.2.3 Cost Model 92

5.3 Backhaul Packet Delay Analysis 93

5.3.1 Mean Backhaul Packet Delay 93

5.3.2 Delay‐limited Success Probability 95

5.3.3 Performance Evaluation 97

5.4 Backhaul Deployment Cost Analysis 101

5.5 Backhaul‐aware BS Association Policy 103

5.5.1 Mean Network Packet Delay 103

5.5.2 BS Association Policy 107

5.5.3 Numerical Results 109

5.6 Conclusions 115

References 115

6 Dynamic Enhanced Inter‐cell Interference Coordination

StrategywithQuality ofService Guarantees for Heterogeneous Networks 119

Wei‐Sheng Lai, Tsung‐Hui Chang, Kuan‐Hsuan Yeh and Ta‐Sung Lee

6.1 Introduction 119

6.2 System Model andProblem Statement 121

6.2.1 Network Environments 121

6.2.2 QoS Constraint 124

6.2.3 Problem Statements 125

6.3 Dynamic Interference Coordination Strategy 126

6.3.1 SMDP Analysis 126

6.3.2 Admission Control withaQoS Constraint 128

6.3.3 Joint Dynamic eICIC andAdmission Control for

SumRateMaximization 129

6.3.4 Joint Dynamic eICIC andAdmission Control for

ProportionalFairness Maximization 130

6.4 Numerical Results 132

6.5 Conclusion 140

References 140

7 Cell Selection forJoint Optimization oftheRadio Access

and Backhaul inHeterogeneous Cellular Networks 143

Antonio De Domenico, Valentin Savin and Dimitri Ktenas

7.1 Introduction 143

7.2 System Model andProblem Statement 145

7.2.1 Joint RAN/BH Capacity 146

7.2.2 Problem Statement 151

7.3 Proposed Solutions 151

7.3.1 Evolve 151

7.3.2 Relax 154

7.3.3 Practical Implementation oftheProposed Algorithms 156

www.ebook3000.com

viii Contents

7.4 Simulation Results 157

7.5 Conclusion 165 References 165

8 Multiband andMultichannel Aggregation forHigh‐speed Wireless

Backhaul: Challenges andSolutions 167

Xiaojing Huang

8.1 Introduction 167

8.2 Spectrum forWireless Backhaul 170

8.2.1 Microwave Band andChannel Allocation 170

8.2.2 Millimetre‐wave Band andUsage Trend 171

8.3 Multiband andMultichannel Aggregation 172

8.3.1 Band andChannel Aggregation Overview 172

8.3.2 System Architecture 174

8.3.3 Subband Aggregation andImplementations 177

8.3.4 Full SDR Approach forBand andChannel Aggregation 183

8.4 Spectrally Efficient Channel Aggregation 185

8.4.1 System Overview 185

8.4.2 Frequency‐domain Multiplexing Without aGuard Band 186

8.4.3 Digital IF Signal Generation andReception 188

8.4.4 High-performance OFDM Transmission 188

8.5 Practical System Examples 189

8.5.1 CSIRO Ngara Backhaul 190

8.5.2 CSIRO High‐speed E‐band Systems 191

8.6 Conclusions 194 References 194

9 Security Challenges forCloud Radio Access Networks 195

Victor Sucasas, Georgios Mantas and Jonathan Rodriguez

9.1 Introduction 195

9.2 Overview ofC‐RAN Architecture 196

9.3 Intrusion Attacks intheC‐RAN Environment 197

9.3.1 Entry Points forIntrusion Attacks 198

9.3.2 Technical Challenges forIntrusion Detection Counter‐mechanisms 201

9.3.3 Insider Attacks 203

9.4 Distributed Denial ofService (DDoS) Attacks Against C‐RAN 205

9.4.1 DDoS Attacks Using Signalling Amplification 206

9.4.2 DDoS Attacks Against External Entities Over the Mobile Network 207

9.4.3 DDoS Attacks fromExternal Compromised IP Networks OvertheMobile Network 208

9.5 Conclusions 209 References 209

Index 213

www.ebook3000.com

List of Contributors

Jens Bartelt

Technische Universität Dresden, Vodafone Chair MNS, Dresden, Germany

Tsung-Hui Chang

School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen,CUHK (SZ), China

Antonio De Domenico

CEA, LETI, MINATEC, Grenoble, France

Gerhard Fettweis

Technische Universität Dresden, Vodafone Chair MNS, Dresden, Germany

Ummy Habiba

The Department of Electrical and Computer Engineering, University of Manitoba, Canada

Ekram Hossain

The Department of Electrical and Computer Engineering, University of Manitoba, Canada

Aiping Huang

College of Information Science and Electronic Engineering, Zhejiang University, China

Xiaojing Huang

Faculty of Engineering and Information Technology, University of Technology Sydney (UTS), Australia

Kazi Mohammed Saidul Huq

Instituto de Telecomunicações, Aveiro, Portugal

Marios Kountouris

Mathematical and Algorithmic Sciences Lab, France Research Centre, Huawei Technologies, France

Dimitri Ktenas

CEA, LETI, MINATEC, Grenoble, France

www.ebook3000.com

x List of Contributors

Wei-Sheng Lai

Department of Electrical and Computer Engineering, National Chiao Tung University, Hsinchu, Taiwan

Ta-Sung Lee

Department of Electrical and Computer Engineering, National Chiao Tung University, Hsinchu, Taiwan

Johannes Lessmann

NEC Laboratories Europe, Heidelberg, Germany

Georgios Mantas

Instituto de Telecomunicações, Aveiro, Portugal

Shahid Mumtaz

Instituto de Telecomunicações, Aveiro, Portugal

Tony Q. S. Quek

Information Systems Technology and Design Pillar, Singapore University of Technology and Design, Singapore

Jonathan Rodriguez

Instituto de Telecomunicações, Aveiro, Portugal

Peter Rost

Nokia Networks, Munich, Germany

Valentin Savin

CEA, LETI, MINATEC, Grenoble, France

Hangguan Shan

College of Information Science and Electronic Engineering, Zhejiang University, China

Victor Sucasas

Instituto de Telecomunicações, Aveiro, Portugal

Hina Tabassum

The Department of Electrical and Computer Engineering, University of Manitoba, Canada

Dirk Wübben

University of Bremen, Department of Communications Engineering, Bremen, Germany

Kuan-Hsuan Yeh

ASUSTeK Computer Inc., Taipei, Taiwan

Gongzheng Zhang

College of Information Science and Electronic Engineering, Zhejiang University, China

www.ebook3000.com

Preface

In a mobile communication system, the segment that connects the core to the access networks is termed the ‘backhaul’. The edges of any telecommunication network are connected through backhauling. The importance of backhaul research is spurred by the need for increasing data capacity and coverage to cater for the ever‐growing population of electronic devices–smartphones, tablets and laptops–which is foreseen to hit unprecedented levels by 2020. The backhaul is anticipated to play a critical role in handling large volumes of traffic, its handling capability driven by stringent demands from both mobile broadband and the introduction of heterogeneous networks (HetNets). Backhaul technology has been extensively investigated for legacy mobile systems, but is still a topic that will dominate the research arena for next generation mobile systems; it is clear that without proper backhauling, the benefits introduced by any new radio access network technologies and protocols would be overshadowed.

Traditionally, the backhaul segment connects the RAN (radio access network) to the rest of the network where the baseband processing takes place at the cell site. However, with the onset of next generation networks, the notion of ‘fronthaul access’ is also gaining momentum. The future technology roadmap points towards SDN (software‐defined networks) and network virtualization as means of effectively sharing resources on demand between different mobile operators, thus taking a step towards reducing the operational and capital expenditure in future networks. Moreover, the baseband processing will be centralized, allowing the operators tocompletely manage interference through coordinated resource‐management strategies. In fact, 3GPP are today visualizing a C‐RAN (cloud-RAN) architecture, where the evolved base stations are connected to the C‐RAN unit through communication hauls, to what is referred to as the ‘fronthaul network’. Traditionally, fibre‐optic technology is used to roll out the deployment of base stations; however, this comes along with inherent limitations, including cost and lack of availability at many small sites. This provides the impetus for radio solutions that can handle large volumes of traffic on the fronthaul access, triggering the research community at large to find alternative and advanced solutions that can supersede fibre.

The current work on backhaul and fronthaul technology is fragmented, and still in its infancy. There are still giant steps to be taken towards developing concrete

www.ebook3000.com

xii Preface

solutions to provide a modern communication haul for next generation networks, which is also commonly referred to as 5G. This book aims to be the first of its kind to hinge together the related discussions on the fronthaul and backhaul access under the umbrella of 5G networks, which we will often refer to as the ‘communication haul’. We aim to discuss these pivotal building blocks of the communication infrastructure and provide a view of where it all started, where we are now in terms of LTE/LTE‐A networking and the future challenges that lie ahead for 5G. In addition, this book presents a comprehensive analysis of different types of backhaul/fronthaul technologies while introducing innovative protocol architectures.

In the compilation of this book, the editors have drawn on their vast experience in international research and being at the forefront of the communication haul research arena and standardization. This book aims to be the first to talk openly about next generation communication hauls, and will hopefully serve as a useful reference not only for postgraduate students to learn more about this evolving field, but also to stimulate mobile communication researchers towards taking further innovative strides in this field and marking their legacy in the 5G arena.

Kazi Mohammed Saidul Huq

Jonathan Rodríguez

Instituto de Telecomunicações, Aveiro, Portugal

Acknowledgements

This book is the first of its kind tackling the research challenge on the communication haul for legacy and emerging mobile communication networks, and the authors hope that it will serve as a source of inspiration for researchers to drive new breakthroughs on this topic. The inspiration for this book stems from the editors’ vast experience at the forefront of European research on backhaul/fronthaul architecture for future wireless systems, including the E-COOP project (UID/EEA/50008/2013), an interdisciplinary research initiative funded by the Instituto de Telecomunicações (Portugal). However, this work would not be complete if it weren’t for those who contributed along the way. The editors would first like to thank all the collaborators that have contributed with chapters toward the compilation of this book, providing complementary ideas towards building a complete vision of the communication haul. Moreover, a heartfelt acknowledgement is due to the members of the 4TELL Research Group at the Instituto de Telecomunicações who contributed with useful suggestions and revisions. Furthermore, the editors would like to acknowledge the Fundação para a Ciência e a Tecnologia (FCT‐ Portugal) for the grant (reference number: SFRH/BPD/110104/2015) that supported this work.

Kazi Mohammed Saidul Huq

Jonathan Rodríguez

Instituto de Telecomunicações, Aveiro, Portugal

Introduction: The Communication Haul Challenge

Kazi Mohammed Saidul Huq and Jonathan Rodriguez

Instituto de Telecomunicações, Aveiro, Portugal

1.1 Introduction

Nowadays, the mobile Internet is a pervasive phenomenon that is changing social trends and playing a pivotal role in creating a digital economy. This, in part, is driven by advancements in semiconductor technology, which are enabling faster and more energy‐compliant devices, such as smartphones, tablets and sensor devices, among others. However, a truly smart digital world is still in its infancy and the current trends are set to continue, leading to an unprecedented rise in mobile data traffic and intelligent devices. In fact, according to an Ericsson report [1], a typical laptop will generate 11 GB, a tablet 3.1 GB and a smartphone 2 GB per month by the end of 2018. These figures represent the changing communication paradigm, where the end user will not only receive data but generate data; in other words, the end user will become a ‘prosumer’ running data‐hungry applications, for example, high‐definition wireless video streaming, machine‐to‐machine communication, health‐monitoring applications and social networking. Therefore, existing technology requires a radical engineering design upgrade in order to compete with ever‐growing user expectations and to accommodate the foreseen increase in traffic. The change will be driven by market expectations, and the new technology being considered is fifth generation (5G) communications [2].

Experts anticipate that 5G will deliver and meet the expectations of a new era in

wireless connectivity, and will play a key role in enabling this so‐called digital world.

2 Backhauling/Fronthauling for Future Wireless Systems

In contrast to legacy fourth generation (4G) systems, the widely accepted consensus on the 5G requirement includes [3, 4]:

• Capacity: 1000x increase in area capacity;

• Latency: Less than 1 millisecond (ms) round trip time (RTT) latency;

• Energy: 100x improvement in energy efficiency in terms of Joules/bit;

• Cost: 10–100x reduction in cost of deployment;

• Mobility: Mobility support and always‐on connectivity of users that have high throughput requirements.

To achieve these targets, all the key mobile stakeholders, such as operators, vendors and the mobile research community, are contriving to reengineer the mobile architecture in order to support higher‐speed data connectivity.

Small‐cell technology is an emerging deployment that is providing promising results in terms of delivering fast connectivity due to the small distance between the base station (BS) and the end user, whilst reducing energy consumption. Market use cases of small cells such as the indoor femto cell have already become a success story, so the question is, can we extrapolate the femto cell paradigm to the outdoor world? In fact, current trends are suggesting that this is the way forward, with multi‐tier heterogeneous networks being a new design addition to the LTE‐Advanced standard [5,6]. Here, multi‐tier radio networks (small‐cell tiers) play a pivotal role, coupled with network coexistence approaches to reduce the interference between tiers. Moreover, mobile technology will continue to evolve in this direction with the hyper‐ dense deployment of small cells providing hotspot islands of high data connectivity coverage zones. This context will ask new questions from the research community in terms of how to tunnel this traffic from the local serving base station towards the core network. Typically, in legacy networks, the segment of the network that interconnects the BS to the RAN (radio access network) to the EPC (evolved packet core) is called the backhaul. Fibre optic lines or microwave links have fulfilled this role, with limitations in terms of deployment cost and limited coverage area. However, mobile technology is heading towards an era of virtualization and software‐defined networking, where radio resources are allocated from a common pool to different providers, and their management is centralized. This new era is, in fact, reflecting parallels in the cloud computing world, with the onset of cloud services. Emerging mobile networks are heading towards a C‐RAN (cloud radio access network) approach [7, 8], where RRUs (remote radio units) and a centralized processing RAN core work in synergy to provide coordinated scheduling, or, in other words, interference management. This paradigm is changing the perception of the communication haul in the network, from backhauling to incorporating both a back and fronthaul segment. In this context, the backhaul dictates how the information is parried from the base stations to the core network, whilst the fronthaul refers to the connectivity segment between the C‐RAN core network and the small cell. Figure1.1 shows definitions of

BBU

(a)

(b)

X2 Sync

Layer 3

Layer 2

Layer 1

MME = Mobility management entity

SGW = Serving gateway

PGW = Packet data gateway

EPC = Evolved packed core

BS 1

BS 2

BS N

RRU 1

RRU 2

RRU N

RAN fronthaul

RAN backhaul

Aggregation switch Router

Aggregation point

BBU N

BBU 2

BBU 1

X2 Sync

Layer 3

Layer 2

Layer 1

BBU pool cloud

MME

Transpor t backhaul

PGW

SGW

Core network (EPC)

MME

Transpor t backhaul

PGW

SGW

Core network (EPC)

Figure1.1 Communication haul segments of (a) legacy and (b) emerging C‐RAN mobile network

4 Backhauling/Fronthauling for Future Wireless Systems

the backhaul and fronthaul segments pertaining to legacy and emerging C‐RAN architectures.

The future enhanced communication haul (be it backhaul or fronthaul) for 5G is expected to be deployed around 2020 in order to support the exponential growth in wireless data that is forecast over the next decade. Therefore, there is substantial market interest in the development of ground‐breaking backhaul and fronthaul solutions that can not only enhance today’s networks, but also provide a coherent interference management approach in emerging technologies such as C‐RAN and beyond. This communication haul challenge provided the inspiration for this book and its title: Backhauling/Fronthauling for Future Wireless Systems.

The book intends to bring together all mobile stakeholders, from academia and industry, to identify and promote technical challenges and recent results related to smart backhaul/fronthaul research for future communication systems such as 5G. It provides an overview of current approaches to backhauling legacy communication systems and explains the rationale for deploying future smart and efficient backhauling/fronthauling infrastructure from architectural, technical and business points of view using real‐life applications and use cases. The book is intended to inspire researchers, operators and manufacturers to render ground‐breaking ideas in the newly emerging discipline of smart backhauling/fronthauling over future, ultra‐dense wireless systems. Moreover, detailed security challenges are presented to analyse the performance of smart backhauling/fronthauling for future wireless. It is clear that smart backhauling/fronthauling deployment can offer a palette of interesting colours capable of painting new business opportunities for mobile stakeholders for next generation wireless communication systems. This is the first book of its kind on smart backhauling/fronthauling for future wireless systems which updates the research community on the communication haul roadmap, reflecting current and emerging features emanating from the 3GPP group.

To guide the reader through this adventure, the book has the following layout. In Chapter 2, a reference architecture for the future radio communication haul is presented from a 5G perspective. 5G networks are anticipated to obtain Shannon‐ level and beyond throughput and almost zero latency. However, there are several challenges to solve if 5G is to outperform legacy mobile platforms; one of these is the design of the communication ‘haul’. Traditionally, the backhaul segment connects the radio access network (RAN) to the rest of the network where the baseband processing takes place at the cell site. However, in this chapter, we will use the concept of ‘ fronthaul access,’ which is recently gaining significant interest since it has the potential to support remote baseband processing based on adopting a cloud radio access network (C‐RAN) architecture that aims to mitigate (or coordinate) interference in operator‐deployed infrastructures; this eases significantly the requirements in interference‐aware transceivers. To do this, we provide a reference architecture that also includes a network and protocol architecture and proposes a so‐called ‘cloud resource optimizer’. This integrated solution will be the enabler for

Introduction: The Communication Haul Challenge 5

RAN‐as‐a‐Service, not only paving the way for effective radio resource management, but opening up new business opportunities for virtual mobile service providers.

Emerging channel transmission approaches and the possibility of using higher frequency bands, such as massive MIMO and millimetre‐wave (mmWave), respectively, are of paramount importance for future wireless systems and for the communication haul. Chapter3 introduces the fundamentals with regard to massive MIMO and mmWave communication, and their suitability for small‐cell backhauling and fronthauling. Furthermore, a performance analysis model for wireless backhauling ofsmall cells with massive MIMO and mmWave communication is outlined. Using this model, some numerical results on the performance of massive‐MIMO‐ and/or mmWave‐based wireless backhaul networks are presented.

C‐RAN promises considerable benefits compared to decentralized network architectures. Centralizing the baseband processing enables smaller radio access points as well as cooperative signal processing and ease of upgrade and maintenance. Further, by realizing the processing not on dedicated hardware, but on dynamic and flexible general‐purpose processors, cloud‐based networks enable load balancing between processing elements to enhance energy and cost efficiency. However, centralization also places challenging requirements on the fronthaul network in terms of latency and data rate. This is especially critical if a heterogeneous fronthaul is considered, consisting not only of dedicated fibre but also of, for example, mmWave links. A flexible centralization approach can relax these requirements by adaptively assigning different parts of the processing chain either to the centralized baseband processors or the base stations based on the load situation, user scenario and the availability of the fronthaul links. This not only reduces the requirements in terms of latency and data rate, but also couples the data rate to the actual user traffic. In Chapter 4, a comprehensive overview of different decentralization approaches is given, and we analyse their specific requirements in terms of latency and data rate. Furthermore, we demonstrate the performance of flexible centralization and providedesign guidelines on how to set up the fronthaul network to avoid over‐ or under‐dimensioning.

Heterogeneous backhaul deployment using different wired and wireless technologies is a potential solution to meet the demand in small‐cell and ultra‐dense networks. Therefore, it is of cardinal importance to evaluate and compare the performance characteristics of various backhaul technologies in order to understand their effect on the network aggregate performance and provide guidelines for system design. In Chapter5, the authors propose relevant backhaul models and study the delay performance of various backhaul technologies with different capabilities and characteristics, including fibre, xDSL, mmWave and sub‐6 GHz. Using these models, the authors aim to optimize the base station (BS) association so as to minimize the mean network packet delay in a macro‐cell network overlaid with small cells. Furthermore, the authors model and analyse the backhaul deployment cost and show that there exists an optimal gateway density that minimizes the mean backhaul cost

6 Backhauling/Fronthauling for Future Wireless Systems

per small‐cell base station. Numerical results are presented to show the delay performance characteristics of different backhaul solutions. Comparisons between the proposed and traditional BS association policies show the significant effects of backhaul on network performance, which demonstrates the importance of joint system design and optimization for both the radio access and backhaul networks.

The small‐cell network (also called a HetNet) has been recognized as a potential solution to offer better service coverage and higher spectral efficiency. However, the dense deployment of small cells could cause inter‐cell interference problems and reduce the performance gains of HetNets. Various techniques have been developed in 4G for tackling inter‐cell interference. In particular, the inter-cell interference coordination (ICIC) technique can coordinate the data transmission and interference in two neighbouring cells. In Chapter6, the authors consider a HetNet consisting of macro‐cell networks overlaid with small‐cell networks that access the same spectrum simultaneously. Here, the HetNet architecture assumes macro cells and small cells interconnected via a high‐speed fronthaul/backhaul connection. In particular, due to the mobility of wireless subscribers, the load and data traffic are different in every active macro and small cell. The conventional static enhanced ICIC (eICIC) mechanism cannot ensure that adapting the almost blank subframes (ABS) duty cycle corresponds to the dynamic network condition. Only the dynamic eICIC mechanism is suitable for this non‐static network traffic. Therefore, the authors aim to develop a dynamic interference coordination strategy for eICIC for maximizing system utilities under given QoS constraints. In contrast to the traditional eICIC mechanism, the proposed method does not add any backhaul requirements. Computer simulations show that the performance in various scenarios of the dynamic eICIC mechanism with QoS requirements is better than a static eICIC approach and the conventional dynamic eICIC mechanism.

Cell selection for joint optimization considering backhauling technology is needed for future wireless systems. In this regard, Chapter 7 provides a comprehensive analysis for joint optimization considering the backhaul in terms of cell selection. This chapter considers heterogeneous cellular networks, where clusters of small cells are locally deployed to create hotspot regions inside the macro‐cell area. Most of theresearch on this topic has focused on mitigating co‐channel interference; however, the wireless backhaul has recently emerged as an urgent challenge to enable ubiquitous broadband wireless services in small cells. In realistic scenarios, the backhaul may limit the amount of signalling that can be exchanged amongst neighbouring cells, which aims to coordinate their operations in real time; furthermore, in highly loaded cells (such as hotspots), the backhaul can limit the data rate experienced by the end users. Here, the authors develop a novel cell‐association framework, which aims to balance the users amongst heterogeneous cells to improve the overall radio and backhaul resource usage and increase the system performance. The authors describe the relationship between cell load, resource management and backhaul capacity constraints. Then, the cell‐selection problem is expressed as a combinatorial

Introduction: The Communication Haul Challenge 7

optimization problem and two heuristic algorithms–called Evolve and Relax–are presented to solve this dilemma. The analysis shows that Evolve converges to a near‐ optimal solution, leading to notable improvements with respect to the classic SINR‐ based association scheme in terms of throughput and resource utilization efficiency.

High‐speed and long‐range wireless backhaul is a cost‐effective alternative to a fibre network. The ever‐increasing demand for high‐speed broadband services mandates higher spectral efficiency and wider bandwidth to be adopted in the wireless backhauls. As wireless mobile networks evolve toward 5G, employing higher‐order modulation and performing multiband and multichannel aggregation for wireless backhauling have become industry trends. However, commercially available wireless backhaul systems do not meet the stringent requirements for both high speed and long range at the same time. In Chapter8, the various system architectures for multiband and multichannel aggregation are discussed. The challenges for achieving high‐speed wireless transmission in multiband and multichannel systems are addressed. These challenges include: how to improve spectrum efficiency and power efficiency; how to prevent inter‐channel interference; and how to ensure low latency in order to ensure resilient packet delivery and load balancing.

Despite the significant benefits of C‐RAN technology in 5G mobile communication systems, C‐RAN technology has to face multiple inherent security challenges associated with virtual systems and cloud computing technology, which may hinder its successful establishment in the market. Thus, it is critical to address these challenges in order for C‐RAN technology to reach its full potential and foster the deployment of future 5G mobile communication systems. Therefore, Chapter9 presents representative examples of possible threats and attacks against the main components in the C‐RAN architecture in order to shed light on the security challenges of C‐RAN technology and provide a roadmap to overcome the security bottleneck.

In conclusion, we firmly believe this book will serve as a useful reference for early‐ stage researchers and academics embarking on this radio communication haul odyssey, but beyond that, it targets all major 5G stakeholders who are working at the forefront of this technology to provide inspiration towards rendering ground‐breaking ideas in the design of new communication hauls for next‐generation systems.

References

[1] Ericsson (2013) Mobility report, June. [2] Andrews, J. G., Buzzi, S., Choi, W., Hanly, S. V., Lozano, A., Soong, A. C. K. and Zhang, J. C.

(2014) What Will 5G Be? IEEE Journal on Selected Areas on Communication, 32(6), 1065–1082. [3] Huawei Technologies Co. (2013) 5G: A technology vision. White paper. [4] Osseiran, A., Boccardi, F., Braun, V., Kusume, K., Marsch, P., Maternia, M., Queseth, O., Schellmann,

M., Schotten, H., Taoka, H., Tullberg, H., Uusitalo, M. A., Timus, B. and Fallgren, M. (2014)

Scenarios for 5G mobile and wireless communications: The vision of the METIS project. IEEE

Communications Magazine, 52(5), 26–35.

8 Backhauling/Fronthauling for Future Wireless Systems

[5] Parkvall, S., Dahlman, E., Furuskär, A., Jading, Y., Olsson, M., Wanstedt, S. and Zangi, K. (2008)

‘LTE Advanced–Evolving LTE towards IMT‐Advanced,’ Vehicular Technology Conference, 21–24 September, pp. 1–5.

[6] 3GPP (2011) ‘Feasibility Study for Further Advancements for E‐UTRA (LTE‐Advanced) (Release

10),’ TR 36.912, V10.0.0, March.

[7] China Mobile Research Institute (2011) ‘C‐RAN: The Road Towards Green RAN’. Technical report,

April. Available at: http://labs.chinamobile.com/cran/wp‐content/uploads/CRAN_white_paper_ v2_5_EN.pdf.

[8] Checko, A., Christiansen, H. L., Yan, Y., Scolari, L., Kardaras, G., Berger, M. S. and Dittmann, L.

(2015) Cloud RAN for Mobile Networks–A Technology Overview. IEEE Communications Surveys Tutorials, 17(1), 405–426.

www.ebook3000.com

A C‐RAN Approach for5G Applications

Kazi Mohammed Saidul Huq, Shahid Mumtaz and Jonathan Rodriguez

Instituto de Telecomunicações, Aveiro, Portugal

2.1 Introduction

Nowadays mobile Internet is a pervasive phenomenon. In the last decade, this phenomenon, along with the market drive for novel software applications spurred by the availability of smartphone handsets, has led to an unprecedented increase in data traffic. Researchers and experts predict that this upward trend will continue as the 5G community envisions new usage scenarios that involve connecting people, machines and applications through a mobile infrastructure. For this reason, the current technology requires a radical change to cater for this new tidal wave of mobile data, which has led us to the fifth generation (5G) communications era [1]. 5G will be expected to deliver a new era of wireless broadband connectivity, shaped by emerging use cases that aim to interconnect devices (the Internet of Things– IoT), enhance quality of experience (QoE) for the end user in terms of traditional mobile connectivity and be the main platform for addressing critical emergency infrastructures. 5G will play a role in the digitalization of Europe, and key targets include: increasing the peak data rate by 100 times, enhancing network capacity by 1000 times, increasing energy efficiency by 10 times and reducing latency by 30 times [2], all of which represent significant and challenging design requirements in contrast to the legacy 4G system. To achieve these targets, mobile stakeholders (such as operators, carriers and manufacturers) are contriving to incorporate macro cells and small cells into the design of the radio access infrastructure. This has forced system designers to reconsider the

10 Backhauling/Fronthauling for Future Wireless Systems

existing backhaul design of legacy 4G radio networks and to consider both a new backhaul and fronthaul design for ultra‐dense heterogeneous networks (HetNets).

5G networks are increasingly perceived as carriers to support a fully fledged, data‐ centric application rather than voice‐centric applications. Hence, one of the principal dilemmas operators are coming across nowadays is how to transform the existing

backhaul/fronthaul

infrastructure into an Internet Protocol (IP)‐based backhaul/fronthaul solution for hyper‐dense small‐cell deployment. With regard to the hauling of data, the continued use of fibre will give rise to the same problems as experienced today, which are mainly economic but also involve restrictions on deployment due to the geographical locations of transceiver cell sites. Millimetre‐wave (mmWave) backhaul/fronthaul is an option, but technological and regulatory challenges are yet to be addressed for its successful deployment. Another emerging solution is to exploit the interworking and joint design of open access and backhaul/fronthaul network architecture for hyper‐dense small cells based on cloud radio access networks (C‐RANs) [3]. This requires smart backhauling/fronthauling solutions that optimize their operations jointly with the access network optimization protocol. The availability, convergence and economics of smart backhauling/fronthauling systems arethe most important factors in selecting the appropriate backhaul/fronthaul technologies for multiple radio access technologies (including small cells, relays and distributed antennas) and heterogeneous types of excessive traffic in the future cellular network. However, in this chapter, we will use the concept of ‘fronthaul access’, which is recently gaining significant interest since it has the potential to support remote baseband processing based on adopting a C‐RAN architecture that aims to mitigate (or coordinate) interference in operator‐deployed infrastructures; this eases significantly the requirements in interference‐aware transceivers. Under the umbrella of a C‐RAN scenario, we introduce the notion of a ‘cloud resource optimizer’, which requires reengineering the medium access control (MAC) to provide a unified solution. The emergence of wireless fronthaul solutions widens the appeal for small‐cell deployments, because a fibre‐only solution–the technology typically used for fronthaul–is too expensive or just not available at many small‐cell sites. Moreover, we will also present a few ideas of potential applications for C‐RAN‐based mobile systems such as virtualization of device‐to‐device (D2D) services.

Following the introduction, this chapter is organized as follows. In Section2.2, we provide a brief overview of different types of backhauling/fronthauling technologies, and in particular, guide the interested reader through the transition from existing to emerging communication haul technologies. In Section2.3, we present network and protocol architecture for the baseline 3GPP coordinated multi‐point (CoMP) system, as a starting point, and then evolve this towards the emerging C‐RAN‐based architecture in Section2.4, which is widely seen as the next step on the mobile evolutionary landscape and indeed one step towards the 5G communication platform.

The terms backhaul and fronthaul are used interchangeably in this chapter.

A C‐RAN Approach for5G Applications 11

Based on this platform, we develop an integrated solution for the cloud resource optimizer, which defines a unified MAC. Section2.5 takes this design to the next level by using device‐to‐device (D2D) communication as a use‐case application by introducing a new small‐cell paradigm based on ‘on‐demand’ virtual small cells for coping with the dynamic variations in mobile traffic throughout the day; which is also an emerging scenario within the context of 5G. Finally, Section2.6 summarizes and concludes this chapter.

2.2 From Wired toWireless Backhaul/Fronthaul Technologies

In this section we provide a brief summary of the different kinds of backhaul/fronthaul technologies which are widely accepted and used by operators and service providers. According to [4, 5] hauling technologies are divided into two major categories: wired and wireless. Figure 2.1 shows the classification of backhaul technologies. For example, in the case of the wired backhaul, copper cables are the conventional medium whereas optical fibres are touted as an emerging hauling medium.

In wired backhaul, two types of physical media are widely used: copper cables and optical fibres. Copper cables are the conventional hauling medium between base transceiver stations (BTSs) and the base station controller (BSC) [4]. Currently, copper cables are being replaced by optical fibres due to their higher rates and low latency. Traditional copper‐based backhauling is used in digital subscriber line (DSL) access networks [6]. The alternative to copper for mobile backhaul is fibre‐based solutions that can provide almost unlimited capacity. The main fibre access options include GPON (gigabit passive optical network), carrier Ethernet and point‐to‐point (PTP) fibre [7].

There is another type of backhaul: wireless backhaul. This type of communication haul can be distinguished by the different frequency bands. Although the channel traits are different in this type of backhaul owing to different bands, each technology has its own merits and demerits. One very significant similarity amongst these technologies over wired backhaul is fast and relatively cheap deployment. For example, free space optics (FSO) use light to transmit data, but unlike relying on fibre as a transmission medium, free space propagation is applied [8]. FSO links

Copper

Wired

Optical fibre

Backhaul/fronthaul

Wireless

Free space optics

Figure2.1 Different types of backhaul/fronthaul

SatelliteMicrowave mmWave Relaying

12 Backhauling/Fronthauling for Future Wireless Systems

also create nearly zero interference between each other; the reason being the narrow beam width. Microwave communication haul technologies utilize different bands of carrier frequencies, ranging from 6 GHz to 42 GHz [5]. Microwave uses licensed spectrum which, in turn, enhances deployment time and cost [9]. Recently, a new paradigm is emerging under the wireless backhaul category: millimetre‐wave (mmWave) technology [10]. The explosive developments in circuit technologies have led to mmWave now being considered a viable option, and indeed foreseen as shaping next‐ generation small‐cell wireless backhaul. There are three types of frequency bands available for mmWave –60, 70/80 and 90 GHz [10]. These high carrier frequencies can enable multi‐Gbps data rates [5]. As the 60 GHz band is unlicensed and the higher bands only require an easy and inexpensive licensing process, the links can be deployed much faster and at lower cost [11]. The relay backhaul is another alternative, and is mainly used in the access link. Its inherent advantage isthat relays use the same transmission technology and licences as the access link.However, they also have similar shortcomings in terms of range (up to a few kilometres), capacity (a few hundred Mbps) and interference [5]. Satellite backhaul provides an answer for certain terrain where no other backhaul technologies are viable to deploy [4]. In general, T1/E1 is the physical transmission medium over satellite links for cellular backhaul [12].

2.3 Architecture forCoordinated Systems According toBaseline 3GPP

The C‐RAN incorporates both a joint signal processing capability and the resource optimization of data belonging to different users which conventional coordinated 3GPP techniques cannot carry out due to high complexity and signal overhead during coordination. Data and signalling are exchanged between different base stations (BSs) through links which are usually capacity limited. This sometimes makes the signalling exchange infeasible. In this section, we describe network and protocol architecture of a coordinated system according to 3GPP.

Figure2.2 shows the network architecture of a coordinated 3GPP system. This baseline scenario is based on BS cooperation, which recently attracted much interest from the research community. In the 3GPP LTE‐Advanced, it is referred to as coordinated multi‐point (CoMP) transmission and is being studied actively in LTE release 11 [13].

The inter‐BS cooperation has been presented as an effective approach to mitigate inter‐cell interference and hence improve cell edge throughput performance. Among the several categories of CoMP technologies [14], we focus only on downlink joint transmission (JT) CoMP in this chapter. In JT CoMP, downlink data can be simultaneously transmitted from multiple BSs to user equipment (UE). It is well known that the cell‐edge performance is dramatically improved by JT CoMP. However,

A C‐RAN Approach for5G Applications 13

RRC IP RLC MAC

Cell

S1 S1

Layer 3 Layer 2 Layer 1PHY

X2-AP SCTP

RRC IP

MAC

RLC

PHY

EPC

GTP-U

UDP

RRC

RLC MAC

PHY

Figure2.2 Network architecture of baseline 3GPP CoMP system

Cell

Figure2.3 Depiction of a JT CoMP use case

RRC RLC MAC

Cell

Layer 3 Layer 2 Layer 1 PHY

theperformance of JT CoMP can be degraded in the absence of a high‐speed and low‐latency backhaul network [15].

This scenario is based on a distributed approach, where each BS has its own layers of LTE protocol stack (i.e., physical (PHY), medium access control (MAC), radio link control (RLC), packet data convergence protocol (PDCP)) and each BS scheduler controls its own UE in the cell. The BSs are connected via an IP‐based X2 interface, which acts as an asynchronous communication link for managing JT CoMP operation; this interface is also used for distributing downlink data between BSs. These BSs are attached to the core network via the S1/S5 interface. Moreover, we assume that two BSs are synchronized by a global positioning system (GPS).

To understand the underlying mechanics of CoMP, Figure 2.3 illustrates a JT CoMP use case, where a user migrates between cells in an LTE network. Assume that the UE is located at the cell centre in cell receiving a downlink signal from BS between cell

and cell2, the UE automatically triggers JT CoMP to improve the

performance at the cell edge by receiving a downlink signal from BS

. Finally, when the UE moves to cell2, the UE automatically terminates JT CoMP

operation and BS

becomes the communication link.

, initially. The UE is attached to BS1 and

. However, as the UE moves to the cell edge

in addition to

14 Backhauling/Fronthauling for Future Wireless Systems

EPC

Data

Local sch

Resource

management

Co-ord sch

PDCP

RLC

MAC

PHY

Data

resource

management

Co-ordinated

signalling

Data forwarding

ControlMAC Synchronization

Figure2.4 Protocol architecture of baseline 3GPP CoMP system

PDCP

RLC

MAC

PHY

Local sch

Resource

Co-ord sch

management

Figure2.4 shows the protocol architecture to realize the simultaneous transmission scheme based on the LTE standard. The UE reports two kinds of reference signal received power (RSRP) messages to BS the difference between RSRP

and RSRP2 (in dBm) is smaller than the predefined

: RSRP1 from BS1 and RSRP2 from BS2. If

CoMP threshold, then JT CoMP is started; if the difference exceeds the predefined CoMP threshold, then JT CoMP is terminated.

When JT CoMP is triggered, the scheduler in BS part in BS

to make sure that the radio resources are available for JT CoMP (see the

will first check with its counter-

heavy black line in Figure2.4). During JT CoMP, the downlink data are processed in the following manner (see black arrows). First, PDCP, RLC and MAC are applied tothe downlink data in BS At the same time, the scheduler in BS

and the MAC protocol data unit (MAC‐PDU) is created.

provides the joint transmission time as well as

control information regarding MCS (modulation and coding scheme), radio resource to be used and antenna mapping for the MAC‐PDU. The joint transmission time and

A C‐RAN Approach for5G Applications 15

the control information are then attached to the MAC‐PDU and duplicated; one of them is sent to PHY in BS

and other is sent to PHY in BS2 via the X2 interface. The

PHY processing is carried out at both BSs in parallel. Finally, the MAC‐PDU is simultaneously transmitted from the two synchronized BSs at the specified joint transmission time.

To transport a MAC‐PDU from BS

to BS2, the MAC‐PDU is encapsulated by the

GTP tunnelling protocol. The joint transmission time and the control information that should be attached to this MAC‐PDU are included in a MAC‐control element (MAC‐ CE) in the MAC‐PDU.

2.4 Reference Architecture forC‐RAN

To overcome the limitations of CoMP, a holistic architectural change is expected via connecting the BSs to central clouds. Unlike the baseline CoMP scenario described in the previous section, in the C‐RAN most of the signalling takes place in the cloud and is shared among sites in a pool of virtualized baseband processing units (BBUs). Due to the fact that fewer BBUs are required in the C‐RAN compared to the traditional architecture (legacy 3GPP scenario), C‐RAN also has the potential to reduce the cost of network operation. This type of network architecture also improves scalability and makes BBU maintenance easier. Different operators can share this cloud BBU pool, which allows some to rent the RAN as a cloud service. Since BBUs from different sites are co‐located in one pool, they can communicate with lower delays. This brings to the forefront many other advantages, since existing mechanisms introduced in LTE‐A to increase spectral efficiency, interference management and throughput, such as enhanced inter‐cell interference coordination (eICIC) and CoMP, are greatly facilitated here.

2.4.1 System Architecture forFronthaul‐based C‐RAN

Emerging scenarios in cell deployment are heading towards the notion of cloud radio. In this section we provide the reference system model for the C‐RAN scenario with the description of its components. C‐RAN is a novel mobile technology that separates baseband processing units (BBUs) from radio front‐ends such as remote radio units (RRUs). In this technology, BBUs of several BSs are positioned in a central entity to form aBBUpool where the radio front‐ends of those BSs are deployed at the cell sites [16–18]. Therefore, this new framework unfolds a new paradigm for algorithms/ techniques that require centralized and cooperative processing. However, the deployment of this new technology faces several potential research challenges, which include latency, efficient fronthaul design and radio resource management for a converged network.

Fronthaul enables a C‐RAN architecture in which all the BBUs are placed at a

distance from the cell site. The fronthaul transports the unprocessed RF signal from

16 Backhauling/Fronthauling for Future Wireless Systems

the antennas to the remote BBUs. While the fronthaul requires higher bandwidth, lower latency and more accurate synchronization than the backhaul, it does support more efficient use of RAN resources; when coupled with legacy interference and mobility management tools, this can significantly minimize interference in the structured part of the network, including multi‐tier cell interference.

The general system model of the fronthaul‐based C‐RAN scenario is illustrated in Figure 2.5, and consists of three main components [18], namely: (i) a centralized BBU pool, (ii) remote radio units (RRUs) with antennas and (iii) a transport link, that is a fronthaul network which connects the RRUs to the BBU pool. The RRU provides the interface to the fibre as well as performing digital processing, digital‐to‐analogue conversion, analogue‐to‐digital conversion, power amplification and filtering [16]. The distance between the RRU and the BBU can be extended up to 40 km, where the ceiling range emanates from the processing and propagation delay. Optical fibre, mmWave and microwave connections can be used. In the downlink, the RRUs transmit the RF signals to the UEs, or in the uplink the RRUs carry the baseband signals from the UEs to the BBU pool for further processing. The BBU pool is composed of BBUs which operate as virtual base stations to process baseband signals and optimize thenetwork resource allocation for one RRU or a set of RRUs. The fronthaul links can constitute different technologies, namely wired (fiber → ideal) and wireless (mmWave → non‐ideal). One can easily add or update any number of BBUs in thiscloud depending on the needs and cell planning of the network operator. This C‐ RAN‐based architecture is also more energy efficient than the CoMP‐based scenario due to reduced power consumption at the cell sites. In the C‐RAN network architecture, no additional power is needed in cell sites other than for RRU operation.

By enabling joint processing in the cloud, key research challenges emerge related to joint provisioning of resources between the different BBUs. This leads us to the design of a so‐called ‘cloud resource optimizer’.

2.4.2 Cloud Resource Optimizer

In this section we present the proposed cloud resource optimizer for the C‐RAN. Interconnections and functions split between BBUs and RRUs are depicted in Figure 2.6. Unlike a CoMP resource management module, where all the resource management entities are separated for different BSs, this resource optimizer unifies all the resource management operation including allocation, interference management and signalling for different BBUs in the cloud pool. Inside this cloud resource optimizer, the PHYs from different RRUs are merged into one common MAC, control (Ctrl) and Synchronization (Sync) entity. This operation prompts us to develop a new MAC approach for this cloud‐based system. The MAC works as an enabler between different types of radio access technologies (RAT) such as LTE (IMT technology) and WiFi (non‐IMT technology).

Microwave/mmWave

mmWave/optical fibre

Fronthaul

BBU N BBU 2

BBU 1

X2 Sync

Layer 3

Layer 2

Layer 1

Fronthaul

BBU pool cloud

MME

Copper

PGW

SGW

Core network (EPC)

Fronthaul

RRU 1

RRU 2

RRU N

Figure2.5 Operator’s perspective on the fronthaul‐based C‐RAN scenario

18 Backhauling/Fronthauling for Future Wireless Systems

EPC

Data

PDCP

RLC

Unified MAC

MAC Ctrl Sync

Cloud Resource Optimizer

PHY

I/Q I/Q

RRU

PHY

RRU

Figure2.6 Architecture of the cloud resource optimizer

We consider a novel, unified MAC frame for our C‐RAN scenario in Figure2.7, unlike in legacy CoMP where each RAN has its own MAC. The shift in engineering design to introduce the presence of a global MAC entity will not only improve the efficiency (both spectrum and energy) of the system, but take a step towards reducing the overall interference in the network. This unified MAC will be a modified version of an existing LTE MAC frame described in [19].

As can be seen in Figure2.7, there are several MAC‐CEs in both the downlink and uplink MAC. Following Table1 and Table2 from the 36.321 standard [19] (shown here in Tables2.1 and 2.2), we can see the logical channel ID (LCID) types of MAC header. The parts indicated by the bold rectangle emphasize the LCID values for the various MAC‐CEs.

We define a new MAC‐CE for this purpose. We use the reserved element field for specifying the unified frame, and this is indexed in the MAC‐PDU sub‐header by an LCID value equal to 11001 in the uplink. The new element is called a unified frame and is appended to the existing LCID values, such as the common control channel (CCCH), cell radio network temporary identifier (C‐RNTI) and the padding.

A C‐RAN Approach for5G Applications 19

MAC PDU/transport block

Header

Sub-header1Sub-header

RR ELCID = 11001 Unified frame

Table2.1 Values ofLCID forDL‐SCH (adapted from[19])

Index LCID values

00000 CCCH 00001–01010 Identity of the logical channel 01011–11011 Reserved 11100 UE Contention Resolution Identity 11101 Timing Advance Command 11110 DRX (Dynamic Reception) Command 11111 Padding

Ctrl

elements

Sub-header

Payload (SDU)Padding

Ctrl

Element 1

Ctrl

Element 2

Figure2.7 Unied MAC frame for C‐RAN

Ctrl

Element N

Table2.2 Values ofLCID forUL‐SCH (adapted from[19])

Index LCID values

00000 CCCH 00001–01010 Identity of the logical channel 01011–11001 Reserved 11010 Power Headroom Report 11011 C‐RNTI 11100 Truncated BSR (Buffer Status Report) 11101 Short BSR 11110 Long BSR 11111 Padding

20 Backhauling/Fronthauling for Future Wireless Systems

This unified MAC in the cloud resource optimizer can provide differentiated services to dual‐band devices which connect simultaneously to different kindsof networks. Unlike the current convention where the UE selects either licensed→LTE or unlicensed→WiFi, the cloud resource optimizer makes dynamic decisions in the unified MAC framework, benefiting from the global knowledge available in terms of congestion level on the different radio networks and the quality of service (QoS) requirements of the UE’s traffic. This novel MACscheme has the potential to open up new opportunities in terms of traffic‐ orientated applications.

2.5 Potential Applications forC‐RAN‐based Mobile Systems

The evolution towards 5G is considered to be the convergence of Internet services with existing mobile networking standards, leading to the commonly used term ‘ mobile Internet’ over small cell, with very high connectivity speeds. In addition, green communications seem to play a pivotal role in this evolutionary path, with key mobile stakeholders driving momentum towards a greener society through cost‐effective design approaches. In fact, it is becoming increasingly clear from emerging services and technological trends that energy and cost per bit reduction, service ubiquity and high‐speed connectivity are becoming desirable traits for next‐ generation networks. Providing a step towards this vision, small cells are envisaged as the vehicle for ubiquitous 5G services providing cost‐effective, high‐speed communications.

These small cells are set up on demand and constitute a novel paradigm toward a 5G wireless system in two ways:

• A wireless network of cooperative small cells (CSC);

• Virtual small cells (VSC).

These novel, on‐demand small cells (CSC, VSC) have the ability to cope with today’s wireless traffic dynamics and adjust their RF parameters accordingly. Moreover, these on‐demand small cells can be used in various applications and scenarios en route to the accomplishment of the 5G goal. Among others, one of these is D2D‐based C‐RAN, which will be discussed in the next section.

2.5.1 Virtualization ofD2D Services

D2D is widely considered an effective and efficient candidate approach for very low latency communications in 5G [20], as well as being the enabler for improving spectrum efficiency. In fact, by reusing the spectrum, two D2D users can form a direct data link without exploiting explicitly the communication infrastructure

A C‐RAN Approach for5G Applications 21

(BSand core networks). D2D communications will also be spurred on from the application perspective, as D2D is considered an ideal deployment for proximity‐ based services, an example being social networking.

However, coexistence approaches are required in order for D2D users not to interfere with macro‐cell users, but at the same time to exploit spectrum appropriately inanera where spectral resources are at a premium. In this context, we propose an integrated solution where we combine technology paradigms such as C‐RAN and virtual small cells in synergy to provide an enabler for effective D2D‐based communications [21]. This novel architecture has the potential to solve most of the challenges related to emerging 5G systems (capacity, latency, energy efficiency, CAPEX/ OPEX and mobility). Moreover, these D2D networks will be created on demand, for example, if there are certain users at the cell edge with physically low battery levels, then a user with a high battery level will become the cluster head and other users with low battery levels will communicate in a D2D manner with the cluster head, while the cluster head communicates directly with the C‐RAN.

In this architecture, we first split the control/data plane where the RRU provides the signalling service for the whole coverage area and exploit these virtual small cells for delivering data services for high‐rate transmission, complemented by a light control overhead and a selection option on the most appropriate air interface (mmWave could be the best option), which is illustrated in Figure2.8.

2.5.2 Numerical Analysis

In order to analyse the performance of D2D‐based C‐RAN efficiently, we have enhanced an existing system‐level simulator (SLS) with a centralized cloud entity to control all the baseband processing, an RRU which acts as an antenna and a D2D user pair. Moreover, we have also enhanced the following key performance indicators (KPIs) to evaluate the performance of the proposed system.

• Throughput (with ideal fronthaul): The average throughput per cell is defined as the sum of the total amount of bits being successfully received by all active users in the system divided by the product of the number of cells being simulated in thesystem and the total amount of time spent in the transmission of these packets (the simulation time for LTE is TTI = 1 ms).

• Throughput (with non‐ideal fronthaul): The average throughput per cell is defined as the sum of the total amount of bits being successfully received by all active users in the system divided by the product of the number of cells being simulated in the system, the total amount of time spent in the transmission of these packets (the simulation time for LTE is TTI = 1 ms) and the delay of the fronthaul link (10 ms).

Table2.3 shows the full list of simulation parameters.

22 Backhauling/Fronthauling for Future Wireless Systems

Channel condition

BBU1

Control

Transport

Network

MAC

Physical

BBU2

Control

Transport

Network

MAC

Physical

Fronthaul

BBU N

Control

Transport

Network

MAC

Physical

Network

MAC

Physical

mmWave cell 30 GHz

LTE cell 800 MHz

RRU

D2D users

Cellular users

Figure2.8 On‐demand, device‐centric advanced C‐RAN

Routing: finding the best way of each flow

Assign resource

Figure2.9 shows the average throughput of the system with and without ideal fronthaul. For this simulation, we consider 20 D2D pairs and 20 cellular users (CUs). We also assume a fixed resource allocation between CUs and D2D users. There are 100 resource blocks (RBs) in the LTE 20 MHz band, which are divided equally among CUs and D2D users (50 RBs each). Each of these RBs is then assigned to its corresponding users via a proportional fairness (PF) scheduler. CUs on 50 RBs communicate using a cellular link (UE1↔RRU↔UE2), while D2D users use a direct link (UE1↔UE2). When only CUs with ideal fronthaul are deployed, the average throughput of the system is around 10 Mbit/s, but when this scenario is enhanced with D2D, the average throughput rises to 20 Mbit/s. This increase in throughput is due to the inclusion in the C‐RAN network of D2D, which, thanks to its direct communication capability, enhances the average throughput of the system. Moreover, if resource‐allocation schemes between CUs and D2D users with some interference‐cancellation mechanism are considered, the average throughput of the system increases even further.

A C‐RAN Approach for5G Applications 23

Average throughput (Mb/s)

fronthaul)

Table2.3 Simulation parameters

Name Parameter

System LTE‐A, 20 MHz, 2.6 GHz Resource block (RB) 100 Duplexing method Cellular: FDD (downlink)

D2D: FDD (uplink using TDD timeslot) Mode selection Shortest distance (cellular or D2D) Resource allocation Fixed allocation Channel estimation Perfect Channel models Between D2D 40log10d[m] + 30 + 30log10(f [Mhz] + 49)

RRU→ D2D 36.7log10d[m] + 40.9 + 26log10(f [Mhz]/5) + α RRU→ CU 36.7log10d[m] + 40.9 + 26log10(f [Mhz]/5) + α

shadowing

Retransmission HARQ Scheduler of eNB Proportional fairness (PF) Power control Adaptive power Traffic Full buffer Fronthaul Ideal (no delay)/non‐ideal (10 ms delay) Maximum transmit power RRU = 30 dBm,

Cellular Tx_Power = 24 dBm

D2D Tx_Power = 9 dBm Noise figure 5 dBm for base station/9 dBm for D2D receiver Thermal noise density −174 dBm/Hz User speed Static

(ideal fronthaul)

D2D

(Ideal fronthaul)

(Non-Ideal

D2D

(Non-Ideal

Figure2.9 Throughput vs. ideal/non‐ideal fronthaul

24 Backhauling/Fronthauling for Future Wireless Systems

When simulations with a non‐ideal fronthaul are run, a delay of 10 ms is experienced, as shown in Figure2.9, and the throughput of CUs drops to around 4 Mbit/s. This is due to the fronthaul delay: the greater the delay, the lower the throughput. But for the D2D case, throughput remains the same, because in D2D, data are transferred directly between devices and therefore there is a sort of ‘zero delay’ [22], but this is still under the control of the C‐RAN, which is a benefit in terms of mobility and handover.

Virtual small cells not only need to be ubiquitous and cost‐effective, but have to deliver emerging services in a secure fashion in an era where applications will handle ‘extremely confidential data’ and money transactions. Therefore, 5G networks must deliver a framework with a palette of security tools that are the enablers for a cross‐ system, end‐to‐end secure link which is fast and lightweight in nature. Virtual small cells also open up the possibility for network operators to invest innetwork‐sharing scenarios whereby operators can accommodate the foreseen increase in traffic whilst reducing their investment in new infrastructure, and beyond that significantly reduce their energy bill. These enhancements are currently addressed by 3GPP; however, this raises new research challenges when applied to the virtual small environment.

Table2.4 shows the comparison of D2D‐based C‐RAN with existing LTE‐Advanced (LTE‐A) technologies like CoMP, taking into consideration the main architectural blocks of the communication network, like the evolved node B (eNB) (in LTE, the BS is called the eNB).

2.6 Conclusion

To take a step towards the 5G vision, this chapter has described a C‐RAN reference system architecture as the fundamental building block that has the potential to evolve and to anticipate disruptive changes in users’ demands and small‐cell deployment. The first part of this chapter introduces the C‐RAN architecture, which exploits RRU technology connected to the core network using backhaul technology based on fibre‐optic links. C‐RAN was engineered so as to mitigate one of the key disabling features affecting mobile communications since their inception: user interference. In C‐RAN, the key is to harness all the baseband processing from all users within a common unit, thus providing the operator complete control over the network and the ability to coordinate signal transmissions, providing a significant step towards mitigating interference in the network and alleviating interference‐aware transceivers. However, the key aim of the chapter was to go beyond C‐RAN and examine the fronthaul component. In particular, we use the C‐RAN approach as a fundamental building block and build on this to provide a more flexible platform that is able to support emerging use cases under the 5G umbrella. In this context, we introduce the notion of a cloud resource optimizer which works in synergy with a unified MAC, allocating virtual radio resources on demand to support various applications, and potentially acting as the lynchpin to

(continued )

bands

20 m 100 m/1000 m

Combination of

C‐RAN and D2D.

Improved capacity and

coverage at cell centre

and edge. More energy

efficient than CoMP

and C‐RAN. Relaying

Improved capacity and

coverage at cell centre

and edge. More energy

efficient than CoMP.

in cellular networks,

safety in public

services, sharing in

context‐based

applications.

Transferring of data

between users occurs

in licensed bands

directly. It is

managed by a cloud

central controller.

Transferring of data

between users occurs

directly, be it licensed

or unlicensed bands.

Within licensed bands,

a radio controlling

cloud is used for

transferring data.

3GPP release 11–12 IEEE 3GPP release 11–12 3GPP and IEEE

Table2.4 Comparison oftechnologies

Characteristic CoMP C‐RAN D2D [23] D2D:C‐RAN

Standardization

Frequency band Licenced bands Licenced bands Licenced bands Licenced/Unlicenced

Good (500–600 Mbps) Good (500–600 Mbps) Very good (1 Gbps) Excellent (2 Gbps)

Max transmission distance 500 m 100 m/1000 m

Latency >1 ms >1 ms ‘Zero latency’ ‘Zero latency’

Capacity

No No Yes Ye s

Uniformity of service

provision

Application Improved capacity at

cell edge. Longer

battery life of hand-

sets. Complete

operator control.

a central controlling

unit such as eNB is

used for transferring

data.

Infrastructure Within licensed bands,

Combination of

C‐RAN and D2D.

CAPEX: No expenses

since users are using

their devices with all

types of technologies

such as WiFi, LTE and

D2D.

CAPEX: Subsidized

C‐RAN (BBU)

hardware. Installing

new RRUs.

OPEX:

Communication haul,

OPEX: Usage of

battery of the devices.

i.e., fronthaul. Leasing

and electric power

consumption of the cell

site. Operational cost

of RRUs.

Characteristic CoMP C‐RAN D2D [23] D2D:C‐RAN

Table2.4 (continued )

hardware.

Commissioning new

cell sites and towers.

Expenses CAPEX: Subsidized

OPEX:

Communication haul,

i.e., fronthaul. Leasing

and electric power

consumption of the

cell site. Operational

cost of eNBs.

These values are calculated under ideal channel and traffic conditions and with 2x2 MIMO using 20 Mhz of bandwidth.

100 m is for mmWave at 60 GHz and 1000 m is for fibre.

A C‐RAN Approach for5G Applications 27

support the design of new 5G protocols/algorithms. In particular, we show how this platform can not only support the emerging D2D paradigm, but is also able to support the deployment of small‐cell technology which is seen as pivotal to the 5G story.

References

[1] Andrews, J. G., Buzzi, S., Choi, W., Hanly, S. V., Lozano, A., Soong, A. C. K. and Zhang, J. C.

(2014) What Will 5G Be? IEEE Journal on Selected Areas in Communications, 32(6), 1065–1082.

[2] Osseiran, A., Boccardi, F., Braun, V., Kusume, K., Marsch, P., Maternia, M., Queseth, O.,

Schellmann, M., Schotten, H., Taoka, H., Tullberg, H., Uusitalo, M. A., Timus, B. and Fallgren, M. (2014) Scenarios for 5G mobile and wireless communications: the vision of the METIS project. IEEE Communications Magazine, 52(5), 26–35.

[3] China Mobile Research Institute (2011) ‘C‐RAN: The Road Towards Green RAN.’ Technical

report, April. Available at: http://labs.chinamobile.com/cran/wp‐content/uploads/CRAN_white_ paper_v2_5_EN.pdf.

[4] Tipmongkolsilp, O., Zaghloul, S. and Jukan, A. (2011) The Evolution of Cellular Backhaul

Technologies: Current Issues and Future Trends. IEEE Communications Surveys Tutorials, 13(1), 97–113.

[5] Bartelt, J., Fettweis, G., Wubben, D., Boldi, M. and Melis, B. (2013) ‘Heterogeneous Backhaul

for Cloud‐Based Mobile Networks.’ Paper presented at the Vehicular Technology Conference (VTC Fall), pp. 1–5.

[6] Eriksson, P. and Odenhammar, B. (2006) ‘VDSL2: Next important broadband technology,’

Ericsson Review No. 1.

[7] Orphanoudakis, T., Kosmatos, E., Angelopoulos, J. and Stavdas, A. (2013) Exploiting PONs for

mobile backhaul. IEEE Communications Magazine, 51(2), S27–S34.

[8] LightPointe Communications Inc. (2009) ‘Understanding the performance of free space optics.’

White paper.

[9] Giesken, K. (2002) Application of wireless technology in the mobile backhaul network. Bechtel

Telecommunications Technical Journal, 1(1), 62–70.

[10] Rappaport, T. S., Sun, S., Mayzus, R., Zhao, H., Azar, Y., Wang, K., Wong, G. N., Schulz, J. K.,

Samimi, M. and Gutierrez, F. (2013) Millimeter Wave Mobile Communications for 5G Cellular: It Will Work! IEEE Access, 1, 335–349.

[11] Rangan, S., Rappaport, T. S. and Erkip, E. (2014) Millimeter‐Wave Cellular Wireless Networks:

Potentials and Challenges. Proceedings of the IEEE, 102(3), 366–385.

[12] Owens, J. (2002) Satellite Backhaul Viability. Bechtel Telecommunications Technical Journal,

1(1), 58–61.

[13] RP‐111117 Work Item Description, ‘Coordinated Multi‐Point Operation for LTE,’ Samsung, 3GPP

TSG RAN meeting #53, Fukuoka, Japan, September 13–16, 2011.

[14] Zhang, L., Nagai, Y., Okamawari, T. and Fujii, T. (2013)‘Field Experiment of Network Control

Architecture for CoMP JT in LTE‐Advanced over Asynchronous X2 Interface.’ Paper presented at the Vehicular Technology Conference (VTC Spring), 2nd–5th June, pp. 1, 5.

[15] Okamawari, T., Zhang, L., Nagate, A., Hayashi, H. and Fujii, T. (2011) ‘Design of Control

Architecture for Downlink CoMP Joint Transmission with Inter‐BS Coordination in Next Generation Cellular Systems.’ Paper presented at the Vehicular Technology Conference (VTC Fall), 5th–8th September, pp. 1–5.

[16] Checko, A., Christiansen, H. L., Yan, Y., Scolari, L., Kardaras, G., Berger, M. S. and Dittmann, L.

(2015) Cloud RAN for Mobile Networks – A Technology Overview. IEEE Communications

Surveys Tutorials, 17(1), 405–426.

28 Backhauling/Fronthauling for Future Wireless Systems

[17] Beyene, Y. D., Jantti, R. and Ruttik, K. (2014) Cloud‐RAN Architecture for Indoor DAS. IEEE

Access, 2, 1205–1212.

[18] Wang, R., Hu, H. and Yang, X. (2014) Potentials and Challenges of C‐RAN Supporting Multi‐

RATs Toward 5G Mobile Networks. IEEE Access, 2, 1187–1195.

[19] http://www.etsi.org/deliver/etsi_TS/136300_136399/136321/09.00.00_60/ts_136321v090000p.

pdf.

[20] Mumtaz, S., Huq, K. M. S. and Rodriguez, J. (2014) Direct mobile‐to‐mobile communication:

Paradigm for 5G. IEEE Wireless Communications, 21(5), 14–23.

[21] Huq, K. M. S., Mumtaz, S., Rodriguez, J., Marques, P., Okyere, B. and Frascolla, V. (2016)

Enhanced C‐RAN using D2D Network. IEEE Wireless Communications, submitted January, 2016

(under review). [22] Huawei Technologies Co. (2013) ‘5G: A technology vision.’ White paper. [23] Feng, D., Lu, L., Yuan‐Wu, Y., Li, G., Li, S. and Feng, G. (2014) Device‐to‐device communications

in cellular networks. IEEE Communications Magazine, 52(4), 49–55.

Backhauling 5G Small Cells withMassive‐MIMO‐Enabled mmWave Communication

Ummy Habiba, Hina Tabassum and Ekram Hossain

Department of Electrical and Computer Engineering, University of Manitoba, Canada

3.1 Introduction

The exponential growth in the number of mobile subscribers and their corresponding wireless data traffic has demanded the emergence of fifth generation (5G) cellular networks. One of the key requirements for 5G is to increase the data rate radically, to approximately 1000 times that of the current 4G technology. In this regard, network densification is a straightforward way to increase the data rates. The idea of network densification is to increase the density of small base stations (SBSs) in order to enhance the number of users supported in a geographic region [1]. In principle, the density of SBSs can be increased indefinitely until there is only one user supported per SBS with its transmission and backhaul connections [1]. This extreme densification raises a variety of challenges that include determining the appropriate cell associations, managing inter‐tier and intra‐tier interference and providing high‐capacity backhaul connectivity to SBSs at the same time.

Provisioning of cost‐effective and scalable backhaul solutions for 5G ultra‐dense

networks is challenging due to the need for ample resources for the backhaul transmissions to/from a massive number of SBSs. Consequently, it is crucial to consider both wired and wireless options for backhauling infrastructure of 5G cellular networks depending on the location of the small cells and quality‐of‐service (QoS)

30 Backhauling/Fronthauling for Future Wireless Systems

requirements of the users. Even though the existing wired backhaul ensures high reliability, it may not be an economical and ascendable solution for densely deployed SBSs. On the other hand, wireless backhauling solutions are cost‐efficient as well as easily scalable to provide connectivity to small cells. Reasons include the possibility of frequency reuse, easy operation and management and increased flexibility of backhaul resource allocation.

There exist several wireless backhauling solutions for small cells, such as TVWS

800 MHz), sub‐6 GHz (licensed and unlicensed), microwave spectrum between

( 6 GHz and 60 GHz and FSO (free space optical) spectrum within the laser spectrum. Moreover, the spectrum available between 30 and 300 GHz can offer 200 times greater resources than the resources used for current cellular communications [2]. This frequency band with the wavelengths ranging from 1 to 10 mm is referred to as millimetre‐wave (mmWave) spectrum. Among various wireless backhauling options, mmWave spectrum between 60 and 90 GHz is considered suitable for 5G networks due to its small wavelength which allows the use of a large number of antenna elements in a compact form, line‐of‐sight (LOS) propagation with larger channel bandwidth and ample unlicensed spectrum.

Nonetheless, mmWave spectrum has a number of propagation challenges due to its susceptibility to shadowing, rain attenuation and molecular absorption. The mmWave signals cannot penetrate building walls and other blockages. Therefore, indoor and outdoor users need to be served by separate BSs operating in mmWave frequencies with large directivity gains. Consequently, highly directive antennas are essential for mmWave BSs to generate narrow beams that suppress the impact of environmental attenuation and interference arriving from neighbouring cells.

In this context, large‐scale antenna systems (often referred to as massive MIMO) can be deployed at the transmitter and receiver to enhance the directivity. The massive MIMO technology is well suited to the short‐wavelength mmWave signals, as a large number of antenna elements can be installed within a small area of mmWave transceiver nodes, enabling highly directional beams with high capacity. Massive MIMO technology coupled with efficient beamforming strategies diminishes the uncorrelated noise and short‐term fading effects irrespective of the number of users or BSs in the network [3]. However, the performance of beamforming remains limited by the channel estimation and acquisition techniques.

The rest of the chapter is organized as follows. In Section3.2, we first compare different wireless backhauling solutions that currently exist. Section3.3 describes the key features of mmWave propagation and massive MIMO technology. In Section 3.4, we review the state-of-the-art in designing backhaul systems using mmWave spectrum and discuss different LOS and NLOS topologies for mmWave backhauling. In this section, we also outline the design and implementation issues for massive‐MIMO‐enabled mmWave backhauling systems. Next, in Section3.5, we model a massive‐MIMO‐enabled mmWave backhauling system and consider maximizing the network rate through an efficient stable‐matching‐based user

Backhauling with Massive-MIMO-Enabled mmWave Communication 31

association scheme. Numerical results are also presented in Section3.5 before the chapter is concluded in Section3.6.

3.2 Existing Wireless Backhauling Solutions for5G Small Cells

The wireless backhaul connectivity of small cells is a function of the location and density of the SBSs, line‐of‐sight (LOS) conditions between the wireless backhaul hubs and SBSs, data rate requirements and additional costs for spectrum licensing. Appropriate backhauling spectrum thus needs to be chosen as per the network requirements. To this end, in this section, based on the studies in [4–6], we detail the features of different spectrum options for wireless backhauling.

• TVWS (600–800 MHz): The unused TV spectrum, known as TV white spaces (TVWS), can be a suitable option for sparsely populated areas where communication is limited to voice, video and real‐time gaming. This unlicensed spectrum can be accessed opportunistically to backhaul small cells in a cognitive manner. TVWS allows NLOS backhaul connectivity over a wider area (1–5 km) without any constraint for antenna alignment. Due to opportunistic access, there is a risk of unavailability of the spectrum. In addition, the backhaul connectivity must ensure that it does not impose any interference on primary TV transmission. The data rates provided by TVWS‐based backhauls may not be sufficient for backhauling 5G networks.

• Sub‐6 GHz: The licensed sub‐6 GHz spectrum (800 MHz– 6 GHz) is suitable for backhauling SBSs in rural as well as urban areas. Without any additional hardware or antenna alignment requirement, this spectrum allows NLOS communication over a wider distance. In urban areas, it can provide backhaul coverage from 1.5 to

2.5 km whereas it can support backhaul links up to 10 km in rural areas. However, the sub‐6 GHz spectrum is costly and provides narrow channel bandwidth which limits the capacity. In dense areas, sub‐6 GHz spectrum is already in use; therefore, it can suffer from severe interference. Sub‐6 GHz spectrum is thus not recommended for backhauling 5G small cells in dense urban areas.

• Microwave (6–60 GHz): The microwave spectrum is a common choice for back- hauling SBSs in urban and rural areas as it can provide 1 Gbps+ data rates for real‐time and non‐real‐time services. However, the spectrum is costly and it requires antenna alignment for LOS communication to achieve high directivity gain. The size of the antennas designed for these frequency bands is relatively large, which might not be feasible for SBSs. The microwave signals can provide backhaul coverage up to 4 km. Therefore, microwave backhauls may be beneficial from the perspective of data rate but may not be cost‐effective.

• Unlicensed millimetre wave (60–90 GHz): The unlicensed mmWave spectrum is another promising candidate for backhauling 5G small cells in dense urban areas. Since mmWave signals suffer from high propagation loss, a coverage distance

32 Backhauling/Fronthauling for Future Wireless Systems

beyond 1 km may not be possible. However, this may work for densely deployed small cells because of the reduced interference from adjacent links and, in turn, increased capability of frequency reuse. These extremely high frequency signals enable the installation of a large number of antenna elements within a small area which enable high antenna array gain. The wider channel bandwidth offered by mmWave spectrum and high directivity can provide multi‐Gbps data rates. Although mmWave spectrum requires antenna alignment and LOS connectivity, there is no additional spectrum cost.

• Free space optical (FSO) technology: FSO can be an alternative solution to traditional radio frequency (RF) spectrum. The FSO link (laser beam) between laser photo‐ detector transceivers is not susceptible to electromagnetic interference. FSO links use wavelengths in the range of micrometre and are capable of providing data rates in the order of 10 Gbps over 1 km [7]. There is no licensing cost for FSO transmission. However, there will be a hardware cost which is comparatively lower than optical fibre links. For FSO links, LOS is required and laser beams are sensitive to weather conditions such as rain, snow and fog. To minimize the effects of atmospheric attenuation, several techniques can be used that include aperture averaging, wave division multiplexing (WDM), larger receive aperture, fine pointing mirrors (FPM), etc.

Among the existing wireless backhaul solutions, mmWave spectrum has the potential to meet the requirements of 5G networks as it can provide wider channel bandwidth with highly directive narrow beams and, in turn, low interference. The small wavelength of mmWave also has the flexibility to incorporate the massive MIMO technology and, in turn, support a large number of users with a high data rate. As such, in dense urban areas, massive‐MIMO‐enabled mmWave links are highly suitable for backhauling 5G small cells. In the following section we discuss the key features of mmWave and massive MIMO technologies.

3.3 Fundamentals ofmmWave andMassive MIMO Technologies

3.3.1 MmWave Communication

As mentioned in the previous section, the capability of providing larger spectrum with reduced interference and a high data rate makes mmWave spectrum attractive forbackhauling ultra‐dense cellular networks. Nonetheless, the challenging features of mmWave propagation include the following:

• Atmospheric attenuation: The extremely high frequencies in the mmWave spec- trum are vulnerable to high propagation losses due to rain attenuation, oxygen or othermolecular absorption. However, these environmental factors may not cause significant propagation loss when considered for short‐range communications, for example, ultra‐dense small‐cell networks. The results in [8] imply a minimal impact

Backhauling with Massive-MIMO-Enabled mmWave Communication 33

of rain attenuation on mmWave frequencies. In times of heavy rainfall, the rain attenuation for 28 GHz is only 1.4 dB over 200 m distance [9]. The attenuation due to atmospheric absorption is also very low for mmWave frequencies, particularly for 28 GHz and 73 GHz [9]. In the case of 60 GHz, oxygen absorption may cause attenuation from 15 to 30 dB/km [10].

• Penetration losses: The mmWave frequencies experience high penetration loss on the exterior surfaces of urban buildings whereas the penetration loss for indoor materials is relatively low [9]. Thus, the outdoor BSs in the mmWave network can hardly serve indoor users.

• Reflection factor: Large numbers of obstructions may affect the reflected multi- paths of mmWave signals with large delays for both LOS and NLOS situations. Even then, strong signals can be received within 200 m distance in a highly NLOS environment [9]. Path‐loss exponents for the mmWave frequencies are within the following range: 3.2–4.58 for NLOS and 1.68–2.3 for LOS environments [9, 11].

For a viable implementation of mmWave backhauling, the existing system designs need to be improved. For instance, the high‐gain and electrically steerable antennas at the user and BSs with appropriate beamforming can produce highly directive beams and can combat large‐scale and small‐scale fading in the channel. The directivity gain can be improved further by using spatial multiplexing. To achieve high antenna gain in dense small‐cell networks, mmWave communication can combine polarization, adaptive beamforming and new spatial‐processing techniques such as massive MIMO. Recent advances in the design of integrated circuits and high‐directivity antennas confirm the availability of appropriate hardware solutions for achieving high gain with mmWave frequencies [12, 13].

3.3.2 MU‐MIMO withLarge Antenna Arrays

In dense small‐cell networks, MU‐MIMO technology can be used to provide backhaul connectivity to a large number of small cells simultaneously. Also, to ensure a high data rate for mmWave backhaul links, MU‐MIMO with a large number of antennas is highly beneficial. Ideally, there can be an infinite number of antennas that can support an infinite number of users at the same time [3]. However, in practice, the number of serving users is limited by the finite coherence time of the channel. The increased number of antennas effectively mitigates the fast fading and uncorrelated noise; however, the impact of interference cannot be ignored. The primary types of interference in a massive MIMO system include the following:

• Pilot contamination: In a massive MIMO system, orthogonal pilot sequences are typically used to train the massive MIMO BSs via uplink transmissions of the users. Due to uplink–downlink reciprocity in a time‐division duplex (TDD)

34 Backhauling/Fronthauling for Future Wireless Systems

massive MIMO system, the estimated channel information in the uplink is exploited for the downlink transmissions. Note that the fundamental limitation of the massive MIMO BSs arises from the reuse of pilot sequences in the neighbouring BSs, that is, transmissions of users in the neighbouring cells on the same pilot sequence. This pilot reuse contaminates the channel estimates of a user in a given cell. This phenomenon is referred to as coherent inter‐cell interference (or pilot contamination). Nonetheless, even in the presence of pilot contamination, massive MIMO BSs can serve a large number of users simultaneously, with asymptotic SIR coverage probability given in closed form in [14].

Note that the impact of pilot contamination tends to be more severe in multi‐tier networks; therefore, efficient pilot decontamination schemes will be of significant importance. Some of the pioneering works on pilot decontamination include [15] where multicast transmissions are used to eliminate the effects of pilot contamination.

• Traditional inter‐cell interference: Apart from pilot contamination, a user rate in a massive MIMO system is also affected by the traditional inter‐cell interference (i.e., non‐coherent interference).

• Intra‐cell or multi‐access interference: In MU‐MIMO, a BS serves multiple users simultaneously using spatial multiplexing. Thus, users form clusters within a cell where they share the same scattering environment with similar spatial channel correlations. Due to imperfect channel estimation and, in turn, imperfect precoding matrices, one communication stream can interfere with another stream within the same cluster. Nonetheless, efficient precoding/beamforming schemes can eliminate intra‐cell interference in MU‐MIMO systems.

With the increasing number of antennas in massive MIMO systems, the computational complexity for designing the precoders also increases. In this regard, there are several research works that aim to reduce the complexity of precoding design for massive MIMO systems. A hierarchical interference mitigation technique proposed in [16] utilizes an inner precoder for intra‐cell spatial multiplexing and an outer precoder for inter‐cell interference cancellation. The iterative algorithms for precoding can reduce the computational complexity significantly for massive MIMO systems. In [17], the authors propose a low‐complexity precoding technique based on truncated polynomial expansion (TPE) that provides a better approximation of matrix inversion and can be optimized to maximize the weighted max–min fairness for massive MIMOsystems.

3.4 MmWave Backhauling: State oftheArt andResearch Issues

Moving toward dense small cells in the 5G network requires a combination of backhauling techniques that may vary according to the locations, target QoS requirements and traffic load of the different SBSs. For example, indoor SBSs can obtain

Backhauling with Massive-MIMO-Enabled mmWave Communication 35

high‐capacity backhauls from the existing wired infrastructure. By contrast, it can be more complex and expensive to set up wired backhaul connectivity for outdoor SBSs as they can possibly be installed below roof‐top level, on exterior walls of buildings, street lamps or other street fixtures. In such scenarios, wireless backhauling solutions can serve the purpose. In this section, we will discuss and review the state of the art of LOS and non‐LOS mmWave backhauling in terms of feasibility, topology and implementation and will then detail the fundamental challenges that exist in massive‐MIMO‐enabled mmWave backhaul systems.

3.4.1 LOS mmWave Backhauling

3.4.1.1 Point‐to‐Point (PtP) Topology

PtP topology is the traditional method of wireless backhauling in cellular networks. The mmWave frequency ranging from 60–90 GHz can be used to form PtP beams mitigating the deleterious effects of oxygen/molecular absorption and rain attenuation. Due to the limitation of mmWave propagation, the inter‐site distance of picocells needs to be within 100 m for reliable PtP X2‐based backhaul links [18]. The mmWave frequencies allow the formation of two or more PtP links at the same location using highly directive antenna arrays.

3.4.1.2 Point‐to‐Multipoint (PtMP) Topology

PtMP topology could be another suitable alternative to PtP backhauling. The PtMP wireless links are based on a hub and remote concept [19]. For example, a small cell with fibre‐based backhaul connection can serve as a wireless backhaul hub and can support the backhaul transmission of six to eight small cells at a time. A PtMP in‐band backhauling system was proposed in [20], where mmWave spectrum was used in both the backhaul and access links. A time‐division multiplexing (TDM)‐based scheduling algorithm was proposed for PtMP mmWave backhauling, where SBSs are partitioned into three sectors. In this scheduling algorithm, the backhaul links and access links are scheduled simultaneously in an adaptive manner and the hub steers beams toward the neighbouring SBSs in one sector in each time slot.

3.4.1.3 Mesh Topology

In the case of some physical obstructions, the small cells might need multiple links for reliable backhauling; therefore, the flexibility of the backhauling system can be improved further by using a multi‐hop mesh network. In the multi‐hop mesh topology, long‐distance backhaul links are replaced by multiple short links, ensuring the

36 Backhauling/Fronthauling for Future Wireless Systems

reliability of the backhauling system. However, the processing and accessing delays at each hop may affect the performance of the backhauling links.

Recently, [5] considered a flexible mesh connectivity in mmWave backhauls with a strict bound on latency. Electrically steerable antenna arrays are used in this directed‐mesh connection operating at mmWave frequencies. Each node is capable of self‐tuning its parameters to obtain an optimal path that provides maximal throughput and minimal latency. To achieve this optimal performance, a TDM‐based joint scheduling and routing algorithm is implemented where the link parameters are updated in real time. The joint scheduling of transmission over access and backhaul links using mmWave frequency maximizes the spatial reuse while managing the intra‐cell and inter‐cell interference efficiently [21]. In addition to the densely deployed SBSs, the density of the wireless devices can also be very high in a 5G network. For such a scenario, the centralized MAC scheduling algorithm proposed in [21] suggests enabling direct device‐to‐device (D2D) transmission for optimal path selection and concurrent transmission scheduling over the access link and backhaul links. In some cases, their proposed algorithm can achieve near‐optimal performance in terms of throughput and latency.

The self‐backhauling architecture proposed in [22] demonstrates an improved multi‐hop mesh networking where a fraction of SBSs have wired backhaul and others are backhauled wirelessly. Each wired SBS provides backhaul links to multiple SBSs using mmWave frequency without imposing any interference. The authors characterized the mmWave network as noise‐limited where the interference power does not cause any harm for a moderate density of SBSs. The authors analysed network performance in terms of coverage rate for different combinations of the fraction of wired backhauled SBSs and mmWave backhauled SBSs. The results presented in [22] show that increasing the fraction of wired backhauled SBSs can improve the coverage rate significantly. However, if the density of the wired backhauled SBSs is kept constant, then the rate will eventually saturate for increasing density of the wireless backhauled SBSs. In the same self‐backhauling mmWave network, the authors also investigated the impact of the co‐existence of an ultra‐high‐frequency (UHF) network with a mmWave network.

3.4.2 NLOS mmWave Backhauling

In practice, it is likely that LOS backhaul links will be blocked by buildings or other surrounding objects. This renders the use of the mmWave frequency for backhauling more difficult. Also, such links are susceptible to rain attenuation, oxygen absorption and beam misalignment (due to wind, vibration and other environmental factors). As such, it is important to have accurate directed beamforming capability with high gains and subtle beam alignment for mmWave backhaul links. In the case of non‐LOS wireless backhaul links, the diffracted ray gives the propagation loss for the desired

Backhauling with Massive-MIMO-Enabled mmWave Communication 37

link and other reflected rays are treated as interfering links. To determine the gain for the desired link, the antenna array for each link is steered toward the point of diffraction. The mmWave frequency therefore enables a very narrow beam with high antenna gain, which, in principle, diminishes the spatial interference. However, the interference cannot be completely ignored in a scenario with ultra‐dense deployment of small cells where the probability of spatial interference is high.

The NLOS PtP backhauling model considered in [23] includes the effect of rain, oxygen absorption (for 60 GHz) and antenna misalignment. For their system model, the simulation results show that the high‐frequency links (60 and 73 GHz) form high‐gain narrower beams with a fading and implementation margin that can compensate for additional propagation losses (due to rain and other factors) to some extent. In [18], the authors proposed a high‐gain beam‐alignment technique using a hierarchical beamforming codebook which is computationally efficient. Their proposed framework adaptively samples the subspace and forms an optimal beam that maximizes the received SNR. In order to validate their framework they also investigated the wind effects on beam alignment using pole movement analysis. Their analysis also shows how frequently beam alignment needs to be performed. The large antenna arrays are sensitive to beam misalignment. Therefore, more research work on mmWave backhaul links is required to investigate the trade‐off between array size and achievable beamforming gain.

3.4.3 Research Challenges forBackhauling in5G Networks

As mentioned earlier, for ultra‐dense 5G networks, the mmWave frequency is envisioned as a key technology that can provide Gbps backhaul connectivity due to ample spectrum resources available in this frequency band. Additionally, due to highly directive beamforming gains, the integration of massive MIMO to the backhauling infrastructure can further enhance the reliability of the mmWave wireless backhaul links. Nonetheless, the successful roll‐out of the massive MIMO and mmWave technologies for wireless backhauling is hindered by several design and implementation issues. To this end, in this section, we discuss some of the existing and anticipated research issues in the design of massive‐MIMO‐enabled mmWave systems.

3.4.3.1 Provision ofSimultaneous Backhaul toMultiple SBSs

In ultra‐dense wireless‐backhauled small‐cell networks, system operators need to support the backhauls of several SBSs at the same time. This requires a pool of spectrum resources that can be efficiently allocated to various small cells at the same time such that the interference among backhaul streams remains below a prescribed limit. In this context, the ample amount of spectrum in the mmWave bands and its noise‐ limited nature could potentially serve the purpose, especially when closely located SBSs need to be backhauled. On the other hand, massive‐MIMO‐based backhauling

38 Backhauling/Fronthauling for Future Wireless Systems

is another technique which could potentially support the backhauls of multiple SBSs within the coverage of a massive‐MIMO‐enabled backhaul hub. This solution is more suitable for backhaul transmissions between SBSs and the core network since it exploits PtMP transmissions in the same time and frequency resource.

For multi‐user mmWave systems, multiple beams need to be formed at the same time, which necessitates efficient precoding schemes. Also, developing multi‐user hybrid analogue–digital precoding for mmWave is very challenging since it requires more processing at the digital layer to manage inter‐cell/intra‐cell interference [24]. mmWave transceivers are thus expected to be expensive and complex in design. Recently, a digitally controlled phase shifter network (DPSN)‐based hybrid precoding/combining scheme proposed in [25] was shown to be capable of reducing the cost and complexity of mmWave transceivers.

3.4.3.2 Acquisition ofChannel State Information (CSI)

With multi‐user massive MIMO technology and efficient beamforming, it becomes possible to serve the backhauls of a large number of SBSs with large degrees of freedom. However, CSI estimation is an underlying requirement which strongly impacts the performance of beamforming in massive MIMO systems. Traditionally, in massive MIMO systems, the channels between transmitters and receivers are estimated from orthogonal pilot sequences. These sequences are, however, limited in number due to the finite coherence time of the channel. Therefore, it becomes crucial to reuse the same pilot sequences in a multi‐tier network. This reuse of pilot sequences in different cells leads to pilot contamination that limits the rate gains of a massive MIMO system. To overcome this issue, coordinated multi‐point transmissions (CoMP) can be employed. Also, a set of antenna elements in a massive MIMO system can be leveraged to mitigate the impact of pilot contamination.

3.4.3.3 Adaptive Backhaul/Access Spectrum Selection

Traditionally, the network operators optimize the subchannels of a typical RF spectrum (sub‐6 GHz) that are allocated to a given user at its access links given the wired backhaul at BSs. However, the possible use of a combination of frequency bands in 5G access/backhaul networks such as microwave, mmWave and sub‐6 GHz renders this task particularly challenging. The reason is that the incurred interference and offered network capacity at different frequency bands can be different for various indoor/outdoor environments. For instance, compared to the traditional sub‐6 GHz band, the mmWave frequencies have high penetration/attenuation losses that can vary significantly for indoor and outdoor propagation environments. A careful system‐level analysis is therefore required to adaptively select an appropriate combination of frequencies for access/backhauls taking into account the crucial

www.ebook3000.com

Backhauling with Massive-MIMO-Enabled mmWave Communication 39

factors such asinterference conditions at a given frequency, locations of the SBSs, their surrounding environment, transmit/receive BS antenna characteristics and the beamforming gains.

Moreover, the possibility of reusing the same frequency at both the backhaul and access links (i.e., in‐band backhauling) with minimal interference is also crucial and can vary for different spectrum bands. For example, the high directivity and noise‐limited nature of mmWave spectrum can strongly support in‐band backhauling compared to traditional sub‐6 GHz frequency spectrum. Consequently, choosing a feasible spectrum band that can enhance data rates via in‐band backhauling is also crucial.

3.4.3.4 Backhaul Spectrum‐aware User Association

As mentioned previously, different frequency bands may result in significantly varying data rates even in similar system settings. Therefore, optimal spectrum selection to serve a user under given circumstances is crucial and requires the operation of SBSs in a variety of frequency bands [26]. Nonetheless, deploying such SBSs may not be a trivial task due to hardware modifications as well as deployment costs. Therefore, from the perspective of network operators, the significance of determining an efficient, low‐complexity, traffic‐offloading criterion is evident. Also, from the perspective of users, it is important to choose a user‐association criterion that considers propagation losses for the frequency used and antenna‐specific parameters of the SBSs.

3.4.3.5 Backhaul/Access Link Scheduling

In practice, the massive‐MIMO‐enabled backhaul system can serve a limited number of backhaul streams at a given time and frequency resource. Consequently, the significance of backhaul scheduling becomes evident in dense cellular networks. In addition to traditional time‐division‐based scheduling, more sophisticated scheduling techniques can be implemented for mmWave massive MIMO networks. For example, in [25], the proposed beam‐division‐multiplexing (BDM)‐based scheduling is shown to improve performance gain for in‐band mmWave backhauling. Therefore, the combination of TDM and BDM can make the user‐scheduling procedure much more flexible. Moreover, unlike traditional user scheduling, the backhaul scheduling requires consideration of the traffic load of a serving SBS and average achievable data rate at the access links in addition to the backhaul channel conditions.

3.4.3.6 Number ofRF Chains

Typically, MIMO systems are equipped with few antennas and, in turn, the number of employed RF chains, digital‐to‐analogue converters (DACs) and analogue‐to‐digital converters (ADCs) can be comparable to the number of antennas. However, in a

40 Backhauling/Fronthauling for Future Wireless Systems

massive MIMO system, deploying RF chains in a comparable quantity is not practically feasible. Due to this limitation, the orthogonal channel estimation approach may not be used even without pilot contamination. Also, the energy consumption of a MIMO transceiver increases as a function of the number of active RF chains due to the high energy consumption of ADCs. Thus, the channel estimation and beamforming algorithms should be designed taking into account the constraints on the number of RF chains.

3.4.3.7 Other Implementation Issues

The directed multi‐gigabit (DMG) PHY specification in the IEEE standard 802.11ad suggests OFDM modulation for high‐data‐rate applications. As mmWave propagation characteristics are quite different from those for the microwave, the OFDM parameters for 3GPP LTE need to be modified. For mmWave communication, OFDM PHY has different frame structures and can be implemented using QPSK, 16‐QAM and 64‐QAM. The bandwidth of OFDM subcarriers and guard intervals needs to be larger. The basic TDD‐based mmWave frame has 10 subframes and each subframe contains 14 OFDM symbols [27]. When mmWave is used for wireless backhauling, subframes can be configured to support multi‐hop transmission utilizing spatial multiplexing [27]. The configuration of mmWave subframing needs further investigation.

3.5 Case Study: Massive‐MIMO‐based mmWave BackhaulingSystem

As depicted in Figure3.1, we consider wireless backhaul hubs installed within macro base stations (MBSs) to provide backhaul connectivity to SBSs or access points (APs) over mmWave links utilizing massive MIMO technology. The APs use different mmWave frequencies in the access and backhaul links. We assume that the APs and user equipment (UEs) are equipped with directional antennas with sectorized gain pattern. Each hub supports N time block, each AP also schedules N

and Pa, respectively. During the training phase, each AP sends a preassigned

are P

orthogonal pilot sequence to the hub, which is estimated perfectly by the hub and the pilot sequence is not used by any other neighbouring hub (i.e., no pilot contamination is assumed). The channel estimation is facilitated with time‐division duplexing at the hubs so that the channel reciprocity is guaranteed.

Since the walls of buildings are impenetrable to mmWave signals, indoor APs can neither serve outdoor users nor interfere with outdoor AP transmission. In the outdoor environment, the APs are deployed on the exterior of buildings or street fixtures where users are more likely to be NLOS to the hubs and users. We focus on outdoor user performance and assume that the hub–AP and AP–user channels fade independently across time slots.

APs simultaneously with Mh antennas. Within the same

UEs. The transmission power of a hub and AP

Backhauling with Massive-MIMO-Enabled mmWave Communication 41

{}H

{}

()

{}

()

i()

j()

Core network

MBS/

wireless

backhaul

hub

Massive MIMO

based mmWave

backhaul link

Macro cell

Small cell

Blockage

NLOS link

MBS/

wireless

backhaul

hub

Figure3.1 Small‐cell network architecture incorporating massive‐MIMO‐based mmWave backhauling system

3.5.1 System Model

We consider that backhaul hubs and APs are distributed according to homogeneous Poisson point processes (PPPs) in a cellular region ℝ and APs are

with density λH and

The locations of UEs are approximated by a PPP denote by

the jth AP backhauled by the ith hub and

jth AP. The buildings and other outdoor blockages are modelled using the Boolean

scheme of rectangles [28]. In this model, the blockages are considered as a process of random rectangles where the centres of the rectangles form a PPP Φ The length, width and orientation of the rectangles are determined as independent and identically distributed (i.i.d.) random variables. We also assume that the distribution of blockages is stationary and does not vary with any translation or rotation. Over the geographic area, the probability that a network node will be in the LOS region is determined based on the size and density of the blockages [29]. Considering downlink transmission, we denote the distance between a transmitting node {X

. The PPPs formed by the hubs

with density λA, respectively.

with density λU. We

as the kth UE served by the

with density λb.

} and

42 Backhauling/Fronthauling for Future Wireless Systems

tHA

,{}

rAU

{,}

DD D

tttrrr

()

() ()

tr tr,

()

tr,

()

b nn({ })1 234,,,

j()

i()

()

j()

()

DagG



tr,

c c

()1

receiving node {Yr} as Rl, where

and

. The LOS proba-

bility for the link is given by:

pR e

(3.1)

where ρ is the average LOS distance.

3.5.1.1 Directivity Gain

We model directional beamforming gains as marks of the PPPs and approximate the actual beamforming pattern using the sectored gain pattern presented in [22, 28, 30] for ad hoc and mmWave networks. The antenna pattern is represented as D

(ϕ), where G

G,g,θ

is the main lobe directivity gain, g is the side‐lobe directivity gain and θ is the beam width of the main lobe with the antenna boresight direction ϕ at the node. For a given angle of departure

gain for the link R

) between the target nodes, the directivity gain is

at the transmitter and angle of arrival

is determined as

at thereceiver, the total directivity

,, ,,

. For the desired link

. The number of beams

from interfering nodes is considered to be a discrete random variable.

Considering the probability distribution for four possible scenarios, the directivity

gain of the interfering links is

and bn constants as proposed in [29]. The probability distribution for the inter-

use a

fering links is shown in Table 3.1. Here,

with probability

and

. For example,

, where we

associated with in the downlink represents that the main lobe of

and the random distribution of interference link from

is overlapped by a side lobe of

. The directivity gain of the interfering link will then be

probability

()

with

3.5.1.2 Path‐loss Model

MmWave links may experience either LOS or NLOS propagation. Therefore, the path loss experienced by the link R

RpRKRpRKR

Table3.1 Probability distribution function of

i 1 2 3 4

GtG

ctc

can be determined as:

1 (3.2)

gtG

Gtg

gtg

Backhauling with Massive-MIMO-Enabled mmWave Communication 43

(( ))pR

N M

�

i()

where

is the Bernoulli RV for the LOS probability p(Rl) and αL and αN are the LOS and NLOS path‐loss exponents. The log‐normal shadowing and path loss at a reference distance are approximated as K respectively. We assume small‐scale fading h

and KN for LOS and NLOS,

as a normalized gamma random

variable.

3.5.1.3 User Association

We consider that each UE can associate with one AP which is backhauled by at most one hub. The association between AP and UE is indicated by:

10,

if associated with

otherwise

Resources at the APs are uniformly allocated to a maximum of Q

UEs according to

the effective load. The percentage of resource consumption for a UE is denoted as

. The hub–AP association indicator is considered as:

10,

if associated with

otherwise

i1()

APs. The

Resources at the hubs are uniformly allocated among a maximum of Q

percentage of resource consumption for an AP is denoted as

3.5.1.4 Interference Model

As we focus on a massive MIMO network with a large number of antennas implemented in hubs, where

, the large‐scale antenna arrays diminish the effects of uncorrelated noise and the system is considered to be interference limited [3]. We assume that efficient beamforming and a pilot decontamination scheme eliminate the effects of intra‐cell interference and pilot contamination. Thus, an AP will receive interference only from other hubs that are transmitting in the downlink. The interference received by

(the jth AP which is backhauled by the ith hub) is expressed as

follows:

lH H

02:\

Ii i

Ph DLR

lij

. (3.3)

44 Backhauling/Fronthauling for Future Wireless Systems

j()

i()

j()

The access link interference between the user element

and access point

be expressed as follows:

lA A

02:\

Ph DLR

ljkll

kj kj

3.5.1.5 SINR andRate Calculation

In the massive MIMO regime, the SINR at the backhaul link for AP imated as follows [31]:

lij

(3.5b)

The SINR received at

where

is the thermal noise power. The user rate in the downlink is therefore cal-

in the access link is given by:

Ph DLR

ljk

INR

Ph DLR

culated as:

can

(3.4)

can be approx-

(3.5a)

RBkjlog min

1SINR,SINR

(3.6)

where B is the bandwidth of the channel assigned to the user.

3.5.2 Maximizing User Rate

In the downlink transmission, the hubs forward the traffic to the APs over the backhaul links and then the APs forward the traffic to the desired users. The hubs can support a fixed number of APs simultaneously and APs also have the constraint to support a limited number of users at a time. Therefore, the downlink rate is greatly affected by the association at the backhaul and access links. In this context, we formulate a user‐association problem for the downlink transmission to maximize the overall user rate as follows:

ji kj

kj kj kj

(3.7a)

Backhauling with Massive-MIMO-Enabled mmWave Communication 45

12,,...,

 

aa() ,1

aa()

() ,

h ab()

ha()

.. ,, ,

11 (3.7b)

,,, (3.7c)

kj j

Since solving this optimization problem can be computationally complex in real time, we focus on devising less complex distributed user‐association solutions. The theory of matching can be used to design efficient distributed solutions for user associations [32,33].

3.5.3 Matching Theory forUser Association

We address the combinatorial problem of user association in the downlink by transforming it to a two‐sided matching game. For two disjoint sets of players in the matching game, a matching is performed between the two sets of players according to each player’s preference metric. To obtain a stable matching, each player is assigned a fixed quota which is referred to as the maximum number of players that each player can be matched to. To solve the user‐association problem in our system model, the association in the backhaul and access links is performed using the matching framework presented in [33–35].

3.5.3.1 Matching Game forHub–AP Association

At first we consider a many‐to‐one matching game for the association between the hubs and APs. We denote the set of hubs by

H and the set of APs by

A . Each AP can be matched with at most one hub and each hub can

have matching with one or more APs depending on its quota Q

. The preference

relations for hubs and APs depend on the channel state of each backhaul link. Each AP aims to maximize its utility function given by its achievable rate. Considering a random association, each AP calculates the rate that it can receive from each hub and makes its preference list

its preference list

using the rates calculated from the initial random

over the hubs. Similarly, each hub constructs

association. Each hub gives a ranking score to each AP based on its preference list.

After constructing the preference lists, matching among the disjoint sets of hubs and APs is performed iteratively until a stable match is found. A matching μ

that

represents the hub–AP association can be expressed as a function from the set

•

hQ h

•

into the set

and

if and only if

such that [33, 35]:

46 Backhauling/Fronthauling for Future Wireless Systems



{, ,..., }12 A

{, ,..., }12 U

 

{}

Here, the matching function μb(a) denotes the matched APs and μb(h) represents the matched hubs. As described in Algorithm 1 below, the APs first apply to their most preferred hubs according to the preference list. If the quota of the hub Q

is not

overloaded, it will schedule the AP. If the quota is exceeded, the hub checks the ranking for the applicant AP. If the ranking of the AP is better than the rank of other scheduled APs, then it discards the worst AP with maximum rank and associates with the applicant AP. This will continue until all APs are associated and their preference lists become empty. Thus, the algorithm will end with a stable matching

Algorithm 1: Matching algorithm forhub–AP association

1. input: ΦH, ΦA and Q

2. initialization: calculate the preference lists

and

respectively

3. each hub

4. while (at least one AP is free AND its preference list

gives a ranking score to APs based on

is not empty) do

5. each unassociated AP applies to its most preferred hub

6. if the Qh is not overloaded then

7. associate with the applicant AP

8. else

9. compare the ranking of the applicant with the

currently associated APs

10. if

then

11. discard the worst AP from the hub's current associations

12. associate with the applicant AP

13. discard the hub from the preference list of the

discarded AP

14. set the discarded AP as free

15. end if

16. end if

17. end while

18. output:

3.5.3.2 Matching Game forAP–UE Association

Once we have the stable matching

for the backhaul connectivity, we perform another many‐to‐one matching game to find association among APs and UEs at the access end. We consider the set of APs as

and the set of UEs as

. This matching game aims to match each UE with one AP and each

hub with one or more UEs depending on its quota Q

. Using the stable matching

at the backhaul end and a random association at the access end, each UE calculates the rate that it can receive from each AP using Equation (3.6) and makes its preference list

over the APs. APs also construct their preference list

Backhauling with Massive-MIMO-Enabled mmWave Communication 47

 

{}

uu() ,1

uu()

() ,

()

au()



depending on the total rate they can provide to the associated users. Each AP also gives a ranking score to UEs based on its preference. To find a stable matching described in Algorithm 2. In this case, a matching μ association is expressed as a function from the set

for the AP–UE association, matching is performed iteratively as

representing the AP–UE

into the set

such that:

•

aQ a

•

if and only if

and

Here, the matching function μ

(u) denotes the matched UEs and μb(a) represents

the matched APs. The two‐sided matching algorithms to find the hub–AP and AP– UE association are guaranteed to converge as the applicant (AP or UE) never applies to its preferred node (hub or AP) twice and thus the algorithm will have a finite number of iterations [32]. The algorithms return the stable matching

and

whenever any unassociated AP or UE has no hub or UE left in its preference list,

respectively.

Algorithm 2: Matching algorithm forAP–UE association

1. input: ΦA, ΦU and Q

2. initialization: calculate the preference lists

respectively

3. each AP

gives a ranking score to UEs based on

4. while (at least one UE is free AND its preference list

is not empty) do

5. each free UE Uk applies to its most preferred AP in

6. if the Qa is not overloaded then

7. associate with the applicant UE

8. else

9. compare the ranking of the applicant with the

currently associated UEs

10. if

11. discard the worst UE from the AP's current

associations

12. associate with the applicant UE

13. discard the AP from the preference list of the

discarded UE

14. set the discarded UE as free

15. end if

16. end if

17. end while

18. output:

and

then

48 Backhauling/Fronthauling for Future Wireless Systems

141 4.

100

300

0.8

SINR threshold in dB

SINR coverage probability

3.5.4 Numerical Results

In this section, we present numerical results for the performance of a typical user in the downlink with traditional distance‐based association and distributed stable matching algorithms. The massive‐MIMO‐based mmWave backhaul hub is assumed to be equipped with 256 antennas and operates at 60 GHz with channel bandwidth of 2 GHz. The AP operates at 73 GHz and the channel bandwidth is 2 GHz. The transmission power for the hub and AP are 46 dBm and 30 dBm, respectively. To approximate the directivity gains for the hubs, AP and UE, we consider a 10 degree beam width for the hub and AP with a main‐lobe gain of 18 dB and a side‐lobe gain of −4 dB for both the mmWave hubs and APs. The directive beamforming for users is approximated by 10 dB beams with 90 degree beam width. We assume the average LOS distance access and backhaul) are taken as 2 and 3.5, respectively.

First, in Figure3.2, we compare the SINR coverage probability for a massive‐ MIMO‐based mmWave backhaul system for the backhaul link and access link with both microwave and mmWave frequencies. Here, we consider

per sq. km with ilar blockage for microwave transmission. As shown in Figure 3.2, mmWave frequencies offer a better SINR coverage than microwave transmission in the access links.

m. The path‐loss exponents for LOS and NLOS links (both

and

. For fair comparison, we include sim-

Figure3.2 SINR coverage probability for mmWave backhaul link, mmWave and microwave access link

0.7

0.6

0.5

0.4

0.3

0.2

0.1

−10 0 10 20 30 40 50

microwave access link (2.7 GHz) mmWave access link (73 GHz) mmWave backhaul link (60 GHz)

Backhauling with Massive-MIMO-Enabled mmWave Communication 49

100

(a) (b)

AP density (per sq. km)

Average user rate (in Gbps)

800

AP density (per sq. km)

Since mmWave transmission uses a wider channel bandwidth than microwave transmission, the user rate will be higher for mmWave communication. The comparison of average user rate between mmWave and microwave transmission shown in Figure3.3 indicates that microwave transmission is unable to provide a Gbps data rate to the user end as expected from 5G small‐cell networks. The mmWave frequency can provide a multi‐Gbps data rate on average even for densely deployed SBSs. We also observe that the user rate depends on the backhaul quota (Q

) or

maximum limit of the hubs. In Figure3.3, it is shown that a higher average user rate can be achieved by increasing the backhaul quota which enables hubs to provide more backhaul links to the APs.

To analyse the impact of association schemes on user rates, we compare the downlink rates of all users for conventional nearest AP association and stable matching association. Figure3.4 shows the total network rate and average user rates for different backhaul quotas (Q with AP quota

) and association schemes. Here, we consider

for simulation. We observe that when the competition for

and

AP–UE association is less for lower user density, the nearest AP association performs better than the stable‐matching‐based association. As user density increases, the competition for association with the desired AP also increases. In such scenarios, the overall user rate increases for the stable matching association when compared to the distance‐dependent association due to several factors.

0 200 400 600 800

mmWave rate with Qh = 20

mmWave rate with Qh = 10

Figure3.3 Average user rate for (a) mmWave and (b) microwave access link with different backhaul quotas (Qh)

300

250

200

150

100

Average user rate (in Mbps)

0 200 400 600

Microwave rate with Qh = 20

Microwave rate with Qh = 10

50 Backhauling/Fronthauling for Future Wireless Systems

(a) (b)

User density (per sq. km) User density (per sq. km)

Rate of all users (in Gbps)

600

7000

6000

5000

4000

3000

2000

1000

100 200 300 400 500 600

Stable matching, Qh =20 Stable matching, Qh =10 Nearest AP association, Qh = 20 Nearest AP association, Qh = 10

Figure3.4 (a) Network rate and (b) average user rate for different user‐association schemes

In the case of distance‐dependent backhaul link association, densely deployed APs compete to get associated with the nearest backhaul hub. The hub associates with the APs according to the arrival of the association requests. Once the quota is exceeded, the hub does not have the flexibility to alter the association even if it knows that a better rate can be achieved. An AP cannot get backhaul connectivity if the nearest hub is overloaded. Similarly, for access link association, the UEs send their association requests to their nearest APs and get associated if the AP quota is not exceeded. When the competition for user association increases, more users are rejected. Also, the distance‐dependent association may result in poor SINR at the receivers in some scenarios. For example, if the target AP’s nearest backhaul hub is NLOS or quite far away, then downlink transmission to the AP may experience poor SINR.

On the other hand, stable matching is performed utilizing the channel state information. The preference lists for hubs, APs and UEs are calculated based on the rate information. During backhaul link association, the stable matching algorithm allows the overloaded hub to recheck the ranking or the rates it can deliver to APs it is currently associated with. The hub can update its association if it can provide a better rate to the requesting AP. The stable matching algorithm also provides an opportunity for an AP to apply to the next preferred hub if it is rejected by its most preferred hub. The next preferred hub may be able to support this requesting AP with a better rate. Consequently, this will allow the AP to serve more UEs in the network. Similarly, the stable matching algorithm for the AP–UE association, which exploits the channel state information, will also enable UEs to achieve better user rates at the access links when compared to those with the nearest AP association.

Average user rate (in Gbps)

100 200 300 400 500

Stable matching, Qh =20 Stable matching, Qh =10 Nearest AP association, Qh = 20 Nearest AP association, Qh = 10

Backhauling with Massive-MIMO-Enabled mmWave Communication 51

3.6 Conclusion

We have discussed wireless backhaul solutions for future 5G small‐cell networks. In order to provide backhaul connectivity to a large number of SBSs in such a network in a cost‐effective way, mmWave spectrum is beneficial due to the ample amount of unlicensed spectrum and high directivity. The performance of mmWave propagation can be enhanced by using massive MIMO through installing a large number of antennas in the backhaul system. We have described a tractable system model for a massive‐MIMO‐enabled mmWave backhauling network. To maximize the overall user rate, we have formulated a matching game which provides stable matching association at the backhaul and access ends. Numerical results have shown that the massive‐MIMO‐enabled mmWave backhaul solution provides a reliable SINR coverage probability and the use of mmWave in the access link increases the average user rate compared to traditional microwave. Analysis of the stable matching user association has shown that channel‐state‐aware matching can provide a better user rate compared to distance‐dependent user association in a dense small‐cell network.

The insights from the results motivate the design of a backhaul‐aware user association scheme so that the overall network is maximized in a dense small‐cell network. To improve the network performance further, massive MIMO can be incorporated at the APs so that more users can be served simultaneously with higher antenna gain.

Acknowledgement

This work was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC).

References

[1] Andrews, J. G., Buzzi, B., Choi, W., Hanly, S. V., Lozano, A., Soong, A. C. K. and Zhang, J. C.

(2014) What will 5G be?. IEEE Journal on Selected Areas in Communication, 32(6), 1065–1082.

[2] Rangan, S., Rappaport, T. and Erkip, E. (2014) Millimeter‐wave cellular wireless networks:

Potentials and challenges. Proceedings of the IEEE, 102(3), 366–385.

[3] Marzetta, T. L. (2010) Noncooperative cellular wireless with unlimited numbers of base station

antennas. IEEE Transactions on Wireless Communications, 9(11), 3590–3600.

[4] Siddique, U., Tabassum, H., Hossain, E. and Kim, D. I. (2015) Wireless backhauling of 5G small

cells: Challenges and solution approaches. IEEE Wireless Communications, Special Issue on ‘Smart Backhauling and Fronthauling for 5G Networks’, 22(5), 22–31.

[5] Interdigital (2013) ‘Small Cell Millimeter Wave Mesh Backhaul.’ White paper. Interdigital,

Wilmington, DE, USA. Available at: http://goo.gl/Dl2Z6V.

[6] Li, Y., Pappas, N., Angelakis, V., Pioro, M. and Yuan, D. (2015) Optimization of free space optical

wireless network for cellular backhauling. IEEE Journal on Selected Areas in Communication,

33(9), 1841–1854.

52 Backhauling/Fronthauling for Future Wireless Systems

[7] Demers, F., Yanikomeroglu, H. and St‐Hilaire, M. (2011) A survey of opportunities for free space

optics in next generation cellular networks. In Proceedings of the Communication Networks and Services Research Conference, pp. 210–216, May.

[8] Zhao, Q. and Li, J. (2006) Rain attenuation in millimeter wave ranges. In Proceedings of the IEEE

International Symposium on Antennas, Propagation and EM Theory, pp. 1–4, October.

[9] Rappaport, T., Sun, S., Mayzus, R., Zhao, H., Azar, Y., Wang, K., Wong, G., Schulz, J., Samimi, M.

and Gutierrez, F. (2013) Millimeter wave mobile communications for 5G cellular: It will work! IEEE Access, 1, 335–349.

[10] Daniels, R. C. and Heath, R. W. (2007) 60 GHz wireless communications: Emerging requirements

and design recommendations. IEEE Vehicular Technology Magazine, 2(3), 41–50.

[11] Roh, W., Seol, J., Park, J., Lee, B., Lee, J., Kim, Y., Cho, J., Cheun, K. and Aryanfar, F. (2014)

Millimeter‐wave beamforming as an enabling technology for 5G cellular communications: Theoretical feasibility and prototype results. IEEE Communications Magazine, 52(2), 106–113.

[12] Rappaport, T. S., Murdock, J. N. and Gutierrez, F. (2011) State of the art in 60‐GHz integrated

circuits and systems for wireless communications. Proceedings of the IEEE, 99(8), 1390–1436.

[13] Dussopt, L., Bouayadi, O. E., Luna, J. A. Z., Dehos, C. and Lamy, Y. (2015) Millimeter‐wave

antennas for radio access and backhaul in 5G heterogeneous mobile networks. European Conference on Antennas and Propagation, pp. 1–4, April.

[14] Bai, T. and Heath, R. W. (2014) Asymptotic coverage probability and rate in massive MIMO net-

works. In Proceedings of the IEEE GlobalSIP, pp. 602–606, December.

[15] Xiang, Z., Tao, M. and Wang, X. (2014) Massive MIMO multicasting in noncooperative cellular

networks. IEEE Journal on Selected Areas in Communication, 32(6), 1180–1193.

[16] Liu, A. and Lau, V. (2014) Hierarchical interference mitigation for massive MIMO cellular net-

works. In IEEE Transactions on Signal Processing, 62(18), 4786–4797.

[17] Kammoun, A., Muller, A., Bjornson, E. and Debbah, M. (2014) Low‐complexity linear precoding

for multi‐cell massive MIMO systems. In Proceedings of EUSIPCO, pp. 2150–2154, September.

[18] Hur, S., Kim, T., Love, D. J., Krogmeier, J. V., Thomas, T. A. and Ghosh, A. (2013) Millimeter

wave beamforming for wireless backhaul and access in small cell networks. IEEE Transactions on Communication, 61(10), 4391–4403.

[19] Wireless 20/20 (2012) ‘Rethinking Small Cell Backhaul: A Business Case Analysis of Cost‐

Effective Small Cell Backhaul Network Solutions.’ White paper, July. Available at: http://www. wireless2020.com/docs/RethinkingSmallCellBackhaulWP.pdf

[20] Taori, R. and Sridharan, A. (2015) Point‐to‐multipoint in‐band mmWave backhaul for 5G net-

works. IEEE Communications Magazine, 53(1), 195–201.

[21] Niu, Y., Gao, C., Li, Y., Su, L., Jin, D. and Vasilakos, A. V. (2015) Exploiting device‐to‐device

transmissions in joint scheduling of access and backhaul for small cells in 60 GHz band. IEEE Journal on Selected Areas in Communication, 33(10), 2052–2069.

[22] Singh, S., Kulkarni, M. N., Ghosh, A. and Andrews, J. G. (2015) Tractable model for rate in self‐

backhauled millimeter wave cellular networks. IEEE Journal on Selected Areas in Communication, 31(10), 2196–2211.

[23] Coldrey, M., Koorapaty, H., Berg, J., Ghebretensa, Z., Hansryd, J., Derneryd, A. and Falahati, S.

(2012) Small‐cell wireless backhauling: A non‐line‐of‐sight approach for point‐to‐point microwave links. In Proceedings of the IEEE Vehicular Technology Conference, pp. 1–5, September.

[24] Alkhateeb, A., Mo, J., Gonzalez‐Prelcic, N. and Heath, R. W. (2014) MIMO precoding and

combining solutions for millimeter‐wave systems. IEEE Communications Magazine, 52(12), 122–131.

[25] Gao, Z., Dai, L., Mi, D., Wang, Z., Imran, M. A. and Shakir, M. Z. (2015) MmWave massive‐

MIMO‐based wireless backhaul for the 5G ultra‐dense network. IEEE Wireless Communications, 22(5), 13–21.

Backhauling with Massive-MIMO-Enabled mmWave Communication 53

[26] Wang, N., Hossain, E. and Bhargava, V. K. (2015) Backhauling 5G small cells: A radio resource

management perspective. IEEE Wireless Communications, 22(5), 41–49.

[27] Zheng, K., Zhao, L., Mei, J., Dohler, M., Xiang, W. and Peng, Y. (2015) 10Gbps‐HetSNets with

millimeter‐wave communications: Access & networking challenges and protocols. IEEE Communications Magazine, 53(1), 222–231.

[28] Bai, T., Vaze, R. and Heath, R. W. (2014) Analysis of blockage effects on urban cellular networks.

IEEE Transactions on Wireless Communication, 13(9), 5070–5083.

[29] Bai, T. and Heath, R. W. (2015) Coverage and rate analysis for millimeter‐wave cellular networks.

IEEE Transactions on Wireless Communication, 14(2), 1100–1114.

[30] Hunter, A., Andrews, J. and Weber, S. (2008) Transmission capacity of ad hoc networks with

spatial diversity. IEEE Transactions on Wireless Communication, 7(12), 5058–5071.

[31] Bethanabhotla, D., Bursalioglu, O., Papadopoulos, H. C. and Caire, G. (2014) User association and

load balancing for cellular massive MIMO. In Proceedings of Information Theory and its Applications (ITA), pp. 1–10, February.

[32] Roth, A. E. and Sotomayor, M. A. O. (1992) Two‐sided matching: A study in game‐theoretic mod-

eling and analysis, Cambridge University Press, pp. 485–541.

[33] Jorswieck, E. (2011) Stable matchings for resource allocation in wireless networks. In Proceedings

of the 17th International Conference on Digital Signal Processing (DSP), pp. 1–8, July.

[34] Semiari, O., Saad, W., Dawy, Z. and Bennis, M. (2015) Matching theory for backhaul management

in small cell networks with mmWave capabilities. In Proceedings of the IEEE ICC, pp. 3460–3465, June.

[35] Sekander, S., Tabassum, H. and Hossain, E. (2015) A matching game for decoupled uplink–down-

link user association in full‐duplex small cell networks. In IEEE Globecom’15, December.

Fronthaul foraFlexible Centralization inCloud Radio Access Networks

Jens Bartelt,1 Dirk Wübben,2 Peter Rost,3 Johannes Lessmann4 andGerhardFettweis

Technische Universität Dresden, Dresden, Germany

University of Bremen, Bremen, Germany

Nokia Networks, Munich, Germany

NEC Laboratories Europe, Heidelberg, Germany

4.1 Introduction

The architecture of fourth generation (4G) mobile networks, for example 3GPP long‐term evolution (LTE), is organized in a decentralized way, such that the complete baseband processing including the physical (PHY) layer, medium access (MAC) layer and parts of the network layer processing are performed at the base stations (BSs). Internet Protocol (IP)‐layer user data are then forwarded between the BS and the network core, which requires a relatively modest transport network, usually known as the backhaul (BH) network.

An alternative to this decentralized concept is to centralize radio access network (RAN) functionalities. First introduced in [1], the so‐called centralized‐RAN or cloud‐RAN (C‐RAN) architecture proposes to reduce the functionality of BSs to so‐ called remote radio heads (RRHs), which only perform analogue processing and forward digital samples between the RRH and centralized baseband units (BBUs). Such a centralized architecture is already utilized in some 4G networks and is actively

56 Backhauling/Fronthauling for Future Wireless Systems

being considered for future mobile networks as it offers several advantages. First among these advantages is a reduction in operational and capital expenditure. By reducing the size, the sites can be smaller and the energy consumption can be reduced, particularly because no active cooling is required. As all BBUs are located in a centralized location, maintenance also becomes much easier.

Next, the centralization of BBUs makes cooperative processing techniques much easier to implement. Techniques such as coordinated multi‐point (CoMP) [2, 3] or multi‐point turbo detection (MPTD) [4] require an extensive exchange of signals between BSs and suffer heavily from delays on BH links. Hence, they are much easier to implement if all signals are forwarded and processed in a centralized BBU.

Finally, current advances in processor technology and virtualization have enabled the implementation of baseband processing on general‐purpose processors (GPPs). The early deployments of C‐RAN utilized centralized BBUs composed of dedicated hardware such as ASICs (application‐specific integrated circuits), FPGAs (field‐ programmable gate arrays) and DSPs (digital signal processors). However, this approach merely amounts to a physical separation of the analogue front‐ends from the digital basebands that would be co‐located in the decentralized architecture. As a result, a separate BBU is required per BS and signals need to be exchanged between the BBUs, which still makes joint processing difficult. More recent approaches [5] therefore advocate the use of more flexible GPPs, commonly found in PCs or servers. Thereby, virtualization and the concept of cloud computing as used in the IT industry [6] can be facilitated. While this further simplifies maintenance, upgrades and organization, it also introduces the economy of scale, as standardized hardware can be used, further lowering capital expenditure (CAPEX). Additionally, virtualization allows balancing processing load between the different BSs according to their load variations, thereby lowering the total processing power that needs to be deployed.

These benefits of C‐RAN come at the price of a demanding transport network. In the C‐RAN architecture, the transport network forwards samples between the BBUs and the RRHs, and is commonly known as the fronthaul (FH) network. In order to enable efficient, centralized and cooperative processing, the FH network must offer a huge capacity and low total latency and jitter. In fact, if not dimensioned correctly, the FH network could become a bottleneck for the performance of future networks. Furthermore, FH networks are very expensive to deploy, thus making a simple overprovisioning not economically viable. Several approaches have been investigated to reduce these FH requirements, for example by compressing FH data [7] or reducing costs by a joint optimization of the RAN and BH/FH network [8]. In this chapter, we want to describe another promising approach to address the FH challenge in future networks: a flexible centralization of RAN functionality [5, 9]. As indicated above, the current fully decentralized or fully centralized architectures are two extreme approaches. Utilizing a flexible functional split, a part of the baseband processing would be located in the BS and another in a centralized processing unit. In this chapter,

Fronthaul for a Flexible Centralization in C-RANs 57

User Radio

Base

Fronthaul

Cloud

Backhaul Core

we will discuss how this approach can reduce the requirements on the FH network while still partially maintaining the benefits of centralization [10]. For this, we will first introduce the considered network architecture and explain different options for a flexible centralization. We will further describe the technologies that will help to enable this approach, and describe how such a flexible split calls for a convergence of the FH and BH network to a unified transport network.

4.2 Radio Access Network Architecture

Figure4.1 shows the architecture of a contemporary mobile network and illustrates the difference between fully decentralized, fully centralized and flexible architectures. Users (called user equipment (UE) in LTE) are located on the edge of the network and ultimately want to communicate with the core on the other edge of the network. The core then routes the traffic to the destination via gateways, and is also responsible for network management and policy control. To reach the core, the UE communicates with a BS via a RAN link. In a fully decentralized network, the BS performs baseband processing on the PHY layer and MAC layer as well as packet data convergence (PDCP, packet data convergence protocol). The traffic can then be forwarded as IP packets over BH links to the core. In the fully centralized architecture,

access

station

MAC+PDCP

architecture

Decentralized

Flexible

architecture

Centralized

PHY

‘last mile’

aggregation

processing

centre

MAC+PDCP

PHY

CPRICPRI

Figure4.1 System architecture and protocol stack of decentralized, exible and centralized networks

58 Backhauling/Fronthauling for Future Wireless Systems

by contrast, all PHY, MAC and PDCP processing is performed not at the BS but in a cloud processing centre, which is connected to the core via BH. The cloud processing centre exchanges digital samples with the BS over the FH, usually using the Common Public Radio Interface (CPRI) [11] standard. The FH can be separated into an aggregation network that collects and distributes data to many BSs, and the so‐called ‘last mile’ that represents the final link to the BSs. For the FH as well as for the BH, a number of topologies (star, chain, ring and tree) can be used [12].

The two currently used concepts of either fully decentralized or fully centralized baseband processing are the two extreme sides of a basic trade‐off. The more processing is centralized, the easier it is to implement cooperative processing, the higher are the potential gains in cost savings, and the easier are maintenance and upgrading. This is traded off with a demanding and expensive FH network. The decentralized approach, on the other hand, requires a much simpler BH network but makes the aforementioned operations much more difficult or expensive. For future mobile network architectures, it therefore makes sense to look at intermediate options that still offer high centralization gains but at reduced FH requirements. In the next section, four of these intermediate split options that offer a large reduction in FH requirements will be described.

4.3 Functional Split Options

Figure4.2 shows the more detailed PHY layer signal processing chain in the BS of a typical mobile network, using the example of LTE [13]. In the downlink (DL), the MAC layer user data are first encoded for forward error correction (FEC) before being modulated and precoded. These operations are performed according to the current channel quality and channel state, which have to be made available by measurements. Next, the user and control data are mapped to the physical resources, for example, subcarriers and time slots, thereby multiplexing different logical channels, for example, control and data channels. Additional signals for synchronization and channel measurements are added at this stage as well. In LTE, this resource mapping is performed in the frequency domain. After transforming the signal to the time domain, a cyclic prefix (CP) is added and the data are digitally filtered. Finally, the data are D/A converted, up‐converted to the carrier frequency and then transmitted via the antennas.

In the uplink (UL), the process is reversed. The radio frequency signal received from the UEs is first down‐converted to baseband, then digitized by sampling and quantization and digitally filtered. After the CP has been removed, the data are transformed to the frequency domain, where the different channels are demapped from the physical resources and the signals are equalized. After converting the signals back to the time domain, the symbols are detected and decoded. The resulting MAC data are then forwarded to the higher layers. In this work, we will view the hybrid automatic

Fronthaul for a Flexible Centralization in C-RANs 59

Downlink

Centralized

MAC-layer

processing

FEC encoding / decoding

Mod. + precoding / detection + equal.

Split D

Split C

Split B

Local

processing

IFFT + mapping / FFT+demapping

Split A

D/A conv. + filtering / A/D conv.

Radio frequency processing

Antenna

Uplink

Figure4.2 Functional split options. Reproduced from [10] with permission from the IEEE

repeat request (HARQ) as part of the decoding process, although it is often seen as part of the MAC layer.

Within this processing chain, four options of splitting the processing between the BS and the central processing unit are very promising. These are indicated in Figure4.2 as split options A–D.

Split option A corresponds to the split used in C‐RAN as implemented today. The FH interface for this split is standardized in CPRI [11]. In the DL, the complete baseband processing is performed centrally and digital samples are forwarded to theBSs. In the UL, conversely, the received signals are only digitized, filtered and then forwarded. As all baseband processing is centralized, there is no drawback in terms of what types of joint processing can be performed.

In split option B, the mapping/demapping is decentralized. In the DL, this means that the user and control data are forwarded separately and in the frequency domain. Synchronization and reference signals can be either forwarded from the central processing unit to the BSs or they may be generated and added at the BS. The different signals are combined at the BS, converted to the time domain and the CP is added before it is D/A converted. In the UL, the CP is removed, the signals are converted to the frequency domain and the separate channels are demapped at the BS.

60 Backhauling/Fronthauling for Future Wireless Systems

Synchronization and channel estimation can either be performed at the BS or centrally. Due to the fact that the decentralized processing mainly involves transforming and restructuring the received signal, all types of cooperative processing can still be performed centrally without any disadvantage. However, for split B and all subsequent splits, digital processing resources are required at the BS, which makes them bigger and more power‐hungry.

For split option C, modulation and precoding are also decentralized in the DL. Accordingly, bit data instead of complex amplitude samples need to be forwarded from the central processing unit to the BS. In the UL, equalization and demodulation are performed at the BSs so that soft information on user bits, for example, log‐ likelihood ratios (LLRs), needs to be forwarded to the central processing unit. As detection is now performed decentrally, techniques like CoMP or MPTD can no longer be performed easily, so this split does not offer as many centralization gains as the previous ones. Still, techniques like, for example, joint decoding [14] can be performed centrally. Also, techniques such as in‐network processing have been proposed [15] that enable a decentralized, yet joint detection. However, these methods require extensive inter‐BS links to exchange signals, thereby again increasing the requirements of the FH.

In split option D, coding and decoding are also performed decentrally, that is, all PHY‐layer processing is performed at the BSs. Accordingly, centralization gains can only come from higher‐layer processing. Techniques like joint scheduling [16] and connection control [17] still offer benefits from such a partly centralized architecture. However, the requirements of these higher‐layer splits are very similar to those of split D.

4.4 Requirements ofFlexible Functional Splits

The main benefit of an only partial centralization is the reduced requirements on the FH network in terms of data rate, latency and jitter. In the following, the requirements of the four introduced split options will be discussed. While the required data rates can be relatively easily derived, the requirements on latency and jitter are difficult to estimate, as they depend on numerous physical parameters as well as standardization and implementation aspects. The data rates in this section are derived for a single sector and a single carrier/frequency band. If a site incorporates multiple sectors and carriers–as is quite common in today’s network–the data rates scale linearly with that number. This is sometimes identified as the number of ‘sector‐carriers’. Furthermore, LTE is used as a baseline technology, as it has the most demanding requirements when compared to third generation (3G) and second generation (2G) technologies. However, the derivations of other previous or future standards are expected to be very similar. Some numerical examples for contemporary and future networks are given at the end of this section.

Fronthaul for a Flexible Centralization in C-RANs 61

DNfN

4.4.1 Split A

Split A marks the limit between analogue and digital signals. Accordingly, the data exchanged via the FH correspond to digitized I/Q samples and the data rate required for split option A in bit/s can be calculated as follows:

where N

AAsQA,,

is the number of antennas, fs is the sampling frequency and N

(4.1)

is the reso-

Q,A

lution of the quantizer in bits. The factor 2 accounts for the I‐ and Q‐ phase of the signal and γ is the overhead introduced by the FH, that is the overhead of line coding or an FEC, or additional control signals. The number of antennas N

indicates that a

separate data stream needs to be fronthauled for each antenna, while the sampling frequency f

accounts for the system’s bandwidth. The quantizer resolution N

Q,A

and

the factor 2 account for the word length per digital sample.

The dependence on the number of antennas N

of this split may become critical

in future mobile networks, when massive multiple‐input/multiple‐output (MIMO) techniques [18] with 100 or more antenna elements may be introduced, as this linearly scales up the FH data rate. Similarly, f

depends on the total bandwidth, which is also

foreseen to increase in future networks. The quantizer resolution needs to be quite high, usually around 15 bits per dimension, due to the high dynamics of the time domain signal in LTE [19]. However, the main disadvantage of this split is the fact that it does not depend on the actual user traffic, that is even when no user is connected to the BS, the full FH data rate needs to be forwarded.

In terms of latency, CPRI defines a maximum round‐trip time of 5 microseconds (µs) for processing, which is added to the propagation time to amount, typically, to a few hundred µs. According to the standard, this is mainly motivated by the inner loop power control for the UL in UTRAN (universal terrestrial radio access network, the mobile network of 3G) [11]. To compensate for the fast fading and near–far effects, the latency must be in the order of the coherence time of the channel, which can be as low as one millisecond (ms) for fast users in current 4G systems. As the coherence time decreases with higher carrier frequencies, the requirements can be expected to be even stricter in 5G systems which utilize millimetre‐wave (mmWave) frequencies. Other works [19, 20] state that the time allowed in LTE for HARQ acknowledgements is the most critical latency constraint. The UE expects an acknowledgment after 3 ms. When subtracting the time typically required for baseband processing, this leaves a few hundred ms for the FH. However, this is a direct result of the LTE standard and not of physical constraints, and thus could be modified in future standards. For split C, we describe methods for mitigating the delay requirement of HARQ.

In order to enable CoMP and distributed MIMO techniques, samples of different BSs need to be correctly aligned and the timing advance, that is, the different propagation times of different UEs, need to be known precisely. For this, the total delay on

62 Backhauling/Fronthauling for Future Wireless Systems

BASC SF QB

NSC1200

the FH must be measured and jitter must not be too high. Hence, the accuracy of the timing measurements and the jitter need to be in the order of magnitude of the sample duration, that is, a few tens of nanoseconds (ns) for LTE. In future networks utilizing bandwidths in the GHz range, these requirements will increase still further as the sample duration decreases. This could lead to a maximum tolerable jitter in the order of hundreds of picoseconds.

4.4.2 Split B

The data exchanged when utilizing split option B corresponds to frequency domain samples. Additionally, the different physical channels are separately available. These different channels carry data responsible for, for example, synchronization, channel estimation, control signals or user data. Some of these channels, for example, synchronization and reference symbols, do not have to be fronthauled as they can be generated at the BSs. These signals do not benefit from centralized processing and can be calculated based on a few parameters like the cell identifier, which can be assigned permanently. The required data rate for split B can be calculated as:

where N frame, T

Q,B

is the number of utilized subcarriers, NS is the number of symbols per

is the frame duration, η the percentage of actually occupied resources and

the number of quantization bits per sample for this split. NSC points to the fact

NN NT N

(4.2)

that usually guard carriers are used to avoid interference with neighbouring bands. As they do not carry any data, they can be added after/discarded during the mapping/ demapping process. To give an example, a 20 MHz LTE system has 2048 subcarriers in total, of which 848 are used as guard bands, leading to riers. Similarly, N

is the number of symbols that carry user or control channels, as

utilized subcar-

synchronization and reference signals also potentially do not have to be forwarded between the BS and the central processing unit. The load factor η accounts for the fact that before/after mapping/demapping only those physical resources need to be forwarded that actually carry user data, that is, if the BS is only 50% loaded, the FH data will also be reduced by a factor of 2. As the dynamics of the frequency domain signal are much lower compared to time domain signals, the number of bits per sample N

Q,B

can be reduced to 7–9 bits [19], thereby further reducing the FH data rate.

Split B offers the option of performing channel estimation decentrally, as the reference symbols are available after demapping. This would remove the power control constraints on the latency for this split, as the signal-to-interference-plus-noise ratio (SINR) could be calculated in the BS. However, the centralized precoding still requires that up‐to‐date channel information is available at the central processing unit. As a consequence, the latency requirements could be slightly relaxed, depending

Fronthaul for a Flexible Centralization in C-RANs 63

CLSC SF QC

log

on the coherence time of the channel. Jitter is less critical for the higher‐layer splits, as the samples are aligned at the BS. Still, processing in the central unit can only start once the symbols from all BSs are available. While the symbols can be buffered to compensate for different latencies, a high difference in arrival times of different BSs would lead to a large latency.

4.4.3 Split C

Split C is the only split that exhibits a significant asymmetry between UL and DL. In the DL, encoded user bits are transported, while in the UL, LLR values have to be forwarded to enable turbo decoding. Still, the required data rate for both UL and DL can similarly be calculated as:

where N

is the number of layers (spatial streams), M is the modulation order and N

(4.3)

Q,C

is the number of quantization bits for split C.

The difference between UL and DL lies in N

. In the DL, only encoded user bits

Q,C

are forwarded, so the number of bits per symbol is coupled to the utilized modulation scheme, that is 2 bits for 4‐QAM, 4 for 16‐QAM and so on. In the UL, one LLR per information bit has to be forwarded and each LLR is typically represented by 3 bits, that is, 6 bits for 4‐QAM, 12 bits for 16‐QAM and so on.

The main advantage of this split is that the number of antennas is mapped to spatial streams and vice versa. To give an example, if a BS is equipped with four antennas but due to the channel state can only transmit one spatial stream to a user, only this one stream has to be forwarded instead of one stream for each of the four antennas. This will become of importance in massive MIMO systems [18], when a high number of antennas is mostly used for beamforming, and only a limited number of independent user streams are transmitted. Both the number of layers and the modulation scheme depend on the current channel quality of a certain user. These two dependencies result in the fact that the FH traffic is even more coupled with the actual user traffic, that is, when a user faces bad channel conditions and can transmit only little data, this is reflected in the FH traffic.

As channel decoding is still performed centrally, the HARQ scheme of LTE can be limiting, as UEs require an acknowledgement to be sent within 3 ms, so split C needs to meet that latency requirement. To overcome this, local feedback schemes have been proposed [21] in which the BSs send feedback without waiting for confirmation about the decoding outcome from the central processing unit. However, this will reduce the performance to some degree. Alternatively, it has been proposed to suspend the HARQ process if the acknowledgement cannot be sent in time and simply not schedule new traffic until the decoding is finished [19]. For future systems, increasing the allowed HARQ delay could also be considered. However, this requires

64 Backhauling/Fronthauling for Future Wireless Systems

DLSC SF cQD

log

higher available memory resources at the UEs, as the data of the different HARQ processes will have to be kept until the acknowledgement arrives.

4.4.4 Split D

Using split D, bit‐level user data are transported and thus correspond very closely to what is considered classical BH in current networks. The required data rate can be calculated as:

where R

DNNNTMRN

is the code rate and N

is the number of quantization bits for this split.

Q,D

(4.4)

Coding/decoding adds/removes redundant bits to/from the actual information bits. These redundant bits do not have to be forwarded in this split, which further decreases the FH data rate. Also, the code rate is coupled to the channel quality and therefore to the actual user traffic. As information bits are forwarded, the number of quantization bits can be set to one for this split.

Split D terminates PHY‐layer processing. Hence, the latency requirement is determined by the higher layers. If centralized scheduling is performed and intended to exploit time diversity, the latency must be small enough to follow the fading of the channel. Otherwise it can be relaxed to the application‐layer requirements which are typically in the order of a few tens of ms.

4.4.5 Summary andExamples

Table 4.1 summarizes the main parameters that determine the requirements of the different splits.

The different data rate requirements are further illustrated in Figure 4.3. The requirements for the four different splits are depicted for four different parameter sets. The baseline is a 4G system, that is, LTE, which is further differentiated in a maximum data rate, utilizing the highest parameters possible and an exemplary parameter set using more representative values. The exemplary parameter set was chosen to illustrate additional reductions in data rate that would not be visible in the maximum data rate. To give a few examples, the maximum load of a BS is 100%, while in real deployments such a utilization would actually indicate a capacity problem of the network as a BS should be loaded less than 100%. Similarly, the maximum code rate is

1.0 (i.e., uncoded transmission), while real systems have to employ a lower code rate according to the channel quality.

As we discussed previously, the requirements are expected to increase in future networks mainly through three advancements: the introduction of massive MIMO and the utilization of higher carrier frequencies and bandwidths. Accordingly, Figure4.3

Fronthaul for a Flexible Centralization in C-RANs 65

Data rate

Table4.1 Parameters impacting FH requirements fordifferent functional splits

Split Data rate Latency Jitter

• Number of antennas

• Bandwidth

• Quantization in time

domain (high number of bits)

B • Number of antennas

• Number of utilized

subcarriers

• Load

• Quantization in fre-

quency domain (low number of bits)

C • Number of spatial layers

• Load

• Modulation order

• Number of quantization

bits 1 (DL) or 3 (UL)

D • Number of spatial layers

• Load

• Modulation order

• Code rate

• Channel coherence time

• UL power control needs

to be able to follow fast fading

• Channel coherence time

• Precoding must be able

to follow channel

• Maximum delay of HARQ acknowledgement

• Requirements of higher‐layer applications

• For joint scheduling, channel coherence time can still be determining factor

• Sample duration

• Timing advance

needs to be measured exactly

• Queueing should not lead to increased total latency

1 Tbps

1 Gbps

1 Mbps

7.2 Tbps

4.9 Gbps

2.5 Gbps

Split A Split B

2.7 Tbps

5G maximum

4G maximum

960 Gbps

1.6 Gbps

311 Mbps

5G exemplary

4G exemplary

1.7 Tbps

52 Mbps

38 Gbps

1.6 Gbps

Split C Split D

560 Gbps

6.4 Gbps 538 Mbps

8.6 Mbps

Figure 4.3 Data rate requirements of different functional splits for current and future networks

66 Backhauling/Fronthauling for Future Wireless Systems

also shows data rates for a potential 5G system with 100 antennas and a sampling frequency of 1.5 GHz. As higher‐order modulation schemes are also under discussion, 1024‐QAM was assumed. To account for the lower symbol distance in 1024‐QAM, a higher quantizer resolution was also taken into account. The full list of parameters can be found in Table4.2. In rows where only one parameter is listed, the exemplary and maximum values are identical.

From the chosen parameters it is obvious that a potential 5G system would further increase the already demanding requirements. In particular, the scaling with number of antennas illustrates clearly that a full centralization with per‐antenna backhauling is infeasible. In fact, the data rate is increased by more than three orders of magnitude. While it can be expected that transport network technologies will also advance in the future, such an increase cannot be expected in the timeframe considered for 5G mobile networks. It can be further seen from Figure4.3 that while the higher‐layer splits lead to a reduction in general, there is also an important difference between the maximum possible data rate and the exemplary rate. This difference is discussed in detail in Section4.5.

Furthermore, Table4.3 shows the impact of higher carrier frequencies, larger bandwidth and higher UE speeds on the channel coherence time and sample duration, which, in turn, determine the maximum tolerable delay and the delay accuracy for a fully centralized system (split A), respectively. From the forecast numbers for a 5G system, it is clear that these requirements would be even more challenging to fulfil than those of a 4G system.

Table4.2 Parameters forthecalculation ofdata rate requirements

Symbol Description 5G max. / exemplary 4G max. / exemplary

Number of antennas 100 / 100 4 / 2 Number of spatial

50 / 8 4 / 1

layers

Number of subcarriers 60 k / 50 k 1200 / 1080 Number of data

14 / 12 14 / 12

symbols per frame

Sampling frequency

1.5 GHz 30.72 MHz

(bandwidth)

Number of quantiza-

18, 12, 3, 1 bit 15, 9, 3, 1 bit tion bits per I / Q dimension for split A, B, C, D

γ FH overhead 1.33 1.33

Frame duration 1 ms 1 ms

η Utilization (load) 1.0 / 0.5 1.0 / 0.5

M Modulation order 1024 / 16 64 / 4 R

Code rate 1.0 / 0.5 1.0 / 0.5

Fronthaul for a Flexible Centralization in C-RANs 67

N mM

,, log

Table4.3 Delay requirements of4G andexemplary 5G systems forsplit A

Parameter 4G 5G

Carrier frequency f Max. UE speed v 250 km/h 500 km/h Channel coherence time/max. delay

023.4

(

Bandwidth 20 MHz 1 GHz Sampling rate f Sample duration/delay accuracy

) [22]

2 GHz 70 GHz

914 µs 13 µs

30.72 MHz 1.5 GHz

32.6 ns 0.67 ns

4.5 Statistical Multiplexing inaFlexibly Centralized Network

The previous section observed how a flexible split in general reduces the data rate requirements by not forwarding certain parts of the signal. One main disadvantage of split A, which corresponds to the currently used split in C‐RAN, is identified with the fact that the FH data rate is always constant and does not vary with the actual user traffic. With higher‐layer splits, this coupling is progressively increased, which gives rise to another important aspect of a flexible functional split: the statistical multiplexing gain.

As described in Section4.2, FH networks typically consist of two parts: the so‐ called ‘last mile’ that connects the individual BSs and an aggregation, or ‘metro’, network that aggregates that traffic from numerous BSs and forwards it to the core network. The observations made in Section4.4 are mainly valid for the last mile, as the data rates for single BSs are derived. In the aggregation network, multiples of the thus‐described data streams have to be forwarded over a single link. Because the individual last mile traffic is time variant, the dimensioning of these aggregation links now poses a trade‐off: on the one hand, the aggregation network has to be able to forward peak traffic, but on the other hand, peak traffic will only occur in a very limited number of times, resulting in an underutilization of the expensively deployed network for most of the time. In the following, we describe how this can be addressed to dimension the network appropriately.

4.5.1 Distribution ofFH Data Rate per Base Station

The variance of the FH traffic is different for the four functional splits and depends on the respective parameters in Equations (4.1)–(4.4). The variant parameters in those equations are

and Rc. While the load η varies due to the changing

68 Backhauling/Fronthauling for Future Wireless Systems

CDF

FH data rate in Mbps

0.8

0.6

0.4

0.2

Split A Split B Split C Split D Full load Variable load

–1

Figure4.4 Cumulative distribution function (CDF) of the data rate for different functional split options with full load (solid lines) and an exemplary variable load (dashed lines). Reproduced from [10] with permission from the IEEE

traffic demand generated by the users, the remaining parameters depend on the channel quality of the users. In order to use a high number of layers, the channel needs to offer a high spatial separation, while the remaining parameters depend on the SINR. To illustrate the impact of those varying parameters, Figure 4.4 shows the cumulative probability function (CDF) of a single BS for the different split options. The SINR distribution was taken from system‐level simulations in a dense urban scenario. As the load distribution has the largest impact, we show the CDFs both for a full load and a variable load with a uniform distribution between 0 and 1. The remaining parameters are chosen identically to the exemplary 4G system in Table4.2.

It can be observed that the data rate for split A is constant, which is a major disadvantage. Split B only varies with the load and is therefore constant in the case of a full load. Split C additionally varies with the modulation scheme, which can be seen from the three steps in the fully loaded case, corresponding to the three modulation schemes4‐QAM, 16‐QAM and 64‐QAM. Similarly, split D depends on the modulation and coding schemes (MCS), so 28 steps can be observed, corresponding to the 28 utilized MCSs in this example. For the case of varying load, these distributions are additionally convoluted with the load distribution. In general, we can observe that, on the one hand, the maximum data rate in each split decreases, and that the data rate becomes more variant. This variance is exploited for statistical multiplexing.

4.5.2 Outage Rate

To avoid having to dimension the network for peak traffic, usually a certain outage probability is defined [23], that is, it is conceded that for a certain percentage of time, the network cannot transport the offered traffic. This will potentially lead to a reduced

Fronthaul for a Flexible Centralization in C-RANs 69

D ~( )

ODD



 09.

QoE (quality of experience) for the users, but is accepted by the operators as it can reduce the required FH capacity dramatically. As an example, assume the traffic of one

BS is approximately Gaussian distributed

with

that the actual data rate D exceeds the deployed capacity D

. Then the probability

can be calculated as:

erfc

(4.5)

with erfc being the complementary error function. This probability is also known as the outage probability.

Conversely, we can calculate the capacity that has to be deployed to yield a certain

outage probability as:

erfc

22 (4.6)

As this data rate is identical to percentiles of the corresponding data rate distribution, we write write D

to identify the data rate that needs to be deployed for an outage probability

to easily identify a data rate with its outage probability, that is, we

Po1

of 1%. We also call this data rate the outage rate.

Due to congestion effects, outages can already occur at loads below 100%. To account for this, the outage rate is usually divided by a safety factor  to calculate the actually deployed data rate D

For example, an outage probability of 1% and a safety factor

[23]:

(4.7)

imply that the

link will be loaded less than 90% for 99% of the time.

Similar observations, of course, can be made for distributions other than the Gaussian distribution used here as an example. However, the outage capacity is only relevant if the data rate follows a varying distribution. Hence, the outage capacity for the constant data rate of split A is always equal to the maximum data rate.

4.5.3 Statistical Multiplexing onAggregation Links

The statistical multiplexing gain comes into effect in the aggregation network, when multiple streams with varying data rates are added up. The most straightforward way

In fact, the load distribution can never be truly Gaussian, as this would allow for negative loads. However, the

assumption will make the comparison in Section4.5.3 easier.

www.ebook3000.com

70 Backhauling/Fronthauling for Future Wireless Systems

DNN

~( ,)

DPNN

PDF

FH data rate in Mbps

to dimension the aggregation link would be simply to scale the data rate for one BS with the number of BSs being the aggregate, for example, if N BSs are aggregated:

erfc

daggr,nomux do

22 (4.8)

However, this neglects the statistical multiplexing. The aggregation of several varying data streams can be seen as a summation of random variables. From the central limit theorem [24], it is known that the distribution of the sum of random variables will converge to a Gaussian distribution. If all data rates follow the same distribution, the sum will, in fact, be distributed as

daggr

. Accordingly, the outage

probability can now be calculated as:

ioaggr oaggr

erfc

oaggr D

(4.9)

and the outage rate as:

erfc

daggr

(4.10)

The convergence of probability density functions (PDFs) to a Gaussian distribution is illustrated in Figure4.5, using split B as an example. The load is uniformly distributed between 0 and 1, which yields a rectangular PDF for a single BS. When adding multiple varying data streams, the corresponding PDFs are convoluted. This yields a triangular distribution for two BSs and increasingly Gaussian‐like distributions for four and eight BSs, as depicted. Figure4.5 additionally illustrates the rationale behind

–3

× 10

1 BS

1.5

0.5

2 BS

4 BS

8 BS

100

1000

Figure4.5 PDFs of aggregated FH trafc of one, two, four and eight BSs, and data rate percentiles

2000 3000 4000 5000

Fronthaul for a Flexible Centralization in C-RANs 71

()

Po99 %

 09.

No. of supported BS

Deployed FH capacity in Mbps

the outage rates. While the maximum data rate D the 1% outage rate D

is approximately 3.6 Gbps. Consequently, about 1.4 Gbps of

for eight BSs is close to 5 Gbps,

100

aggregation capacity can be saved by accepting a mere 1% of outage probability.

As can be seen from Equation (4.10), the aggregated outage rate scales with

of the standard deviation, while the outage rate in Equation (4.8) scales with N times the standard deviation. This difference corresponds to the statistical multiplexing gain. Practically, this gain occurs because it is very unlikely that a large number of streams will carry peak traffic at the same time, hence the aggregated probability distribution is flattened out. The resulting multiplexing gain can be calculated as:

(4.11)

mux

daggr,nomux

daggr

with

erfc

P . For a large number of BSs this converges to:

limNg

mux

1 (4.12)

Figure4.6 illustrates this multiplexing gain using the data rate distribution from

Section4.5.1. The x‐axis shows the deployed FH data rate D

d,aggr

/ϵ for

and

and the y‐axis the number of BSs that could be supported with a single

aggregation link of corresponding capacity.

Split A

Split B

Split C

Split D

w/o mux gain

w/ mux gain

Figure4.6 Number of supported BSs versus deployed FH capacity of different functional split options with and without multiplexing gain (solid and dashed lines, respectively). Reproduced from [10] with permission from the IEEE

72 Backhauling/Fronthauling for Future Wireless Systems

smR

05.

Deployed FH capacity in Mbps

No. of supported BS

For split A there is no aggregation gain because the FH data rate is constant. For the higher splits, a statistical multiplexing gain of up to a factor of 3 can be observed, with the gain being more pronounced for a larger number of BSs and for higher‐layer splits. This can again be explained with probabilistic methods.

Equations (4.1)–(4.4) contain four factors that are varying: N

,η,m and Rc. The

modulation scheme and code rate are typically chosen together and are therefore not independent. We therefore define the spectral efficiency:

Otherwise we can expect that N

,η and s are independent. With this, the mean of

(4.13)

the distribution of the product of these variables can be calculated as:

(4.14)

and the variance is given by:

222222 22

NN ss

LL L

(4.15)

From Equation (4.12) and Equation (4.15) we now see that the total multiplexing gain depends on the ratio of σ variances of the parameters N

and μD, which, in turn, depend on the means and

, η, m and Rc. In summary, it can be said that the

statistical multiplexing gain is larger the more the individual parameters vary. An illustration for this is given in Figure4.7. It shows the number of supported BSs versus the deployed FH capacity for split D, but for different load distributions. The load distributions are uniformly distributed between 0 and 1, 0.2 and 0.8, 0.4 and 0.6, and constant with

, respectively. This yields four distributions with the same

w/o mux gain w/ mux gain

Load variance

Figure4.7 Number of supported BSs versus deployed FH capacity of split option D for uniform load distributions with different variances

Fronthaul for a Flexible Centralization in C-RANs 73

05.

mean load tively. As can be seen, the multiplexing gain is higher for larger variances, as predicted by Equation (4.12). In other words, the over‐provisioning of not considering the multiplexing gain is much worse if the traffic is varying heavily.

but different variances of 0.0833, 0.0300, 0.0033 and 0, respec-

4.6 Convergence ofFronthaul andBackhaul Technologies

The concept of the flexible functional split describes a gradual variation between what is currently known as FH and BH. Due to the very different requirements of these two ‘extreme’ versions of a split, they are, in fact, deployed as two different segments of the network. This means that not only is separate hardware deployed but also that completely different and largely incompatible standards have evolved. While this has to be owing to historic development, it is a highly suboptimal solution, as it not only drives up costs but also makes the management of the network more complex. Even more importantly, the complexity would further increase if the same approach was followed for a flexible functional split, that is, developing different hardware and standards for each of the split options. Instead, it is highly desirable to converge the technologies of the current two options and design a unified transport network. This unified transport network should not only be able to support any number of different functional splits but should also be agnostic with respect to the underlying access technology to make it future proof for the coming generations of mobile networks. In the following, an overview of the existing solutions across different layers is given and convergence to a unified technology is discussed.

4.6.1 Physical Layer Technologies

PHY‐layer technologies pose an upper bound on the performance that can be achieved in FH networks. Additionally, they have a large impact on the cost of the FH links, as specific hardware is required for each technology. In general, PHY‐layer technologies can be grouped into wired and wireless technologies. The wired technologies can be further divided into fibre/optical technologies and copper/electrical technologies, while wireless technologies are usually distinguished by the carrier frequency they utilize.

The most intuitive – and hence most widely used – option in current C‐RAN deployments is fibre. It offers very large capacities of tens of Gbps, low latency and high reliability. The range is only limited by the tolerable latency and the cost of repeaters. Dedicated point‐to‐point links have been used so far for FH in fully centralized C‐RAN, meaning that one fibre core or one wavelength channel is dedicated to an FH link from RRH to BBU. In order to exploit statistical multiplexing and to avoid a heavy underutilization of the fibre in a flexible C‐RAN deployment, time‐shared optical networks could be used [25] in the future. The main disadvantage

74 Backhauling/Fronthauling for Future Wireless Systems

of fibre technology is that it is expensive and slow to deploy due to the extensive civil works required. Adding right‐of‐way issues to that prohibits fibre connectivity toevery BS, especially in the case of dense deployments of small cells. While some operators have their own fibre networks, others– especially non‐incumbent operators – have to rely on expensive third‐party leases, further complicating the business case.

In the context of classical BH, that is, very high‐layer splits, a number of other PHY technologies are in use, making them principal candidates for a flexible C‐RAN solution. Copper‐based solutions like coax cables face the same economic problems as fibre but in addition offer less capacity. In some cases, existing deployments can be utilized to mitigate that effect. For example, digital subscriber line (DSL) connection of shop owners could be used to provide FH for indoor BSs if a corresponding incentive was offered. However, due to the limited capacity and protocol overhead of DSL technology, even with new technology generations such as G.fast [26], this is only an option for higher‐layer splits.

Wireless options, in general, offer the benefit that they are cheaper and faster to deploy, as only comparatively little civil work is required. On the other hand, their range and reliability are limited by the much higher path loss.

One category of wireless transport solutions uses similar carrier frequencies to the access links, that is, sub‐6 GHz bands. These are more suitable for high‐layer splits, as their respective capacities are comparable with the access link technology. The latency can also be quite high because similar protocols to the access link have to be observed. While some of these technologies can, in principle, use the same hardware as the access link and therefore benefit from economies of scale, additional licensed frequency bands have to be acquired or already available ones have to be assigned, which again makes the overall deployment quite expensive. On the other hand, sub‐6 GHz technologies offer point‐to‐multipoint connections, which can significantly reduce the number of required antennas and thus reduce CAPEX.

Microwave technology up to a few tens of GHz is already widely used in classical BH and could also be an option for lower‐layer splits. With capacities of currently up to 1 Gbps and a range of a few tens of kilometres, it can be utilized for the ‘last mile’, although not for the aggregation network. The main disadvantage is that, again, licensed frequency bands are utilized, which increases cost and deployment time. Also, the limited range requires the set‐up of potentially multiple solid masts for the intermediate antenna dishes along the transmission path. This results in the need for real estate or roof access, all of which comes with administrative and cost‐related overhead.

Millimetre‐wave technology, also sometimes referred to as V‐band or E‐band technology depending on the corresponding frequency bands between 60 and 90 GHz, is another good option for wireless FH. While still a relatively new technology, it already offers a capacity of up to 10 Gbps and per‐hop latencies as low as

Fronthaul for a Flexible Centralization in C-RANs 75

10 ns. In addition, it utilizes unlicensed or lightly licensed bands, which is advantageous in terms of cost and deployment time. However, mmWave frequencies suffer from relatively high free‐space path loss. Although this can be partially compensated with high‐gain antennas, it still limits the total range to a few kilometres at most for E‐band and a few hundred metres for V‐band. With current research into large‐scale antenna arrays with hundreds of individual antenna elements, beamforming techniques could be used to enable a flexible selection of point‐to‐point links. Thereby, time sharing or quick reconfiguration of interconnections in cases of large changes in traffic distribution could be enabled. Due to the low wavelength at mmWave frequencies (5 mm at 60 GHz), the corresponding antenna arrays can be built relatively small.

Finally, free space optics (FSO) is a wireless technology that utilizes laser light for communication. While it offers similar capacities and latencies to mmWave technology, it is susceptible to misalignment and wind sway due to its narrow beam width. Weather effects like fog or snow can also limit the range or decrease the reliability. Still, it is a viable option for the last mile in scenarios where these effects can be neglected.

In summary, a more flexible approach to the functional split can benefit from a more heterogeneous selection of FH technologies. Lower cost and higher deployment speed make wireless technologies attractive on the last mile and the first level of aggregation. Range limitation can be overcome by multi‐hop chains at the cost of increased latency. Still, it is undisputed that the higher levels of the aggregation network have to be composed of fibre, which alone offers the capacity to aggregate hundreds of BSs.

To further illustrate the suitability of the discussed technologies for the different functional splits, Figure4.8 shows the requirements of the four splits A–D in terms of latency and capacity (for a single BS) versus the different technologies, based on a market study in [27]. As can be seen, the utilization of split B or C would already enable a more heterogeneous selection of applicable PHY technologies. It should be noted that all discussed FH technologies can be expected to advance in the future; however, access link technology will similarly progress, so that the mapping can still be considered viable for future networks.

4.6.2 Data/MAC Layer Technologies

On the MAC layer, there currently exist two completely different standards corresponding to the two different extremes of the functional split: full centralization and full decentralization. In fully centralized C‐RAN architectures, CPRI has evolved as the de facto industry standard for digital FH, while the classical BH in decentralized networks utilizes Ethernet. In terms of the data plane protocol, the incompatibility of these two standards arises mainly from the different frame formats that are used and

76 Backhauling/Fronthauling for Future Wireless Systems

Data rate in Mbps

Latency (round trip time per hop) in s

–4

–3

–2

–1

Split A

Split B Split C

Split D

Fibre(CWDM/dark)

mmWave

μ-Wave (PtP/PtMP)

FSO

Metro optical

Passive optical network (PON)

Sub-6 GHz

xDSL

Figure4.8 Functional split requirements versus physical layer technologies. Reproduced from [10] with permission from the IEEE

the strict requirements on timing and jitter that were introduced for CPRI. This incompatibility has resulted in the fact that two separate transport networks have to be deployed, which is a suboptimal solution as it increases deployment cost and management overhead.

At first glance, the introduction of a flexible split seems to further complicate this issue, as neither of the two existing standards will be compatible with all intermediate splits. However, this offers the chance to design a unified transport technology that is not only compatible with the existing two standards, but with a wide range of functional splits. Such a unified technology does not need to be designed from scratch but could be based on one of the existing standards. Investigations are already under way to enable CPRI over Ethernet [28], while also taking the possibility into account of mapping arbitrary data to Ethernet frames. The main challenge in such a unified frame format is to meet the strict requirements of the lower splits in terms of latency and jitter. To achieve them, large improvements in time synchronization as well as a reduction in deterministic delays will have to be achieved. GPS‐assisted synchronization or synchronous Ethernet using the precision time protocol [29] are available and could be utilized for this. In terms of capacity, 10 Gbps Ethernet is already standardized (802.3.ae, ak, an [30]) and study groups are investigating data rates up to 400 Gbps [31]. On the other hand, a recent initiative aims to standardize a more flexible FH in the form of the next generation fronthaul interface (NGFI) [20]. However, while its goal is to make the FH more flexible and support higher‐layer splits, it is currently not considering utilizing it for BH as well.

Fronthaul for a Flexible Centralization in C-RANs 77

4.6.3 Network Layer Technologies

The network layer is responsible for switching and routing traffic between network nodes. As such, it has a decisive impact on a converged transport network architecture. While BH networks inherently work with packet switching, FH networks commonly use dedicated point‐to‐point connections. A converged transport network will have to support packet switching and point‐to‐multipoint connections, as this simplifies implementation and management and allows for arbitrary placement of the centralized processing elements, as discussed in Section 4.7. Currently, they can be located only at sites with sufficient dedicated links to support the RRHs, leading to a concentration in a low number of large data centres. However, to enable the fast migration of RAN functionality from one data centre to another, and to adapt to varying load profiles across BSs, packet switching is the only efficient option.

One challenge for this is that additional delays are introduced due to the processing time of switches and routers. Especially for very large networks with many hops, this could make low‐layer functional splits infeasible. However, recent advances in low‐latency switching have decreased the switching delay to about 100 ns, which is three orders of magnitude lower than the requirements for a fully centralized network.

A more important impact comes from queueing, as this not only introduces additional delay but it is also indeterminate as the length of the queue will vary with the variation in packet arrival. Prioritization of packets can be used to reduce the total wait time for certain packets. This is managed via quality of service (QoS) class identifiers; however, the means of choice for providing fairness among QoS classes are probabilistic scheduling methods, adding another layer of unpredictability. For a converged transport network architecture, new methods would have to be found to make the delay more predictable and guarantee maximum latencies. Special QoS classes for RAN functionalities could be introduced that prioritize corresponding packets above everything else. For times of congestion, the dropping of other packets could be considered. The requirements for higher‐layer applications are, in fact, many orders of magnitude larger than RAN packages, for example, Voice‐over‐IP requires latencies of tens of ms as compared to the hundreds of µs required for a fully centralized RAN architecture. Hence, RAN requirements instead of application‐layer requirements will dominate the design of a converged BH/FH network. On the other hand, the application layer could also benefit from greatly reduced latencies, enabling ‘tactile Internet’ applications [32] like virtual reality, automation control or remote vehicle steering. Still, this would require an update of the whole BH network. While the queueing policy can usually be modified via software or firmware updates, low‐ latency switches require new hardware at substantial cost. Also, switches need to enable network synchronization as discussed in the section above, which is not widely supported by the currently deployed hardware.

78 Backhauling/Fronthauling for Future Wireless Systems

4.6.4 Control andManagement Plane

One common requirement from the sections above is that future networks need to be more flexible. The introduction of more heterogeneous PHY technologies, a unified frame format able to carry traffic from different splits and packet‐switched FH all support a transport network that can be adapted to follow, for example, hourly or daily traffic variations. Still, the evolution to 5G networks will bring along a further process of updating existing deployments or adding more hardware or functionalities. In particular, with the advent of mmWave access technology, massive MIMO and the tactile Internet, the requirements on the transport network will change and potentially become even more demanding. In order to adapt the network both in the short and in the long term, a universal management plane is required. This management plane will be in charge of policies ranging from connection control on the RAN and FH links over congestion control and load balancing to routing and QoS policies.

The centralization of RAN functionality also poses new challenges in terms of reliability, as the data centres hosting such functionalities are potential single points of failure, for example in the case of localized power outage or malicious attacks from hackers. As such, the control and management plane also has to be in charge of outage protection and failover control. This will include functionalities ranging from rerouting traffic in the case of a single link outage to migrating complete centralized RAN functionality from one data centre to another in the case of more severe malfunctions. In order to provide low failover times, 1 + 1 protection schemes could be used on the transport network at the cost of additional deployed capacities.

For the implementation of such a universal control and management plane, software‐defined networking (SDN) [33] and network function virtualization (NFV) [34] are good candidate technologies as they provide–among other things–unification via abstraction layers. The main challenge arises from the scale to handle a complete mobile network, including BSs, the transport networks, centralized RAN functionality and a possibly virtualized network core.

4.7 Enablers ofaFlexible Functional Split

The utilization of a flexible functional split impacts not only the FH network but needs to be supported by the BSs and the central processing entity. The increased flexibility that is offered by a larger number of split options comes at the price of higher hardware complexity at the respective endpoints. In this section, a short overview is given of the technologies that are needed from the baseband processing perspective to enable a flexible functional split.

As the functional split determines the type of joint processing that can be performed, the split has to be adapted to the current scenario. For example, in a dense deployment with a large number of cell‐edge users, CoMP techniques requiring a lower‐layer split might yield large gains in user throughput. On the other hand, macro‐cell deployments

Fronthaul for a Flexible Centralization in C-RANs 79

might not benefit from a high degree of centralization and therefore a higher‐layer split could be employed, reducing the load on the aggregation network. This means that the functional split will have to be adapted in space and also in time to fit the scenario. This not only requires more flexible interfaces on the transport network, as discussed in the previous section, but also that the hardware both in the BS and in the central entity can perform the respective processing. This brings two problems: first, in order to support higher‐layer splits, BSs would need to be equipped with the same hardware as fully decentralized BSs. As one of the intended advantages of a centralized RAN is smaller BSs, this effect would be nullified. Second, baseband processing is usually implemented on dedicated hardware like FPGAs or ASICs, which cannot be flexibly reconfigured to match any functional split. As a consequence, a more flexible hardware architecture is required in addition to the converged transport network. Recent progress in the IT industry provides very promising solutions for this with the approach of RAN virtualization and ‘cloudification’.

Although the term ‘cloud’ was part of even the first considerations on C‐RAN [1], the first deployments did not follow the principles of cloud computing as they are known in the IT industry, namely providing a ‘shared pool of configurable computing resources that can be rapidly provisioned’ [35]. Instead, the first deployments followed the approach of simply splitting up the conventional base stations by separating the baseband hardware from the RF front‐ends [36]. However, this simple hardware centralization does not embrace the concept of cloud computing. Many advantages of centralized processing, like load balancing and energy savings, are only possible if the baseband hardware can be dynamically allocated and configured according to demand. This requires a virtualization of the available hardware, which, in general, requires the utilization of GPPs. Only if the hardware can be quickly and easily reconfigured to perform smaller or larger parts of the baseband processing can a flexible functional split efficiently be employed. The main challenge with this is that the implementation of baseband processing requires far higher throughput and lower latencies than traditional IT applications, which is why dedicated hardware is used in conventional BBUs. Additionally, the virtualization of the hardware introduces extra overhead in the form of a so‐called hypervisor that oversees and controls the provisioning and operation of the virtual machines.

In order to reduce the overhead of virtualization, newer approaches to virtualized hardware employ so‐called ‘bare metal’ servers. These do not require an additional operating system for each virtual machine. Instead, the hypervisor can directly communicate with the physical hardware. This also has the advantage that the physical hardware can be fully dedicated to a certain task, for example, a physical processor core can be assigned to the processing of a specific BS. This makes it far easier to guarantee performance when compared to a conventional virtualized system, where a physical core might have to process several virtual machines. Some recent works show that it is possible to implement computationally complex processing like turbo decoding on general‐purpose hardware in real time [37]. Additionally, the amount of

80 Backhauling/Fronthauling for Future Wireless Systems

required processor cores can be predicted, allowing for precise provisioning in order to avoid outage by under‐provisioning or wasting resources by over‐provisioning [38]. However, the jury is still out as to whether the implementation of the full baseband stack on GPP is cost‐ and energy‐efficient.

As an alternative, certain parts of the baseband processing could be outsourced to dedicated hardware accelerators while maintaining the virtualization approach. The newest generation of servers can not only be equipped with GPPs but can also be supplemented with DSP cores [39]. The operation of fast Fourier transformation (FFT) and inverse FFT (IFFT) and channel coding/decoding, in particular, can be much more efficiently performed by dedicated hardware. Field trials have already shown the validity of such a hardware‐accelerator‐assisted GPP implementation [40]. While these approaches are mainly designed for the central processing entity, a similar architecture can be employed on the BS side. The deployment of micro‐ servers or ‘cloudlets’ at BSs has already been envisioned, albeit mainly for the purpose of application processing [41]. However, this perfectly complements the approach of the flexible split. GPP cloudlets with additional hardware accelerators could be deployed and used for baseband processing when a higher‐layer split is configured, and when a lower‐layer split is configured, the now idle hardware could be utilized for user‐application processing, thereby avoiding underutilization of the hardware at the BSs.

In summary, the flexibility offered by a virtualized cloud‐RAN implementation will also be the basis for the flexible functional split that can be dynamically adapted in time and space to optimally reflect the scenario in terms of traffic density, fronthaul load and hardware utilization.

4.8 Summary

Taking all of the above into account, it becomes clear that the design of the FH network will be a major challenge in future networks. While the approach of C‐RAN offers tremendous advantages, the FH could become a major bottleneck, both in terms of performance and cost efficiency. A more flexible functional split will help to mitigate this problem. A partial centralization can reduce the requirements on the FH dramatically in scenarios where full centralization offers no advantage. In particular, the data rate can be easily lowered and– more importantly – coupled to the actual user traffic. Network operators, therefore, need to carefully decide whether the traffic demand justifies the expensive FH for high centralization and how much gain can be expected from joint processing. In the aggregation network, the multiplexing gain plays an additional important role. As it makes use of the temporal and special traffic variation within a network, it is beneficial to aggregate traffic with a highly variable distribution. As described in Section4.5, the variance mainly results from variable BS load and a variable SINR. Therefore, base stations exhibiting different load patterns

Fronthaul for a Flexible Centralization in C-RANs 81

Split A

Split B Split C Split D

No. of supported BS

should be aggregated, for example rural areas and inner cities, and similarly those with different SNR distributions, for example macro cells and dense small cells. The effect will be even more pronounced in a converged BH/FH network. By adding further traffic in the form of BH and control signalling, the traffic becomes more diverse, thereby increasing the benefit of statistical multiplexing.

The main benefit of the reduced FH requirements will come in the form of reduced deployment costs. Less FH capacity per BS means that less capacity needs to be deployed. To summarize the effect of statistical multiplexing and the general data rate reduction of higher‐layer splits on the deployment cost, Figure 4.9 illustrates the number of BSs that can be supported by a single link using exemplary FH technologies. While, for example, a separate fibre core with 20 Gbps capacity is required for every seven BSs in a fully centralized network corresponding to split A, more than 800 can be supported per fibre core when utilizing split D. Similarly, more heterogeneous technologies like wireless FH can only be utilized with higher‐layer splits, thereby replacing the expensive fibre entirely on the last mile.

Starting from the physical layer technologies, future mobile networks have to aim to converge the BH and FH technologies and utilize unified methods across all network layers. They have to transport different types of traffic, meet a set of greatly different QoS parameters and be flexibly reconfigurable while utilizing different physical technologies. While the implementation of such a network will be challenging in the beginning, it offers the chance to reduce the complexity of network management and operation substantially while additionally reducing deployment cost, culminating in a virtualization of the transport network. In fact, a fully virtualized RAN should not only aim to virtualize a single part of the network like baseband processing, but all elements including base stations, fronthaul, baseband processors

00 00

xDSL

Fibre

μ-Wave

xDSL

Fibre

μ-Wave

xDSL

134

Fibre

μ-Wave

xDSL

811

Fibre

μ-Wave

Figure4.9 Number of supported BSs for different aggregation technologies: xDSL at 200 Mbps; μ‐wave at 2 Gbps; bre at 20 Gbps. Reproduced from [10] with permission from the IEEE

82 Backhauling/Fronthauling for Future Wireless Systems

and backhaul. Numerous tools such as a flexible functional split, a converged BH/FH network, SDN, NFV and cloudlets will have to come together to achieve the flexibility that is required to improve performance, cost efficiency and adaptability of future mobile networks.

Acknowledgement

The research leading to these results has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 671551 and the European Union’s Seventh Framework Programme (FP7/2007‐2013)

under grant agreement No 317941. The European Union and its agencies are not liable or otherwise responsible for the contents of this document; its content reflects the view of its authors only.

References

[1] Guan, H., Kolding, T. and Merz, P. (2010) Discovery of cloud‐RAN. [2] Irmer, R., Droste, H., Marsch, P., Grieger, M., Fettweis, G., Brueck, S., Mayer, H.‐P., Thiel, L. and

Jungnickel, V. (2011) Coordinated multipoint: Concepts, performance, and field trial results. IEEE Communications Magazine, 49(2), 102–111.

[3] Sawahashi, M., Kishiyama, Y., Morimoto, A., Nishikawa, D. and Tanno, M. (2010) Coordinated

multipoint transmission/reception techniques for LTE‐advanced. IEEE Wireless Communications, 17(3), 26–34.

[4] Caire, G. and Müller, R. (2001) The Optimal Received Power Distribution of IC‐Based Iterative

Multiuser Joint Decoders. Proceedings of the 39th Annual Allerton Conference on Communication, Control and Computing, Monticello, IL, October.

[5] Rost, P., Bernardos, C. J., De Domenico, A., Di Girolamo, M., Lalam, M., Maeder, A., Sabella, D.

and Wübben, D. (2014) Cloud Technologies for Flexible 5G Radio Access Networks. IEEE Communications Magazine, 52(5), 68–76.

[6] Armbrust, M., Fox, A., Griffith, R., Joseph, A. D., Katz, R., Konwinski, A., Lee, G., Patterson, D.,

Rabkin, A., Stoica, I. and Zaharia, M. (2010) A view of cloud computing. Communications ACM, 53(4), 58.

[7] Guo, B., Cao, W., Tao, A. and Samardzija, D. (2013) LTE/LTE‐A Signal Compression on the CPRI

Interface. Bell Labs Technical Journal, 18(2), 117–133.

[8] Suryaprakash, V., Rost, P. and Fettweis, G. (2015) Are Heterogeneous Cloud‐Based Radio Access

Networks Cost Effective? IEEE Journal on Selected Areas in Communication, 33(10), 2239–2251.

[9] Wübben, D., Rost, P., Bartelt, J., Lalam, M., Savin, V., Gorgoglione, M., Dekorsy, M. and Fettweis,

G. (2014) Benefits and Impact of Cloud Computing on 5G Signal Processing: Flexible centralization through cloud‐RAN. IEEE Signal Processing Magazine, 31(6), 35–44.

[10] Bartelt, J., Rost, P., Wübben, D., Lessmann, J., Melis, B. and Fettweis, G. (2015) Fronthaul and

Backhaul Requirements of Flexibly Centralized Radio Access Networks. IEEE Wireless Communications, 22(5), 105–111.

[11] Common Public Radio Interface, CPRI Specification V6.0, 2013. Available at: http://www.cpri.

info/. Accessed: September 1, 2015.

Fronthaul for a Flexible Centralization in C-RANs 83

[12] Nadiv, R. and Naveh, T. (2010) Wireless Backhaul Topologies: Analyzing Backhaul Topology

Strategies. Ceragon white paper, August. Available at: https://www.ceragon.com/images/ Reasource_Center/White_Papers/Ceragon_Wireless_Backhaul_Topologies_Tree_vs_%20Ring_ White_Paper.pdf. Accessed: October 8, 2015.

[13] 3GPP (2015) 3GPP TS 36.213, ‘Evolved Universal Terrestrial Radio Access (E‐UTRA); Physical

layer procedures (Release 12),’ v12.5.0, April.

[14] Valenti, M. C. (1999) Iterative Detection and Decoding for Wireless Communications. PhD disser-

tation, Virginia Polytechnic Institute and State University, Blacksburg, VA.

[15] Paul, H., Shin, B.‐S., Wübben, D. and Dekorsy, A. (2013) In‐network‐processing for small cell

cooperation in dense networks. Proceedings of the IEEE 78th Vehicular Technology Conference, Las Vegas, NV, September.

[16] Fritzsche, R., Rost, P. and Fettweis, G. (2015) Robust Rate Adaptation and Proportional Fair

Scheduling With Imperfect CSI. IEEE Transactions on Wireless Communication, 14(8), 4417–4427.

[17] De Domenico, A., Savin, V. and Ktenas, D. (2013) A backhaul‐aware cell selection algorithm for

heterogeneous cellular networks. IEEE 24th International Symposium on Personal Indoor and Mobile Radio Communications, London, September.

[18] Larsson, E., Edfors, O., Tufvesson, F. and Marzetta, T. (2014) Massive MIMO for next generation

wireless systems. IEEE Communications Magazine, 52(2), 186–195.

[19] Dötsch, U., Doll, M., Mayer, H. P., Schaich, F., Segel, J. and Sehier, P. (2013) Quantitative Analysis

of Split Base Station Processing and Determination of Advantageous Architectures for LTE. Bell Labs Technical Journal, 18(1), 105–128.

[20] Huang, J., Yuan, Y. etal. (2015) White Paper of Next Generation Fronthaul Interface. White paper,

June. Available at: http://labs.chinamobile.com/cran/wp‐content/uploads/White%20Paper%20of%20 Next%20Generation%20Fronthaul%20Interface.PDF. Accessed: October 8, 2015.

[21] Khalili, S. and Simeone, O. (2015) Uplink HARQ for C‐RAN via Low‐Latency Local Feedback

over MIMO Finite‐Blocklength Links. arXiv preprint, arXiv:1508.06570.

[22] Rappaport, T. S. (2002) Wireless Communications: Principles and Practice, 2nd edition.

NewJersey: Prentice Hall.

[23] d’Halluin, Y., Forsyth, P. A. and Vetzal. K. R. (2007) Wireless Network Capacity Management:

AReal Options Approach. European Journal of Operations Research, 176(1), 584–609.

[24] Rice, J. (2006) Mathematical Statistics and Data Analysis, 2nd edition, Belmont: Duxbury Press,

pp. 181–187.

[25] Zervas, G., Triay, J., Amaya, N., Qin, Y., Cervelló‐Pastor, C. and Simeonidou, D. (2011) Time

Shared Optical Network (TSON): a novel metro architecture for flexible multi‐granular services. Optics Express, 19(26), B509–B514.

[26] ITU‐T Recommendation G.9701, ‘Fast access to subscriber terminals (G.fast) – Physical layer

specification,’ 2014.

[27] NFSO‐ICT‐317941 iJOIN, ‘D5.2– Final Definition of Requirements and Scenarios,’ November

2014. Available at: http://www.ict‐ijoin.eu/wp‐content/uploads/2012/10/D5.2.pdf. Accessed: September 1, 2015.

[28] IEEE (2014) ‘Standard for Radio over Ethernet Encapsulations and Mappings.’ IEEE Standard

P1904.3, 2014. Available at: http://standards.ieee.org/develop/project/1904.3.html. Accessed: September 1, 2015.

[29] IEEE (2013) ‘Standard for a Precision Clock Synchronization Protocol for Networked Measurement

and Control Systems.’ IEEE Standard P1588‐2008, 2013. Available at: https://standards.ieee.org/ develop/project/1588.html. Accessed: September 1, 2015.

[30] IEEE (2014) ‘Standard for Ethernet.’ IEEE Standard P802.3‐2012, 2014. Available at: http://standards.

ieee.org/findstds/standard/802.3‐2012.html. Accessed: September 1, 2015.

84 Backhauling/Fronthauling for Future Wireless Systems

[31] IEEE(2014) ‘Standard for Ethernet Amendment: Media Access Control Parameters, Physical

Layers and Management Parameters for 400 Gb/s Operation.’ IEEE Standard P802.3bs, 2014. Available at: http://www.ieee802.org/3/bs/index.html. Accessed: September 1, 2015.

[32] Fettweis, G. (2014) The Tactile Internet: Applications and Challenges. IEEE Vehicular Technology

Magazine, 9(1), 64–70.

[33] Open Networking Foundation (2012) Software‐Defined Networking: The New Norm for Networks.

White Paper, April. Available at: https://www.opennetworking.org/images/stories/downloads/ white‐papers/wp‐sdn‐newnorm.pdf. Accessed: September 1, 2015.

[34] ETSI Industry Specification Group for Network Functions Virtualisation (n.d.). Available at: http://

www.etsi.org/technologies‐clusters/technologies/nfv. Accessed: September 1, 2015.

[35] Mell, P. and Grance, T. (2011) The NIST Definition of Cloud Computing. US National Institute of

Science and Technology. Available at: http://csrc.nist.gov/publications/nistpubs/800‐145/ SP800‐145.pdf. Accessed: September 1, 2015.

[36] Li, L., Liu, J., Xion, K. and Butovitsch, P. (2012) Field test of uplink CoMP joint processing with

C‐RAN testbed. 7th International ICST Conference on Communication and Networking China, August.

[37] Paul, H., Wübben, D. and Rost, P. (2015) Implementation and Analysis of Forward Error Correction

Decoding for Cloud‐RAN Systems. 2nd International Workshop on Cloud‐Process. Heterogeneous Mobile Communication Networks, London, UK, June.

[38] Rost, P., Talarico, S. and Valenti, M. C. (2015) The Complexity‐Rate Tradeoff of Centralized Radio

Access Networks. arXiv preprint, arXiv:1503.08585.

[39] Hewlett‐Packard Company (2014) Efficient deployment of virtual network functions on HP

ProLiant m800. Technical white paper. Available at: http://h20195.www2.hp.com/V2/getpdf. aspx/4AA5‐5395ENW.pdf. Accessed: September 1, 2015.

[40] Huang, C. I. J., Duan, R., Cui, C., Jiang, J. X. and Li, L. (2014) Recent Progress on C‐RAN

Centralization and Cloudification. IEEE Access, 2, 1030–1039.

[41] Satyanarayanan, M., Bahl, P., Caceres, R. and Davies, N. (2009) The case for VM‐based cloudlets

in mobile computing. IEEE Pervasive Computing, 8(4), 14–23.

Analysis andOptimization forHeterogeneous Backhaul Technologies

Gongzheng Zhang,1 Tony Q. S. Quek,2 Marios Kountouris,3 Aiping Huang1 and Hangguan Shan

College of Information Science and Electronic Engineering, Zhejiang University,

China

Information Systems Technology and Design Pillar, Singapore University of

Technology and Design, Singapore

Mathematical and Algorithmic Sciences Laboratory, France Research Centre,

Huawei Technologies, France

5.1 Introduction

Densifying the cellular network via deploying ultra‐dense small‐cell base stations (BSs) is a promising way to meet the tremendous demand for cellular data as we approach next generation cellular networks [1, 2]. To carry the traffic from BSs to the core network and vice versa, the backhaul network needs to be enhanced proportionally. Meanwhile, low delay on the radio access and backhaul links becomes essential to deliver a wide range of services and applications in future cellular networks, for example, VoIP and online gaming with an acceptable quality of service (QoS) [3]. As a result, backhaul has become the next big challenge for providing reliable and timely connectivity between BSs and the core network [4, 5], especially for delay‐sensitive services or network functionalities.

86 Backhauling/Fronthauling for Future Wireless Systems

Unlike traditional macro‐cell BSs, which are usually directly connected to the operator’s core network through fibre with very low latency or microwave links with high reliability, small‐cell BSs are not always in easy‐to‐reach locations, for example near street level or lampposts rather than rooftops, which makes conventional fibre or microwave links impractical or cost‐inefficient. Many wired and wireless technologies have been proposed as backhaul solutions for small cells [6–10]. Wired backhaul technologies have the advantages of high reliability, high data rates and high availability. However, they may sustain long and variable delays in the backbone routes or switches due to multiple hops, especially for xDSL, which can only reach up to 200–400 metres per single hop [11]. Wireless backhaul can be deployed more easily and at lower cost. Sub‐6 GHz wireless backhaul has the advantage of non‐line‐of‐sight (NLOS) transmission, while the presence of interference due to the co‐existence issue makes the wireless links unreliable and introduces unpredictable delay. Millimetre‐wave (mmWave) technologies of 60 GHz and 70–80 GHz are another potential backhaul solution, as they offer high capacity and reliability based on line‐of‐sight (LOS) links. Due tothe small carrier wavelength and possibility of directional beamforming, the mmWave links can indeed be modelled as pseudo‐wired without interference, which makes them extremely suitable for dense small‐cell networks [12]. However, multi‐ hop implementation is needed in the absence of LOS, which causes additional delay. Due to these diverse characteristics, heterogeneous backhaul deployment willbe a potential solution. It is essential to model and compare the performance ofthese different types of backhaul technologies to provide guidelines for such a system design.

Besides their capabilities, the cost of these backhaul technologies behaves differently in terms of both deployment and operation, which is another aspect that may limit the deployment of small‐cell networks. Specifically, the deployment of a wired link (e.g. fibre or cable) is much more expensive than a wireless link, while the operational cost (e.g. the power consumption) of a wired link is much lower than a wireless link. As a result, it is quite a challenging task to identify the most appropriate and efficient solution for backhaul infrastructure, especially for dense small‐cell networks. Furthermore, optimizing the configurations so as to minimize the cost is critical for the operators.

Furthermore, in a two‐tier heterogeneous cellular network where a small‐cell network is overlaid with a macro‐cell network, BS association is another challenging problem. From a signal quality perspective, users will prefer to associate with macro‐ cell BSs due to their larger transmit power. By contrast, from a load point of view, users will prefer to associate with small‐cell BSs, which are usually underutilized. Furthermore, imperfections in the backhaul link may result in packet delay increases if users associate with small‐cell BSs. Therefore, the BS association policy should take the signal quality, load and backhaul into consideration to optimize the overall network performance.

As a timely and practically relevant topic, backhaul has been drawing much attention in recent years from many angles, including transmission, cost and system design. Various

Analysis andOptimization forHeterogeneous Backhaul Technologies 87

backhaul technologies are introduced in [4–10]. Modelling the delay perfor mance of wired networks through measurements has a long history, and the results for routers and switches can be found in [13–17]. The transmission characteristics of mmWave andsub‐6 GHz form another hot topic, and initial results for mmWave can befound in [12, 18–21]. There is some system design work taking backhaul into consideration, where the backhaul is considered as a capacity constraint [22, 23]. Some other work tries to estimate the backhaul cost via listing all the components of the network and their prices,which is practical but lacks theoretical analysis [24–27]. Cost models for some specifictechnologies are presented in [28–30]. The BS association problem is studied for heterogeneous cellular networks in [31–33]. Different joint spectrum-partitioning and bias-based BS association algorithms are proposed to maximize the user rate or rate coverage in [34–36]. However, backhaul is not considered in thatwork, even though it has a significant effect, as shown in our work, and can change the whole picture. As mentioned previously, a general backhaul model is needed, especially for studying the delay performance and designing the backhaul network to minimize cost.

The main contributions of this work are twofold. First, we propose a backhaul model for four promising backhaul technologies as a means to investigate the effect of different backhaul technologies on the network performance. Specifically:

• The packet delay in backhaul links is modelled for four technologies, includ- ing fibre, xDSL, mmWave and sub‐6 GHz, each with distinct transmission characteristics.

• The mean packet delay and delay‐limited success probability are analysed and compared for the above‐mentioned technologies. Our results show that fibre is always the best choice, sub‐6 GHz and xDSL are suitable for links with modest link length, while mmWave is a very competitive candidate for short links in terms of delay performance.

• A tractable model for quantifying the backhaul cost is presented and the mean backhaul cost per small‐cell BS is analysed. A key result is that there exists an optimal gateway density which minimizes the mean cost, and the optimal operating point depends on the ratio of per gateway cost to per unit length link cost.

The second contribution of our work is to propose a backhaul‐aware BS association policy for two‐tier cellular networks. The aim is to minimize the mean network packet delay coming from both the backhaul and radio access links. Some remarkable conclusions are obtained from numerical results. Specifically:

• Backhaul delay may dominate the mean network packet delay from gateways to users when the backhaul network does not scale proportionally to the increasing number of small cells.

• The proposed backhaul‐aware BS association policy outperforms conventional association policies with or without biasing, in terms of mean network packet delay.

88 Backhauling/Fronthauling for Future Wireless Systems

• BS association without biasing may even outperform biased ones without consid- ering backhaul, which implies that users may be misled into associating with small‐ cell BSs, thus deteriorating the system performance.

The remainder of this chapter is organized as follows. Section5.2 presents the network model, the packet delay model and the cost model. The mean packet delay and delay‐limited success probability of backhaul links are analysed in Section5.3. The backhaul cost is analysed in Section5.4. In Section5.5, the mean network packet delay is analysed, based on which the backhaul‐aware BS association is proposed and evaluated for a two‐tier cellular network. Section5.6 summarizes the chapter.

5.2 Backhaul Model

5.2.1 Network Model

We consider a two‐tier cellular network consisting of radio access and backhaul networks, with gateways, hubs, macro‐cell and small‐cell BSs and users as components, as illustrated in Figure 5.1. The macro‐cell BSs are always co‐located with gateways, while the small‐cell BSs connect to the gateways via various backhaul technologies. We denote the macro‐cell network and small‐cell network as tiers one and two, respectively. The locations of the gateways, macro‐cell BSs, small‐cell BSs and users are modelled as independent homogeneous Poisson point processes (PPPs), Φ

, Φ

and Φu of density λg, λ

b,1

b,2

, λ

and λu, respectively. Without loss of generality,

b,1

b,2

Small cell gateway

Hub

Small-cell BS

Macro cell gateway

Macro-cell BS

User

Fibre

Sub-6 GHz

xDSL

mmWave

Radio access

Figure 5.1 Heterogeneous network model embracing various backhaul technologies. Reproduced from [43] with permission from the IEEE

Analysis andOptimization forHeterogeneous Backhaul Technologies 89

n dr/

the links from gateways to the core network are assumed to be a common infrastructure for all four backhaul technologies and thus are neglected in the following.

The radio access network connects users with (macro‐cell or small‐cell) BSs through wireless links, which usually have only one hop. The backhaul network is composed of links connecting small‐cell BSs with gateways, which may be multi‐hop links depending on the type of technologies used. The backhaul links using different technologies may differ in the number of hops due to different transmission ranges per single hop. We denote the transmission range of one hop in the backhaul link by r, which is the distance to guarantee a certain minimum capacity. For instance, the transmission ranges of popular backhaul technologies can generally be ordered as

fibresub- GHz xDSL mmWav

which is determined by the link length d as

[7]. We denote the number of hops along the link by n,

. In the following, we consider that the small‐cell BS is associated with the nearest gateway. In this case, the length of a backhaul link follows a Rayleigh distribution with probability density function (pdf) given by [37]:

f dd d

().2

exp (5.1)

Therefore, the average length of the backhaul link can be obtained as

the average number of hops in a backhaul link

(1) can be estimated as [37]:

, and

(5.2)

r12

This is an optimistic estimation since the link length is defined as the physical dis-

tance between the BS and the gateway.

5.2.2 Delay Model

In this work, we focus on the packet delay from the gateway to the user, that is, the downlink scenario. The packet delay has a significant impact on the queueing and end‐to‐end delays, and it consists of the packet transmission and propagation delays along the link as well as of the processing delay at each node.

• For wired backhaul, the packet delay mainly comes from the processing time in the gateway and hubs, which means that the transmission and propagation delays can be ignored due to the relatively large capacity and highly reliable transmission of the wired backhaul.

• For wireless backhaul, the packet delay along the link mainly comes from the trans- mission time in each hop in case of retransmissions, because the decode‐and‐forward procedure is typically applied in each hop.

WILEY Backhauling Fronthauling for Future Wireless Systems User Manual

Specifications and Main Features

Frequently Asked Questions

User Manual