Sherif Adeshina Ondas milim etricas e MIMO massivo para ...

Universidade de AveiroDepartamento deElectronica Telecomunicacoes e Informatica

2020

Sherif AdeshinaBusari

Ondas milimetricas e MIMO massivo paraotimizacao da capacidade e cobertura de redesheterogeneas de 5G

Millimeter-wave Massive MIMO forCapacity-Coverage Optimization in 5GHeterogeneous Networks

Programa de Doutoramento em Telecomunicacoesdas Universidades do Minho Aveiro e Porto


2020



Tese apresentada as Universidades do Minho Aveiro e Porto para cumpri-mento dos requesitos necessarios a obtencao do grau de Doutor em Tele-comunicacoes no ambito do programa doutoral MAP-Tele realizada sob aorientacao cientıfica do Doutor Jonathan Rodrıguez Gonzalez InvestigadorPrincipal do Instituto de Telecomunicacoes Universidade de Aveiro

Apoio financeiro da Fundacao para a Ciencia e a Tecnologia (FCT) Portugalatraves da bolsa com a referencia PDBD1138232015


2020



A thesis submitted to the Universities of Minho Aveiro and Porto in partialfulfilment of the requirements for Doctoral degree in Telecommunicationsunder the MAP-Tele doctoral program The thesis was conducted under thesupervision of Doctor Jonathan Rodrıguez Gonzalez Principal ResearcherInstituto de Telecomunicacoes Universidade de Aveiro

This work is supported by the Fundacao para a Ciencia e a Tec-nologia (FCT) Portugal under the grant with reference numberPDBD1138232015

o juri the jury

presidente president Doutor Vitor Bras de Sequeira AmaralProfessor Catedratico

Universidade de Aveiro

(por delegacao do Reitor da Universidade de Aveiro)

vogais examiners committee Doutora Noelia Susana Costa CorreiaProfessora Auxiliar

Departamento de Engenharia Electronica e Informatica

Universidade do Algarve

Doutor Ramiro Samano RoblesResearch Associate

CISTER Research Centre

Instituto Superior de Engenharia do Porto

Doutor Anıbal Manuel de Oliveira DuarteProfessor Catedratico

Departamento de Electronica Telecomunicacoes e Informatica


Doutor Alexandre Julio Teixeira SantosProfessor Associado com Agregacao

Departamento de Informatica

Universidade do Minho

Doutor Jonathan Rodrıguez GonzalezInvestigador Principal (Orientador)

Instituto de Telecomunicacoes


agradecimentos acknowledgments

This PhD journey has been both exciting and challenging and many peopleand entities have contributed in varied capacities towards this phase of mylife First my gratitude goes to my Supervisor Prof Jonathan RodrıguezGonzalez for granting me the opportunity to undertake the PhD programunder his worthy supervision His mentorship and support have been ofgreat benefit to me He always provides timely painstaking and valuablefeedbacks on my work and I have benefitted immensely from his pool ofknowledge and expertise Next I would like to express my sincere appreci-ation to Dr Shahid Mumtaz and Dr Kazi Mohammed Saidul Huq for allthe help guidance supervision and motivation throughout the period Yourtechnical and non-technical support were invaluable It has been a greathonor to work with both of you Also my appreciation goes to Ms ClaudiaBarbosa for her assistance and prompt actions in resolving all administrativeissues and to Antonio Morgado for his timely help in translating the titleand abstract of this thesis to portuguese The chair and members of thejury are highly appreciated for their valuable time in reviewing this thesisand providing insightful comments

I acknowledge and deeply appreciate the financial support (PhD Scholar-ship - PDBD1138232015) that I received from the Fundacao para aCiencia e a Tecnologia (FCT) Portugal Also the work in this thesis hasbeen supported by two European Commission H2020 projects (SPEED-5Gand 5GENESIS) the THz-BEGUN project of FCTMEC and the EuropeanRegional Development Fund (FEDER) through COMPETE 2020 POR AL-GARVE 2020 FCT under i-Five Project (POCI-01-0145-FEDER-030500)Next I extend my gratitude to the management and staff of the Institutode Telecomunicacoes (IT) Aveiro Portugal for providing the enabling envi-ronment to undertake the research and also to the MAP-Tele Doctoral pro-gram scientific committee All the members of the Mobile Systems4TELLResearch Group of IT are appreciated I also appreciate many friends andcolleagues who have contributed in one way or the other during the pe-riod particularly Dr Isiaka Alimi Dr Oluyomi Aboderin Engr AkeemMufutau and Engr Stephen Ogodo as well as their respective families Iam very grateful to the Federal University of Technology Akure (FUTA)Nigeria for granting me study leave to pursue the PhD program All staff ofthe Electrical and Electronics Engineering department of FUTA are deeplyappreciated for their support Also I thank Prof Buliaminu Kareem andProf Akinlabi Oyetunji of FUTA for their immense help

I appreciate the tremendous love prayers and support of my parents MrLateef Ayoola Busari and Mrs Musiliat Kehinde Busari my siblings andin-laws I am highly and deeply indebted to my wife Hafsah OlajumokeSulaimon Thank you for your love support understanding encourage-ment and prayers I am also indebted to my lovely children HameedahMuhammad and Mahfuz for tolerating my busy schedule during the pro-gram Above all my gratitude goes to God the Almighty for His countlessblessings He says ldquoAnd He gave you of all that you asked for and ifyou count the Blessings of Allah never will you be able to count them rdquoQurrsquoan 14 vs 34

Palavras-chave Canal 3D 5a geracao agregados com formatacao de feixe hıbrida MIMOmassivo ondas milimetricas redes celulares ultra densas

Resumo As redes LTE-A atuais nao sao capazes de suportar o crescimento expo-nencial de trafego que esta previsto para a proxima decada De acordocom a previsao da Ericsson espera-se que em 2020 a nıvel global 6 milmillhoes de subscritores venham a gerar mensalmente 46 exabytes de trafegode dados a partir de 24 mil milhoes de dispositivos ligados a rede movelsendo os telefones inteligentes e dispositivos IoT de curto alcance os prin-cipais responsaveis por tal nıvel de trafego Em resposta a esta exigenciaespera-se que as redes de 5a geracao (5G) tenham um desempenho substan-cialmente superior as redes de 4a geracao (4G) atuais Desencadeado peloUIT (Uniao Internacional das Telecomunicacoes) no ambito da iniciativaIMT-2020 o 5G ira suportar tres grandes tipos de utilizacoes banda largamovel capaz de suportar aplicacoes com debitos na ordem de varios Gbpscomunicacoes de baixa latencia e alta fiabilidade indispensaveis em cenariosde emergencia comunicacoes massivas maquina-a-maquina para conectivi-dade generalizada Entre as varias tecnologias capacitadoras que estao a serexploradas pelo 5G as comunicacoes atraves de ondas milimetricas os agre-gados MIMO massivo e as redes celulares ultra densas (RUD) apresentam-se como sendo as tecnologias fundamentais Antecipa-se que o conjuntodestas tecnologias venha a fornecer as redes 5G um aumento de capacidadede 1000times atraves da utilizacao de maiores larguras de banda melhoria daeficiencia espectral e elevada reutilizacao de frequencias respectivamenteEmbora estas tecnologias possam abrir caminho para as redes sem fioscom debitos na ordem dos gigabits existem ainda varios desafios que temque ser resolvidos para que seja possıvel aproveitar totalmente a largura debanda disponıvel de maneira eficiente utilizando abordagens de formatacaode feixe e de modelacao de canal adequadas Nesta tese investigamos amelhoria de desempenho do sistema conseguida atraves da utilizacao deondas milimetricas e agregados MIMO massivo em cenarios de redes celu-lares ultradensas de 5a geracao e em cenarios rsquoinfrastrutura celular-para-qualquer coisarsquo (do ingles cellular infrastructure-to-everything) envolvendoutilizadores pedestres e veıculares Como um componente fundamental dassimulacoes de sistema utilizadas nesta tese e o canal de propagacao im-plementamos modelos de canal tridimensional (3D) para caracterizar deforma precisa o canal de propagacao nestes cenarios e assim conseguir umaavaliacao de desempenho mais condizente com a realidade Para resolver osproblemas associados ao custo do equipamento complexidade e consumode energia das arquiteturas MIMO massivo propomos um modelo inovadorde agregados com formatacao de feixe hıbrida Este modelo generico rev-ela as oportunidades que podem ser aproveitadas atraves da sobreposicaode sub-agregados no sentido de obter um compromisso equilibrado entreeficiencia espectral (ES) e eficiencia energetica (EE) nas redes 5G Os prin-cipais resultados desta investigacao mostram que a utilizacao conjunta deondas milimetricas e de agregados MIMO massivo possibilita a obtencao emsimultaneo de taxas de transmissao na ordem de varios Gbps e a operacaode rede de forma energeticamente eficiente

Keywords 3D channel 5G hybrid beamforming massive MIMO mmWave UDN

Abstract Todayrsquos Long Term Evolution Advanced (LTE-A) networks cannot supportthe exponential growth in mobile traffic forecast for the next decade By2020 according to Ericsson 6 billion mobile subscribers worldwide are pro-jected to generate 46 exabytes of mobile data traffic monthly from 24 billionconnected devices smartphones and short-range Internet of Things (IoT)devices being the key prosumers In response 5G networks are foreseento markedly outperform legacy 4G systems Triggered by the InternationalTelecommunication Union (ITU) under the IMT-2020 network initiative 5Gwill support three broad categories of use cases enhanced mobile broadband(eMBB) for multi-Gbps data rate applications ultra-reliable and low la-tency communications (URLLC) for critical scenarios and massive machinetype communications (mMTC) for massive connectivity Among the sev-eral technology enablers being explored for 5G millimeter-wave (mmWave)communication massive MIMO antenna arrays and ultra-dense small cellnetworks (UDNs) feature as the dominant technologies These technologiesin synergy are anticipated to provide the 1000times capacity increase for 5Gnetworks (relative to 4G) through the combined impact of large additionalbandwidth spectral efficiency (SE) enhancement and high frequency reuserespectively However although these technologies can pave the way to-wards gigabit wireless there are still several challenges to solve in terms ofhow we can fully harness the available bandwidth efficiently through appro-priate beamforming and channel modeling approaches In this thesis weinvestigate the system performance enhancements realizable with mmWavemassive MIMO in 5G UDN and cellular infrastructure-to-everything (C-I2X)application scenarios involving pedestrian and vehicular users As a criticalcomponent of the system-level simulation approach adopted in this thesiswe implemented 3D channel models for the accurate characterization of thewireless channels in these scenarios and for realistic performance evaluationTo address the hardware cost complexity and power consumption of themassive MIMO architectures we propose a novel generalized framework forhybrid beamforming (HBF) array structures The generalized model revealsthe opportunities that can be harnessed with the overlapped subarray struc-tures for a balanced trade-off between SE and energy efficiency (EE) of 5Gnetworks The key results in this investigation show that mmWave mas-sive MIMO can deliver multi-Gbps rates for 5G whilst maintaining energy-efficient operation of the network

Table of Contents

Table of Contents i

List of Acronyms iv

List of Symbols vii

List of Figures ix

List of Tables xi

List of Algorithms xiii

1 Introduction 1

11 Introduction 1

12 Overview of the Big Three Enablers 4

121 Millimeter-Wave Communications 4

122 Massive MIMO 4

123 Ultra-Dense Networks 5

13 Thesis Motivation 6

14 Thesis Objectives 7

15 Scientific Methodology Applied 9

16 Thesis Contributions 10

17 Organization of the Thesis 14

2 Millimeter-wave Massive MIMO UDN for 5G Networks 17

21 Evolution towards mmWave Massive MIMO UDNs 17

211 SISO to massive MIMO 18

212 Microwave to mmWave Communication 23

213 Legacy Macrocell to Ultra-Dense Small Cell Deployment 24

22 Dawn of mmWave Massive MIMO 25

221 Architecture 26

222 Propagation Characteristics 27

223 Health and Safety Issues 31

224 Standardization Activities 31

23 5G Channel Measurement and Modeling 32

231 mmWave Massive MIMO Channels 34

232 3GPP 3D Channel Models 36

i

233 NYUSIM Channel Model 38

24 Beamforming Techniques 39

241 Analog Beamforming 39

242 Digital Beamforming 41

243 Hybrid Beamforming 42

25 Conclusions 45

3 3D Channel Modeling for 5G UDN and C-I2X 47

31 Background 47

32 microWave and mmWave Channels Individual Performance 48

321 System Model 48

322 Map-based Simulation Framework 50

323 Simulation Results 54

33 Joint Channel Performance for 5G UDN 59

331 Deployment Layout 59

332 Simulation Results and Analyses 61

333 Challenges and Proposed Solutions 65

34 C-I2V Channel Performance 68

341 Network Deployment 69

342 Channel Model 70

343 Antenna Model 71


35 Conclusions 83

4 Novel Generalized Framework for Hybrid Beamforming 85

41 Background 85

42 Hybrid Beamforming Schemes 86

421 Fully-connected HBF architecture 87

422 Sub-connected HBF architecture 88

423 Overlapped subarray HBF architecture 88

424 Proposed Generalized Framework for HBF architectures 90

43 Precoding and Postcoding 93

431 Analog-only Beamsteering 94

432 Hybrid Precoding with Baseband Zero Forcing 95

433 Singular Value Decomposition Precoding 95

44 Power Consumption Model 96

45 Spectral and Energy Efficiency 98

451 Spectral Efficiency and Achievable Rate 98

452 Energy Efficiency 98

46 Hybrid Beamforming for C-I2X 99

461 System Model and Parameters 99


47 Hybrid Beamforming for C-I2P 106



48 Conclusions 112

ii

5 Conclusions and Future Work 11551 Thesis Summary 115

511 Channel Modeling 116512 Hybrid Beamforming 117

52 Future Research Directions 118521 THz Channel Modeling 118522 Ultra-massive MIMO 119523 6G for Energy Efficiency 120524 Quantum Machine Learning 120

Bibliography 122

iii

List of Acronyms

1G First generation

2D Two-dimensional

3D Three-dimensional

3G Third generation

3GPP Third generation partnership project

4G Fourth generation

5G Fifth generation

6G Sixth generation

ABF Analog beamforming

AE Antenna element

AI Artificial intelligence

AoA Angle of arrival

AoD Angle of departure

AP Access point

AS Angular spread

AWGN Additive white Gaussian noise

B5G Beyond-5G

BBU Baseband unit

BS Base station

C-I2P Cellular infrastructure-to-pedestrian

C-I2V Cellular infrastructure-to-vehicle

C-I2X Cellular infrastructure-to-everything

C-V2X Cellular vehicle-to-everything

CL Coupling loss

CN Condition number

CSI Channel state information

DBF Digital beamforming

DL Downlink

DoF Degree of freedom

DS Delay spread

DSRC Dedicated short-range communication

ECDF Empirical cumulative distribution function

iv

EE Energy efficiencyeMBB Enhanced mobile broadband

GF Geometry factor

HBF Hybrid beamformingHetNet Heterogeneous networkHPBW Half power beamwidth

iid Independent and identically distributedI2I Indoor-to-indoorICI Inter-cell interferenceIMT International mobile telecommunicationsIoT Internet of thingsISD Inter-site distanceITS Intelligent transport systemITU International telecommunication union

KF K-FactorKPI Key performance indicator

LIS Large intelligent surfaceLLS Link level simulatorLOS Line of sightLSP Large-scale parameterLTE-A Long term evolution-advanced

MAC Medium access controlMC MacrocellMIMO Multiple-input multiple-outputML Machine learningmMTC Massive machine type communicationsmmWave Millimeter-waveMPC Multipath componentMU-MIMO Multi-user MIMOMUI Multi-user interference

NF Noise figureNGMN Next-generation mobile networkNLOS Non-LOSNLS Network level simulatorNR New radio

O2I Outdoor-to-indoorO2O Outdoor-to-outdoorOFDM Orthogonal frequency division multiplexing

PDP Power delay profilePHY Physical layer

v

PL Path lossPLE Path loss exponentPS Phase shifter

QML Quantum machine learning

RAN Radio access networkRB Resource blockRF Radio frequencyRIS Reconfigurable intelligent surfaceRMS Root-mean-squareRSRP Reference signal received powerRX Receiver

SC Small cellSCM Spatial channel modelSE Spectral efficiencySF Shadow fadingSINR Signal to interference plus noise ratioSIR Signal to interference ratioSISO Single-input single-outputSLS System level simulatorSNR Signal to noise ratioSP SubpathSSCM Statistical spatial channel modelSSF Small-scale fadingSSP Small-scale parameterSVD Singular value decomposition

THz TerahertzTTI Transmission time intervalTX Transmitter

UDN Ultra-dense networkUE User equipmentULA Uniform linear arrayUMa Urban macrocellUMi Urban microcellUPA Uniform planar arrayURLLC Ultra-reliable and low latency communications

V2I Vehicle-to-infrastructureV2V Vehicle-to-vehicle

WiFi Wireless fidelityWRC World radiocommunications conference

ZF Zero forcing

vi

List of Symbols

B BandwidthBc Coherence bandwidthGo Maximum boresight gainGAE Antenna element gainGRX RX antenna gainGTX TX antenna gainJ Number of BSsK Number of UEsL Number of rays or MPCsNRX Number of RX antenna elementsNRFRX Number of RX RF chains

NTX Number of TX antenna elementsNRFTX Number of TX RF chains

NUE Number of UEsNcl Number of clustersNo Thermal noise power densityNsp Number of subpathsMPCsNRXsub Number of RX elements in a subarray

NTXsub Number of TX elements in a subarray

Ns Number of streamsPLeff Effective PLPLmax Maximum PLPBH Power consumption of the backhaulPLOS LOS probabilityPRAN Power consumption of the RANPRX Receive powerPT Transmit powerPtotal Total power consumptionn of the networkTc Channel correlation timeTt Data transmission timeTu Channel update timeΩTX Density of TXsn Path loss exponentH Channel matrixn Noise vectorx Transmit signal vector

vii

4N Subarray spacingηEE Energy efficiencyηSE Spectral efficiencyλ Wavelength(middot)H Conjugate transpose operatorφ3dB Azimuth 3dB HPBWρ Normalized transmit powerσ Noiseσ2n Noise powerτ Propagation time delayθ3dB Elevation 3dB HPBWΘ Distance-dependent phase changeϕ Phaseϑ Velocity-induced Doppler shiftξ Average antenna efficiencyd Distanced2Dminusin 2D indoor distanced2Dminusout 2D outdoor distanced3D 3D distancedr Road distancefD Doppler frequencyfc Carrier frequencyhRX RX heighthTX TX heighthdir Directional CIRvRX Velocity of users

viii

List of Figures

11 Evolution of mobile wireless communication from 1G to 6G 2

12 The ten key enabling technologies for 5G 3

13 The symbiotic cycle of the three prominent 5G enablers 5

14 Mobile traffic forecast for 2015-2024 6

15 Network layout for 5G UDN (Scenario 1) 8

16 Network layout for C-I2X (Scenario 2) 8

17 Evolved 5G system level simulator 11

18 C-I2X system level simulator 11

19 Thesis organization 15

21 Generic system model 19

22 Directional communication with mmWave massive MIMO 24

23 Candidate 5G architecture based on microWavemmWave massive MIMO UDN 26

24 Atmospheric and molecular absorption at mmWave frequencies 28

25 Rain attenuation at mmWave frequencies 28

26 Classification of 5G channel models based on modeling approach 34

27 Illustration of spherical wavefront phenomena for mmWave massive MIMO 36

28 Flow chart for the 3GPP 3D geometry-based SCM 37

29 Analog beamforming architecture 40

210 Digital beamforming architecture 41

211 Hybrid beamforming architecture 43

31 Cellular deployment layout for channel performance 49

32 ECDF of coupling loss 55

33 ECDF of geometry factor 57

34 Impact of UE height (floor level) on SINR 58

35 Impact of BS downtilt angle on SINR 58

36 5G UDN deployment layout 59

37 Impact of bandwidth on cell capacity with all users outdoors 62

38 Impact of bandwidth on cell capacity with 80 of the UEs indoors 63

39 Impact of bandwidth on SC user throughput 64

310 Impact of bandwidth on SC spectral efficiency 64

311 Impact of transmit power on cell performance 65

312 Impacts of antenna directivity and traffic type on cell performance 67

313 Impact of carrier frequency on cell performance 67

314 C-I2V deployment layout 70

ix

315 CDF of path loss for the three C-I2V technologies 73316 Path loss variation for the coverage area of one AP 74317 Path loss variation for the entire route 74318 CDF of K-Factor for the three C-I2V systems 76319 CDF of RMS delay spreads for the three C-I2V systems 77320 CDF for the number of clusters for the three C-I2V systems 78321 CDF for the number of MPCs for the three C-I2V systems 78322 Power Delay Profile snapshot from the mth AP 79323 Power Delay Profile snapshot from the nth AP 79324 Channel rank for the three C-I2V systems 81325 Channel condition number for the three C-I2V systems 81326 Data rates for the three C-I2V systems 82

41 Fully-connected HBF architecture 8842 Sub-connected HBF architecture 8943 Overlapped subarray HBF architecture 8944 Generalized HBF architecture 9145 Number of PSs required for different structures (NTX = 64) 10246 Power consumption of components for different 4N (PT = 1 W) 10247 Sum Spectral efficiency versus total transmit power 10348 Energy efficiency versus total transmit power 10449 Energy efficiency versus spectral efficiency 104410 Spectral efficiency vs transmit power for each user (4N = 4) 105411 Energy efficiency vs transmit power for each user (4N = 4) 105412 Deployment layout for C-I2P scenario 106413 Hybrid beamforming and analog combining system architecture 107414 EE-SE performance as a function of PT (baseline) 109415 EE-SE performance as a function of PT [fc = 28 GHz] 109416 EE-SE performance as a function of PT [B = 5 GHz] 110417 EE-SE performance as a function of PT [GTX = 18 dBi] 111418 EE-SE performance as a function of PT [TX = UPA] 111

x

List of Tables

11 Performance comparison of 4G 5G and 6G networks 3

21 Summary of benefits and challenges for antenna technologies 2222 Available bandwidth for mmWave frequency bands 2323 Available bandwidth for THz frequency bands 2324 Comparison of microWave and mmWave massive MIMO propagation properties 2925 Comparison of microWave and mmWave massive MIMO use cases 3026 Comparison of adapted 5G channel models 3327 Comparison of other representative 5G channel models 3528 Distribution Parameters for 3D CIR Generation 3829 Comparison of beamforming techniques 44

31 Simulation parameters for channel performance 4932 LOS probability 5033 Path loss models 5134 Antenna radiation pattern 5435 Simulation parameters for system performance 6036 Comparison of microWave MC and mmWave SC 6137 Multi-class traffic model 6638 Key simulation parameters for C-I2V 72

41 Power consumption of components 9742 HBF array structure components 10043 Power allocation for the array structures 10044 Simulation parameters for C-I2X scenario 10145 Baseline simulation parameters for C-I2P scenario 108

xi

xii

List of Algorithms

1 Map-based simulation framework 53

2 Two-Stage Multi-User Hybrid Beamforming with Baseband Zero Forcing 96

xiii

xiv

Chapter 1

Introduction

This chapter introduces the main concepts investigated in this thesis It provides anoverview on massive MIMO millimeter-wave communication and ultra-dense small cell net-work as three key technologies for enhancing the performance of next-generation mobile net-works building on the initial release of 5G This provides the context for the motivation andobjectives of this thesis the scientific methodology applied for the investigation and the the-sis contributions The chapter ends with the organization of the thesis which presents anexecutive summary of the concepts covered and elaborated in later chapters

11 Introduction

Since the early 80rsquos operators and regulators of mobile wireless communication systemsroll out a new generation of cellular networks almost every decade The year 2020 is expectedto herald a new dawn with the introduction and possible commercial deployment of thefifth generation (5G) cellular networks that will significantly outperform prior generations(ie from the first generation (1G) to the fourth generation (4G)) Widespread adoptionof 5G networks is anticipated by 2025 The 5G era is foreseen to usher in next-generationmobile networks (NGMNs) that will deliver super high speed connectivity coupled with higherreliability and spectral efficiency (SE) and lower energy consumption than todayrsquos legacynetworks The aim is to evolve a cellular network that remarkably pushes forward the limitsof legacy mobile networks across all key performance indicators (KPIs) This is motivatedby a mix of economic demands (mobile traffic growth cost energy etc) and socio-technicalconcerns (health environment technological advances etc) which render current standardsand systems unsustainable [1 2]

The 5G networks also known as the international mobile telecommunications (IMT)-2020are anticipated to support three broad categories of use cases namely enhanced mobile broad-band (eMBB) ultra-reliable and low latency communications (URLLC) and massive machinetype communications (mMTC) [3] The eMBB use case targets mobile broadband servicesrequiring high data rates and cell capacities (eg cellular mobile and vehicular networks)URLLC is for scenarios with stringent reliability and latency requirements (eg medical andpublic safety applications) and mMTC targets massive connectivity based on the internet ofthings (IoT) paradigm [4] The minimum technical requirements for 5G have been approvedin [5] and a summary of the KPIs for the uses cases is given in [3]

For the downlink (DL) eMBB use case in dense urban 5G scenarios which is the main

1

focal point of this thesis the minimum target values according to the international telecom-munication union (ITU) radiocommuication standard [5] include 20 Gbps (peak data rate)100 Mbps (user experienced data rate) 30 bpsHz (peak SE) 0225 bpsHz (5 user SE) and78 bpsHzTRxP (average SE) among others [3] The ambitious goals set for 5G networksas compared to the 4G long term evolution-advanced (LTE-A) systems include 1000times highermobile data traffic per geographical area 100times higher typical user data 100times more connecteddevices 10times lower network energy consumption and 5times reduced end-to-end (E2E) latency[6 7]

Between 1G and 4G mobile networks have moved from analog to digital voice-only tomultimedia (voice and data) circuit-switched to packet-switched networks and from 24kbps throughput to a peak data rate of 100 Mbps (for highly-mobile users) and up to 1 Gbps(for stationarypedestrian users) [6 8] With the introduction of several architectural andtechnology changes 5G aims to markedly surpass the performance of legacy networks and itsevolution has been highly dynamic migrating from research and field trials to standardizationand real deployments around 2020 and beyond Moreover sixth generation (6G) networksresearch is starting to ramp up The 6G networks are expected to address the shortcomingsof 5G networks and surpass their performance across all KPIs The evolution from 1G to 6Gis illustrated in Figure 11 A quantitative comparison of the key performance metrics of 4Gnetworks with the corresponding targets for 5G and 6G networks are shown in Table 11

Mobile Broadband ServicesSmart amp

Green WorldIntelligent Networks

Today



Today

6G amp

Beyond

1G 2G 3G 4G 5G1G 2G 3G 4G 5G

Foundation of Mobile Telephonybull Advanced Mobile

Phone Service (AMPS)

bull Total Access Communication System (TACS)

Mobile Telephony for Everyonebull Global System for

Mobile Communication (GSM)

bull Digital-AMPSbull IS-95

Foundation of Mobile Broadbandbull Wideband ndash Code

Division Multiple Access (W-CDMA)

bull High Speed Packet Access (HSPA)

bull CDMA-2000

Future of Mobile Broadbandbull Long-Term

Evolution (LTE)bull LTE-Advanced










bull CDMA-2000



Networked Societybull 5G New Radio

(NR)bull IMT-2020










bull CDMA-2000




(NR)bull IMT-2020

Testbeds Prototypes Trials CommercializationTestbeds Prototypes Trials CommercializationTestbeds Prototypes Trials Commercialization

Rel 14 Rel 15 Rel 16Rel 14 Rel 15 Rel 163GPP Rel 14 Rel 15 Rel 163GPP

mMTC ITU

URLLC

eMBBmMTC ITU

URLLC

eMBB



Today

6G amp

Beyond

1G 2G 3G 4G 5G










bull CDMA-2000




(NR)bull IMT-2020

Testbeds Prototypes Trials Commercialization

Rel 14 Rel 15 Rel 163GPP

mMTC ITU

URLLC

eMBB

Figure 11 Evolution of mobile wireless communication from 1G to 6G (adapted from [9])

2

Table 11 Performance comparison of 4G 5G and 6G networks (adapted from [12 10ndash13])

Performance metrics 4G 5G 6GPeak data rate (Gbps) 1 20 1000User experienced data rate (Gbps) 001 1 100Connection density (deviceskm2) 105 106 1016

Mobility support (kmph) 350 500 1000Area traffic capacity (Mbitsm2) 01 10 50Latency (ms) 10 5 01Reliability () 99 99999 9999999Positioning accuracy (m) 1 01 001Spectral efficiency (bpsHz) 3 10 100Network energy efficiency (Jbit)lowast 1 001 0001lowastNormalized with 4G value

To realize the promising set targets several enabling technologies are being exploredfor 5G The ten key enablers are shown in Figure 12 The dominant technology that consis-tently features in the list of enablers is the millimeter-wave (mmWave) massive multiple-inputmultiple-output (MIMO) system It is a promising technology that combines the prospectsof huge available bandwidth in the mmWave spectrum (30-300 GHz) with the expected gainsfrom massive MIMO arrays (with several tens or hundreds of antenna elements (AEs)) en-abling the opportunity to deliver the anticipated and stringent peak data rates envisaged for5G [2]

5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing

5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing

Figure 12 The ten key enabling technologies for 5G (adapted from [14])

3

When the mmWave massive MIMO technology is used in the heterogeneous network(HetNet) topology (involving a dense deployment of small cells (SCs) within the coverage areaof the umbrella macrocell (MC) otherwise referred to as the ultra-dense network (UDN))5G networks can be projected to reap the benefits of the three enablers on a very large scaleand thereby support a plethora of high-speed services and bandwidth-hungry applications nothitherto possible [15] These three enablers (mmWave massive MIMO and UDN) constitutethe subject of investigation in this thesis

12 Overview of the Big Three Enablers

Future cellular systems (5G and beyond) will employ the so-called ldquobig threerdquo technologies(i) mmWave communications (employing large quantities of new bandwidth) (ii) massiveMIMO (using many more antennas to facilitate throughput gains in the spatial dimension)and (iii) UDN (featuring extreme densification of infrastructure) The expected capacitygains from these technologies are due to the combined impact of large additional spectrumSE enhancement and high frequency reuse respectively [16] These three enablers will lead toseveral orders of magnitude in throughput gain with the goal to support the explosive demandfor mobile broadband services foreseen for the next decade In the following we provide abrief overview of the three technologies

121 Millimeter-Wave Communications

The mmWave frequency band (ie the extremely high frequency (EHF) range of theelectromagnetic (EM) spectrum representing 30-300 GHz and corresponding to wavelength1-10 mm) has an abundant bandwidth of up to 252 GHz With a reasonable assumptionof 40 availability over time [17] these mmWave bands will possibly open up sim100 GHznew spectrum for mobile broadband applications In this spectrum block about 23 GHzbandwidth is being identified for mmWave cellular in the 30-100 GHz bands excluding the 57-64 GHz oxygen absorption band which is best suited for indoor fixed wireless communications(ie the unlicensed 60 GHz band) As a key enabler for the multi-gigabit-per-second (Gbps)wireless access in NGMNs mmWave wireless connectivity offers extremely high data ratesto support many applications such as short-range communications vehicular networks andwireless in-band fronthaulingbackhauling among others [18]

122 Massive MIMO

Massive MIMO is a technology which scales up the number of AEs by several orders ofmagnitude in constrast to conventional MIMO systems (ie from up to 8 to ge 64) [19] thuscapitalizing on the benefits of MIMO on a much larger scale [20] It has the potential toincrease the capacity of mobile networks in manifolds through aggressive spatial multiplexingwhile simultaneously improving the radiated energy efficiency (EE) Using the excess degree offreedom (DoF) resulting from the large number of antennas massive MIMO can harness theavailable space resources to improve SE Moreover with the aid of beamforming it can alsosuppress interference by directing energy to desired terminals only This avoids fading dipsand thereby reduces the latency on the air interface [2122] When deployed in the mmWaveregime the corresponding downscaling of the massive MIMO antennas drastically reducescost and power not only by using low-cost low-power components but also by eliminating

4

expensive and bulky components (such as large coaxial cables) and high-power radio frequency(RF) amplifiers at the front-end [8 20]

123 Ultra-Dense Networks

UDN refers to the hyper-dense deployment of SC base stations (BSs) within the coverageareas of the MC BSs It has been identified as the single most effective way to increase networkcapacity based on its potentials to significantly raise throughput increase SE and EE as wellas enhance seamless coverage for cellular networks [15] In decreasing order of capabilities SCsare classified as metro- micro- pico- or femtocells based on power range coverage distanceand the number of concurrent users to be served The rationale behind them is to bringusers physically close to their serving BSs to enable higher data rates The overlay of SCs ontraditional MCs leads to a multi-tier HetNet where the host MC BSs handle more efficientlycontrol plane signaling (eg resource allocation synchronization mobility management etc)while the SC BSs provide high-capacity and spectrally-efficient data plane services to the users[6] This HetNet topology has the potential to deliver many benefits It increases networkcapacity based on increased cell density and high spatial and frequency reuse Moreoverit enhances SE based on improved average signal to interference plus noise ratio (SINR)with tighter interference control It also improves EE based on reduced transmission powerand lower path loss (PL) resulting from the shorter distance between each user equipment(UE) and its serving BS [61523ndash26] The three enablers exhibit a symbiotic relationship asillustrated in Figure 13

Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength

smaller antenna size

High beamforming gain

lower pathloss

Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss

Figure 13 The symbiotic cycle of the three prominent 5G enablers

5

Overall these three technology enablers are complementary in many respects Largeswathes of bandwidth required for 5G needs the mobile network to migrate to higher frequen-cies especially the promising mmWave (and terahertz (THz)) bands These high frequenciesrequire many antennas to overcome the PL in such an environment Furthermore higherfrequencies need smaller cells to mitigate blockage and PL effects and the effects in turncause the interference due to densification to decay quickly [23] The amalgam of the threefeatures produces the HetNet architecture and the mmWave massive MIMO paradigm thathave emerged as key subjects of research for 5G and beyond This promising architecture ispoised to open up new frontiers of services and applications for NGMNs It shows potentialsto significantly raise user throughput enhance the systemsrsquo SE and EE as well as increasethe capacity of mobile networks using the joint capabilities of the three enablers

13 Thesis Motivation

Several emerging use cases are being identified to address the explosive traffic demand inNGMNs where the worldwide monthly traffic is projected to grow from 5 exabytes (EB) in2015 to 136 EB by 2024 as shown in Figure 14 A large chunk of the traffic will be generatedindoors and through video applications [27]

Figure 14 Mobile traffic forecast for 2015-2024 (adapted from [28])

6

In urban metropolitan cities with dense high-rise buildings UEs are distributed on differ-ent floors Thus the propagation environment becomes three-dimensional (3D) where usersneed to be separated not only in the azimuth (horizontal) domain but also in the eleva-tion (vertical) domain [29] In addition SCs will be densely deployed to serve as the highrate hotspot tier while the MC serves as the coverage tier This UDN architecture that isillustrated in Figure 15 (on next page) is one of the two scenarios investigated in this thesis

While the system model in Figure 15 requires 3D channel models the legacy networks andthe traditional massive MIMO systems employ two-dimensional (2D) channel models withantenna array elements arranged linearly in the azimuth direction only [30] However 2Dchannel models underestimate system performance [31ndash34] as they do not account for arrayfeatures such as elevation beamforming The extra spatial DoF provided by the elevationdomain enables interference mitigation leading to considerable increase in SE [30] Thus3D channel models are critical for the accurate and realistic performance characterization of5G networks The challenge therefore is to extend the modeling approach to the elevationdomain and evaluate the system performance of the 3D 5G UDNs

Another important scenario for 5G is the cellular vehicle-to-everything (C-V2X) paradigmwhere ldquoXrdquo could be a vehicle infrastructure grid network pedestrian user etc This thesishowever focuses on the cellular infrastructure-to-everything (C-I2X) paradigm that is shownin Figure 16n (on next page) In this second scenario lampost-based access points (APs)are used to provide ultra-broadband connectivity to pedestrian users high mobility vehiclesandor a combination of both Applying mmWave spectrum at street-level sites is currentlybeing explored for capacity improvement in 5G networks and for offloading traffic from theregular BSs [35]

While the 5G UDN and C-I2X architectures can provide higher system capacities theyintuitively imply higher overall power consumption due to dense deployment of infrastructureHowever 5G networks are required to be green soft and super-fast (ie energy-efficient self-organizing and high-rate respectively) [36] Thus energy-efficient design schemes that willminimize the power consumption in order to lower the networksrsquo energy utilization and carbondioxide (CO2) gas emission footprints are critical [37] Therefore antenna array architecturesand algorithms that can deliver ultra-high data rates whilst maintaining a balanced trade-offin EE as well as hardware cost and complexity of the network are critical design goals andthus key research challenges

Motivated by these challenges the subject of investigation of this thesis is the interplayof mmWave communication and massive MIMO that targets towards capacity and coverageoptimization of 5G networks We focus specifically on the implementation of 3D microWave andmmWave channels design and analyses of beamforming schemes with massive MIMO arraysand system-level performance evaluation of the networks in UDN and C-I2X applicationscenarios with a view to enhancing cell throughputs and delivering cost-effective energy-efficient and super high speed connectivity in realizing the 5G goals

14 Thesis Objectives

The goal of this research is to optimize the capacity and coverage of 5G cellular networksfor cost-effective and ultra-high speed wireless connectivity Motivated by this goal the re-search objectives are to

7

Figure 15 Network layout for 5G UDN (Scenario 1)

Figure 16 Network layout for C-I2X (Scenario 2)

8

bull Investigate the joint application of mmWave and massive MIMO for enhanc-ing 5G cell throughputs The legacy UDNs employing microWave MIMO cannot supportthe next-generation applications due to limited bandwidth low SE and high cross-tierinterference from dense cells To address this challenge this thesis investigates the jointapplication of mmWave and massive MIMO in delivering ultra-high system capacitiesby using large mmWave bandwidth enhancing SE with massive MIMO and eliminatingcross-tier interference using a two-tier (microWave-mmWave) architecture whilst enhancingcoverage with densely-deployed SCs In the resulting 5G HetNets the mmWave SCsthat are anticipated to provide multi-Gbps throughput suffer from SINR bottleneckdue to the degrading impact of noise (due to larger bandwidth) opportunistic natureof mmWave propagation (resulting from the susceptibility to blockage) high PL andpotentially high interference due to the density of the SCs This is addressed in thisthesis through appropriate 3D beamforming for 5G UDN and C-I2X channels

bull Investigate the SE and EE trade-off of mmWave massive MIMO architec-tures Digital beamforming (DBF) (used in conventional MIMO systems) is costly andpower-hungry for massive MIMO especially at mmWave bands On the other handanalog beamforming (ABF) suffers SE limitations despite its low cost and complexityConsequently hybrid beamforming (HBF) is being extensively explored as the candi-date architecture for mmWave massive MIMO for balanced SE-EE trade-off In HBFthe mapping from the low-dimensional RF chains to the high-dimensional antennas im-pacts on the SE-EE performance Until now only the sub-connected and fully-connectedmapped structures are popular with limited investigation on the overlapped subarraystructure Consequently a novel generalized framework for HBF array structure isproposed in this thesis for a systematic analysis of performance of the different arraystructures

This thesis therefore explores the intersection of massive MIMO mmWave communi-cations and UDN in delivering a disruptive and adaptive HetNet platform for capacity andcoverage optimization of 5G networks whilst ensuring energy-efficient cost-effective and low-complexity operation of the networks

15 Scientific Methodology Applied

The main objective of this research is to provide a system-level performance evaluation onthe three key 5G technology enablers investigated in this thesis Numerical simulations math-ematical analyses and field trials are the three main approaches employed in evaluating theperformance of any new communication system Though analytically tractable mathematicalmethods are often constrained with simplifying assumptions that potentially limit their usein modeling large-scale highly-complex and dynamic networks Realistic performance can bemeasured in live operating environments However the economic and operational require-ment of such field or live tests are costly as well as practically infeasible for the early designand development stages Hence simulations have become an increasingly important approachfor the assessment of networksrsquo performance due to obvious cost and implementation advan-tages Simulations can provide the verification and validation of the key characteristics andbehaviors of the overall system [38ndash40]

9

Depending on the performance metrics under investigation simulators can be catego-rized into three link level simulator (LLS) system level simulator (SLS) and network levelsimulator (NLS) [41] While the LLS examines detailed bit-level physical layer (PHY) func-tionalities of a single link the SLS evaluates performance of links involving many BSs andUEs at the medium access control (MAC) layer with the PHY abstracted usually throughlook-up tables SLS focuses on the radio access networks (RANs) or air interface and facil-itates analyses on resource allocation capacity coverage SE and EE among others Theperformance of protocols across all layers of the network including control signaling andbackhaulfronthaul issues are evaluated with a NLS using metrics such as latency packetloss etc Besides metric-based classification simulators can also be grouped based on ra-dio access technologies supported (cellular vehicular wireless fidelity (WiFi) etc) codinglanguage (MATLAB python C++ etc) licensing option (open source proprietary freeof charge for academic use) or network scenario capabilities (LTE 5G beyond-5G (B5G)dedicated short-range communication (DSRC) etc) [39]

For this thesis the SLS approach has been used as the main performance assessmentmethodology supported by theoretical analysis to understand the performance of the systemon a wide scale deployment In terms of implementation we adopted a baseline LTE-A SLS[38] and added advanced 5G features and functionalities The evolved 5G-compliant SLS(illustrated in Figure 17 on next page) is employed for the 5G UDN performance evalua-tions in this thesis Similarly we implemented and added advanced features on a baselinechannel-only simulator [42] and transformed it into a 5G-compliant SLS for the investigationof performance pertaining to the C-I2X scenarios The evolved C-I2X SLS is illustrated inFigure 18 (shown on next page)

16 Thesis Contributions

This PhD study mainly contributes towards providing system-level and analytical under-standing of 5G UDN and C-I2X scenarios with respect to enhancing system capacity andcoverage as well as the SE and EE performance of networks using mmWave massive MIMOThe main contributions and novelty of this PhD study lies in the following This thesis

(i) contributes to the development of two 5G-compliant SLSs as tools for investigatingthe performance of advanced 5G scenarios features algorithms and models Withthe implementation of the 5G new radio (NR) frame structure 3D channel modelsmulti-tiermulti-band HetNet spatial consistency blockage and mobility models andprecodingbeamforming algorithms the tools enable the characterization analyses andperformance evaluation of 3D microWave and mmWave channels both individually andjointly for the sake of coexistence or interoperability for future emerging 5G UDN andC-I2X application scenarios involving cellular and vehicular users

(ii) analyzes the performance enhancements realizable with mmWave massive MIMO rel-ative to legacy systems such as LTE-A and DSRC in C-I2X scenarios Also the per-formance trade-offs realizable with the overlapped subarray HBF structures when com-pared to the fully-connected and sub-connected structures are investigated in this thesisIn addition we evaluate the performance of zero forcing (ZF)-based HBF in multi-userMIMO setups in contrast to analog beamsteering and singular value decomposition(SVD) precoding algorithms

10

Legend

SLS baseline available

SLS baseline adopted

Newly implemented for thesis

Concurrent Implementation

Under ImplementationOutlook

2- UE Distribution

- Constant UEs per Cell- Variable UEs per Cell- Constant UEs per Area- Stochastic -- Stochastic- Trace -- Trace

System Level Simulator

10- Applications

- Cellular WiFi Vehicular Satellite- Dual Mobility amp Connectivity DSA- Ray Tracing THz CRAN SON LAA- 5GNR Uplink

9- Techniques amp Technologies

- OFDM NOMA- FDD TDD- Microwave mmWave - MIMO Massive MIMO- LTE-A 5G Frame structure

4- Path Loss and Shadowing Model

- 2D (COST231 Dual Slope TS36942 )- 3GPP 3D (TR36873)- 3GPP 3D (TR38900)- Claussen Shadow Fading

3- Antenna Directivity

- Tri-sector (ULA UPA UCA)- Six-sector- Omnidirectional- Smart Adaptive

8- Schedulers

- SU-MIMO (RR Best CQIPF max Throughput CoMP FFR)- MU-MIMO (RR Best CQI PF max Throughput )- Admission Control Load balancing- QoS-based EE-SE co-design

7- HetNet + UDN

- Multi-tier (Femtocell RRH CoMP)- Multi-frequency (HPN Small cells)- Multi-RAT (WiFi + Cellular)

1- BS Layout

- Hexagonal- Circular- Stochastic -- Stochastic- Hybrid- Trace -- Trace

6- Mobility + Handover

- Same speed per user- Simple walking and handover- Variable user speed- Variable user direction- QoS-based handover- Inter-tier handover

5- Fast Fading Model

- 2D PDP TDL (TU PedA(B)VehA(B)Rayleigh)- Winner II+ - 3GPP 3D (TR36873)- 3GPP 3D (TR38900)

Figure 17 Evolved 5G system level simulator

mmWave Massive MIMO Channel-only

Simulator

Mobility Model

Spatial Consistency

Blockage Model

5G NR Frame Structure

Precoding and Combining

SE-EE Performance

I2X Network Layout

Channel-to-System Transformation

Generalized Array Structure

C-I2X System Level

Simulator

Figure 18 C-I2X system level simulator

11

(iii) proposes a novel generalized framework for the design and analysis of HBF antennaarray structures for any massive MIMO hybrid configuration The generalized modelenables the investigation of the SE and EE as well as the power and hardware costsof the system together with the performance trade-offs The proposed model providesinsights for the cost-effective high-rate and energy-efficient operation of next-generationnetworks and beyond

The results of this PhD study have been disseminated in the underlisted scientific publi-cations six international peer-reviewed journal articles and seven conference papers a bookchapter and five posters

bull Journal Articles

J1 S A Busari K M S Huq S Mumtaz L Dai and J Rodriguez ldquoMillimeter-WaveMassive MIMO Communication for Future Wireless Systems A Surveyrdquo IEEE Com-munications Surveys and Tutorials vol 20 no 2 pp 836-869 May 2018

J2 S A Busari S Mumtaz S Al-Rubaye and J Rodriguez ldquo5G Millimeter-Wave MobileBroadband Performance and Challengesrdquo IEEE Communications Magazine vol 56no 6 pp 137-143 June 2018

J3 S A Busari M A Khan K M S Huq S Mumtaz and J Rodriguez ldquoMillimetre-wave massive MIMO for cellular vehicle-to-infrastructure communicationrdquo IET Intelli-gent Transport Systems vol 13 no 6 pp 983-990 June 2019

J4 S A Busari K M S Huq S Mumtaz J Rodriguez Y Fang D C Sicker S Al-Rubaye and A Tsourdos ldquoGeneralized Hybrid Beamforming for Vehicular Connectivityusing THz Massive MIMOrdquo IEEE Transactions on Vehicular Technology vol 68 no9 pp 8372-8383 Sept 2019

J5 K M S Huq S A Busari J Rodriguez V Frascolla W Buzzi and D C SickerldquoTerahertz-enabled Wireless System for Beyond-5G Ultra-Fast Network A Brief Sur-veyrdquo IEEE Network vol 33 no 4 pp 89-95 JulyAugust 2019

J6 M A Khan S A Busari K M S Huq S Mumtaz S Al-Rubaye J Rodriguez andA Al-Dulaimi ldquoA novel mapping technique for ray tracer to system-level simulationrdquoComputer Communications vol 150 pp 378-383 Jan 2020

bull Conference Papers

C1 S A Busari S Mumtaz and J Rodriguez ldquoHybrid Precoding Techniques for THzMassive MIMO in Hotspot Network Deploymentrdquo IEEE Vehicular Technology Confer-ence (VTC-Spring) Workshop 2020 Antwerp Belgium pp 1-6 May 2020

C2 S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoTerahertz Massive MIMOfor Beyond-5G Wireless Communicationrdquo IEEE International Conference on Commu-nications (ICC) 2019 Shanghai PR China pp 1-6 May 2019

12

C3 S A Busari K M S Huq G Felfel and J Rodriguez ldquoAdaptive Resource Allo-cation for Energy-Efficient Millimeter-Wave Massive MIMO Networksrdquo IEEE GlobalCommunications Conference (GLOBECOM) 2018 Dubai United Arab Emirates pp1-6 Dec 2018

C4 S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoImpact of 3D ChannelModeling for ultra-high speed Beyond-5G Networksrdquo IEEE Global CommunicationsConference (GLOBECOM) Workshop 2018 Dubai United Arab Emirates pp 1-6Dec 2018

C5 S A Busari S Mumtaz K M S Huq J Rodriguez and H Gacanin ldquoSystem-LevelPerformance Evaluation for 5G mmWave Cellular Networkrdquo IEEE Global Communi-cations Conference (GLOBECOM) 2017 Singapore pp 1-7 Dec 2017

C6 M Neija S Mumtaz K M S Huq S A Busari J Rodriguez Z Zhou ldquoAn IoTBased E-Health Monitoring System Using ECG Signalrdquo IEEE Global CommunicationsConference (GLOBECOM) 2017 Singapore pp 1-6 Dec 2017

C7 S A Busari S Mumtaz K M S Huq and J Rodriguez ldquoX2-Based Handover Per-formance in LTE Ultra-Dense Networks using NS-3rdquo IARIA The Seventh InternationalConference on Advances in Cognitive Radio COCORA 2017 Venice Italy Vol 7 pp31 - 36 April 2017

bull Book Chapter

B1 S A Busari S Mumtaz K M S Huq and J Rodriguez ldquoMillimeter Wave ChannelMeasurerdquo Chapter in Encyclopedia of Wireless Networks Shen X Lin X Zhang K(Eds) Springer International Publishing Berlin May 2018

bull Posters

P1 S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoA Novel GeneralizedHybrid Beamforming for THz Massive MIMO Networksrdquo 2019 MAP-Tele WorkshopPorto Portugal Sept 2019

P2 S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoMillimeter-wave MassiveMIMO for Capacity and Coverage Optimizationrdquo Science and Technology Summit inPortugal (Ciencia 2019) Lisbon Portugal July 2019

P3 S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoTHz Massive MIMO forC-I2X in B5G networksrdquo Research Summit - Universidade de Aveiro Aveiro PortugalJuly 2019

P4 S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoMillimeter-wave Mas-sive MIMO for Capacity and Coverage Optimizationrdquo Instituto de TelecomunicacoesResearch Day Aveiro Portugal Sept 2018

P5 S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoMillimeter-wave Mo-bile Broadband for 5G Heterogeneous Cellular Networksrdquo 2017 MAP-Tele WorkshopAveiro Portugal Sept 2017

13

17 Organization of the Thesis

The remainder of this PhD thesis is organized as follows

Chapter 2 Millimeter-wave Massive MIMO UDN for 5G NetworksThe first two sections of this chapter provide a background on the evolution and trend to-wards the mmWave massive MIMO system for 5G UDNs The background connects howthe networks have transformed from employing single antennas to deploying massive MIMOantenna arrays from operating in the sub-6 GHz microWave bands to moving to the mmWaveand THz bands and from using the legacy umbrella MCs to transitioning to the ultra-denseSC network topology The overview of these so-called ldquobig threerdquo 5G technologies and theirinterplay for NGMNs are given The third section then provides the state of the art onmmWave massive MIMO channel models as the basis for the implementation of the 3D chan-nel models that are used for the investigation in later chapters The fourth section of thischapter is dedicated to beamforming techniques and array structures for mmWave massiveMIMO which are the focus of Chapter 4 where a generalized HBF framework is proposedThe last section of the chapter presents the concluding remarks

Chapter 3 3D Channel Modeling for 5G UDN and C-I2XThis chapter focuses on the performance of 5G UDNs and C-I2X networks using 3D channelmodels The chapter principally covers three aspects The first part evaluates the individualperformance of 3D microWave and mmWave channels in order to provide insights for coexistencein NGMNs The second part considers the joint performance of the 3D microWave and mmWavechannels in the light of 5G UDNs The third principal part is dedicated to the performance ofcellular infrastructure-to-vehicle (C-I2V) channels involving street-level lampost-mount APsand vehicles This part compares the mmWave massive MIMO channel in this propagationenvironment with the DSRC and LTE-A channels The challenges in these 5G scenarios arethen highlighted and analysed

Chapter 4 Novel Generalized Framework for Hybrid BeamformingThis chapter focuses on beamforming techniques and the SE and EE analyses of 5G UDNsand C-I2X networks The chapter principally covers three aspects The first part proposes anovel generalized framework for HBF in mmWave massive MIMO networks The proposedframework facilitates the comparative analysis and performance evaluation of the differentHBF configurations the fully-connected sub-connected and the overlapped subarray archi-tectures The performance of the different array structures is evaluated using a C-I2X appli-cation scenario involving lampost-mount APs and a combination of pedestrian and vehicularusers The second part considers the cellular infrastructure-to-pedestrian (C-I2P) use casebetween street-level APs and pedestrian users and evaluates the performance of the systemusing three beamforming approaches analog beamsteering (AN-BST) hybrid precoding withzero forcing (HYB-ZF) and the SVD as upper bound (SVD-UB) precoding The impacts ofvarious system parameters such as transmit power bandwidth carrier frequency antennagain number of RF chains and data streams are analyzed Also the SE and EE performanceand trade-off are presented with a view to providing useful insights for enabling high ratespectrally-efficient and green operation of next-generation mobile networks

Chapter 5 Conclusions and Future WorksThis chapter provides the summary principal findings and recommendations on the conceptsinvestigated in this thesis channel modeling beamforming and system-level performance ofmmWave massive MIMO for 5G UDN and C-I2X scenarios The future research directions arethen presented as related open issues for NGMNs in the areas of THz channel modeling ultra-

14

massive MIMO quantum machine learning (QML) and EE optimization for the forthcoming6G era

The schematic diagram for the overall organization of the thesis is shown in Figure 19

Chapter 1

Introduction

Aim and Objectives

Motivation and Methodology

Thesis Contributions

Chapter 1

Introduction

Aim and Objectives



Chapter 2

mmWave Massive MIMO UDN

5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling

UDN microWave Channel

UDN mmWave Channel

C-I2X mmWave Channel

Performance Evaluation

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4

HBF array Structures

Generalized HBF Framework

Precoding Algorithms

SE-EE Analysis


Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary

Future Research Directions

Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Figure 19 Thesis organization

15

16

Chapter 2

Millimeter-wave Massive MIMOUDN for 5G Networks

The first two sections of this chapter provide a background on the evolution and trendtowards the mmWave massive MIMO system for 5G UDNs The background elaborates onhow the networks have transformed from employing single antennas to deploying massiveMIMO antenna arrays from operating in the sub-6 GHz microWave bands to moving to themmWave and THz bands and from using the legacy umbrella macrocells to transitioning to theultra-dense SC network topology The overview of these so-called ldquobig threerdquo 5G technologiesand their interplay for next-generation mobile networks are given The third section thenprovides the state of the art on mmWave massive MIMO channel models as the basis forthe implementation of the 3D channel models that play a pivotal role in later chapters Thefourth section of this chapter is dedicated to beamforming techniques and array structures formmWave massive MIMO which are the focus of Chapter 4 where a generalized HBF frameworkis proposed The last section of the chapter presents the concluding remarks

21 Evolution towards mmWave Massive MIMO UDNs

Over time mobile network technologies have continued to evolve pushing legacy systemstowards their theoretical limits and motivating research for next-generation networks withbetter performance capabilities in terms of reliability latency throughput SE EE and costamong others [23] Thus cellular networks have undergone remarkable transitions from SISOat microWave frequencies to the latest legacy system with massive MIMO at microWave frequencies(hereafter referred to as conventional massive MIMO) However the projections of explosivegrowth in mobile traffic unprecedented increase in connected wireless devices and proliferationof increasingly smart applications and broadband services have motivated research for thedevelopment of 5G mobile networks These networks are expected to deliver super high speedconnectivity and high data rates provide seamless coverage support diverse use cases andsatisfy a wide range of performance requirements where legacy cellular networks have reachedtheir theoretical limits This has prompted the academic and industrial communities toconsider mmWave massive MIMO with a view to exploiting the joint capabilities of mmWavecommunications and massive MIMO antenna arrays [2]

Notably from 4G the performance of cellular networks has significantly improved since theadoption of the HetNet architecture such that the UDN layout has been identified as the single

17

most effective way to increase system capacity in next-generation networks [15] The overlayof SCs on traditional MC BSs provides multi-dimensional benefits for mobile networks Itleads to enhanced coverage and capacity due to increased cell density leading to higher spatialand frequency reuse It also improves SE due to higher SINR with tighter interference controland increases EE due to lower transmission powers and reduced inter-site distance (ISD) of theSCs (when compared with the MCs) [4344] In addition cross-tier interference is eliminatedwhen the MC and SC tiers operate on different frequency bands thereby further increasingthe potentials of the topology Intra-tier interference can be controlled by maintaining theoptimal cell density threshold and through advanced interference mitigation techniques For5G networks therefore UDNs employing mmWave massive MIMO are expected to providethe 1000-fold network capacity increase (when compared to LTE-A) for meeting the foreseenexplosive traffic demands of mobile networks [23] In the following sub-sections we presentan overview of the road towards mmWave massive MIMO UDNs

Notations Throughout this thesis we use the following notations X is a matrix x isa vector and x is a scalar |middot| is the magnitude of a vector or determinant of a matrix middot isthe norm of a vector middotF denotes the Frobenius norm while [X]ij represents the elements

of X in the ith row and jth column The identity matrix is I while the inverse transpose andconjugate transpose operators are denoted with (middot)minus1 (middot)T and (middot)H respectively X Ymeans X is far greater than Y and tr (middot) is the trace operator

211 SISO to massive MIMO

Single-input single-output (SISO) systems employ single antennas one positioned at thetransmitter (TX) or BS and another at the receiver (RX) or UE MIMO systems use multipleantennas at both sides MIMO offers higher capacity and reliability than SISO systems as itschannels have considerable advantages over SISO channels in terms of multiplexing diversityand array gains [4546] The diversity gains of MIMO scale with the number of independentchannels between the TX and the RX while the maximum multiplexing gain is given by thelesser number of antennas on either the TX or RX units (ie min(NRX NTX)) Howevermaximum multiplexing and diversity gains cannot be simultaneously harnessed from MIMOsystems as a fundamental trade-off exists between both The best of the two gains cannot beachieved concurrently [46] Massive MIMO on the other hand uses a much larger numberof antennas (eg NTX = 64 1024) than those used in conventional MIMO systems withNTX = 2 8 To analyze the different configurations consider a generic system modelshown in Figure 21 (on next page)

The multi-cellular DL system consists of J BSs each equipped with NTX antennas and KUEs with NRX antennas per UE For forallj isin 1 2 J and forallk isin 1 2 K the receivedsignal vector y can be modeled as (21) and (22)

y = Hx + n (21)

ykj1

ykjNRX

=

hkj11 middot middot middot hkj1NTX

hkjNRX 1 middot middot middot hkjNRX NTX

xj1

xjNRX

+

nk1

nkNRX

(22)

18

where H x and n represent the channel matrix transmit signal vector and the noise vectorrespectively and n is assumed to be the additive white Gaussian noise (AWGN) followinga complex normal distribution CN (0σ) with zero mean and σ standard deviation For an-alytical tractability we focus on a single cell (ie J = 1) under the following four scenar-ios SISO point-to-point or single-user MIMO (SU-MIMO) (conventional) multi-user MIMO(MU-MIMO) and massive MIMO

BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1UE1

1

NRXUE1

1

NRX

UE2UE2

1

NRXUE2

1

NRX

UEKUEK

1

NRXUEK

1

NRX

BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1

1

NRX

UE2

1

NRX

UEK

1

NRX

Figure 21 Generic system model

(a) SISO [NTX = 1 NRX = 1K = 1]

For the SISO scenario H x and n in (21) become scalars The received signal y thusreduces to (23) The SE (bitssHz) of the single link can thus be expressed as (24)

y1 = h11x1 + n1 (23)

ηSISOSE = log2 (1 + SNR) = log2

(1 + h2PT

σ2n

) (24)

where PT is the transmit power SNR is the signal to noise ratio σ2n is the noise power and

h is the channel coefficient

(b) SU-MIMO [NTX gt 1 NRX gt 1K = 1]

Here both the TX and RX are equipped with multiple antennas For the SU-MIMOsystem the received signal vector y isin CNRXtimes1 can be expressed as (25)

19

y =radicρHx + n (25)

where x isin CNTXtimes1 n isin CNRXtimes1 H isin CNRXtimesNTX and ρ is a scalar representing the normal-ized transmit power Assuming perfect channel state information (CSI) at the receiver theSE (bitssHz) for the SU-MIMO is given by (26)

ηSUminusMIMOSE = log2

∣∣∣∣(INRX +ρ

NTXHHH

)∣∣∣∣ (26)

The expression in (26) is bounded by (27) where the actual achievable SE depends on thedistribution of the singular values of HHH [45]

log2 (1 + ρNRX) le ηSUminusMIMOSE le min (NRX NTX) log2

(1 +

ρmax (NRX NTX)

NTX

) (27)

The SU-MIMO configuration leads to increased data rates and average SE without anyincrease in the SNR or the bandwidth of such systems as required by the Shannon capacitytheorem on the theoretical limit for SISO systems This additional increase in capacity comesfrom spatial multiplexing through multi-stream transmissions from the multiple antennas Itis also possible to use the additional antennas for beamforming (and thus increase the SNR)or for diversity (so as to increase the reliability of the system) [15 47] While the channelcapacity of SISO systems increases logarithmically with an increase in systemrsquos SNR MIMOsystemsrsquo capacities increase linearly with increasing number of antennas (ie scales with thesmaller of the number of TX or RX antenna when the channel is full rank) The gains can belimited (ie not exactly linear increase) when the channel is not full rank [48] This increaseis however at the expense of the additional cost of deployment of multiple antennas spaceconstraints (particularly for mobile terminals) and increased signal processing complexities[49]

(c) MU-MIMO [NTX gt 1 NRX ge 1K gt 1]

MIMO systems have two basic configurations SU-MIMO and MU-MIMO [45 50] MU-MIMO offers greater advantages [20] over SU-MIMO and these include

bull MU-MIMO exploits multi-user diversity in the spatial domain by allocating a time-frequency resource to multiple users while SU-MIMO transmissions dedicate all time-frequency resources to a single terminal This results in significant gains over SU-MIMOparticularly when the channels are spatially correlated [50]

bull MU-MIMO BS antennas simultaneously serve many users where relatively cheap single-antenna devices can be employed at user terminals while expensive equipment is onlyneeded at the BS thereby bringing down cost

20

bull MU-MIMO system is less sensitive to the propagation environment when comparedto the SU-MIMO system Therefore rich scattering is generally not required in theMU-MIMO case

For the MU-MIMO scenarios the BS transmits simultaneously to multiple users In thesimple case where each UE has a single antenna the receive transmit and noise vectors in(25) become y isin CKtimes1 x isin CNTXtimes1 and n isin CKtimes1 respectively The channel matrixbecomes H isin CKtimesNTX and the achievable SE can be expressed as (28)

ηMUminusMIMOSE = max log2

∣∣(INRX + ρHPHH)∣∣ (28)

where P is a positive diagonal matrix with power allocations P = p1 p2 middot middot middot pK whichmaximizes the sum transmission rate [45]

Overall MIMO is a smart technology aimed at improving the performance of wirelesscommunication links [47] Compared to the SISO systems MIMO systems have been shownfrom studies and commercial deployment scenarios in different wireless standards such asthe IEEE 80211 (WiFi) IEEE 80216 (Worldwide Interoperability for Microwave Access(WiMAX)) the third generation (3G) Universal Mobile Telecommunications System (UMTS)and the High Speed Packet Access (HSPA) family series as well as the LTE to offer significantimprovements in the performance of cellular systems with respect to both capacity andreliability [4551] In addition MIMO system implementations have shifted to MU-MIMO inrecent years due to its superior benefits which have made it the candidate for several wirelessstandards [2045] However despite its great significant advantages over SISO and SU-MIMOantenna systems MU-MIMO has been identified as a non-scalable technology and as a sequelmassive MIMO is evolving to scale up the benefits of MIMO significantly Unlike MU-MIMOwhich has roughly the same number of terminals and service antennas massive MIMO hasan excess of service antennas over active terminals which can be used for enhancements suchas beamforming in order to bring about improved throughput and EE [20]

(d) Massive MIMO [NRX NTX NRX rarrinfin or NTX NRX NTX rarrinfinK gt 1]

Massive MIMO is also known as large-scale antenna system full dimension MIMO verylarge MIMO and hyper MIMO It employs an antenna array system with a few hundredBS antennas simultaneously serving many tens of user terminals on the same time-frequencyresource [20 50] It has the potential to enormously improve SE by using its large numberof BS transmit antennas to exploit spatial domain DoF for high-resolution beamforming andfor providing diversity and compensating PL thereby improving the EE data rates and linkreliability [1952] With the practical acquisition of CSI massive MIMO achieves an order-of-magnitude higher SE in real life than the small-scale conventional MIMO system The thirdgeneration partnership project (3GPP) has been steadily increasing the maximum number ofantennas in progressive LTE releases and massive MIMO will be a key ingredient in 5G [53]

When the number of antennas grows large such that NTX NRX NTX rarr infin theachievable SE for the massive MIMO system in (26) tends to (29) and when NRX NTX NRX rarrinfin it approximates to (210) Equations (29) and (210) assume that the row or col-

umn vectors of the channel H are asymptotically orthogonalHHH

NTXasymp INRX and demonstrate

21

the advantages of massive MIMO where the capacity grows linearly with the number of theemployed antenna at the BS or the UE as the case may be [45] Even at mmWave frequen-cies it is still possible to employ massive MIMO arrays for spatial multiplexing alongsidebeamforming in order to increase system capacity [54 55] However the multiplexing gainreduces in line of sight (LOS) propagation environments [45]

ηMassiveminusMIMOSE asymp NRX log2 (1 + ρ) (29)

ηMassiveminusMIMOSE asymp NTX log2

(1 + ρ

NRX

NTX

) (210)

Massive MIMO requires CSI for both uplink (UL) and DL and depends on phase coherentsignals from all the antennas at the BS It is an enabling technology for enhancing the EESE reliability security and robustness of future broadband networks both fixed and mobileHowever despite these obvious benefits massive MIMO implementation is faced with somechallenges which have been subjects of research studies These challenges among othersinclude the following [8 192050]

bull need for simple linear and real-time techniques and hardware for optimized processing ofthe vast amounts of generated baseband data at associated internal power consumption

bull need for new and realistic characterization and modeling of radio channels taking intoaccount the number geometry and distribution of the antennas

bull need for accurate CSI acquisition and feedback mechanisms and techniques to combatpilot contamination and effects of hardware impairments due to the use of low-costlow-power components

bull need for the development of commercial and scalable prototypes and deployment sce-narios to engineer the heterogeneous network solutions for future mobile systems

These challenges are being addressed by various works with a view to enabling the practicaluse of massive MIMO for 5G and beyond As of 2020 practical massive MIMO systemsare already being tested trialed and deployed [56ndash58] The summary of the benefits andchallenges of SISO SU-MIMO MU-MIMO and massive MIMO are shown in Table 21 (whereX means benefit times means challenge and the numberamount of the symbols signifies thenormalized quantity relative to SISO) [59]

Table 21 Summary of benefits and challenges for antenna technologies

Features SISO SU-MIMO MU-MIMO Massive MIMODiversity gain times X XX XXXXMultiplexing gain times XX XXX XXXXArray gain times XX XX XXXXComputational complexity times timestimes timestimestimes timestimestimestimesChannel estimation times timestimes timestimestimes timestimestimestimesPilot contamination times timestimes timestimestimes timestimestimestimes

22

212 Microwave to mmWave Communication

In legacy networks the operation of cellular networks has been mainly limited to the con-gested sub-6 GHz microWave frequency bands Though these bands have favorable propagationcharacteristics however the total available bandwidth of 1-2 GHz is grossly insufficient tosupport the foreseen traffic demand of next-generation mobile services and applications Thischallenge pushes for the exploration of under-utilized higher frequency bands where there isan abundant amount of bandwidth In the 30-100 GHz mmWave band there is more than10times bandwidth than that available at microWave bands as shown in Table 22 In the 01-10 THzbands there is even much more (contiguous) bandwidth available as shown in Table 23

Table 22 Available bandwidth for mmWave frequency bands (adapted from [819232460])

Frequency (GHz) 225-235 275-312 386-400 405-425 455-469Bandwidth (GHz) 10 13 14 20 14Frequency (GHz) 472-482 482-502 710-760 810-860 920-950Bandwidth (GHz) 10 20 50 50 29

Table 23 Available bandwidth for THz frequency bands (adapted from [6162])

Frequency (THz) 01-02 02-027 027-032 033-037 038-044 044-049Bandwidth (GHz) 100 70 50 35 65 56Frequency (THz) 049-052 052-066 066-072 084-094 066-084 094-103Bandwidth (GHz) 40 123 60 142 47 58Frequency (THz) 103-13 13-135 135-149 149-156 156-183 183-198Bandwidth (GHz) 38 51 92 29 25 56

When compared to the congested sub-3 GHz microWave bands used by the second generation(2G)-4G cellular networks and the additional television white space (TVWS) and other sub-6GHz microWave frequencies approved at the ITUrsquos world radiocommunications conference (WRC)2015 the mmWave (and THz) bands offer several advantages in terms of larger bandwidthwhich translate directly to higher capacity and data rates and smaller wavelengths enablingmassive MIMO and adaptive beamforming techniques The relatively closer spectral alloca-tions in the mmWave bands lead to a more homogeneous propagation unlike the disjointedspectrum in legacy networks On the other hand mmWave signals are prone to higher PLhigher penetration loss severe atmospheric absorption and more attenuation due to rainwhen compared with microWave signals In addition they are vulnerable to blockages by objectsThus directional communication is being employed in mmWave systems to limit the severepropagation losses and mitigate interference thereby ensuring higher throughput and betterEE [8606364]

However recent research studies and channel measurement campaigns carried out at dif-ferent mmWave frequencies (eg 28 38 60 and 73 GHz) have revealed that mmWave signalscan mitigate the aforementioned challenges by employing adaptive beamforming techniquesto suppress interference and use relay stations to circumvent obstacles thereby avoiding block-ages [19 65ndash67] as shown in Figure 22 More so for the 50-200 m cell size envisaged formmWave SCs the expected 14 dB attenuation (ie 7 dBkm) due to heavy rainfalls has aminimal effect [60] The high PL limits inter-cell interference (ICI) and allows more frequencyreuse which improves the overall system capacity [68] In addition the huge spectrum of-

23

fered by the mmWave band will enable radio access (BS-UE) fronthaul (BS-baseband unit(BBU)) and backhaul (BBU-core network) links to support much higher capacity than legacy4G networks [60] In fact the supposed drawback of short-distance or short-range mmWavecommunication perfectly fits into the UDN trend and opens up new avenues for short-rangeapplications such as the potential use in data centers and C-I2X use cases

Reflector

Obstacle

Relay

BS

UE 2

UE 3UE 3

UE 1UE 1UE 1

UE 4UE 4

Figure 22 Directional communication with mmWave massive MIMO (adapted from [65])

213 Legacy Macrocell to Ultra-Dense Small Cell Deployment

The early generations of cellular networks had cell sizes on the order of hundreds ofsquare kilometers However the cell sizes have been increasingly shrinking as this has beenamply demonstrated the most-effective way to increase system capacity [3166970] For 5Gextreme densification of SCs within the coverage area of the MC is targeted as one of thecore methods to improve the area SE ((bitssHz)m2) towards realizing the 1000times increasein network capacity relative to legacy 4G system [16] This UDN topology enables trafficoffloading to the SCs (with coverage in the range of tens of meters) particularly for indoorhotspot and dense urban SCs The smaller cell sizes of these SCs allow the reuse of spectrumacross a geographical area and reduces the number of users competing for resources at eachBS [316] In addition the shorter ISD between the SCs and their UEs leads to higher SINRand increased user throughputs [4344]

In the first deployment phase of 5G the orthogonal deployment of cells is envisaged wherethe MCs and SCs operate at different sub-6 GHz frequencies For the second phase of 5G thedeployment of UDNs at microWave mmWave and even THz frequencies is expected to provide

24

much larger peak data rates in the range of multi-Gbps or even Tbps [156970] By employingthe enabling technologies such as mmWave and massive MIMO for example 5G UDNs willsupport the foreseen explosive traffic demands of future mobile networks whilst satisfyingperformance requirements such as better reliability reduced latency and higher SE and EEamong others [23] Despite the anticipated BS densification gains increasing cell density mayalso result in increased other-cell interference (OCI) thus necessitating interference mitigationtechniques such as cooperative scheduling coordinated multipoint etc [3] Also due to OCIan unlimited increase in the number of SCs is counter-productive Therefore the optimaldensity threshold must be maintained in order to reap the full benefits of UDN More soextreme densification will also lead to other challenges in the areas of mobility support userassociation load balancing costs of installation maintenance backhauling etc [16]

22 Dawn of mmWave Massive MIMO

MmWave massive MIMO is a promising candidate technology for exploring new frontiersfor NGMNs starting with 5G networks It benefits from the combination of large avail-able bandwidth (mmWave frequency bands) and high antenna gains (achievable with massiveMIMO antenna arrays) enhanced SE and EE increased reliability compactness flexibilityand improved overall system capacity This is expected to break away from todayrsquos techno-logical shackles taking a step towards addressing the challenges of the explosively-growingmobile data demand and open up new scenarios for future applications [2] For mmWavemassive MIMO systems maximum benefits can be achieved when different TX-RX antennapairs experience independently-fading channel coefficients This is realizable when the AEsrsquospacing is at least 05λ where λ reduces with increasing carrier frequency (fc) a higher num-ber of elements in antenna arrays of same physical dimension can be realized at mmWavethan at microWave frequencies [59]

At mmWave frequencies the dimensions of the AEs (as well as the inter-antenna spacing)become incredibly small (due to their dependence on λ) It thus becomes possible to pack alarge number of AEs in a physically-limited space thereby enabling compact massive MIMOantenna array not only at the BSs but also at the UEs [71 72] As of 2019 the maximumnumbers of antennas under consideration by 3GPP (for example at 70 GHz) are 1024 for theBSs and 64 for the UEs As for the RF chains the maximum numbers are 32 and 8 for theBSs and UEs respectively [3 4]

Several works have shown that λ2 element spacing leads to low spatial correlation How-ever inter-element spacing less than λ2 can facilitate a reduction in the total area of thearray The potentially resulting higher spatial correlation can however be suppressed byproper tuning of the antennas mutual coupling using inter-element spacing of 037λ (incontrast to the conventional 05λ) the authors in [73] showed that mutual coupling can becontrolled by placing a slot between each pair of antenna elements This leads to improvementin SNR as well as reduction in the size of antenna array With low radiation power (due tosmall antenna size) and high propagation attenuation it becomes necessary to use highly-directional steerable configurable or smart antenna arrays for mmWave massive MIMO inorder to ensure high received signal power for successful detection [74] An overview of thearchitecture and the propagation characteristics of mmWave massive MIMO is given in thefollowing subsections

25

221 Architecture

A representative architecture for the 5G network is shown in Figure 23 The archi-tecture is a multi-tier HetNet composed of the MC and SC BSs with massive MIMO andmicroWavemmWave communication capabilities It features several scenarios that are subject toongoing research in realizing the 5G goals Some of these scenarios are highlighted as follows[59]

bull Split control and data plane framework where the control signals are handled by thelong-range microWave massive MIMO MC BS (for efficient mobility and other control sig-naling) while data signals are handled by the mmWave massive MIMO SCs for highcapacity

bull Dual-mode or dual-band SCs where close-by users are served by mmWave access linkswhile further-away users are served at microWave frequency band thereby serving as adynamic cell and emulating the ldquocell breathingrdquo concept from legacy networks

bull In-band backhauling where both the access and backhaul links are on the same (mmWave)frequency band to reduce cost and optimize spectrum utilization

bull Vehicular communication where the microWave MC BSs serve the long-range and highlymobile users while the SCs serve the pedestrianlow-mobility users

bull Virtual cells where a user particularly cell-edge user chooses its serving BS without be-ing constrained to be served only by the closest BS as in the legacy user association andhandover approach based on coverage zone This also extends to the multi-connectivityapproach where a user is attached to more than one BS at the same time [75]

microwave

mmWave

Control

Data

X2-interface

Vehicular Comm cell

Virtual cell

Macro BS

Dual-mode small cell

Long range macro BS user

Split control amp data small cell

Massive MIMO antenna array

Figure 23 Candidate 5G architecture based on microWavemmWave massive MIMO UDN

26

The architecture in Figure 23 brings to the forefront many opportunities in terms ofpossible scenarios use cases and applications Some of the scenarios have been consideredlately for legacy systems and will be advanced in 5G while new ones will also emerge Therealization of the architecture in Figure 23 is being pursued for 5G through multi-disciplinaryand cross-layer approaches

222 Propagation Characteristics

Marked differences exist in the propagation characteristics of mmWave massive MIMOnetworks and those of legacyconventional cellular systems At mmWave frequencies andas the number of BS antennas goes to infinity the channel characteristics generally becomedeterministic Different users experience asymptotic channel orthogonality and fewer userterminals can be supported due to reduced coverage area [2] Also signals propagated atmmWave frequencies experience higher PL (which increases with increase in fc) [17] and alsohave reduced penetrating power through solids and buildings Thus they are significantlymore prone to the effects of shadowing diffraction and blockage as λ is typically less thanthe physical dimensions of the obstacles [76ndash78] In addition mmWave signals suffer moreattenuation due to rain [79] have increased susceptibility to atmospheric absorption [80] andexperience higher attenuation due to foliage than microWave signals [81] The attenuations dueto atmosphericmolecular absorption and rain as a function of fc are presented in Figures 24and 25 (shown on next page) respectively

As shown in Figure 24 the specific attenuation due to atmospheric and molecular ab-sorption have peaks around 60 and 180 GHz The authors in [82] using air composition andatmospheric data have shown that the peaks are due to the high absorption coefficient ofoxygen (O2) and water vapor (H2O) at 60 GHz and 180 GHz respectively These two fre-quency bands are thus best suited for short distance indoor applications and the unlicensed60 GHz mmWave WiFi has already taken the lead in this direction through its standard-ization Further the experienced attenuation and molecular noise at mmWave frequencies(excluding the 60 GHz band) vary with the time of the day and the season of the year beingmore pronounced during the night than the day and more during winter than in summerdue to the combined effect of the fall in temperature and the corresponding rise in humidity[82] Though the impact of these attenuation effects limits communication coverage and linkquality the impact is however minimal for the average SC sizes of 50-200 m envisaged for5G SC networks More so beamforming is being employed to increase the array gains andimprove the SNR in order to counter the effects of the comparatively higher PL at mmWavefrequencies (when compared to the microWave propagation) [4 60]

Overall the losses in mmWave systems are higher than those of microWave systems How-ever the smaller wavelength (which enables massive antenna arrays) and the huge availablebandwidth in the bands can compensate for the losses to maintain and even drastically boostperformance gains with respect to SE and EE provided evolving computational complexitysignal processing and other implementation issues are addressed [5383] The marked differ-ences in propagation characteristics at microWave and mmWave frequencies necessitate changesin the architecture and applications of cellular networks as evident in the candidate archi-tecture earlier shown in Figure 23

27

50 100 150 200 250 300

Frequency (GHz)

10-3

10-2

10-1

100

101

102

Spe

cific

Atte

nuat

ion

(dB

km

)

Figure 24 Atmospheric and molecular absorption at mmWave frequencies [4]

50 100 150 200 250 300

Frequency (GHz)

10-2

10-1

100

101

102

Rai

n A

ttenu

atio

n (d

Bk

m)

025 mmh25 mmh125 mmh25 mmh50 mmh100 mmh150 mmh200 mmhr

Figure 25 Rain attenuation at mmWave frequencies [4]

28

In Table 24 we present a summary of the fundamental differences between conventional(microWave or sub-6 GHz) massive MIMO and mmWave massive MIMO The comparison in Table24 shows the challenges which have to be addressed or exploited in order to realize the antic-ipated benefits of mmWave massive MIMO networks Table 25 compares the potential usecases based on the propagation characteristics in LOS non-LOS (NLOS) outdoor-to-outdoor(O2O) indoor-to-indoor (I2I) and outdoor-to-indoor (O2I) environments [53] Neglecting thesmall-scale fading (SSF) the received power (as a function of the separation distance (d))can be modeled as (211) and (212) for the microWave and mmWave massive MIMO systemsrespectively Unlike in microWave (211) it can be seen that that the shadow fading (SF) andthe path loss exponent (PLE) in the mmWave case (212) are functions of blockage () [2]

Table 24 Comparison of microWave and mmWave massive MIMO propagation properties

Properties microWave massive MIMO mmWave massive MIMOPath loss Lower PL compared to mmWave at

same d At fc = 2 GHz and d = 500m typical of MCs PLmax = 9341368 and 1693 dB can be expectedin LOS NLOS and O2I scenariosrespectively [84]

Higher pathloss compared to microWaveat same d At fc = 28 GHz and ford=100 m typical of SCs maximumPLmax of 1033 1238 and 1542 dBcan be expected in LOS NLOS andO2I scenarios respectively [85]

Shadow Fading Independent random variable smalland independent of blockage NLOSpropagation Typical values are 46 and 7 for LOS NLOS and O2Irespectively [84]

Large dependent on other ran-dom variables and mainly caused byblockage LOSnear-LOS propaga-tion Typical values are 31 78 and9 for LOS NLOS and O2I respec-tively [85]

Interference Distance-dependent Dominated bya few nearby ones leads to back-ground interference floor for largenumber of interferers

Not really distance-dependent as-sumes an ON-OFF type of behaviorStrongly attenuated by randomly-aligned antenna gain patterns andblockage

SINR Changes slowly from cell center tocell edge

Undergoes extremely-random rapidfluctuations assumes an ON-OFFdepending on the beam steering effi-ciency blockage and random beamalignment

Antenna Array Singly-massive only the BSs havemassive antenna array

Double-massive both the BSs andthe UEs have massive antenna arraycapability

Signal Processing Moderately-complex particularlyCSI acquisition and feedback

Highly-complex due very largeamount of CSI

Handover Usually done at cell edges based onsignal strength for load balancingconsiderations

Occurs more frequently because ofblockage beam alignment and highnetwork density

PmicroWaveRX (d) asymp PTGTRxPXSF (d)

(λ

4π

)2

dminusα (211)

PmmWaveRX (d ) asymp PTGTRxPXSF (d )

(λ

4π

)2

dminusα(d) (212)

29

where PT is the TX power PRX is the receive power GTRxP is the combined gains of the TXand RX XSF is the SF function and α is the PLE Other differences between microWave massiveMIMO and mmWave massive MIMO are further highlighted in Table 25

Table 25 Comparison of microWave and mmWave massive MIMO use cases (adapted from [53])

Use case microWave massive MIMO mmWave massive MIMOBroadband access High data rates in most prop-

agation scenarios (eg sim100Mbpsuser using 40 MHz ofbandwidth) with uniformlygood quality of service (QoS)

Huge data rates (eg 10Gbpsuser using several GHz ofbandwidth) in some propagationscenarios

IoT mMTC Beamforming gain gives power-saving and better coverage thanlegacy networks

Not fit for low data rate applica-tions which will incur significantpower overhead

URLLC Channel hardening improves re-liability over legacy networks

Difficult due to unreliable prop-agation

Mobility support Same great support as in legacynetworks

Theoretically possible but verychallenging

High throughput fixed link Narrow beamforming is possiblewith 100 antennas 20 dB beam-forming gain is achievable onlyarray size limits the gain

Possibly even higher beamform-ing gain than at sub-6 GHz sincemore antennas fit into a givenarea

High user density Spatial multiplexing of tens ofUEs is feasible and has beendemonstrated in field-trials

Same capability as at sub-6 GHzin theory but practically limitedif hybrid implementation is used

O2O I2I communication High data rates and reliability inboth LOS and NLOS scenarios

Huge data rates in LOShotspots but unreliable dueto blockage phenomena

O2I communication High data rates and reliability Highly difficultinfeasible due topropagation losses

Backhaul fronthaul Can multiplex many links butrelatively modest data rates perlink

Great for LOS links particularlyfor fixed antenna deploymentsbut less suitable for NLOS links

Operational regime Mainly interference-limited incellular networks due to highSNR from beamforming gainsand substantial inter-user inter-ference

Mainly noise-limited in indoorscenaros due to the huge band-width and limited ICI but canbe interference-limited in out-door scenarios

In terms of benefits the larger bandwidth available in the mmWave bands when comparedto the microWave bands enables new applications such as wireless fronthauling the higher PLfavors high-rate short-range communications while the shorter wavelength allows the anten-nas at both the BSs and UEs to scale up (ie go massive) due to the dramatic reduction inantenna sizes However new challenges evolve The combination of the huge bandwidth andmassive antenna arrays translate to heavy computational load and signal processing due tothe large amount of CSI that have to be processed for channel estimation channel feedbackprecoding etc Also mmWave signaling will lead to a higher frequency of handovers in UDNif not properly managed

30

Notwithstanding the promising potentials of mmWave massive MIMO technology a broadrange of challenges spanning the length and breadth of communications theory and engineer-ing has to be addressed The challenges arise due to the differences in the architecture andpropagation characteristics of mmWave massive MIMO networks when compared to priorsystems Among others the key challenges include channel modeling antenna and RFtransceiver architecture design waveforms and multiple access schemes information theo-retic issues channel estimation techniques modulation and EE issues MAC layer designinterference management mobility management health and safety issues system-level mod-eling experimental demonstrations tests and characterization standardization and businessmodels [2]

223 Health and Safety Issues

With the mmWave bands being promising candidates for future broadband mobile com-munication networks it is essential to understand the impacts of mmWave radiation on thehuman body and the potential health effects related to its exposure In addition the currentsafety rules regarding RF exposure do not specify limits above 100 GHz whereas spectrumuse will inevitably move to these bands over time hence the need for further investigationsto codify safety metrics at these frequencies [2] The mmWave band constitutes RF spec-trum with fc between 30-300 GHz The photon energy in these bands ranges from 01 to12 milli-electron volts (meV) Unlike the ultraviolet (UV) X-ray and gamma rays mmWaveradiation is non-ionizing and so cannot cause cancer Therefore the main safety concernis heating of the eyes and skin caused by the absorption of mmWave energy in the humanbody and constitutes the major biological effect that can be caused by the absorption of EMmmWave energy by tissues cells and biological fluid [2 86]

Based on findings from mmWave radiation studies the authors in [86] concluded thatboth the eyes and the skin (whose tissues would receive the most radiation) do not appearprone to damage from exposure experienced from mmWave communication technologies in thefar-field while more studies are required regarding exposure to communication devices (suchas mobile devices with high-gain adaptive and smart antennas) in the near-field Similarlyaccording to the authors of [87] more than 90 of the transmitted EM power is absorbedwithin the epidermis and dermis layers and little power penetrates further into deeper tissuesHowever heating of human tissue may extend deeper than the epidermis and dermis layersAlso the steady state temperature elevations at different body locations may vary even whenthe intensities of EM radiations are the same The authors concluded that power density (PD)is not likely to be as useful as specific absorption rate (SAR) for assessing safety especiallyin the near-field and the proposed temperature-based technique is an acceptable dosimetricquantity for demonstrating safety and setting exposure limits for mmWave radiations [8687]

224 Standardization Activities

The attractive capabilities and high commercial potentials of mmWave communicationshave spurred several international activities aimed at standardizing it for wireless personalarea networks (WPANs) and wireless local area networks (WLANs) These efforts includethe IEEE 802153c WirelessHD and WiGiG (IEEE 80211ad and IEEE 80211ay) [606364]Early efforts focused on the 60 GHz band for WiFi and WiGiG standards due to the earliernotion that mmWave bands are unsuitable for mobile communications [16] However several

31

new findings from research trials and deployment (such as MiWEBA [88 89] and MiWaveS[90 91]) have established the suitability of incorporating mmWave communications into cel-lular networks [65 92] Massive MIMO on the other hand has been under consideration forstandardization by 3GPP since LTE-A Release 12 [16] The release work programme whichwas largely completed in March 2015 considered four areas of significant enhancements andenablers SCs and HetNets multi-antennas (eg massive MIMO and elevation beamforming)proximity services and procedures for supporting diverse traffic types [93] 3GPPrsquos Release14 (completed June 2017) featured major 5G enablers including MIMO enhancements Fur-ther works on massive MIMO standardization featured in 3GPPrsquos 5G New Radio (NR) (alsoreferred to as Release 15 or 5G Phase 1) completed in 2018 and enhancements will continuethrough Release 16 (5G Phase 2) expected to be finalized by December 2019 Standardiza-tion of both mmWave communications and massive MIMO will open up new opportunitiesfor next-generation cellular networks [3 4]

The three major players for 5G standardization are the ITU 3GPP and the Institute ofElectrical and Electronics Engineers (IEEE) Candidate technologies are being evaluated byITU-Radiocommunication through its 5D working party (WP) The allocation of mmWavespectrum for cellular applications is expected to be finalized at WRC 2019 Also ITU-T hasbeen conducting system review and proof-of-concept studies on IMT-2020 by its Study Group13 [3416] As the third major player IEEE has recently undertaken a 5G track to enhanceexisting standards and to evolve new technologies for the mmWave unlicensed bands Theseefforts include IEEE 80211 adajay IEEE 802153c ECMA-387 P19181 P19143 andIEEE 80222 among others The completion timelines for these activities vary among thedifferent specification groups These activities will lead to the first certified 5G standardsThe standardization is expected to be completed by late 2019 ahead of the much-anticipated2020 timeline for commercial deployment [3 4 16]

23 5G Channel Measurement and Modeling

Several new technologies are being explored for 5G systems in order to provide anywhereand anytime connectivity for anyone and anything Each of these technologies introducesnew propagation properties and sets specific requirements on 5G channel modeling For ex-ample the mmWave massive MIMO channel is intrinsically an ultra-broadband channel withhuge bandwidth and spatial multiplexing capability to significantly enhance wireless accessand improve cell and user throughputs [68 94] Inspecting from a propagation perspectivemmWave signals exhibit LOS or near-LOS propagation with absolute increase in PL withincreasing fc [95] specular reflection attenuation [96] diffuse scattering [97 98] very highdiffraction attenuation [99] and frequency dispersion effect which with the prospect of hugebandwidth allows the propagation to be considered as frequency-dependent [21 99] In gen-eral 5G channel models should support wide frequency range (eg 350 MHz-100 GHz)broad bandwidths (10 MHz-4 GHz) wide range of scenarios (indoor urban suburban ru-ral etc) double-directional 3D antenna and propagation modeling frequency dependencysmooth time evolution spatial and frequency consistency large antenna arrays and highmobility among others [100]

Typically channel measurement campaigns are undertaken at different times cities en-vironments and under different scenarios Following the channel measurement activitieschannel parameters (such as PL PLE SF penetration loss power delay profile (PDP) delay

32

spreads (DSs) angular spreads (ASs) coherence bandwidth etc) are estimated from theobtained data or field results to develop channel models Considering all necessary factors(such as frequency propagation environment scenario etc) the developed channel modelsprovide statistical mathematical and analytical frameworks for simulation studies and per-formance evaluation of wireless communication networks Such frameworks are also used tocompare andor validate empirical data from field deployment and operation tests Thus ef-ficient and accurate channel models are central to system design and performance evaluationUsing channel sounding techniques several measurement campaigns have been undertaken bydifferent groups in different cities at different frequencies (10-100 GHz) and for diverse sce-narios and setups to characterize the mmWave channel References [4100] provide excellentsummaries of the results (with respect to the PL PLE SF PDP DS AS Rician K-Factor(KF) coherence bandwidth etc) of the cross-continent channel measurement efforts between2012 and 2018 A summary of the capabilities of the 5G channel models adapted for thisthesis are given in Table 26

Table 26 Comparison of adapted 5G channel models (adapted from [100])

Features 3GPP TR 36873[84]

3GPP TR 38900[85]

NYUSIM[42 101102]

Modeling approach GBSM map-basedhybrid model

GBSM map-basedhybrid model

Statistical

Frequency range (GHz) lt 6 6-100 05-100Bandwidth [lt6 gt 6] GHz 10 of fc 10 of fc [100 MHz 2 GHz]Support large array yes yes yesSupport spherical waves no no noSupport dual mobility no no noSupport 3D propagation yes yes yesSupport mmWave no yes yesDynamic modeling yes yes yesSpatial consistency yes yes yesHigh mobility limited limited limitedBlockage modeling yes yes yesGaseous absorption yes yes yes

Based on the modeling approach adopted the channel models are classified as shown inFigure 26 (shown on next page) The respective models take their name (as used in Tables 26)using a bottom-up naming approach For example RS-GBSM represents a Regular-ShapedGeometry-Based Stochastic Model while TDL-NGSM is a Tapped Delay Line Non-Geometrybased Stochastic Model Deterministic channel models characterize the physical propagationparameters in a deterministic manner by solving the Maxwellrsquos equations or approximatedpropagation equations These models are site-specific and they rely on the detailed or preciseinformation of the propagation environment including the location of the BSs UEs andscatterers Thus they have high accuracy but introduce high computational complexity Anexample of deterministic channel models is ray tracing (RT) where the positions of the TXand RX are specified and then all possible rays are predicted The map-based deterministicmodels use RT and 3D maps The stochastic models on the other hand describe channelparameters using certain statistical or probability distributions The stochastic models aremore analytically tractable and can be adapted to various scenarios and settings However

33

they have lower accuracy when compared to the deterministic models [100] Hybrid models usea combination of the deterministic and stochastic approaches [75] Other representative 5Gchannel models that have resulted from the respective measurement campaigns are comparedin Table 27 (shown on next page) [100]

5G Channel Models

Stochastic Model (SM)

Describes channel parameters using

certain probability distributions




Deterministic Model (DM)

Predicts the propagation waves by

solving the Maxwellrsquos equations




Ray tracing-based (RT)

Rays are determined according

to the theory of

approximations of EM fields



to the theory of


Non-Geometry based (NG)

Scatterers are not geometrically

distributed



distributed

Geometry-based (GB)

Scatterers are geometrically

distributed

Geometry-based (GB)


distributed

Map-based (MB)

Based on RT method using a

simplified 3D map

Map-based (MB)


simplified 3D map

Regular-shaped (RS)

Scatterers are distributed

according to a regular shape

Regular-shaped (RS)



Irregular-shaped (IS)

Scatterers are Irregularly

distributed



distributed

Tapped Delay Line (TDL)

Uses a summation of a series of

complex gains with different

time delays




time delays

Cluster Delay Line (CDL)

Models using the correlation

between paths



between paths

5G Channel Models









to the theory of




distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Regular-shaped (RS)





distributed




time delays



between paths

Figure 26 Classification of 5G channel models based on modeling approach (adapted from[100])

In this section we provide a background to 5G channel modeling involving the interplayof massive MIMO and mmWave communication in cellular and C-I2X scenarios

231 mmWave Massive MIMO Channels

At mmWave frequencies the assumption of asymptotic pair-wise orthogonality betweenchannel vectors under independent and identically distributed (iid) Rayleigh fading channelwhich is valid for conventional massive MIMO no longer holds The number of independentmultipath components (MPCs) becomes limited and so channel vectors exhibit correlated fad-ing [68112] For a mutuallly orthogonal channel every pair of column vectors of the channelH satisfies the condition in (213) With the assumption of iid Rayleigh fading channelthe condition in (213) is asymptotically achieved with very large NTX which by the law oflarge numbers gives (214)

hHmhn = 0 forallm 6= n (213)

1

NTXhHmhn minusrarr 0 NTX minusrarrinfin forallm 6= n (214)

34

Table 27 Comparison of other representative 5G channel models (adapted from [100])

Features 3GPP TR38901

QuaDRiGa mmMAGIC 5GCMSIG MG5GCM

[103] [104105] [106] [107] [108]

Modeling approach GBSMmap-based

hybrid

GBSM GBSMGGBSM

GBSMGGBSM

GBSM

Frequency range(GHz)

05-100 045-100 6-100 05-100 -

Bandwidth [lt 6 gt6] GHz

10 of fc 1 GHz 2 GHz [100 MHz 2GHz]

-

Support large array yes yes yes limited yesSupport sphericalwaves

no yes yes no yes

Support dual mobil-ity

no no no no yes

Support 3D propa-gation

yes yes yes yes yes

Support mmWave yes yes yes yes yesDynamic modeling yes yes yes yes yesSpatial consistency yes yes yes yes noHigh mobility limited yes yes yes yesBlockage modeling yes yes no yes noGaseous absorption yes yes no no no

Features METIS COST2100

MiWEBA IMT-2020

[109] [110] [88] [111]

Modeling approach Stochastic Map-based GBSM Q-D based GBSMmap-based


up to 70 up to 100 lt 6 57-66 05-100


[100 MHz 1GHz]

10 of fc - 216 GHz [100 MHz10 of fc]

Support large array no yes - yes yesSupport sphericalwaves

no yes no yes no


limited yes no yes no


yes yes yes yes yes

Support mmWave partly yes no yes yesDynamic modeling no yes yes limited yesSpatial consistency SF only yes yes yes yesHigh mobility limited no yes no limitedBlockage modeling no yes no yes yesGaseous absorption no yes no yes yes

35

However mmWave massive MIMO channels are neither iid nor is the NTX infiniteTherefore the condition for mutual orthogonality in (213) cannot be satisfied in reality [112]MmWave massive MIMO channel models therefore have to consider this non-orthogonalityfor propagation in realistic environments Similarly the assumption of planar waves in con-ventional massive MIMO would have to be replaced with spherical waves representation asdepicted in Figure 27 Also spatial non-stationarity which becomes more severe has tobe considered in mmWave massive MIMO channel models [95 112 113] Also for practi-cal realization of such channels the corresponding high computational rate required for CSIestimation and feedback in high-mobility channels have to be factored into the design [68]Extensive indoor and outdoor channel measurement campaigns and simulations have beencarried out by [60 94 114] among others to deduce these outcomes Many of these issueshave been factored into the different 5G channel models along their evolutionary trails asshown in Tables 26 and 27 It should be noted however that the channel models employedin this work do not consider spherical waves and dual mobility support as earlier shown inTable 26 The thesis assumes plane waves propagation and considers static BSsAPs butmobile users

UE 3UE 2

eNodeBSpherical

wavefronts

UE 1

Cluster 1

Cluster 2

Cluster 3

Cluster 5 Cluster 4

Figure 27 Illustration of spherical wavefront phenomena for mmWave massive MIMO(adapted from [112])

232 3GPP 3D Channel Models

Over time several channel models have evolved These include the COST series (231259 and 273) the WINNER family (I and II) and the spatial channel models (SCMs) Thesechannel models were 2D models However studies such as [31ndash33] among others have shownthat 2D channel models underestimate system performance 3D channel models give a morerealistic outlook as they consider the elevation (zenithvertical) angles alongside the azimuth

36

(horizontal) angles used by the 2D models Thus 3D channel models are essential for theaccurate and realistic performance evaluation of mobile networks In addition three signif-icant paradigms are shifting interest away from the 2D microWave channel models The firstparadigm is the growing interest in the amazing spectral prospects achievable with mmWavecommunication Elevation beamforming enabled by mmWave massive MIMO and planar an-tenna arrays constitutes the second paradigm Interest in O2I and indoor-to-outdoor (I2O)propagation modeling that account for building blockage and other limiting effects is thethird [8485115] Accordingly newer (3D) channel models which address these interests havecontinued to evolve as earlier shown in Tables 26 and 27

Set

scenario

network layout

antenna parameters

Assign propagation

condition

LOS

NLOS

Calculate pathloss

Generate correlated

LSPs

delay spread

angular spread

shadow fading

K-factor

Generate

cross-polarisation

ratios

Generate

AoAs

AoDs

Perform random

coupling of rays

Generate delays

Generate

cluster powers

Draw random

initial phases

Generate

channel

coefficients

Apply

pathloss and

shadowing

Figure 28 Flow chart for the 3GPP 3D geometry-based SCM (adapted from [328485103])

Most recent channel model efforts are adopting the 3D geometry-based stochastic model(GBSM) approach References [76104105116] and [101117ndash123] provide a good backgroundon mmWave channel modeling using the GBSM approach The three 3D 3GPP channelmodels (ie TR 36873 [84] TR 38900 [85] and TR 38901 [103]) are GBSM models andthey generally follow the flow chart in Figure 28 to estimate channel parameters [3234] The3GPP TR 36873 [84] is defined for the sub-6 GHz band for LTE while 3GPP TR 38900 [85]models the mmWave band from above 6 GHz up to 100 GHz The features of these two modelshave been shown in Table 26 The recent 3GPP TR 38901 [103] generalizes the channel modelfrom 05-100 GHz as shown in Table 27 The 3GPP channel models consider the common usecases urban macrocell (UMa) urban microcell (UMi) or street canyon rural macrocell (RMa)and indoor office scenarios for LOS NLOS and O2I propagation environments The core of themodels (ie the SSF) is identical to the WINNER+ model The models consider effects such

37

as spatial consistency blockage atmospheric attenuation and oxygen absorption at mmWavebands However the models have limited capabilities for dual mobility spherical waves andnon-stationarity for massive MIMO antenna arrays Typically the mmWave models supportlarger bandwidths (up to 10 of fc) [4 100]

The 3GPP 3D channel models parameterize the DS AS and cluster powers as frequency-dependent and support multiple frequencies in the same scenario The number of rays withina cluster are generated within a given range based on the intra-cluster DS and AS as wellas the array size Each MPC may have different delays powers and angles with the offsetangles within a cluster generated randomly [8485100103] The models also proposed a map-based hybrid (stochastic-deterministic) model using a combination of the described stochasticmodel and a deterministic model based on RT and 3D map considering the influences of theenvironmental structures and materials [100] The methodology from these 3GPP channelsmodels have been adopted in our channel implementation and the parameters and values havebeen extensively used in the baseline SLS [38] used for some of the performance evaluationand results in this investigation

233 NYUSIM Channel Model

Following extensive mmWave channel measurements at 28 38 60 and 73 GHz bands a3GPP-like 5G channel simulator known as NYUSIM [42101102] was developed by the NewYork University wireless team The NYUSIM is a 3D statistical spatial channel model (SSCM)and employs the time cluster-spatial lobe modeling approach [101] The time cluster (TC)and spatial lobe (SL) concepts describe the temporal and spatial statistics independently andtheir description somewhat differs from the cluster structure in the 3GPPWINNER modelA TC refers to a number of MPCs or rays with close delays and where different clusters areseparated by an inter-cluster void interval larger than 25 ns The different MPCs forming thecluster can however arrive from different directions The SL on the other hand describes a setof MPCs with close angle of arrival (AoA) or angle of departure (AoD) whose powers spreadin the azimuth and elevation planes but whose rays could possibly arrive over hundreds ofnanoseconds The channel impulse response (CIR) is generated for the respective TCs andcluster subpaths (SPs) [100 101121] The statistical distributions for the TC-SL generationare given in Table 28

Table 28 Distribution Parameters for 3D CIR Generation (adapted from [101])

TC SL Symbol Name of Parameter DistributionNcl Number of Time Clusters Discrete Uniform [1 6]Nsp Number of Sub-paths Discrete Uniform [1 30]τcl Cluster Delays Exponential

TC Pcl Cluster Powers Lognormalρclsp Sub-path Delays ExponentialP clsp Sub-path Powers Lognormalψclsp Sub-path Phases Uniform (0 2π)Nlobe Number of Spatial Lobes (AoD amp AoA) Poissonφlobe Lobe Azimuth Angles (AoD amp AoA) Uniform (0 360)

SL θlobe Lobe Elevation Angles (AoD amp AoA) Gaussianσφ RMS Lobe Azimuth Spread (AoD amp AoA) Gaussianσθ RMS Lobe Elevation Spread (AoD amp AoA) Laplacian

38

The channel simulator (NYUSIM) has support for large arrays mmWave communicationgaseous absorption large bandwidth and 3D propagation [42 101] and through its updateshas a graphical user interface (GUI) supports wideband communication [102] and proposessupport for spatial consistency [115] It however has limited or no support for mobilityblockage and spherical wave modeling The NYUSIM model has a similar representationto the extended Saleh-Valenzuela model commonly employed for mmWave massive MIMOThe extended model is a narrowband clustered channel representation which allows accuratecapturing of the characteristics of mmWave channels Under this clustered model in (215)the discrete-time narrowband channel matrix H is assumed to be a sum of the contributionsof L propagation paths or MPCs [67124]

H =

radicNRXNTX

L

Lsuml=1

αlmiddotandRX(φRXl θRXl

)middotandHTX

(φTXl θTXl

)middotaRX

(φRXl θRXl

)middotaHTX

(φTXl θTXl

)

(215)

where αl is the complex gain of the lth path φTXl θTXl are the azimuth and elevation AoDsrespectively while φRXl θRXl represent the azimuth and elevation AoAs respectively Thevectors aRX

(φRXl θRXl

)and aTX

(φTXl θTXl

)represent the normalized RX and TX array

response vectors at the azimuth (φ) and elevation (θ) angles respectively andRX(φRXl θRXl

)and andTX

(φTXl θTXl

)are the RX and TX gains respectively which can be set to one within

the range of the AoAs and AoDs for simplicity and without loss of generality [125126] Thechannel simulator updated with advanced 5G features was used for some of the performanceevaluation and results in this investigation

24 Beamforming Techniques

The design of beamforming schemes is highly essential for mmWave massive MIMO sys-tems Beamforming optimizes the system performance using the concept of interference can-cellation in advance by controlling the phases andor magnitudes of the original signals Itaims at transmitting pencil-shaped beams that point directly at targeted terminals with min-imal or no interference projected to non-users of interest Beamforming schemes can generallybe classified into three analog beamforming (ABF) digital beamforming (DBF) and hybridbeamforming (HBF) While ABF can only be employed for single-stream single-user MIMOsystems both DBF and HBF schemes can be used for single-user as well as multi-user systems[127] The three beamforming schemes are overviewed in the following subsections Detaileddescription and implementation details are provided in the subsequent chapters

241 Analog Beamforming

This is used to control the phases of signals (using phase shifters (PSs)) with single datastream (using a single RF chain) in order to realize significant antenna array gain and effectiveSNR With perfect knowledge of the CSI available at both the BS and the UE the analogbeamformer employs NTX antennas at the BS with only one RF chain to send a single datastream to a terminal (ie UE) with NRX antennas and only one RF chain too [128] Thisconcept is also called beam steering and aims at the design of the analog precoder (also

39

known as TX RF beamformer) vector fRF and analog combiner (alternatively called RX RFbeamformer or postcoder) vector wRF that maximize the effective channelSNR

∣∣wHRFHfRF

∣∣where H isin CNRXtimesNTX is the channel The optimization problem for ABF is thus formulatedas in (216) where ϕi and ϕl are the AoAs and AoDs respectively [127] The analog TX isshown in Figure 29 and the analog RX can be similarly illustrated Here only one beam iscreated and is particularly useful in LOS propagation scenarios [53](

wopt fopt)

= arg max∣∣wH

RFHfRF

∣∣subject to

wi =

radic1

NRXmiddot ejϕi foralli

fl =

radic1

NTXmiddot ejϕl foralll

(216)

RF

chain

Analog TX beamformer (fRF)

PAPA

PAPA

PAPA

1

2

NTX

PSs

s

RF

chain


PA

PA

PA

1

2

NTX

PSs

s

Figure 29 Analog beamforming architecture

Perfect CSI is unrealistic in practical systems thus necessitating beam training where boththe UE and the BS collaborate in selecting the best beamformer (at the BS end) and combiner(at the user end) pair (|f |minus |w| pair) from pre-defined codebooks in order to optimize systemperformance [127] For mmWave massive MIMO systems the codebook sizes could be verylarge due to a large number of antennas together with the accompanying huge overheads Insolving this challenge a systematic beam training scheme was proposed by Wang et al [129]which reduces the overhead and limits the potentially exhaustive codebooks search using ahierarchical approach Similarly Cordeiro et al in [130] proposed a single-sided beam trainingscheme using a two-step approach which was adopted by the IEEE 80211ad standard wherethe combiner is first fixed to search for the best beamformer exhaustively and subsequentlythe best beamformer is fixed to search for the best combiner exhaustively Though the ABFscheme has simple hardware requirement (only one RF chain) it suffers severe performanceloss as only the phases of transmit signals can be controlled More so an extension of thescheme to the multi-user system is not trivial [127] Therefore its use in mmWave massiveMIMO systems is rather limited as mmWave massive MIMO targets the multi-user case toboost system capacity

40

242 Digital Beamforming

This can control both the phases and amplitudes of transmit signals and it can be em-ployed in both single-user and multi-user MIMO systems For the single-user case the digitalbeamformer (DBF) FDBF isin CNTXtimesNs uses NTX antennas and NTX

RF RF chains at the BS totransmit Ns data streams (where Ns le NTX

RF and NTXRF = NTX) to a user with NRX antennas

and NRXRF RF chains (where NRX

RF ge Ns and NRXRF = NRX) The transmit DBF is shown

in Figure 210 and RX DBF can be similarly illustrated by replacing the TX componentswith the corresponding RX components The DBF can create a number of beams and hasflexibility of adapting the beams to multipath and frequency selecting fading [53128]

PA

PA

1

2

NTX

PAPA

RF chainRF chain

RF chainRF chain

RF chainRF chain

Ns

Baseband

Digital

Beamformer

FDBF

PA

PA

1

2

NTX

PA

RF chain

RF chain

RF chain

Ns

Baseband

Digital

Beamformer

FDBF

Figure 210 Digital beamforming architecture

Examples of linear digital beamformers include the matched filter (MF) zero forcing (ZF)and the Wiener filter (WF) beamformer in increasing order of complexity and performanceThe digital beamformer models are expressed in (217)-(219) respectively [127] where His the NRX times NTX channel matrix with normalized power PRX represents the average re-ceived power and σ2

n denotes the noise power The digital combinerspostcodersbeamformers(WDBF ) can be similarly formulated

FMFDBF = HH (217)

FZFDBF = HH

(HHH

)minus1 (218)

FWFDBF = HH

(HHH +

σ2nNs

PRXINRX

)minus1

(219)

For digital beamforming the optimal TX beamformer and RX beamformer can be real-ized via the singular value decomposition (SVD) of the channel matrix H since H can bedecomposed as H = UΣVH With this decomposition and using (220) the optimal TX

41

beamformer Vopt and RX beamformer Uopt can be set as the first Ns columns of FDBF andWDBF respectively [131]

[WDBF ΣDBF FDBF ] = svd(H) (220)

On the other hand multi-user systems employing digital beamformers simultaneouslytransmit to K mobile terminals where each terminal is equipped with NRX antennas andthe BS is equipped with NTX antennas NTX

RF RF chains and FDBF isin CNTXtimesNsK beamform-ers such that (Ns le NTX) and the kth user has (NTX timesNs) digital beamformer Fk

DBF isinCNTXtimesNs forallk isin 1 2 K with total transmit power constraint

∥∥FkDBF

∥∥2

F= Ns The

received signal by the kth terminal is thus expressed as (221)

yk = Hk

Ksumn=1

FnDBF sn + nk (221)

where sn of size (Nstimes1) is the original signal vector with normalized power before beamform-ing Hk which is of size (NRXtimesNTX) is the channel matrix between the BS and the kth UEand nk is the AWGN vector with entries following iid distribution CN (0 σ2

n) From (221)the terms HkF

nDBFxn for n 6= k are interferences to the kth UE For Block Diagonization (BD)

beamformer design [132] FnDBF is chosen to satisfy the condition HkF

nDBF sn = 0 foralln 6= k

Generally digital beamformers have best performance in terms of SE (as compared toABF and HBF) as they are able to control both the phases and amplitudes of transmitsignals However their use of one dedicated RF chain per antenna can lead to higher energyconsumption and prohibitive hardware cost making them somewhat impractical for mmWavemassive MIMO systems As technology matures and components consume much less powerthan what they consume today DBF may become practically feasible for mmWave massiveMIMO systems [5357131133] There are several non-linear digital beamformers such as theoptimal Dirty Paper Coding (DPC) [134] and the near optimal Tomlinson-Harashima (TH)beamformers [135] which have superior performance than the aforementioned linear digitalbeamformers (eg MF ZF and WF) but they also have higher computational complexity[127]

243 Hybrid Beamforming

This is a promising scheme for mmWave massive MIMO as it offers a significant reduc-tion in the number of required RF chains and the associated energy consumption and cost(when compared with DBF) yet achieving near-optimal performance It realizes this hybridconfiguration by employing a small-size digital beamformer FBB with a small number of RFchains (1 lt NRF lt NTX) to cancel interference in the first stage and a large-size analogbeamformer FRF with a large number of phase shifters (PSs) in the second stage to increasethe antenna array gain [127] This two-stage hybrid beamformer [136] employs the BS analogbeamformer and the UE analog combiner to jointly maximize the desired signal power of eachuser in the first stage and in the second stage employs BS digital beamformer to managemulti-user interference (MUI) The TX HBF is illustrated in Figure 211 and the RX HBFstructure can be similarly illustrated with WRF and WBB as the analog RF and digital base-band combiner respectively as investigated in [125131137138] The HYB configuration is

42

capable of creating multiple beams (limited by NRF ) which can be adapted to the multipathand frequency-selective fading environment [53128]

Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX

Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX

Analog Beamformer FRF

PSsNs

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSs

Figure 211 Hybrid beamforming architecture

For a multi-user system with K users served a BS the RF stage of the HBF architectureessentially involves a coordinated beamforming approach that maximizes the effective channelgain of all users as in (222) [125131137138]

maxFRF WRFk forallk

Ksumk=1

∥∥WHRFkHkFRF

∥∥2

F

subject to

FHRFFRF = INTX WH

RFkWRFk = INRXk forallk isin 1 2 K

(222)

Several approaches in developing FRF and WRF have been proposed over time includingthe popular orthogonal matching pursuit (OMP) [125] and the generalized low rank approx-imation of matrices (GLRAM) [137] among others For the digital baseband stage popularmethods in developing FBB and WBB include the maximum ratio transmissioncombining(MRTMRC) and the zero forcing (ZF) and block diagonalization (BD) techniques [138] Forthis multi-user case the baseband beamformers follow from the operations on the concate-nated effective channel matrix that is given by (223)

Heffk = [HTeff1 H

Teffk H

TeffK ]T (223)

where Heffk = WHk HkFRF FMRT

BB and FZFBB are respectively given by (224) and (225)

respectively [138] WBB can be similarly determined

FMRTBB = HH

effk (224)

FZFBB = HH

effk (HeffkHHeffk )minus1 (225)

43

The received signal vector rk observed by the kth terminal after beamforming can thenbe expressed as (226) and becomes yk after being combined with the RX combiners WRFk

and WBBk as in (227) where n = k is the desired signal and n 6= k are interference termsfor the kth UE

rk = Hk

Ksumn=1

FRFn FBB

n sn + nk (226)

yk = WHBBkWH

RFkHk

Ksumn=1

FRFn FBB

n sn + WHBBkWH

RFknk (227)

In [127] the authors established that hybrid beamformers outperform analog beamform-ers in terms of achievable rate (bpsHz) and approaches the optimal performance of digitalbeamformers particularly as the number of BS antennas increases significantly which is ofbenefit for mmWave massive MIMO systems Thus hybrid beamforming is currently beingextensively explored for mmWave massive MIMO as the digital beamforming counterpart isseen as impractical due to its prohibitive cost and high energy consumption However thisassumption is based on current hardware limitations The development trends show that thecost and power consumption of baseband processing can be reduced in the future when digitalbeamforming will be the mainstream [57133]

Other HBF schemes such as the minimal Euclidean hybrid beamformer [139] mean-squared error (MSE)-based beamforming [140] HBF based on the 1-bit Analog to DigitalConverters (ADCs) [141] and beamspace MIMO [142] have recently been proposed with aview to reducing the energy consumption and hardware cost of beamformers This is in orderto make them realistic and practical for mmWave massive MIMO systems while keeping com-plexity at a modest level A comparison of the three beamforming techniques is summarizedin Table 29

Table 29 Comparison of beamforming techniques

Features Analog Digital HybridNumber of streams Single-stream Multi-stream Multi-streamNumber of users Single-user Multi-user Multi-userSignal control capability Phase control only Phase and

amplitude controlPhase and

amplitude controlHardware requirement Least one RF chain

onlyHighest number of

RF chains equalnumber of transmit

antennas

Intermediatenumber of RF

chains less thannumber of transmit

antennasEnergy consumption Least Highest IntermediateCost Least Highest IntermediatePerformance Least Optimal Near-optimalSuitability for mmWavemassive MIMO

Highly-limited noamplitude control

no multi-user

Impracticalprohibitive cost and

high energyconsumption

Practical andrealistic

44

25 Conclusions

The mmWave massive MIMO UDN technology represents an attempt to harness thepromising benefits realizable with the huge available bandwidth in the mmWave bands theSE improvement that can be obtained with the massive MIMO antenna array systems andthe high capacity gains achievable with UDN In this chapter we have presented an overviewof the concepts and techniques being proposed for mmWave massive MIMO UDNs principallywith respect to 3D channel modeling and beamforming techniques In doing so we highlightedthe requirements of the emerging network technology in contrast to legacy systems and elabo-rated on key use cases and research challenges for 5G networks and beyond In particular thisthesis identified the enhanced mobile broadband connectivity use case (5G UDN and C-I2X)as a vehicle for evaluating the key contributions on mmWave massive MIMO UDN The 2020+experience is expected to represent a balanced networking ecosystem where technical require-ments synchronize well with socio-economic and environmental concerns Therefore a higherlevel of safety improved health lower cost and limited energy and CO2 footprints are amongprincipal drivers for the B5G era alongside the much-anticipated boost in network capacityuser throughput and SE Thus EE is employed as a critical performance metric alongsideSE that will be extensively employed throughout this thesis As we march towards 2020research on mmWave massive MIMO will continue to mature and new trends will emergeNo doubt mmWave massive MIMO UDN demonstates amazing potentials in realizing the1000-fold capacity objective for 5G networks The technology will usher in new paradigms forNGMNs and open up new frontiers for cellular services and applications The backgroundchallenges and open issues on mmWave massive MIMO leading to the compilation of thischapter were the results of the survey that were published in IEEE Communications Surveysand Tutorials1

1S A Busari K M S Huq S Mumtaz L Dai and J Rodriguez ldquoMillimeter-Wave Massive MIMOCommunication for Future Wireless Systems A Surveyrdquo IEEE Communications Surveys and Tutorials vol20 no 2 pp 836-869 May 2018

45

46

Chapter 3

3D Channel Modeling for 5G UDNand C-I2X

This chapter focuses on 3D channel modeling for 5G UDNs and C-I2X networks whichis considered one of the main novelties in this thesis The 3D channel models are pivotalfor realistic characterization and evaluation of mmWave massive MIMO performance Thechapter principally covers three aspects (i) evaluates the individual performance of 3D microWaveand mmWave channels in order to provide insights for coexistence in NGMNs (ii) assesses thejoint performance of the 3D microWave and mmWave channels in the light of 5G UDNs and (iii)investigates the performance of cellular infrastructure-to-vehicle (C-I2V) channels involvingstreet-level lampost-mount APs and vehicles This part compares the mmWave massive MIMOchannel in this propagation environment with the DSRC and LTE-A channels The challengesin these 5G scenarios are then highlighted and analyzed

31 Background

The future networks (starting with 5G) are anticipated to be multi-tier HetNets Forthe first phase of 5G the cost-efficient and practical deployment option being exployed is thecombination of 5G NR with the legacy LTE-A In this architecture the microWave LTE-A tier willprovide coverage and signaling while the 5G mmWave tier provides the anticipated capacityboost [35] There is a consensus in the academia and the telecommunications industry thatadding new frequency bands to existing deployments is a future-proof and cost-efficient wayto improve performance meet the growing needs of mobile broadband subscribers and delivernew 5G-based services More so 5G at mid and high bands is well suited for deployment atexisting site grids especially when combined with low-band LTE [35] In order to evaluatethe performance of such a network architecture accurate characterization of the wirelesschannel is fundamental Consequently in this chapter we first characterize and comparethe individual performance of two 3GPP 3D channel models the LTE channel model forsub-6 GHz [84] and the mmWave channel for frequencies above 6 GHz [85] to provide thebasis for coexistence in 5G UDNs Then in the light of 5G HetNets we investigate the jointperformance of the two channels using the UMa and UMi scenarios for LOS NLOS and O2Ipropagation environments

Similarly applying mmWave spectrum at street-level sites is also a good option to providehigh rate connectivity to users (cellular andor vehicular) By placing antennas on lamposts

47

outer walls and the likes it is possible to avoid typical diffraction losses from rooftops andachieve shorter distances to users located in outdoor hotspots or in targeted buildings Thistopology also offloads traffic from the regular cellular BSs thereby enhancing the capacity ofthe network Along this line there is a recent partnership of the automotive and telecommu-nications industries (ie the 5G Automotive Association) in the area of intelligent transportsystems (ITSs) on the C-V2X paradigm While the 3GPP has standardized C-V2X in Release14 its extension to support the 5G NR is anticipated to be finalized in Release 16 for moreadvanced use cases [143] While the prospect of 5G NR C-V2X is amazing the mmWavechannel exhibits challenging propagation properties markedly different from the sub-6 GHzchannels where both DSRC and LTE-A operate [4 59] The differences become even morepronounced for mmWave vehicular channels due to the impact of high mobility [18130144]In this chapter using the enhanced 3D channel model for C-I2X adapted from the imple-mentation in [42101102] we characterize the mmWave massive MIMO channel between thestreet-level lampost-mount APs and vehicular users

32 microWave and mmWave Channels Individual Performance

Many recent studies have attempted system performance assessment of candidate 5Gmobile networks However most of the studies use simplified channel models (eg 2D LOS-only outdoor-only etc) [145 146] These simplified models do not provide accurate andrealistic characterization of the radio environment In this chapter however we provide acomprehensive analysis using the 3GPP 3D SCMs We adopt the 3GPP TR 36873 channelmodel for LTE [84] and the 3GPP TR 38900 channel model for spectrum above 6 GHz [85] forthe microWave and mmWave channels respectively These models capture diverse channel effectsand cater for LOS NLOS and O2I propagation as well as the building and indoor effectsAs 3D channel models account for a high volume of parameters variables and dependenciesanalyses of the system performance require systematic investigation Thus in this sectionwe characterize the large-scale fading (PL and SF) statistics and use it to characterize thesignal to interference ratio (SIR) SNR and SINR performance Following this we extend theinvestigation to important calibration metrics such as coupling loss (CL) and geometry factor(GF) and study the impacts of UE height and BS downtilt angle on the channel performance

In the following subsections we describe the considered network layout outline the systemparameters for the simulation of the network and present the propagation models employedExcept otherwise stated we follow strictly the 3GPP-compliant baseline for evaluation in [84]and [85] for the microWave and mmWave bands respectively Using CL and GF (SIR SNR andSINR) as metrics we present the simulation results for the individual channel performanceunder four scenarios UMa-microWave UMi-microWave UMa-mmWave and UMi-mmWave

321 System Model

We show in Figure 31 (on next page) the considered system layout It is a square gridwith 57 BSs composed of 19 tri-sectored sites The considered are is further divided intonX times Y smaller squares Users are deployed in the mapped area with a UE per small squareThe four scenarios considered are the UMa and UMi scenarios for both the microWave [84] andthe mmWave bands [85] The simulation parameters for the respective scenario is highlightedin Table 31 (shown on next page)

48

55 timesISD (m)

55

timesIS

D (

m)

BuildingBase

Station

Outdoor

User

Signal

link

Interfering

linkIndoor

User

Figure 31 Cellular deployment layout for channel performance

Table 31 Simulation parameters for channel performance

Parameters UMa-microWave UMi-microWave UMa-mmWave UMi-mmWave

fc (GHz) 2 28PT (dBm) 46 35B (MHz) 10 125hTX (m) 25 10 25 10ISD (m) 500 200 500 200

The simulation parameters employed are based on Tables (61 71-1 82-1 and 82-2) in[84] for the microWave bands and Tables (72-1 73-1 78-1 and 78-2) in [85] for the mmWavebands For all scenarios the BSs have uniform planar arrays (UPAs) with 4times 10 AEs whilethe UEs have uniform linear arrays (ULAs) with 2times1 AEs All elements are spaced 05λ apartalong the horizontal and vertical as the case may be All antenna ports are co-polarized

49

Each element has a gain of 8 dBi with 102o downtilt at the BS and a gain of 0 dBi 9 dB noisefigure (NF) and -174 dBmHz thermal noise power density (No) for the omnidirectional UEsThe scenario-specific parameters are as earlier given in Table 31 where B is the bandwidthand hTX is the TX height

322 Map-based Simulation Framework

The simulation flow follows the framework described in this subsection The inputs arethe BSsrsquo transmit powers (P jT ) and the 2D coordinates of the BSs (xj yj) and the UEs(xk yk) forallj isin 1 2 J and forallk isin 1 2 K The output is the reference signal receivedpower (RSRP) for all users in the mapped area Using the parameters in Tables 31 thePLOS PL and SF are computed using Table 32 (on bottom of page) and Table 33 (on nextpage) based on Tables 72-1 [84] and 741-1 [85] for the microWave and mmWave bands respec-tively The transmit antenna gains (GTX) are calculated based on Tables 71-1 [84] and 73-1[85] for the microWave and mmWave respectively The SF is modeled as a distance-dependentlog-normal distribution N (0 σ) with zero mean and standard deviation (σ) following theClaussen correlated SF map [147] implementation in [38] Claussen in [147] proposed an ef-ficient low-complexity method where each new fading value is generated based only on thecorrelation with a small number of selected neighboring values in the SF map This leads toa significant reduction in computational complexity and memory requirements in contrast toother classical methods

Table 32 LOS probability (adapted from [8485])

Scenario LOS probability (distances and heights are in meters)

UMa-microWave PLOS =

(min

(18

d2D 1

)(1minus exp

(minusd2D

63

))+ exp

(minusd2D

63

))(1 + C (d2DhRX

))

UMi-microWave PLOS =

(min

(18

d2D 1

)(1minus exp

(minusd2D

36

))+ exp

(minusd2D

36

))

UMa-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

63

)(1minus 18

d2D

))(1 + C (d2DhRX

)) 18 lt d2D

UMi-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

36

)(1minus 18

d2D

)) 18 lt d2D

where C (d2D hRX) =

1 d2D le 18

1 + 125C prime (hRX)(d2D100

)3exp

(minusd2D150

) 18 lt d2D

C prime (hRX) =

0 hRX le 13(hRX minus 13

10

)15

13 lt hRX le 23

and d2D (is the outdoor separation distance hereafter referred to as d2Dminusout) [84] [85]

50

Tab

le3

3P

ath

loss

mod

els

(ad

apte

dfr

om[8

485]

)

Scen

ari

oP

ath

loss

(dB

)σSF

(dB

)

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

5000

4

UM

a-

microW

ave

PLNLOS

=69

51+

390

9lo

g10(d

3D

)+

20lo

g10(fc)

+75

log

10(hRX

)minus

06hRXminus

00

825(hRX

)26

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

3

UM

i-micro

Wav

ePLNLOS

=23

15+

367

log

10(d

3D

)+

26lo

g10(fc)minus

03hUE

4

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

32

4+

20

log

10(d

3D

)+

20lo

g10(fc)

d10ltd

2DltdBP

32

4+

40

log

10(d

3D

)+

20lo

g10(fc)minus

10lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

500

04

UM

a-

mm

Wav

ePLNLOS

=14

44+

390

8lo

g10(d

3D

)+

20lo

g10(fc)minus

06h

RX

6

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

PLLOS

=

32

4+

21

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

324

+40

log

10(d

3D

)+

20lo

g10(fc)minus

95

log

10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

4

UM

i-m

mW

ave

PLNLOS

=22

85+

353

log

10(d

3D

)+

213

log

10(fc)minus

03hUE

78

2

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

NO

TE

S

f cis

inG

Hzd

2Dd

3D

andhRX

are

inm

eter

sdBP

=32

0((hRXminus

1)f c

)fo

rU

Ma

anddBP

=120

((hRXminus

1)fc)

for

UM

idBP

isth

eb

reak

-poi

nt

dis

tance

[84]

[8

5]an

dPLwall

isth

ew

all

pen

etra

tion

loss

base

don

Tab

les

74

3-(1

amp2)

in[8

5]

51

For each scenario a UE may be in LOS or NLOS in either O2O or O2I environment toeach BS Consequently the various UE maps (indoor distance (d2Dminusin)) outdoor distance(d2Dminusout) UE height (hRX) LOS probability (PLOS) etc) are calculated as follows Notethat j and k subscripts represent BS and UE respectively Also X and Y represented thelength and breadth of the mapped area respectively (scaled according to a 101 map resolutionfor complexity reduction) where binornd randi and rand denote the binomial random num-ber uniformly distributed pseudorandom integer and uniformly distributed random numberoperators respectively

(i) Generate UE indoor-outdoor mapThis is generated based on the indoor (rin) - outdoor (rout) ratio According to the3GPP baseline in [84] and [85] rin = 08 and rout = 02 representing 80 indoor and20 outdoor users

χ = binornd(1 rin X Y ) (31)

χk =

0 k rarr outdoor

1 k rarr indoor(32)

(ii) Generate building floor mapThe number of floors the indoor users are distributed into is computed as follows

nmapfl = randi([nmin

fl nmaxfl ]) (33)

nfl = randi(nmapfl X Y ) (34)

where nminfl = 4 and nmin

fl = 8 [84 85] The actual maximum number of floors used for

the height distribution of the UEs is nmapfl where the floor number of each user k in the

X-Y mapped area falls between 1 and nkfl forallk isin 1 2 K

(iii) Generate indoor distance mapThe indoor distance of the users are computed as

d2Dminusin = 25times rand(XY ) (35)

Using (31)-(35) and Table 31 as inputs the computation of RSRP for the users fol-lows the framework in Algorithm 1 Users are attached to the respective BSs based on themaximum RSRP criterion The values of the variable parameters depend on the state of thesimulation with respect to the scenario and user location in each run andor transmissiontime interval (TTI) For each scenario 1000 simulation runs are performed and averagedThe empirical cumulative distribution function (ECDF) is used to analyze the results

52

Algorithm 1 Map-based simulation framework

Inputs P jT hj forallj isin 1 2 middot middot middot J larr Table 31χ nfl and d2Dminusin larr (31)-(35)

OutputRSRPk forallk isin 1 2 middot middot middot Kfor k rarr 1 to K do

I- Compute UE height map

hk = 15 + (χk times [3(nkfl minus 1)]) (36)

for j rarr 1 to J doII- Compute 2D amp 3D distances to all BSs

dkj2D =radic

(xj minus xk)2 + (yj minus yk)2 (37)

dkj3D =

radic(dkj2D)2 + (hj minus hk)2 (38)

dkj2Dminusout = dkj2D minus (χk times dk2Dminusin) (39)

III- Compute PLOS using Table 32

ξkj = binornd(1 P kjLOS(dkj2Dminusout)) (310)

ξkj =

0 k j rarr NLOS

1 k j rarr LOS(311)

IV- Compute PLkj using Table 33

χk ξkj =

0 0 rarr PLNLOS

0 1 rarr PLLOS

1 0 rarr PLO2IminusNLOS

1 1 rarr PLO2IminusLOS

(312)

V- Compute SFkj using Table 33

VI- Compute GkjTX Gkj

RX using Table 34 (shown on next page) [8485]VII- Compute RSRPkj

RSRPkj = P kjT +Gkj

TX +GkjRX minus PLkj minus SFkj (313)

RSRPk = max(RSRPk(1J)) (314)

Assign user k rarr jth BS with maxRSRPk

53

Table 34 Antenna radiation pattern (adapted from [8485])

Parameter Values

Antenna element horizontal radia-tion pattern (dB)

AEH = minusmin

12

(φ

65o 30 dB

)Antenna element vertical radiationpattern (dB)

AEV = minusmin

12

(θ minus 90o

65o 30 dB

)Combining method for 3D antennaelement pattern (dB)

G(φθ) = minusmin minus [AEH(φ) +AEV (θ)] 30 dB

Maximum directional gain of an an-tenna element (GAEmax)

8 dBi

323 Simulation Results

In this section we present the simulation results for the four scenarios considered usingthe CL and GF (using the SINR SNR and SIR) as performance metrics

(a) Coupling Loss

The difference between the received signal and the transmitted signal along the LOS di-rection is the CL [31] For each user and its attached BS the CL (dB) and the RSRP (dB)are defined as (315) and (316) respectively

CLk = RSRPk minus P kjT (315)

RSRPkj = P kjT +Gkj


CL depends only on the slow fading (ie large-scale) parameters It captures all attenua-tion sources between a UE and its attached BS As can be seen from substituting (316) into(315) CL is independent of PT [148] It is used in the Phase 1 calibration by standardizationbodies such as 3GPP to bring companies and organizations involved in channel measurementsand modeling to a common ground with respect to reported channel results [33] In Figure32 we show results for CL (ie the LOS curves) for the four scenarios

In Figure 32 the minimum CL is sim45 dB for both microWave-UMa and microWave-UMi andsim35 dB and sim75 dB for mmWave-UMa and mmWave-UMi respectively The results inFigures 32(a) and 32(b) are consistent with [84] and Figure 32(d) is consistent with thecorresponding case in [148] Further in Figures 32(a)-(d) we provide loss curves for the NLOSUEs and the overall loss curves involving all UEs (both LOS and NLOS UEs combineddenoted on Figures 32(a)-(d) as LOS+NLOS) A penalty of around 20-35 dB is observedbetween the LOS and NLOS cases for the microWave band and around 20-50 dB for the mmWavecase

54

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF

(a) UMa - Wave (2 GHz)

LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF

(b) UMi - Wave (2 GHz)

LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF

(c) UMa - mmWave (28 GHz)

LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF

(d) UMi - mmWave (28 GHz)

LOS+NLOSLOSNLOS

Figure 32 ECDF of coupling loss

(b) Geometry Factor

The GF captures the statistics of the SINR [148] or SIR [31] or either of the two [85]These are necessary in order to compute the SE UE throughput and cell capacity TheSINR SIR and SNR for a given UE are defined as (317) (318) and (319) respectively GFmeasures the performance of users with respect to the received signal strength relative to theinterference from other BSs (as SINR (with noise) or SIR (without noise)) It is also usedin Phase 1 calibration to assess performance as a measure of UEsrsquo SE and throughput [31]The SNR performance on the other hand characterizes the maximum achievable capacity ininterference-free scenarios [4]

55

SINRk = RSRPkj minus (Jsum

j=1 k 6rarrjRSRPkj +Nk) (317)

SIRk = RSRPkj minus (Jsum

j=1 k 6rarrjRSRPkj) (318)

SNRk = RSRPkj minusNk (319)

Nk = No + 10 log10Bk +NF (320)

The SNRSIRSINR performance for the four scenarios are shown in Figure 33 In allcases the SINR curves expectedly characterize the worst-case performance as they incorporateboth the noise and interference terms The SIR curves in Figures 33(a) and 33(b) overlapthe SINR curves for the microWave UMa and microWave UMi scenarios respectively This outcomeshows that microWave network is interference-limited As for the mmWave scenarios it is theSNR curves that overlap the SINR curves as shown in Figures 33(c) and 33(d) for mmWaveUMa and mmWave UMi cases respectively It reveals that the mmWave network is noise-limited for the considered scenario However for ultra-dense mmWave network with muchshorter ISDs the mmWave network could be interference-limited also due to transition fromNLOS to LOS interference [149]

For all scenarios the GF follows a similar trend consistent with the results in [31] and [84]for the microWave scenarios and [148] for the mmWave bands In comparing the SNRSIRSINRperformance for the four scenarios we use the 0 dB point which translates to an SE of 1bpsHz The point where the ECDF curve first crosses the 0 dB point gives the percentageof users that will achieve a SE of 1 bpsHz or less For SNR 45 13 689 725 of usersachieve up to 0 dB (or SE of 1 bpsHz) for the UMa-microWave UMi-microWave UMa-mmWave andUMi-mmWave scenarios respectively Therefore while more than 95 of microWave users achieveSE greater than 1 bpsHz in the microWave networks only sim30 of mmWave users will achievethe same feat As for SIR 71 88 64 76 of users achieve up to 0 dB while in termsof SINR 120 89 684 746 of users achieve up to 0 dB for the same order in scenarioAgain mmWave systems perform poorly with only sim30 of users achieving the 1 bpsHzSINR compared to sim90 in the microWave set-ups This SINR bottleneck is of great concernfor mmWave networks expected to provide the much-anticipated multi-Gbps throughputs in5GB5G networks The results shown here are with 125 MHz mmWave bandwidth basedon [145] as 100 MHz is projected as the practical size for a component carrier in mmWavesystems [3] SINR degradation will therefore be of more significant consequences if muchlarger mmWave bandwidths (up to 1-2 GHz) are used due to the expected increase in noisewith increasing bandwidth as can be seen from (320)

56

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1E

CD

F(a) UMa - Wave (2 GHz)

SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF

(d) UMi - mmWave (28 GHz)SNRSIRSINR

Figure 33 ECDF of geometry factor

The individual performance of the microWave and mmWave channels for UMa and UMiscenarios have been shown in Figures 32 and 33 However 5G HetNets is anticipated tofeature UMa-microWave (LTE) and UMi-mmWave (5G) channels as the feasible and cost-effectiveoption for increasing cell capacities in early 5G deployments [35] The UMa-microWave is expectedto provide coverage Low data rates from such cells are acceptable and the research on them ismore established On the other hand the UMi-mmWave tier is foreseen to provide the much-anticipated capacity to meet the explosive data rate demands projected for the 5GB5G eraThe results from Figure 33(d) however show low performance In the following subsectionswe investigate further the factors responsible for low SINR in 3D UMi mmWave channels

(c) Impact of UE height and BS downtilt

In Figure 34 we show the effect of wallpenetration losses and indoor losses on the SINRperformance The two extremes are the cases when all UEs are indoors and when all UEs areoutdoors For the other cases 20 of UEs are outdoors while the remaining 80 indoor usersare randomly distributed according to the maximum number of floors For example the curvenamed ldquo3 floorsrdquo means that the 80 indoor users are randomly distributed in floors 1-3 thecurve named ldquo7 floorsrdquo means that the 80 indoor users are randomly distributed in floors1-7 and so on A significant 20-30 dB gap exists between when all the UEs are outdoors andwhen they are all indoors Further in the 3GPPrsquos case with 20 of UEs outdoors and 80indoors the distribution of the indoor UEs across floors (ie the effect of UE heights) does

57

not amount to significant impact on average performance The differences are paled by thehigh number of UEs at system-level Further we show in Figure 35 the results of simulationsfor the two extreme cases only (ie all UEs indoors and all UEs outdoors) with different BSdowntilt angles

-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF

1 floor2 floors3 floors4 floors5 floors6 floors7 floors8 floorsall indoorsall outdoors

Figure 34 Impact of UE height (floor level) on SINR

-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF

Downtilt 0o UEs indoors



Downtilt 0o UEs outdoors



Figure 35 Impact of BS downtilt angle on SINR

58

The results show that the BS downtilt angle does not significantly impact average perfor-mance The observable gap between when all UEs are outdoors and when all UEs are indoorsis attributable again to the wall and indoor losses

33 Joint Channel Performance for 5G UDN

In this section we present results for the system-level performance for the joint microWave-mmWave UDNs featuring both outdoor and indoor users For the evaluation we employ the3GPP 3D channel model [84] and [85] for the microWave and mmWave bands respectively Firstwe describe the considered network layout outline the system parameters for the simulation ofthe network and then present the system-level performance of the coexisting microWave-mmWaveUDN in terms of SE UE throughput and cell capacity The challenges and solutions for the5G eMBB setups are also highlighted

331 Deployment Layout

As shown in Figure 36 the considered two-tier UDN consists of the microWave massiveMIMO-based MCs and the mmWave massive MIMO-based SCs densely deployed in eachMC Buildings with four to eight floors are randomly situated within the coverage area Alsothe UEs are randomly and uniformly distributed in the cells with a probability following theindoor-to-outdoor ratio Outdoor UEs are at a height of 15 m Each indoor UE is at arandom height between 15 and 225 m evaluated using hUE = 15 + 3(nfl minus 1) where nfl isthe floor number of the user [8485]

Figure 36 5G UDN deployment layout

59

The simulation parameters are presented in Table 35 The simulations consider a multi-user DL scenario and employ the MIMO-orthogonal frequency division multiplexing (OFDM)PHY numerology for LTE [38] and mmWave [145] cells For the channel model the MC tierfollows the 3GPP UMa scenario of TR 36873 [84] while the SC tier follows the 3GPP UMiscenario of TR 38900 [85] For the most part the simulation parameters are in line withthe 3GPP evaluation guidelines in [84] and [85] for the microWave and mmWave frequenciesrespectively

Further to the simulation parameters in Table 35 the simulations employ the closed loopspatial multiplexing (CLSM) transmit mode and frequency division duplexing (FDD) Alsohomogeneous power allocation (ie equal power per resource block (RB)) was employed Forresource allocation the proportional fair (PF) scheduler is adopted as it strikes a balancebetween maximizing system throughput and maintaining fairness among users For a timeslot t the PF algorithm schedules (ie allocates RB(s)) to user klowast with the largest Rk(t)

Tk(t)

among all active users in the system where Rk(t) is the user requested rate and Tk(t) is theaverage user throughput updated at specified time window [38]

Table 35 Simulation parameters for system performance

Network ele-ments

Parameters Macrocell(MC)

Small cell (SC)

Number of cells 3 9 (3 SCsMCsector)

Sector layout Tri-sector RadialInter-site Distance [ISD] (m) 500 200Height [hTX ] (m) 25 10

BS Planar array elements (vertical x horizontal) 10times 4 10times 4Carrier frequency [fc] (GHz) 2 28Bandwidth (MHz) 20 125 250 500

1000Transmit Power [PT ] (dBm) 49 35Minimum Coupling Loss [MCL] (dB) 70 45Number of UEs per MC sector and SC 20 20Outdoor height [hRX ] (m) 15 15Indoor height [hRX ] (m) 15-225 15-225

UE Linear array elements (vertical x horizontal) 1times 4 1times 4UEsrsquo speed (kmph) 3 3Noise Figure [NF] (dB) 9 9Noise power spectral density[No] (dBmHz) -174 -174

3D Channel Pathloss Shadow fading and fast fading 3GPP TR 36873[84]

3GPP TR 38900[85]

Cell capacity is employed as the main performance metric for the network The capacity(C) of each cell is evaluated as C = B middot log2 (1 + SINR) where SINR is evaluated using (321)

SINR (dB) = (PT +GTX +GRX minus PLminus SF )minus

[(Nintsumn=1

In

)+ (No + 10 log10B +NF )

]

(321)

where In is the interference from non-target links operating on same frequency band The PLincludes the penetrationwall and indoor losses while the TX and RX antenna element gains

60

are 8 dBi and 0 dBi respectively The SE expressed as CB (bpsHz) suggests that theabundant bandwidth results in significant capacity gains only when sufficient SINR is alreadyguaranteed

Table 36 compares and contrasts the propagation characteristics of the MC with the SCtier following [84] and [85] respectively The comparison is done using (321) The presentedvalues are obtained by plugging the respective ISD values in the 3GPP models [84] and [85]for the MC and SC respectively However the actual values for each user will vary dependingon the user location Overall as illustrated in Table 36 the SINR in the mmWave SC appearssmaller than that in the microWave MC tier (when compared at the same ISD) due to the higherlosses and noise level at mmWave If all other parameters are kept constant the SINR (andby extension the SE) continues to decrease with increasing amount of mmWave bandwidthemployed However the capacity continues to grow but not at the same rate as the increasingbandwidth

Table 36 Comparison of microWave MC and mmWave SC

Properties Macrocell (ISD = 500 m fc =2 GHz) [84]

Small cell (ISD = 200 m fc =28 GHz) [85]

PT 49 dBm 35 dBmSignal LOSmax = 934 dB LOSmax = 1033 dB

PLNLOSmax = 1368 dB NLOSmax = 1238 dB

O2Imax = 1693 dB (includingpenetration and indoor loss)

O2Imax = 1542 dB (includingindoor and penetration losses using

low-loss model (ie glass andconcrete walls))

SF Zero mean log-normal distributionwith std of 4 6 and 7 for LOS

NLOS and O2I respectively

Zero mean log-normal distributionwith std of 31 78 and 9 for LOS


InterferencesumNint

n=1 In Less number of interferers buthigher interference due to directed

interference and longer range

More interferers due to highersmall cell density but with less

interference due to strongattenuation resulting from higher

path lossesNo -174 dBmHz -174 dBmHz

Noise NF 9 dB 9 dB10 log10B 73 dB at B = 20 MHz Noise here

is lower than in mmWave SCs withlarger bandwidths

B = 125 250 500 1000 MHzNoise increases with increasing

bandwidth

332 Simulation Results and Analyses

We present the simulation results and analyze the performance of the network The resultsare from simulations using the parameters in Table 35 It is instructive to note that all resultsare given per MC sector

(a) Average Cell Capacity

The average MC capacity (ACMC) average SC capacity (ACSC) and average cell capacity(ACcell) when all the UEs are outdoor and when 80 of the UEs are indoor (according tothe 3GPP in [84] and [85]) are shown in Figures 37 and 38 respectively In both cases

61

ACcell = ACMC + (3 times ACSC) This is the direct result of deploying three SCs within eachMC sector In Figure 37 the ACMC from the 20 MHz LTE bandwidth is sim40 Mbps TheACSC on the other hand increased from roughly 120 Mbps to 500 Mbps by moving from125 MHz to 1 GHz mmWave bandwidth respectively Correspondingly the ACcell increasedfrom 400 Mbps to 153 Gbps It can be observed that the ACcell increased dramatically withUDN (40times for the 1 GHz bandwidth case) compared to the 40 Mbps realizable if it were aMC-only set-up However the ACSC does not increase linearly with increasing bandwidthFor example doubling the SC bandwidth (eg from 500 MHz to 1 GHz) only brought about446 increase in ACSC (ie 3433 to 4964 Mbps)

Small cell bandwidth (MHz)

0

200

400

600

800

Cel

l cap

acit

y (M

bp

s)

1000

1200

1400

125

1600

250500

1000

average macro

average small

total

Figure 37 Impact of bandwidth on cell capacity with all users outdoors

The results for the 3GPP case (with 80 of the UEs indoors) are shown in Figure 38Compared to the outdoor scenario the performance in this case is highly degraded Usingthe 1 GHz case for example the ACMC ACSC and ACcell are 373 4159 and 12851 Mbpsrespectively Despite the 1 GHz bandwidth the performance is even much lower than the125 MHz case for the outdoor scenario As highlighted earlier the aggregate PL in O2Ienvironment is higher than in outdoor cases This is due to the additional wall and indoorlosses experienced by indoor users These losses further reduce the received signal strengththereby degrading the SINR and the cell capacity Our simulations show that the networkperformance decreases as the percentage of indoor users increases from outdoor-only (0) toindoor-only (100)

62


0

20

40

60

80

Cel

l cap

acit

y (M

bp

s) 100

120

140

125250

5001000

average macroaverage smalltotal

Figure 38 Impact of bandwidth on cell capacity with 80 of the UEs indoors

(b) Average UE Throughput

The average UE throughputs for the SCs in indoor 3GPP and outdoor environments arepresented in Figure 39 In each case the UE throughput increased with increasing mmWavebandwidth Again in all cases the throughput increase is not proportional to the increasein bandwidth For the same reasons explained earlier there is significant degradation inperformance for the fully indoor and the 3GPP scenarios as compared to the fully outdoorenvironment In Figure 39 the three scenarios correspond to the 100 80 and 0 of in-door UEs respectively For the 1 GHz mmWave bandwidth for example the average UEthroughput for the indoor 3GPP and outdoor scenarios are 396 104 and 12409 Mbpsrespectively

(c) Spectral Efficiency

For the SC tier the average SE for the three environments considered are shown in Figure310 In each environment the SE decreased with increasing bandwidth moving from 025046 148 bpsHz at 125 MHz to 007 024 103 bpsHz at 1 GHz for the indoor 3GPPand outdoor environment respectively With all parameters remaining constant in (321)increasing the bandwidth expectedly leads to a reduction in SINR and SE The emphasisis on the SCs due to the investigation of the impact of increasing bandwidth on networkperformance For the MCs the 20 MHz LTE bandwidth was employed for all scenarios andso not considered in the analysis

63

Small cell bandwidth (MHz) of indoor U

Es

0

20

40

60

0

80

Ave

rag

e U

E T

hro

ug

hp

ut

(Mb

ps)

125

100

120

140

250 80500

1001000

Figure 39 Impact of bandwidth on SC user throughput

of indoor U

EsSmall cell bandwidth (MHz)

0

05

Sp

ectr

al e

ffic

ien

cy (

bp

sH

z)

125

1

15

0250

80500

1001000

Figure 310 Impact of bandwidth on SC spectral efficiency

64

333 Challenges and Proposed Solutions

The SINR is degraded in mmWave systems due to the higher PL increased noise andother additional losses such as wallpenetration and indoor losses (for indoor users) and theabsorption losses (for the 57-63 GHz bands) [5985150] In scenarios with very low receivedsignals it is not unusual for the SCs to have poorer performance than the MCs (or evencomplete outage) despite the much larger bandwidth in the SCs This situation in additionto the high susceptibility to blockage makes higher frequency (ie mmWave and THz) linkshighly opportunistic and potentially unreliable despite the amazing spectral prospects [4]Two possible options to overcome the SINR bottleneck in mmWave systems are signal powerenhancement and interference mitigation

(a) Signal Power Enhancement

Increasing the transmit powers of the mmWave SCs can enhance the received signalstrength In Figure 311 we show that the capacity of mmWave SCs increases with increas-ing transmit power (in steps from 30 to 49 dBm) The capacity increase with the increasingtransmit power in outdoor scenario is markedly significant On the contrary the performancewith both the 80 and 100 indoor UEs are poor For the 49 dBm case for example theACSC is only 200 Mbps compared to almost 14 Gbps in the outdoor scenario

30 32 34 36 38 40 42 44 46 48

Small cell transmit power (dBm)

0

200

400

600

800

1000

1200

1400

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)

All UEs outdoor80 of UEs indoorAll UEs indoor

f = 28 GHz B = 1 GHz

Figure 311 Impact of transmit power on cell performance

65

However the SC transmit powers cannot practically be increased indiscriminately due tothe following constraints among others (i) transmit powers are limited by regulation (ii)excessive power will increase the interference level as well in addition to enhancing the signaland (iii) more transmit power will increase the energy consumption of the network which isundesirable for 5G networks Therefore the optimal transmit power is a critical design goal

(b) Interference Mitigation

Beamforming to target UEs mitigates interference from other BSs This will also enhancethe signal level through its transmit and receive beamforming gains However this requirespencil-like beams whose performance heavily depends on beam steering beam tracking andbeam alignment efficiencies alongside other complexity challenges In Figure 312 we showthe performance of the SCs with directional (8 dBi gain 65o half power beamwidth (HPBW))and omnidirectional antenna (0 dBi) for both full buffer and mixed traffic scenarios Theimplemented mixed traffic is a multi-class traffic type composed of the hypertext transferprotocol (HTTP) file transfer protocol (FTP) gaming voice over internet protocol (VoIP)and video The distribution of each of the traffic type is given in Table 37 following the3GPP traffic model evaluation methodology [48] and represents a more realistic user patternthan the full buffer which assumes that users have infinite buffers and are active all the time

Table 37 Multi-class traffic model (adapted from [48])

Application Traffic Category Percentage of Users ()FTP Best effort 10HTTP Interactive 20Video Streaming 20VoIP Real-time 30Gaming Interactive real-time 20

With respect to directivity the results are shown in Figure 312 The network perfor-mance with directional antenna is poorer than that with omnidirectional antenna despite thehigher antenna gain in the directional case This is due to the fact that we employed staticbeamforming [32] where only a part of the users (ie those within the coverage zone of thebeams) is served In radial SCs with users randomly in all directions dynamic beamformingwhere the locations of the users are continuously scanned and tracked is the solution Insuch a case a better performance can be achieved with narrower beams and higher gainsAlthough this would result in a much higher complexity due to beam tracking requirements

As for the traffic type the performance with the mixed traffic is slightly better than thefull buffer scenario in each case as shown in Figure 312 Users in the mixed traffic scenariodo not request data all the time as they have finite buffers In such inactive time instantsthe interference level in the network is reduced This accounts for the increased cell capacitywhen averaged over the entire transmission period The results when fc increases from 28GHz to 01 THz is shown in Figure 313

66

0 20 40 60 80 100

Percentage of Indoor UEs ()

0

100

200

300

400

500

600

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)

Omnidirectional full bufferDirectional full bufferOmnidirectional mixed trafficDirectional mixed traffic

f = 28 GHz B = 1 GHz PTX

= 35 dBm

Figure 312 Impacts of antenna directivity and traffic type on cell performance

125 250 375 500 625 750 875 1000


0

50

100

150

200

250

300

350

400

450

500

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)

28 GHz - 80 indoor UEs01 THz - 80 indoor UEs28 GHz - All UEs outdoor01 THz - All UEs outdoor

PTX

= 35 dBm

Figure 313 Impact of carrier frequency on cell performance

67

In Figure 313 the performance of the SC network gets poorer as higher frequencies areemployed This trend is due to the increasing PL with increasing fc At 01 THz (100 GHz)the ACSC for the outdoor scenario is sim130 Mbps using 1 GHz bandwidth compared to sim500Mbps at 28 GHz In the 80 indoor UEs case the performance is much worse The ACSC issim20 Mbps at 01 THz compared to sim45 Mbps at 28 GHz

It is evident from the results in Figures 311-313 that appropriate transmit power coupledwith adequate beamforming gains will overcome the SINR bottleneck This will consequentlyboost the performance of indoor users served from outdoor mmWave SC networks and furtherboost the performance of outdoor users In addition as the mmWave frequency increases upto the THz bands the ISD would have to be correspondingly decreased As of now onlya coverage range (or ISD) up to 10 m can be seen as realistic for THz band application incellular systems as compared to the maximum of 200 m for mmWave systems [151]

34 C-I2V Channel Performance

Autonomous driving is an innovative and revolutionary paradigm for future intelligenttransport systems (ITS) To be fully-functional and efficient vehicles will use hundreds ofsensors and generate terabytes of data that will be used and shared for safety infotainment andallied services Vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communicationthus requires data rate latency and reliability far beyond what the legacy DSRC and LTE-Asystems can support This has motivated the use of mmWave massive MIMO to facilitategigabits-per-second (Gbps) communication for C-V2X scenarios Focusing on C-I2V thissection characterizes the mmWave massive MIMO vehicular channel using metrics such asPL root-mean-square (RMS) DS KF cluster and ray distribution PDP channel rank andcondition number (CN) as well as data rate We then compare the mmWave performancewith the DSRC and LTE-A capabilities and offer useful insights on vehicular channels

Currently DSRC (known as ITS-G5 in Europe) is the legacy system for vehicular commu-nication and safety It operates on the 59 GHz band using transceivers based on the IEEE80211p standard [152] Unfortunately DSRC can only support data rates up to 27 Mbpswith typical average of 2-6 Mbps This is grossly inadequate for next-generation applicationsforeseen to require multi-Gbps rates Similarly legacy 4G systems can only support a maxi-mum data rate of 100 Mbps in high mobility (vehicular) scenarios The same story goes for thebandwidth reliability and latency requirements Consequently the industrial and academicresearch communities have identified the mmWave bands to come to the rescue [18152153]Fortunately the mmWave bands are being extensively explored for 5G services and applica-tions due to their amazing spectral opportunities The mmWave band is expected to supportthe high rate vehicular applications to facilitate future emerging use cases in infrastructure-assisted and autonomous driving and enable high-rate infotainment and ultra-reliable safetyservices This is in addition to the anticipated benefits of enhanced vehicular safety bettertraffic management more efficient toll collection and commute time reduction among othersThe mmWave spectrum is already being used in standardized systems such as IEEE 80211ad(60 GHz) and radar (76 GHz) [18]

ITS-supported vehicles will be equipped with tens to hundreds of sensors (eg LIDARultrasonic radar camera etc) which together with the on-board communication chipsetswill enable diverse services such as live map download videos streaming environmental datagathering and broadcast for different vehicular applications like sensor data sharing cloud pro-

68

cessing cooperative perception platooning intenttrajectory sharing real-time local updatespath planning collision avoidance blind-spot removal and general warnings [18152153] ForV2I links the infrastructure can gather sensing data (about the vehicles or the surroundingtraffic) from the vehicles The sensed data can be processed in the cloud and used to providelive images or real-time maps of the environment These maps can be used by the transporta-tion control system for congestion avoidance general warnings (such as dangerous situations)and overall traffic efficiency improvement Also automakers can use the sensed data for faultor potential failure diagnosis of the vehicles The infrastructure can also be used to providehigh rate internet access to the vehicles for automated driving and infotainment services suchas media download and video streaming [18 143] Similarly the potential V2V applicationsinclude cooperative perception where perceptual data from neighbouring vehicles can be usedto create a satellite view of the surrounding traffic The view can be used to extend theperception range of each vehicle in order to reveal hidden objects cover blind spots and avoida collision with other vehicles Shared data among the vehicles can also be used for otherapplications such as path planning and trajectory sharing among others [18152]

Many authors have attempted to address different aspects of the challenges Majority ofthe works centre on the V2V and V2I scenarios in urban street [154155] highway [152153]and high speed rail (HSR) [156] environments The works on I2V consider the infrastructuresas BSs or SCs with typical range 200-500 m which translate to sub-6 GHz and mmWavechannels with many clustered blockers and scatterers and where models such as [102 103]can be readily adopted However [102] does not consider mobility while [103] supports onlypedestrian mobility and is reported to have excessive number of clusters and SPs that isunsupported by measurement [157] This section however motivates the use of mmWavemassive MIMO lampost-mount APs spaced at very short intervals typically 20 m for denseroad side unit (RSU) deployment [18] We then characterize the mmWave vehicular channeland compare its performance with the DSRC and LTE-A (sub-6 GHz) vehicular channels toprovide insights for 5G NR C-I2V

341 Network Deployment

In this section we present the network layout antenna and channel models as well as theprecoding technique employed We show in Figure 314 the deployment layout for the C-I2Vscenario We consider a dr = 500 m-long section of the road in an UMi street environmentStationary APs are mounted a height hTX = 5 m on street lampost with a density of ΩTX

= 50 BSkm This corresponds to 25 evenly-spaced APs for the considered distance Thevehicles traverse the route at a speed of vRX = 36 kmh = 10 ms and have roof-mountantennas with height hRX = 15 m above the reference ground level Further we assumethat the APs are connected by high rate backhaul links The DL connectivity is by LOSwith the roof-top positioning of vehicle antenna Therefore the relatively high position ofthe APs (compared to a V2V scenario) ensures good link [158] Also the 3D separationdistance (d3D) between the vehicle (RX) and its serving AP (TX) at each time instant ofthe considered scenario gives a LOS probability PLOS asymp 1 according to (322) [101] As aresult the I2V communication in this scenario is by LOS It should however be noted thatLOS here does not mean pure LOS as there is still sparse blockage and scattering effects frompedestrians trees and road signs as illustrated in Figure 314

69

Figure 314 C-I2V deployment layout

PLOS(d3D) =

[min

(27

d 1

)(1minus eminus

d71

)+ eminus

d71

]2

(322)

342 Channel Model

We consider a clustered 3D SSCM based on the implementation in [42 101 102] butadapted and enhanced for C-I2V The fast-fading double-directional CIR (hdir) with Ncl

clusters and Nsp SPs rays or MPCs per cluster for each transmission link is given by (323)

hdir(t φ θ) =

Nclsumc=1

Nspcsums=1

PRXcs middotejϕcs middotδ(tminusτcs) middotGTX(φminusφcs θminusθcs) middotGRX(φminusφcs θminusθst)

(323)

where in (323) PRXcs ϕcs and τ cs denote the received power magnitude phase andpropagation time delay of the cluster-SP combinations respectively t is time φ and θ arethe angle offsets from the boresight direction for the azimuth and elevation respectively Foreach ray φcs and θcs are the azimuth AoD and elevation AoD at the AP (or azimuth AoAand elevation AoA for the vehicle as the case may be) GTX and GRX are the TX and RXantenna gains modeled as in (324) and (325) [101]

G(φ θ) = max

(G0e

αφ2+βθ2 Go100

) (324)

Go =41253 middot ξφ2

3dBθ23dB

α =4 ln(2)

φ23dB

β =4 ln(2)

θ23dB

(325)

70

where Go is the maximum directive boresight gain ξ is the average antenna efficiency φ3dB

and θ3dB are the azimuth and elevation HPBW respectively The variables α and β areevaluated using (325)

It should be noted that due to the high vehicular mobility (relative to static and pedestriancases) the channel becomes time-variant (ie the channel coherence time becomes smallerthan the observation window) The resulting phase ϕcs given by (326)-(328) is now com-posed of the distance-dependent phase change Θcs and the velocity-induced Doppler shiftϑD(cs) (caused by the Doppler frequency due to the relative motion between the TX andRX)

ϕcs = Θcs + ϑD(cs) (326)

ϕcs = 2π(fcτcs + fD(cs) 4 t

) (327)

ϕcs = 2π

(fcτcs +

vRX cos (φcs)

λ4 t

) (328)

where fD(cs) is the Doppler frequency which is positive when the vehicle is moving towardsthe AP and negative when moving away from it [49159]

343 Antenna Model

The APs and vehicles are equipped with massive MIMO arrays with NTX and NRX

AEs respectively We consider ULAs with inter-element spacing dTX = dRX = λ2 whereλ = cfc is the wavelength and c = 3times 108 ms is the speed of light For the massive MIMOthe hdir(t φ θ) in (323)-(328) is extended to H(t τ) in (329) where aRX and aTX are RXand TX array response vectors given by (330) and (331) respectively

H =

radicNRXNTXsumNclc=1Nspc

middotNclsumc=1

Nspcsums=1

radicPLcs(dcs) middot e

j2π

(fcτcs+

vRX cos(φcs)λ

4t)middot aRX

(φRXcs

)aHTX

(φTXcs

)

(329)

aRX(φRXcs

)=

1radicNRX

ej 2πλ dRX(nrminus1) sin(φRXcs )

forallnr = 1 2 NRX (330)

aTX(φTXcs

)=

1radicNTX

ej 2πλ dTX(ntminus1) sin(φTXcs )

forallnt = 1 2 NTX (331)

DSRC and LTE-A use limited numbers of AEs Typical MIMO configurations are 2times 22times 4 4times 4 and 2times 8 for NRX timesNTX On the other hand mmWave massive MIMO arrays

71

employ large number of AEs At 70 GHz for example the arrays can go up to 64 times 1024(with typical configurations being 16times 64 16times 128 and 32times 256 for NRX timesNTX accordingto the 3GPP The maximum number of RF chains or number of streams (Ns) at the RXand TX at such frequency are 8 and 32 respectively [3 4] The large arrays offer amazingopportunities to beamform highly-directive beam (through ABF) for enhanced signal strengthor to multiplex multiple streams (via DBF and HBF) for higher data rates Many studiesadvocate for HBF for mmWave massive MIMO for its balanced trade-off between SE and EErelative to the power-exhaustive DBF and the low-rate ABF [160] However for the evaluationin this subsection we employ ABF while the HBF analysis is employed and presented ingreater detail in Chapter 4 We note that we employ ABF for two reasons First the shortTX-RX separation distance LOS propagation and high level of antenna correlation due toSU-MIMO and sparse scattering [125] potentially guarantees near-optimal performance withABF Second single-stream beamforming ensures a fair comparison of the performance ofmmWave massive MIMO with the DSRC and LTE-A that use a modest number of antennasThe extension to the MU-MIMO case is presented in Chapter 4


In this subsection we present the simulation results for the performance of the threeconsidered technologies We simulate for 50000 TTIs and average the results over 100 channelrealizations For a fair comparison we set NF = 6 dB No = -174 dBmHz PT = 30 dBm andthe PLE = 2 The technology-dependent key simulation parameters for the three systemsconsidered are given in Table 38 Using the cumulative distribution function (CDF) wepresent results for the entire route traversed by vehicle We also show the results for arepresentative distance (ie 20 m) covered by each AP in the scenario

Table 38 Key simulation parameters for C-I2V

Parameter DSRC LTE-A mmWavefc (GHz) 59 26 73B (MHz) 10 20 396NTX 2 4 64NRX 2 4 16φTX3dB 65 65 10

φRX3dB 65 65 10

To compare the performance of DSRC and LTE-A with the mmWave massive MIMOadvocated in this work for the considered scenario we characterize the vehicular channelusing PL KF RMS DS (τRMS) number of clusters number of resolvable MPCs PDPchannel rank channel condition number and data rate as follows

(a) Path Loss

The effective PL (PLeff) which combines the PL and SF is given by (332) and (333)

PLeff = PL+ SF (332)

PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (333)

72

where n is the PLE and X is the log-normal random SF variable with zero mean and σstandard deviation [101] We note that blockage is modeled inherently in (333) as it matchesthe blockage-dependent PL model (334) in [18] when there is randi(01) number of blockersat each time instant This appropriately models the LOS and sparse blockage regime of theconsidered scenario where κ and Υ are parameters determined by the number of blockers(see Table 72 in [18])

PL = 10κ log10 (d3D) + Υ + 15

(d3D

1000

) (334)

Using (333) the CDFs of PLeff results per link for the three technologies are shown inFigure 315 As can be deduced from (333) PL expectedly increases with increasing fcHence the mmWave system at 73 GHz exhibits a penalty of up to 30 dB in PL comparedto the sub-6 GHz DSRC and LTE-A at 59 GHz and 26 GHz respectively The mmWavesystem however compensates for its high PL with large beamforming gains from highly-directive antennas in order to bring the received powers to levels comparable to that of sub-6GHz systems or even higher [18125]

50 55 60 65 70 75 80 85 90 95

Path Loss (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave

Figure 315 CDF of path loss for the three C-I2V technologies

In Figure 316 the PL variation as a function of the distance travelled for an AP-vehicleconnection time for the three systems is shown with and without spatial consistency Simi-larly the PL variations as a function of the distance for the entire 500 m route are shown inFigure 317 It should be noted that the plots in Figure 317 are periodic in nature due to theinter-AP handovers as the vehicle moves towards and away from the coverage area of each ofthe densely-deployed APs

73

0 2 4 6 8 10 12 14 16 18 20

Travelled distance (m)

40

50

60

70

80

90

100

110

120

Path

Loss (

dB

)DSRC without spatial consistency

DSRC with spatial consistency

LTE-A without spatial consistency

LTE-A with spatial consistency

mmWave without spatial consistency

mmWave with spatial consistency

Figure 316 Path loss variation for the coverage area of one AP

0 50 100 150 200 250 300 350 400 450 500


40

50

60

70

80

90

100

110

120

Path

Loss (

dB

)

DSRC without spatial consistency






Figure 317 Path loss variation for the entire route

74

As shown in the Figures 316 and 317 PL variation is random without spatial consis-tency due to the impact of SF With spatial consistency the variation is more uniform andsystematic We note that it is important for channel models to incorporate spatial consis-tency where the channel parameters vary in a realistic and continuous manner as a functionof position and by which closely-placed users have similar channel characteristics as againstrandomized values [100 161] We note that the large-scale parameters (LSPs) such as SF(and as a consequence the PLeff vary more slowly than the fast-fading small-scale parameters(SSPs) The spatial consistency phenomenon leads to three time scales channel correlationtime (Tc) for LSPs channel update time (Tu) for SSPs and then the data transmission time(Tt) The time scales are related by (335)

Tc = χ middot Tu = χ middot ε middot Tt (335)

where χ and ε are integer values We note further that Tt is standardized as 1 ms for 4Gand 5G systems However it is more realistic for channel-aware schedulers to employ Tu thatis used for updating the fast-fading channel parameters as the basis for scheduling Inspiredby [103 144 152 161] we adopt 01 m and 1 m as the update and correlation distancesrespectively At vRX = 10 ms employed for the simulation in this subsection this correspondsto χ = 10 ε = 10 Therefore Tu = 10 TTIs = 10 ms and Tc = 100 TTIs = 100 ms Thismeans that data is transmitted every 1 ms the SSPs are updated every 10 ms while the LSPsare updated every 100 ms

(b) Rician K-Factor

The KF statistics is a measure of the ratio of LOS-to-NLOS strength and its value affectsthe performance of MIMO systems significantly [162] Higher LOS translates to stronger prop-agation and higher SNR [157] The KF is evaluated using (336) [162] where the numeratorin (336) is the LOS component and the denominator is the sum of all NLOS components

KF =PLOSsumPNLOS

=PRXc=1s=1(sumNcl

c=1

sumNspcs=1 PRXcs

)minus PRXc=1s=1

(336)

Figure 318 shows the CDFs of the KF for the three systems evaluated using (336) Itcan be readily seen that mmWave has a higher KF than DSRC and LTE-A This indicateslarger LOS strength and higher directivity Figure 318 also shows that the powers of NLOScomponents dominate only about 20 of time (at 02 CDF points) where the KF is lessthan 0 dB while for the remaining 80 LOS component dominates It can also be observedthat the curves do not reach the 100 CDF points The gaps indicate the percentage ofpure LOS where only the LOS component is present (ie KF= infin) Figure 318 shows thatmmWave has higher percentage of pure LOS than the DSRC and LTE-A as can be seen atthe saturation points of the CDF curves

75

-20 0 20 40 60 80 100 120

K-Factor (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave

Figure 318 CDF of K-Factor for the three C-I2V systems

(c) RMS Delay Spreads

The RMS DS (τRMS) is a measure of the delay dispersion of the channel and is evalu-ated using (337) [115] It is also related to the channel coherence bandwidth (Bc) whichcharacterizes the frequency selectivity of the channel If Bc B (as is the case in widebandsystems like mmWave) the channel will be frequency-selective thereby leading to inter-symbolinterference (ISI) To combat this OFDM is employed in 4G and 5G systems to convert thefrequency-selective wideband channels to flat-fading channels The metrics (τRMS) and Bcare related by (338)

τRMS =

radicradicradicradicsumNclc=1

sumNspcs=1 τ2

csPRXcssumNclc=1

sumNspcs=1 PRXcs

minus

(sumNclc=1

sumNspcs=1 τcsPRXcssumNcl

c=1

sumNspcs=1 PRXcs

)2

(337)

Bc asymp1

2πτRMS (338)

The CDFs for the τRMS for the three systems are shown in Figure 319 The mmWavesystem has lower τRMS due to its narrower beams According to [18 144] highly-directivenarrow beams can reduce both the delay and Doppler spreads and increase the coherence timein mmWave channels This resulting outcome lessens the severity of the impact of Dopplerspread More so the values from the τRMS CDFs in Figure 319 when plugged into (338)gives Bc with the range [7160] MHz which are far higher than the 15625 kHz [163] 180

76

kHz [164] and 144 MHz [165] for one OFDM RB for DSRC LTE-A and mmWave systemsrespectively Similarly the τRMS with range [1 22] ns in Figure 319 is far less than theOFDM cyclic prefix duration of 16 micros [163] 52 micros [164] and 44 micros 057 micros [165] for theDSRC LTE-A and mmWave (5G NR) standards respectively Therefore ISI is not a problemwith OFDM as the CP duration is larger than the DS

2 4 6 8 10 12 14 16 18 20 22

RMS Delay Spread (ns)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave

Figure 319 CDF of RMS delay spreads for the three C-I2V systems

(d) Number of Clusters and Sub-paths

The number of clusters (Ncl) and SPs (Nsp) per cluster are randomly generated as uniformdiscrete distributions respectively The cluster (and sub-paths) powers delays and phasesfollow lognormal exponential and uniform (02π) distributions respectively [101] The CDFsfor the Ncl and Nsp are given in Figures 320 and 321 respectively Figure 320 showsthat for the considered scenario the mmWave system has two clusters on average while theDSRC and LTE-A have between 3 and 4 clusters Similarly as shown in Figure 321 themmWave system has 6 SPs while DSRC and LTE-A both have 24 SPs for the 50 CDFpoints The maximum number of resolvable rays are 9 40 and 42 for mmWave LTE-A andDSRC respectively Also Figure 321 shows that the mmWave system has limited numberof resolvable rays compared to the sub-6 GHz systems This outcome buttresses the sparsenature of mmWave systems

77

1 2 3 4 5 6

Number of Clusters

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave

Figure 320 CDF for the number of clusters for the three C-I2V systems

0 5 10 15 20 25 30 35 40 45

Number of resolvable MPCs

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave

Figure 321 CDF for the number of MPCs for the three C-I2V systems

78

(e) Power Delay Profile

In Figures 322 and 323 we show two snapshots of the PDPs for the three systemsconsidered PDPs show the distribution of the received signal powers of the MPCs with theircorresponding time delays which are used to characterize the channel with respect to the DSand coherence bandwidth [115] From Figures 322 and 323 it can be observed that mmWavehas lower number of clusters and overall number of rays when compared to the sub-6 GHzDSRC and LTE-A (Figures 320 and 321) as well as lower number of rays per cluster

0

-150

-100

mmWave

Re

ce

ive

d P

ow

er

(dB

)

-50

Absolute Time Delay (ns)

500

Technology

0

LTE-A

1000 DSRC

Figure 322 Power Delay Profile snapshot from the mth AP

0

-150

-100

mmWave200

Re

ce

ive

d P

ow

er

(dB

)

-50


Technology

0

LTE-A400

600 DSRC

Figure 323 Power Delay Profile snapshot from the nth AP

79

It should be noted that the first arriving ray is the LOS component With the sparsenature of the MPCs in mmWave the LOS-to-NLOS strength will be higher This is anindication of the higher directivity of the mmWave system as compared to the other twosystems with more diffuse scattering and lower directivity The MPCs with longer delaysgenerally have lower received powers

While the corresponding MPC (such as the LOS component) in the three systems havecomparable received directional powers (since the mmWave antenna gain compensates for thehigher PLeff) the sum of received component powers is lower in the mmWave systems due tohigher absorption and blockage such that the powers of many rays are below the noise leveland so they are filtered out

(f) Channel Rank and Condition Number

The rank of a channel matrix determines how many data streams can be sent across thechannel A full rank channel for example has rank = min(NRX NTX) While the channelrank is a pointer to the quantitative multiplexing capacity of the channel it does not indicatethe relative strength of the streams On the other hand the channel condition number (CN)is the qualitative measure of the MIMO channel

Using the singular values of the channel matrix CN indicates the ratio of the maximumto minimum singular values resulting from the singular value decomposition (SVD) of thechannel matrix A channel with CN = 0 dB has full rank and thus has equal gains across thechannel eigenmodes With 0 lt CN le 20 dB the channel is rank-deficient with comparablegains across the eigenmodes while CN gt 20 dB shows that the minimum singular value isclose to zero The rank of the channel gives the measure of how many data streams can bemultiplexed while the channel CN is an indicator of the quality of the wireless channel [102]In Figures 324 and 325 we show the variations of channel rank and CN respectively withthe indicated MIMO configurations

Connecting Figures 324 and 325 DSRC with rank 2 has 0 lt CN le 40 dB With therelative variation around 20 dB the channel strengths of the two eigenvalues are comparableThus two streams can be multiplexed over the channel with the water-filling power allocationgiving the optimal performance For LTE-A with 4 times 4 channel the rank varies between 3and 4 while the CN is typically higher than 20 dB thereby suggesting the multiplexing ofstreams even lower than the rank for good performance

Similarly according to Figures 324 and 325 the mmWave massive MIMO system with16 times 64 antenna configuration has rapid fluctuations in rank between 3 and 7 Howeverits extremely high CN (ie CN gt 160 dB) suggests relatively few dominant eigenmodesresulting from the high correlation of the tightly-spaced antennas at mmWave This outcomereveals the rank deficiency of mmWave SU-MIMO where single-stream ABF or HBF witha few streams will give good or optimal performance For MU-MIMO where the users aremore separated in space many users can be multiplexed Thus HBF or DBF becomes moreappropriate for higher system capacity

80

2 4 6 8 10 12 14 16 18 20


1

2

3

4

5

6

7

8C

hannel R

ank

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)

Figure 324 Channel rank for the three C-I2V systems

2 4 6 8 10 12 14 16 18 20


0

20

40

60

80

100

120

140

160

Channel C

onditio

n N

um

ber

(dB

)

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)

Figure 325 Channel condition number for the three C-I2V systems

81

(g) Data Rate

Using transceivers with NRFTX = NRF

RX = 1 RF chain for the ABF processing consideredthe beamforming and combining matrices reduce to vectors f isin CNTXtimes1 and w isin CNRXtimes1respectively The received signal y is then given by (339) The achievable data rate is givenby (340)

y =radicρwHHfs+ wHn (339)

R = B middot log2

(1 + ρRminus1

n wHHf times fHHw) (340)

where Rn = σ2nw

Hw is the noise covariance after combining

0 2 4 6 8 10 12

Data Rate (Gbps)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave

Figure 326 Data rates for the three C-I2V systems

Finally the data rate performance of the three systems are shown in Figure 326 The rateis evaluated using (340) and consistently shows the mmWave massive MIMO system realizingGbps rates compared to the DSRC and LTE-A While a direct comparison is unrealistic aswe employed different configurations for the three systems based on their respective baselinevalues according to the operating standards the results in Figure 326 motivates the usemmWave massive MIMO for Gbps vehicular communication particularly for 5G NR C-I2Vinvestigated in this section The data rates in Figure 326 results from using a bandwidth of10 MHz for DSRC [163] 20 MHz for LTE-A [164] and 396 MHz for mmWave system [165]as stated in the simulation parameters of Table 38

82

35 Conclusions

This chapter has presented the interplay of massive MIMO mmWave and UDN as thethree big technologies to support the 5G eMBB use case Using system-level simulations with3GPP 3D channel models we showed the performance of 3D microWave and mmWave channelmodels for UMa and UMi scenarios The results show that mmWave systems have highercoupling losses than microWave systems The results also show for the most part that microWavesystems are interference-limited while mmWave systems are noise-limited for the consideredscenarios Also indoor users served by outdoor BSs show highly degraded performanceThis is due to the additional 20-30 dB wall and indoor losses experienced by indoor usersThe degradation will become even more pronounced if much larger mmWave bandwidths areemployed due to the expected increase in noise

For the purpose of coexistence we also showed the performance of a representative 5GUDN with microWave MCs and mmWave SCs The impact of large bandwidth antenna directiv-ity traffic type and carrier frequency on the SE UE throughput and cell capacity were alsoinvestigated The results show that without proper design the performance of mmWave SCscan be highly degraded despite the abundance of available bandwidth in the mmWave bandsWe also showed that while the performance is highly promising for outdoor users the through-put experienced by indoor users is still generally poor This results from the SINR bottleneckarising from the noise-limited condition of mmWave systems and the high propagation lossesat such frequencies To overcome this challenge the design and operation parameters of 5Gnetworks (such as bandwidth transmit power beamforming configuration etc) have to becarefully considered This has to be done in a way that optimizes performance and efficiencywhile maintaining a balance in complexity and cost By mitigating the SINR challenge thespectral benefits at mmWave bands can be fully tapped This will unleash their potential forsupporting future cellular networksrsquo services and applications

We also characterized the vehicular channel for C-I2V communication where the infras-tructure are APs mounted on street-level lamposts in a UMi type environment Using diversechannel metrics we compared the channel statistics of I2V using mmWave massive MIMOwith that of legacy DSRC and LTE-A systems With modest system configurations weshowed that mmWave massive MIMO system can enable Gbps data rates for C-I2V commu-nication in order to support the anticipated explosive rate demands of future ITS Finallywe remark that the results in this chapter have been published in the proceedings of IEEEGlobecom conference2 IEEE Communication Magazine3 and the IET Intelligent TransportSystems journal4

2S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoImpact of 3D Channel Modeling for ultra-high speed Beyond-5G Networksrdquo IEEE Global Communications Conference (GLOBECOM) Workshop 2018Dubai United Arab Emirates pp 1-6 Dec 2018

3S A Busari S Mumtaz S Al-Rubaye and J Rodriguez ldquo5G Millimeter-Wave Mobile BroadbandPerformance and Challengesrdquo IEEE Communications Magazine vol 56 no 6 pp 137-143 June 2018

4S A Busari M A Khan K M S Huq S Mumtaz and J Rodriguez ldquoMillimetre-wave massive MIMOfor cellular vehicle-to-infrastructure communicationrdquo IET Intelligent Transport Systems vol 13 no 6 pp983-990 June 2019

83

84

Chapter 4

Novel Generalized Framework forHybrid Beamforming

This chapter focuses on beamforming techniques and the spectral and energy efficienciesanalyses for 5G UDNs and C-I2X networks In the first part of the chapter we propose a novelgeneralized framework for HBF in mmWave massive MIMO networks The proposed frame-work facilitates the comparative analysis and performance evaluation of the different HBFconfigurations the fully-connected the sub-connected and the overlapped subarray architec-tures The performance of the different array structures is evaluated using a C-I2X applicationscenario involving lampost-mount APs and a combination of pedestrian and vehicular usersThe second part considers the C-I2P use case between the street-level APs and pedestrianusers and evaluates the performance of the system using three beamforming approaches ana-log beamsteering hybrid precoding with zero forcing and the SVD precoding The SE and EEperformance and trade-off are presented with a view to providing useful insights for enablinghigh rate and energy-efficient operation for next-generation networks

41 Background

Massive soft super-fast and green are the key attributes that will describe NGMNs (5Gand beyond) The future networks are envisaged to be massive by concurrently featuringdenser cells (UDN) larger bandwidths (mmWave) and a higher number of antennas (massiveMIMO) in contrast to legacy cellular networks This will provide a platform for deliveringhuge performance gains across all fronts The soft and super-fast nature of the networks isreflected by their ability to be self-organizing and virtual However EE is becoming a criticaldesign factor in addition to SE This will raise significant design requirements that proliferatethroughout the system that includes efficient beamforming structure that is a key energyconsumer in next-generation transceiver designs [36]

Using appropriate antenna array structure and beamforming5 (or precoding) techniquesmassive MIMO can provide the needed array gains to counter the severe effects of the highPL at high frequencies (eg mmWave and THz) and thus increase the SNR for user devicesIt also provides the opportunity for highly-directional beams to mitigate MUI The large

5Beamforming is used in its widest meaning throughout this thesis such that beamforming andprecoding refer to exactly the same thing and can be used interchangeably (cf httpsma-mimoellintechse20171003what-is-the-difference-between-beamforming-and-precoding)

85

arrays can also be leveraged to provide spatial multiplexing gains by transmitting multiplestreams so as to boost SE [125 136] The combined benefits from the huge bandwidth inmmWave bands as well as the array and multiplexing gains from massive MIMO will lead tosignificant enhancement in user throughput QoE and system capacity Since EE is a criticalKPI the research community has been investigating different system architectures with aview to identifying the optimal model not only in terms of the SE-EE performance but alsowith respect to the hardware requirement cost and implementation complexity [166ndash168]

In this chapter we propose a novel generalized HBF framework for maintaining a balancedSE-EE trade-off in mmWave massive MIMO networks We focus on the MU-MIMO set-upsemploying different HBF architectures at the BSs (or TXs) and comparing their performancewith the ABF and DBF schemes In all cases the UEs (or RXs) employ the ABF architec-ture Performance of the proposed schemes are evaluated in 5G UDN and C-I2X applicationscenarios First we introduce the HBF schemes followed by the performance metrics andthen the system models performance evaluation results and discussions

42 Hybrid Beamforming Schemes

For massive MIMO three beamforming architectures (ie ABF DBF and HBF) havebeen identified [169] In the ABF implementation all the AEs are connected to a singleRF chain via a network of PSs For DBF each AE is connected to a dedicated RF chainThe HBF architecture uses a reduced number of RF chains It divides its structure into twostages the large-sized ABF stage for increasing array gain and the small-sized DBF stage formitigating MUI [127136170] Comparing the three architectures HBF maintains a balancedperformance between the ABF (with low SE power consumption cost and complexity) onthe one hand and the DBF (with high SE power consumption cost and complexity) on theother From the perspectives of SE EE cost and hardware complexity HBF thus strikesa balanced performance trade-off when compared to the fully-analog and the fully-digitalimplementations As a result HBF has been copiously demonstrated as a practically-feasiblearray structure for massive MIMO [167] It is thus the architecture of interest in this thesis6

Using the HBF architecture it is possible to realize three different array structures specif-ically the fully-connected the sub-connected and the overlapped subarray structures In thisthesis we present a novel generalized framework for the design and performance analysis ofthe HBF architectures The different subarray configurations can be realized by varying theparameter known as the subarray spacing which leads to the corresponding changes in systemperformance To proceed we recall the mmWave massive MIMO channel model and the ULAantenna array responses that will be used throughout this chapter where L is the number ofrays or MPCs and all other parameters are as defined in Chapters 2 and 3

PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (41)

6It is worth noting that the development trends show that the cost and power consumption of the fully-digitaltransceiver can be reduced in the future thereby making it the preferred choice for system implementation dueto its superior flexibility and performance [53133] For example the authors in [57] already report interestingresults using testbeds based on the fully-digital architecture for mmWave massive MIMO

86

hdir(t φ) =Lsuml=1

PRXl middot ejϕl middot δ(tminus τl) middotGTX(φminus φTXl ) middotGRX(φminus φRXl ) (42)

GTXRX(dB) = GAE(dB) + 10 log10(NTXRXsub ) (43)

H =

radicNRXNTX

LmiddotLsuml=1

radicPLl(d) middot e

j2π

(fcτl+

vRX cos(φl)λ

4t)middot aRX

(φRXl

)aHTX

(φTXl

) (44)

aRX(φRXl

)=

1radicNRX

ej 2πλ dRX(nrminus1) sin(φRXl )


aTX(φTXl

)=

1radicNTX

ej 2πλ dTX(ntminus1) sin(φTXl )


where in (43) GTX and GRX are calculated based on the antenna element gain (GAE) and

the number of AEs in the respective TX and RX subarrays (NTXRXsub )

421 Fully-connected HBF architecture

The fully-connected HBF architecture is shown in Figure 41 In this structure each RFchain is connected to all the AEs via a network of PSs [4] The user signal is first digitallyprecoded by the baseband precoder (FBB) then processed by every RF chain then fed to theanalog beamformer (FRF ) and thereafter transmitted using all AEs in the array [169] Byadjusting the phases of the transmitted signals on all antennas a highly-directive signal canbe achieved Here all the elements of FRF have the same amplitude thereby achieving thefull beamforming gain [4168] This structure is built at the cost of large number of PSs andcombiners where signal processing is carried out over the entire array in RF domain [166]Thus it consumes a high amount of energy for the excitation of the phased array networkand compensation of the insertion losses of the PSs It also has a high computational andhardware complexity [167]

87

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA

Analog beamformer (FRF)

s1

s2

sK

1

NTX

PSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs

Figure 41 Fully-connected HBF architecture

422 Sub-connected HBF architecture

The sub-connected HBF architecture is shown in Figure 42 In this configuration eachRF chain is only connected to a subarray via a network of PSs [4] Each userrsquos signal is firstdigitally precoded by FBB then processed by a single dedicated RF chain then fed to the FRF

and thereafter transmitted using only a set of AEs constituting a subarray [169] For each RFchain only the transmitted signals on a subarray can be adjusted Thus the beamforminggain and directivity is reduced by a factor equal to the number of subarrays AdditionallyFRF is a block diagonal (BD) matrix in this case where all non-zero elements have the samefixed amplitude [4 168] This structure employs less number of PSs (as compared to thefully-connected structure) and does not use combiners as there is no need to add the analogsignals [4] Thus it consumes less amount of energy for the excitation of the phased arraynetwork and compensation in terms of reduced insertion losses of the PSs It also has a lowercomputational and hardware complexity when compared to the fully-connected structure[167]

423 Overlapped subarray HBF architecture

The overlapped subarray HBF architecture is shown in Figure 43 In this structure eachRF chain is connected to a subarray via a network of PSs but the subarrays are allowed tooverlap [168 171] Here each RF chain is connected to antennas in more than one subar-ray This configuration is different from the sub-connected structure where each RF chain ismapped to only antennas in a single subarray It is also different from the fully-connectedstructure with each RF connected to all the antennas Therefore each userrsquos signal is first dig-itally precoded by FBB then processed by more than one RF chain then fed to the FRF andthereafter transmitted using only a set of AEs constituting the mapped subarrays With thisconfiguration the hardware complexity and the required number of components are reducedcompared to the fully-connected structure whilst maintaining system performance typically

88

in between the fully-connected and the sub-connected architectures [168] It is noteworthyto mention that the overlapped structure offers a broad range of opportunities that can beexplored as many configurations are realizable in the overlapped structure (ie all possibili-ties between the two extremes of the fully-connected and the sub-connected structures) Theopportunities even become wider as the systemrsquos NTX and NRF increase

Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs

Figure 42 Sub-connected HBF architecture

Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K

Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K

Figure 43 Overlapped subarray HBF architecture

89

424 Proposed Generalized Framework for HBF architectures

With respect to the HBF architecture three classes of structures can be realized depend-ing on the interconnection among the RF chains PSs and AEs (ie the RF-PS-AE map-ping) The three structures that have been proposed are the (i) fully-connected structure[167170] (ii) sub-connected structure [167169170] (also known as the partially-connectedor array-of-subarrays [166]) and (iii) overlapped subarray structure [168 171] In additionthe fully-connected and the sub-connected array structures have received considerable atten-tion in the literature [136 166 167] However the investigation of the overlapped subarraystructure is rather limited [168 171] A parameter known as the subarray spacing (4Ms)in [171] and (4N) in [168] was introduced for the overlapped subarray structure such thatvarying its value leads to the different subarray configurations and the consequent changesin system performance However no generalized framework is available in the literature fora comprehensive comparative analysis of the different structures in a unified and systematicmanner

In this section a novel generalized framework for the design and analysis of HBF antennaarray structures is proposed Beyond the state of the art the proposed generalized modelfacilitates not only the SE analysis of the different HBF array structures but also the EEanalysis Using the generalized framework it is possible to realize the three different HBFstructures and quantify the required number of the different components for the architecturesas well as the beam power allocation The generalized HBF architecture is shown in the TXside7 of Figure 44 (on next page) and the step-wise procedures for the design and analysis ofthe generalized framework are outlined as follows

(i) Set the transmit power (PT ) for the BS noting the appropriate regulation on the effec-tive isotropic radiated power (EIRP) limits

(ii) Set the number of AEs (NTX) for the BS of the massive MIMO system under consid-eration

(iii) Determine the number of RF chains (NRF ) based on the expected maximum multi-plexing capability of the BS and such that NTX

NRFis an integer Note that for any HBF

structure 1 lt NRF lt NTX

(iv) Set the inter-subarray spacing (4N) where 4N isin

0 1 2 NTXNRF

For the gener-

alized framework proposed in this work note that 4N is the critical parameter thatdetermines the specific array structure as given by (47)

4N =

0 rarr fully-connected[1 2

(NTXNRF

minus 1)]

rarr overlapped

NTXNRF

rarr sub-connected

(47)

7The receiversUEs of the considered systems in this chapter employ ABF leading to the (multi-beam) TXhybrid and RX analog (single stream per user) configuration For the short-range C-I2X scenarios investigatedthe RXsUEs are assumed to exhibit LOS communication and spatial correlation and are power-constrainedsuch that ABF becomes optimal for each UE

90

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Fig

ure

44

Gen

eral

ized

HB

Far

chit

ectu

re

91

(v) Determine the required number of PSs for each RF chain(NkPS forallk isin 1 2 NRF

)using (48) The total number of required PSs (NPS) for a BS is then determined using(49)

NkPS = NTX minus4N (NRF minus 1) (48)

NPS =

NRFsumk=1

NkPS = NRF timesNk

PS (49)

(vi) For each RF chain develop the mapping index vector mk using (410) where forallk isin1 2 NRF The entries in mk give the indices of the AEs connected to the kth RFchain

mk = [(k minus 1)4N + 1 NTX minus4N (NRF minus k) ] (410)

Note that the number of AEs in a subarray(NTXsub

)equals the number of PSs connected

to each RF chain where NTXsub = Nk

PS = |mk|

(vii) Develop the NTX timesNRF mapping index matrix (M) by stacking the mkrsquos Elements ofM are given by (411) and each element shows the connection index of the kth RF chainto the jth AE forallk isin 1 2 middot middot middot K forallj isin 1 2 J by letting K = NRF and J = NTX We note that M becomes a block diagonal matrix (blkdiag) for the sub-connected arraystructure

M =

m11 0 middot middot middot 0

m4N+12 middot middot middot 0

mNsub1 middot middot middot mJminusNsub+1K

0 m4N+Nsub2

0 0 0 mJK

(411)

(viii) Develop the NTX timesNRF (or J timesK) booleanbinary matrix B by replacing all nonzeroelements of M in (411) with ones (1primes) as shown in (412)

B =

1 0 middot middot middot 0

1 middot middot middot 0


0 1

0 0 0 1

(412)

92

(ix) Develop the NTX times 1 combiner vector (g) given by (413) indicating the number of RFchains connected to the jth AE

g =

g1

g2

gNTX

gj =

NRFsumk=1

Bjk forallj isin 1 2 NTX (413)

(x) Determine the number of combiners (Ncomb) needed for the specific array structureusing (414) Note that AEs connected to just a single RF chain do not need combinersAs a result the sub-connected structure has zero combiners as each AE is connected toa single RF chain while the fully-connected has the maximum number of combiners asevery AE is connected to all RF chains

Ncomb =∣∣∣[gj gt 1]NTXj=1

∣∣∣ =

NTX rarr fully-connected

NTX minus 24N rarr overlapped

0 rarr sub-connected

(414)

(xi) Determine the transmit beam power(P kb)

for each of the K beams using (415) where(middot) is the weighting factor and gj is evaluated using (413) The total transmit powerconstraint set in step (i) is enforced by (416)

P kb =PTNTX

sumjisinmk

(gj)minus1

(415)

PT =

NRFsumk=1

P kb (416)

Based on the enumerated design procedures the required number of components for thedifferent subarray configurations can be determined depending on the choice of 4N Weremark that while the design procedures outlined above focus on the BS or AP the gener-alized framework is equally applicable for the RXs or UEs by replacing the appropriate TXcomponents with the corresponding RX components

43 Precoding and Postcoding

Since HBF is considered at the TX the AP uses a baseband precoder or beamformerFBB isin CKtimesK followed by an RF precoder FRF isin CNTXtimesNTX

RF The transmit symbol vectors isin CKtimes1 and the sampled transmit signal vector x isin CNTXtimes1 in (417) are related by (418)

s = [s1 s2 middot middot middot sK ]T x = [x1 x2 middot middot middot xNTX ]T (417)

93

x = FRFFBBs (418)

Since ABF is considered at each of the users each UE employs an RF postcoder8 orequalizer wk

RF isin CNRXtimes1 The received signal vector (after the precoding and postcoding

operations) is y = [y1 y2 middot middot middot yK ]T where yk for each user forallk isin 1 2 middot middot middot K is given by(419) and n = k is the desired signal while n 6= k are interference terms for the kth UE

yk = wHk Hk

Ksumn=1

FRF fBBn sn + wHk nk (419)

In (419) Hk isin CNkRXtimesNTX is the channel matrix between the AP and the kth UE and nk is

the AWGN following a complex normal distribution CN (0 σ) with zero mean and σ standarddeviation The precoding and postcoding schemes considered in this thesis are described asfollows

431 Analog-only Beamsteering

In the analog-only beamsteering (AN-BST) scheme the RF beamformer fkRF isin CNTXtimes1

at the BSAP and the RF postcoder at the UE wkRF isin CNRXtimes1 forallk isin 1 2 middot middot middot K are all

implemented with analog PSs using (420) and (421) respectively

[fRFk

]=

1radicNTX

ejφk = aTX(φTXl

) (420)

[wRFk

]=

1radicNRX

ejφk = aRX(φRXl

) (421)

where FRF isin CNTXtimesK and WRF isin CNRXtimesK are given by (422) and (423) respectivelyThe elements of both matrices (FRF and WRF ) are constrained to have constant magnitudethough with variable phases As given by (420) and (421) the beamsteering codebooksFRF and WRF are parameterized by the TX and RX array response vectors aTX

(φTXl

)and

aRX(φRXl

) respectively The beamsteering codebooks are particularly suitable for sparse

and single-path channels (ie mmWave and THz channels) [125 166] as opposed to theGrassmanian codebooks which are usually designed for the rich channels of traditional MIMOsystems [136172]

FRF =[fRF1 fRF2 fRFK

] (422)

WRF =[wRF

1 wRF2 wRF

K

] (423)

8Recall that we use beamforming and precoding interchangeably Also we use postcoding to mean combin-ing or equalization at the UEs However the use of term ldquocombinerrdquo was avoided in order to avoid confusionwith the combiner used for adding the phase-shifted signals at the TX

94

432 Hybrid Precoding with Baseband Zero Forcing

The analog stage of the hybrid precoding with baseband zero forcing (HYB-ZF) schemeemploys FRF and WRF as in (422) and (423) respectively The digital stage then employsthe popular zero forcing (ZF) precoder FBB isin CKtimesK given by (424) and (425) to mitigateMUI

FBB =[fBB1 fBB2 middot middot middot fBBK

] (424)

FBB = HHeff

(HeffHH

eff

)minus1 (425)

where Heff = wHk HkFRF is the effective channel after applying the RF beamformer and

combiner to the channel The total transmit power (PT ) constraint is enforced by normalizingFBB according to (426) and (427) The two-stage hybrid precoding algorithm is given inAlgorithm 2

fBBk =fBBk

||FRFFBB||F forallk = 1 2 K (426)

||FRFFBB||2F = K (427)

433 Singular Value Decomposition Precoding

For the singular value decomposition precoding as upper bound (SVD-UB) precoding andcombining employ the unitary matrices (Fk and WH

k respectively) resulting from the SVD

of Hk isin CNkRXtimesNTX forallk isin 1 2 K - the channel matrix between the AP and the kth

UE according to (428)

[WkΣkFk] = svd(Hk) (428)

where Hk = WkΣkFHk and Σk represents the diagonal matrix for the channelrsquos singular val-

ues SVD-UB uses Σmaxk value for calculating the single-user rate which denotes the upper

bound It also represents the case with no MUI

95

Algorithm 2 Two-Stage Multi-User Hybrid Beamforming with Baseband Zero Forcing

Inputs Hk akRX akTX forallk isin 1 2 K larr (44)-(46)

OutputsFBB FRF WRF

First Stage Analog Beamforming

for k rarr 1 to K do- Set the RF beamformers fkRF and wk

RF to the beamsteering codebooks or arraysteering vectors akTX and akRX respectively

fRFk = akTX

wRFk = akRX

- Set FRF =[fRF1 fRF2 fRFK

]and WRF =

[wRF

1 wRF2 wRF

K

] according to

(422) and (423) respectively

Second Stage Digital Beamforming

for k rarr 1 to K do- Estimate the effective channel

Hkeff = wH

k HkFRF

- The linear baseband zero forcing precoder FBB =[fBB1 fBB2 middot middot middot fBBK

]is designed

using (425)- The total power constraint is enforced using the fBBk normalizations in (426)forallk isin 1 2 K

44 Power Consumption Model

The power consumption model of the generalized HBF array structure in Figure 41 isgiven in this subsection We model a realistic power consumption framework considering notonly the power consumption at the RAN (PRAN ) but also the backhaul power consumption(PBH) The total consumed power (Ptotal) is thus given by (429) The components of PRANand PBH are further described as follows

Ptotal = PRAN + PBH (429)

(i) RAN Power (PRAN ) For the coverage of a single BS the PRAN given by (430)consists of the transmit power of the BS (PT ) the circuit power of the BS (P TXcct ) andthe combined RX circuit powers (PRXcct ) of all the NUE users served by the BS Differentfrom most existing studies such as [37166167] we include the power consumed by theRXs in the model

PRAN = PT + P TXcct +NUE

(PRXcct

) (430)

96

(ii) Backhaul Power (PBH) This involves the power consumed for the communication be-tween the BS and the core network It is given by (431) and it is dependent on thedata rate (R) or the amount of data transferred per unit time (bitss)

PBH = LBH middotR (431)

In (431) LBH = 250 mW(Gbitss) [173] is the power per unit data rate while theP TXRFC and PRXRFC are given by (432) and (433) respectively [166] The P TXcct PRXcct andPtotal are correspondingly given by (434)-(436) The breakdown of the values of the differentcomponents are given in Table 41 We assume a fixed miscellaneous power PFIX = 1 W(noting that SC BSs or APs do not have any active cooling system [174])

Table 41 Power consumption of components

TX Component Notation Unitpower[mW]

RX Component Notation UnitPower[mW]

Digital to Analog Con-verter

PDAC [175] 110 Analog to Digital Con-verter

PADC [175] 200

Mixer PMIX [166] 23 Mixer PMIX [166] 23Local Oscillator PLO [166] 5 Local Oscillator PLO [166] 5Low Pass Filter PLPF [166] 15 Low Pass Filter PLPF [166] 15Phase Shifter PPS [175] 30 Phase Shifter PPS [175] 30Power Amplifier PPA [175] 16 Low Noise Amplifier PLNA [175] 30Baseband precoder PBB [175] 243 - - -Combiner Pcomb [173] 195 - - -

P TXRFC = PDAC + PMIX + PLO + PLPF (432)

PRXRFC = PADC + PMIX + PLO + PLPF (433)

P TXcct = NTXRF P

TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX

combPTXcomb + P TXBB + PFIX (434)

PRXcct = NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA (435)

Ptotal =PT +[NTXRF P

TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX

combPTXcomb + P TXBB + PFIX

]NUE

[NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA

]+ LBH middotR

(436)

97

45 Spectral and Energy Efficiency

Following the given generalized HBF antenna structure and the channel beamformingand power consumption models we give the expressions for the performance of the systemin this section in terms of the achievable sum data rate (R) spectral efficiency (ηSE) andenergy efficiency (ηEE) The main objective is to efficiently design FRF FBB and WRF tomaximize the system performance

451 Spectral Efficiency and Achievable Rate

Given the received signal yk in (419) the ηkSE [(bitss)Hz] Rk (bitss) of the kth userand R for sum data rate of all users are given by (437)-(439) where Bk is the bandwidthallocated to the kth user and the SINR for the respective precoding techniques are given in(440)-(442) forallk = 1 2 K

ηkSE = log2 (1 + SINRk) (437)

Rk = Bk times ηkSE (438)

R =

Ksumk=1

Rk (439)

SINRANminusBSTk =P kT middot

∣∣wHk Hkf

RFk

∣∣2sumn6=k P

nT middot∣∣wH

k HkfRFn∣∣2 + σ2

k

(440)

SINRHY BminusZFk =P kT middot

∣∣wHk HkFRF fBBk

∣∣2sumn6=k P

nT middot∣∣wH

k HkFRF fBBn∣∣2 + σ2

k

(441)

SINRSV DminusUBk = P kT middot |Σmaxk |2 (442)

452 Energy Efficiency

The EE (bitsJ) of the system is the ratio of the system throughput or sum data rate R(given by (439)) to the total power consumption Ptotal (expressed as (436)

ηEE =sum rate

total power consumed=

R

Ptotal (443)

98

The optimal EE as a function of the SE ηlowastEE(ηSE) is given according to the fundamentalEE-SE relation [176] by (444) ∣∣∣∣dηlowastEE(ηSE)

dηSE

∣∣∣∣ηSE=

R

BW

= 0 (444)

where ηlowastEE(ηSE) = max (ηEE(ηSE)) is strictly quasiconcave in ηSE when Ptotal includes boththe transmit power PT and the circuit power Pcct [176177]

46 Hybrid Beamforming for C-I2X

The performance analyses of the HBF structures have been investigated for diverse sce-narios The authors in [125167178] considered single-cell single-user multi-stream commu-nication while the authors in [136 171] analyzed for the single-cell multi-user multi-streamsystem In [138] the HBF evaluation was extended to the multi-cell multi-user multi-stream scenario However these evaluations consider the typical cellular deployments withISD ge 500 m for the microWave setups and 50-200 m for the mmWave scenarios Extension tothe short-range domain using street-level lampost-mount APs with ISDs le 10 m particularlyfor outdoor applications is still missing Using the generalized HBF framework we evaluatethe performance of different HBF configurations using a C-I2X application scenario involvingboth cellular and vehicular users We employ a realistic power consumption model to assessthe performance of the respective systems not only in terms of the SE and EE but also thepower and hardware cost for the network Different from most existing works our compre-hensive power consumption model includes the TX power the TX circuit power RX circuitpower as well as the backhaul power as given in Section 44

461 System Model and Parameters

We consider the network deployment layout for C-I2X already introduced in Figure 16We focus on the single-cell multi-user DL scenario where a massive MIMO AP communicatesNUE user devices (i e the sum of cellular and vehicular users being scheduled in each TTI)We consider an urban street deployment where the APs are mounted on street lamposts alonga road 500 m long The APs are evenly spaced at 10 m interval and are mounted at a heighthTX = 5 m on the walkway lamposts All UEs are at a height of hRX = 15 m Each cellularuser (cUE) traverses the route at a pedestrian speed of vcRX = 36 kmh while each vehicularuser (vUE) moves at vvRX = 36 kmh respectively The width (w) of the walkway for cUEsis 2 m while the vUEs are at a further 3 m from the walkway At each time instant the I2XDL connectivity is by LOS as the 3D separation distance between each user and its servingAP gives a LOS probability PLOS asymp 1 according to (321) [101] We consider the channelmodel earlier given by (41)-(46) Given that GAE is the gain of an AE and NTX

sub and NRXsub

are the number of TX and RX AEs in a subarray respectively GTX and GRX are given by(445) and (446) respectively [179]

GTX(dB) = GAE(dB) + 10 log10(NTXsub ) (445)

GRX(dB) = GAE(dB) + 10 log10(NRXsub ) (446)

99

Using the generalized structure introduced in Section 424 a table such as Table 42can be populated with the appropriate entries based on the values set and determined using(47)-(416) In Table 42 we give the values of the required number of components for thesample scenario considered in this work where the AP is equipped with NTX = 64 NTX

RF = 8and 4N = 0 2 4 6 8 The AP employs the generalized HBF structure as shown in Figure44 This leads to the fully-connected structure when 4N = 0 and leads to the sub-connectedstructure when4N = NTXNRF = 8 while4N = 2 4 6 represent the overlapped subarraystructure

For the UEs we consider NRX = 8 NRXRF = 1 and 4N = 0 for all the K = 8 users This

corresponds to a fully-connected structure for the single stream per user ABF configurationat the UEs The numbers of the respective required components for each UE are also givenin Table 42 Thus we focus on the multi-user beamforming case with a single stream peruser Therefore the total number of streams is Ns = NUE = NTX

RF = 8

Table 42 HBF array structure components

TX 4N = 0 4N = 2 4N = 4 4N = 6 4N = 8 RX 4N = 0NTX 64 64 64 64 64 NRX 8NPA 64 64 64 64 64 NLNA 8NRF 8 8 8 8 8 NRF 1Nsub 64 50 36 22 8 Nsub 8NPS 512 400 288 176 64 NPS 8Ncomb 64 60 56 52 0 Ncomb 0

Different from most works in the literature where P kT = PT K we note that this is simplynot the case for P kT for the generalized framework explored in this work particularly for theoverlapped subarray structure as already given in (415) In the literature where the fully-connected or sub-connected subarray structures are typically employed it is easy to assumethe uniform power allocation In such settings each AE is connected to either all the RFchains (as in the fully-connected case) or to one RF chain only (for the sub-connected subarrayconfiguration) However for the overlapped subarray structure each of the AEs is connectedto different amount of RF chains depending on the configuration (based on 4N) Hence thebeam power is dependent on the RF chain-AE mapping

Table 43 Power allocation for the array structures

Beam 4N = 0 4N = 2 4N = 4 4N = 6 4N = 81 1 12107 14214 15000 12 1 09964 09928 09583 13 1 09130 08261 07917 14 1 08797 07595 07500 15 1 08797 07595 07500 16 1 09130 08261 07916 17 1 09964 09928 09583 18 1 12107 14214 15000 1

PT (W) 8 8 8 8 8

To illustrate (415) for the configurations considered in this work where the APBS isequipped with NTX = 64 NTX

RF = 8 4N = 0 2 4 6 8 K = 8 and PT = 8 W the beam

100

power to each of the 8 users for the different values of 4N are given in Table 43 In eachcase the total power constraint is enforced As evident from the Table 43 the fully-connected(4N = 0) and the sub-connected (4N = 8) structures have equal power allocation while theoverlapped structures have unequal power allocation


In this section simulation results are provided to illustrate the system performance interms of ηSE ηEE and hardware cost and to compare the performance of the different sub-array configurations Further to the given deployment parameters the other key simulationparameters are further given in Table 44 In each run or channel realization users arerandomly deployed for the first TTI Thereafter each UE follows its mobility course (withrespect to speed and direction) throughout subsequent TTIs The results are averaged overthe simulations runs and TTIs

Table 44 Simulation parameters for C-I2X scenario

Parameter Description Valuefc Carrier frequency 100 GHzB Bandwidth 2 GHz

X(micro σ) Shadow fading (0 7) dBNo Noise power spectral density -174 dBmHzNF Noise figure 6 dBPT Transmit power [01 middot middot middot 8] WGAE Antenna element gain 8 dBi

GTXmax Maximum transmit gain 26 dBiGRXmax Maximum receive gain 17 dBinRuns Number of channel realizations 1000

(a) Power and Hardware Costs

First we provide a comparative performance analysis of the structures with respect tothe hardware requirement and power consumption In Table 42 we provided a breakdownof the hardware components needed to realize each of the structures for the case whereNTX = 64 NTX

RF = 8 for the AP and NRX = 8 NRXRF = 1 for each of the K = 8 UEs For

the phased array network the number of required PSs (NPS) changes significantly with thespecific structure In Figure 45 we show how NPS varies with the NRF for a fixed NTX

With a PPS = 30 mW per unit PS the overlapped subarray structures offer a window ofopportunity between the fully-connected and the sub-connected structures at the two extremeends in terms of power consumption The case is similar with respect to the number ofcombiners Ncomb required for each of the structures However the contribution of the PSs tothe Ptotal is far greater than that of the combiners both in terms of the number required aswell as the unit component power cost as can be seen in Tables 41 and 42 respectively

With respect to the overall power consumption of the network Figure 46 shows the Ptotalfor the different structures as well as the proportions of the contributing components (iethe consumed power at the TX (PTX) the consumed power at all RX (PRXs) and the powerconsumed for backhauling (PBH) when PT = 1 W Note that Ptotal = PTX + PRXs + PBH where PT is included in PTX

101

1 2 3 4 5 6 7 8

50

100

150

200

250

300

350

400

450

500

550

Figure 45 Number of PSs required for different structures (NTX = 64)

Figure 46 Power consumption of components for different 4N (PT = 1 W)

From Figure 46 it can be seen that Ptotal decreases as we move from the fully-connected(4N = 0) through the overlapped subarray to the sub-connected structure (4N = 8)

102

Similarly as we move from 4N = 0 to 4N = 8 the contribution of PTX and PBH reduceswith PTX reducing at a faster rate than PBH Except for the fully-connected structurePBH gt PTX for all structures As noted in [174] the computation and backhaul powerconsumption will constitute the largest of the total power consumption of future networksFor all cases the PRXs maintain steady values constituting roughly 10 and 15 of Ptotalfor the fully-connected and sub-connected structures respectively

(b) Spectral and Energy Efficiency

The sum SE and the EE performance for different transmit powers (PT ) for all the consid-ered structures are shown in Figures 47 and 48 respectively For all the subarray structuresin Figure 47 the SE increases logarithmically as PT increases The trend for the EE ishowever different As shown in Figure 48 the EE first increases peaks (at the optimalvalue) and then decreases as PT increases The optimal EE point is critical for the design ofenergy-efficient networks as the increase in PT beyond the optimal value leads to performancedegradation of the system with respect to the EE though the SE continues to increase

0 1 2 3 4 5 6 7 8

Total TX Power (W)

140

160

180

200

220

240

260

Su

m S

pe

ctr

al E

ffic

ien

cy (

bitss

Hz)

Figure 47 Sum Spectral efficiency versus total transmit power

Figure 49 shows the EE-SE performance trade-off curves for the different subarray struc-tures For each structure the EE starts to increase as the SE increases It then reachesthe optimal point (given by (446)) and thereafter continues to decrease as SE increasesEach curve follows a quasi-concave (bell-shaped) trend which is consistent with the resultsin [166176177] Further it can be observed that each of the structures has different perfor-mance The fully-connected structure has the highest SE and lowest EE on one end while thesub-connected structure has the lowest SE and highest EE on the other end In between thesetwo ends the different overlapped subarray structures show varying performance depending

103

on4N For example the overlapped structure with4N = 4 strikes a good balance in EE-SEperformance for the considered scenario

0 1 2 3 4 5 6 7 8

Total TX Power (W)

11

115

12

125

13

135

14

145

15

En

erg

y E

ffic

ien

cy (

Gb

J)

Figure 48 Energy efficiency versus total transmit power

18 20 22 24 26 28 30 32 34

Average Spectral Efficiency (bitssHz)

11

115

12

125

13

135

14

145

15

Energ

y E

ffic

iency (

GbJ

)

Figure 49 Energy efficiency versus spectral efficiency

104

(c) User Performance

In Figure 410 we show the SE performance for all the users as the PT increases Similarlythe EE performance for all users with increasing PT is shown in Figure 411 We show theresults for only the overlapped subarray case with 4N = 4 which was previously shown asthe structure having a balanced performance trade-off with respect to SE and EE

22

0

24

26

8

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

28

05 76

Transmit Power (W)

30

5

User

32

41 32

115

24

25

26

27

28

29

30

31

Figure 410 Spectral efficiency vs transmit power for each user (4N = 4)

11

0

12

8

Energ

y E

ffic

iency (

GbJ

)

05 7

13

6

Transmit Power (W)

5

User

14

41 32

115

114

116

118

12

122

124

126

128

13

132

134

Figure 411 Energy efficiency vs transmit power for each user (4N = 4)

105

In Figures 410 and 411 it can be seen that the performance of each of the users arevery close in values at each PT point thus guaranteeing a level of fairness among users Astypical Figure 410 follows the logarithmically-increasing trend in SE as PT increases for eachof the users Similarly the curves in Figure 411 follows the quasi-concave trend in EE asthe PT increases for each of the users In addition with equal bandwidth allocation per userRk = ηkSE times (BK) Therefore each user is able to reach a data rate of more than 5 Gbpswith a minimum SE of 20 bitssHz in a 2 GHz bandwidth for 8 users

47 Hybrid Beamforming for C-I2P

While the analysis in Section 46 involves both cellular and vehicular users (C-I2X) thissection provides discussion on the C-I2P scenario Among several use cases the authors in[151] proposed that mmWaveTHz APs mounted on street lamposts can be used to provideultra-broadband connectivity to low-mobility (pedestrian) users along street walkways andsimilar hotspots The proximity of users to these APs and the abundant bandwidth atmmWave and THz frequencies make the setup a viable candidate to offload traffic fromthe BSs in B5G networks Unlike in the C-I2X scenario where the focal point was on theperformance of the different HBF configurations in C-I2P we direct the attention towardsthe performance of the different precoder designs (ie AN-BST HYB-ZF and SVD-UB) ashighlighted in Section 43


The deployment layout for the massive MIMO network is shown in Figure 412 For thewalkway scenario we focus on the DL where the evenly-spaced APs provide connectivity tothe pedestrian users We assume the APs are connected by high-rate wireless backhaul linksand that each user is connected to a single AP at each time instant For the walkway scenariothe AP is mounted at a height hTX = 5 m on a street lampost thus stationary Each useris at a height of hRX = 15 m and traverses the route at a pedestrian speed of vRX = 36kmh The width (w) of the walkway is 2 m With the short TX-RX separation distance dwe consider a single-path channel L = 1 and assume LOS connectivity between the AP andthe UEs as PLOS asymp 1 according to [101]

U1

U7

00

1

U5

AP

U8 U2

2

10

3

AP

UE

he

igh

t (m

)

05 8

4

U3

5

Walkway width (m)

6

U6

1

Walkway length (m)

U4

415 22 0

Figure 412 Deployment layout for C-I2P scenario

106

Using the two-stage multi-user HBF scheme consider the system with NTX AP antennasand NTX

RF RF chains such that NTXRF lt NTX (TX HBF) and K terminals each with NRX

antennas and only one RF chain (RX ABF) For a fully connected hybrid beamformer system

the AP employs a baseband (BB) digital beamformer FBB isin CNTXRF timesNs followed by the analog

RF beamformer FRF isin CNTXtimesNTXRF such that the transmitted signal becomes x = FRFFBBs

The received signal vector rk observed by the kth terminal after beamforming can then beexpressed as (447) After being combined with the analog combiner wk where wk has similarconstraints as the analog beamformer FRF the signal yk becomes (448) The multibeam TXhybrid-analog RX configuration thus described is shown in Figure 413 where the HBF TXemploys the fully-connected array structure

rk = Hk

Ksumn=1

FRF fBBn sn + nk (447)

yk = wHk Hk

Ksumn=1


Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Transmitter

1

NRX

1

NRX

Analog beamformer (WRF)

User K

LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

RF

chain

RF

chain

LNA

LNA

LNA

RF

chain

yK

RF

chain

RF

chain

y1

Receivers

User 1LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

1

NRX

LNA

LNA

LNA

1

NRX

1

NRX


User K

LNA

LNA

LNA

RF

chain

yK

RF

chain

y1

Receivers

User 1LNA

LNA

LNA

1

NRX

GTX GRX

Channel

Figure 413 Hybrid (multi-beam) beamforming and analog combining (single beam per user)system architecture

In each run the users are randomly deployed for the first TTI Thereafter each UE followsits mobility course (with respect to speed and direction) throughout subsequent TTIs At1 ms speed and TTI length of 1 ms it takes 10000 TTIs to traverse the coverage areaof the AP (10 m end-to-end) The results are averaged over 200 simulations runs (whereeach run is 10000 TTIs) First we evaluate the system performance using the baselinesimulation parameters given in Table 45 and thereafter we investigate the impacts of otherkey parameters such as carrier frequency bandwidth antenna gain etc

107

Table 45 Baseline simulation parameters for C-I2P scenario

Parameter Value Parameter Valuefc 100 GHz B 1 GHz

X(micro σ) (0 4) dB n 2No -174 dBmHz NF 6 dBNTX 64 NRX 8hTX 5 m hRX 15 mGTX 25 dBi GRX 9 dBiK 8 vRX 36 kmhPT [050] dBm PRF 153 mWPPS 30 mW PPA 16 mWPBB 243 mW LBH 250 mW(Gbs)nTTIs 10000 nRuns 200


In this section we present the simulation results using the SE and EE performance forthe three sets of precoding techniques described in Section 43

(a) Baseline Performance

The baseline results realized with the key simulation parameters and values in Table 45are shown in Figure 414 As PT increases the average (ie per user) SE for the HYB-ZFand SVD-UB schemes increase almost linearly A gap of about 4 bitssHz exists between theHYB-ZF and the SVD-UB that serves as the upper bound on the achievable rate While theSVD-UB considers no interference at all the HYB-ZF only mitigates the interference As forAN-BST the SE performance is almost flat This results from the inability of the techniqueto mitigate MUI thereby leading to a somewhat low SINR relative to the other two schemesIn this kind of scenario where the users are very close to the TX and thus have high SNRsinterference mitigation is critical in order to lower the interference levels and guarantee goodSINR levels

The EE curves in Figure 414 show the typical quasi-concave trend where the EE firstincreases as PT increases then reach the peak points and thereafter continue to reduce [177]From the performance curves in Figure 414 the optimal PT for the joint EE-SE optimizationare the points where the EE and SE curves intersect for the respective precoding techniqueThe optimal PT is in the range 40-41 dBm for both HYB-ZF and SVD-UB Increasing thePT beyond the optimal points leads to increase in SE but at the expense of reduction in theEE of the system More so the average SE of sim27 bitssHz translates to up to 337 Gbpsthroughput per user and sim27 GbitsJ in terms of EE for HYB-ZF at the joint EE-SE optimalPT of 40 dBm for example However at the peak EE point of 29 GbitsJ for HYB-ZF (wherePT = 30 dBm) the SE reduces to 24 bitssHz (ie 3 Gbps per-user throughput)

It is instructive to note however that EE would be a more critical design goal than SE inorder to facilitate the green operation of future networks The performance of AN-BST followsa markedly different trend With the relatively-low flat average SE of 4 bitssHz lowerPT appears more energy-efficient as increase in PT does not bring about any correspondingincrease in SE This is due to the limiting impact of interference on the achievable rate Thisagain provides the impetus for interference mitigation in MU-MIMO systems

108

0 5 10 15 20 25 30 35 40 45 50

Transmit Power (dBm)

0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB

Figure 414 EE-SE performance as a function of PT (baseline)

0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB

Figure 415 EE-SE performance as a function of PT [fc = 28 GHz]

(b) Impact of carrier frequency

To show the impact of fc on the baseline results we changed fc from 100 GHz to 28GHz while keeping all other baseline system parameters constant The system performanceat 28 GHz is shown in Figure 415 It can be observed that the trends of the EE and SE

109

curves are the same as those of the baseline results in Figure 414 However the SE plots forSVD-UB and HYB-ZF are around 3 bitssHz higher at 28 GHz in Figure 415 in contrastto the 100 GHz baseline plots of Figure 414 This outcome results from the expected effectof reduction in PL as fc decreases However the larger available bandwidth and the higherantenna gain realizable at higher mmWave bands offer gains that will translate to higheroverall throughputs in the higher mmWave bands (100 GHz) than at the lower mmWavefrequencies (28 GHz)

(c) Impact of bandwidth

With respect to the effect of bandwidth on system performance Figure 416 shows theperformance when the bandwidth of the 100 GHz setup is increased from 1 GHz (Figure 414)to 5 GHz (Figure 416) while all other baseline parameters remain constant The performancetrend remains the same for both figures and for all techniques and evidently for both the EEand SE metrics However a decrease of 2-3 bitssHz can be observed on the SE performancefor SVD-UB and HYB-ZF as the bandwidth increased from 1 GHz to 5 GHz This reductionin SE is attributable to the increase in noise as the bandwidth increased Nevertheless thelarger bandwidth leads to increased user throughput and capacity for the 100 GHz mmWavesystem

0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB

Figure 416 EE-SE performance as a function of PT [B = 5 GHz]

(d) Impact of antenna gain

In Figure 417 we show the SE-EE performance of the system when the TX antenna gainis reduced from 25 dBi (Figure 414) to 18 dBi (Figure 417) while keeping the UE gain at 9dBi as before The results follow a similar pattern as the baseline though with a reduction inthe SE The lower GTX at the AP translates to wider beams that potentially causes higherinterference leading to reduced performance in Figure 417 relative to the baseline in Figure

110

414 Therefore sharper and narrower beams are more advantageous in scenarios like the oneconsidered in this work

0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB

Figure 417 EE-SE performance as a function of PT [GTX = 18 dBi]

0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

GbitsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB

Figure 418 EE-SE performance as a function of PT [TX = UPA]

111

(e) Impact of antenna geometry

Keeping the carrier frequency at 100 GHz bandwidth at 1 GHz and other parametersemployed for Figure 414 constant we changed the AP antenna geometry from ULA to UPAThe resulting plots in Figure 418 show similar trends (ie linear in SE and quasi-concave inEE) as those in the baseline results (Figure 414) While the HYB-ZF and SVD-UB results inboth Figures 414 and 418 are relatively the same the EE and SE values for the AN-BST arelower in Figure 418 than in Figure 414 With the antenna elements arranged in planar forminter-beam interference is higher with UPA than in ULA for the scenario under considerationThis interference effect is not mitigated in the AN-BST (unlike the HYB-ZF and SVD-UB)schemes hence lower the lower SE and EE performance It is instructive to note that theimpact of antenna geometry would be obvious with scenarios having users at different heightswhere 3D beamforming would provide the opportunity to discriminate users at same azimuthbut different elevation angles [13]

48 Conclusions

In this chapter we proposed a novel generalized HBF array structure for the DL multi-user mmWave massive MIMO network The generalized framework enables the design andcomparative performance analysis of different possible subarray configurations (ie the fully-connected the sub-connected and the overlapped subarray structures) The performance ofthe proposed model was analyzed within a C-I2X application scenario where ldquoXrdquo is com-bination of pedestrian users and high-mobility vehicular users The results show that theoverlapped subarray structure can provide a balanced performance trade-off in terms of SEEE and hardware costs in contrast to the popular fully-connected structure (with high SEand limited EE) and the sub-connected structure (with reduced SE and high EE)

In particular the overlapped subarray structures (depending on4N) can provide SE gainsin the range of 11-25 over the sub-connected array structure while approaching the 275gain in SE of the fully-connected architecture In a similar vein the overlapped subarraystructure suffers between 34 to 149 reduction in EE relative to the sub-connected structurein contrast to the 196 loss in EE of the fully-connected architecture With a balanced SE-EE trade-off the overlapped subarray structure therefore shows potential for NGMNs thattargets both high-rate and energy-efficient operation of the network

Similarly using a C-I2P application scenario we have shown the impacts of carrier fre-quency bandwidth antenna gain and antenna geometry on the EE and SE performanceusing a candidate 5G scenario (with street-level lampost-mount APs providing connectivityto pedestrian users) Using a single-cell multi-user setup where a massive MIMO AP com-municates to multiple users with single stream per user we compared the performance ofthe three precoding schemes AN-BST HYB-ZF and SVD-UB The results show that theAN-BST scheme (with no baseband precoder for MUI mitigation) shows poor performancewhen compared to the other two schemes

On the other hand the HYB-ZF scheme which employs a baseband ZF precoder forMUI mitigation approaches the performance of the upper bound SVD-UB with a gap of sim5bitssHz in SE and a gap of sim1 GbitsJ in EE at the optimal operating points for the energyand spectrally-efficient network operation Finally we note that the results in this chapter

112

have been published in the IEEE Transactions on Vehicular Technology9 and the proceedingsof IEEE International Conference on Communications (ICC)10

9S A Busari K M S Huq S Mumtaz J Rodriguez Y Fang D C Sicker S Al-Rubaye and ATsourdos ldquoGeneralized Hybrid Beamforming for Vehicular Connectivity using THz Massive MIMOrdquo IEEETransactions on Vehicular Technology pp 1-12 June 2019

10S A Busari K M S Huq S Mumtaz and J Rodriguez ldquoTerahertz Massive MIMO for Beyond-5GWireless Communicationrdquo IEEE International Conference on Communications (ICC) 2019 Shanghai PRChina pp 1-6 May 2019

113

114

Chapter 5

Conclusions and Future Work

This chapter provides the summary of the principal findings and design recommendationson the concepts investigated in this thesis channel modeling beamforming and system-levelsimulations of mmWave massive MIMO for 5G UDNs and C-I2X The future research di-rections are then presented as related open issues for NGMNs in the areas of THz channelmodeling ultra-massive MIMO quantum machine learning and EE optimization for the forth-coming 6G era

51 Thesis Summary

The data traffic forecasts for the 5G era (2020 onwards) imply that the available band-widths in the sub-6 GHz microWave bands will be insufficient to meet the data rate demandsof users Next-generation cellular and vehicular applications for example are envisaged torequire DL data rates in the order of multi-Gbps For these two domains of mobile wire-less connectivity the mmWave spectrum can provide the multi-GHz contiguous bandwidthsrequired to meet the projected throughput demands [180181] As a result mmWave commu-nication has received considerable attention in the research community in the present decadeThis trend is expected to continue as the progress being made in various areas of mmWavecommunication (such as electronic components channel modeling spectrum allocation stan-dardization use cases etc) continue to spur further research activities [151166]

To enable mmWave communications the TXs (and RXs) must use massive MIMO antennaarrays [151] This is because the free space path loss (FSPL) increases as the fc increasesaccording to the fundamental Friis equation [18] Fortunately a 10times increase in fc (egfrom 28 GHz to 28 GHz) correspondingly leads to a 10-fold reduction in λ As a resultthe dimensions of the AEs as well as their inter-element spacing become incredibly small(due to their dependence on λ) It thus becomes possible to pack a large number of AEs ina physically-limited space thus enabling the mmWave massive MIMO paradigm The highPL at mmWave frequencies constrains the systems employing mmWave massive MIMO todistances much shorter than the ISDs of legacy cellular networks This enables the applicationof scenarios such as 5G UDN and short-range C-I2X use cases The use of mmWave massiveMIMO in this ultra-dense domain has been the central focus of the investigation in this thesis

With the huge available bandwidth in the mmWave spectrum the aggressive spatial mul-tiplexing realizable with massive MIMO arrays and the expected high capacity gains from thedensely-deployed BSs or APs the amalgam of the three technologies can meet the projected

115

explosive user data demand and thereby support the 5G eMBB use case However despitethe potential of this architecture a number of challenges surface regarding system implemen-tation Some of these challenges are investigated in this thesis as highlighted hereunderSome others that are subject to future investigations are presented in the next section

511 Channel Modeling

The accurate characterization of the wireless channel is fundamental to the realistic evalua-tion of system performance Different from legacy systems 5G networks bring to the forefrontmany different new perspectives for the wireless channel modeling Moving to the mmWavespectrum introduces new dynamics to channel modeling Among the newly-introduced chal-lenges are the need to model and support massive MIMO 3D beamforming wide frequencyrange broad bandwidth smooth time evolution spatial and frequency consistency mobilitycoexistence and diverse use cases or scenarios Consequently in Chapter 3 of this thesis

bull We adopted the 3GPP 3D channel models for LTE (TR 36873 [84]) and mmWave (TR38900 [85]) frequencies as legacy 2D channel models have been shown to underesti-mate performance [31ndash34] and do not cater for future emerging 5G scenarios such assky-rise buildings and vehicular scenarios that require a 3D perspective These were im-plemented and form part of an integrated 5G SLS with enhanced functionalities basedon LTE-A SLS [38] Some of the added features include 3GPP 3D mmWave channelmodel (TR 38900 [85]) 5G NR frame structure [165] multi-tier and multi-frequencyHetNet inter-tier handover (leading to uneven cell loading) among others

bull We calibrated the channels using the appropriate metrics standards and referencesWith the evolved 5G simulator we characterized the individual performance of the 3DmicroWave and mmWave channels using the UMa and UMi scenarios in LOS NLOS andO2I propagation environments Focusing on the mmWave SC tier that is particularlyrequired to provide the anticipated capacity boost for 5G we investigated parameterssuch as BS downtilt and UE height that could impact system performance Our resultsshow that the mmWave channel was underwhelming in terms of expected multi-Gbpsdata rate due to SINR bottleneck particularly for indoor users served by outdoor BSsThe SINR statistics reveal that indoor users experience up to 30 dB additional lossesfrom wall and in-building objects It also reveals the degrading impact of the highernoise levels resulting from the larger bandwidths employed in mmWave systems

bull The performance of the joint use of the two channel models in a 5G UDN deploymentframework with microWave MCs and densely-deployed mmWave SCs was also evaluatedThis multi-tier scenario investigates the impact and interplay of the so-called ldquobig threerdquotechnologies for 5G UDN massive MIMO and mmWave communications The resultsshow that much higher capacity can be realized with UDNs than in MC-only set-upsThe results also reveal that performance does not scale proportionally with increasein the employed mmWave bandwidth The corresponding increase in noise (due tolarger bandwidths) reduces the SINR Outdoor users experience promising data ratesnotwithstanding but the throughputs of indoor users are highly degraded This is dueto the additional wall and indoor losses (on top of the inherently high PL at mmWavefrequencies) which further reduce the SINR of indoor users

116

bull In the last part of Chapter 3 we adopted the measurement-based 3D mmWave massiveMIMO channel-only simulator [42] and enhanced its capabilities for the investigation ofthe C-I2V scenario involving street-level lampost-based APs (or infrastructure or BS)and vehicles with roof-mount antennas as receivers We upgraded the channel simulatorby implementing blockage model spatial consistency mobility and advanced 5G featureswith respect to the 5G NR frame structure beamforming and scheduling Using theupgraded simulator we then fully characterized the mmWave massive MIMO vehicularchannel using metrics such as PL RMS-DS Rician KF cluster and ray distributionPDP channel rank channel condition number and data rate We also compared themmWave performance with the DSRC and LTE-A capabilities and offered useful insightson vehicular channels in such scenarios

Given the 3D channel implementation and characterization for 5G UDN and C-I2V scenar-ios the aggregate results indicate that outdoor users show promising performance in mmWave3D channels for 5G eMBB use cases On the other hand the additional wall and indoor losses(on top of the inherently high PL at mmWave frequencies) significantly degrade the perfor-mance of indoor users served by outdoor BSs Therefore overcoming the collective impactof the increasing noise higher PL and indoor losses remains a challenge for indoor users inthe mmWave tier of 5G HetNets where high rates are anticipated Thus the practical op-tion advocated in this thesis and supported by recent findings and deployment [35 53] is toemploy the UMa-microWave for coverage whilst using the UMi-mmWave for high-rate outdoorUDN users and serving indoor users using the indoor femtocells or WiFi APs SimilarlymmWave massive MIMO can deliver Gbps rates for supporting vehicular safety infotainmentand allied services for applications such as autonomous driving The mmWave access can bemade available by using the 5G cellular infrastructure dedicated mmWave V2XI2XI2V ormodified IEEE 80211ad unlicensed band as advocated for example by 3GPP Release 15 for5G-V2X [182]


Beamforming is required in mmWave massive MIMO systems to provide large array gainsto overcome the high PL enable highly-directional beams to mitigate ICI provide spatialmultiplexing gains to boost system capacity as well as facilitate the mitigation of MUI [136]With possibilities for ABF HBF and DBF architectures the HBF array structure has beenextensively demonstrated as a practically-feasible architecture for massive MIMO Consideringthe SE EE cost and hardware complexity the HBF approach strikes a balanced performancetrade-off when compared to the fully-analog and the fully-digital implementations Using theHBF architecture it is possible to realize three different subarray structures specificallythe fully-connected sub-connected and the overlapped subarray structures However nogeneralized framework exists for the comparative performance of these structures More sothe performance of HBF schemes in 5G UDN and short-range C-I2P and C-I2X scenarioshave not received adequate attention

Therefore in Chapter 4 of this thesis

bull We developed a generalized model for the design and analysis of any HBF array structureor configuration using mmWave massive MIMO We outlined in a step-wise mannerthe procedures for the design and then analyzed the performance of quintessential con-figurations comparatively using the C-I2X application scenario involving both cellular

117

and vehicular users A parameter known as the subarray spacing is introduced suchthat varying its value leads to the different subarray configurations and the consequentchanges in system performance

bull Using a realistic power consumption model we assessed the performance of the sys-tem not only in terms of the SE and EE but also the power and hardware cost forthe network Different from most existing works our comprehensive power consump-tion model includes the TX power TX circuit power RX circuit power as well as thebackhaul power Our results reveal that the backhaul power constitute the largest per-centage of the total power consumption in the 5G scenario as ultra-high data rates areexchanged between the TX and the core network This result is contrary to the sit-uation with relatively low-rate legacy networks where the TX power takes the largestchunk of the total power consumption

bull Moreover using the HBF structure we investigated the performance of the hybridprecoding with baseband zero forcing for MUI mitigation in C-I2P scenario and assessedits superiority over the analog-only beamsteering approach and how its performanceapproaches the SVD precoding as the upper bound The impacts of system parameterssuch as carrier frequency bandwidth antenna gain and antenna geometry on the SE-EEperformance of the network were also assessed

Thus Chapter 4 presented a novel generalized framework for the design and performanceanalysis of the different HBF architectures The objective of this work is two-fold generalizedframework and performance trade-off with respect to the existing solutions Firstly while thefully-connected and the sub-connected structures are popular and widely investigated theoverlapped subarray structure has not received significant attention Thus this work providesa generalized framework to realize any of the configurations for performance comparisonSecondly as the results show the overlapped subarray implementation maintains a balancedtrade-off in terms of SE EE complexity and hardware cost when compared to the popularfully-connected and the sub-connected structures The overlapped structure therefore offerspromising potential for 5G networks employing mmWave massive MIMO to deliver ultra-highdata rates whilst maintaining a balance in the EE of the network

52 Future Research Directions

As we march towards 2020 activities on 5G networks are moving from research field trialsand standardization to real deployments The 5G Phase 1 has been finalized 5G Phase 2has recently been defined by the 3GPP and the research on 6G networks is picking up paceSince 6G networks are expected to address the limitations and challenges of 5G and surpassits performance the future research directions presented in this thesis are in the following 6Gresearch areas

521 THz Channel Modeling

Spectrum use will undoubtedly move to the 03-10 THz bands in the 6G mobile systemera With enormous bandwidths far greater than the amount available in the sub-6 GHz andmmWave bands combined THz band communications (THzBC) will open up new frontiers forexciting services and applications requiring ultra-broadband (Tbps) connectivity To combat

118

high PL at THz frequencies directional and dynamic ultra-massive MIMO antennas areexpected to be used High directionality and dynamic ultra-massive MIMO antennas lead tonarrow beamwidth and very limited interference Thus very high data rate per area or perlink can be expected However there are also critical fundamental challenges for applyingTHz communications in mobile networks For example the channel modeling is still largelyunknown in the THz band [183] except those below 300 GHz for stationary indoor scenarios[184] Moreover ultra-high rates lead to ultra-high energy consumption Thus energy-efficient communications are needed on both the digital signal processing and radio interfacelevels

Since the future ultra-fast 6G THz network will be modeled in ultra-dense setups con-sisting of numerous hotspots stochastic modeling approaches have to be extended towardsthe 3D channel modeling to account for effects such as 3D beamforming in THz networksMoreover analytical validation would have to be complemented with real experimentationin order to assess the feasibility of the design within the 6G networking scenario that in-cludes practical models for characterizing the channel data traffic and mobility within amulti-user environment Modeling the impact of mobility and dual mobility for THz cellularand vehicular networks is still a fundamental challenge for the forthcoming 6G system Inaddition extension of the mmWave channel models to cover new and extreme applicationscenarios such as tunnel underground underwater human body molecular and unmannedaerial vehicle (UAV) communication is still largely missing or sparingly explored Moreover6G will integrate terrestrial airborne and satellite networks leading to future emerging usecases and extreme scenarios that will benefit from THzBC [12 100] How to exploit THzBCand performance analysis will be a topic for future research Also design and development ofintegrated multi-band transceivers that will facilitate the coexistence of microWave mmWave andTHz communication will be a fundamental component of 6G and is thus a research outlook

522 Ultra-massive MIMO

The extremely small size of THz antennas will enable ultra-massive MIMO (UM-MIMO)or extra-large scale massive MIMO (XL-MIMO) arrays with elements far greater than thenumber expected in 5G systems UM-MIMO is being explored as the practical technology tocombat the distance or range challenge in THz systems while offering amazing opportunitiesfor beamforming and spatial multiplexing in delivering ultra-high capacities for 6G systemsThe ultra-large arrays however bring about new set of challenges with respect to THz UM-MIMO fabrication channel modeling modulation waveform design and beamforming multi-carrier antenna configurations spatial modulation and other challenges across all layers ofthe protocol stack [151185ndash187]

In UM-MIMO or XL-MIMO the array dimension is pushed to the extreme For exampleXL-MIMO arrays can be integrated into large structures such as the walls of buildings in amegacity in airports large shopping malls or along the structure of a stadium and therebyserve a large number of user devices This use case is considered a distinct operating regime ofmassive MIMO and comes with its own challenges and opportunities [188] The prospectiveuse cases that would be the subject of 6G research activities include the use of THz UM-MIMO for communication and sensing on-chipchip-to-chip communication and data centersand other cellular vehicular biological and molecular applications [185189]

119

523 6G for Energy Efficiency

Hardware cost complexity and power consumption have largely limited 5G antenna andbeamforming designs to the hybrid architectures which have reduced flexibility and rate whencompared to the fully-digital implementation However the development trends show thatthe cost and power consumption of fully-digital transceivers can be reduced in the futurethereby making digital precoding a good choice for spectral- and energy-efficient 6G systemimplementation [53 57] Another technology being considered for EE optimization in 6Gsystems is wireless communication with reconfigurable intelligent surfaces (RISs) or largeintelligent surfaces (LISs) for centralized distributed or cell-free UM-MIMO systems [12188190191]

LISs generically denote active large electromagnetic surfaces that possesses communicationcapabilities The walls of buildings room and factory celings laptop cases and human cloth-ing among others can be used as intelligent surfaces for smart environment in 6G They willbe equipped with metamaterial-based antennas programmable metasurfaces fluid antennasor software-defined material for wireless communication [11] RIS refers to a meta-surfaceequipped with integrated electronic circuits that can be programmed to alter an incomingelectromagnetic field in a customizable way It can be readily fabricated using lithographyand nano-printing methods and is attractive from an energy consumption standpoint as itamplifies and forwards the incoming signals without employing any power amplifier therebyconsuming less energy than legacy transceiver designs [190 191] In harnessing the benefitsof LISRIS technology new transceiver designs are being considered for energy-efficient 6Gnetworks and are thus continuing topics for future research

524 Quantum Machine Learning

An ultra-massive and complex networking scenario is foreseen for 6G systems from extra-large scale MIMO and ultra-wideband spectrum to ultra-dense small and tiny cells LIS andmassive IoTinternet of everything (IoE) deployment The anticipated huge scale of data nodoubt necessitates the use of artificial intelligence (AI) tools for intelligent adaptive learningaccurate prediction and reliable decision making for efficient network management Someof the areas where AI would be applied include channel estimationdetection radio re-source management energy modeling cellchannel selection association location predictionbeam tracking and management celluser clustering switch and handover among HetNetsspectrum sensingaccess signal dimension reduction user behavior analysis mobility man-agement intrusionfaultanomaly detection complexity reduction and system optimization[9 192193]

Due to the advances in AI techniques especially deep learning and the availability ofmassive training data the interest in using AI for the design and optimization of wirelessnetworks has significantly increased and it is widely accepted that AI will be at the heartof 6G [11] Machine learning (ML) as a subbranch of AI enables machines to learn per-form and improve their operations by exploiting the operational knowledge and experiencegained in the form of data ML (via supervised unsupervised or reinforcement learning) canpotentially assist big data analytics to realize self-sustaining proactive and efficient wirelessnetworks Also the tremendous potential of parallelism offered by quantum computing (QC)and related quantum technologies over classical computing paradigms is further enabling MLQuantum machine learning (QML) has therefore emerged as a technology paradigm to ad-

120

dress the evolving challenges of the increasing human and machine interconnectivity big dataautonomous management and self-organization demands in 6G networks [9]

In summary the above-named subjects constitute the principal future research directionsas a natural evolution from the 5G technology enablers investigated in this thesis whosesolution can be considered the first building block of 6G Like the 5G enablers these 6Gconcepts are inter-connected THz spectrum enables UM-MIMO and their joint use demandsQML in the foreseen complex communication scenarios (cellular vehicular etc) where EEwould be a critical performance metric Therefore the interplay of these technologies togetherwith the associated issues of use cases standardization health and safety business model andperformance optimization remain interesting open issues that are subjects for future works6G would be a stimulating research area with both evolutionary and revolutionary paradigmsthat would deliver innovative solutions for NGMNs Finally we remark that the open issuesand future research directions presented in this chapter were the results of the surveys thatwere published in IEEE Network11 and IEEE Communications Surveys and Tutorials12

11K M S Huq S A Busari J Rodriguez V Frascolla W Buzzi and D C Sicker ldquoTerahertz-enabledWireless System for Beyond-5G Ultra-Fast Network A Brief Surveyrdquo IEEE Network vol 33 no 4 pp89-95 JulyAugust 2019


121

Bibliography

[1] ITU-R ldquoIMT vision - Framework and overall objectives of the future developmentof IMT for 2020 and beyond - Recommendation ITU-R M2083-0rdquo ITU-R GenevaSwitzerland Tech Rep Sep 2015 last accessed 26-02-2020 [Online] Availablehttpswwwituintdms pubrecitu-rrecmR-REC-M2083-0-201509-IPDF-Epdf

[2] S Mumtaz J Rodriguez and L Dai ldquoIntroduction to mmWave massive MIMOrdquo inmmWave Massive MIMO A Paradigm for 5G S Mumtaz J Rodriguez and L DaiEds Academic Press 2017 pp 1ndash18

[3] M Shafi A F Molisch P J Smith T Haustein P Zhu P D Silva F Tufves-son A Benjebbour and G Wunder ldquo5G A Tutorial Overview of Standards TrialsChallenges Deployment and Practicerdquo IEEE Journal on Selected Areas in Communi-cations vol 35 no 6 pp 1201ndash1221 June 2017

[4] M Xiao S Mumtaz Y Huang L Dai Y Li M Matthaiou G K KaragiannidisE Bjornson K Yang I Chih-Lin and A Ghosh ldquoMillimeter Wave Communicationsfor Future Mobile Networksrdquo IEEE Journal on Selected Areas in Communicationsvol 35 no 9 pp 1909ndash1935 Sep 2017

[5] ITU ldquoMinimum Requirements Related to Technical Performance for IMT-2020 Radio Interface(s) document ITU-R M [IMT-2020TECH PERF REQ]rdquoITU Tech Rep Oct 2016 last accessed 26-02-2020 [Online] Availablehttpswwwituintdms pubitu-ropbrepR-REP-M2410-2017-PDF-Epdf

[6] F B Saghezchi et al Drivers for 5G The rsquoPervasive Connected Worldrsquo John Wileyamp Sons Ltd UK 2015 chapter 1 pp 1ndash27 in Rodriguez J (Ed) Fundamentals of5G Mobile Networks

[7] T Chen M Matinmikko X Chen X Zhou and P Ahokangas ldquoSoftware definedmobile networks concept survey and research directionsrdquo IEEE CommunicationsMagazine vol 53 no 11 pp 126ndash133 Nov 2015

[8] A Gupta and R K Jha ldquoA Survey of 5G Network Architecture and Emerging Tech-nologiesrdquo IEEE Access vol 3 pp 1206ndash1232 2015

[9] S J Nawaz S K Sharma S Wyne M N Patwary and M Asaduzzaman ldquoQuantumMachine Learning for 6G Communication Networks State-of-the-Art and Vision forthe Futurerdquo IEEE Access vol 7 pp 46 317ndash46 350 2019

[10] ldquo5G vision requirements and enabling technologiesrdquo 5G South Korea South KoreaTech Rep Mar 2016

122

[11] F Tariq M R A Khandaker K Wong M A Imran M Bennis and M DebbahldquoA Speculative Study on 6Grdquo CoRR vol abs190206700 2019 [Online] Availablehttparxivorgabs190206700

[12] W Saad M Bennis and M Chen ldquoA Vision of 6G Wireless Systems ApplicationsTrends Technologies and Open Research Problemsrdquo IEEE Network pp 1ndash9 2019

[13] E Calvanese Strinati S Barbarossa J L Gonzalez-Jimenez D Ktenas N CassiauL Maret and C Dehos ldquo6G The Next Frontier From Holographic Messaging toArtificial Intelligence Using Subterahertz and Visible Light Communicationrdquo IEEEVehicular Technology Magazine vol 14 no 3 pp 42ndash50 Sep 2019

[14] I F Akyildiz S Nie S-C Lin and M Chandrasekaran ldquo5G roadmap 10 key enablingtechnologiesrdquo Computer Networks vol 106 pp 17ndash48 2016

[15] S Vahid R Tafazolli and M Filo Small Cells for 5G Mobile Networks John Wileyamp Sons Ltd UK 2015 chapter 3 pp 63ndash104 in Rodriguez J (Ed) Fundamentals of5G Mobile Networks

[16] J G Andrews S Buzzi W Choi S V Hanly A Lozano A C K Soong and J CZhang ldquoWhat Will 5G Berdquo IEEE Journal on Selected Areas in Communicationsvol 32 no 6 pp 1065ndash1082 June 2014

[17] F Khan and Z Pi ldquommWave mobile broadband (MMB) Unleashing the 3300GHzspectrumrdquo in 34th IEEE Sarnoff Symposium May 2011 pp 1ndash6

[18] V Va T Shimizu G Bansal and R W H Jr ldquoMillimeter Wave Vehicular Commu-nications A Surveyrdquo Foundations and Trends in Networking vol 10 no 1 pp 1ndash1132016

[19] W H Chin Z Fan and R Haines ldquoEmerging technologies and research challengesfor 5G wireless networksrdquo IEEE Wireless Communications vol 21 no 2 pp 106ndash112April 2014

[20] E G Larsson O Edfors F Tufvesson and T L Marzetta ldquoMassive MIMO for nextgeneration wireless systemsrdquo IEEE Communications Magazine vol 52 no 2 pp 186ndash195 February 2014

[21] T L Marzetta ldquoNoncooperative Cellular Wireless with Unlimited Numbers of BaseStation Antennasrdquo IEEE Transactions on Wireless Communications vol 9 no 11pp 3590ndash3600 November 2010

[22] F Rusek D Persson B K Lau E G Larsson T L Marzetta O Edfors andF Tufvesson ldquoScaling Up MIMO Opportunities and Challenges with Very Large Ar-raysrdquo IEEE Signal Processing Magazine vol 30 no 1 pp 40ndash60 Jan 2013

[23] Q C Li H Niu A T Papathanassiou and G Wu ldquo5G Network Capacity KeyElements and Technologiesrdquo IEEE Vehicular Technology Magazine vol 9 no 1 pp71ndash78 March 2014

123

[24] Z S Bojkovic M R Bakmaz and B M Bakmaz ldquoResearch challenges for 5G cellulararchitecturerdquo in 2015 12th International Conference on Telecommunication in ModernSatellite Cable and Broadcasting Services (TELSIKS) Oct 2015 pp 215ndash222

[25] D Feng C Jiang G Lim L J Cimini G Feng and G Y Li ldquoA survey of energy-efficient wireless communicationsrdquo IEEE Communications Surveys amp Tutorials vol 15no 1 pp 167ndash178 2013

[26] E Hossain M Rasti H Tabassum and A Abdelnasser ldquoEvolution toward 5G multi-tier cellular wireless networks An interference management perspectiverdquo IEEE Wire-less Communications vol 21 no 3 pp 118ndash127 June 2014

[27] ldquoEricsson Mobility Report November 2018rdquo Ericsson Tech Rep Nov 2018 lastaccessed 26-02-2020 [Online] Available httpswwwericssoncomassetslocalmobility-reportdocuments2018ericsson-mobility-report-november-2018pdf

[28] Ericsson Traffic Exploration Tool Online Ericsson Accessed 27-02-2020 [Online]Available httpwwwericssoncomTETtrafficViewloadBasicEditorericsson

[29] E Bjornson J Hoydis and L Sanguinetti ldquoMassive MIMO Networks Spectral En-ergy and Hardware Efficiencyrdquo Foundations and Trends Rcopy in Signal Processing vol 11no 3-4 pp 154ndash655 2017

[30] W Liu and Z Wang ldquoNon-Uniform Full-Dimension MIMO New Topologies and Op-portunitiesrdquo IEEE Wireless Communications vol 26 no 2 pp 124ndash132 April 2019

[31] A Kammoun H Khanfir Z Altman M Debbah and M Kamoun ldquoPreliminaryResults on 3D Channel Modeling From Theory to Standardizationrdquo IEEE Journal onSelected Areas in Communications vol 32 no 6 pp 1219ndash1229 June 2014

[32] F Ademaj M Taranetz and M Rupp ldquo3GPP 3D MIMO channel model a holis-tic implementation guideline for open source simulation toolsrdquo EURASIP Journal onWireless Communications and Networking vol 2016 no 1 p 55 Feb 2016

[33] R N Almesaeed A S Ameen E Mellios A Doufexi and A Nix ldquo3D ChannelModels Principles Characteristics and System Implicationsrdquo IEEE CommunicationsMagazine vol 55 no 4 pp 152ndash159 April 2017

[34] F Ademaj M Taranetz and M Rupp ldquoImplementation validation and applicationof the 3GPP 3D MIMO channel model in open source simulation toolsrdquo in 2015 In-ternational Symposium on Wireless Communication Systems (ISWCS) Aug 2015 pp721ndash725

[35] F Kronestedt H Asplund A Furuskar D H Kang M Lundevall and K WallstedtldquoThe advantages of combining 5G NR with LTErdquo Ericsson Tech Rep Nov2018 last accessed 26-02-2020 [Online] Available httpswwwericssoncomenericsson-technology-reviewarchive2018the-advantages-of-combining-5g-nr-with-lte

[36] C-L I ldquoSeven fundamental rethinking for next-generation wireless communicationsrdquoAPSIPA Transactions on Signal and Information Processing vol 6 p e10 2017

124

[37] K M S Huq S Mumtaz J Bachmatiuk J Rodriguez X Wang and R L AguiarldquoGreen HetNet CoMP Energy Efficiency Analysis and Optimizationrdquo IEEE Transac-tions on Vehicular Technology vol 64 no 10 pp 4670ndash4683 Oct 2015

[38] M Rupp S Schwarz and M Taranetz The Vienna LTE-Advanced Simulators Up andDownlink Link and System Level Simulation 1st ed ser Signals and CommunicationTechnology Springer Singapore 2016

[39] M K Muller F Ademaj T Dittrich A Fastenbauer B Ramos Elbal A NabaviL Nagel S Schwarz and M Rupp ldquoFlexible multi-node simulation of cellular mo-bile communications the Vienna 5G System Level Simulatorrdquo EURASIP Journal onWireless Communications and Networking vol 2018 no 1 p 227 Sep 2018

[40] S Pratschner B Tahir L Marijanovic M Mussbah K Kirev R Nissel S Schwarzand M Rupp ldquoVersatile mobile communications simulation the Vienna 5G Link LevelSimulatorrdquo EURASIP Journal on Wireless Communications and Networking vol 2018no 1 p 226 Sep 2018

[41] P Alvarez C Galiotto J van de Belt D Finn H Ahmadi and L DaSilva ldquoSimulatingdense small cell networksrdquo in 2016 IEEE Wireless Communications and NetworkingConference April 2016 pp 1ndash6

[42] New York University ldquoNYUSIM v16rdquo 2017 last ac-cessed 26-02-2020 [Online] Available httpwirelessengineeringnyuedu5g-millimeter-wave-channel-modeling-software

[43] S Chen X Ji C Xing Z Fei and H Wang ldquoSystem-level performance evaluationof ultra-dense networks for 5Grdquo in TENCON 2015 IEEE Region 10 Conference 2015pp 1ndash4

[44] X Ge S Tu G Mao C X Wang and T Han ldquo5G Ultra-Dense Cellular NetworksrdquoIEEE Wireless Communications vol 23 no 1 pp 72ndash79 2016

[45] L Lu G Y Li A L Swindlehurst A Ashikhmin and R Zhang ldquoAn Overview ofMassive MIMO Benefits and Challengesrdquo IEEE Journal of Selected Topics in SignalProcessing vol 8 no 5 pp 742ndash758 Oct 2014

[46] L G Ordez D P Palomar and J R Fonollosa ldquoFundamental diversity multiplexingand array gain tradeoff under different MIMO channel modelsrdquo in 2011 IEEE Inter-national Conference on Acoustics Speech and Signal Processing (ICASSP) May 2011pp 3252ndash3255

[47] S Mattigiri and C Warty ldquoA study of fundamental limitations of small antennasMIMO approachrdquo in 2013 IEEE Aerospace Conference March 2013 pp 1ndash8

[48] F Khan LTE for 4G Mobile Broadband Air interface technologies and performanceCambridge University Press New York USA 2009

[49] A Goldsmith Wireless Communications Cambridge University Press CambridgeUnited Kingdom 2005

125

[50] Q Li G Li W Lee M Lee D Mazzarese B Clerckx and Z Li ldquoMIMO techniquesin WiMAX and LTE a feature overviewrdquo IEEE Communications Magazine vol 48no 5 pp 86ndash92 May 2010

[51] H Inanolu ldquoMultiple-Input Multiple-Output System Capacity Antenna and Propaga-tion Aspectsrdquo IEEE Antennas and Propagation Magazine vol 55 no 1 pp 253ndash273Feb 2013

[52] S Wang Y Xin S Chen W Zhang and C Wang ldquoEnhancing spectral efficiency forlte-advanced and beyond cellular networks [Guest Editorial]rdquo IEEE Wireless Commu-nications vol 21 no 2 pp 8ndash9 April 2014

[53] E Bjornson L Van der Perre S Buzzi and E G Larsson ldquoMassive MIMO in Sub-6 GHz and mmWave Physical Practical and Use-Case Differencesrdquo IEEE WirelessCommunications vol 26 no 2 pp 100ndash108 April 2019

[54] J Lota S Sun T S Rappaport and A Demosthenous ldquo5G Uniform Linear ArraysWith Beamforming and Spatial Multiplexing at 28 37 64 and 71 GHz for Outdoor Ur-ban Communication A Two-Level Approachrdquo IEEE Transactions on Vehicular Tech-nology vol 66 no 11 pp 9972ndash9985 Nov 2017

[55] S Sun T S Rappaport R W Heath A Nix and S Rangan ldquoMimo for millimeter-wave wireless communications beamforming spatial multiplexing or bothrdquo IEEECommunications Magazine vol 52 no 12 pp 110ndash121 December 2014

[56] G Liu X Hou J Jin F Wang Q Wang Y Hao Y Huang X Wang X Xiao andA Deng ldquo3-D-MIMO With Massive Antennas Paves the Way to 5G Enhanced MobileBroadband From System Design to Field Trialsrdquo IEEE Journal on Selected Areas inCommunications vol 35 no 6 pp 1222ndash1233 June 2017

[57] B Yang Z Yu J Lan R Zhang J Zhou and W Hong ldquoDigital Beamforming-Based Massive MIMO Transceiver for 5G Millimeter-Wave Communicationsrdquo IEEETransactions on Microwave Theory and Techniques vol 66 no 7 pp 3403ndash3418 July2018

[58] C Chen V Volski L Van der Perre G A E Vandenbosch and S Pollin ldquoFiniteLarge Antenna Arrays for Massive MIMO Characterization and System Impactrdquo IEEETransactions on Antennas and Propagation vol 65 no 12 pp 6712ndash6720 Dec 2017

[59] S Mumtaz J Rodriguez and L Dai Eds mmWave massive MIMO A Paradigm for5G Academic Press UK 2017

[60] T S Rappaport S Sun R Mayzus H Zhao Y Azar K Wang G N Wong J KSchulz M Samimi and F Gutierrez ldquoMillimeter Wave Mobile Communications for5G Cellular It Will Workrdquo IEEE Access vol 1 pp 335ndash349 2013

[61] ITU-R (2015 Jun) Technology trends of active services in the frequencyrange 275-3000 GHz Online Last accessed 05-03-2020 [Online] AvailablehttpswwwituintpubR-REP-SM2352

126

[62] L Zakrajsek D Pados and J M Jornet ldquoDesign and performance analysis of ultra-massive multi-carrier multiple input multiple output communication in the terahertzbandrdquo in Proc SPIE vol 10209 Apr 2017 pp 1ndash12

[63] L Wei R Q Hu Y Qian and G Wu ldquoKey elements to enable millimeter wavecommunications for 5G wireless systemsrdquo IEEE Wireless Communications vol 21no 6 pp 136ndash143 December 2014

[64] S Hur T Kim D J Love J V Krogmeier T A Thomas and A Ghosh ldquoMillimeterWave Beamforming for Wireless Backhaul and Access in Small Cell Networksrdquo IEEETransactions on Communications vol 61 no 10 pp 4391ndash4403 October 2013

[65] H Shokri-Ghadikolaei C Fischione P Popovski and M Zorzi ldquoDesign aspects ofshort-range millimeter-wave networks A MAC layer perspectiverdquo IEEE Networkvol 30 no 3 pp 88ndash96 May 2016

[66] T S Rappaport G R MacCartney M K Samimi and S Sun ldquoWideband Millimeter-Wave Propagation Measurements and Channel Models for Future Wireless Communi-cation System Designrdquo IEEE Transactions on Communications vol 63 no 9 pp3029ndash3056 Sep 2015

[67] Z Pi and F Khan ldquoAn introduction to millimeter-wave mobile broadband systemsrdquoIEEE Communications Magazine vol 49 no 6 pp 101ndash107 June 2011

[68] A L Swindlehurst E Ayanoglu P Heydari and F Capolino ldquoMillimeter-wave mas-sive MIMO the next wireless revolutionrdquo IEEE Communications Magazine vol 52no 9 pp 56ndash62 Sep 2014

[69] M Ding D Lopez-Perez H Claussen and M A Kaafar ldquoOn the Fundamental Char-acteristics of Ultra-Dense Small Cell Networksrdquo IEEE Network vol 32 no 3 pp92ndash100 May 2018

[70] D Lopez-Perez M Ding H Claussen and A H Jafari ldquoTowards 1 GbpsUE inCellular Systems Understanding Ultra-Dense Small Cell Deploymentsrdquo IEEE Com-munications Surveys Tutorials vol 17 no 4 pp 2078ndash2101 Fourthquarter 2015

[71] S Buzzi and C DrsquoAndrea ldquoDoubly Massive mmWave MIMO Systems Using VeryLarge Antenna Arrays at Both Transmitter and Receiverrdquo in 2016 IEEE Global Com-munications Conference (GLOBECOM) Dec 2016 pp 1ndash6

[72] S Buzzi and C DAndrea ldquoThe doubly massive MIMO regime in mmWavecommunicationsrdquo Sep 2016 proc Tyrrhenian Int Workshop Digit Communpp 1-28 [Online] Available Availablehttptyrr2016cnitittyrr-contentuploadsThe-doubly-massive-MIMOregime-in-mmWave DAndrea TIW16pdf

[73] V Ezzati M Fakharzadeh F Farzaneh and M R Naeini ldquoEffect of Antenna Cou-pling on the SNR Improvement of Mm-Wave Massive MIMO for 5Grdquo in 2019 IEEEInternational Symposium on Antennas and Propagation and USNC-URSI Radio ScienceMeeting July 2019 pp 417ndash418

127

[74] J A Zhang X Huang V Dyadyuk and Y J Guo ldquoHybrid antenna array for mmWavemassive MIMOrdquo in mmWave Massive MIMO A Paradigm for 5G S Mumtaz J Ro-driguez and L Dai Eds Academic Press 2017 pp 39 ndash 61

[75] V Petrov D Solomitckii A Samuylov M A Lema M Gapeyenko D MoltchanovS Andreev V Naumov K Samouylov M Dohler and Y Koucheryavy ldquoDynamicMulti-Connectivity Performance in Ultra-Dense Urban mmWave Deploymentsrdquo IEEEJournal on Selected Areas in Communications vol 35 no 9 pp 2038ndash2055 Sep 2017

[76] M R Akdeniz Y Liu M K Samimi S Sun S Rangan T S Rappaport and E ErkipldquoMillimeter Wave Channel Modeling and Cellular Capacity Evaluationrdquo IEEE Journalon Selected Areas in Communications vol 32 no 6 pp 1164ndash1179 June 2014

[77] Z Shi R Lu J Chen and X S Shen ldquoThree-dimensional spatial multiplexing fordirectional millimeter-wave communications in multi-cubicle office environmentsrdquo in2013 IEEE Global Communications Conference (GLOBECOM) Dec 2013 pp 4384ndash4389

[78] C Kourogiorgas S Sagkriotis and A D Panagopoulos ldquoCoverage and outage capacityevaluation in 5G millimeter wave cellular systems impact of rain attenuationrdquo in 20159th European Conference on Antennas and Propagation (EuCAP) April 2015 pp 1ndash5

[79] T S Rappaport J N Murdock and F Gutierrez ldquoState of the Art in 60-GHz Inte-grated Circuits and Systems for Wireless Communicationsrdquo Proceedings of the IEEEvol 99 no 8 pp 1390ndash1436 Aug 2011

[80] Z Qingling and J Li ldquoRain Attenuation in Millimeter Wave Rangesrdquo in 2006 7thInternational Symposium on Antennas Propagation EM Theory Oct 2006 pp 1ndash4

[81] S Joshi and S Sancheti ldquoFoliage loss measurements of tropical trees at 35 GHzrdquo in2008 International Conference on Recent Advances in Microwave Theory and Applica-tions Nov 2008 pp 531ndash532

[82] S A Hosseini E Zarepour M Hassan and C T Chou ldquoAnalyzing diurnal variationsof millimeter wave channelsrdquo in 2016 IEEE Conference on Computer CommunicationsWorkshops (INFOCOM WKSHPS) April 2016 pp 377ndash382

[83] T E Bogale X Wang and L B Le ldquommWave communication enabling techniquesfor 5G wireless systems A link level perspectiverdquo in mmWave Massive MIMO AParadigm for 5G S Mumtaz J Rodriguez and L Dai Eds Academic Press 2017pp 195 ndash 225

[84] ldquoStudy on 3D channel model for LTE v1220 3GPP TR 36873rdquo 3GPP Tech Rep2014 last accessed 26-02-2020 [Online] Available httpwww3gpporgftpSpecsarchive36 series36873

[85] ldquoStudy on channel model for frequency spectrum above 6 GHz v1420 3GPP TR38900rdquo 3GPP Tech Rep 2016 last accessed 26-02-2020 [Online] Availablehttpwww3gpporgftpSpecsarchive38 series38900

128

[86] T Wu T S Rappaport and C M Collins ldquoSafe for Generations to Come Consider-ations of Safety for Millimeter Waves in Wireless Communicationsrdquo IEEE MicrowaveMagazine vol 16 no 2 pp 65ndash84 March 2015

[87] mdashmdash ldquoThe human body and millimeter-wave wireless communication systems Inter-actions and implicationsrdquo in 2015 IEEE International Conference on Communications(ICC) June 2015 pp 2423ndash2429

[88] ldquoEuropean 7th Framework Programme Project MiWEBArdquo MiWEBA Tech RepJan 2018 [Online] Available Availablehttpwwwmiwebaeu

[89] K Sakaguchi G K Tran H Shimodaira S Nanba T Sakurai K Takinami I SiaudE C Strinati A Capone I Karls R Arefi and T Haustein ldquoMillimeter-Wave Evo-lution for 5G Cellular Networksrdquo IEICE Transactions on Communications vol E98Bno 3 pp 388ndash402 2015

[90] ldquoEuropean 7 th Framework Programme Project MiWaveSrdquo MiWaveS Tech RepJan 2018 [Online] Available Availablehttpwwwmiwaveseu

[91] V Frascolla M Faerber E C Strinati L Dussopt V Kotzsch E Ohlmer M ShariatJ Putkonen and G Romano ldquoMmWave use cases and prototyping A way towards5G standardizationrdquo in 2015 European Conference on Networks and Communications(EuCNC) June 2015 pp 128ndash132

[92] H Shokri-Ghadikolaei C Fischione G Fodor P Popovski and M Zorzi ldquoMillimeterWave Cellular Networks A MAC Layer Perspectiverdquo IEEE Transactions on Commu-nications vol 63 no 10 pp 3437ndash3458 Oct 2015

[93] ldquoUnderstanding 3GPP Release 12 Standards for HSPA+ and LTE EnhancementsrdquoBellevue WA USA Tech Rep Feb 2015 last accessed 26-02-2020 [Online]Available Availablehttpwww5gamericasorgfiles6614235904574G Americas- 3GPP Release 12 Executive Summary - February 2015pdf

[94] S L H Nguyen K Haneda J Jarvelainen A Karttunen and J Putkonen ldquoOn theMutual Orthogonality of Millimeter-Wave Massive MIMO Channelsrdquo in 2015 IEEE81st Vehicular Technology Conference (VTC Spring) May 2015 pp 1ndash5

[95] K Zheng S Ou and X Yin ldquoMassive MIMO Channel Models A Surveyrdquo Interna-tional Journal of Antennas and Propagation vol 2014 no 848071 p 10 pages 2014

[96] J M Jornet and I F Akyildiz ldquoChannel Modeling and Capacity Analysis for Elec-tromagnetic Wireless Nanonetworks in the Terahertz Bandrdquo IEEE Transactions onWireless Communications vol 10 no 10 pp 3211ndash3221 October 2011

[97] T Kurner and S Priebe ldquoTowards THz Communications - Status in Research Stan-dardization and Regulationrdquo Journal of Infrared Millimeter and Terahertz Wavesvol 35 no 1 pp 53ndash62 Jan 2014

[98] R Piesiewicz C Jansen D Mittleman T Kleine-Ostmann M Koch and T KurnerldquoScattering Analysis for the Modeling of THz Communication Systemsrdquo IEEE Trans-actions on Antennas and Propagation vol 55 no 11 pp 3002ndash3009 Nov 2007

129

[99] M Jacob S Priebe R Dickhoff T Kleine-Ostmann T Schrader and T KurnerldquoDiffraction in mm and Sub-mm Wave Indoor Propagation Channelsrdquo IEEE Transac-tions on Microwave Theory and Techniques vol 60 no 3 pp 833ndash844 March 2012

[100] C Wang J Bian J Sun W Zhang and M Zhang ldquoA Survey of 5G Channel Mea-surements and Modelsrdquo IEEE Communications Surveys Tutorials vol 20 no 4 pp3142ndash3168 Fourthquarter 2018

[101] M K Samimi and T S Rappaport ldquo3-D Millimeter-Wave Statistical Channel Modelfor 5G Wireless System Designrdquo IEEE Transactions on Microwave Theory and Tech-niques vol 64 no 7 pp 2207ndash2225 July 2016

[102] S Sun G R MacCartney and T S Rappaport ldquoA novel millimeter-wave channel sim-ulator and applications for 5G wireless communicationsrdquo in 2017 IEEE InternationalConference on Communications (ICC) May 2017 pp 1ndash7

[103] ldquoStudy on channel model for frequencies from 05 to 100 GHz v1430 3GPP TR38901rdquo 3GPP Tech Rep Dec 2017 last accessed 26-02-2020 [Online] Availablehttpwww3gpporgftpSpecsarchive38 series38901[lastaccessed14Jan2019]

[104] S Jaeckel L Raschkowski K Brner and L Thiele ldquoQuaDRiGa A 3-D Multi-CellChannel Model With Time Evolution for Enabling Virtual Field Trialsrdquo IEEE Trans-actions on Antennas and Propagation vol 62 no 6 pp 3242ndash3256 2014

[105] S Jaeckel M Peter K Sakaguchi W Keusgen and J Medbo ldquo5G Channel Modelsin mm-Wave Frequency Bandsrdquo in European Wireless 2016 22nd European WirelessConference 2016 pp 1ndash6

[106] ldquoH2020-ICT-671650-mmMAGICD22 v1 Measurement results and final mmMAGICchannel modelsrdquo Tech Rep Dec 2017

[107] A Ghosh ldquo5G Channel Model for Bands up to 100 GHzrdquo San Diego CA USA TechRep Dec 2015 [Online] Available Availablehttpwww5gworkshopscom5GCMhtml

[108] S Wu C Wang e M Aggoune M M Alwakeel and X You ldquoA General 3-DNon-Stationary 5G Wireless Channel Modelrdquo IEEE Transactions on Communicationsvol 66 no 7 pp 3065ndash3078 July 2018

[109] METIS ldquoMETIS channel models deliverable 14 version 3rdquo Sweden Tech Rep Jul2015

[110] L Liu C Oestges J Poutanen K Haneda P Vainikainen F Quitin F Tufvessonand P D Doncker ldquoThe COST 2100 MIMO channel modelrdquo IEEE Wireless Commu-nications vol 19 no 6 pp 92ndash99 December 2012

[111] ITU-R ldquoPreliminary draft new report ITU-R M [IMT-2020EVAL] R15-WP5D-170613-TD-0332rdquo Tech Rep Jun 2017

[112] B Ai K Guan G Li and S Mumtaz ldquommWave massive MIMO channel modelingrdquoin mmWave Massive MIMO A Paradigm for 5G S Mumtaz J Rodriguez and L DaiEds Academic Press 2017 pp 169 ndash 194

130

[113] I F Akyildiz J M Jornet and C Han ldquoTeraNets ultra-broadband communicationnetworks in the terahertz bandrdquo IEEE Wireless Communications vol 21 no 4 pp130ndash135 August 2014

[114] S Priebe M Jacob and T Krner ldquoCalibrated broadband ray tracing for the simulationof wave propagation in mm and sub-mm wave indoor communication channelsrdquo inEuropean Wireless 2012 18th European Wireless Conference 2012 April 2012 pp 1ndash10

[115] B Ai K Guan R He J Li G Li D He Z Zhong and K M S Huq ldquoOn In-door Millimeter Wave Massive MIMO Channels Measurement and Simulationrdquo IEEEJournal on Selected Areas in Communications vol 35 no 7 pp 1678ndash1690 July 2017

[116] X Zhao S Li Q Wang M Wang S Sun and W Hong ldquoChannel MeasurementsModeling Simulation and Validation at 32 GHz in Outdoor Microcells for 5G RadioSystemsrdquo IEEE Access vol 5 pp 1062ndash1072 2017

[117] N A Muhammad P Wang Y Li and B Vucetic ldquoAnalytical Model for OutdoorMillimeter Wave Channels Using Geometry-Based Stochastic Approachrdquo IEEE Trans-actions on Vehicular Technology vol 66 no 2 pp 912ndash926 Feb 2017

[118] M K Samimi G R MacCartney S Sun and T S Rappaport ldquo28 GHz Millimeter-Wave Ultrawideband Small-Scale Fading Models in Wireless Channelsrdquo in 2016 IEEE83rd Vehicular Technology Conference (VTC Spring) May 2016 pp 1ndash6

[119] M K Samimi and T S Rappaport ldquoLocal multipath model parameters for generat-ing 5G millimeter-wave 3GPP-like channel impulse responserdquo in 2016 10th EuropeanConference on Antennas and Propagation (EuCAP) April 2016 pp 1ndash5

[120] M K Samimi S Sun and T S Rappaport ldquoMIMO channel modeling and capacityanalysis for 5G millimeter-wave wireless systemsrdquo in 2016 10th European Conferenceon Antennas and Propagation (EuCAP) April 2016 pp 1ndash5

[121] M K Samimi and T S Rappaport ldquoStatistical Channel Model with Multi-Frequencyand Arbitrary Antenna Beamwidth for Millimeter-Wave Outdoor Communicationsrdquo in2015 IEEE Globecom Workshops (GC Wkshps) Dec 2015 pp 1ndash7

[122] A Torabi S A Zekavat and A Al-Rasheed ldquoMillimeter wave directional channelmodelingrdquo in 2015 IEEE International Conference on Wireless for Space and ExtremeEnvironments (WiSEE) Dec 2015 pp 1ndash6

[123] S Hur S Baek B Kim Y Chang A F Molisch T S Rappaport K Haneda andJ Park ldquoProposal on Millimeter-Wave Channel Modeling for 5G Cellular SystemrdquoIEEE Journal of Selected Topics in Signal Processing vol 10 no 3 pp 454ndash469 April2016

[124] T Bai A Alkhateeb and R W Heath ldquoCoverage and capacity of millimeter-wavecellular networksrdquo IEEE Communications Magazine vol 52 no 9 pp 70ndash77 Sep2014

131

[125] O E Ayach S Rajagopal S Abu-Surra Z Pi and R W Heath ldquoSpatially SparsePrecoding in Millimeter Wave MIMO Systemsrdquo IEEE Transactions on Wireless Com-munications vol 13 no 3 pp 1499ndash1513 March 2014

[126] A Alkhateeb O E Ayach G Leus and R W Heath ldquoChannel Estimation and HybridPrecoding for Millimeter Wave Cellular Systemsrdquo IEEE Journal of Selected Topics inSignal Processing vol 8 no 5 pp 831ndash846 Oct 2014

[127] X Gao L Dai Z Gao T Xie and Z Wang ldquoPrecoding for mmWave massive MIMOrdquoin mmWave Massive MIMO A Paradigm for 5G S Mumtaz J Rodriguez and L DaiEds Academic Press 2017 pp 79 ndash 111

[128] I Ahmed H Khammari A Shahid A Musa K S Kim E De Poorter and I Moer-man ldquoA Survey on Hybrid Beamforming Techniques in 5G Architecture and SystemModel Perspectivesrdquo IEEE Communications Surveys Tutorials vol 20 no 4 pp 3060ndash3097 Fourthquarter 2018

[129] J Wang Z Lan C woo Pyo T Baykas C sean Sum M A Rahman J Gao R Fu-nada F Kojima H Harada and S Kato ldquoBeam codebook based beamforming protocolfor multi-Gbps millimeter-wave WPAN systemsrdquo IEEE Journal on Selected Areas inCommunications vol 27 no 8 pp 1390ndash1399 October 2009

[130] C Cordeiro D Akhmetov and M Park ldquoIEEE 80211ad Introduction and Perfor-mance Evaluation of the First Multi-gbps Wifi Technologyrdquo in Proceedings of the 2010ACM International Workshop on mmWave Communications From Circuits to Net-works ser mmCom rsquo10 New York NY USA ACM 2010 pp 3ndash8

[131] S Sun and T S Rappaport ldquoChannel Modeling and Multi-Cell HybridBeamforming for Fifth-Generation MillimeterWave Wireless CommunicationsrdquoNYU Wireless Brooklyn New York Technical Report TR 2018-001 May2018 [Online] Available httpswirelessengineeringnyuedustatic-homepagetech-reportsTR2018-001 ShuSunpdf

[132] Z Shen R Chen J G Andrews R W Heath and B L Evans ldquoLow complexity userselection algorithms for multiuser MIMO systems with block diagonalizationrdquo IEEETransactions on Signal Processing vol 54 no 9 pp 3658ndash3663 Sep 2006

[133] E Bjrnson L Sanguinetti H Wymeersch J Hoydis and T L Marzetta ldquoMassiveMIMO is a realityWhat is next Five promising research directions for antenna arraysrdquoDigital Signal Processing vol 94 pp 3ndash20 2019

[134] M Costa ldquoWriting on dirty paper (Corresp)rdquo IEEE Transactions on InformationTheory vol 29 no 3 pp 439ndash441 May 1983

[135] R D Wesel and J M Cioffi ldquoAchievable rates for Tomlinson-Harashima precodingrdquoIEEE Transactions on Information Theory vol 44 no 2 pp 824ndash831 March 1998

[136] A Alkhateeb G Leus and R W Heath ldquoLimited Feedback Hybrid Precoding forMulti-User Millimeter Wave Systemsrdquo IEEE Transactions on Wireless Communica-tions vol 14 no 11 pp 6481ndash6494 Nov 2015

132

[137] N Song H Sun and T Yang ldquoCoordinated Hybrid Beamforming for Millimeter WaveMulti-User Massive MIMO Systemsrdquo in 2016 IEEE Global Communications Conference(GLOBECOM) Dec 2016 pp 1ndash6

[138] S Sun T S Rappaport and M Shafi ldquoHybrid beamforming for 5G millimeter-wavemulti-cell networksrdquo in IEEE INFOCOM 2018 - IEEE Conference on Computer Com-munications Workshops (INFOCOM WKSHPS) April 2018 pp 589ndash596

[139] T E Bogale and L B Le ldquoBeamforming for multiuser massive MIMO systems Digitalversus hybrid analog-digitalrdquo in 2014 IEEE Global Communications Conference Dec2014 pp 4066ndash4071

[140] M Kim and Y H Lee ldquoMSE-Based Hybrid RFBaseband Processing for Millimeter-Wave Communication Systems in MIMO Interference Channelsrdquo IEEE Transactionson Vehicular Technology vol 64 no 6 pp 2714ndash2720 June 2015

[141] A Alkhateeb J Mo N Gonzalez-Prelcic and R W Heath ldquoMIMO Precoding andCombining Solutions for Millimeter-Wave Systemsrdquo IEEE Communications Magazinevol 52 no 12 pp 122ndash131 December 2014

[142] J Brady N Behdad and A M Sayeed ldquoBeamspace MIMO for Millimeter-WaveCommunications System Architecture Modeling Analysis and Measurementsrdquo IEEETransactions on Antennas and Propagation vol 61 no 7 pp 3814ndash3827 July 2013

[143] Qualcomm ldquo5G NR based C-V2Xrdquo last accessed 26-02-2020 [Online] Available httpswwwqualcommcommediadocumentsfiles5g-nr-based-c-v2x-presentationpdf[lastaccessed14-01-2019]

[144] V Va J Choi and R W Heath ldquoThe Impact of Beamwidth on Temporal ChannelVariation in Vehicular Channels and Its Implicationsrdquo IEEE Transactions on VehicularTechnology vol 66 no 6 pp 5014ndash5029 June 2017

[145] F Khan Z Pi and S Rajagopal ldquoMillimeter-wave mobile broadband with large scalespatial processing for 5G mobile communicationrdquo in 2012 50th Annual Allerton Confer-ence on Communication Control and Computing (Allerton) Oct 2012 pp 1517ndash1523

[146] N F Abdullah D Berraki A Ameen S Armour A Doufexi A Nix and M BeachldquoChannel Parameters and Throughput Predictions for mmWave and LTE-A Networksin Urban Environmentsrdquo in 2015 IEEE 81st Vehicular Technology Conference (VTCSpring) May 2015 pp 1ndash5

[147] H Claussen ldquoEfficient modelling of channel maps with correlated shadow fading inmobile radio systemsrdquo in 2005 IEEE 16th International Symposium on Personal Indoorand Mobile Radio Communications vol 1 Sept 2005 pp 512ndash516

[148] N Rupasinghe Y Kakishima and Gven ldquoSystem-level performance of mmWavecellular networks for urban micro environmentsrdquo in 2017 XXXIInd General Assemblyand Scientific Symposium of the International Union of Radio Science (URSI GASS)Aug 2017 pp 1ndash4

133

[149] A H Jafari J Park and R W Heath ldquoAnalysis of interference mitigation in mmWavecommunicationsrdquo in 2017 IEEE International Conference on Communications (ICC)May 2017 pp 1ndash6

[150] G Yang J Du and M Xiao ldquoMaximum Throughput Path Selection With RandomBlockage for Indoor 60 GHz Relay Networksrdquo IEEE Transactions on Communicationsvol 63 no 10 pp 3511ndash3524 Oct 2015

[151] I F Akyildiz J M Jornet and C Han ldquoTerahertz band Next frontier for wirelesscommunicationsrdquo Physical Communication vol 12 pp 16ndash32 2014

[152] C Perfecto J D Ser and M Bennis ldquoMillimeter-Wave V2V Communications Dis-tributed Association and Beam Alignmentrdquo IEEE Journal on Selected Areas in Com-munications vol 35 no 9 pp 2148ndash2162 Sept 2017

[153] A Tassi M Egan R J Piechocki and A Nix ldquoModeling and Design of Millimeter-Wave Networks for Highway Vehicular Communicationrdquo IEEE Transactions on Vehic-ular Technology vol 66 no 12 pp 10 676ndash10 691 Dec 2017

[154] Y Wang K Venugopal A F Molisch and R W Heath ldquoAnalysis of Urban MillimeterWave Microcellular Networksrdquo in 2016 IEEE 84th Vehicular Technology Conference(VTC-Fall) Sept 2016 pp 1ndash5

[155] mdashmdash ldquoMmWave Vehicle-to-Infrastructure Communication Analysis of Urban Micro-cellular Networksrdquo IEEE Transactions on Vehicular Technology vol 67 no 8 pp7086ndash7100 Aug 2018

[156] M Gao B Ai Y Niu Z Zhong Y Liu G Ma Z Zhang and D Li ldquoDynamicmmWave beam tracking for high speed railway communicationsrdquo in 2018 IEEE Wire-less Communications and Networking Conference Workshops (WCNCW) April 2018pp 278ndash283

[157] T S Rappaport S Sun and M Shafi ldquoInvestigation and Comparison of 3GPP andNYUSIM Channel Models for 5G Wireless Communicationsrdquo in 2017 IEEE 86th Ve-hicular Technology Conference (VTC-Fall) Sept 2017 pp 1ndash5

[158] J Choi V Va N Gonzalez-Prelcic R Daniels C R Bhat and R W HeathldquoMillimeter-Wave Vehicular Communication to Support Massive Automotive SensingrdquoIEEE Communications Magazine vol 54 no 12 pp 160ndash167 December 2016

[159] S Buzzi and C DrsquoAndrea ldquoOn clustered statistical MIMO millimeter wavechannel simulationrdquo May 2016 last accessed 26-02-2020 [Online] Availablehttpsarxivorgabs160400648

[160] J Zhang Y Huang T Yu J Wang and M Xiao ldquoHybrid Precoding for Multi-Subarray Millimeter-Wave Communication Systemsrdquo IEEE Wireless CommunicationsLetters vol 7 no 3 pp 440ndash443 June 2018

[161] S Ju and T S Rappaport ldquoMillimeter-wave Extended NYUSIM Channel Modelfor Spatial Consistencyrdquo in 2018 IEEE Global Communications Conference (GLOBE-COM) IEEE Aug 2018 pp 1ndash6

134

[162] S Mukherjee S S Das A Chatterjee and S Chatterjee ldquoAnalytical Calculation ofRician K-Factor for Indoor Wireless Channel Modelsrdquo IEEE Access vol 5 pp 19 194ndash19 212 2017

[163] M S A Abdulgader and L Wu ldquoThe Physical layer of the IEEE 80211p WAVE Com-munication Standard The Specifications and Challengesrdquo in Proceedings of the WorldCongress on Engineering and Computer Science (WCECS) vol II San Fransisco USAOct 2014 pp 1ndash8

[164] 3GPP LTE physical layer framework for performance verification TSG RAN1 48R1070674 Last accessed 26-02-2020 [Online] Available httpwww3gpporgDynaReportTDocExMtg--R1-48--26033htm

[165] mdashmdash ldquo TS 38211 Technical Specification Group Radio Access Network NR Physicalchannels and modulation v1510rdquo 2018 last accessed 26-02-2020 [Online] Availablehttpwww3gpporgftpSpecsarchive38 series38211

[166] C Lin and G Y Li ldquoEnergy-Efficient Design of Indoor mmWave and Sub-THz SystemsWith Antenna Arraysrdquo IEEE Transactions on Wireless Communications vol 15 no 7pp 4660ndash4672 July 2016

[167] X Gao L Dai S Han C L I and R W Heath ldquoEnergy-Efficient Hybrid Analog andDigital Precoding for MmWave MIMO Systems With Large Antenna Arraysrdquo IEEEJournal on Selected Areas in Communications vol 34 no 4 pp 998ndash1009 April 2016

[168] Z Wang J Zhu J Wang and G Yue ldquoAn Overlapped Subarray Structure in HybridMillimeter-Wave Multi-User MIMO Systemrdquo in 2018 IEEE Global CommunicationsConference (GLOBECOM 2018) Abu Dhabi United Arab Emirates Dec 2018 pp1ndash6

[169] S Kutty and D Sen ldquoBeamforming for Millimeter Wave Communications An Inclu-sive Surveyrdquo IEEE Communications Surveys Tutorials vol 18 no 2 pp 949ndash973Secondquarter 2016

[170] S Han C L I Z Xu and C Rowell ldquoLarge-scale antenna systems with hybrid analogand digital beamforming for millimeter wave 5Grdquo IEEE Communications Magazinevol 53 no 1 pp 186ndash194 Jan 2015

[171] N Song T Yang and H Sun ldquoOverlapped Subarray Based Hybrid Beamforming forMillimeter Wave Multiuser Massive MIMOrdquo IEEE Signal Processing Letters vol 24no 5 pp 550ndash554 May 2017

[172] N Zorba and H S Hassanein ldquoGrassmannian beamforming for Coordinated Multipointtransmission in multicell systemsrdquo in 38th Annual IEEE Conference on Local ComputerNetworks - Workshops Oct 2013 pp 65ndash69

[173] A Pizzo and L Sanguinetti ldquoOptimal design of energy-efficient millimeter wave hybridtransceivers for wireless backhaulrdquo in 2017 15th International Symposium on Modelingand Optimization in Mobile Ad Hoc and Wireless Networks (WiOpt) May 2017 pp1ndash8

135

[174] X Ge J Yang H Gharavi and Y Sun ldquoEnergy Efficiency Challenges of 5G Small CellNetworksrdquo IEEE Communications Magazine vol 55 no 5 pp 184ndash191 May 2017

[175] S Buzzi and C DrsquoAndrea ldquoAre mmWave Low-Complexity Beamforming StructuresEnergy-Efficient Analysis of the Downlink MU-MIMOrdquo in 2016 IEEE GlobecomWorkshops (GC Wkshps) Dec 2016 pp 1ndash6

[176] C Xiong G Y Li S Zhang Y Chen and S Xu ldquoEnergy- and Spectral-EfficiencyTradeoff in Downlink OFDMA Networksrdquo IEEE Transactions on Wireless Communi-cations vol 10 no 11 pp 3874ndash3886 November 2011

[177] K M S Huq S Mumtaz J Rodriguez and R Aguiar ldquoOverview of Spectral- andEnergy-Efficiency Trade-off in OFDMA Wireless Systemrdquo in Green Communication for4G Wireless Systems S Mumtaz and J Rodriguez Eds River Publishers Aalborg2013

[178] O E Ayach R W Heath S Rajagopal and Z Pi ldquoMultimode precoding in millime-ter wave MIMO transmitters with multiple antenna sub-arraysrdquo in 2013 IEEE GlobalCommunications Conference (GLOBECOM) Dec 2013 pp 3476ndash3480

[179] S Sun T S Rappaport M Shafi P Tang J Zhang and P J Smith ldquoPropagationModels and Performance Evaluation for 5G Millimeter-Wave Bandsrdquo IEEE Transac-tions on Vehicular Technology vol 67 no 9 pp 8422ndash8439 Sep 2018

[180] S Mumtaz J M Jornet J Aulin W H Gerstacker X Dong and B Ai ldquoTerahertzCommunication for Vehicular Networksrdquo IEEE Transactions on Vehicular Technologyvol 66 no 7 pp 5617ndash5625 July 2017

[181] K M S Huq J M Jornet W H Gerstacker A Al-Dulaimi Z Zhou and J AulinldquoTHz Communications for Mobile Heterogeneous Networksrdquo IEEE CommunicationsMagazine vol 56 no 6 pp 94ndash95 June 2018

[182] L Zhao X Li B Gu Z Zhou S Mumtaz V Frascolla H Gacanin M I AshrafJ Rodriguez M Yang and S Al-Rubaye ldquoVehicular Communications Standardiza-tion and Open Issuesrdquo IEEE Communications Standards Magazine vol 2 no 4 pp74ndash80 December 2018

[183] C Han and Y Chen ldquoPropagation Modeling for Wireless Communications in the Ter-ahertz Bandrdquo IEEE Communications Magazine vol 56 no 6 pp 96ndash101 June 2018

[184] S Priebe and T Kurner ldquoStochastic Modeling of THz Indoor Radio Channelsrdquo IEEETransactions on Wireless Communications vol 12 no 9 pp 4445ndash4455 Sep 2013

[185] I F Akyildiz and J M Jornet ldquoRealizing Ultra-Massive MIMO (1024 x 1024) com-munication in the (006-10) Terahertz bandrdquo Nano Communication Networks vol 8pp 46ndash54 2016 electromagnetic Communication in Nano-scale

[186] K Tekbiyik A R Ekti G K Kurt and A Grin ldquoTerahertz band communicationsystems Challenges novelties and standardization effortsrdquo Physical Communicationvol 35 p 100700 2019

136

[187] C Han J M Jornet and I Akyildiz ldquoUltra-Massive MIMO Channel Modeling forGraphene-Enabled Terahertz-Band Communicationsrdquo in 2018 IEEE 87th VehicularTechnology Conference (VTC Spring) June 2018 pp 1ndash5

[188] E de Carvalho A Ali A Amiri M Angjelichinoski and R W Heath JrldquoNon-Stationarities in Extra-Large Scale Massive MIMOrdquo CoRR vol abs1903030852019 [Online] Available httparxivorgabs190303085

[189] A Faisal H Sarieddeen H Dahrouj T Y Al-Naffouri and M-S Alouini ldquoUltra-Massive MIMO Systems at Terahertz Bands Prospects and Challengesrdquo arXiv e-printsp arXiv190211090 Feb 2019

[190] C Huang A Zappone G C Alexandropoulos M Debbah and C Yuen ldquoReconfig-urable Intelligent Surfaces for Energy Efficiency in Wireless Communicationrdquo arXive-prints p arXiv181006934 Oct 2018

[191] A Taha M Alrabeiah and A Alkhateeb ldquoEnabling Large Intelligent Surfaces withCompressive Sensing and Deep Learningrdquo CoRR vol abs190410136 2019 [Online]Available httparxivorgabs190410136

[192] M G Kibria K Nguyen G P Villardi O Zhao K Ishizu and F Kojima ldquoBig DataAnalytics Machine Learning and Artificial Intelligence in Next-Generation WirelessNetworksrdquo IEEE Access vol 6 pp 32 328ndash32 338 2018

[193] C Jiang H Zhang Y Ren Z Han K Chen and L Hanzo ldquoMachine LearningParadigms for Next-Generation Wireless Networksrdquo IEEE Wireless Communicationsvol 24 no 2 pp 98ndash105 April 2017

137

Table of Contents

List of Acronyms

List of Symbols

List of Figures

List of Tables

List of Algorithms

Introduction

Introduction

Overview of the Big Three Enablers

Millimeter-Wave Communications

Massive MIMO

Ultra-Dense Networks

Thesis Motivation

Thesis Objectives

Scientific Methodology Applied


Organization of the Thesis

Millimeter-wave Massive MIMO UDN for 5G Networks

Evolution towards mmWave Massive MIMO UDNs

SISO to massive MIMO

Microwave to mmWave Communication

Legacy Macrocell to Ultra-Dense Small Cell Deployment

Dawn of mmWave Massive MIMO

Architecture

Propagation Characteristics

Health and Safety Issues

Standardization Activities

5G Channel Measurement and Modeling

mmWave Massive MIMO Channels

3GPP 3D Channel Models

NYUSIM Channel Model

Beamforming Techniques

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Conclusions

3D Channel Modeling for 5G UDN and C-I2X

Background

Wave and mmWave Channels Individual Performance

System Model

Map-based Simulation Framework

Simulation Results

Joint Channel Performance for 5G UDN

Deployment Layout

Simulation Results and Analyses

Challenges and Proposed Solutions

C-I2V Channel Performance

Network Deployment

Channel Model

Antenna Model


Conclusions

Novel Generalized Framework for Hybrid Beamforming

Background

Hybrid Beamforming Schemes

Fully-connected HBF architecture

Sub-connected HBF architecture

Overlapped subarray HBF architecture

Proposed Generalized Framework for HBF architectures

Precoding and Postcoding

Analog-only Beamsteering

Hybrid Precoding with Baseband Zero Forcing

Singular Value Decomposition Precoding

Power Consumption Model

Spectral and Energy Efficiency

Spectral Efficiency and Achievable Rate

Energy Efficiency

Hybrid Beamforming for C-I2X

System Model and Parameters

Simulation Results

Hybrid Beamforming for C-I2P


Simulation Results

Conclusions


Thesis Summary

Channel Modeling

Hybrid Beamforming


THz Channel Modeling

Ultra-massive MIMO

6G for Energy Efficiency

Quantum Machine Learning

Bibliography


2020



Tese apresentada as Universidades do Minho Aveiro e Porto para cumpri-mento dos requesitos necessarios a obtencao do grau de Doutor em Tele-comunicacoes no ambito do programa doutoral MAP-Tele realizada sob aorientacao cientıfica do Doutor Jonathan Rodrıguez Gonzalez InvestigadorPrincipal do Instituto de Telecomunicacoes Universidade de Aveiro

Apoio financeiro da Fundacao para a Ciencia e a Tecnologia (FCT) Portugalatraves da bolsa com a referencia PDBD1138232015


2020





o juri the jury



























Table of Contents

Table of Contents i

List of Acronyms iv

List of Symbols vii

List of Figures ix

List of Tables xi


1 Introduction 1

11 Introduction 1



122 Massive MIMO 4













221 Architecture 26







i






25 Conclusions 45


31 Background 47


321 System Model 48












35 Conclusions 83


41 Background 85




















48 Conclusions 112

ii




Bibliography 122

iii

List of Acronyms

1G First generation

2D Two-dimensional


3G Third generation



5G Fifth generation

6G Sixth generation


AE Antenna element




AP Access point

AS Angular spread


B5G Beyond-5G

BBU Baseband unit

BS Base station





CL Coupling loss

CN Condition number



DL Downlink


DS Delay spread



iv


GF Geometry factor









v









ZF Zero forcing

vi

List of Symbols






vii


viii

List of Figures



































ix



x

List of Tables





xi

xii

List of Algorithms



xiii

xiv

Chapter 1

Introduction


11 Introduction




1





Today



Today

6G amp

Beyond

1G 2G 3G 4G 5G1G 2G 3G 4G 5G










bull CDMA-2000












bull CDMA-2000




(NR)bull IMT-2020










bull CDMA-2000




(NR)bull IMT-2020



mMTC ITU

URLLC

eMBBmMTC ITU

URLLC

eMBB



Today

6G amp

Beyond

1G 2G 3G 4G 5G










bull CDMA-2000




(NR)bull IMT-2020



mMTC ITU

URLLC

eMBB


2





5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing

5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing


3






122 Massive MIMO


4




Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss

Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss


5





6








7



8






9







10

Legend






2- UE Distribution



10- Applications








8- Schedulers


7- HetNet + UDN


1- BS Layout








Simulator

Mobility Model

Spatial Consistency

Blockage Model



SE-EE Performance

I2X Network Layout



C-I2X System Level

Simulator


11













12






bull Book Chapter


bull Posters






13







14



Chapter 1

Introduction

Aim and Objectives



Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary



15

16

Chapter 2






17







y = Hx + n (21)

ykj1

ykjNRX

=



xj1

xjNRX

+

nk1

nkNRX

(22)

18


BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1UE1

1

NRXUE1

1

NRX

UE2UE2

1

NRXUE2

1

NRX

UEKUEK

1

NRXUEK

1

NRX

BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1

1

NRX

UE2

1

NRX

UEK

1

NRX


(a) SISO [NTX = 1 NRX = 1K = 1]


y1 = h11x1 + n1 (23)


(1 + h2PT

σ2n

) (24)





19





NTXHHH

)∣∣∣∣ (26)



(1 +

ρmax (NRX NTX)

NTX

) (27)






20




∣∣(INRX + ρHPHH)∣∣ (28)








21




(1 + ρ

NRX

NTX

) (210)









22









23


Reflector

Obstacle

Relay

BS

UE 2

UE 3UE 3

UE 1UE 1UE 1

UE 4UE 4





24






25

221 Architecture







microwave

mmWave

Control

Data

X2-interface

Vehicular Comm cell

Virtual cell

Macro BS






26






27

50 100 150 200 250 300

Frequency (GHz)

10-3

10-2

10-1

100

101

102

Spe

cific

Atte

nuat

ion

(dB

km

)


50 100 150 200 250 300

Frequency (GHz)

10-2

10-1

100

101

102

Rai

n A

ttenu

atio

n (d

Bk

m)



28



















(λ

4π

)2

dminusα (211)


(λ

4π

)2

dminusα(d) (212)

29
























30







31






32




3GPP TR 38900[85]

NYUSIM[42 101102]



Statistical



33


5G Channel Models















to the theory of




to the theory of




distributed



distributed

Geometry-based (GB)


distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Map-based (MB)


simplified 3D map

Regular-shaped (RS)



Regular-shaped (RS)





distributed



distributed




time delays




time delays



between paths



between paths

5G Channel Models









to the theory of




distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Regular-shaped (RS)





distributed




time delays



between paths






1


34




[103] [104105] [106] [107] [108]


hybrid

GBSM GBSMGGBSM

GBSMGGBSM

GBSM


05-100 045-100 6-100 05-100 -



-


no yes yes no yes


no no no no yes


yes yes yes yes yes



MiWEBA IMT-2020

[109] [110] [88] [111]



up to 70 up to 100 lt 6 57-66 05-100


[100 MHz 1GHz]

10 of fc - 216 GHz [100 MHz10 of fc]


no yes no yes no




yes yes yes yes yes


35


UE 3UE 2

eNodeBSpherical

wavefronts

UE 1

Cluster 1

Cluster 2

Cluster 3

Cluster 5 Cluster 4




36


Set

scenario

network layout

antenna parameters

Assign propagation

condition

LOS

NLOS

Calculate pathloss

Generate correlated

LSPs

delay spread

angular spread

shadow fading

K-factor

Generate

cross-polarisation

ratios

Generate

AoAs

AoDs

Perform random

coupling of rays

Generate delays

Generate

cluster powers

Draw random

initial phases

Generate

channel

coefficients

Apply

pathloss and

shadowing



37









38


H =

radicNRXNTX

L

Lsuml=1


)middotandHTX

(φTXl θTXl

)middotaRX

(φRXl θRXl

)middotaHTX

(φTXl θTXl

)

(215)


(φRXl θRXl

)and aTX

(φTXl θTXl



)and andTX

(φTXl θTXl







39


∣∣wHRFHfRF


wopt fopt)

= arg max∣∣wH

RFHfRF

∣∣subject to

wi =

radic1


fl =

radic1


(216)

RF

chain


PAPA

PAPA

PAPA

1

2

NTX

PSs

s

RF

chain


PA

PA

PA

1

2

NTX

PSs

s



40








PA

PA

1

2

NTX

PAPA

RF chainRF chain

RF chainRF chain

RF chainRF chain

Ns

Baseband

Digital

Beamformer

FDBF

PA

PA

1

2

NTX

PA

RF chain

RF chain

RF chain

Ns

Baseband

Digital

Beamformer

FDBF




FMFDBF = HH (217)

FZFDBF = HH

(HHH

)minus1 (218)

FWFDBF = HH

(HHH +

σ2nNs

PRXINRX

)minus1

(219)


41






∥∥FkDBF

∥∥2

F= Ns The


yk = Hk

Ksumn=1

FnDBF sn + nk (221)









42


Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX

Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSsNs

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSs



maxFRF WRFk forallk

Ksumk=1

∥∥WHRFkHkFRF

∥∥2

F

subject to

FHRFFRF = INTX WH


(222)


Heffk = [HTeff1 H

Teffk H

TeffK ]T (223)




FMRTBB = HH

effk (224)

FZFBB = HH


43



rk = Hk

Ksumn=1

FRFn FBB

n sn + nk (226)

yk = WHBBkWH

RFkHk

Ksumn=1

FRFn FBB

n sn + WHBBkWH

RFknk (227)









antennas





no multi-user




44

25 Conclusions



45

46

Chapter 3



31 Background



47





321 System Model


48

55 timesISD (m)

55

timesIS

D (

m)

BuildingBase

Station

Outdoor

User

Signal

link

Interfering

linkIndoor

User






49







(min

(18

d2D 1

)(1minus exp

(minusd2D

63

))+ exp

(minusd2D

63

))(1 + C (d2DhRX

))


(min

(18

d2D 1

)(1minus exp

(minusd2D

36

))+ exp

(minusd2D

36

))

UMa-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

63

)(1minus 18

d2D

))(1 + C (d2DhRX

)) 18 lt d2D

UMi-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

36

)(1minus 18

d2D

)) 18 lt d2D

where C (d2D hRX) =

1 d2D le 18


)3exp

(minusd2D150

) 18 lt d2D

C prime (hRX) =


10

)15

13 lt hRX le 23


50

Tab

le3

3P

ath

loss

mod

els

(ad

apte

dfr

om[8

485]

)

Scen

ari

oP

ath

loss

(dB

)σSF

(dB

)

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

5000

4

UM

a-

microW

ave

PLNLOS

=69

51+

390

9lo

g10(d

3D

)+

20lo

g10(fc)

+75

log

10(hRX

)minus

06hRXminus

00

825(hRX

)26

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

3

UM

i-micro

Wav

ePLNLOS

=23

15+

367

log

10(d

3D

)+

26lo

g10(fc)minus

03hUE

4

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

32

4+

20

log

10(d

3D

)+

20lo

g10(fc)

d10ltd

2DltdBP

32

4+

40

log

10(d

3D

)+

20lo

g10(fc)minus

10lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

500

04

UM

a-

mm

Wav

ePLNLOS

=14

44+

390

8lo

g10(d

3D

)+

20lo

g10(fc)minus

06h

RX

6

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

PLLOS

=

32

4+

21

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

324

+40

log

10(d

3D

)+

20lo

g10(fc)minus

95

log

10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

4

UM

i-m

mW

ave

PLNLOS

=22

85+

353

log

10(d

3D

)+

213

log

10(fc)minus

03hUE

78

2

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

NO

TE

S

f cis

inG

Hzd

2Dd

3D

andhRX

are

inm

eter

sdBP

=32

0((hRXminus

1)f c

)fo

rU

Ma

anddBP

=120

((hRXminus

1)fc)

for

UM

idBP

isth

eb

reak

-poi

nt

dis

tance

[84]

[8

5]an

dPLwall

isth

ew

all

pen

etra

tion

loss

base

don

Tab

les

74

3-(1

amp2)

in[8

5]

51




χk =

0 k rarr outdoor

1 k rarr indoor(32)



fl nmaxfl ]) (33)









52







dkj2D =radic


dkj3D =





ξkj =

0 k j rarr NLOS

1 k j rarr LOS(311)


χk ξkj =

0 0 rarr PLNLOS

0 1 rarr PLLOS



(312)




RSRPkj = P kjT +Gkj




53


Parameter Values


AEH = minusmin

12

(φ

65o 30 dB


AEV = minusmin

12

(θ minus 90o

65o 30 dB




8 dBi



(a) Coupling Loss



RSRPkj = P kjT +Gkj




54

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS


(b) Geometry Factor


55






Nk = No + 10 log10Bk +NF (320)



56

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1E

CD


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF






57


-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF



-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF








58







59



Tk(t)



Network ele-ments


Small cell (SC)







3GPP TR 38900[85]



[(Nintsumn=1

In

)+ (No + 10 log10B +NF )

]

(321)


60















InterferencesumNint









bandwidth





61



0

200

400

600

800

Cel

l cap

acit

y (M

bp

s)

1000

1200

1400

125

1600

250500

1000

average macro

average small

total



62


0

20

40

60

80

Cel

l cap

acit

y (M

bp

s) 100

120

140

125250

5001000







63


Es

0

20

40

60

0

80

Ave

rag

e U

E T

hro

ug

hp

ut

(Mb

ps)

125

100

120

140

250 80500

1001000


of indoor U


0

05

Sp

ectr

al e

ffic

ien

cy (

bp

sH

z)

125

1

15

0250

80500

1001000


64





30 32 34 36 38 40 42 44 46 48


0

200

400

600

800

1000

1200

1400

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)




65








66

0 20 40 60 80 100


0

100

200

300

400

500

600

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)



= 35 dBm


125 250 375 500 625 750 875 1000


0

50

100

150

200

250

300

350

400

450

500

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)


PTX

= 35 dBm


67







68






69


PLOS(d3D) =

[min

(27

d 1

)(1minus eminus

d71

)+ eminus

d71

]2

(322)

342 Channel Model



hdir(t φ θ) =

Nclsumc=1

Nspcsums=1


(323)


G(φ θ) = max

(G0e

αφ2+βθ2 Go100

) (324)


3dBθ23dB

α =4 ln(2)

φ23dB

β =4 ln(2)

θ23dB

(325)

70






) (327)

ϕcs = 2π

(fcτcs +

vRX cos (φcs)

λ4 t

) (328)


343 Antenna Model



H =


middotNclsumc=1

Nspcsums=1


j2π

(fcτcs+

vRX cos(φcs)λ

4t)middot aRX

(φRXcs

)aHTX

(φTXcs

)

(329)

aRX(φRXcs

)=

1radicNRX



aTX(φTXcs

)=

1radicNTX




71






φRX3dB 65 65 10


(a) Path Loss



PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (333)

72


PL = 10κ log10 (d3D) + Υ + 15

(d3D

1000

) (334)


50 55 60 65 70 75 80 85 90 95

Path Loss (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave



73

0 2 4 6 8 10 12 14 16 18 20


40

50

60

70

80

90

100

110

120

Path

Loss (

dB








0 50 100 150 200 250 300 350 400 450 500


40

50

60

70

80

90

100

110

120

Path

Loss (

dB

)








74




(b) Rician K-Factor


KF =PLOSsumPNLOS

=PRXc=1s=1(sumNcl

c=1

sumNspcs=1 PRXcs

)minus PRXc=1s=1

(336)


75

-20 0 20 40 60 80 100 120

K-Factor (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave




τRMS =


sumNspcs=1 τ2

csPRXcssumNclc=1

sumNspcs=1 PRXcs

minus

(sumNclc=1


c=1

sumNspcs=1 PRXcs

)2

(337)

Bc asymp1

2πτRMS (338)


76


2 4 6 8 10 12 14 16 18 20 22


0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave




77

1 2 3 4 5 6

Number of Clusters

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave


0 5 10 15 20 25 30 35 40 45


0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave


78



0

-150

-100

mmWave

Re

ce

ive

d P

ow

er

(dB

)

-50


500

Technology

0

LTE-A

1000 DSRC


0

-150

-100

mmWave200

Re

ce

ive

d P

ow

er

(dB

)

-50


Technology

0

LTE-A400

600 DSRC


79








80

2 4 6 8 10 12 14 16 18 20


1

2

3

4

5

6

7

8C

hannel R

ank

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)


2 4 6 8 10 12 14 16 18 20


0

20

40

60

80

100

120

140

160

Channel C

onditio

n N

um

ber

(dB

)

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)


81

(g) Data Rate




R = B middot log2

(1 + ρRminus1


where Rn = σ2nw


0 2 4 6 8 10 12

Data Rate (Gbps)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave



82

35 Conclusions







83

84

Chapter 4



41 Background




85






PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (41)


86

hdir(t φ) =Lsuml=1



H =

radicNRXNTX

LmiddotLsuml=1


j2π

(fcτl+

vRX cos(φl)λ

4t)middot aRX

(φRXl

)aHTX

(φTXl

) (44)

aRX(φRXl

)=

1radicNRX



aTX(φTXl

)=

1radicNTX







87

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs







88


Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs


Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K

Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K


89










0 1 2 NTXNRF

For the gener-


4N =


(NTXNRF

minus 1)]

rarr overlapped

NTXNRF

rarr sub-connected

(47)


90

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Fig

ure

44

Gen

eral

ized

HB

Far

chit

ectu

re

91




NPS =

NRFsumk=1

NkPS = NRF timesNk

PS (49)






PS = |mk|


M =




0 m4N+Nsub2

0 0 0 mJK

(411)


B =




0 1

0 0 0 1

(412)

92


g =

g1

g2

gNTX

gj =

NRFsumk=1




∣∣∣ =




(414)



P kb =PTNTX

sumjisinmk

(gj)minus1

(415)

PT =

NRFsumk=1

P kb (416)






93

x = FRFFBBs (418)




yk = wHk Hk

Ksumn=1








[fRFk

]=

1radicNTX

ejφk = aTX(φTXl

) (420)

[wRFk

]=

1radicNRX

ejφk = aRX(φRXl

) (421)


(φTXl

)and

aRX(φRXl




] (422)

WRF =[wRF

1 wRF2 wRF

K

] (423)


94




] (424)

FBB = HHeff

(HeffHH

eff

)minus1 (425)



fBBk =fBBk


||FRFFBB||2F = K (427)










95



OutputsFBB FRF WRF




fRFk = akTX

wRFk = akRX


]and WRF =

[wRF

1 wRF2 wRF

K

] according to




Hkeff = wH

k HkFRF


]is designed







(PRXcct

) (430)

96









PADC [175] 200




P TXcct = NTXRF P

TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX


PRXcct = NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA (435)


TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX


]NUE

[NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA

]+ LBH middotR

(436)

97







R =

Ksumk=1

Rk (439)


∣∣wHk Hkf

RFk

∣∣2sumn6=k P

nT middot∣∣wH


k

(440)



∣∣2sumn6=k P

nT middot∣∣wH


k

(441)




ηEE =sum rate


R

Ptotal (443)

98


dηSE

∣∣∣∣ηSE=

R

BW

= 0 (444)






sub and NRXsub




99





RF = 8





Beam 4N = 0 4N = 2 4N = 4 4N = 6 4N = 81 1 12107 14214 15000 12 1 09964 09928 09583 13 1 09130 08261 07917 14 1 08797 07595 07500 15 1 08797 07595 07500 16 1 09130 08261 07916 17 1 09964 09928 09583 18 1 12107 14214 15000 1

PT (W) 8 8 8 8 8



100














101

1 2 3 4 5 6 7 8

50

100

150

200

250

300

350

400

450

500

550




102




0 1 2 3 4 5 6 7 8

Total TX Power (W)

140

160

180

200

220

240

260

Su

m S

pe

ctr

al E

ffic

ien

cy (

bitss

Hz)



103


0 1 2 3 4 5 6 7 8

Total TX Power (W)

11

115

12

125

13

135

14

145

15

En

erg

y E

ffic

ien

cy (

Gb

J)


18 20 22 24 26 28 30 32 34


11

115

12

125

13

135

14

145

15

Energ

y E

ffic

iency (

GbJ

)


104



22

0

24

26

8

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

28

05 76

Transmit Power (W)

30

5

User

32

41 32

115

24

25

26

27

28

29

30

31


11

0

12

8

Energ

y E

ffic

iency (

GbJ

)

05 7

13

6

Transmit Power (W)

5

User

14

41 32

115

114

116

118

12

122

124

126

128

13

132

134


105






U1

U7

00

1

U5

AP

U8 U2

2

10

3

AP

UE

he

igh

t (m

)

05 8

4

U3

5

Walkway width (m)

6

U6

1

Walkway length (m)

U4

415 22 0


106







rk = Hk

Ksumn=1


yk = wHk Hk

Ksumn=1


Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Transmitter

1

NRX

1

NRX


User K

LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

RF

chain

RF

chain

LNA

LNA

LNA

RF

chain

yK

RF

chain

RF

chain

y1

Receivers

User 1LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

1

NRX

LNA

LNA

LNA

1

NRX

1

NRX


User K

LNA

LNA

LNA

RF

chain

yK

RF

chain

y1

Receivers

User 1LNA

LNA

LNA

1

NRX

GTX GRX

Channel



107










108

0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB




109




0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB




110


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

GbitsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


111



48 Conclusions





112




113

114

Chapter 5



51 Thesis Summary




115







116







117









118






119







120





121

Bibliography











122














123














124














125













126













127













128














129















130













131













132













133














134













135














136








137

Table of Contents

List of Acronyms

List of Symbols

List of Figures

List of Tables

List of Algorithms

Introduction

Introduction



Massive MIMO


Thesis Motivation

Thesis Objectives










Architecture









Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Conclusions


Background


System Model


Simulation Results


Deployment Layout




Network Deployment

Channel Model

Antenna Model


Conclusions


Background













Energy Efficiency



Simulation Results



Simulation Results

Conclusions


Thesis Summary

Channel Modeling

Hybrid Beamforming



Ultra-massive MIMO



Bibliography


2020





o juri the jury



























Table of Contents

Table of Contents i

List of Acronyms iv

List of Symbols vii

List of Figures ix

List of Tables xi


1 Introduction 1

11 Introduction 1



122 Massive MIMO 4













221 Architecture 26







i






25 Conclusions 45


31 Background 47


321 System Model 48












35 Conclusions 83


41 Background 85




















48 Conclusions 112

ii




Bibliography 122

iii

List of Acronyms

1G First generation

2D Two-dimensional


3G Third generation



5G Fifth generation

6G Sixth generation


AE Antenna element




AP Access point

AS Angular spread


B5G Beyond-5G

BBU Baseband unit

BS Base station





CL Coupling loss

CN Condition number



DL Downlink


DS Delay spread



iv


GF Geometry factor









v









ZF Zero forcing

vi

List of Symbols






vii


viii

List of Figures



































ix



x

List of Tables





xi

xii

List of Algorithms



xiii

xiv

Chapter 1

Introduction


11 Introduction




1





Today



Today

6G amp

Beyond

1G 2G 3G 4G 5G1G 2G 3G 4G 5G










bull CDMA-2000












bull CDMA-2000




(NR)bull IMT-2020










bull CDMA-2000




(NR)bull IMT-2020



mMTC ITU

URLLC

eMBBmMTC ITU

URLLC

eMBB



Today

6G amp

Beyond

1G 2G 3G 4G 5G










bull CDMA-2000




(NR)bull IMT-2020



mMTC ITU

URLLC

eMBB


2





5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing

5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing


3






122 Massive MIMO


4




Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss

Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss


5





6








7



8






9







10

Legend






2- UE Distribution



10- Applications








8- Schedulers


7- HetNet + UDN


1- BS Layout








Simulator

Mobility Model

Spatial Consistency

Blockage Model



SE-EE Performance

I2X Network Layout



C-I2X System Level

Simulator


11













12






bull Book Chapter


bull Posters






13







14



Chapter 1

Introduction

Aim and Objectives



Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary



15

16

Chapter 2






17







y = Hx + n (21)

ykj1

ykjNRX

=



xj1

xjNRX

+

nk1

nkNRX

(22)

18


BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1UE1

1

NRXUE1

1

NRX

UE2UE2

1

NRXUE2

1

NRX

UEKUEK

1

NRXUEK

1

NRX

BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1

1

NRX

UE2

1

NRX

UEK

1

NRX


(a) SISO [NTX = 1 NRX = 1K = 1]


y1 = h11x1 + n1 (23)


(1 + h2PT

σ2n

) (24)





19





NTXHHH

)∣∣∣∣ (26)



(1 +

ρmax (NRX NTX)

NTX

) (27)






20




∣∣(INRX + ρHPHH)∣∣ (28)








21




(1 + ρ

NRX

NTX

) (210)









22









23


Reflector

Obstacle

Relay

BS

UE 2

UE 3UE 3

UE 1UE 1UE 1

UE 4UE 4





24






25

221 Architecture







microwave

mmWave

Control

Data

X2-interface

Vehicular Comm cell

Virtual cell

Macro BS






26






27

50 100 150 200 250 300

Frequency (GHz)

10-3

10-2

10-1

100

101

102

Spe

cific

Atte

nuat

ion

(dB

km

)


50 100 150 200 250 300

Frequency (GHz)

10-2

10-1

100

101

102

Rai

n A

ttenu

atio

n (d

Bk

m)



28



















(λ

4π

)2

dminusα (211)


(λ

4π

)2

dminusα(d) (212)

29
























30







31






32




3GPP TR 38900[85]

NYUSIM[42 101102]



Statistical



33


5G Channel Models















to the theory of




to the theory of




distributed



distributed

Geometry-based (GB)


distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Map-based (MB)


simplified 3D map

Regular-shaped (RS)



Regular-shaped (RS)





distributed



distributed




time delays




time delays



between paths



between paths

5G Channel Models









to the theory of




distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Regular-shaped (RS)





distributed




time delays



between paths






1


34




[103] [104105] [106] [107] [108]


hybrid

GBSM GBSMGGBSM

GBSMGGBSM

GBSM


05-100 045-100 6-100 05-100 -



-


no yes yes no yes


no no no no yes


yes yes yes yes yes



MiWEBA IMT-2020

[109] [110] [88] [111]



up to 70 up to 100 lt 6 57-66 05-100


[100 MHz 1GHz]

10 of fc - 216 GHz [100 MHz10 of fc]


no yes no yes no




yes yes yes yes yes


35


UE 3UE 2

eNodeBSpherical

wavefronts

UE 1

Cluster 1

Cluster 2

Cluster 3

Cluster 5 Cluster 4




36


Set

scenario

network layout

antenna parameters

Assign propagation

condition

LOS

NLOS

Calculate pathloss

Generate correlated

LSPs

delay spread

angular spread

shadow fading

K-factor

Generate

cross-polarisation

ratios

Generate

AoAs

AoDs

Perform random

coupling of rays

Generate delays

Generate

cluster powers

Draw random

initial phases

Generate

channel

coefficients

Apply

pathloss and

shadowing



37









38


H =

radicNRXNTX

L

Lsuml=1


)middotandHTX

(φTXl θTXl

)middotaRX

(φRXl θRXl

)middotaHTX

(φTXl θTXl

)

(215)


(φRXl θRXl

)and aTX

(φTXl θTXl



)and andTX

(φTXl θTXl







39


∣∣wHRFHfRF


wopt fopt)

= arg max∣∣wH

RFHfRF

∣∣subject to

wi =

radic1


fl =

radic1


(216)

RF

chain


PAPA

PAPA

PAPA

1

2

NTX

PSs

s

RF

chain


PA

PA

PA

1

2

NTX

PSs

s



40








PA

PA

1

2

NTX

PAPA

RF chainRF chain

RF chainRF chain

RF chainRF chain

Ns

Baseband

Digital

Beamformer

FDBF

PA

PA

1

2

NTX

PA

RF chain

RF chain

RF chain

Ns

Baseband

Digital

Beamformer

FDBF




FMFDBF = HH (217)

FZFDBF = HH

(HHH

)minus1 (218)

FWFDBF = HH

(HHH +

σ2nNs

PRXINRX

)minus1

(219)


41






∥∥FkDBF

∥∥2

F= Ns The


yk = Hk

Ksumn=1

FnDBF sn + nk (221)









42


Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX

Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSsNs

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSs



maxFRF WRFk forallk

Ksumk=1

∥∥WHRFkHkFRF

∥∥2

F

subject to

FHRFFRF = INTX WH


(222)


Heffk = [HTeff1 H

Teffk H

TeffK ]T (223)




FMRTBB = HH

effk (224)

FZFBB = HH


43



rk = Hk

Ksumn=1

FRFn FBB

n sn + nk (226)

yk = WHBBkWH

RFkHk

Ksumn=1

FRFn FBB

n sn + WHBBkWH

RFknk (227)









antennas





no multi-user




44

25 Conclusions



45

46

Chapter 3



31 Background



47





321 System Model


48

55 timesISD (m)

55

timesIS

D (

m)

BuildingBase

Station

Outdoor

User

Signal

link

Interfering

linkIndoor

User






49







(min

(18

d2D 1

)(1minus exp

(minusd2D

63

))+ exp

(minusd2D

63

))(1 + C (d2DhRX

))


(min

(18

d2D 1

)(1minus exp

(minusd2D

36

))+ exp

(minusd2D

36

))

UMa-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

63

)(1minus 18

d2D

))(1 + C (d2DhRX

)) 18 lt d2D

UMi-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

36

)(1minus 18

d2D

)) 18 lt d2D

where C (d2D hRX) =

1 d2D le 18


)3exp

(minusd2D150

) 18 lt d2D

C prime (hRX) =


10

)15

13 lt hRX le 23


50

Tab

le3

3P

ath

loss

mod

els

(ad

apte

dfr

om[8

485]

)

Scen

ari

oP

ath

loss

(dB

)σSF

(dB

)

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

5000

4

UM

a-

microW

ave

PLNLOS

=69

51+

390

9lo

g10(d

3D

)+

20lo

g10(fc)

+75

log

10(hRX

)minus

06hRXminus

00

825(hRX

)26

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

3

UM

i-micro

Wav

ePLNLOS

=23

15+

367

log

10(d

3D

)+

26lo

g10(fc)minus

03hUE

4

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

32

4+

20

log

10(d

3D

)+

20lo

g10(fc)

d10ltd

2DltdBP

32

4+

40

log

10(d

3D

)+

20lo

g10(fc)minus

10lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

500

04

UM

a-

mm

Wav

ePLNLOS

=14

44+

390

8lo

g10(d

3D

)+

20lo

g10(fc)minus

06h

RX

6

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

PLLOS

=

32

4+

21

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

324

+40

log

10(d

3D

)+

20lo

g10(fc)minus

95

log

10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

4

UM

i-m

mW

ave

PLNLOS

=22

85+

353

log

10(d

3D

)+

213

log

10(fc)minus

03hUE

78

2

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

NO

TE

S

f cis

inG

Hzd

2Dd

3D

andhRX

are

inm

eter

sdBP

=32

0((hRXminus

1)f c

)fo

rU

Ma

anddBP

=120

((hRXminus

1)fc)

for

UM

idBP

isth

eb

reak

-poi

nt

dis

tance

[84]

[8

5]an

dPLwall

isth

ew

all

pen

etra

tion

loss

base

don

Tab

les

74

3-(1

amp2)

in[8

5]

51




χk =

0 k rarr outdoor

1 k rarr indoor(32)



fl nmaxfl ]) (33)









52







dkj2D =radic


dkj3D =





ξkj =

0 k j rarr NLOS

1 k j rarr LOS(311)


χk ξkj =

0 0 rarr PLNLOS

0 1 rarr PLLOS



(312)




RSRPkj = P kjT +Gkj




53


Parameter Values


AEH = minusmin

12

(φ

65o 30 dB


AEV = minusmin

12

(θ minus 90o

65o 30 dB




8 dBi



(a) Coupling Loss



RSRPkj = P kjT +Gkj




54

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS


(b) Geometry Factor


55






Nk = No + 10 log10Bk +NF (320)



56

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1E

CD


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF






57


-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF



-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF








58







59



Tk(t)



Network ele-ments


Small cell (SC)







3GPP TR 38900[85]



[(Nintsumn=1

In

)+ (No + 10 log10B +NF )

]

(321)


60















InterferencesumNint









bandwidth





61



0

200

400

600

800

Cel

l cap

acit

y (M

bp

s)

1000

1200

1400

125

1600

250500

1000

average macro

average small

total



62


0

20

40

60

80

Cel

l cap

acit

y (M

bp

s) 100

120

140

125250

5001000







63


Es

0

20

40

60

0

80

Ave

rag

e U

E T

hro

ug

hp

ut

(Mb

ps)

125

100

120

140

250 80500

1001000


of indoor U


0

05

Sp

ectr

al e

ffic

ien

cy (

bp

sH

z)

125

1

15

0250

80500

1001000


64





30 32 34 36 38 40 42 44 46 48


0

200

400

600

800

1000

1200

1400

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)




65








66

0 20 40 60 80 100


0

100

200

300

400

500

600

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)



= 35 dBm


125 250 375 500 625 750 875 1000


0

50

100

150

200

250

300

350

400

450

500

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)


PTX

= 35 dBm


67







68






69


PLOS(d3D) =

[min

(27

d 1

)(1minus eminus

d71

)+ eminus

d71

]2

(322)

342 Channel Model



hdir(t φ θ) =

Nclsumc=1

Nspcsums=1


(323)


G(φ θ) = max

(G0e

αφ2+βθ2 Go100

) (324)


3dBθ23dB

α =4 ln(2)

φ23dB

β =4 ln(2)

θ23dB

(325)

70






) (327)

ϕcs = 2π

(fcτcs +

vRX cos (φcs)

λ4 t

) (328)


343 Antenna Model



H =


middotNclsumc=1

Nspcsums=1


j2π

(fcτcs+

vRX cos(φcs)λ

4t)middot aRX

(φRXcs

)aHTX

(φTXcs

)

(329)

aRX(φRXcs

)=

1radicNRX



aTX(φTXcs

)=

1radicNTX




71






φRX3dB 65 65 10


(a) Path Loss



PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (333)

72


PL = 10κ log10 (d3D) + Υ + 15

(d3D

1000

) (334)


50 55 60 65 70 75 80 85 90 95

Path Loss (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave



73

0 2 4 6 8 10 12 14 16 18 20


40

50

60

70

80

90

100

110

120

Path

Loss (

dB








0 50 100 150 200 250 300 350 400 450 500


40

50

60

70

80

90

100

110

120

Path

Loss (

dB

)








74




(b) Rician K-Factor


KF =PLOSsumPNLOS

=PRXc=1s=1(sumNcl

c=1

sumNspcs=1 PRXcs

)minus PRXc=1s=1

(336)


75

-20 0 20 40 60 80 100 120

K-Factor (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave




τRMS =


sumNspcs=1 τ2

csPRXcssumNclc=1

sumNspcs=1 PRXcs

minus

(sumNclc=1


c=1

sumNspcs=1 PRXcs

)2

(337)

Bc asymp1

2πτRMS (338)


76


2 4 6 8 10 12 14 16 18 20 22


0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave




77

1 2 3 4 5 6

Number of Clusters

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave


0 5 10 15 20 25 30 35 40 45


0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave


78



0

-150

-100

mmWave

Re

ce

ive

d P

ow

er

(dB

)

-50


500

Technology

0

LTE-A

1000 DSRC


0

-150

-100

mmWave200

Re

ce

ive

d P

ow

er

(dB

)

-50


Technology

0

LTE-A400

600 DSRC


79








80

2 4 6 8 10 12 14 16 18 20


1

2

3

4

5

6

7

8C

hannel R

ank

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)


2 4 6 8 10 12 14 16 18 20


0

20

40

60

80

100

120

140

160

Channel C

onditio

n N

um

ber

(dB

)

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)


81

(g) Data Rate




R = B middot log2

(1 + ρRminus1


where Rn = σ2nw


0 2 4 6 8 10 12

Data Rate (Gbps)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave



82

35 Conclusions







83

84

Chapter 4



41 Background




85






PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (41)


86

hdir(t φ) =Lsuml=1



H =

radicNRXNTX

LmiddotLsuml=1


j2π

(fcτl+

vRX cos(φl)λ

4t)middot aRX

(φRXl

)aHTX

(φTXl

) (44)

aRX(φRXl

)=

1radicNRX



aTX(φTXl

)=

1radicNTX







87

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs







88


Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs


Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K

Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K


89










0 1 2 NTXNRF

For the gener-


4N =


(NTXNRF

minus 1)]

rarr overlapped

NTXNRF

rarr sub-connected

(47)


90

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Fig

ure

44

Gen

eral

ized

HB

Far

chit

ectu

re

91




NPS =

NRFsumk=1

NkPS = NRF timesNk

PS (49)






PS = |mk|


M =




0 m4N+Nsub2

0 0 0 mJK

(411)


B =




0 1

0 0 0 1

(412)

92


g =

g1

g2

gNTX

gj =

NRFsumk=1




∣∣∣ =




(414)



P kb =PTNTX

sumjisinmk

(gj)minus1

(415)

PT =

NRFsumk=1

P kb (416)






93

x = FRFFBBs (418)




yk = wHk Hk

Ksumn=1








[fRFk

]=

1radicNTX

ejφk = aTX(φTXl

) (420)

[wRFk

]=

1radicNRX

ejφk = aRX(φRXl

) (421)


(φTXl

)and

aRX(φRXl




] (422)

WRF =[wRF

1 wRF2 wRF

K

] (423)


94




] (424)

FBB = HHeff

(HeffHH

eff

)minus1 (425)



fBBk =fBBk


||FRFFBB||2F = K (427)










95



OutputsFBB FRF WRF




fRFk = akTX

wRFk = akRX


]and WRF =

[wRF

1 wRF2 wRF

K

] according to




Hkeff = wH

k HkFRF


]is designed







(PRXcct

) (430)

96









PADC [175] 200




P TXcct = NTXRF P

TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX


PRXcct = NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA (435)


TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX


]NUE

[NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA

]+ LBH middotR

(436)

97







R =

Ksumk=1

Rk (439)


∣∣wHk Hkf

RFk

∣∣2sumn6=k P

nT middot∣∣wH


k

(440)



∣∣2sumn6=k P

nT middot∣∣wH


k

(441)




ηEE =sum rate


R

Ptotal (443)

98


dηSE

∣∣∣∣ηSE=

R

BW

= 0 (444)






sub and NRXsub




99





RF = 8





Beam 4N = 0 4N = 2 4N = 4 4N = 6 4N = 81 1 12107 14214 15000 12 1 09964 09928 09583 13 1 09130 08261 07917 14 1 08797 07595 07500 15 1 08797 07595 07500 16 1 09130 08261 07916 17 1 09964 09928 09583 18 1 12107 14214 15000 1

PT (W) 8 8 8 8 8



100














101

1 2 3 4 5 6 7 8

50

100

150

200

250

300

350

400

450

500

550




102




0 1 2 3 4 5 6 7 8

Total TX Power (W)

140

160

180

200

220

240

260

Su

m S

pe

ctr

al E

ffic

ien

cy (

bitss

Hz)



103


0 1 2 3 4 5 6 7 8

Total TX Power (W)

11

115

12

125

13

135

14

145

15

En

erg

y E

ffic

ien

cy (

Gb

J)


18 20 22 24 26 28 30 32 34


11

115

12

125

13

135

14

145

15

Energ

y E

ffic

iency (

GbJ

)


104



22

0

24

26

8

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

28

05 76

Transmit Power (W)

30

5

User

32

41 32

115

24

25

26

27

28

29

30

31


11

0

12

8

Energ

y E

ffic

iency (

GbJ

)

05 7

13

6

Transmit Power (W)

5

User

14

41 32

115

114

116

118

12

122

124

126

128

13

132

134


105






U1

U7

00

1

U5

AP

U8 U2

2

10

3

AP

UE

he

igh

t (m

)

05 8

4

U3

5

Walkway width (m)

6

U6

1

Walkway length (m)

U4

415 22 0


106







rk = Hk

Ksumn=1


yk = wHk Hk

Ksumn=1


Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Transmitter

1

NRX

1

NRX


User K

LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

RF

chain

RF

chain

LNA

LNA

LNA

RF

chain

yK

RF

chain

RF

chain

y1

Receivers

User 1LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

1

NRX

LNA

LNA

LNA

1

NRX

1

NRX


User K

LNA

LNA

LNA

RF

chain

yK

RF

chain

y1

Receivers

User 1LNA

LNA

LNA

1

NRX

GTX GRX

Channel



107










108

0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB




109




0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB




110


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

GbitsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


111



48 Conclusions





112




113

114

Chapter 5



51 Thesis Summary




115







116







117









118






119







120





121

Bibliography











122














123














124














125













126













127













128














129















130













131













132













133














134













135














136








137

Table of Contents

List of Acronyms

List of Symbols

List of Figures

List of Tables

List of Algorithms

Introduction

Introduction



Massive MIMO


Thesis Motivation

Thesis Objectives










Architecture









Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Conclusions


Background


System Model


Simulation Results


Deployment Layout




Network Deployment

Channel Model

Antenna Model


Conclusions


Background













Energy Efficiency



Simulation Results



Simulation Results

Conclusions


Thesis Summary

Channel Modeling

Hybrid Beamforming



Ultra-massive MIMO



Bibliography

o juri the jury



























Table of Contents

Table of Contents i

List of Acronyms iv

List of Symbols vii

List of Figures ix

List of Tables xi


1 Introduction 1

11 Introduction 1



122 Massive MIMO 4













221 Architecture 26







i






25 Conclusions 45


31 Background 47


321 System Model 48












35 Conclusions 83


41 Background 85




















48 Conclusions 112

ii




Bibliography 122

iii

List of Acronyms

1G First generation

2D Two-dimensional


3G Third generation



5G Fifth generation

6G Sixth generation


AE Antenna element




AP Access point

AS Angular spread


B5G Beyond-5G

BBU Baseband unit

BS Base station





CL Coupling loss

CN Condition number



DL Downlink


DS Delay spread



iv


GF Geometry factor









v









ZF Zero forcing

vi

List of Symbols






vii


viii

List of Figures



































ix



x

List of Tables





xi

xii

List of Algorithms



xiii

xiv

Chapter 1

Introduction


11 Introduction




1





Today



Today

6G amp

Beyond

1G 2G 3G 4G 5G1G 2G 3G 4G 5G










bull CDMA-2000












bull CDMA-2000




(NR)bull IMT-2020










bull CDMA-2000




(NR)bull IMT-2020



mMTC ITU

URLLC

eMBBmMTC ITU

URLLC

eMBB



Today

6G amp

Beyond

1G 2G 3G 4G 5G










bull CDMA-2000




(NR)bull IMT-2020



mMTC ITU

URLLC

eMBB


2





5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing

5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing


3






122 Massive MIMO


4




Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss

Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss


5





6








7



8






9







10

Legend






2- UE Distribution



10- Applications








8- Schedulers


7- HetNet + UDN


1- BS Layout








Simulator

Mobility Model

Spatial Consistency

Blockage Model



SE-EE Performance

I2X Network Layout



C-I2X System Level

Simulator


11













12






bull Book Chapter


bull Posters






13







14



Chapter 1

Introduction

Aim and Objectives



Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary



15

16

Chapter 2






17







y = Hx + n (21)

ykj1

ykjNRX

=



xj1

xjNRX

+

nk1

nkNRX

(22)

18


BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1UE1

1

NRXUE1

1

NRX

UE2UE2

1

NRXUE2

1

NRX

UEKUEK

1

NRXUEK

1

NRX

BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1

1

NRX

UE2

1

NRX

UEK

1

NRX


(a) SISO [NTX = 1 NRX = 1K = 1]


y1 = h11x1 + n1 (23)


(1 + h2PT

σ2n

) (24)





19





NTXHHH

)∣∣∣∣ (26)



(1 +

ρmax (NRX NTX)

NTX

) (27)






20




∣∣(INRX + ρHPHH)∣∣ (28)








21




(1 + ρ

NRX

NTX

) (210)









22









23


Reflector

Obstacle

Relay

BS

UE 2

UE 3UE 3

UE 1UE 1UE 1

UE 4UE 4





24






25

221 Architecture







microwave

mmWave

Control

Data

X2-interface

Vehicular Comm cell

Virtual cell

Macro BS






26






27

50 100 150 200 250 300

Frequency (GHz)

10-3

10-2

10-1

100

101

102

Spe

cific

Atte

nuat

ion

(dB

km

)


50 100 150 200 250 300

Frequency (GHz)

10-2

10-1

100

101

102

Rai

n A

ttenu

atio

n (d

Bk

m)



28



















(λ

4π

)2

dminusα (211)


(λ

4π

)2

dminusα(d) (212)

29
























30







31






32




3GPP TR 38900[85]

NYUSIM[42 101102]



Statistical



33


5G Channel Models















to the theory of




to the theory of




distributed



distributed

Geometry-based (GB)


distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Map-based (MB)


simplified 3D map

Regular-shaped (RS)



Regular-shaped (RS)





distributed



distributed




time delays




time delays



between paths



between paths

5G Channel Models









to the theory of




distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Regular-shaped (RS)





distributed




time delays



between paths






1


34




[103] [104105] [106] [107] [108]


hybrid

GBSM GBSMGGBSM

GBSMGGBSM

GBSM


05-100 045-100 6-100 05-100 -



-


no yes yes no yes


no no no no yes


yes yes yes yes yes



MiWEBA IMT-2020

[109] [110] [88] [111]



up to 70 up to 100 lt 6 57-66 05-100


[100 MHz 1GHz]

10 of fc - 216 GHz [100 MHz10 of fc]


no yes no yes no




yes yes yes yes yes


35


UE 3UE 2

eNodeBSpherical

wavefronts

UE 1

Cluster 1

Cluster 2

Cluster 3

Cluster 5 Cluster 4




36


Set

scenario

network layout

antenna parameters

Assign propagation

condition

LOS

NLOS

Calculate pathloss

Generate correlated

LSPs

delay spread

angular spread

shadow fading

K-factor

Generate

cross-polarisation

ratios

Generate

AoAs

AoDs

Perform random

coupling of rays

Generate delays

Generate

cluster powers

Draw random

initial phases

Generate

channel

coefficients

Apply

pathloss and

shadowing



37









38


H =

radicNRXNTX

L

Lsuml=1


)middotandHTX

(φTXl θTXl

)middotaRX

(φRXl θRXl

)middotaHTX

(φTXl θTXl

)

(215)


(φRXl θRXl

)and aTX

(φTXl θTXl



)and andTX

(φTXl θTXl







39


∣∣wHRFHfRF


wopt fopt)

= arg max∣∣wH

RFHfRF

∣∣subject to

wi =

radic1


fl =

radic1


(216)

RF

chain


PAPA

PAPA

PAPA

1

2

NTX

PSs

s

RF

chain


PA

PA

PA

1

2

NTX

PSs

s



40








PA

PA

1

2

NTX

PAPA

RF chainRF chain

RF chainRF chain

RF chainRF chain

Ns

Baseband

Digital

Beamformer

FDBF

PA

PA

1

2

NTX

PA

RF chain

RF chain

RF chain

Ns

Baseband

Digital

Beamformer

FDBF




FMFDBF = HH (217)

FZFDBF = HH

(HHH

)minus1 (218)

FWFDBF = HH

(HHH +

σ2nNs

PRXINRX

)minus1

(219)


41






∥∥FkDBF

∥∥2

F= Ns The


yk = Hk

Ksumn=1

FnDBF sn + nk (221)









42


Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX

Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSsNs

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSs



maxFRF WRFk forallk

Ksumk=1

∥∥WHRFkHkFRF

∥∥2

F

subject to

FHRFFRF = INTX WH


(222)


Heffk = [HTeff1 H

Teffk H

TeffK ]T (223)




FMRTBB = HH

effk (224)

FZFBB = HH


43



rk = Hk

Ksumn=1

FRFn FBB

n sn + nk (226)

yk = WHBBkWH

RFkHk

Ksumn=1

FRFn FBB

n sn + WHBBkWH

RFknk (227)









antennas





no multi-user




44

25 Conclusions



45

46

Chapter 3



31 Background



47





321 System Model


48

55 timesISD (m)

55

timesIS

D (

m)

BuildingBase

Station

Outdoor

User

Signal

link

Interfering

linkIndoor

User






49







(min

(18

d2D 1

)(1minus exp

(minusd2D

63

))+ exp

(minusd2D

63

))(1 + C (d2DhRX

))


(min

(18

d2D 1

)(1minus exp

(minusd2D

36

))+ exp

(minusd2D

36

))

UMa-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

63

)(1minus 18

d2D

))(1 + C (d2DhRX

)) 18 lt d2D

UMi-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

36

)(1minus 18

d2D

)) 18 lt d2D

where C (d2D hRX) =

1 d2D le 18


)3exp

(minusd2D150

) 18 lt d2D

C prime (hRX) =


10

)15

13 lt hRX le 23


50

Tab

le3

3P

ath

loss

mod

els

(ad

apte

dfr

om[8

485]

)

Scen

ari

oP

ath

loss

(dB

)σSF

(dB

)

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

5000

4

UM

a-

microW

ave

PLNLOS

=69

51+

390

9lo

g10(d

3D

)+

20lo

g10(fc)

+75

log

10(hRX

)minus

06hRXminus

00

825(hRX

)26

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

3

UM

i-micro

Wav

ePLNLOS

=23

15+

367

log

10(d

3D

)+

26lo

g10(fc)minus

03hUE

4

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

32

4+

20

log

10(d

3D

)+

20lo

g10(fc)

d10ltd

2DltdBP

32

4+

40

log

10(d

3D

)+

20lo

g10(fc)minus

10lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

500

04

UM

a-

mm

Wav

ePLNLOS

=14

44+

390

8lo

g10(d

3D

)+

20lo

g10(fc)minus

06h

RX

6

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

PLLOS

=

32

4+

21

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

324

+40

log

10(d

3D

)+

20lo

g10(fc)minus

95

log

10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

4

UM

i-m

mW

ave

PLNLOS

=22

85+

353

log

10(d

3D

)+

213

log

10(fc)minus

03hUE

78

2

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

NO

TE

S

f cis

inG

Hzd

2Dd

3D

andhRX

are

inm

eter

sdBP

=32

0((hRXminus

1)f c

)fo

rU

Ma

anddBP

=120

((hRXminus

1)fc)

for

UM

idBP

isth

eb

reak

-poi

nt

dis

tance

[84]

[8

5]an

dPLwall

isth

ew

all

pen

etra

tion

loss

base

don

Tab

les

74

3-(1

amp2)

in[8

5]

51




χk =

0 k rarr outdoor

1 k rarr indoor(32)



fl nmaxfl ]) (33)









52







dkj2D =radic


dkj3D =





ξkj =

0 k j rarr NLOS

1 k j rarr LOS(311)


χk ξkj =

0 0 rarr PLNLOS

0 1 rarr PLLOS



(312)




RSRPkj = P kjT +Gkj




53


Parameter Values


AEH = minusmin

12

(φ

65o 30 dB


AEV = minusmin

12

(θ minus 90o

65o 30 dB




8 dBi



(a) Coupling Loss



RSRPkj = P kjT +Gkj




54

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS


(b) Geometry Factor


55






Nk = No + 10 log10Bk +NF (320)



56

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1E

CD


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF






57


-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF



-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF








58







59



Tk(t)



Network ele-ments


Small cell (SC)







3GPP TR 38900[85]



[(Nintsumn=1

In

)+ (No + 10 log10B +NF )

]

(321)


60















InterferencesumNint









bandwidth





61



0

200

400

600

800

Cel

l cap

acit

y (M

bp

s)

1000

1200

1400

125

1600

250500

1000

average macro

average small

total



62


0

20

40

60

80

Cel

l cap

acit

y (M

bp

s) 100

120

140

125250

5001000







63


Es

0

20

40

60

0

80

Ave

rag

e U

E T

hro

ug

hp

ut

(Mb

ps)

125

100

120

140

250 80500

1001000


of indoor U


0

05

Sp

ectr

al e

ffic

ien

cy (

bp

sH

z)

125

1

15

0250

80500

1001000


64





30 32 34 36 38 40 42 44 46 48


0

200

400

600

800

1000

1200

1400

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)




65








66

0 20 40 60 80 100


0

100

200

300

400

500

600

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)



= 35 dBm


125 250 375 500 625 750 875 1000


0

50

100

150

200

250

300

350

400

450

500

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)


PTX

= 35 dBm


67







68






69


PLOS(d3D) =

[min

(27

d 1

)(1minus eminus

d71

)+ eminus

d71

]2

(322)

342 Channel Model



hdir(t φ θ) =

Nclsumc=1

Nspcsums=1


(323)


G(φ θ) = max

(G0e

αφ2+βθ2 Go100

) (324)


3dBθ23dB

α =4 ln(2)

φ23dB

β =4 ln(2)

θ23dB

(325)

70






) (327)

ϕcs = 2π

(fcτcs +

vRX cos (φcs)

λ4 t

) (328)


343 Antenna Model



H =


middotNclsumc=1

Nspcsums=1


j2π

(fcτcs+

vRX cos(φcs)λ

4t)middot aRX

(φRXcs

)aHTX

(φTXcs

)

(329)

aRX(φRXcs

)=

1radicNRX



aTX(φTXcs

)=

1radicNTX




71






φRX3dB 65 65 10


(a) Path Loss



PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (333)

72


PL = 10κ log10 (d3D) + Υ + 15

(d3D

1000

) (334)


50 55 60 65 70 75 80 85 90 95

Path Loss (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave



73

0 2 4 6 8 10 12 14 16 18 20


40

50

60

70

80

90

100

110

120

Path

Loss (

dB








0 50 100 150 200 250 300 350 400 450 500


40

50

60

70

80

90

100

110

120

Path

Loss (

dB

)








74




(b) Rician K-Factor


KF =PLOSsumPNLOS

=PRXc=1s=1(sumNcl

c=1

sumNspcs=1 PRXcs

)minus PRXc=1s=1

(336)


75

-20 0 20 40 60 80 100 120

K-Factor (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave




τRMS =


sumNspcs=1 τ2

csPRXcssumNclc=1

sumNspcs=1 PRXcs

minus

(sumNclc=1


c=1

sumNspcs=1 PRXcs

)2

(337)

Bc asymp1

2πτRMS (338)


76


2 4 6 8 10 12 14 16 18 20 22


0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave




77

1 2 3 4 5 6

Number of Clusters

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave


0 5 10 15 20 25 30 35 40 45


0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave


78



0

-150

-100

mmWave

Re

ce

ive

d P

ow

er

(dB

)

-50


500

Technology

0

LTE-A

1000 DSRC


0

-150

-100

mmWave200

Re

ce

ive

d P

ow

er

(dB

)

-50


Technology

0

LTE-A400

600 DSRC


79








80

2 4 6 8 10 12 14 16 18 20


1

2

3

4

5

6

7

8C

hannel R

ank

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)


2 4 6 8 10 12 14 16 18 20


0

20

40

60

80

100

120

140

160

Channel C

onditio

n N

um

ber

(dB

)

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)


81

(g) Data Rate




R = B middot log2

(1 + ρRminus1


where Rn = σ2nw


0 2 4 6 8 10 12

Data Rate (Gbps)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave



82

35 Conclusions







83

84

Chapter 4



41 Background




85






PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (41)


86

hdir(t φ) =Lsuml=1



H =

radicNRXNTX

LmiddotLsuml=1


j2π

(fcτl+

vRX cos(φl)λ

4t)middot aRX

(φRXl

)aHTX

(φTXl

) (44)

aRX(φRXl

)=

1radicNRX



aTX(φTXl

)=

1radicNTX







87

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs







88


Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs


Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K

Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K


89










0 1 2 NTXNRF

For the gener-


4N =


(NTXNRF

minus 1)]

rarr overlapped

NTXNRF

rarr sub-connected

(47)


90

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Fig

ure

44

Gen

eral

ized

HB

Far

chit

ectu

re

91




NPS =

NRFsumk=1

NkPS = NRF timesNk

PS (49)






PS = |mk|


M =




0 m4N+Nsub2

0 0 0 mJK

(411)


B =




0 1

0 0 0 1

(412)

92


g =

g1

g2

gNTX

gj =

NRFsumk=1




∣∣∣ =




(414)



P kb =PTNTX

sumjisinmk

(gj)minus1

(415)

PT =

NRFsumk=1

P kb (416)






93

x = FRFFBBs (418)




yk = wHk Hk

Ksumn=1








[fRFk

]=

1radicNTX

ejφk = aTX(φTXl

) (420)

[wRFk

]=

1radicNRX

ejφk = aRX(φRXl

) (421)


(φTXl

)and

aRX(φRXl




] (422)

WRF =[wRF

1 wRF2 wRF

K

] (423)


94




] (424)

FBB = HHeff

(HeffHH

eff

)minus1 (425)



fBBk =fBBk


||FRFFBB||2F = K (427)










95



OutputsFBB FRF WRF




fRFk = akTX

wRFk = akRX


]and WRF =

[wRF

1 wRF2 wRF

K

] according to




Hkeff = wH

k HkFRF


]is designed







(PRXcct

) (430)

96









PADC [175] 200




P TXcct = NTXRF P

TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX


PRXcct = NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA (435)


TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX


]NUE

[NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA

]+ LBH middotR

(436)

97







R =

Ksumk=1

Rk (439)


∣∣wHk Hkf

RFk

∣∣2sumn6=k P

nT middot∣∣wH


k

(440)



∣∣2sumn6=k P

nT middot∣∣wH


k

(441)




ηEE =sum rate


R

Ptotal (443)

98


dηSE

∣∣∣∣ηSE=

R

BW

= 0 (444)






sub and NRXsub




99





RF = 8





Beam 4N = 0 4N = 2 4N = 4 4N = 6 4N = 81 1 12107 14214 15000 12 1 09964 09928 09583 13 1 09130 08261 07917 14 1 08797 07595 07500 15 1 08797 07595 07500 16 1 09130 08261 07916 17 1 09964 09928 09583 18 1 12107 14214 15000 1

PT (W) 8 8 8 8 8



100














101

1 2 3 4 5 6 7 8

50

100

150

200

250

300

350

400

450

500

550




102




0 1 2 3 4 5 6 7 8

Total TX Power (W)

140

160

180

200

220

240

260

Su

m S

pe

ctr

al E

ffic

ien

cy (

bitss

Hz)



103


0 1 2 3 4 5 6 7 8

Total TX Power (W)

11

115

12

125

13

135

14

145

15

En

erg

y E

ffic

ien

cy (

Gb

J)


18 20 22 24 26 28 30 32 34


11

115

12

125

13

135

14

145

15

Energ

y E

ffic

iency (

GbJ

)


104



22

0

24

26

8

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

28

05 76

Transmit Power (W)

30

5

User

32

41 32

115

24

25

26

27

28

29

30

31


11

0

12

8

Energ

y E

ffic

iency (

GbJ

)

05 7

13

6

Transmit Power (W)

5

User

14

41 32

115

114

116

118

12

122

124

126

128

13

132

134


105






U1

U7

00

1

U5

AP

U8 U2

2

10

3

AP

UE

he

igh

t (m

)

05 8

4

U3

5

Walkway width (m)

6

U6

1

Walkway length (m)

U4

415 22 0


106







rk = Hk

Ksumn=1


yk = wHk Hk

Ksumn=1


Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Transmitter

1

NRX

1

NRX


User K

LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

RF

chain

RF

chain

LNA

LNA

LNA

RF

chain

yK

RF

chain

RF

chain

y1

Receivers

User 1LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

1

NRX

LNA

LNA

LNA

1

NRX

1

NRX


User K

LNA

LNA

LNA

RF

chain

yK

RF

chain

y1

Receivers

User 1LNA

LNA

LNA

1

NRX

GTX GRX

Channel



107










108

0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB




109




0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB




110


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

GbitsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


111



48 Conclusions





112




113

114

Chapter 5



51 Thesis Summary




115







116







117









118






119







120





121

Bibliography











122














123














124














125













126













127













128














129















130













131













132













133














134













135














136








137

Table of Contents

List of Acronyms

List of Symbols

List of Figures

List of Tables

List of Algorithms

Introduction

Introduction



Massive MIMO


Thesis Motivation

Thesis Objectives










Architecture









Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Conclusions


Background


System Model


Simulation Results


Deployment Layout




Network Deployment

Channel Model

Antenna Model


Conclusions


Background













Energy Efficiency



Simulation Results



Simulation Results

Conclusions


Thesis Summary

Channel Modeling

Hybrid Beamforming



Ultra-massive MIMO



Bibliography









Table of Contents

Table of Contents i

List of Acronyms iv

List of Symbols vii

List of Figures ix

List of Tables xi


1 Introduction 1

11 Introduction 1



122 Massive MIMO 4













221 Architecture 26







i






25 Conclusions 45


31 Background 47


321 System Model 48












35 Conclusions 83


41 Background 85




















48 Conclusions 112

ii




Bibliography 122

iii

List of Acronyms

1G First generation

2D Two-dimensional


3G Third generation



5G Fifth generation

6G Sixth generation


AE Antenna element




AP Access point

AS Angular spread


B5G Beyond-5G

BBU Baseband unit

BS Base station





CL Coupling loss

CN Condition number



DL Downlink


DS Delay spread



iv


GF Geometry factor









v









ZF Zero forcing

vi

List of Symbols






vii


viii

List of Figures



































ix



x

List of Tables





xi

xii

List of Algorithms



xiii

xiv

Chapter 1

Introduction


11 Introduction




1





Today



Today

6G amp

Beyond

1G 2G 3G 4G 5G1G 2G 3G 4G 5G










bull CDMA-2000












bull CDMA-2000




(NR)bull IMT-2020










bull CDMA-2000




(NR)bull IMT-2020



mMTC ITU

URLLC

eMBBmMTC ITU

URLLC

eMBB



Today

6G amp

Beyond

1G 2G 3G 4G 5G










bull CDMA-2000




(NR)bull IMT-2020



mMTC ITU

URLLC

eMBB


2





5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing

5G

Massive

MIMO

Wireless

Software Defined

Networking

(WSDN)

Device-to-Device

Communications

(D2D)

Network

Function

Virtualization

(NFV)

Millimeter-wave

amp

Terahertz bands

(mmWave THz)

Internet of

Things (IoT)Green

Communications

Ultra-Dense

Networks

(UDN)

Radio Access

Techniques

Big Data amp

Mobile Cloud

Computing


3






122 Massive MIMO


4




Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss

Ultra-Dense

Network

Massive

MIMO

mmWave

Communication

Short wavelength



lower pathloss


5





6








7



8






9







10

Legend






2- UE Distribution



10- Applications








8- Schedulers


7- HetNet + UDN


1- BS Layout








Simulator

Mobility Model

Spatial Consistency

Blockage Model



SE-EE Performance

I2X Network Layout



C-I2X System Level

Simulator


11













12






bull Book Chapter


bull Posters






13







14



Chapter 1

Introduction

Aim and Objectives



Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary


Chapter 1

Introduction

Aim and Objectives



Chapter 2


5G Channel Models

Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Chapter 3

3D Channel Modeling


UDN mmWave Channel



Chapter 4




SE-EE Analysis


Chapter 5

Conclusions

Thesis Summary



15

16

Chapter 2






17







y = Hx + n (21)

ykj1

ykjNRX

=



xj1

xjNRX

+

nk1

nkNRX

(22)

18


BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1UE1

1

NRXUE1

1

NRX

UE2UE2

1

NRXUE2

1

NRX

UEKUEK

1

NRXUEK

1

NRX

BS1

Wireless

Channel

1

NTX

BSJ

1

NTX

UE1

1

NRX

UE2

1

NRX

UEK

1

NRX


(a) SISO [NTX = 1 NRX = 1K = 1]


y1 = h11x1 + n1 (23)


(1 + h2PT

σ2n

) (24)





19





NTXHHH

)∣∣∣∣ (26)



(1 +

ρmax (NRX NTX)

NTX

) (27)






20




∣∣(INRX + ρHPHH)∣∣ (28)








21




(1 + ρ

NRX

NTX

) (210)









22









23


Reflector

Obstacle

Relay

BS

UE 2

UE 3UE 3

UE 1UE 1UE 1

UE 4UE 4





24






25

221 Architecture







microwave

mmWave

Control

Data

X2-interface

Vehicular Comm cell

Virtual cell

Macro BS






26






27

50 100 150 200 250 300

Frequency (GHz)

10-3

10-2

10-1

100

101

102

Spe

cific

Atte

nuat

ion

(dB

km

)


50 100 150 200 250 300

Frequency (GHz)

10-2

10-1

100

101

102

Rai

n A

ttenu

atio

n (d

Bk

m)



28



















(λ

4π

)2

dminusα (211)


(λ

4π

)2

dminusα(d) (212)

29
























30







31






32




3GPP TR 38900[85]

NYUSIM[42 101102]



Statistical



33


5G Channel Models















to the theory of




to the theory of




distributed



distributed

Geometry-based (GB)


distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Map-based (MB)


simplified 3D map

Regular-shaped (RS)



Regular-shaped (RS)





distributed



distributed




time delays




time delays



between paths



between paths

5G Channel Models









to the theory of




distributed

Geometry-based (GB)


distributed

Map-based (MB)


simplified 3D map

Regular-shaped (RS)





distributed




time delays



between paths






1


34




[103] [104105] [106] [107] [108]


hybrid

GBSM GBSMGGBSM

GBSMGGBSM

GBSM


05-100 045-100 6-100 05-100 -



-


no yes yes no yes


no no no no yes


yes yes yes yes yes



MiWEBA IMT-2020

[109] [110] [88] [111]



up to 70 up to 100 lt 6 57-66 05-100


[100 MHz 1GHz]

10 of fc - 216 GHz [100 MHz10 of fc]


no yes no yes no




yes yes yes yes yes


35


UE 3UE 2

eNodeBSpherical

wavefronts

UE 1

Cluster 1

Cluster 2

Cluster 3

Cluster 5 Cluster 4




36


Set

scenario

network layout

antenna parameters

Assign propagation

condition

LOS

NLOS

Calculate pathloss

Generate correlated

LSPs

delay spread

angular spread

shadow fading

K-factor

Generate

cross-polarisation

ratios

Generate

AoAs

AoDs

Perform random

coupling of rays

Generate delays

Generate

cluster powers

Draw random

initial phases

Generate

channel

coefficients

Apply

pathloss and

shadowing



37









38


H =

radicNRXNTX

L

Lsuml=1


)middotandHTX

(φTXl θTXl

)middotaRX

(φRXl θRXl

)middotaHTX

(φTXl θTXl

)

(215)


(φRXl θRXl

)and aTX

(φTXl θTXl



)and andTX

(φTXl θTXl







39


∣∣wHRFHfRF


wopt fopt)

= arg max∣∣wH

RFHfRF

∣∣subject to

wi =

radic1


fl =

radic1


(216)

RF

chain


PAPA

PAPA

PAPA

1

2

NTX

PSs

s

RF

chain


PA

PA

PA

1

2

NTX

PSs

s



40








PA

PA

1

2

NTX

PAPA

RF chainRF chain

RF chainRF chain

RF chainRF chain

Ns

Baseband

Digital

Beamformer

FDBF

PA

PA

1

2

NTX

PA

RF chain

RF chain

RF chain

Ns

Baseband

Digital

Beamformer

FDBF




FMFDBF = HH (217)

FZFDBF = HH

(HHH

)minus1 (218)

FWFDBF = HH

(HHH +

σ2nNs

PRXINRX

)minus1

(219)


41






∥∥FkDBF

∥∥2

F= Ns The


yk = Hk

Ksumn=1

FnDBF sn + nk (221)









42


Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX

Ns

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSsNs

Baseband

Digital

Beamformer

FBB

RF chain 1

RF chain K

PA

PA

PA

1

2

NTX


PSs



maxFRF WRFk forallk

Ksumk=1

∥∥WHRFkHkFRF

∥∥2

F

subject to

FHRFFRF = INTX WH


(222)


Heffk = [HTeff1 H

Teffk H

TeffK ]T (223)




FMRTBB = HH

effk (224)

FZFBB = HH


43



rk = Hk

Ksumn=1

FRFn FBB

n sn + nk (226)

yk = WHBBkWH

RFkHk

Ksumn=1

FRFn FBB

n sn + WHBBkWH

RFknk (227)









antennas





no multi-user




44

25 Conclusions



45

46

Chapter 3



31 Background



47





321 System Model


48

55 timesISD (m)

55

timesIS

D (

m)

BuildingBase

Station

Outdoor

User

Signal

link

Interfering

linkIndoor

User






49







(min

(18

d2D 1

)(1minus exp

(minusd2D

63

))+ exp

(minusd2D

63

))(1 + C (d2DhRX

))


(min

(18

d2D 1

)(1minus exp

(minusd2D

36

))+ exp

(minusd2D

36

))

UMa-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

63

)(1minus 18

d2D

))(1 + C (d2DhRX

)) 18 lt d2D

UMi-mmWave PLOS =

1 d2D le 18(

18d2D

+ exp(minusd2D

36

)(1minus 18

d2D

)) 18 lt d2D

where C (d2D hRX) =

1 d2D le 18


)3exp

(minusd2D150

) 18 lt d2D

C prime (hRX) =


10

)15

13 lt hRX le 23


50

Tab

le3

3P

ath

loss

mod

els

(ad

apte

dfr

om[8

485]

)

Scen

ari

oP

ath

loss

(dB

)σSF

(dB

)

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

5000

4

UM

a-

microW

ave

PLNLOS

=69

51+

390

9lo

g10(d

3D

)+

20lo

g10(fc)

+75

log

10(hRX

)minus

06hRXminus

00

825(hRX

)26

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

28

+22

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

28

+40

log

10(d

3D

)+

20lo

g10(fc)minus

9lo

g10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

3

UM

i-micro

Wav

ePLNLOS

=23

15+

367

log

10(d

3D

)+

26lo

g10(fc)minus

03hUE

4

PLO

2I

=PLLOSN

LOS

+20

+0

5(d

2Dminusin

)7

PLLOS

=

32

4+

20

log

10(d

3D

)+

20lo

g10(fc)

d10ltd

2DltdBP

32

4+

40

log

10(d

3D

)+

20lo

g10(fc)minus

10lo

g10((dBP

)2+

(25minushRX

)2)

dBPltd

2Dlt

500

04

UM

a-

mm

Wav

ePLNLOS

=14

44+

390

8lo

g10(d

3D

)+

20lo

g10(fc)minus

06h

RX

6

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

PLLOS

=

32

4+

21

log

10(d

3D

)+

20lo

g10(fc)

10ltd

2DltdBP

324

+40

log

10(d

3D

)+

20lo

g10(fc)minus

95

log

10((dBP

)2+

(10minushRX

)2)

dBPltd

2Dlt

5000

4

UM

i-m

mW

ave

PLNLOS

=22

85+

353

log

10(d

3D

)+

213

log

10(fc)minus

03hUE

78

2

PLO

2I

=PLLOSN

LOS

+PLwall

+05

(d2Dminusin

)7

NO

TE

S

f cis

inG

Hzd

2Dd

3D

andhRX

are

inm

eter

sdBP

=32

0((hRXminus

1)f c

)fo

rU

Ma

anddBP

=120

((hRXminus

1)fc)

for

UM

idBP

isth

eb

reak

-poi

nt

dis

tance

[84]

[8

5]an

dPLwall

isth

ew

all

pen

etra

tion

loss

base

don

Tab

les

74

3-(1

amp2)

in[8

5]

51




χk =

0 k rarr outdoor

1 k rarr indoor(32)



fl nmaxfl ]) (33)









52







dkj2D =radic


dkj3D =





ξkj =

0 k j rarr NLOS

1 k j rarr LOS(311)


χk ξkj =

0 0 rarr PLNLOS

0 1 rarr PLLOS



(312)




RSRPkj = P kjT +Gkj




53


Parameter Values


AEH = minusmin

12

(φ

65o 30 dB


AEV = minusmin

12

(θ minus 90o

65o 30 dB




8 dBi



(a) Coupling Loss



RSRPkj = P kjT +Gkj




54

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS

-250 -200 -150 -100 -50 0Coupling Loss (dB)

0

02

04

06

08

1

EC

DF


LOS+NLOSLOSNLOS


(b) Geometry Factor


55






Nk = No + 10 log10Bk +NF (320)



56

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1E

CD


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF


SNRSIRSINR

-100 -50 0 50 100SNRSIRSINR (dB)

0

02

04

06

08

1

EC

DF






57


-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF



-80 -60 -40 -20 0 20 40SINR (dB)

0

01

02

03

04

05

06

07

08

09

1

EC

DF








58







59



Tk(t)



Network ele-ments


Small cell (SC)







3GPP TR 38900[85]



[(Nintsumn=1

In

)+ (No + 10 log10B +NF )

]

(321)


60















InterferencesumNint









bandwidth





61



0

200

400

600

800

Cel

l cap

acit

y (M

bp

s)

1000

1200

1400

125

1600

250500

1000

average macro

average small

total



62


0

20

40

60

80

Cel

l cap

acit

y (M

bp

s) 100

120

140

125250

5001000







63


Es

0

20

40

60

0

80

Ave

rag

e U

E T

hro

ug

hp

ut

(Mb

ps)

125

100

120

140

250 80500

1001000


of indoor U


0

05

Sp

ectr

al e

ffic

ien

cy (

bp

sH

z)

125

1

15

0250

80500

1001000


64





30 32 34 36 38 40 42 44 46 48


0

200

400

600

800

1000

1200

1400

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)




65








66

0 20 40 60 80 100


0

100

200

300

400

500

600

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)



= 35 dBm


125 250 375 500 625 750 875 1000


0

50

100

150

200

250

300

350

400

450

500

Ave

rag

e sm

all c

ell c

apac

ity

(Mb

ps)


PTX

= 35 dBm


67







68






69


PLOS(d3D) =

[min

(27

d 1

)(1minus eminus

d71

)+ eminus

d71

]2

(322)

342 Channel Model



hdir(t φ θ) =

Nclsumc=1

Nspcsums=1


(323)


G(φ θ) = max

(G0e

αφ2+βθ2 Go100

) (324)


3dBθ23dB

α =4 ln(2)

φ23dB

β =4 ln(2)

θ23dB

(325)

70






) (327)

ϕcs = 2π

(fcτcs +

vRX cos (φcs)

λ4 t

) (328)


343 Antenna Model



H =


middotNclsumc=1

Nspcsums=1


j2π

(fcτcs+

vRX cos(φcs)λ

4t)middot aRX

(φRXcs

)aHTX

(φTXcs

)

(329)

aRX(φRXcs

)=

1radicNRX



aTX(φTXcs

)=

1radicNTX




71






φRX3dB 65 65 10


(a) Path Loss



PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (333)

72


PL = 10κ log10 (d3D) + Υ + 15

(d3D

1000

) (334)


50 55 60 65 70 75 80 85 90 95

Path Loss (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave



73

0 2 4 6 8 10 12 14 16 18 20


40

50

60

70

80

90

100

110

120

Path

Loss (

dB








0 50 100 150 200 250 300 350 400 450 500


40

50

60

70

80

90

100

110

120

Path

Loss (

dB

)








74




(b) Rician K-Factor


KF =PLOSsumPNLOS

=PRXc=1s=1(sumNcl

c=1

sumNspcs=1 PRXcs

)minus PRXc=1s=1

(336)


75

-20 0 20 40 60 80 100 120

K-Factor (dB)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave




τRMS =


sumNspcs=1 τ2

csPRXcssumNclc=1

sumNspcs=1 PRXcs

minus

(sumNclc=1


c=1

sumNspcs=1 PRXcs

)2

(337)

Bc asymp1

2πτRMS (338)


76


2 4 6 8 10 12 14 16 18 20 22


0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave




77

1 2 3 4 5 6

Number of Clusters

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave


0 5 10 15 20 25 30 35 40 45


0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave


78



0

-150

-100

mmWave

Re

ce

ive

d P

ow

er

(dB

)

-50


500

Technology

0

LTE-A

1000 DSRC


0

-150

-100

mmWave200

Re

ce

ive

d P

ow

er

(dB

)

-50


Technology

0

LTE-A400

600 DSRC


79








80

2 4 6 8 10 12 14 16 18 20


1

2

3

4

5

6

7

8C

hannel R

ank

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)


2 4 6 8 10 12 14 16 18 20


0

20

40

60

80

100

120

140

160

Channel C

onditio

n N

um

ber

(dB

)

DSRC (2 2)

LTE-A (4 4)

mmWave (16 64)


81

(g) Data Rate




R = B middot log2

(1 + ρRminus1


where Rn = σ2nw


0 2 4 6 8 10 12

Data Rate (Gbps)

0

01

02

03

04

05

06

07

08

09

1

CD

F

DSRC

LTE-A

mmWave



82

35 Conclusions







83

84

Chapter 4



41 Background




85






PLeff = 20 log10

(4πfcc

)+ 10n log10 (d3D) +X(0 σ) (41)


86

hdir(t φ) =Lsuml=1



H =

radicNRXNTX

LmiddotLsuml=1


j2π

(fcτl+

vRX cos(φl)λ

4t)middot aRX

(φRXl

)aHTX

(φTXl

) (44)

aRX(φRXl

)=

1radicNRX



aTX(φTXl

)=

1radicNTX







87

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

PSs







88


Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs

Baseband

precoder

(FBB)

RF chain

1

RF chain

K


s1

s2

sK

PA

PA

PA

PA

PA

PA

Subarray 1

Subarray KPSs


Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PAPA

PAPA

PAPA

PAPA

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K

Baseband

precoder

(FBB)

RF chain

1


s1

s2

sK

PA

PA

PA

PA

Subarray 1

Subarray K

RF chain

K


89










0 1 2 NTXNRF

For the gener-


4N =


(NTXNRF

minus 1)]

rarr overlapped

NTXNRF

rarr sub-connected

(47)


90

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

AL

NA

LN

AL

NA

LN

AL

NA

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Ba

se

ban

d

Beam

form

er

(FB

B)

s1

s2

sK

RF

Ch

ain

1

DA

CX

LP

F

LO

MIX

RF

Ch

ain

2

DA

CX

LP

F

LO

MIX

RF

Ch

ain

K

DA

CX

LP

F

LO

MIX

N N

Su

barr

ay

sp

acin

g

Sp

litte

rC

om

bin

er

PA

1 2

NT

X

PS

RF

Be

am

form

er

(FR

F)

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

y1

User

1 R

F C

ha

in

LN

A

LN

A

LN

A

LP

FX

AD

C

LO

MIX

1 2

NR

X

yK

User

K R

F C

ha

in

RF

Beam

form

er

(WR

F)

Tra

nsm

itte

r em

plo

yin

g H

BF

R

eceiv

ers

em

plo

yin

g A

BF

Fig

ure

44

Gen

eral

ized

HB

Far

chit

ectu

re

91




NPS =

NRFsumk=1

NkPS = NRF timesNk

PS (49)






PS = |mk|


M =




0 m4N+Nsub2

0 0 0 mJK

(411)


B =




0 1

0 0 0 1

(412)

92


g =

g1

g2

gNTX

gj =

NRFsumk=1




∣∣∣ =




(414)



P kb =PTNTX

sumjisinmk

(gj)minus1

(415)

PT =

NRFsumk=1

P kb (416)






93

x = FRFFBBs (418)




yk = wHk Hk

Ksumn=1








[fRFk

]=

1radicNTX

ejφk = aTX(φTXl

) (420)

[wRFk

]=

1radicNRX

ejφk = aRX(φRXl

) (421)


(φTXl

)and

aRX(φRXl




] (422)

WRF =[wRF

1 wRF2 wRF

K

] (423)


94




] (424)

FBB = HHeff

(HeffHH

eff

)minus1 (425)



fBBk =fBBk


||FRFFBB||2F = K (427)










95



OutputsFBB FRF WRF




fRFk = akTX

wRFk = akRX


]and WRF =

[wRF

1 wRF2 wRF

K

] according to




Hkeff = wH

k HkFRF


]is designed







(PRXcct

) (430)

96









PADC [175] 200




P TXcct = NTXRF P

TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX


PRXcct = NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA (435)


TXRFC +NTX

RF NTXsub P

TXPS +NTXPPA +NTX


]NUE

[NRXRF P

RXRFC +NRX

RF NRXsub P

RXPS +NRXPLNA

]+ LBH middotR

(436)

97







R =

Ksumk=1

Rk (439)


∣∣wHk Hkf

RFk

∣∣2sumn6=k P

nT middot∣∣wH


k

(440)



∣∣2sumn6=k P

nT middot∣∣wH


k

(441)




ηEE =sum rate


R

Ptotal (443)

98


dηSE

∣∣∣∣ηSE=

R

BW

= 0 (444)






sub and NRXsub




99





RF = 8





Beam 4N = 0 4N = 2 4N = 4 4N = 6 4N = 81 1 12107 14214 15000 12 1 09964 09928 09583 13 1 09130 08261 07917 14 1 08797 07595 07500 15 1 08797 07595 07500 16 1 09130 08261 07916 17 1 09964 09928 09583 18 1 12107 14214 15000 1

PT (W) 8 8 8 8 8



100














101

1 2 3 4 5 6 7 8

50

100

150

200

250

300

350

400

450

500

550




102




0 1 2 3 4 5 6 7 8

Total TX Power (W)

140

160

180

200

220

240

260

Su

m S

pe

ctr

al E

ffic

ien

cy (

bitss

Hz)



103


0 1 2 3 4 5 6 7 8

Total TX Power (W)

11

115

12

125

13

135

14

145

15

En

erg

y E

ffic

ien

cy (

Gb

J)


18 20 22 24 26 28 30 32 34


11

115

12

125

13

135

14

145

15

Energ

y E

ffic

iency (

GbJ

)


104



22

0

24

26

8

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

28

05 76

Transmit Power (W)

30

5

User

32

41 32

115

24

25

26

27

28

29

30

31


11

0

12

8

Energ

y E

ffic

iency (

GbJ

)

05 7

13

6

Transmit Power (W)

5

User

14

41 32

115

114

116

118

12

122

124

126

128

13

132

134


105






U1

U7

00

1

U5

AP

U8 U2

2

10

3

AP

UE

he

igh

t (m

)

05 8

4

U3

5

Walkway width (m)

6

U6

1

Walkway length (m)

U4

415 22 0


106







rk = Hk

Ksumn=1


yk = wHk Hk

Ksumn=1


Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Baseband

precoder

(FBB)

RF chain

1

RF chain

K

PA

PA

PA


s1

s2

sK

1

NTX

Transmitter

1

NRX

1

NRX


User K

LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

RF

chain

RF

chain

LNA

LNA

LNA

RF

chain

yK

RF

chain

RF

chain

y1

Receivers

User 1LNALNA

LNALNA

LNALNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

LNA

1

NRX

LNA

LNA

LNA

1

NRX

1

NRX


User K

LNA

LNA

LNA

RF

chain

yK

RF

chain

y1

Receivers

User 1LNA

LNA

LNA

1

NRX

GTX GRX

Channel



107










108

0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB




109




0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB




110


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

Gb

itsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


0 5 10 15 20 25 30 35 40 45 50


0

05

1

15

2

25

3

35

4

En

erg

y E

ffic

ien

cy (

GbitsJ

)

0

5

10

15

20

25

30

35

40

Sp

ectr

al E

ffic

ien

cy [

(bitss

)H

z]

EE - AN-BST

EE - HYB-ZF

EE - SVD-UB

SE - AN-BST

SE - HYB-ZF

SE - SVD-UB


111



48 Conclusions





112




113

114

Chapter 5



51 Thesis Summary




115







116







117









118






119







120





121

Bibliography











122














123














124














125













126













127













128














129















130













131













132













133














134













135














136








137

Table of Contents

List of Acronyms

List of Symbols

List of Figures

List of Tables

List of Algorithms

Introduction

Introduction



Massive MIMO


Thesis Motivation

Thesis Objectives










Architecture









Analog Beamforming

Digital Beamforming

Hybrid Beamforming

Conclusions


Background


System Model


Simulation Results


Deployment Layout




Network Deployment

Channel Model

Antenna Model


Conclusions


Background













Energy Efficiency



Simulation Results



Simulation Results

Conclusions


Thesis Summary

Channel Modeling

Hybrid Beamforming



Ultra-massive MIMO



Bibliography

Sherif Adeshina Ondas milim etricas e MIMO massivo para ...

Documents

Transcript of Sherif Adeshina Ondas milim etricas e MIMO massivo para ...